linux/mm
Johannes Weiner 51b8c1fe25 vfs: keep inodes with page cache off the inode shrinker LRU
Historically (pre-2.5), the inode shrinker used to reclaim only empty
inodes and skip over those that still contained page cache.  This caused
problems on highmem hosts: struct inode could put fill lowmem zones
before the cache was getting reclaimed in the highmem zones.

To address this, the inode shrinker started to strip page cache to
facilitate reclaiming lowmem.  However, this comes with its own set of
problems: the shrinkers may drop actively used page cache just because
the inodes are not currently open or dirty - think working with a large
git tree.  It further doesn't respect cgroup memory protection settings
and can cause priority inversions between containers.

Nowadays, the page cache also holds non-resident info for evicted cache
pages in order to detect refaults.  We've come to rely heavily on this
data inside reclaim for protecting the cache workingset and driving swap
behavior.  We also use it to quantify and report workload health through
psi.  The latter in turn is used for fleet health monitoring, as well as
driving automated memory sizing of workloads and containers, proactive
reclaim and memory offloading schemes.

The consequences of dropping page cache prematurely is that we're seeing
subtle and not-so-subtle failures in all of the above-mentioned
scenarios, with the workload generally entering unexpected thrashing
states while losing the ability to reliably detect it.

To fix this on non-highmem systems at least, going back to rotating
inodes on the LRU isn't feasible.  We've tried (commit a76cf1a474
("mm: don't reclaim inodes with many attached pages")) and failed
(commit 69056ee6a8 ("Revert "mm: don't reclaim inodes with many
attached pages"")).

The issue is mostly that shrinker pools attract pressure based on their
size, and when objects get skipped the shrinkers remember this as
deferred reclaim work.  This accumulates excessive pressure on the
remaining inodes, and we can quickly eat into heavily used ones, or
dirty ones that require IO to reclaim, when there potentially is plenty
of cold, clean cache around still.

Instead, this patch keeps populated inodes off the inode LRU in the
first place - just like an open file or dirty state would.  An otherwise
clean and unused inode then gets queued when the last cache entry
disappears.  This solves the problem without reintroducing the reclaim
issues, and generally is a bit more scalable than having to wade through
potentially hundreds of thousands of busy inodes.

Locking is a bit tricky because the locks protecting the inode state
(i_lock) and the inode LRU (lru_list.lock) don't nest inside the
irq-safe page cache lock (i_pages.xa_lock).  Page cache deletions are
serialized through i_lock, taken before the i_pages lock, to make sure
depopulated inodes are queued reliably.  Additions may race with
deletions, but we'll check again in the shrinker.  If additions race
with the shrinker itself, we're protected by the i_lock: if find_inode()
or iput() win, the shrinker will bail on the elevated i_count or
I_REFERENCED; if the shrinker wins and goes ahead with the inode, it
will set I_FREEING and inhibit further igets(), which will cause the
other side to create a new instance of the inode instead.

