]> git.proxmox.com Git - mirror_zfs.git/log
mirror_zfs.git
2 weeks agoetc/init.d: decide which variant to use at build time. master
Benda Xu [Mon, 8 Apr 2024 23:52:24 +0000 (07:52 +0800)]
etc/init.d: decide which variant to use at build time.

Let Debian use the sysv-rc variant of the script, even when OpenRC is
installed. Unlike on Gentoo, OpenRC on Debian consumes both the
sysv-rc scripts and OpenRC ones. ZFS initscripts on Debian should be
the sysv-rc version to provide most compatibility and to integrate
with the rest of initscripts for dependency tracking.

Restrict the substitution in the Makefile to the dedicated list.

This construct is inspired by Mo Zhou's detection of the execution
shell and follows the strategy of Peter in 6ef28c526ba7.

As of 2024, the initscripts are mostly relevant on Debian, Gentoo and
their derivatives.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Benda Xu <orv@debian.org>
Issue #8063
Issue #8204
Issue #8359
Closes #15977

2 weeks agoFix locale-specific time
Maxim Filimonov [Mon, 8 Apr 2024 22:37:41 +0000 (02:37 +0400)]
Fix locale-specific time

In `zpool status -t`, scrub date/time is reported using the C locale,
while trim time is reported using the current one. This is inconsistent.
This patch fixes that.

Reviewed-by: Tino Reichardt <milky-zfs@mcmilk.de>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Maxim Filimonov <che@bein.link>
Closes #15878
Closes #15879

2 weeks agoRemove db_state DB_NOFILL checks from syncing context
Alexander Motin [Mon, 8 Apr 2024 22:23:43 +0000 (18:23 -0400)]
Remove db_state DB_NOFILL checks from syncing context

Syncing context should not depend on current state of dbuf, which
could already change several times in later transaction groups,
but rely solely on dirty record for the transaction group being
synced. Some of the checks seem already impossible, while instead
of others I think we should better check for absence of data in
the specific dirty record rather than DB_NOFILL.

Reviewed-by: Robert Evans <evansr@google.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes #16057

2 weeks agoSpeculative prefetch for reordered requests
Alexander Motin [Mon, 8 Apr 2024 22:13:27 +0000 (18:13 -0400)]
Speculative prefetch for reordered requests

Before this change speculative prefetcher was able to detect a stream
only if all of its accesses are perfectly sequential.  It was easy to
implement and is perfectly fine for single-threaded applications.
Unfortunately multi-threaded network servers, such as iSCSI, SMB or
NFS usually have plenty of threads and may often reorder requests,
preventing successful speculation and prefetch.

This change allows speculative prefetcher to detect streams even if
requests are reordered by introducing a list of 9 non-contiguous
ranges up to 16MB ahead of current stream position and filling the
gaps as more requests arrive.  It also allows stream to proceed
even with holes up to a certain configurable threshold (25%).

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes #16022

2 weeks agoFix read errors race after block cloning
Alexander Motin [Mon, 8 Apr 2024 19:03:18 +0000 (15:03 -0400)]
Fix read errors race after block cloning

Investigating read errors triggering panic fixed in #16042 I've
found that we have a race in a sync process between the moment
dirty record for cloned block is removed and the moment dbuf is
destroyed.  If dmu_buf_hold_array_by_dnode() take a hold on a
cloned dbuf before it is synced/destroyed, then dbuf_read_impl()
may see it still in DB_NOFILL state, but without the dirty record.
Such case is not an error, but equivalent to DB_UNCACHED, since
the dbuf block pointer is already updated by dbuf_write_ready().
Unfortunately it is impossible to safely change the dbuf state
to DB_UNCACHED there, since there may already be another cloning
in progress, that dropped dbuf lock before creating a new dirty
record, protected only by the range lock.

Reviewed-by: Rob Norris <robn@despairlabs.com>
Reviewed-by: Robert Evans <evansr@google.com>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes #16052

2 weeks agozinject: inject device errors into ioctls
Rob N [Mon, 8 Apr 2024 18:59:04 +0000 (04:59 +1000)]
zinject: inject device errors into ioctls

Adds 'ioctl' as a valid IO type for device error injection, so we can
simulate a flush error (which OpenZFS currently ignores, but that's by
the by).

To support this, adding ZIO_STAGE_VDEV_IO_DONE to ZIO_IOCTL_PIPELINE,
since that's where device error injection happens. This needs a small
exclusion to avoid the vdev_queue, since flushes are not queued, and I'm
assuming that the various failure responses are still reasonable for
flush failures (probes, media change, etc). This seems reasonable to me,
as a flush failure is not unlike a write failure in this regard, however
this may be too aggressive or subtle to assume in just this change.

Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Closes #16061

2 weeks agovdev_disk: ensure trim errors are returned immediately
Rob N [Mon, 8 Apr 2024 18:50:24 +0000 (04:50 +1000)]
vdev_disk: ensure trim errors are returned immediately

After 06e25f9c4, the discard issuing code was organised such that if
requesting an async discard or secure erase failed before the IO was
issued (that is, calling __blkdev_issue_discard() returned an error),
the failed zio would never be executed, resulting in txg_sync hanging
forever waiting for IO to finish.

This commit fixes that by immediately executing a failed zio on error.
To handle the successful synchronous op case, we fake an async op by,
when not using an asynchronous submission method, queuing the successful
result zio as part of the discard handler.

Since it was hard to understand the differences between discard and
secure erase, and sync and async, across different kernel versions, I've
commented and reorganised the code a bit to try and make everything more
contained and linear.

Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Reviewed-by: Ameer Hamza <ahamza@ixsystems.com>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Closes #16070

2 weeks agozvol_os: fix compile with blk-mq on Linux 4.x
Rob N [Mon, 8 Apr 2024 18:38:49 +0000 (04:38 +1000)]
zvol_os: fix compile with blk-mq on Linux 4.x

99741bde5 accesses a cached blk-mq hardware context through the mq_hctx
field of struct request. However, this field did not exist until 5.0.
Before that, the private function blk_mq_map_queue() was used to dig it
out of broader queue context. This commit detects this situation, and
handles it with a poor-man's simulation of that function.

Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Reviewed-by: Ameer Hamza <ahamza@ixsystems.com>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Closes #16069

2 weeks agozvol_os: fix build on Linux <3.13
Rob N [Mon, 8 Apr 2024 17:13:27 +0000 (03:13 +1000)]
zvol_os: fix build on Linux <3.13

99741bde5 introduced zvol_num_taskqs, but put it behind the HAVE_BLK_MQ
define, preventing builds on versions of Linux that don't have it
(<3.13, incl EL7).

Nothing about it seems dependent on blk-mq, so this just moves it out
from behind that define and so fixes the build.

Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Reviewed-by: Ameer Hamza <ahamza@ixsystems.com>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Closes #16062

3 weeks agozvol: use multiple taskq
Ameer Hamza [Thu, 4 Apr 2024 01:21:25 +0000 (06:21 +0500)]
zvol: use multiple taskq

Currently, zvol uses a single taskq, resulting in throughput bottleneck
under heavy load due to lock contention on the single taskq. This patch
addresses the performance bottleneck under heavy load conditions by
utilizing multiple taskqs, thus mitigating lock contention. The number
of taskqs scale dynamically based on the available CPUs in the system,
as illustrated below:

                taskq   total
cpus    taskqs  threads threads
------- ------- ------- -------
1       1       32       32
2       1       32       32
4       1       32       32
8       2       16       32
16      3       11       33
32      5       7        35
64      8       8        64
128     11      12       132
256     16      16       256

Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Tony Nguyen <tony.nguyen@delphix.com>
Signed-off-by: Ameer Hamza <ahamza@ixsystems.com>
Closes #15992

3 weeks agoFix panics when truncating/deleting files
Pavel Snajdr [Thu, 4 Apr 2024 01:09:19 +0000 (03:09 +0200)]
Fix panics when truncating/deleting files

There's an union in dbuf_dirty_record_t; dr_brtwrite could evaluate
to B_TRUE if the dirty record is of another type than dl. Adding
more explicit dr type check before trying to access dr_brtwrite.

