We no longer allow upper filesystems to be unregistered from the base
mount while vfs_notify_upper() or any other upper operation is pending.
New upper mounts can still be registered during this period, but they
will be added at the end of the upper mount tailq. We therefore no
longer need to allocate marker nodes during vfs_notify_upper() to keep
our place in the iteration.
Reviewed by: kib, mckusick
Tested by: pho
Differential Revision: https://reviews.freebsd.org/D31016
In certain emergency cases such as media failure or removal, UFS will
initiate a forced unmount in order to prevent dirty buffers from
accumulating against the no-longer-usable filesystem. The presence
of a stacked filesystem such as nullfs or unionfs above the UFS mount
will prevent this forced unmount from succeeding.
This change addreses the situation by allowing stacked filesystems to
be recursively unmounted on a taskqueue thread when the MNT_RECURSE
flag is specified to dounmount(). This call will block until all upper
mounts have been removed unless the caller specifies the MNT_DEFERRED
flag to indicate the base filesystem should also be unmounted from the
taskqueue.
To achieve this, the recently-added vfs_pin_from_vp()/vfs_unpin() KPIs
have been combined with the existing 'mnt_uppers' list used by nullfs
and renamed to vfs_register_upper_from_vp()/vfs_unregister_upper().
The format of the mnt_uppers list has also been changed to accommodate
filesystems such as unionfs in which a given mount may be stacked atop
more than one lower mount. Additionally, management of lower FS
reclaim/unlink notifications has been split into a separate list
managed by a separate set of KPIs, as registration of an upper FS no
longer implies interest in these notifications.
Reviewed by: kib, mckusick
Tested by: pho
Differential Revision: https://reviews.freebsd.org/D31016
This is aimed at preventing stacked filesystems like nullfs and unionfs
from "losing" their lower mounts due to forced unmount. Otherwise,
VFS operations that are passed through to the lower filesystem(s) may
crash or otherwise cause unpredictable behavior.
Introduce two new functions: vfs_pin_from_vp() and vfs_unpin().
which are intended to be called on the lower mount(s) when the stacked
filesystem is mounted and unmounted, respectively.
Much as registration in the mnt_uppers list previously did, pinning
will prevent even forced unmount of the lower FS and will allow the
stacked FS to freely operate on the lower mount either by direct
use of the struct mount* or indirect use through a properly-referenced
vnode's v_mount field.
vfs_pin_from_vp() is modeled after vfs_ref_from_vp() in that it uses
the mount interlock coupled with re-checking vp->v_mount to ensure
that it will fail in the face of a pending unmount request, even if
the concurrent unmount fully completes.
Adopt these new functions in both nullfs and unionfs.
Reviewed By: kib, markj
Differential Revision: https://reviews.freebsd.org/D30401
Return EBUSY instead and let caller to handle the issue.
For vgone()/vnode reclamation, caller first does vinvalbuf(V_SAVE),
which return EBUSY in case dirty buffers where not flushed. Then caller
calls vinvalbuf(0) due to non-zero return, which gets rid of all dirty
buffers without dependencies.
PR: 238565
Reviewed by: asomers, mckusick
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Differential revision: https://reviews.freebsd.org/D30555
There is no need to own vnode interlock, since v_object is type stable
and can only change to/from NULL, and no other checks in the function
access fields protected by the interlock. Remove the need variable, the
result of the test is directly usable as return value.
Tested by: mav, pho
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
The original filesystem release (4.2BSD) had no embedded sysmlinks.
Historically symbolic links were just a different type of file, so
the content of the symbolic link was contained in a single disk block
fragment. We observed that most symbolic links were short enough that
they could fit in the area of the inode that normally holds the block
pointers. So we created embedded symlinks where the content of the
link was held in the inode's pointer area thus avoiding the need to
seek and read a data fragment and reducing the pressure on the block
cache. At the time we had only UFS1 with 32-bit block pointers,
so the test for a fastlink was:
di_size < (NDADDR + NIADDR) * sizeof(daddr_t)
(where daddr_t would be ufs1_daddr_t today).
