Disable history_004_pos

Occasionally observed failure of history_004_pos due to the test
case not being 100% reliable.  In order to prevent false positives
disable this test case until it can be made reliable.

Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #7026
Closes #7028

Fix incompatibility with Reiser4 patched kernels

In ZFSOnLinux, our sources and build system are self contained such that
we do not need to make changes to the Linux kernel sources. Reiser4 on
the other hand exists solely as a kernel tree patch and opts to make
changes to the kernel rather than adapt to it. After Linux 4.1 made a
VFS change that replaced new_sync_read with do_sync_read, Reiser4's
maintainer decided to modify the kernel VFS to export the old function.
This caused our autotools check to misidentify the kernel API as
predating Linux 4.1 on kernels that have been patched with Reiser4
support, which breaks our build.

Reiser4 really should be patched to stop doing this, but lets modify our
check to be more strict to help the affected users of both filesystems.

Also, we were not checking the types of arguments and return value of
new_sync_read() and new_sync_write() . Lets fix that too.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Signed-off-by: Richard Yao <ryao@gentoo.org>
Closes #6241
Closes #7021

Use zap_count instead of cached z_size for unlink

As a performance optimization Lustre does not strictly update
the SA_ZPL_SIZE when adding/removing from non-directory entries.
This results in entries which cannot be removed through the ZPL
layer even though the ZAP is empty and safe to remove.

Resolve this issue by checking the zap_count() directly instead
on relying on the cached SA_ZPL_SIZE.  Micro-benchmarks show no
significant performance impact due to the additional overhead
of using zap_count().

Reviewed-by: Olaf Faaland <faaland1@llnl.gov>
Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Signed-off-by: Alex Zhuravlev <alexey.zhuravlev@intel.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #7019

Revert raidz_map and _col structure types

As part of the refactoring of ab9f4b0b824ab4cc64a4fa382c037f4154de12d6,
several uint64_t-s and uint8_t-s were changed to other types.  This
caused ZoL github issue #6981, an overflow of a size_t on a 32-bit ARM
machine.  In absense of any strong motivation for the type changes, this
simply puts them back, modulo the changes accumulated for ABD.

Compile-tested on amd64 and run-tested on armhf.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Gvozden Neskovic <neskovic@gmail.com>
Signed-off-by: Nathaniel Wesley Filardo <nwf@cs.jhu.edu>
Closes #6981
Closes #7023

Fix unused variable warnings

Resolved unused variable warnings observed after restricting
-Wno-unused-but-set-variable to only libzfs and libzpool.

Reviewed-by: DHE <git@dehacked.net>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #6941

Fix ztest_verify_dnode_bt() test case

In ztest_verify_dnode_bt the ztest_object_lock must be held in
order to safely verify the unused bonus space.

Reviewed-by: DHE <git@dehacked.net>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #6941

Fix -fsanitize=address memory leak

kmem_alloc(0, ...) in userspace returns a leakable pointer.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: DHE <git@dehacked.net>
Issue #6941

Fix percentage styling in zfs-module-parameters.5

Replace "percent" with "%", add bold to default values.

Reviewed-by: bunder2015 <omfgbunder@gmail.com>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: George Amanakis <gamanakis@gmail.com>
Closes #7018

Reduce codecov PR comments

Attempt to reduce the number of comments posted by codecov
to PR requests.  Based on the codecov documenation setting
"require_changes=yes" and "behavior=once" should result in
a single comment under most circumstances.

https://docs.codecov.io/v4.3.6/docs/pull-request-comments

Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #7022
Closes #7025

zhack: fix getopt return type

This fixes zhack's command processing on ARM.  On ARM char
is unsigned, and so, in promotion to an int, it will never
compare equal to -1.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Nathaniel Wesley Filardo <nwf@cs.jhu.edu>
Closes #7016

Fix ARC hit rate

When the compressed ARC feature was added in commit d3c2ae1
the method of reference counting in the ARC was modified.  As
part of this accounting change the arc_buf_add_ref() function
was removed entirely.

This would have be fine but the arc_buf_add_ref() function
served a second undocumented purpose of updating the ARC access
information when taking a hold on a dbuf.  Without this logic
in place a cached dbuf would not migrate its associated
arc_buf_hdr_t to the MFU list.  This would negatively impact
the ARC hit rate, particularly on systems with a small ARC.

This change reinstates the missing call to arc_access() from
dbuf_hold() by implementing a new arc_buf_access() function.

Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Tim Chase <tim@chase2k.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #6171
Closes #6852
Closes #6989

Fix 'zpool add' handling of nested interior VDEVs

When replacing a faulted device which was previously handled by a spare
multiple levels of nested interior VDEVs will be present in the pool
configuration; the following example illustrates one of the possible
situations:

   NAME                          STATE     READ WRITE CKSUM
   testpool                      DEGRADED     0     0     0
     raidz1-0                    DEGRADED     0     0     0
       spare-0                   DEGRADED     0     0     0
         replacing-0             DEGRADED     0     0     0
           /var/tmp/fault-dev    UNAVAIL      0     0     0  cannot open
           /var/tmp/replace-dev  ONLINE       0     0     0
         /var/tmp/spare-dev1     ONLINE       0     0     0
       /var/tmp/safe-dev         ONLINE       0     0     0
   spares
     /var/tmp/spare-dev1         INUSE     currently in use

This is safe and allowed, but get_replication() needs to handle this
situation gracefully to let zpool add new devices to the pool.

Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: loli10K <ezomori.nozomu@gmail.com>
Closes #6678
Closes #6996

OpenZFS 8909 - 8585 can cause a use-after-free kernel panic

Authored by: Prakash Surya <prakash.surya@delphix.com>
Reviewed by: John Kennedy <jwk404@gmail.com>
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Brad Lewis <brad.lewis@delphix.com>
Reviewed by: Igor Kozhukhov <igor@dilos.org>
Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov>
Approved by: Robert Mustacchi <rm@joyent.com>
Ported-by: Prakash Surya <prakash.surya@delphix.com>
PROBLEM
=======

There's a race condition that exists if `zil_free_lwb` races with either
`zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`.

Here's an example panic due to this bug:

    > ::status
    debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40
    operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc)
    image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513
    panic message:
    BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference
    dump content: kernel pages only

    > $c
    zio_shrink+0x12()
    zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20)
    zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8)
    zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8)
    zil_commit+0x80(ffffff03dcd15cc0, 9a9)
    zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0)
    fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0)
    write+0x250(42, fffffd7ff4832000, 2000)
    sys_syscall+0x177()

If there's an outstanding lwb that's in `zil_commit_waiter_timeout`
waiting to timeout, waiting on it's waiter's CV, we must be sure not to
call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB
may be freed and can result in a use-after-free situation where the
stale lwb pointer stored in the `zil_commit_waiter_t` structure of the
thread waiting on the waiter's CV is used.

