2005-04-21 16:53:53 +02:00
|
|
|
/* When a needed block is not in the cache, it must be fetched from the disk.
|
|
|
|
* Special character files also require I/O. The routines for these are here.
|
|
|
|
*
|
|
|
|
* The entry points in this file are:
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
* dev_open: open a character device
|
|
|
|
* dev_reopen: reopen a character device after a driver crash
|
|
|
|
* dev_close: close a character device
|
|
|
|
* bdev_open: open a block device
|
|
|
|
* bdev_close: close a block device
|
2005-04-21 16:53:53 +02:00
|
|
|
* dev_io: FS does a read or write on a device
|
2005-07-27 15:08:52 +02:00
|
|
|
* dev_status: FS processes callback request alert
|
2005-04-21 16:53:53 +02:00
|
|
|
* gen_opcl: generic call to a task to perform an open/close
|
|
|
|
* gen_io: generic call to a task to perform an I/O operation
|
|
|
|
* no_dev: open/close processing for devices that don't exist
|
2005-10-05 17:38:15 +02:00
|
|
|
* no_dev_io: i/o processing for devices that don't exist
|
2005-04-21 16:53:53 +02:00
|
|
|
* tty_opcl: perform tty-specific processing for open/close
|
|
|
|
* ctty_opcl: perform controlling-tty-specific processing for open/close
|
|
|
|
* ctty_io: perform controlling-tty-specific processing for I/O
|
2012-02-13 16:28:04 +01:00
|
|
|
* pm_setsid: perform VFS's side of setsid system call
|
2005-04-21 16:53:53 +02:00
|
|
|
* do_ioctl: perform the IOCTL system call
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include "fs.h"
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
#include <string.h>
|
2005-04-21 16:53:53 +02:00
|
|
|
#include <fcntl.h>
|
2006-06-20 12:12:09 +02:00
|
|
|
#include <assert.h>
|
2011-03-25 11:56:43 +01:00
|
|
|
#include <sys/stat.h>
|
2005-04-21 16:53:53 +02:00
|
|
|
#include <minix/callnr.h>
|
|
|
|
#include <minix/com.h>
|
endpoint-aware conversion of servers.
'who', indicating caller number in pm and fs and some other servers, has
been removed in favour of 'who_e' (endpoint) and 'who_p' (proc nr.).
In both PM and FS, isokendpt() convert endpoints to process slot
numbers, returning OK if it was a valid and consistent endpoint number.
okendpt() does the same but panic()s if it doesn't succeed. (In PM,
this is pm_isok..)
pm and fs keep their own records of process endpoints in their proc tables,
which are needed to make kernel calls about those processes.
message field names have changed.
fs drivers are endpoints.
fs now doesn't try to get out of driver deadlock, as the protocol isn't
supposed to let that happen any more. (A warning is printed if ELOCKED
is detected though.)
fproc[].fp_task (indicating which driver the process is suspended on)
became an int.
PM and FS now get endpoint numbers of initial boot processes from the
kernel. These happen to be the same as the old proc numbers, to let
user processes reach them with the old numbers, but FS and PM don't know
that. All new processes after INIT, even after the generation number
wraps around, get endpoint numbers with generation 1 and higher, so
the first instances of the boot processes are the only processes ever
to have endpoint numbers in the old proc number range.
More return code checks of sys_* functions have been added.
IS has become endpoint-aware. Ditched the 'text' and 'data' fields
in the kernel dump (which show locations, not sizes, so aren't terribly
useful) in favour of the endpoint number. Proc number is still visible.
Some other dumps (e.g. dmap, rs) show endpoint numbers now too which got
the formatting changed.
PM reading segments using rw_seg() has changed - it uses other fields
in the message now instead of encoding the segment and process number and
fd in the fd field. For that it uses _read_pm() and _write_pm() which to
_taskcall()s directly in pm/misc.c.
PM now sys_exit()s itself on panic(), instead of sys_abort().
RS also talks in endpoints instead of process numbers.
2006-03-03 11:20:58 +01:00
|
|
|
#include <minix/endpoint.h>
|
2006-06-20 12:12:09 +02:00
|
|
|
#include <minix/ioctl.h>
|
2006-11-27 15:21:43 +01:00
|
|
|
#include <minix/u64.h>
|
2005-04-21 16:53:53 +02:00
|
|
|
#include "file.h"
|
|
|
|
#include "fproc.h"
|
2012-02-13 16:28:04 +01:00
|
|
|
#include "scratchpad.h"
|
|
|
|
#include "dmap.h"
|
2006-10-25 15:40:36 +02:00
|
|
|
#include <minix/vfsif.h>
|
|
|
|
#include "vnode.h"
|
|
|
|
#include "vmnt.h"
|
2005-04-21 16:53:53 +02:00
|
|
|
#include "param.h"
|
|
|
|
|
2012-03-25 20:25:53 +02:00
|
|
|
static void restart_reopen(int major);
|
|
|
|
static int safe_io_conversion(endpoint_t, cp_grant_id_t *, int *,
|
2012-03-24 16:16:34 +01:00
|
|
|
endpoint_t *, void **, size_t, u32_t *);
|
2012-02-13 16:28:04 +01:00
|
|
|
|
2012-03-25 20:25:53 +02:00
|
|
|
static int dummyproc;
|
2005-04-21 16:53:53 +02:00
|
|
|
|
2006-10-25 15:40:36 +02:00
|
|
|
|
2005-04-21 16:53:53 +02:00
|
|
|
/*===========================================================================*
|
|
|
|
* dev_open *
|
|
|
|
*===========================================================================*/
|
2012-03-25 20:25:53 +02:00
|
|
|
int dev_open(
|
2010-04-13 12:58:41 +02:00
|
|
|
dev_t dev, /* device to open */
|
2012-02-13 16:28:04 +01:00
|
|
|
endpoint_t proc_e, /* process to open for */
|
2010-04-13 12:58:41 +02:00
|
|
|
int flags /* mode bits and flags */
|
|
|
|
)
|
2005-04-21 16:53:53 +02:00
|
|
|
{
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
/* Open a character device. */
|
2005-04-21 16:53:53 +02:00
|
|
|
int major, r;
|
|
|
|
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
/* Determine the major device number so as to call the device class specific
|
2005-04-21 16:53:53 +02:00
|
|
|
* open/close routine. (This is the only routine that must check the
|
|
|
|
* device number for being in range. All others can trust this check.)
|
|
|
|
*/
|
2012-02-13 16:28:04 +01:00
|
|
|
major = major(dev);
|
|
|
|
if (major < 0 || major >= NR_DEVICES) major = 0;
|
|
|
|
if (dmap[major].dmap_driver == NONE) return(ENXIO);
|
|
|
|
r = (*dmap[major].dmap_opcl)(DEV_OPEN, dev, proc_e, flags);
|
2008-02-22 16:41:07 +01:00
|
|
|
return(r);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/*===========================================================================*
|
|
|
|
* dev_reopen *
|
|
|
|
*===========================================================================*/
|
2012-03-25 20:25:53 +02:00
|
|
|
int dev_reopen(
|
2010-04-13 12:58:41 +02:00
|
|
|
dev_t dev, /* device to open */
|
|
|
|
int filp_no, /* filp to reopen for */
|
|
|
|
int flags /* mode bits and flags */
|
|
|
|
)
|
2008-02-22 16:41:07 +01:00
|
|
|
{
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
/* Reopen a character device after a failing device driver. */
|
2012-02-13 16:28:04 +01:00
|
|
|
|
2008-02-22 16:41:07 +01:00
|
|
|
int major, r;
|
|
|
|
struct dmap *dp;
|
|
|
|
|
2012-02-13 16:28:04 +01:00
|
|
|
/* Determine the major device number and call the device class specific
|
|
|
|
* open/close routine. (This is the only routine that must check the device
|
|
|
|
* number for being in range. All others can trust this check.)
|
2008-02-22 16:41:07 +01:00
|
|
|
*/
|
2012-02-13 16:28:04 +01:00
|
|
|
|
|
|
|
major = major(dev);
|
|
|
|
if (major < 0 || major >= NR_DEVICES) major = 0;
|
2008-02-22 16:41:07 +01:00
|
|
|
dp = &dmap[major];
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
if (dp->dmap_driver == NONE) return(ENXIO);
|
2008-02-22 16:41:07 +01:00
|
|
|
r = (*dp->dmap_opcl)(DEV_REOPEN, dev, filp_no, flags);
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
if (r == SUSPEND) r = OK;
|
2005-04-21 16:53:53 +02:00
|
|
|
return(r);
|
|
|
|
}
|
|
|
|
|
2006-10-25 15:40:36 +02:00
|
|
|
|
2005-04-21 16:53:53 +02:00
|
|
|
/*===========================================================================*
|
|
|
|
* dev_close *
|
|
|
|
*===========================================================================*/
|
2012-03-25 20:25:53 +02:00
|
|
|
int dev_close(
|
2010-04-13 12:58:41 +02:00
|
|
|
dev_t dev, /* device to close */
|
|
|
|
int filp_no
|
|
|
|
)
|
2005-04-21 16:53:53 +02:00
|
|
|
{
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
/* Close a character device. */
|
2012-02-13 16:28:04 +01:00
|
|
|
int r, major;
|
2008-02-22 16:41:07 +01:00
|
|
|
|
2005-10-21 21:31:38 +02:00
|
|
|
/* See if driver is roughly valid. */
|
2012-02-13 16:28:04 +01:00
|
|
|
major = major(dev);
|
|
|
|
if (major < 0 || major >= NR_DEVICES) return(ENXIO);
|
|
|
|
if (dmap[major].dmap_driver == NONE) return(ENXIO);
|
|
|
|
r = (*dmap[major].dmap_opcl)(DEV_CLOSE, dev, filp_no, 0);
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
return(r);
|
2005-04-21 16:53:53 +02:00
|
|
|
}
|
|
|
|
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
/*===========================================================================*
|
2012-02-13 16:28:04 +01:00
|
|
|
* dev_open *
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
*===========================================================================*/
|
2012-03-25 20:25:53 +02:00
|
|
|
int bdev_open(dev_t dev, int access)
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
{
|
|
|
|
/* Open a block device. */
|
|
|
|
int major;
|
|
|
|
|
|
|
|
major = major(dev);
|
|
|
|
if (major < 0 || major >= NR_DEVICES) return(ENXIO);
|
|
|
|
if (dmap[major].dmap_driver == NONE) return(ENXIO);
|
|
|
|
|
|
|
|
return (*dmap[major].dmap_opcl)(BDEV_OPEN, dev, 0, access);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/*===========================================================================*
|
|
|
|
* bdev_close *
|
|
|
|
*===========================================================================*/
|
2012-03-25 20:25:53 +02:00
|
|
|
int bdev_close(dev_t dev)
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
{
|
|
|
|
/* Close a block device. */
|
|
|
|
int major;
|
|
|
|
|
|
|
|
major = major(dev);
|
|
|
|
if (major < 0 || major >= NR_DEVICES) return(ENXIO);
|
|
|
|
if (dmap[major].dmap_driver == NONE) return(ENXIO);
|
|
|
|
|
|
|
|
return (*dmap[major].dmap_opcl)(BDEV_CLOSE, dev, 0, 0);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/*===========================================================================*
|
|
|
|
* bdev_ioctl *
|
|
|
|
*===========================================================================*/
|
2012-03-25 20:25:53 +02:00
|
|
|
static int bdev_ioctl(dev_t dev, endpoint_t proc_e, int req, void *buf)
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
{
|
|
|
|
/* Perform an I/O control operation on a block device. */
|
|
|
|
struct dmap *dp;
|
|
|
|
u32_t dummy;
|
|
|
|
cp_grant_id_t gid;
|
|
|
|
message dev_mess;
|
2012-02-13 16:28:04 +01:00
|
|
|
int op, major_dev, minor_dev;
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
|
|
|
|
major_dev = major(dev);
|
|
|
|
minor_dev = minor(dev);
|
|
|
|
|
|
|
|
/* Determine task dmap. */
|
|
|
|
dp = &dmap[major_dev];
|
|
|
|
if (dp->dmap_driver == NONE) {
|
|
|
|
printf("VFS: dev_io: no driver for major %d\n", major_dev);
|
|
|
|
return(ENXIO);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Set up a grant if necessary. */
|
|
|
|
op = VFS_DEV_IOCTL;
|
2012-04-18 11:01:15 +02:00
|
|
|
(void) safe_io_conversion(dp->dmap_driver, &gid, &op, &proc_e, &buf, req,
|
2012-02-13 16:28:04 +01:00
|
|
|
&dummy);
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
|
|
|
|
/* Set up the message passed to the task. */
|
|
|
|
memset(&dev_mess, 0, sizeof(dev_mess));
|
|
|
|
|
|
|
|
dev_mess.m_type = BDEV_IOCTL;
|
|
|
|
dev_mess.BDEV_MINOR = minor_dev;
|
|
|
|
dev_mess.BDEV_REQUEST = req;
|
|
|
|
dev_mess.BDEV_GRANT = gid;
|
|
|
|
dev_mess.BDEV_ID = 0;
|
|
|
|
|
|
|
|
/* Call the task. */
|
|
|
|
(*dp->dmap_io)(dp->dmap_driver, &dev_mess);
|
|
|
|
|
|
|
|
/* Clean up. */
|
2012-02-13 16:28:04 +01:00
|
|
|
if (GRANT_VALID(gid)) cpf_revoke(gid);
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
|
|
|
|
if (dp->dmap_driver == NONE) {
|
|
|
|
printf("VFS: block driver gone!?\n");
|
|
|
|
return(EIO);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Return the result. */
|
2012-02-13 16:28:04 +01:00
|
|
|
return(dev_mess.BDEV_STATUS);
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2005-07-27 15:08:52 +02:00
|
|
|
/*===========================================================================*
|
2012-02-13 16:28:04 +01:00
|
|
|
* find_suspended_ep *
|
2006-06-20 12:12:09 +02:00
|
|
|
*===========================================================================*/
|
2012-02-13 16:28:04 +01:00
|
|
|
endpoint_t find_suspended_ep(endpoint_t driver, cp_grant_id_t g)
|
2006-06-20 12:12:09 +02:00
|
|
|
{
|
2012-02-13 16:28:04 +01:00
|
|
|
/* A process is suspended on a driver for which VFS issued a grant. Find out
|
|
|
|
* which process it was.
|
2006-06-20 12:12:09 +02:00
|
|
|
*/
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
struct fproc *rfp;
|
|
|
|
for (rfp = &fproc[0]; rfp < &fproc[NR_PROCS]; rfp++) {
|
2012-02-13 16:28:04 +01:00
|
|
|
if(rfp->fp_pid == PID_FREE)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
if(rfp->fp_blocked_on == FP_BLOCKED_ON_OTHER &&
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
rfp->fp_task == driver && rfp->fp_grant == g)
|
2012-02-13 16:28:04 +01:00
|
|
|
return(rfp->fp_endpoint);
|
|
|
|
}
|
2006-10-25 15:40:36 +02:00
|
|
|
|
2012-02-13 16:28:04 +01:00
|
|
|
return(NONE);
|
2006-06-20 12:12:09 +02:00
|
|
|
}
|
|
|
|
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
|
2006-06-20 12:12:09 +02:00
|
|
|
/*===========================================================================*
|
|
|
|
* dev_status *
|
2005-07-27 15:08:52 +02:00
|
|
|
*===========================================================================*/
|
VFS: make all IPC asynchronous
By decoupling synchronous drivers from VFS, we are a big step closer to
supporting driver crashes under all circumstances. That is, VFS can't
become stuck on IPC with a synchronous driver (e.g., INET) and can
recover from crashing block drivers during open/close/ioctl or during
communication with an FS.
In order to maintain serialized communication with a synchronous driver,
the communication is wrapped by a mutex on a per driver basis (not major
numbers as there can be multiple majors with identical endpoints). Majors
that share a driver endpoint point to a single mutex object.
In order to support crashes from block drivers, the file reopen tactic
had to be changed; first reopen files associated with the crashed
driver, then send the new driver endpoint to FSes. This solves a
deadlock between the FS and the block driver;
- VFS would send REQ_NEW_DRIVER to an FS, but he FS only receives it
after retrying the current request to the newly started driver.
- The block driver would refuse the retried request until all files
had been reopened.
- VFS would reopen files only after getting a reply from the initial
REQ_NEW_DRIVER.
When a character special driver crashes, all associated files have to
be marked invalid and closed (or reopened if flagged as such). However,
they can only be closed if a thread holds exclusive access to it. To
obtain exclusive access, the worker thread (which handles the new driver
endpoint event from DS) schedules a new job to garbage collect invalid
files. This way, we can signal the worker thread that was talking to the
crashed driver and will release exclusive access to a file associated
with the crashed driver and prevent the garbage collecting worker thread
from dead locking on that file.
Also, when a character special driver crashes, RS will unmap the driver
and remap it upon restart. During unmapping, associated files are marked
invalid instead of waiting for an endpoint up event from DS, as that
event might come later than new read/write/select requests and thus
cause confusion in the freshly started driver.
When locking a filp, the usage counters are no longer checked. The usage
counter can legally go down to zero during filp invalidation while there
are locks pending.
DS events are handled by a separate worker thread instead of the main
thread as reopening files could lead to another crash and a stuck thread.
An additional worker thread is then necessary to unlock it.
Finally, with everything asynchronous a race condition in do_select
surfaced. A select entry was only marked in use after succesfully sending
initial select requests to drivers and having to wait. When multiple
select() calls were handled there was opportunity that these entries
were overwritten. This had as effect that some select results were
ignored (and select() remained blocking instead if returning) or do_select
tried to access filps that were not present (because thrown away by
secondary select()). This bug manifested itself with sendrecs, but was
very hard to reproduce. However, it became awfully easy to trigger with
asynsends only.
2012-08-28 16:06:51 +02:00
|
|
|
void dev_status(endpoint_t drv_e)
|
2005-07-27 15:08:52 +02:00
|
|
|
{
|
2012-02-13 16:28:04 +01:00
|
|
|
/* A device sent us a notification it has something for us. Retrieve it. */
|
|
|
|
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
message st;
|
2012-02-13 16:28:04 +01:00
|
|
|
int major, get_more = 1;
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
endpoint_t endpt;
|
2005-07-27 15:08:52 +02:00
|
|
|
|
2012-02-13 16:28:04 +01:00
|
|
|
for (major = 0; major < NR_DEVICES; major++)
|
VFS: make all IPC asynchronous
By decoupling synchronous drivers from VFS, we are a big step closer to
supporting driver crashes under all circumstances. That is, VFS can't
become stuck on IPC with a synchronous driver (e.g., INET) and can
recover from crashing block drivers during open/close/ioctl or during
communication with an FS.
In order to maintain serialized communication with a synchronous driver,
the communication is wrapped by a mutex on a per driver basis (not major
numbers as there can be multiple majors with identical endpoints). Majors
that share a driver endpoint point to a single mutex object.
In order to support crashes from block drivers, the file reopen tactic
had to be changed; first reopen files associated with the crashed
driver, then send the new driver endpoint to FSes. This solves a
deadlock between the FS and the block driver;
- VFS would send REQ_NEW_DRIVER to an FS, but he FS only receives it
after retrying the current request to the newly started driver.
- The block driver would refuse the retried request until all files
had been reopened.
- VFS would reopen files only after getting a reply from the initial
REQ_NEW_DRIVER.
When a character special driver crashes, all associated files have to
be marked invalid and closed (or reopened if flagged as such). However,
they can only be closed if a thread holds exclusive access to it. To
obtain exclusive access, the worker thread (which handles the new driver
endpoint event from DS) schedules a new job to garbage collect invalid
files. This way, we can signal the worker thread that was talking to the
crashed driver and will release exclusive access to a file associated
with the crashed driver and prevent the garbage collecting worker thread
from dead locking on that file.
Also, when a character special driver crashes, RS will unmap the driver
and remap it upon restart. During unmapping, associated files are marked
invalid instead of waiting for an endpoint up event from DS, as that
event might come later than new read/write/select requests and thus
cause confusion in the freshly started driver.
When locking a filp, the usage counters are no longer checked. The usage
counter can legally go down to zero during filp invalidation while there
are locks pending.
DS events are handled by a separate worker thread instead of the main
thread as reopening files could lead to another crash and a stuck thread.
An additional worker thread is then necessary to unlock it.
Finally, with everything asynchronous a race condition in do_select
surfaced. A select entry was only marked in use after succesfully sending
initial select requests to drivers and having to wait. When multiple
select() calls were handled there was opportunity that these entries
were overwritten. This had as effect that some select results were
ignored (and select() remained blocking instead if returning) or do_select
tried to access filps that were not present (because thrown away by
secondary select()). This bug manifested itself with sendrecs, but was
very hard to reproduce. However, it became awfully easy to trigger with
asynsends only.
2012-08-28 16:06:51 +02:00
|
|
|
if (dmap_driver_match(drv_e, major))
|
2012-02-13 16:28:04 +01:00
|
|
|
break; /* 'major' is the device that sent the message */
|
2005-07-27 15:08:52 +02:00
|
|
|
|
2012-02-13 16:28:04 +01:00
|
|
|
if (major >= NR_DEVICES) /* Device endpoint not found; nothing to do */
|
|
|
|
return;
|
2008-02-22 16:41:07 +01:00
|
|
|
|
2012-02-07 16:41:25 +01:00
|
|
|
if (dev_style_asyn(dmap[major].dmap_style)) {
|
VFS: make all IPC asynchronous
By decoupling synchronous drivers from VFS, we are a big step closer to
supporting driver crashes under all circumstances. That is, VFS can't
become stuck on IPC with a synchronous driver (e.g., INET) and can
recover from crashing block drivers during open/close/ioctl or during
communication with an FS.
In order to maintain serialized communication with a synchronous driver,
the communication is wrapped by a mutex on a per driver basis (not major
numbers as there can be multiple majors with identical endpoints). Majors
that share a driver endpoint point to a single mutex object.
In order to support crashes from block drivers, the file reopen tactic
had to be changed; first reopen files associated with the crashed
driver, then send the new driver endpoint to FSes. This solves a
deadlock between the FS and the block driver;
- VFS would send REQ_NEW_DRIVER to an FS, but he FS only receives it
after retrying the current request to the newly started driver.
- The block driver would refuse the retried request until all files
had been reopened.
- VFS would reopen files only after getting a reply from the initial
REQ_NEW_DRIVER.
When a character special driver crashes, all associated files have to
be marked invalid and closed (or reopened if flagged as such). However,
they can only be closed if a thread holds exclusive access to it. To
obtain exclusive access, the worker thread (which handles the new driver
endpoint event from DS) schedules a new job to garbage collect invalid
files. This way, we can signal the worker thread that was talking to the
crashed driver and will release exclusive access to a file associated
with the crashed driver and prevent the garbage collecting worker thread
from dead locking on that file.
Also, when a character special driver crashes, RS will unmap the driver
and remap it upon restart. During unmapping, associated files are marked
invalid instead of waiting for an endpoint up event from DS, as that
event might come later than new read/write/select requests and thus
cause confusion in the freshly started driver.
When locking a filp, the usage counters are no longer checked. The usage
counter can legally go down to zero during filp invalidation while there
are locks pending.
DS events are handled by a separate worker thread instead of the main
thread as reopening files could lead to another crash and a stuck thread.
An additional worker thread is then necessary to unlock it.
