- 11 May, 2022 5 commits
-
-
Allison Henderson authored
This is a clean up patch that merges xfs_delattr_context into xfs_attr_item. Now that the refactoring is complete and the delayed operation infrastructure is in place, we can combine these to eliminate the extra struct Signed-off-by:
Allison Henderson <allison.henderson@oracle.com> Reviewed-by:
Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by:
Chandan Babu R <chandanrlinux@gmail.com> Signed-off-by:
Dave Chinner <david@fromorbit.com>
-
Allison Henderson authored
This patch adds a debug option to enable log attribute replay. Eventually this can be removed when delayed attrs becomes permanent. Signed-off-by:
Darrick J. Wong <djwong@kernel.org> Signed-off-by:
-
Allison Henderson authored
This patch adds an error tag that we can use to test log attribute recovery and replay Signed-off-by:
-
Allison Henderson authored
Remove xfs_attr_set_args, xfs_attr_remove_args, and xfs_attr_trans_roll. These high level loops are now driven by the delayed operations code, and can be removed. Additionally collapse in the leaf_bp parameter of xfs_attr_set_iter since we only have one caller that passes dac->leaf_bp Signed-off-by:
-
Allison Henderson authored
These routines set up and queue a new deferred attribute operations. These functions are meant to be called by any routine needing to initiate a deferred attribute operation as opposed to the existing inline operations. New helper function xfs_attr_item_init also added. Finally enable delayed attributes in xfs_attr_set and xfs_attr_remove. Signed-off-by:
-
- 09 May, 2022 2 commits
-
-
Allison Henderson authored
This is a clean up patch that skips the flip flag logic for delayed attr renames. Since the log replay keeps the inode locked, we do not need to worry about race windows with attr lookups. So we can skip over flipping the flag and the extra transaction roll for it Signed-off-by:
-
Allison Henderson authored
This patch adds the needed routines to create, log and recover logged extended attribute intents. Signed-off-by:
-
- 04 May, 2022 22 commits
-
-
Allison Henderson authored
Currently attributes are modified directly across one or more transactions. But they are not logged or replayed in the event of an error. The goal of log attr replay is to enable logging and replaying of attribute operations using the existing delayed operations infrastructure. This will later enable the attributes to become part of larger multi part operations that also must first be recorded to the log. This is mostly of interest in the scheme of parent pointers which would need to maintain an attribute containing parent inode information any time an inode is moved, created, or removed. Parent pointers would then be of interest to any feature that would need to quickly derive an inode path from the mount point. Online scrub, nfs lookups and fs grow or shrink operations are all features that could take advantage of this. This patch adds two new log item types for setting or removing attributes as deferred operations. The xfs_attri_log_item will log an intent to set or remove an attribute. The corresponding xfs_attrd_log_item holds a reference to the xfs_attri_log_item and is freed once the transaction is done. Both log items use a generic xfs_attr_log_format structure that contains the attribute name, value, flags, inode, and an op_flag that indicates if the operations is a set or remove. [dchinner: added extra little bits needed for intent whiteouts] Signed-off-by:
-
Allison Henderson authored
During an attr rename operation, blocks are saved for later removal as rmtblkno2. The rmtblkno is used in the case of needing to alloc more blocks if not enough were available. However, in the case that no further blocks need to be added or removed, we can return as soon as xfs_attr_node_addname completes, rather than rolling the transaction with an -EAGAIN return. This extra loop does not hurt anything right now, but it will be a problem later when we get into log items because we end up with an empty log transaction. So, add a simple check to cut out the unneeded iteration. Signed-off-by:
-
Allison Henderson authored
The new deferred attr patch set uncovered a double unlock in the recent port of the defer ops capture and continue code. During log recovery, we're allowed to hold buffers to a transaction that's being used to replay an intent item. When we capture the resources as part of scheduling a continuation of an intent chain, we call xfs_buf_hold to retain our reference to the buffer beyond the transaction commit, but we do /not/ call xfs_trans_bhold to maintain the buffer lock. This means that xfs_defer_ops_continue needs to relock the buffers before xfs_defer_restore_resources joins then tothe new transaction. Additionally, the buffers should not be passed back via the dres structure since they need to remain locked unlike the inodes. So simply set dr_bufs to zero after populating the dres structure. Signed-off-by:
Chandan Babu R <chandan.babu@oracle.com> Signed-off-by:
-
Dave Chinner authored
-
Dave Chinner authored
