Commit Graph

1573 Commits

Author SHA1 Message Date
Konstantin Belousov
269d02f171 The dqrele() function syncs the dq, then acquires the dqh lock, and then
does final drop of the the dq reference to put it onto the free list.
There is a possibility that the dq would be found by another thread
after sync and before the dqh lock is acquired. If that other thread
drops the dq before we have taken the dqh lock, the dirty dq is put on
the free list.

Recheck the DQ_MOD after the dqh lock is relocked. Repeat dqsync() if
the dq is dirty. This ensures that up to date dq is written in the quota
file and fixes assertion in dqget().

Reported and tested by:	Frode Nordahl <frode nordahl net>
MFC after:	3 days
2008-12-08 11:04:17 +00:00
Konstantin Belousov
7b603a4ac7 Improve usefulness of the panic by printing the pointer to the problematic
dquot. In-tree gdb is often unable to get the dq value, so supply it in
panic message.

MFC after:	3 days
2008-12-07 13:25:06 +00:00
Konstantin Belousov
ffdaeffe21 Do not lock vnode interlock around reading of v_iflag to check VI_DOOMED.
Read of the pointer is atomic, and flag cannot be set while vnode lock
is held.

Requested by:	jhb
MFC after:	1 month
2008-12-02 11:12:50 +00:00
Konstantin Belousov
11c68fb23f Busy ufs filesystem around block of code that does ".." lookup. Since
mnt_lock is before lock of any vnode on the mp, it uses LK_NOWAIT. Since
MNTK_UNMOUNT may be transient, pdp lock is dropped when vfs_busy()
failed, and operation is retried after some time. This way, ffs_vget()
is not called on the mp that may be in the process of being destroyed by
unmount.

Check for the VI_DOOMED flag on pdp after its lock is reacquired, to
better detect some situations where directory containing ".."
entry is removed during the lookup.

Reviewed by:	tegge, attilio (previous version)
Tested by:	pho
MFC after:	1 month
2008-11-22 13:11:11 +00:00
John Baldwin
25c398ee9b Fix typo. 2008-11-19 20:06:59 +00:00
Doug Ambrisko
f1c1cdbb9c For now on every 10 cyclinder groups flush the buffer cache to free
up space.  If the buffer cache fills up then the disk systems can
grind to a halt.  Better tuning can be figured out later.

Tested by:	Tim, others and work
Reviewed by:	Kostik Belousov
PR:		128832
2008-11-13 17:40:21 +00:00
John Baldwin
2ef42c06d6 Quiet a WITNESS warning with the dirhash sx locks by setting the DUPOK
flag.  Specifically, if two threads race to create a dirhash for a
directory, then one might already have created a private dirhash
structure (and locked it) when it realizes the directory now has a
structure and tries to lock that one.
2008-11-04 18:56:12 +00:00
Edward Tomasz Napierala
4a4f18ed37 In UFS, when reading EA that contains ACL fails for some reason, include
inode number and filesystem name, so the administrator can fix the problem.

Approved by:	rwatson (mentor)
2008-11-04 12:30:31 +00:00
Attilio Rao
83b3bdbc8a Improve VFS locking:
- Implement real draining for vfs consumers by not relying on the
  mnt_lock and using instead a refcount in order to keep track of lock
  requesters.
- Due to the change above, remove the mnt_lock lockmgr because it is now
  useless.
- Due to the change above, vfs_busy() is no more linked to a lockmgr.
  Change so its KPI by removing the interlock argument and defining 2 new
  flags for it: MBF_NOWAIT which basically replaces the LK_NOWAIT of the
  old version (which was unlinked from the lockmgr alredy) and
  MBF_MNTLSTLOCK which provides the ability to drop the mountlist_mtx
  once the mnt interlock is held (ability still desired by most consumers).
- The stub used into vfs_mount_destroy(), that allows to override the
  mnt_ref if running for more than 3 seconds, make it totally useless.
  Remove it as it was thought to work into older versions.
  If a problem of "refcount held never going away" should appear, we will
  need to fix properly instead than trust on such hackish solution.
- Fix a bug where returning (with an error) from dounmount() was still
  leaving the MNTK_MWAIT flag on even if it the waiters were actually
  woken up. Just a place in vfs_mount_destroy() is left because it is
  going to recycle the structure in any case, so it doesn't matter.
- Remove the markercnt refcount as it is useless.