Finally, with everything asynchronous a race condition in do_select
surfaced. A select entry was only marked in use after succesfully sending
initial select requests to drivers and having to wait. When multiple
select() calls were handled there was opportunity that these entries
were overwritten. This had as effect that some select results were
ignored (and select() remained blocking instead if returning) or do_select
tried to access filps that were not present (because thrown away by
secondary select()). This bug manifested itself with sendrecs, but was
very hard to reproduce. However, it became awfully easy to trigger with
asynsends only.
2012-08-28 16:06:51 +02:00
|
|
|
printf("VFS: not doing dev_status for async driver %d\n", drv_e);
|
2012-02-13 16:28:04 +01:00
|
|
|
return;
|
|
|
|
}
|
2005-07-27 15:08:52 +02:00
|
|
|
|
2012-02-13 16:28:04 +01:00
|
|
|
/* Continuously send DEV_STATUS messages until the device has nothing to
|
|
|
|
* say to us anymore. */
|
|
|
|
do {
|
|
|
|
int r;
|
|
|
|
st.m_type = DEV_STATUS;
|
VFS: make all IPC asynchronous
By decoupling synchronous drivers from VFS, we are a big step closer to
supporting driver crashes under all circumstances. That is, VFS can't
become stuck on IPC with a synchronous driver (e.g., INET) and can
recover from crashing block drivers during open/close/ioctl or during
communication with an FS.
In order to maintain serialized communication with a synchronous driver,
the communication is wrapped by a mutex on a per driver basis (not major
numbers as there can be multiple majors with identical endpoints). Majors
that share a driver endpoint point to a single mutex object.
In order to support crashes from block drivers, the file reopen tactic
had to be changed; first reopen files associated with the crashed
driver, then send the new driver endpoint to FSes. This solves a
deadlock between the FS and the block driver;
- VFS would send REQ_NEW_DRIVER to an FS, but he FS only receives it
after retrying the current request to the newly started driver.
- The block driver would refuse the retried request until all files
had been reopened.
- VFS would reopen files only after getting a reply from the initial
REQ_NEW_DRIVER.
When a character special driver crashes, all associated files have to
be marked invalid and closed (or reopened if flagged as such). However,
they can only be closed if a thread holds exclusive access to it. To
obtain exclusive access, the worker thread (which handles the new driver
endpoint event from DS) schedules a new job to garbage collect invalid
files. This way, we can signal the worker thread that was talking to the
crashed driver and will release exclusive access to a file associated
with the crashed driver and prevent the garbage collecting worker thread
from dead locking on that file.
Also, when a character special driver crashes, RS will unmap the driver
and remap it upon restart. During unmapping, associated files are marked
invalid instead of waiting for an endpoint up event from DS, as that
event might come later than new read/write/select requests and thus
cause confusion in the freshly started driver.
When locking a filp, the usage counters are no longer checked. The usage
counter can legally go down to zero during filp invalidation while there
are locks pending.
DS events are handled by a separate worker thread instead of the main
thread as reopening files could lead to another crash and a stuck thread.
An additional worker thread is then necessary to unlock it.
Finally, with everything asynchronous a race condition in do_select
surfaced. A select entry was only marked in use after succesfully sending
initial select requests to drivers and having to wait. When multiple
select() calls were handled there was opportunity that these entries
were overwritten. This had as effect that some select results were
ignored (and select() remained blocking instead if returning) or do_select
tried to access filps that were not present (because thrown away by
secondary select()). This bug manifested itself with sendrecs, but was
very hard to reproduce. However, it became awfully easy to trigger with
asynsends only.
2012-08-28 16:06:51 +02:00
|
|
|
r = drv_sendrec(drv_e, &st);
|
2012-02-13 16:28:04 +01:00
|
|
|
if (r == OK && st.REP_STATUS == ERESTART) r = EDEADEPT;
|
|
|
|
if (r != OK) {
|
VFS: make all IPC asynchronous
By decoupling synchronous drivers from VFS, we are a big step closer to
supporting driver crashes under all circumstances. That is, VFS can't
become stuck on IPC with a synchronous driver (e.g., INET) and can
recover from crashing block drivers during open/close/ioctl or during
communication with an FS.
In order to maintain serialized communication with a synchronous driver,
the communication is wrapped by a mutex on a per driver basis (not major
numbers as there can be multiple majors with identical endpoints). Majors
that share a driver endpoint point to a single mutex object.
In order to support crashes from block drivers, the file reopen tactic
had to be changed; first reopen files associated with the crashed
driver, then send the new driver endpoint to FSes. This solves a
deadlock between the FS and the block driver;
- VFS would send REQ_NEW_DRIVER to an FS, but he FS only receives it
after retrying the current request to the newly started driver.
- The block driver would refuse the retried request until all files
had been reopened.
- VFS would reopen files only after getting a reply from the initial
REQ_NEW_DRIVER.
When a character special driver crashes, all associated files have to
be marked invalid and closed (or reopened if flagged as such). However,
they can only be closed if a thread holds exclusive access to it. To
obtain exclusive access, the worker thread (which handles the new driver
endpoint event from DS) schedules a new job to garbage collect invalid
files. This way, we can signal the worker thread that was talking to the
crashed driver and will release exclusive access to a file associated
with the crashed driver and prevent the garbage collecting worker thread
from dead locking on that file.
Also, when a character special driver crashes, RS will unmap the driver
and remap it upon restart. During unmapping, associated files are marked
invalid instead of waiting for an endpoint up event from DS, as that
event might come later than new read/write/select requests and thus
cause confusion in the freshly started driver.
When locking a filp, the usage counters are no longer checked. The usage
counter can legally go down to zero during filp invalidation while there
are locks pending.
DS events are handled by a separate worker thread instead of the main
thread as reopening files could lead to another crash and a stuck thread.
An additional worker thread is then necessary to unlock it.
Finally, with everything asynchronous a race condition in do_select
surfaced. A select entry was only marked in use after succesfully sending
initial select requests to drivers and having to wait. When multiple
select() calls were handled there was opportunity that these entries
were overwritten. This had as effect that some select results were
ignored (and select() remained blocking instead if returning) or do_select
tried to access filps that were not present (because thrown away by
secondary select()). This bug manifested itself with sendrecs, but was
very hard to reproduce. However, it became awfully easy to trigger with
asynsends only.
2012-08-28 16:06:51 +02:00
|
|
|
printf("VFS: DEV_STATUS failed to %d: %d\n", drv_e, r);
|
2012-02-13 16:28:04 +01:00
|
|
|
if (r == EDEADSRCDST || r == EDEADEPT) return;
|
|
|
|
panic("VFS: couldn't sendrec for DEV_STATUS: %d", r);
|
|
|
|
}
|
2005-07-27 15:08:52 +02:00
|
|
|
|
2012-02-13 16:28:04 +01:00
|
|
|
switch(st.m_type) {
|
|
|
|
case DEV_REVIVE:
|
|
|
|
/* We've got results for a read/write/ioctl call to a
|
|
|
|
* synchronous character driver */
|
|
|
|
endpt = st.REP_ENDPT;
|
|
|
|
if (endpt == VFS_PROC_NR) {
|
VFS: make all IPC asynchronous
By decoupling synchronous drivers from VFS, we are a big step closer to
supporting driver crashes under all circumstances. That is, VFS can't
become stuck on IPC with a synchronous driver (e.g., INET) and can
recover from crashing block drivers during open/close/ioctl or during
communication with an FS.
In order to maintain serialized communication with a synchronous driver,
the communication is wrapped by a mutex on a per driver basis (not major
numbers as there can be multiple majors with identical endpoints). Majors
that share a driver endpoint point to a single mutex object.
In order to support crashes from block drivers, the file reopen tactic
had to be changed; first reopen files associated with the crashed
driver, then send the new driver endpoint to FSes. This solves a
deadlock between the FS and the block driver;
- VFS would send REQ_NEW_DRIVER to an FS, but he FS only receives it
after retrying the current request to the newly started driver.
- The block driver would refuse the retried request until all files
had been reopened.
- VFS would reopen files only after getting a reply from the initial
REQ_NEW_DRIVER.
When a character special driver crashes, all associated files have to
be marked invalid and closed (or reopened if flagged as such). However,
they can only be closed if a thread holds exclusive access to it. To
obtain exclusive access, the worker thread (which handles the new driver
endpoint event from DS) schedules a new job to garbage collect invalid
files. This way, we can signal the worker thread that was talking to the
crashed driver and will release exclusive access to a file associated
with the crashed driver and prevent the garbage collecting worker thread
from dead locking on that file.
Also, when a character special driver crashes, RS will unmap the driver
and remap it upon restart. During unmapping, associated files are marked
invalid instead of waiting for an endpoint up event from DS, as that
event might come later than new read/write/select requests and thus
cause confusion in the freshly started driver.
When locking a filp, the usage counters are no longer checked. The usage
counter can legally go down to zero during filp invalidation while there
are locks pending.
DS events are handled by a separate worker thread instead of the main
thread as reopening files could lead to another crash and a stuck thread.
An additional worker thread is then necessary to unlock it.
Finally, with everything asynchronous a race condition in do_select
surfaced. A select entry was only marked in use after succesfully sending
initial select requests to drivers and having to wait. When multiple
select() calls were handled there was opportunity that these entries
were overwritten. This had as effect that some select results were
ignored (and select() remained blocking instead if returning) or do_select
tried to access filps that were not present (because thrown away by
secondary select()). This bug manifested itself with sendrecs, but was
very hard to reproduce. However, it became awfully easy to trigger with
asynsends only.
2012-08-28 16:06:51 +02:00
|
|
|
endpt = find_suspended_ep(drv_e, st.REP_IO_GRANT);
|
2012-04-13 14:50:38 +02:00
|
|
|
if (endpt == NONE) {
|
2012-02-13 16:28:04 +01:00
|
|
|
printf("VFS: proc with grant %d from %d not found\n",
|
|
|
|
st.REP_IO_GRANT, st.m_source);
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
revive(endpt, st.REP_STATUS);
|
|
|
|
break;
|
|
|
|
case DEV_IO_READY:
|
|
|
|
/* Reply to a select request: driver is ready for I/O */
|
|
|
|
select_reply2(st.m_source, st.DEV_MINOR, st.DEV_SEL_OPS);
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
printf("VFS: unrecognized reply %d to DEV_STATUS\n",st.m_type);
|
|
|
|
/* Fall through. */
|
|
|
|
case DEV_NO_STATUS:
|
|
|
|
get_more = 0;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
} while(get_more);
|
2005-07-27 15:08:52 +02:00
|
|
|
}
|
|
|
|
|
2006-06-20 12:12:09 +02:00
|
|
|
/*===========================================================================*
|
|
|
|
* safe_io_conversion *
|
|
|
|
*===========================================================================*/
|
2012-03-25 20:25:53 +02:00
|
|
|
static int safe_io_conversion(driver, gid, op, io_ept, buf, bytes, pos_lo)
|
2006-06-20 12:12:09 +02:00
|
|
|
endpoint_t driver;
|
|
|
|
cp_grant_id_t *gid;
|
|
|
|
int *op;
|
|
|
|
endpoint_t *io_ept;
|
2006-06-29 16:24:07 +02:00
|
|
|
void **buf;
|
2012-02-13 16:28:04 +01:00
|
|
|
size_t bytes;
|
2007-04-24 15:27:33 +02:00
|
|
|
u32_t *pos_lo;
|
2006-06-20 12:12:09 +02:00
|
|
|
{
|
2012-02-13 16:28:04 +01:00
|
|
|
/* Convert operation to the 'safe' variant (i.e., grant based) if applicable.
|
|
|
|
* If no copying of data is involved, there is also no need to convert. */
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
|
2012-02-13 16:28:04 +01:00
|
|
|
int access = 0;
|
|
|
|
size_t size;
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
|
2012-02-13 16:28:04 +01:00
|
|
|
*gid = GRANT_INVALID; /* Grant to buffer */
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
|
|
|
|
switch(*op) {
|
2012-02-13 16:28:04 +01:00
|
|
|
case VFS_DEV_READ:
|
|
|
|
case VFS_DEV_WRITE:
|
|
|
|
/* Change to safe op. */
|
|
|
|
*op = (*op == VFS_DEV_READ) ? DEV_READ_S : DEV_WRITE_S;
|
|
|
|
*gid = cpf_grant_magic(driver, *io_ept, (vir_bytes) *buf, bytes,
|
|
|
|
*op == DEV_READ_S ? CPF_WRITE : CPF_READ);
|
|
|
|
if (*gid < 0)
|
|
|
|
panic("VFS: cpf_grant_magic of READ/WRITE buffer failed");
|
|
|
|
break;
|
|
|
|
case VFS_DEV_IOCTL:
|
|
|
|
*pos_lo = *io_ept; /* Old endpoint in POSITION field. */
|
|
|
|
*op = DEV_IOCTL_S;
|
2012-04-13 14:50:38 +02:00
|
|
|
/* For IOCTLs, the bytes parameter encodes requested access method
|
|
|
|
* and buffer size */
|
|
|
|
if(_MINIX_IOCTL_IOR(bytes)) access |= CPF_WRITE;
|
|
|
|
if(_MINIX_IOCTL_IOW(bytes)) access |= CPF_READ;
|
|
|
|
if(_MINIX_IOCTL_BIG(bytes))
|
|
|
|
size = _MINIX_IOCTL_SIZE_BIG(bytes);
|
2012-02-13 16:28:04 +01:00
|
|
|
else
|
2012-04-13 14:50:38 +02:00
|
|
|
size = _MINIX_IOCTL_SIZE(bytes);
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
|
2012-02-13 16:28:04 +01:00
|
|
|
/* Grant access to the buffer even if no I/O happens with the ioctl, in
|
|
|
|
* order to disambiguate requests with DEV_IOCTL_S.
|
|
|
|
*/
|
|
|
|
*gid = cpf_grant_magic(driver, *io_ept, (vir_bytes) *buf, size, access);
|
|
|
|
if (*gid < 0)
|
|
|
|
panic("VFS: cpf_grant_magic IOCTL buffer failed");
|
|
|
|
|
|
|
|
break;
|
|
|
|
case VFS_DEV_SELECT:
|
|
|
|
*op = DEV_SELECT;
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
panic("VFS: unknown operation %d for safe I/O conversion", *op);
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
}
|
2006-06-20 12:12:09 +02:00
|
|
|
|
2012-02-13 16:28:04 +01:00
|
|
|
/* If we have converted to a safe operation, I/O endpoint becomes VFS if it
|
|
|
|
* wasn't already.
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
*/
|
|
|
|
if(GRANT_VALID(*gid)) {
|
2010-06-08 15:58:01 +02:00
|
|
|
*io_ept = VFS_PROC_NR;
|
2012-02-13 16:28:04 +01:00
|
|
|
return(1);
|
|
|
|
}
|
2006-06-20 12:12:09 +02:00
|
|
|
|
2012-02-13 16:28:04 +01:00
|
|
|
/* Not converted to a safe operation (because there is no copying involved in
|
|
|
|
* this operation).
|
|
|
|
*/
|
|
|
|
return(0);
|
2006-06-20 12:12:09 +02:00
|
|
|
}
|
|
|
|
|
2012-03-25 20:25:53 +02:00
|
|
|
static int cancel_nblock(struct dmap * dp,
|
2012-02-07 16:33:32 +01:00
|
|
|
int minor,
|
|
|
|
int call,
|
|
|
|
endpoint_t ioproc,
|
|
|
|
cp_grant_id_t gid)
|
|
|
|
{
|
|
|
|
message dev_mess;
|
|
|
|
|
|
|
|
dev_mess.m_type = CANCEL;
|
|
|
|
dev_mess.USER_ENDPT = ioproc;
|
|
|
|
dev_mess.IO_GRANT = (char *) gid;
|
|
|
|
|
|
|
|
/* This R_BIT/W_BIT check taken from suspend()/unpause()
|
|
|
|
* logic. Mode is expected in the COUNT field.
|
|
|
|
*/
|
|
|
|
dev_mess.COUNT = 0;
|
|
|
|
if (call == READ)
|
|
|
|
dev_mess.COUNT = R_BIT;
|
|
|
|
else if (call == WRITE)
|
|
|
|
dev_mess.COUNT = W_BIT;
|
|
|
|
dev_mess.DEVICE = minor;
|
|
|
|
(*dp->dmap_io)(dp->dmap_driver, &dev_mess);
|
|
|
|
|
|
|
|
return dev_mess.REP_STATUS;
|
|
|
|
}
|
|
|
|
|
2005-04-21 16:53:53 +02:00
|
|
|
/*===========================================================================*
|
|
|
|
* dev_io *
|
|
|
|
*===========================================================================*/
|
2012-03-25 20:25:53 +02:00
|
|
|
int dev_io(
|
2010-04-13 12:58:41 +02:00
|
|
|
int op, /* DEV_READ, DEV_WRITE, DEV_IOCTL, etc. */
|
|
|
|
dev_t dev, /* major-minor device number */
|
|
|
|
int proc_e, /* in whose address space is buf? */
|
|
|
|
void *buf, /* virtual address of the buffer */
|
|
|
|
u64_t pos, /* byte position */
|
2012-02-13 16:28:04 +01:00
|
|
|
size_t bytes, /* how many bytes to transfer */
|
2010-04-13 12:58:41 +02:00
|
|
|
int flags, /* special flags, like O_NONBLOCK */
|
|
|
|
int suspend_reopen /* Just suspend the process */
|
|
|
|
)
|
2005-04-21 16:53:53 +02:00
|
|
|
{
|
2012-02-13 16:28:04 +01:00
|
|
|
/* Read from or write to a device. The parameter 'dev' tells which one. */
|
2005-04-21 16:53:53 +02:00
|
|
|
struct dmap *dp;
|
2007-04-24 15:27:33 +02:00
|
|
|
u32_t pos_lo, pos_high;
|
2005-04-21 16:53:53 +02:00
|
|
|
message dev_mess;
|
2006-06-20 12:12:09 +02:00
|
|
|
cp_grant_id_t gid = GRANT_INVALID;
|
2012-02-13 16:28:04 +01:00
|
|
|
int safe, minor_dev, major_dev;
|
2006-06-29 16:24:07 +02:00
|
|
|
void *buf_used;
|
2006-08-10 13:51:11 +02:00
|
|
|
endpoint_t ioproc;
|
2012-02-20 14:19:29 +01:00
|
|
|
int ret, is_asyn;
|
2005-04-21 16:53:53 +02:00
|
|
|
|
2012-02-13 16:28:04 +01:00
|
|
|
pos_lo = ex64lo(pos);
|
|
|
|
pos_high = ex64hi(pos);
|
|
|
|
major_dev = major(dev);
|
|
|
|
minor_dev = minor(dev);
|
2006-11-27 15:21:43 +01:00
|
|
|
|
2005-04-21 16:53:53 +02:00
|
|
|
/* Determine task dmap. */
|
2012-02-13 16:28:04 +01:00
|
|
|
dp = &dmap[major_dev];
|
2005-04-21 16:53:53 +02:00
|
|
|
|
2005-10-05 17:38:15 +02:00
|
|
|
/* See if driver is roughly valid. */
|
2012-02-17 14:36:57 +01:00
|
|
|
if (dp->dmap_driver == NONE) return(ENXIO);
|
endpoint-aware conversion of servers.
'who', indicating caller number in pm and fs and some other servers, has
been removed in favour of 'who_e' (endpoint) and 'who_p' (proc nr.).
In both PM and FS, isokendpt() convert endpoints to process slot
numbers, returning OK if it was a valid and consistent endpoint number.
okendpt() does the same but panic()s if it doesn't succeed. (In PM,
this is pm_isok..)
pm and fs keep their own records of process endpoints in their proc tables,
which are needed to make kernel calls about those processes.
message field names have changed.
fs drivers are endpoints.
fs now doesn't try to get out of driver deadlock, as the protocol isn't
supposed to let that happen any more. (A warning is printed if ELOCKED
is detected though.)
fproc[].fp_task (indicating which driver the process is suspended on)
became an int.
PM and FS now get endpoint numbers of initial boot processes from the
kernel. These happen to be the same as the old proc numbers, to let
user processes reach them with the old numbers, but FS and PM don't know
that. All new processes after INIT, even after the generation number
wraps around, get endpoint numbers with generation 1 and higher, so
the first instances of the boot processes are the only processes ever
to have endpoint numbers in the old proc number range.
More return code checks of sys_* functions have been added.
IS has become endpoint-aware. Ditched the 'text' and 'data' fields
in the kernel dump (which show locations, not sizes, so aren't terribly
useful) in favour of the endpoint number. Proc number is still visible.
Some other dumps (e.g. dmap, rs) show endpoint numbers now too which got
the formatting changed.
PM reading segments using rw_seg() has changed - it uses other fields
in the message now instead of encoding the segment and process number and
fd in the fd field. For that it uses _read_pm() and _write_pm() which to
_taskcall()s directly in pm/misc.c.
PM now sys_exit()s itself on panic(), instead of sys_abort().
RS also talks in endpoints instead of process numbers.
2006-03-03 11:20:58 +01:00
|
|
|
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
if (suspend_reopen) {
|
2008-02-22 16:41:07 +01:00
|
|
|
/* Suspend user. */
|
|
|
|
fp->fp_grant = GRANT_INVALID;
|
|
|
|
fp->fp_ioproc = NONE;
|
2009-09-22 23:48:26 +02:00
|
|
|
wait_for(dp->dmap_driver);
|
2012-02-13 16:28:04 +01:00
|
|
|
fp->fp_flags |= FP_SUSP_REOPEN;
|
2008-02-22 16:41:07 +01:00
|
|
|
return(SUSPEND);
|
|
|
|
}
|
|
|
|
|
endpoint-aware conversion of servers.
'who', indicating caller number in pm and fs and some other servers, has
been removed in favour of 'who_e' (endpoint) and 'who_p' (proc nr.).
In both PM and FS, isokendpt() convert endpoints to process slot
numbers, returning OK if it was a valid and consistent endpoint number.
okendpt() does the same but panic()s if it doesn't succeed. (In PM,
this is pm_isok..)
pm and fs keep their own records of process endpoints in their proc tables,
which are needed to make kernel calls about those processes.
message field names have changed.
fs drivers are endpoints.
fs now doesn't try to get out of driver deadlock, as the protocol isn't
supposed to let that happen any more. (A warning is printed if ELOCKED
is detected though.)
fproc[].fp_task (indicating which driver the process is suspended on)
became an int.
PM and FS now get endpoint numbers of initial boot processes from the
kernel. These happen to be the same as the old proc numbers, to let
user processes reach them with the old numbers, but FS and PM don't know
that. All new processes after INIT, even after the generation number
wraps around, get endpoint numbers with generation 1 and higher, so
the first instances of the boot processes are the only processes ever
to have endpoint numbers in the old proc number range.
More return code checks of sys_* functions have been added.
IS has become endpoint-aware. Ditched the 'text' and 'data' fields
in the kernel dump (which show locations, not sizes, so aren't terribly
useful) in favour of the endpoint number. Proc number is still visible.
Some other dumps (e.g. dmap, rs) show endpoint numbers now too which got
the formatting changed.
PM reading segments using rw_seg() has changed - it uses other fields
in the message now instead of encoding the segment and process number and
fd in the fd field. For that it uses _read_pm() and _write_pm() which to
_taskcall()s directly in pm/misc.c.
PM now sys_exit()s itself on panic(), instead of sys_abort().
RS also talks in endpoints instead of process numbers.
