Add zfs-test facility to automatically rerun failing tests

This was a project proposed as part of the Quality theme for the
hackthon for the 2021 OpenZFS Developer Summit. The idea is to improve
the usability of the automated tests that get run when a PR is created
by having failing tests automatically rerun in order to make flaky
tests less impactful.

Reviewed-by: John Kennedy <john.kennedy@delphix.com>
Reviewed-by: Tony Nguyen <tony.nguyen@delphix.com>
Signed-off-by: Paul Dagnelie <pcd@delphix.com>
Closes #12740

Linux 5.16: wait_on_page_bit() no longer available to modules

Instead, linux/pagemap.h offers a number of folio-specific functions to
be called instead. In this case, module/os/linux/zfs/zfs_vnops_os.c
wants to call wait_on_page_bit(pp, PG_writeback). This gets replaced
with folio_wait_bit(folio_page(pp), PG_writeback). This change modifies
the code to conditionally compile that if configure identifies th
presence of the folio_wait_bit() function.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Coleman Kane <ckane@colemankane.org>
Closes #12800

Iterate encrypted clones at zvol_create_minor

Userland figures out which encryption-root keys are required to load,
and issues ZFS_IOC_LOAD_KEY.
The tail section of spa_keystore_load_wkey() will call
zvol_create_minors() on the encryption-root object.

Any clones of the encrypted zvol will not be plumbed. This commits
adds additional logic to detect if zvol has clones, and is encrypted,
then adds these to the list of zvols to call zvol_create_minors() on.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Ryan Moeller <ryan@iXsystems.com>
Signed-off-by: Jorgen Lundman <lundman@lundman.net>
Closes #12471

Exclude zfs_copies_003_pos on Linux

This test case may fail on 5.13 and newer Linux kernels if the
/dev/zvol/ device is not created by udev.

Reviewed-by: Rich Ercolani <rincebrain@gmail.com>
Reviewed-by: John Kennedy <john.kennedy@delphix.com>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #12301
Closes  #12738

Restore dirty dnode detection logic

In addition to flushing memory mapped regions when checking holes,
commit de198f2d95 modified the dirty dnode detection logic to check
the dn->dn_dirty_records instead of the dn->dn_dirty_link.  Relying
on the dirty record has not be reliable, switch back to the previous
method.

Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #11900
Closes #12745

Fix lseek(SEEK_DATA/SEEK_HOLE) mmap consistency

When using lseek(2) to report data/holes memory mapped regions of
the file were ignored.  This could result in incorrect results.
To handle this zfs_holey_common() was updated to asynchronously
writeback any dirty mmap(2) regions prior to reporting holes.

Additionally, while not strictly required, the dn_struct_rwlock is
now held over the dirty check to prevent the dnode structure from
changing.  This ensures that a clean dnode can't be dirtied before
the data/hole is located.  The range lock is now also taken to
ensure the call cannot race with zfs_write().

Furthermore, the code was refactored to provide a dnode_is_dirty()
helper function which checks the dnode for any dirty records to
determine its dirtiness.

Reviewed-by: Matthew Ahrens <mahrens@delphix.com>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Rich Ercolani <rincebrain@gmail.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #11900
Closes #12724

Upgrade to libabigail 2.0.0

Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Dimitri John Ledkov <dimitri.ledkov@canonical.com>
Closes #12722
Closes #12739

zed: Control NVMe fault LEDs

The ZED code currently can only turn on the fault LED for
a faulted disk in a JBOD enclosure.  This extends support
for faulted NVMe disks as well.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tony Hutter <hutter2@llnl.gov>
Closes #12648
Closes #12695

Linux 5.16 compat: submit_bio()

The submit_bio() prototype has changed again.  The version is 5.16
still only expects a single argument but the return type has changed
to void.  Since we never used the returned value before update the
configure check to detect both single arg versions.

Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Alexander Lobakin <alobakin@pm.me>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #12725

Linux 5.16 compat: linux/elevator.h

Commit https://github.com/torvalds/linux/commit/2e9bc346 moved
the elevator.h header under the block/ directory as part of some
refactoring.  This turns out not to be a problem since there's
no longer anything we need from the header.  This has been the
case for some time, this change removes the elevator.h include
and replaces it with a major.h include.

Reviewed-by: Alexander Lobakin <alobakin@pm.me>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #12725

vdev_id: Fix PHY sorting

One of our developers noticed a bug in vdev_id where we were incorrectly
sorting PHYs using alphabetical sorting (which usually works) instead
of natural sorting (-v).  For example:

[port-0:0]# ls -d phy*
phy-0:10  phy-0:11  phy-0:8  phy-0:9

[port-0:0]# ls -vd phy*
phy-0:8  phy-0:9  phy-0:10  phy-0:11

This fixes the issue.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tony Hutter <hutter2@llnl.gov>
Closes #12699

vdev_id: Fix enclosure_symlinks feature

The vdev_id.conf "enclosure_symlinks" option persistently creates
and maps /dev/by-enclosure symlinks to dynamic /dev/sg* devices.

This patch fixes two issues:

1. The enclosure_symlinks feature was accidentally broken in:

   vdev_id: Support daisy-chained JBODs in multipath mode

2. Even when working, the feature numbered the enclosure
   sequentially rather than by HBA port number.  That meant that
   if a port was down or didn't appear in sysfs, then the
   enclosure_sumlinks numbers would be numbered wrong.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Arshad Hussain <arshad.hussain@aeoncomputing.com>
Signed-off-by: Tony Hutter <hutter2@llnl.gov>
Closes #12660

Rescan enclosure sysfs path on import

When you create a pool, zfs writes vd->vdev_enc_sysfs_path with the
enclosure sysfs path to the fault LEDs, like:

    vdev_enc_sysfs_path = /sys/class/enclosure/0:0:1:0/SLOT8

However, this enclosure path doesn't get updated on successive imports
even if enclosure path to the disk changes.  This patch fixes the issue.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tony Hutter <hutter2@llnl.gov>
Closes #11950
Closes #12095

FreeBSD: Catch up with recent VFS changes

cn_thread is always curthread.

https://cgit.freebsd.org/src/commit?id=b4a58fbf640409a1e507d9f7b411c83a3f83a2f3
https://cgit.freebsd.org/src/commit?id=2b68eb8e1dbbdaf6a0df1c83b26f5403ca52d4c3

Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Alan Somers <asomers@gmail.com>
Signed-off-by: Ryan Moeller <freqlabs@FreeBSD.org>
Closes #12668

FreeBSD: fix compilation of FreeBSD world after 29274c9f6

prng32_bounded() is available to kernel only on FreeBSD 13+.

