- 09 Dec, 2020 32 commits
-
-
Boris Burkov authored
If we are not using space cache v1, we should not create the free space object or free space inodes. This comes up when we delete the existing free space objects/inodes when migrating to v2, only to see them get recreated for every dirtied block group. Reviewed-by:
Josef Bacik <josef@toxicpanda.com> Signed-off-by:
Boris Burkov <boris@bur.io> Reviewed-by:
David Sterba <dsterba@suse.com> Signed-off-by:
-
Boris Burkov authored
When the filesystem transitions from space cache v1 to v2 or to nospace_cache, it removes the old cached data, but does not remove the FREE_SPACE items nor the free space inodes they point to. This doesn't cause any issues besides being a bit inefficient, since these items no longer do anything useful. To fix it, when we are mounting, and plan to disable the space cache, destroy each block group's free space item and free space inode. The code to remove the items is lifted from the existing use case of removing the block group, with a light adaptation to handle whether or not we have already looked up the free space inode. Reviewed-by:
-
Boris Burkov authored
If the remount is ro->ro, rw->ro, or rw->rw, we will not create or clear the free space tree. This can be surprising, so print a warning to dmesg to make the failure more visible. It is also important to ensure that the space cache options (SPACE_CACHE, FREE_SPACE_TREE) are consistent, so ensure those are set to properly match the current on disk state (which won't be changing). Reviewed-by:
-
Boris Burkov authored
To make the contents of /proc/mounts better match the actual state of the filesystem, base the display of the space cache mount options off the contents of the super block rather than the last mount options passed in. Since there are many scenarios where the mount will ignore a space cache option, simply showing the passed in option is misleading. For example, if we mount with -o remount,space_cache=v2 on a read-write file system without an existing free space tree, we won't build a free space tree, but /proc/mounts will read space_cache=v2 (until we mount again and it goes away) cache_generation is set iff space_cache=v1, FREE_SPACE_TREE is set iff space_cache=v2, and if neither is the case, we print nospace_cache. Reviewed-by:
-
Boris Burkov authored
When mounting, btrfs uses the cache_generation in the super block to determine if space cache v1 is in use. However, by mounting with nospace_cache or space_cache=v2, it is possible to disable space cache v1, which does not result in un-setting cache_generation back to 0. In order to base some logic, like mount option printing in /proc/mounts, on the current state of the space cache rather than just the values of the mount option, keep the value of cache_generation consistent with the status of space cache v1. We ensure that cache_generation > 0 iff the file system is using space_cache v1. This requires committing a transaction on any mount which changes whether we are using v1. (v1->nospace_cache, v1->v2, nospace_cache->v1, v2->v1). Since the mechanism for writing out the cache generation is transaction commit, but we want some finer grained control over when we un-set it, we can't just rely on the SPACE_CACHE mount option, and introduce an fs_info flag that mount can use when it wants to unset the generation. Reviewed-by:
-
Boris Burkov authored
A user might want to revert to v1 or nospace_cache on a root filesystem, and much like turning on the free space tree, that can only be done remounting from ro->rw. Support clearing the free space tree on such mounts by moving it into the shared remount logic. Since the CLEAR_CACHE option sticks around across remounts, this change would result in clearing the tree for ever on every remount, which is not desirable. To fix that, add CLEAR_CACHE to the oneshot options we clear at mount end, which has the other bonus of not cluttering the /proc/mounts output with clear_cache. Reviewed-by:
-
Boris Burkov authored
Some options only apply during mount time and are cleared at the end of mount. For now, the example is USEBACKUPROOT, but CLEAR_CACHE also fits the bill, and this is a preparation patch for also clearing that option. One subtlety is that the current code only resets USEBACKUPROOT on rw mounts, but the option is meaningfully "consumed" by a ro mount, so it feels appropriate to clear in that case as well. A subsequent read-write remount would not go through open_ctree, which is the only place that checks the option, so the change should be benign. Reviewed-by:
-
Boris Burkov authored
When a user attempts to remount a btrfs filesystem with 'mount -o remount,space_cache=v2', that operation silently succeeds. Unfortunately, this is misleading, because the remount does not create the free space tree. /proc/mounts will incorrectly show space_cache=v2, but on the next mount, the file system will revert to the old space_cache. For now, we handle only the easier case, where the existing mount is read-only and the new mount is read-write. In that case, we can create the free space tree without contending with the block groups changing as we go. Reviewed-by:
-
Boris Burkov authored
