* Renamed struct timer to struct minix_timer
* Renamed timer_t to minix_timer_t
* Ensured all the code uses the minix_timer_t typedef
* Removed ifdef around _BSD_TIMER_T
* Removed include/timers.h and merged it into include/minix/timers.h
* Resolved prototype conflict by renaming kernel's (re)set_timer
to (re)set_kernel_timer.
Change-Id: I56f0f30dfed96e1a0575d92492294cf9a06468a5
This call copies a file descriptor from a remote process into the
calling process. The call is for the VND driver only, and in the
future, ACLs will prevent any other process from using this call.
Change-Id: Ib16fdd1f1a12cb38a70d7e441dad91bc86898f6d
Previously, VFS would reopen a character device after a driver crash
if the associated file descriptor was opened with the O_REOPEN flag.
This patch removes support for this feature. The code was complex,
full of uncovered corner cases, and hard to test. Moreover, it did not
actually hide the crash from user applications: they would get an
error code to indicate that something went wrong, and have to decide
based on the nature of the underlying device how to continue.
- remove support for O_REOPEN, and make playwave(1) reopen its device;
- remove support for the DEV_REOPEN protocol message;
- remove all code in VFS related to reopening character devices;
- no longer change VFS filp reference count and FD bitmap upon filp
invalidation; instead, make get_filp* fail all calls on invalidated
FDs except when obtained with the locktype VNODE_OPCL which is used
by close_fd only;
- remove the VFS fproc file descriptor bitmap entirely, returning to
the situation that a FD is in use if its slot points to a filp; use
FILP_CLOSED as single means of marking a filp as invalidated.
Change-Id: I34f6bc69a036b3a8fc667c1f80435ff3af56558f
- block the calling thread on character device close;
- fully separate block and character open/close routines;
- reuse generic open/close code for the cloning case;
- zero all messages to drivers before filling them;
- use appropriate types for major/minor device numbers.
Change-Id: Ia90e6fe5688f212f835c5ee1bfca831cb249cf51
- prefix them with VFS_ as they are going to VFS;
- give these calls normal call numbers;
- give them their own set of message field aliases;
- also make do_mapdriver a regular call.
Change-Id: I2140439f288b06d699a1f65438bd8306509b259e
- check each file descriptor's open access mode (filp_mode);
- treat an error returned by a character driver as a select error;
- check all filps in each set before finishing select;
- do not copy back file descriptor sets if an error occurred;
- remove the hardcoded list of supported character major devices,
since all drivers should now be capable of responding properly;
- add tests to test40 and fix its error count aggregation.
Change-Id: I57ef58d3afb82640fc50b59c859ee4b25f02db17
I/O control requests now come with the endpoint of the user process
that initiated the ioctl(2) call. It is stored in a new BDEV_USER
field, which is an alias for BDEV_FLAGS. The contents of this field
are to be used only in highly specific situations. It should be
preserved (not replaced!) by services that forward IOCTL requests,
and may be set to NONE for service-initiated IOCTL requests.
Change-Id: I68a01b9ce43eca00e61b985a9cf87f55ba683de4
The original R_BIT and W_BIT definitions have nothing to do with the
way these bits are used. Their distinct usage is more apparent when
they have different names.
Change-Id: Ia984457f900078b2e3502ceed565fead4e5bb965
Previously, reading from or writing to a character device would not
update the file position on the corresponding filp object. Performing
this update correctly is not trivial: during and after the I/O
operation, the filp object must not be locked. Ideally, read/write
requests on a filp that is already involved in a read/write operation,
should be queued. For now, we optimistically update the file position
at the start of the I/O; this works under the assumptions listed in
the corresponding comment.
Change-Id: I172a61781850423709924390ae3df1f2d1f94707
POSIX states that when interrupted, partially successful pipe
operations should return the partial result rather than EINTR. VFS
previously wouldn't look at the partial result, and not clear it
either, which would result in a panic upon the next pipe operation.