Link: https://lkml.kernel.org/r/20210614211904.14420-4-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Roman Gushchin <guro@fb.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Dave Chinner <david@fromorbit.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-11-09 10:02:48 -08:00
..
damon mm/damon: remove return value from before_terminate callback 2021-11-06 13:30:46 -07:00
kasan kasan: arm64: fix pcpu_page_first_chunk crash with KASAN_VMALLOC 2021-11-06 13:30:37 -07:00
kfence kfence: always use static branches to guard kfence_alloc() 2021-11-06 13:30:43 -07:00
backing-dev.c mm/vmscan: throttle reclaim until some writeback completes if congested 2021-11-06 13:30:40 -07:00
balloon_compaction.c mm: fix typos in comments 2021-05-07 00:26:35 -07:00
bootmem_info.c mm/bootmem_info.c: mark __init on register_page_bootmem_info_section 2021-09-03 09:58:14 -07:00
cleancache.c
cma_debug.c mm/cma: change cma mutex to irq safe spinlock 2021-05-05 11:27:21 -07:00
cma_sysfs.c mm: cma: support sysfs 2021-05-05 11:27:24 -07:00
cma.c memblock: rename memblock_free to memblock_phys_free 2021-11-06 13:30:41 -07:00
cma.h mm: cma: support sysfs 2021-05-05 11:27:24 -07:00
compaction.c mm/vmscan: centralise timeout values for reclaim_throttle 2021-11-06 13:30:40 -07:00
debug_page_ref.c
debug_vm_pgtable.c mm: debug_vm_pgtable: don't use __P000 directly 2021-11-06 13:30:33 -07:00
debug.c mm/migrate: de-duplicate migrate_reason strings 2021-11-06 13:30:41 -07:00
dmapool.c mm/dmapool: use DEVICE_ATTR_RO macro 2021-06-29 10:53:52 -07:00
early_ioremap.c mm/early_ioremap.c: remove redundant early_ioremap_shutdown() 2021-09-08 11:50:24 -07:00
fadvise.c
failslab.c
filemap.c vfs: keep inodes with page cache off the inode shrinker LRU 2021-11-09 10:02:48 -08:00
frontswap.c mm/mempool: minor coding style tweaks 2021-05-05 11:27:27 -07:00
gup_test.c selftests/vm: gup_test: test faulting in kernel, and verify pinnable pages 2021-05-05 11:27:26 -07:00
gup_test.h selftests/vm: gup_test: fix test flag 2021-05-05 11:27:26 -07:00
gup.c mm/gup: further simplify __gup_device_huge() 2021-11-06 13:30:34 -07:00
highmem.c mm/highmem: remove deprecated kmap_atomic 2021-11-06 13:30:43 -07:00
hmm.c mm/hmm: bypass devmap pte when all pfn requested flags are fulfilled 2021-09-08 18:45:52 -07:00
huge_memory.c mm: filemap: check if THP has hwpoisoned subpage for PMD page fault 2021-10-28 17:18:55 -07:00
hugetlb_cgroup.c hugetlb_cgroup: remove unused hugetlb_cgroup_from_counter macro 2021-11-06 13:30:39 -07:00
hugetlb_vmemmap.c mm: hugetlb: introduce CONFIG_HUGETLB_PAGE_FREE_VMEMMAP_DEFAULT_ON 2021-06-30 20:47:26 -07:00
hugetlb_vmemmap.h mm: hugetlb: introduce nr_free_vmemmap_pages in the struct hstate 2021-06-30 20:47:25 -07:00
hugetlb.c hugetlbfs: extend the definition of hugepages parameter to support node allocation 2021-11-06 13:30:41 -07:00
hwpoison-inject.c mm: hwpoison: don't drop slab caches for offlining non-LRU page 2021-09-03 09:58:15 -07:00
init-mm.c mm: add setup_initial_init_mm() helper 2021-07-08 11:48:21 -07:00
internal.h mm/vmscan: centralise timeout values for reclaim_throttle 2021-11-06 13:30:40 -07:00
interval_tree.c mm/interval_tree: add comments to improve code readability 2021-04-30 11:20:38 -07:00
io-mapping.c mm: add a io_mapping_map_user helper 2021-04-30 11:20:39 -07:00
ioremap.c mm: move ioremap_page_range to vmalloc.c 2021-09-08 11:50:24 -07:00
Kconfig mm/memory_hotplug: restrict CONFIG_MEMORY_HOTPLUG to 64 bit 2021-11-06 13:30:42 -07:00
Kconfig.debug
