- 09 Feb, 2021 16 commits
-
-
Naohiro Aota authored
Factor out adding a page to a bio from submit_extent_page(). The page is added only when bio_flags are the same, contiguous and the added page fits in the same stripe as pages in the bio. Condition checks are reordered to allow early return to avoid possibly heavy btrfs_bio_fits_in_stripe() calling. Reviewed-by:
Josef Bacik <josef@toxicpanda.com> Signed-off-by:
Naohiro Aota <naohiro.aota@wdc.com> Reviewed-by:
David Sterba <dsterba@suse.com> Signed-off-by:
-
Naohiro Aota authored
We must reset the zones of a deleted unused block group to rewind the zones' write pointers to the zones' start. To do this, we can use the DISCARD_SYNC code to do the reset when the filesystem is running on zoned devices. Reviewed-by:
Anand Jain <anand.jain@oracle.com> Signed-off-by:
-
Naohiro Aota authored
Since the allocation info of a tree log node is not recorded in the extent tree, calculate_alloc_pointer() cannot detect this node, so the pointer can be over a tree node. Replaying the log calls btrfs_remove_free_space() for each node in the log tree. So, advance the pointer after the node to not allocate over it. Reviewed-by:
-
Naohiro Aota authored
Tree manipulating operations like merging nodes often release once-allocated tree nodes. Such nodes are cleaned so that pages in the node are not uselessly written out. On zoned volumes, however, such optimization blocks the following IOs as the cancellation of the write out of the freed blocks breaks the sequential write sequence expected by the device. Introduce a list of clean and unwritten extent buffers that have been released in a transaction. Redirty the buffers so that btree_write_cache_pages() can send proper bios to the devices. Besides it clears the entire content of the extent buffer not to confuse raw block scanners e.g. 'btrfs check'. By clearing the content, csum_dirty_buffer() complains about bytenr mismatch, so avoid the checking and checksum using newly introduced buffer flag EXTENT_BUFFER_NO_CHECK. Reviewed-by:
-
Naohiro Aota authored
Implement a sequential extent allocator for zoned filesystems. This allocator only needs to check if there is enough space in the block group after the allocation pointer to satisfy the extent allocation request. Therefore the allocator never manages bitmaps or clusters. Also, add assertions to the corresponding functions. As zone append writing is used, it would be unnecessary to track the allocation offset, as the allocator only needs to check available space. But by tracking and returning the offset as an allocated region, we can skip modification of ordered extents and checksum information when there is no IO reordering. Reviewed-by:
-
Naohiro Aota authored
In a zoned filesystem a once written then freed region is not usable until the underlying zone has been reset. So we need to distinguish such unusable space from usable free space. Therefore we need to introduce the "zone_unusable" field to the block group structure, and "bytes_zone_unusable" to the space_info structure to track the unusable space. Pinned bytes are always reclaimed to the unusable space. But, when an allocated region is returned before using e.g., the block group becomes read-only between allocation time and reservation time, we can safely return the region to the block group. For the situation, this commit introduces "btrfs_add_free_space_unused". This behaves the same as btrfs_add_free_space() on regular filesystem. On zoned filesystems, it rewinds the allocation offset. Because the read-only bytes tracks free but unusable bytes when the block group is read-only, we need to migrate the zone_unusable bytes to read-only bytes when a block group is marked read-only. Reviewed-by:
-
Naohiro Aota authored
Conventional zones do not have a write pointer, so we cannot use it to determine the allocation offset for sequential allocation if a block group contains a conventional zone. But instead, we can consider the end of the highest addressed extent in the block group for the allocation offset. For new block group, we cannot calculate the allocation offset by consulting the extent tree, because it can cause deadlock by taking extent buffer lock after chunk mutex, which is already taken in btrfs_make_block_group(). Since it is a new block group anyways, we can simply set the allocation offset to 0. Reviewed-by:
-
Naohiro Aota authored