2006-03-03 11:20:58 +01:00
|
|
|
if(isokendpt(dp->dmap_driver, &dummyproc) != OK) {
|
2012-02-13 16:28:04 +01:00
|
|
|
printf("VFS: dev_io: old driver for major %x (%d)\n", major_dev,
|
|
|
|
dp->dmap_driver);
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
return(ENXIO);
|
2005-10-21 21:31:38 +02:00
|
|
|
}
|
2005-10-05 17:38:15 +02:00
|
|
|
|
2006-06-20 12:12:09 +02:00
|
|
|
/* By default, these are right. */
|
Server/driver protocols: no longer allow third-party copies.
Before safecopies, the IO_ENDPT and DL_ENDPT message fields were needed
to know which actual process to copy data from/to, as that process may
not always be the caller. Now that we have full safecopy support, these
fields have become useless for that purpose: the owner of the grant is
*always* the caller. Allowing the caller to supply another endpoint is
in fact dangerous, because the callee may then end up using a grant
from a third party. One could call this a variant of the confused
deputy problem.
From now on, safecopy calls should always use the caller's endpoint as
grant owner. This fully obsoletes the DL_ENDPT field in the
inet/ethernet protocol. IO_ENDPT has other uses besides identifying the
grant owner though. This patch renames IO_ENDPT to USER_ENDPT, not only
because that is a more fitting name (it should never be used for I/O
after all), but also in order to intentionally break any old system
source code outside the base system. If this patch breaks your code,
fixing it is fairly simple:
- DL_ENDPT should be replaced with m_source;
- IO_ENDPT should be replaced with m_source when used for safecopies;
- IO_ENDPT should be replaced with USER_ENDPT for any other use, e.g.
when setting REP_ENDPT, matching requests in CANCEL calls, getting
DEV_SELECT flags, and retrieving of the real user process's endpoint
in DEV_OPEN.
The changes in this patch are binary backward compatible.
2011-04-11 19:35:05 +02:00
|
|
|
dev_mess.USER_ENDPT = proc_e;
|
2006-06-20 12:12:09 +02:00
|
|
|
dev_mess.ADDRESS = buf;
|
|
|
|
|
|
|
|
/* Convert DEV_* to DEV_*_S variants. */
|
2006-06-29 16:24:07 +02:00
|
|
|
buf_used = buf;
|
2012-02-13 16:28:04 +01:00
|
|
|
safe = safe_io_conversion(dp->dmap_driver, &gid, &op,
|
|
|
|
(endpoint_t *) &dev_mess.USER_ENDPT, &buf_used,
|
|
|
|
bytes, &pos_lo);
|
2006-06-20 12:12:09 +02:00
|
|
|
|
2012-02-20 14:19:29 +01:00
|
|
|
is_asyn = dev_style_asyn(dp->dmap_style);
|
|
|
|
|
2012-02-13 16:28:04 +01:00
|
|
|
/* If the safe conversion was done, set the IO_GRANT to
|
2006-06-20 12:12:09 +02:00
|
|
|
* the grant id.
|
|
|
|
*/
|
|
|
|
if(safe) dev_mess.IO_GRANT = (char *) gid;
|
|
|
|
|
|
|
|
/* Set up the rest of the message passed to task. */
|
2005-04-21 16:53:53 +02:00
|
|
|
dev_mess.m_type = op;
|
2012-02-13 16:28:04 +01:00
|
|
|
dev_mess.DEVICE = minor_dev;
|
2007-04-24 15:27:33 +02:00
|
|
|
dev_mess.POSITION = pos_lo;
|
2005-04-21 16:53:53 +02:00
|
|
|
dev_mess.COUNT = bytes;
|
2007-04-24 15:27:33 +02:00
|
|
|
dev_mess.HIGHPOS = pos_high;
|
2012-02-20 14:19:29 +01:00
|
|
|
dev_mess.FLAGS = 0;
|
|
|
|
|
|
|
|
if (flags & O_NONBLOCK)
|
|
|
|
dev_mess.FLAGS |= FLG_OP_NONBLOCK;
|
2006-06-20 12:12:09 +02:00
|
|
|
|
2006-08-10 13:51:11 +02:00
|
|
|
/* This will be used if the i/o is suspended. */
|
Server/driver protocols: no longer allow third-party copies.
Before safecopies, the IO_ENDPT and DL_ENDPT message fields were needed
to know which actual process to copy data from/to, as that process may
not always be the caller. Now that we have full safecopy support, these
fields have become useless for that purpose: the owner of the grant is
*always* the caller. Allowing the caller to supply another endpoint is
in fact dangerous, because the callee may then end up using a grant
from a third party. One could call this a variant of the confused
deputy problem.
From now on, safecopy calls should always use the caller's endpoint as
grant owner. This fully obsoletes the DL_ENDPT field in the
inet/ethernet protocol. IO_ENDPT has other uses besides identifying the
grant owner though. This patch renames IO_ENDPT to USER_ENDPT, not only
because that is a more fitting name (it should never be used for I/O
after all), but also in order to intentionally break any old system
source code outside the base system. If this patch breaks your code,
fixing it is fairly simple:
- DL_ENDPT should be replaced with m_source;
- IO_ENDPT should be replaced with m_source when used for safecopies;
- IO_ENDPT should be replaced with USER_ENDPT for any other use, e.g.
when setting REP_ENDPT, matching requests in CANCEL calls, getting
DEV_SELECT flags, and retrieving of the real user process's endpoint
in DEV_OPEN.
The changes in this patch are binary backward compatible.
2011-04-11 19:35:05 +02:00
|
|
|
ioproc = dev_mess.USER_ENDPT;
|
2005-04-21 16:53:53 +02:00
|
|
|
|
|
|
|
/* Call the task. */
|
|
|
|
(*dp->dmap_io)(dp->dmap_driver, &dev_mess);
|
|
|
|
|
2006-03-15 22:55:32 +01:00
|
|
|
if(dp->dmap_driver == NONE) {
|
2012-02-13 16:28:04 +01:00
|
|
|
/* Driver has vanished. */
|
|
|
|
printf("VFS: driver gone?!\n");
|
|
|
|
if(safe) cpf_revoke(gid);
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
return(EIO);
|
2006-03-15 22:55:32 +01:00
|
|
|
}
|
endpoint-aware conversion of servers.
'who', indicating caller number in pm and fs and some other servers, has
been removed in favour of 'who_e' (endpoint) and 'who_p' (proc nr.).
In both PM and FS, isokendpt() convert endpoints to process slot
numbers, returning OK if it was a valid and consistent endpoint number.
okendpt() does the same but panic()s if it doesn't succeed. (In PM,
this is pm_isok..)
pm and fs keep their own records of process endpoints in their proc tables,
which are needed to make kernel calls about those processes.
message field names have changed.
fs drivers are endpoints.
fs now doesn't try to get out of driver deadlock, as the protocol isn't
supposed to let that happen any more. (A warning is printed if ELOCKED
is detected though.)
fproc[].fp_task (indicating which driver the process is suspended on)
became an int.
PM and FS now get endpoint numbers of initial boot processes from the
kernel. These happen to be the same as the old proc numbers, to let
user processes reach them with the old numbers, but FS and PM don't know
that. All new processes after INIT, even after the generation number
wraps around, get endpoint numbers with generation 1 and higher, so
the first instances of the boot processes are the only processes ever
to have endpoint numbers in the old proc number range.
More return code checks of sys_* functions have been added.
IS has become endpoint-aware. Ditched the 'text' and 'data' fields
in the kernel dump (which show locations, not sizes, so aren't terribly
useful) in favour of the endpoint number. Proc number is still visible.
Some other dumps (e.g. dmap, rs) show endpoint numbers now too which got
the formatting changed.
PM reading segments using rw_seg() has changed - it uses other fields
in the message now instead of encoding the segment and process number and
fd in the fd field. For that it uses _read_pm() and _write_pm() which to
_taskcall()s directly in pm/misc.c.
PM now sys_exit()s itself on panic(), instead of sys_abort().
RS also talks in endpoints instead of process numbers.
2006-03-03 11:20:58 +01:00
|
|
|
|
2012-02-07 16:33:32 +01:00
|
|
|
ret = dev_mess.REP_STATUS;
|
|
|
|
|
2005-04-21 16:53:53 +02:00
|
|
|
/* Task has completed. See if call completed. */
|
2012-02-07 16:33:32 +01:00
|
|
|
if (ret == SUSPEND) {
|
2012-02-20 14:19:29 +01:00
|
|
|
if ((flags & O_NONBLOCK) && !is_asyn) {
|
2005-04-21 16:53:53 +02:00
|
|
|
/* Not supposed to block. */
|
2012-04-13 14:50:38 +02:00
|
|
|
ret = cancel_nblock(dp, minor_dev, job_call_nr, ioproc, gid);
|
2012-02-07 16:33:32 +01:00
|
|
|
if (ret == EINTR)
|
|
|
|
ret = EAGAIN;
|
2005-04-21 16:53:53 +02:00
|
|
|
} else {
|
2009-05-08 15:50:29 +02:00
|
|
|
/* select() will do suspending itself. */
|
|
|
|
if(op != DEV_SELECT) {
|
|
|
|
/* Suspend user. */
|
2009-09-22 23:48:26 +02:00
|
|
|
wait_for(dp->dmap_driver);
|
2009-05-08 15:50:29 +02:00
|
|
|
}
|
2006-06-20 12:12:09 +02:00
|
|
|
assert(!GRANT_VALID(fp->fp_grant));
|
|
|
|
fp->fp_grant = gid; /* revoke this when unsuspended. */
|
2006-08-10 13:51:11 +02:00
|
|
|
fp->fp_ioproc = ioproc;
|
2008-02-22 16:41:07 +01:00
|
|
|
|
2012-02-20 14:19:29 +01:00
|
|
|
if ((flags & O_NONBLOCK) && !is_asyn) {
|
2008-02-22 16:41:07 +01:00
|
|
|
/* Not supposed to block, send cancel message */
|
2012-04-13 14:50:38 +02:00
|
|
|
cancel_nblock(dp, minor_dev, job_call_nr, ioproc, gid);
|
2012-02-07 16:33:32 +01:00
|
|
|
/*
|
|
|
|
* FIXME Should do something about EINTR -> EAGAIN
|
|
|
|
* mapping
|
2008-02-22 16:41:07 +01:00
|
|
|
*/
|
|
|
|
}
|
2005-04-21 16:53:53 +02:00
|
|
|
return(SUSPEND);
|
|
|
|
}
|
|
|
|
}
|
2006-06-20 12:12:09 +02:00
|
|
|
|
|
|
|
/* No suspend, or cancelled suspend, so I/O is over and can be cleaned up. */
|
2012-02-13 16:28:04 +01:00
|
|
|
if(safe) cpf_revoke(gid);
|
2006-06-20 12:12:09 +02:00
|
|
|
|
2012-02-07 16:33:32 +01:00
|
|
|
return ret;
|
2005-04-21 16:53:53 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
/*===========================================================================*
|
|
|
|
* gen_opcl *
|
|
|
|
*===========================================================================*/
|
2012-03-25 20:25:53 +02:00
|
|
|
int gen_opcl(
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
int op, /* operation, (B)DEV_OPEN or (B)DEV_CLOSE */
|
2010-04-13 12:58:41 +02:00
|
|
|
dev_t dev, /* device to open or close */
|
2012-02-13 16:28:04 +01:00
|
|
|
endpoint_t proc_e, /* process to open/close for */
|
2010-04-13 12:58:41 +02:00
|
|
|
int flags /* mode bits and flags */
|
|
|
|
)
|
2005-04-21 16:53:53 +02:00
|
|
|
{
|
2012-02-13 16:28:04 +01:00
|
|
|
/* Called from the dmap struct on opens & closes of special files.*/
|
|
|
|
int r, minor_dev, major_dev, is_bdev;
|
2005-04-21 16:53:53 +02:00
|
|
|
struct dmap *dp;
|
|
|
|
message dev_mess;
|
|
|
|
|
|
|
|
/* Determine task dmap. */
|
2012-02-13 16:28:04 +01:00
|
|
|
major_dev = major(dev);
|
|
|
|
minor_dev = minor(dev);
|
|
|
|
assert(major_dev >= 0 && major_dev < NR_DEVICES);
|
|
|
|
dp = &dmap[major_dev];
|
|
|
|
assert(dp->dmap_driver != NONE);
|
2005-04-21 16:53:53 +02:00
|
|
|
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
is_bdev = IS_BDEV_RQ(op);
|
|
|
|
|
|
|
|
if (is_bdev) {
|
|
|
|
memset(&dev_mess, 0, sizeof(dev_mess));
|
|
|
|
dev_mess.m_type = op;
|
2012-02-13 16:28:04 +01:00
|
|
|
dev_mess.BDEV_MINOR = minor_dev;
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
dev_mess.BDEV_ACCESS = flags;
|
|
|
|
dev_mess.BDEV_ID = 0;
|
|
|
|
} else {
|
|
|
|
dev_mess.m_type = op;
|
2012-02-13 16:28:04 +01:00
|
|
|
dev_mess.DEVICE = minor_dev;
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
dev_mess.USER_ENDPT = proc_e;
|
|
|
|
dev_mess.COUNT = flags;
|
|
|
|
}
|
2005-04-21 16:53:53 +02:00
|
|
|
|
|
|
|
/* Call the task. */
|
2012-02-13 16:28:04 +01:00
|
|
|
r = (*dp->dmap_io)(dp->dmap_driver, &dev_mess);
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
if (r != OK) return(r);
|
2005-04-21 16:53:53 +02:00
|
|
|
|
2012-02-13 16:28:04 +01:00
|
|
|
if (op == DEV_OPEN && dp->dmap_style == STYLE_DEVA) {
|
|
|
|
fp->fp_task = dp->dmap_driver;
|
|
|
|
worker_wait();
|
|
|
|
}
|
|
|
|
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
if (is_bdev)
|
|
|
|
return(dev_mess.BDEV_STATUS);
|
|
|
|
else
|
|
|
|
return(dev_mess.REP_STATUS);
|
2005-04-21 16:53:53 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
/*===========================================================================*
|
|
|
|
* tty_opcl *
|
|
|
|
*===========================================================================*/
|
2012-03-25 20:25:53 +02:00
|
|
|
int tty_opcl(
|
2010-04-13 12:58:41 +02:00
|
|
|
int op, /* operation, DEV_OPEN or DEV_CLOSE */
|
|
|
|
dev_t dev, /* device to open or close */
|
2012-02-13 16:28:04 +01:00
|
|
|
endpoint_t proc_e, /* process to open/close for */
|
2010-04-13 12:58:41 +02:00
|
|
|
int flags /* mode bits and flags */
|
|
|
|
)
|
2005-04-21 16:53:53 +02:00
|
|
|
{
|
|
|
|
/* This procedure is called from the dmap struct on tty open/close. */
|
2012-02-13 16:28:04 +01:00
|
|
|
|
2005-04-21 16:53:53 +02:00
|
|
|
int r;
|
|
|
|
register struct fproc *rfp;
|
|
|
|
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
assert(!IS_BDEV_RQ(op));
|
|
|
|
|
2005-04-21 16:53:53 +02:00
|
|
|
/* Add O_NOCTTY to the flags if this process is not a session leader, or
|
|
|
|
* if it already has a controlling tty, or if it is someone elses
|
|
|
|
* controlling tty.
|
|
|
|
*/
|
2012-02-13 16:28:04 +01:00
|
|
|
if (!(fp->fp_flags & FP_SESLDR) || fp->fp_tty != 0) {
|
2005-04-21 16:53:53 +02:00
|
|
|
flags |= O_NOCTTY;
|
|
|
|
} else {
|
|
|
|
for (rfp = &fproc[0]; rfp < &fproc[NR_PROCS]; rfp++) {
|
2006-03-15 16:34:12 +01:00
|
|
|
if(rfp->fp_pid == PID_FREE) continue;
|
2005-04-21 16:53:53 +02:00
|
|
|
if (rfp->fp_tty == dev) flags |= O_NOCTTY;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
endpoint-aware conversion of servers.
'who', indicating caller number in pm and fs and some other servers, has
been removed in favour of 'who_e' (endpoint) and 'who_p' (proc nr.).
In both PM and FS, isokendpt() convert endpoints to process slot
numbers, returning OK if it was a valid and consistent endpoint number.
okendpt() does the same but panic()s if it doesn't succeed. (In PM,
this is pm_isok..)
pm and fs keep their own records of process endpoints in their proc tables,
which are needed to make kernel calls about those processes.
message field names have changed.
fs drivers are endpoints.
fs now doesn't try to get out of driver deadlock, as the protocol isn't
supposed to let that happen any more. (A warning is printed if ELOCKED
is detected though.)
fproc[].fp_task (indicating which driver the process is suspended on)
became an int.
PM and FS now get endpoint numbers of initial boot processes from the
kernel. These happen to be the same as the old proc numbers, to let
user processes reach them with the old numbers, but FS and PM don't know
that. All new processes after INIT, even after the generation number
wraps around, get endpoint numbers with generation 1 and higher, so
the first instances of the boot processes are the only processes ever
to have endpoint numbers in the old proc number range.
More return code checks of sys_* functions have been added.
IS has become endpoint-aware. Ditched the 'text' and 'data' fields
in the kernel dump (which show locations, not sizes, so aren't terribly
useful) in favour of the endpoint number. Proc number is still visible.
Some other dumps (e.g. dmap, rs) show endpoint numbers now too which got
the formatting changed.
PM reading segments using rw_seg() has changed - it uses other fields
in the message now instead of encoding the segment and process number and
fd in the fd field. For that it uses _read_pm() and _write_pm() which to
_taskcall()s directly in pm/misc.c.
PM now sys_exit()s itself on panic(), instead of sys_abort().
RS also talks in endpoints instead of process numbers.
2006-03-03 11:20:58 +01:00
|
|
|
r = gen_opcl(op, dev, proc_e, flags);
|
2005-04-21 16:53:53 +02:00
|
|
|
|
|
|
|
/* Did this call make the tty the controlling tty? */
|
|
|
|
if (r == 1) {
|
|
|
|
fp->fp_tty = dev;
|
|
|
|
r = OK;
|
|
|
|
}
|
2012-02-13 16:28:04 +01:00
|
|
|
|
2005-04-21 16:53:53 +02:00
|
|
|
return(r);
|
|
|
|
}
|
|
|
|
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
|
2005-04-21 16:53:53 +02:00
|
|
|
/*===========================================================================*
|
|
|
|
* ctty_opcl *
|
|
|
|
*===========================================================================*/
|
2012-03-25 20:25:53 +02:00
|
|
|
int ctty_opcl(
|
2010-04-13 12:58:41 +02:00
|
|
|
int op, /* operation, DEV_OPEN or DEV_CLOSE */
|
2012-02-13 16:28:04 +01:00
|
|
|
dev_t UNUSED(dev), /* device to open or close */
|
|
|
|
endpoint_t UNUSED(proc_e), /* process to open/close for */
|
|
|
|
int UNUSED(flags) /* mode bits and flags */
|
2010-04-13 12:58:41 +02:00
|
|
|
)
|
2005-04-21 16:53:53 +02:00
|
|
|
{
|
2012-02-13 16:28:04 +01:00
|
|
|
/* This procedure is called from the dmap struct on opening or closing
|
2005-04-21 16:53:53 +02:00
|
|
|
* /dev/tty, the magic device that translates to the controlling tty.
|
|
|
|
*/
|
2012-02-13 16:28:04 +01:00
|
|
|
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
assert(!IS_BDEV_RQ(op));
|
|
|
|
|
2005-04-21 16:53:53 +02:00
|
|
|
return(fp->fp_tty == 0 ? ENXIO : OK);
|
|
|
|
}
|
|
|
|
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
|
2005-04-21 16:53:53 +02:00
|
|
|
/*===========================================================================*
|
2006-05-11 16:57:23 +02:00
|
|
|
* pm_setsid *
|
2005-04-21 16:53:53 +02:00
|
|
|
*===========================================================================*/
|
2012-04-13 14:50:38 +02:00
|
|
|
void pm_setsid(endpoint_t proc_e)
|
2005-04-21 16:53:53 +02:00
|
|
|
{
|
2012-02-13 16:28:04 +01:00
|
|
|
/* Perform the VFS side of the SETSID call, i.e. get rid of the controlling
|
2005-04-21 16:53:53 +02:00
|
|
|
* terminal of a process, and make the process a session leader.
|
|
|
|
*/
|
|
|
|
register struct fproc *rfp;
|
endpoint-aware conversion of servers.
'who', indicating caller number in pm and fs and some other servers, has
been removed in favour of 'who_e' (endpoint) and 'who_p' (proc nr.).
In both PM and FS, isokendpt() convert endpoints to process slot
numbers, returning OK if it was a valid and consistent endpoint number.
okendpt() does the same but panic()s if it doesn't succeed. (In PM,
this is pm_isok..)
pm and fs keep their own records of process endpoints in their proc tables,
which are needed to make kernel calls about those processes.
message field names have changed.
fs drivers are endpoints.
fs now doesn't try to get out of driver deadlock, as the protocol isn't
supposed to let that happen any more. (A warning is printed if ELOCKED
is detected though.)
fproc[].fp_task (indicating which driver the process is suspended on)
became an int.
PM and FS now get endpoint numbers of initial boot processes from the
kernel. These happen to be the same as the old proc numbers, to let
user processes reach them with the old numbers, but FS and PM don't know
that. All new processes after INIT, even after the generation number
wraps around, get endpoint numbers with generation 1 and higher, so
the first instances of the boot processes are the only processes ever
to have endpoint numbers in the old proc number range.
More return code checks of sys_* functions have been added.
IS has become endpoint-aware. Ditched the 'text' and 'data' fields
in the kernel dump (which show locations, not sizes, so aren't terribly
useful) in favour of the endpoint number. Proc number is still visible.
Some other dumps (e.g. dmap, rs) show endpoint numbers now too which got
the formatting changed.
PM reading segments using rw_seg() has changed - it uses other fields
in the message now instead of encoding the segment and process number and
fd in the fd field. For that it uses _read_pm() and _write_pm() which to
_taskcall()s directly in pm/misc.c.
PM now sys_exit()s itself on panic(), instead of sys_abort().
RS also talks in endpoints instead of process numbers.
2006-03-03 11:20:58 +01:00
|
|
|
int slot;
|
2005-04-21 16:53:53 +02:00
|
|
|
|
|
|
|
/* Make the process a session leader with no controlling tty. */
|
2006-05-11 16:57:23 +02:00
|
|
|
okendpt(proc_e, &slot);
|
endpoint-aware conversion of servers.
'who', indicating caller number in pm and fs and some other servers, has
been removed in favour of 'who_e' (endpoint) and 'who_p' (proc nr.).
In both PM and FS, isokendpt() convert endpoints to process slot
numbers, returning OK if it was a valid and consistent endpoint number.
okendpt() does the same but panic()s if it doesn't succeed. (In PM,
this is pm_isok..)
pm and fs keep their own records of process endpoints in their proc tables,
which are needed to make kernel calls about those processes.
message field names have changed.
fs drivers are endpoints.
fs now doesn't try to get out of driver deadlock, as the protocol isn't
supposed to let that happen any more. (A warning is printed if ELOCKED
is detected though.)
fproc[].fp_task (indicating which driver the process is suspended on)
became an int.
PM and FS now get endpoint numbers of initial boot processes from the
kernel. These happen to be the same as the old proc numbers, to let
user processes reach them with the old numbers, but FS and PM don't know
that. All new processes after INIT, even after the generation number
wraps around, get endpoint numbers with generation 1 and higher, so
the first instances of the boot processes are the only processes ever
to have endpoint numbers in the old proc number range.