Fixes two similar panics:

[ 1373.806119] VERIFY0(db->db_level) failed (0 == 1)
[ 1373.807232] PANIC at dbuf.c:2549:dbuf_undirty()
[ 1373.814979]  dump_stack_lvl+0x71/0x90
[ 1373.815799]  spl_panic+0xd3/0x100 [spl]
[ 1373.827709]  dbuf_undirty+0x62a/0x970 [zfs]
[ 1373.829204]  dmu_buf_will_dirty_impl+0x1e9/0x5b0 [zfs]
[ 1373.831010]  dnode_free_range+0x532/0x1220 [zfs]
[ 1373.833922]  dmu_free_long_range+0x4e0/0x930 [zfs]
[ 1373.835277]  zfs_trunc+0x75/0x1e0 [zfs]
[ 1373.837958]  zfs_freesp+0x9b/0x470 [zfs]
[ 1373.847236]  zfs_setattr+0x161a/0x3500 [zfs]
[ 1373.855267]  zpl_setattr+0x125/0x320 [zfs]
[ 1373.856725]  notify_change+0x1ee/0x4a0
[ 1373.859207]  do_truncate+0x7f/0xd0
[ 1373.859968]  do_sys_ftruncate+0x28e/0x2e0
[ 1373.860962]  do_syscall_64+0x38/0x90
[ 1373.861751]  entry_SYSCALL_64_after_hwframe+0x6e/0xd8

[ 1822.381337] VERIFY0(db->db_level) failed (0 == 1)
[ 1822.382376] PANIC at dbuf.c:2549:dbuf_undirty()
[ 1822.389232]  dump_stack_lvl+0x71/0x90
[ 1822.389920]  spl_panic+0xd3/0x100 [spl]
[ 1822.399567]  dbuf_undirty+0x62a/0x970 [zfs]
[ 1822.400583]  dmu_buf_will_dirty_impl+0x1e9/0x5b0 [zfs]
[ 1822.401752]  dnode_free_range+0x532/0x1220 [zfs]
[ 1822.402841]  dmu_object_free+0x74/0x120 [zfs]
[ 1822.403869]  zfs_znode_delete+0x75/0x120 [zfs]
[ 1822.404906]  zfs_rmnode+0x3f6/0x7f0 [zfs]
[ 1822.405870]  zfs_inactive+0xa3/0x610 [zfs]
[ 1822.407803]  zpl_evict_inode+0x3e/0x90 [zfs]
[ 1822.408831]  evict+0xc1/0x1c0
[ 1822.409387]  do_unlinkat+0x147/0x300
[ 1822.410060]  __x64_sys_unlinkat+0x33/0x60
[ 1822.410802]  do_syscall_64+0x38/0x90
[ 1822.411458]  entry_SYSCALL_64_after_hwframe+0x6e/0xd8

Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Brian Atkinson <batkinson@lanl.gov>
Signed-off-by: Pavel Snajdr <snajpa@snajpa.net>
Closes #15983

3 weeks agoman: move zfs_prepare_disk.8 to nodist_man_MANS
Shengqi Chen [Thu, 4 Apr 2024 01:04:15 +0000 (09:04 +0800)]
man: move zfs_prepare_disk.8 to nodist_man_MANS

The commit b53077a added zfs_prepare_disk.8 to the wrong list
dist_man_MANS, in which @zfsexecdir@ will not be properly substituted.
This leads to wrong path in the manpage in generated release tarballs.

Reported-by: Benda Xu <orv@debian.org>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Shengqi Chen <harry-chen@outlook.com>
Closes #15979

3 weeks agovdev props comment and manpage should include zfsd and FreeBSD mentions
Alek P [Thu, 4 Apr 2024 00:56:34 +0000 (20:56 -0400)]
vdev props comment and manpage should include zfsd and FreeBSD mentions

Reviewed-by: Tino Reichardt <milky-zfs@mcmilk.de>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Rob Norris <robn@despairlabs.com>
Reviewed-by: Allan Jude <allan@klarasystems.com>
Signed-off-by: Alek Pinchuk <apinchuk@axcient.com>
Closes #15968

3 weeks agozap_leaf: make l_hash[] variable length to silence UBSAN
Rob N [Wed, 3 Apr 2024 23:38:18 +0000 (10:38 +1100)]
zap_leaf: make l_hash[] variable length to silence UBSAN

When UBSAN is active and OpenZFS is a debug build, the l_hash assert at
the bottom of zap_open_leaf() causes UBSAN to complain.

This follows the example in 786641dcf to shut it up.

Sponsored-by: https://despairlabs.com/sponsor/
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <robn@despairlabs.com>
Closes #15964

3 weeks agoGive a better message from 'zpool get' with invalid pool name
Paul Dagnelie [Wed, 3 Apr 2024 23:34:46 +0000 (16:34 -0700)]
Give a better message from 'zpool get' with invalid pool name

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Don Brady <don.brady@klarasystems.com>
Reviewed-by: Tony Nguyen <tony.nguyen@delphix.com>
Signed-off-by: Paul Dagnelie <pcd@delphix.com>
Closes #15942

3 weeks agotests: simple zinject disk fault arg check
Rob Norris [Wed, 27 Mar 2024 05:15:48 +0000 (16:15 +1100)]
tests: simple zinject disk fault arg check

Just making sure the valid values for disk faults are accepted.
Obviously we can do a lot more, but this will do to get us started.

Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Brian Atkinson <batkinson@lanl.gov>
Reviewed-by: Tino Reichardt <milky-zfs@mcmilk.de>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Closes #15953

3 weeks agozinject: show more device fault fields
Rob Norris [Thu, 29 Feb 2024 23:38:41 +0000 (10:38 +1100)]
zinject: show more device fault fields

Once there's a few different kinds injected, its pretty hard to see them
otherwise.

So, lets show IO type, error type and frequency fields in the table too.

Since we now have to convert from error code to pretty string, refactor
the error names into a table and add lookup functions.

Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Brian Atkinson <batkinson@lanl.gov>
Reviewed-by: Tino Reichardt <milky-zfs@mcmilk.de>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Closes #15953

3 weeks agoMakefile.bsd: sort and cleanup source file list
Rob N [Wed, 3 Apr 2024 22:49:22 +0000 (09:49 +1100)]
Makefile.bsd: sort and cleanup source file list

All files now in their correct sections, and all sections match on-disk
dir layout, and all sorted.

Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Tino Reichardt <milky-zfs@mcmilk.de>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Closes #15943

3 weeks agoLinux 6.9 compat: blk_alloc_disk() now takes two args
Rob Norris [Wed, 27 Mar 2024 00:24:57 +0000 (11:24 +1100)]
Linux 6.9 compat: blk_alloc_disk() now takes two args

There's an extra nullable arg for queue limits. Detect it, and set it to
NULL. Similar change for blk_mq_alloc_disk(), now three args, same
treatment.

Error return now has error encoded in the return, so detect with
IS_ERR() and explicitly NULL our own return.

Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <robn@despairlabs.com>
Sponsored-by: https://despairlabs.com/sponsor/
Closes #16027
Closes #16033

3 weeks agoLinux 6.9 compat: bdev handles are now struct file
Rob Norris [Tue, 26 Mar 2024 23:07:50 +0000 (10:07 +1100)]
Linux 6.9 compat: bdev handles are now struct file

bdev_open_by_path() is replaced by bdev_file_open_by_path(), which
returns a plain old struct file*. Release function is gone entirely; the
regular file release function fput() will take care of the bdev
specifics.

Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <robn@despairlabs.com>
Sponsored-by: https://despairlabs.com/sponsor/
Closes #16027
Closes #16033

3 weeks agovdev_disk: don't touch vbio after its handed off to the kernel
Rob N [Wed, 3 Apr 2024 22:17:07 +0000 (09:17 +1100)]
vdev_disk: don't touch vbio after its handed off to the kernel

After IO is unplugged, it may complete immediately and vbio_completion
be called on interrupt context. That may interrupt or deschedule our
task. If its the last bio, the vbio will be freed. Then, we get
rescheduled, and try to write to freed memory through vbio->.

This patch just removes the the cleanup, and the corresponding assert.
These were leftovers from a previous iteration of vbio_submit() and were
always "belt and suspenders" ops anyway, never strictly required.

Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc
Reported-by: Rich Ercolani <rincebrain@gmail.com>
Reviewed-by: Laurențiu Nicola <lnicola@dend.ro>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: George Wilson <george.wilson@delphix.com>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Closes #16045
Closes #16050
Closes #16049

3 weeks agoxdr: header cleanup
Rob N [Wed, 3 Apr 2024 22:13:27 +0000 (09:13 +1100)]
xdr: header cleanup

#16047 notes that include/os/freebsd/spl/rpc/xdr.h carried an
(apparently) incompatible license. While looking into it, it seems that
this file is actually unnecessary these days - FreeBSD's kernel XDR has
XDR_CONTROL, xdrmem_control and XDR_GET_BYTES_AVAIL, while userspace has
XDR_CONTROL and xdrmem_control, and our implementation of
XDR_GET_BYTES_AVAIL for libspl works nicely with it. So this removes
that file outright.