A zoned filesystem must allocate blocks at the zones' write pointer. The device's write pointer position can be mapped to a logical address within a block group. To facilitate this, add an "alloc_offset" to the block-group to track the logical addresses of the write pointer. This logical address is populated in btrfs_load_block_group_zone_info() from the write pointers of corresponding zones. For now, zoned filesystems the single profile. Supporting non-single profile with zone append writing is not trivial. For example, in the DUP profile, we send a zone append writing IO to two zones on a device. The device reply with written LBAs for the IOs. If the offsets of the returned addresses from the beginning of the zone are different, then it results in different logical addresses. We need fine-grained logical to physical mapping to support such separated physical address issue. Since it should require additional metadata type, disable non-single profiles for now. This commit supports the case all the zones in a block group are sequential. The next patch will handle the case having a conventional zone. Reviewed-by:
-
Naohiro Aota authored
Add a check in verify_one_dev_extent() to ensure that a device extent on a zoned block device is aligned to the respective zone boundary. If it isn't, mark the filesystem as unclean. Reviewed-by:
-
Naohiro Aota authored
Implement a zoned chunk and device extent allocator. One device zone becomes a device extent so that a zone reset affects only this device extent and does not change the state of blocks in the neighbor device extents. To implement the allocator, we need to extend the following functions for a zoned filesystem. - init_alloc_chunk_ctl - dev_extent_search_start - dev_extent_hole_check - decide_stripe_size init_alloc_chunk_ctl_zoned() is mostly the same as regular one. It always set the stripe_size to the zone size and aligns the parameters to the zone size. dev_extent_search_start() only aligns the start offset to zone boundaries. We don't care about the first 1MB like in regular filesystem because we anyway reserve the first two zones for superblock logging. dev_extent_hole_check_zoned() checks if zones in given hole are either conventional or empty sequential zones. Also, it skips zones reserved for superblock logging. With the change to the hole, the new hole may now contain pending extents. So, in this case, loop again to check that. Finally, decide_stripe_size_zoned() should shrink the number of devices instead of stripe size because we need to honor stripe_size == zone_size. Reviewed-by:
-
Johannes Thumshirn authored
Run a zoned filesystem on non-zoned devices. This is done by "slicing up" the block device into static sized chunks and fake a conventional zone on each of them. The emulated zone size is determined from the size of device extent. This is mainly aimed at testing of zoned filesystems, i.e. the zoned chunk allocator, on regular block devices. Reviewed-by:
Johannes Thumshirn <johannes.thumshirn@wdc.com> Signed-off-by:
-
Naohiro Aota authored
The implementation of fitrim depends on space cache, which is not used and disabled for zoned extent allocator. So the current code does not work with zoned filesystem. In the future, we can implement fitrim for zoned filesystems by enabling space cache (but, only for fitrim) or scanning the extent tree at fitrim time. For now, disallow fitrim on zoned filesystems. Reviewed-by:
-
Johannes Thumshirn authored
Don't set the zoned flag in fs_info as soon as we're encountering the incompat filesystem flag for a zoned filesystem on mount. The zoned flag in fs_info is in a union together with the zone_size, so setting it too early will result in setting an incorrect zone_size as well. Once the correct zone_size is read from the device, we can rely on the zoned flag in fs_info as well to determine if the filesystem is zoned. Reviewed-by:
-
Johannes Thumshirn authored
Since we have no write pointer in conventional zones, we cannot determine the allocation offset from it. Instead, we set the allocation offset after the highest addressed extent. This is done by reading the extent tree in btrfs_load_block_group_zone_info(). However, this function is called from btrfs_read_block_groups(), so the read lock for the tree node could be recursively taken. To avoid this unsafe locking scenario, release the path before reading the extent tree to get the allocation offset. Reviewed-by:
-
Naohiro Aota authored
