summaryrefslogtreecommitdiff
path: root/Documentation
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2021-02-21 10:34:36 -0800
committerLinus Torvalds <torvalds@linux-foundation.org>2021-02-21 10:34:36 -0800
commitb52bb135aad99deea9bfe5f050c3295b049adc87 (patch)
treef5282961c881e3077cf14ae42bef9f800bfee6aa /Documentation
parent4f016a316f2243efb0d1c0e7259f07817eb99e67 (diff)
parent1cd738b13ae9b29e03d6149f0246c61f76e81fcf (diff)
Merge tag 'xfs-5.12-merge-5' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux
Pull xfs updates from Darrick Wong: "There's a lot going on this time, which seems about right for this drama-filled year. Community developers added some code to speed up freezing when read-only workloads are still running, refactored the logging code, added checks to prevent file extent counter overflow, reduced iolock cycling to speed up fsync and gc scans, and started the slow march towards supporting filesystem shrinking. There's a huge refactoring of the internal speculative preallocation garbage collection code which fixes a bunch of bugs, makes the gc scheduling per-AG and hence multithreaded, and standardizes the retry logic when we try to reserve space or quota, can't, and want to trigger a gc scan. We also enable multithreaded quotacheck to reduce mount times further. This is also preparation for background file gc, which may or may not land for 5.13. We also fixed some deadlocks in the rename code, fixed a quota accounting leak when FSSETXATTR fails, restored the behavior that write faults to an mmap'd region actually cause a SIGBUS, fixed a bug where sgid directory inheritance wasn't quite working properly, and fixed a bug where symlinks weren't working properly in ecryptfs. We also now advertise the inode btree counters feature that was introduced two cycles ago. Summary: - Fix an ABBA deadlock when renaming files on overlayfs. - Make sure that we can't overflow the inode extent counters when adding to or removing extents from a file. - Make directory sgid inheritance work the same way as all the other filesystems. - Don't drain the buffer cache on freeze and ro remount, which should reduce the amount of time if read-only workloads are continuing during the freeze. - Fix a bug where symlink size isn't reported to the vfs in ecryptfs. - Disentangle log cleaning from log covering. This refactoring sets us up for future changes to the log, though for now it simply means that we can use covering for freezes, and cleaning becomes something we only do at unmount. - Speed up file fsyncs by reducing iolock cycling. - Fix delalloc blocks leaking when changing the project id fails because of input validation errors in FSSETXATTR. - Fix oversized quota reservation when converting unwritten extents during a DAX write. - Create a transaction allocation helper function to standardize the idiom of allocating a transaction, reserving blocks, locking inodes, and reserving quota. Replace all the open-coded logic for file creation, file ownership changes, and file modifications to use them. - Actually shut down the fs if the incore quota reservations get corrupted. - Fix background block garbage collection scans to not block and to actually clean out CoW staging extents properly. - Run block gc scans when we run low on project quota. - Use the standardized transaction allocation helpers to make it so that ENOSPC and EDQUOT errors during reservation will back out, invoke the block gc scanner, and try again. This is preparation for introducing background inode garbage collection in the next cycle. - Combine speculative post-EOF block garbage collection with speculative copy on write block garbage collection. - Enable multithreaded quotacheck. - Allow sysadmins to tweak the CPU affinities and maximum concurrency levels of quotacheck and background blockgc worker pools. - Expose the inode btree counter feature in the fs geometry ioctl. - Cleanups of the growfs code in preparation for starting work on filesystem shrinking. - Fix all the bloody gcc warnings that the maintainer knows about. :P - Fix a RST syntax error. - Don't trigger bmbt corruption assertions after the fs shuts down. - Restore behavior of forcing SIGBUS on a shut down filesystem when someone triggers a mmap write fault (or really, any buffered write)" * tag 'xfs-5.12-merge-5' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux: (85 commits) xfs: consider shutdown in bmapbt cursor delete assert xfs: fix boolreturn.cocci warnings xfs: restore shutdown check in mapped write fault path xfs: fix rst syntax error in admin guide xfs: fix incorrect root dquot corruption error when switching group/project quota types xfs: get rid of xfs_growfs_{data,log}_t xfs: rename `new' to `delta' in xfs_growfs_data_private() libxfs: expose inobtcount in xfs geometry xfs: don't bounce the iolock between free_{eof,cow}blocks xfs: expose the blockgc workqueue knobs publicly xfs: parallelize block preallocation garbage collection xfs: rename block gc start and stop functions xfs: only walk the incore inode tree once per blockgc scan xfs: consolidate the eofblocks and cowblocks workers xfs: consolidate incore inode radix tree posteof/cowblocks tags xfs: remove trivial eof/cowblocks functions xfs: hide xfs_icache_free_cowblocks xfs: hide xfs_icache_free_eofblocks xfs: relocate the eofb/cowb workqueue functions xfs: set WQ_SYSFS on all workqueues in debug mode ...
Diffstat (limited to 'Documentation')
-rw-r--r--Documentation/admin-guide/xfs.rst42
1 files changed, 42 insertions, 0 deletions
diff --git a/Documentation/admin-guide/xfs.rst b/Documentation/admin-guide/xfs.rst
index 86de8a1ad91c..6178153d3320 100644
--- a/Documentation/admin-guide/xfs.rst
+++ b/Documentation/admin-guide/xfs.rst
@@ -495,3 +495,45 @@ the class and error context. For example, the default values for
"metadata/ENODEV" are "0" rather than "-1" so that this error handler defaults
to "fail immediately" behaviour. This is done because ENODEV is a fatal,
unrecoverable error no matter how many times the metadata IO is retried.
+
+Workqueue Concurrency
+=====================
+
+XFS uses kernel workqueues to parallelize metadata update processes. This
+enables it to take advantage of storage hardware that can service many IO
+operations simultaneously. This interface exposes internal implementation
+details of XFS, and as such is explicitly not part of any userspace API/ABI
+guarantee the kernel may give userspace. These are undocumented features of
+the generic workqueue implementation XFS uses for concurrency, and they are
+provided here purely for diagnostic and tuning purposes and may change at any
+time in the future.
+
+The control knobs for a filesystem's workqueues are organized by task at hand
+and the short name of the data device. They all can be found in:
+
+ /sys/bus/workqueue/devices/${task}!${device}
+
+================ ===========
+ Task Description
+================ ===========
+ xfs_iwalk-$pid Inode scans of the entire filesystem. Currently limited to
+ mount time quotacheck.
+ xfs-blockgc Background garbage collection of disk space that have been
+ speculatively allocated beyond EOF or for staging copy on
+ write operations.
+================ ===========
+
+For example, the knobs for the quotacheck workqueue for /dev/nvme0n1 would be
+found in /sys/bus/workqueue/devices/xfs_iwalk-1111!nvme0n1/.
+
+The interesting knobs for XFS workqueues are as follows:
+
+============ ===========
+ Knob Description
+============ ===========
+ max_active Maximum number of background threads that can be started to
+ run the work.
+ cpumask CPUs upon which the threads are allowed to run.
+ nice Relative priority of scheduling the threads. These are the
+ same nice levels that can be applied to userspace processes.
+============ ===========