Call inline random_get_pseudo_bytes() with correct pointer type.
To be consistent, apply to Linux as well.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Signed-off-by: Martin Matuska <mm@FreeBSD.org>
Closes #12282

ZTS: Standardize use of destroy_dataset in cleanup

When cleaning up a test case standardize on using the convention:

    datasetexists $ds && destroy_dataset $ds <flags>

By using 'destroy_dataset' instead of 'log_must zfs destroy' we ensure
that the destroy is retried in the event that a ZFS volume is busy.
This helps ensures ensure tests are fully cleaned up and prevents false
positive test failures on Linux.

Note that all of the tests which used 'zfs destroy' in cleanup have
been updated even if they don't use volumes.  This was done to
clearly establish the expected convention.

Reviewed-by: Rich Ercolani <rincebrain@gmail.com>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: John Kennedy <john.kennedy@delphix.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #12663

Workaround cloud-init hotplug issue

cloud-init added a hook which triggers on every device add/rm
event, which results in holding open devices for a while after
they're created/destroyed.

So let's shove an exclusion rule for that into the GH workflows
until it gets fixed.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: John Kennedy <john.kennedy@delphix.com>
Signed-off-by: Rich Ercolani <rincebrain@gmail.com>
Closes #12644
Closes #12669

Use fallthrough macro

As of the Linux 5.9 kernel a fallthrough macro has been added which
should be used to anotate all intentional fallthrough paths.  Once
all of the kernel code paths have been updated to use fallthrough
the -Wimplicit-fallthrough option will because the default.  To
avoid warnings in the OpenZFS code base when this happens apply
the fallthrough macro.

Additional reading: https://lwn.net/Articles/794944/

Reviewed-by: Tony Nguyen <tony.nguyen@delphix.com>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #12441

Detect HAVE_LARGE_STACKS at compile time (#12584)

Move HAVE_LARGE_STACKS definitions to header and set when appropriate.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Allan Jude <allan@klarasystems.com>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Ryan Moeller <ryan@iXsystems.com>
Signed-off-by: Kevin Bowling <kbowling@FreeBSD.org>
Closes #12350

Correct a flaw in the Python 3 version checking (#12636)

It turns out the ax_python_devel.m4 version check assumes that
("3.X+1.0" >= "3.X.0") is True in Python, which is not when X+1
is 10 or above and X is not. (Also presumably X+1=100 and ...)

So let's remake the check to behave consistently, using the
"packaging" or (if absent) the "distlib" modules.

(Also, update the Github workflows to use the new packages.)

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: John Kennedy <john.kennedy@delphix.com>
Signed-off-by: Rich Ercolani <rincebrain@gmail.com>
Closes: #12073

Tag 2.1.1

Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>

Linux 5.14 compat: META

Increase the Linux-Maximum version in the META file to 5.14.
All of the required compatibility patches have been merged
and the 5.14 kernel has been officially released.

Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #12565

Fixed data integrity issue when underlying disk returns error

Errors in zil_lwb_write_done() are not propagated to
zil_lwb_flush_vdevs_done() which can result in zil_commit_impl()
not returning an error to applications even when zfs was not able
to write data to the disk.

Remove the ZIO_FLAG_DONT_PROPAGATE flag from zio_rewrite() to
allow errors to propagate and consolidate the error handling for
flush and write errors to a single location (rather than having
error handling split between the "write done" and "flush done"
handlers).

Reviewed-by: George Wilson <gwilson@delphix.com>
Reviewed-by: Prakash Surya <prakash.surya@delphix.com>
Signed-off-by: Arun KV <arun.kv@datacore.com>
Closes #12391
Closes #12443

ZTS: Waiting for zvols to be available

This is a follow up patch for PR #12515 which addresses some
additional ZTS tests which are unreliable are should explicitly
wait for the required zvols to be available.

Reviewed-by: John Kennedy <john.kennedy@delphix.com>
Reviewed-by: Ryan Moeller <ryan@iXsystems.com>
Reviewed-by: @Theo13111
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #12553

Verify embedded blkptr's in arc_read()

The block pointer verification check in arc_read() should also
cover embedded block pointers.  While highly unlikely, accessing
a damaged block pointer can result in panic.  To further harden
the code extend the existing check to include embedded block
pointers and add a comment explaining the rational for this
sanity check.  Lastly, correct a flaw in zfs_blkptr_verify()
so the error count is checked even when checking a untrusted
config to verify the non-pool-specific portions of a block
pointer.

Reviewed-by: Matthew Ahrens <mahrens@delphix.com>
Reviewed-by: Tony Nguyen <tony.nguyen@delphix.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #12535

Linux 5.15 compat: get_acl()

Kernel commits

332f606b32b6 ovl: enable RCU'd ->get_acl()
0cad6246621b vfs: add rcu argument to ->get_acl() callback

Added compatibility code to detect the new ->get_acl() interface
and correctly handle the case where the new rcu argument is set.

Reviewed-by: Coleman Kane <ckane@colemankane.org>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #12548

Allow sending corrupt snapshots even if metadata is corrupted

When zfs_send_corrupt_data is set, use the TRAVERSE_HARD flag,
so traverse_visitbp() will not fail with ECKSUM if a blockpointer
cannot be read, but rather will continue and send the objects it can.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: John Kennedy <john.kennedy@delphix.com>
Signed-off-by: Allan Jude <allan@klarasystems.com>
Sponsored-By: Klara Inc.
Sponsored-By: WHC Online Solutions Inc.
Closes #12541

arc: Drop an incorrect assert

Unfortunately, there was an overzealous assertion that was (in pretty
specific circumstances) false, causing failure.  This assertion was
added in error, so we're removing it.

