Personal/linuxsmiths/wait for ongoing flush must drain flush callback #162
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
added 30 commits
October 12, 2024 01:09
Since write can get a bc allocated by readahead, we need to enforce rsize <= wsize
Misc fixes and comments added.
This makes progress very slow with file reads/writes which need to get the bc vector.
… directory cache purge of the current directory inode
…e not changed. w/o this read only scenarios, we can have attr expire and then get_file_size() will return -1 and we will not do readahead.
…of whether it could or could not add the entry to directory cache. This is because if a file/dir is created inside a directory, our cache technically becomes out of sync with the server. Note that w emay fail to add to the directory cache because it may be full or some other error.
…) and send_readdir_or_readdirplus_response() to fix an inode leak Now we hold inode lookupcnt ref for all entries with a valid inode. This way, send_readdir_or_readdirplus_response() can call decref() for the inodes it doens't pass to fuse. Previously we were decrementing dircachecnt and there was a possibility of that becoming 0 while lookupcnt was already 0. Those inodes would be sitting in inode_map with lookupcnt=dircachecnt=0.
We rename all locks to the form <context>_lock_N. N is the order of the lock. A thread can only hold a higher order lock (greater N) then the highest order lock it's currently holding, i.e., a thread holding a lock *_lock_N cannot hold any lock from *_lock_0 to *_lock_N-1 (it can only hold *_lock_N+1 and higher order locks). Later I plan to add assertions to catch violations of the above rule. nfs_client::inode_map_lock -> nfs_client::inode_map_lock_0 nfs_inode::ilock -> nfs_inode::ilock_1 nfs_inode::readdircache_lock -> nfs_inode::readdircache_lock_2 nfs_client::jukebox_seeds_lock -> nfs_client::jukebox_seeds_lock_39 ra_state::ra_lock -> ra_state::ra_lock_40 rpc_task_helper::task_index_lock -> rpc_task_helper::task_index_lock_41 rpc_stats_az::stats_lock -> rpc_stats_az::stats_lock_42 bytes_chunk_cache::chunkmap_lock -> bytes_chunk_cache::chunkmap_lock_43 membuf::mb_lock -> membuf::mb_lock_44
…correctly as we were clearing the nfs_inode.
Added a TODO to fix it. Also fixed a log for readdir.
… at the server. w/o this we might return stale data for a file that's truncated by some other client.
* Inline flush to free the fuse thread. * Fix * Addressed the review comments. * Audit changes. * Refactored flushing and inline flushing code, with good comments. --------- Co-authored-by: Nagendra Tomar <[email protected]>
(cherry picked from commit 222c5db)
* Invalidate attribute cache if post-op attributes are missing * Address comment * address comment --------- Co-authored-by: amulyan13 <root@amnNfsCanaryVM3.vnwjqyduitfelpwtytwaokslhe.cbnx.internal.cloudapp.net>
* Working change of inline writes handling. * Code Refactoring. * Changed to vector pointer. * Nitpiks.
* Removing tenant and subid from config * Minor fixes * Fixed run command * Fixes * Audit changes. * Updated with clientid processing and ten+sub in cb * Addressed comments * Updated * Updated * Revert "Updated" This reverts commit 44693ce. * Update * Update * Update --------- Co-authored-by: Nagendra Tomar <[email protected]>
This can help us see fi any specific fuse callback handler is blockign or taking more time.
… cache enabled status etc
…side sync_membufs() is senn to fail sometimes.
…side sync_membufs() is senn to fail sometimes.
… before starting the first write in sync_membufs()
Co-authored-by: Ubuntu <shubham@shubham808VM-westeurope.2vwp434kux0ehoz4y3jbj5wcwa.ax.internal.cloudapp.net>
This can be called when cache is being shutdown on file close. This will delete dirty/uncommitted membufs too.
Co-authored-by: Nagendra Tomar <[email protected]>
…e flush to complete. We cannot hold flush_lock while waiting for any WRITE RPCs to complete as write_iov_callback()->on_flush_complete() also grabs the flush_lock. This can cause a deadlock.
…the flush to complete We have a rule that anyone waiting for a write to complete must not hold the flush_lock as the write_iov_callback() also may hold the flush lock.
Co-authored-by: Nagendra Tomar <[email protected]>
else we can have a deadlock where write_iov_callback() first releases the membuf lock which make wait_for_ongoing_flush() believe that all flush are done and it can take the flush_lock safely, but write_iov_callback() calls on_flush_complete() later which grabs the flush_lock. This can cause a deadlock. The fix is to let the callback drain. Since wait_for_ongoing_flush() is a rarely used function, we make that check while write_iov_callback() can run w/o any additional check or complexity. Another way of solving this would be to release the membuf lock after on_flush_complete() runs, but then it would have added unncessary restrictions on what we can do insode on_flush_complete().
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.