Add TRIM support - replaces #5925 and #7363 #8255

dweeezil · 2019-01-08T22:52:05Z

Motivation and Context

Frequently-requested feature to help performance on SSDs and on various other SAN-like storage back-ends which support UNMAP/TRIM.

Description

This is a port of Nextenta's TRIM support to the current master branch - post-device removal and post-device initialization.

How Has This Been Tested?

New ZTS tests. Also additions to ztest.

Types of changes

Bug fix (non-breaking change which fixes an issue)
New feature (non-breaking change which adds functionality)
Performance enhancement (non-breaking change which improves efficiency)
Code cleanup (non-breaking change which makes code smaller or more readable)
Breaking change (fix or feature that would cause existing functionality to change)
Documentation (a change to man pages or other documentation)

Checklist:

My code follows the ZFS on Linux code style requirements.
I have updated the documentation accordingly.
I have read the contributing document.
I have added tests to cover my changes.
All new and existing tests passed.
All commit messages are properly formatted and contain Signed-off-by.

codecov · 2019-01-09T07:19:55Z

Codecov Report

Merging #8255 into master will decrease coverage by 0.04%.
The diff coverage is 88.01%.

@@            Coverage Diff             @@
##           master    #8255      +/-   ##
==========================================
- Coverage   78.33%   78.29%   -0.05%     
==========================================
  Files         380      382       +2     
  Lines      115719   116678     +959     
==========================================
+ Hits        90653    91350     +697     
- Misses      25066    25328     +262

Flag	Coverage Δ
#kernel	`78.93% <86.25%> (+0.04%)`	⬆️
#user	`65.8% <87.99%> (-1.21%)`	⬇️

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 0a10863...3a184b8. Read the comment docs.

dweeezil · 2019-01-14T04:25:31Z

This has been refreshed with fixes to issues discovered during testing. There are still some deadlock issues which occur mainly when the autotrim property is changed and another pool management is performed (mainly adding and possibly removing vdevs). The deadlocks are being actively investigated.

behlendorf · 2019-01-17T00:38:47Z

cmd/zpool/zpool_main.c

@@ -6649,6 +6700,53 @@ zpool_do_resilver(int argc, char **argv)
 	return (for_each_pool(argc, argv, B_TRUE, NULL, scrub_callback, &cb));
 }

+/*
+ * zpool trim [-s|-r <rate>] <pool> ...


nit: missing -p flag

behlendorf · 2019-01-17T00:52:57Z

cmd/zpool/zpool_main.c

+			print_trim_status(trim_prog, (vs->vs_space -
+			    vs->vs_alloc) + zpool_slog_space(nvroot),
+			    trim_rate, trim_start_time, trim_stop_time);
+		}


Let's model getting the trim stats off the way this is handled for checkpoint, scan, and removal stats. Specifically, with a new ZPOOL_CONFIG_TRIM_STATS key containing a uint64 array. For consistency let's try and get all this code to work in basically the same way.

behlendorf · 2019-01-17T00:58:12Z

include/linux/blkdev_compat.h

@@ -539,6 +539,40 @@ bio_is_fua(struct bio *bio)
 #endif
 }

+/*
+ * bio_set_discard - Set the appropriate flags in a bio to indicate
+ * that the specific random of sectors should be discarded.


s/random/range

behlendorf · 2019-01-17T01:23:19Z

include/sys/dmu.h

@@ -383,6 +383,8 @@ typedef struct dmu_buf {
 #define	DMU_POOL_OBSOLETE_BPOBJ		"com.delphix:obsolete_bpobj"
 #define	DMU_POOL_CONDENSING_INDIRECT	"com.delphix:condensing_indirect"
 #define	DMU_POOL_ZPOOL_CHECKPOINT	"com.delphix:zpool_checkpoint"
+#define	DMU_POOL_TRIM_START_TIME	"trim_start_time"
+#define	DMU_POOL_TRIM_STOP_TIME		"trim_stop_time"


This need to converted this to a single entry containing an array of uint64's for TRIM. Just like DMU_POOL_SCAN or DMU_POOL_REMOVING. Something like this (I may have missed some needed entries).