Reviewed-by: Matthew Ahrens <mahrens@delphix.com>
Reviewed-by: George Wilson <gwilson@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rich Ercolani <rincebrain@gmail.com>
Closes #9897
Closes #12020
Closes #12246

Compressed receive with different ashift can result in incorrect PSIZE on disk

We round up the psize to the nearest multiple of the asize or to the
lsize, whichever is smaller. Once that's done, we allocate a new
buffer of the appropriate size, zero the tail, and copy the data
into it. This adds a small performance cost to these kinds of writes,
but fixes the bookkeeping problems.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Matthew Ahrens <mahrens@delphix.com>
Co-authored-by: Matthew Ahrens <matthew.ahrens@delphix.com>
Signed-off-by: Paul Dagnelie <pcd@delphix.com>
Closes #12522
Closes #8462

Linux 5.15 compat: standalone <linux/stdarg.h>

Kernel commits

39f75da7bcc8 ("isystem: trim/fixup stdarg.h and other headers")
c0891ac15f04 ("isystem: ship and use stdarg.h")
564f963eabd1 ("isystem: delete global -isystem compile option")

(for now can be found in linux-next.git tree, will land into the
 Linus' tree during the ongoing 5.15 cycle with one of akpm merges)

removed the -isystem flag and disallowed the inclusion of any
compiler header files. They also introduced a minimal
<linux/stdarg.h> as a replacement for <stdarg.h>.
include/os/linux/spl/sys/cmn_err.h in the ZFS source tree includes
<stdarg.h> unconditionally. Introduce a test for <linux/stdarg.h>
and include it instead of the compiler's one to prevent module
build breakage.

Reviewed-by: Tony Nguyen <tony.nguyen@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Lobakin <alobakin@pm.me>
Closes #12531

Linux 5.15 compat: block device readahead

The 5.15 kernel moved the backing_dev_info structure out of
the request queue structure which causes a build failure.

Rather than look in the new location for the BDI we instead
detect this upstream refactoring by the existance of either
the blk_queue_update_readahead() or disk_update_readahead()
functions.  In either case, there's no longer any reason to
manually set the ra_pages value since it will be overridden
with a reasonable default (2x the block size) when
blk_queue_io_opt() is called.

Therefore, we update the compatibility wrapper to do nothing
for 5.9 and newer kernels.  While it's tempting to do the
same for older kernels we want to keep the compatibility
code to preserve the existing behavior.  Removing it would
effectively increase the default readahead to 128k.

Reviewed-by: Tony Nguyen <tony.nguyen@delphix.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #12532

Detect iSCSI in the zpool cmd vdev media script

Reviewed-by: Serapheim Dimitropoulos <serapheim@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Tony Nguyen <tony.nguyen@delphix.com>
Signed-off-by: Don Brady <don.brady@delphix.com>
Closes #12206

CI: don't install abigail-tools

We use docker image instead.

Reviewed-by: John Kennedy <john.kennedy@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: George Melikov <mail@gmelikov.ru>
Closes #12529

Update ABI files via new libabigail version

Reviewed-by: John Kennedy <john.kennedy@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: George Melikov <mail@gmelikov.ru>
Closes #12529

Libabigail: make .abi files more consistent

Reviewed-by: John Kennedy <john.kennedy@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: George Melikov <mail@gmelikov.ru>
Closes #12529

CI: use fresh libabigail via docker image

Reviewed-by: John Kennedy <john.kennedy@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: George Melikov <mail@gmelikov.ru>
Closes #12529

Check for libabigail version

We need to use 1.8.0+ version, older versions
may segfault and give inconsistent results.

Reviewed-by: John Kennedy <john.kennedy@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: George Melikov <mail@gmelikov.ru>
Closes #12529

ZTS: Remove exceptions for flaky zhack on FreeBSD

Issue #11854 has been resolved, so we can remove the exceptions for it.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: John Kennedy <john.kennedy@delphix.com>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Signed-off-by: Ryan Moeller <ryan@iXsystems.com>
Closes #12527

FreeBSD: Don't remove SA xattr if not SA znode

We attempt to remove an existing SA xattr when setting a dir xattr, but
this only makes sense if the znode has been upgraded to the SA format.
Otherwise, we will hit an assert in zfs_sa_get_xattr.

Make sure this is an SA znode before attempting to remove the SA xattr.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Signed-off-by: Ryan Moeller <ryan@iXsystems.com>
Closes #12514

Fix cross-endian interoperability of zstd

It turns out that layouts of union bitfields are a pain, and the
current code results in an inconsistent layout between BE and LE
systems, leading to zstd-active datasets on one erroring out on
the other.

Switch everyone over to the LE layout, and add compatibility code
to read both.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Matthew Ahrens <mahrens@delphix.com>
Signed-off-by: Rich Ercolani <rincebrain@gmail.com>
Closes #12008
Closes #12022

ZTS: Waiting for zvols to be available

The ZTS block_device_wait helper function should use -e when waiting
for a file to appear since it will be either a block special device
or a symlink.  This didn't cause any failures but when a device path
was specified the function would wait longer than needed.

Additionally update the most flakey test cases to pass the file path
to block_device_wait to try and improve the test reliability.  The
udev behavior on Fedora in particular can result in frequent false
positives.

Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: John Kennedy <john.kennedy@delphix.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #12515

Correct checking bdev_check_media_change message

We're not looking for bdev_disk_changed.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Signed-off-by: Ryan Moeller <ryan@iXsystems.com>
Closes #12492

Make 'zpool labelclear -f' work on offlined disks

This patch allows you to clear the label on offlined disks in an active
pool with `-f`.  Previously, labelclear wouldn't let you do that.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Tony Nguyen <tony.nguyen@delphix.com>
Signed-off-by: Tony Hutter <hutter2@llnl.gov>
Closes #12511

vdev_id: Return an error if config file is not found

Signed-off-by: Anton Gubarkov <anton.gubarkov@gmail.com>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>

zpool-remove.8: describe top-level vdev sector size limitation

Document that top-level vdevs cannot be removed unless all top-level
vdevs have the same sector size.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Signed-off-by: Sam Hathaway <sam@sam-hathaway.com>
Closes #11339
Closes #12472

Initialize parity blocks before RAID-Z reconstruction benchmarking

benchmark_raidz() allocates a row to benchmark parity calculation and
reconstruction.  In the latter case, the parity blocks are left
uninitialized, leading to reports from KMSAN.

Initialize parity blocks to 0xAA as we do for the data earlier in the
function.  This does not affect the selected RAID-Z implementation on
any of several systems tested.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Mark Johnston <markj@FreeBSD.org>
Closes #12473

ZTS: Add tests for creation time

Reviewed-by: Tony Nguyen <tony.nguyen@delphix.com>
Reviewed-by: Allan Jude <allan@klarasystems.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Signed-off-by: Ryan Moeller <ryan@iXsystems.com>
Closes #12432

Linux 4.11 compat: statx support

Linux 4.11 added a new statx system call that allows us to expose crtime
as btime. We do this by caching crtime in the znode to match how atime,
ctime and mtime are cached in the inode.

statx also introduced a new way of reporting whether the immutable,
append and nodump bits have been set. It adds support for reporting
compression and encryption, but the semantics on other filesystems is
not just to report compression/encryption, but to allow it to be turned
on/off at the file level. We do not support that.

We could implement semantics where we refuse to allow user modification
of the bit, but we would need to do a dnode_hold() in zfs_znode_alloc()
to find out encryption/compression information. That would introduce
locking that will have a minor (although unmeasured) performance cost.
It also would be inferior to zdb, which reports far more detailed
information. We therefore omit reporting of encryption/compression
through statx in favor of recommending that users interested in such
information use zdb.

Reviewed-by: Tony Nguyen <tony.nguyen@delphix.com>
Reviewed-by: Allan Jude <allan@klarasystems.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Ryan Moeller <ryan@iXsystems.com>
Signed-off-by: Richard Yao <ryao@gentoo.org>
Closes #8507

zfs.4: Fix typo s/compatiblity/compatibility/

Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Ryan Moeller <ryan@ixsystems.com>
Reviewed-by: Tony Nguyen <tony.nguyen@delphix.com>
Signed-off-by: Gordon Bergling <gbergling@googlemail.com>
Closes #12464

Remove b_pabd/b_rabd allocation from arc_hdr_alloc()

When a header is allocated for full overwrite it is a waste of time
to allocate b_pabd/b_rabd for it, since arc_write() will free them
without ever being touched.  If it is a read or a partial overwrite
then arc_read() and arc_hdr_decrypt() allocate them explicitly.

Reduced memory allocation in user threads also reduces ARC eviction
throttling there, proportionally increasing it in ZIO threads, that
is not good.  To minimize or even avoid it introduce ARC allocation
reserve, allowing certain arc_get_data_abd() callers to allocate a
bit longer in situations where user threads will already throttle.

Reviewed-by: George Wilson <gwilson@delphix.com>
Reviewed-by: Mark Maybee <mark.maybee@delphix.com>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Closes #12398

Optimize arc_l2c_only lists assertions

It is very expensive and not informative to call multilist_is_empty()
for each arc_change_state() on debug builds to check for impossible.
Instead implement special index function for arc_l2c_only->arcs_list,
multilists, panicking on any attempt to use it.

Reviewed-by: Mark Maybee <mark.maybee@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Closes #12421

Fix/improve dbuf hits accounting

Instead of clearing stats inside arc_buf_alloc_impl() do it inside
arc_hdr_alloc() and arc_release().  It fixes statistics being wiped
every time a new dbuf is filled from the ARC.

Remove b_l1hdr.b_l2_hits. L2ARC hits are accounted at b_l2hdr.b_hits.
Since the hits are accounted under hash lock, replace atomics with
simple increments.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: George Wilson <george.wilson@delphix.com>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Closes #12422

Avoid vq_lock drop in vdev_queue_aggregate()

vq_lock is already too congested for two more operations per I/O.
Instead of dropping and reacquiring it inside vdev_queue_aggregate()
delegate the zio_vdev_io_bypass() and zio_execute() calls for parent
I/Os to callers, that drop the lock any way to execute the new I/O.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Mark Maybee <mark.maybee@delphix.com>
Reviewed-by: Brian Atkinson <batkinson@lanl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Closes #12297

Use more atomics in refcounts

Use atomic_load_64() for zfs_refcount_count() to prevent torn reads
on 32-bit platforms.  On 64-bit ones it should not change anything.

When built with ZFS_DEBUG but running without tracking enabled use
atomics instead of mutexes same as for builds without ZFS_DEBUG.
Since rc_tracked can't change live we can check it without lock.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Matthew Ahrens <mahrens@delphix.com>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Closes #12420

ZTS: Avoid unset $tmpdir in redacted_panic

The redacted_send tests make use of a $tmpdir variable, except in
redacted_send/redacted_panic the variable is never defined.

Use $TEST_BASE_DIR instead.

Clean up the stream file after the test.

Reviewed-by: John Kennedy <john.kennedy@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Ryan Moeller <ryan@iXsystems.com>
Closes #12455

Restore FreeBSD sysctl processing for arc.min and arc.max

Before OpenZFS 2.0, trying to set the FreeBSD sysctl vfs.zfs.arc_max
to a disallowed value would return an error.
Since the switch, it instead only generates WARN_IF_TUNING_IGNORED

Keep the ability to set the sysctl's specifically to 0, even though
that is less than the minimum, because some tests depend on this.