To keep the includes in nvpair.c tidy, I've made a few small adjustments
to the Linux headers. By definition, rpc/types.h provides bool_t and is
included before rpc/xdr.h, so I've created rpc/types.h for Linux. This
isn't necessary for userspace; both FreeBSD native and tirpc on Linux
already have these headers set up correctly.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <robn@despairlabs.com>
Sponsored-by: https://despairlabs.com/sponsor/
Closes #16047
Closes #16051

3 weeks agoImprove dbuf_read() error reporting
Alexander Motin [Wed, 3 Apr 2024 22:04:26 +0000 (18:04 -0400)]
Improve dbuf_read() error reporting

Previous code reported non-ZIO errors only via return value, but
not via parent ZIO.  It could cause NULL-dereference panics due
to dmu_buf_hold_array_by_dnode() ignoring the return value,
relying solely on parent ZIO status.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Ameer Hamza <ahamza@ixsystems.com>
Reported by: Ameer Hamza <ahamza@ixsystems.com>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes #16042

3 weeks agoLinux 5.18+ compat: Detect filemap_range_has_page
Robert Evans [Sat, 30 Mar 2024 00:11:52 +0000 (20:11 -0400)]
Linux 5.18+ compat: Detect filemap_range_has_page

In v5.18 `filemap_range_has_page` moved to `pagemap.h`

`pagemap.h` has been around since 3.10 so just include both

Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Rob Norris <robn@despairlabs.com>
Signed-off-by: Robert Evans <evansr@google.com>
Closes #16034

3 weeks agoFix buffer underflow if sysfs file is empty
Robert Evans [Fri, 29 Mar 2024 21:59:23 +0000 (17:59 -0400)]
Fix buffer underflow if sysfs file is empty

Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Jason Lee <jasonlee@lanl.gov>
Signed-off-by: Robert Evans <evansr@google.com>
Closes #16028
Closes #16035

3 weeks agovdev_disk: clean up spa/bdev mode conversion
Rob N [Fri, 29 Mar 2024 21:51:33 +0000 (08:51 +1100)]
vdev_disk: clean up spa/bdev mode conversion

43e8f6e37 introduced a subtle API misuse, in that it passed the output
from vdev_bdev_mode() back into itself. Fortunately, the
SPA_MODE_(READ|WRITE) bit values exactly map to the FMODE_(READ|WRITE) &
BLK_OPEN_(READ|WRITE) bit values, so it didn't result in a bug, but it
was hard to read and understand, so I cleaned it up.

In doing so, I noticed that the only call to vdev_bdev_mode() without
the "exclusive" flag set was in that misuse, and actually, we never do a
non-exclusive blkdev_get_by_path(). So I've just made exclusive be
always-on.

Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Allan Jude <allan@klarasystems.com>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Closes #15995

4 weeks agozvols: prevent overflow of minor device numbers
Fabian-Gruenbichler [Fri, 29 Mar 2024 21:37:40 +0000 (22:37 +0100)]
zvols: prevent overflow of minor device numbers

currently, the linux kernel allows 2^20 minor devices per major device
number.  ZFS reserves blocks of 2^4 minors per zvol: 1 for the zvol
itself, the other 15 for the first partitions of that zvol. as a result,
only 2^16 such blocks are available for use.

there are no checks in place to avoid overflowing into the major device
number when more than 2^16 zvols are allocated (with volmode=dev or
default). instead of ignoring this limit, which comes with all sorts of
weird knock-on effects, detect this situation and simply fail allocating
the zvol block device early on.

without this safeguard, the kernel will reject the attempt to create an
already existing block device, but ZFS doesn't handle this error and
gets confused about which zvol occupies which minor slot, potentially
resulting in kernel NULL derefs and other issues later on.

Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Fabian Grünbichler <f.gruenbichler@proxmox.com>
Closes #16006

4 weeks agoAdd ashift validation when adding devices to a pool
George Wilson [Fri, 29 Mar 2024 19:15:56 +0000 (15:15 -0400)]
Add ashift validation when adding devices to a pool

Currently, zpool add allows users to add top-level vdevs that have
different ashifts but doing so prevents users from being able to
perform a top-level vdev removal. Often times consumers may not realize
that they have mismatched ashifts until the top-level removal fails.

This feature adds ashift validation to the zpool add command and will
fail the operation if the sector size of the specified vdev does not
match the existing pool. This behavior can be disabled by using the -f
flag. In addition, new flags have been added to provide fine-grained
control to disable specific checks. These flags
are:

--allow-in-use
--allow-ashift-mismatch
--allow-replicaton-mismatch

The force flag will disable all of these checks.

Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Mark Maybee <mmaybee@delphix.com>
Signed-off-by: George Wilson <gwilson@delphix.com>
Closes #15509

4 weeks agoZTS: fix flakiness in cp_files_002_pos
Robert Evans [Wed, 27 Mar 2024 21:59:16 +0000 (17:59 -0400)]
ZTS: fix flakiness in cp_files_002_pos

Fix RANDOM to not return zero.

Overwriting with `dd ... count=0` does not test anything.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Allan Jude <allan@klarasystems.com>
Signed-off-by: Robert Evans <evansr@google.com>
Closes #16029

4 weeks agoBRT: Check pool clone stats in more tests
Alexander Motin [Tue, 19 Mar 2024 17:08:05 +0000 (13:08 -0400)]
BRT: Check pool clone stats in more tests

This should allow to catch some leaks, if those happen.

While there fix some cosmetic issues.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes #16007

4 weeks agoBRT: Fix tests to work on non-empty pools
Alexander Motin [Tue, 19 Mar 2024 16:25:14 +0000 (12:25 -0400)]
BRT: Fix tests to work on non-empty pools

It should not normally happen, but if it does, better to not fail
everything for no good reason, or it may be hard to debug.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes #16007

4 weeks agoBRT: Fix holes cloning.
Alexander Motin [Mon, 18 Mar 2024 18:19:53 +0000 (14:19 -0400)]
BRT: Fix holes cloning.

 - When reading L0 block pointers handle buffers without ones and
without dirty records as a holes.  Those appear when dnode size
was increased, but the end was never written, so there are no new
indirection levels to store the pointers.  It makes no sense to
return EAGAIN here, since sync won't create new indirection levels
until there will be actual writes.
 - When cloning blocks set destination hole logical birth time
to the current TXG.  Otherwise if we are cloning over existing
data, newly created holes may not be properly replicated later.
Use BP_SET_BIRTH() when possible to not replicate its logic.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes #15994
Closes #16007

4 weeks agoBRT: Skip getting length in brt_entry_lookup()
Alexander Motin [Tue, 26 Mar 2024 00:13:45 +0000 (20:13 -0400)]
BRT: Skip getting length in brt_entry_lookup()

Unlike DDT, where ZAP values may have different lengths due to
compression, all BRT entries are identical 8-byte counters.  It
does not make sense to first fetch the length only to assert it.
zap_lookup_uint64() is specifically designed to work with counters
of different size and should return error if something odd found.
Calling it straight allows to save some measurable CPU time.

Reviewed-by: Pawel Jakub Dawidek <pawel@dawidek.net>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Rob Norris <robn@despairlabs.com>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes #15950

4 weeks agoabd_iter_page: don't use compound heads on Linux <4.5
Rob Norris [Wed, 13 Mar 2024 23:57:30 +0000 (10:57 +1100)]
abd_iter_page: don't use compound heads on Linux <4.5

Before 4.5 (specifically, torvalds/linux@ddc58f2), head and tail pages
in a compound page were refcounted separately. This means that using the
head page without taking a reference to it could see it cleaned up later
before we're finished with it. Specifically, bio_add_page() would take a
reference, and drop its reference after the bio completion callback
returns.

If the zio is executed immediately from the completion callback, this is
usually ok, as any data is referenced through the tail page referenced
by the ABD, and so becomes "live" that way. If there's a delay in zio
execution (high load, error injection), then the head page can be freed,
along with any dirty flags or other indicators that the underlying
memory is used. Later, when the zio completes and that memory is
accessed, its either unmapped and an unhandled fault takes down the
entire system, or it is mapped and we end up messing around in someone
else's memory. Both of these are very bad.

The solution on these older kernels is to take a reference to the head
page when we use it, and release it when we're done. There's not really
a sensible way under our current structure to do this; the "best" would
be to keep a list of head page references in the ABD, and release them
when the ABD is freed.

Since this additional overhead is totally unnecessary on 4.5+, where
head and tail pages share refcounts, I've opted to simply not use the
compound head in ABD page iteration there. This is theoretically less
efficient (though cleaning up head page references would add overhead),
but its safe, and we still get the other benefits of not mapping pages
before adding them to a bio and not mis-splitting pages.

There doesn't appear to be an obvious symbol name or config option we
can match on to discover this behaviour in configure (and the mm/page
APIs have changed a lot since then anyway), so I've gone with a simple
version check.

Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Closes #15533
Closes #15588

4 weeks agovdev_disk: use bio_chain() to submit multiple BIOs
Rob Norris [Wed, 21 Feb 2024 00:07:21 +0000 (11:07 +1100)]
vdev_disk: use bio_chain() to submit multiple BIOs

Simplifies our code a lot, so we don't have to wait for each and
reassemble them.

Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Closes #15533
Closes #15588

4 weeks agovdev_disk: add module parameter to select BIO submission method
Rob Norris [Tue, 9 Jan 2024 02:28:57 +0000 (13:28 +1100)]
vdev_disk: add module parameter to select BIO submission method

This makes the submission method selectable at module load time via the
`zfs_vdev_disk_classic` parameter, allowing this change to be backported
to 2.2 safely, and disabled in favour of the "classic" submission method
if new problems come up.

Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Closes #15533
Closes #15588

4 weeks agovdev_disk: rewrite BIO filling machinery to avoid split pages
Rob Norris [Tue, 18 Jul 2023 01:11:29 +0000 (11:11 +1000)]
vdev_disk: rewrite BIO filling machinery to avoid split pages

This commit tackles a number of issues in the way BIOs (`struct bio`)
are constructed for submission to the Linux block layer.

The kernel has a hard upper limit on the number of pages/segments that
can be added to a BIO, as well as a separate limit for each device
(related to its queue depth and other scheduling characteristics).

ZFS counts the number of memory pages in the request ABD
(`abd_nr_pages_off()`, and then uses that as the number of segments to
put into the BIO, up to the hard upper limit. If it requires more than
the limit, it will create multiple BIOs.

Leaving aside the fact that page count method is wrong (see below), not
limiting to the device segment max means that the device driver will
need to split the BIO in half. This is alone is not necessarily a
problem, but it interacts with another issue to cause a much larger
problem.

The kernel function to add a segment to a BIO (`bio_add_page()`) takes a
`struct page` pointer, and offset+len within it. `struct page` can
represent a run of contiguous memory pages (known as a "compound page").
In can be of arbitrary length.

The ZFS functions that count ABD pages and load them into the BIO
(`abd_nr_pages_off()`, `bio_map()` and `abd_bio_map_off()`) will never
consider a page to be more than `PAGE_SIZE` (4K), even if the `struct
page` is for multiple pages. In this case, it will load the same `struct
page` into the BIO multiple times, with the offset adjusted each time.

With a sufficiently large ABD, this can easily lead to the BIO being
entirely filled much earlier than it could have been. This is also
further contributes to the problem caused by the incorrect segment limit
calculation, as its much easier to go past the device limit, and so
require a split.

Again, this is not a problem on its own.

The logic for "never submit more than `PAGE_SIZE`" is actually a little
more subtle. It will actually never submit a buffer that crosses a 4K
page boundary.

In practice, this is fine, as most ABDs are scattered, that is a list of
complete 4K pages, and so are loaded in as such.

Linear ABDs are typically allocated from slabs, and for small sizes they
are frequently not aligned to page boundaries. For example, a 12K
allocation can span four pages, eg:

     -- 4K -- -- 4K -- -- 4K -- -- 4K --
    |        |        |        |        |
          :## ######## ######## ######:    [1K, 4K, 4K, 3K]

Such an allocation would be loaded into a BIO as you see:

    [1K, 4K, 4K, 3K]

This tends not to be a problem in practice, because even if the BIO were
filled and needed to be split, each half would still have either a start
or end aligned to the logical block size of the device (assuming 4K at
least).

---

In ideal circumstances, these shortcomings don't cause any particular
problems. Its when they start to interact with other ZFS features that
things get interesting.

Aggregation will create a "gang" ABD, which is simply a list of other
ABDs. Iterating over a gang ABD is just iterating over each ABD within
it in turn.

Because the segments are simply loaded in order, we can end up with
uneven segments either side of the "gap" between the two ABDs. For
example, two 12K ABDs might be aggregated and then loaded as:

    [1K, 4K, 4K, 3K, 2K, 4K, 4K, 2K]

Should a split occur, each individual BIO can end up either having an
start or end offset that is not aligned to the logical block size, which
some drivers (eg SCSI) will reject. However, this tends not to happen
because the default aggregation limit usually keeps the BIO small enough
to not require more than one split, and most pages are actually full 4K
pages, so hitting an uneven gap is very rare anyway.

If the pool is under particular memory pressure, then an IO can be
broken down into a "gang block", a 512-byte block composed of a header
and up to three block pointers. Each points to a fragment of the
original write, or in turn, another gang block, breaking the original
data up over and over until space can be found in the pool for each of
them.

Each gang header is a separate 512-byte memory allocation from a slab,
that needs to be written down to disk. When the gang header is added to
the BIO, its a single 512-byte segment.

Pulling all this together, consider a large aggregated write of gang
blocks. This results a BIO containing lots of 512-byte segments. Given
our tendency to overfill the BIO, a split is likely, and most possible
split points will yield a pair of BIOs that are misaligned. Drivers that
care, like the SCSI driver, will reject them.

---

This commit is a substantial refactor and rewrite of much of `vdev_disk`
to sort all this out.

`vdev_bio_max_segs()` now returns the ideal maximum size for the device,
if available. There's also a tuneable `zfs_vdev_disk_max_segs` to
override this, to assist with testing.

We scan the ABD up front to count the number of pages within it, and to
confirm that if we submitted all those pages to one or more BIOs, it
could be split at any point with creating a misaligned BIO.  If the
pages in the BIO are not usable (as in any of the above situations), the
ABD is linearised, and then checked again. This is the same technique
used in `vdev_geom` on FreeBSD, adjusted for Linux's variable page size
and allocator quirks.

`vbio_t` is a cleanup and enhancement of the old `dio_request_t`. The
idea is simply that it can hold all the state needed to create, submit
and return multiple BIOs, including all the refcounts, the ABD copy if
it was needed, and so on. Apart from what I hope is a clearer interface,
the major difference is that because we know how many BIOs we'll need up
front, we don't need the old overflow logic that would grow the BIO
array, throw away all the old work and restart. We can get it right from
the start.

Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Closes #15533
Closes #15588

4 weeks agovdev_disk: make read/write IO function configurable
Rob Norris [Tue, 9 Jan 2024 01:29:19 +0000 (12:29 +1100)]
vdev_disk: make read/write IO function configurable

This is just setting up for the next couple of commits, which will add a
new IO function and a parameter to select it.

Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Closes #15533
Closes #15588

4 weeks agovdev_disk: reorganise vdev_disk_io_start
Rob Norris [Tue, 9 Jan 2024 01:23:30 +0000 (12:23 +1100)]
vdev_disk: reorganise vdev_disk_io_start

Light reshuffle to make it a bit more linear to read and get rid of a
bunch of args that aren't needed in all cases.

Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Closes #15533
Closes #15588

4 weeks agovdev_disk: rename existing functions to vdev_classic_*
Rob Norris [Tue, 9 Jan 2024 01:12:56 +0000 (12:12 +1100)]
vdev_disk: rename existing functions to vdev_classic_*

This is just renaming the existing functions we're about to replace and
grouping them together to make the next commits easier to follow.

Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Closes #15533
Closes #15588

4 weeks agoabd: add page iterator
Rob Norris [Mon, 11 Dec 2023 05:05:54 +0000 (16:05 +1100)]
abd: add page iterator

The regular ABD iterators yield data buffers, so they have to map and
unmap pages into kernel memory. If the caller only wants to count
chunks, or can use page pointers directly, then the map/unmap is just
unnecessary overhead.

This adds adb_iterate_page_func, which yields unmapped struct page
instead.

Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Closes #15533
Closes #15588

4 weeks agolinux 5.4 compat: page_size()
Rob Norris [Mon, 13 Nov 2023 06:55:29 +0000 (17:55 +1100)]
linux 5.4 compat: page_size()

Before 5.4 we have to do a little math.

Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Closes #15533
Closes #15588

4 weeks agoBRT: Make BRT block sizes configurable
Alexander Motin [Mon, 25 Mar 2024 22:02:38 +0000 (18:02 -0400)]
BRT: Make BRT block sizes configurable

Similar to DDT make BRT data and indirect block sizes configurable
via module parameters.  I am not sure what would be the best yet,
but similar to DDT 4KB blocks kill all chances of compression on
vdev with ashift=12 or more, that on my tests reaches 3x.

While here, fix documentation for respective DDT parameters.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes #15967

4 weeks agoProvide macros for setting and getting blkptr birth times
George Wilson [Mon, 25 Mar 2024 22:01:54 +0000 (18:01 -0400)]
Provide macros for setting and getting blkptr birth times

There exist a couple of macros that are used to update the blkptr birth
times but they can often be confusing. For example, the
BP_PHYSICAL_BIRTH() macro will provide either the physical birth time
if it is set or else return back the logical birth time. The
complement to this macro is BP_SET_BIRTH() which will set the logical
birth time and set the physical birth time if they are not the same.
Consumers may get confused when they are trying to get the physical
birth time and use the BP_PHYSICAL_BIRTH() macro only to find out that
the logical birth time is what is actually returned.