#define DMU_POOL_TRIM "org.open-zfs:trim" ... /* * All members must be uint64_t, for byteswap purposes. */ typedef struct spa_trim_phys { uint64_t tr_state; /* dsl_scan_state_t */ uint64_t tr_start_time; uint64_t tr_end_time; uint64_t tr_rate; uint64_t tr_to_trim; /* bytes that need to be trimmed */ uint64_t tr_trimmed; /* bytes that have been trimmed */ } spa_trim_phys_t;

behlendorf · 2019-01-17T01:26:54Z

include/sys/fs/zfs.h

+#define	ZPOOL_CONFIG_TRIM_PROG		"trim_prog"
+#define	ZPOOL_CONFIG_TRIM_RATE		"trim_rate"
+#define	ZPOOL_CONFIG_TRIM_START_TIME	"trim_start_time"
+#define	ZPOOL_CONFIG_TRIM_STOP_TIME	"trim_stop_time"


These can be dropped since we want to move this all in to spa_trim_phys_t.

behlendorf · 2019-01-17T01:44:50Z

include/sys/spa_impl.h

+	kcondvar_t	spa_man_trim_done_cv;	/* manual trim has completed */
+	/* For details on trim start/stop times see spa_get_trim_prog. */
+	uint64_t	spa_man_trim_start_time;
+	uint64_t	spa_man_trim_stop_time;


We should probably add a spa_trim_phys_t here instead and use those values. Like spa_removing_phys above. We might also want to consider moving the manual and auto trim state in to their own structures as well.

behlendorf · 2019-01-17T01:47:38Z

include/sys/sysevent/eventdefs.h

@@ -118,6 +118,8 @@ extern "C" {
 #define	ESC_ZFS_BOOTFS_VDEV_ATTACH	"bootfs_vdev_attach"
 #define	ESC_ZFS_POOL_REGUID		"pool_reguid"
 #define	ESC_ZFS_HISTORY_EVENT		"history_event"
+#define	ESC_ZFS_TRIM_START		"trim_start"
+#define	ESC_ZFS_TRIM_FINISH		"trim_finish"


This is fine, but it reminds be that we forget to add these for removal and initialize.

behlendorf · 2019-01-17T01:57:42Z

include/sys/vdev_impl.h

+	kmutex_t	vdev_trim_zios_lock;
+	kcondvar_t	vdev_trim_zios_cv;
+	uint64_t	vdev_trim_zios;	/* # of in-flight async trim zios */
+	boolean_t	vdev_trim_zios_stop;	/* see zio_trim_should_bypass */


It'd be nice to move these up under the vdev_initialize_* members, and even slightly renaming the new fields for consistency would be help but isn't critical.

behlendorf · 2019-01-17T02:05:10Z

lib/libspl/include/sys/dkio.h

+	 * a couple unused pointers is too low to justify that risk.
+	 */
+	void			(*dfl_ck_func)(uint64_t, uint64_t, void *);
+	void			*dfl_ck_arg;


This constraint really doesn't apply to us, nothing on Linux will call this, We should either drop this or wrap it and the assigned zio_trim_check functions with __sun__ to make it clear this is dead code.

behlendorf · 2019-01-17T02:08:21Z

lib/libzfs/libzfs_pool.c

+	zc.zc_cookie = (uintptr_t)&tci;
+
+	if (zfs_ioctl(hdl, ZFS_IOC_POOL_TRIM, &zc) == 0)
+		return (0);


This needs to be converted to a new in/out nvlist style ioctl and use libzfs_core.c and lzc_ioctl.

ahrens · 2019-01-20T16:11:10Z

module/zfs/zio.c

+	metaslab_t *ms = msp;
+	boolean_t held = MUTEX_HELD(&ms->ms_lock);
+	if (!held)
+		mutex_enter(&ms->ms_lock);


since this routine is a no-op on nondebug bits, can we also make it not grab/drop the mutex on nondebug?

ahrens · 2019-01-20T16:19:48Z

module/zfs/metaslab.c

- * ==========================================================================
- */
-static uint64_t
-metaslab_ff_alloc(metaslab_t *msp, uint64_t size)


How are the changes to metaslab_*_alloc() related to TRIM? Could you break them out into a separate PR? (FYI, I would support this change.)