A zoned filesystem currently has a superblock at the beginning of the superblock logging zones if the zones are conventional. This difference in superblock position causes a chicken-and-egg problem for filesystems with emulated zones. Since the device is a regular (non-zoned) device, we cannot know if the filesystem is regular or zoned while reading the superblock. But, to load the superblock, we need to see if it is emulated zoned or not. Place the superblocks at the same location as they are on regular filesystem on regular devices to solve the problem. It is possible because it's ensured that all the superblock locations are at an (emulated) conventional zone on regular devices. Reviewed-by:
-
Naohiro Aota authored
This is a preparation patch to implement zone emulation on a regular device. To emulate a zoned filesystem on a regular (non-zoned) device, we need to decide an emulated zone size. Instead of making it a compile-time static value, we'll make it configurable at mkfs time. Since we have one zone == one device extent restriction, we can determine the emulated zone size from the size of a device extent. We can extend btrfs_get_dev_zone_info() to show a regular device filled with conventional zones once the zone size is decided. The current call site of btrfs_get_dev_zone_info() during the mount process is earlier than loading the file system trees so that we don't know the size of a device extent at this point. Thus we can't slice a regular device to conventional zones. This patch introduces btrfs_get_dev_zone_info_all_devices to load the zone info for all the devices. And, it places this function in open_ctree() after loading the trees. Reviewed-by:
-
- 08 Feb, 2021 24 commits
-
-
Naohiro Aota authored
A ZONE_APPEND bio must follow hardware restrictions (e.g. not exceeding max_zone_append_sectors) not to be split. bio_iov_iter_get_pages builds such restricted bio using __bio_iov_append_get_pages if bio_op(bio) == REQ_OP_ZONE_APPEND. To utilize it, we need to set the bio_op before calling bio_iov_iter_get_pages(). This commit introduces IOMAP_F_ZONE_APPEND, so that iomap user can set the flag to indicate they want REQ_OP_ZONE_APPEND and restricted bio. Reviewed-by:
Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by:
Christoph Hellwig <hch@lst.de> Reviewed-by:
Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com> Signed-off-by:
-
Johannes Thumshirn authored
Add bio_add_zone_append_page(), a wrapper around bio_add_hw_page() which is intended to be used by file systems that directly add pages to a bio instead of using bio_iov_iter_get_pages(). Reviewed-by:
Jens Axboe <axboe@kernel.dk> Signed-off-by:
-
Filipe Manana authored
At btrfs_copy_root(), if the call to btrfs_inc_ref() fails we end up returning without unlocking and releasing our reference on the extent buffer named "cow" we previously allocated with btrfs_alloc_tree_block(). So fix that by unlocking the extent buffer and dropping our reference on it before returning. Fixes: be20aa9d ("Btrfs: Add mount option to turn off data cow") CC: stable@vger.kernel.org # 4.4+ Signed-off-by:
Filipe Manana <fdmanana@suse.com> Reviewed-by:
-
Qu Wenruo authored
In read_extent_buffer_pages(), if we failed to lock the page atomically, we just exit with return value 0. This is counter-intuitive, as normally if we can't lock what we need, we would return something like EAGAIN. But that return hides under (wait == WAIT_NONE) branch, which only gets triggered for readahead. And for readahead, if we failed to lock the page, it means the extent buffer is either being read by other thread, or has been read and is under modification. Either way the eb will or has been cached, thus readahead has no need to wait for it. Add comment on this counter-intuitive behavior. Reported-by:
Dan Carpenter <dan.carpenter@oracle.com> Reviewed-by:
Qu Wenruo <wqu@suse.com> Signed-off-by:
-
Qu Wenruo authored
This adds the basic RO mount ability for 4K sector size on 64K page system. Currently we only plan to support 4K and 64K page system. Reviewed-by:
-
Qu Wenruo authored
