mirror of
https://git.proxmox.com/git/mirror_zfs.git
synced 2026-01-14 17:22:05 +03:00
3d6ee9a68c
Page writeback is considered completed when the associated itx callback completes. A syncing writeback will receive the error in its callback directly, but an in-flight async writeback that was promoted to sync by the ZIL may also receive an error. Writeback errors, even syncing writeback errors, are not especially serious on their own, because the error will ultimately be returned to the zil_commit() caller, either zfs_fsync() for an explicit sync op (eg msync()) or to zfs_putpage() itself for a syncing (WB_SYNC_ALL) writeback (kernel housekeeping or sync_file_range(SYNC_FILE_RANGE_WAIT_AFTER). The only thing we need to do when a page writeback fails is to re-mark the page dirty, since we don't know if it made it to disk yet. This will ensure that it gets written out again in the future, either some scheduled async writeback or another explicit syncing call. On the other side, we need to make sure that if a syncing op arrives, any changes on dirty pages are written back to the DMU and/or the ZIL first. We do this by starting an _async_ (WB_SYNC_NONE) writeback on the file mapping at the start of the sync op (fsync(), msync(), etc). An async op will get an async itx created and logged, ready for the followup zfs_fsync()->zil_commit() to find, while avoiding a zil_commit() call for every page in the range. Sponsored-by: Klara, Inc. Sponsored-by: Wasabi Technology, Inc. Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Alexander Motin <alexander.motin@TrueNAS.com> Signed-off-by: Rob Norris <rob.norris@klarasystems.com> Closes #17398 |
||
|---|---|---|
| .. | ||
| avl | ||
| icp | ||
| lua | ||
| nvpair | ||
| os | ||
| unicode | ||
| zcommon | ||
| zfs | ||
| zstd | ||
| .gitignore | ||
| Kbuild.in | ||
| Makefile.bsd | ||
| Makefile.in | ||