If you're referring to the removal of metaslab_ff_alloc(), this appears to have been done as part of porting your openzfs/openzfs@8629fdc patch from the OpenZFS PR. The removal of this function does certainly seem like something that ought to be in its own standalone PR.

ahrens · 2019-01-20T16:25:25Z

module/zfs/metaslab.c

 			}
 		}
+		/*
+		 * If there's a trim ongoing, punch out the holes that will
+		 * be filled back in in metaslab_trim_done.


how about be filled in by metaslab_trim_done?

ahrens · 2019-01-20T16:26:32Z

module/zfs/metaslab.c

+		 */
+		if (msp->ms_trimming_ts != NULL) {
+			range_tree_walk(msp->ms_trimming_ts, range_tree_remove,
+			    msp->ms_allocatable);


It looks like we aren't able to allocate space that is free but in the process of being trimmed. Do we update ms_weight and ms_max_size to reflect that (and be consistent with ms_allocatable)? If not, I think metaslab_group_alloc_normal() will get confused.

Rather than deal with the implications of ms_trimming_ts everywhere, could we instead prevent allocations from metaslabs that we are in the middle of trimming? Like ms_initializing does? (see vdev_initialize_ms_mark())

ahrens · 2019-01-20T16:33:32Z

module/zfs/metaslab.c

@@ -3265,8 +3287,7 @@ metaslab_group_alloc_normal(metaslab_group_t *mg, zio_alloc_list_t *zal,
 		 * we may end up in an infinite loop retrying the same
 		 * metaslab.
 		 */
-		ASSERT(!metaslab_should_allocate(msp, asize));
-
+		/* ASSERT(!metaslab_should_allocate(msp, asize)); XXX */


yeah, this is where we will get confused because some space is not allocatable, but that is not reflected in all the metadata about the metaslab.

ahrens · 2019-01-20T16:37:14Z

module/zfs/metaslab.c

@@ -3937,6 +3969,7 @@ metaslab_claim_concrete(vdev_t *vd, uint64_t offset, uint64_t size,
 	VERIFY3U(range_tree_space(msp->ms_allocatable) - size, <=,
 	    msp->ms_size);
 	range_tree_remove(msp->ms_allocatable, offset, size);
+	metaslab_trim_remove(msp, offset, size);


I would think that the trimsets should all be empty at this point. Can we assert that, instead of trying to remove this?

ahrens · 2019-01-20T16:57:06Z

module/zfs/vdev.c

+ * 1) Keeping the trim commands reasonably small.
+ * 2) Keeping the ability to rollback back for as many txgs as possible.
+ * 3) Waiting around too long that the user starts to get uneasy about not
+ *	seeing any space being freed after they remove some files.


My understanding is that space which has been freed but not yet trimmed is in fact available for allocation (and thus "free"). That's good. However, this comment implies that this space is not counted as free in the user-visible accounting somewhere. That seems wrong to me - if the space is available for allocation then it is by definition free and should be displayed as such.

dweeezil · 2019-01-21T14:23:59Z

Status update: This has just been refreshed with some fixes to the various deadlocks observed while running ztest (which have been left as separate commits). Reviewers should look at 1eda737 and the related follow-on commits in particular. With these changes in place, zloop can now generally survive an overnight run. I've not yet done any actual testing to make sure this plays well (i.e. actually works as opposed to simply not panicking) with allocation classes or non-trivial cases of device removal (other than those which are performed via ztest, which are plenty). We (@behlendorf and myself) will begin addressing the review feedback shortly. Also, some of the new ZTS tests need fixing.

[EDIT]: Note that there were actually 3 (force) pushes in rapid succession because there was an experimental fix attempt left in by accident and also to rebase on current master with recent changes to metaslab.c. Also, I'll note that as new fixes are added, they're commited to the dweeezil:ntrim3-next branch.