If we attempt to create a free space tree while any block groups have needs_free_space set, we will double add the new free space item and hit EEXIST. Previously, we only created the free space tree on a new mount, so we never hit the case, but if we try to create it on a remount, such block groups could exist and trip us up. We don't do anything with this field unless the free space tree is enabled, so there is no harm in not setting it. Reviewed-by:
-
Boris Burkov authored
When we mount a rw filesystem, we start the orphan cleanup process in tree root and filesystem tree. However, when we remount a ro file system rw, we only clean the former. Move the calls to btrfs_orphan_cleanup() on tree_root and fs_root to the shared rw mount routine to effectively add them on ro->rw remount. Reviewed-by:
-
Boris Burkov authored
Mounting rw and remounting from ro to rw naturally share invariants and functionality which result in a correctly setup rw filesystem. Luckily, there is even a strong unity in the code which implements them. In mount's open_ctree, these operations mostly happen after an early return for ro file systems, and in remount, they happen in a section devoted to remounting ro->rw, after some remount specific validation passes. However, there are unfortunately a few differences. There are small deviations in the order of some of the operations, remount does not start orphan cleanup in root_tree or fs_tree, remount does not create the free space tree, and remount does not handle "one-shot" mount options like clear_cache and uuid tree rescan. Since we want to add building the free space tree to remount, and also to start the same orphan cleanup process on a filesystem mounted as ro then remounted rw, we would benefit from unifying the logic between the two code paths. This patch only lifts the existing common functionality, and leaves a natural path for fixing the discrepancies. Reviewed-by:
-
Filipe Manana authored
Early on during a transaction commit we acquire the tree_log_mutex and hold it until after we write the super blocks. But before writing the extent buffers dirtied by the transaction and the super blocks we unblock the transaction by setting its state to TRANS_STATE_UNBLOCKED and setting fs_info->running_transaction to NULL. This means that after that and before writing the super blocks, new transactions can start. However if any transaction wants to log an inode, it will block waiting for the transaction commit to write its dirty extent buffers and the super blocks because the tree_log_mutex is only released after those operations are complete, and starting a new log transaction blocks on that mutex (at start_log_trans()). Writing the dirty extent buffers and the super blocks can take a very significant amount of time to complete, but we could allow the tasks wanting to log an inode to proceed with most of their steps: 1) create the log trees 2) log metadata in the trees 3) write their dirty extent buffers They only need to wait for the previous transaction commit to complete (write its super blocks) before they attempt to write their super blocks, otherwise we could end up with a corrupt filesystem after a crash. So change start_log_trans() to use the root tree's log_mutex to serialize for the creation of the log root tree instead of using the tree_log_mutex, and make btrfs_sync_log() acquire the tree_log_mutex before writing the super blocks. This allows for inode logging to wait much less time when there is a previous transaction that is still committing, often not having to wait at all, as by the time when we try to sync the log the previous transaction already wrote its super blocks. This patch belongs to a patch set that is comprised of the following patches: btrfs: fix race causing unnecessary inode logging during link and rename btrfs: fix race that results in logging old extents during a fast fsync btrfs: fix race that causes unnecessary logging of ancestor inodes btrfs: fix race that makes inode logging fallback to transaction commit btrfs: fix race leading to unnecessary transaction commit when logging inode btrfs: do not block inode logging for so long during transaction commit The following script that uses dbench was used to measure the impact of the whole patchset: $ cat test-dbench.sh #!/bin/bash DEV=/dev/nvme0n1 MNT=/mnt/btrfs MOUNT_OPTIONS="-o ssd" echo "performance" | \ tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor mkfs.btrfs -f -m single -d single $DEV mount $MOUNT_OPTIONS $DEV $MNT dbench -D $MNT -t 300 64 umount $MNT The test was run on a machine with 12 cores, 64G of ram, using a NVMe device and a non-debug kernel configuration (Debian's default). Before patch set: Operation Count AvgLat MaxLat ---------------------------------------- NTCreateX 11277211 0.250 85.340 Close 8283172 0.002 6.479 Rename 477515 1.935 86.026 Unlink 2277936 0.770 87.071 Deltree 256 15.732 81.379 Mkdir 128 0.003 0.009 Qpathinfo 10221180 0.056 44.404 Qfileinfo 1789967 0.002 4.066 Qfsinfo 1874399 0.003 9.176 Sfileinfo 918589 0.061 10.247 Find 3951758 0.341 54.040 WriteX 5616547 0.047 85.079 ReadX 17676028 0.005 9.704 LockX 36704 0.003 1.800 UnlockX 36704 0.002 0.687 Flush 790541 14.115 676.236 Throughput 1179.19 MB/sec 64 clients 64 procs max_latency=676.240 ms After patch set: Operation Count AvgLat MaxLat ---------------------------------------- NTCreateX 12687926 0.171 86.526 Close 9320780 0.002 8.063 Rename 537253 1.444 78.576 Unlink 2561827 0.559 87.228 Deltree 374 11.499 73.549 Mkdir 187 0.003 0.005 Qpathinfo 11500300 0.061 36.801 Qfileinfo 2017118 0.002 7.189 Qfsinfo 2108641 0.003 4.825 Sfileinfo 1033574 0.008 8.065 Find 4446553 0.408 47.835 WriteX 6335667 0.045 84.388 ReadX 19887312 0.003 9.215 LockX 41312 0.003 1.394 UnlockX 41312 0.002 1.425 Flush 889233 13.014 623.259 Throughput 1339.32 MB/sec 64 clients 64 procs max_latency=623.265 ms +12.7% throughput, -8.2% max latency Signed-off-by:Filipe Manana <fdmanana@suse.com> Signed-off-by:
-
Filipe Manana authored
When logging an inode we may often have to fallback to a full transaction commit, either because a new block group was allocated, there is some case we can not deal with without a transaction commit or some error like an ENOMEM happened. However after we fallback to a transaction commit, we have a time window where we can make the next attempt to log any inode commit the next transaction unnecessarily, adding additional overhead and increasing latency. A sequence of steps that leads to this issue is the following: 1) The current open transaction has a generation of 1000; 2) A new block group is allocated, and as a consequence we must make sure any attempts to commit a log fallback to a transaction commit, so btrfs_set_log_full_commit() is called from btrfs_make_block_group(). This sets fs_info->last_trans_log_full_commit to 1000; 3) Task A is holding a handle on transaction 1000 and tries to log inode X. Once it gets to start_log_trans(), it calls btrfs_need_log_full_commit() which returns true, since fs_info->last_trans_log_full_commit has a value of 1000. So we end up returning EAGAIN and propagating it up to btrfs_sync_file(), where we commit transaction 1000; 4) The transaction commit task (task A) sets the transaction state to unblocked (TRANS_STATE_UNBLOCKED); 5) Some other task, task B, starts a new transaction with a generation of 1001; 6) Some stuff is done with transaction 1001, some btree blocks COWed, etc; 7) Transaction 1000 has not fully committed yet, we are still writing all the extent buffers it created; 8) Some new task, task C, starts an fsync of inode Y, gets a handle for transaction 1001, and it gets to btrfs_log_inode_parent() which does the following check: if (fs_info->last_trans_log_full_commit > last_committed) { ret = 1; goto end_no_trans; } At that point last_trans_log_full_commit has a value of 1000 and last_committed (value of fs_info->last_trans_committed) has a value of 999, since transaction 1000 has not yet committed - it is either still writing out dirty extent buffers, its super blocks or unpinning extents. As a consequence we return 1, which gets propagated up to btrfs_sync_file(), which will then call btrfs_commit_transaction() for transaction 1001. As a consequence we have an unnecessary second transaction commit, we previously committed transaction 1000 and now commit transaction 1001 as well, resulting in more overhead and increased latency. So fix this double transaction commit issue simply by removing that check, because all we need to do is wait for the previous transaction to finish its commit, which we already do later when starting the log transaction at start_log_trans(), because there we acquire the tree_log_mutex lock, which is held by a transaction commit and only released after the transaction commits its super blocks. Another issue that check has is that it reads last_trans_log_full_commit without using READ_ONCE(), which is incorrect since that member of struct btrfs_fs_info is always updated with WRITE_ONCE() through the helper btrfs_set_log_full_commit(). This double transaction commit issue can actually be triggered quite often in long runs of dbench, since besides the creation of new block groups that force inode logging to fallback to a transaction commit, there are cases where dbench asks to fsync a directory which had files in it that were previously renamed or subdirectories that were removed, resulting in the inode logging to fallback to a full transaction commit. This patch belongs to a patch set that is comprised of the following patches: btrfs: fix race causing unnecessary inode logging during link and rename btrfs: fix race that results in logging old extents during a fast fsync btrfs: fix race that causes unnecessary logging of ancestor inodes btrfs: fix race that makes inode logging fallback to transaction commit btrfs: fix race leading to unnecessary transaction commit when logging inode btrfs: do not block inode logging for so long during transaction commit Performance results are mentioned in the change log of the last patch. Signed-off-by: -
Filipe Manana authored
