write_bytes_to_xdr_buf() is pretty expensive to use for inserting
an XDR data item that is always 1 XDR_UNIT at an address that is
always XDR word-aligned.
Since both the readv and splice read paths encode EOF and maxcount
values, move both to a common code path.
Reviewed-by: Jeff Layton <jlayton@kernel.org> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
write_bytes_to_xdr_buf() is a generic way to place a variable-length
data item in an already-reserved spot in the encoding buffer.
However, it is costly. In nfsd4_encode_fattr(), it is unnecessary
because the data item is fixed in size and the buffer destination
address is always word-aligned.
Reviewed-by: Jeff Layton <jlayton@kernel.org> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
write_bytes_to_xdr_buf() is a generic way to place a variable-length
data item in an already-reserved spot in the encoding buffer.
However, it is costly, and here, it is unnecessary because the
data item is fixed in size, the buffer destination address is
always word-aligned, and the destination location is already in
@p.
Reviewed-by: Jeff Layton <jlayton@kernel.org> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
This printk pops every time nfsd.ko gets plugged in. Most kmods don't do
that and this one is not very informative. Olaf's email address seems to
be defunct at this point anyway. Just drop it.
Cc: Olaf Kirch <okir@suse.com> Signed-off-by: Jeff Layton <jlayton@kernel.org> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Currently there is no limit on how many v4 clients are supported
by the system. This can be a problem in systems with small memory
configuration to function properly when a very large number of
clients exist that creates memory shortage conditions.
This patch enforces a limit of 1024 NFSv4 clients, including courtesy
clients, per 1GB of system memory. When the number of the clients
reaches the limit, requests that create new clients are returned
with NFS4ERR_DELAY and the laundromat is kicked start to trim old
clients. Due to the overhead of the upcall to remove the client
record, the maximun number of clients the laundromat removes on
each run is limited to 128. This is done to ensure the laundromat
can still process the other tasks in a timely manner.
Since there is now a limit of the number of clients, the 24-hr
idle time limit of courtesy client is no longer needed and was
removed.
Signed-off-by: Dai Ngo <dai.ngo@oracle.com> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
The documenting comment for struct nf_file states:
/*
* A representation of a file that has been opened by knfsd. These are hashed
* in the hashtable by inode pointer value. Note that this object doesn't
* hold a reference to the inode by itself, so the nf_inode pointer should
* never be dereferenced, only used for comparison.
*/
Replace the two existing dereferences to make the comment always
true.
Reviewed-by: Jeff Layton <jlayton@kernel.org> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
The last close of a file should enable other accessors to open and
use that file immediately. Leaving the file open in the filecache
prevents other users from accessing that file until the filecache
garbage-collects the file -- sometimes that takes several seconds.
Reported-by: Wang Yugui <wangyugui@e16-tech.com> Link: https://bugzilla.linux-nfs.org/show_bug.cgi?387 Reviewed-by: Jeff Layton <jlayton@kernel.org> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Enable the filecache hash table to start small, then grow with the
workload. Smaller server deployments benefit because there should
be lower memory utilization. Larger server deployments should see
improved scaling with the number of open files.
Suggested-by: Jeff Layton <jlayton@kernel.org> Suggested-by: Dave Chinner <david@fromorbit.com> Reviewed-by: Jeff Layton <jlayton@kernel.org> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
In a moment, the nfsd_file_hashtbl global will be replaced with an
rhashtable. Replace the one or two spots that need to check if the
hash table is available. We can easily reuse the SHUTDOWN flag for
this purpose.
Document that this mechanism relies on callers to hold the
nfsd_mutex to prevent init, shutdown, and purging to run
concurrently.
Reviewed-by: Jeff Layton <jlayton@kernel.org> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
The code that computes the hashval is the same in both callers.
To prevent them from going stale, reframe the documenting comments
to remove descriptions of the underlying hash table structure, which
is about to be replaced.
Reviewed-by: Jeff Layton <jlayton@kernel.org> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
I'm about to replace nfsd_file_hashtbl with an rhashtable. The
individual hash values will no longer be visible or relevant, so
remove them from the tracepoints.