Update dkio.h from Nexenta's version to pick up DKIOCFREE and add their dkioc_free_util.h header for TRIM support. Requires-builders: none

Ported by: Tim Chase <[email protected]> Porting notes: The trim kstats are in zfs/<pool> along with the other per-pool stats. The kstats can be cleared by writing to the kstat file. Null format parameters to strftime() were replaced with "%c". Added vdev trace support. New dfl_alloc() function in the SPL is used to allocate arrays of dkioc_free_list_t objects since they may be large enough to require virtual memory. Other changes: Suppressed kstat creation for pools with "$" names. The changes to vdev_raidz_map_alloc() have been minimized in order to allow more conflict-free merging with future changes (ABD). Added the following module parameters: zfs_trim - Enable TRIM zfs_trim_min_ext_sz - Minimum size to trim zfs_txgs_per_trim - Transaction groups over which to batch trims Requires-builders: none

Requires-builders: none

The extended zpool iostat options -wlqr will display information about automatic and manual TRIMs. This commit also fixes a completely unrelated bug in which the IOS_LATENCY row in the vsx_type_to_nvlist array was missing an entry for the scrub nvlist. Requires-builders: none

…ents. Requires-builders: none

The blkdev_issue_discard() function has been available for a long time by the kernel but it only supports synchronous discards. The __blkdev_issue_discard() function provides an asynchronous interface but was added in the 4.6 kernel. Only supporting synchronously discards can potentially limit performance when processing a large number of small extents. To avoid this an asynchronous discard implementation has been added to vdev_disk.c which builds on existing functionality. The kernel provided synchronous version remains the default pending additional functional and performance testing. Due to different mechamism used for submitting TRIM commands there were not being properly accounted for in the extended statistics. Resolve this by allow for aggregated stats to be returned as part of the TRIM zio. This allows for far better visibility in to the discard request sizes. Minor documentation updates. Signed-off-by: Brian Behlendorf <[email protected]> Requires-builders: none

Requires-builders: none

Signed-off-by: Brian Behlendorf <[email protected]> Requires-builders: none

Porting Notes: Man page changes dropped for the moment. This can be reconsiled when the final version is merged to OpenZFS. They are accurate now, only worded a little differently. Requires-builders: none

1) Removed the first-fit allocator. 2) Moved the autotrim metaslab scheduling logic into vdev_auto_trim. 2a) As a consequence of openzfs#2, metaslab_trimset_t was rendered superfluous. New trimsets are simple range_tree_t's. 3) Made ms_trimming_ts remove extents it is working on from ms_tree and then add them back in. 3a) As a consequence of openzfs#3, undone all the direct changes to the allocators and removed metaslab_check_trim_conflict and range_tree_find_gap. Porting Notes: * Removed WITH_*_ALLOCATOR macros and aligned remaining allocations with OpenZFS. Unused wariables warnings resolved with the gcc __attribute__ ((unused__ keyword. * Added missing calls for ms_condensing_cv. Signed-off-by: Brian Behlendorf <[email protected]> Requires-builders: none

Porting Notes: * metaslab_sync changes already applied. * resync of test cases needed Requires-builders: none

1) Simplified the SM_FREE spacemap writing while a trim is active. 2) Simplified the range_tree_verify in metaslab_check_free. 3) Clarified comment above metaslab_trim_all. 4) Substituted 'flust out' with 'drop' in comment in metaslab_trim_all. 5) Moved ms_prev_ts clearing up to ms_cur_ts claring in metaslab_trim_all. 6) Added recomputation of metaslab weight when metaslab is loaded. 7) Moved dmu_tx_commit inside of spa_trim_update_time. 8) Made the smallest allowable manual trim rate 1/1000th of a metaslab size. 9) Switched to using hrtime_t in manual trim timing logic. 10) Changed "limited" to "preserve_spilled" in vdev_auto_trim. 11) Moved vdev_notrim setting into zio_vdev_io_assess.a Porting Notes: * vdev_disk.c and zio.c hunks already applied. * nsec_per_tick -> MSEC2NSEC(1) Requires-builders: none

… wanting to condense. Requires-builders: none

Requires-builders: none

Some storage backends such as large thinly-provisioned SANs are very slow for large trims. Manual trim now supports "zpool trim -p" (partial trim) to skip metaslabs for which there is no spacemap. Signed-off-by: Tim Chase <[email protected]> Requires-builders: none

