minix/servers/vfs/dmap.c

376 lines
11 KiB
C
Raw Normal View History

2005-04-21 16:53:53 +02:00
/* This file contains the table with device <-> driver mappings. It also
* contains some routines to dynamically add and/ or remove device drivers
2012-02-13 16:28:04 +01:00
* or change mappings.
2005-04-21 16:53:53 +02:00
*/
#include "fs.h"
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
#include <assert.h>
2005-04-21 16:53:53 +02:00
#include <string.h>
#include <stdlib.h>
#include <ctype.h>
#include <unistd.h>
2005-04-21 16:53:53 +02:00
#include <minix/com.h>
2007-08-07 14:52:47 +02:00
#include <minix/ds.h>
2012-02-13 16:28:04 +01:00
#include "fproc.h"
#include "dmap.h"
#include "param.h"
2005-04-21 16:53:53 +02:00
/* The order of the entries in the table determines the mapping between major
* device numbers and device drivers. Character and block devices
* can be intermixed at random. The ordering determines the device numbers in
2012-02-13 16:28:04 +01:00
* /dev. Note that the major device numbers used in /dev are NOT the same as
* the process numbers of the device drivers. See <minix/dmap.h> for mappings.
2005-04-21 16:53:53 +02:00
*/
struct dmap dmap[NR_DEVICES];
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
#define DT_EMPTY { no_dev, no_dev_io, NONE, "", 0, STYLE_NDEV, NULL, NONE, \
0, NULL, 0}
/*===========================================================================*
* lock_dmap *
*===========================================================================*/
void lock_dmap(struct dmap *dp)
{
/* Lock a driver */
struct worker_thread *org_self;
struct fproc *org_fp;
int r;
assert(dp != NULL);
assert(dp->dmap_driver != NONE);
org_fp = fp;
org_self = self;
if ((r = mutex_lock(dp->dmap_lock_ref)) != 0)
panic("unable to get a lock on dmap: %d\n", r);
fp = org_fp;
self = org_self;
}
/*===========================================================================*
* unlock_dmap *
*===========================================================================*/
void unlock_dmap(struct dmap *dp)
{
/* Unlock a driver */
int r;
assert(dp != NULL);
if ((r = mutex_unlock(dp->dmap_lock_ref)) != 0)
panic("unable to unlock dmap lock: %d\n", r);
}
2005-04-21 16:53:53 +02:00
2007-08-07 14:52:47 +02:00
/*===========================================================================*
* do_mapdriver *
*===========================================================================*/
2012-03-25 20:25:53 +02:00
int do_mapdriver()
2007-08-07 14:52:47 +02:00
{
2012-02-13 16:28:04 +01:00
/* Create a device->driver mapping. RS will tell us which major is driven by
* this driver, what type of device it is (regular, TTY, asynchronous, clone,
* etc), and its label. This label is registered with DS, and allows us to
* retrieve the driver's endpoint.
*/
int r, flags, major, style;
2012-02-13 16:28:04 +01:00
endpoint_t endpoint;
vir_bytes label_vir;
size_t label_len;
char label[LABEL_MAX];
2007-08-07 14:52:47 +02:00
2012-02-13 16:28:04 +01:00
/* Only RS can map drivers. */
if (who_e != RS_PROC_NR) return(EPERM);
2007-08-07 14:52:47 +02:00
label_vir = (vir_bytes) job_m_in.md_label;
label_len = (size_t) job_m_in.md_label_len;
major = job_m_in.md_major;
flags = job_m_in.md_flags;
style = job_m_in.md_style;
2007-08-07 14:52:47 +02:00
/* Get the label */
2012-02-13 16:28:04 +01:00
if (label_len+1 > sizeof(label)) { /* Can we store this label? */
printf("VFS: do_mapdriver: label too long\n");
return(EINVAL);
}
r = sys_vircopy(who_e, label_vir, SELF, (vir_bytes) label, label_len);
2012-02-13 16:28:04 +01:00
if (r != OK) {
printf("VFS: do_mapdriver: sys_vircopy failed: %d\n", r);
return(EINVAL);
}
label[label_len] = '\0'; /* Terminate label */
2007-08-07 14:52:47 +02:00
2012-02-13 16:28:04 +01:00
/* Now we know how the driver is called, fetch its endpoint */
r = ds_retrieve_label_endpt(label, &endpoint);
if (r != OK) {
printf("VFS: do_mapdriver: label '%s' unknown\n", label);
return(EINVAL);
}
2007-08-07 14:52:47 +02:00
2012-02-13 16:28:04 +01:00
/* Try to update device mapping. */
return map_driver(label, major, endpoint, style, flags);
2007-08-07 14:52:47 +02:00
}
2005-04-21 16:53:53 +02:00
/*===========================================================================*
* map_driver *
*===========================================================================*/
2012-03-25 20:25:53 +02:00
int map_driver(label, major, proc_nr_e, style, flags)
2012-07-13 18:08:06 +02:00
const char label[LABEL_MAX]; /* name of the driver */
2007-08-07 14:52:47 +02:00
int major; /* major number of the device */
endpoint_t proc_nr_e; /* process number of the driver */
int style; /* style of the device */
int flags; /* device flags */
2007-08-07 14:52:47 +02:00
{
2012-02-13 16:28:04 +01:00
/* Add a new device driver mapping in the dmap table. If the proc_nr is set to
* NONE, we're supposed to unmap it.