Reviewed-by: Jeff Layton <jlayton@kernel.org> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
The checks in nfsd_file_acquire() and nfsd_file_put() that directly
invoke filecache garbage collection are intended to keep cache
occupancy between a low- and high-watermark. The reason to limit the
capacity of the filecache is to keep filecache lookups reasonably
fast.
However, invoking garbage collection at those points has some
undesirable negative impacts. Files that are held open by NFSv4
clients often push the occupancy of the filecache over these
watermarks. At that point:
- Every call to nfsd_file_acquire() and nfsd_file_put() results in
an LRU walk. This has the same effect on lookup latency as long
chains in the hash table.
- Garbage collection will then run on every nfsd thread, causing a
lot of unnecessary lock contention.
- Limiting cache capacity pushes out files used only by NFSv3
clients, which are the type of files the filecache is supposed to
help.
To address those negative impacts, remove the direct calls to the
garbage collector. Subsequent patches will address maintaining
lookup efficiency as cache capacity increases.
Suggested-by: Wang Yugui <wangyugui@e16-tech.com> Suggested-by: Dave Chinner <david@fromorbit.com> Reviewed-by: Jeff Layton <jlayton@kernel.org> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Without LRU item rotation, the shrinker visits only a few items on
the end of the LRU list, and those would always be long-term OPEN
files for NFSv4 workloads. That makes the filecache shrinker
completely ineffective.
Adopt the same strategy as the inode LRU by using LRU_ROTATE.
Suggested-by: Dave Chinner <david@fromorbit.com> Reviewed-by: Jeff Layton <jlayton@kernel.org> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
There have been reports of problems when running fstests generic/531
against Linux NFS servers with NFSv4. The NFS server that hosts the
test's SCRATCH_DEV suffers from CPU soft lock-ups during the test.
Analysis shows that:
fs/nfsd/filecache.c
482 ret = list_lru_walk(&nfsd_file_lru,
483 nfsd_file_lru_cb,
484 &head, LONG_MAX);
causes nfsd_file_gc() to walk the entire length of the filecache LRU
list every time it is called (which is quite frequently). The walk
holds a spinlock the entire time that prevents other nfsd threads
from accessing the filecache.
What's more, for NFSv4 workloads, none of the items that are visited
during this walk may be evicted, since they are all files that are
held OPEN by NFS clients.
Address this by ensuring that open files are not kept on the LRU
list.
Reported-by: Frank van der Linden <fllinden@amazon.com> Reported-by: Wang Yugui <wangyugui@e16-tech.com> Link: https://bugzilla.linux-nfs.org/show_bug.cgi?id=386 Suggested-by: Trond Myklebust <trond.myklebust@hammerspace.com> Reviewed-by: Jeff Layton <jlayton@kernel.org> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
There has always been the capability of exporting filecache metrics
via /proc, but it was never hooked up. Let's surface these metrics
to enable better observability of the filecache.
Reviewed-by: Jeff Layton <jlayton@kernel.org> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Count the number of successful acquisitions that did not create a
file (ie, acquisitions that do not result in a compulsory cache
miss). This count can be compared directly with the reported hit
count to compute a hit ratio.
Reviewed-by: Jeff Layton <jlayton@kernel.org> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Capture file handles and how they map to local inodes. In particular,
NFSv4 PUTFH uses fh_verify() so we can now observe which file handles
are the target of OPEN, LOOKUP, RENAME, and so on.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Instead of trusting that struct file_lock returns completely unchanged
after vfs_test_lock() when there's no conflicting lock, stash away our
nlm_lockowner reference so we can properly release it for all cases.
This defends against another file_lock implementation overwriting fl_owner
when the return type is F_UNLCK.
Reported-by: Roberto Bergantinos Corpas <rbergant@redhat.com> Tested-by: Roberto Bergantinos Corpas <rbergant@redhat.com> Signed-off-by: Benjamin Coddington <bcodding@redhat.com> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
This flag is a new way to configure ignore mask which allows adding and
removing the event flags FAN_ONDIR and FAN_EVENT_ON_CHILD in ignore mask.
