from the buffer pages to buffer. Combine the code to set buffer
dirty range (previously in vfs_setdirty()) and to clean the pages
(vfs_clean_pages()) into new function vfs_clean_pages_dirty_buf(). Now
the vm object lock is acquired only once.
Drain the VPO_BUSY bit of the buffer pages before setting valid
and clean bits in vfs_clean_pages_dirty_buf() with new helper
vfs_drain_busy_pages(). pmap_clear_modify() asserts that page is not
busy.
In vfs_busy_pages(), move the wait for draining of VPO_BUSY before
the dirtyness handling, to follow the structure of
vfs_clean_pages_dirty_buf().
Reported and tested by: pho
Suggested and reviewed by: alc
MFC after: 2 weeks
independent code. Move this code into mincore(), and eliminate the
page queues lock from pmap_mincore().
Push down the page queues lock into pmap_clear_modify(),
pmap_clear_reference(), and pmap_is_modified(). Assert that these
functions are never passed an unmanaged page.
Eliminate an inaccurate comment from powerpc/powerpc/mmu_if.m:
Contrary to what the comment says, pmap_mincore() is not simply an
optimization. Without a complete pmap_mincore() implementation,
mincore() cannot return either MINCORE_MODIFIED or MINCORE_REFERENCED
because only the pmap can provide this information.
Eliminate the page queues lock from vfs_setdirty_locked_object(),
vm_pageout_clean(), vm_object_page_collect_flush(), and
vm_object_page_clean(). Generally speaking, these are all accesses
to the page's dirty field, which are synchronized by the containing
vm object's lock.
Reduce the scope of the page queues lock in vm_object_madvise() and
vm_page_dontneed().
Reviewed by: kib (an earlier version)
eliminate it.
Assert that the object containing the page is locked in
vm_page_test_dirty(). Perform some style clean up while I'm here.
Reviewed by: kib
vm_page_try_to_free(). Consequently, push down the page queues lock into
pmap_enter_quick(), pmap_page_wired_mapped(), pmap_remove_all(), and
pmap_remove_write().
Push down the page queues lock into Xen's pmap_page_is_mapped(). (I
overlooked the Xen pmap in r207702.)
Switch to a per-processor counter for the total number of pages cached.
to unconditionally set PG_REFERENCED on a page before sleeping. In many
cases, it's perfectly ok for the page to disappear, i.e., be reclaimed by
the page daemon, before the caller to vm_page_sleep() is reawakened.
Instead, we now explicitly set PG_REFERENCED in those cases where having
the page persist until the caller is awakened is clearly desirable. Note,
however, that setting PG_REFERENCED on the page is still only a hint,
and not a guarantee that the page should persist.
architecture from page queue lock to a hashed array of page locks
(based on a patch by Jeff Roberson), I've implemented page lock
support in the MI code and have only moved vm_page's hold_count
out from under page queue mutex to page lock. This changes
pmap_extract_and_hold on all pmaps.
Supported by: Bitgravity Inc.
Discussed with: alc, jeffr, and kib
brings in support for an optional intent log which eliminates the need
for background fsck on unclean shutdown.
Sponsored by: iXsystems, Yahoo!, and Juniper.
With help from: McKusick and Peter Holm
In r205860 I missed the fact that there is code that strongly assumes
that devvp bo_bsize is equal to underlying provider's sectorsize.
In those places it is hard to obtain the sectorsize in an alternative
way if devvp bo_bsize is set to something else.
So, I am reverting bo_bsize assigment in g_vfs_open.
Instead, in getblk I use DEV_BSIZE block size for b_offset calculation
if vp is a disk vp as reported by vn_isdisk. This should coinside with
vp being a devvp.
Reported by: Mykola Dzham <i@levsha.me>
Tested by: Mykola Dzham <i@levsha.me>
Pointyhat to: avg
MFC after: 2 weeks
X-ToDo: convert bread(devvp) in all fs to use bo_bsize-d blocks
portion of the page that was written. Among other problems, this
page might be picked up by pagedaemon, with failed assertion in
vm_pageout_flush() about validity of the page.
Reported and tested by: pho
Approved by: re (kensmith)
MFC after: 3 weeks
Add support for kernel fault injection using KFAIL_POINT_* macros and
fail_point_* infrastructure. Add example fail point in vfs_bio.c to
simulate VM buf pressure.
Approved by: dfr (mentor)
Provide a more descriptive comment.
Eliminate dead code. The page cannot possibly have PG_ZERO set.
Eliminate unnecessary blank lines.
Reviewed by: tegge
following changes:
Rename vfs_page_set_valid() to vfs_page_set_validclean() to reflect
what this function actually does. Suggested by: tegge
Introduce a new version of vfs_page_set_valid() that does no more than
what the function's name implies. Specifically, it does not update
the page's dirty mask, and thus it does not require the page queues
lock to be held.
Update two of the three callers to the old vfs_page_set_valid() to
call vfs_page_set_validclean() instead because they actually require
the page's dirty mask to be cleared.
Introduce vm_page_set_valid().
Reviewed by: tegge
suffer from the race condition that motivated revision 1.94. Consequently,
the work-around that was implemented by revision 1.94 is no longer needed.
Moreover, reverting this work-around eliminates the need for
vfs_busy_pages() to acquire the page queues lock when preparing a buffer
for read.
Reviewed by: tegge
B_DELWRI cleanup and vnode disassociation should happen just before to
assign the buffer to a queue.
Reported by: miwi, Volker <volker at vwsoft dot com>,
Ben Kaduk <minimarmot at gmail dot com>,
Christopher Mallon <christoph dot mallon at gmx dot de>
Tested by: lulf, miwi
the "nbufkv" sleep.
First, ffs background cg group block write requests a new buffer for
the shadow copy. When ffs_bufwrite() is called from the bufdaemon due
to buffers shortage, requesting the buffer deadlock bufdaemon.
