| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
See protocol documentation in MS-SMB2 section 2.2.42.2.2
Signed-off-by: Enzo Matsumiya <[email protected]>
Signed-off-by: Steve French <[email protected]>
|
|
Change "compress=" mount option to a boolean flag, that, if set,
will enable negotiating compression algorithms with the server.
Do not de/compress anything for now.
Signed-off-by: Enzo Matsumiya <[email protected]>
Signed-off-by: Steve French <[email protected]>
|
|
It can be helpful in debugging to know which ioctls are called to better
correlate them with smb3 fsctls (and opens). Add a dynamic trace point
to trace ioctls into cifs.ko
Here is sample output:
TASK-PID CPU# ||||| TIMESTAMP FUNCTION
| | | ||||| | |
new-inotify-ioc-90418 [001] ..... 142157.397024: smb3_ioctl: xid=18 fid=0x0 ioctl cmd=0xc009cf0b
new-inotify-ioc-90457 [007] ..... 142217.943569: smb3_ioctl: xid=22 fid=0x389bf5b6 ioctl cmd=0xc009cf0b
Signed-off-by: Steve French <[email protected]>
|
|
cifs writeback doesn't correctly handle the case where
cifs_extend_writeback() hits a point where it is considering an additional
folio, but this would overrun the wsize - at which point it drops out of
the xarray scanning loop and calls xas_pause(). The problem is that
xas_pause() advances the loop counter - thereby skipping that page.
What needs to happen is for xas_reset() to be called any time we decide we
don't want to process the page we're looking at, but rather send the
request we are building and start a new one.
Fix this by copying and adapting the netfslib writepages code as a
temporary measure, with cifs writeback intending to be offloaded to
netfslib in the near future.
This also fixes the issue with the use of filemap_get_folios_tag() causing
retry of a bunch of pages which the extender already dealt with.
This can be tested by creating, say, a 64K file somewhere not on cifs
(otherwise copy-offload may get underfoot), mounting a cifs share with a
wsize of 64000, copying the file to it and then comparing the original file
and the copy:
dd if=/dev/urandom of=/tmp/64K bs=64k count=1
mount //192.168.6.1/test /mnt -o user=...,pass=...,wsize=64000
cp /tmp/64K /mnt/64K
cmp /tmp/64K /mnt/64K
Without the fix, the cmp fails at position 64000 (or shortly thereafter).
Fixes: d08089f649a0 ("cifs: Change the I/O paths to use an iterator rather than a page list")
Signed-off-by: David Howells <[email protected]>
cc: Steve French <[email protected]>
cc: Paulo Alcantara <[email protected]>
cc: Ronnie Sahlberg <[email protected]>
cc: Shyam Prasad N <[email protected]>
cc: Tom Talpey <[email protected]>
cc: Jeff Layton <[email protected]>
cc: [email protected]
cc: [email protected]
cc: [email protected]
cc: [email protected]
Signed-off-by: Steve French <[email protected]>
|
|
Add support for returning reparse mount option in /proc/mounts.
Reported-by: kernel test robot <[email protected]>
Closes: https://lore.kernel.org/oe-kbuild-all/[email protected]/
Signed-off-by: Paulo Alcantara <[email protected]>
Signed-off-by: Steve French <[email protected]>
|
|
Set correct dirent->d_type for IO_REPARSE_TAG_DFS{,R} and
IO_REPARSE_TAG_MOUNT_POINT reparse points.
Signed-off-by: Paulo Alcantara <[email protected]>
Signed-off-by: Steve French <[email protected]>
|
|
Parse the extended attributes from WSL reparse points to correctly
report uid, gid mode and dev from ther instantiated inodes.
Signed-off-by: Paulo Alcantara <[email protected]>
Signed-off-by: Steve French <[email protected]>
|
|
Add a new command to smb2_compound_op() for querying WSL extended
attributes from reparse points.
Signed-off-by: Paulo Alcantara <[email protected]>
Signed-off-by: Steve French <[email protected]>
|
|
This was intended to be an IS_ERR() check. The ea_create_context()
function doesn't return NULL.
Fixes: 1eab17fe485c ("smb: client: add support for WSL reparse points")
Reviewed-by: Paulo Alcantara <[email protected]>
Signed-off-by: Dan Carpenter <[email protected]>
Signed-off-by: Steve French <[email protected]>
|
|
Add support for creating special files via WSL reparse points when
using 'reparse=wsl' mount option. They're faster than NFS reparse
points because they don't require extra roundtrips to figure out what
->d_type a specific dirent is as such information is already stored in
query dir responses and then making getdents() calls faster.