When logging an inode and the previous transaction is still committing, we have a time window where we can end up incorrectly think an inode has its last_unlink_trans field with a value greater than the last transaction committed, which results in the logging to fallback to a full transaction commit, which is usually much more expensive than doing a log commit. The race is described by the following steps: 1) We are at transaction 1000; 2) We modify an inode X (a directory) using transaction 1000 and set its last_unlink_trans field to 1000, because for example we removed one of its subdirectories; 3) We create a new inode Y with a dentry in inode X using transaction 1000, so its generation field is set to 1000; 4) The commit for transaction 1000 is started by task A; 5) The task committing transaction 1000 sets the transaction state to unblocked, writes the dirty extent buffers and the super blocks, then unlocks tree_log_mutex; 6) Some task starts a new transaction with a generation of 1001; 7) We do some modification to inode Y (using transaction 1001); 8) The transaction 1000 commit starts unpinning extents. At this point fs_info->last_trans_committed still has a value of 999; 9) Task B starts an fsync on inode Y, and gets a handle for transaction 1001. When it gets to check_parent_dirs_for_sync() it does the checking of the ancestor dentries because the following check does not evaluate to true: if (S_ISREG(inode->vfs_inode.i_mode) && inode->generation <= last_committed && inode->last_unlink_trans <= last_committed) goto out; The generation value for inode Y is 1000 and last_committed, which has the value read from fs_info->last_trans_committed, has a value of 999, so that check evaluates to false and we proceed to check the ancestor inodes. Once we get to the first ancestor, inode X, we call btrfs_must_commit_transaction() on it, which evaluates to true: static bool btrfs_must_commit_transaction(...) { struct btrfs_fs_info *fs_info = inode->root->fs_info; bool ret = false; mutex_lock(&inode->log_mutex); if (inode->last_unlink_trans > fs_info->last_trans_committed) { /* * Make sure any commits to the log are forced to be full * commits. */ btrfs_set_log_full_commit(trans); ret = true; } (...) because inode's X last_unlink_trans has a value of 1000 and fs_info->last_trans_committed still has a value of 999, it returns true to check_parent_dirs_for_sync(), making it return 1 which is propagated up to btrfs_sync_file(), causing it to fallback to a full transaction commit of transaction 1001. We should have not fallen back to commit transaction 1001, since inode X had last_unlink_trans set to 1000 and the super blocks for transaction 1000 were already written. So while not resulting in a functional problem, it leads to a lot more work and higher latencies for a fsync since committing a transaction is usually more expensive than committing a log (if other filesystem changes happened under that transaction). Similar problem happens when logging directories, for the same reason as btrfs_must_commit_transaction() returns true on an inode with its last_unlink_trans having the generation of the previous transaction and that transaction is still committing, unpinning its freed extents. So fix this by comparing last_unlink_trans with the id of the current transaction instead of fs_info->last_trans_committed. This case is often hit when running dbench for a long enough duration, as it does lots of rename and rmdir operations (both update the field last_unlink_trans of an inode) and fsyncs of files and directories. This patch belongs to a patch set that is comprised of the following patches: btrfs: fix race causing unnecessary inode logging during link and rename btrfs: fix race that results in logging old extents during a fast fsync btrfs: fix race that causes unnecessary logging of ancestor inodes btrfs: fix race that makes inode logging fallback to transaction commit btrfs: fix race leading to unnecessary transaction commit when logging inode btrfs: do not block inode logging for so long during transaction commit Performance results are mentioned in the change log of the last patch. Signed-off-by: -
Filipe Manana authored
When logging an inode and we are checking if we need to log ancestors that are new, if the previous transaction is still committing we have a time window where we can unnecessarily log ancestor inodes that were created in the previous transaction. The race is described by the following steps: 1) We are at transaction 1000; 2) Directory inode X is created, its generation is set to 1000; 3) The commit for transaction 1000 is started by task A; 4) The task committing transaction 1000 sets the transaction state to unblocked, writes the dirty extent buffers and the super blocks, then unlocks tree_log_mutex; 5) Inode Y, a regular file, is created under directory inode X, this results in starting a new transaction with a generation of 1001; 6) The transaction 1000 commit is unpinning extents. At this point fs_info->last_trans_committed still has a value of 999; 7) Task B calls fsync on inode Y and gets a handle for transaction 1001; 8) Task B ends up at log_all_new_ancestors() and then because inode Y has only one hard link, ends up at log_new_ancestors_fast(). There it reads a value of 999 from fs_info->last_trans_committed, and sees that the parent inode X has a generation of 1000, so we end up logging inode X: if (inode->generation > fs_info->last_trans_committed) { ret = btrfs_log_inode(trans, root, inode, LOG_INODE_EXISTS, ctx); (...) which is not necessary since it was created in the past transaction, with a generation of 1000, and that transaction has already committed its super blocks - it's still unpinning extents so it has not yet updated fs_info->last_trans_committed from 999 to 1000. So this just causes us to spend more time logging and allocating and writing more tree blocks for the log tree. So fix this by comparing an inode's generation with the generation of the transaction our transaction handle refers to - if the inode's generation matches the generation of the current transaction than we know it is a new inode we need to log, otherwise don't log it. This case is often hit when running dbench for a long enough duration. This patch belongs to a patch set that is comprised of the following patches: btrfs: fix race causing unnecessary inode logging during link and rename btrfs: fix race that results in logging old extents during a fast fsync btrfs: fix race that causes unnecessary logging of ancestor inodes btrfs: fix race that makes inode logging fallback to transaction commit btrfs: fix race leading to unnecessary transaction commit when logging inode btrfs: do not block inode logging for so long during transaction commit Performance results are mentioned in the change log of the last patch. Signed-off-by: -
