ctdb-common: Fix a warning in the pcap code

[173/416] Compiling ctdb/common/system_socket.c
../../common/system_socket.c: In function ‘ctdb_sys_read_tcp_packet’:
../../common/system_socket.c:1016:15: error: cast discards ‘const’ qualifier from pointer target type [-Werror=cast-qual]
 1016 |         eth = (struct ether_header *)buffer;
      |               ^

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>

ctdb-common: Do not use raw socket when ENABLE_PCAP is defined

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>

ctdb-common: Move a misplaced comment

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>

ctdb-build: Add --enable-pcap configure option

This forces the use pcap for packet capture on Linux.

It appears that using a raw socket for capture does not work with
infiniband - pcap support for that to come.

Don't (yet?) change the default capture method to pcap.  On some
platforms (e.g. my personal Intel NUC, running Debian testing), pcap
is much less reliable than the raw socket.  However, pcap seems fine
on most other platforms.

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>

ctdb-build: Use pcap-config when available

The build currently fails on AIX, which can't find the pcap headers
because they're installed in a non-standard place.  However, there is
a pcap-config script available.

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>

s3:locking: remove unused get_share_mode_lock()

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>
Autobuild-User(master): Jeremy Allison <jra@samba.org>
Autobuild-Date(master): Tue Sep 20 01:34:55 UTC 2022 on sn-devel-184

s3:smbd: make use of share_mode_entry_prepare_{lock_add,unlock}() in open_{file_ntcreate,directory}()

This gives a nice speed up...

The following test with 256 commections all looping with open/close
on the same inode (share root) is improved drastically:

  smbtorture //127.0.0.1/m -Uroot%test smb2.bench.path-contention-shared \
         --option='torture:bench_path=' \
         --option="torture:timelimit=60" \
         --option="torture:nprocs=256" \
         --option="torture:qdepth=1"

From something like this:

    open[num/s=11536,avslat=0.011450,minlat=0.000039,maxlat=0.052707]
    close[num/s=11534,avslat=0.010878,minlat=0.000022,maxlat=0.052342]

(only this commit with the close part reverted) to:

    open[num/s=12722,avslat=0.009548,minlat=0.000051,maxlat=0.054338]
    close[num/s=12720,avslat=0.010701,minlat=0.000033,maxlat=0.054372]

(with both patches) to:

    open[num/s=37680,avslat=0.003471,minlat=0.000040,maxlat=0.061411]
    close[num/s=37678,avslat=0.003440,minlat=0.000022,maxlat=0.051536]

So we are finally perform similar like we did in Samba 4.12,
which resulted in:

    open[num/s=36846,avslat=0.003574,minlat=0.000043,maxlat=0.020378]
    close[num/s=36844,avslat=0.003552,minlat=0.000026,maxlat=0.020321]

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:smbd: let open_file_ntcreate() calculate info = FILE_WAS_* before get_share_mode_lock()

This will simplify further changes.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:smbd: make use of share_mode_entry_prepare_{lock_del,unlock}() in close_{remove_share_mode,directory}()

This gives a nice speed up...

The following test with 256 commections all looping with open/close
on the same inode (share root) is improved drastically:

  smbtorture //127.0.0.1/m -Uroot%test smb2.bench.path-contention-shared \
         --option='torture:bench_path=' \
         --option="torture:timelimit=60" \
         --option="torture:nprocs=256" \
         --option="torture:qdepth=1"

From some like this:

    open[num/s=11536,avslat=0.011450,minlat=0.000039,maxlat=0.052707]
    close[num/s=11534,avslat=0.010878,minlat=0.000022,maxlat=0.052342]

to:
    open[num/s=13225,avslat=0.010504,minlat=0.000042,maxlat=0.054023]
    close[num/s=13223,avslat=0.008971,minlat=0.000022,maxlat=0.053838]

But this is only half of the solution, the next commits will
add a similar optimization to the open code, at the end we'll
perform like we did in Samba 4.12:

    open[num/s=37680,avslat=0.003471,minlat=0.000040,maxlat=0.061411]
    close[num/s=37678,avslat=0.003440,minlat=0.000022,maxlat=0.051536]

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:smbd: make use of close_share_mode_lock_{prepare,cleanup}() in close_directory()

It's good to have this in common as close_remove_share_mode()
and in the end we'll avoid get_existing_share_mode_lock()
and call them via share_mode_entry_prepare_{lock,unlock}(),
so that they can run under a tdb chainlock.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:smbd: split out some generic code from close_remove_share_mode()

close_share_mode_lock_prepare() will operates on share_mode_lock
in order to check if the object needs to be deleted or if
we can remove the share_mode_entry directly.

close_share_mode_lock_cleanup() will finish after the object
has been deleted.