Change-Id: Ia1eb72b4b77394051444e63a1390d49bb315eb04
The main purpose of this patch is to fix handling of unpause calls
from PM while another call is ongoing. The solution to this problem
sparked a full revision of the threading model, consisting of a large
number of related changes:
- all active worker threads are now always associated with a process,
and every process has at most one active thread working for it;
- the process lock is always held by a process's worker thread;
- a process can now have both normal work and postponed PM work
associated to it;
- timer expiry and non-postponed PM work is done from the main thread;
- filp garbage collection is done from a thread associated with VFS;
- reboot calls from PM are now done from a thread associated with PM;
- the DS events handler is protected from starting multiple threads;
- support for a system worker thread has been removed;
- the deadlock recovery thread has been replaced by a parameter to the
worker_start() function; the number of worker threads has
consequently been increased by one;
- saving and restoring of global but per-thread variables is now
centralized in worker_suspend() and worker_resume(); err_code is now
saved and restored in all cases;
- the concept of jobs has been removed, and job_m_in now points to a
message stored in the worker thread structure instead;
- the PM lock has been removed;
- the separate exec lock has been replaced by a lock on the VM
process, which was already being locked for exec calls anyway;
- PM_UNPAUSE is now processed as a postponed PM request, from a thread
associated with the target process;
- the FP_DROP_WORK flag has been removed, since it is no longer more
than just an optimization and only applied to processes operating on
a pipe when getting killed;
- assignment to "fp" now takes place only when obtaining new work in
the main thread or a worker thread, when resuming execution of a
thread, and in the special case of exiting processes during reboot;
- there are no longer special cases where the yield() call is used to
force a thread to run.
Change-Id: I7a97b9b95c2450454a9b5318dfa0e6150d4e6858
The T_DUMPCORE implementation was not only broken - it would currently
produce a coredump of the tracer process rather than the traced
process - but also deeply flawed, and fixing it would require serious
alteration of PM's internal state machine. It should be possible to
implement the same functionality in userland, and that is now the
suggested way forward. For now, also remove the (identical) utilities
using T_DUMPCORE: dumpcore(1) and gcore(1).
Change-Id: I1d51be19c739362b8a5833de949b76382a1edbcc
Previously, processing of some replies coming from character drivers
could block on locks, and therefore, such processing was done from
threads that were associated to the character driver process. The
hidden consequence of this was that if all threads were in use, VFS
could drop replies coming from the driver. This patch returns VFS to
a situation where the replies from character drivers are processed
instantly from the main thread, by removing the situations that may
cause VFS to block while handling those replies.
- change the locking model for select, so that it will never block
on any processing that happens after the select call has been set
up, in particular processing of character driver select replies;
- clearly mark all select routines that may never block;
- protect against race conditions in do_select as result of the
locking that still does happen there (as is required for pipes);
- also handle select timers from the main thread;
- move processing of character driver replies into device.c.
Change-Id: I4dc8e69f265cbd178de0fbf321d35f58f067cc57
These days, DEV_OPEN calls to character drivers block the calling
thread until completion or failure, and thus never return SUSPEND to
the caller. The same already applied to BDEV_OPEN calls to block
drivers. It has thus become impossible for a process to enter a state
of being blocked on a device open call.
There is currently no support for restarting device open calls to
restarted character drivers. This support was present in the _DOPEN
logic, but was already no longer triggering. In the future, this case
should be handled by the thread performing the open request.
Change-Id: I6cc1e7b4c9ed116c6ce160b315e6e060124dce00
- change all sync char drivers into async drivers;
- retire support for the sync protocol in libchardev;
- remove async dev style, as this is now the default;
- remove dev_status from VFS;
- clean up now-unused protocol messages.
Change-Id: I6aacff712292f6b29f2ccd51bc1e7d7003723e87
The async char protocol already has this, so this patch closes the
gap between the two protocols a bit. Support for this flag has been
added to all sync char drivers that support CANCEL at all.
The LOG driver was already using the asynchronous protocol, but it
did not support the nonblocking transfer flag. This has been fixed
as well.
Change-Id: Ia55432c9f102765b59ad3feb45a8bd47a782c93f
As with w_task, this ensures that the field remains cleared if it is
not used. Without this, worker_stop could mistakenly identify a thread
as talking to a device driver rather than a (crashed) file server.
Change-Id: I7d3ebed3efc3cd4f5c891f61c67a6463109b6376
It was always set, but not always cleared, when talking to asynchronous
drivers. This could cause erratic behavior upon a driver crash.
Normally, a worker thread's w_task field is set when it's about to
communicate with a driver or FS. Then upon receiving a reply we can
do sanity checks (that the thread we want to wake up was actually
waiting for a reply). Also, when a driver/FS crashes, we can identify
which worker threads were talking to the crashed endpoint and handle
the error gracefully.