The existing test cases were split in to multiple test cases and refactored. There are now test cases for the following: zpool_trim_001_pos - Verify manual TRIM zpool_trim_002_pos - Verify manual trim can be interrupted zpool_trim_003_pos - Verify 'zpool trim -s' rate limiting zpool_trim_004_pos - Verify 'zpool trim -p' partial TRIM works zpool_trim_005_neg - Verify bad parameters to 'zpool trim' zpool_trim_006_neg - Verify bad parameters to 'zpool trim -r' autotrim_001_pos - Verify 'autotrim=on' pool data integrity autotrim_002_pos - Verify various pool geometries manualtrim_001_pos - Verify manual trim pool data integrity manualtrim_002_pos - Verify various pool geometries manualtrim_003_pos - Verify 'zpool import|export' manualtrim_004_pos - Verify 'zpool online|offline|replace' manualtrim_005_pos - Verify TRIM and scrub run concurrently Signed-off-by: Brian Behlendorf <[email protected]> Requires-builders: none

* Rename TRIM taskq threads to be more concise for Linux. * Fix divide by zero panic Signed-off-by: Brian Behlendorf <[email protected]> Requires-builders: none

* Fixed missing taskq_destroy when exporting a pool which is being actively trimmed. * Add auto/manual TRIM coverage to ztest. * Temporarily disable manualtrim_004_pos. Signed-off-by: Brian Behlendorf <[email protected]> Requires-builders: none

Signed-off-by: Chunwei Chen <[email protected]> Requires-builders: none

Signed-off-by: Tim Chase <[email protected]> Requires-builders: none

Requires-builders: none

NOTE: should be squashed into a previous commit Requires-builders: none

Also, comment-out the ASSERT(!metaslab_should_allocate(msp, asize)); in metaslab_group_alloc_normal(). It seems that the additional metaslab_group_sort() performed by trim makes this assertion invalid. Requires-builders: none

Requires-builders: none

Also, stop trimming in places where initialization is stopped: spa_vdev_remove_log() and spa_vdev_remove_top(). Requires-builders: none

NOTE: This should be its own separate PR. It was discovered during debugging of the TRIM work. Requires-builders: none

Requires-builders: none

In trim_stop_set() and trim_stop_wait(). Requires-builders: none

... and related locking Don't take the spa_auto_trim_lock in the sync path: The auto trim taskq is now stopped and started in response to changes in the "autotrim" property asynchronously by the spa async task. If the spa_auto-trim_lock is taken in the sync task, the system can deadlock as follows: Non-sync context task: Acquires spa_auto_trim_lock via spa_vdev_enter() or some other path. Sync task: syncing a change to the "autotrim" property attempts to take spa_auto_trim_lock and blocks. Non-sync context task: blocks attempting to take a spa config lock which won't be released until the sync task completes. Deadlock. Also, since the auto trim taskq is now started asynchronously, it may not yet be ready yet when the sync task calls spa_auto_trim(). Modified spa_auto_trim so it will simply return if the taskq is not available yet (it will do its thing on the next pass). Also, avoid starting auto trim taskqs for the "import$" pseudo-pool. Also, don't attempt to create the taskq if a spa unload is in-progress. Requires-builders: none

The change in the previous commit to not attempt to create the taskq if a spa unload is in-progress caused the auto trim taskq to not be started upon import. Requires-builders: none

Rather than using the default value for the property, we need to use the value which might be set by "zpool create -o autotrim=". Requires-builders: none

Long zloop runs were occasionally hitting the "ASSERT(spa->spa_man_trim_taskq != NULL)" in spa_man_trim_taskq_destroy(). It's not clear to me how this was happening because the only place "spa->spa_man_trim_taskq" is cleared is in spa_man_trim_taskq_destroy() itself which is only called from the (single) spa_async_thread() or from spa_unload(). To that end, this commit adds a non-NULL check in spa_async_thread() analagous to the tests which were added when support for stopping/starting the auto trim taskqs were added to spa_async_thread(). NOTE: Yes, that means I consider this to be a band-aid.

tonyhutter · 2019-01-24T23:01:47Z

man/man8/zpool.8

@@ -2110,6 +2159,70 @@ again.
 Starts a resilver. If an existing resilver is already running it will be
 restarted from the beginning. Any drives that were scheduled for a deferred
 resilver will be added to the new one.