The legacy FAN_MARK_IGNORED_MASK flag would always ignore events on
directories and would ignore events on children depending on whether
the FAN_EVENT_ON_CHILD flag was set in the (non ignored) mask.
FAN_MARK_IGNORE can be used to ignore events on children without setting
FAN_EVENT_ON_CHILD in the mark's mask and will not ignore events on
directories unconditionally, only when FAN_ONDIR is set in ignore mask.
The new behavior is non-downgradable. After calling fanotify_mark() with
FAN_MARK_IGNORE once, calling fanotify_mark() with FAN_MARK_IGNORED_MASK
on the same object will return EEXIST error.
Setting the event flags with FAN_MARK_IGNORE on a non-dir inode mark
has no meaning and will return ENOTDIR error.
The meaning of FAN_MARK_IGNORED_SURV_MODIFY is preserved with the new
FAN_MARK_IGNORE flag, but with a few semantic differences:
1. FAN_MARK_IGNORED_SURV_MODIFY is required for filesystem and mount
marks and on an inode mark on a directory. Omitting this flag
will return EINVAL or EISDIR error.
2. An ignore mask on a non-directory inode that survives modify could
never be downgraded to an ignore mask that does not survive modify.
With new FAN_MARK_IGNORE semantics we make that rule explicit -
trying to update a surviving ignore mask without the flag
FAN_MARK_IGNORED_SURV_MODIFY will return EEXIST error.
The conveniene macro FAN_MARK_IGNORE_SURV is added for
(FAN_MARK_IGNORE | FAN_MARK_IGNORED_SURV_MODIFY), because the
common case should use short constant names.
Setting flags FAN_ONDIR FAN_EVENT_ON_CHILD in ignore mask has no effect.
The FAN_EVENT_ON_CHILD flag in mask implicitly applies to ignore mask and
ignore mask is always implicitly applied to events on directories.
Define a mark flag that replaces this legacy behavior with logic of
applying the ignore mask according to event flags in ignore mask.
Implement the new logic to prepare for supporting an ignore mask that
ignores events on children and ignore mask that does not ignore events
on directories.
To emphasize the change in terminology, also rename ignored_mask mark
member to ignore_mask and use accessors to get only the effective
ignored events or the ignored events and flags.
This change in terminology finally aligns with the "ignore mask"
language in man pages and in most of the comments.
NFSD has advertised support for the NFSv4 time_create attribute
since commit e377a3e698fb ("nfsd: Add support for the birth time
attribute").
Igor Mammedov reports that Mac OS clients attempt to set the NFSv4
birth time attribute via OPEN(CREATE) and SETATTR if the server
indicates that it supports it, but since the above commit was
merged, those attempts now fail.
Table 5 in RFC 8881 lists the time_create attribute as one that can
be both set and retrieved, but the above commit did not add server
support for clients to provide a time_create attribute. IMO that's
a bug in our implementation of the NFSv4 protocol, which this commit
addresses.
Whether NFSD silently ignores the new birth time or actually sets it
is another matter. I haven't found another filesystem service in the
Linux kernel that enables users or clients to modify a file's birth
time attribute.
This commit reflects my (perhaps incorrect) understanding of whether
Linux users can set a file's birth time. NFSD will now recognize a
time_create attribute but it ignores its value. It clears the
time_create bit in the returned attribute bitmask to indicate that
the value was not used.
Reported-by: Igor Mammedov <imammedo@redhat.com> Fixes: e377a3e698fb ("nfsd: Add support for the birth time attribute") Tested-by: Igor Mammedov <imammedo@redhat.com> Reviewed-by: Jeff Layton <jlayton@kernel.org> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Commit ceaf69f8eadc ("fanotify: do not allow setting dirent events in
mask of non-dir") added restrictions about setting dirent events in the
mask of a non-dir inode mark, which does not make any sense.
For backward compatibility, these restictions were added only to new
(v5.17+) APIs.
It also does not make any sense to set the flags FAN_EVENT_ON_CHILD or
FAN_ONDIR in the mask of a non-dir inode. Add these flags to the
dir-only restriction of the new APIs as well.
Move the check of the dir-only flags for new APIs into the helper
fanotify_events_supported(), which is only called for FAN_MARK_ADD,
because there is no need to error on an attempt to remove the dir-only
flags from non-dir inode.