Also lost, was the ability to set vfs.zfs.arc_max to a value less
than the default vfs.zfs.arc_min at boot time. Restore this as well.

Reviewed-by: Tony Nguyen <tony.nguyen@delphix.com>
Reviewed-by: Ryan Moeller <ryan@ixsystems.com>
Signed-off-by: Allan Jude <allan@klarasystems.com>
Closes #12161

zfs: add missed dependency of zfs module on zlib

Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Martin Matuska <mm@FreeBSD.org>
Co-authored-by: Konstantin Belousov <kib@FreeBSD.org>
Signed-off-by: Ryan Moeller <ryan@iXsystems.com>
External-issue: https://reviews.freebsd.org/D31207
Closes #12442

Add zfs.sh -r flag to reload modules

zfs.sh already can load and unload, so why not both?

This is convenient when developing changes to the module and you want
to rapidly make some changes, rebuild the module, reload the module,
and test the changes.

Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Ryan Moeller <ryan@iXsystems.com>
Closes #12450

Fix usage of find in tests/Makefile.am

The path is not optional on FreeBSD.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Tony Nguyen <tony.nguyen@delphix.com>
Signed-off-by: Ryan Moeller <ryan@iXsystems.com>
Closes #12453

Run arc_evict thread at higher priority

Run arc_evict thread at higher priority, nice=0, to give it more CPU
time which can improve performance for workload with high ARC evict
activities.

On mixed read/write and sequential read workloads, I've seen between
10-40% better performance.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Signed-off-by: Tony Nguyen <tony.nguyen@delphix.com>
Closes #12397

Make get_key_material_file fail more verbosely

It turns out, there are a lot of possible reasons for fopen to fail.
Let's share which reason we failed for today.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Tony Nguyen <tony.nguyen@delphix.com>
Signed-off-by: Rich Ercolani <rincebrain@gmail.com>
Closes #12410

Enable /proc/diskstats for zvols

The /proc/diskstats accounting needs to be explicitly enabled
for block devices which do not use multi-queue.

Reviewed-by: Tony Nguyen <tony.nguyen@delphix.com>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #12440
Closes #12066

Man zpool-scrub.8: describe sequential scrub

Describe sequential scrub and add examples of scrub status.

Reviewed-by: Richard Laager <rlaager@wiktel.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Tony Nguyen <tony.nguyen@delphix.com>
Signed-off-by: George Melikov <mail@gmelikov.ru>
Closes #12429

Modify checksum obtain method of QAT

CpaDcGeneratefooter function that obtain the checksum code
does not support the CPA_DC_STATELESS mode. So we get the
adler32 chencksum of the end of the zlib from dc_results.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Chengfei Zhu <chengfeix.zhu@intel.com>
Signed-off-by: hedong.zhang <h_d_zhang@163.com>
Closes #12343

Allow disabling of unmapped I/O on FreeBSD

We have a tunable which permits one to disable the use of unmapped I/O
for the buffer cache.  Respect it in ZFS as well.  This is useful for
KMSAN, which cannot easily maintain shadow state for unmapped pages.

No functional change intended, as unmapped I/O is permitted by default
and there's no real reason to disable it in practice except for
debugging.

Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Tony Nguyen <tony.nguyen@delphix.com>
Reviewed-by: Ryan Moeller <ryan@iXsystems.com>
Signed-off-by: Mark Johnston <markj@FreeBSD.org>
Closes #12446

Add comment on metaslab_class_throttle_reserve() locking

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Issue #12314
Closes #12419

Assorted fixes for the performance tests

- Bail out early if we're running the perf tests and forget to
  specify disks.
- Allow perf tests to run with any number of disks.
- Remove weekly vs. nightly settings
- Move variables with common values to perf.shlib
- Use zinject to clear the ARC over export/import
- Fix dbuf cache size calculation

When the meaning of `dbuf_cache_max_bytes` changed, the performance
test that covers the dbuf cache started to fail. The test would try to
write files for the test using the max possible size of the cache,
inevitably filling the pool and failing. This change uses
`dbuf_cache_shift` to correctly calculate the dbuf cache size.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Tony Nguyen <tony.nguyen@delphix.com>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Signed-off-by: John Kennedy <john.kennedy@delphix.com>
Closes #12408

Read past end of argv array in zpool_do_import()

`zpool_do_import()` passes `argv[0]`, (optionally) `argv[1]`, and
`pool_specified` to `import_pools()`.  If `pool_specified==FALSE`, the
`argv[]` arguments are not used.  However, these values may be off the
end of the `argv[]` array, so loading them could dereference unmapped
memory.  This error is reported by the asan build:

```
=================================================================
==6003==ERROR: AddressSanitizer: heap-buffer-overflow
READ of size 8 at 0x6030000004a8 thread T0
    #0 0x562a078b50eb in zpool_do_import zpool_main.c:3796
    #1 0x562a078858c5 in main zpool_main.c:10709
    #2 0x7f5115231bf6 in __libc_start_main
    #3 0x562a07885eb9 in _start

0x6030000004a8 is located 0 bytes to the right of 24-byte region
allocated by thread T0 here:
    #0 0x7f5116ac6b40 in __interceptor_malloc
    #1 0x562a07885770 in main zpool_main.c:10699
    #2 0x7f5115231bf6 in __libc_start_main
```

This commit passes NULL for these arguments if they are off the end
of the `argv[]` array.

Reviewed-by: George Wilson <gwilson@delphix.com>
Reviewed-by: John Kennedy <john.kennedy@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Allan Jude <allan@klarasystems.com>
Signed-off-by: Matthew Ahrens <mahrens@delphix.com>
Closes #12339

Add missing properties to zfs allow manpage

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Václav Skála <skala@vshosting.cz>
Closes #12402

Fixes in persistent L2ARC

In l2arc_add_vdev() first decide whether the device is eligible for
L2ARC rebuild or whole device trim and then add it to the list of cache
devices. Otherwise l2arc_feed_thread() might already start writing on
the device invalidating previous content as l2ad_hand = l2ad_start.
However l2arc_rebuild_vdev() needs the device present in the cache
device list to figure out its l2arc_dev_t. Fix this by moving most of
l2arc_rebuild_vdev() in a new function l2arc_rebuild_dev() which does
not need to search in the cache device list.