Asynchronous drivers are a bit special, though. In most cases, the
sender of the request is not interested in the reply (the sender was
suspended and only wants to know whether the request was successfully
caried out or not). However, the open request is special, as the reply
carries information needed by the sender. This is the only request
where a worker thread actually yields and waits for the result. This is
also the only case where we're interested in setting w_task for
asynchronous drivers.
Change-Id: Ia1ce2747937df376122b5e13b6a069de27fcc379
* Removed startup code patches in lib/csu regarding kernel to userland
ABI.
* Aligned stack layout on NetBSD stack layout.
* Generate valid stack pointers instead of offsets by taking into account
_minix_kerninfo->kinfo->user_sp.
* Refactored stack generation, by moving part of execve in two
functions {minix_stack_params(), minix_stack_fill()} and using them
in execve(), rs and vm.
* Changed load offset of rtld (ld.so) to:
execi.args.stack_high - execi.args.stack_size - 0xa00000
which is 10MB below the main executable stack.
Change-Id: I839daf3de43321cded44105634102d419cb36cec
The main motivation for this change is that only Loris supports
multithreading, and Loris supports dynamic thread allocation, so the
number of supported threads can be implemented as a bit flag (i.e.,
either 1 or "at least as many as VFS has"). The ABI break obviates the
need to support file system versioning at this time, and several
other aspects are better implemented as flags as well. Other changes:
- replace peek/bpeek test upon mount with FS flag as well;
- mark libsffs as 64-bit file size capable;
- remove old (3.2.1) getdents support.
Change-Id: I313eace9c50ed816656c31cd47d969033d952a03
- pass in file system type through mount(2), and return this type in
statvfs structures as generated by [f]statvfs(2);
- align mount flags field with NetBSD's, splitting out service flags
which are not to be passed to VFS;
- remove limitation of mount ABI to 16-byte labels, so that labels
can be made larger in the future;
- introduce new m11 message union type for mount(2) as side effect.
Change-Id: I88b7710e297e00a5e4582ada5243d3d5c2801fd9
This is a requirement for implementing calls such as getmntinfo(3).
VFS is now responsible for filling in some of the structure's fields.
Change-Id: I0c1fa78019587efefd2949b3be38cd9a7ddc2ced
The memory-mapped files implementation (mmap() etc.) is implemented with
the help of the filesystems using the in-VM FS cache. Filesystems tell it
about all cached blocks and their metadata. Metadata is: device offset and,
if any (and known), inode number and in-inode offset. VM can then map in
requested memory-mapped file blocks, and request them if necessary.
A limitation of this system is that filesystem block sizes that are not
a multiple of the VM system (and VM hardware) page size are not possible;
we can't map blocks in partially. (We can copy, but then the benefits of
mapping and sharing the physical pages is gone.) So until before this
commit various pieces of caching code assumed page size multiple
blocksizes. This isn't strictly necessary as long as mmap() needn't be
supported on that FS.
This change allows the in-FS cache code (libminixfs) to allocate any-sized
blocks, and will not interact with the VM cache for non-pagesize-multiple
blocks. In that case it will also signal requestors, by failing 'peek'
requests, that mmap() should not be supported on this FS. VM and VFS
will then gracefully fail all file-mapping mmap() calls, and exec() will
fall back to copying executable blocks instead of mmap()ping executables.
As a result, 3 diagnostics that signal file-mapped mmap()s failing
(hitherto an unusual occurence) are disabled, as ld.so does file-mapped
mmap()s to map in objects it needs. On FSes not supporting it this situation
is legitimate and shouldn't cause so much noise. ld.so will revert to its own
minix-specific allocate+copy style of starting executables if mmap()s fail.
Change-Id: Iecb1c8090f5e0be28da8f5181bb35084eb18f67b
The VM server now manages its call masks such that all user processes
share the same call mask. As a result, an update for the call mask of
any user process will apply to all user processes. This is similar to
the privilege infrastructure employed by the kernel, and may serve as
a template for similar fine-grained restrictions in other servers.
Concretely, this patch fixes the problem of "service edit init" not
applying the given VM call mask to user processes started from RC
scripts during system startup.
In addition, this patch makes RS set a proper VM call mask for each
recovery script it spawns.
Change-Id: I520a30d85a0d3f3502d2b158293a2258825358cf
-By adding MKGCC=yes and MKGCCCMDS=yes on the make commandline
it is now possible to compile and install GCC on the system.
Before doing this, if you are not using the build.sh script,
you will need to call the fetch scripts in order to retrieve
the sources of GCC and its dependencies.