A similar situation can occur if an lwb is issued to disk, and thus in
the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the
disk is servicing that lwb. In this situation, the lwb will be freed by
`zil_free_lwb`, which will result in a use-after-free situation when the
lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called.

This race condition is prevented in `zil_close` by calling `zil_commit`
before `zil_free_lwb` is called, which will ensure all outstanding (i.e.
all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states)
reach the `LWB_STATE_DONE` state before the lwb's are freed
(`zil_commit` will not return untill all the lwb's are
`LWB_STATE_DONE`).

Further, this race condition is prevented in `zil_sync` by only calling
`zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set.
All lwb's not in the `LWB_STATE_DONE` state will have a non-null value
for this pointer; the pointer is only cleared in
`zil_lwb_flush_vdevs_done`, at which point the lwb's state will be
changed to `LWB_STATE_DONE`.

This race *is* present in `zil_suspend`, leading to this bug.

At first glance, it would appear as though this would not be true
because `zil_suspend` will call `zil_commit`, just like `zil_close`, but
the problem is that `zil_suspend` will set the zilog's `zl_suspend`
field prior to calling `zil_commit`. Further, in `zil_commit`, if
`zl_suspend` is set, `zil_commit` will take a special branch of logic
and use `txg_wait_synced` instead of performing the normal `zil_commit`
logic.

This call to `txg_wait_synced` might be good enough for the data to
reach disk safely before it returns, but it does not ensure that all
outstanding lwb's reach the `LWB_STATE_DONE` state before it returns.
This is because, if there's an lwb "stuck" in
`zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will
maintain a non-null value for it's `lwb_buf` field and thus `zil_sync`
will not free that lwb. Thus, even though the lwb's data is already on
disk, the lwb will be left lingering, waiting on the CV, and will
eventually timeout and be issued to disk even though the write is
unnecessary.

So, after `zil_commit` is called from `zil_suspend`, we incorrectly
assume that there are not outstanding lwb's, and proceed to free all
lwb's found on the zilog's lwb list. As a result, we free the lwb that
will later be used `zil_commit_waiter_timeout`.

SOLUTION
========

The solution to this, is to ensure all outstanding lwb's complete before
calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch
accomplishes this goal by forcing the normal `zil_commit` logic when
called from `zil_sync`.

Now, `zil_suspend` will call `zil_commit_impl` which will always use the
normal logic of waiting/issuing lwb's to disk before it returns. As a
result, any lwb's outstanding when `zil_commit_impl` is called will be
guaranteed to reach the `LWB_STATE_DONE` state by the time it returns.

Further, no new lwb's will be created via `zil_commit` since the zilog's
`zl_suspend` flag will be set. This will force all new callers of
`zil_commit` to use `txg_wait_synced` instead of creating and issuing
new lwb's.

Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is
called will be in the `LWB_STATE_DONE` state, and we'll avoid this race
condition.

OpenZFS-issue: https://www.illumos.org/issues/8909
OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d
Closes #6940

Call commit callbacks from the tail of the list

Our zfs backed Lustre MDT had soft lockups while under heavy metadata
workloads while handling transaction callbacks from osd_zfs.

The problem is zfs is not taking advantage of the fast path in
Lustre's trans callback handling, where Lustre will skip the calls
to ptlrpc_commit_replies() when it already saw a higher transaction
number.

This patch corrects this, it also has a positive impact on metadata
performance on Lustre with osd_zfs, plus some cleanup in the headers.

A similar issue for ext4/ldiskfs is described on:
https://jira.hpdd.intel.com/browse/LU-6527

Reviewed-by: Olaf Faaland <faaland1@llnl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Li Dongyang <dongyang.li@anu.edu.au>
Closes #6986

Remove empty files accidentally added by a8b2e306

This patch simply removes 2 empty files that were accidentally
added a part of the scrub priority patch.

Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Signed-off-by: Tom Caputi <tcaputi@datto.com>
Closes #6990

Support re-prioritizing asynchronous prefetches

When sequential scrubs were merged, all calls to arc_read()
(including prefetch IOs) were given ZIO_PRIORITY_ASYNC_READ.
Unfortunately, this behaves badly with an existing issue where
prefetch IOs cannot be re-prioritized after the issue. The
result is that synchronous reads end up in the same vdev_queue
as the scrub IOs and can have (in some workloads) multiple
seconds of latency.

This patch incorporates 2 changes. The first ensures that all
scrub IOs are given ZIO_PRIORITY_SCRUB to allow the vdev_queue
code to differentiate between these I/Os and user prefetches.
Second, this patch introduces zio_change_priority() to provide
the missing capability to upgrade a zio's priority.

Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tom Caputi <tcaputi@datto.com>
Closes #6921
Closes #6926

vdev_id: new slot type ses

This extends vdev_id to support a new slot type, ses, for SCSI Enclosure
Services.  With slot type ses, the disk slot numbers are determined by
using the device slot number reported by sg_ses for the device with
matching SAS address, found by querying all available enclosures.

This is primarily of use on systems with a deficient driver omitting
support for bay_identifier in /sys/devices.  In my testing, I found that
the existing slot types of port and id were not stable across disk
replacement, so an alternative was required.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Simon Guest <simon.guest@tesujimath.org>
Closes #6956

Handle broken pipes in arc_summary

Using a command similar to 'arc_summary.py | head' causes
a broken pipe exception. Gracefully exit in the case of a
broken pipe in arc_summary.py.

Reviewed-by: Richard Elling <Richard.Elling@RichardElling.com>
Reviewed-by: loli10K <ezomori.nozomu@gmail.com>
Signed-off-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Closes #6965
Closes #6969

Handle invalid options in arc_summary

If an invalid option is provided to arc_summary.py we handle any error
thrown from the getopt Python module and print the usage help message.

Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: loli10K <ezomori.nozomu@gmail.com>
Closes #6983

OpenZFS 8794 - cstyle generates warnings with recent perl

Authored by: Dominik Hassler <hadfl@omniosce.org>
Reviewed by: Andy Fiddaman <andy@omniosce.org>
Reviewed by: Igor Kozhukhov <igor@dilos.org>
Reviewed by: Toomas Soome <tsoome@me.com>
Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov>
Approved by: Dan McDonald <danmcd@joyent.com>
Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov>
OpenZFS-issue: https://www.illumos.org/issues/8794
OpenZFS-commit: https://github.com/openzfs/openzfs/commit/578f67364c
Closes #6973

ZTS: Fix create-o_ashift test case

The function that fills the uberblock ring buffer on every device label
has been reworked to avoid occasional failures caused by a race
condition that prevents 'zpool sync' from writing some uberblock
sequentially: this happens when the pool sync ioctl dispatch code calls
txg_wait_synced() while we're already waiting for a TXG to sync.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: loli10K <ezomori.nozomu@gmail.com>
Closes #6924
Closes #6977

Fix multihost stale cache file import

When the multihost property is enabled it should be impossible to
import an active pool even using the force (-f) option.  This patch
prevents a forced import from succeeding when importing with a
stale cache file.