When embedded symlinks were added, a spare field in the superblock
with a known zero value became fs_maxsymlinklen. New filesystems
set this field to (NDADDR + NIADDR) * sizeof(daddr_t). Embedded
symlinks were assumed when di_size < fs->fs_maxsymlinklen. Thus
filesystems that preceeded this change always read from blocks
(since fs->fs_maxsymlinklen == 0) and newer ones used embedded
symlinks if they fit. Similarly symlinks created on pre-embedded
symlink filesystems always spill into blocks while newer ones will
embed if they fit.
At the same time that the embedded symbolic links were added, the
on-disk directory structure was changed splitting the former
u_int16_t d_namlen into u_int8_t d_type and u_int8_t d_namlen.
Thus fs_maxsymlinklen <= 0 (as used by the OFSFMT() macro) can
be used to distinguish old directory formats. In retrospect that
should have just been an added flag, but we did not realize we
needed to know about that change until it was already in production.
Code was split into ufs/ffs so that the log structured filesystem could
use ufs functionality while doing its own disk layout. This meant
that no ffs superblock fields could be used in the ufs code. Thus
ffs superblock fields that were needed in ufs code had to be copied
to fields in the mount structure. Since ufs_readlink needed to know
if a link was embedded, fs_maxlinklen gets copied to mnt_maxsymlinklen.
The kernel panic that arose to making this fix was triggered when a
disk error created an inode of type symlink with no allocated data
blocks but a large size. When readlink was called the uiomove was
attempted which segment faulted.
static int
ufs_readlink(ap)
struct vop_readlink_args /* {
struct vnode *a_vp;
struct uio *a_uio;
struct ucred *a_cred;
} */ *ap;
{
struct vnode *vp = ap->a_vp;
struct inode *ip = VTOI(vp);
doff_t isize;
isize = ip->i_size;
if ((isize < vp->v_mount->mnt_maxsymlinklen) ||
DIP(ip, i_blocks) == 0) { /* XXX - for old fastlink support */
return (uiomove(SHORTLINK(ip), isize, ap->a_uio));
}
return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
}
The second part of the "if" statement that adds
DIP(ip, i_blocks) == 0) { /* XXX - for old fastlink support */
is problematic. It never appeared in BSD released by Berkeley because
as noted above mnt_maxsymlinklen is 0 for old format filesystems, so
will always fall through to the VOP_READ as it should. I had to dig
back through `git blame' to find that Rodney Grimes added it as
part of ``The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch.''
He must have brought it across from an earlier FreeBSD. Unfortunately
the source-control logs for FreeBSD up to the merger with the
AT&T-blessed 4.4BSD-Lite conversion were destroyed as part of the
agreement to let FreeBSD remain unencumbered, so I cannot pin-point
where that line got added on the FreeBSD side.
The one change needed here is that mnt_maxsymlinklen is declared as
an `int' and should be changed to be `u_int64_t'.
This discovery led us to check out the code that deletes symbolic
links. Specifically
if (vp->v_type == VLNK &&
(ip->i_size < vp->v_mount->mnt_maxsymlinklen ||
datablocks == 0)) {
if (length != 0)
panic("ffs_truncate: partial truncate of symlink");
bzero(SHORTLINK(ip), (u_int)ip->i_size);
ip->i_size = 0;
DIP_SET(ip, i_size, 0);
UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE | IN_UPDATE);
if (needextclean)
goto extclean;
return (ffs_update(vp, waitforupdate));
}
Here too our broken symlink inode with no data blocks allocated
and a large size will segment fault as we are incorrectly using the
test that we have no data blocks to decide that it is an embdedded
symbolic link and attempting to bzero past the end of the inode.
The test for datablocks == 0 is unnecessary as the test for
ip->i_size < vp->v_mount->mnt_maxsymlinklen will do the right
thing in all cases.
The test for datablocks == 0 was added by David Greenman in this commit:
Author: David Greenman <dg@FreeBSD.org>
Date: Tue Aug 2 13:51:05 1994 +0000
Completed (hopefully) the kernel support for old style "fastlinks".
Notes:
svn path=/head/; revision=1821
I am guessing that he likely earlier added the incorrect test in the
ufs_readlink code.
I asked David if he had any recollection of why he made this change.
Amazingly, he still had a recollection of why he had made a one-line
change more than twenty years ago. And unsurpisingly it was because
he had been stuck between a rock and a hard place.
