I received a report about suspicious growth of unreclaimable slabs on
some machines. I've found that it happens on machines with low memory
pressure, and these unreclaimable slabs are external names attached to
dentries.
External names are allocated using generic kmalloc() function, so they
are accounted as unreclaimable. But they are held by dentries, which
are reclaimable, and they will be reclaimed under the memory pressure.
In particular, this breaks MemAvailable calculation, as it doesn't take
unreclaimable slabs into account. This leads to a silly situation, when
a machine is almost idle, has no memory pressure and therefore has a big
dentry cache. And the resulting MemAvailable is too low to start a new
workload.
To address the issue, the NR_INDIRECTLY_RECLAIMABLE_BYTES counter is
used to track the amount of memory, consumed by external names. The
counter is increased in the dentry allocation path, if an external name
structure is allocated; and it's decreased in the dentry freeing path.
To reproduce the problem I've used the following Python script:
import os
for iter in range (0, 10000000):
try:
name = ("/some_long_name_%d" % iter) + "_" * 220
os.stat(name)
except Exception:
pass
Without this patch:
$ cat /proc/meminfo | grep MemAvailable
MemAvailable: 7811688 kB
$ python indirect.py
$ cat /proc/meminfo | grep MemAvailable
MemAvailable: 2753052 kB
With the patch:
$ cat /proc/meminfo | grep MemAvailable
MemAvailable: 7809516 kB
$ python indirect.py
$ cat /proc/meminfo | grep MemAvailable
MemAvailable: 7749144 kB
[guro@fb.com: fix indirectly reclaimable memory accounting for CONFIG_SLOB]
Link: http://lkml.kernel.org/r/20180312194140.19517-1-guro@fb.com
[guro@fb.com: fix indirectly reclaimable memory accounting]
Link: http://lkml.kernel.org/r/20180313125701.7955-1-guro@fb.com
Link: http://lkml.kernel.org/r/20180305133743.12746-5-guro@fb.com
Signed-off-by: Roman Gushchin <guro@fb.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
As part of the effort to remove VLAs from the kernel[1], this changes
the allocation of the bhs and pages arrays from being on the stack to being
kcalloc()ed. This also allows for the removal of the explicit zeroing
of bhs.
https://lkml.org/lkml/2018/3/7/621
Signed-off-by: Kyle Spiers <ksspiers@google.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Forcing the log to disk after reading the agf is wrong, we might be
calling xfs_log_force with XFS_LOG_SYNC with a metadata lock held.
This can cause a deadlock when racing a fstrim with a filesystem
shutdown.
The deadlock has been identified due a miscalculation bug in device-mapper
dm-thin, which returns lack of space to its users earlier than the device itself
really runs out of space, changing the device-mapper volume into an error state.
The problem happened while filling the filesystem with a single file,
triggering the bug in device-mapper, consequently causing an IO error
and shutting down the filesystem.
If such file is removed, and fstrim executed before the XFS finishes the
shut down process, the fstrim process will end up holding the buffer
lock, and going to sleep on the cil wait queue.
At this point, the shut down process will try to wake up all the threads
waiting on the cil wait queue, but for this, it will try to hold the
same buffer log already held my the fstrim, locking up the filesystem.
Signed-off-by: Carlos Maiolino <cmaiolino@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
In nfs[34]_decode_dirent, the cookie is advanced as soon as it is
read, but decoding may still fail later in the function, returning
an error. Because the cookie has been advanced, the failing entry
is not re-requested from the server, resulting in a missing directory
entry.
In addition, nfs v3 and v4 read the cookie at different locations
in the xdr_stream, so the behavior of the two can be inconsistent.
Fix these by reading the cookie into a temporary variable, and
only advancing the cookie once the entire entry has been decoded
from the xdr_stream successfully.
Signed-off-by: Frank Sorenson <sorenson@redhat.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
When we use EXCLUSIVE4_1 mode, the server returns an attribute mask where
all the bits indicate which attributes were set, and where the verifier
was stored. In order to figure out which attribute we have to resend,
we need to clear out the attributes that are set in exclcreat_bitmask.