Commit f49169c97fce ("NFSD: Remove svc_serv_ops::svo_module") removed
calls to module_put_and_kthread_exit() from threads that acted as SUNRPC
servers and had a related svc_serv_ops structure. This was correct.
It ALSO removed the module_put_and_kthread_exit() call from
nfs4_run_state_manager() which is NOT a SUNRPC service.
Consequently every time the NFSv4 state manager runs the module count
increments and won't be decremented. So the nfsv4 module cannot be
unloaded.
So restore the module_put_and_kthread_exit() call.
Fixes: f49169c97fce ("NFSD: Remove svc_serv_ops::svo_module") Signed-off-by: NeilBrown <neilb@suse.de> Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Now that there are no more callers of nfsd_file_put() that might
hold a spin lock, ensure the lockdep infrastructure can catch
newly introduced calls to nfsd_file_put() made while a spinlock
is held.
Refactor: Use existing helpers that other lock operations use. This
change removes several automatic variables, so re-organize the
variable declarations for readability.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
This can be reproduce by concurrent operations:
1. fsopen(nfsd)/fsconfig
2. insmod/rmmod nfsd
Since the nfsd file system is registered before than nfsd_net allocated,
the caller may get the file_system_type and use the nfsd_net before it
allocated, then null-ptr-deref occurred.
So init_nfsd() should call register_filesystem() last.
Fixes: bd5ae9288d64 ("nfsd: register pernet ops last, unregister first") Signed-off-by: Zhang Xiaoxu <zhangxiaoxu5@huawei.com> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Once an svc thread is scheduled and executing an RPC, no other
processes will touch svc_rqst::rq_flags. Thus bus-locked atomics are
not needed outside the svc thread scheduler.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
There have been reports of races that cause NFSv4 OPEN(CREATE) to
return an error even though the requested file was created. NFSv4
does not provide a status code for this case.
To mitigate some of these problems, reorganize the NFSv4
OPEN(CREATE) logic to allocate resources before the file is actually
created, and open the new file while the parent directory is still
locked.
Two new APIs are added:
+ Add an API that works like nfsd_file_acquire() but does not open
the underlying file. The OPEN(CREATE) path can use this API when it
already has an open file.
+ Add an API that is kin to dentry_open(). NFSD needs to create a
file and grab an open "struct file *" atomically. The
alloc_empty_file() has to be done before the inode create. If it
fails (for example, because the NFS server has exceeded its
max_files limit), we avoid creating the file and can still return
an error to the NFS client.
BugLink: https://bugzilla.linux-nfs.org/show_bug.cgi?id=382 Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Tested-by: JianHong Yin <jiyin@redhat.com> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
The NFSv3 CREATE and NFSv4 OPEN(CREATE) use cases are about to
diverge such that it makes sense to split do_nfsd_create() into one
version for NFSv3 and one for NFSv4.
As a first step, copy do_nfsd_create() to nfs3proc.c and remove
NFSv4-specific logic.
One immediate legibility benefit is that the logic for handling
NFSv3 createhow is now quite straightforward. NFSv4 createhow
has some subtleties that IMO do not belong in generic code.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
I'd like to move do_nfsd_create() out of vfs.c. Therefore
nfsd_create_setattr() needs to be made publicly visible.
Note that both call sites in vfs.c commit both the new object and
its parent directory, so just combine those common metadata commits
into nfsd_create_setattr().
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Clean up: The "out" label already invokes fh_drop_write().
Note that fh_drop_write() is already careful not to invoke
mnt_drop_write() if either it has already been done or there is
nothing to drop. Therefore no change in behavior is expected.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
As near as I can tell, mode bit masking and setting S_IFREG is
already done by do_nfsd_create() and vfs_create(). The NFSv4 path
(do_open_lookup), for example, does not bother with this special
processing.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
This patch allows expired client with lock state to be in COURTESY
state. Lock conflict with COURTESY client is resolved by the fs/lock
code using the lm_lock_expirable and lm_expire_lock callback in the
struct lock_manager_operations.
