Ensure ioend->io_error gets propagated back to e.g. AIO completions.
Signed-off-by: Christoph Hellwig <[email protected]>
Reviewed-by: Alex Elder <[email protected]>
The log item ops aren't nessecarily the biggest exploit vector, but marking
them const is easy enough. Also remove the unused xfs_item_ops_t typedef
while we're at it.
Signed-off-by: Christoph Hellwig <[email protected]>
Reviewed-by: Dave Chinner <[email protected]>
Reviewed-by: Alex Elder <[email protected]>
Instead of calling into transport_emulate_control_cdb from
__transport_execute_tasks for some CDBs always set up ->exectute_tasks
in the command sequence and use it uniformly.
(nab: Add default passthrough break for SERVICE_ACTION_IN)
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nicholas Bellinger <[email protected]>
All ->execute_task instances now need to complete the I/O explicitly,
which can either happen synchronously or asynchronously.
Note that a lot of the CDB emulations appear to return success even if
some lowlevel operations failed. Given that this is an existing issue
this patch doesn't change that fact.
(nab: Adding missing switch breaks in PR-IN + PR_OUT)
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nicholas Bellinger <[email protected]>
Encapsulate each CDB emulation into a function of its own, to prepare
setting ->exectute_task to these routines.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nicholas Bellinger <[email protected]>
We want to be able to handle all CDBs through it and remove hacks like
always using the first task in a CDB in target_report_luns.
Also rename the callback to ->execute_task to better describe its use.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nicholas Bellinger <[email protected]>
Split core_scsi2_emulate_crh into one routine each for the
PERSISTENT_RESERVE_IN and PERSISTENT_RESERVE_OUT side.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nicholas Bellinger <[email protected]>
Split core_scsi2_emulate_crh into one routine each for the reserve and
release side. The common code now is in a helper called by both
routines.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nicholas Bellinger <[email protected]>
If we only have a single task per command (which at least in my testing
is the by far most common case) we do not have to allocate a new per-task
S/G list but can reuse the one from the command.
(nab: Fix BIDI handling in transport_free_dev_tasks)
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nicholas Bellinger <[email protected]>
There were only two callers, and one of them always wants the call
to transport_allocate_data_tasks anyway. Also drop the constant
lba argument to transport_allocate_data_tasks and move the variables
inside it into the minimum required scope.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nicholas Bellinger <[email protected]>
These are two fairly small functions, and merging them gives a much
more readable control flow, and opportunities for more useful comments.
It also moves all code related to resources allocation closer together
and allows to remove a forward declaration for transport_allocate_tasks.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nicholas Bellinger <[email protected]>
This field is never used given that BIDI handling happens at the
command and not the task level. Remove it and the dead code in
pscsi that tries to work on it.
It also prevents pSCSI passthrough for the two currently enabled BIDI
commands now that task->task_sg_bidi support has been removed.
Signed-off-by: Christoph Hellwig <[email protected]>
Cc: Boaz Harrosh <[email protected]>
Signed-off-by: Nicholas Bellinger <[email protected]>
Instead of abusing the target processing thread for offloading I/O
completion in the backends to user context add a new workqueue. This means
completions can be processed as fast as available CPU time allows it,
including in parallel with other completions and more importantly I/O
submission or QUEUE FULL retries. This should give much better performance
especially on loaded systems.
As a fallout we can merge all the completed states into a single
one.
On the downside this change complicates lun reset handling a bit by
requiring us to cancel a work item only for those states that have it
initialized. The alternative would be to either always initialize the work
item to a dummy handler, or always use the same handler and do a switch on
the state. The long term solution will be a flag that says that the command
has an initialized work item, but that's only going to be useful once we
have more users.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nicholas Bellinger <[email protected]>
We never check for this state, and it makes testing for a completed
state much harder given that it overrides the existing state.