We can reuse these function in close_directory() soon and
in the end we'll avoid get_existing_share_mode_lock()
and call them via share_mode_entry_prepare_{lock,unlock}(),
so that they can run under a tdb chainlock.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:smbd: avoid remove_oplock() in close_remove_share_mode()

This inlines remove_oplock() into close_remove_share_mode() and
calls remove_share_oplock() and release_file_oplock() directly.

The idea is that we'll soon call remove_share_oplock()
under a tdb chainlock, while release_file_oplock() needs to be called outside.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:smbd: let close_directory() only change the user if needed

The logic is now similar to close_remove_share_mode().

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:smbd: remove one indentation level in close_directory()

We now use a goto done in order to skip the deletion part.

This means the code flow is now almost idential compared to
close_remove_share_mode().

It prepares to split common code to be shared by
close_remove_share_mode() and close_directory().

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:smbd: let close_directory() use the same delete_dir logic as close_remove_share_mode()

This will make further changes simpler.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:smbd: improve !delete_file logic in close_remove_share_mode()

This makes it much easier to understand the logic (at least for me).

It will make the following changes easier.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:smbd: let close_directory() hold the lock during delete_all_streams/rmdir_internals

Now that we're using g_lock, it doesn't mean we're holding a tdb
chainlock.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:smbd: remove static from release_file_oplock()

It will be used in close.c in the next commit.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:smbd: maintain all SHARE_MODE_LEASE_* flags not only _READ

Remember SMB2 Create is the only was to upgrade a lease.

The strategy is that opening of a file will always result
in storing the total lease bits.

But we're lazy clearing the flags on close.

We'll only clear them by traversing all entries when
we break a NONE or when opening a new handle.

We don't do any decision on SHARE_MODE_LEASE_{HANDLE,WRITE},
maybe we'll do in future, but at least it should be much more
sane for debugging now!

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:smbd: split out check_and_store_share_mode()

This shows that the code in open_file_ntcreate() and
open_directory() is basically the same now, which
simplifies things a lot.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:smbd: also call handle_share_mode_lease for directories

It means we call open_mode_check() now only via handle_share_mode_lease()
and the fact that we never grant any directory leases (yet), means
that delay_for_oplocks() avoids the share_mode_forall_entries() loop.

This is a way into supporting directory leases, but that's not
the point for this commit, the point is that.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:smbd: prepare delay_for_oplock() for directories

We don't support directory leases yet, so it should be
an noop for now.

The point is that we want to call
delay_for_oplock(oplock_request=NO_OPLOCK)
for directories soon.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:smbd: call set_file_oplock() after set_share_mode()

The important part is the call to get a kernel oplock is deferred
until after set_share_mode(). The goal is to get the code
between get_share_mode_lock() and set_share_mode() free of any
blocking operation.

As we were optimistic to get the oplock that was asked for,
we need to remove_share_oplock() in order to set NO_OPLOCK
also in the share_mode entry.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:smbd: call grant_fsp_lease() after set_share_mode()

This means we don't have to call remove_lease_if_stale() if
set_share_mode() fails. It's easier to cleanup the share mode entry.

And it makes the code flow easier to the following changes.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:smbd: move grant_fsp_lease()/set_file_oplock() out of handle_share_mode_lease()

The aim is to call set_file_oplock() after set_share_mode(), so that we
only ask for kernel oplocks after set_share_mode().

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:smbd: move grant_fsp_lease()/set_file_oplock() out of delay_for_oplocks()

It means delay_for_oplocks() is no longer asking for kernel oplocks.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:smbd: add more detailed debugging to delay_for_oplock()

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: add share_mode_entry_prepare_{lock,unlock}() infrastructure

When adding or deleting share mode entries elements, we typically
have a pattern like this:

1. get the g_lock via get_[existing_]share_mode_lock()
2. do some checking of the existing record
3. add/delete a share_mode_entry to the record
4. do some vfs operations still protected by the g_lock
5. (optional) cleanup of the record on failure
6. release the g_lock

We can optimize this to:

- Run 1-3. under a tdb chainlock
- Only protect vfs operations with the g_lock
  if a new file was created/will be deleted
- Regrab the g_lock for a cleanup.