Filipe Manana authored
When logging the extents of an inode during a fast fsync, we have a time window where we can log extents that are from the previous transaction and already persisted. This only makes us waste time unnecessarily. The following sequence of steps shows how this can happen: 1) We are at transaction 1000; 2) An ordered extent E from inode I completes, that is it has gone through btrfs_finish_ordered_io(), and it set the extent maps' generation to 1000 when we unpin the extent, which is the generation of the current transaction; 3) The commit for transaction 1000 starts by task A; 4) The task committing transaction 1000 sets the transaction state to unblocked, writes the dirty extent buffers and the super blocks, then unlocks tree_log_mutex; 5) Some change is made to inode I, resulting in creation of a new transaction with a generation of 1001; 6) The transaction 1000 commit starts unpinning extents. At this point fs_info->last_trans_committed still has a value of 999; 7) Task B starts an fsync on inode I, and when it gets to btrfs_log_changed_extents() sees the extent map for extent E in the list of modified extents. It sees the extent map has a generation of 1000 and fs_info->last_trans_committed has a value of 999, so it proceeds to logging the respective file extent item and all the checksums covering its range. So we end up wasting time since the extent was already persisted and is reachable through the trees pointed to by the super block committed by transaction 1000. So just fix this by comparing the extent maps generation against the generation of the transaction handle - if it is smaller then the id in the handle, we know the extent was already persisted and we do not need to log it. This patch belongs to a patch set that is comprised of the following patches: btrfs: fix race causing unnecessary inode logging during link and rename btrfs: fix race that results in logging old extents during a fast fsync btrfs: fix race that causes unnecessary logging of ancestor inodes btrfs: fix race that makes inode logging fallback to transaction commit btrfs: fix race leading to unnecessary transaction commit when logging inode btrfs: do not block inode logging for so long during transaction commit Performance results are mentioned in the change log of the last patch. Signed-off-by:
-
Filipe Manana authored
When we are doing a rename or a link operation for an inode that was logged in the previous transaction and that transaction is still committing, we have a time window where we incorrectly consider that the inode was logged previously in the current transaction and therefore decide to log it to update it in the log. The following steps give an example on how this happens during a link operation: 1) Inode X is logged in transaction 1000, so its logged_trans field is set to 1000; 2) Task A starts to commit transaction 1000; 3) The state of transaction 1000 is changed to TRANS_STATE_UNBLOCKED; 4) Task B starts a link operation for inode X, and as a consequence it starts transaction 1001; 5) Task A is still committing transaction 1000, therefore the value stored at fs_info->last_trans_committed is still 999; 6) Task B calls btrfs_log_new_name(), it reads a value of 999 from fs_info->last_trans_committed and because the logged_trans field of inode X has a value of 1000, the function does not return immediately, instead it proceeds to logging the inode, which should not happen because the inode was logged in the previous transaction (1000) and not in the current one (1001). This is not a functional problem, just wasted time and space logging an inode that does not need to be logged, contributing to higher latency for link and rename operations. So fix this by comparing the inodes' logged_trans field with the generation of the current transaction instead of comparing with the value stored in fs_info->last_trans_committed. This case is often hit when running dbench for a long enough duration, as it does lots of rename operations. This patch belongs to a patch set that is comprised of the following patches: btrfs: fix race causing unnecessary inode logging during link and rename btrfs: fix race that results in logging old extents during a fast fsync btrfs: fix race that causes unnecessary logging of ancestor inodes btrfs: fix race that makes inode logging fallback to transaction commit btrfs: fix race leading to unnecessary transaction commit when logging inode btrfs: do not block inode logging for so long during transaction commit Performance results are mentioned in the change log of the last patch. Signed-off-by:
-
David Sterba authored