-Reduce difference with NetBSD share/mk
Move Minix-specific parameters from bsd.gcc.mk to bsd.own.mk,
which is anyway patched, so that bsd.gcc.mk is now aligned
on the NetBSD version.
-Clean libraries dependencies, compiles stdc++ only if gcc is
also compiled (it is part of the gcc sources)
-Correct minix.h header sequence, cleanup spec headers.
-Fix cross-compilation from a 32bit host targeting MINIX/arm
Change-Id: I1b234af18eed4ab5675188244e931b2a2b7bd943
Implement getrusage.
These fields of struct rusage are not supported and always set to zero at this time
long ru_nswap; /* swaps */
long ru_inblock; /* block input operations */
long ru_oublock; /* block output operations */
long ru_msgsnd; /* messages sent */
long ru_msgrcv; /* messages received */
long ru_nvcsw; /* voluntary context switches */
long ru_nivcsw; /* involuntary context switches */
test75.c is the unit test for this new function
Change-Id: I3f1eb69de1fce90d087d76773b09021fc6106539
kernel:
. modules can be as big as the space (8MB) between them
instead of 4MB; memory is slightly bigger with DBG=-g
arm ucontext:
. r4 is clobbered by the restore function, as it's
used as a scratch register, causing problems for the
DBG=-g build
. r1-r3 are safe for scratch registers, as they are
caller-save, so use r3 instead; and don't bother
restoring r1-r3, but preserve r4
vfs:
. improve TLL pointer sanity check a bit
Change-Id: I0e3cfc367fdc14477e40d04b5e044f288ca4cc7d
. unpause() and revive() can race - revive() can run during
a device i/o unblock, causing two sendnb()s to occur, and the
2nd one to fail
. this can easily happen when a process is blocking on tty and
is then killed by a signal - tty cancels the i/o and then
kills the process by a signal
Change-Id: Ia319acaedfa336b78c030a2c4af7246959bdcf87
. libc: add vfs_mmap, a way for vfs to initiate mmap()s.
This is a good special case to have as vfs is a slightly
different client from regular user processes. It doesn't do it
for itself, and has the dev & inode info already so the callback
to VFS for the lookup isn't necessary. So it has different info
to have to give to VM.
. libc: also add minix_mmap64() that accepts a 64-bit offset, even
though our off_t is still 32 bit now.
. On exec() time, try to mmap() in the executable if available.
(It is not yet available in this commit.)
. To support mmap(), add do_vm_call that allows VM to lookup
(to ino+dev), do i/o from and close FD's on behalf of other
processes.
Change-Id: I831551e45a6781c74313c450eb9c967a68505932
This commit introduces a new request type called REQ_BPEEK. It
requests minor device blocks from the FS. Analogously to REQ_PEEK,
it requests the filesystem to get the requested blocks into its
cache, without actually copying the result anywhere.
Change-Id: If1d06645b0e17553a64b3167091e9d12efeb3d6f
In libexec, split the memory allocation method into cleared and
non-cleared. Cleared gives zeroed memory, non-cleared gives 'junk'
memory (that will be overwritten anyway, and so needn't be cleared)
that is faster to get.
Also introduce the 'memmap' method that can be used, if available,
to map code and data from executables into a process using the
third-party mmap() mode.
Change-Id: I26694fd3c21deb8b97e01ed675dfc14719b0672b
. vfs read_only() assumes vnode->v_vmnt is non-NULL, but it can
be NULL sometimes
. e.g. fchmod() on UDS triggered NULL deref; add a check and
add REQ_CHMOD to pfs so unix domain sockets can be fchmod()ded
. add to test56
Change-Id: I83c840f101b647516897cc99fcf472116d762012
m_out is shared between threads as the reply message, and it can happen
results get overwritten by another thread before the reply is sent. This
change
. makes m_out local to the message handling function,
declared on the stack of the caller
. forces callers of reply() to give it a message, or
declare the reply message has no significant fields except
for the return code by calling replycode()
Change-Id: Id06300083a63c72c00f34f86a5c7d96e4bbdf9f6
Variant of utime(2) with struct timespec (with ns precision)
instead of time_t values; also allows for tv_nsec members
the values UTIME_NOW (force update to current time) or
UTIME_OMIT (allow to set either atim or mtim independently.)
Provides a superset of utimes(2), futimes(2), lutimes(2),
and futimens(2).
Provides the same subset of utimensat(2) as does NetBSD 6.