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
[Anna: Fixed typo NFS4_CREATE_EXCLUSIVE4 -> NFS4_CREATE_EXCLUSIVE]
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
When we've changed the file size, then ensure we declare it to be
up to date in the inode attributes.
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Allocate the owner_id when we allocate the state and free it when we free
the state. That lets us get rid of a gnarly ida_pre_get() / ida_get_new()
loop.
Signed-off-by: Matthew Wilcox <mawilcox@microsoft.com>
Reviewed-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Neither nfs_inode_set_delegation() nor nfs_inode_reclaim_delegation() are
generic code. They have no business delving into NFSv4 OPEN xdr structures,
so let's replace the "struct nfs_openres" parameter.
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Don't bother even recording an invalid change attribute if we hold a
delegation since we already know the state of our attribute cache.
We can rely on the fact that we will pick up a copy from the server
when we return the delegation.
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Currently, if the NFS_INO_INVALID_ATTR flag is set, for instance by
a call to nfs_post_op_update_inode_locked(), then it will not be cleared
until all the attributes have been revalidated. This means, for instance,
that NFSv4 writes will always force a full attribute revalidation.
Track the ctime, mtime, size and change attribute separately from the
other attributes so that we can have nfs_post_op_update_inode_locked()
set them correctly, and later have the cache consistency bitmask be
able to clear them.
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
If we managed to revalidate all the attributes, then there is no reason
to mark them as invalid again. We do, however want to ensure that we
set nfsi->attrtimeo correctly.
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
If we received weak cache consistency data from the server, then those
attributes are up to date, and there is no reason to mark them as
dirty in the attribute cache.
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Even if the change attribute is missing, it is still OK to mark the other
attributes as being up to date.
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Starting with NFSv4.1, the server is able to deduce the client id from
the SEQUENCE op which means it can always figure out whether or not
the client is holding a delegation on a file that is being changed.
For that reason, RFC5661 does not require a delegation to be unconditionally
recalled on operations such as SETATTR, RENAME, or REMOVE.
Note that for now, we continue to return READ delegations since that is
still expected by the Linux knfsd server.
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
The 'fmode' argument can take an FMODE_EXEC value, which we want to
filter out before comparing to the delegation type.
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
If we have a RECLAIM_COMPLETE with a populated cl_lock_waitq, then
that implies that a reconnect has occurred. Since we can't expect a
CB_NOTIFY_LOCK callback at that point, just wake up the entire queue
so that all the tasks can re-poll for their locks.
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
The task is expected to sleep for a while here, and it's possible that
a new EXCHANGE_ID has occurred in the interim, and we were assigned a
new clientid. Since this is a per-client list, there isn't a lot of
value in vetting the clientid on the incoming request.
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
We may get a notification and lose the race to another client. Ensure
that we wait again for a notification in that case.
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Pull ceph updates from Ilya Dryomov:
"The big ticket items are:
- support for rbd "fancy" striping (myself).
The striping feature bit is now fully implemented, allowing mapping
v2 images with non-default striping patterns. This completes
support for --image-format 2.
- CephFS quota support (Luis Henriques and Zheng Yan).
This set is based on the new SnapRealm code in the upcoming v13.y.z
("Mimic") release. Quota handling will be rejected on older
filesystems.
- memory usage improvements in CephFS (Chengguang Xu).
Directory specific bits have been split out of ceph_file_info and
some effort went into improving cap reservation code to avoid OOM
crashes.