This patch modifies VFS ABI and breaks KPI for vfs_busy() so manpages and
__FreeBSD_version will be modified accordingly.

Discussed with:	kib
Tested by:	pho
2008-11-02 10:15:42 +00:00
Edward Tomasz Napierala
15bc6b2bd8 Introduce accmode_t. This is required for NFSv4 ACLs - it will be neccessary
to add more V* constants, and the variables changed by this patch were often
being assigned to mode_t variables, which is 16 bit.

Approved by:	rwatson (mentor)
2008-10-28 13:44:11 +00:00
Konstantin Belousov
5400fa16b4 Provide an explanation for getinoquota() call in the ufs_access vop.
MFC after:	3 days
2008-10-28 12:00:28 +00:00
Dag-Erling Smørgrav
e11e3f187d Fix a number of style issues in the MALLOC / FREE commit. I've tried to
be careful not to fix anything that was already broken; the NFSv4 code is
particularly bad in this respect.
2008-10-23 20:26:15 +00:00
Dag-Erling Smørgrav
1ede983cc9 Retire the MALLOC and FREE macros. They are an abomination unto style(9).
MFC after:	3 months
2008-10-23 15:53:51 +00:00
Konstantin Belousov
016f98f947 Assert that v_holdcnt is non-zero before entering lockmgr in vn_lock
and ffs_lock. This cannot catch situations where holdcnt is incremented
not by curthread, but I think it is useful.

Reviewed by:	tegge, attilio
Tested by:	pho
MFC after:	2 weeks
2008-10-20 10:11:33 +00:00
Konstantin Belousov
4560452f01 Sync up summary information for cylinder groups while data is already
in memory during snapshot creation. This improves the results of the
background fsck.

Submitted by: tegge
MFC after: 1 week
2008-10-13 14:05:01 +00:00
Attilio Rao
0d7935fd01 Remove the struct thread unuseful argument from bufobj interface.
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>
2008-10-10 21:23:50 +00:00
John Baldwin
f888634792 Enable shared lookups on UFS. There are some remaining issues with forced
unmounts, but those are in the VFS lookup code are not UFS specific.

Tested by:	pho, kris
2008-09-24 18:53:04 +00:00
John Baldwin
10fbe6292c Close a race between concurrent calls to ufsdirhash_recycle() and
ufsdirhash_free() introduced in my last commit by removing the dirhash
about to be free'd in ufsdirhash_free() from the global dirhash list
before dropping the sx lock.

Tested by:	kris
2008-09-22 20:53:22 +00:00
Konstantin Belousov
86dacdfe2b Initialize va_flags and va_filerev properly in VOP_GETATTR(). Don't
initialize va_vaflags and va_spare because they are not part of the
VOP_GETATTR() API. Also don't initialize birthtime to ctime or zero.

Submitted by:   Jaakko Heinonen <jh saunalahti fi>
Reviewed by:	bde
Discussed on:   freebsd-fs
MFC after:	1 month
2008-09-20 19:46:45 +00:00
John Baldwin
c481551083 Retire the 'i_reclen' field from the in-memory i-node. Previously,
during a DELETE lookup operation, lookup would cache the length of the
directory entry to be deleted in 'i_reclen'.  Later, the actual VOP to
remove the directory entry (ufs_remove, ufs_rename, etc.) would call
ufs_dirremove() which extended the length of the previous directory
entry to "remove" the deleted entry.

However, we always read the entire block containing the directory
entry when doing the removal, so we always have the directory entry to
be deleted in-memory when doing the update to the directory block.
Also, we already have to figure out where the directory entry that is
being removed is in the block so that we can pass the component name
to the dirhash code to update the dirhash.  So, instead of passing
'i_reclen' from ufs_lookup() to the ufs_dirremove() routine, just read
the 'd_reclen' field directly out of the entry being removed when
updating the length of the previous entry in the block.

This avoids a cosmetic issue of writing to 'i_reclen' while holding a
shared vnode lock.  It also slightly reduces the amount of side-band
data passed from ufs_lookup() to operations updating a directory via
the directory's i-node.