Introduce a new flag for getnewbuf(), GB_NOWAIT_BD, to request getblk
to not block while allocating the buffer, and return failure
instead. Add a flag argument to the geteblk to allow to pass the flags
to getblk(). Do not repeat the getnewbuf() call from geteblk if buffer
allocation failed and either GB_NOWAIT_BD is specified, or geteblk()
is called from bufdaemon (or its helper, see below). In
ffs_bufwrite(), fall back to synchronous cg block write if shadow
block allocation failed.
Since r107847, buffer write assumes that vnode owning the buffer is
locked. The second problem is that buffer cache may accumulate many
buffers belonging to limited number of vnodes. With such workload,
quite often threads that own the mentioned vnodes locks are trying to
read another block from the vnodes, and, due to buffer cache
exhaustion, are asking bufdaemon for help. Bufdaemon is unable to make
any substantial progress because the vnodes are locked.
Allow the threads owning vnode locks to help the bufdaemon by doing
the flush pass over the buffer cache before getnewbuf() is going to
uninterruptible sleep. Move the flushing code from buf_daemon() to new
helper function buf_do_flush(), that is called from getnewbuf(). The
number of buffers flushed by single call to buf_do_flush() from
getnewbuf() is limited by new sysctl vfs.flushbufqtarget. Prevent
recursive calls to buf_do_flush() by marking the bufdaemon and threads
that temporarily help bufdaemon by TDP_BUFNEED flag.
In collaboration with: pho
Reviewed by: tegge (previous version)
Tested by: glebius, yandex ...
MFC after: 3 weeks
address space sizes to be longs instead of ints. Specifically, the follow
values are now longs: runningbufspace, bufspace, maxbufspace,
bufmallocspace, maxbufmallocspace, lobufspace, hibufspace, lorunningspace,
hirunningspace, maxswzone, maxbcache, and maxpipekva. Previously, a
relatively small number (~ 44000) of buffers set in kern.nbuf would result
in integer overflows resulting either in hangs or bogus values of
hidirtybuffers and lodirtybuffers. Now one has to overflow a long to see
such problems. There was a check for a nbuf setting that would cause
overflows in the auto-tuning of nbuf. I've changed it to always check and
cap nbuf but warn if a user-supplied tunable would cause overflow.
Note that this changes the ABI of several sysctls that are used by things
like top(1), etc., so any MFC would probably require a some gross shims
to allow for that.
MFC after: 1 month
- Align the fifo output in fifo_print() with other vn_printf() output.
- Remove the leading space from lockmgr_printinfo() so its output lines up
in vn_printf().
- lockmgr_printinfo() now ends with a newline, so remove an extra newline
from vn_printf().
In particular following functions KPI results modified:
- bufobj_invalbuf()
- bufsync()
and BO_SYNC() "virtual method" of the buffer objects set.
Main consumers of bufobj functions are affected by this change too and,
in particular, functions which changed their KPI are:
- vinvalbuf()
- g_vfs_close()
Due to the KPI breakage, __FreeBSD_version will be bumped in a later
commit.
As a side note, please consider just temporary the 'curthread' argument
passing to VOP_SYNC() (in bufsync()) as it will be axed out ASAP
Reviewed by: kib
Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
Show the b_dep value for the buffer in the show buffer command.
Add a comand to dump the dirty/clean buffer list for vnode.
Reviewed by: tegge
Tested and used by: pho
MFC after: 1 month
dropped after the call to lockmgr() so just revert this approach using
something similar to the precedent one:
BUF_LOCKWAITERS() just checks if there are waiters (not the actual number
of them) and it is based on newly introduced lockmgr_waiters() which
returns if the lockmgr has waiters or not. The name has been choosen
differently by old lockwaiters() in order to not confuse them.
KPI results enriched by this commit so __FreeBSD_version bumping and
manpage update will be happening soon.
'struct buf' also changes, so kernel ABI is disturbed.
Bug found by: jeff
Approved by: jeff, kib
BO_LOCK/UNLOCK/MTX when manipulating the bufobj.
- Create a new lock in the bufobj to lock bufobj fields independently.
This leaves the vnode interlock as an 'identity' lock while the bufobj
is an io lock. The bufobj lock is ordered before the vnode interlock
and also before the mnt ilock.
- Exploit this new lock order to simplify softdep_check_suspend().
- A few sync related functions are marked with a new XXX to note that
we may not properly interlock against a non-zero bv_cnt when
attempting to sync all vnodes on a mountlist. I do not believe this
race is important. If I'm wrong this will make these locations easier
to find.
Reviewed by: kib (earlier diff)
Tested by: kris, pho (earlier diff)
around the check for the BV_BKGRDINPROG in the brelse() and bqrelse().
See the comment for the explanation why it is safe.
Tested by: pho
Submitted by: jeff
after each SYSINIT() macro invocation. This makes a number of
lightweight C parsers much happier with the FreeBSD kernel
source, including cflow's prcc and lxr.
MFC after: 1 month
Discussed with: imp, rink
than rely on the lockmgr support [1]:
* bump the waiters only if the interlock is held
* let brelvp() return the waiters count
* rely on brelvp() instead than BUF_LOCKWAITERS() in order to check
for the waiters number
- Remove a namespace pollution introduced recently with lockmgr.h
including lock.h by including lock.h directly in the consumers and
making it mandatory for using lockmgr.
- Modify flags accepted by lockinit():
* introduce LK_NOPROFILE which disables lock profiling for the
specified lockmgr
* introduce LK_QUIET which disables ktr tracing for the specified
lockmgr [2]
* disallow LK_SLEEPFAIL and LK_NOWAIT to be passed there so that it
can only be used on a per-instance basis
- Remove BUF_LOCKWAITERS() and lockwaiters() as they are no longer
used
This patch breaks KPI so __FreBSD_version will be bumped and manpages
updated by further commits. Additively, 'struct buf' changes results in
a disturbed ABI also.