Signed-off-by: Paulo Alcantara <[email protected]>
Signed-off-by: Steve French <[email protected]>
|
|
Replace @desired_access, @create_disposition, @create_options and
@mode parameters with a single @oparms.
No functional changes.
Signed-off-by: Paulo Alcantara <[email protected]>
Signed-off-by: Steve French <[email protected]>
|
|
Now that smb2_compound_op() can accept up to 5 commands in a single
compound request, set the appropriate NextCommand and related flags to
all subsequent commands as well as handling the case where a valid
@cfile is passed and therefore skipping create and close requests in
the compound chain.
This fix a potential broken compound request that could be sent from
smb2_get_reparse_inode() if the client found a valid open
file (@cfile) prior to calling smb2_compound_op().
Signed-off-by: Paulo Alcantara <[email protected]>
Signed-off-by: Steve French <[email protected]>
|
|
In preparation to add support for creating special files also via WSL
reparse points in next commits.
Signed-off-by: Paulo Alcantara <[email protected]>
Signed-off-by: Steve French <[email protected]>
|
|
Allow the user to create special files and symlinks by choosing
between WSL and NFS reparse points via 'reparse={nfs,wsl}' mount
options. If unset or 'reparse=default', the client will default to
creating them via NFS reparse points.
Creating WSL reparse points isn't supported yet, so simply return
error when attempting to mount with 'reparse=wsl' for now.
Signed-off-by: Paulo Alcantara <[email protected]>
Signed-off-by: Steve French <[email protected]>
|
|
There is a shortcoming in the current implementation of the file
lease mechanism exposed when the lease keys were attempted to be
reused for unlink, rename and set_path_size operations for a client. As
per MS-SMB2, lease keys are associated with the file name. Linux smb
client maintains lease keys with the inode. If the file has any hardlinks,
it is possible that the lease for a file be wrongly reused for an
operation on the hardlink or vice versa. In these cases, the mentioned
compound operations fail with STATUS_INVALID_PARAMETER.
This patch adds a fallback to the old mechanism of not sending any
lease with these compound operations if the request with lease key fails
with STATUS_INVALID_PARAMETER.
Resending the same request without lease key should not hurt any
functionality, but might impact performance especially in cases where
the error is not because of the usage of wrong lease key and we might
end up doing an extra roundtrip.
Signed-off-by: Meetakshi Setiya <[email protected]>
Signed-off-by: Steve French <[email protected]>
|
|
When a file/dentry has been deleted before closing all its open
handles, currently, closing them can add them to the deferred
close list. This can lead to problems in creating file with the
same name when the file is re-created before the deferred close
completes. This issue was seen while reusing a client's already
existing lease on a file for compound operations and xfstest 591
failed because of the deferred close handle that remained valid
even after the file was deleted and was being reused to create a
file with the same name. The server in this case returns an error
on open with STATUS_DELETE_PENDING. Recreating the file would
fail till the deferred handles are closed (duration specified in
closetimeo).
This patch fixes the issue by flagging all open handles for the
deleted file (file path to be precise) by setting
status_file_deleted to true in the cifsFileInfo structure. As per
the information classes specified in MS-FSCC, SMB2 query info
response from the server has a DeletePending field, set to true
to indicate that deletion has been requested on that file. If
this is the case, flag the open handles for this file too.
When doing close in cifs_close for each of these handles, check the
value of this boolean field and do not defer close these handles
if the corresponding filepath has been deleted.
Signed-off-by: Meetakshi Setiya <[email protected]>
Signed-off-by: Steve French <[email protected]>
|
|
Currently, when a rename, unlink or set path size compound operation
is requested on a file that has a lot of dirty pages to be written
to the server, we do not send the lease key for these requests. As a
result, the server can assume that this request is from a new client, and
send a lease break notification to the same client, on the same
connection. As a response to the lease break, the client can consume
several credits to write the dirty pages to the server. Depending on the
server's credit grant implementation, the server can stop granting more
credits to this connection, and this can cause a deadlock (which can only
be resolved when the lease timer on the server expires).
One of the problems here is that the client is sending no lease key,
even if it has a lease for the file. This patch fixes the problem by
reusing the existing lease key on the file for rename, unlink and set path
size compound operations so that the client does not break its own lease.