Reviewed by:	jeff
2008-09-16 19:06:44 +00:00
John Baldwin
b2ef6b1833 Fix a race with shared lookups on UFS. If the the dirhash code reached the
cap on memory usage, then shared LOOKUP operations could start free'ing
dirhash structures.  Without these fixes, concurrent free's on the same
directory could result in one of the threads blocked on a lock in a dirhash
structure free'd by the other thread.
- Replace the lockmgr lock in the dirhash structure with an sx lock.
- Use a reference count managed with ufsdirhash_hold()/drop() to determine
  when to free the dirhash structures.  The directory i-node holds a
  reference while the dirhash is attached to an i-node.  Code that wishes
  to lock the dirhash while holding a shared vnode lock must first
  acquire a private reference to the dirhash while holding the vnode
  interlock before acquiring the dirhash sx lock.  After acquiring the sx
  lock, it drops the private reference after checking to see if the
  dirhash is still used by the directory i-node.
2008-09-16 16:23:56 +00:00
John Baldwin
5316d529ec - Only set i_offset in the parent directory's i-node during a lookup for
non-LOOKUP operations.
- Relax a VOP assertion for a DELETE lookup.  rename() uses WANTPARENT
  instead of LOCKPARENT when looking up the source pathname.  ufs_rename()
  uses a relookup() to lock the parent directory when it decides to finally
  remove the source path.  Thus, it is ok for a DELETE with WANTPARENT set
  instead of LOCKPARENT to use a shared vnode lock rather than an exclusive
  vnode lock.

Reported by:	kris (2)
Reviewed by:	jeff
2008-09-16 16:18:36 +00:00
John Baldwin
1b7cf11b00 vdropl() drops the vnode interlock. Thus, the code in the QUOTA case that
upgrades the vnode lock if it is share locked was dropping the interlock
before actually checking VI_DOOMED.  Fix this by do the vdropl() after the
check and relying on it to drop the vnode interlock.

Reported by:	pho
Reviewed by:	kib
MFC after:	1 week
2008-09-16 16:15:38 +00:00
Konstantin Belousov
6fecb4e41e Suspend the write operations on the UFS filesystem being unmounted or
remounted from rw to ro.

Proposed and reviewed by:  tegge
In collaboration with:	pho
MFC after:	 1 month
2008-09-16 11:55:53 +00:00
Konstantin Belousov
2814d5ba5f When attempt is made to suspend a filesystem that is already syspended,
wait until the current suspension is lifted instead of silently returning
success immediately. The consequences of calling vfs_write() resume when
not owning the suspension are not well-defined at best.

Add the vfs_susp_clean() mount method to be called from
vfs_write_resume(). Set it to process_deferred_inactive() for ffs, and
stop calling it manually.

Add the thread flag TDP_IGNSUSP that allows to bypass the suspension
point in the vn_start_write. It is intended for use by VFS in the
situations where the suspender want to do some i/o requiring calls to
vn_start_write(), and this i/o cannot be done later.

Reviewed by:	tegge
In collaboration with:	pho
MFC after:	 1 month
2008-09-16 11:51:06 +00:00
Konstantin Belousov
52dfc8d7da Add the ffs structures introspection functions for ddb.
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
2008-09-16 11:19:38 +00:00
Konstantin Belousov
90446e360c When downgrading the read-write mount to read-only, do_unmount() sets
MNT_RDONLY flag before the VFS_MOUNT() is called. In ufs_inactive()
and ufs_itimes_locked(), UFS verifies whether the fs is read-only by
checking MNT_RDONLY, but this may cause loss of the IN_MODIFIED flag
for inode on the fs being remounted rw->ro.

Introduce UFS_RDONLY() struct ufsmount' method that reports the value
of the fs_ronly. The later is set to 1 only after the remount is
finished.

Reviewed by:	tegge
In collaboration with:	pho
MFC after:	 1 month
2008-09-16 10:59:35 +00:00
Konstantin Belousov
0411d79138 The struct inode *ip supplied to softdep_freefile is not neccessary the
inode having number ino. In r170991, the ip was marked IN_MODIFIED, that
is not quite correct.

Mark only the right inode modified by checking inode number.

Reviewed by:	tegge
In collaboration with:	pho
MFC after:	 1 month
2008-09-16 10:52:25 +00:00
Edward Tomasz Napierala
86a0c0aa7b When calling extattr_check_cred, use V{READ,WRITE}, not I{READ,WRITE}.
Approved by:	rwatson (mentor)
2008-09-03 12:46:09 +00:00
Attilio Rao
59d4932531 Decontextualize vfs_busy(), vfs_unbusy() and vfs_mount_alloc() functions.
Manpages are updated accordingly.

Tested by:	Diego Sardina <siarodx at gmail dot com>
2008-08-31 14:26:08 +00:00
Attilio Rao
0359a12ead Decontextualize the couplet VOP_GETATTR / VOP_SETATTR as the passed thread
was always curthread and totally unuseful.

Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
2008-08-28 15:23:18 +00:00
Konstantin Belousov
acd05e468a In ffs_valloc(), ffs_vget() may fail because insmntque() refused to
insert new vnode into the mount vnode list. Then, for the SU-enabled
mount, ffs_vfree could create freefile dependency. This dependency can
hang around forever since inode is not marked as IN_MODIFIED and
correspondingly inodeblock may be not marked as dirty.

After ffs_vget() fails, retry with FFSV_FORCEINSMQ, mark the inode as
modified, and vput() it immediately. Take care of the dup alloc.

Tested by:	pho
Reviewed by:	tegge
MFC after:	1 month
2008-08-28 09:19:50 +00:00
Konstantin Belousov
7b7ed832e4 Softdep code may need to instantiate vnode when processing
dependencies. In particular, it may need this while syncing filesystem
being unmounted. Since during unmount MNTK_NOINSMNTQUE flag is set,
that could sometimes disallow insertion of the vnode into the vnode
mount list, softdep code needs to overwrite the MNTK_NOINSMNTQUE flag.

Create the ffs_vgetf() function that sets the VV_FORCEINSMQ flag for
new vnode and use it consistently from the softdep code instead of
ffs_vget().

Add the retry logic to the softdep_flushfiles() to flush the vnodes
that could be instantiated while flushing softdep dependencies.

Tested by:	pho, kris
Reviewed by:	tegge
MFC after:	1 month
2008-08-28 09:18:20 +00:00
Konstantin Belousov
f2228325de Put the relocked variable from the r182111 into the #ifdef QUOTA braces
to prevent warning about unused var on the !QUOTA kernels.

Reported by:	ed
MFC after:	1 week
2008-08-24 19:06:19 +00:00
Konstantin Belousov
689eae1d90 Revert the r167541: "Remove unneeded getinoquota() call in the
ufs_access()." The call to getinoquota in ufs_access() serves the
purpose of instantiating inode dquot from the vn_open(). Since quotas
are accounted only for the inodes with already attached dquot, removal
of the call prevented opened inodes from participation in the quota
calculations.

Since ufs_access() may be called with the vnode being only shared
locked, upgrade (and then downgrade) vnode lock if calling
getinoquota().

Reported by:	simon at optinet com
In collaboration with:	pho
MFC after:	1 week
2008-08-24 17:24:22 +00:00
Konstantin Belousov
e792b09be2 Revert r181345.
Move the NULL pointer check to the vfs_deleteopt() function.

Discussed with:	rodrigc
MFC after:	3 days
2008-08-10 12:15:36 +00:00
Konstantin Belousov
a1a917e029 User may do "mount -o snapshot ...", that causes new FFS mount to be
performed with snapshot option, while the mp->mnt_opt is NULL.
Protect against NULL pointer dereference.

Noted by:	Mateusz Guzik <mjguzik gmail com>
MFC after:	3 days
2008-08-06 14:47:19 +00:00
Dag-Erling Smørgrav
20ed1beeb5 ufsmount.h uses "struct\tfoo *bar;", except where it doesn't.
quota.h uses "struct foo\t*bar;", except where it doesn't.
Try to make them both agree with themselves (though not with eachother)
2008-08-05 15:24:07 +00:00
Dag-Erling Smørgrav
1ac541a69a Whitespace, prototypes 2008-08-05 10:25:55 +00:00
John Baldwin
4a67a0d994 Whitespace tweak. 2008-07-30 21:07:56 +00:00
Konstantin Belousov
89672c6337 The ffs_balloc_ufs{1,2} functions call bdwrite() while having several
vnode buffers locked at once. In particular, there are indirect buffers
among locked ones. The bdwrite() may start the flushing to keep dirty
buffer list at the bounds. If any buffer on the dirty list requires
translation from logical to physical block number, code may ends up
trying to lock an indirect buffer already locked in ffs_balloc_ufsX.

Prevent the bdflush() activity when several buffers are locked at once
by setting the TDP_INBDFUSH for the problematic code blocks.

