Pull smb client fixes from Steve French:

 - SMB3.1.1 POSIX extensions fix for char remapping

 - Fix for repeated directory listings when directory leases enabled

 - deferred close handle reuse fix

* tag 'v6.16-rc1-smb3-client-fixes' of git://git.samba.org/sfrench/cifs-2.6:
  smb: improve directory cache reuse for readdir operations
  smb: client: fix perf regression with deferred closes
  smb: client: disable path remapping with POSIX extensions

Currently, cached directory contents were not reused across subsequent
'ls' operations because the cache validity check relied on comparing
the ctx pointer, which changes with each readdir invocation. As a
result, the cached dir entries was not marked as valid and the cache was
not utilized for subsequent 'ls' operations.

This change uses the file pointer, which remains consistent across all
readdir calls for a given directory instance, to associate and validate
the cache. As a result, cached directory contents can now be
correctly reused, improving performance for repeated directory listings.

Performance gains with local windows SMB server:

Without the patch and default actimeo=1:
 1000 directory enumeration operations on dir with 10k files took 135.0s

With this patch and actimeo=0:
 1000 directory enumeration operations on dir with 10k files took just 5.1s

Signed-off-by: Bharath SM <bharathsm@microsoft.com>
Reviewed-by: Shyam Prasad N <sprasad@microsoft.com>
Cc: stable@vger.kernel.org
Signed-off-by: Steve French <stfrench@microsoft.com>

Customer reported that one of their applications started failing to
open files with STATUS_INSUFFICIENT_RESOURCES due to NetApp server
hitting the maximum number of opens to same file that it would allow
for a single client connection.

It turned out the client was failing to reuse open handles with
deferred closes because matching ->f_flags directly without masking
off O_CREAT|O_EXCL|O_TRUNC bits first broke the comparision and then
client ended up with thousands of deferred closes to same file.  Those
bits are already satisfied on the original open, so no need to check
them against existing open handles.

Reproducer:

 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
 #include <unistd.h>
 #include <fcntl.h>
 #include <pthread.h>

 #define NR_THREADS      4
 #define NR_ITERATIONS   2500
 #define TEST_FILE       "/mnt/1/test/dir/foo"

 static char buf[64];

 static void *worker(void *arg)
 {
         int i, j;
         int fd;

         for (i = 0; i < NR_ITERATIONS; i++) {
                 fd = open(TEST_FILE, O_WRONLY|O_CREAT|O_APPEND, 0666);
                 for (j = 0; j < 16; j++)
                         write(fd, buf, sizeof(buf));
                 close(fd);
         }
 }

 int main(int argc, char *argv[])
 {
         pthread_t t[NR_THREADS];
         int fd;
         int i;

         fd = open(TEST_FILE, O_WRONLY|O_CREAT|O_TRUNC, 0666);
         close(fd);
         memset(buf, 'a', sizeof(buf));
         for (i = 0; i < NR_THREADS; i++)
                 pthread_create(&t[i], NULL, worker, NULL);
         for (i = 0; i < NR_THREADS; i++)
                 pthread_join(t[i], NULL);
         return 0;
 }

Before patch:

$ mount.cifs //srv/share /mnt/1 -o ...
$ mkdir -p /mnt/1/test/dir
$ gcc repro.c && ./a.out
...
number of opens: 1391

After patch:

$ mount.cifs //srv/share /mnt/1 -o ...
$ mkdir -p /mnt/1/test/dir
$ gcc repro.c && ./a.out
...
number of opens: 1

Cc: linux-cifs@vger.kernel.org
Cc: David Howells <dhowells@redhat.com>
Cc: Jay Shin <jaeshin@redhat.com>
Cc: Pierguido Lambri <plambri@redhat.com>
Fixes: b8ea3b1ff544 ("smb: enable reuse of deferred file handles for write operations")
Acked-by: Shyam Prasad N <sprasad@microsoft.com>
Signed-off-by: Paulo Alcantara (Red Hat) <pc@manguebit.org>
Signed-off-by: Steve French <stfrench@microsoft.com>

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

We normally can't create a new directory with the case-insensitive
option already set - except when we're creating a snapshot.

And if casefolding is enabled filesystem wide, we should still set it
even though not strictly required, for consistency.

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Previously, we only ever logged the filesystem UUID.

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

We have to be able to print superblock sections even if they fail to
validate (for debugging), so we have to calculate the number of entries
from the field size.

Reported-by: syzbot+5138f00559ffb3cb3610@syzkaller.appspotmail.com
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

It seems btree node scan picked up a partially overwritten btree node,
and corrected the "bset version older than sb version_min" error -
resulting in an invalid superblock with a bad version_min field.

Don't run this check at all when we're in btree node scan, and when we
do run it, do something saner if the bset version is totally crazy.

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Multiple ioctl handlers individually use a lot of stack space, and clang chooses
to inline them into the bch2_fs_ioctl() function, blowing through the warning
limit:

fs/bcachefs/chardev.c:655:6: error: stack frame size (1032) exceeds limit (1024) in 'bch2_fs_ioctl' [-Werror,-Wframe-larger-than]
  655 | long bch2_fs_ioctl(struct bch_fs *c, unsigned cmd, void __user *arg)

By marking the largest two of them as noinline_for_stack, no indidual code path
ends up using this much, which avoids the warning and reduces the possible
total stack usage in the ioctl handler.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

fsck_err() can return a transaction restart if passed a transaction
object - this has always been true when it has to drop locks to prompt
for user input, but we're seeing this more now that we're logging the
error being corrected in the journal.

gc_accounting_done() doesn't call fsck_err() from an actual commit loop,
and it doesn't need to be holding btree locks when it calls fsck_err(),
so the easy fix here for the unhandled transaction restart is to just
not pass it the transaction object. We'll miss out on the fancy new
logging, but that's ok.

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>