After removing the inode number cache that was using the free space cache code, we can remove at least the recalc_thresholds callback from the ops. Both code and tests use the same callback function. It's moved before its first use. The use_bitmaps callback is still needed by tests to create some extents/bitmap setup. Signed-off-by: -
Nikolay Borisov authored
Since it's being used solely for the freespace cache unconditionally set the flags required for it. Signed-off-by:
Nikolay Borisov <nborisov@suse.com> Reviewed-by:
-
Nikolay Borisov authored
Following removal of the ino cache io_ctl_init will be called only on behalf of the freespace inode. In this case we always want to check CRCs so conditional code that depended on io_ctl::check_crc can be removed. Signed-off-by:
-
Nikolay Borisov authored
It's been deprecated since commit b547a88e ("btrfs: start deprecation of mount option inode_cache") which enumerates the reasons. A filesystem that uses the feature (mount -o inode_cache) tracks the inode numbers in bitmaps, that data stay on the filesystem after this patch. The size is roughly 5MiB for 1M inodes [1], which is considered small enough to be left there. Removal of the change can be implemented in btrfs-progs if needed. [1] https://lore.kernel.org/linux-btrfs/20201127145836.GZ6430@twin.jikos.cz/Signed-off-by:
-
Nikolay Borisov authored
The former is going away as part of the inode map removal so switch callers to btrfs_find_free_objectid. No functional changes since with INODE_MAP disabled (default) find_free_objectid was called anyway. Signed-off-by:
-
Nikolay Borisov authored
Those functions are going to be used even after inode cache is removed so moved them to a more appropriate place. Signed-off-by:
-
David Sterba authored
Since commit 72deb455 ("block: remove CONFIG_LBDAF") (5.2) the sector_t type is u64 on all arches and configs so we don't need to typecast it. It used to be unsigned long and the result of sector size shifts were not guaranteed to fit in the type. Reviewed-by:
Johannes Thumshirn <johannes.thumshirn@wdc.com> Signed-off-by:
-
Naohiro Aota authored
Superblock (and its copies) is the only data structure in btrfs which has a fixed location on a device. Since we cannot overwrite in a sequential write required zone, we cannot place superblock in the zone. One easy solution is limiting superblock and copies to be placed only in conventional zones. However, this method has two downsides: one is reduced number of superblock copies. The location of the second copy of superblock is 256GB, which is in a sequential write required zone on typical devices in the market today. So, the number of superblock and copies is limited to be two. Second downside is that we cannot support devices which have no conventional zones at all. To solve these two problems, we employ superblock log writing. It uses two adjacent zones as a circular buffer to write updated superblocks. Once the first zone is filled up, start writing into the second one. Then, when both zones are filled up and before starting to write to the first zone again, it reset the first zone. We can determine the position of the latest superblock by reading write pointer information from a device. One corner case is when both zones are full. For this situation, we read out the last superblock of each zone, and compare them to determine which zone is older. The following zones are reserved as the circular buffer on ZONED btrfs. - The primary superblock: zones 0 and 1 - The first copy: zones 16 and 17 - The second copy: zones 1024 or zone at 256GB which is minimum, and next to it If these reserved zones are conventional, superblock is written fixed at the start of the zone without logging. Signed-off-by:
Naohiro Aota <naohiro.aota@wdc.com> Reviewed-by:
-
Naohiro Aota authored
Placing both data and metadata in a block group is impossible in ZONED mode. For data, we can allocate a space for it and write it immediately after the allocation. For metadata, however, we cannot do that, because the logical addresses are recorded in other metadata buffers to build up the trees. As a result, a data buffer can be placed after a metadata buffer, which is not written yet. Writing out the data buffer will break the sequential write rule. Check and disallow MIXED_BG with ZONED mode. Reviewed-by:
Anand Jain <anand.jain@oracle.com> Signed-off-by:
-
Naohiro Aota authored
fallocate() is implemented by reserving actual extent instead of reservations. This can result in exposing the sequential write constraint of host-managed zoned block devices to the application, which would break the POSIX semantic for the fallocated file. To avoid this, report fallocate() as not supported when in ZONED mode for now. In the future, we may be able to implement "in-memory" fallocate() in ZONED mode by utilizing space_info->bytes_may_use or similar, so this returns EOPNOTSUPP. Reviewed-by:
-
Naohiro Aota authored
NODATACOW implies overwriting the file data on a device, which is impossible in sequential required zones. Disable NODATACOW globally with mount option and per-file NODATACOW attribute by masking FS_NOCOW_FL. Reviewed-by:
-
Naohiro Aota authored