2007-08-07 14:52:47 +02:00
*/
2012-02-13 16:28:04 +01:00
int slot, s;
2007-08-07 14:52:47 +02:00
size_t len;
struct dmap *dp;
struct fproc *rfp;
2007-08-07 14:52:47 +02:00
/* Get pointer to device entry in the dmap table. */
if (major < 0 || major >= NR_DEVICES) return(ENODEV);
2012-02-13 16:28:04 +01:00
dp = &dmap[major];
2007-08-07 14:52:47 +02:00
/* Check if we're supposed to unmap it. */
2012-02-13 16:28:04 +01:00
if (proc_nr_e == NONE) {
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
/* Even when a driver is now unmapped and is shortly to be mapped in
* due to recovery, invalidate associated filps if they're character
* special files. More sophisticated recovery mechanisms which would
* reduce the need to invalidate files are possible, but would require
* cooperation of the driver and more recovery framework between RS,
* VFS, and DS.
*/
invalidate_filp_by_char_major(major);
2007-08-07 14:52:47 +02:00
dp->dmap_opcl = no_dev;
dp->dmap_io = no_dev_io;
dp->dmap_driver = NONE;
dp->dmap_flags = flags;
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
dp->dmap_lock_ref = &dp->dmap_lock;
2007-08-07 14:52:47 +02:00
return(OK);
}
2012-02-13 16:28:04 +01:00
/* Check process number of new driver if it was alive before mapping */
s = isokendpt(proc_nr_e, &slot);
if (s != OK) {
/* This is not a problem only when we force this driver mapping */
if (! (flags & DRV_FORCED))
2007-08-07 14:52:47 +02:00
return(EINVAL);
} else {
2012-02-13 16:28:04 +01:00
rfp = &fproc[slot];
rfp->fp_flags |= FP_SYS_PROC; /* Process is a driver */
2007-08-07 14:52:47 +02:00
}
if (label != NULL) {
2012-02-13 16:28:04 +01:00
len = strlen(label);
if (len+1 > sizeof(dp->dmap_label))
2012-02-13 16:28:04 +01:00
panic("VFS: map_driver: label too long: %d", len);
2012-07-13 18:08:06 +02:00
strlcpy(dp->dmap_label, label, LABEL_MAX);
}
2007-08-07 14:52:47 +02:00
2012-02-13 16:28:04 +01:00
/* Store driver I/O routines based on type of device */
2007-08-07 14:52:47 +02:00
switch (style) {
2012-02-13 16:28:04 +01:00
case STYLE_DEV:
dp->dmap_opcl = gen_opcl;
dp->dmap_io = gen_io;
break;
2012-02-13 16:28:04 +01:00
case STYLE_DEVA:
dp->dmap_opcl = gen_opcl;
dp->dmap_io = asyn_io;
break;
2012-02-13 16:28:04 +01:00
case STYLE_TTY:
dp->dmap_opcl = tty_opcl;
dp->dmap_io = gen_io;
break;
2012-02-13 16:28:04 +01:00
case STYLE_CTTY:
dp->dmap_opcl = ctty_opcl;
dp->dmap_io = ctty_io;
break;
2012-02-13 16:28:04 +01:00
case STYLE_CLONE:
dp->dmap_opcl = clone_opcl;
dp->dmap_io = gen_io;
break;
2012-02-13 16:28:04 +01:00
case STYLE_CLONE_A:
dp->dmap_opcl = clone_opcl;
dp->dmap_io = asyn_io;
break;
default:
return(EINVAL);
2007-08-07 14:52:47 +02:00
}
2012-02-13 16:28:04 +01:00
2007-08-07 14:52:47 +02:00
dp->dmap_driver = proc_nr_e;
dp->dmap_flags = flags;
dp->dmap_style = style;
2012-02-13 16:28:04 +01:00
return(OK);
2007-08-07 14:52:47 +02:00
}
/*===========================================================================*
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
* dmap_unmap_by_endpt *
*===========================================================================*/
2012-03-25 20:25:53 +02:00
void dmap_unmap_by_endpt(endpoint_t proc_e)
{
2012-02-13 16:28:04 +01:00
/* Lookup driver in dmap table by endpoint and unmap it */
int major, r;
for (major = 0; major < NR_DEVICES; major++) {
if (dmap_driver_match(proc_e, major)) {
/* Found driver; overwrite it with a NULL entry */
if ((r = map_driver(NULL, major, NONE, 0, 0)) != OK) {
printf("VFS: unmapping driver %d for major %d failed:"
" %d\n", proc_e, major, r);
}
}
}
}
/*===========================================================================*
* map_service *
*===========================================================================*/
2012-03-25 20:25:53 +02:00
int map_service(struct rprocpub *rpub)
{
/* Map a new service by storing its device driver properties. */
int 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
struct dmap *fdp, *sdp;
/* Not a driver, nothing more to do. */
2012-02-13 16:28:04 +01:00
if(rpub->dev_nr == NO_DEV) return(OK);
/* Map driver. */
2012-02-13 16:28:04 +01:00
r = map_driver(rpub->label, rpub->dev_nr, rpub->endpoint, rpub->dev_style,
rpub->dev_flags);
if(r != OK) return(r);
/* If driver has two major numbers associated, also map the other one. */
if(rpub->dev_style2 != STYLE_NDEV) {
r = map_driver(rpub->label, rpub->dev_nr+1, rpub->endpoint,
2012-02-13 16:28:04 +01:00
rpub->dev_style2, rpub->dev_flags);
if(r != OK) return(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
/* To ensure that future dmap lock attempts always lock the same driver
* regardless of major number, refer the second dmap lock reference
* to the first dmap entry.