If conflict client is in COURTESY state, set it to EXPIRABLE and
schedule the laundromat to run immediately to expire the client. The
callback lm_expire_lock waits for the laundromat to flush its work
queue before returning to caller.
Reviewed-by: J. Bruce Fields <bfields@fieldses.org> Signed-off-by: Dai Ngo <dai.ngo@oracle.com> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Add 2 new callbacks, lm_lock_expirable and lm_expire_lock, to
lock_manager_operations to allow the lock manager to take appropriate
action to resolve the lock conflict if possible.
A new field, lm_mod_owner, is also added to lock_manager_operations.
The lm_mod_owner is used by the fs/lock code to make sure the lock
manager module such as nfsd, is not freed while lock conflict is being
resolved.
lm_lock_expirable checks and returns true to indicate that the lock
conflict can be resolved else return false. This callback must be
called with the flc_lock held so it can not block.
lm_expire_lock is called to resolve the lock conflict if the returned
value from lm_lock_expirable is true. This callback is called without
the flc_lock held since it's allowed to block. Upon returning from
this callback, the lock conflict should be resolved and the caller is
expected to restart the conflict check from the beginnning of the list.
Lock manager, such as NFSv4 courteous server, uses this callback to
resolve conflict by destroying lock owner, or the NFSv4 courtesy client
(client that has expired but allowed to maintains its states) that owns
the lock.
Reviewed-by: J. Bruce Fields <bfields@fieldses.org> Signed-off-by: Dai Ngo <dai.ngo@oracle.com> Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Reviewed-by: Jeff Layton <jlayton@kernel.org> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Add helper locks_owner_has_blockers to check if there is any blockers
for a given lockowner.
Reviewed-by: J. Bruce Fields <bfields@fieldses.org> Signed-off-by: Dai Ngo <dai.ngo@oracle.com> Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Reviewed-by: Jeff Layton <jlayton@kernel.org> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
This patch moves create/destroy of laundry_wq from nfs4_state_start
and nfs4_state_shutdown_net to init_nfsd and exit_nfsd to prevent
the laundromat from being freed while a thread is processing a
conflicting lock.
Reviewed-by: J. Bruce Fields <bfields@fieldses.org> Signed-off-by: Dai Ngo <dai.ngo@oracle.com> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
This patch allows expired client with open state to be in COURTESY
state. Share/access conflict with COURTESY client is resolved by
setting COURTESY client to EXPIRABLE state, schedule laundromat
to run and returning nfserr_jukebox to the request client.
Reviewed-by: J. Bruce Fields <bfields@fieldses.org> Signed-off-by: Dai Ngo <dai.ngo@oracle.com> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
This patch provides courteous server support for delegation only.
Only expired client with delegation but no conflict and no open
or lock state is allowed to be in COURTESY state.
Delegation conflict with COURTESY/EXPIRABLE client is resolved by
setting it to EXPIRABLE, queue work for the laundromat and return
delay to the caller. Conflict is resolved when the laudromat runs
and expires the EXIRABLE client while the NFS client retries the
OPEN request. Local thread request that gets conflict is doing the
retry in _break_lease.
Client in COURTESY or EXPIRABLE state is allowed to reconnect and
continues to have access to its state. Access to the nfs4_client by
the reconnecting thread and the laundromat is serialized via the
client_lock.
Reviewed-by: J. Bruce Fields <bfields@fieldses.org> Signed-off-by: Dai Ngo <dai.ngo@oracle.com> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
nfsd_splice_actor() checks that the page being spliced does not
match the previous element in the svc_rqst::rq_pages array. We
believe this is to prevent a double put_page() in cases where the
READ payload is partially contained in the xdr_buf's head buffer.
However, the NFSD READ proc functions no longer place any part of
the READ payload in the head buffer, in order to properly support
NFS/RDMA READ with Write chunks. Therefore, simplify the logic in
nfsd_splice_actor() to remove this unnecessary check.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Fixes sparce warnings:
fs/notify/fanotify/fanotify_user.c:267:63: sparse:
warning: restricted fmode_t degrades to integer
fs/notify/fanotify/fanotify_user.c:1351:28: sparse:
warning: restricted fmode_t degrades to integer
FMODE_NONTIFY have bitwise fmode_t type and requires __force attribute
for any casts.