Also remove the unused deferred_t_state which is related to it.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nicholas Bellinger <[email protected]>
We never queue an command with this state, and only set it in a completely
bogus place in tcm_fc.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nicholas Bellinger <[email protected]>
We only need to decrement dev->depth_left if failing a command from
__transport_execute_tasks. Instead of doing it first thing in
transport_generic_request_failure and requiring a pseudo-flag argument
for it just opencode the decrement in the two callers (which should
be factored into a single one anyway)
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nicholas Bellinger <[email protected]>
Currently we stop the timers for all tasks in a command fairly late during
I/O completion, which is fairly pointless and requires all kinds of safety
checks.
Instead delete pending timers early on in transport_complete_task, thus
ensuring no new timers firest after that. We take t_state_lock a bit later
in that function thus making sure currenly running timers are out of the
criticial section. To be completely sure the timer has finished we also
add another del_timer_sync call when freeing the task.
This also allows removing TF_TIMER_RUNNING as it would be equivalent
to TF_ACTIVE now.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nicholas Bellinger <[email protected]>
TF_TIMER_STOP is useless as it only helps to mitigate a tiny race during
deleting the timer. But given that we have cleared TF_ACTIVE at this point
we already have another mitigation a few lines down the function.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nicholas Bellinger <[email protected]>
list_for_each_entry_safe only protects against deletions from the list,
but not against any concurrent modifications. Given that we drop
t_state_lock inside the loop it is not safe in transport_free_dev_tasks.
Instead of use a local dispose_list that we move all tasks that are
to be deleted to. This is safe because we never do list_emptry checks
on t_list to check if a command is on the list anywhere.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nicholas Bellinger <[email protected]>
Change one remaining user of transport_cmd_check_stop(cmd, 2, 0) to the
transport_cmd_check_stop_to_fabric wrapper.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nicholas Bellinger <[email protected]>
We always operated on the same queue, so move finding it into the function,
just like we do for all other helpers operating on it.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nicholas Bellinger <[email protected]>
Add a new boolean at_head parameter to transport_add_cmd_to_queue and thus
obsolete the SCF_EMULATE_QUEUE_FULL flag.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nicholas Bellinger <[email protected]>
Remove the need for the transport_qf_callback callback by making
sure we have specific states with specific handlers for the two
queue full cases.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nicholas Bellinger <[email protected]>
Remove the dpo_emulated, fua_write_emulated, fua_read_emulated and
write_cache_emulated methods, and replace them with a simple bitfields in
se_subsystem_api in those cases where they ever returned one.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nicholas Bellinger <[email protected]>
Do not block the submitting thread when handling a SYNCHRONIZE CACHE command,
but implement it asynchronously by sending the FLUSH command ourself and
calling transport_complete_sync_cache from the completion handler.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nicholas Bellinger <[email protected]>
Add a switch statement implementing the CDB LBA/len update directly
in target_get_task_cdb and remove the old ->transport_split_cdb
callback and all its implementations.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nicholas Bellinger <[email protected]>
Instead of calling out to the backends from the core to get a per-task
CDB and then modify it for the LBA/len pair used for this CDB provide
a helper that writes the adjusted CDB into a provided buffer and call
this method from ->do_task in pscsi.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nicholas Bellinger <[email protected]>
The most commonly used file, iblock and rd backends have no use for
a per-task CDB and thus don't need a method to copy it into their
otherwise unused CDB fields.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nicholas Bellinger <[email protected]>
Rearrange the fields in se_task to avoid holes. Also increase the
flags field to 16 bits as we have the space for it, and this makes
adding new flags safer.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nicholas Bellinger <[email protected]>
This is a squashed version of the following unnecessary se_task structure
member removal patches:
target: remove the task_execute_queue field in se_task
Instead of using a separate flag we can simply do list_emptry checks
on t_execute_list if we make sure to always use list_del_init to remove
a task from the list. Also factor some duplicate code into a new
__transport_remove_task_from_execute_queue helper.
target: remove the read-only task_no field in se_task
The task_no field never was initialized and only used in debug printks,
so kill it.
target: remove the task_padded_sg field in se_task
This field is only check in one place and not actually needed there.