As updates to the space cache v1 are in-place, the space cache cannot be located over sequential zones and there is no guarantees that the device will have enough conventional zones to store this cache. Resolve this problem by disabling completely the space cache v1. This does not introduce any problems with sequential block groups: all the free space is located after the allocation pointer and no free space before the pointer. There is no need to have such cache. Note: we can technically use free-space-tree (space cache v2) on ZONED mode. But, since ZONED mode now always allocates extents in a block group sequentially regardless of underlying device zone type, it's no use to enable and maintain the tree. For the same reason, NODATACOW is also disabled. In summary, ZONED will disable: | Disabled features | Reason | |-------------------+-----------------------------------------------------| | RAID/DUP | Cannot handle two zone append writes to different | | | zones | |-------------------+-----------------------------------------------------| | space_cache (v1) | In-place updating | | NODATACOW | In-place updating | |-------------------+-----------------------------------------------------| | fallocate | Reserved extent will be a write hole | |-------------------+-----------------------------------------------------| | MIXED_BG | Allocated metadata region will be write holes for | | | data writes | Reviewed-by:
-
Naohiro Aota authored
The zone append write command has a maximum IO size restriction it accepts. This is because a zone append write command cannot be split, as we ask the device to place the data into a specific target zone and the device responds with the actual written location of the data. Introduce max_zone_append_size to zone_info and fs_info to track the value, so we can limit all I/O to a zoned block device that we want to write using the zone append command to the device's limits. Reviewed-by:
-
Naohiro Aota authored
Introduce function btrfs_check_zoned_mode() to check if ZONED flag is enabled on the file system and if the file system consists of zoned devices with equal zone size. Reviewed-by:
Damien Le Moal <damien.lemoal@wdc.com> Signed-off-by:
-
Naohiro Aota authored
If a zoned block device is found, get its zone information (number of zones and zone size). To avoid costly run-time zone report commands to test the device zones type during block allocation, attach the seq_zones bitmap to the device structure to indicate if a zone is sequential or accept random writes. Also it attaches the empty_zones bitmap to indicate if a zone is empty or not. This patch also introduces the helper function btrfs_dev_is_sequential() to test if the zone storing a block is a sequential write required zone and btrfs_dev_is_empty_zone() to test if the zone is a empty zone. Reviewed-by:
-
- 08 Dec, 2020 8 commits
-
-
Naohiro Aota authored
This patch introduces the ZONED incompat flag. The flag indicates that the volume management will satisfy the constraints imposed by host-managed zoned block devices (aligned chunk allocation, append-only updates, reset zone after filled). As the zoned support will happen incrementally due to enhancing some core infrastructure like super block writes, tree-log, raid support, the feature will appear in sysfs only on debug builds. It will be enabled once the support is feature complete and applications can reliably check whether zoned support is present or not. Reviewed-by:
-
Nikolay Borisov authored
Results in slightly smaller code. add/remove: 0/0 grow/shrink: 0/1 up/down: 0/-11 (-11) Function old new delta btrfs_should_end_transaction 96 85 -11 Total: Before=20070, After=20059, chg -0.05% Signed-off-by:
-
Nikolay Borisov authored
Signed-off-by:
-
Nikolay Borisov authored
It simply gets assigned to 'ret' in case of errors. The flow of the while loop is not changed by this commit since the few call sites that 'goto next' will simply break from the loop. Signed-off-by:
-
Nikolay Borisov authored
In most cases when an error is returned from a function 'ret' is simply assigned to 'err'. There is only one case where walk_up_reloc_tree can return a positive value - in this case the code breaks from the loop and ret is going to get its return value from btrfs_cow_block - either 0 or negative. This retains the old logic of how 'err' used to be set at this call site. Signed-off-by:
-
Nikolay Borisov authored
Use only a single 'ret' to control whether we should abort the transaction or not. That's fine, because if we abort a transaction then btrfs_end_transaction will return the same value as passed to btrfs_abort_transaction. No semantic changes. Signed-off-by:
-
Filipe Manana authored
When we are attempting to start writeback for an existing extent in NOCOW mode, at run_delalloc_nocow(), we must check if the extent is shared, and if it is, fallback to a COW write. However we do such check while still holding a read lock on the leaf that contains the file extent item, and that check, the call to btrfs_cross_ref_exist(), can take some time because: 1) It needs to do a search on the extent tree, which obviously takes some time, specially if delayed references are being run at the moment, as we can block when trying to lock currently write locked btree nodes; 2) It needs to check the delayed references for any existing reference for our data extent, this requires acquiring the delayed references' spinlock and maybe block on the mutex of a delayed reference head in the case where there is a delayed reference for our