In btrfs data page read path, the page status update are handled in two different locations: btrfs_do_read_page() { while (cur <= end) { /* No need to read from disk */ if (HOLE/PREALLOC/INLINE){ memset(); set_extent_uptodate(); continue; } /* Read from disk */ ret = submit_extent_page(end_bio_extent_readpage); } end_bio_extent_readpage() { endio_readpage_uptodate_page_status(); } This is fine for sectorsize == PAGE_SIZE case, as for above loop we should only hit one branch and then exit. But for subpage, there is more work to be done in page status update: - Page Unlock condition Unlike regular page size == sectorsize case, we can no longer just unlock a page. Only the last reader of the page can unlock the page. This means, we can unlock the page either in the while() loop, or in the endio function. - Page uptodate condition Since we have multiple sectors to read for a page, we can only mark the full page uptodate if all sectors are uptodate. To handle both subpage and regular cases, introduce a pair of functions to help handling page status update: - begin_page_read() For regular case, it does nothing. For subpage case, it updates the reader counters so that later end_page_read() can know who is the last one to unlock the page. - end_page_read() This is just endio_readpage_uptodate_page_status() renamed. The original name is a little too long and too specific for endio. The new thing added is the condition for page unlock. Now for subpage data, we unlock the page if we're the last reader. This does not only provide the basis for subpage data read, but also hide the special handling of page read from the main read loop. Also, since we're changing how the page lock is handled, there are two existing error paths where we need to manually unlock the page before calling begin_page_read(). Signed-off-by: -
Qu Wenruo authored
To support subpage sector size, data also need extra info to make sure which sectors in a page are uptodate/dirty/... This patch will make pages for data inodes get btrfs_subpage structure attached, and detached when the page is freed. This patch also slightly changes the timing when set_page_extent_mapped() is called to make sure: - We have page->mapping set page->mapping->host is used to grab btrfs_fs_info, thus we can only call this function after page is mapped to an inode. One call site attaches pages to inode manually, thus we have to modify the timing of set_page_extent_mapped() a bit. - As soon as possible, before other operations Since memory allocation can fail, we have to do extra error handling. Calling set_page_extent_mapped() as soon as possible can simply the error handling for several call sites. The idea is pretty much the same as iomap_page, but with more bitmaps for btrfs specific cases. Currently the plan is to switch iomap if iomap can provide sector aligned write back (only write back dirty sectors, but not the full page, data balance require this feature). So we will stick to btrfs specific bitmap for now. Signed-off-by:
-
Qu Wenruo authored
For subpage metadata validation check, there are some differences: - Read must finish in one bvec Since we're just reading one subpage range in one page, it should never be split into two bios nor two bvecs. - How to grab the existing eb Instead of grabbing eb using page->private, we have to go search radix tree as we don't have any direct pointer at hand. Reviewed-by:
-
Qu Wenruo authored
To handle subpage status update, add the following: - Use btrfs_page_*() subpage-aware helpers to update page status Now we can handle both cases well. - No page unlock for subpage metadata Since subpage metadata doesn't utilize page locking at all, skip it. For subpage data locking, it's handled in later commits. Reviewed-by:
-
Qu Wenruo authored
Introduce a helper, read_extent_buffer_subpage(), to do the subpage extent buffer read. The difference between regular and subpage routines are: - No page locking Here we completely rely on extent locking. Page locking can reduce the concurrency greatly, as if we lock one page to read one extent buffer, all the other extent buffers in the same page will have to wait. - Extent uptodate condition Despite the existing PageUptodate() and EXTENT_BUFFER_UPTODATE check, We also need to check btrfs_subpage::uptodate_bitmap. - No page iteration Just one page, no need to loop, this greatly simplified the subpage routine. This patch only implements the bio submit part, no endio support yet. Reviewed-by:
-
Qu Wenruo authored
Unlike the original try_release_extent_buffer(), try_release_subpage_extent_buffer() will iterate through all the ebs in the page, and try to release each. We can release the full page only after there's no private attached, which means all ebs of that page have been released as well. Reviewed-by:
-
Qu Wenruo authored
For btrfs_clone_extent_buffer(), it's mostly the same code of __alloc_dummy_extent_buffer(), except it has extra page copy. So to make it subpage compatible, we only need to: - Call set_extent_buffer_uptodate() instead of SetPageUptodate() This will set correct uptodate bit for subpage and regular sector size cases. Since we're calling set_extent_buffer_uptodate() which will also set EXTENT_BUFFER_UPTODATE bit, we don't need to manually set that bit either. Reviewed-by:
-
Qu Wenruo authored
To support subpage in set_extent_buffer_uptodate and clear_extent_buffer_uptodate we only need to use the subpage-aware helpers to update the page bits. Reviewed-by:
-
Qu Wenruo authored
Introduce the following functions to handle subpage error status: - btrfs_subpage_set_error() - btrfs_subpage_clear_error() - btrfs_subpage_test_error() These helpers can only be called when the page has subpage attached and the range is ensured to be inside the page. - btrfs_page_set_error() - btrfs_page_clear_error() - btrfs_page_test_error() These helpers can handle both regular sector size and subpage without problem. Reviewed-by:
-
Qu Wenruo authored
Introduce the following functions to handle subpage uptodate status: - btrfs_subpage_set_uptodate() - btrfs_subpage_clear_uptodate() - btrfs_subpage_test_uptodate() These helpers can only be called when the page has subpage attached and the range is ensured to be inside the page. - btrfs_page_set_uptodate() - btrfs_page_clear_uptodate() - btrfs_page_test_uptodate() These helpers can handle both regular sector size and subpage. Although caller should still ensure that the range is inside the page. Reviewed-by:
-
Qu Wenruo authored
There are locations where we allocate dummy extent buffers for temporary usage, like in tree_mod_log_rewind() or get_old_root(). These dummy extent buffers will be handled by the same eb accessors, and if they don't have page::private subpage eb accessors could fail. To address such problems, make __alloc_dummy_extent_buffer() attach page private for dummy extent buffers too. Reviewed-by:
-
Qu Wenruo authored
In btrfs_release_extent_buffer_pages(), we need to add extra handling for subpage. Introduce a helper, detach_extent_buffer_page(), to do different handling for regular and subpage cases. For subpage case, handle detaching page private. For unmapped (dummy or cloned) ebs, we can detach the page private immediately as the page can only be attached to one unmapped eb. For mapped ebs, we have to ensure there are no eb in the page range before we delete it, as page->private is shared between all ebs in the same page. But there is a subpage specific race, where we can race with extent buffer allocation, and clear the page private while new eb is still being utilized, like this: Extent buffer A is the new extent buffer which will be allocated, while extent buffer B is the last existing extent buffer of the page. T1 (eb A) | T2 (eb B) -------------------------------+------------------------------ alloc_extent_buffer() | btrfs_release_extent_buffer_pages() |- p = find_or_create_page() | | |- attach_extent_buffer_page() | | | | |- detach_extent_buffer_page() | | |- if (!page_range_has_eb()) | | | No new eb in the page range yet | | | As new eb A hasn't yet been | | | inserted into radix tree. | | |- btrfs_detach_subpage() | | |- detach_page_private(); |- radix_tree_insert() | Then we have a metadata eb whose page has no private bit. To avoid such race, we introduce a subpage metadata-specific member, btrfs_subpage::eb_refs. In alloc_extent_buffer() we increase eb_refs in the critical section of private_lock. Then page_range_has_eb() will return true for detach_extent_buffer_page(), and will not detach page private. The section is marked by: - btrfs_page_inc_eb_refs() - btrfs_page_dec_eb_refs() Signed-off-by:
-
Qu Wenruo authored
For subpage case, grab_extent_buffer() can't really get an extent buffer just from btrfs_subpage. We have radix tree lock protecting us from inserting the same eb into the tree. Thus we don't really need to do the extra hassle, just let alloc_extent_buffer() handle the existing eb in radix tree. Now if two ebs are being allocated as the same time, one will fail with -EEIXST when inserting into the radix tree. So for grab_extent_buffer(), just always return NULL for subpage case. Reviewed-by:
-
Qu Wenruo authored
For subpage case, we need to allocate additional memory for each metadata page. So we need to: - Allow attach_extent_buffer_page() to return int to indicate allocation failure - Allow manually pre-allocate subpage memory for alloc_extent_buffer() As we don't want to use GFP_ATOMIC under spinlock, we introduce btrfs_alloc_subpage() and btrfs_free_subpage() functions for this purpose. (The simple wrap for btrfs_free_subpage() is for later convert to kmem_cache. Already internally tested without problem) - Preallocate btrfs_subpage structure for alloc_extent_buffer() We don't want to call memory allocation with spinlock held, so do preallocation before we acquire mapping->private_lock. - Handle subpage and regular case differently in attach_extent_buffer_page() For regular case, no change, just do the usual thing. For subpage case, allocate new memory or use the preallocated memory. For future subpage metadata, we will make use of radix tree to grab extent buffer. Reviewed-by:
-
Qu Wenruo authored
For sectorsize < page size support, we need a structure to record extra status info for each sector of a page. Introduce the skeleton structure, all subpage related code would go to subpage.[ch]. Reviewed-by:
-
Qu Wenruo authored
For the incoming subpage support, UNMAPPED extent buffer will have different behavior in btrfs_release_extent_buffer(). This means we need to set UNMAPPED bit early before calling btrfs_release_extent_buffer(). Currently there is only one caller which relies on btrfs_release_extent_buffer() in its error path while set UNMAPPED bit late: - btrfs_clone_extent_buffer() Make it subpage compatible by setting the UNMAPPED bit early, since we're here, also move the UPTODATE bit early. There is another caller, __alloc_dummy_extent_buffer(), setting UNMAPPED bit late, but that function clean up the allocated page manually, thus no need for any modification. Reviewed-by:
-
Qu Wenruo authored
PAGE_CLEAR_DIRTY and PAGE_SET_WRITEBACK are two defines used in __process_pages_contig(), to let the function know to clear page dirty bit and then set page writeback. However page writeback and dirty bits are conflicting (at least for sector size == PAGE_SIZE case), this means these two have to be always updated together. This means we can merge PAGE_CLEAR_DIRTY and PAGE_SET_WRITEBACK to PAGE_START_WRITEBACK. Reviewed-by:
-
Filipe Manana authored
Often an fsync needs to fallback to a transaction commit for several reasons (to ensure consistency after a power failure, a new block group was allocated or a temporary error such as ENOMEM or ENOSPC happened). In that case the log is marked as needing a full commit and any concurrent tasks attempting to log inodes or commit the log will also fallback to the transaction commit. When this happens they all wait for the task that first started the transaction commit to finish the transaction commit - however they wait until the full transaction commit happens, which is not needed, as they only need to wait for the superblocks to be persisted and not for unpinning all the extents pinned during the transaction's lifetime, which even for short lived transactions can be a few thousand and take some significant amount of time to complete - for dbench workloads I have observed up to 4~5 milliseconds of time spent unpinning extents in the worst cases, and the number of pinned extents was between 2 to 3 thousand. So allow fsync tasks to skip waiting for the unpinning of extents when they call btrfs_commit_transaction() and they were not the task that started the transaction commit (that one has to do it, the alternative would be to offload the transaction commit to another task so that it could avoid waiting for the extent unpinning or offload the extent unpinning to another task). This patch is part of a patchset comprised of the following patches: btrfs: remove unnecessary directory inode item update when deleting dir entry btrfs: stop setting nbytes when filling inode item for logging btrfs: avoid logging new ancestor inodes when logging new inode btrfs: skip logging directories already logged when logging all parents btrfs: skip logging inodes already logged when logging new entries btrfs: remove unnecessary check_parent_dirs_for_sync() btrfs: make concurrent fsyncs wait less when waiting for a transaction commit After applying the entire patchset, dbench shows improvements in respect to throughput and latency. The script used to measure it is the following: $ cat dbench-test.sh #!