Also import utimens() and lutimeNS() from NetBSD-current.
This also adds the sys_settime() kernel call which allows for the adjusting
of the clock named realtime in the kernel. The existing sys_stime()
function is still needed for a separate job (setting the boottime). The
boottime is set in the readclock driver. The sys_settime() interface is
meant to be flexible and will support both clock_settime() and adjtime()
when adjtime() is implemented later.
settimeofday() was adjusted to use the clock_settime() interface.
One side note discovered during testing: uptime(1) (part of the last(1)),
uses wtmp to determine boottime (not Minix's times(2)). This leads `uptime`
to report odd results when you set the time to a time prior to boottime.
This isn't a new bug introduced by my changes. It's been there for a while.
In order to make it more clear that ticks should be used for timers
and realtime should be used for timestamps / displaying the date/time,
getuptime() was renamed to getticks() and getuptime2() was renamed to
getuptime().
Servers, drivers, libraries, tests, etc that use getuptime()/getuptime2()
have been updated. In instances where a realtime was calculated, the
calculation was changed to use realtime.
System calls clock_getres() and clock_gettime() were added to PM/libc.
When you provided a string with junk after the terminating nul to a
UNIX domain socket and used bind(2), the canonical path function would
not properly terminate the new string. This caused VFS to return
ENAMETOOLONG on an otherwise valid path name.
Test case is added to test56.
Change-Id: I883b6be23d9e4ea13c3cee28cbb3726343df037f
REQ_PEEK behaves just like REQ_READ except that it does not copy
data anywhere, just obtains the blocks from the FS into the cache.
To be used by the future mmap implementation.
Change-Id: I1b56de304f0a7152b69a72c8962d04258adb44f9
The build system distinction between "bootprog" and "service" is
meaningless as boot programs are standard services.
As minix.service.mk simply imports minix.bootprog.mk, reduce confusion
by removing minix.bootprog.mk and placing the rules in minix.service.mk.
Change-Id: I4056b1e574bed59a8c890239b41b1a7c7cad63e8
Remove old versions of system calls and system calls that don't have
a libc api interface anymore (dup, dup2, creat).
VFS still contains support for old system call numbers for the new stat
system calls (i.e., 65, 66, 67) to keep supporting old binaries built for
MINIX 3.2.1 (prior to the release).
Change-Id: I721779b58a50c7eeae20669de24658d55d69b25b
libchardriver does not support DEV_REOPEN and will return ERESTART
when you do try it. This made VFS unhappy and concluded erroneously
that the driver was EDEADEPT.
if an exec() fails partway through reading in the sections, the target
process is already gone and a defunct process remains. sanity checking
the binary beforehand helps that.
test10 mutilates binaries and exec()s them on purpose; making an exec()
fail cleanly in such cases seems like acceptable behaviour.
fixes test10 on ARM.
Change-Id: I1ed9bb200ce469d4d349073cadccad5503b2fcb0
* Updating common/lib
* Updating lib/csu
* Updating lib/libc
* Updating libexec/ld.elf_so
* Corrected test on __minix in featuretest to actually follow the
meaning of the comment.
* Cleaned up _REENTRANT-related defintions.
* Disabled -D_REENTRANT for libfetch
* Removing some unneeded __NBSD_LIBC defines and tests
Change-Id: Ic1394baef74d11b9f86b312f5ff4bbc3cbf72ce2
This patch uses stricter locking for REQ_LINK, REQ_MKDIR, REQ_MKNOD,
REQ_RENAME, REQ_RMDIR, REQ_SLINK and REQ_UNLINK. For all requests, VFS
locks the directory in which we add or remove an inode with VNODE_WRITE.
I.e., the operations have exclusive access to that directory.
Furthermore, REQ_CHOWN, REQ_CHMOD, and REQ_FTRUNC now lock the vmnt
VMNT_READ; VMNT_WRITE was unnecessary.
Because pipes have no file position. VFS maintained (file) offsets into a
buffer internal to PFS and stored them in vnodes for simplicity, mixing
the responsibilities of filp and vnode objects.
With this patch PFS ignores the position field in REQ_READ and REQ_WRITE
requests making VFS' job a lot simpler.