Also included a bunch of assorted fixes all over the place from
Chengguang and others"
* tag 'ceph-for-4.17-rc1' of git://github.com/ceph/ceph-client: (67 commits)
ceph: quota: report root dir quota usage in statfs
ceph: quota: add counter for snaprealms with quota
ceph: quota: cache inode pointer in ceph_snap_realm
ceph: fix root quota realm check
ceph: don't check quota for snap inode
ceph: quota: update MDS when max_bytes is approaching
ceph: quota: support for ceph.quota.max_bytes
ceph: quota: don't allow cross-quota renames
ceph: quota: support for ceph.quota.max_files
ceph: quota: add initial infrastructure to support cephfs quotas
rbd: remove VLA usage
rbd: fix spelling mistake: "reregisteration" -> "reregistration"
ceph: rename function drop_leases() to a more descriptive name
ceph: fix invalid point dereference for error case in mdsc destroy
ceph: return proper bool type to caller instead of pointer
ceph: optimize memory usage
ceph: optimize mds session register
libceph, ceph: add __init attribution to init funcitons
ceph: filter out used flags when printing unused open flags
ceph: don't wait on writeback when there is no more dirty pages
...
Pull libnvdimm updates from Dan Williams:
"This cycle was was not something I ever want to repeat as there were
several late changes that have only now just settled.
Half of the branch up to commit d2c997c0f1 ("fs, dax: use
page->mapping to warn...") have been in -next for several releases.
The of_pmem driver and the address range scrub rework were late
arrivals, and the dax work was scaled back at the last moment.
The of_pmem driver missed a previous merge window due to an oversight.
A sense of obligation to rectify that miss is why it is included for
4.17. It has acks from PowerPC folks. Stephen reported a build failure
that only occurs when merging it with your latest tree, for now I have
fixed that up by disabling modular builds of of_pmem. A test merge
with your tree has received a build success report from the 0day robot
over 156 configs.
An initial version of the ARS rework was submitted before the merge
window. It is self contained to libnvdimm, a net code reduction, and
passing all unit tests.
The filesystem-dax changes are based on the wait_var_event()
functionality from tip/sched/core. However, late review feedback
showed that those changes regressed truncate performance to a large
degree. The branch was rewound to drop the truncate behavior change
and now only includes preparation patches and cleanups (with full acks
and reviews). The finalization of this dax-dma-vs-trnucate work will
need to wait for 4.18.
Summary:
- A rework of the filesytem-dax implementation provides for detection
of unmap operations (truncate / hole punch) colliding with
in-progress device-DMA. A fix for these collisions remains a
work-in-progress pending resolution of truncate latency and
starvation regressions.
- The of_pmem driver expands the users of libnvdimm outside of x86
and ACPI to describe an implementation of persistent memory on
PowerPC with Open Firmware / Device tree.
- Address Range Scrub (ARS) handling is completely rewritten to
account for the fact that ARS may run for 100s of seconds and there
is no platform defined way to cancel it. ARS will now no longer
block namespace initialization.
- The NVDIMM Namespace Label implementation is updated to handle
label areas as small as 1K, down from 128K.
- Miscellaneous cleanups and updates to unit test infrastructure"
* tag 'libnvdimm-for-4.17' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdimm/nvdimm: (39 commits)
libnvdimm, of_pmem: workaround OF_NUMA=n build error
nfit, address-range-scrub: add module option to skip initial ars
nfit, address-range-scrub: rework and simplify ARS state machine
nfit, address-range-scrub: determine one platform max_ars value
powerpc/powernv: Create platform devs for nvdimm buses
doc/devicetree: Persistent memory region bindings
libnvdimm: Add device-tree based driver
libnvdimm: Add of_node to region and bus descriptors
libnvdimm, region: quiet region probe
libnvdimm, namespace: use a safe lookup for dimm device name
libnvdimm, dimm: fix dpa reservation vs uninitialized label area
libnvdimm, testing: update the default smart ctrl_temperature
libnvdimm, testing: Add emulation for smart injection commands
nfit, address-range-scrub: introduce nfit_spa->ars_state
libnvdimm: add an api to cast a 'struct nd_region' to its 'struct device'
nfit, address-range-scrub: fix scrub in-progress reporting
dax, dm: allow device-mapper to operate without dax support
dax: introduce CONFIG_DAX_DRIVER
fs, dax: use page->mapping to warn if truncate collides with a busy page
ext2, dax: introduce ext2_dax_aops
...