[2] Really, currently there is no ktr tracing in the lockmgr, but it
will be added soon.
[1] Submitted by: kib
Tested by: pho, Andrea Barberio <insomniac at slackware dot it>
A couple of notes for this:
* WITNESS support, when enabled, is only used for shared locks in order
to avoid problems with the "disowned" locks
* KA_HELD and KA_UNHELD only exists in the lockmgr namespace in order
to assert for a generic thread (not curthread) owning or not the
lock. Really, this kind of check is bogus but it seems very
widespread in the consumers code. So, for the moment, we cater this
untrusted behaviour, until the consumers are not fixed and the
options could be removed (hopefully during 8.0-CURRENT lifecycle)
* Implementing KA_HELD and KA_UNHELD (not surported natively by
WITNESS) made necessary the introduction of LA_MASKASSERT which
specifies the range for default lock assertion flags
* About other aspects, lockmgr_assert() follows exactly what other
locking primitives offer about this operation.
- Build real assertions for buffer cache locks on the top of
lockmgr_assert(). They can be used with the BUF_ASSERT_*(bp)
paradigm.
- Add checks at lock destruction time and use a cookie for verifying
lock integrity at any operation.
- Redefine BUF_LOCKFREE() in order to not use a direct assert but
let it rely on the aforementioned destruction time check.
KPI results evidently broken, so __FreeBSD_version bumping and
manpage update result necessary and will be committed soon.
Side note: lockmgr_assert() will be used soon in order to implement
real assertions in the vnode namespace replacing the legacy and still
bogus "VOP_ISLOCKED()" way.
Tested by: kris (earlier version)
Reviewed by: jhb
lockmgr lkp, when held in exclusive mode, is recursed
- Introduce the function BUF_RECURSED() which does the same for bufobj
locks based on the top of lockmgr_recursed()
- Introduce the function BUF_ISLOCKED() which works like the counterpart
VOP_ISLOCKED(9), showing the state of lockmgr linked with the bufobj
BUF_RECURSED() and BUF_ISLOCKED() entirely replace the usage of bogus
BUF_REFCNT() in a more explicative and SMP-compliant way.
This allows us to axe out BUF_REFCNT() and leaving the function
lockcount() totally unused in our stock kernel. Further commits will
axe lockcount() as well as part of lockmgr() cleanup.
KPI results, obviously, broken so further commits will update manpages
and freebsd version.
Tested by: kris (on UFS and NFS)
conjuction with 'thread' argument passing which is always curthread.
Remove the unuseful extra-argument and pass explicitly curthread to lower
layer functions, when necessary.
KPI results broken by this change, which should affect several ports, so
version bumping and manpage update will be further committed.
Tested by: kris, pho, Diego Sardina <siarodx at gmail dot com>
Remove this argument and pass curthread directly to underlying
VOP_LOCK1() VFS method. This modify makes the code cleaner and in
particular remove an annoying dependence helping next lockmgr() cleanup.
KPI results, obviously, changed.
Manpage and FreeBSD_version will be updated through further commits.
As a side note, would be valuable to say that next commits will address
a similar cleanup about VFS methods, in particular vop_lock1 and
vop_unlock.
Tested by: Diego Sardina <siarodx at gmail dot com>,
Andrea Di Pasquale <whyx dot it at gmail dot com>
mounted FS' problems. These are more along the lines of 'avoiding an
avoidable panic' than a complete solution to removable devices. We
now close the barn door after the horse has gotten lose and has been
hit by a truck, as it were. The barn no longer catches fire in this
case, but the horse is still dead :-).
The vfs_bio.c fix causes us not to put a failed write back into the
dirty pool if the error returned was ENXIO. In that case, the buffer
is treated like any other clean buffer that's being retured. ENXIO
means the device isn't there anymore and will never be there again in
the future, so retrying is futile.
The vfs_mount.c fix treats 'ENXIO' as success for unmounting a file
system. If the device is gone, retrying later won't help and we'll
never be able to unmount the device.
These two are part of a larger patch set submitted by the author. The
other patches will be forth coming. I added comments to these two
patches.
Submitted by: Henrik Gulbrandsen
Reviewed by: phk@
PR: usb/46176 (partial)
to kproc_xxx as they actually make whole processes.
Thos makes way for us to add REAL kthread_create() and friends
that actually make theads. it turns out that most of these
calls actually end up being moved back to the thread version
when it's added. but we need to make this cosmetic change first.
I'd LOVE to do this rename in 7.0 so that we can eventually MFC the
new kthread_xxx() calls.
ways:
(1) Cached pages are no longer kept in the object's resident page
splay tree and memq. Instead, they are kept in a separate per-object
splay tree of cached pages. However, access to this new per-object
splay tree is synchronized by the _free_ page queues lock, not to be
confused with the heavily contended page queues lock. Consequently, a
cached page can be reclaimed by vm_page_alloc(9) without acquiring the
object's lock or the page queues lock.
This solves a problem independently reported by tegge@ and Isilon.
Specifically, they observed the page daemon consuming a great deal of
CPU time because of pages bouncing back and forth between the cache
queue (PQ_CACHE) and the inactive queue (PQ_INACTIVE). The source of
this problem turned out to be a deadlock avoidance strategy employed
when selecting a cached page to reclaim in vm_page_select_cache().
However, the root cause was really that reclaiming a cached page
required the acquisition of an object lock while the page queues lock
was already held. Thus, this change addresses the problem at its
root, by eliminating the need to acquire the object's lock.
Moreover, keeping cached pages in the object's primary splay tree and
memq was, in effect, optimizing for the uncommon case. Cached pages
are reclaimed far, far more often than they are reactivated. Instead,
this change makes reclamation cheaper, especially in terms of
synchronization overhead, and reactivation more expensive, because
reactivated pages will have to be reentered into the object's primary
splay tree and memq.