khugepaged.c mm: khugepaged: recalculate min_free_kbytes after stopping khugepaged 2021-11-06 13:30:39 -07:00
kmemleak.c mm/kmemleak: allow __GFP_NOLOCKDEP passed to kmemleak's gfp 2021-09-08 18:45:53 -07:00
ksm.c mm/ksm: remove old GCC 4.9+ check 2021-09-13 10:18:28 -07:00
list_lru.c mm: list_lru: only add memcg-aware lrus to the global lru list 2021-11-06 13:30:35 -07:00
maccess.c ARM: 9115/1: mm/maccess: fix unaligned copy_{from,to}_kernel_nofault 2021-08-20 11:39:25 +01:00
madvise.c Merge branch 'akpm' (patches from Andrew) 2021-09-03 10:08:28 -07:00
Makefile mm: introduce Data Access MONitor (DAMON) 2021-09-08 11:50:24 -07:00
mapping_dirty_helpers.c mm/mapping_dirty_helpers: remove double Note in kerneldoc 2021-07-01 11:06:02 -07:00
memblock.c memblock: add MEMBLOCK_DRIVER_MANAGED to mimic IORESOURCE_SYSRAM_DRIVER_MANAGED 2021-11-06 13:30:42 -07:00
memcontrol.c mm/vmscan: throttle reclaim when no progress is being made 2021-11-06 13:30:40 -07:00
memfd.c Reimplement RLIMIT_MEMLOCK on top of ucounts 2021-04-30 14:14:02 -05:00
memory_hotplug.c mm/memory_hotplug: indicate MEMBLOCK_DRIVER_MANAGED with IORESOURCE_SYSRAM_DRIVER_MANAGED 2021-11-06 13:30:42 -07:00
memory-failure.c mm: hwpoison: handle non-anonymous THP correctly 2021-11-06 13:30:38 -07:00
memory.c mm: remove redundant smp_wmb() 2021-11-06 13:30:36 -07:00
mempolicy.c mm: migrate: make demotion knob depend on migration 2021-11-06 13:30:41 -07:00
mempool.c kasan: use separate (un)poison implementation for integrated init 2021-06-04 19:32:21 +01:00
memremap.c mm/memory_hotplug: remove nid parameter from arch_remove_memory() 2021-09-08 11:50:23 -07:00
memtest.c
migrate.c mm: migrate: make demotion knob depend on migration 2021-11-06 13:30:41 -07:00
mincore.c
mlock.c mm: introduce memfd_secret system call to create "secret" memory areas 2021-07-08 11:48:21 -07:00
mm_init.c include/linux/page-flags-layout.h: cleanups 2021-04-30 11:20:42 -07:00
mmap_lock.c mm: mmap_lock: fix disabling preemption directly 2021-07-23 17:43:28 -07:00
mmap.c mm/mmap.c: fix a data race of mm->total_vm 2021-11-06 13:30:35 -07:00
mmu_gather.c
mmu_notifier.c
mmzone.c
mprotect.c mm/mprotect.c: avoid repeated assignment in do_mprotect_pkey() 2021-11-06 13:30:36 -07:00
mremap.c mm, hugepages: add mremap() support for hugepage backed vma 2021-11-06 13:30:39 -07:00
msync.c mm/msync: exit early when the flags is an MS_ASYNC and start < vm_start 2021-04-30 11:20:37 -07:00
nommu.c mm: nommu: kill arch_get_unmapped_area() 2021-11-06 13:30:41 -07:00
oom_kill.c mm: mark the OOM reaper thread as freezable 2021-11-06 13:30:41 -07:00
page_alloc.c mm/page_alloc: remove the throttling logic from the page allocator 2021-11-06 13:30:40 -07:00
page_counter.c mm: page_counter: mitigate consequences of a page_counter underflow 2021-04-30 11:20:38 -07:00
page_ext.c mm/page_ext.c: fix a comment 2021-11-06 13:30:34 -07:00
page_idle.c mm/idle_page_tracking: make PG_idle reusable 2021-09-08 11:50:24 -07:00
page_io.c
page_isolation.c mm/page_isolation: guard against possible putback unisolated page 2021-11-06 13:30:40 -07:00
page_owner.c mm: remove pfn_valid_within() and CONFIG_HOLES_IN_ZONE 2021-09-08 11:50:22 -07:00
page_poison.c mm: page_poison: print page info when corruption is caught 2021-04-30 11:20:36 -07:00
page_reporting.c mm/page_reporting: allow driver to specify reporting order 2021-06-29 10:53:47 -07:00
page_reporting.h mm/page_reporting: export reporting order as module parameter 2021-06-29 10:53:47 -07:00
page_vma_mapped.c mm: device exclusive memory access 2021-07-01 11:06:03 -07:00