The root cause of the problem is that the kernel modules trusted
the hostid provided in configuration.  This is always correct when
the configuration is generated by scanning for the pool.  However,
when using an existing cache file the hostid could be stale which
would result in the activity check being skipped.

Resolve the issue by always using the hostid read from the label
configuration where the best uberblock was found.

Reviewed-by: Olaf Faaland <faaland1@llnl.gov>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #6933
Closes #6971

Honor --with-mounthelperdir where applicable

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Signed-off-by: loli10K <ezomori.nozomu@gmail.com>
Closes #6962

Fix --with-systemd on Debian-based distributions (#6963)

These changes propagate the "--with-systemd" configure option to the
RPM spec file, allowing Debian-based distributions to package
systemd-related files.

Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: loli10K <ezomori.nozomu@gmail.com>
Closes #6591
Closes #6963

Remove lib/libspl/include/sys/frame.h

The functionality provided by this header is not required by any
of the ZFS user space code.  Minimal functionality was provided
in commit c28a677 which added include/sys/frame.h.

Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #6960
Closes #6972

Enable QAT support in zfs-dkms RPM

Enable QAT accelerated gzip compression in zfs-dkms RPM package when
environment variant ICP_ROOT is set to QAT drive source code folder
and QAT hardware presence.  Otherwise, use default gzip compression.

Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: David Qian <david.qian@intel.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #6932

Add zfs-import.target services in spec file

Add missing zfs-import.target to list of systemd services in zfs
RPM spec file.

Reviewed-by: Niklas Wagner <Skaro@Skaronator.com>
Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Ralf Ertzinger <ralf@skytale.net>
Issue #6953
Closes #6955

Various ZED fixes

* Teach ZED to handle spares usingi the configured ashift: if the zpool
   'ashift' property is set then ZED should use its value when kicking
   in a hotspare; with this change 512e disks can be used as spares
   for VDEVs that were created with ashift=9, even if ZFS natively
   detects them as 4K block devices.

 * Introduce an additional auto_spare test case which verifies that in
   the face of multiple device failures an appropiate number of spares
   are kicked in.

 * Fix zed_stop() in "libtest.shlib" which did not correctly wait the
   target pid.

 * Fix ZED crashing on startup caused by a race condition in libzfs
   when used in multi-threaded context.

 * Convert ZED over to using the tpool library which is already present
   in the Illumos FMA code.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: loli10K <ezomori.nozomu@gmail.com>
Closes #2562
Closes #6858

Disable vdev_zaps_004_pos

Occasionally observed failure of vdev_zaps_004_pos due to the test
case not being 100% reliable.  In order to prevent false positives
disable this test case until it can be made reliable.

Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #6935
Closes #6936

Suppress incorrect objtool warnings

Suppress incorrect warnings from versions of objtool which are not
aware of x86 EVEX prefix instructions used for AVX512.

  module/zfs/vdev_raidz_math_avx512bw.o: warning:
  objtool: <func+offset>: can't find jump dest instruction at .text

Reviewed-by: Don Brady <don.brady@delphix.com>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #6928

Fix segfault in zpool iostat when adding VDEVs

Fix a segfault when running 'zpool iostat -v 1' while adding
a VDEV.

Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tony Hutter <hutter2@llnl.gov>
Closes #6748
Closes #6872

OpenZFS 8603 - rename zilog's "zl_writer_lock" to "zl_issuer_lock"

This is a purely cosmetic change. The zilog's "zl_writer_lock" field is
being renamed to "zl_issuer_lock" to try and make the code easier to
understand; no other changes are made.

Authored by: Prakash Surya <prakash.surya@delphix.com>
Reviewed by: C Fraire <cfraire@me.com>
Approved by: Dan McDonald <danmcd@joyent.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov>
OpenZFS-issue: https://www.illumos.org/issues/8603
OpenZFS-commit: https://github.com/openzfs/openzfs/commit/2daf06546b
Closes #6927

Disable create-o_ashift

Occasionally observed failure of create-o_ashift due to the test
case not being 100% reliable.  In order to prevent false positives
disable this test case until it can be made reliable.

Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Reviewed-by: loli10K <ezomori.nozomu@gmail.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #6924
Closes #6925

OpenZFS 8585 - improve batching done in zil_commit()

Authored by: Prakash Surya <prakash.surya@delphix.com>
Reviewed by: Brad Lewis <brad.lewis@delphix.com>
Reviewed by: Matt Ahrens <mahrens@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Approved by: Dan McDonald <danmcd@joyent.com>
Ported-by: Prakash Surya <prakash.surya@delphix.com>
Problem
=======

The current implementation of zil_commit() can introduce significant
latency, beyond what is inherent due to the latency of the underlying
storage. The additional latency comes from two main problems:

 1. When there's outstanding ZIL blocks being written (i.e. there's
    already a "writer thread" in progress), then any new calls to
    zil_commit() will block waiting for the currently oustanding ZIL
    blocks to complete. The blocks written for each "writer thread" is
    coined a "batch", and there can only ever be a single "batch" being
    written at a time. When a batch is being written, any new ZIL
    transactions will have to wait for the next batch to be written,
    which won't occur until the current batch finishes.

    As a result, the underlying storage may not be used as efficiently
    as possible. While "new" threads enter zil_commit() and are blocked
    waiting for the next batch, it's possible that the underlying
    storage isn't fully utilized by the current batch of ZIL blocks. In
    that case, it'd be better to allow these new threads to generate
    (and issue) a new ZIL block, such that it could be serviced by the
    underlying storage concurrently with the other ZIL blocks that are
    being serviced.

 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch"
    to complete, prior to zil_commit() returning. The size of any given
    batch is proportional to the number of ZIL transaction in the queue
    at the time that the batch starts processing the queue; which
    doesn't occur until the previous batch completes. Thus, if there's a
    lot of transactions in the queue, the batch could be composed of
    many ZIL blocks, and each call to zil_commit() will have to wait for
    all of these writes to complete (even if the thread calling
    zil_commit() only cared about one of the transactions in the batch).

To further complicate the situation, these two issues result in the
following side effect:

 3. If a given batch takes longer to complete than normal, this results
    in larger batch sizes, which then take longer to complete and
    further drive up the latency of zil_commit(). This can occur for a
    number of reasons, including (but not limited to): transient changes
    in the workload, and storage latency irregularites.

Solution
========

The solution attempted by this change has the following goals:

 1. no on-disk changes; maintain current on-disk format.
 2. modify the "batch size" to be equal to the "ZIL block size".
 3. allow new batches to be generated and issued to disk, while there's
    already batches being serviced by the disk.
 4. allow zil_commit() to wait for as few ZIL blocks as possible.
 5. use as few ZIL blocks as possible, for the same amount of ZIL
    transactions, without introducing significant latency to any
    individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks.