(2) Cached pages are now stored alongside free pages in the physical
memory allocator's buddy queues, increasing the likelihood that large
allocations of contiguous physical memory (i.e., superpages) will
succeed.
Finally, as a result of this change long-standing restrictions on when
and where a cached page can be reclaimed and returned by
vm_page_alloc(9) are eliminated. Specifically, calls to
vm_page_alloc(9) specifying VM_ALLOC_INTERRUPT can now reclaim and
return a formerly cached page. Consequently, a call to malloc(9)
specifying M_NOWAIT is less likely to fail.
Discussed with: many over the course of the summer, including jeff@,
Justin Husted @ Isilon, peter@, tegge@
Tested by: an earlier version by kris@
Approved by: re (kensmith)
td_ru. This removes the requirement for per-process synchronization in
statclock() and mi_switch(). This was previously supported by
sched_lock which is going away. All modifications to rusage are now
done in the context of the owning thread. reads proceed without locks.
- Aggregate exiting threads rusage in thread_exit() such that the exiting
thread's rusage is not lost.
- Provide a new routine, rufetch() to fetch an aggregate of all rusage
structures from all threads in a process. This routine must be used
in any place requiring a rusage from a process prior to it's exit. The
exited process's rusage is still available via p_ru.
- Aggregate tick statistics only on demand via rufetch() or when a thread
exits. Tick statistics are kept in the thread and protected by sched_lock
until it exits.
Initial patch by: attilio
Reviewed by: attilio, bde (some objections), arch (mostly silent)
Probabilly, a general approach is not the better solution here, so we should
solve the sched_lock protection problems separately.
Requested by: alc
Approved by: jeff (mentor)
vmcnts. This can be used to abstract away pcpu details but also changes
to use atomics for all counters now. This means sched lock is no longer
responsible for protecting counts in the switch routines.
Contributed by: Attilio Rao <attilio@FreeBSD.org>
a thread is an idle thread, just see if it has the IDLETD
flag set. That flag will probably move to the pflags word
as it's permenent and never chenges for the life of the
system so it doesn't need locking.
file are after snaplock, while other ffs device buffers are before
snaplock in global lock order. By itself, this could cause deadlock
when bdwrite() tries to flush dirty buffers on snapshotted ffs. If,
during the flush, COW activity for snapshot needs to allocate block
and ffs_alloccg() selects the cylinder group that is being written
by bdwrite(), then kernel would panic due to recursive buffer lock
acquision.
Avoid dealing with buffers in bdwrite() that are from other side of
snaplock divisor in the lock order then the buffer being written. Add
new BOP, bop_bdwrite(), to do dirty buffer flushing for same vnode in
the bdwrite(). Default implementation, bufbdflush(), refactors the code
from bdwrite(). For ffs device buffers, specialized implementation is
used.
Reviewed by: tegge, jeff, Russell Cattelan (cattelan xfs org, xfs changes)
Tested by: Peter Holm
X-MFC after: 3 weeks (if ever: it changes ABI)
vnode v_flag. For cluster buffers this would result in dereferencing NULL
b_vp. To prevent the panic, cache relevant vnode flag before calling
bstrategy.
Reported by: Peter Holm, kris
Tested by: Peter Holm
Reviewed by: tegge
Pointy hat to: kib
by vnode. Allow for md thread and the thread that owns lock on vnode
backing the md device to do the write even when runningbufspace is
exhausted.
Tested by: Peter Holm
Reviewed by: tegge
MFC after: 2 weeks
Call vfs_setdirty_locked_object() from vfs_busy_pages() instead of
vfs_setdirty(), thereby eliminating a second acquisition and release
of the same vm object lock.
queues lock to BIO_READ operations. Recent changes to the implementation
of the per-page flags have eliminated the need for the page queues lock
in the other cases.
synchronized by the lock on the object containing the page.
Transition PG_WANTED and PG_SWAPINPROG to use the new field,
eliminating the need for holding the page queues lock when setting
or clearing these flags. Rename PG_WANTED and PG_SWAPINPROG to
VPO_WANTED and VPO_SWAPINPROG, respectively.
Eliminate the assertion that the page queues lock is held in
vm_page_io_finish().
Eliminate the acquisition and release of the page queues lock
around calls to vm_page_io_finish() in kern_sendfile() and
vfs_unbusy_pages().
requires Giant. It is set in bgetvp and cleared in brelvp.
- Create QUEUE_DIRTY_GIANT for dirty buffers that require giant.
- In the buf daemon, only grab giant when processing QUEUE_DIRTY_GIANT and
only if we think there are buffers in that queue.
Sponsored by: Isilon Systems, Inc.
the system when brelse() was called with B_RELBUF set on the buffer. This
could be a problem when the system was low on memory, had many buffers on
QUEUE_EMPTYKVA and started to traverse directories. For each getnewbuf(),
pages were allocated from the system, driving the free reserve downwards.
For each brelse(), the system put the buffer on QUEUE_CLEAN, with B_INVAL
set.
This commit changes the semantics of B_RELBUF to also free pages from
non-VMIO buffers.
Reviewed by: alc
- provide an interface (macros) to the page coloring part of the VM system,
this allows to try different coloring algorithms without the need to
touch every file [1]
- make the page queue tuning values readable: sysctl vm.stats.pagequeue
- autotuning of the page coloring values based upon the cache size instead
of options in the kernel config (disabling of the page coloring as a
kernel option is still possible)
MD changes:
- detection of the cache size: only IA32 and AMD64 (untested) contains
cache size detection code, every other arch just comes with a dummy
function (this results in the use of default values like it was the
case without the autotuning of the page coloring)
- print some more info on Intel CPU's (like we do on AMD and Transmeta
CPU's)
Note to AMD owners (IA32 and AMD64): please run "sysctl vm.stats.pagequeue"
and report if the cache* values are zero (= bug in the cache detection code)
or not.