FreeBSD was up to 1.1.5 before the switch to the 4.4BSD-Lite code
base. Prior to that, there were three years of development in all
areas of the kernel, including the filesystem code, from the combined
set of people including Bill Jolitz, Patchkit contributors, and
FreeBSD Project members. The compatibility issue at hand was caused
by the FASTLINKS patches from Curt Mayer. In merging in the 4.4BSD-Lite
changes David had to find a way to provide compatibility with both
the changes that had been made in FreeBSD 1.1.5 and with 4.4BSD-Lite.
He felt that these changes would provide compatibility with both systems.
In his words:
``My recollection is that the 'FASTLINKS' symlinks support in
FreeBSD-1.x, as implemented by Curt Mayer, worked differently than
4.4BSD. He used a spare field in the inode to duplicately store the
length. When the 4.4BSD-Lite merge was done, the optimized symlinks
support for existing filesystems (those that were initialized in
FreeBSD-1.x) were broken due to the FFS on-disk structure of
4.4BSD-Lite differing from FreeBSD-1.x. My commit was needed to
restore the backward compatibility with FreeBSD-1.x filesystems.
I think it was the best that could be done in the somewhat urgent
circumstances of the post Berkeley-USL settlement. Also, regarding
Rod's massive commit with little explanation, some context: John
Dyson and I did the initial re-port of the 4.4BSD-Lite kernel to
the 386 platform in just 10 days. It was by far the most intense
hacking effort of my life. In addition to the porting of tons of
FreeBSD-1 code, I think we wrote more than 30,000 lines of new code
in that time to deal with the missing pieces and architectural
changes of 4.4BSD-Lite. We didn't make many notes along the way.
There was a lot of pressure to get something out to the rest of the
developer community as fast as possible, so detailed discrete commits
didn't happen - it all came as a giant wad, which is why Rod's
commit message was worded the way it was.''
Reported by: Chuck Silvers
Tested by: Chuck Silvers
History by: David Greenman Lawrence
MFC after: 1 week
Sponsored by: Netflix
vnodes are a bit special in that they may exist on per-CPU lists even
while free. Add a KASAN-only destructor that poisons regions of each
vnode that are not expected to be accessed after a free.
MFC after: 2 weeks
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D29459
The global list has a marker with an invariant that free vnodes are
placed somewhere past that. A caller which performs filtering (like ZFS)
can move said marker all the way to the end, across free vnodes which
don't match. Then a caller which does not perform filtering will fail to
find them. This makes vn_alloc_hard sleep for 1 second instead of
reclaiming, resulting in significant stalls.
Fix the problem by requiring an explicit marker by callers which do
filtering.
As a temporary measure extend vnlru_free to restart if it fails to
reclaim anything.
Big thanks go to the reporter for testing several iterations of the
patch.
Reported by: Yamagi <lists yamagi.org>
Tested by: Yamagi <lists yamagi.org>
Reviewed by: kib
Differential Revision: https://reviews.freebsd.org/D29324
Nullfs vnode which shares vm_object and pages with the lower vnode should
not be exempt from the reclaim just because lower vnode cached a lot.
Their reclamation is actually very cheap and should be preferred over
real fs vnodes, but this change is already useful.
Reported and tested by: pho
Reviewed by: mckusick
Sponsored by: The FreeBSD Foundation
MFC after: 2 weeks
Differential revision: https://reviews.freebsd.org/D29178
Respect filter-specific flags for the EVFILT_FS filter.
When a kevent is registered with the EVFILT_FS filter, it is always
triggered when an EVFILT_FS event occurs, regardless of the
filter-specific flags used. Fix that.
Reviewed by: kib
Differential Revision: https://reviews.freebsd.org/D28974
The data is only needed by filesystems that
1. use buffer cache
2. utilize clustering write support.
Requested by: mjg
Reviewed by: asomers (previous version), fsu (ext2 parts), mckusick
Tested by: pho
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D28679
When possible, relock the vnode and retry inactivation. Only vunref() is
required not to drop the vnode lock, so handle it specially by not retrying.
This is a part of the efforts to ensure that unlinked not referenced vnode
does not prevent inode from reusing.
Reviewed by: chs, mckusick
Tested by: pho
MFC after: 2 weeks
Sponsored by: The FreeBSD Foundation
They were only modified to accomodate a redundant assertion.