Rationale:
- transport_do_task_sg_chain asserts that we have task_sg_chaining
set early on
- we only make use of the sg_prev_nents field we calculate based on it
if there is another sg list that gets chained onto this one, which
never happens for the last (or only) task.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nicholas Bellinger <[email protected]>
Replace various atomic_t variables that were mostly under t_state_lock
with new flags in task_flags. Note that the execution error path
didn't take t_state_lock before, so add it there.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nicholas Bellinger <[email protected]>
This is a squashed version of the following se_task cleanup patches:
target: remove the unused task_state_flags field in se_task
target: remove the unused se_obj_ptr field in se_task
target: remove the se_dev field in se_task
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nicholas Bellinger <[email protected]>
This is a squashed version of the following target_core_base.h
cleanup patches:
target: remove the unused SHUTDOWN_SIGS defintion
target: remove unused se_mem leftovers
target: remove the unused map_func_t typedef
target: move TRANSPORT_IOV_DATA_BUFFER to the iscsi-specific code
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nicholas Bellinger <[email protected]>
The cdb_none, map_data_SG and map_control_SG methods have no callers left
and can be removed now.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nicholas Bellinger <[email protected]>
Move the entirely request allocation, mapping and submission into ->do_task.
This
a) avoids blocking the I/O submission thread unessecarily, and
b) simplifies the code greatly
Note that the code seems to have various error handling issues, mostly
related to bidi handling in the current form. I've added comments about
those but not tried to fix them in this commit.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nicholas Bellinger <[email protected]>
Move the entirely bio allocation, mapping and submission into ->do_task.
This
a) avoids blocking the I/O submission thread unessecarily, and
b) simplifies the code greatly
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nicholas Bellinger <[email protected]>
Currently the loop device tries to call directly into write_begin/write_end
instead of going through ->write if it can. This is a fairly nasty shortcut
as write_begin and write_end are only callbacks for the generic write code
and expect to be called with filesystem specific locks held.
This code currently causes various issues for clustered filesystems as it
doesn't take the required cluster locks, and it also causes issues for XFS
as it doesn't properly lock against the swapext ioctl as called by the
defragmentation tools. This in case causes data corruption if
defragmentation hits a busy loop device in the wrong time window, as
reported by RH QA.
The reason why we have this shortcut is that it saves a data copy when
doing a transformation on the loop device, which is the technical term
for using cryptoloop (or an XOR transformation). Given that cryptoloop
has been deprecated in favour of dm-crypt my opinion is that we should
simply drop this shortcut instead of finding complicated ways to to
introduce a formal interface for this shortcut.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Jens Axboe <[email protected]>
When we call xfs_flush_buftarg (generally from sync or umount) it already
is too late to flush the data workqueues, as I/O completion is signalled
for them and we are thus already done with the data we would flush here.
There are places where flushing them might be useful, but the current
sync interface doesn't give us that opportunity.
Signed-off-by: Christoph Hellwig <[email protected]>
Reviewed-by: Dave Chinner <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
The calling convention that returns a pointer to a static buffer is
fairly nasty, so just opencode it in the only caller that is left.
Signed-off-by: Christoph Hellwig <[email protected]>
Reviewed-by: Dave Chinner <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
Use xfs_ioerror_alert instead of opencoding a very similar error
message.
Signed-off-by: Christoph Hellwig <[email protected]>
Reviewed-by: Dave Chinner <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
Instead of passing the block number and mount structure explicitly
get them off the bp and fix make the argument order more natural.
Also move it to xfs_buf.c and stop printing the device name given
that we already get the fs name as part of xfs_alert, and we know
what device is operates on because of the caller that gets printed,
finally rename it to xfs_buf_ioerror_alert and pass __func__ as
argument where it makes sense.
Signed-off-by: Christoph Hellwig <[email protected]>
Reviewed-by: Dave Chinner <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
Change _xfs_buf_initialize to allocate the buffer directly and rename it to
xfs_buf_alloc now that is the only buffer allocation routine. Also remove
the xfs_buf_deallocate wrapper around the kmem_zone_free calls for buffers.
Signed-off-by: Christoph Hellwig <[email protected]>
Reviewed-by: Dave Chinner <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
For each call to xfs_buf_stale we call xfs_buf_delwri_dequeue either
directly before or after it, or are guaranteed by the surrounding
conditionals that we are never called on delwri buffers. Simply
this situation by moving the call to xfs_buf_delwri_dequeue into
xfs_buf_stale.