page-writeback.c mm/vmscan: centralise timeout values for reclaim_throttle 2021-11-06 13:30:40 -07:00
pagewalk.c mm: pagewalk: fix walk for hugepage tables 2021-06-29 10:53:49 -07:00
percpu-internal.h Merge branch 'for-5.14' of git://git.kernel.org/pub/scm/linux/kernel/git/dennis/percpu 2021-07-01 17:17:24 -07:00
percpu-km.c percpu: flush tlb in pcpu_reclaim_populated() 2021-07-04 18:30:17 +00:00
percpu-stats.c percpu: rework memcg accounting 2021-06-05 20:43:15 +00:00
percpu-vm.c percpu: flush tlb in pcpu_reclaim_populated() 2021-07-04 18:30:17 +00:00
percpu.c memblock: use memblock_free for freeing virtual pointers 2021-11-06 13:30:41 -07:00
pgalloc-track.h mm: fix typos in comments 2021-05-07 00:26:35 -07:00
pgtable-generic.c mm/thp: fix __split_huge_pmd_locked() on shmem migration entry 2021-06-16 09:24:42 -07:00
process_vm_access.c mm/process_vm_access.c: remove duplicate include 2021-05-05 11:27:27 -07:00
ptdump.c
readahead.c mm/readahead.c: fix incorrect comments for get_init_ra_size 2021-11-06 13:30:41 -07:00
rmap.c mm/rmap.c: avoid double faults migrating device private pages 2021-11-06 13:30:43 -07:00
rodata_test.c
secretmem.c mm/secretmem: avoid letting secretmem_users drop to zero 2021-10-28 17:18:55 -07:00
shmem.c mm: shmem: don't truncate page if memory failure happens 2021-11-06 13:30:38 -07:00
shuffle.c
shuffle.h mm/shuffle: fix section mismatch warning 2021-05-22 15:09:07 -10:00
slab_common.c mm: remove HARDENED_USERCOPY_FALLBACK 2021-11-06 13:30:43 -07:00
slab.c mm: remove HARDENED_USERCOPY_FALLBACK 2021-11-06 13:30:43 -07:00
slab.h mm/memcg: fix NULL pointer dereference in memcg_slab_free_hook() 2021-07-30 10:14:39 -07:00
slob.c
slub.c mm: remove HARDENED_USERCOPY_FALLBACK 2021-11-06 13:30:43 -07:00
sparse-vmemmap.c mm: remove redundant smp_wmb() 2021-11-06 13:30:36 -07:00
sparse.c memblock: use memblock_free for freeing virtual pointers 2021-11-06 13:30:41 -07:00
swap_cgroup.c
swap_slots.c mm: Replace deprecated CPU-hotplug functions. 2021-08-28 01:46:17 +02:00
swap_state.c Revert "mm: swap: check if swap backing device is congested or not" 2021-08-20 11:31:42 -07:00
swap.c mm: optimise put_pages_list() 2021-11-06 13:30:35 -07:00
swapfile.c mm/swapfile: fix an integer overflow in swap_show() 2021-11-06 13:30:35 -07:00
truncate.c vfs: keep inodes with page cache off the inode shrinker LRU 2021-11-09 10:02:48 -08:00
usercopy.c
userfaultfd.c mm: shmem: don't truncate page if memory failure happens 2021-11-06 13:30:38 -07:00
util.c mm: fix uninitialized use in overcommit_policy_handler 2021-09-24 16:13:35 -07:00
vmacache.c
vmalloc.c mm/vmalloc: introduce alloc_pages_bulk_array_mempolicy to accelerate memory allocation 2021-11-06 13:30:37 -07:00
vmpressure.c mm/vmpressure: fix data-race with memcg->socket_pressure 2021-11-06 13:30:40 -07:00
vmscan.c vfs: keep inodes with page cache off the inode shrinker LRU 2021-11-09 10:02:48 -08:00
vmstat.c mm: vmstat.c: make extfrag_index show more pretty 2021-11-06 13:30:42 -07:00
workingset.c vfs: keep inodes with page cache off the inode shrinker LRU 2021-11-09 10:02:48 -08:00
z3fold.c mm/z3fold: add kerneldoc fields for z3fold_pool 2021-07-01 11:06:03 -07:00
zbud.c mm/zbud: add kerneldoc fields for zbud_pool 2021-07-01 11:06:03 -07:00
zpool.c mm: fix typos in comments 2021-05-07 00:26:35 -07:00
zsmalloc.c mm/zsmalloc.c: close race window between zs_pool_dec_isolated() and zs_unregister_migration() 2021-11-06 13:30:43 -07:00
zswap.c mm/zswap.c: fix two bugs in zswap_writeback_entry() 2021-06-30 20:47:31 -07:00