.sync and fsync used unnecessarily restrictive locking type
.fsync violated locking order by obtaining a vmnt lock after a filp lock
.fsync contained a TOCTOU bug
.new_node violated locking rules (didn't upgrade lock upon file creation)
.do_pipe used unnecessarily restrictive locking type
.always lock pipes exclusively; even a read operation might require to do
a write on a vnode object (update pipe size)
.when opening a file with O_TRUNC, upgrade vnode lock when truncating
.utime used unnecessarily restrictive locking type
.path parsing:
.always acquire VMNT_WRITE or VMNT_EXCL on vmnt and downgrade to
VMNT_READ if that was what was actually requested. This prevents the
following deadlock scenario:
thread A:
lock_vmnt(vmp, TLL_READSER);
lock_vnode(vp, TLL_READSER);
upgrade_vmnt_lock(vmp, TLL_WRITE);
thread B:
lock_vmnt(vmp, TLL_READ);
lock_vnode(vp, TLL_READSER);
thread A will be stuck in upgrade_vmnt_lock and thread B is stuck in
lock_vnode. This happens when, for example, thread A tries create a
new node (open.c:new_node) and thread B tries to do eat_path to
change dir (stadir.c:do_chdir). When the path is being resolved, a
vnode is always locked with VNODE_OPCL (TLL_READSER) and then
downgraded to VNODE_READ if read-only is actually requested. Thread
A locks the vmnt with VMNT_WRITE (TLL_READSER) which still allows
VMNT_READ locks. Thread B can't acquire a lock on the vnode because
thread A has it; Thread A can't upgrade its vmnt lock to VMNT_WRITE
(TLL_WRITE) because thread B has a VMNT_READ lock on it.
By serializing vmnt locks during path parsing, thread B can only
acquire a lock on vmp when thread A has completely finished its
operation.
mount.c: In function 'mount_pfs':
mount.c:395:17: error: variable 'rfp' set but not used [-Werror=unused-but-set-variable]
Change-Id: I2f22590ab4e3a4a1678e9096626ebca53d2660e6
new_node makes the assumption that when it does last_dir on a path, a
successive advance would not yield a lock on a vmnt, because last_dir
already locked the vmnt. This is true except when last_dir resolves
to a directory on the parent vmnt of the file that was the result of
advance. For example,
# cd /
# echo foo > home
where home is on a different (sub) partition than / is (default
install). last_dir would resolve to / and advance would resolve to
/home.
With this change, last_dir resolves to the root node on the /home
partition, making the assumption valid again.
The VFS/FS protocol does not require the file server to supply a
special device node number in response to a REQ_CREATE request, as
this call creates only regular files. Therefore, VFS should not
erroneously save this piece of information from the REQ_CREATE reply
either.
Upon reboot VFS semi-exits all processes and unmounts the file system.
However, upon unmount, exiting FUSE file systems might need service from
the file system (due to libc). As the FUSE process is halfway the exit
procedure, it doesn't have a valid root directory and working directory.
Trying to do system calls then triggers a sanity check in VFS.
This fix first exits normal processes which should then allow for
unmounting FUSE file systems. Then VFS exits all processes including
File Servers and unmounts the rest of the file system.
There is a deadlock vulnerability when there are no worker threads
available and all of them blocked on a worker thread that's waiting for a
reply from a driver or a reply from an FS that needs to make a back call. In
these cases the deadlock resolver thread should kick in, but didn't in all
cases. Moreover, POSIX calls from File Servers weren't handled properly
anymore, which also could lead to deadlocks.
The check_bsf() macro uses assert(mutex_trylock(&bsf_lock)) and
assumes bsf_lock is locked afterwards. This breaks when compiling
with NOASSERTS="yes". Also: macro to function transition.
. whenever this function is called, pm will expect
the process to be cleaned up
. so don't abort the process entirely on error
. fixes a later 'forking on top of in-use child' vfs panic
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.
. ld.so is linked at 0 but it can relocate itself; we
wish to load ld.so higher though to trap NULL dereferences.
if we know we have to execute ld.so, vfs tells libexec to put it
higher.
When VFS runs out of vnodes after closing a vnode in opcl, common_open
will try to unlock a vnode through unlock_filp that has already been
unlocked in clone_opcl. By first obtaining and locking a new vnode this
situation is prevented; if there are no free vnodes, common_open will
unlock a still locked vnode.
. some strncpy/strcpy to strlcpy conversions
. new <minix/param.h> to avoid including other minix headers
that have colliding definitions with library and commands code,
causing parse warnings
. removed some dead code / assignments
This commit removes all traces of Minix segments (the text/data/stack
memory map abstraction in the kernel) and significance of Intel segments
(hardware segments like CS, DS that add offsets to all addressing before
page table translation). This ultimately simplifies the memory layout
and addressing and makes the same layout possible on non-Intel
architectures.