Signed-off-by: Christoph Hellwig <[email protected]>
Reviewed-by: Dave Chinner <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
The code is unused and under a config option that doesn't exist, remove it.
Signed-off-by: Christoph Hellwig <[email protected]>
Reviewed-by: Dave Chinner <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
The code to flush buffers in the umount code is a bit iffy: we first
flush all delwri buffers out, but then might be able to queue up a
new one when logging the sb counts. On a normal shutdown that one
would get flushed out when doing the synchronous superblock write in
xfs_unmountfs_writesb, but we skip that one if the filesystem has
been shut down.
Fix this by moving the delwri list flushing until just before unmounting
the log, and while we're at it also remove the superflous delwri list
and buffer lru flusing for the rt and log device that can never have
cached or delwri buffers.
Signed-off-by: Christoph Hellwig <[email protected]>
Reported-by: Amit Sahrawat <[email protected]>
Tested-by: Amit Sahrawat <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
Directories are only updated transactionally, which means fsync only
needs to flush the log the inode is currently dirty, but not bother
with checking for dirty data, non-transactional updates, and most
importanly doesn't have to flush disk caches except as part of a
transaction commit.
While the first two optimizations can't easily be measured, the
latter actually makes a difference when doing lots of fsync that do
not actually have to commit the inode, e.g. because an earlier fsync
already pushed the log far enough.
The new xfs_dir_fsync is identical to xfs_nfs_commit_metadata except
for the prototype, but I'm not sure creating a common helper for the
two is worth it given how simple the functions are.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
There is no reason to keep a reference to the inode even if we unlock
it during transaction commit because we never drop a reference between
the ijoin and commit. Also use this fact to merge xfs_trans_ijoin_ref
back into xfs_trans_ijoin - the third argument decides if an unlock
is needed now.
I'm actually starting to wonder if allowing inodes to be unlocked
at transaction commit really is worth the effort. The only real
benefit is that they can be unlocked earlier when commiting a
synchronous transactions, but that could be solved by doing the
log force manually after the unlock, too.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
Let the transaction commit unlock the inode before it potentially causes
a synchronous log force.
Signed-off-by: Christoph Hellwig <[email protected]>
Reviewed-by: Dave Chinner <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
Only read the LSN we need to push to with the ilock held, and then release
it before we do the log force to improve concurrency.
Signed-off-by: Christoph Hellwig <[email protected]>
Reviewed-by: Dave Chinner <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
Only read the LSN we need to push to with the ilock held, and then release
it before we do the log force to improve concurrency.
This also removes the only direct caller of _xfs_trans_commit, thus
allowing it to be merged into the plain xfs_trans_commit again.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
XFS_TRANS_SWAPEXT is a transaction type, not a flag for xfs_trans_commit, so
don't pass it in xfs_swap_extents.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
Convert all function prototypes to the short form used elsewhere,
and remove duplicates of comments already placed at the function
body.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
Fix a case in xfs_bmap_add_extent_unwritten_real where we aren't
passing the returned error on.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
All the parameters passed to xfs_bmap_add_extent_hole_real() are in
the xfs_bmalloca structure now. Just pass the bmalloca parameter to
the function instead of 8 separate parameters.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
All the parameters passed to xfs_bmap_add_extent_delay_real() are in
the xfs_bmalloca structure now. Just pass the bmalloca parameter to
the function instead of 8 separate parameters.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
There is no real need to the xfs_bmap_add_extent, as the callers
know what kind of extents they need to it. Removing it means
duplicating the extents to btree conversion logic in three places,
but overall it's still much simpler code and quite a bit less code.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
Add a common helper for finding the last extent in a file.