More return code checks of sys_* functions have been added.
IS has become endpoint-aware. Ditched the 'text' and 'data' fields
in the kernel dump (which show locations, not sizes, so aren't terribly
useful) in favour of the endpoint number. Proc number is still visible.
Some other dumps (e.g. dmap, rs) show endpoint numbers now too which got
the formatting changed.
PM reading segments using rw_seg() has changed - it uses other fields
in the message now instead of encoding the segment and process number and
fd in the fd field. For that it uses _read_pm() and _write_pm() which to
_taskcall()s directly in pm/misc.c.
PM now sys_exit()s itself on panic(), instead of sys_abort().
RS also talks in endpoints instead of process numbers.
2006-03-03 11:20:58 +01:00
|
|
|
rfp = &fproc[slot];
|
2012-02-13 16:28:04 +01:00
|
|
|
rfp->fp_flags |= FP_SESLDR;
|
2005-04-21 16:53:53 +02:00
|
|
|
rfp->fp_tty = 0;
|
|
|
|
}
|
|
|
|
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
|
2005-04-21 16:53:53 +02:00
|
|
|
/*===========================================================================*
|
|
|
|
* do_ioctl *
|
|
|
|
*===========================================================================*/
|
2012-03-25 20:25:53 +02:00
|
|
|
int do_ioctl()
|
2005-04-21 16:53:53 +02:00
|
|
|
{
|
2012-04-13 14:50:38 +02:00
|
|
|
/* Perform the ioctl(ls_fd, request, argx) system call */
|
2005-04-21 16:53:53 +02:00
|
|
|
|
2012-04-13 14:50:38 +02:00
|
|
|
int r = OK, suspend_reopen, ioctlrequest;
|
2005-04-21 16:53:53 +02:00
|
|
|
struct filp *f;
|
2006-10-25 15:40:36 +02:00
|
|
|
register struct vnode *vp;
|
2005-04-21 16:53:53 +02:00
|
|
|
dev_t dev;
|
2012-04-13 14:50:38 +02:00
|
|
|
void *argx;
|
2005-04-21 16:53:53 +02:00
|
|
|
|
2012-04-13 14:50:38 +02:00
|
|
|
scratch(fp).file.fd_nr = job_m_in.ls_fd;
|
|
|
|
ioctlrequest = job_m_in.REQUEST;
|
|
|
|
argx = job_m_in.ADDRESS;
|
2012-02-13 16:28:04 +01:00
|
|
|
|
|
|
|
if ((f = get_filp(scratch(fp).file.fd_nr, VNODE_READ)) == NULL)
|
|
|
|
return(err_code);
|
2006-10-25 15:40:36 +02:00
|
|
|
vp = f->filp_vno; /* get vnode pointer */
|
2012-04-25 14:44:42 +02:00
|
|
|
if (!S_ISCHR(vp->v_mode) && !S_ISBLK(vp->v_mode)) {
|
2012-02-13 16:28:04 +01:00
|
|
|
r = ENOTTY;
|
|
|
|
}
|
2005-04-21 16:53:53 +02:00
|
|
|
|
2012-02-13 16:28:04 +01:00
|
|
|
if (r == OK) {
|
VFS: make all IPC asynchronous
By decoupling synchronous drivers from VFS, we are a big step closer to
supporting driver crashes under all circumstances. That is, VFS can't
become stuck on IPC with a synchronous driver (e.g., INET) and can
recover from crashing block drivers during open/close/ioctl or during
communication with an FS.
In order to maintain serialized communication with a synchronous driver,
the communication is wrapped by a mutex on a per driver basis (not major
numbers as there can be multiple majors with identical endpoints). Majors
that share a driver endpoint point to a single mutex object.
In order to support crashes from block drivers, the file reopen tactic
had to be changed; first reopen files associated with the crashed
driver, then send the new driver endpoint to FSes. This solves a
deadlock between the FS and the block driver;
- VFS would send REQ_NEW_DRIVER to an FS, but he FS only receives it
after retrying the current request to the newly started driver.
- The block driver would refuse the retried request until all files
had been reopened.
- VFS would reopen files only after getting a reply from the initial
REQ_NEW_DRIVER.
When a character special driver crashes, all associated files have to
be marked invalid and closed (or reopened if flagged as such). However,
they can only be closed if a thread holds exclusive access to it. To
obtain exclusive access, the worker thread (which handles the new driver
endpoint event from DS) schedules a new job to garbage collect invalid
files. This way, we can signal the worker thread that was talking to the
crashed driver and will release exclusive access to a file associated
with the crashed driver and prevent the garbage collecting worker thread
from dead locking on that file.
Also, when a character special driver crashes, RS will unmap the driver
and remap it upon restart. During unmapping, associated files are marked
invalid instead of waiting for an endpoint up event from DS, as that
event might come later than new read/write/select requests and thus
cause confusion in the freshly started driver.
When locking a filp, the usage counters are no longer checked. The usage
counter can legally go down to zero during filp invalidation while there
are locks pending.
DS events are handled by a separate worker thread instead of the main
thread as reopening files could lead to another crash and a stuck thread.
An additional worker thread is then necessary to unlock it.
Finally, with everything asynchronous a race condition in do_select
surfaced. A select entry was only marked in use after succesfully sending
initial select requests to drivers and having to wait. When multiple
select() calls were handled there was opportunity that these entries
were overwritten. This had as effect that some select results were
ignored (and select() remained blocking instead if returning) or do_select
tried to access filps that were not present (because thrown away by
secondary select()). This bug manifested itself with sendrecs, but was
very hard to reproduce. However, it became awfully easy to trigger with
asynsends only.
2012-08-28 16:06:51 +02:00
|
|
|
suspend_reopen = (f->filp_state & FS_NEEDS_REOPEN);
|
2012-02-13 16:28:04 +01:00
|
|
|
dev = (dev_t) vp->v_sdev;
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
|
2012-04-25 14:44:42 +02:00
|
|
|
if (S_ISBLK(vp->v_mode))
|
2012-04-13 14:50:38 +02:00
|
|
|
r = bdev_ioctl(dev, who_e, ioctlrequest, argx);
|
2012-02-13 16:28:04 +01:00
|
|
|
else
|
2012-04-13 14:50:38 +02:00
|
|
|
r = dev_io(VFS_DEV_IOCTL, dev, who_e, argx, cvu64(0),
|
|
|
|
ioctlrequest, f->filp_flags, suspend_reopen);
|
2012-02-13 16:28:04 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
unlock_filp(f);
|
|
|
|
|
|
|
|
return(r);
|
2005-04-21 16:53:53 +02:00
|
|
|
}
|
|
|
|
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
|
2005-04-21 16:53:53 +02:00
|
|
|
/*===========================================================================*
|
|
|
|
* gen_io *
|
|
|
|
*===========================================================================*/
|
2012-03-25 20:25:53 +02:00
|
|
|
int gen_io(driver_e, mess_ptr)
|
2012-02-13 16:28:04 +01:00
|
|
|
endpoint_t driver_e; /* which endpoint to call */
|
2005-04-21 16:53:53 +02:00
|
|
|
message *mess_ptr; /* pointer to message for task */
|
|
|
|
{
|
|
|
|
/* All file system I/O ultimately comes down to I/O on major/minor device
|
|
|
|
* pairs. These lead to calls on the following routines via the dmap table.
|
|
|
|
*/
|
VFS: make all IPC asynchronous
By decoupling synchronous drivers from VFS, we are a big step closer to
supporting driver crashes under all circumstances. That is, VFS can't
become stuck on IPC with a synchronous driver (e.g., INET) and can
recover from crashing block drivers during open/close/ioctl or during
communication with an FS.
In order to maintain serialized communication with a synchronous driver,
the communication is wrapped by a mutex on a per driver basis (not major
numbers as there can be multiple majors with identical endpoints). Majors
that share a driver endpoint point to a single mutex object.
In order to support crashes from block drivers, the file reopen tactic
had to be changed; first reopen files associated with the crashed
driver, then send the new driver endpoint to FSes. This solves a
deadlock between the FS and the block driver;
- VFS would send REQ_NEW_DRIVER to an FS, but he FS only receives it
after retrying the current request to the newly started driver.
- The block driver would refuse the retried request until all files
had been reopened.
- VFS would reopen files only after getting a reply from the initial
REQ_NEW_DRIVER.
When a character special driver crashes, all associated files have to
be marked invalid and closed (or reopened if flagged as such). However,
they can only be closed if a thread holds exclusive access to it. To
obtain exclusive access, the worker thread (which handles the new driver
endpoint event from DS) schedules a new job to garbage collect invalid
files. This way, we can signal the worker thread that was talking to the
crashed driver and will release exclusive access to a file associated
with the crashed driver and prevent the garbage collecting worker thread
from dead locking on that file.
Also, when a character special driver crashes, RS will unmap the driver
and remap it upon restart. During unmapping, associated files are marked
invalid instead of waiting for an endpoint up event from DS, as that
event might come later than new read/write/select requests and thus
cause confusion in the freshly started driver.
When locking a filp, the usage counters are no longer checked. The usage
counter can legally go down to zero during filp invalidation while there
are locks pending.
DS events are handled by a separate worker thread instead of the main
thread as reopening files could lead to another crash and a stuck thread.
An additional worker thread is then necessary to unlock it.
Finally, with everything asynchronous a race condition in do_select
surfaced. A select entry was only marked in use after succesfully sending
initial select requests to drivers and having to wait. When multiple
select() calls were handled there was opportunity that these entries
were overwritten. This had as effect that some select results were
ignored (and select() remained blocking instead if returning) or do_select
tried to access filps that were not present (because thrown away by
secondary select()). This bug manifested itself with sendrecs, but was
very hard to reproduce. However, it became awfully easy to trigger with
asynsends only.
2012-08-28 16:06:51 +02:00
|
|
|
int r, status, proc_e = NONE, is_bdev, retry_count;
|
|
|
|
message mess_retry;
|
2008-11-19 13:26:10 +01:00
|
|
|
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
is_bdev = IS_BDEV_RQ(mess_ptr->m_type);
|
VFS: make all IPC asynchronous
By decoupling synchronous drivers from VFS, we are a big step closer to
supporting driver crashes under all circumstances. That is, VFS can't
become stuck on IPC with a synchronous driver (e.g., INET) and can
recover from crashing block drivers during open/close/ioctl or during
communication with an FS.
In order to maintain serialized communication with a synchronous driver,
the communication is wrapped by a mutex on a per driver basis (not major
numbers as there can be multiple majors with identical endpoints). Majors
that share a driver endpoint point to a single mutex object.
In order to support crashes from block drivers, the file reopen tactic
had to be changed; first reopen files associated with the crashed
driver, then send the new driver endpoint to FSes. This solves a
deadlock between the FS and the block driver;
- VFS would send REQ_NEW_DRIVER to an FS, but he FS only receives it
after retrying the current request to the newly started driver.
- The block driver would refuse the retried request until all files
had been reopened.
- VFS would reopen files only after getting a reply from the initial
REQ_NEW_DRIVER.
When a character special driver crashes, all associated files have to
be marked invalid and closed (or reopened if flagged as such). However,
they can only be closed if a thread holds exclusive access to it. To
obtain exclusive access, the worker thread (which handles the new driver
endpoint event from DS) schedules a new job to garbage collect invalid
files. This way, we can signal the worker thread that was talking to the
crashed driver and will release exclusive access to a file associated
with the crashed driver and prevent the garbage collecting worker thread
from dead locking on that file.
Also, when a character special driver crashes, RS will unmap the driver
and remap it upon restart. During unmapping, associated files are marked
invalid instead of waiting for an endpoint up event from DS, as that
event might come later than new read/write/select requests and thus
cause confusion in the freshly started driver.
When locking a filp, the usage counters are no longer checked. The usage
counter can legally go down to zero during filp invalidation while there
are locks pending.
DS events are handled by a separate worker thread instead of the main
thread as reopening files could lead to another crash and a stuck thread.
An additional worker thread is then necessary to unlock it.
Finally, with everything asynchronous a race condition in do_select
surfaced. A select entry was only marked in use after succesfully sending
initial select requests to drivers and having to wait. When multiple
select() calls were handled there was opportunity that these entries
were overwritten. This had as effect that some select results were
ignored (and select() remained blocking instead if returning) or do_select
tried to access filps that were not present (because thrown away by
secondary select()). This bug manifested itself with sendrecs, but was
very hard to reproduce. However, it became awfully easy to trigger with
asynsends only.
2012-08-28 16:06:51 +02:00
|
|
|
mess_retry = *mess_ptr;
|
|
|
|
retry_count = 0;
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
|
|
|
|
if (!is_bdev) proc_e = mess_ptr->USER_ENDPT;
|
2010-08-30 15:44:07 +02:00
|
|
|
|
VFS: make all IPC asynchronous
By decoupling synchronous drivers from VFS, we are a big step closer to
supporting driver crashes under all circumstances. That is, VFS can't
become stuck on IPC with a synchronous driver (e.g., INET) and can
recover from crashing block drivers during open/close/ioctl or during
communication with an FS.
In order to maintain serialized communication with a synchronous driver,
the communication is wrapped by a mutex on a per driver basis (not major
numbers as there can be multiple majors with identical endpoints). Majors
that share a driver endpoint point to a single mutex object.
In order to support crashes from block drivers, the file reopen tactic
had to be changed; first reopen files associated with the crashed
driver, then send the new driver endpoint to FSes. This solves a
deadlock between the FS and the block driver;
- VFS would send REQ_NEW_DRIVER to an FS, but he FS only receives it
after retrying the current request to the newly started driver.
- The block driver would refuse the retried request until all files
had been reopened.
- VFS would reopen files only after getting a reply from the initial
REQ_NEW_DRIVER.
When a character special driver crashes, all associated files have to
be marked invalid and closed (or reopened if flagged as such). However,
they can only be closed if a thread holds exclusive access to it. To
obtain exclusive access, the worker thread (which handles the new driver
endpoint event from DS) schedules a new job to garbage collect invalid
files. This way, we can signal the worker thread that was talking to the
crashed driver and will release exclusive access to a file associated
with the crashed driver and prevent the garbage collecting worker thread
from dead locking on that file.
Also, when a character special driver crashes, RS will unmap the driver
and remap it upon restart. During unmapping, associated files are marked
invalid instead of waiting for an endpoint up event from DS, as that
event might come later than new read/write/select requests and thus
cause confusion in the freshly started driver.
When locking a filp, the usage counters are no longer checked. The usage
counter can legally go down to zero during filp invalidation while there
are locks pending.
DS events are handled by a separate worker thread instead of the main
thread as reopening files could lead to another crash and a stuck thread.
An additional worker thread is then necessary to unlock it.
Finally, with everything asynchronous a race condition in do_select
surfaced. A select entry was only marked in use after succesfully sending
initial select requests to drivers and having to wait. When multiple
select() calls were handled there was opportunity that these entries
were overwritten. This had as effect that some select results were
ignored (and select() remained blocking instead if returning) or do_select
tried to access filps that were not present (because thrown away by
secondary select()). This bug manifested itself with sendrecs, but was
very hard to reproduce. However, it became awfully easy to trigger with
asynsends only.
2012-08-28 16:06:51 +02:00
|
|
|
do {
|
|
|
|
r = drv_sendrec(driver_e, mess_ptr);
|
|
|
|
if (r == OK) {
|
|
|
|
if (is_bdev)
|
|
|
|
status = mess_ptr->BDEV_STATUS;
|
|
|
|
else
|
|
|
|
status = mess_ptr->REP_STATUS;
|
|
|
|
if (status == ERESTART) {
|
|
|
|
r = EDEADEPT;
|
|
|
|
*mess_ptr = mess_retry;
|
|
|
|
retry_count++;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
} while (r == EDEADEPT && retry_count < 5);
|
|
|
|
|
2012-02-13 16:28:04 +01:00
|
|
|
if (r != OK) {
|
|
|
|
if (r == EDEADSRCDST || r == EDEADEPT) {
|
|
|
|
printf("VFS: dead driver %d\n", driver_e);
|
|
|
|
dmap_unmap_by_endpt(driver_e);
|
|
|
|
return(r);
|
|
|
|
} else if (r == ELOCKED) {
|
|
|
|
printf("VFS: ELOCKED talking to %d\n", driver_e);
|
|
|
|
return(r);
|
2005-04-21 16:53:53 +02:00
|
|
|
}
|
2012-02-13 16:28:04 +01:00
|
|
|
panic("call_task: can't send/receive: %d", r);
|
|
|
|
}
|
2010-08-30 15:44:07 +02:00
|
|
|
|
2012-02-13 16:28:04 +01:00
|
|
|
/* Did the process we did the sendrec() for get a result? */
|
VFS: make all IPC asynchronous
By decoupling synchronous drivers from VFS, we are a big step closer to
supporting driver crashes under all circumstances. That is, VFS can't
become stuck on IPC with a synchronous driver (e.g., INET) and can
recover from crashing block drivers during open/close/ioctl or during
communication with an FS.
In order to maintain serialized communication with a synchronous driver,
the communication is wrapped by a mutex on a per driver basis (not major
numbers as there can be multiple majors with identical endpoints). Majors
that share a driver endpoint point to a single mutex object.
In order to support crashes from block drivers, the file reopen tactic
had to be changed; first reopen files associated with the crashed
driver, then send the new driver endpoint to FSes. This solves a
deadlock between the FS and the block driver;
- VFS would send REQ_NEW_DRIVER to an FS, but he FS only receives it
after retrying the current request to the newly started driver.
- The block driver would refuse the retried request until all files
had been reopened.
- VFS would reopen files only after getting a reply from the initial
REQ_NEW_DRIVER.
When a character special driver crashes, all associated files have to
be marked invalid and closed (or reopened if flagged as such). However,
they can only be closed if a thread holds exclusive access to it. To
obtain exclusive access, the worker thread (which handles the new driver
endpoint event from DS) schedules a new job to garbage collect invalid
files. This way, we can signal the worker thread that was talking to the
crashed driver and will release exclusive access to a file associated
with the crashed driver and prevent the garbage collecting worker thread
from dead locking on that file.
Also, when a character special driver crashes, RS will unmap the driver
and remap it upon restart. During unmapping, associated files are marked
invalid instead of waiting for an endpoint up event from DS, as that
event might come later than new read/write/select requests and thus
cause confusion in the freshly started driver.
When locking a filp, the usage counters are no longer checked. The usage
counter can legally go down to zero during filp invalidation while there
are locks pending.
DS events are handled by a separate worker thread instead of the main
thread as reopening files could lead to another crash and a stuck thread.
An additional worker thread is then necessary to unlock it.
Finally, with everything asynchronous a race condition in do_select
surfaced. A select entry was only marked in use after succesfully sending
initial select requests to drivers and having to wait. When multiple
select() calls were handled there was opportunity that these entries
were overwritten. This had as effect that some select results were
ignored (and select() remained blocking instead if returning) or do_select
tried to access filps that were not present (because thrown away by
secondary select()). This bug manifested itself with sendrecs, but was
very hard to reproduce. However, it became awfully easy to trigger with
asynsends only.
2012-08-28 16:06:51 +02:00
|
|
|
if (!is_bdev && mess_ptr->REP_ENDPT != proc_e && mess_ptr->m_type != EIO) {
|
2012-02-13 16:28:04 +01:00
|
|
|
printf("VFS: strange device reply from %d, type = %d, "
|
|
|
|
"proc = %d (not %d) (2) ignored\n", mess_ptr->m_source,
|
2010-08-30 15:44:07 +02:00
|
|
|
mess_ptr->m_type, proc_e, mess_ptr->REP_ENDPT);
|
|
|
|
|
2012-02-13 16:28:04 +01:00
|
|
|
return(EIO);
|
VFS: make all IPC asynchronous
By decoupling synchronous drivers from VFS, we are a big step closer to
supporting driver crashes under all circumstances. That is, VFS can't
become stuck on IPC with a synchronous driver (e.g., INET) and can
recover from crashing block drivers during open/close/ioctl or during
communication with an FS.
In order to maintain serialized communication with a synchronous driver,
the communication is wrapped by a mutex on a per driver basis (not major
numbers as there can be multiple majors with identical endpoints). Majors
that share a driver endpoint point to a single mutex object.
In order to support crashes from block drivers, the file reopen tactic
had to be changed; first reopen files associated with the crashed
driver, then send the new driver endpoint to FSes. This solves a
deadlock between the FS and the block driver;
- VFS would send REQ_NEW_DRIVER to an FS, but he FS only receives it
after retrying the current request to the newly started driver.
- The block driver would refuse the retried request until all files
had been reopened.
- VFS would reopen files only after getting a reply from the initial
REQ_NEW_DRIVER.
When a character special driver crashes, all associated files have to
be marked invalid and closed (or reopened if flagged as such). However,
they can only be closed if a thread holds exclusive access to it. To
obtain exclusive access, the worker thread (which handles the new driver
endpoint event from DS) schedules a new job to garbage collect invalid
files. This way, we can signal the worker thread that was talking to the
crashed driver and will release exclusive access to a file associated
with the crashed driver and prevent the garbage collecting worker thread
from dead locking on that file.
Also, when a character special driver crashes, RS will unmap the driver
and remap it upon restart. During unmapping, associated files are marked
invalid instead of waiting for an endpoint up event from DS, as that
event might come later than new read/write/select requests and thus
cause confusion in the freshly started driver.
When locking a filp, the usage counters are no longer checked. The usage
counter can legally go down to zero during filp invalidation while there
are locks pending.
DS events are handled by a separate worker thread instead of the main
thread as reopening files could lead to another crash and a stuck thread.
An additional worker thread is then necessary to unlock it.
Finally, with everything asynchronous a race condition in do_select
surfaced. A select entry was only marked in use after succesfully sending
initial select requests to drivers and having to wait. When multiple
select() calls were handled there was opportunity that these entries
were overwritten. This had as effect that some select results were
ignored (and select() remained blocking instead if returning) or do_select
tried to access filps that were not present (because thrown away by
secondary select()). This bug manifested itself with sendrecs, but was
very hard to reproduce. However, it became awfully easy to trigger with
asynsends only.
2012-08-28 16:06:51 +02:00
|
|
|
} else if (!IS_DRV_REPLY(mess_ptr->m_type))
|
|
|
|
return(EIO);
|
2006-03-15 14:37:20 +01:00
|
|
|
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
return(OK);
|
2005-04-21 16:53:53 +02:00
|
|
|
}
|
|
|
|
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
|
2008-02-22 16:41:07 +01:00
|
|
|
/*===========================================================================*
|
|
|
|
* asyn_io *
|
|
|
|
*===========================================================================*/
|
VFS: make all IPC asynchronous
By decoupling synchronous drivers from VFS, we are a big step closer to
supporting driver crashes under all circumstances. That is, VFS can't
become stuck on IPC with a synchronous driver (e.g., INET) and can
recover from crashing block drivers during open/close/ioctl or during
communication with an FS.
In order to maintain serialized communication with a synchronous driver,
the communication is wrapped by a mutex on a per driver basis (not major
numbers as there can be multiple majors with identical endpoints). Majors
that share a driver endpoint point to a single mutex object.
In order to support crashes from block drivers, the file reopen tactic
had to be changed; first reopen files associated with the crashed
driver, then send the new driver endpoint to FSes. This solves a
deadlock between the FS and the block driver;
- VFS would send REQ_NEW_DRIVER to an FS, but he FS only receives it
after retrying the current request to the newly started driver.