The new share_mode_entry_prepare_lock()
allows the caller to run a function within a tdb chainlock
similar to share_mode_do_locked_vfs_denied() where vfs calls are denied
and the execution is done within a tdb chainlock.

But the callback function is allowed to decide if it wants to
keep the lock at the g_lock layer on return.

The decision is kept in struct share_mode_entry_prepare_state,
which is then passed to share_mode_entry_prepare_unlock()
with an optional callback to do some cleanup under the
still existing g_lock or a regrabed g_lock.

In the ideal case the callback function passed to
share_mode_entry_prepare_lock() is able to decide that
it can drop the g_lock and the share_mode_entry_prepare_unlock().
gets a NULL callback as there's nothing to cleanup.
In this case share_mode_entry_prepare_unlock() is a noop.

This will allow us to avoid fallbacks to the dbwrap_watch based
waiting for the g_lock in the SMB2 Create and Close code paths.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: optimize share_mode_do_locked_vfs_denied() with g_lock_lock callback

It means that in callers function will run under a single tdb chainlock,
which means callers from the outside will never see the record being
locked at g_lock level, as the g_lock is only held in memory.
within the single tdb chainlock. As a result we'll very unlikely hit
the case where we need to wait for a g_lock using the dbwrap_watch
logic.

Review with: git show -w

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: add current_share_mode_glck helper functions

We'll soon make use of callback functions passed to g_lock_lock(),
during these callback function we'll only be allowed to
call 'g_lock_lock_cb_state' based functions.

Given that nesting of share_mode call, we need to
make it transparent to the callers and the detail
that we optimize using g_lock_lock() callbacks.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:g_lock: add callback function to g_lock_lock()

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:g_lock: add callback function to g_lock_lock_send()

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:g_lock: add callback function to g_lock_lock_simple_fn()

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:g_lock: add callback function to g_lock_trylock()

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:g_lock: reorder the logic in g_lock_trylock()

We now have only one code path that stores the fully
granted lock.

This is not change in behavior, but it will simplify further
changes.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:g_lock: remove redundant code in g_lock_trylock()

g_lock_cleanup_shared() handles lck.num_shared == 0 just fine...

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:g_lock: add g_lock_ctx->busy and assert it to false

This prepares some helper functions in order to
allow callers of g_lock_lock() to pass in a callback function
that will run under the tdb chainlock when G_LOCK_WRITE was granted.

The idea is that the callers callback function would run with
g_lock_ctx->busy == true and all key based function are not be allowed
during the execution of the callback function. Only the
g_lock_lock_cb_state based helper function are allowed to be used.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:g_lock: add g_lock_lock_cb_state infrastructure

This prepares some helper functions in order to
allow callers of g_lock_lock() to pass in a callback function
that will run under the tdb chainlock when G_LOCK_WRITE was granted.

The idea is that the callers callback function would only be allowed
to run these new helper functions, while all key based function are
not to be allowed.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:g_lock: reorder the logic in g_lock_lock_simple_fn()

First we fully check if we'll get the lock
and then store the lock.

This is not change in behavior, but it will simplify further changes.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

lib/dbwrap: allow dbwrap_merge_dbufs() to update an existing buffer

This will be useful in future...

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:dbwrap_watch: add dbwrap_watched_watch_force_alerting()

This is useful when we want to wakeup the next watcher
without modifying the record.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:dbwrap_watch: add dbwrap_watched_watch_reset_alerting() helper

This can be used if the decision of using dbwrap_watched_watch_skip_alerting()
needs to be reverted...

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:dbwrap_watch: let dbwrap_watched_watch_skip_alerting() also clear the selected watcher

If a watcher was already selected for a wakeup notification reset it...

