deletion is active, specifically after a call to ffs_blkrelease_start()
but before the call to ffs_blkrelease_finish(), ffs_blkrelease_start()
will have handed out SINGLETON_KEY rather than starting a collection
sequence. Thus if we get a SINGLETON_KEY passed to ffs_blkrelease_finish(),
we just return rather than trying to finish the nonexistent sequence.
Reported by: Warner Losh (imp@)
Sponsored by: Netflix
superblock has a check-hash error, an error message noting the
superblock check-hash failure is printed and the mount fails. The
administrator then runs fsck to repair the filesystem and when
successful, the filesystem can once again be mounted.
This approach fails if the filesystem in question is a root filesystem
from which you are trying to boot. Here, the loader fails when trying
to access the filesystem to get the kernel to boot. So it is necessary
to allow the loader to ignore the superblock check-hash error and make
a best effort to read the kernel. The filesystem may be suffiently
corrupted that the read attempt fails, but there is no harm in trying
since the loader makes no attempt to write to the filesystem.
Once the kernel is loaded and starts to run, it attempts to mount its
root filesystem. Once again, failure means that it breaks to its prompt
to ask where to get its root filesystem. Unless you have an alternate
root filesystem, you are stuck.
Since the root filesystem is initially mounted read-only, it is
safe to make an attempt to mount the root filesystem with the failed
superblock check-hash. Thus, when asked to mount a root filesystem
with a failed superblock check-hash, the kernel prints a warning
message that the root filesystem superblock check-hash needs repair,
but notes that it is ignoring the error and proceeding. It does
mark the filesystem as needing an fsck which prevents it from being
enabled for writing until fsck has been run on it. The net effect
is that the reboot fails to single user, but at least at that point
the administrator has the tools at hand to fix the problem.
Reported by: Rick Macklem (rmacklem@)
Discussed with: Warner Losh (imp@)
Sponsored by: Netflix
predates metadata check hashes so that it is done before deciding
whether to compute a check-hash of the superblock.
Reported by: Rick Macklem <rmacklem@uoguelph.ca>
Sponsored by: Netflix
This corrects a bug that prevented snapshots from being mounted due to a
superblock check-hash failure.
Reported by: Brennan Vincent <brennan@umanwizard.com>
Tested by: Peter Holm (pho@)
Sponsored by: Netflix
Directory entries must be padded to maintain alignment; in many
filesystems the padding was not initialized, resulting in stack
memory being copied out to userspace. With the ino64 work there
are also some explicit pad fields in struct dirent. Add a subroutine
to clear these bytes and use it in the in-tree filesystems. The
NFS client is omitted for now as it was fixed separately in r340787.
Reported by: Thomas Barabosch, Fraunhofer FKIE
Reviewed by: kib
MFC after: 3 days
Sponsored by: The FreeBSD Foundation
The d_off field has been added to the dirent structure recently.
Currently filesystems don't support this feature. Support has been
added and tested for zfs, ufs, ext2fs, fdescfs, msdosfs and unionfs.
A stub implementation is available for cd9660, nandfs, udf and
pseudofs but hasn't been tested.
Motivation for this feature: our usecase is for a userspace nfs server
(nfs-ganesha) with zfs. At the moment we cache direntry offsets by
calling lseek once per entry, with this patch we can get the offset
directly from getdirentries(2) calls which provides a significant
speedup.
Submitted by: Jack Halford <jack@gandi.net>
Reviewed by: mckusick, pfg, rmacklem (previous versions)
Sponsored by: Gandi.net
MFC after: 1 week
Differential revision: https://reviews.freebsd.org/D17917
document the libufs interface for fetching and storing inodes.
The undocumented getino / putino interface has been replaced
with a new getinode / putinode interface.
Convert the utilities that had been using the undocumented
interface to use the new documented interface.
No functional change (as for now the libufs library does not
do inode check-hashes).
Reviewed by: kib
Tested by: Peter Holm
Sponsored by: Netflix
This will enable callers to take const paths as part of syscall
decleration improvements.
Where doing so is easy and non-distruptive carry the const through
implementations. In UFS the value is passed to an interface that must
take non-const values. In ZFS, const poisoning would touch code shared
with upstream and it's not worth adding diffs.
Bump __FreeBSD_version for external API consumers.
Reviewed by: kib (prior version)
Obtained from: CheriBSD
Sponsored by: DARPA, AFRL
Differential Revision: https://reviews.freebsd.org/D17805
Flags prevent open(2) and *at(2) vfs syscalls name lookup from
escaping the starting directory. Supposedly the interface is similar
to the same proposed Linux flags.