Merge tag 'reflink-speedups-5.19_2022-04-28' of git://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux into xfs-5.19-for-next xfs: fix reflink inefficiencies As Dave Chinner has complained about on IRC, there are a couple of things about reflink that are very inefficient. First of all, we limited the size of all bunmapi operations to avoid flooding the log with defer ops in the worst case, but recent changes to the defer ops code have solved that problem, so get rid of the bunmapi length clamp. Second, the log reservations for reflink operations are far far larger than they need to be. Shrink them to exactly what we need to handle each deferred RUI and CUI log item, and no more. Also reduce logcount because we don't need 8 rolls per operation. Introduce a transaction reservation compatibility layer to avoid changing the minimum log size calculations. Signed-off-by:
-
Dave Chinner authored
Merge tag 'rmap-speedups-5.19_2022-04-28' of git://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux into xfs-5.19-for-next xfs: fix rmap inefficiencies Reduce the performance impact of the reverse mapping btree when reflink is enabled by using the much faster non-overlapped btree lookup functions when we're searching the rmap index with a fully specified key. If we find the exact record we're looking for, great! We don't have to perform the full overlapped scan. For filesystems with high sharing factors this reduces the xfs_scrub runtime by a good 15%%. This has been shown to reduce the fstests runtime for realtime rmap configurations by 30%%, since the lack of AGs severely limits scalability. Signed-off-by:
-
Dave Chinner authored
-
Dave Chinner authored
-
Dave Chinner authored
We don't check that the v4 feature flags taht v5 requires to be set are actually set anywhere. Do this check when we see that the filesystem is a v5 filesystem. Signed-off-by:
Dave Chinner <dchinner@redhat.com> Reviewed-by:
Christoph Hellwig <hch@lst.de> Reviewed-by:
-
Dave Chinner authored
While xfs_has_nlink() is not used in kernel, it is used in userspace (e.g. by xfs_db) so we need to set the XFS_FEAT_NLINK flag correctly in xfs_sb_version_to_features(). Signed-off-by:
-
Dave Chinner authored
xfs_repair catches fork size/format mismatches, but the in-kernel verifier doesn't, leading to null pointer failures when attempting to perform operations on the fork. This can occur in the xfs_dir_is_empty() where the in-memory fork format does not match the size and so the fork data pointer is accessed incorrectly. Note: this causes new failures in xfs/348 which is testing mode vs ftype mismatches. We now detect a regular file that has been changed to a directory or symlink mode as being corrupt because the data fork is for a symlink or directory should be in local form when there are only 3 bytes of data in the data fork. Hence the inode verify for the regular file now fires w/ -EFSCORRUPTED because the inode fork format does not match the format the corrupted mode says it should be in. Signed-off-by:
-
Dave Chinner authored
To catch the obvious graph cycle problem and hence potential endless looping. Signed-off-by:
-
Dave Chinner authored
When we log modifications based on intents, we add both intent and intent done items to the modification being made. These get written to the log to ensure that the operation is re-run if the intent done is not found in the log. However, for operations that complete wholly within a single checkpoint, the change in the checkpoint is atomic and will never need replay. In this case, we don't need to actually write the intent and intent done items to the journal because log recovery will never need to manually restart this modification. Log recovery currently handles intent/intent done matching by inserting the intent into the AIL, then removing it when a matching intent done item is found. Hence for all the intent-based operations that complete within a checkpoint, we spend all that time parsing the intent/intent done items just to cancel them and do nothing with them. Hence it follows that the only time we actually need intents in the log is when the modification crosses checkpoint boundaries in the log and so may only be partially complete in the journal. Hence if we commit and intent done item to the CIL and the intent item is in the same checkpoint, we don't actually have to write them to the journal because log recovery will always cancel the intents. We've never really worried about the overhead of logging intents unnecessarily like this because the intents we log are generally very much smaller than the change being made. e.g. freeing an extent involves modifying at lease two freespace btree blocks and the AGF, so the EFI/EFD overhead is only a small increase in space and processing time compared to the overall cost of freeing an extent. However, delayed attributes change this cost equation dramatically, especially for inline attributes. In the case of adding an inline attribute, we only log the inode core and attribute fork at present. With delayed attributes, we now log the attr intent which includes the name and value, the inode core adn attr fork, and finally the attr intent done item. We increase the number of items we log from 1 to 3, and the number of log vectors (regions) goes up from 3 to 7. Hence we tripple the number of objects that the CIL has to