/bin/bash DEV=/dev/sdk MNT=/mnt/sdk MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="-m single -d single" echo "performance" | tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor umount $DEV &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT dbench -D $MNT -t 300 64 umount $MNT The test was run on a physical machine with 12 cores (Intel corei7), 64G of ram, using a NVMe device and a non-debug kernel configuration (Debian's default configuration). Before applying patchset, 32 clients: Operation Count AvgLat MaxLat ---------------------------------------- NTCreateX 9627107 0.153 61.938 Close 7072076 0.001 3.175 Rename 407633 1.222 44.439 Unlink 1943895 0.658 44.440 Deltree 256 17.339 110.891 Mkdir 128 0.003 0.009 Qpathinfo 8725406 0.064 17.850 Qfileinfo 1529516 0.001 2.188 Qfsinfo 1599884 0.002 1.457 Sfileinfo 784200 0.005 3.562 Find 3373513 0.411 30.312 WriteX 4802132 0.053 29.054 ReadX 15089959 0.002 5.801 LockX 31344 0.002 0.425 UnlockX 31344 0.001 0.173 Flush 674724 5.952 341.830 Throughput 1008.02 MB/sec 32 clients 32 procs max_latency=341.833 ms After applying patchset, 32 clients: After patchset, with 32 clients: Operation Count AvgLat MaxLat ---------------------------------------- NTCreateX 9931568 0.111 25.597 Close 7295730 0.001 2.171 Rename 420549 0.982 49.714 Unlink 2005366 0.497 39.015 Deltree 256 11.149 89.242 Mkdir 128 0.002 0.014 Qpathinfo 9001863 0.049 20.761 Qfileinfo 1577730 0.001 2.546 Qfsinfo 1650508 0.002 3.531 Sfileinfo 809031 0.005 5.846 Find 3480259 0.309 23.977 WriteX 4952505 0.043 41.283 ReadX 15568127 0.002 5.476 LockX 32338 0.002 0.978 UnlockX 32338 0.001 2.032 Flush 696017 7.485 228.835 Throughput 1049.91 MB/sec 32 clients 32 procs max_latency=228.847 ms --> +4.1% throughput, -39.6% max latency Before applying patchset, 64 clients: Operation Count AvgLat MaxLat ---------------------------------------- NTCreateX 8956748 0.342 108.312 Close 6579660 0.001 3.823 Rename 379209 2.396 81.897 Unlink 1808625 1.108 131.148 Deltree 256 25.632 172.176 Mkdir 128 0.003 0.018 Qpathinfo 8117615 0.131 55.916 Qfileinfo 1423495 0.001 2.635 Qfsinfo 1488496 0.002 5.412 Sfileinfo 729472 0.007 8.643 Find 3138598 0.855 78.321 WriteX 4470783 0.102 79.442 ReadX 14038139 0.002 7.578 LockX 29158 0.002 0.844 UnlockX 29158 0.001 0.567 Flush 627746 14.168 506.151 Throughput 924.738 MB/sec 64 clients 64 procs max_latency=506.154 ms After applying patchset, 64 clients: Operation Count AvgLat MaxLat ---------------------------------------- NTCreateX 9069003 0.303 43.193 Close 6662328 0.001 3.888 Rename 383976 2.194 46.418 Unlink 1831080 1.022 43.873 Deltree 256 24.037 155.763 Mkdir 128 0.002 0.005 Qpathinfo 8219173 0.137 30.233 Qfileinfo 1441203 0.001 3.204 Qfsinfo 1507092 0.002 4.055 Sfileinfo 738775 0.006 5.431 Find 3177874 0.936 38.170 WriteX 4526152 0.084 39.518 ReadX 14213562 0.002 24.760 LockX 29522 0.002 1.221 UnlockX 29522 0.001 0.694 Flush 635652 14.358 422.039 Throughput 990.13 MB/sec 64 clients 64 procs max_latency=422.043 ms --> +6.8% throughput, -18.1% max latency Reviewed-by:
-
Filipe Manana authored
Whenever we fsync an inode, if it is a directory, a regular file that was created in the current transaction or has last_unlink_trans set to the generation of the current transaction, we check if any of its ancestor inodes (and the inode itself if it is a directory) can not be logged and need a fallback to a full transaction commit - if so, we return with a value of 1 in order to fallback to a transaction commit. However we often do not need to fallback to a transaction commit because: 1) The ancestor inode is not an immediate parent, and therefore there is not an explicit request to log it and it is not needed neither to guarantee the consistency of the inode originally asked to be logged (fsynced) nor its immediate parent; 2) The ancestor inode was already logged before, in which case any link, unlink or rename operation updates the log as needed. So for these two cases we can avoid an unnecessary transaction commit. Therefore remove check_parent_dirs_for_sync() and add a check at the top of btrfs_log_inode() to make us fallback immediately to a transaction commit when we are logging a directory inode that can not be logged and needs a full transaction commit. All we need to protect is the case where after renaming a file someone fsyncs only the old directory, which would result is losing the renamed file after a log replay. This patch is part of a patchset comprised of the following patches: btrfs: remove unnecessary directory inode item update when deleting dir entry btrfs: stop setting nbytes when filling inode item for logging btrfs: avoid logging new ancestor inodes when logging new inode btrfs: skip logging directories already logged when logging all parents btrfs: skip logging inodes already logged when logging new entries btrfs: remove unnecessary check_parent_dirs_for_sync() btrfs: make concurrent fsyncs wait less when waiting for a transaction commit Performance results, after applying all patches, are mentioned in the change log of the last patch. Reviewed-by:
-