Signed-off-by: Vasily Averin <vvs@openvz.org> Reviewed-by: Christian Brauner (Microsoft) <brauner@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/9adfd6ac-1b89-791e-796b-49ada3293985@openvz.org Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
The logic for handling events on child in groups that have a mark on
the parent inode, but without FS_EVENT_ON_CHILD flag in the mask is
duplicated in several places and inconsistent.
Move the logic into the preparation of mark type iterator, so that the
parent mark type will be excluded from all mark type iterations in that
case.
This results in several subtle changes of behavior, hopefully all
desired changes of behavior, for example:
- Group A has a mount mark with FS_MODIFY in mask
- Group A has a mark with ignore mask that does not survive FS_MODIFY
and does not watch children on directory D.
- Group B has a mark with FS_MODIFY in mask that does watch children
on directory D.
- FS_MODIFY event on file D/foo should not clear the ignore mask of
group A, but before this change it does
And if group A ignore mask was set to survive FS_MODIFY:
- FS_MODIFY event on file D/foo should be reported to group A on account
of the mount mark, but before this change it is wrongly ignored
fsnotify_foreach_iter_mark_type() is used to reduce boilerplate code
of iterating all marks of a specific group interested in an event
by consulting the iterator report_mask.
Use an open coded version of that iterator in fsnotify_iter_next()
that collects all marks of the current iteration group without
consulting the iterator report_mask.
At the moment, the two iterator variants are the same, but this
decoupling will allow us to exclude some of the group's marks from
reporting the event, for example for event on child and inode marks
on parent did not request to watch events on children.
Fixes: 2f02fd3fa13e ("fanotify: fix ignore mask logic for events on child and on dir") Reported-by: Jan Kara <jack@suse.com> Signed-off-by: Amir Goldstein <amir73il@gmail.com> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20220511190213.831646-2-amir73il@gmail.com Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
When an inode mark is created with flag FAN_MARK_EVICTABLE, it will not
pin the marked inode to inode cache, so when inode is evicted from cache
due to memory pressure, the mark will be lost.
When an inode mark with flag FAN_MARK_EVICATBLE is updated without using
this flag, the marked inode is pinned to inode cache.
When an inode mark is updated with flag FAN_MARK_EVICTABLE but an
existing mark already has the inode pinned, the mark update fails with
error EEXIST.
Evictable inode marks can be used to setup inode marks with ignored mask
to suppress events from uninteresting files or directories in a lazy
manner, upon receiving the first event, without having to iterate all
the uninteresting files or directories before hand.
The evictbale inode mark feature allows performing this lazy marks setup
without exhausting the system memory with pinned inodes.
Handle FAN_MARK_IGNORED_SURV_MODIFY flag change in a helper that
is called after updating the mark mask.
Replace the added and removed return values and help variables with
bool recalc return values and help variable, which makes the code a
bit easier to follow.
Rename flags argument to fan_flags to emphasize the difference from
mark->flags.
fsnotify_add_mark() and variants implicitly take a reference on inode
when attaching a mark to an inode.
Make that behavior opt-out with the mark flag FSNOTIFY_MARK_FLAG_NO_IREF.
Instead of taking the inode reference when attaching connector to inode
and dropping the inode reference when detaching connector from inode,
take the inode reference on attach of the first mark that wants to hold
an inode reference and drop the inode reference on detach of the last
mark that wants to hold an inode reference.
Backends can "upgrade" an existing mark to take an inode reference, but
cannot "downgrade" a mark with inode reference to release the refernce.
This leaves the choice to the backend whether or not to pin the inode
when adding an inode mark.
This is intended to be used when adding a mark with ignored mask that is
used for optimization in cases where group can afford getting unneeded
events and reinstate the mark with ignored mask when inode is accessed
again after being evicted.
Before commit 9542e6a643fc6 ("nfsd: Containerise filecache laundrette")
nfsd would close open files in direct reclaim context. There is no
guarantee that others memory shrinkers don't do the same and no
guarantee that future shrinkers won't do that.