process, and more than double the number of log vectors that need to be written to the journal. At scale, this means delayed attributes cause a non-pipelined CIL to become CPU bound processing all the extra items, resulting in a > 40% performance degradation on 16-way file+xattr create worklaods. Pipelining the CIL (as per 5.15) reduces the performance degradation to 20%, but now the limitation is the rate at which the log items can be written to the iclogs and iclogs be dispatched for IO and completed. Even log IO completion is slowed down by these intents, because it now has to process 3x the number of items in the checkpoint. Processing completed intents is especially inefficient here, because we first insert the intent into the AIL, then remove it from the AIL when the intent done is processed. IOWs, we are also doing expensive operations in log IO completion we could completely avoid if we didn't log completed intent/intent done pairs. Enter log item whiteouts. When an intent done is committed, we can check to see if the associated intent is in the same checkpoint as we are currently committing the intent done to. If so, we can mark the intent log item with a whiteout and immediately free the intent done item rather than committing it to the CIL. We can basically skip the entire formatting and CIL insertion steps for the intent done item. However, we cannot remove the intent item from the CIL at this point because the unlocked per-cpu CIL item lists do not permit removal without holding the CIL context lock exclusively. Transaction commit only holds the context lock shared, hence the best we can do is mark the intent item with a whiteout so that the CIL push can release it rather than writing it to the log. This means we never write the intent to the log if the intent done has also been committed to the same checkpoint, but we'll always write the intent if the intent done has not been committed or has been committed to a different checkpoint. This will result in correct log recovery behaviour in all cases, without the overhead of logging unnecessary intents. This intent whiteout concept is generic - we can apply it to all intent/intent done pairs that have a direct 1:1 relationship. The way deferred ops iterate and relog intents mean that all intents currently have a 1:1 relationship with their done intent, and hence we can apply this cancellation to all existing intent/intent done implementations. For delayed attributes with a 16-way 64kB xattr create workload, whiteouts reduce the amount of journalled metadata from ~2.5GB/s down to ~600MB/s and improve the creation rate from 9000/s to 14000/s. Signed-off-by:
-
Dave Chinner authored
When we release an intent that a whiteout applies to, it will not have been committed to the journal and so won't be in the AIL. Hence when we drop the last reference to the intent, we do not want to try to remove it from the AIL as that will trigger a filesystem shutdown. Hence make the removal of intents from the AIL conditional on them actually being in the AIL so we do the correct thing. Signed-off-by:
-
Dave Chinner authored
To apply a whiteout to an intent item when an intent done item is committed, we need to be able to retrieve the intent item from the the intent done item. Add a log item op method for doing this, and wire all the intent done items up to it. Signed-off-by:
-
Dave Chinner authored
In preparation for adding support for intent item whiteouts. Signed-off-by:
-
Dave Chinner authored
Intent whiteouts will require extra work to be done during transaction commit if the transaction contains an intent done item. To determine if a transaction contains an intent done item, we want to avoid having to walk all the items in the transaction to check if they are intent done items. Hence when we add an intent done item to a transaction, tag the transaction to indicate that it contains such an item. We don't tag the transaction when the defer ops is relogging an intent to move it forward in the log. Whiteouts will never apply to these cases, so we don't need to bother looking for them. Signed-off-by:
-
Dave Chinner authored
We currently have a couple of helper functions that try to infer whether the log item is an intent or intent done item from the combinations of operations it supports. This is incredibly fragile and not very efficient as it requires checking specific combinations of ops. We need to be able to identify intent and intent done items quickly and easily in upcoming patches, so simply add intent and intent done type flags to the log item ops flags. These are static flags to begin with, so intent items should have been typed like this from the start. Signed-off-by:
-
Dave Chinner authored
If the first operation in a string of defer ops has no intents, then there is no reason to commit it before running the first call to xfs_defer_finish_one(). This allows the defer ops to be used effectively for non-intent based operations without requiring an unnecessary extra transaction commit when first called. This fixes a regression in per-attribute modification transaction count when delayed attributes are not being used. Signed-off-by:
-
Dave Chinner authored