In theory, with these goals met, the new allgorithm will allow the
following improvements:

 1. new ZIL blocks can be generated and issued, while there's already
    oustanding ZIL blocks being serviced by the storage.
 2. the latency of zil_commit() should be proportional to the underlying
    storage latency, rather than the incoming synchronous workload.

Porting Notes
=============

Due to the changes made in commit 119a394ab0, the lifetime of an itx
structure differs than in OpenZFS. Specifically, the itx structure is
kept around until the data associated with the itx is considered to be
safe on disk; this is so that the itx's callback can be called after the
data is committed to stable storage. Since OpenZFS doesn't have this itx
callback mechanism, it's able to destroy the itx structure immediately
after the itx is committed to an lwb (before the lwb is written to
disk).

To support this difference, and to ensure the itx's callbacks can still
be called after the itx's data is on disk, a few changes had to be made:

  * A list of itxs was added to the lwb structure. This list contains
    all of the itxs that have been committed to the lwb, such that the
    callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(),
    after the data for the itxs is committed to disk.

  * A list of itxs was added on the stack of the zil_process_commit_list()
    function; the "nolwb_itxs" list. In some circumstances, an itx may
    not be committed to an lwb (e.g. if allocating the "next" ZIL block
    on disk fails), so this list is used to keep track of which itxs
    fall into this state, such that their callbacks can be called after
    the ZIL's writer pipeline is "stalled".

  * The logic to actually call the itx's callback was moved into the
    zil_itx_destroy() function. Since all consumers of zil_itx_destroy()
    were effectively performing the same logic (i.e. if callback is
    non-null, call the callback), it seemed like useful code cleanup to
    consolidate this logic into a single function.

Additionally, the existing Linux tracepoint infrastructure dealing with
the ZIL's probes and structures had to be updated to reflect these code
changes. Specifically:

  * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be
    removed from "trace_zil.h" as well.

  * Some of the zilog structure's fields were removed, which affected
    the tracepoint definitions of the structure.

  * New tracepoints had to be added for the following 3 new probes:
      * zil__process__commit__itx
      * zil__process__normal__itx
      * zil__commit__io__error

OpenZFS-issue: https://www.illumos.org/issues/8585
OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a
Closes #6566

Fix NFS sticky bit permission denied error

When zfs_sticky_remove_access() was originally adapted for Linux
a typo was made which altered the intended behavior.  As described
in the block comment, the intended behavior is that permission
should be granted when the entry is a regular file and you have
write access.  That is, S_ISREG should have been used instead of
S_ISDIR.

Restricting permission to regular files made good sense for older
systems where setting the bit on executable files would instruct
the system to save the program's text segment on the swap device.

On modern systems this behavior has been replaced by the sticky
bit acting as a restricted deletion flag and the plain file
restriction has been relaxed.

Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #6889
Closes #6910

Add /usr/bin/env to COPY_EXEC_LIST initramfs hook

5dc1ff29 changed the user space program to mount a zfs snapshot
from /bin/sh to /usr/bin/env.  If the executable is not present
in the initramfs then snapshots cannot be automounted.

Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Richard Laager <rlaager@wiktel.com>
Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Signed-off-by: James Dingwall <james.dingwall@zynstra.com>
Closes #5360
Closes #6913

Fix 'zpool create|add' replication level check

When the pool configuration contains a hole due to a previous device
removal ignore this top level vdev.  Failure to do so will result in
the current configuration being assessed to have a non-uniform
replication level and the expected warning will be disabled.

The zpool_add_010_pos test case was extended to cover this scenario.

Reviewed-by: George Melikov <mail@gmelikov.ru>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #6907
Closes #6911

Preserve itx alloc size for zio_data_buf_free()

Using zio_data_buf_alloc() to allocate the itx's may be unsafe
because the itx->itx_lr.lrc_reclen field is not constant from
allocation to free.  Using a different itx->itx_lr.lrc_reclen
size in zio_data_buf_free() can result in the allocation being
returned to the wrong kmem cache.

This issue can be avoided entirely by storing the allocation size
in itx->itx_size and using that for zio_data_buf_free().

Reviewed by: Prakash Surya <prakash.surya@delphix.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #6912

Unbreak the scan status ABI

When d4a72f23 was merged, pss_pass_issued was incorrectly
added to the middle of the pool_scan_stat_t structure
instead of the end. This patch simply corrects this issue.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Signed-off-by: Tom Caputi <tcaputi@datto.com>
Closes #6909

Fix 'zfs get {user|group}objused@' functionality

Fix a regression accidentally introduced in 1b81ab4 that prevents
'zfs get {user|group}objused@' from correctly reporting the requested
value.

Update "userspace_003_pos.ksh" and "groupspace_003_pos.ksh" to verify
this functionality.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Signed-off-by: loli10K <ezomori.nozomu@gmail.com>
Closes #6908

Linux 4.14 compat: CONFIG_GCC_PLUGIN_RANDSTRUCT

Fix build errors with gcc 7.2.0 on Gentoo with kernel 4.14
built with CONFIG_GCC_PLUGIN_RANDSTRUCT=y such as:

module/nvpair/nvpair.c:2810:2:error:
positional initialization of field in ?struct? declared with
'designated_init' attribute [-Werror=designated-init]
  nvs_native_nvlist,
  ^~~~~~~~~~~~~~~~~

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Mark Wright <gienah@gentoo.org>
Closes #5390
Closes #6903

initramfs: Honor canmount=off

The initramfs script was not honoring canmount=off.  With this change,
it does.  If the administrator has asked that a filesystem not be
mounted, that should be honored.

As an exception, the initramfs script ignores canmount=off on the
rootfs.  The rootfs should not have canmount=off set either.  However,
mounting it anyway seems harmless because it is being asked for
explicitly.  The point of this exception is to avoid the risk of
breaking existing systems, just in case someone has canmount=off set on
their rootfs.

The initramfs still mounts filesystems with canmount=noauto.  This is
necessary because it is typical to set that on the rootfs so that it can
be cloned.  Without canmount=noauto, the clones' duplicate mountpoints
would conflict.

This is the remainder of the fix for:
https://github.com/zfsonlinux/pkg-zfs/issues/221

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Signed-off-by: Richard Laager <rlaager@wiktel.com>
Closes #6897

initramfs: Honor mountpoint=none/legacy

For filesystems that are children of the rootfs, when mountpoint=none or
mountpoint=legacy, the initrafms script would assume a mountpoint based
on the dataset path.  Given that the rootfs should have mountpoint=/ and
mountpoint inheritance is is the default behavior of ZFS, this behavior
seems unnecessary.  In any event, it turns mountpoint=none into a no-op.
That removes this option from the administrator, and if someone uses it,
it does not work as expected.  Worse yet, if the mountpoint directory
does not exist (which is the typical case for mountpoint=none), the
mounting and thus the boot process will fail.  For the case of
mountpoint=legacy, the assumed mountpoint may not be the correct value
set in /etc/fstab.