This change cleans up these macros and makes them symmetrical. The same
functionally is preserved but the name is changed. Instead of calling
BP_PHYSICAL_BIRTH(), consumer can now call BP_GET_BIRTH(). In
additional to cleaning up this naming conventions, two new sets of
macros are introduced -- BP_[SET|GET]_LOGICAL_BIRTH() and
BP_[SET|GET]_PHYSICAL_BIRTH.  These new macros allow the consumer to
get and set the specific birth time.

As part of the cleanup, the unused GRID macros have been removed and
that portion of the blkptr are currently unused.

Reviewed-by: Matthew Ahrens <mahrens@delphix.com>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Mark Maybee <mark.maybee@delphix.com>
Signed-off-by: George Wilson <gwilson@delphix.com>
Closes #15962

4 weeks agoBRT: Relax brt_pending_apply() locking
Alexander Motin [Mon, 25 Mar 2024 21:59:55 +0000 (17:59 -0400)]
BRT: Relax brt_pending_apply() locking

Since brt_pending_apply() is running in syncing context, no other
brt_pending_tree accesses are possible for the TXG.  We don't need
to acquire brt_pending_lock here.

Reviewed-by: Pawel Jakub Dawidek <pawel@dawidek.net>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Brian Atkinson <batkinson@lanl.gov>
Reviewed-by: Rob Norris <robn@despairlabs.com>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes #15955

4 weeks agoZAP: Massively switch to _by_dnode() interfaces
Alexander Motin [Mon, 25 Mar 2024 21:58:50 +0000 (17:58 -0400)]
ZAP: Massively switch to _by_dnode() interfaces

Before this change ZAP called dnode_hold() for almost every block
access, that was clearly visible in profiler under heavy load, such
as BRT.  This patch makes it always hold the dnode reference between
zap_lockdir() and zap_unlockdir().  It allows to avoid most of dnode
operations between those.  It also adds several new _by_dnode() APIs
to ZAP and uses them in BRT code.  Also adds dmu_prefetch_by_dnode()
variant and uses it in the ZAP code.

After this there remains only one call to dmu_buf_dnode_enter(),
which seems to be unneeded.  So remove the call and the functions.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes #15951

4 weeks agoBRT: Skip duplicate BRT prefetches
Alexander Motin [Mon, 25 Mar 2024 21:58:04 +0000 (17:58 -0400)]
BRT: Skip duplicate BRT prefetches

If there is a pending entry for this block, then we've already
issued BRT prefetch for it within this TXG, so don't do it again.
BRT vdev lookup and following zap_prefetch_uint64() call can be
pretty expensive and should be avoided when not necessary.

Reviewed-by: Pawel Jakub Dawidek <pawel@dawidek.net>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes #15941

4 weeks agoFix corruption caused by mmap flushing problems
Robert Evans [Mon, 25 Mar 2024 21:56:49 +0000 (17:56 -0400)]
Fix corruption caused by mmap flushing problems

1) Make mmap flushes synchronous. Linux may skip flushing dirty pages
   already in writeback unless data-integrity sync is requested.

2) Change zfs_putpage to use TXG_WAIT. Otherwise dirty pages may be
   skipped due to DMU pushing back on TX assign.

3) Add missing mmap flush when doing block cloning.

4) While here, pass errors from putpage to writepage/writepages.

This change fixes corruption edge cases, but unfortunately adds
synchronous ZIL flushes for dirty mmap pages to llseek and bclone
operations. It may be possible to avoid these sync writes later
but would need more tricky refactoring of the writeback code.

Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Robert Evans <evansr@google.com>
Closes #15933
Closes #16019

5 weeks agoZAP: Some cleanups/micro-optimizations
Alexander Motin [Thu, 21 Mar 2024 23:43:53 +0000 (19:43 -0400)]
ZAP: Some cleanups/micro-optimizations

- Remove custom zap_memset(), use regular memset().
- Use PANIC() instead of opaque cmn_err(CE_PANIC).
- Provide entry parameter to zap_leaf_rehash_entry().
- Reduce branching in zap_leaf_array_create() inner loop.
- Remove signedness where it should not be.

Should be no function changes.

Reviewed-by: Brian Atkinson <batkinson@lanl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes #15976

5 weeks agoudev: correctly handle partition #16 and later
Fabian-Gruenbichler [Thu, 21 Mar 2024 23:38:24 +0000 (00:38 +0100)]
udev: correctly handle partition #16 and later

If a zvol has more than 15 partitions, the minor device number exhausts
the slot count reserved for partitions next to the zvol itself. As a
result, the minor number cannot be used to determine the partition
number for the higher partition, and doing so results in wrong named
symlinks being generated by udev.

Since the partition number is encoded in the block device name anyway,
let's just extract it from there instead.

Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Tino Reichardt <milky-zfs@mcmilk.de>
Signed-off-by: Fabian Grünbichler <f.gruenbichler@proxmox.com>
Closes #15904
Closes #15970

5 weeks agoBRT: Change brt_pending_tree sorting order
Alexander Motin [Thu, 21 Mar 2024 22:42:21 +0000 (18:42 -0400)]
BRT: Change brt_pending_tree sorting order

It does not look important how exactly brt_pending_tree is sorted.
When cloning large file, it is quite likely that all of its blocks
have identical physical birth times, so comparing them first does
not provide useful entropy, while accesses additional cache line.
In most cases combination of vdev and offset provides unique result
and physical birth time comparison is not even needed.  Meanwhile,
when traversing the tree inside brt_pending_apply(), it can be
beneficial for dbuf cache and CPU cache hits to group processing
by vdev and so by the per-VDEV BRT ZAPs.

Reviewed-by: Rob Norris <robn@despairlabs.com>
Reviewed-by: Brian Atkinson <batkinson@lanl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes #15954

5 weeks agozio: update ZIO type x stage documentation
Rob N [Thu, 21 Mar 2024 19:10:04 +0000 (06:10 +1100)]
zio: update ZIO type x stage documentation

- add column for TRIM ZIOs
- remove R from ZIO_STAGE_ISSUE_ASYNC, never happened
- remove I from ZIO_STAGE_VDEV_IO_DONE, never happened

Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Closes #15959

5 weeks agoFix option string, adding -e and fixing order
Cameron Harr [Thu, 21 Mar 2024 16:00:29 +0000 (09:00 -0700)]
Fix option string, adding -e and fixing order

The recently added '-e' option (PR #15769) missed adding the
new option in the online `zpool status` help command. This
adds the options and reorders a couple of the other options
that were not listed alphabetically.

Reviewed-by: Brian Atkinson <batkinson@lanl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Cameron Harr <harr1@llnl.gov>
Closes #16008

5 weeks agoUpdate resume token at object receive.
Alexander Motin [Thu, 21 Mar 2024 00:22:36 +0000 (20:22 -0400)]
Update resume token at object receive.

Before this change resume token was updated only on data receive.
Usually it is enough to resume replication without much overlap.
But we've got a report of a curios case, where replication source
was traversed with recursive grep, which through enabled atime
modified every object without modifying any data.  It produced
several gigabytes of replication traffic without a single data
write and so without a single resume point.

While the resume token was not designed to resume from an object,
I've found that the send implementation always sends object before
any data. So by requesting resume from offset 0 we are effectively
resuming from the object, followed (or not) by the data at offset
0, just as we need it.

Reviewed-by: Allan Jude <allan@klarasystems.com>
Reviewed-by: Paul Dagnelie <pcd@delphix.com>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes #15927

5 weeks agoLinux 6.8 compat: use splice_copy_file_range() for fallback
Rob N [Wed, 20 Mar 2024 23:46:15 +0000 (10:46 +1100)]
Linux 6.8 compat: use splice_copy_file_range() for fallback

Linux 6.8 removes generic_copy_file_range(), which had been reduced to a
simple wrapper around splice_copy_file_range(). Detect that function
directly and use it if generic_ is not available.

Sponsored-by: https://despairlabs.com/sponsor/
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <robn@despairlabs.com>
Closes #15930
Closes #15931

5 weeks agofreebsd: fix missing headers in distribution tarball
Rob N [Wed, 20 Mar 2024 17:08:50 +0000 (04:08 +1100)]
freebsd: fix missing headers in distribution tarball

arc_os.h and freebsd_event.h aren't included in release tarballs, so the
build fails on FreeBSD. This fixes it.

Sponsored-by: https://despairlabs.com/sponsor/
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Tino Reichardt <milky-zfs@mcmilk.de>
Signed-off-by: Rob Norris <robn@despairlabs.com>
Closes #15963

8 weeks agoddt: reduce DDT_NAMELEN
Rob Norris [Mon, 19 Feb 2024 10:19:32 +0000 (21:19 +1100)]
ddt: reduce DDT_NAMELEN

This is the buffer size passed to ddt_object_name(), to expand the
DMU_POOL_DDT format. That format inserts the table checksum, class and
type names, which as I write this are max 6, 9 and 3, respectively.

Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <robn@despairlabs.com>
Sponsored-by: https://despairlabs.com/sponsor/
Closes #15908

8 weeks agoconfig: use -Wno-format-truncation globally
Rob Norris [Mon, 19 Feb 2024 10:13:59 +0000 (21:13 +1100)]
config: use -Wno-format-truncation globally

-Wformat-truncation looks for places where the return code of snprintf()
is unchecked and the provided buffer might be too short. This is based
on a heuristic that can change between compiler versions.

It has been seen to get this wrong in ddt_object_name(), leading to
DDT_NAMELEN being increased somewhat arbitrarily.

There's no good reason to have this warning enabled, so here we disable
it everywhere. Truncation may be undesirable, but snprintf() is
guaranteed to emit a trailing null, so at worst we get a short string,
not a buffer overrun.

Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <robn@despairlabs.com>
Sponsored-by: https://despairlabs.com/sponsor/
Closes #15908

8 weeks agoFixed parameter passing error when calling zfs_acl_chmod
Quartz [Mon, 26 Feb 2024 19:41:44 +0000 (03:41 +0800)]
Fixed parameter passing error when calling zfs_acl_chmod

Follow up to 99495ba6abbf0bb726324d03212c6f5ffa00043e which
accidentally introduce this regression.

Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Quartz <yyhran@163.com>
Closes #15907

2 months agoCheck for minimum partition size
Brian Behlendorf [Fri, 16 Feb 2024 17:07:32 +0000 (09:07 -0800)]
Check for minimum partition size

On Linux block devices used for vdevs will by partitioned.  The block
device must be large enough for an 64M partition starting at offset
of 2048 sectors (part1), and a second 64M reserved partition at the
end of the device (part9).

This commit adds a capacity check when creating the GPT label to
immediately detect a device which is too small.  With the existing
code this would be caught slightly latter when attempting to use
the partition.  Catching it sooner let's us print a more useful error.

Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #15898

2 months agoZTS: Skip cross-fs bclone tests if FreeBSD < 14.0
Tony Hutter [Fri, 16 Feb 2024 16:59:56 +0000 (08:59 -0800)]
ZTS: Skip cross-fs bclone tests if FreeBSD < 14.0

Skip cross filesystem block cloning tests on FreeBSD if running
less than version 14.0.  Cross filesystem copy_file_range() was
added in FreeBSD 14.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tony Hutter <hutter2@llnl.gov>
Closes #15901

2 months agoddt: document the theory and the key data structures
Rob Norris [Mon, 27 Nov 2023 23:43:36 +0000 (10:43 +1100)]
ddt: document the theory and the key data structures

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Sponsored-by: Klara, Inc.
Sponsored-by: iXsystems, Inc.
Closes #15887

2 months agoddt: only create tables for dedup-capable checksums
Rob Norris [Thu, 1 Feb 2024 00:05:18 +0000 (11:05 +1100)]
ddt: only create tables for dedup-capable checksums

Most values in zio_checksum can never be used for dedup, partly because
the dedup= property only offers a limited list, but also some values (eg
ZIO_CHECKSUM_OFF) aren't real and will never be seen.

A true flag would be better than a hardcoded list, but thats more
cleanup elsewhere than I want to do right now.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Sponsored-by: Klara, Inc.
Sponsored-by: iXsystems, Inc.
Closes #15887

2 months agoddt: simplify entry load and flags
Rob Norris [Tue, 5 Dec 2023 03:28:39 +0000 (14:28 +1100)]
ddt: simplify entry load and flags

Only a single bit is needed to track entry state, and definitely not two
whole bytes. Some light refactoring in ddt_lookup() is needed to support
this, but it reads a lot better now.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Sponsored-by: Klara, Inc.
Sponsored-by: iXsystems, Inc.
Closes #15887

2 months agoddt: remove ddt_node
Rob Norris [Mon, 31 Jul 2023 07:42:34 +0000 (17:42 +1000)]
ddt: remove ddt_node

Nothing uses it.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Sponsored-by: Klara, Inc.
Sponsored-by: iXsystems, Inc.
Closes #15887

2 months agoddt: rework ops interface in terms of keys and values
Rob Norris [Mon, 3 Jul 2023 13:28:46 +0000 (23:28 +1000)]
ddt: rework ops interface in terms of keys and values

Store objects store keys and values, so have them take those types and
nothing more. This way, they don't need to be concerned about the "kind"
of entry being operated on; the dispatch layer can take care of the
appropriate conversions.

This adds a "contains" op to see if a particular entry exists without
loading it, which makes a couple of things easier to do; in particular,
it allows us to avoid an allocation in ddt_class_contains().

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Sponsored-by: Klara, Inc.
Sponsored-by: iXsystems, Inc.
Closes #15887

2 months agoddt: ensure ddt objects exist before trying to get stats from them
Rob Norris [Thu, 15 Jun 2023 06:10:00 +0000 (16:10 +1000)]
ddt: ensure ddt objects exist before trying to get stats from them

ddt_get_dedup_histogram() was actually checking it, just in an extremely
cursed way. ddt_get_dedup_object_stats() wasn't, but wasn't being called
from a dangerous place so no one noticed.

These checks are necessary, because spa_ddt[] is not populated until
spa_load(), but the spa can exist before that, while being created, and
as vdevs and metaslabs are initialised the space accounting functions
will be called to update pool space counts.

Probably the whole create path doesn't need to go asking for space
accounting from metadata subsystems until after the pool is created.
This will at least catch misuse.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Sponsored-by: Klara, Inc.
Sponsored-by: iXsystems, Inc.
Closes #15887

2 months agoddt: remove struct names and forward declarations
Rob Norris [Mon, 3 Jul 2023 02:43:37 +0000 (12:43 +1000)]
ddt: remove struct names and forward declarations

Things get confused when there's more than one name for a thing.

Note that we don't do this for ddt_object_t, ddt_histogram_t and
ddt_stat_t because they're part of the public ZFS interface.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Sponsored-by: Klara, Inc.
Sponsored-by: iXsystems, Inc.
Closes #15887

2 months agoddt: typedef ddt_type and ddt_class
Rob Norris [Mon, 3 Jul 2023 02:32:53 +0000 (12:32 +1000)]
ddt: typedef ddt_type and ddt_class

Mostly for consistency, so the reader is less likely to wonder why these
things look different.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Sponsored-by: Klara, Inc.
Sponsored-by: iXsystems, Inc.
Closes #15887

2 months agoddt: split internal DDT API into separate header
Rob Norris [Fri, 30 Jun 2023 03:35:18 +0000 (13:35 +1000)]
ddt: split internal DDT API into separate header

Just to make it easier to know which bits to pay attention to.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Sponsored-by: Klara, Inc.
Sponsored-by: iXsystems, Inc.
Closes #15887

2 months agoddt: remove DDE_GET_NDVAS macro
Rob Norris [Mon, 3 Jul 2023 05:25:06 +0000 (15:25 +1000)]
ddt: remove DDE_GET_NDVAS macro

It was a weird and confusing name, because it wasn't actually returning
the number of DVAs in the entry (as in, in the value/phys part) but the
maximum number of possible DVAs in a BP generated from the entry, based
on the encrypt bit in the key. This is unlike the similarly named
BP_GET_NDVAS, which really does return the number of DVAs.

Since its only used in this one place, and for a specific purpose, it
seemed more sensible to just write it in-place and remove the name.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Sponsored-by: Klara, Inc.
Sponsored-by: iXsystems, Inc.
Closes #15887

2 months agoddt: lift dedup stats out to separate file
Rob Norris [Tue, 16 May 2023 03:30:26 +0000 (13:30 +1000)]
ddt: lift dedup stats out to separate file

We want to add other kinds of dedup-related objects and keep stats for
them. This makes those functions easier to use from outside ddt.c.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Sponsored-by: Klara, Inc.
Sponsored-by: iXsystems, Inc.
Closes #15887

2 months agoddt: compare keys, not entries
Rob Norris [Fri, 9 Jun 2023 00:14:42 +0000 (10:14 +1000)]
ddt: compare keys, not entries

We're about to have different kinds of things that we'll compare on key,
so generalise this function to support that.

(It actually worked fine because of the way the casts work out, but it
requires the key to be at the start of the object so the cast through
ddt_entry_t works, and even then it reads strangely for anything that's
not a ddt_entry_t).