Add for newlines:

Starts a resilver. If an existing resilver is already running it will be restarted from the beginning. Any drives that were scheduled for a deferred resilver will be added to the new one. +.It Xo +.Nm .Cm trim .Op Fl p .Op Fl r Ar rate | Fl s

tonyhutter · 2019-01-24T23:19:57Z

Include atrim and mtrim latency histograms in the man page

--- a/man/man8/zpool.8
+++ b/man/man8/zpool.8
@@ -1820,6 +1820,10 @@ Amount of time IO spent in asynchronous priority queues.  Does not include
 disk time.
 .Ar scrub :
 Amount of time IO spent in scrub queue. Does not include disk time.
+.Ar atrim :
+Amount of time IO spent doing a trim operation during an autotrim.
+.Ar mtrim :
+Amount of time IO spent doing a trim operation during a manual trim.
 .It Fl l
 Include average latency statistics:
 .Pp

tonyhutter · 2019-01-24T23:41:44Z

You'll want to add the new trim_start and trim_end events to man/man5/zfs-events.5.

richardelling · 2019-01-25T23:18:06Z

tests/zfs-tests/tests/functional/cli_root/zpool_trim/cleanup.ksh

+# You may not use this file except in compliance with the License.
+#
+# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+# or http://www.opensolaris.org/os/licensing.


opensolaris.org is no more. Please refer to http://opensource.org/licenses/CDDL-1.0

tonyhutter · 2019-01-25T23:38:46Z

tests/runfiles/linux.run

+tests = ['autotrim_001_pos', 'autotrim_002_pos', 'manualtrim_001_pos',
+    'manualtrim_002_pos', 'manualtrim_003_pos', 'manualtrim_004_pos',
+    'manualtrim_005_pos']
+


Add tags = ['functional', 'trim']

dweeezil · 2019-02-15T17:54:24Z

I'm going to be closing this PR. @behlendorf has been doing some major re-working of the original TRIM work to, among other things:

Harmonize its operation with that of device initialization so it's possible, for example, to TRIM individual vdevs.
Add per-vdev trim stats
Re-work the underlying discard mechanism to use a new ZIO type rather than the DKIOCFREE ioctl.
Utilize the new "xlat" vdev method from device initialization where appropriate.
Switch from the legacy ioctl interface to the new nvlist-style ioctl interface.
Plenty of other restructuring, in particular w.r.t. the management of the trim threads.

Expect a new PR to be posted soon.

UNMAP/TRIM support is a frequently-requested feature to help prevent performance from degrading on SSDs and on various other SAN-like storage back-ends. By issuing UNMAP/TRIM commands for sectors which are no longer allocated the underlying device can often more efficiently manage itself. This TRIM implementation is modeled on the `zpool initialize` feature which writes a pattern to all unallocated space in the pool. The new `zpool trim` command uses the same vdev_xlate() code to calculate what sectors are unallocated, the same per- vdev TRIM thread model and locking, and the same basic CLI for a consistent user experience. The core difference is that instead of writing a pattern it will issue UNMAP/TRIM commands for those extents. The zio pipeline was updated to accommodate this by adding a new ZIO_TYPE_TRIM type and associated spa taskq. This new type makes is straight forward to add the platform specific TRIM/UNMAP calls to vdev_disk.c and vdev_file.c. These new ZIO_TYPE_TRIM zios are handled largely the same way as ZIO_TYPE_READs or ZIO_TYPE_WRITEs. This makes it possible to largely avoid changing the pipieline, one exception is that TRIM zio's may exceed the 16M block size limit since they contain no data. In addition to the manual `zpool trim` command, a background automatic TRIM was added and is controlled by the 'autotrim' property. It relies on the exact same infrastructure as the manual TRIM. However, instead of relying on the extents in a metaslab's ms_allocatable range tree, a ms_trim tree is kept per metaslab. When 'autotrim=on', ranges added back to the ms_allocatable tree are also added to the ms_free tree. The ms_free tree is then periodically consumed by an autotrim thread which systematically walks a top level vdev's metaslabs. Since the automatic TRIM will skip ranges it considers too small there is value in occasionally running a full `zpool trim`. This may occur when the freed blocks are small and not enough time was allowed to aggregate them. An automatic TRIM and a manual `zpool trim` may be run concurrently, in which case the automatic TRIM will yield to the manual TRIM. Reviewed-by: Jorgen Lundman <[email protected]> Reviewed-by: Tim Chase <[email protected]> Reviewed-by: Matt Ahrens <[email protected]> Reviewed-by: George Wilson <[email protected]> Reviewed-by: Serapheim Dimitropoulos <[email protected]> Contributions-by: Saso Kiselkov <[email protected]> Contributions-by: Tim Chase <[email protected]> Contributions-by: Chunwei Chen <[email protected]> Signed-off-by: Brian Behlendorf <[email protected]> Closes #8419 Closes #598