A very trivial example could be a set of commands by a client that
maintains open handle (for write) to a file and then tries to copy the
contents of that file to another one, eg.,
tail -f /dev/null > myfile &
mv myfile myfile2
Presently, the network capture on the client shows that the move (or
rename) would trigger a lease break on the same client, for the same file.
With the lease key reused, the lease break request-response overhead is
eliminated, thereby reducing the roundtrips performed for this set of
operations.
The patch fixes the bug described above and also provides perf benefit.
Signed-off-by: Meetakshi Setiya <[email protected]>
Signed-off-by: Steve French <[email protected]>
|
|
Changes to allocation size are approximated for extending writes of cached
files until the server returns the actual value (on SMB3 close or query info
for example), but it was setting the estimated value for number of blocks
to larger than the file size even if the file is likely sparse which
breaks various xfstests (e.g. generic/129, 130, 221, 228).
When i_size and i_blocks are updated in write completion do not increase
allocation size more than what was written (rounded up to 512 bytes).
Signed-off-by: Steve French <[email protected]>
|
|
The SLAB_MEM_SPREAD flag is already a no-op as of 6.8-rc1, remove
its usage so we can delete it from slab. No functional change.
Link: https://lore.kernel.org/all/[email protected]/
Signed-off-by: Chengming Zhou <[email protected]>
Signed-off-by: Steve French <[email protected]>
|
|
There are cases where a session is disconnected and password has changed
on the server (or expired) for this user and this currently can not
be fixed without unmount and mounting again. This patch allows
remount to change the password (for the non Kerberos case, Kerberos
ticket refresh is handled differently) when the session is disconnected
and the user can not reconnect due to still using old password.
Future patches should also allow us to setup the keyring (cifscreds)
to have an "alternate password" so we would be able to change
the password before the session drops (without the risk of races
between when the password changes and the disconnect occurs -
ie cases where the old password is still needed because the new
password has not fully rolled out to all servers yet).
Cc: [email protected]
Signed-off-by: Steve French <[email protected]>
|
|
In cases of large directories, the readdir operation may span multiple
round trips to retrieve contents. This introduces a potential race
condition in case of concurrent write and readdir operations. If the
readdir operation initiates before a write has been processed by the
server, it may update the file size attribute to an older value.
Address this issue by avoiding file size updates from readdir when we
have read/write lease.
Scenario:
1) process1: open dir xyz
2) process1: readdir instance 1 on xyz
3) process2: create file.txt for write
4) process2: write x bytes to file.txt
5) process2: close file.txt
6) process2: open file.txt for read
7) process1: readdir 2 - overwrites file.txt inode size to 0
8) process2: read contents of file.txt - bug, short read with 0 bytes
Cc: [email protected]
Reviewed-by: Shyam Prasad N <[email protected]>
Signed-off-by: Bharath SM <[email protected]>
Signed-off-by: Steve French <[email protected]>
|
|
When a dirent points to a missing inode, we really should print out the
dirent.
This requires quite a bit of refactoring, but there's some other
benefits: we now do the entire looup (dirent and inode) in a single
btree transaction, and copy to the VFS inode with btree locks still
held, like the create path.
Signed-off-by: Kent Overstreet <[email protected]>
|
|
Signed-off-by: Kent Overstreet <[email protected]>
|
|
Signed-off-by: Kent Overstreet <[email protected]>
|
|
Going to be adding more code here for checking subvol structure.
Signed-off-by: Kent Overstreet <[email protected]>
|
|
Repurposing standard error codes in bcachefs code is banned in new code,
and we need to get rid of the remaining ones - private error codes give
us much better error messages.
Signed-off-by: Kent Overstreet <[email protected]>
|
|
WQ_UNBOUND with max_active 1 means ordered workqueue, but we don't
actually need or want ordered semantics - and probably want a higher
concurrency limit anyways.
Signed-off-by: Kent Overstreet <[email protected]>
|
|
Signed-off-by: Kent Overstreet <[email protected]>
|
|
Need to fix this oversight for the new FS_IOC_(GET|SET)UUID ioctls.
Signed-off-by: Kent Overstreet <[email protected]>
|
|
switch the statfs code from something horrible and open coded to the
more standard uuid_to_fsid()
Signed-off-by: Kent Overstreet <[email protected]>
|
|
Files within a subvolume cannot be renamed into another subvolume, but
subvolumes themselves were intended to be.