Based upon work by: Chad David <davidc@acns.ab.ca> [1]
Reviewed by: alc, arch (in 2004)
Discussed with: alc, Chad David, arch (in 2004)
- Prefer '_' to ' ', as it results in more easily parsed results in
memory monitoring tools such as vmstat.
- Remove punctuation that is incompatible with using memory type names
as file names, such as '/' characters.
- Disambiguate some collisions by adding subsystem prefixes to some
memory types.
- Generally prefer lower case to upper case.
- If the same type is defined in multiple architecture directories,
attempt to use the same name in additional cases.
Not all instances were caught in this change, so more work is required to
finish this conversion. Similar changes are required for UMA zone names.
Add a new private thread flag to indicate that the thread should
not sleep if runningbufspace is too large.
Set this flag on the bufdaemon and syncer threads so that they skip
the waitrunningbufspace() call in bufwrite() rather than than
checking the proc pointer vs. the known proc pointers for these two
threads. A way of preventing these threads from being starved for
I/O but still placing limits on their outstanding I/O would be
desirable.
Set this flag in ffs_copyonwrite() to prevent bufwrite() calls from
blocking on the runningbufspace check while holding snaplk. This
prevents snaplk from being held for an arbitrarily long period of
time if runningbufspace is high and greatly reduces the contention
for snaplk. The disadvantage is that ffs_copyonwrite() can start
a large amount of I/O if there are a large number of snapshots,
which could cause a deadlock in other parts of the code.
Call runningbufwakeup() in ffs_copyonwrite() to decrement runningbufspace
before attempting to grab snaplk so that I/O requests waiting on
snaplk are not counted in runningbufspace as being in-progress.
Increment runningbufspace again before actually launching the
original I/O request.
Prior to the above two changes, the system could deadlock if enough
I/O requests were blocked by snaplk to prevent runningbufspace from
falling below lorunningspace and one of the bawrite() calls in
ffs_copyonwrite() blocked in waitrunningbufspace() while holding
snaplk.
See <http://www.holm.cc/stress/log/cons143.html>
bio may have been freed and reassigned by the wakeup before being
tested after releasing the bdonelock.
There's a non-zero chance this is the cause of a few of the crashes
knocking around with biodone() sitting in the stack backtrace.
Reviewed By: phk@
make the b_iodone callback responsible for setting it if it is needed.
Previously, it was set unconditionally by bufdone() without holding
whichever lock is shared by the b_iodone callback and the corresponding
top-half function. Consequently, in a race, the top-half function could
conclude that operation was done before the b_iodone callback finished.
See, for example, aio_physwakeup() and aio_fphysio().
Note: I don't believe that the other, more widely-used b_iodone callbacks
are affected.
Discussed with: jeff
Reviewed by: phk
MFC after: 2 weeks
atomic write request, it can fill the buffer cache with the entirety
of that write in order to handle retries. However, it never drops
the vnode lock, or else it wouldn't be atomic, so it ends up waiting
indefinitely for more buf memory that cannot be gotten as it has it
all, and it waits in an uncancellable state.
To fix this, hibufspace is exported and scaled to a reasonable
fraction. This is used as the limit of how much of an atomic write
request by the NFS client will be handled asynchronously. If the
request is larger than this, it will be turned into a synchronous
request which won't deadlock the system. It's possible this value is
far off from what is required by some, so it shall be tunable as soon
as mount_nfs(8) learns of the new field.
The slowdown between an asynchronous and a synchronous write on NFS
appears to be on the order of 2x-4x.
General nod by: gad
MFC after: 2 weeks
More testing: wes
PR: kern/79208
milliseconds due to what is essentially n^2 algorithmic complexity. This
change makes the algorithm N*2 instead. This heavy processing manifested
itself as skipping in audio and video playback due to the long scheduling
latencies and contention on giant by pcm.
- flushbufqueues() is now responsible for flushing multiple buffers
rather than one at a time. This allows us to save our progress in the
list by using a sentinal. We must do the numdirtywakeup() and
waitrunningbufspace() here now rather than in buf_daemon().
- Also add a uio_yield() after we have processed the list once for bufs
without deps and again for bufs with deps. This is to release Giant
and allow any other giant locked code to proceed.
Tested by: Many users on current@
Revealed by: schedgraph traces sent by Emil Mikulic & Anthony Ginepro
Give FFS vnodes a specific bufwrite method which contains all the
background write stuff and then calls into the default bufwrite()
for the rest of the job.
Remove all the background write related stuff from the normal bufwrite.
This drags the softdep_move_dependencies() back into FFS.
Long term, it is worth looking at simply copying the data into
allocated memory and issuing the bio directly and not create the
"shadow buf" in the first place (just like copy-on-write is done
in snapshots for instance). I don't think we really gain anything
but complexity from doing this with a buf.
- Remove some KASSERTs which are invalid if the appropriate lock is
not held.
- Slightly restructure bremfree() so that it is more sane.
- Change the flush code in bdwrite() to avoid acquiring a mutex
whenever possible.
- Change the flush code in bdwrite() to avoid holding the bufobj mutex
while calling buf_countdeps(). This introduces a lock-order
relationship with the softdep lock that can not otherwise be resolved.
- Don't set B_DONE until bufdone() is complete, otherwise another
processor may believe the buf is done before it is.
- Only acquire Giant if the caller has set b_iodone. Don't grab giant
around normal bufdone() calls.
Sponsored By: Isilon Systems, Inc.
I'm not sure why a credential was added to these in the first place, it is
not used anywhere and it doesn't make much sense:
The credentials for syncing a file (ability to write to the
file) should be checked at the system call level.
Credentials for syncing one or more filesystems ("none")
should be checked at the system call level as well.
If the filesystem implementation needs a particular credential
to carry out the syncing it would logically have to the
cached mount credential, or a credential cached along with
any delayed write data.