This was referenced Jan 8, 2019

OpenZFS - 6363 Add UNMAP/TRIM functionality #5925

Closed

OpenZFS - 6363 Add UNMAP/TRIM functionality (v2) #7363

Closed

behlendorf added the Status: Code Review Needed Ready for review and testing label Jan 9, 2019

dweeezil requested a review from behlendorf January 14, 2019 04:25

behlendorf reviewed Jan 17, 2019

View reviewed changes

behlendorf added the Type: Feature Feature request or new feature label Jan 18, 2019

ahrens reviewed Jan 20, 2019

View reviewed changes

dweeezil force-pushed the ntrim3 branch 2 times, most recently from d70be42 to 9efe6c0 Compare January 21, 2019 14:16

dweeezil and others added 18 commits January 21, 2019 08:31

Refresh dkio.h and add dkioc_free_util.h

21306c0

Update dkio.h from Nexenta's version to pick up DKIOCFREE and add their dkioc_free_util.h header for TRIM support. Requires-builders: none

Trimming an offlined vdev asserts in zio_create.

1748965

Requires-builders: none

Matt Ahrens' review comments, round 2. Brian Behlendorf's review comm…

0f0c103

…ents. Requires-builders: none

Review feedback

f05d5e0

Requires-builders: none

Fix abd_alloc_sametype() panic

957d56f

Signed-off-by: Brian Behlendorf <[email protected]> Requires-builders: none

Matt Ahrens' review comments.

d27976d

Porting Notes: Man page changes dropped for the moment. This can be reconsiled when the final version is merged to OpenZFS. They are accurate now, only worded a little differently. Requires-builders: none

Tim Chase's review comments, round 2.

76439d5

Porting Notes: * metaslab_sync changes already applied. * resync of test cases needed Requires-builders: none

Deadlockiness associated with doing postponing trimming on a metaslab…

5614f2b

… wanting to condense. Requires-builders: none

Matt Ahrens' review comments, round 5.

feae3c2

Requires-builders: none

Deadlockiness in autotrim due to recent changes.

7d12663

Requires-builders: none

Review feedback

46d31de

* Rename TRIM taskq threads to be more concise for Linux. * Fix divide by zero panic Signed-off-by: Brian Behlendorf <[email protected]> Requires-builders: none

behlendorf and others added 19 commits January 21, 2019 08:32

Review feedback 2

ae1457d

* Fixed missing taskq_destroy when exporting a pool which is being actively trimmed. * Add auto/manual TRIM coverage to ztest. * Temporarily disable manualtrim_004_pos. Signed-off-by: Brian Behlendorf <[email protected]> Requires-builders: none

Add trim manpage

8b88b3e

Signed-off-by: Chunwei Chen <[email protected]> Requires-builders: none

Fix wrong logical operator

977c20e

Signed-off-by: Chunwei Chen <[email protected]> Requires-builders: none

Wait for 1 sec before check trim status

3670596

Signed-off-by: Chunwei Chen <[email protected]> Requires-builders: none

Clean-ups following rebase to master

f19d6f6

Signed-off-by: Tim Chase <[email protected]> Requires-builders: none

ZIO_PIPELINE_CONTINUE fix

ec4e894

Requires-builders: none

More fixups

0b8cc39

Requires-builders: none

Add tags to trim test cases

26fdfdb

NOTE: should be squashed into a previous commit Requires-builders: none

Preserve activation flags when sorting metaslabs

3c5eb9c

Also, comment-out the ASSERT(!metaslab_should_allocate(msp, asize)); in metaslab_group_alloc_normal(). It seems that the additional metaslab_group_sort() performed by trim makes this assertion invalid. Requires-builders: none