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Sponsored-by: Klara, Inc.
Sponsored-by: iXsystems, Inc.
Closes #15887

2 months agoddt_zap: standardise temp buffer allocations
Rob Norris [Wed, 17 Jan 2024 22:51:41 +0000 (09:51 +1100)]
ddt_zap: standardise temp buffer allocations

Always do them on the heap, and when we know how much we need, only that
much.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Sponsored-by: Klara, Inc.
Sponsored-by: iXsystems, Inc.
Closes #15887

2 months agoddt: move entry compression into ddt_zap
Rob Norris [Fri, 30 Jun 2023 02:48:45 +0000 (12:48 +1000)]
ddt: move entry compression into ddt_zap

I think I can say with some confidence that anyone making a new storage
type in 2023 is doing their own thing with compression, not this.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Sponsored-by: Klara, Inc.
Sponsored-by: iXsystems, Inc.
Closes #15887

2 months agoddt: modernise assertions
Rob Norris [Thu, 15 Feb 2024 08:37:38 +0000 (19:37 +1100)]
ddt: modernise assertions

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Sponsored-by: Klara, Inc.
Sponsored-by: iXsystems, Inc.
Closes #15887

2 months agoLinux: Cleanup taskq threads spawn/exit
Alexander Motin [Tue, 13 Feb 2024 19:15:16 +0000 (14:15 -0500)]
Linux: Cleanup taskq threads spawn/exit

This changes taskq_thread_should_stop() to limit maximum exit rate
for idle threads to one per 5 seconds.  I believe the previous one
was broken, not allowing any thread exits for tasks arriving more
than one at a time and so completing while others are running.

Also while there:
 - Remove taskq_thread_spawn() calls on task allocation errors.
 - Remove extra taskq_thread_should_stop() call.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Rich Ercolani <rincebrain@gmail.com>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes #15873

2 months agozdb: Fix false leak report for BRT objects
Bi11 [Tue, 13 Feb 2024 00:58:47 +0000 (08:58 +0800)]
zdb: Fix false leak report for BRT objects

Fix a misreport in 'zdb -d' where it falsely marked
BRT objects as leaked.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Signed-off-by: Yuxin Wang <yuxinwang9999@gmail.com>
Closes #15882

2 months agoBRT: Fix slop space calculation with block cloning
Bi11 [Mon, 12 Feb 2024 21:53:33 +0000 (05:53 +0800)]
BRT: Fix slop space calculation with block cloning

Similar to deduplication, the size of data duplicated by block cloning
should not be included in the slop space calculation.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Signed-off-by: Yuxin Wang <yuxinwang9999@gmail.com>
Closes #15874

2 months agoAllowing PERFPOOL to be defined by zfs-test users
Kevin Greene [Fri, 9 Feb 2024 18:02:46 +0000 (10:02 -0800)]
Allowing PERFPOOL to be defined by zfs-test users

Reviewed-by: John Wren Kennedy <john.kennedy@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Kevin Greene <kevin.greene@delphix.com>
Closes #15868

2 months agoUpdate zfs-snapshot.8
Shawn Bayern [Thu, 8 Feb 2024 21:06:12 +0000 (16:06 -0500)]
Update zfs-snapshot.8

Fixes a small inaccuracy in the description of snapshot
atomicity

zfs-snapshot(8) appears to contain a small error.  The existing
version reads "Snapshots are taken atomically, so that all
snapshots correspond to the same moment in time."  Per
zfs_main.c, which in do_snapshot() simply loops over argv, this
does not appear to be correct when multiple snapshots are
specified explicitly on the command line.  I believe the intent
of the man page was to say that *recursive* snapshots are all
created atomically.

This proposed change fixes that error.  Because the existing
statement may confuse some readers anyway, the commit also also
adds a small amount of general explanatory information that may
be helpful.

The change also adds an introductory sentence that summarizes
what 'zfs snapshot' does in the first place.  In that sentence,
the text "different datasets" is intended to indicate that
(again per the code) the same dataset cannot be specified
multiple times on the command line.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Shawn Bayern <sbayern@law.fsu.edu>
Closes #15857

2 months agozfs list: add '-t fs' and '-t vol' options
Rob N [Thu, 8 Feb 2024 18:22:58 +0000 (05:22 +1100)]
zfs list: add '-t fs' and '-t vol' options

Because "filesystem" and "volume" are just too long!

Sponsored-by: https://despairlabs.com/sponsor/
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <robn@despairlabs.com>
Closes #15864

2 months agoAdd slow disk diagnosis to ZED
Don Brady [Thu, 8 Feb 2024 17:19:52 +0000 (10:19 -0700)]
Add slow disk diagnosis to ZED

Slow disk response times can be indicative of a failing drive. ZFS
currently tracks slow I/Os (slower than zio_slow_io_ms) and generates
events (ereport.fs.zfs.delay).  However, no action is taken by ZED,
like is done for checksum or I/O errors.  This change adds slow disk
diagnosis to ZED which is opt-in using new VDEV properties:
  VDEV_PROP_SLOW_IO_N
  VDEV_PROP_SLOW_IO_T

If multiple VDEVs in a pool are undergoing slow I/Os, then it skips
the zpool_vdev_degrade().

Sponsored-By: OpenDrives Inc.
Sponsored-By: Klara Inc.
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Allan Jude <allan@klarasystems.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Co-authored-by: Rob Wing <rob.wing@klarasystems.com>
Signed-off-by: Don Brady <don.brady@klarasystems.com>
Closes #15469

2 months agoLUA: Backport CVE-2020-24370's patch
the-Chain-Warden-thresh [Wed, 7 Feb 2024 19:53:05 +0000 (03:53 +0800)]
LUA: Backport CVE-2020-24370's patch

CVE-2020-24370 is a security vulnerability in lua. Although the CVE
description in CVE-2020-24370 said that this CVE only affected lua
5.4.0, according to lua this CVE actually existed since lua 5.2. The
root cause of this CVE is the negation overflow that occurs when you
try to take the negative of 0x80000000. Thus, this CVE also exists in
openzfs. Try to backport the fix to the lua in openzfs since the
original fix is for 5.4 and several functions have been changed.

https://github.com/advisories/GHSA-gfr4-c37g-mm3v
https://nvd.nist.gov/vuln/detail/CVE-2020-24370
https://www.lua.org/bugs.html#5.4.0-11
https://github.com/lua/lua/commit/a585eae6e7ada1ca9271607a4f48dfb1786

Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: ChenHao Lu <18302010006@fudan.edu.cn>
Closes #15847

2 months agoAdd 'zpool status -e' flag to see unhealthy vdevs
Cameron Harr [Wed, 7 Feb 2024 17:12:12 +0000 (09:12 -0800)]
Add 'zpool status -e' flag to see unhealthy vdevs

When very large pools are present, it can be laborious to find
reasons for why a pool is degraded and/or where an unhealthy vdev
is. This option filters out vdevs that are ONLINE and with no errors
to make it easier to see where the issues are. Root and parents of
unhealthy vdevs will always be printed.

Testing:
ZFS errors and drive failures for multiple vdevs were simulated with
zinject.

Sample vdev listings with '-e' option
- All vdevs healthy
    NAME        STATE     READ WRITE CKSUM
    iron5       ONLINE       0     0     0

- ZFS errors
    NAME        STATE     READ WRITE CKSUM
    iron5       ONLINE       0     0     0
      raidz2-5  ONLINE       1     0     0
        L23     ONLINE       1     0     0
        L24     ONLINE       1     0     0
        L37     ONLINE       1     0     0

- Vdev faulted
    NAME        STATE     READ WRITE CKSUM
    iron5       DEGRADED     0     0     0
      raidz2-6  DEGRADED     0     0     0
        L67     FAULTED      0     0     0  too many errors

- Vdev faults and data errors
    NAME        STATE     READ WRITE CKSUM
    iron5       DEGRADED     0     0     0
      raidz2-1  DEGRADED     0     0     0
        L2      FAULTED      0     0     0  too many errors
      raidz2-5  ONLINE       1     0     0
        L23     ONLINE       1     0     0
        L24     ONLINE       1     0     0
        L37     ONLINE       1     0     0
      raidz2-6  DEGRADED     0     0     0
        L67     FAULTED      0     0     0  too many errors

- Vdev missing
    NAME        STATE     READ WRITE CKSUM
    iron5       DEGRADED     0     0     0
      raidz2-6  DEGRADED     0     0     0
        L67     UNAVAIL      3     1     0

- Slow devices when -s provided with -e
    NAME        STATE     READ WRITE CKSUM  SLOW
    iron5       DEGRADED     0     0     0     -
      raidz2-5  DEGRADED     0     0     0     -
        L10     FAULTED      0     0     0     0  external device fault
        L51     ONLINE       0     0     0    14

Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Cameron Harr <harr1@llnl.gov>
Closes #15769

2 months agoBRT: Fix FICLONE/FICLONERANGE shortened copy
Brian Behlendorf [Tue, 6 Feb 2024 00:44:45 +0000 (16:44 -0800)]
BRT: Fix FICLONE/FICLONERANGE shortened copy

On Linux the ioctl_ficlonerange() and ioctl_ficlone() system calls
are expected to either fully clone the specified range or return an
error.  The range may be for an entire file.  While internally ZFS
supports cloning partial ranges there's no way to return the length
cloned to the caller so we need to make this all or nothing.