Largely based on a patch from Dave Chinner.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
We can just call xfs_bmap_add_extent_hole_delay directly to add a
delayed allocated regions to the extent tree, instead of going
through all the complexities of xfs_bmap_add_extent that aren't
needed for this simple case.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
Delalloc reservations are much simpler than allocations, so give
them a separate bmapi-level interface. Using the previously added
xfs_bmapi_reserve_delalloc we get a function that is only minimally
more complicated than xfs_bmapi_read, which is far from the complexity
in xfs_bmapi. Also remove the XFS_BMAPI_DELAY code after switching
over the only user to xfs_bmapi_delay.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
Move the reservation of delayed allocations, and addition of delalloc
regions to the extent trees into a new helper function. For now
this adds some twisted goto logic to xfs_bmapi, but that will be
cleaned up in the following patches.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
Instead of using a local variable that needs to updated when we modify
the extent map just check ifp->if_bytes directly where we use it.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
We already have the worst case blocks reserved, so xfs_icsb_modify_counters
won't fail in xfs_bmap_add_extent_delay_real. In fact we've had an assert
to catch this case since day and it never triggered. So remove the code
to try smaller reservations, and just return the error for that case in
addition to keeping the assert.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
Both xfs_bmap_add_extent_hole_delay and xfs_bmap_add_extent_hole_real
already contain code to handle the case where there is no extent to
merge with, which is effectively the same as the code duplicated here.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
Return unwritten extent conversion errors to aio_complete.
Skip both unwritten extent conversion and size updates if we had an
I/O error or the filesystem has been shut down.
Return -EIO to the aio/buffer completion handlers in case of a
forced shutdown.
Signed-off-by: Christoph Hellwig <[email protected]>
Reviewed-by: Dave Chinner <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
Currently a buffered reader or writer can add pages to the pagecache
while we are waiting for the iolock in xfs_file_dio_aio_write. Prevent
this by re-checking mapping->nrpages after we got the iolock, and if
nessecary upgrade the lock to exclusive mode. To simplify this a bit
only take the ilock inside of xfs_file_aio_write_checks.
Signed-off-by: Christoph Hellwig <[email protected]>
Reviewed-by: Dave Chinner <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
Currently xfs_attr_inactive causes a synchronous transactions if we are
removing a file that has any extents allocated to the attribute fork, and
thus makes XFS extremely slow at removing files with out of line extended
attributes. The code looks a like a relict from the days before the busy
extent list, but with the busy extent list we avoid reusing data and attr
extents that have been freed but not commited yet, so this code is just
as superflous as the synchronous transactions for data blocks.
Signed-off-by: Christoph Hellwig <[email protected]>
Reported-by: Bernd Schubert <[email protected]>
Reviewed-by: Dave Chinner <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
We now have an i_dio_count filed and surrounding infrastructure to wait
for direct I/O completion instead of i_icount, and we have never needed
to iocount waits for buffered I/O given that we only set the page uptodate
after finishing all required work. Thus remove i_iocount, and replace
the actually needed waits with calls to inode_dio_wait.
Signed-off-by: Christoph Hellwig <[email protected]>
Reviewed-by: Dave Chinner <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
The current code relies on the xfs_ioend_wait call later on to make sure
all I/O actually has completed. The xfs_ioend_wait call will go away soon,
so prepare for that by using the waiting filemap function.
Signed-off-by: Christoph Hellwig <[email protected]>
Reviewed-by: Dave Chinner <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
There is no reason to queue up ioends for processing in user context
unless we actually need it. Just complete ioends that do not convert
unwritten extents or need a size update from the end_io context.
Signed-off-by: Christoph Hellwig <[email protected]>
Reviewed-by: Dave Chinner <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
We really shouldn't complete AIO or DIO requests until we have finished
the unwritten extent conversion and size update. This means fsync never
has to pick up any ioends as all work has been completed when signalling
I/O completion.
Signed-off-by: Christoph Hellwig <[email protected]>
Reviewed-by: Dave Chinner <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
No driver returns ENODEV from it bio completion handler, not has this
ever been documented. Remove the dead code dealing with it.
Signed-off-by: Christoph Hellwig <[email protected]>
Reviewed-by: Dave Chinner <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
And also remove the strange local lock and delwri list pointers in a few
functions.
Signed-off-by: Christoph Hellwig <[email protected]>
Reviewed-by: Dave Chinner <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
Remove the xfs_buf_relse from xfs_bwrite and let the caller handle it to
mirror the delwri and read paths.