Discussed with: rwatson
setting the B_REMFREE flag in the buf. This is done to prevent lock order
reversals with code that must call bremfree() with a local lock held.
This also reduces overhead by removing two lock operations per buf for
fsync() and similar.
- Check for the B_REMFREE flag in brelse() and bqrelse() after the bqlock
has been acquired so that we may remove ourself from the free-list.
- Provide a bremfreef() function to immediately remove a buf from a
free-list for use only by NFS. This is done because the nfsclient code
overloads the b_freelist queue for its own async. io queue.
- Simplify the numfreebuffers accounting by removing a switch statement
that executed the same code in every possible case.
- getnewbuf() can encounter locked bufs on free-lists once Giant is removed.
Remove a panic associated with this condition and delay asserts that
inspect the buf until after it is locked.
Reviewed by: phk
Sponsored by: Isilon Systems, Inc.
need for most calls to vm_page_busy(). Specifically, most calls to
vm_page_busy() occur immediately prior to a call to vm_page_remove().
In such cases, the containing vm object is locked across both calls.
Consequently, the setting of the vm page's PG_BUSY flag is not even
visible to other threads that are following the synchronization
protocol.
This change (1) eliminates the calls to vm_page_busy() that
immediately precede a call to vm_page_remove() or functions, such as
vm_page_free() and vm_page_rename(), that call it and (2) relaxes the
requirement in vm_page_remove() that the vm page's PG_BUSY flag is
set. Now, the vm page's PG_BUSY flag is set only when the vm object
lock is released while the vm page is still in transition. Typically,
this is when it is undergoing I/O.
buf->b-dev.
Put a bio between the buf passed to dev_strategy() and the device driver
strategy routine in order to not clobber fields in the buf.
Assert copyright on vfs_bio.c and update copyright message to canonical
text. There is no legal difference between John Dysons two-clause
abbreviated BSD license and the canonical text.
Give ffs it's own bufobj->bo_ops vector and create a private strategy
routine, (currently misnamed for forwards compatibility), which is
just a copy of the generic bufstrategy routine except we call
softdep_disk_prewrite() directly instead of through the buf_prewrite()
indirection.
Teach UFS about the need for softdep_disk_prewrite() and call the
function directly in FFS.
Remove buf_prewrite() from the default bufstrategy() and from the
global bio_ops method vector.
We keep si_bsize_phys around for now as that is the simplest way to pull
the number out of disk device drivers in devfs_open(). The correct solution
would be to do an ioctl(DIOCGSECTORSIZE), but the point is probably mooth
when filesystems sit on GEOM, so don't bother for now.
and release of the global page queues lock required to make the call.
Remove GIANT_REQUIRED from vm_hold_free_pages(). All of its VM operations
are properly synchronized.
Extend it with a strategy method.
Add bufstrategy() which do the usual VOP_SPECSTRATEGY/VOP_STRATEGY
song and dance.
Rename ibwrite to bufwrite().
Move the two NFS buf_ops to more sensible places, add bufstrategy
to them.
Add inlines for bwrite() and bstrategy() which calls through
buf->b_bufobj->b_ops->b_{write,strategy}().
Replace almost all VOP_STRATEGY()/VOP_SPECSTRATEGY() calls with bstrategy().
Initialize b_bufobj for all buffers.
Make incore() and gbincore() take a bufobj instead of a vnode.
Make inmem() local to vfs_bio.c
Change a lot of VI_[UN]LOCK(bp->b_vp) to BO_[UN]LOCK(bp->b_bufobj)
also VI_MTX() to BO_MTX(),
Make buf_vlist_add() take a bufobj instead of a vnode.
Eliminate other uses of bp->b_vp where bp->b_bufobj will do.
Various minor polishing: remove "register", turn panic into KASSERT,
use new function declarations, TAILQ_FOREACH_SAFE() etc.
Add bufobj_wref(), bufobj_wdrop() and bufobj_wwait() to handle the write
count on a bufobj. Bufobj_wdrop() replaces vwakeup().
Use these functions all relevant places except in ffs_softdep.c where
the use if interlocked_sleep() makes this impossible.
Rename b_vnbufs to b_bobufs now that we touch all the relevant files anyway.
of the number of threads which are inside whatever is behind the
cdevsw for this particular cdev.
Make the device mutex visible through dev_lock() and dev_unlock().
We may want finer granularity later.
Replace spechash_mtx use with dev_lock()/dev_unlock().
preparation for integration of p4::phk_bufwork. In the future,
local filesystems will talk to GEOM directly and they will consequently
be able to issue BIO_DELETE directly. Since the removal of the fla
driver, BIO_DELETE has effectively been a no-op anyway.
When avoiding the zeroing of "bogus_page" when it appears in a buf,
be sure to advance the pointers into the data for successive pages.
The bug caused file corruption when read(2)ing from a "hole" in a
file where a previous page of the read block had already been faulted
in: fsx tripped up on this pretty quickly. The particular access
pattern is probably pretty unusual, so other applications probably
wouldn't have had problems, but you'd never know.
Reviewed By: alc@
around in the vnodes surroundings when we allocate a block.
Assign a blocksize when we create a vnode, and yell a warning (and ignore it)
if we got the wrong size.
Please email all such warnings to me.
The big lines are:
NODEV -> NULL
NOUDEV -> NODEV
udev_t -> dev_t
udev2dev() -> findcdev()
Various minor adjustments including handling of userland access to kernel
space struct cdev etc.
allocation and deallocation. This flag's principal use is shortly after
allocation. For such cases, clearing the flag is pointless. The only
unusual use of PG_ZERO is in vfs_bio_clrbuf(). However, allocbuf() never
requests a prezeroed page. So, vfs_bio_clrbuf() never sees a prezeroed
page.
Reviewed by: tegge@
silences an annoying warning in getblk() when VMIO'ing on a directory
vnode, which can happen when vfs.vmiodirenable is 1.