- The block driver would refuse the retried request until all files
had been reopened.
- VFS would reopen files only after getting a reply from the initial
REQ_NEW_DRIVER.
When a character special driver crashes, all associated files have to
be marked invalid and closed (or reopened if flagged as such). However,
they can only be closed if a thread holds exclusive access to it. To
obtain exclusive access, the worker thread (which handles the new driver
endpoint event from DS) schedules a new job to garbage collect invalid
files. This way, we can signal the worker thread that was talking to the
crashed driver and will release exclusive access to a file associated
with the crashed driver and prevent the garbage collecting worker thread
from dead locking on that file.
Also, when a character special driver crashes, RS will unmap the driver
and remap it upon restart. During unmapping, associated files are marked
invalid instead of waiting for an endpoint up event from DS, as that
event might come later than new read/write/select requests and thus
cause confusion in the freshly started driver.
When locking a filp, the usage counters are no longer checked. The usage
counter can legally go down to zero during filp invalidation while there
are locks pending.
DS events are handled by a separate worker thread instead of the main
thread as reopening files could lead to another crash and a stuck thread.
An additional worker thread is then necessary to unlock it.
Finally, with everything asynchronous a race condition in do_select
surfaced. A select entry was only marked in use after succesfully sending
initial select requests to drivers and having to wait. When multiple
select() calls were handled there was opportunity that these entries
were overwritten. This had as effect that some select results were
ignored (and select() remained blocking instead if returning) or do_select
tried to access filps that were not present (because thrown away by
secondary select()). This bug manifested itself with sendrecs, but was
very hard to reproduce. However, it became awfully easy to trigger with
asynsends only.
2012-08-28 16:06:51 +02:00
|
|
|
int asyn_io(endpoint_t drv_e, message *mess_ptr)
|
2008-02-22 16:41:07 +01:00
|
|
|
{
|
|
|
|
/* All file system I/O ultimately comes down to I/O on major/minor device
|
2012-02-13 16:28:04 +01:00
|
|
|
* pairs. These lead to calls on the following routines via the dmap table.
|
2008-02-22 16:41:07 +01:00
|
|
|
*/
|
|
|
|
|
2010-02-19 11:00:32 +01:00
|
|
|
int r;
|
2008-02-22 16:41:07 +01:00
|
|
|
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
assert(!IS_BDEV_RQ(mess_ptr->m_type));
|
VFS: make all IPC asynchronous
By decoupling synchronous drivers from VFS, we are a big step closer to
supporting driver crashes under all circumstances. That is, VFS can't
become stuck on IPC with a synchronous driver (e.g., INET) and can
recover from crashing block drivers during open/close/ioctl or during
communication with an FS.
In order to maintain serialized communication with a synchronous driver,
the communication is wrapped by a mutex on a per driver basis (not major
numbers as there can be multiple majors with identical endpoints). Majors
that share a driver endpoint point to a single mutex object.
In order to support crashes from block drivers, the file reopen tactic
had to be changed; first reopen files associated with the crashed
driver, then send the new driver endpoint to FSes. This solves a
deadlock between the FS and the block driver;
- VFS would send REQ_NEW_DRIVER to an FS, but he FS only receives it
after retrying the current request to the newly started driver.
- The block driver would refuse the retried request until all files
had been reopened.
- VFS would reopen files only after getting a reply from the initial
REQ_NEW_DRIVER.
When a character special driver crashes, all associated files have to
be marked invalid and closed (or reopened if flagged as such). However,
they can only be closed if a thread holds exclusive access to it. To
obtain exclusive access, the worker thread (which handles the new driver
endpoint event from DS) schedules a new job to garbage collect invalid
files. This way, we can signal the worker thread that was talking to the
crashed driver and will release exclusive access to a file associated
with the crashed driver and prevent the garbage collecting worker thread
from dead locking on that file.
Also, when a character special driver crashes, RS will unmap the driver
and remap it upon restart. During unmapping, associated files are marked
invalid instead of waiting for an endpoint up event from DS, as that
event might come later than new read/write/select requests and thus
cause confusion in the freshly started driver.
When locking a filp, the usage counters are no longer checked. The usage
counter can legally go down to zero during filp invalidation while there
are locks pending.
DS events are handled by a separate worker thread instead of the main
thread as reopening files could lead to another crash and a stuck thread.
An additional worker thread is then necessary to unlock it.
Finally, with everything asynchronous a race condition in do_select
surfaced. A select entry was only marked in use after succesfully sending
initial select requests to drivers and having to wait. When multiple
select() calls were handled there was opportunity that these entries
were overwritten. This had as effect that some select results were
ignored (and select() remained blocking instead if returning) or do_select
tried to access filps that were not present (because thrown away by
secondary select()). This bug manifested itself with sendrecs, but was
very hard to reproduce. However, it became awfully easy to trigger with
asynsends only.
2012-08-28 16:06:51 +02:00
|
|
|
self->w_drv_sendrec = mess_ptr; /* Remember where result should be stored */
|
|
|
|
self->w_task = drv_e;
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
|
VFS: make all IPC asynchronous
By decoupling synchronous drivers from VFS, we are a big step closer to
supporting driver crashes under all circumstances. That is, VFS can't
become stuck on IPC with a synchronous driver (e.g., INET) and can
recover from crashing block drivers during open/close/ioctl or during
communication with an FS.
In order to maintain serialized communication with a synchronous driver,
the communication is wrapped by a mutex on a per driver basis (not major
numbers as there can be multiple majors with identical endpoints). Majors
that share a driver endpoint point to a single mutex object.
In order to support crashes from block drivers, the file reopen tactic
had to be changed; first reopen files associated with the crashed
driver, then send the new driver endpoint to FSes. This solves a
deadlock between the FS and the block driver;
- VFS would send REQ_NEW_DRIVER to an FS, but he FS only receives it
after retrying the current request to the newly started driver.
- The block driver would refuse the retried request until all files
had been reopened.
- VFS would reopen files only after getting a reply from the initial
REQ_NEW_DRIVER.
When a character special driver crashes, all associated files have to
be marked invalid and closed (or reopened if flagged as such). However,
they can only be closed if a thread holds exclusive access to it. To
obtain exclusive access, the worker thread (which handles the new driver
endpoint event from DS) schedules a new job to garbage collect invalid
files. This way, we can signal the worker thread that was talking to the
crashed driver and will release exclusive access to a file associated
with the crashed driver and prevent the garbage collecting worker thread
from dead locking on that file.
Also, when a character special driver crashes, RS will unmap the driver
and remap it upon restart. During unmapping, associated files are marked
invalid instead of waiting for an endpoint up event from DS, as that
event might come later than new read/write/select requests and thus
cause confusion in the freshly started driver.
When locking a filp, the usage counters are no longer checked. The usage
counter can legally go down to zero during filp invalidation while there
are locks pending.
DS events are handled by a separate worker thread instead of the main
thread as reopening files could lead to another crash and a stuck thread.
An additional worker thread is then necessary to unlock it.
Finally, with everything asynchronous a race condition in do_select
surfaced. A select entry was only marked in use after succesfully sending
initial select requests to drivers and having to wait. When multiple
select() calls were handled there was opportunity that these entries
were overwritten. This had as effect that some select results were
ignored (and select() remained blocking instead if returning) or do_select
tried to access filps that were not present (because thrown away by
secondary select()). This bug manifested itself with sendrecs, but was
very hard to reproduce. However, it became awfully easy to trigger with
asynsends only.
2012-08-28 16:06:51 +02:00
|
|
|
r = asynsend3(drv_e, mess_ptr, AMF_NOREPLY);
|
2012-02-13 16:28:04 +01:00
|
|
|
|
|
|
|
if (r != OK) panic("VFS: asynsend in asyn_io failed: %d", r);
|
2008-02-22 16:41:07 +01:00
|
|
|
|
|
|
|
/* Fake a SUSPEND */
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
mess_ptr->REP_STATUS = SUSPEND;
|
|
|
|
return(OK);
|
2008-02-22 16:41:07 +01:00
|
|
|
}
|
|
|
|
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
|
2005-04-21 16:53:53 +02:00
|
|
|
/*===========================================================================*
|
|
|
|
* ctty_io *
|
|
|
|
*===========================================================================*/
|
2012-03-25 20:25:53 +02:00
|
|
|
int ctty_io(
|
2012-02-13 16:28:04 +01:00
|
|
|
endpoint_t UNUSED(task_nr), /* not used - for compatibility with dmap_t */
|
|
|
|
message *mess_ptr /* pointer to message for task */
|
|
|
|
)
|
2005-04-21 16:53:53 +02:00
|
|
|
{
|
|
|
|
/* This routine is only called for one device, namely /dev/tty. Its job
|
|
|
|
* is to change the message to use the controlling terminal, instead of the
|
|
|
|
* major/minor pair for /dev/tty itself.
|
|
|
|
*/
|
|
|
|
|
|
|
|
struct dmap *dp;
|
|
|
|
|
|
|
|
if (fp->fp_tty == 0) {
|
|
|
|
/* No controlling tty present anymore, return an I/O error. */
|
|
|
|
mess_ptr->REP_STATUS = EIO;
|
|
|
|
} else {
|
|
|
|
/* Substitute the controlling terminal device. */
|
2012-02-13 16:28:04 +01:00
|
|
|
dp = &dmap[major(fp->fp_tty)];
|
|
|
|
mess_ptr->DEVICE = minor(fp->fp_tty);
|
endpoint-aware conversion of servers.
'who', indicating caller number in pm and fs and some other servers, has
been removed in favour of 'who_e' (endpoint) and 'who_p' (proc nr.).
In both PM and FS, isokendpt() convert endpoints to process slot
numbers, returning OK if it was a valid and consistent endpoint number.
okendpt() does the same but panic()s if it doesn't succeed. (In PM,
this is pm_isok..)
pm and fs keep their own records of process endpoints in their proc tables,
which are needed to make kernel calls about those processes.
message field names have changed.
fs drivers are endpoints.
fs now doesn't try to get out of driver deadlock, as the protocol isn't
supposed to let that happen any more. (A warning is printed if ELOCKED
is detected though.)
fproc[].fp_task (indicating which driver the process is suspended on)
became an int.
PM and FS now get endpoint numbers of initial boot processes from the
kernel. These happen to be the same as the old proc numbers, to let
user processes reach them with the old numbers, but FS and PM don't know
that. All new processes after INIT, even after the generation number
wraps around, get endpoint numbers with generation 1 and higher, so
the first instances of the boot processes are the only processes ever
to have endpoint numbers in the old proc number range.
More return code checks of sys_* functions have been added.
IS has become endpoint-aware. Ditched the 'text' and 'data' fields
in the kernel dump (which show locations, not sizes, so aren't terribly
useful) in favour of the endpoint number. Proc number is still visible.
Some other dumps (e.g. dmap, rs) show endpoint numbers now too which got
the formatting changed.
PM reading segments using rw_seg() has changed - it uses other fields
in the message now instead of encoding the segment and process number and
fd in the fd field. For that it uses _read_pm() and _write_pm() which to
_taskcall()s directly in pm/misc.c.
PM now sys_exit()s itself on panic(), instead of sys_abort().
RS also talks in endpoints instead of process numbers.
2006-03-03 11:20:58 +01:00
|
|
|
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
if (dp->dmap_driver == NONE) {
|
|
|
|
printf("FS: ctty_io: no driver for dev\n");
|
|
|
|
return(EIO);
|
|
|
|
}
|
endpoint-aware conversion of servers.
'who', indicating caller number in pm and fs and some other servers, has
been removed in favour of 'who_e' (endpoint) and 'who_p' (proc nr.).
In both PM and FS, isokendpt() convert endpoints to process slot
numbers, returning OK if it was a valid and consistent endpoint number.
okendpt() does the same but panic()s if it doesn't succeed. (In PM,
this is pm_isok..)
pm and fs keep their own records of process endpoints in their proc tables,
which are needed to make kernel calls about those processes.
message field names have changed.
fs drivers are endpoints.
fs now doesn't try to get out of driver deadlock, as the protocol isn't
supposed to let that happen any more. (A warning is printed if ELOCKED
is detected though.)
fproc[].fp_task (indicating which driver the process is suspended on)
became an int.
PM and FS now get endpoint numbers of initial boot processes from the
kernel. These happen to be the same as the old proc numbers, to let
user processes reach them with the old numbers, but FS and PM don't know
that. All new processes after INIT, even after the generation number
wraps around, get endpoint numbers with generation 1 and higher, so
the first instances of the boot processes are the only processes ever
to have endpoint numbers in the old proc number range.
More return code checks of sys_* functions have been added.
IS has become endpoint-aware. Ditched the 'text' and 'data' fields
in the kernel dump (which show locations, not sizes, so aren't terribly
useful) in favour of the endpoint number. Proc number is still visible.
Some other dumps (e.g. dmap, rs) show endpoint numbers now too which got
the formatting changed.
PM reading segments using rw_seg() has changed - it uses other fields
in the message now instead of encoding the segment and process number and
fd in the fd field. For that it uses _read_pm() and _write_pm() which to
_taskcall()s directly in pm/misc.c.
PM now sys_exit()s itself on panic(), instead of sys_abort().
RS also talks in endpoints instead of process numbers.
2006-03-03 11:20:58 +01:00
|
|
|
|
2012-02-13 16:28:04 +01:00
|
|
|
if (isokendpt(dp->dmap_driver, &dummyproc) != OK) {
|
|
|
|
printf("VFS: ctty_io: old driver %d\n", dp->dmap_driver);
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
return(EIO);
|
endpoint-aware conversion of servers.
'who', indicating caller number in pm and fs and some other servers, has
been removed in favour of 'who_e' (endpoint) and 'who_p' (proc nr.).
In both PM and FS, isokendpt() convert endpoints to process slot
numbers, returning OK if it was a valid and consistent endpoint number.
okendpt() does the same but panic()s if it doesn't succeed. (In PM,
this is pm_isok..)
pm and fs keep their own records of process endpoints in their proc tables,
which are needed to make kernel calls about those processes.
message field names have changed.
fs drivers are endpoints.
fs now doesn't try to get out of driver deadlock, as the protocol isn't
supposed to let that happen any more. (A warning is printed if ELOCKED
is detected though.)
fproc[].fp_task (indicating which driver the process is suspended on)
became an int.
PM and FS now get endpoint numbers of initial boot processes from the
kernel. These happen to be the same as the old proc numbers, to let
user processes reach them with the old numbers, but FS and PM don't know
that. All new processes after INIT, even after the generation number
wraps around, get endpoint numbers with generation 1 and higher, so
the first instances of the boot processes are the only processes ever
to have endpoint numbers in the old proc number range.
More return code checks of sys_* functions have been added.
IS has become endpoint-aware. Ditched the 'text' and 'data' fields
in the kernel dump (which show locations, not sizes, so aren't terribly
useful) in favour of the endpoint number. Proc number is still visible.
Some other dumps (e.g. dmap, rs) show endpoint numbers now too which got
the formatting changed.
PM reading segments using rw_seg() has changed - it uses other fields
in the message now instead of encoding the segment and process number and
fd in the fd field. For that it uses _read_pm() and _write_pm() which to
_taskcall()s directly in pm/misc.c.
PM now sys_exit()s itself on panic(), instead of sys_abort().
RS also talks in endpoints instead of process numbers.
2006-03-03 11:20:58 +01:00
|
|
|
}
|
|
|
|
|
2005-04-21 16:53:53 +02:00
|
|
|
(*dp->dmap_io)(dp->dmap_driver, mess_ptr);
|
|
|
|
}
|
2012-02-13 16:28:04 +01:00
|
|
|
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
return(OK);
|
2005-04-21 16:53:53 +02:00
|
|
|
}
|
|
|
|
|
2006-10-25 15:40:36 +02:00
|
|
|
|
2005-04-21 16:53:53 +02:00
|
|
|
/*===========================================================================*
|
|
|
|
* no_dev *
|
|
|
|
*===========================================================================*/
|
2012-03-25 20:25:53 +02:00
|
|
|
int no_dev(
|
2010-04-13 12:58:41 +02:00
|
|
|
int UNUSED(op), /* operation, DEV_OPEN or DEV_CLOSE */
|
|
|
|
dev_t UNUSED(dev), /* device to open or close */
|
|
|
|
int UNUSED(proc), /* process to open/close for */
|
|
|
|
int UNUSED(flags) /* mode bits and flags */
|
|
|
|
)
|
2005-04-21 16:53:53 +02:00
|
|
|
{
|
|
|
|
/* Called when opening a nonexistent device. */
|
|
|
|
return(ENODEV);
|
|
|
|
}
|
|
|
|
|
2005-10-05 17:38:15 +02:00
|
|
|
/*===========================================================================*
|
|
|
|
* no_dev_io *
|
|
|
|
*===========================================================================*/
|
2012-03-25 20:25:53 +02:00
|
|
|
int no_dev_io(endpoint_t UNUSED(proc), message *UNUSED(m))
|
2005-10-05 17:38:15 +02:00
|
|
|
{
|
|
|
|
/* Called when doing i/o on a nonexistent device. */
|
2006-10-25 15:40:36 +02:00
|
|
|
printf("VFS: I/O on unmapped device number\n");
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
return(EIO);
|
2005-10-05 17:38:15 +02:00
|
|
|
}
|
|
|
|
|
2006-10-25 15:40:36 +02:00
|
|
|
|
2005-04-21 16:53:53 +02:00
|
|
|
/*===========================================================================*
|
|
|
|
* clone_opcl *
|
|
|
|
*===========================================================================*/
|
2012-03-25 20:25:53 +02:00
|
|
|
int clone_opcl(
|
2010-04-13 12:58:41 +02:00
|
|
|
int op, /* operation, DEV_OPEN or DEV_CLOSE */
|
|
|
|
dev_t dev, /* device to open or close */
|
|
|
|
int proc_e, /* process to open/close for */
|
|
|
|
int flags /* mode bits and flags */
|
|
|
|
)
|
2005-04-21 16:53:53 +02:00
|
|
|
{
|
|
|
|
/* Some devices need special processing upon open. Such a device is "cloned",
|
|
|
|
* i.e. on a succesful open it is replaced by a new device with a new unique
|
|
|
|
* minor device number. This new device number identifies a new object (such
|
|
|
|
* as a new network connection) that has been allocated within a task.
|
|
|
|
*/
|
|
|
|
struct dmap *dp;
|
2012-02-13 16:28:04 +01:00
|
|
|
int r, minor_dev, major_dev;
|
2005-04-21 16:53:53 +02:00
|
|
|
message dev_mess;
|
|
|
|
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
assert(!IS_BDEV_RQ(op));
|
|
|
|
|
2005-04-21 16:53:53 +02:00
|
|
|
/* Determine task dmap. */
|
2012-02-13 16:28:04 +01:00
|
|
|
minor_dev = minor(dev);
|
|
|
|
major_dev = major(dev);
|
|
|
|
assert(major_dev >= 0 && major_dev < NR_DEVICES);
|
|
|
|
dp = &dmap[major_dev];
|
|
|
|
assert(dp->dmap_driver != NONE);
|
2005-04-21 16:53:53 +02:00
|
|
|
|
|
|
|
dev_mess.m_type = op;
|
2012-02-13 16:28:04 +01:00
|
|
|
dev_mess.DEVICE = minor_dev;
|
Server/driver protocols: no longer allow third-party copies.
Before safecopies, the IO_ENDPT and DL_ENDPT message fields were needed
to know which actual process to copy data from/to, as that process may
not always be the caller. Now that we have full safecopy support, these
fields have become useless for that purpose: the owner of the grant is
*always* the caller. Allowing the caller to supply another endpoint is
in fact dangerous, because the callee may then end up using a grant
from a third party. One could call this a variant of the confused
deputy problem.
From now on, safecopy calls should always use the caller's endpoint as
grant owner. This fully obsoletes the DL_ENDPT field in the
inet/ethernet protocol. IO_ENDPT has other uses besides identifying the
grant owner though. This patch renames IO_ENDPT to USER_ENDPT, not only
because that is a more fitting name (it should never be used for I/O
after all), but also in order to intentionally break any old system
source code outside the base system. If this patch breaks your code,
fixing it is fairly simple:
- DL_ENDPT should be replaced with m_source;
- IO_ENDPT should be replaced with m_source when used for safecopies;
- IO_ENDPT should be replaced with USER_ENDPT for any other use, e.g.
when setting REP_ENDPT, matching requests in CANCEL calls, getting
DEV_SELECT flags, and retrieving of the real user process's endpoint
in DEV_OPEN.
The changes in this patch are binary backward compatible.
2011-04-11 19:35:05 +02:00
|
|
|
dev_mess.USER_ENDPT = proc_e;
|
2005-04-21 16:53:53 +02:00
|
|
|
dev_mess.COUNT = flags;
|
|
|
|
|
endpoint-aware conversion of servers.
'who', indicating caller number in pm and fs and some other servers, has
been removed in favour of 'who_e' (endpoint) and 'who_p' (proc nr.).
In both PM and FS, isokendpt() convert endpoints to process slot
numbers, returning OK if it was a valid and consistent endpoint number.
okendpt() does the same but panic()s if it doesn't succeed. (In PM,
this is pm_isok..)
pm and fs keep their own records of process endpoints in their proc tables,
which are needed to make kernel calls about those processes.
message field names have changed.
fs drivers are endpoints.
fs now doesn't try to get out of driver deadlock, as the protocol isn't
supposed to let that happen any more. (A warning is printed if ELOCKED
is detected though.)
fproc[].fp_task (indicating which driver the process is suspended on)
became an int.
PM and FS now get endpoint numbers of initial boot processes from the
kernel. These happen to be the same as the old proc numbers, to let
user processes reach them with the old numbers, but FS and PM don't know
that. All new processes after INIT, even after the generation number
wraps around, get endpoint numbers with generation 1 and higher, so
the first instances of the boot processes are the only processes ever
to have endpoint numbers in the old proc number range.
More return code checks of sys_* functions have been added.
IS has become endpoint-aware. Ditched the 'text' and 'data' fields
in the kernel dump (which show locations, not sizes, so aren't terribly
useful) in favour of the endpoint number. Proc number is still visible.
Some other dumps (e.g. dmap, rs) show endpoint numbers now too which got
the formatting changed.
PM reading segments using rw_seg() has changed - it uses other fields
in the message now instead of encoding the segment and process number and
fd in the fd field. For that it uses _read_pm() and _write_pm() which to
_taskcall()s directly in pm/misc.c.
PM now sys_exit()s itself on panic(), instead of sys_abort().
RS also talks in endpoints instead of process numbers.
2006-03-03 11:20:58 +01:00
|
|
|
if(isokendpt(dp->dmap_driver, &dummyproc) != OK) {
|
2012-02-13 16:28:04 +01:00
|
|
|
printf("VFS clone_opcl: bad driver endpoint for major %d (%d)\n",
|
|
|
|
major_dev, dp->dmap_driver);
|
|
|
|
return(ENXIO);
|
endpoint-aware conversion of servers.
'who', indicating caller number in pm and fs and some other servers, has
been removed in favour of 'who_e' (endpoint) and 'who_p' (proc nr.).
In both PM and FS, isokendpt() convert endpoints to process slot
numbers, returning OK if it was a valid and consistent endpoint number.
okendpt() does the same but panic()s if it doesn't succeed. (In PM,
this is pm_isok..)
pm and fs keep their own records of process endpoints in their proc tables,
which are needed to make kernel calls about those processes.
message field names have changed.
fs drivers are endpoints.
fs now doesn't try to get out of driver deadlock, as the protocol isn't
supposed to let that happen any more. (A warning is printed if ELOCKED
is detected though.)
fproc[].fp_task (indicating which driver the process is suspended on)
became an int.