Also remove the mount pointer passed to xfs_bwrite, which is superflous now
that we have a mount pointer in the buftarg.
Signed-off-by: Christoph Hellwig <[email protected]>
Reviewed-by: Dave Chinner <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
Unify the ways we add buffers to the delwri queue by always calling
xfs_buf_delwri_queue directly. The xfs_bdwrite functions is removed and
opencoded in its callers, and the two places setting XBF_DELWRI while a
buffer is locked and expecting xfs_buf_unlock to pick it up are converted
to call xfs_buf_delwri_queue directly, too. Also replace the
XFS_BUF_UNDELAYWRITE macro with direct calls to xfs_buf_delwri_dequeue
to make the explicit queuing/dequeuing more obvious.
Signed-off-by: Christoph Hellwig <[email protected]>
Reviewed-by: Dave Chinner <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
Do not transfer a reference held by the caller to the buffer on the list,
or decrement it in xfs_buf_delwri_queue, but instead grab a new reference
if needed, and let the caller drop its own reference. Also move setting
of the XBF_DELWRI and XBF_ASYNC flags into xfs_buf_delwri_queue, and
only do it if needed. Note that for now xfs_buf_unlock already has
XBF_DELWRI, but that will change in the following patches.
Signed-off-by: Christoph Hellwig <[email protected]>
Reviewed-by: Dave Chinner <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
We can just unlock the buffer in the caller, and the decrement of b_hold
would also be needed in the !unlock, we just never hit that case currently
given that the caller handles that case.
Signed-off-by: Christoph Hellwig <[email protected]>
Reviewed-by: Dave Chinner <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
We cannot ever reach xfs_buf_iorequest for a buffer with XBF_DELWRI set,
given that all write handlers make sure that the buffer is remove from
the delwri queue before, and we never do reads with the XBF_DELWRI flag
set (which the code would not handle correctly anyway).
Signed-off-by: Christoph Hellwig <[email protected]>
Reviewed-by: Dave Chinner <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
Currently we have a few issues with the way the workqueue code is used to
implement AIL pushing:
- it accidentally uses the same workqueue as the syncer action, and thus
can be prevented from running if there are enough sync actions active
in the system.
- it doesn't use the HIGHPRI flag to queue at the head of the queue of
work items
At this point I'm not confident enough in getting all the workqueue flags and
tweaks right to provide a perfectly reliable execution context for AIL
pushing, which is the most important piece in XFS to make forward progress
when the log fills.
Revert back to use a kthread per filesystem which fixes all the above issues
at the cost of having a task struct and stack around for each mounted
filesystem. In addition this also gives us much better ways to diagnose
any issues involving hung AIL pushing and removes a small amount of code.
Signed-off-by: Christoph Hellwig <[email protected]>
Reported-by: Stefan Priebe <[email protected]>
Tested-by: Stefan Priebe <[email protected]>
Reviewed-by: Dave Chinner <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
We need to check for pinned buffers even in .iop_pushbuf given that inode
items flush into the same buffers that may be pinned directly due operations
on the unlinked inode list operating directly on buffers. To do this add a
return value to .iop_pushbuf that tells the AIL push about this and use
the existing log force mechanisms to unpin it.
Signed-off-by: Christoph Hellwig <[email protected]>
Reported-by: Stefan Priebe <[email protected]>
Tested-by: Stefan Priebe <[email protected]>
Reviewed-by: Dave Chinner <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
If an item was locked we should not update xa_last_pushed_lsn and thus skip
it when restarting the AIL scan as we need to be able to lock and write it
out as soon as possible. Otherwise heavy lock contention might starve AIL
pushing too easily, especially given the larger backoff once we moved
xa_last_pushed_lsn all the way to the target lsn.
Signed-off-by: Christoph Hellwig <[email protected]>
Reported-by: Stefan Priebe <[email protected]>
Tested-by: Stefan Priebe <[email protected]>
Reviewed-by: Dave Chinner <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
Move all the checks performed on a bio into a new helper, and call it as
soon as bio is submitted even if it is a re-submission from ->make_request.