In contrast to l2arc_add_vdev() we do not have to worry about
l2arc_feed_thread() invalidating previous content when onlining a
cache device. The device parameters (l2ad*) are not cleared when
offlining the device and writing new buffers will not invalidate
all previous content. In worst case only buffers that have not had
their log block written to the device will be lost.

Retire persist_l2arc_00{4,5,8} tests since they cover code already
covered by the remaining ones. Test persist_l2arc_006 is renamed to
persist_l2arc_004 and persist_l2arc_007 is renamed to persist_l2arc_005.

Fix a typo in persist_l2arc_004, and remove an assertion that is not
always true from l2arc_arcstats_pos. Also update an assertion in
persist_l2arc_005 and explain why in a comment.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: George Amanakis <gamanakis@gmail.com>
Closes #12365

Initialize dn_next_type[] in the dnode constructor

It seems nothing ensures that this array is zeroed when a dnode is
freshly allocated, so in principle it retains the values from the
previous allocation.  In practice it seems to be the case that the
fields should end up zeroed, but we can zero the field anyway for
consistency.

This was found using KMSAN.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Signed-off-by: Mark Johnston <markj@FreeBSD.org>
Closes #12383

Zero pad bytes following TX_WRITE log data

When logging a TX_WRITE record in the case where file data has to be
copied from the DMU, we pad the log record size to a multiple of 8
bytes.  In this case, any padding bytes should be zeroed, otherwise the
contents of uninitialized memory are written to the ZIL.

This was found using KMSAN.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Signed-off-by: Mark Johnston <markj@FreeBSD.org>
Closes #12383

Zero pad bytes when allocating a ZIL record

When allocating a record, we round up the allocation size to a multiple
of 8.  In this case, any padding bytes should be zeroed, otherwise the
contents of uninitialized memory are written to the ZIL.

This was found using KMSAN.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Signed-off-by: Mark Johnston <markj@FreeBSD.org>
Closes #12383

Initialize all fields in zfs_log_xvattr()

When logging TX_SETATTR, we could otherwise fail to initialize part of
the corresponding ZIL record depending on which fields are present in
the xvattr.  Initialize the creation time and the AV scan timestamp to
zero so that uninitialized bytes are not written to the ZIL.

This was found using KMSAN.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Signed-off-by: Mark Johnston <markj@FreeBSD.org>
Closes #12383

Initialize "autoreplace" in spa_ld_get_props()

spa_prop_find() may fail to find the specified property, in which case
it suppresses ENOENT from zap_lookup().  In this case, the return value
is left uninitialized, so spa_autoreplace was being initialized using an
uninitialized stack variable.

This was found using KMSAN.  It appears to be a regression from commit
9eb7b46ed0, which removed the initialization of "autoreplace" from the
definition.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Signed-off-by: Mark Johnston <markj@FreeBSD.org>
Closes #12383

Linux 5.14 compat: explicity assign set_page_dirty

Kernel 5.14 introduced a change where set_page_dirty of
struct address_space_operations is no longer implicitly set to
__set_page_dirty_buffers(), which ended up resulting in a NULL
pointer deref in the kernel when it is attempted to be called.
This change sets .set_page_dirty in the structure to
__set_page_dirty_nobuffers(), which was introduced with the
related patch set. The breaking change was introduce in commit
0af573780b0b13fceb7fabd49dc1b073cee9a507 to torvalds/linux.git.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Coleman Kane <ckane@colemankane.org>
Closes #12427

Fix unfortunate NULL in spa_update_dspace

After 1325434b, we can in certain circumstances end up calling
spa_update_dspace with vd->vdev_mg NULL, which ends poorly during
vdev removal.

So let's not do that further space adjustment when we can't.

Reviewed-by: Matthew Ahrens <mahrens@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rich Ercolani <rincebrain@gmail.com>
Closes #12380
Closes #12428

Linux 5.14 compat: blk_alloc_disk()

In Linux 5.14, blk_alloc_queue is no longer exported, and its usage
has been superseded by blk_alloc_disk, which returns a gendisk struct
from which we can still retrieve the struct request_queue* that is
needed in the one place where it is used. This also replaces the call
to alloc_disk(minors), and minors is now set via struct member
assignment.

Reviewed-by: Tony Nguyen <tony.nguyen@delphix.com>
Reviewed-by: Olaf Faaland <faaland1@llnl.gov>
Reviewed-by: Coleman Kane <ckane@colemankane.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #12362
Closes #12409

zloop: Add a max iterations option, use default run/pass times

It is useful to have control over the number of iterations of zloop so
we can easily produce "x core dumps found *in y iterations*" metrics.

Using random values for run/pass times doesn't improve coverage in a
meaningful way.

Randomizing run time could be seen as a compromise between running a
greater variety of shorter tests versus a smaller variety of longer
tests within a fixed time span.  However, it is not desirable when
running a fixed number of iterations.

Pass time already incorporates randomness within ztest.

Either parameter can be passed to ztest explicitly if the defaults are
not satisfactory.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: John Kennedy <john.kennedy@delphix.com>
Signed-off-by: Ryan Moeller <ryan@iXsystems.com>
Closes #12411

FreeBSD: Ignore make_dev_s() errors

Since errors returned by zvol_create_minor_impl() are ignored by the
common code, it is more convenient to ignore make_dev_s() errors there.
It allows, for example, to get device created for the zvol after later
rename instead of having it further stuck in half-created state.
zvol_rename_minor() already ignores those errors.