PM and FS now get endpoint numbers of initial boot processes from the
kernel. These happen to be the same as the old proc numbers, to let
user processes reach them with the old numbers, but FS and PM don't know
that. All new processes after INIT, even after the generation number
wraps around, get endpoint numbers with generation 1 and higher, so
the first instances of the boot processes are the only processes ever
to have endpoint numbers in the old proc number range.
More return code checks of sys_* functions have been added.
IS has become endpoint-aware. Ditched the 'text' and 'data' fields
in the kernel dump (which show locations, not sizes, so aren't terribly
useful) in favour of the endpoint number. Proc number is still visible.
Some other dumps (e.g. dmap, rs) show endpoint numbers now too which got
the formatting changed.
PM reading segments using rw_seg() has changed - it uses other fields
in the message now instead of encoding the segment and process number and
fd in the fd field. For that it uses _read_pm() and _write_pm() which to
_taskcall()s directly in pm/misc.c.
PM now sys_exit()s itself on panic(), instead of sys_abort().
RS also talks in endpoints instead of process numbers.
2006-03-03 11:20:58 +01:00
|
|
|
}
|
|
|
|
|
2005-04-21 16:53:53 +02:00
|
|
|
/* Call the task. */
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
r = (*dp->dmap_io)(dp->dmap_driver, &dev_mess);
|
|
|
|
if (r != OK) return(r);
|
2005-04-21 16:53:53 +02:00
|
|
|
|
2012-02-07 16:41:25 +01:00
|
|
|
if (op == DEV_OPEN && dev_style_asyn(dp->dmap_style)) {
|
2012-02-13 16:28:04 +01:00
|
|
|
/* Wait for reply when driver is asynchronous */
|
|
|
|
fp->fp_task = dp->dmap_driver;
|
|
|
|
worker_wait();
|
|
|
|
}
|
|
|
|
|
2005-04-21 16:53:53 +02:00
|
|
|
if (op == DEV_OPEN && dev_mess.REP_STATUS >= 0) {
|
2012-02-13 16:28:04 +01:00
|
|
|
if (dev_mess.REP_STATUS != minor_dev) {
|
2012-07-26 14:28:03 +02:00
|
|
|
struct vnode *vp;
|
|
|
|
struct node_details res;
|
2007-08-07 14:52:47 +02:00
|
|
|
|
2006-10-25 15:40:36 +02:00
|
|
|
/* A new minor device number has been returned.
|
2012-07-26 14:28:03 +02:00
|
|
|
* Request PFS to create a temporary device file to hold it.
|
|
|
|
*/
|
2012-02-13 16:28:04 +01:00
|
|
|
|
2012-07-26 14:28:03 +02:00
|
|
|
/* Device number of the new device. */
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
dev = (dev & ~(BYTE << MINOR)) | (dev_mess.REP_STATUS << MINOR);
|
2006-10-25 15:40:36 +02:00
|
|
|
|
2012-07-26 14:28:03 +02:00
|
|
|
/* Issue request */
|
2010-03-30 17:00:09 +02:00
|
|
|
r = req_newnode(PFS_PROC_NR, fp->fp_effuid, fp->fp_effgid,
|
2012-07-26 14:28:03 +02:00
|
|
|
ALL_MODES | I_CHAR_SPECIAL, dev, &res);
|
|
|
|
if (r != OK) {
|
2012-02-13 16:28:04 +01:00
|
|
|
(void) clone_opcl(DEV_CLOSE, dev, proc_e, 0);
|
|
|
|
return r;
|
2012-07-26 14:28:03 +02:00
|
|
|
}
|
2012-02-13 16:28:04 +01:00
|
|
|
|
2012-07-26 14:28:03 +02:00
|
|
|
/* Drop old node and use the new values */
|
2012-02-13 16:28:04 +01:00
|
|
|
if ((vp = get_free_vnode()) == NULL)
|
|
|
|
return(err_code);
|
|
|
|
lock_vnode(vp, VNODE_OPCL);
|
|
|
|
|
2012-07-26 14:28:03 +02:00
|
|
|
assert(FD_ISSET(scratch(fp).file.fd_nr, &fp->fp_filp_inuse));
|
|
|
|
unlock_vnode(fp->fp_filp[scratch(fp).file.fd_nr]->filp_vno);
|
|
|
|
put_vnode(fp->fp_filp[scratch(fp).file.fd_nr]->filp_vno);
|
|
|
|
|
2006-10-25 15:40:36 +02:00
|
|
|
vp->v_fs_e = res.fs_e;
|
2010-05-10 15:26:00 +02:00
|
|
|
vp->v_vmnt = NULL;
|
2012-02-13 16:28:04 +01:00
|
|
|
vp->v_dev = NO_DEV;
|
2007-08-07 14:52:47 +02:00
|
|
|
vp->v_fs_e = res.fs_e;
|
2006-10-25 15:40:36 +02:00
|
|
|
vp->v_inode_nr = res.inode_nr;
|
2012-02-13 16:28:04 +01:00
|
|
|
vp->v_mode = res.fmode;
|
2006-10-25 15:40:36 +02:00
|
|
|
vp->v_sdev = dev;
|
2007-01-05 17:36:55 +01:00
|
|
|
vp->v_fs_count = 1;
|
|
|
|
vp->v_ref_count = 1;
|
2012-02-13 16:28:04 +01:00
|
|
|
fp->fp_filp[scratch(fp).file.fd_nr]->filp_vno = vp;
|
2005-04-21 16:53:53 +02:00
|
|
|
}
|
|
|
|
dev_mess.REP_STATUS = OK;
|
|
|
|
}
|
|
|
|
return(dev_mess.REP_STATUS);
|
|
|
|
}
|
2005-07-27 15:08:52 +02:00
|
|
|
|
2007-08-07 14:52:47 +02:00
|
|
|
|
2005-10-20 21:39:32 +02:00
|
|
|
/*===========================================================================*
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
* bdev_up *
|
2005-10-20 21:39:32 +02:00
|
|
|
*===========================================================================*/
|
2012-03-25 20:25:53 +02:00
|
|
|
void bdev_up(int maj)
|
2005-10-20 21:39:32 +02:00
|
|
|
{
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
/* A new block device driver has been mapped in. This may affect both mounted
|
|
|
|
* file systems and open block-special files.
|
|
|
|
*/
|
2011-11-30 19:05:26 +01:00
|
|
|
int r, found, bits;
|
2012-02-13 16:28:04 +01:00
|
|
|
struct filp *rfilp;
|
2008-02-22 16:41:07 +01:00
|
|
|
struct vmnt *vmp;
|
|
|
|
struct vnode *vp;
|
2011-11-30 19:05:26 +01:00
|
|
|
char *label;
|
2008-02-22 16:41:07 +01:00
|
|
|
|
2012-02-13 16:28:04 +01:00
|
|
|
if (maj < 0 || maj >= NR_DEVICES) panic("VFS: out-of-bound major");
|
2011-11-30 19:05:26 +01:00
|
|
|
label = dmap[maj].dmap_label;
|
VFS: make all IPC asynchronous
By decoupling synchronous drivers from VFS, we are a big step closer to
supporting driver crashes under all circumstances. That is, VFS can't
become stuck on IPC with a synchronous driver (e.g., INET) and can
recover from crashing block drivers during open/close/ioctl or during
communication with an FS.
In order to maintain serialized communication with a synchronous driver,
the communication is wrapped by a mutex on a per driver basis (not major
numbers as there can be multiple majors with identical endpoints). Majors
that share a driver endpoint point to a single mutex object.
In order to support crashes from block drivers, the file reopen tactic
had to be changed; first reopen files associated with the crashed
driver, then send the new driver endpoint to FSes. This solves a
deadlock between the FS and the block driver;
- VFS would send REQ_NEW_DRIVER to an FS, but he FS only receives it
after retrying the current request to the newly started driver.
- The block driver would refuse the retried request until all files
had been reopened.
- VFS would reopen files only after getting a reply from the initial
REQ_NEW_DRIVER.
When a character special driver crashes, all associated files have to
be marked invalid and closed (or reopened if flagged as such). However,
they can only be closed if a thread holds exclusive access to it. To
obtain exclusive access, the worker thread (which handles the new driver
endpoint event from DS) schedules a new job to garbage collect invalid
files. This way, we can signal the worker thread that was talking to the
crashed driver and will release exclusive access to a file associated
with the crashed driver and prevent the garbage collecting worker thread
from dead locking on that file.
Also, when a character special driver crashes, RS will unmap the driver
and remap it upon restart. During unmapping, associated files are marked
invalid instead of waiting for an endpoint up event from DS, as that
event might come later than new read/write/select requests and thus
cause confusion in the freshly started driver.
When locking a filp, the usage counters are no longer checked. The usage
counter can legally go down to zero during filp invalidation while there
are locks pending.
DS events are handled by a separate worker thread instead of the main
thread as reopening files could lead to another crash and a stuck thread.
An additional worker thread is then necessary to unlock it.
Finally, with everything asynchronous a race condition in do_select
surfaced. A select entry was only marked in use after succesfully sending
initial select requests to drivers and having to wait. When multiple
select() calls were handled there was opportunity that these entries
were overwritten. This had as effect that some select results were
ignored (and select() remained blocking instead if returning) or do_select
tried to access filps that were not present (because thrown away by
secondary select()). This bug manifested itself with sendrecs, but was
very hard to reproduce. However, it became awfully easy to trigger with
asynsends only.
2012-08-28 16:06:51 +02:00
|
|
|
found = 0;
|
2006-10-25 15:40:36 +02:00
|
|
|
|
2011-03-25 11:56:43 +01:00
|
|
|
/* For each block-special file that was previously opened on the affected
|
|
|
|
* device, we need to reopen it on the new driver.
|
|
|
|
*/
|
2012-02-13 16:28:04 +01:00
|
|
|
for (rfilp = filp; rfilp < &filp[NR_FILPS]; rfilp++) {
|
|
|
|
if (rfilp->filp_count < 1 || !(vp = rfilp->filp_vno)) continue;
|
|
|
|
if (major(vp->v_sdev) != maj) continue;
|
|
|
|
if (!S_ISBLK(vp->v_mode)) continue;
|
2011-03-25 11:56:43 +01:00
|
|
|
|
|
|
|
/* Reopen the device on the driver, once per filp. */
|
2012-02-13 16:28:04 +01:00
|
|
|
bits = mode_map[rfilp->filp_mode & O_ACCMODE];
|
VFS: make all IPC asynchronous
By decoupling synchronous drivers from VFS, we are a big step closer to
supporting driver crashes under all circumstances. That is, VFS can't
become stuck on IPC with a synchronous driver (e.g., INET) and can
recover from crashing block drivers during open/close/ioctl or during
communication with an FS.
In order to maintain serialized communication with a synchronous driver,
the communication is wrapped by a mutex on a per driver basis (not major
numbers as there can be multiple majors with identical endpoints). Majors
that share a driver endpoint point to a single mutex object.
In order to support crashes from block drivers, the file reopen tactic
had to be changed; first reopen files associated with the crashed
driver, then send the new driver endpoint to FSes. This solves a
deadlock between the FS and the block driver;
- VFS would send REQ_NEW_DRIVER to an FS, but he FS only receives it
after retrying the current request to the newly started driver.
- The block driver would refuse the retried request until all files
had been reopened.
- VFS would reopen files only after getting a reply from the initial
REQ_NEW_DRIVER.
When a character special driver crashes, all associated files have to
be marked invalid and closed (or reopened if flagged as such). However,
they can only be closed if a thread holds exclusive access to it. To
obtain exclusive access, the worker thread (which handles the new driver
endpoint event from DS) schedules a new job to garbage collect invalid
files. This way, we can signal the worker thread that was talking to the
crashed driver and will release exclusive access to a file associated
with the crashed driver and prevent the garbage collecting worker thread
from dead locking on that file.
Also, when a character special driver crashes, RS will unmap the driver
and remap it upon restart. During unmapping, associated files are marked
invalid instead of waiting for an endpoint up event from DS, as that
event might come later than new read/write/select requests and thus
cause confusion in the freshly started driver.
When locking a filp, the usage counters are no longer checked. The usage
counter can legally go down to zero during filp invalidation while there
are locks pending.
DS events are handled by a separate worker thread instead of the main
thread as reopening files could lead to another crash and a stuck thread.
An additional worker thread is then necessary to unlock it.
Finally, with everything asynchronous a race condition in do_select
surfaced. A select entry was only marked in use after succesfully sending
initial select requests to drivers and having to wait. When multiple
select() calls were handled there was opportunity that these entries
were overwritten. This had as effect that some select results were
ignored (and select() remained blocking instead if returning) or do_select
tried to access filps that were not present (because thrown away by
secondary select()). This bug manifested itself with sendrecs, but was
very hard to reproduce. However, it became awfully easy to trigger with
asynsends only.
2012-08-28 16:06:51 +02:00
|
|
|
if ((r = bdev_open(vp->v_sdev, bits)) != OK) {
|
2011-03-25 11:56:43 +01:00
|
|
|
printf("VFS: mounted dev %d/%d re-open failed: %d.\n",
|
|
|
|
maj, minor(vp->v_sdev), r);
|
VFS: make all IPC asynchronous
By decoupling synchronous drivers from VFS, we are a big step closer to
supporting driver crashes under all circumstances. That is, VFS can't
become stuck on IPC with a synchronous driver (e.g., INET) and can
recover from crashing block drivers during open/close/ioctl or during
communication with an FS.
In order to maintain serialized communication with a synchronous driver,
the communication is wrapped by a mutex on a per driver basis (not major
numbers as there can be multiple majors with identical endpoints). Majors
that share a driver endpoint point to a single mutex object.
In order to support crashes from block drivers, the file reopen tactic
had to be changed; first reopen files associated with the crashed
driver, then send the new driver endpoint to FSes. This solves a
deadlock between the FS and the block driver;
- VFS would send REQ_NEW_DRIVER to an FS, but he FS only receives it
after retrying the current request to the newly started driver.
- The block driver would refuse the retried request until all files
had been reopened.
- VFS would reopen files only after getting a reply from the initial
REQ_NEW_DRIVER.
When a character special driver crashes, all associated files have to
be marked invalid and closed (or reopened if flagged as such). However,
they can only be closed if a thread holds exclusive access to it. To
obtain exclusive access, the worker thread (which handles the new driver
endpoint event from DS) schedules a new job to garbage collect invalid
files. This way, we can signal the worker thread that was talking to the
crashed driver and will release exclusive access to a file associated
with the crashed driver and prevent the garbage collecting worker thread
from dead locking on that file.
Also, when a character special driver crashes, RS will unmap the driver
and remap it upon restart. During unmapping, associated files are marked
invalid instead of waiting for an endpoint up event from DS, as that
event might come later than new read/write/select requests and thus
cause confusion in the freshly started driver.
When locking a filp, the usage counters are no longer checked. The usage
counter can legally go down to zero during filp invalidation while there
are locks pending.
DS events are handled by a separate worker thread instead of the main
thread as reopening files could lead to another crash and a stuck thread.
An additional worker thread is then necessary to unlock it.
Finally, with everything asynchronous a race condition in do_select
surfaced. A select entry was only marked in use after succesfully sending
initial select requests to drivers and having to wait. When multiple
select() calls were handled there was opportunity that these entries
were overwritten. This had as effect that some select results were
ignored (and select() remained blocking instead if returning) or do_select
tried to access filps that were not present (because thrown away by
secondary select()). This bug manifested itself with sendrecs, but was
very hard to reproduce. However, it became awfully easy to trigger with
asynsends only.
2012-08-28 16:06:51 +02:00
|
|
|
dmap[maj].dmap_recovering = 0;
|
|
|
|
return; /* Give up entirely */
|
|
|
|
}
|
2011-03-25 11:56:43 +01:00
|
|
|
|
|
|
|
found = 1;
|
|
|
|
}
|
|
|
|
|
VFS: make all IPC asynchronous
By decoupling synchronous drivers from VFS, we are a big step closer to
supporting driver crashes under all circumstances. That is, VFS can't
become stuck on IPC with a synchronous driver (e.g., INET) and can
recover from crashing block drivers during open/close/ioctl or during
communication with an FS.
In order to maintain serialized communication with a synchronous driver,
the communication is wrapped by a mutex on a per driver basis (not major
numbers as there can be multiple majors with identical endpoints). Majors
that share a driver endpoint point to a single mutex object.
In order to support crashes from block drivers, the file reopen tactic
had to be changed; first reopen files associated with the crashed
driver, then send the new driver endpoint to FSes. This solves a
deadlock between the FS and the block driver;
- VFS would send REQ_NEW_DRIVER to an FS, but he FS only receives it
after retrying the current request to the newly started driver.
- The block driver would refuse the retried request until all files
had been reopened.
- VFS would reopen files only after getting a reply from the initial
REQ_NEW_DRIVER.
When a character special driver crashes, all associated files have to
be marked invalid and closed (or reopened if flagged as such). However,
they can only be closed if a thread holds exclusive access to it. To
obtain exclusive access, the worker thread (which handles the new driver
endpoint event from DS) schedules a new job to garbage collect invalid
files. This way, we can signal the worker thread that was talking to the
crashed driver and will release exclusive access to a file associated
with the crashed driver and prevent the garbage collecting worker thread
from dead locking on that file.
Also, when a character special driver crashes, RS will unmap the driver
and remap it upon restart. During unmapping, associated files are marked
invalid instead of waiting for an endpoint up event from DS, as that
event might come later than new read/write/select requests and thus
cause confusion in the freshly started driver.
When locking a filp, the usage counters are no longer checked. The usage
counter can legally go down to zero during filp invalidation while there
are locks pending.
DS events are handled by a separate worker thread instead of the main
thread as reopening files could lead to another crash and a stuck thread.
An additional worker thread is then necessary to unlock it.
Finally, with everything asynchronous a race condition in do_select
surfaced. A select entry was only marked in use after succesfully sending
initial select requests to drivers and having to wait. When multiple
select() calls were handled there was opportunity that these entries
were overwritten. This had as effect that some select results were
ignored (and select() remained blocking instead if returning) or do_select
tried to access filps that were not present (because thrown away by
secondary select()). This bug manifested itself with sendrecs, but was
very hard to reproduce. However, it became awfully easy to trigger with
asynsends only.
2012-08-28 16:06:51 +02:00
|
|
|
/* Tell each affected mounted file system about the new endpoint.
|
|
|
|
*/
|
|
|
|
for (vmp = &vmnt[0]; vmp < &vmnt[NR_MNTS]; ++vmp) {
|
|
|
|
if (major(vmp->m_dev) != maj) continue;
|
|
|
|
|
|
|
|
/* Send the driver label to the mounted file system. */
|
|
|
|
if (OK != req_newdriver(vmp->m_fs_e, vmp->m_dev, label))
|
|
|
|
printf("VFS dev_up: error sending new driver label to %d\n",
|
|
|
|
vmp->m_fs_e);
|
|
|
|
}
|
|
|
|
|
2011-03-25 11:56:43 +01:00
|
|
|
/* If any block-special file was open for this major at all, also inform the
|
2012-02-13 16:28:04 +01:00
|
|
|
* root file system about the new driver. We do this even if the
|
|
|
|
* block-special file is linked to another mounted file system, merely
|
2011-03-25 11:56:43 +01:00
|
|
|
* because it is more work to check for that case.
|
|
|
|
*/
|
|
|
|
if (found) {
|
2011-11-30 19:05:26 +01:00
|
|
|
if (OK != req_newdriver(ROOT_FS_E, makedev(maj, 0), label))
|
|
|
|
printf("VFSdev_up: error sending new driver label to %d\n",
|
2012-02-13 16:28:04 +01:00
|
|
|
ROOT_FS_E);
|
2011-03-25 11:56:43 +01:00
|
|
|
}
|
VFS: make all IPC asynchronous
By decoupling synchronous drivers from VFS, we are a big step closer to
supporting driver crashes under all circumstances. That is, VFS can't
become stuck on IPC with a synchronous driver (e.g., INET) and can
recover from crashing block drivers during open/close/ioctl or during
communication with an FS.
In order to maintain serialized communication with a synchronous driver,
the communication is wrapped by a mutex on a per driver basis (not major
numbers as there can be multiple majors with identical endpoints). Majors
that share a driver endpoint point to a single mutex object.
In order to support crashes from block drivers, the file reopen tactic
had to be changed; first reopen files associated with the crashed
driver, then send the new driver endpoint to FSes. This solves a
deadlock between the FS and the block driver;
- VFS would send REQ_NEW_DRIVER to an FS, but he FS only receives it
after retrying the current request to the newly started driver.
- The block driver would refuse the retried request until all files
had been reopened.
- VFS would reopen files only after getting a reply from the initial
REQ_NEW_DRIVER.
When a character special driver crashes, all associated files have to
be marked invalid and closed (or reopened if flagged as such). However,
they can only be closed if a thread holds exclusive access to it. To
obtain exclusive access, the worker thread (which handles the new driver
endpoint event from DS) schedules a new job to garbage collect invalid
files. This way, we can signal the worker thread that was talking to the
crashed driver and will release exclusive access to a file associated
with the crashed driver and prevent the garbage collecting worker thread
from dead locking on that file.
Also, when a character special driver crashes, RS will unmap the driver
and remap it upon restart. During unmapping, associated files are marked
invalid instead of waiting for an endpoint up event from DS, as that
event might come later than new read/write/select requests and thus
cause confusion in the freshly started driver.
When locking a filp, the usage counters are no longer checked. The usage
counter can legally go down to zero during filp invalidation while there
are locks pending.
DS events are handled by a separate worker thread instead of the main
thread as reopening files could lead to another crash and a stuck thread.
An additional worker thread is then necessary to unlock it.
Finally, with everything asynchronous a race condition in do_select
surfaced. A select entry was only marked in use after succesfully sending
initial select requests to drivers and having to wait. When multiple
select() calls were handled there was opportunity that these entries
were overwritten. This had as effect that some select results were
ignored (and select() remained blocking instead if returning) or do_select
tried to access filps that were not present (because thrown away by
secondary select()). This bug manifested itself with sendrecs, but was
very hard to reproduce. However, it became awfully easy to trigger with
asynsends only.
2012-08-28 16:06:51 +02:00
|
|
|
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
}
|
|
|
|
|
2011-03-25 11:56:43 +01:00
|
|
|
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
/*===========================================================================*
|
|
|
|
* cdev_up *
|
|
|
|
*===========================================================================*/
|
2012-03-25 20:25:53 +02:00
|
|
|
void cdev_up(int maj)
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
{
|
|
|
|
/* A new character device driver has been mapped in.
|
2012-02-13 16:28:04 +01:00
|
|
|
*/
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
int needs_reopen, fd_nr;
|
2012-02-13 16:28:04 +01:00
|
|
|
struct filp *rfilp;
|
Split block/character protocols and libdriver
This patch separates the character and block driver communication
protocols. The old character protocol remains the same, but a new
block protocol is introduced. The libdriver library is replaced by
two new libraries: libchardriver and libblockdriver. Their exposed
API, and drivers that use them, have been updated accordingly.
Together, libbdev and libblockdriver now completely abstract away
the message format used by the block protocol. As the memory driver
is both a character and a block device driver, it now implements its
own message loop.