Trim should skip removed devices

014e259

Requires-builders: none

Don't dereference null vdev_ms

af4a108

Requires-builders: none

Account for non-concrete vdevs in spa_num_auto_trimming

c641ee6

Also, stop trimming in places where initialization is stopped: spa_vdev_remove_log() and spa_vdev_remove_top(). Requires-builders: none

Use proper tag for spa config refcounts

f8e5760

NOTE: This should be its own separate PR. It was discovered during debugging of the TRIM work. Requires-builders: none

Re-instate tracked spa config refcounts

c18f20a

Requires-builders: none

Skip non-concrete vdevs...

562bb92

In trim_stop_set() and trim_stop_wait(). Requires-builders: none

Move the auto trim taskq start-up on import

e38d791

The change in the previous commit to not attempt to create the taskq if a spa unload is in-progress caused the auto trim taskq to not be started upon import. Requires-builders: none

Set spa->spa_auto_trim on pool creation

8d4a118

Rather than using the default value for the property, we need to use the value which might be set by "zpool create -o autotrim=". Requires-builders: none

dweeezil force-pushed the ntrim3 branch from 9efe6c0 to 3a184b8 Compare January 21, 2019 14:40

tonyhutter reviewed Jan 24, 2019

View reviewed changes

richardelling reviewed Jan 25, 2019

View reviewed changes

This comment has been minimized.

Sign in to view

tonyhutter reviewed Jan 25, 2019

View reviewed changes

This comment has been minimized.

Sign in to view

dweeezil closed this Feb 15, 2019

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add TRIM support - replaces #5925 and #7363 #8255

Add TRIM support - replaces #5925 and #7363 #8255

dweeezil commented Jan 8, 2019

codecov bot commented Jan 9, 2019 •

edited

Loading

dweeezil commented Jan 14, 2019

behlendorf Jan 17, 2019

behlendorf Jan 17, 2019

behlendorf Jan 17, 2019

behlendorf Jan 17, 2019

behlendorf Jan 17, 2019

behlendorf Jan 17, 2019

behlendorf Jan 17, 2019

behlendorf Jan 17, 2019

behlendorf Jan 17, 2019

behlendorf Jan 17, 2019

ahrens Jan 20, 2019

ahrens Jan 20, 2019

dweeezil Jan 23, 2019

ahrens Jan 20, 2019

ahrens Jan 20, 2019

ahrens Jan 20, 2019

ahrens Jan 20, 2019

ahrens Jan 20, 2019

dweeezil commented Jan 21, 2019 •

edited

Loading

tonyhutter Jan 24, 2019

tonyhutter commented Jan 24, 2019

tonyhutter commented Jan 24, 2019

richardelling Jan 25, 2019

This comment has been minimized.

tonyhutter Jan 25, 2019

This comment has been minimized.

dweeezil commented Feb 15, 2019

Add TRIM support - replaces #5925 and #7363 #8255

Add TRIM support - replaces #5925 and #7363 #8255

Conversation

dweeezil commented Jan 8, 2019

Motivation and Context

Description

How Has This Been Tested?

Types of changes

Checklist:

codecov bot commented Jan 9, 2019 • edited Loading

Codecov Report

dweeezil commented Jan 14, 2019

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

dweeezil commented Jan 21, 2019 • edited Loading

Choose a reason for hiding this comment

tonyhutter commented Jan 24, 2019

tonyhutter commented Jan 24, 2019

Choose a reason for hiding this comment

This comment has been minimized.

Choose a reason for hiding this comment

This comment has been minimized.

dweeezil commented Feb 15, 2019

codecov bot commented Jan 9, 2019 •

edited

Loading

dweeezil commented Jan 21, 2019 •

edited

Loading