Reviewed by: jilles (code, previous version of manpages), 0mp (manpages)
Discussed with: allanjude, emaste, jonathan
Sponsored by: The FreeBSD Foundation
Differential revision: https://reviews.freebsd.org/D17547
check hash to the superblock. If a check hash fails when an attempt
is made to mount a filesystem, the mount fails with EINVAL (Invalid
argument). This avoids a class of filesystem panics related to
corrupted superblocks. The hash is done using crc32c.
Check hases are added only to UFS2 and not to UFS1 as UFS1 is primarily
used in embedded systems with small memories and low-powered processors
which need as light-weight a filesystem as possible.
Reviewed by: kib
Tested by: Peter Holm
Sponsored by: Netflix
UFS quotaoff iterates over all mp vnodes, and derefences and clears
the pointers to corresponding dquots. If SU work items transiently
reference some of dquots,quotaoff() would eventually fail, but all
processed vnodes are already stripped from dquots. The state is
problematic, since quotas are left enabled, but there is no dquots
where blocks and inodes can be accounted. The result is assertion
failures and NULL pointer dereferences.
Fix it by suspending writes around quotaoff() call. Since the
filesystem is synced, no dandling references to dquots from SU
workitems can left behind, which means that quotaoff succeeds.
The complication there is that quotaoff VFS op is performed with the
mount point busied, while to suspend, we need to start write on the
mp. If vn_start_write() is called on busied mp, system might deadlock
against parallel unmount request. Handle this by unbusy-ing mp before
starting write, which in turn requires changing the quotaoff()
interface to return with the mount point not busied, same as was done
for quotaon().
Reviewed by: mckusick
Reported and tested by: pho
Sponsored by: The FreeBSD Foundation
Approved by: re (gjb)
MFC after: 1 week
Differential revision: https://reviews.freebsd.org/D17208
Doing so can deadlock when the thread already owns another vnode lock,
e.g. during a rename, as was demonstrated by the reporter. In fact,
there seems to be no need to force the call to getinoquota() always,
because vn_open() locks vnode exclusively, and this is the most
important case. To add to the point, directories where the dirent is
added or removed, are locked exclusively as well.
Reported by: bwidawsk
Tested by: bwidawsk, pho (as part of the larger patch)
Sponsored by: The FreeBSD Foundation
Approved by: re (gjb)
MFC after: 1 week
found that performance was no worse and usually better when running
with TRIM consolidation. Performance improvement was most noticable
when multiple large files are released in a short period of time.
Thus, TRIM consolidation is being enabled by default. Should
operational problems be found, it can be disabled using the command
`sysctl vfs.ffs.dotrimcons=0'. This variable can also be set as a
tunable if early disabling is necessary.
Approved by: re (gjb)
Sponsored by: Netflix
given in random(4).
This includes updating of the relevant man pages, and no-longer-used
harvesting parameters.
Ensure that the pseudo-unit-test still does something useful, now also
with the "other" algorithm instead of Yarrow.
PR: 230870
Reviewed by: cem
Approved by: so(delphij,gtetlow)
Approved by: re(marius)
Differential Revision: https://reviews.freebsd.org/D16898
blocks of a file as contiguously as possible. Since the filesystem
does not know how large a file will grow when it is first being
written, it initially places the file in a set of blocks in which
it currently fits. As it grows, it is relocated to areas with
larger contiguous blocks. In this way it saves its large contiguous
sets of blocks for the files that need them and thus avoids
unnecessaily fragmenting its disk space.
We used to skip reallocating the blocks of a file into a contiguous
sequence if the underlying flash device requested BIO_DELETE
notifications, because devices that benefit from BIO_DELETE also
benefit from not moving the data. However, in the algorithm described
above that reallocates the blocks, the destination for the data is
usually moved before the data is written to the initially allocated
location. So we rarely suffer the penalty of extra writes. With
the addition of the consolodation of contiguous blocks into single
BIO_DELETE operations, having fewer but larger contiguous blocks
reduces the number of (slow and expensive) BIO_DELETE operations.
So when doing BIO_DELETE consolodation, we do block reallocation.
Reviewed by: kib
Tested by: Peter Holm
Sponsored by: Netflix
When deleting files on filesystems that are stored on flash-memory
(solid-state) disk drives, the filesystem notifies the underlying
disk of the blocks that it is no longer using. The notification
allows the drive to avoid saving these blocks when it needs to
flash (zero out) one of its flash pages. These notifications of
no-longer-being-used blocks are referred to as TRIM notifications.
In FreeBSD these TRIM notifications are sent from the filesystem
to the drive using the BIO_DELETE command.