Reported and tested by:	pho, Josef Buchsteiner at Juniper
In collaboration with:	kan
MFC after:	1 month
2008-07-23 14:32:44 +00:00
Pawel Jakub Dawidek
a80d8caa74 Say hi to svn, by simplifing ffs_vget() function a bit - there is no need for
a variable that is used only once.
2008-07-19 22:29:44 +00:00
Craig Rodrigues
8e7a2353ec Fix comments to replace SBSIZE with SBLOCKSIZE, since SBSIZE
was renamed to SBLOCKSIZE in version 1.33

Reviewed by:	mckusick
2008-05-24 20:44:14 +00:00
Craig Rodrigues
fb77e0af12 After converting the "snapshot" mount option to the MNT_SNAPSHOT flag,
delete "snapshot" from the persistent mount options list.
This should fix problems with doing a mount -o snapshot of a file system, followed by
an NFS export of the same file system.

PR:		122833
Reported by:	Leon Kos <leon.kos lecad fs uni-lj si>,
		Jaakko Heinonen <jh saunalahti fi>
MFC after:	1 month
2008-05-24 00:41:32 +00:00
Craig Rodrigues
02a871f1ea For the following mount options, do not perform the string to flag conversions
here, because we already do them further up in vfs_donmount() in vfs_mount.c

async -> MNT_ASYNC
force -> MNT_FORCE
multilabel -> MNT_MULTILABEL
noatime -> MNT_NOATIME
noclusterr -> MNT_NOCLUSTERR
noclusterw -> MNT_NOCLUSTERW

MFC after:  1 month
2008-05-24 00:02:12 +00:00
Stephan Uphoff
2ac78f0e1a Allow VM object creation in ufs_lookup. (If vfs.vmiodirenable is set)
Directory IO without a VM object will store data in 'malloced' buffers
severely limiting caching of the data. Without this  change VM objects for
directories are only created on an open() of the directory.
TODO: Inline test if VM object already exists to avoid locking/function call
overhead.

Tested by: kris@
Reviewed by: jeff@
Reported by: David Filo
2008-05-20 19:05:43 +00:00
Jeff Roberson
721cc5664f - Use a local variable for i_ino in ufs_lookup. It is only used to
communicate between two parts of this one function.  This was causing
   problems with shared lookups as each would trash the ino value in the
   inode.
 - Remove the unused i_ino field from the inode structure.
2008-04-22 12:34:16 +00:00
Konstantin Belousov
eab626f110 Move the head of byte-level advisory lock list from the
filesystem-specific vnode data to the struct vnode. Provide the
default implementation for the vop_advlock and vop_advlockasync.
Purge the locks on the vnode reclaim by using the lf_purgelocks().
The default implementation is augmented for the nfs and smbfs.
In the nfs_advlock, push the Giant inside the nfs_dolock.

Before the change, the vop_advlock and vop_advlockasync have taken the
unlocked vnode and dereferenced the fs-private inode data, racing with
with the vnode reclamation due to forced unmount. Now, the vop_getattr
under the shared vnode lock is used to obtain the inode size, and
later, in the lf_advlockasync, after locking the vnode interlock, the
VI_DOOMED flag is checked to prevent an operation on the doomed vnode.

The implementation of the lf_purgelocks() is submitted by dfr.

Reported by:	kris
Tested by:	kris, pho
Discussed with:	jeff, dfr
MFC after:	2 weeks
2008-04-16 11:33:32 +00:00
Jeff Roberson
b300d706ea - Use a lockmgr lock rather than a mtx to protect dirhash. This lock
may be held for the duration of the various dirhash operations which
   avoids many complex unlock/lock/revalidate sequences.
 - Permit shared locks on lookup.  To protect the ip->i_dirhash pointer we
   use the vnode interlock in the shared case.  Callers holding the
   exclusive vnode lock can run without fear of concurrent modification to
   i_dirhash.
 - Hold an exclusive dirhash lock when creating the dirhash structure for
   the first time or when re-creating a dirhash structure which has been
   recycled.

Tested by:	kris, pho
2008-04-11 09:48:12 +00:00
Jeff Roberson
eb1314a249 - cache dp->i_offset in the local 'i_offset' variable for use in loop
indexes so directory lookup becomes shared lock safe.  In the modifying
   cases an exclusive lock is held here so the commit routine may
   rely on the state of i_offset.
 - Similarly handle i_diroff by fetching at the start and setting only once
   the operation is complete.  Without the exclusive lock these are only
   considered hints.
 - Assert that an exclusive lock is held when we're preparing for a commit
   routine.
 - Honor the lock type request from lookup instead of always using exclusive
   locking.

Tested by:	pho, kris
2008-04-11 09:44:25 +00:00