This change makes the initramfs script not mount the filesystem in
either case.  For mountpoint=none, this means we are correctly honoring
the setting.  For mountpoint=legacy, there are two scenarios:  If
canmount=on, the filesystem will be mounted by the normal mechanisms
later in the boot process.  If canmount=noauto, the filesystem will not
be mounted at all, unless the administrator has done something special.
If they're not doing something special and they want it mounted by the
initramfs, they can simply not set mountpoint=legacy.

This is part of the fix for:
https://github.com/zfsonlinux/pkg-zfs/issues/221

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Signed-off-by: Richard Laager <rlaager@wiktel.com>
Closes #6897

zpool(8): Fix "zpool import -t"

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Signed-off-by: DHE <git@dehacked.net>
Closes #6894

Update for cppcheck v1.80

Resolve new warnings and errors from cppcheck v1.80.

* [lib/libshare/libshare.c:543]: (warning)
  Possible null pointer dereference: protocol
* [lib/libzfs/libzfs_dataset.c:2323]: (warning)
  Possible null pointer dereference: srctype
* [lib/libzfs/libzfs_import.c:318]: (error)
  Uninitialized variable: link
* [module/zfs/abd.c:353]: (error) Uninitialized variable: sg
* [module/zfs/abd.c:353]: (error) Uninitialized variable: i
* [module/zfs/abd.c:385]: (error) Uninitialized variable: sg
* [module/zfs/abd.c:385]: (error) Uninitialized variable: i
* [module/zfs/abd.c:553]: (error) Uninitialized variable: i
* [module/zfs/abd.c:553]: (error) Uninitialized variable: sg
* [module/zfs/abd.c:763]: (error) Uninitialized variable: i
* [module/zfs/abd.c:763]: (error) Uninitialized variable: sg
* [module/zfs/abd.c:305]: (error) Uninitialized variable: tmp_page
* [module/zfs/zpl_xattr.c:342]: (warning)
   Possible null pointer dereference: value
* [module/zfs/zvol.c:208]: (error) Uninitialized variable: p

Convert the following suppression to inline.

* [module/zfs/zfs_vnops.c:840]: (error)
  Possible null pointer dereference: aiov

Exclude HAVE_UIO_ZEROCOPY and HAVE_DNLC from analysis since
these macro's will never be defined until this functionality
is implemented.

Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #6879

Fix data on evict_skips in arc_summary.py

Display correct data from kstat arcstats for evict_skips,
which is currently repeating the data from mutex_misses.
Fixes #6882

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Signed-off-by: Scot W. Stevenson <scot.stevenson@gmail.com>
Closes #6882
Closes #6883

Fix ARC pointer overrun

Only access the `b_crypt_hdr` field of an ARC header if the content
is encrypted.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Tom Caputi <tcaputi@datto.com>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Signed-off-by: DHE <git@dehacked.net>
Closes #6877

Sequential scrub and resilvers

Currently, scrubs and resilvers can take an extremely
long time to complete. This is largely due to the fact
that zfs scans process pools in logical order, as
determined by each block's bookmark. This makes sense
from a simplicity perspective, but blocks in zfs are
often scattered randomly across disks, particularly
due to zfs's copy-on-write mechanisms.

This patch improves performance by splitting scrubs
and resilvers into a metadata scanning phase and an IO
issuing phase. The metadata scan reads through the
structure of the pool and gathers an in-memory queue
of I/Os, sorted by size and offset on disk. The issuing
phase will then issue the scrub I/Os as sequentially as
possible, greatly improving performance.

This patch also updates and cleans up some of the scan
code which has not been updated in several years.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Authored-by: Saso Kiselkov <saso.kiselkov@nexenta.com>
Authored-by: Alek Pinchuk <apinchuk@datto.com>
Authored-by: Tom Caputi <tcaputi@datto.com>
Signed-off-by: Tom Caputi <tcaputi@datto.com>
Closes #3625
Closes #6256

Minor code cleanups in arc_python.py

Remove unused library re and associated variable kstat_pobj. Add note
to documentation at start of program about required support for old
versions of Python. Change variable "format" (which is a built-in
function) to "fmt".

Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Scot W. Stevenson <scot.stevenson@gmail.com>
Closes #6869

Fix dirty check in dmu_offset_next()

The correct way to determine if a dnode is dirty is to check
if any of the dn->dn_dirty_link's are active.  Relying solely
on the dn->dn_dirtyctx can result in the dnode being mistakenly
reported as clean.

Reviewed-by: Chunwei Chen <tuxoko@gmail.com>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #3125
Closes #6867

Disable automatic dependencies in zfs-test package

All of the ZTS test scripts specify /bin/ksh as the interpreter.
Unfortunately, as of Fedora 27 only /usr/bin/ksh is provided by
the package manager.  Rather than change all the scripts to
accommodate the latest Fedora disable automatic dependencies
for the zfs-test package.  Functionally this will not cause
any problems since /bin is a symlink to /usr/bin.

Reviewed-by: George Melikov <mail@gmelikov.ru>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #6868

Disable zvol_ENOSPC_001_pos on 32-bit systems

Occasionally observed failure of zvol_ENOSPC_001_pos due to the
test case taking too long to complete.  Disable the test case until
it can be improved.

Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #5848
Closes #6862

Fix truncate(2) mtime and ctime handling

On Linux, ftruncate(2) always changes the file timestamps, even if the
file size is not changed. However, in case of a successfull
truncate(2), the timestamps are updated only if the file size changes.
This translates to the VFS calling the ZFS Posix Layer "setattr"
function (zpl_setattr) with ATTR_MTIME and ATTR_CTIME unconditionally
set on the iattr mask only when doing a ftruncate(2), while the
truncate(2) is left to the filesystem implementation to be dealt with.

This behaviour is consistent with POSIX:2004/SUSv3 specifications
where there's no explicit requirement for file size changes to update
the timestamps only for ftruncate(2):

http://pubs.opengroup.org/onlinepubs/009695399/functions/truncate.html
http://pubs.opengroup.org/onlinepubs/009695399/functions/ftruncate.html

This has been later updated in POSIX:2008/SUSv4 where, for both
truncate(2)/ftruncate(2), there's no mention of this size change
requirement:

http://austingroupbugs.net/view.php?id=489
http://pubs.opengroup.org/onlinepubs/9699919799/functions/truncate.html
http://pubs.opengroup.org/onlinepubs/9699919799/functions/ftruncate.html

Unfortunately the Linux VFS is still calling into the ZPL without
ATTR_MTIME/ATTR_CTIME set in the truncate(2) case: we fix this by
explicitly updating the timestamps when detecting the ATTR_SIZE bit,
which is always set in do_truncate(), on the iattr mask.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: loli10K <ezomori.nozomu@gmail.com>
Closes #6811
Closes #6819

Add .travis.yml

Travis builders have maximum work time ~49 minutes,
so we have to use 5 builders and spread the ZTS over
them using test group tags.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Signed-off-by: George Melikov <mail@gmelikov.ru>
Closes #6829

Fix arc_summary.py -d crash with Python3

Prevents arc_summary.py crashing when called with parameter -d or
long form --description with Python3.

Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Signed-off-by: Scot W. Stevenson <scot.stevenson@gmail.com>
Closes #6849
Closes #6850

OpenZFS 7531 - Assign correct flags to prefetched buffers

Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: Prakash Surya <prakash.surya@delphix.com>
Authored by: abraunegg <alex.braunegg@gmail.com>
Approved by: Dan McDonald <danmcd@joyent.com>
Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>
OpenZFS-issue: https://www.illumos.org/issues/7531
OpenZFS-commit: https://github.com/openzfs/openzfs/commit/468008cb

Long hold the dataset during upgrade

If the receive or rollback is performed while filesystem is upgrading
the objset may be evicted in `dsl_dataset_clone_swap_sync_impl`. This
will lead to NULL pointer dereference when upgrade tries to access
evicted objset.

This commit adds long hold of dataset during whole upgrade process.
The receive and rollback will return an EBUSY error until the
upgrade is not finished.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Arkadiusz Bubała <arkadiusz.bubala@open-e.com>
Closes #5295
Closes #6837

Fix encryption root hierarchy issue

After doing a recursive raw receive, zfs userspace performs
a final pass to adjust the encryption root hierarchy as
needed. Unfortunately, the FORCE_INHERIT ioctl had a bug
which caused the encryption root to always be assigned to
the direct parent instead of the inheriting parent. This
patch simply fixes this issue.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alek Pinchuk <apinchuk@datto.com>
Signed-off-by: Tom Caputi <tcaputi@datto.com>
Closes #6847
Closes #6848

Handle compressed buffers in __dbuf_hold_impl()

In __dbuf_hold_impl(), if a buffer is currently syncing and is still
referenced from db_data, a copy is made in case it is dirtied again in
the txg.  Previously, the buffer for the copy was simply allocated with
arc_alloc_buf() which doesn't handle compressed or encrypted buffers
(which are a special case of a compressed buffer).  The result was
typically an invalid memory access because the newly-allocated buffer
was of the uncompressed size.

This commit fixes the problem by handling the 2 compressed cases,
encrypted and unencrypted, respectively, with arc_alloc_raw_buf() and
arc_alloc_compressed_buf().

Although using the proper allocation functions fixes the invalid memory
access by allocating a buffer of the compressed size, another unrelated
issue made it impossible to properly detect compressed buffers in the
first place.  The header's compression flag was set to ZIO_COMPRESS_OFF
in arc_write() when it was possible that an attached buffer was actually
compressed.  This commit adds logic to only set ZIO_COMPRESS_OFF in
the non-ZIO_RAW case which wil handle both cases of compressed buffers
(encrypted or unencrypted).

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tim Chase <tim@chase2k.com>
Closes #5742
Closes #6797

Refresh TEST file to include new variables

Add any missing variables for CI control to TEST.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Closes #6846

Cleanup systemd dependencies

Some redundancy is present in the systemd dependencies, as
noticed in PR#6764. Existing setups might rely on these quirks,
so these cleanups have been moved to the development branch.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Signed-off-by: Antonio Russo <antonio.e.russo@gmail.com>
Closes #6822

Fix undefined %{systemd_svcs} in RPM scriptlets

This allows RPM-based systems to properly control package installation
and removal when using systemd.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Signed-off-by: loli10K <ezomori.nozomu@gmail.com>
Closes #6838
Closes #6841

OpenZFS 8607 - variable set but not used

Reviewed by: Yuri Pankov <yuripv@gmx.com>
Reviewed by: Igor Kozhukhov <igor@dilos.org>
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Approved by: Robert Mustacchi <rm@joyent.com>
Authored by: Toomas Soome <tsoome@me.com>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>
OpenZFS-issue: https://www.illumos.org/issues/8607
OpenZFS-commit: https://github.com/openzfs/openzfs/commit/b852c2f5
Closes #6842

Provide tags in perf-regression.run

A prior commit changed test-runner to enable tagging
of testgroups within a test suite runfile. They must
be specified in each runfile. Update the runfile for
performance regressions so it is properly tagged.

Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Closes #6830

Fix zfs-tests.sh single test functionality

Without any tag specified into the runtime-generated runfile the
test-runner will not execute the test provided from the command line:
fix this by adding tag information to the custom runfile.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Signed-off-by: loli10K <ezomori.nozomu@gmail.com>
Closes #6826

contrib/initramfs: switch to automake

Use automake to build initramfs scripts and hooks.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: loli10K <ezomori.nozomu@gmail.com>
Closes #6761

Bug fix in qat_compress.c when compressed size is < 4KB

When the 128KB block is compressed to less than 4KB, the pointer
to the Footer is not in the end of the compressed buffer, that's
because the Header offset was added twice for this case. So there
is a gap between the Footer and the compressed buffer.
1. Always compute the Footer pointer address from the start of the
last page.
2. Remove the un-used workaroud code which has been verified fixed
with the latest driver and this fix.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Signed-off-by: Weigang Li <weigang.li@intel.com>
Closes #6827

Sort output of tunables in arc_summary.py

Sort list of tunables printed by _tunable_summary()
alphabetically

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Signed-off-by: Scot W. Stevenson <scot.stevenson@gmail.com>
Closes #6828

Disable automatic dependencies in DKMS package

By default additional dependencies are generated automatically for
packages.  This is normally a good thing because it helps ensure
things just work.  It doesn't make sense for the DKMS package which
requires minimal dependencies that can be easily listed.

Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: loli10K <ezomori.nozomu@gmail.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #6467
Closes #6835

Build regression in c89 cleanups

Fixed build regression in non-debug builds from recent cleanups of
c89 workarounds.

Reviewed-by: Tim Chase <tim@chase2k.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Don Brady <don.brady@delphix.com>
Closes #6832

Disable zpool_import_missing_003_pos

Rarely observed failure of zpool_import_missing_003_pos during
automated testing due to timeout.  Disable the test case until
it can be improved.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Signed-off-by: George Melikov <mail@gmelikov.ru>
Issue #6839
Closes #6840

Add documentation strings to arc_summary.py

Include docstrings (PEP8, PEP257) for module and all functions.
Separately, remove outdated section in comment at start of
module. Separately, remove unused global constant "usetunable".

Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Signed-off-by: Scot W. Stevenson <scot.stevenson@gmail.com>
Closes #6818

Fix column alignment with long zpool names

`zpool status` normally aligns NAME/STATE/etc columns:

    NAME                       STATE     READ WRITE CKSUM
    dummy                      ONLINE       0     0     0
      mirror-0                 ONLINE       0     0     0
        /tmp/dummy-long-1.bin  ONLINE       0     0     0
        /tmp/dummy-long-2.bin  ONLINE       0     0     0
      mirror-1                 ONLINE       0     0     0
        /tmp/dummy-long-3.bin  ONLINE       0     0     0
        /tmp/dummy-long-4.bin  ONLINE       0     0     0

However, if the zpool name is longer than the zvol names, alignment
issues arise:

    NAME                  STATE     READ WRITE CKSUM
    dummy-very-very-long-zpool-name  ONLINE       0     0     0
      mirror-0            ONLINE       0     0     0
        /tmp/dummy-1.bin  ONLINE       0     0     0
        /tmp/dummy-2.bin  ONLINE       0     0     0
      mirror-1            ONLINE       0     0     0
        /tmp/dummy-3.bin  ONLINE       0     0     0
        /tmp/dummy-4.bin  ONLINE       0     0     0

`zpool iostat` and `zpool import` are also affected:

                  capacity     operations     bandwidth
    pool        alloc   free   read  write   read  write
    ----------  -----  -----  -----  -----  -----  -----
    dummy        104K  1.97G      0      0    152  9.84K
    dummy-very-very-long-zpool-name   152K  1.97G      0      1    144  13.1K
    ----------  -----  -----  -----  -----  -----  -----

    dummy-very-very-long-zpool-name  ONLINE
      mirror-0            ONLINE
        /tmp/dummy-1.bin  ONLINE
        /tmp/dummy-2.bin  ONLINE
      mirror-1            ONLINE
        /tmp/dummy-3.bin  ONLINE
        /tmp/dummy-4.bin  ONLINE

Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: George Gaydarov <git@gg7.io>
Closes #6786

Rewrite fHits() in arc_summary.py with SI units

Complete rewrite of fHits(). Move units from non-standard English
abbreviations to SI units, thereby avoiding confusion because of
"long scale" and "short scale" numbers. Remove unused parameter
"Decimal". Add function string. Aim to confirm to PEP8.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Signed-off-by: Scot W. Stevenson <scot.stevenson@gmail.com>
Closes #6815

Undo c89 workarounds to match with upstream

With PR 5756 the zfs module now supports c99 and the
remaining past c89 workarounds can be undone.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Signed-off-by: Don Brady <don.brady@delphix.com>
Closes #6816

Minor code cleanup in arc_summary.py

Simplify and inline single-use function div1(); inline twice-used
function div2(); add function comment to zfs_header(); replace
variable "unused" in get_Kstat() with "_" following convention.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Signed-off-by: Scot W. Stevenson <scot.stevenson@gmail.com>
Closes #6802

Initramfs fixes

* initramfs: Fix inconsistent whitespace
* initramfs: Fix a spelling error
* initramfs: Set elevator=noop on the rpool's disks

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Signed-off-by: Richard Laager <rlaager@wiktel.com>
Closes #6807

Allow test-runner to filter test groups by tag

Enable test-runner to accept a list of tags to identify
which test groups the user wishes to run.

Also allow test-runner to perform multiple iterations
of a test run.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: John Wren Kennedy <john.kennedy@delphix.com>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Signed-off-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Closes #6788

initramfs: Set elevator=noop on the rpool's disks

ZFS already sets elevator=noop for wholedisk vdevs (for all pools), but
typical root-on-ZFS installations use partitions.  This sets
elevator=noop on the disks in the root pool.

Ubuntu 16.04 and 16.10 had this.  It was lost in 17.04 due to Debian
switching to this upstream initramfs script.

Signed-off-by: Richard Laager <rlaager@wiktel.com>

initramfs: Fix a spelling error

This fixes a typo in a comment.

Signed-off-by: Richard Laager <rlaager@wiktel.com>

initramfs: Fix inconsistent whitespace

This fixes one instance of inconsistent whitespace.

Signed-off-by: Richard Laager <rlaager@wiktel.com>

Add scan.coverity.com badge to README

Include the scan.coverity.com status badge in the top level README.

Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #6801

OpenZFS 640 - number_to_scaled_string is duplicated in several commands

Porting Notes:
- The OpenZFS patch added nicenum_scale() and nicenum() to a
  library not used by ZFS.  Rather than pull in a new dependency
  the version of nicenum in lib/libzpool/util.c was simply
  replaced with the new one.

Reviewed by: Sebastian Wiedenroth <wiedi@frubar.net>
Reviewed by: Robert Mustacchi <rm@joyent.com>
Reviewed by: Yuri Pankov <yuripv@gmx.com>
Approved by: Dan McDonald <danmcd@joyent.com>
Authored by: Jason King <jason.brian.king@gmail.com>
Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>
OpenZFS-issue: https://www.illumos.org/issues/640
OpenZFS-commit: https://github.com/openzfs/openzfs/commit/0a055120
Closes #6796

Rewrite of function fBytes() in arc_summary.py

Replace if-elif-else construction with shorter loop;
remove unused parameter "Decimal"; centralize format
string; add function documentation string; conform to
PEP8.

Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Scot W. Stevenson <scot.stevenson@gmail.com>
Closes #6784

systemd zfs-import.target and documentation

zfs-import-{cache,scan}.service must complete before any mounting of
filesystems can occur. To simplify this dependency, create a target
that is reached After (in the systemd sense) the pool is imported.

Additionally, recommend that legacy zfs mounts use the option

x-systemd.requires=zfs-import.target

to codify this requirement.

Reviewed-by: Fabian Grünbichler <f.gruenbichler@proxmox.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Signed-off-by: Antonio Russo <antonio.e.russo@gmail.com>
Closes #6764

Update zfs module parameters man5

Update zfs module parameters man5 with missing parameter details
for multiple tunings.

Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Signed-off-by: Alex Braunegg <alex.braunegg@gmail.com>
Closes #6785

Fix status command options in zpool(8)

The 'zpool status' command supports the -P option for printing full
path names.  It does not support the -p parsable option for printing
exact values.

Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: loli10K <ezomori.nozomu@gmail.com>
Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #6792
Closes #6794

OpenZFS 8081 - Compiler warnings in zdb

Fix compiler warnings in zdb.  With these changes, FreeBSD can compile
zdb with all compiler warnings enabled save -Wunused-parameter.

usr/src/cmd/zdb/zdb.c
usr/src/cmd/zdb/zdb_il.c
usr/src/uts/common/fs/zfs/sys/sa.h
usr/src/uts/common/fs/zfs/sys/spa.h
Fix numerous warnings, including:
* const-correctness
* shadowing global definitions
* signed vs unsigned comparisons
* missing prototypes, or missing static declarations
* unused variables and functions
* Unreadable array initializations
* Missing struct initializers

usr/src/cmd/zdb/zdb.h
Add a header file to declare common symbols

usr/src/lib/libzpool/common/sys/zfs_context.h
usr/src/uts/common/fs/zfs/arc.c
usr/src/uts/common/fs/zfs/dbuf.c
usr/src/uts/common/fs/zfs/spa.c
usr/src/uts/common/fs/zfs/txg.c
Add a function prototype for zk_thread_create, and ensure that every
callback supplied to this function actually matches the prototype.

usr/src/cmd/ztest/ztest.c
usr/src/uts/common/fs/zfs/sys/zil.h
usr/src/uts/common/fs/zfs/zfs_replay.c
usr/src/uts/common/fs/zfs/zvol.c
Add a function prototype for zil_replay_func_t, and ensure that
every function of this type actually matches the prototype.

usr/src/uts/common/fs/zfs/sys/refcount.h
Change FTAG so it discards any constness of __func__, necessary
since existing APIs expect it passed as void *.