Bring the warning message in line with reality at the same time.
Submitted by: hmp
were a rather overwhelming task. I soon learned that if you don't know
where you're going to store something, at least try to pile it next to
something slightly related in the hope that a pattern emerges.
Apply the same principle to the ffs/snapshot/softupdates code which have
leaked into specfs: Add yet a buf-quasi-method and call it from the
only two places I can see it can make a difference and implement the
magic in ffs_softdep.c where it belongs.
It's not pretty, but at least it's one less layer violated.
This is what we came here for: Hang dev_t's from their cdevsw,
refcount cdevsw and dev_t and generally keep track of things a lot
better than we used to:
Hold a cdevsw reference around all entrances into the device driver,
this will be necessary to safely determine when we can unload driver
code.
Hold a dev_t reference while the device is open.
KASSERT that we do not enter the driver on a non-referenced dev_t.
Remove old D_NAG code, anonymous dev_t's are not a problem now.
When destroy_dev() is called on a referenced dev_t, move it to
dead_cdevsw's list. When the refcount drops, free it.
Check that cdevsw->d_version is correct. If not, set all methods
to the dead_*() methods to prevent entrance into driver. Print
warning on console to this effect. The device driver may still
explode if it is also incompatible with newbus, but in that case
we probably didn't get this far in the first place.
accurate reporting of multi-terabyte filesystem sizes.
You should build and boot a new kernel BEFORE doing a `make world'
as the new kernel will know about binaries using the old statfs
structure, but an old kernel will not know about the new system
calls that support the new statfs structure. Running an old kernel
after a `make world' will cause programs such as `df' that do a
statfs system call to fail with a bad system call.
Reviewed by: Bruce Evans <bde@zeta.org.au>
Reviewed by: Tim Robbins <tjr@freebsd.org>
Reviewed by: Julian Elischer <julian@elischer.org>
Reviewed by: the hoards of <arch@freebsd.org>
Sponsored by: DARPA & NAI Labs.
consistency initialized. Consequently, a number of conditionals that
checked the validity of b_object before passing it to VM_OBJECT_LOCK()
and VM_OBJECT_UNLOCK() are no longer needed.
waitrunningbufspace() calls so that they are always able to
proceed and clean up buffer space.
Submitted by: Brian Fundakowski Feldman <green@freebsd.org>
the point where it being a macro is no longer sensible, and it will
only be more so in days to come.
BIO_STRATEGY() is now only used from DEV_STRATEGY() and should not
be used directly anymore.
Put the contents of both in the new function dev_strategy() and
make DEV_STRATEGY() call that function.
In addition, this allows us to make the rather magic bufdonebio()
helper function static.
This alse saves hunderedandsome bytes of code in a typical kernel.
caused snapshot related problems.
- The vp can not be NULL here or we would panic in vfs_bio_awrite(). Stop
confusing the logic by checking for it in several places.
Submitted by: kirk and then rototilled by me to remove vp == NULL checks.
freed belong to the kernel object.)
- Increase the granularity of the vm object locking in vm_hold_load_pages()
in order to reduce the number of times that we acquire and release the
same lock.
pmap_extract_and_hold(). Note, however, that GIANT_REQUIRED should not be
removed until all platforms fully implement the "prot" parameter to
pmap_extract_and_hold().
Reviewed by: tegge
bail out if the buffer is not already present.
- The buffer returned by incore() is not locked and should not be sent to
brelse(). Use getblk() with the new GB_NOCREAT flag to preserve the
desired semantics.
- Surround all accesses of the BKGRD{WAIT,INPROG} flags with the vnode
interlock.
- Don't use the B_LOCKED flag and QUEUE_LOCKED for background write
buffers. Check for the BKGRDINPROG flag before recycling or throwing
away a buffer. We do this instead because it is not safe for us to move
the original buffer to a new queue from the callback on the background
write buffer.
- Remove the B_LOCKED flag and the locked buffer queue. They are no longer
used.
- The vnode interlock is used around checks for BKGRDINPROG where it may
not be strictly necessary. If we hold the buf lock the a back-ground
write will not be started without our knowledge, one may only be
completed while we're not looking. Rather than remove the code, Document
two of the places where this extra locking is done. A pass should be
done to verify and minimize the locking later.
where physical addresses larger than virtual addresses, such as i386s
with PAE.
- Use this to represent physical addresses in the MI vm system and in the
i386 pmap code. This also changes the paddr parameter to d_mmap_t.
- Fix printf formats to handle physical addresses >4G in the i386 memory
detection code, and due to kvtop returning vm_paddr_t instead of u_long.
Note that this is a name change only; vm_paddr_t is still the same as
vm_offset_t on all currently supported platforms.
Sponsored by: DARPA, Network Associates Laboratories
Discussed with: re, phk (cdevsw change)
- Create a new function bdone() which sets B_DONE and calls wakup(bp). This
is suitable for use as b_iodone for buf consumers who are not going
through the buf cache.
- Create a new function bwait() which waits for the buf to be done at a set
priority and with a specific wmesg.
- Replace several cases where the above functionality was implemented
without locking with the new functions.
requiring locked bufs in vfs_bio_awrite(). Previously the buf could
have been written out by fsync before we acquired the buf lock if it
weren't for giant. The cluster_wbuild() handles this race properly but
the single write at the end of vfs_bio_awrite() would not.
- Modify flushbufqueues() so there is only one copy of the loop. Pass a
parameter in that says whether or not we should sync bufs with deps.
- Call flushbufqueues() a second time and then break if we couldn't find
any bufs without deps.
- Define one flag GB_LOCK_NOWAIT that tells getblk() to pass the LK_NOWAIT
flag to the initial BUF_LOCK(). This will eventually be used in cases
were we want to use a buffer only if it is not currently in use.
- Convert all consumers of the getblk() api to use this extra parameter.