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: let _share_mode_do_locked_vfs_* use get/put_share_mode_lock_internal

This avoids calling talloc(mem_ctx, struct share_mode_lock)
and uses stack variables instead.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: split out put_share_mode_lock_internal()

This pairs with get_share_mode_lock_internal() and will allow us
to use a struct share_mode_lock stack variable in future,
which will be much cheaper.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: split out get_share_mode_lock_internal()

This detaches the logic from the talloc(mem_ctx, struct share_mode_lock).
In future we will have cases where we use a stack variable instead,
which will be much cheaper.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: remove static_share_mode_data_refcount

The effective value of share_mode_lock_key_refcount
is 'share_mode_lock_key_refcount + static_share_mode_data_refcount',
which is quite confusing.

This complexity is not needed and we can just use
share_mode_lock_key_refcount.

This will also simplify further changes.

Review with: git show -U15 -w

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: move from uint8_t share_mode_lock_key_data[] to struct file_id

This will allow us to have better debug messages and will also make
further changes easier.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:smb2_trans2: make use of share_mode_do_locked_vfs_allowed() in smb_posix_unlink()

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:smbd: let update_write_time_on_close() use share_mode_do_locked_vfs_denied()

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:smbd: let update_write_time_on_close() use share_mode_{old,changed}_write_time()

We're already holding a share_mode_lock, so we can use it directly.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: make use of share_mode_do_locked_vfs_denied() in set_sticky_write_time()

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: make use of share_mode_do_locked_vfs_denied() in set_write_time()

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: make use of share_mode_do_locked_vfs_denied() in file_has_open_streams()

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:smbd: let lease_match() use share_mode_do_locked_vfs_denied()

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: let set_delete_on_close() use share_mode_do_locked_vfs_denied()

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: make use of new share_mode_set_{changed,old}_write_time() helpers

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: add share_mode_set_{changed,old}_write_time() helpers

These will be used in future to call them unter an existing share mode
lock...

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:smbd: let setup_poll_open() use share_mode_do_locked_vfs_denied()

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:smbd: move get_existing_share_mode_lock() into setup_poll_open()

This will simplify the next steps...

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: make 'struct share_mode_lock' private to share_mode_lock.c

There are no callers left dereferencing it.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: make use of share_mode_lock_access_private_data() in reset_share_mode_entry()

We should avoid dereference 'struct share_mode_lock' as much as possible.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: pass struct share_mode_data to share_mode_entry_do()

We should avoid dereference 'struct share_mode_lock' as much as possible.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: make use of share_mode_lock_access_private_data() in share_mode_forall_entries()

We should avoid dereference 'struct share_mode_lock' as much as possible.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: make use of share_mode_lock_file_id() in share_mode_watch_send()

We should avoid dereference 'struct share_mode_lock' as much as possible.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: add and use share_mode_lock_assert_private_data()

We should avoid dereference 'struct share_mode_lock' as much as possible.

In some places we just rely on share_mode_lock_access_private_data()
to work, if needed the caller should already check it's result...

Note that share_mode_lock_assert_private_data() can't fail up to
now, but we want to change that in future and only load it on
demand.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: let get_share_mode_write_time() use share_mode_lock_access_private_data()

We should avoid dereference 'struct share_mode_lock' as much as possible.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: let set_write_time() use share_mode_lock_access_private_data()

We should avoid dereference 'struct share_mode_lock' as much as possible.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: let set_sticky_write_time() use share_mode_lock_access_private_data()

We should avoid dereference 'struct share_mode_lock' as much as possible.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: let is_delete_on_close_set() use share_mode_lock_access_private_data()

We should avoid dereference 'struct share_mode_lock' as much as possible.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: let get_delete_on_close_token() use share_mode_lock_access_private_data()

We should avoid dereference 'struct share_mode_lock' as much as possible.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: let set_delete_on_close_lck() use share_mode_lock_access_private_data()

We should avoid dereference 'struct share_mode_lock' as much as possible.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: let reset_delete_on_close_lck() use share_mode_lock_access_private_data()

We should avoid dereference 'struct share_mode_lock' as much as possible.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: let remove_lease_if_stale() use share_mode_lock_file_id()

We should avoid dereference 'struct share_mode_lock' as much as possible.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: add share_mode_lock_file_id()

This will simplify some (mostly debug) code soon.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: make use of share_mode_lock_access_private_data() in rename_share_filename()

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: let rename_share_filename_state maintain a struct share_mode_data pointer

We only need to access lck->data once...