Until now, the filesystem would send a separate message to the drive
for each block of the file that was deleted. Each Gigabyte of file
size resulted in over 3000 TRIM messages being sent to the drive.
This burst of messages can overwhelm the drive's task queue causing
multiple second delays for read and write requests.
This implementation collects runs of contiguous blocks in the file
and then consolodates them into a single BIO_DELETE command to the
drive. The BIO_DELETE command describes the run of blocks as a
single large block being deleted. Each Gigabyte of file size can
result in as few as two BIO_DELETE commands and is typically less
than ten. Though these larger BIO_DELETE commands take longer to
run, they do not clog the drive task queue, so read and write
commands can intersperse effectively with them.
Though this new feature has been throughly reviewed and tested, it
is being added disabled by default so as to minimize the possibility
of disrupting the upcoming 12.0 release. It can be enabled by running
``sysctl vfs.ffs.dotrimcons=1''. Users are encouraged to test it.
If no problems arise, we will consider requesting that it be enabled
by default for 12.0.
Reviewed by: kib
Tested by: Peter Holm
Sponsored by: Netflix
driven by problems found with the algorithms being tested for TRIM
consolodation.
Reported by: Peter Holm
Suggested by: kib
Reviewed by: kib
Sponsored by: Netflix
jails since FreeBSD 7.
Along with the system call, put the various security.jail.allow_foo and
security.jail.foo_allowed sysctls partly under COMPAT_FREEBSD11 (or
BURN_BRIDGES). These sysctls had two disparate uses: on the system side,
they were global permissions for jails created via jail(2) which lacked
fine-grained permission controls; inside a jail, they're read-only
descriptions of what the current jail is allowed to do. The first use
is obsolete along with jail(2), but keep them for the second-read-only use.
Differential Revision: D14791
a smaller number of larger TRIM requests. The hope had been to have
the full TRIM consolodation in place for 12.0, but the algorithms
are still under development and need further testing. With this
framework in place it will be possible to easily add TRIM consolodation
once the optimal strategy has been found.
The only functional change with this patch is the elimination of TRIM
requests for blocks that are freed before they have been likely to
have been written.
Reviewed by: kib
Discussed with: Warner Losh and Chuck Silvers
Sponsored by: Netflix
(incorrectly) called while another suspension is already active.
PR: 230220
Reported by: dexuan
Tested by: pho
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
The timespecadd(3) family of macros were imported from NetBSD back in
r35029. However, they were initially guarded by #ifdef _KERNEL. In the
meantime, we have grown at least 28 syscalls that use timespecs in some
way, leading many programs both inside and outside of the base system to
redefine those macros. It's better just to make the definitions public.
Our kernel currently defines two-argument versions of timespecadd and
timespecsub. NetBSD, OpenBSD, and FreeDesktop.org's libbsd, however, define
three-argument versions. Solaris also defines a three-argument version, but
only in its kernel. This revision changes our definition to match the
common three-argument version.
Bump _FreeBSD_version due to the breaking KPI change.
Discussed with: cem, jilles, ian, bde
Differential Revision: https://reviews.freebsd.org/D14725
It can occur during ordinary use of softupdates, or perhaps if writes to the
underlying media fail (causing bufs to be redirtied). Either way, it is not
particularly actionable.
Reviewed by: imp, kib
Sponsored by: Dell EMC Isilon
Differential Revision: https://reviews.freebsd.org/D16258
Avoid Undefined Behavior in ffs_clusteracct()
Change the type of 'bit' variable from int to unsigned int and use unsigned
values consistently.
sys/ufs/ffs/ffs_subr.c:336:10, shift exponent -1 is negative
Detected with Kernel Undefined Behavior Sanitizer.
Reported by <Harry Pantazis>
Submitted by: Pedro Giffuni
Convert integer structure flags to use um_flags:
int um_candelete; /* devvp supports TRIM */
int um_writesuspended; /* suspension in progress */
become:
#define UM_CANDELETE 0x00000001 /* devvp supports TRIM */
#define UM_WRITESUSPENDED 0x00000002 /* suspension in progress */
This is in preparation for adding other flags to indicate forcible
unmount in progress after a disk failure and possibly forcible
downgrade to read-only.
No functional change intended.
Sponsored by: Netflix
The reference counting that's done in the geom_vfs layer to prevent
delivery of requests to defunct devices only works if all requests go
through that layer. UFS was bypassing that layer for BIO_DELETE requests,
sending them to the geom_consumer directly with g_io_request. Allocate
a buf, fill it in, and call strategy on it instead.
Submitted by: Chuck Silvers
Reviewed by: scottl, imp, kirk
Sponsored by: Netflix
Differential: https://reviews.freebsd.org/D15456
- Create getblkx(9) variant of getblk(9) which can return error.