While there, switch from MAXPHYS to maxphys in FreeBSD 13+.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Tony Nguyen <tony.nguyen@delphix.com>
Reviewed-by: Ryan Moeller <ryan@iXsystems.com>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Closes #12375

Remove old orig_fd variable from zfs send

Possibly required in the past, but is currently fills no purpose.
Ordinarily such tiny cleanup is not generally worth it, however
on the macOS port, in a future commit, we do unspeakable things to the
"fd" for send/recv, and it would be easier to only have to deal with
one "fd" instead of two.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Tony Nguyen <tony.nguyen@delphix.com>
Signed-off-by: Jorgen Lundman <lundman@lundman.net>
Closes #12404

Optimize allocation throttling

Remove mc_lock use from metaslab_class_throttle_*().  The math there
is based on refcounts and so atomic, so the only race possible there
is between zfs_refcount_count() and zfs_refcount_add().  But in most
cases metaslab_class_throttle_reserve() is called with the allocator
lock held, which covers the race.  In cases where the lock is not
held, GANG_ALLOCATION() or METASLAB_MUST_RESERVE are set, and so we
do not use zfs_refcount_count().  And even if we assume some other
non-existing scenario, the worst that may happen from this race is
few more I/Os get to allocation earlier, that is not a problem.

Move locks and data of different allocators into different cache
lines to avoid false sharing.  Group spa_alloc_* arrays together
into single array of aligned struct spa_alloc spa_allocs.  Align
struct metaslab_class_allocator.

Reviewed-by: Paul Dagnelie <pcd@delphix.com>
Reviewed-by: Ryan Moeller <ryan@iXsystems.com>
Reviewed-by: Don Brady <don.brady@delphix.com>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Closes #12314

CI: generate ABI files if changed

So commit author can just download them as
artifacts and commit.

Reviewed-by: Ryan Moeller <ryan@ixsystems.com>
Reviewed-by: John Kennedy <john.kennedy@delphix.com>
Signed-off-by: George Melikov <mail@gmelikov.ru>
Closes #12379

Minor ARC optimizations

Remove unneeded global, practically constant, state pointer variables
(arc_anon, arc_mru, etc.), replacing them with macros of real state
variables addresses (&ARC_anon, &ARC_mru, etc.).

Change ARC_EVICT_ALL from -1ULL to UINT64_MAX, not requiring special
handling in inner loop of ARC reclamation.  Respectively change bytes
argument of arc_evict_state() from int64_t to uint64_t.

Reviewed-by: Matthew Ahrens <mahrens@delphix.com>
Reviewed-by: Mark Maybee <mark.maybee@delphix.com>
Reviewed-by: Ryan Moeller <ryan@iXsystems.com>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Closes #12348

dmu_redact.c does not call bqueue_destroy

Ensure all calls to bqueue_init() has a corresponding call to bqueue_destroy()

Reviewed-by: Paul Dagnelie <pcd@delphix.com>
Co-authored-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Jorgen Lundman <lundman@lundman.net>
Closes #12118

A few fixes of callback typecasting (for the upcoming ClangCFI)

* zio: avoid callback typecasting
* zil: avoid zil_itxg_clean() callback typecasting
* zpl: decouple zpl_readpage() into two separate callbacks
* nvpair: explicitly declare callbacks for xdr_array()
* linux/zfs_nvops: don't use external iput() as a callback
* zcp_synctask: don't use fnvlist_free() as a callback
* zvol: don't use ops->zv_free() as a callback for taskq_dispatch()

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Mark Maybee <mark.maybee@delphix.com>
Signed-off-by: Alexander Lobakin <alobakin@pm.me>
Closes #12260

Remove unused fields from zvol_task_t

We don't use or need the pool name or value source in the zvol tasks.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Signed-off-by: Ryan Moeller <ryan@iXsystems.com>
Closes #12361

FreeBSD: Switch from MAXPHYS to maxphys on FreeBSD 13+

Reviewed-by: Allan Jude <allan@klarasystems.com>
Reviewed-by: Ryan Moeller <ryan@iXsystems.com>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Closes #12378

zpool_influxdb: fix -Werror=stringop-truncation

Use strlcpy instead of problematic strncpy

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Richard Elling <Richard.Elling@RichardElling.com>
Reviewed-by: Ryan Moeller <ryan@iXsystems.com>
Signed-off-by: George Melikov <mail@gmelikov.ru>
Closes #12344

Correct zfs-send(8) on readonly sends

zfs-send(8) claimed in the flags list you could use -pR when sending
a readonly filesystem or volume. You cannot.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Tony Nguyen <tony.nguyen@delphix.com>
Signed-off-by: Rich Ercolani <rincebrain@gmail.com>
Closes #12336

Introduce dsl_dir_diduse_transfer_space()

Most of dsl_dir_diduse_space() and dsl_dir_transfer_space() CPU time
is a dd_lock overhead and time spent in dmu_buf_will_dirty(). Calling
them one after another is a waste of time and even more contention.
Doing that twice for each rewritten block within dbuf_write_done()
via dsl_dataset_block_kill() and dsl_dataset_block_born() created one
of the biggest CPU overheads in case of small blocks rewrite.

dsl_dir_diduse_transfer_space() combines functionality of these two
functions for cases where it is needed, but without double overhead,
practically for the cost of dsl_dir_diduse_space() or even cheaper.

While there, optimize dsl_dir_phys() calls in dsl_dir_diduse_space()
and dsl_dir_transfer_space().  It seems Clang detects some aliasing
there, repeating dd->dd_dbuf->db_data dereference multiple times,
increasing dd_lock scope and contention.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Matthew Ahrens <mahrens@delphix.com>
Author: Ryan Moeller <ryan@iXsystems.com>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Closes #12300

config/libatomic: require -latomic iff atomic.c doesn't link w/o it

In absence of LTO, and dynamic libatomic, la.so ends up in the needs
section of every toolchain executable; some consider this an issue.

Reviewed-by: Ryan Moeller <ryan@iXsystems.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Ahelenia Ziemiańska <nabijaczleweli@nabijaczleweli.xyz>
Closes #12345
Closes #12359

Tinker with slop space accounting with dedup

* Tinker with slop space accounting with dedup

Do not include the deduplicated space usage in the slop space
reservation, it leads to surprising outcomes.