The most important semantic change made to the block protocol is that
it is no longer possible to return both partial results and an error
for a single transfer. This simplifies the interaction between the
caller and the driver, as the I/O vector no longer needs to be copied
back. Also, drivers are now no longer supposed to decide based on the
layout of the I/O vector when a transfer should be cut short. Put
simply, transfers are now supposed to either succeed completely, or
result in an error.
After this patch, the state of the various pieces is as follows:
- block protocol: stable
- libbdev API: stable for synchronous communication
- libblockdriver API: needs slight revision (the drvlib/partition API
in particular; the threading API will also change shortly)
- character protocol: needs cleanup
- libchardriver API: needs cleanup accordingly
- driver restarts: largely unsupported until endpoint changes are
reintroduced
As a side effect, this patch eliminates several bugs, hacks, and gcc
-Wall and -W warnings all over the place. It probably introduces a
few new ones, too.
Update warning: this patch changes the protocol between MFS and disk
drivers, so in order to use old/new images, the MFS from the ramdisk
must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
|
|
|
struct fproc *rfp;
|
|
|
|
struct vnode *vp;
|
|
|
|
|
VFS: make all IPC asynchronous
By decoupling synchronous drivers from VFS, we are a big step closer to
supporting driver crashes under all circumstances. That is, VFS can't
become stuck on IPC with a synchronous driver (e.g., INET) and can
recover from crashing block drivers during open/close/ioctl or during
communication with an FS.
In order to maintain serialized communication with a synchronous driver,
the communication is wrapped by a mutex on a per driver basis (not major
numbers as there can be multiple majors with identical endpoints). Majors
that share a driver endpoint point to a single mutex object.
In order to support crashes from block drivers, the file reopen tactic
had to be changed; first reopen files associated with the crashed
driver, then send the new driver endpoint to FSes. This solves a
deadlock between the FS and the block driver;
- VFS would send REQ_NEW_DRIVER to an FS, but he FS only receives it
after retrying the current request to the newly started driver.
- The block driver would refuse the retried request until all files
had been reopened.
- VFS would reopen files only after getting a reply from the initial
REQ_NEW_DRIVER.
When a character special driver crashes, all associated files have to
be marked invalid and closed (or reopened if flagged as such). However,
they can only be closed if a thread holds exclusive access to it. To
obtain exclusive access, the worker thread (which handles the new driver
endpoint event from DS) schedules a new job to garbage collect invalid
files. This way, we can signal the worker thread that was talking to the
crashed driver and will release exclusive access to a file associated
with the crashed driver and prevent the garbage collecting worker thread
from dead locking on that file.
Also, when a character special driver crashes, RS will unmap the driver
and remap it upon restart. During unmapping, associated files are marked
invalid instead of waiting for an endpoint up event from DS, as that
event might come later than new read/write/select requests and thus
cause confusion in the freshly started driver.
When locking a filp, the usage counters are no longer checked. The usage
counter can legally go down to zero during filp invalidation while there
are locks pending.
DS events are handled by a separate worker thread instead of the main
thread as reopening files could lead to another crash and a stuck thread.
An additional worker thread is then necessary to unlock it.
Finally, with everything asynchronous a race condition in do_select
surfaced. A select entry was only marked in use after succesfully sending
initial select requests to drivers and having to wait. When multiple
select() calls were handled there was opportunity that these entries
were overwritten. This had as effect that some select results were
ignored (and select() remained blocking instead if returning) or do_select
tried to access filps that were not present (because thrown away by
secondary select()). This bug manifested itself with sendrecs, but was
very hard to reproduce. However, it became awfully easy to trigger with
asynsends only.
2012-08-28 16:06:51 +02:00
|
|
|
/* Look for processes that are suspended in an OPEN call. Set FP_SUSP_REOPEN
|
2008-02-22 16:41:07 +01:00
|
|
|
* to indicate that this process was suspended before the call to dev_up.
|
|
|
|
*/
|
|
|
|
for (rfp = &fproc[0]; rfp < &fproc[NR_PROCS]; rfp++) {
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
if(rfp->fp_pid == PID_FREE) continue;
|
|
|
|
if(rfp->fp_blocked_on != FP_BLOCKED_ON_DOPEN) continue;
|
2008-02-22 16:41:07 +01:00
|
|
|
|
2012-02-13 16:28:04 +01:00
|
|
|
fd_nr = scratch(rfp).file.fd_nr;
|
|
|
|
printf("VFS: dev_up: found process in FP_BLOCKED_ON_DOPEN, fd %d\n",
|
|
|
|
fd_nr);
|
|
|
|
rfilp = rfp->fp_filp[fd_nr];
|
|
|
|
vp = rfilp->filp_vno;
|
|
|
|
if (!vp) panic("VFS: cdev_up: no vp");
|
2012-04-25 14:44:42 +02:00
|
|
|
if (!S_ISCHR(vp->v_mode)) continue;
|
2012-02-13 16:28:04 +01:00
|
|
|
if (major(vp->v_sdev) != maj) continue;
|
2008-02-22 16:41:07 +01:00
|
|
|
|
2012-02-13 16:28:04 +01:00
|
|
|
rfp->fp_flags |= FP_SUSP_REOPEN;
|
2008-02-22 16:41:07 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
needs_reopen= FALSE;
|
2012-02-13 16:28:04 +01:00
|
|
|
for (rfilp = filp; rfilp < &filp[NR_FILPS]; rfilp++) {
|
|
|
|
if (rfilp->filp_count < 1 || !(vp = rfilp->filp_vno)) continue;
|
|
|
|
if (major(vp->v_sdev) != maj) continue;
|
|
|
|
if (!S_ISCHR(vp->v_mode)) continue;
|
2008-02-22 16:41:07 +01:00
|
|
|
|
VFS: make all IPC asynchronous
By decoupling synchronous drivers from VFS, we are a big step closer to
supporting driver crashes under all circumstances. That is, VFS can't
become stuck on IPC with a synchronous driver (e.g., INET) and can
recover from crashing block drivers during open/close/ioctl or during
communication with an FS.
In order to maintain serialized communication with a synchronous driver,
the communication is wrapped by a mutex on a per driver basis (not major
numbers as there can be multiple majors with identical endpoints). Majors
that share a driver endpoint point to a single mutex object.
In order to support crashes from block drivers, the file reopen tactic
had to be changed; first reopen files associated with the crashed
driver, then send the new driver endpoint to FSes. This solves a
deadlock between the FS and the block driver;
- VFS would send REQ_NEW_DRIVER to an FS, but he FS only receives it
after retrying the current request to the newly started driver.
- The block driver would refuse the retried request until all files
had been reopened.
- VFS would reopen files only after getting a reply from the initial
REQ_NEW_DRIVER.
When a character special driver crashes, all associated files have to
be marked invalid and closed (or reopened if flagged as such). However,
they can only be closed if a thread holds exclusive access to it. To
obtain exclusive access, the worker thread (which handles the new driver
endpoint event from DS) schedules a new job to garbage collect invalid
files. This way, we can signal the worker thread that was talking to the
crashed driver and will release exclusive access to a file associated
with the crashed driver and prevent the garbage collecting worker thread
from dead locking on that file.
Also, when a character special driver crashes, RS will unmap the driver
and remap it upon restart. During unmapping, associated files are marked
invalid instead of waiting for an endpoint up event from DS, as that
event might come later than new read/write/select requests and thus
cause confusion in the freshly started driver.
When locking a filp, the usage counters are no longer checked. The usage
counter can legally go down to zero during filp invalidation while there
are locks pending.
DS events are handled by a separate worker thread instead of the main
thread as reopening files could lead to another crash and a stuck thread.
An additional worker thread is then necessary to unlock it.
Finally, with everything asynchronous a race condition in do_select
surfaced. A select entry was only marked in use after succesfully sending
initial select requests to drivers and having to wait. When multiple
select() calls were handled there was opportunity that these entries
were overwritten. This had as effect that some select results were
ignored (and select() remained blocking instead if returning) or do_select
tried to access filps that were not present (because thrown away by
secondary select()). This bug manifested itself with sendrecs, but was
very hard to reproduce. However, it became awfully easy to trigger with
asynsends only.
2012-08-28 16:06:51 +02:00
|
|
|
rfilp->filp_state |= FS_NEEDS_REOPEN;
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
needs_reopen = TRUE;
|
2008-02-22 16:41:07 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
if (needs_reopen)
|
|
|
|
restart_reopen(maj);
|
|
|
|
}
|
|
|
|
|
2012-02-13 16:28:04 +01:00
|
|
|
/*===========================================================================*
|
|
|
|
* open_reply *
|
|
|
|
*===========================================================================*/
|
2012-03-25 20:25:53 +02:00
|
|
|
void open_reply(void)
|
2012-02-13 16:28:04 +01:00
|
|
|
{
|
|
|
|
struct fproc *rfp;
|
VFS: make all IPC asynchronous
By decoupling synchronous drivers from VFS, we are a big step closer to
supporting driver crashes under all circumstances. That is, VFS can't
become stuck on IPC with a synchronous driver (e.g., INET) and can
recover from crashing block drivers during open/close/ioctl or during
communication with an FS.
In order to maintain serialized communication with a synchronous driver,
the communication is wrapped by a mutex on a per driver basis (not major
numbers as there can be multiple majors with identical endpoints). Majors
that share a driver endpoint point to a single mutex object.
In order to support crashes from block drivers, the file reopen tactic
had to be changed; first reopen files associated with the crashed
driver, then send the new driver endpoint to FSes. This solves a
deadlock between the FS and the block driver;
- VFS would send REQ_NEW_DRIVER to an FS, but he FS only receives it
after retrying the current request to the newly started driver.
- The block driver would refuse the retried request until all files
had been reopened.
- VFS would reopen files only after getting a reply from the initial
REQ_NEW_DRIVER.
When a character special driver crashes, all associated files have to
be marked invalid and closed (or reopened if flagged as such). However,
they can only be closed if a thread holds exclusive access to it. To
obtain exclusive access, the worker thread (which handles the new driver
endpoint event from DS) schedules a new job to garbage collect invalid
files. This way, we can signal the worker thread that was talking to the
crashed driver and will release exclusive access to a file associated
with the crashed driver and prevent the garbage collecting worker thread
from dead locking on that file.
Also, when a character special driver crashes, RS will unmap the driver
and remap it upon restart. During unmapping, associated files are marked
invalid instead of waiting for an endpoint up event from DS, as that
event might come later than new read/write/select requests and thus
cause confusion in the freshly started driver.
When locking a filp, the usage counters are no longer checked. The usage
counter can legally go down to zero during filp invalidation while there
are locks pending.
DS events are handled by a separate worker thread instead of the main
thread as reopening files could lead to another crash and a stuck thread.
An additional worker thread is then necessary to unlock it.
Finally, with everything asynchronous a race condition in do_select
surfaced. A select entry was only marked in use after succesfully sending
initial select requests to drivers and having to wait. When multiple
select() calls were handled there was opportunity that these entries
were overwritten. This had as effect that some select results were
ignored (and select() remained blocking instead if returning) or do_select
tried to access filps that were not present (because thrown away by
secondary select()). This bug manifested itself with sendrecs, but was
very hard to reproduce. However, it became awfully easy to trigger with
asynsends only.
2012-08-28 16:06:51 +02:00
|
|
|
struct worker_thread *wp;
|
2012-02-13 16:28:04 +01:00
|
|
|
endpoint_t proc_e;
|
|
|
|
int slot;
|
|
|
|
|
2012-04-13 14:50:38 +02:00
|
|
|
proc_e = job_m_in.REP_ENDPT;
|
2012-02-13 16:28:04 +01:00
|
|
|
if (isokendpt(proc_e, &slot) != OK) return;
|
|
|
|
rfp = &fproc[slot];
|
VFS: make all IPC asynchronous
By decoupling synchronous drivers from VFS, we are a big step closer to
supporting driver crashes under all circumstances. That is, VFS can't
become stuck on IPC with a synchronous driver (e.g., INET) and can
recover from crashing block drivers during open/close/ioctl or during
communication with an FS.
In order to maintain serialized communication with a synchronous driver,
the communication is wrapped by a mutex on a per driver basis (not major
numbers as there can be multiple majors with identical endpoints). Majors
that share a driver endpoint point to a single mutex object.
In order to support crashes from block drivers, the file reopen tactic
had to be changed; first reopen files associated with the crashed
driver, then send the new driver endpoint to FSes. This solves a
deadlock between the FS and the block driver;
- VFS would send REQ_NEW_DRIVER to an FS, but he FS only receives it
after retrying the current request to the newly started driver.
- The block driver would refuse the retried request until all files
had been reopened.
- VFS would reopen files only after getting a reply from the initial
REQ_NEW_DRIVER.
When a character special driver crashes, all associated files have to
be marked invalid and closed (or reopened if flagged as such). However,
they can only be closed if a thread holds exclusive access to it. To
obtain exclusive access, the worker thread (which handles the new driver
endpoint event from DS) schedules a new job to garbage collect invalid
files. This way, we can signal the worker thread that was talking to the
crashed driver and will release exclusive access to a file associated
with the crashed driver and prevent the garbage collecting worker thread
from dead locking on that file.
Also, when a character special driver crashes, RS will unmap the driver
and remap it upon restart. During unmapping, associated files are marked
invalid instead of waiting for an endpoint up event from DS, as that
event might come later than new read/write/select requests and thus
cause confusion in the freshly started driver.
When locking a filp, the usage counters are no longer checked. The usage
counter can legally go down to zero during filp invalidation while there
are locks pending.
DS events are handled by a separate worker thread instead of the main
thread as reopening files could lead to another crash and a stuck thread.
An additional worker thread is then necessary to unlock it.
Finally, with everything asynchronous a race condition in do_select
surfaced. A select entry was only marked in use after succesfully sending
initial select requests to drivers and having to wait. When multiple
select() calls were handled there was opportunity that these entries
were overwritten. This had as effect that some select results were
ignored (and select() remained blocking instead if returning) or do_select
tried to access filps that were not present (because thrown away by
secondary select()). This bug manifested itself with sendrecs, but was
very hard to reproduce. However, it became awfully easy to trigger with
asynsends only.
2012-08-28 16:06:51 +02:00
|
|
|
wp = worker_get(rfp->fp_wtid);
|
|
|
|
if (wp == NULL || wp->w_task != who_e) {
|
|
|
|
printf("VFS: no worker thread waiting for a reply from %d\n", who_e);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
*wp->w_drv_sendrec = job_m_in;
|
|
|
|
wp->w_drv_sendrec = NULL;
|
|
|
|
wp->w_task = NONE;
|
|
|
|
worker_signal(wp); /* Continue open */
|
|
|
|
}
|
|
|
|
|
|
|
|
/*===========================================================================*
|
|
|
|
* dev_reply *
|
|
|
|
*===========================================================================*/
|
|
|
|
void dev_reply(struct dmap *dp)
|
|
|
|
{
|
|
|
|
struct worker_thread *wp;
|
|
|
|
|
|
|
|
assert(dp != NULL);
|
|
|
|
assert(dp->dmap_servicing != NONE);
|
|
|
|
|
|
|
|
wp = worker_get(dp->dmap_servicing);
|
|
|
|
if (wp == NULL || wp->w_task != who_e) {
|
|
|
|
printf("VFS: no worker thread waiting for a reply from %d\n",
|
|
|
|
who_e);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
assert(wp->w_drv_sendrec != NULL);
|
|
|
|
*wp->w_drv_sendrec = m_in;
|
|
|
|
wp->w_drv_sendrec = NULL;
|
|
|
|
worker_signal(wp);
|
2012-02-13 16:28:04 +01:00
|
|
|
}
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
|
2008-02-22 16:41:07 +01:00
|
|
|
/*===========================================================================*
|
|
|
|
* restart_reopen *
|
|
|
|
*===========================================================================*/
|
2012-03-25 20:25:53 +02:00
|
|
|
static void restart_reopen(maj)
|
2008-02-22 16:41:07 +01:00
|
|
|
int maj;
|
|
|
|
{
|
2012-02-13 16:28:04 +01:00
|
|
|
int n, r, minor_dev, major_dev, fd_nr;
|
2008-02-22 16:41:07 +01:00
|
|
|
endpoint_t driver_e;
|
|
|
|
struct vnode *vp;
|
2012-02-13 16:28:04 +01:00
|
|
|
struct filp *rfilp;
|
2008-02-22 16:41:07 +01:00
|
|
|
struct fproc *rfp;
|
|
|
|
|
2012-02-13 16:28:04 +01:00
|
|
|
if (maj < 0 || maj >= NR_DEVICES) panic("VFS: out-of-bound major");
|
|
|
|
for (rfilp = filp; rfilp < &filp[NR_FILPS]; rfilp++) {
|
|
|
|
if (rfilp->filp_count < 1 || !(vp = rfilp->filp_vno)) continue;
|
VFS: make all IPC asynchronous
By decoupling synchronous drivers from VFS, we are a big step closer to
supporting driver crashes under all circumstances. That is, VFS can't
become stuck on IPC with a synchronous driver (e.g., INET) and can
recover from crashing block drivers during open/close/ioctl or during
communication with an FS.
In order to maintain serialized communication with a synchronous driver,
the communication is wrapped by a mutex on a per driver basis (not major
numbers as there can be multiple majors with identical endpoints). Majors
that share a driver endpoint point to a single mutex object.
In order to support crashes from block drivers, the file reopen tactic
had to be changed; first reopen files associated with the crashed
driver, then send the new driver endpoint to FSes. This solves a
deadlock between the FS and the block driver;
- VFS would send REQ_NEW_DRIVER to an FS, but he FS only receives it
after retrying the current request to the newly started driver.
- The block driver would refuse the retried request until all files
had been reopened.
- VFS would reopen files only after getting a reply from the initial
REQ_NEW_DRIVER.
When a character special driver crashes, all associated files have to
be marked invalid and closed (or reopened if flagged as such). However,
they can only be closed if a thread holds exclusive access to it. To
obtain exclusive access, the worker thread (which handles the new driver
endpoint event from DS) schedules a new job to garbage collect invalid
files. This way, we can signal the worker thread that was talking to the
crashed driver and will release exclusive access to a file associated
with the crashed driver and prevent the garbage collecting worker thread
from dead locking on that file.
Also, when a character special driver crashes, RS will unmap the driver
and remap it upon restart. During unmapping, associated files are marked
invalid instead of waiting for an endpoint up event from DS, as that
event might come later than new read/write/select requests and thus
cause confusion in the freshly started driver.
When locking a filp, the usage counters are no longer checked. The usage
counter can legally go down to zero during filp invalidation while there
are locks pending.
DS events are handled by a separate worker thread instead of the main
thread as reopening files could lead to another crash and a stuck thread.
An additional worker thread is then necessary to unlock it.
Finally, with everything asynchronous a race condition in do_select
surfaced. A select entry was only marked in use after succesfully sending
initial select requests to drivers and having to wait. When multiple
select() calls were handled there was opportunity that these entries
were overwritten. This had as effect that some select results were
ignored (and select() remained blocking instead if returning) or do_select
tried to access filps that were not present (because thrown away by
secondary select()). This bug manifested itself with sendrecs, but was
very hard to reproduce. However, it became awfully easy to trigger with
asynsends only.
2012-08-28 16:06:51 +02:00
|
|
|
if (!(rfilp->filp_state & FS_NEEDS_REOPEN)) continue;
|
2012-04-25 14:44:42 +02:00
|
|
|
if (!S_ISCHR(vp->v_mode)) continue;
|
2012-02-13 16:28:04 +01:00
|
|
|
|
|
|
|
major_dev = major(vp->v_sdev);
|
|
|
|
minor_dev = minor(vp->v_sdev);
|
|
|
|
if (major_dev != maj) continue;
|
|
|
|
|
VFS: make all IPC asynchronous
By decoupling synchronous drivers from VFS, we are a big step closer to
supporting driver crashes under all circumstances. That is, VFS can't
become stuck on IPC with a synchronous driver (e.g., INET) and can
recover from crashing block drivers during open/close/ioctl or during
communication with an FS.
In order to maintain serialized communication with a synchronous driver,
the communication is wrapped by a mutex on a per driver basis (not major
numbers as there can be multiple majors with identical endpoints). Majors
that share a driver endpoint point to a single mutex object.
In order to support crashes from block drivers, the file reopen tactic
had to be changed; first reopen files associated with the crashed
driver, then send the new driver endpoint to FSes. This solves a
deadlock between the FS and the block driver;
- VFS would send REQ_NEW_DRIVER to an FS, but he FS only receives it
after retrying the current request to the newly started driver.
- The block driver would refuse the retried request until all files
had been reopened.
- VFS would reopen files only after getting a reply from the initial
REQ_NEW_DRIVER.
When a character special driver crashes, all associated files have to
be marked invalid and closed (or reopened if flagged as such). However,
they can only be closed if a thread holds exclusive access to it. To
obtain exclusive access, the worker thread (which handles the new driver
endpoint event from DS) schedules a new job to garbage collect invalid
files. This way, we can signal the worker thread that was talking to the
crashed driver and will release exclusive access to a file associated
with the crashed driver and prevent the garbage collecting worker thread
from dead locking on that file.
Also, when a character special driver crashes, RS will unmap the driver
and remap it upon restart. During unmapping, associated files are marked
invalid instead of waiting for an endpoint up event from DS, as that
event might come later than new read/write/select requests and thus
cause confusion in the freshly started driver.
When locking a filp, the usage counters are no longer checked. The usage
counter can legally go down to zero during filp invalidation while there
are locks pending.
DS events are handled by a separate worker thread instead of the main
thread as reopening files could lead to another crash and a stuck thread.
An additional worker thread is then necessary to unlock it.
Finally, with everything asynchronous a race condition in do_select
surfaced. A select entry was only marked in use after succesfully sending
initial select requests to drivers and having to wait. When multiple
select() calls were handled there was opportunity that these entries
were overwritten. This had as effect that some select results were
ignored (and select() remained blocking instead if returning) or do_select
tried to access filps that were not present (because thrown away by
secondary select()). This bug manifested itself with sendrecs, but was
very hard to reproduce. However, it became awfully easy to trigger with
asynsends only.
2012-08-28 16:06:51 +02:00
|
|
|
if (rfilp->filp_flags & O_REOPEN) {
|
|
|
|
/* Try to reopen a file upon driver restart */
|
|
|
|
r = dev_reopen(vp->v_sdev, rfilp-filp,
|
|
|
|
vp->v_mode & (R_BIT|W_BIT));
|
|
|
|
|
|
|
|
if (r == OK)
|
|
|
|
return;
|
2008-02-22 16:41:07 +01:00
|
|
|
|
2012-09-17 17:29:38 +02:00
|
|
|
printf("VFS: file on dev %d/%d re-open failed: %d\n",
|
|
|
|
major_dev, minor_dev, r);
|
VFS: make all IPC asynchronous
By decoupling synchronous drivers from VFS, we are a big step closer to
supporting driver crashes under all circumstances. That is, VFS can't
become stuck on IPC with a synchronous driver (e.g., INET) and can
recover from crashing block drivers during open/close/ioctl or during
communication with an FS.
In order to maintain serialized communication with a synchronous driver,
the communication is wrapped by a mutex on a per driver basis (not major
numbers as there can be multiple majors with identical endpoints). Majors
that share a driver endpoint point to a single mutex object.
In order to support crashes from block drivers, the file reopen tactic
had to be changed; first reopen files associated with the crashed
driver, then send the new driver endpoint to FSes. This solves a
deadlock between the FS and the block driver;
- VFS would send REQ_NEW_DRIVER to an FS, but he FS only receives it
after retrying the current request to the newly started driver.