Porting Notes:
- Many of these fixes have already been applied to Linux.  For
  consistency the OpenZFS version of a change was applied if the
  warning was addressed in an equivalent but different fashion.

Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: Prakash Surya <prakash.surya@delphix.com>
Authored by: Alan Somers <asomers@gmail.com>
Approved by: Richard Lowe <richlowe@richlowe.net>
Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>
OpenZFS-issue: https://www.illumos.org/issues/8081
OpenZFS-commit: https://github.com/openzfs/openzfs/commit/843abe1b8a
Closes #6787

Correct make mancheck recipe

The current make recipe for mancheck silently ignores errors. Correct
the recipe so errors cause the mancheck recipe fail.

The zpool reopen command in the zpool.8 manpage had a bullet list
without an .El.

Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: loli10K <ezomori.nozomu@gmail.com>
Signed-off-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Closes #6790

ZFS send fails to dump objects larger than 128PiB

When dumping objects larger than 128PiB it's possible for do_dump() to
miscalculate the FREE_RECORD offset due to an integer overflow
condition: this prevents the receiving end from correctly restoring
the dumped object.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Fabian Grünbichler <f.gruenbichler@proxmox.com>
Signed-off-by: loli10K <ezomori.nozomu@gmail.com>
Closes #6760

Allow 'zpool events' filtering by pool name

Additionally add four new tests:

 * zpool_events_clear: verify 'zpool events -c' functionality
 * zpool_events_cliargs: verify command line options and arguments
 * zpool_events_follow: verify 'zpool events -f'
 * zpool_events_poolname: verify events filtering by pool name

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Signed-off-by: loli10K <ezomori.nozomu@gmail.com>
Closes #3285
Closes #6762

OpenZFS 8558, 8602 - lwp_create() returns EAGAIN

8558 lwp_create() returns EAGAIN on system with more than 80K ZFS filesystems

On a system with more than 80K ZFS filesystems, we've seen cases
where lwp_create() will start to fail by returning EAGAIN. The
problem being, for each of those 80K ZFS filesystems, a taskq will
be created for each dataset as part of the ZIL for each dataset.

Porting Notes:
- The new nomem taskq kstat was dropped.
- Added module options and documentation for new tunings
  zfs_zil_clean_taskq_nthr_pct, zfs_zil_clean_taskq_minalloc,
  zfs_zil_clean_taskq_maxalloc, and zfs_sync_taskq_batch_pct.

Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Sebastien Roy <sebastien.roy@delphix.com>
Approved by: Robert Mustacchi <rm@joyent.com>
Authored by: Prakash Surya <prakash.surya@delphix.com>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Chris Dunlop <chris@onthe.net.au>
Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>
OpenZFS-issue: https://www.illumos.org/issues/8558
OpenZFS-commit: https://github.com/openzfs/openzfs/commit/216d772

8602 remove unused "dp_early_sync_tasks" field from "dsl_pool" structure

Reviewed by: Serapheim Dimitropoulos <serapheim@delphix.com>
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Approved by: Robert Mustacchi <rm@joyent.com>
Authored by: Prakash Surya <prakash.surya@delphix.com>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Chris Dunlop <chris@onthe.net.au>
Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>
OpenZFS-issue: https://www.illumos.org/issues/8602
OpenZFS-commit: https://github.com/openzfs/openzfs/commit/2bcb545
Closes #6779

Added no_scrub_restart flag to zpool reopen

Added -n flag to zpool reopen that allows a running scrub
operation to continue if there is a device with Dirty Time Log.

By default if a component device has a DTL and zpool reopen
is executed all running scan operations will be restarted.

Added functional tests for `zpool reopen`

Tests covers following scenarios:
* `zpool reopen` without arguments,
* `zpool reopen` with pool name as argument,
* `zpool reopen` while scrubbing,
* `zpool reopen -n` while scrubbing,
* `zpool reopen -n` while resilvering,
* `zpool reopen` with bad arguments.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Tom Caputi <tcaputi@datto.com>
Signed-off-by: Arkadiusz Bubała <arkadiusz.bubala@open-e.com>
Closes #6076
Closes #6746

arcstat: flush stdout / outfile after each line

Otherwise, if arcstat gets interrupted before the desired number of
iterations is reached, the output file will be empty (both if set via
'-o' or via shell redirection).

Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Signed-off-by: Fabian Grünbichler <f.gruenbichler@proxmox.com>
Closes #6775

commitcheck: Multiple OpenZFS ports in commit

Allow commitcheck.sh to handle multiple OpenZFS ports in
a single commit. This is useful in the cases when a change
upstream has bug fixes and it makes sense to port them with
the original patch.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Chris Dunlop <chris@onthe.net.au>
Signed-off-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Closes #6780

Add Coverity defect fix commit checker support

Enable commitcheck.sh to test if a commit message is
in the expected format for a coverity defect fix.

Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Closes #6777

Ensure arc_size_break is filled in arc_summary.py

Use mfu_size and mru_size pulled from the arcstats
kstat file to calculate the mfu and mru percentages
for arc size breakdown.

Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Richard Elling <Richard.Elling@RichardElling.com>
Reviewed-by: AndCycle <andcycle@andcycle.idv.tw>
Signed-off-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Closes #5526
Closes #6770

Correct flake8 errors after STYLE builder update

Fix new flake8 errors related to bare excepts and ambiguous
variable names due to a STYLE builder update.

Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Closes #6776

ZTS: Add auto-spare tests

The ZED is expected to automatically kick in a hot spare device
when there's one available in the pool and a sufficient number of
read errors have been encountered.  Use zinject to simulate the
failure condition and verify the hot spare is used.

auto_spare_001_pos.ksh: read IO errors, the vdev is FAULTED
auto_spare_002_pos.ksh: read CHECKSUM errors, the vdev is DEGRADE

Reviewed by: Richard Elling <Richard.Elling@RichardElling.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: David Quigley <david.quigley@intel.com>
Closes #6280

Use ashift=12 by default on SSDSC2BW48 disks

Currently the 480GB models of this disk do not use ashift=12 by
default.  SSDSC2BW48 is also optimized for 4k blocks.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Signed-off-by: adisbladis <adis@blad.is>
Closes #6774