* Update spa_dedup_dspace sometimes

Sometimes, we get into spa_get_slop_space() with
spa_dedup_dspace=~0ULL, AKA "unset", while spa_dspace is correctly set.

So call the code to update it before we use it if we hit that case.

Reviewed-by: Matthew Ahrens <mahrens@delphix.com>
Reviewed-by: Mark Maybee <mark.maybee@delphix.com>
Signed-off-by: Rich Ercolani <rincebrain@gmail.com>
Closes #12271

Fix ARC ghost states eviction accounting

arc_evict_hdr() returns number of evicted bytes in scope of specific
state.  For ghost states it does not mean the amount of really freed
memory, but the logical buffer size.  It is correct for the eviction
process, but not for waking up threads waiting for ARC size reduction,
as added in "Revise ARC shrinker algorithm" commit, causing premature
wakeups while ARC is still overflowed, allowing even bigger overflow,
plus processing overhead when next allocation will also get blocked,
probably also for too short time.

To fix that make arc_evict_hdr() also return the amount of really
freed memory, which for the ghost states is only the header, and use
it to update arc_evict_count instead.  Originally I was thinking to
not return it at all, since arc_get_data_impl() does not account for
the headers, but decided that some slow allocation progress is better
than long waits, reaching on my tests up to 100ms.

To reduce negative latency effects of long time periods when reclaim
thread can free little real memory, start reclamation process earlier,
before we actually reached the overflow threshold, when we have to
throttle new allocations.  We can also do it without taking global
arc_evict_lock, reducing the contention.

Reviewed-by: George Wilson <gwilson@delphix.com>
Reviewed-by: Allan Jude <allan@klarasystems.com>
Reviewed-by: Ryan Moeller <ryan@iXsystems.com>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Closes #12279

Update bug report template

- Remove the "SPL Version" line, the repositories have been merged
  since the 0.8 release and we no longer need to ask about this.

- Simply ask for the kernel version / patch level and add a hint
  about how to get this information on Linux and FreeBSD.

- Remove "Status: Triage Needed" from the template, in practice
  we really haven't been using this label so let's step setting it.

Reviewed-by: Matthew Ahrens <mahrens@delphix.com>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Ryan Moeller <ryan@iXsystems.com>
Reviewed-by: John Kennedy <john.kennedy@delphix.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes: #12340

file reference counts can get corrupted

Callers of zfs_file_get and zfs_file_put can corrupt the reference
counts for the file structure resulting in a panic or a soft lockup.
When zfs send/recv runs, it will add a reference count to the
open file, and begin to send or recv the stream. If the file descriptor
is closed, then when dmu_recv_stream() or dmu_send() return we will
call zfs_file_put to remove the reference we placed on the file
structure. Unfortunately, because zfs_file_put() uses the file
descriptor to lookup the file structure, it may end up finding that
the file descriptor table no longer contains the file struct, thus
leaking the file structure. Or it might end up finding a file
descriptor for a different file and blindly updating its reference
counts. Other failure modes probably exists.

This change reworks the zfs_file_[get|put] interface to not rely
on the file descriptor but instead pass the zfs_file_t pointer around.

Reviewed-by: Matthew Ahrens <mahrens@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Mark Maybee <mark.maybee@delphix.com>
Reviewed-by: Ryan Moeller <ryan@iXsystems.com>
Co-authored-by: Allan Jude <allan@klarasystems.com>
Signed-off-by: George Wilson <gwilson@delphix.com>
External-issue: DLPX-76119
Closes #12299

dprintf_dnode: strcpy -> strlcpy

Missed a couple of strcpy() in earlier commit, this is only used with
--enable-debug.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Tony Nguyen <tony.nguyen@delphix.com>
Signed-off-by: Jorgen Lundman <lundman@lundman.net>
Closes #12311

Replace strchrnul() with strrchr()

Could have gone either way with this one, either adding it to
macOS/Windows SPL, or returning it to "classic" usage with strrchr().
Since the new special way isn't really used, and only used once,
we have this commit.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Ryan Moeller <ryan@iXsystems.com>
Signed-off-by: Jorgen Lundman <lundman@lundman.net>
Closes  #12312

FreeBSD: Use unmapped I/O for scattered/gang ABD buffers

Many FreeBSD disk drivers support "unmapped" I/O mode, when data
buffer represented not with a virtually contiguous KVA-mapped address
range, but with a list of physical memory pages.  Originally it was
designed to do I/O from buffers without KVA mapping (unmapped).  But
moving virtual addresses out of equation allows us to operate even
non-contiguous data buffers with one condition: all buffer discon-
tinuities must be aligned to memory page borders.

Doing I/O to capable GEOM device this patch traverses through non-
linear ABD buffers, validating the chunks borders.  If the condition
is met, it supplies GEOM with the list of original physical memory
pages instead of copying the data into temporary contiguous buffer.
On capable hardware on pools with ashift=12 and default ABD chunk of
4KB it should handle all the I/O without additional memory copying.

Reviewed-by: Brian Atkinson <batkinson@lanl.gov>
Reviewed-by: Ryan Moeller <ryan@iXsystems.com>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Closes #12320

FreeBSD: Hardcode abd_chunk_size to PAGE_SIZE

It makes no sense to set it below PAGE_SIZE, since it increases all
overheads and makes returning memory to OS problematic.  It makes no
sense to set it above PAGE_SIZE, since such allocations and especially
frees are too expensive and cause KVA fragmentation to benefit from
fewer chunks.  After that it makes no sense to keep more complicated
math here.

What may have sense though is just a tunable border between linear and
scatter ABDs, previously also controlled by this tunable.  Retain that
functionality by taking abd_scatter_min_size tunable from Linux, just
with different default value.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Brian Atkinson <batkinson@lanl.gov>
Reviewed-by: Ryan Moeller <ryan@iXsystems.com>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Closes #12328