kernel/fs/btrfs/delayed-ref.c, branch linux-5.1.y

btrfs: qgroup: Move reserved data accounting from btrfs_delayed_ref_head to btrfs_qgroup_extent_record

2019-02-25T13:13:39Z

[BUG] Btrfs/139 will fail with a high probability if the testing machine (VM) has only 2G RAM. Resulting the final write success while it should fail due to EDQUOT, and the fs will have quota exceeding the limit by 16K. The simplified reproducer will be: (needs a 2G ram VM) $ mkfs.btrfs -f $dev $ mount $dev $mnt $ btrfs subv create $mnt/subv $ btrfs quota enable $mnt $ btrfs quota rescan -w $mnt $ btrfs qgroup limit -e 1G $mnt/subv $ for i in $(seq -w 1 8); do xfs_io -f -c "pwrite 0 128M" $mnt/subv/file_$i > /dev/null echo "file $i written" > /dev/kmsg done $ sync $ btrfs qgroup show -pcre --raw $mnt The last pwrite will not trigger EDQUOT and final 'qgroup show' will show something like: qgroupid rfer excl max_rfer max_excl parent child -------- ---- ---- -------- -------- ------ ----- 0/5 16384 16384 none none --- --- 0/256 1073758208 1073758208 none 1073741824 --- --- And 1073758208 is larger than > 1073741824. [CAUSE] It's a bug in btrfs qgroup data reserved space management. For quota limit, we must ensure that: reserved (data + metadata) + rfer/excl <= limit Since rfer/excl is only updated at transaction commmit time, reserved space needs to be taken special care. One important part of reserved space is data, and for a new data extent written to disk, we still need to take the reserved space until rfer/excl numbers get updated. Originally when an ordered extent finishes, we migrate the reserved qgroup data space from extent_io tree to delayed ref head of the data extent, expecting delayed ref will only be cleaned up at commit transaction time. However for small RAM machine, due to memory pressure dirty pages can be flushed back to disk without committing a transaction. The related events will be something like: file 1 written btrfs_finish_ordered_io: ino=258 ordered offset=0 len=54947840 btrfs_finish_ordered_io: ino=258 ordered offset=54947840 len=5636096 btrfs_finish_ordered_io: ino=258 ordered offset=61153280 len=57344 btrfs_finish_ordered_io: ino=258 ordered offset=61210624 len=8192 btrfs_finish_ordered_io: ino=258 ordered offset=60583936 len=569344 cleanup_ref_head: num_bytes=54947840 cleanup_ref_head: num_bytes=5636096 cleanup_ref_head: num_bytes=569344 cleanup_ref_head: num_bytes=57344 cleanup_ref_head: num_bytes=8192 ^^^^^^^^^^^^^^^^ This will free qgroup data reserved space file 2 written ... file 8 written cleanup_ref_head: num_bytes=8192 ... btrfs_commit_transaction <<< the only transaction committed during the test When file 2 is written, we have already freed 128M reserved qgroup data space for ino 258. Thus later write won't trigger EDQUOT. This allows us to write more data beyond qgroup limit. In my 2G ram VM, it could reach about 1.2G before hitting EDQUOT. [FIX] By moving reserved qgroup data space from btrfs_delayed_ref_head to btrfs_qgroup_extent_record, we can ensure that reserved qgroup data space won't be freed half way before commit transaction, thus fix the problem. Fixes: f64d5ca86821 ("btrfs: delayed_ref: Add new function to record reserved space into delayed ref") Signed-off-by: Qu Wenruo Signed-off-by: David Sterba

btrfs: introduce delayed_refs_rsv

2018-12-17T13:51:46Z

Traditionally we've had voodoo in btrfs to account for the space that delayed refs may take up by having a global_block_rsv. This works most of the time, except when it doesn't. We've had issues reported and seen in production where sometimes the global reserve is exhausted during transaction commit before we can run all of our delayed refs, resulting in an aborted transaction. Because of this voodoo we have equally dubious flushing semantics around throttling delayed refs which we often get wrong. So instead give them their own block_rsv. This way we can always know exactly how much outstanding space we need for delayed refs. This allows us to make sure we are constantly filling that reservation up with space, and allows us to put more precise pressure on the enospc system. Instead of doing math to see if its a good time to throttle, the normal enospc code will be invoked if we have a lot of delayed refs pending, and they will be run via the normal flushing mechanism. For now the delayed_refs_rsv will hold the reservations for the delayed refs, the block group updates, and deleting csums. We could have a separate rsv for the block group updates, but the csum deletion stuff is still handled via the delayed_refs so that will stay there. Historical background: The global reserve has grown to cover everything we don't reserve space explicitly for, and we've grown a lot of weird ad-hoc heuristics to know if we're running short on space and when it's time to force a commit. A failure rate of 20-40 file systems when we run hundreds of thousands of them isn't super high, but cleaning up this code will make things less ugly and more predictible. Thus the delayed refs rsv. We always know how many delayed refs we have outstanding, and although running them generates more we can use the global reserve for that spill over, which fits better into it's desired use than a full blown reservation. This first approach is to simply take how many times we're reserving space for and multiply that by 2 in order to save enough space for the delayed refs that could be generated. This is a niave approach and will probably evolve, but for now it works. Signed-off-by: Josef Bacik Reviewed-by: David Sterba # high-level review [ added background notes from the cover letter ] Signed-off-by: David Sterba