data extent, in the worst case it makes us release the path on the extent tree and retry the whole process again (going back to step 1). There are other operations we do while holding the leaf locked that can take some significant time as well (specially all together): * btrfs_extent_readonly() - to check if the block group containing the extent is currently in RO mode. This requires taking a spinlock and searching for the block group in a rbtree that can be big on large filesystems; * csum_exist_in_range() - to search if there are any checksums in the csum tree for the extent. Like before, this can take some time if we are in a filesystem that has both COW and NOCOW files, in which case the csum tree is not empty; * btrfs_inc_nocow_writers() - increment the number of nocow writers in the block group that contains the data extent. Needs to acquire a spinlock and search for the block group in a rbtree that can be big on large filesystems. So just unlock the leaf (release the path) before doing all those checks, since we do not need it anymore. In case we can not do a NOCOW write for the extent, due to any of those checks failing, and the writeback range goes beyond that extents' length, we will do another btree search for the next file extent item. The following script that calls dbench was used to measure the impact of this change on a VM with 8 CPUs, 16Gb of ram, using a raw NVMe device directly (no intermediary filesystem on the host) and using a non-debug kernel (default configuration on Debian): $ cat test-dbench.sh #!/bin/bash DEV=/dev/sdk MNT=/mnt/sdk MOUNT_OPTIONS="-o ssd -o nodatacow" MKFS_OPTIONS="-m single -d single" mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT dbench -D $MNT -t 300 64 umount $MNT Before this change: Operation Count AvgLat MaxLat ---------------------------------------- NTCreateX 9326331 0.317 399.957 Close 6851198 0.002 6.402 Rename 394894 2.621 402.819 Unlink 1883131 0.931 398.082 Deltree 256 19.160 303.580 Mkdir 128 0.003 0.016 Qpathinfo 8452314 0.068 116.133 Qfileinfo 1481921 0.001 5.081 Qfsinfo 1549963 0.002 4.444 Sfileinfo 759679 0.084 17.079 Find 3268168 0.396 118.196 WriteX 4653310 0.056 110.993 ReadX 14618818 0.005 23.314 LockX 30364 0.003 0.497 UnlockX 30364 0.002 1.720 Flush 653619 16.954 569.299 Throughput 966.651 MB/sec 64 clients 64 procs max_latency=569.377 ms After this change: Operation Count AvgLat MaxLat ---------------------------------------- NTCreateX 9710433 0.302 232.449 Close 7132948 0.002 11.496 Rename 411144 2.452 131.805 Unlink 1960961 0.893 230.383 Deltree 256 14.858 198.646 Mkdir 128 0.002 0.005 Qpathinfo 8800890 0.066 111.588 Qfileinfo 1542556 0.001 3.852 Qfsinfo 1613835 0.002 5.483 Sfileinfo 790871 0.081 19.492 Find 3402743 0.386 120.185 WriteX 4842918 0.054 179.312 ReadX 15220407 0.005 32.435 LockX 31612 0.003 1.533 UnlockX 31612 0.002 1.047 Flush 680567 16.320 463.323 Throughput 1016.59 MB/sec 64 clients 64 procs max_latency=463.327 ms +5.0% throughput, -20.5% max latency Also, the following test using fio was run: $ cat test-fio.sh #!/bin/bash DEV=/dev/sdk MNT=/mnt/sdk MOUNT_OPTIONS="-o ssd -o nodatacow" MKFS_OPTIONS="-d single -m single" if [ $# -ne 4 ]; then echo "Use $0 NUM_JOBS FILE_SIZE FSYNC_FREQ BLOCK_SIZE" exit 1 fi NUM_JOBS=$1 FILE_SIZE=$2 FSYNC_FREQ=$3 BLOCK_SIZE=$4 cat <<EOF > /tmp/fio-job.ini [writers] rw=randwrite fsync=$FSYNC_FREQ fallocate=none group_reporting=1 direct=0 bs=$BLOCK_SIZE ioengine=sync size=$FILE_SIZE directory=$MNT numjobs=$NUM_JOBS EOF echo echo "Using fio config:" echo cat /tmp/fio-job.ini echo echo "mount options: $MOUNT_OPTIONS" echo mkfs.btrfs -f $MKFS_OPTIONS $DEV > /dev/null mount $MOUNT_OPTIONS $DEV $MNT echo "Creating nodatacow files before fio runs..." for ((i = 0; i < $NUM_JOBS; i++)); do xfs_io -f -c "pwrite -b 128M 0 $FILE_SIZE" "$MNT/writers.$i.0" done sync fio /tmp/fio-job.ini umount $MNT Before this change: $ ./test-fio.sh 16 512M 2 4K (...) WRITE: bw=28.3MiB/s (29.6MB/s), 28.3MiB/s-28.3MiB/s (29.6MB/s-29.6MB/s), io=8192MiB (8590MB), run=289800-289800msec After this change: $ ./test-fio.sh 16 512M 2 4K (...) WRITE: bw=31.2MiB/s (32.7MB/s), 31.2MiB/s-31.2MiB/s (32.7MB/s-32.7MB/s), io=8192MiB (8590MB), run=262845-262845msec +9.7% throughput, -9.8% runtime Signed-off-by: -
David Sterba authored
The tree checker is called many times as it verifies metadata at read/write time. The checks follow a simple pattern: if (error_condition) { report_error(); return -EUCLEAN; } All the error reporting functions are annotated as __cold that is supposed to hint the compiler to move the statement block out of the hot path. This does not seem to happen that often. As the error condition is expected to be false almost always, we can annotate it with 'unlikely' as this satisfies one of the few use cases for the annotation. The expected outcome is a stronger hint to compiler to reorder the checks test jump to exit test jump to exit ... which can be observed in asm of eg. check_dir_item, btrfs_check_chunk_valid, check_root_item or check_leaf. There's a measurable run time improvement reported by Josef, the testing workload went from 655 MiB/s to 677 MiB/s, which is about +3%. There should be no functional changes but some of the conditions have been rewritten to produce more readable result, some lines are longer than 80, for the sake of readability. Signed-off-by:
-