We explicitly mark the new helper as beeing non-inlined as the stack
usage for printing the block device name in the failure case is quite
high and this a patch where we have to be extremely conservative about
stack usage.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Jens Axboe <[email protected]>
There is a window in which the ioend that we call inode_dio_wake on
in xfs_end_io_direct_write is already free. Fix this by storing
the inode pointer in a local variable.
This is a fix for the regression introduced in 3.1-rc by
"fs: move inode_dio_done to the end_io handler".
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
There is very little benefit in allowing to let a ->make_request
instance update the bios device and sector and loop around it in
__generic_make_request when we can archive the same through calling
generic_make_request from the driver and letting the loop in
generic_make_request handle it.
Note that various drivers got the return value from ->make_request and
returned non-zero values for errors.
Signed-off-by: Christoph Hellwig <[email protected]>
Acked-by: NeilBrown <[email protected]>
Signed-off-by: Jens Axboe <[email protected]>
Avoid the hacks need for request based device mappers currently by simply
exporting the symbol instead of trying to get it through the back door.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Jens Axboe <[email protected]>
Currently we always redirty an inode that was attempted to be written out
synchronously but has been cleaned by an AIL pushed internall, which is
rather bogus. Fix that by doing the i_update_core check early on and
return 0 for it. Also include async calls for it, as doing any work for
those is just as pointless. While we're at it also fix the sign for the
EIO return in case of a filesystem shutdown, and fix the completely
non-sensical locking around xfs_log_inode.
Signed-off-by: Christoph Hellwig <[email protected]>
Reviewed-by: Dave Chinner <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
(cherry picked from commit 297db93bb74cf687510313eb235a7aec14d67e97)
Signed-off-by: Alex Elder <[email protected]>
During umount we do not add a dirty inode to the lru and wait for it to
become clean first, but force writeback of data and metadata with
I_WILL_FREE set. Currently there is no way for XFS to detect that the
inode has been redirtied for metadata operations, as we skip the
mark_inode_dirty call during teardown. Fix this by setting i_update_core
nanually in that case, so that the inode gets flushed during inode reclaim.
Alternatively we could enable calling mark_inode_dirty for inodes in
I_WILL_FREE state, and let the VFS dirty tracking handle this. I decided
against this as we will get better I/O patterns from reclaim compared to
the synchronous writeout in write_inode_now, and always marking the inode
dirty in some way from xfs_mark_inode_dirty is a better safetly net in
either case.
Signed-off-by: Christoph Hellwig <[email protected]>
Reviewed-by: Dave Chinner <[email protected]>
Signed-off-by: Alex Elder <[email protected]>
(cherry picked from commit da6742a5a4cc844a9982fdd936ddb537c0747856)
Signed-off-by: Alex Elder <[email protected]>
->queuecommand must return either 0, or one of the SCSI_MLQUEUE_* return
values. Non-transient errors are indicated by setting cmd->result before
calling ->scsi_done and returning 0. Fix libsas to adhere to this calling
convention. Note that the DID_ERROR for returns from the low-level driver
might not be correct for all cases, but it's the best we can do with
the current layering in libsas. I also suspect that the pre-existing
handling of -SAS_QUEUE_FULL should really be SCSI_MLQUEUE_HOST_BUSY, but
I'll leave that for a separate change.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: James Bottomley <[email protected]>
Switch sas_queuecommand to a normal indentation and goto based error handling.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: James Bottomley <[email protected]>
Add a new REQ_PRIO to let requests preempt others in the cfq I/O schedule,
and lave REQ_META purely for marking requests as metadata in blktrace.
All existing callers of REQ_META except for XFS are updated to also
set REQ_PRIO for now.
Signed-off-by: Christoph Hellwig <[email protected]>
Reviewed-by: Namhyung Kim <[email protected]>
Signed-off-by: Jens Axboe <[email protected]>
Replace all occurnanced of the undocumented READ_META with READ | REQ_META
and remove the unused WRITE_META define.
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Jens Axboe <[email protected]>
Christoph Hellwig