Since a28334862993 ("page cache: Finish XArray conversion"), on most
major Linux distributions, the page cache doesn't correctly transition
when the hot data set is changing, and leaves the new pages thrashing
indefinitely instead of kicking out the cold ones.
On a freshly booted, freshly ssh'd into virtual machine with 1G RAM
running stock Arch Linux:
[root@ham ~]# ./reclaimtest.sh
+ dd of=workingset-a bs=1M count=0 seek=600
+ cat workingset-a
+ cat workingset-a
+ cat workingset-a
+ cat workingset-a
+ cat workingset-a
+ cat workingset-a
+ cat workingset-a
+ cat workingset-a
+ ./mincore workingset-a
153600/153600 workingset-a
+ dd of=workingset-b bs=1M count=0 seek=600
+ cat workingset-b
+ cat workingset-b
+ cat workingset-b
+ cat workingset-b
+ ./mincore workingset-a workingset-b
104029/153600 workingset-a
120086/153600 workingset-b
+ cat workingset-b
+ cat workingset-b
+ cat workingset-b
+ cat workingset-b
+ ./mincore workingset-a workingset-b
104029/153600 workingset-a
120268/153600 workingset-b
workingset-b is a 600M file on a 1G host that is otherwise entirely
idle. No matter how often it's being accessed, it won't get cached.
While investigating, I noticed that the non-resident information gets
aggressively reclaimed - /proc/vmstat::workingset_nodereclaim. This is
a problem because a workingset transition like this relies on the
non-resident information tracked in the page cache tree of evicted
file ranges: when the cache faults are refaults of recently evicted
cache, we challenge the existing active set, and that allows a new
workingset to establish itself.
Tracing the shrinker that maintains this memory revealed that all page
cache tree nodes were allocated to the root cgroup. This is a problem,
because 1) the shrinker sizes the amount of non-resident information
it keeps to the size of the cgroup's other memory and 2) on most major
Linux distributions, only kernel threads live in the root cgroup and
everything else gets put into services or session groups:
[root@ham ~]# cat /proc/self/cgroup
0::/user.slice/user-0.slice/session-c1.scope
As a result, we basically maintain no non-resident information for the
workloads running on the system, thus breaking the caching algorithm.
Looking through the code, I found the culprit in the above-mentioned
patch: when switching from the radix tree to xarray, it dropped the
__GFP_ACCOUNT flag from the tree node allocations - the flag that
makes sure the allocated memory gets charged to and tracked by the
cgroup of the calling process - in this case, the one doing the fault.
To fix this, allow xarray users to specify per-tree flag that makes
xarray allocate nodes using __GFP_ACCOUNT. Then restore the page cache
tree annotation to request such cgroup tracking for the cache nodes.
With this patch applied, the page cache correctly converges on new
workingsets again after just a few iterations:
[root@ham ~]# ./reclaimtest.sh
+ dd of=workingset-a bs=1M count=0 seek=600
+ cat workingset-a
+ cat workingset-a
+ cat workingset-a
+ cat workingset-a
+ cat workingset-a
+ cat workingset-a
+ cat workingset-a
+ cat workingset-a
+ ./mincore workingset-a
153600/153600 workingset-a
+ dd of=workingset-b bs=1M count=0 seek=600
+ cat workingset-b
+ ./mincore workingset-a workingset-b
124607/153600 workingset-a
87876/153600 workingset-b
+ cat workingset-b
+ ./mincore workingset-a workingset-b
81313/153600 workingset-a
133321/153600 workingset-b
+ cat workingset-b
+ ./mincore workingset-a workingset-b
63036/153600 workingset-a
153600/153600 workingset-b
Cc: stable@vger.kernel.org # 4.20+
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Add SPDX license identifiers to all files which: - Have no license information of any form - Have EXPORT_.*_SYMBOL_GPL inside which was used in the initial scan/conversion to ignore the file These files fall under the project license, GPL v2 only. The resulting SPDX license identifier is: GPL-2.0-only Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Pull misc vfs updates from Al Viro: "Assorted stuff, with no common topic whatsoever..." * 'work.misc' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs: libfs: document simple_get_link() Documentation/filesystems/Locking: fix ->get_link() prototype Documentation/filesystems/vfs.txt: document how ->i_link works Documentation/filesystems/vfs.txt: remove bogus "Last updated" date fs: use timespec64 in relatime_need_update fs/block_dev.c: remove unused include
Pull vfs inode freeing updates from Al Viro: "Introduction of separate method for RCU-delayed part of ->destroy_inode() (if any). Pretty much as posted, except that destroy_inode() stashes ->free_inode into the victim (anon-unioned with ->i_fops) before scheduling i_callback() and the last two patches (sockfs conversion and folding struct socket_wq into struct socket) are excluded - that pair should go through netdev once davem reopens his tree" * 'work.icache' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs: (58 commits) orangefs: make use of ->free_inode() shmem: make use of ->free_inode() hugetlb: make use of ->free_inode() overlayfs: make use of ->free_inode() jfs: switch to ->free_inode() fuse: switch to ->free_inode() ext4: make use of ->free_inode() ecryptfs: make use of ->free_inode() ceph: use ->free_inode() btrfs: use ->free_inode() afs: switch to use of ->free_inode() dax: make use of ->free_inode() ntfs: switch to ->free_inode() securityfs: switch to ->free_inode() apparmor: switch to ->free_inode() rpcpipe: switch to ->free_inode() bpf: switch to ->free_inode() mqueue: switch to ->free_inode() ufs: switch to ->free_inode() coda: switch to ->free_inode() ...