In xfs_itruncate_extents, only cancel cow blocks and clear the reflink
flag if we were asked to truncate the data fork. Attr fork blocks
cannot be shared, so this makes no sense.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Processes like ld that do lots of small writes that aren't necessarily
contiguous result in a lot of small StoreData operations to the server, the
idea being that if someone else changes the data on the server, we only
write our changes over that and not the space between. Further, we don't
want to write back empty space if we can avoid it to make it easier for the
server to do sparse files.
However, making lots of tiny RPC ops is a lot less efficient for the server
than one big one because each op requires allocation of resources and the
taking of locks, so we want to compromise a bit.
Reduce the load by the following:
(1) If a file is just created locally or has just been truncated with
O_TRUNC locally, allow subsequent writes to the file to be merged with
intervening space if that space doesn't cross an entire intervening
page.
(2) Don't flush the file on ->flush() but rather on ->release() if the
file was open for writing.
Just linking vmlinux.o, without this patch, looking in /proc/fs/afs/stats:
file-wr : n=441 nb=513581204
and after the patch:
file-wr : n=62 nb=513668555
there were 379 fewer StoreData RPC operations at the expense of an extra
87K being written.
Signed-off-by: David Howells <dhowells@redhat.com>
Add statistics to /proc/fs/afs/stats for data transfer RPC operations. New
lines are added that look like:
file-rd : n=55794 nb=10252282150
file-wr : n=9789 nb=3247763645
where n= indicates the number of ops completed and nb= indicates the number
of bytes successfully transferred. file-rd is the counts for read/fetch
operations and file-wr the counts for write/store operations.
Note that directory and symlink downloading are included in the file-rd
stats at the moment.
Signed-off-by: David Howells <dhowells@redhat.com>
Locally edit the contents of an AFS directory upon a successful inode
operation that modifies that directory (such as mkdir, create and unlink)
so that we can avoid the current practice of re-downloading the directory
after each change.
This is viable provided that the directory version number we get back from
the modifying RPC op is exactly incremented by 1 from what we had
previously. The data in the directory contents is in a defined format that
we have to parse locally to perform lookups and readdir, so modifying isn't
a problem.
If the edit fails, we just clear the VALID flag on the directory and it
will be reloaded next time it is needed.
Signed-off-by: David Howells <dhowells@redhat.com>
Adjust the AFS directory XDR structures in a number of superficial ways:
(1) Rename them to all begin afs_xdr_.
(2) Use u8 instead of uint8_t.
(3) Mark the structures as __packed so they don't get rearranged by the
compiler.
(4) Rename the hdr member of afs_xdr_dir_block to meta.
(5) Rename the pagehdr member of afs_xdr_dir_block to hdr.
Signed-off-by: David Howells <dhowells@redhat.com>
Split the directory content definitions into a header file so that they can
be used by multiple .c files.
Signed-off-by: David Howells <dhowells@redhat.com>
AFS directories are structured blobs that are downloaded just like files
and then parsed by the lookup and readdir code and, as such, are currently
handled in the pagecache like any other file, with the entire directory
content being thrown away each time the directory changes.
However, since the blob is a known structure and since the data version
counter on a directory increases by exactly one for each change committed
to that directory, we can actually edit the directory locally rather than
fetching it from the server after each locally-induced change.
What we can't do, though, is mix data from the server and data from the
client since the server is technically at liberty to rearrange or compress
a directory if it sees fit, provided it updates the data version number
when it does so and breaks the callback (ie. sends a notification).
Further, lookup with lookup-ahead, readdir and, when it arrives, local
editing are likely want to scan the whole of a directory.
So directory handling needs to be improved to maintain the coherency of the
directory blob prior to permitting local directory editing.
To this end:
(1) If any directory page gets discarded, invalidate and reread the entire
directory.
(2) If readpage notes that if when it fetches a single page that the
version number has changed, the entire directory is flagged for
invalidation.
(3) Read as much of the directory in one go as we can.
Note that this removes local caching of directories in fscache for the
moment as we can't pass the pages to fscache_read_or_alloc_pages() since
page->lru is in use by the LRU.
Signed-off-by: David Howells <dhowells@redhat.com>