- Add GB_NOSPARSE flag for getblk()/getblkx() which requests that BMAP
was performed before the buffer is created, and EJUSTRETURN returned
in case the requested block does not exist.
- Make ffs_read() use GB_NOSPARSE to avoid instantiating buffer (and
allocating the pages for it), copying from zero_region instead.
The end result is less page allocations and buffer recycling when a
hole is read, which is important for some benchmarks.
Requested and reviewed by: jeff
Tested by: pho
Sponsored by: The FreeBSD Foundation
MFC after: 2 weeks
Differential revision: https://reviews.freebsd.org/D14917
Followup to r313780. Also prefix ext2's and nandfs's versions with
EXT2_ and NANDFS_.
Reported by: kib
Reviewed by: kib, mckusick
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D9623
to fix the memory leak that I introduced in r328426. Instead of
trying to clear up the possible memory leak in all the clients, I
ensure that it gets cleaned up in the source (e.g., ffs_sbget ensures
that memory is always freed if it returns an error).
The original change in r328426 was a bit sparse in its description.
So I am expanding on its description here (thanks cem@ and rgrimes@
for your encouragement for my longer commit messages).
In preparation for adding check hashing to superblocks, r328426 is
a refactoring of the code to get the reading/writing of the superblock
into one place. Unlike the cylinder group reading/writing which
ends up in two places (ffs_getcg/ffs_geom_strategy in the kernel
and cgget/cgput in libufs), I have the core superblock functions
just in the kernel (ffs_sbfetch/ffs_sbput in ffs_subr.c which is
already imported into utilities like fsck_ffs as well as libufs to
implement sbget/sbput). The ffs_sbfetch and ffs_sbput functions
take a function pointer to do the actual I/O for which there are
four variants:
ffs_use_bread / ffs_use_bwrite for the in-kernel filesystem
g_use_g_read_data / g_use_g_write_data for kernel geom clients
ufs_use_sa_read for the standalone code (stand/libsa/ufs.c
but not stand/libsa/ufsread.c which is size constrained)
use_pread / use_pwrite for libufs
Uses of these interfaces are in the UFS filesystem, geoms journal &
label, libsa changes, and libufs. They also permeate out into the
filesystem utilities fsck_ffs, newfs, growfs, clri, dump, quotacheck,
fsirand, fstyp, and quot. Some of these utilities should probably be
converted to directly use libufs (like dumpfs was for example), but
there does not seem to be much win in doing so.
Tested by: Peter Holm (pho@)
Coverity cannot determine that handle_written_indirdep() does not access
uninitialized 'sbp' when flags argument is zero.
So, simply move the initialization slightly sooner to silence the warning.
No functional change.
Reported by: Coverity
Sponsored by: Dell EMC Isilon
ffs_sbget() may return a superblock buffer even if it fails, so the
caller must be prepared to free it in this case. Moreover, when tasting
alternate superblock locations in a loop, ffs_sbget()'s readfunc
callback must free the previously allocated buffer.
Reported and tested by: pho
Reviewed by: kib (previous version)
Differential Revision: https://reviews.freebsd.org/D14390
folks running filesystems created on check-hash enabled kernels
(which I will call "new") on a non-check-hash enabled kernels (which
I will call "old). The idea here is to detect when a filesystem is
run on an old kernel and flag the filesystem so that when it gets
moved back to a new kernel, it will not start getting a slew of
check-hash errors.
Back when the UFS version 2 filesystem was created, it added a file
flag FS_INDEXDIRS that was to be set on any filesystem that kept
some sort of on-disk indexing for directories. The idea was precisely
to solve the issue we have today. Specifically that a newer kernel
that supported indexing would be able to tell that the filesystem
had been run on an older non-indexing kernel and that the indexes
should not be used until they had been rebuilt. Since we have never
implemented on-disk directory indicies, the FS_INDEXDIRS flag is
cleared every time any UFS version 2 filesystem ever created is
mounted for writing.
This commit repurposes the FS_INDEXDIRS flag as the FS_METACKHASH
flag. Thus, the FS_METACKHASH is definitively known to have always
been cleared. The FS_INDEXDIRS flag has been moved to a new block
of flags that will always be cleared starting with this commit
(until they get used to implement some future feature which needs
to detect that the filesystem was mounted on a kernel that predates
the new feature).
If a filesystem with check-hashes enabled is mounted on an old
kernel the FS_METACKHASH flag is cleared. When that filesystem is
mounted on a new kernel it will see that the FS_METACKHASH has been
cleared and clears all of the fs_metackhash flags. To get them
re-enabled the user must run fsck (in interactive mode without the
-y flag) which will ask for each supported check hash whether it
should be rebuilt and enabled. When fsck is run in its default preen
mode, it will just ignore the check hashes so they will remain
disabled.