A lot of ->destroy_inode() instances end with call_rcu() of a callback that does RCU-delayed part of freeing. Introduce a new method for doing just that, with saner signature. Rules: ->destroy_inode ->free_inode f g immediate call of f(), RCU-delayed call of g() f NULL immediate call of f(), no RCU-delayed calls NULL g RCU-delayed call of g() NULL NULL RCU-delayed default freeing IOW, NULL ->free_inode gives the same behaviour as now. Note that NULL, NULL is equivalent to NULL, free_inode_nonrcu; we could mandate the latter form, but that would have very little benefit beyond making rules a bit more symmetric. It would break backwards compatibility, require extra boilerplate and expected semantics for (NULL, NULL) pair would have no use whatsoever... Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
file_remove_privs() might be called for non-regular files, e.g. blkdev inode. There is no reason to do its job on things like blkdev inodes, pipes, or cdevs. Hence, abort if file does not refer to a regular inode. AV: more to the point, for devices there might be any number of inodes refering to given device. Which one to strip the permissions from, even if that made any sense in the first place? All of them will be observed with contents modified, after all. Found by LockDoc (Alexander Lochmann, Horst Schirmeier and Olaf Spinczyk) Reviewed-by: Jan Kara <jack@suse.cz> Signed-off-by: Alexander Lochmann <alexander.lochmann@tu-dortmund.de> Signed-off-by: Horst Schirmeier <horst.schirmeier@tu-dortmund.de> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
For some reason, the conversion of the VFS code away from 'struct timespec' left one function behind that still uses it, for absolutely no reason. Using timespec64 will make the atime update logic work correctly past y2038. Signed-off-by: Arnd Bergmann <arnd@arndb.de> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
It seems that commits 5f16f3225b0624 and 00a1a053ebe5, both with same
commitlog ("ext4: atomically set inode->i_flags in ext4_set_inode_flags()")
introduced the set_mask_bits API, but somehow missed not using it in ext4
in the end.
Also, set_mask_bits() is used in fs quite a bit and we can possibly come
up with a generic llsc based implementation (w/o the cmpxchg loop)
Link: http://lkml.kernel.org/r/1548275584-18096-3-git-send-email-vgupta@synopsys.com
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
Reviewed-by: Anthony Yznaga <anthony.yznaga@oracle.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Theodore Ts'o <tytso@mit.edu>
Cc: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jani Nikula <jani.nikula@intel.com>
Cc: Miklos Szeredi <mszeredi@redhat.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This reverts commit a76cf1a474d7d ("mm: don't reclaim inodes with many
attached pages").
This change causes serious changes to page cache and inode cache
behaviour and balance, resulting in major performance regressions when
combining worklaods such as large file copies and kernel compiles.
https://bugzilla.kernel.org/show_bug.cgi?id=202441
This change is a hack to work around the problems introduced by changing
how agressive shrinkers are on small caches in commit 172b06c32b94 ("mm:
slowly shrink slabs with a relatively small number of objects"). It
creates more problems than it solves, wasn't adequately reviewed or
tested, so it needs to be reverted.
Link: http://lkml.kernel.org/r/20190130041707.27750-2-david@fromorbit.com
Fixes: a76cf1a474d7d ("mm: don't reclaim inodes with many attached pages")
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Cc: Wolfgang Walter <linux@stwm.de>
Cc: Roman Gushchin <guro@fb.com>
Cc: Spock <dairinin@gmail.com>
Cc: Rik van Riel <riel@surriel.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Pull y2038 updates from Arnd Bergmann:
"More syscalls and cleanups
This concludes the main part of the system call rework for 64-bit
time_t, which has spread over most of year 2018, the last six system
calls being
- ppoll
- pselect6
- io_pgetevents
- recvmmsg
- futex
- rt_sigtimedwait
As before, nothing changes for 64-bit architectures, while 32-bit
architectures gain another entry point that differs only in the layout
of the timespec structure. Hopefully in the next release we can wire
up all 22 of those system calls on all 32-bit architectures, which
gives us a baseline version for glibc to start using them.
This does not include the clock_adjtime, getrusage/waitid, and
getitimer/setitimer system calls. I still plan to have new versions of
those as well, but they are not required for correct operation of the
C library since they can be emulated using the old 32-bit time_t based
system calls.
Aside from the system calls, there are also a few cleanups here,
removing old kernel internal interfaces that have become unused after
all references got removed. The arch/sh cleanups are part of this,
there were posted several times over the past year without a reaction
from the maintainers, while the corresponding changes made it into all
other architectures"
* tag 'y2038-for-4.21' of ssh://gitolite.kernel.org:/pub/scm/linux/kernel/git/arnd/playground:
timekeeping: remove obsolete time accessors
vfs: replace current_kernel_time64 with ktime equivalent
timekeeping: remove timespec_add/timespec_del
timekeeping: remove unused {read,update}_persistent_clock
sh: remove board_time_init() callback
sh: remove unused rtc_sh_get/set_time infrastructure
sh: sh03: rtc: push down rtc class ops into driver
sh: dreamcast: rtc: push down rtc class ops into driver
y2038: signal: Add compat_sys_rt_sigtimedwait_time64
y2038: signal: Add sys_rt_sigtimedwait_time32
y2038: socket: Add compat_sys_recvmmsg_time64
y2038: futex: Add support for __kernel_timespec
y2038: futex: Move compat implementation into futex.c
io_pgetevents: use __kernel_timespec
pselect6: use __kernel_timespec
ppoll: use __kernel_timespec
signal: Add restore_user_sigmask()
signal: Add set_user_sigmask()