This runs into problems as lockless lookup can still try to use the vnode
and crash instead of getting an error.
The bug was only present in kernels with INVARIANTS.
Reported by: kevans
- Mask out recently added VV_* bits to avoid printing them twice.
- Keep VI_LOCKed on the same line as the rest of the flags.
Reviewed by: kib
Obtained from: CheriBSD
Sponsored by: DARPA
Differential Revision: https://reviews.freebsd.org/D27261
Restart syscalls and some sync operations when filesystem indicated
ERELOOKUP condition, mostly for VOPs operating on metdata. In
particular, lookup results cached in the inode/v_data is no longer
valid and needs recalculating. Right now this should be nop.
Assert that ERELOOKUP is catched everywhere and not returned to
userspace, by asserting that td_errno != ERELOOKUP on syscall return
path.
In collaboration with: pho
Reviewed by: mckusick (previous version), markj
Tested by: markj (syzkaller), pho
Sponsored by: The FreeBSD Foundation
Differential revision: https://reviews.freebsd.org/D26136
All vnodes allocated by UMA are present on the global list used by
vnlru. getnewvnode modifies the state of the vnode (most notably
altering v_holdcnt) but never locks it. Moreover filesystems also
modify it in arbitrary manners sometimes before taking the vnode
lock or adding any other indicator that the vnode can be used.
Picking up such a vnode by vnlru would be problematic.
To that end there are 2 fixes:
- vlrureclaim, not recycling v_holdcnt == 0 vnodes, takes the
interlock and verifies that v_mount has been set. It is an
invariant that the vnode lock is held by that point, providing
the necessary serialisation against locking after vhold.
- vnlru_free_locked, only wanting to free v_holdcnt == 0 vnodes,
now makes sure to only transition the count 0->1 and newly allocated
vnodes start with v_holdcnt == VHOLD_NO_SMR. getnewvnode will only
transition VHOLD_NO_SMR->1 once more making the hold fail
Tested by: pho
It no longer protects any of tested fields, keeping all the checks racy.
While here make vtryrecycle drop the vnode on its own. Avoids an additional
lock trip.
mkdir -p /foo/bar/baz will mkdir each path component and ignore EEXIST.
The NOCACHE lookup will make the namecache unnecessarily evict the existing entry,
and then fallback to the fs lookup routine eventually leading namei to return an
error as the directory is already there.
For invocations like mkdir -p /usr/obj/usr/src/sys/GENERIC/modules this triggers
fallbacks to the slowpath for concurrently executing lookups.
Tested by: pho
Discussed with: kib
It is a common pattern for filesystems' VOP_INACTIVE() implementation
to forcibly reclaim the vnode when its state is final. For instance,
UFS vnode with zero link count is removed, and since it is
inactivated, the last open reference on it is dropped.
On the other hand, vnode might get spurious usecount reference for
many reasons. If the spurious reference exists while vgonel() checks
for active state of the vnode, it would recurse into VOP_INACTIVE().
Fix it by checking and not doing inactivation when vgone() was called
from inactive VOP.
Reported and tested by: pho
Discussed with: mjg
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
There are several negative side-effects of not calling into VOP layer
at all for page cache reads. The biggest is the missed activation of
EVFILT_READ knotes.
Also, it allows filesystem to make more fine grained decision to
refuse read from page cache.
Keep VIRF_PGREAD flag around, it is still useful for nullfs, and for
asserts.
Reviewed by: markj
Tested by: pho
Discussed with: mjg
Sponsored by: The FreeBSD Foundation
Differential revision: https://reviews.freebsd.org/D26346
If possible, i.e. if the requested range is resident valid in the vm
object queue, and some secondary conditions hold, copy data for
read(2) directly from the valid cached pages, avoiding vnode lock and
instantiating buffers. I intentionally do not start read-ahead, nor
handle the advises on the cached range.
Filesystems indicate support for VMIO reads by setting VIRF_PGREAD
flag, which must not be cleared until vnode reclamation.
Currently only filesystems that use vnode pager for v_objects can
enable it, due to reliance on vnp_size. There is a WIP to handle it
for tmpfs.
Reviewed by: markj
Discussed with: jeff
Tested by: pho
Benchmarked by: mjg
Sponsored by: The FreeBSD Foundation
Differential revision: https://reviews.freebsd.org/D25968