There are only two types of addresses in the world now: virtual
and physical; even the kernel and processes have the same virtual
address space. Kernel and user processes can be distinguished at a
glance as processes won't use 0xF0000000 and above.
No static pre-allocated memory sizes exist any more.
Changes to booting:
. The pre_init.c leaves the kernel and modules exactly as
they were left by the bootloader in physical memory
. The kernel starts running using physical addressing,
loaded at a fixed location given in its linker script by the
bootloader. All code and data in this phase are linked to
this fixed low location.
. It makes a bootstrap pagetable to map itself to a
fixed high location (also in linker script) and jumps to
the high address. All code and data then use this high addressing.
. All code/data symbols linked at the low addresses is prefixed by
an objcopy step with __k_unpaged_*, so that that code cannot
reference highly-linked symbols (which aren't valid yet) or vice
versa (symbols that aren't valid any more).
. The two addressing modes are separated in the linker script by
collecting the unpaged_*.o objects and linking them with low
addresses, and linking the rest high. Some objects are linked
twice, once low and once high.
. The bootstrap phase passes a lot of information (e.g. free memory
list, physical location of the modules, etc.) using the kinfo
struct.
. After this bootstrap the low-linked part is freed.
. The kernel maps in VM into the bootstrap page table so that VM can
begin executing. Its first job is to make page tables for all other
boot processes. So VM runs before RS, and RS gets a fully dynamic,
VM-managed address space. VM gets its privilege info from RS as usual
but that happens after RS starts running.
. Both the kernel loading VM and VM organizing boot processes happen
using the libexec logic. This removes the last reason for VM to
still know much about exec() and vm/exec.c is gone.
Further Implementation:
. All segments are based at 0 and have a 4 GB limit.
. The kernel is mapped in at the top of the virtual address
space so as not to constrain the user processes.
. Processes do not use segments from the LDT at all; there are
no segments in the LDT any more, so no LLDT is needed.
. The Minix segments T/D/S are gone and so none of the
user-space or in-kernel copy functions use them. The copy
functions use a process endpoint of NONE to realize it's
a physical address, virtual otherwise.
. The umap call only makes sense to translate a virtual address
to a physical address now.
. Segments-related calls like newmap and alloc_segments are gone.
. All segments-related translation in VM is gone (vir2map etc).
. Initialization in VM is simpler as no moving around is necessary.
. VM and all other boot processes can be linked wherever they wish
and will be mapped in at the right location by the kernel and VM
respectively.
Other changes:
. The multiboot code is less special: it does not use mb_print
for its diagnostics any more but uses printf() as normal, saving
the output into the diagnostics buffer, only printing to the
screen using the direct print functions if a panic() occurs.
. The multiboot code uses the flexible 'free memory map list'
style to receive the list of free memory if available.
. The kernel determines the memory layout of the processes to
a degree: it tells VM where the kernel starts and ends and
where the kernel wants the top of the process to be. VM then
uses this entire range, i.e. the stack is right at the top,
and mmap()ped bits of memory are placed below that downwards,
and the break grows upwards.
Other Consequences:
. Every process gets its own page table as address spaces
can't be separated any more by segments.
. As all segments are 0-based, there is no distinction between
virtual and linear addresses, nor between userspace and
kernel addresses.
. Less work is done when context switching, leading to a net
performance increase. (8% faster on my machine for 'make servers'.)
. The layout and configuration of the GDT makes sysenter and syscall
possible.
. sys_vircopy always uses D for both src and dst
. sys_physcopy uses PHYS_SEG if and only if corresponding
endpoint is NONE, so we can derive the mode (PHYS_SEG or D)
from the endpoint arg in the kernel, dropping the seg args
. fields in msg still filled in for backwards compatability,
using same NONE-logic in the library
. all invocations were S or D, so can safely be dropped
to prepare for the segmentless world
. still assign D to the SCP_SEG field in the message
to make previous kernels usable
. new mode for sys_memset: include process so memset can be
done in physical or virtual address space.
. add a mode to mmap() that lets a process allocate uninitialized
memory.
. this allows an exec()er (RS, VFS, etc.) to request uninitialized
memory from VM and selectively clear the ranges that don't come
from a file, leaving no uninitialized memory left for the process
to see.