Callers currently have to round out the size of buffers to match the aligment constraints of log iovecs and xlog_write(). They should not need to know this detail, so introduce a new function to calculate the iovec length (for use in ->iop_size implementations). Also modify xlog_finish_iovec() to round up the length to the correct alignment so the callers don't need to do this, either. Convert the only user - inode forks - of this alignment rounding to use the new interface. Signed-off-by:
-
Dave Chinner authored
Ever since we added shadown format buffers to the log items, log items need to handle the item being released with shadow buffers attached. Due to the fact this requirement was added at the same time we added new rmap/reflink intents, we missed the cleanup of those items. In theory, this means shadow buffers can be leaked in a very small window when a shutdown is initiated. Testing with KASAN shows this leak does not happen in practice - we haven't identified a single leak in several years of shutdown testing since ~v4.8 kernels. However, the intent whiteout cleanup mechanism results in every cancelled intent in exactly the same state as this tiny race window creates and so if intents down clean up shadow buffers on final release we will leak the shadow buffer for just about every intent we create. Hence we start with this patch to close this condition off and ensure that when whiteouts start to be used we don't leak lots of memory. Signed-off-by:
-
Dave Chinner authored
When we first allocate or resize an inline inode fork, we round up the allocation to 4 byte alingment to make journal alignment constraints. We don't clear the unused bytes, so we can copy up to three uninitialised bytes into the journal. Zero those bytes so we only ever copy zeros into the journal. Signed-off-by:
-
- 28 Apr, 2022 10 commits
-
-
Darrick J. Wong authored
These functions return the maximum number of blocks that could be logged in a particular transaction. "log count" is confusing since there's a separate concept of a log (operation) count in the reservation code, so let's change it to "block count" to be less confusing. Signed-off-by:
-
Darrick J. Wong authored
Currently, the code that performs CoW remapping after a write has this odd behavior where it walks /backwards/ through the data fork to remap extents in reverse order. Earlier, we rewrote the reflink remap function to use deferred bmap log items instead of trying to cram as much into the first transaction that we could. Now do the same for the CoW remap code. There doesn't seem to be any performance impact; we're just making better use of code that we added for the benefit of reflink. Signed-off-by:
-
Darrick J. Wong authored
Before to the introduction of deferred refcount operations, reflink would try to cram refcount btree updates into the same transaction as an allocation or a free event. Mainline XFS has never actually done that, but we never refactored the transaction reservations to reflect that we now do all refcount updates in separate transactions. Fix this to reduce the transaction reservation size even farther, so that between this patch and the previous one, we reduce the tr_write and tr_itruncate sizes by 66%. Signed-off-by:
-
Darrick J. Wong authored
Back in the early days of reflink and rmap development I set the transaction reservation sizes to be overly generous for rmap+reflink filesystems, and a little under-generous for rmap-only filesystems. Since we don't need *eight* transaction rolls to handle three new log intent items, decrease the logcounts to what we actually need, and amend the shadow reservation computation function to reflect what we used to do so that the minimum log size doesn't change. Signed-off-by:
-
Darrick J. Wong authored
Move the tracepoint that computes the size of the transaction used to compute the minimum log size into xfs_log_get_max_trans_res so that we only have to compute this stuff once. Leave xfs_log_get_max_trans_res as a non-static function so that xfs_db can call it to report the results of the userspace computation of the same value to diagnose mkfs/kernel misinteractions. Signed-off-by:
-
Darrick J. Wong authored
Every time someone changes the transaction reservation sizes, they introduce potential compatibility problems if the changes affect the minimum log size that we validate at mount time. If the minimum log size gets larger (which should be avoided because doing so presents a serious risk of log livelock), filesystems created with old mkfs will not mount on a newer kernel; if the minimum size shrinks, filesystems created with newer mkfs will not mount on older kernels. Therefore, enable the creation of a shadow log reservation structure where we can "undo" the effects of tweaks when computing minimum log sizes. These shadow reservations should never be used in practice, but they insulate us from perturbations in minimum log size. Signed-off-by:
-
Darrick J. Wong authored
This raw call isn't necessary since we can always remove a full delalloc extent. Signed-off-by:
-
Darrick J. Wong authored