- The block driver would refuse the retried request until all files
had been reopened.
- VFS would reopen files only after getting a reply from the initial
REQ_NEW_DRIVER.
When a character special driver crashes, all associated files have to
be marked invalid and closed (or reopened if flagged as such). However,
they can only be closed if a thread holds exclusive access to it. To
obtain exclusive access, the worker thread (which handles the new driver
endpoint event from DS) schedules a new job to garbage collect invalid
files. This way, we can signal the worker thread that was talking to the
crashed driver and will release exclusive access to a file associated
with the crashed driver and prevent the garbage collecting worker thread
from dead locking on that file.
Also, when a character special driver crashes, RS will unmap the driver
and remap it upon restart. During unmapping, associated files are marked
invalid instead of waiting for an endpoint up event from DS, as that
event might come later than new read/write/select requests and thus
cause confusion in the freshly started driver.
When locking a filp, the usage counters are no longer checked. The usage
counter can legally go down to zero during filp invalidation while there
are locks pending.
DS events are handled by a separate worker thread instead of the main
thread as reopening files could lead to another crash and a stuck thread.
An additional worker thread is then necessary to unlock it.
Finally, with everything asynchronous a race condition in do_select
surfaced. A select entry was only marked in use after succesfully sending
initial select requests to drivers and having to wait. When multiple
select() calls were handled there was opportunity that these entries
were overwritten. This had as effect that some select results were
ignored (and select() remained blocking instead if returning) or do_select
tried to access filps that were not present (because thrown away by
secondary select()). This bug manifested itself with sendrecs, but was
very hard to reproduce. However, it became awfully easy to trigger with
asynsends only.
2012-08-28 16:06:51 +02:00
|
|
|
}
|
2008-02-22 16:41:07 +01:00
|
|
|
|
VFS: make all IPC asynchronous
By decoupling synchronous drivers from VFS, we are a big step closer to
supporting driver crashes under all circumstances. That is, VFS can't
become stuck on IPC with a synchronous driver (e.g., INET) and can
recover from crashing block drivers during open/close/ioctl or during
communication with an FS.
In order to maintain serialized communication with a synchronous driver,
the communication is wrapped by a mutex on a per driver basis (not major
numbers as there can be multiple majors with identical endpoints). Majors
that share a driver endpoint point to a single mutex object.
In order to support crashes from block drivers, the file reopen tactic
had to be changed; first reopen files associated with the crashed
driver, then send the new driver endpoint to FSes. This solves a
deadlock between the FS and the block driver;
- VFS would send REQ_NEW_DRIVER to an FS, but he FS only receives it
after retrying the current request to the newly started driver.
- The block driver would refuse the retried request until all files
had been reopened.
- VFS would reopen files only after getting a reply from the initial
REQ_NEW_DRIVER.
When a character special driver crashes, all associated files have to
be marked invalid and closed (or reopened if flagged as such). However,
they can only be closed if a thread holds exclusive access to it. To
obtain exclusive access, the worker thread (which handles the new driver
endpoint event from DS) schedules a new job to garbage collect invalid
files. This way, we can signal the worker thread that was talking to the
crashed driver and will release exclusive access to a file associated
with the crashed driver and prevent the garbage collecting worker thread
from dead locking on that file.
Also, when a character special driver crashes, RS will unmap the driver
and remap it upon restart. During unmapping, associated files are marked
invalid instead of waiting for an endpoint up event from DS, as that
event might come later than new read/write/select requests and thus
cause confusion in the freshly started driver.
When locking a filp, the usage counters are no longer checked. The usage
counter can legally go down to zero during filp invalidation while there
are locks pending.
DS events are handled by a separate worker thread instead of the main
thread as reopening files could lead to another crash and a stuck thread.
An additional worker thread is then necessary to unlock it.
Finally, with everything asynchronous a race condition in do_select
surfaced. A select entry was only marked in use after succesfully sending
initial select requests to drivers and having to wait. When multiple
select() calls were handled there was opportunity that these entries
were overwritten. This had as effect that some select results were
ignored (and select() remained blocking instead if returning) or do_select
tried to access filps that were not present (because thrown away by
secondary select()). This bug manifested itself with sendrecs, but was
very hard to reproduce. However, it became awfully easy to trigger with
asynsends only.
2012-08-28 16:06:51 +02:00
|
|
|
/* File descriptor is to be closed when driver restarts. */
|
2012-02-13 16:28:04 +01:00
|
|
|
n = invalidate_filp(rfilp);
|
|
|
|
if (n != rfilp->filp_count) {
|
2008-02-22 16:41:07 +01:00
|
|
|
printf("VFS: warning: invalidate/count "
|
VFS: make all IPC asynchronous
By decoupling synchronous drivers from VFS, we are a big step closer to
supporting driver crashes under all circumstances. That is, VFS can't
become stuck on IPC with a synchronous driver (e.g., INET) and can
recover from crashing block drivers during open/close/ioctl or during
communication with an FS.
In order to maintain serialized communication with a synchronous driver,
the communication is wrapped by a mutex on a per driver basis (not major
numbers as there can be multiple majors with identical endpoints). Majors
that share a driver endpoint point to a single mutex object.
In order to support crashes from block drivers, the file reopen tactic
had to be changed; first reopen files associated with the crashed
driver, then send the new driver endpoint to FSes. This solves a
deadlock between the FS and the block driver;
- VFS would send REQ_NEW_DRIVER to an FS, but he FS only receives it
after retrying the current request to the newly started driver.
- The block driver would refuse the retried request until all files
had been reopened.
- VFS would reopen files only after getting a reply from the initial
REQ_NEW_DRIVER.
When a character special driver crashes, all associated files have to
be marked invalid and closed (or reopened if flagged as such). However,
they can only be closed if a thread holds exclusive access to it. To
obtain exclusive access, the worker thread (which handles the new driver
endpoint event from DS) schedules a new job to garbage collect invalid
files. This way, we can signal the worker thread that was talking to the
crashed driver and will release exclusive access to a file associated
with the crashed driver and prevent the garbage collecting worker thread
from dead locking on that file.
Also, when a character special driver crashes, RS will unmap the driver
and remap it upon restart. During unmapping, associated files are marked
invalid instead of waiting for an endpoint up event from DS, as that
event might come later than new read/write/select requests and thus
cause confusion in the freshly started driver.
When locking a filp, the usage counters are no longer checked. The usage
counter can legally go down to zero during filp invalidation while there
are locks pending.
DS events are handled by a separate worker thread instead of the main
thread as reopening files could lead to another crash and a stuck thread.
An additional worker thread is then necessary to unlock it.
Finally, with everything asynchronous a race condition in do_select
surfaced. A select entry was only marked in use after succesfully sending
initial select requests to drivers and having to wait. When multiple
select() calls were handled there was opportunity that these entries
were overwritten. This had as effect that some select results were
ignored (and select() remained blocking instead if returning) or do_select
tried to access filps that were not present (because thrown away by
secondary select()). This bug manifested itself with sendrecs, but was
very hard to reproduce. However, it became awfully easy to trigger with
asynsends only.
2012-08-28 16:06:51 +02:00
|
|
|
"discrepancy (%d, %d)\n", n, rfilp->filp_count);
|
2008-02-22 16:41:07 +01:00
|
|
|
}
|
2012-02-13 16:28:04 +01:00
|
|
|
rfilp->filp_count = 0;
|
VFS: make all IPC asynchronous
By decoupling synchronous drivers from VFS, we are a big step closer to
supporting driver crashes under all circumstances. That is, VFS can't
become stuck on IPC with a synchronous driver (e.g., INET) and can
recover from crashing block drivers during open/close/ioctl or during
communication with an FS.
In order to maintain serialized communication with a synchronous driver,
the communication is wrapped by a mutex on a per driver basis (not major
numbers as there can be multiple majors with identical endpoints). Majors
that share a driver endpoint point to a single mutex object.
In order to support crashes from block drivers, the file reopen tactic
had to be changed; first reopen files associated with the crashed
driver, then send the new driver endpoint to FSes. This solves a
deadlock between the FS and the block driver;
- VFS would send REQ_NEW_DRIVER to an FS, but he FS only receives it
after retrying the current request to the newly started driver.
- The block driver would refuse the retried request until all files
had been reopened.
- VFS would reopen files only after getting a reply from the initial
REQ_NEW_DRIVER.
When a character special driver crashes, all associated files have to
be marked invalid and closed (or reopened if flagged as such). However,
they can only be closed if a thread holds exclusive access to it. To
obtain exclusive access, the worker thread (which handles the new driver
endpoint event from DS) schedules a new job to garbage collect invalid
files. This way, we can signal the worker thread that was talking to the
crashed driver and will release exclusive access to a file associated
with the crashed driver and prevent the garbage collecting worker thread
from dead locking on that file.
Also, when a character special driver crashes, RS will unmap the driver
and remap it upon restart. During unmapping, associated files are marked
invalid instead of waiting for an endpoint up event from DS, as that
event might come later than new read/write/select requests and thus
cause confusion in the freshly started driver.
When locking a filp, the usage counters are no longer checked. The usage
counter can legally go down to zero during filp invalidation while there
are locks pending.
DS events are handled by a separate worker thread instead of the main
thread as reopening files could lead to another crash and a stuck thread.
An additional worker thread is then necessary to unlock it.
Finally, with everything asynchronous a race condition in do_select
surfaced. A select entry was only marked in use after succesfully sending
initial select requests to drivers and having to wait. When multiple
select() calls were handled there was opportunity that these entries
were overwritten. This had as effect that some select results were
ignored (and select() remained blocking instead if returning) or do_select
tried to access filps that were not present (because thrown away by
secondary select()). This bug manifested itself with sendrecs, but was
very hard to reproduce. However, it became awfully easy to trigger with
asynsends only.
2012-08-28 16:06:51 +02:00
|
|
|
|
|
|
|
/* We have to clean up this filp and vnode, but can't do that yet as
|
|
|
|
* it's locked by a worker thread. Start a new job to garbage collect
|
|
|
|
* invalidated filps associated with this device driver.
|
|
|
|
*/
|
|
|
|
sys_worker_start(do_filp_gc);
|
2008-02-22 16:41:07 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Nothing more to re-open. Restart suspended processes */
|
2012-02-13 16:28:04 +01:00
|
|
|
driver_e = dmap[maj].dmap_driver;
|
2008-02-22 16:41:07 +01:00
|
|
|
|
|
|
|
for (rfp = &fproc[0]; rfp < &fproc[NR_PROCS]; rfp++) {
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
if(rfp->fp_pid == PID_FREE) continue;
|
2009-09-22 23:48:26 +02:00
|
|
|
if(rfp->fp_blocked_on == FP_BLOCKED_ON_OTHER &&
|
2012-02-13 16:28:04 +01:00
|
|
|
rfp->fp_task == driver_e && (rfp->fp_flags & FP_SUSP_REOPEN)) {
|
|
|
|
rfp->fp_flags &= ~FP_SUSP_REOPEN;
|
2009-09-22 23:48:26 +02:00
|
|
|
rfp->fp_blocked_on = FP_BLOCKED_ON_NONE;
|
2008-02-22 16:41:07 +01:00
|
|
|
reply(rfp->fp_endpoint, ERESTART);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Look for processes that are suspened in an OPEN call */
|
|
|
|
for (rfp = &fproc[0]; rfp < &fproc[NR_PROCS]; rfp++) {
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
if (rfp->fp_pid == PID_FREE) continue;
|
2009-09-22 23:48:26 +02:00
|
|
|
if (rfp->fp_blocked_on == FP_BLOCKED_ON_DOPEN ||
|
2012-02-13 16:28:04 +01:00
|
|
|
!(rfp->fp_flags & FP_SUSP_REOPEN)) continue;
|
2008-02-22 16:41:07 +01:00
|
|
|
|
2012-02-13 16:28:04 +01:00
|
|
|
fd_nr = scratch(rfp).file.fd_nr;
|
|
|
|
printf("VFS: restart_reopen: process in FP_BLOCKED_ON_DOPEN fd=%d\n",
|
|
|
|
fd_nr);
|
|
|
|
rfilp = rfp->fp_filp[fd_nr];
|
2008-02-22 16:41:07 +01:00
|
|
|
|
2012-02-13 16:28:04 +01:00
|
|
|
if (!rfilp) {
|
2008-02-22 16:41:07 +01:00
|
|
|
/* Open failed, and automatic reopen was not requested */
|
2009-09-22 23:48:26 +02:00
|
|
|
rfp->fp_blocked_on = FP_BLOCKED_ON_NONE;
|
2008-02-22 16:41:07 +01:00
|
|
|
FD_CLR(fd_nr, &rfp->fp_filp_inuse);
|
|
|
|
reply(rfp->fp_endpoint, EIO);
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
2012-02-13 16:28:04 +01:00
|
|
|
vp = rfilp->filp_vno;
|
|
|
|
if (!vp) panic("VFS: restart_reopen: no vp");
|
2012-04-25 14:44:42 +02:00
|
|
|
if (!S_ISCHR(vp->v_mode)) continue;
|
2012-02-13 16:28:04 +01:00
|
|
|
if (major(vp->v_sdev) != maj) continue;
|
2008-02-22 16:41:07 +01:00
|
|
|
|
2009-09-22 23:48:26 +02:00
|
|
|
rfp->fp_blocked_on = FP_BLOCKED_ON_NONE;
|
2008-02-22 16:41:07 +01:00
|
|
|
reply(rfp->fp_endpoint, fd_nr);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
|
2008-02-22 16:41:07 +01:00
|
|
|
/*===========================================================================*
|
|
|
|
* reopen_reply *
|
|
|
|
*===========================================================================*/
|
2012-03-25 20:25:53 +02:00
|
|
|
void reopen_reply()
|
2008-02-22 16:41:07 +01:00
|
|
|
{
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
endpoint_t driver_e;
|
|
|
|
int filp_no, status, maj;
|
2012-02-13 16:28:04 +01:00
|
|
|
struct filp *rfilp;
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
struct vnode *vp;
|
|
|
|
struct dmap *dp;
|
2008-02-22 16:41:07 +01:00
|
|
|
|
2012-04-13 14:50:38 +02:00
|
|
|
driver_e = job_m_in.m_source;
|
|
|
|
filp_no = job_m_in.REP_ENDPT;
|
|
|
|
status = job_m_in.REP_STATUS;
|
2008-02-22 16:41:07 +01:00
|
|
|
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
if (filp_no < 0 || filp_no >= NR_FILPS) {
|
2012-02-13 16:28:04 +01:00
|
|
|
printf("VFS: reopen_reply: bad filp number %d from driver %d\n",
|
|
|
|
filp_no, driver_e);
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
return;
|
|
|
|
}
|
2008-02-22 16:41:07 +01:00
|
|
|
|
2012-02-13 16:28:04 +01:00
|
|
|
rfilp = &filp[filp_no];
|
|
|
|
if (rfilp->filp_count < 1) {
|
|
|
|
printf("VFS: reopen_reply: filp number %d not inuse (from driver %d)\n",
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
filp_no, driver_e);
|
|
|
|
return;
|
|
|
|
}
|
2008-02-22 16:41:07 +01:00
|
|
|
|
2012-02-13 16:28:04 +01:00
|
|
|
vp = rfilp->filp_vno;
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
if (!vp) {
|
2012-02-13 16:28:04 +01:00
|
|
|
printf("VFS: reopen_reply: no vnode for filp number %d (from driver "
|
|
|
|
"%d)\n", filp_no, driver_e);
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
return;
|
|
|
|
}
|
2008-02-22 16:41:07 +01:00
|
|
|
|
VFS: make all IPC asynchronous
By decoupling synchronous drivers from VFS, we are a big step closer to
supporting driver crashes under all circumstances. That is, VFS can't
become stuck on IPC with a synchronous driver (e.g., INET) and can
recover from crashing block drivers during open/close/ioctl or during
communication with an FS.
In order to maintain serialized communication with a synchronous driver,
the communication is wrapped by a mutex on a per driver basis (not major
numbers as there can be multiple majors with identical endpoints). Majors
that share a driver endpoint point to a single mutex object.
In order to support crashes from block drivers, the file reopen tactic
had to be changed; first reopen files associated with the crashed
driver, then send the new driver endpoint to FSes. This solves a
deadlock between the FS and the block driver;
- VFS would send REQ_NEW_DRIVER to an FS, but he FS only receives it
after retrying the current request to the newly started driver.
- The block driver would refuse the retried request until all files
had been reopened.
- VFS would reopen files only after getting a reply from the initial
REQ_NEW_DRIVER.
When a character special driver crashes, all associated files have to
be marked invalid and closed (or reopened if flagged as such). However,
they can only be closed if a thread holds exclusive access to it. To
obtain exclusive access, the worker thread (which handles the new driver
endpoint event from DS) schedules a new job to garbage collect invalid
files. This way, we can signal the worker thread that was talking to the
crashed driver and will release exclusive access to a file associated
with the crashed driver and prevent the garbage collecting worker thread
from dead locking on that file.
Also, when a character special driver crashes, RS will unmap the driver
and remap it upon restart. During unmapping, associated files are marked
invalid instead of waiting for an endpoint up event from DS, as that
event might come later than new read/write/select requests and thus
cause confusion in the freshly started driver.
When locking a filp, the usage counters are no longer checked. The usage
counter can legally go down to zero during filp invalidation while there
are locks pending.
DS events are handled by a separate worker thread instead of the main
thread as reopening files could lead to another crash and a stuck thread.
An additional worker thread is then necessary to unlock it.
Finally, with everything asynchronous a race condition in do_select
surfaced. A select entry was only marked in use after succesfully sending
initial select requests to drivers and having to wait. When multiple
select() calls were handled there was opportunity that these entries
were overwritten. This had as effect that some select results were
ignored (and select() remained blocking instead if returning) or do_select
tried to access filps that were not present (because thrown away by
secondary select()). This bug manifested itself with sendrecs, but was
very hard to reproduce. However, it became awfully easy to trigger with
asynsends only.
2012-08-28 16:06:51 +02:00
|
|
|
if (!(rfilp->filp_state & FS_NEEDS_REOPEN)) {
|
2012-02-13 16:28:04 +01:00
|
|
|
printf("VFS: reopen_reply: bad state %d for filp number %d"
|
|
|
|
" (from driver %d)\n", rfilp->filp_state, filp_no, driver_e);
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
return;
|
|
|
|
}
|
2008-02-22 16:41:07 +01:00
|
|
|
|
2012-04-25 14:44:42 +02:00
|
|
|
if (!S_ISCHR(vp->v_mode)) {
|
2012-02-13 16:28:04 +01:00
|
|
|
printf("VFS: reopen_reply: bad mode 0%o for filp number %d"
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
" (from driver %d)\n", vp->v_mode, filp_no, driver_e);
|
|
|
|
return;
|
|
|
|
}
|
2008-02-22 16:41:07 +01:00
|
|
|
|
2012-02-13 16:28:04 +01:00
|
|
|
maj = major(vp->v_sdev);
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
dp = &dmap[maj];
|
|
|
|
if (dp->dmap_driver != driver_e) {
|
2012-02-13 16:28:04 +01:00
|
|
|
printf("VFS: reopen_reply: bad major %d for filp number %d "
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
"(from driver %d, current driver is %d)\n", maj, filp_no,
|
|
|
|
driver_e, dp->dmap_driver);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (status == OK) {
|
VFS: make all IPC asynchronous
By decoupling synchronous drivers from VFS, we are a big step closer to
supporting driver crashes under all circumstances. That is, VFS can't
become stuck on IPC with a synchronous driver (e.g., INET) and can
recover from crashing block drivers during open/close/ioctl or during
communication with an FS.
In order to maintain serialized communication with a synchronous driver,
the communication is wrapped by a mutex on a per driver basis (not major
numbers as there can be multiple majors with identical endpoints). Majors
that share a driver endpoint point to a single mutex object.
In order to support crashes from block drivers, the file reopen tactic
had to be changed; first reopen files associated with the crashed
driver, then send the new driver endpoint to FSes. This solves a
deadlock between the FS and the block driver;
- VFS would send REQ_NEW_DRIVER to an FS, but he FS only receives it
after retrying the current request to the newly started driver.
- The block driver would refuse the retried request until all files
had been reopened.
- VFS would reopen files only after getting a reply from the initial
REQ_NEW_DRIVER.
When a character special driver crashes, all associated files have to
be marked invalid and closed (or reopened if flagged as such). However,
they can only be closed if a thread holds exclusive access to it. To
obtain exclusive access, the worker thread (which handles the new driver
endpoint event from DS) schedules a new job to garbage collect invalid
files. This way, we can signal the worker thread that was talking to the
crashed driver and will release exclusive access to a file associated
with the crashed driver and prevent the garbage collecting worker thread
from dead locking on that file.
Also, when a character special driver crashes, RS will unmap the driver
and remap it upon restart. During unmapping, associated files are marked
invalid instead of waiting for an endpoint up event from DS, as that
event might come later than new read/write/select requests and thus
cause confusion in the freshly started driver.
When locking a filp, the usage counters are no longer checked. The usage
counter can legally go down to zero during filp invalidation while there
are locks pending.
DS events are handled by a separate worker thread instead of the main
thread as reopening files could lead to another crash and a stuck thread.
An additional worker thread is then necessary to unlock it.
Finally, with everything asynchronous a race condition in do_select
surfaced. A select entry was only marked in use after succesfully sending
initial select requests to drivers and having to wait. When multiple
select() calls were handled there was opportunity that these entries
were overwritten. This had as effect that some select results were
ignored (and select() remained blocking instead if returning) or do_select
tried to access filps that were not present (because thrown away by
secondary select()). This bug manifested itself with sendrecs, but was
very hard to reproduce. However, it became awfully easy to trigger with
asynsends only.
2012-08-28 16:06:51 +02:00
|
|
|
rfilp->filp_state &= ~FS_NEEDS_REOPEN;
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
} else {
|
2012-02-13 16:28:04 +01:00
|
|
|
printf("VFS: reopen_reply: should handle error status\n");
|
- Introduce support for sticky bit.
- Revise VFS-FS protocol and update VFS/MFS/ISOFS accordingly.
- Clean up MFS by removing old, dead code (backwards compatibility is broken by
the new VFS-FS protocol, anyway) and rewrite other parts. Also, make sure all
functions have proper banners and prototypes.
- VFS should always provide a (syntactically) valid path to the FS; no need for
the FS to do sanity checks when leaving/entering mount points.
- Fix several bugs in MFS:
- Several path lookup bugs in MFS.
- A link can be too big for the path buffer.
- A mountpoint can become inaccessible when the creation of a new inode
fails, because the inode already exists and is a mountpoint.
- Introduce support for supplemental groups.
- Add test 46 to test supplemental group functionality (and removed obsolete
suppl. tests from test 2).
- Clean up VFS (not everything is done yet).
- ISOFS now opens device read-only. This makes the -r flag in the mount command
unnecessary (but will still report to be mounted read-write).
- Introduce PipeFS. PipeFS is a new FS that handles all anonymous and
named pipes. However, named pipes still reside on the (M)FS, as they are part
of the file system on disk. To make this work VFS now has a concept of
'mapped' inodes, which causes read, write, truncate and stat requests to be
redirected to the mapped FS, and all other requests to the original FS.
2009-12-20 21:27:14 +01:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
restart_reopen(maj);
|
2008-02-22 16:41:07 +01:00
|
|
|
}
|