btrfs: only track ref_heads in delayed_ref_updates

2018-12-17T13:51:46Z

We use this number to figure out how many delayed refs to run, but __btrfs_run_delayed_refs really only checks every time we need a new delayed ref head, so we always run at least one ref head completely no matter what the number of items on it. Fix the accounting to only be adjusted when we add/remove a ref head. In addition to using this number to limit the number of delayed refs run, a future patch is also going to use it to calculate the amount of space required for delayed refs space reservation. Reviewed-by: Nikolay Borisov Signed-off-by: Josef Bacik Signed-off-by: David Sterba

btrfs: add btrfs_delete_ref_head helper

2018-12-17T13:51:46Z

We do this dance in cleanup_ref_head and check_ref_cleanup, unify it into a helper and cleanup the calling functions. Reviewed-by: Omar Sandoval Reviewed-by: Nikolay Borisov Signed-off-by: Josef Bacik Reviewed-by: David Sterba Signed-off-by: David Sterba

btrfs: delayed-ref: extract find_first_ref_head from find_ref_head

2018-10-17T17:21:00Z

The find_ref_head shouldn't return the first entry even if no exact match is found. So move the hidden behavior to higher level. Besides, remove the useless local variables in the btrfs_select_ref_head. Reviewed-by: Nikolay Borisov Signed-off-by: Lu Fengqi [ reformat comment ] Signed-off-by: David Sterba

btrfs: switch return_bigger to bool in find_ref_head

2018-10-15T15:23:41Z

Using bool is more suitable than int here, and add the comment about the return_bigger. Signed-off-by: Lu Fengqi Reviewed-by: Nikolay Borisov Reviewed-by: David Sterba Signed-off-by: David Sterba

btrfs: delayed-ref: pass delayed_refs directly to btrfs_delayed_ref_lock

2018-10-15T15:23:41Z

Since trans is only used for referring to delayed_refs, there is no need to pass it instead of delayed_refs to btrfs_delayed_ref_lock(). No functional change. Reviewed-by: Nikolay Borisov Signed-off-by: Lu Fengqi Reviewed-by: David Sterba Signed-off-by: David Sterba

btrfs: delayed-ref: pass delayed_refs directly to btrfs_select_ref_head

2018-10-15T15:23:40Z

Since trans is only used for referring to delayed_refs, there is no need to pass it instead of delayed_refs to btrfs_select_ref_head(). No functional change. Signed-off-by: Lu Fengqi Reviewed-by: Nikolay Borisov Reviewed-by: David Sterba Signed-off-by: David Sterba

Btrfs: delayed-refs: use rb_first_cached for ref_tree

2018-10-15T15:23:33Z

rb_first_cached() trades an extra pointer "leftmost" for doing the same job as rb_first() but in O(1). Functions manipulating href->ref_tree need to get the first entry, this converts href->ref_tree to use rb_first_cached(). For more details about the optimization see patch "Btrfs: delayed-refs: use rb_first_cached for href_root". Tested-by: Holger Hoffstätte Signed-off-by: Liu Bo Reviewed-by: David Sterba Signed-off-by: David Sterba

Btrfs: delayed-refs: use rb_first_cached for href_root

2018-10-15T15:23:33Z

rb_first_cached() trades an extra pointer "leftmost" for doing the same job as rb_first() but in O(1). Functions manipulating href_root need to get the first entry, this converts href_root to use rb_first_cached(). This patch is first in the sequenct of similar updates to other rbtrees and this is analysis of the expected behaviour and improvements. There's a common pattern: while (node = rb_first) { entry = rb_entry(node) next = rb_next(node) rb_erase(node) cleanup(entry) } rb_first needs to traverse the tree up to logN depth, rb_erase can completely reshuffle the tree. With the caching we'll skip the traversal in rb_first. That's a cached memory access vs looped pointer dereference trade-off that IMHO has a clear winner. Measurements show there's not much difference in a sample tree with 10000 nodes: 4.5s / rb_first and 4.8s / rb_first_cached. Real effects of caching and pointer chasing are unpredictable though. Further optimzations can be done to avoid the expensive rb_erase step. In some cases it's ok to process the nodes in any order, so the tree can be traversed in post-order, not rebalancing the children nodes and just calling free. Care must be taken regarding the next node. Tested-by: Holger Hoffstätte Signed-off-by: Liu Bo Reviewed-by: David Sterba [ update changelog from mail discussions ] Signed-off-by: David Sterba