As part of this change support for the REMAP_FILE_CAN_SHORTEN flag
has been added.  When REMAP_FILE_CAN_SHORTEN is set zfs_clone_range()
will return a shortened range when encountering pending dirty records.
When it's clear zfs_clone_range() will block and wait for the records
to be written out allowing the blocks to be cloned.

Furthermore, the file range lock is held over the region being cloned
to prevent it from being modified while cloning.  This doesn't quite
provide an atomic semantics since if an error is encountered only a
portion of the range may be cloned.  This will be converted to an
error if REMAP_FILE_CAN_SHORTEN was not provided and returned to the
caller.  However, the destination file range is left in an undefined
state.

A test case has been added which exercises this functionality by
verifying that `cp --reflink=never|auto|always` works correctly.

Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #15728
Closes #15842

2 months agolibzdb: Initial breakout of libzdb
Rich Ercolani [Mon, 5 Feb 2024 18:00:41 +0000 (13:00 -0500)]
libzdb: Initial breakout of libzdb

Step 1 in trying to slowly rip the zdb functions out of zdb.c
to allow people to play with more flexible things to leverage
zdb's functionality.

No promises on any functions or structs being stable, now or probably
in general unless someone builds a more polished abstraction, the
goal at the moment is to slowly untangle the global state usage
in zdb...

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rich Ercolani <rincebrain@gmail.com>
Closes #15804

2 months agoImprove performance for zpool trim on linux
Umer Saleem [Fri, 2 Feb 2024 19:51:51 +0000 (00:51 +0500)]
Improve performance for zpool trim on linux

On Linux, ZFS uses blkdev_issue_discard in vdev_disk_io_trim to issue
trim command which is synchronous.

This commit updates vdev_disk_io_trim to use __blkdev_issue_discard,
which is asynchronous. Unfortunately there isn't any asynchronous
version for blkdev_issue_secure_erase, so performance of secure trim
will still suffer.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Signed-off-by: Umer Saleem <usaleem@ixsystems.com>
Closes #15843

2 months agoLinux 6.8 compat: handle mnt_idmap user_namespace change
Rob Norris [Tue, 23 Jan 2024 10:14:06 +0000 (21:14 +1100)]
Linux 6.8 compat: handle mnt_idmap user_namespace change

struct mnt_idmap no longer has a struct user_namespace within it. Work
around this by creating a temporary with the copy of the map we need
taken from the idmap.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Co-authored-by: Youzhong Yang <yyang@mathworks.com>
Signed-off-by: Rob Norris <robn@despairlabs.com>
Sponsored-by: https://despairlabs.com/sponsor/
Closes #15805

2 months agoLinux 6.8 compat: fix inode permission tests
Rob Norris [Tue, 23 Jan 2024 06:43:20 +0000 (17:43 +1100)]
Linux 6.8 compat: fix inode permission tests

The name inode_permission is now defined in the kernel. Rename ours to
test_permission, in line with most of our other tests.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <robn@despairlabs.com>
Sponsored-by: https://despairlabs.com/sponsor/
Closes #15805

2 months agoLinux 6.8 compat: replace MAX_ORDER define
Rob Norris [Tue, 23 Jan 2024 05:41:05 +0000 (16:41 +1100)]
Linux 6.8 compat: replace MAX_ORDER define

MAX_ORDER has been renamed to MAX_PAGE_ORDER. Rather than just
redefining it, instead define our own name and set it consistently from
the start.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <robn@despairlabs.com>
Sponsored-by: https://despairlabs.com/sponsor/
Closes #15805

2 months agoLinux 6.8 compat: implement strlcpy fallback
Rob Norris [Tue, 23 Jan 2024 05:34:49 +0000 (16:34 +1100)]
Linux 6.8 compat: implement strlcpy fallback

Linux has removed strlcpy in favour of strscpy. This implements a
fallback implementation of strlcpy for this case.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <robn@despairlabs.com>
Sponsored-by: https://despairlabs.com/sponsor/
Closes #15805

2 months agoLinux 6.8 compat: update for new bdev access functions
Rob Norris [Tue, 23 Jan 2024 04:42:57 +0000 (15:42 +1100)]
Linux 6.8 compat: update for new bdev access functions

blkdev_get_by_path() and blkdev_put() have been replaced by
bdev_open_by_path() and bdev_release(), which return a "handle" object
with the bdev object itself inside.

This adds detection for the new functions, and macros to handle the old
and new forms consistently.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <robn@despairlabs.com>
Sponsored-by: https://despairlabs.com/sponsor/
Closes #15805

2 months agoLinux 6.8 compat: make test functions static
Rob Norris [Mon, 22 Jan 2024 23:50:53 +0000 (10:50 +1100)]
Linux 6.8 compat: make test functions static

The kernel is now being compiled with -Wmissing-prototypes. Most of our
test stub functions had no prototype, and failed to compile. Since they
don't need to be visible anywhere else, just make them all static.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <robn@despairlabs.com>
Sponsored-by: https://despairlabs.com/sponsor/
Closes #15805

2 months agoLinux 6.7 compat: META
Brian Behlendorf [Mon, 29 Jan 2024 19:35:43 +0000 (11:35 -0800)]
Linux 6.7 compat: META

Update the META file to reflect compatibility with the 6.7 kernel.

Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #15833

2 months agoDon't assert mg_initialized due to device addition race
Paul Dagnelie [Mon, 29 Jan 2024 18:36:42 +0000 (10:36 -0800)]
Don't assert mg_initialized due to device addition race

During device removal stress tests, we noticed that we were tripping
the assertion that mg_initialized was true. After investigation, it was
determined that the mg in question was the embedded log metaslab
group for a newly added vdev; the normal mg had been initialized (by
metaslab_sync_reassess, via vdev_sync_done). However, because the spa
config alloc lock is not held as writer across both calls to
metaslab_sync_reassess, it is possible for an allocation to happen
between the two metaslab_groups being initialized. Because the metaslab
code doesn't check the group in question, just the vdev's main mg, it
is possible to get past the initial check in vdev_allocatable and
later fail due to the assertion.

We simply remove the assertions. We could also consider locking the
ALLOC lock around the reassess calls in vdev_sync_done, but that risks
deadlocks. We could check the actual target mg in vdev_allocatable,
but that risks racing with a passivation that comes in after that
check but before the assertion. We still won't be able to actually
allocate from the metaslab group if no metaslabs are ready, so this
change shouldn't break anything.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: George Wilson <george.wilson@delphix.com>
Signed-off-by: Paul Dagnelie <pcd@delphix.com>
Closes #15818

2 months agolibzfs: use zfs_strerror() in place of strerror()
Richard Kojedzinszky [Mon, 22 Jan 2024 23:28:18 +0000 (00:28 +0100)]
libzfs: use zfs_strerror() in place of strerror()

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Tino Reichardt <milky-zfs@mcmilk.de>
Signed-off-by: Richard Kojedzinszky <richard@kojedz.in>
Closes #15793

2 months agolibzfs: make userquota_propname_decode threadsafe
Richard Kojedzinszky [Wed, 17 Jan 2024 19:48:02 +0000 (20:48 +0100)]
libzfs: make userquota_propname_decode threadsafe

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Tino Reichardt <milky-zfs@mcmilk.de>
Signed-off-by: Richard Kojedzinszky <richard@kojedz.in>
Closes #15793

2 months agolibnvpair.c: replace strstr() with strchr() for a single character
rilysh [Mon, 29 Jan 2024 17:46:13 +0000 (23:16 +0530)]
libnvpair.c: replace strstr() with strchr() for a single character

Since we're looking for a single new-line character in the haystack,
it's better (and slightly more efficient) to use strchr() instead of
strstr().

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Richard Yao <richard.yao@alumni.stonybrook.edu>
Signed-off-by: rilysh <nightquick@proton.me>
Closes #15798

2 months agoUpdate man pages to time(1) from time(2)
Chris Davidson [Mon, 29 Jan 2024 17:44:08 +0000 (12:44 -0500)]
Update man pages to time(1) from time(2)

zpool-iostat.8: Updated time(2) -> time(1) to align to manual page
zpool-list.8: Updated time(2) -> time(1) to align to manual page
zpool-status.8: Updated time(2) -> time(1) to align to manual page
zpool-wait.8: Update time(2) -> time(1) to align to manual page

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Christopher Davidson <christopher.davidson@gmail.com>
Closes #15823