The kernel has always disabled any check hash functions that it
does not support, so as more types of check hashes are added, we
will get a non-surprising result. Specifically if filesystems get
moved to kernels supporting fewer of the check hashes, those that
are not supported will be disabled. If the filesystem is moved back
to a kernel with more of the check-hashes available and fsck is run
interactively to rebuild them, then their checking will resume.
Otherwise just the smaller subset will be checked.
A side effect of this commit is that filesystems running with
cylinder-group check hashes will stop having them checked until
fsck is run to re-enable them (since none of them currently have
the FS_METACKHASH flag set). So, if you want check hashes enabled
on your filesystems after booting a kernel with these changes, you
need to run fsck to enable them. Any newly created filesystems will
have check hashes enabled. If in doubt as to whether you have check
hashes emabled, run dumpfs and look at the list of enabled flags
at the end of the superblock details.
ncookies cannot be negative or the allocator will fail. This should only
happen if a caller is very broken but we can still try to survive the
event.
We should probably also verify for uio_resid > MAXPHYS but in that case
it is not clear that just clipping the ncookies value is an adequate
response.
MFC after: 2 weeks
systems running with a heavy filesystem load. Tracking down this
bug was elusive because there were actually two problems. Sometimes
the in-memory check hash was wrong and sometimes the check hash
computed when doing the read was wrong. The occurrence of either
error caused a check-hash mismatch to be reported.
The first error was that the check hash in the in-memory cylinder
group was incorrect. This error was caused by the following
sequence of events:
- We read a cylinder-group buffer and the check hash is valid.
- We update its cg_time and cg_old_time which makes the in-memory
check-hash value invalid but we do not mark the cylinder group dirty.
- We do not make any other changes to the cylinder group, so we
never mark it dirty, thus do not write it out, and hence never
update the incorrect check hash for the in-memory buffer.
- Later, the buffer gets freed, but the page with the old incorrect
check hash is still in the VM cache.
- Later, we read the cylinder group again, and the first page with
the old check hash is still in the VM cache, but some other pages
are not, so we have to do a read.
- The read does not actually get the first page from disk, but rather
from the VM cache, resulting in the old check hash in the buffer.
- The value computed after doing the read does not match causing the
error to be printed.
The fix for this problem is to only set cg_time and cg_old_time as
the cylinder group is being written to disk. This keeps the in-memory
check-hash valid unless the cylinder group has had other modifications
which will require it to be written with a new check hash calculated.
It also requires that the check hash be recalculated in the in-memory
cylinder group when it is marked clean after doing a background write.
The second problem was that the check hash computed at the end of the
read was incorrect because the calculation of the check hash on
completion of the read was being done too soon.
- When a read completes we had the following sequence:
- bufdone()
-- b_ckhashcalc (calculates check hash)
-- bufdone_finish()
--- vfs_vmio_iodone() (replaces bogus pages with the cached ones)
- When we are reading a buffer where one or more pages are already
in memory (but not all pages, or we wouldn't be doing the read),
the I/O is done with bogus_page mapped in for the pages that exist
in the VM cache. This mapping is done to avoid corrupting the
cached pages if there is any I/O overrun. The vfs_vmio_iodone()
function is responsible for replacing the bogus_page(s) with the
cached ones. But we were calculating the check hash before the
bogus_page(s) were replaced. Hence, when we were calculating the
check hash, we were partly reading from bogus_page, which means
we calculated a bad check hash (e.g., because multiple pages have
been mapped to bogus_page, so its contents are indeterminate).
The second fix is to move the check-hash calculation from bufdone()
to bufdone_finish() after the call to vfs_vmio_iodone() so that it
computes the check hash over the correct set of pages.
With these two changes, the occasional cylinder-group check-hash
errors are gone.
Submitted by: David Pfitzner <dpfitzner@netflix.com>
Reviewed by: kib
Tested by: David Pfitzner
Sanitize the values that will be assigned to ncookies so that we ensure
they are sane and we can handle them.
Let ncookies signed as it was before r328346. The valid range is such
that unsigned values are not required and we are not able to avoid at
least one cast anyways.
Hinted by: bde
Specifically reading is done if ffs_sbget() and writing is done
in ffs_sbput(). These functions are exported to libufs via the
sbget() and sbput() functions which then used in the various
filesystem utilities. This work is in preparation for adding
subperblock check hashes.
No functional change intended.
Reviewed by: kib