Reviwed by: arch
Not objected to by: mckusick
delta 1.371) we must ensure that we do not get ourselves into a
recursive trap endlessly trying to clean up after ourselves.
Reported by: Attila Nagy <bra@fsn.hu>
Sponsored by: DARPA & NAI Labs.
track of the number of dirty buffers held by a vnode. When a
bdwrite is done on a buffer, check the existing number of dirty
buffers associated with its vnode. If the number rises above
vfs.dirtybufthresh (currently 90% of vfs.hidirtybuffers), one
of the other (hopefully older) dirty buffers associated with
the vnode is written (using bawrite). In the event that this
approach fails to curb the growth in it the vnode's number of
dirty buffers (due to soft updates rollback dependencies),
the more drastic approach of doing a VOP_FSYNC on the vnode
is used. This code primarily affects very large and actively
written files such as snapshots. This change should eliminate
hanging when taking snapshots or doing background fsck on
very large filesystems.
Hopefully, one day it will be possible to cache filesystem
metadata in the VM cache as is done with file data. As it
stands, only the buffer cache can be used which limits total
metadata storage to about 20Mb no matter how much memory is
available on the system. This rather small memory gets badly
thrashed causing a lot of extra I/O. For example, taking a
snapshot of a 1Tb filesystem minimally requires about 35,000
write operations, but because of the cache thrashing (we only
have about 350 buffers at our disposal) ends up doing about
237,540 I/O's thus taking twenty-five minutes instead of four
if it could run entirely in the cache.
Reported by: Attila Nagy <bra@fsn.hu>
Sponsored by: DARPA & NAI Labs.
- Remove the buftimelock mutex and acquire the buf's interlock to protect
these fields instead.
- Hold the vnode interlock while locking bufs on the clean/dirty queues.
This reduces some cases from one BUF_LOCK with a LK_NOWAIT and another
BUF_LOCK with a LK_TIMEFAIL to a single lock.
Reviewed by: arch, mckusick
queue lock already held.
- In getblk() and flushbufqueues() use bremfreel() while we still have the
buf queue lock held to keep the lists consistent.
- Add LK_NOWAIT to two cases where we're essentially asserting that the bufs
are not locked while acquiring the locks. This will make sure that we get
the appropriate panic() and not another one for sleeping with a lock held.
that is protected by the vnode lock.
- Move B_SCANNED into b_vflags and call it BV_SCANNED.
- Create a vop_stdfsync() modeled after spec's sync.
- Replace spec_fsync, msdos_fsync, and hpfs_fsync with the stdfsync and some
fs specific processing. This gives all of these filesystems proper
behavior wrt MNT_WAIT/NOWAIT and the use of the B_SCANNED flag.
- Annotate the locking in buf.h
buf lists, synchronization variables, and atomic ops for the counters.
This change does not remove giant from any code although some pushdown
may be possible.
- In vfs_bio_awrite() don't access buf fields without the buf lock.
vm_pageout_deficit:
1. Update vm_pageout_deficit before VM_WAIT. There is no sense in
delaying the update; the sooner the pageout daemon receives this
information the better. Reviewed by: tegge
2. Update vm_pageout_deficit according to the number of pages still
needed to complete the allocation, not the original size of the
allocation. Submitted by: tegge
(These errors have existed since the introduction of vm_pageout_deficit
in revision 1.144.)
portable copy. Note that pmap_extract() must be used instead of
pmap_kextract().
This is precursor work to a reorganization of vmapbuf() to close remaining
user/kernel races (which can lead to a panic).
to sort out disk-io from file-io in the vm/buffer/filesystem space.
The intent is to sort VOP_STRATEGY calls into those which operate
on "real" vnodes and those which operate on VCHR vnodes. For
the latter kind, the call will be changed to VOP_SPECSTRATEGY,
possibly conditionally for those places where dual-use happens.
Add a default VOP_SPECSTRATEGY method which will call the normal
VOP_STRATEGY. First time it is called it will print debugging
information. This will only happen if a normal vnode is passed
to VOP_SPECSTRATEGY by mistake.
Add a real VOP_SPECSTRATEGY in specfs, which does what VOP_STRATEGY
does on a VCHR vnode today.
Add a new VOP_STRATEGY method in specfs to catch instances where
the conversion to VOP_SPECSTRATEGY has not yet happened. Handle
the request just like we always did, but first time called print
debugging information.
Apart up to two instances of console messages per boot, this amounts
to a glorified no-op commit.
If you get any of the messages on your console I would very much
like a copy of them mailed to phk@freebsd.org
they may be the only viable ones to flush. Thus it will now wait for
an inode lock if the other alternatives will result in rollbacks (and
immediate redirtying of the buffer). If only buffers with rollbacks
are available, one will be flushed, but then the buffer daemon will
wait briefly before proceeding. Failing to wait briefly effectively
deadlocks a uniprocessor since every other process writing to that
filesystem will wait for the buffer daemon to clean up which takes
close enough to forever to feel like a deadlock.
Reported by: Archie Cobbs <archie@dellroad.org>
Sponsored by: DARPA & NAI Labs.
Approved by: re
indirectly through vm_page_protect(). The one remaining page flag that
is updated by vm_page_protect() is already being updated by our various
pmap implementations.
Note: A later commit will similarly change the VM_PROT_READ case and
eliminate vm_page_protect().
to help clean up. After selecting a potential buffer to write, this
patch has it acquire a lock on the vnode that owns the buffer before
trying to write it. The vnode lock is necessary to avoid a race with
some other process holding the vnode locked and trying to flush its
dirty buffers. In particular, if the vnode in question is a snapshot
file, then the race can lead to a deadlock. To avoid slowing down the
buf_daemon, it does a non-blocking lock request when trying to lock
the vnode. If it fails to get the lock it skips over the buffer and
continues down its queue looking for buffers to flush.
Sponsored by: DARPA & NAI Labs.