For example, if overlayfs implements inode cache of fscache would
keep open files to cached objects, inode shrinkers could end up closing
open files to underlying fs.
Direct reclaim from dnotify mark allocation context may try to close
open files that have dnotify marks of the same group and hit a deadlock
on mark_mutex.
Set the FSNOTIFY_GROUP_NOFS flag to prevent going into direct reclaim
from allocations under dnotify group lock and use the safe group lock
helpers.
Before commit 9542e6a643fc6 ("nfsd: Containerise filecache laundrette")
nfsd would close open files in direct reclaim context and that could
cause a deadlock when fsnotify mark allocation went into direct reclaim
and nfsd shrinker tried to free existing fsnotify marks.
To avoid issues like this in future code, set the FSNOTIFY_GROUP_NOFS
flag on nfsd fsnotify group to prevent going into direct reclaim from
fsnotify_add_inode_mark().
Create helpers to take and release the group mark_mutex lock.
Define a flag FSNOTIFY_GROUP_NOFS in fsnotify_group that determines
if the mark_mutex lock is fs reclaim safe or not. If not safe, the
lock helpers take the lock and disable direct fs reclaim.
In that case we annotate the mutex with a different lockdep class to
express to lockdep that an allocation of mark of an fs reclaim safe group
may take the group lock of another "NOFS" group to evict inodes.
For now, converted only the callers in common code and no backend
defines the NOFS flag. It is intended to be set by fanotify for
evictable marks support.
Instead of passing the allow_dups argument to fsnotify_add_mark()
as an argument, define the group flag FSNOTIFY_GROUP_DUPS to express
the allow_dups behavior and set this behavior at group creation time
for all calls of fsnotify_add_mark().
Rename the allow_dups argument to generic add_flags argument for future
use.
Add flags argument to fsnotify_alloc_group(), define and use the flag
FSNOTIFY_GROUP_USER in inotify and fanotify instead of the helper
fsnotify_alloc_user_group() to indicate user allocation.
Although the flag FSNOTIFY_GROUP_USER is currently not used after group
allocation, we store the flags argument in the group struct for future
use of other group flags.
Update lock usage of lock_manager_operations' functions to reflect
the changes in commit 6109c85037e5 ("locks: add a dedicated spinlock
to protect i_flctx lists").
Signed-off-by: Dai Ngo <dai.ngo@oracle.com> Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Reviewed-by: Jeff Layton <jlayton@kernel.org> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Dirent events (create/delete/move) are only reported on watched
directory inodes, but in fanotify as well as in legacy inotify, it was
always allowed to set them on non-dir inode, which does not result in
any meaningful outcome.
Until kernel v5.17, dirent events in fanotify also differed from events
"on child" (e.g. FAN_OPEN) in the information provided in the event.
For example, FAN_OPEN could be set in the mask of a non-dir or the mask
of its parent and event would report the fid of the child regardless of
the marked object.
By contrast, FAN_DELETE is not reported if the child is marked and the
child fid was not reported in the events.
Since kernel v5.17, with fanotify group flag FAN_REPORT_TARGET_FID, the
fid of the child is reported with dirent events, like events "on child",
which may create confusion for users expecting the same behavior as
events "on child" when setting events in the mask on a child.
The desired semantics of setting dirent events in the mask of a child
are not clear, so for now, deny this action for a group initialized
with flag FAN_REPORT_TARGET_FID and for the new event FAN_RENAME.
We may relax this restriction in the future if we decide on the
semantics and implement them.
The call to filemap_flush() in nfsd_file_put() is there to ensure that
we clear out any writes belonging to a NFSv3 client relatively quickly
and avoid situations where the file can't be evicted by the garbage
collector. It also ensures that we detect write errors quickly.
The problem is this causes a regression in performance for some
workloads.
So try to improve matters by deferring writeback until we're ready to
close the file, and need to detect errors so that we can force the
client to resend.
Tested-by: Jan Kara <jack@suse.cz> Fixes: b6669305d35a ("nfsd: Reduce the number of calls to nfsd_file_gc()") Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com> Link: https://lore.kernel.org/all/20220330103457.r4xrhy2d6nhtouzk@quack3.lan Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
projects / users / dwmw2 / linux.git / log