. use callbacks for clearing the process, clearing memory in the
process, and copying into the process; so that the libexec code
can be used from rs, vfs, and in the future, kernel (to load vm)
and vm (to load boot-time processes)
. make exec() callers (i.e. vfs and rs) determine the
memory layout by explicitly reserving regions using
mmap() calls on behalf of the exec()ing process,
i.e. handling all of the exec logic, thereby eliminating
all special exec() knowledge from VM.
. the new procedure is: clear the exec()ing process
first, then call third-party mmap()s to reserve memory, then
copy the executable file section contents in, all using callbacks
tailored to the caller's way of starting an executable
. i.e. no more explicit EXEC_NEWMEM-style calls in PM or VM
as with rigid 2-section arguments
. this naturally allows generalizing exec() by simply loading
all ELF sections
. drop/merge of lots of duplicate exec() code into libexec
. not copying the code sections to vfs and into the executable
again is a measurable performance improvement (about 3.3% faster
for 'make' in src/servers/)
justification: soon we won't be able to execute sep I&D aouts at
all (because of the vanishing segments), which was the default mode
to generate them so most binaries will be sep I&D.
this makes the vfs/rs exec() unification work simpler.
after unification, common I&D aout could be added back quite simply.
Only attempt to release blocked processes that are blocked. There is
no use in trying to find more blocked processes than we know that are
blocked (on a pipe).
According to POSIX the st_size field of struct stat is undefined for
fifos and anonymous pipes. Thus we can do anything we want. We save a
copy by not being accurate on pipe sizes.
. vfs: pass execname in aux vectors
. ld.elf_so: use this to expand $ORIGIN
. this requires the executable to reserve more
space at exec() calling time
POSIX mandates that a file's modification and change time be left
untouched upon truncate/ftruncate iff the file size does not change.
However, an open(O_TRUNC) call must always update the modification and
change time of the file, even if it was already zero-sized. VFS uses
the file systems' truncate call to implement O_TRUNC. This patch
replaces git-255ae85, which did not take into account the open case.
The size check is now moved into VFS, so that individual file systems
need not check for this case anymore.
. generalize libexec slightly to get some more necessary information
from ELF files, e.g. the interpreter
. execute dynamically linked executables when exec()ed by VFS
. switch to netbsd variant of elf32.h exclusively, solves some
conflicting headers
Pipes consist of two filps (read filp and write filp) and a shared
vnode. When the writer leaves the filp reference count drops to
zero and subsequent find_filp()s should not find the filp when a
reader looks for it and the reader gets EOF. However, the pipe()
system call tries to find two filps, marks them in use, and only
after a successful node creation on PFS, overwrites the shared
vnode with the new vnode. Consequently, this leaves a small window
where a just closed 'pipe write filp' gets reused and marked as
present, before becoming the actual new 'pipe write filp' for a new
pipe. A reader for the old pipe will think a writer is present and
wait for that writer to write something or to leave; both actions
should revive the suspended reader. This will never happen and the
reader will be stuck forever.
When running out of worker threads to handle device replies a dead
lock resolver thread is used. However, it was only used for FS
endpoints; it is now used for "system processes" (drivers and FS
endpoints). Also, drivers were marked as system process when they
were not "forced" to map (i.e., mapping was done before endpoint was
alive).
By making m_in job local (i.e., each job has its own copy of m_in instead
of refering to the global m_in) we don't have to store and restore m_in
on every thread yield. This reduces overhead. Moreover, remove the
assumption that m_in is preserved. Do_XXX functions have to copy the
system call parameters as soon as possible and only pass those copies to
other functions.
Furthermore, this patch cleans up some code and uses better types in a lot
of places.
- add files needed for acpi, ahci, fbd, vfs to libminc
- remove "-lc" from their respective makefiles
- remove setenv from libminc (requires initialization)
- if an operation (R, W, IOCTL) is non blocking, a flag is set
and sent to the device.
- nothing changes for sync devices
- asyn devices should reply asap if an operation is non-blocking.
We must trust the devices, but we had to trust them anyway to
reply to CANCEL correctly
- we safe sending CANCEL commands to asyn devices. This greatly
simplifies the protocol. Asynchronous devices can always reply
when a reply is ready and do not need to deal with other
situations
- currently, none of our drivers use the flags since they drive
virtual devices which do not block
- select_request_async() returns no ops by default
- wantops in do_select() always set correctly, do_select() does
not need a special case for SUSPEND (and ugly code)