This will simplify further changes.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: add share_mode_lock_access_private_data()

In future we should avoid dereference 'struct share_mode_lock'
as much as possible.

This will also allow us to load struct share_mode_data
only if required in future.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: let share_mode_wakeup_waiters() use share_mode_do_locked_vfs_denied()

This allows us get rid of the otherwise unused share_mode_do_locked().

It means we only have one code path that handles the g_lock handling.

This looks like a performance degradation, but all callers of
share_mode_wakeup_waiters() already took the share_mode_lock,
so we only increment the refcount. Note the additional
talloc(mem_ctx, struct share_mode_lock) will be optimized away
in the next commits.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: make share_mode_do_locked() static

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: protect do_lock() with share_mode_do_locked_vfs_allowed()

share_mode_do_locked() will be make static soon.

Here we just want to avoid concurrent access to brlock.tdb
in order to maintain the lock order, we're not interested in the
locking.tdb content at all, expect that there's at least one
entry.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:smbd: protect smbd_do_unlocking() with share_mode_do_locked_vfs_allowed()

share_mode_do_locked() will be make static soon.

Here we just want to avoid concurrent access to brlock.tdb
in order to maintain the lock order, we're not interested in the
locking.tdb content at all, expect that there's at least one
entry and we want to wake potential watchers.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: add share_mode_do_locked_vfs_{denied,allowed}()

These function will add an abstraction to protect
a function that is not allowed to call vfs functions
or allow vfs functions to be called.

Currently these are implemented similar,
but we'll optimize them in the next commits.

The idea is that share_mode_do_locked_vfs_denied()
will be able to run fast enough in order to run
under a tdb chainlock (just a pthread mutex).

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:smbd: add smb_vfs_assert_allowed() to kernel oplock code

Kernel oplocks can block in the same way vfs operations can do.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:smbd: add helpers to deny vfs calls in some sections

Code denying vfs calls can do:

{
   struct smb_vfs_deny_state vfs_deny = {};

   smb_vfs_deny_push(&vfs_deny);

   VFS calls are not allowed here...

   smb_vfs_deny_pop(&vfs_deny);
}

This will allow us to safely run some code under a
tdb chainlock later...

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:smbd: move VFS_FIND() to smbd/vfs.c

It's only used there...

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:smbd: move locking related vfs functions to smbd/vfs.c

This allows us to make VFS_FIND local to smbd/vfs.c in the
next step.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: just use g_lock_dump() for fsp_update_share_mode_flags()

We don't need to protect this with g_lock_lock/g_lock_unlock
as we just want the current flags, we're still protected by the
dbwrap layer lock.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: move fsp_update_share_mode_flags* related functions further down

It will soon need to use 'struct locking_tdb_data'

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: replace locking_tdb_data_store() with share_mode_data_ltdb_store()

This means we flush share_mode_data at the same time...

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: let share_mode_forall_entries() call TALLOC_FREE(ltdb)

We should free ltdb as soon as possible...

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: split out share_mode_data_ltdb_store()

This will allow us to use it in other places too
and we'll avoid to storing multiple times.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: introduce share_mode_data->not_stored

share_mode_data->fresh was very similar, but only set
and never used.

Now we remember 'not_stored' instead, the 'not_' is easier
as ndr_pull sets [skip] elements to 0.

We use this as indication to move the value to
memcache.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: change some debug messages to level unless we got NT_STATUS_NOT_FOUND

NT_STATUS_NOT_FOUND is not a real error in most cases so we should keep
it on level 10, but all other errors should never be without notice...

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: log all share_mode_forall_entries() errors at level 0

These should never fail without notice...

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: let share_mode_forall_leases() log all errors at level 0

These should never fail without notice...

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: let set_delete_on_close_lck() log errors and panic

Most of the calls in set_delete_on_close_lck() are checked with
asserts, so do panic in all situation where things go wrong in an
unexpected way.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: log g_lock_dump() error in locking_tdb_data_fetch() at level 0

This should never fail without notice...

Note we already checked for NT_STATUS_NOT_FOUND before.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: log add locking_tdb_data_store() errors at level 0

These should never fail without notice...

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3:locking: log g_lock_locks() error at level 0

These should never fail without notice...

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15125

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>