This implements subvolume renaming - we need to ensure that there's only
a single dirent that points to a subvolume key (not multiple versions in
different snapshots), and we need to ensure that dirent.d_parent_subol
and inode.bi_parent_subvol are updated.
Signed-off-by: Kent Overstreet <[email protected]>
|
|
btree_and_journal_iter is old code that we want to get rid of, but we're
not ready to yet.
lack of btree node prefetching is, it turns out, a real performance
issue for fsck on spinning rust, so - add it.
Signed-off-by: Kent Overstreet <[email protected]>
|
|
we now always have a btree_trans when using a btree_and_journal_iter;
prep work for adding prefetching to btree_and_journal_iter
Signed-off-by: Kent Overstreet <[email protected]>
|
|
Recently a severe performance regression was discovered, which bisected
to
a6548c8b5eb5 bcachefs: Avoid flushing the journal in the discard path
It turns out the old behaviour, which issued excessive journal flushes,
worked around a performance issue where queueing delays would cause the
journal to not be able to write quickly enough and stall.
The journal flushes masked the issue because they periodically flushed
the device write cache, reducing write latency for non flushes.
This patch reworks the journalling code to allow more than one
(non-flush) write to be in flight at a time. With this patch, doing 4k
random writes and an iodepth of 128, we are now able to hit 560k iops to
a Samsung 970 EVO Plus - previously, we were stuck in the ~200k range.
Signed-off-by: Kent Overstreet <[email protected]>
|
|
Prep work for having multiple journal writes in flight.
Signed-off-by: Kent Overstreet <[email protected]>
|
|
Prep work for having multiple journal writes in flight.
Signed-off-by: Kent Overstreet <[email protected]>
|
|
This gives us a way to record the date and time every journal entry was
written - useful for debugging.
Signed-off-by: Kent Overstreet <[email protected]>
|
|
Signed-off-by: Kent Overstreet <[email protected]>
|
|
Eliminates some error paths - no longer have a hardcoded
BCH_REPLICAS_MAX limit.
Signed-off-by: Kent Overstreet <[email protected]>
|
|
Drop an unnecessary bch2_subvolume_get_snapshot() call, and drop the __
from the name - this is a normal interface.
Signed-off-by: Kent Overstreet <[email protected]>
|
|
Minor renaming for clarity, bit of refactoring.
Signed-off-by: Kent Overstreet <[email protected]>
|
|
Most bcachefs workqueues are used for completions, and should be
WQ_HIGHPRI - this helps reduce queuing delays, we want to complete
quickly once we can no longer signal backpressure by blocking.
Signed-off-by: Kent Overstreet <[email protected]>
|
|
For DT_SUBVOL, we now print both parent and child subvol IDs.
Signed-off-by: Kent Overstreet <[email protected]>
|
|
Signed-off-by: Kent Overstreet <[email protected]>
|
|
Previously, any time we failed to get a journal reservation we'd retry,
with the journal lock held; but this isn't necessary given
wait_event()/wake_up() ordering.
This avoids performance cliffs when the journal starts to get backed up
and lock contention shoots up.
Signed-off-by: Kent Overstreet <[email protected]>
|
|
Signed-off-by: Kent Overstreet <[email protected]>
|
|
Signed-off-by: Kent Overstreet <[email protected]>
|
|
We don't want journal write completions to be blocked behind btree
transactions - io_complete_wq is used for btree updates after data and
metadata writes.
Signed-off-by: Kent Overstreet <[email protected]>
|
|
Signed-off-by: Kent Overstreet <[email protected]>
|
|
When we are checking whether a subvolume is empty in the specified snapshot,
entries that do not belong to this subvolume should be skipped.
This fixes the following case:
$ bcachefs subvolume create ./sub
$ cd sub
$ bcachefs subvolume create ./sub2
$ bcachefs subvolume snapshot . ./snap
$ ls -a snap
. ..
$ rmdir snap
rmdir: failed to remove 'snap': Directory not empty
As Kent suggested, we pass 0 in may_delete_deleted_inode() to ignore subvols
in the subvol we are checking, because inode.bi_subvol is only set on
subvolume roots, and we can't go through every inode in the subvolume and
change bi_subvol when taking a snapshot. It makes the check less strict, but
that's ok, the rest of fsck will still catch it.
Signed-off-by: Guoyu Ou <[email protected]>
Signed-off-by: Kent Overstreet <[email protected]>
|