In commit e1a4e37c, we clamped the length of bunmapi calls on the data forks of shared files to avoid two failure scenarios: one where the extent being unmapped is so sparsely shared that we exceed the transaction reservation with the sheer number of refcount btree updates and EFI intent items; and the other where we attach so many deferred updates to the transaction that we pin the log tail and later the log head meets the tail, causing the log to livelock. We avoid triggering the first problem by tracking the number of ops in the refcount btree cursor and forcing a requeue of the refcount intent item any time we think that we might be close to overflowing. This has been baked into XFS since before the original e1a4 patch. A recent patchset fixed the second problem by changing the deferred ops code to finish all the work items created by each round of trying to complete a refcount intent item, which eliminates the long chains of deferred items (27dad); and causing long-running transactions to relog their intent log items when space in the log gets low (74f4d). Because this clamp affects /any/ unmapping request regardless of the sharing factors of the component blocks, it degrades the performance of all large unmapping requests -- whereas with an unshared file we can unmap millions of blocks in one go, shared files are limited to unmapping a few thousand blocks at a time, which causes the upper level code to spin in a bunmapi loop even if it wasn't needed. This also eliminates one more place where log recovery behavior can differ from online behavior, because bunmapi operations no longer need to requeue. The fstest generic/447 was created to test the old fix, and it still passes with this applied. Partial-revert-of: e1a4e37c ("xfs: try to avoid blowing out the transaction reservation when bunmaping a shared extent") Depends: 27dada07 ("xfs: change the order in which child and parent defer ops ar finished") Depends: 74f4d6a1 ("xfs: only relog deferred intent items if free space in the log gets low") Signed-off-by:
-
Darrick J. Wong authored
A long time ago, I added to XFS the ability to use deferred reference count operations as part of a transaction chain. This enabled us to avoid blowing out the transaction reservation when the blocks in a physical extent all had different reference counts because we could ask the deferred operation manager for a continuation, which would get us a clean transaction. The refcount code asks for a continuation when the number of refcount record updates reaches the point where we think that the transaction has logged enough full btree blocks due to refcount (and free space) btree shape changes and refcount record updates that we're in danger of overflowing the transaction. We did not previously count the EFIs logged to the refcount update transaction because the clamps on the length of a bunmap operation were sufficient to avoid overflowing the transaction reservation even in the worst case situation where every other block of the unmapped extent is shared. Unfortunately, the restrictions on bunmap length avoid failure in the worst case by imposing a maximum unmap length of ~3000 blocks, even for non-pathological cases. This seriously limits performance when freeing large extents. Therefore, track EFIs with the same counter as refcount record updates, and use that information as input into when we should ask for a continuation. This enables the next patch to drop the clumsy bunmap limitation. Depends: 27dada07 ("xfs: change the order in which child and parent defer ops ar finished") Depends: 74f4d6a1 ("xfs: only relog deferred intent items if free space in the log gets low") Signed-off-by:
-
Darrick J. Wong authored
Reverse mapping on a reflink-capable filesystem has some pretty high overhead when performing file operations. This is because the rmap records for logically and physically adjacent extents might not be adjacent in the rmap index due to data block sharing. As a result, we use expensive overlapped-interval btree search, which walks every record that overlaps with the supplied key in the hopes of finding the record. However, profiling data shows that when the index contains a record that is an exact match for a query key, the non-overlapped btree search function can find the record much faster than the overlapped version. Try the non-overlapped lookup first when we're trying to find the left neighbor rmap record for a given file mapping, which makes unwritten extent conversion and remap operations run faster if data block sharing is minimal in this part of the filesystem. Signed-off-by:
-
- 27 Apr, 2022 1 commit
-
-
Darrick J. Wong authored
Reverse mapping on a reflink-capable filesystem has some pretty high overhead when performing file operations. This is because the rmap records for logically and physically adjacent extents might not be adjacent in the rmap index due to data block sharing. As a result, we use expensive overlapped-interval btree search, which walks every record that overlaps with the supplied key in the hopes of finding the record. However, profiling data shows that when the index contains a record that is an exact match for a query key, the non-overlapped btree search function can find the record much faster than the overlapped version. Try the non-overlapped lookup first, which will make scrub run much faster. Signed-off-by:
-