*/
fdp = get_dmap_by_major(rpub->dev_nr);
sdp = get_dmap_by_major(rpub->dev_nr+1);
assert(fdp != NULL);
assert(sdp != NULL);
assert(fdp != sdp);
sdp->dmap_lock_ref = &fdp->dmap_lock;
}
2012-02-13 16:28:04 +01:00
return(OK);
}
2005-04-21 16:53:53 +02:00
/*===========================================================================*
2012-02-13 16:28:04 +01:00
* init_dmap *
2005-04-21 16:53:53 +02:00
*===========================================================================*/
2012-03-25 20:25:53 +02:00
void init_dmap()
2005-04-21 16:53:53 +02:00
{
/* Initialize the table with empty device <-> driver mappings. */
int i;
struct dmap dmap_default = DT_EMPTY;
2012-02-13 16:28:04 +01:00
for (i = 0; i < NR_DEVICES; i++)
dmap[i] = dmap_default;
2005-04-21 16:53:53 +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
/*===========================================================================*
* init_dmap_locks *
*===========================================================================*/
void init_dmap_locks()
{
int i;
for (i = 0; i < NR_DEVICES; i++) {
if (mutex_init(&dmap[i].dmap_lock, NULL) != 0)
panic("unable to initialize dmap lock");
dmap[i].dmap_lock_ref = &dmap[i].dmap_lock;
}
}
/*===========================================================================*
* dmap_driver_match *
2012-02-13 16:28:04 +01:00
*===========================================================================*/
2012-03-25 20:25:53 +02:00
int dmap_driver_match(endpoint_t proc, int major)
{
2012-02-13 16:28:04 +01:00
if (major < 0 || major >= NR_DEVICES) return(0);
if (dmap[major].dmap_driver != NONE && dmap[major].dmap_driver == proc)
return(1);
return(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
/*===========================================================================*
* dmap_by_major *
*===========================================================================*/
struct dmap *
get_dmap_by_major(int major)
{
if (major < 0 || major >= NR_DEVICES) return(NULL);
if (dmap[major].dmap_driver == NONE) return(NULL);
return(&dmap[major]);
}
/*===========================================================================*
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
* dmap_endpt_up *
2012-02-13 16:28:04 +01:00
*===========================================================================*/
2012-03-25 20:25:53 +02:00
void dmap_endpt_up(endpoint_t proc_e, int is_blk)
{
2012-02-13 16:28:04 +01:00
/* A device driver with endpoint proc_e has been restarted. Go tell everyone
* that might be blocking on it that this device is 'up'.
*/
int 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
struct dmap *dp;
struct worker_thread *worker;
if (proc_e == NONE) return;
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 ((dp = get_dmap_by_major(major)) == NULL) continue;
if (dp->dmap_driver == proc_e) {
if (is_blk) {
if (dp->dmap_recovering) {
printf("VFS: driver recovery failure for"
" major %d\n", major);
if (dp->dmap_servicing != NONE) {
worker = worker_get(dp->dmap_servicing);
worker_stop(worker);
}
dp->dmap_recovering = 0;
continue;
}
dp->dmap_recovering = 1;
2012-02-13 16:28:04 +01:00
bdev_up(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
dp->dmap_recovering = 0;
} else {
if (dp->dmap_servicing != NONE) {
worker = worker_get(dp->dmap_servicing);
worker_stop(worker);
}
2012-02-13 16:28:04 +01:00
cdev_up(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
}
}
2012-02-13 16:28:04 +01:00
}
}
/*===========================================================================*
* get_dmap *
2012-02-13 16:28:04 +01:00
*===========================================================================*/
2012-03-25 20:25:53 +02:00
struct dmap *get_dmap(endpoint_t proc_e)
{
/* See if 'proc_e' endpoint belongs to a valid dmap entry. If so, return a
* pointer */
int major;
for (major = 0; major < NR_DEVICES; major++)
2012-02-13 16:28:04 +01:00
if (dmap_driver_match(proc_e, major))
return(&dmap[major]);
return(NULL);
}