Ubuntu HWE improve ior performance#185
Open
hbirth wants to merge 9 commits into
Open
Conversation
hbirth
force-pushed
the
redfs-ubuntu-hwe-multiple-open-DIO-handling
branch
from
5cf3c59 to
2493f33
Compare
Signed-off-by: Horst Birthelmer <hbirthelmer@ddn.com>
Writes that already match the alignment advertised via FUSE_ALIGN_PG_ORDER gain nothing from the writeback cache and can degrade into page-sized WRITE requests under dirty throttling. Send them through fuse_perform_write() instead, which packs requests up to max_write and keeps them stripe-aligned for the backend. They create no dirty pages, so no DLM write lock needs to be cached for them. Unaligned writes keep using the writeback cache. Also clarify in the uapi header that align_page_order is the log2 of the alignment in bytes, not in pages. Ported from the redfs-ubuntu-noble-writethrough-split branch and adapted to the iomap-based writeback path: the decision gates the writeback bool in fuse_cache_write_iter() (and the DLM write-lock acquisition) instead of branching to a writethrough label. Signed-off-by: Horst Birthelmer <hbirthelmer@ddn.com>
Add a per-connection size threshold, settable via fusectl as writethrough_threshold, that sends buffered writes >= threshold through fuse_perform_write() regardless of alignment. The knob is off by default (0 == disabled) and leaves the existing alignment-based decision in place for writes below the threshold. Ported from the redfs-ubuntu-noble-writethrough-split branch; the fusectl dentry uses this branch's fuse_ctl_add_dentry() signature and the ops struct omits the now-removed no_llseek. Signed-off-by: Horst Birthelmer <hbirthelmer@ddn.com>
fuse_readahead() batches whole folios into a single request, capped at min(fc->max_pages, fc->max_read/PAGE_SIZE) pages, but fuse_init_file_inode() let the page cache build folios up to MAX_PAGECACHE_ORDER. A large sequential read could thus produce a folio bigger than one request can carry: the first loop iteration took the folio_pages > cur_pages path, fired WARN_ON(!pages), and broke with ap->num_folios == 0. fuse_send_readpages() was still called and dereferenced a NULL ap->folios[0] via folio_pos(), oopsing at CR2=0x20 (folio->index). Cap the folio order to the per-request page limit so the page cache can never build an unserviceable folio. Signed-off-by: Horst Birthelmer <hbirthelmer@ddn.com>
A FUSE server that advertises a large max_pages and max_write (e.g. max_pages=256, max_write=1MB) cannot currently obtain matching FUSE_READ request sizes from the kernel. Buffered sequential writes arrive at the server at the negotiated max_write size, but a large buffered read() is split into several smaller FUSE_READ requests. For a buffered read, filemap_get_pages() -> page_cache_sync_ra() sizes the read against ractl_max_pages(): max_pages = ractl->ra->ra_pages; if (req_size > max_pages && bdi->io_pages > max_pages) max_pages = min(req_size, bdi->io_pages); fuse leaves bdi->io_pages at the default VM_READAHEAD_PAGES (128KB), so a 1MB read() (req_size = 256 pages) is clamped to the readahead window (128KB, or 256KB for POSIX_FADV_SEQUENTIAL), producing four 256KB FUSE_READ round-trips instead of one. Set bdi->io_pages to fc->max_pages after feature negotiation. As the code above shows, io_pages only raises the limit when the request size already exceeds the readahead window, so it enlarges explicitly requested reads without enlarging the speculative readahead window. This avoids increasing speculative page-cache readahead on behalf of an unprivileged server. NFS does the same, setting io_pages from rpages while leaving ra_pages at the default. fc->max_pages is already bounded by fc->max_pages_limit (and, for virtio-fs, by the virtqueue descriptor count), so io_pages inherits the same bound. Suggested-by: Joanne Koong <joannelkoong@gmail.com> Signed-off-by: Jim Harris <jim.harris@nvidia.com> Assisted-by: Cursor:claude-opus-4.8
commit 0c58a97 ("fuse: remove tmp folio for writebacks and internal rb tree") removed temp folios for dirty page writeback. Consequently, fuse can now use the default writeback accounting. With switching fuse to use default writeback accounting, there are some added benefits. This updates wb->writeback_inodes tracking as well now and updates writeback throughput estimates after writeback completion. This commit also removes inc_wb_stat() and dec_wb_stat(). These have no callers anymore now that fuse does not call them. Signed-off-by: Joanne Koong <joannelkoong@gmail.com> Reviewed-by: David Hildenbrand <david@redhat.com> Reviewed-by: Bernd Schubert <bschubert@ddn.com> Signed-off-by: Miklos Szeredi <mszeredi@redhat.com> (cherry picked from commit 494d2f5)
When several entities open the same file for writing, the cached write path (fuse_cache_write_iter()) serializes them on the exclusive inode lock, which is held across synchronous server round-trips. Add an opt-in per-connection knob (force_dio_on_contention, off by default, exposed via the fuse control filesystem). When enabled, an inode opened by a second writer is latched into direct IO: reads and writes use the parallel shared-lock dio path, the page cache is flushed and dropped at open (reusing the O_TRUNC scaffolding), and new opens skip caching mode. The latch is cleared when the last writer closes and reverted to caching mode on mmap. Detection uses fi->write_files so only writers count: a reader opening alongside a single writer does not trigger the switch. Signed-off-by: Horst Birthelmer <hbirthelmer@ddn.com>
A FUSE_NOTIFY_INVAL_INODE data invalidation means another (remote) entity is modifying the file. When it is also open for writing here, this is the multi-writer contention already handled at open() time, so latch the inode into direct IO (gated by force_dio_on_contention). Reverted on last writer close / mmap, as for the open-time latch. Latching to direct IO is only coherent if no buffered write can deposit dirty folios into the page cache after it has been dropped: once latched the inode serves reads and writes direct (from the server). Add a dedicated per-inode rw_semaphore, wb_inval_rwsem, to serialize the buffered-write page-cache dirtying against the latch transitions. The writeback path holds it for read around the dirtying and re-checks the latch under it; fuse_reverse_inval_inode(), the mmap revert and the last-writer-close revert hold it for write around their invalidate + latch update. The close revert also gains an invalidate (in fuse_file_release(), a sleepable context) so a clean folio repopulated by a read racing the latch is not served stale once caching resumes; only clean folios exist while latched, so it is server-free. The writer's read-side section must stay free of server round-trips or the down_write() could wait on the server; Signed-off-by: Horst Birthelmer <hbirthelmer@ddn.com>
Extending FOPEN_PARALLEL_DIRECT_WRITES writes were forced onto the
exclusive inode lock, re-serializing the parallel phase. The exclusive
lock only bundled "write + advance i_size + undo-on-failure" into one
unit. But i_size is committed by fuse_write_update_attr() under
fi->lock, only on a successful growing write and independent of the
inode rwsem -- so shared-lock writers commit size correctly and have
nothing to undo. Drop the past-EOF exclusive triggers and gate the
whole-file fuse_do_truncate() rollback on holding the exclusive lock.
Lock mode is passed to __fuse_direct_IO(); i_size is committed at the
same point in every path, only the failure rollback differs:
non-exclusive (relaxed, parallel):
fuse_direct_write_iter
fuse_dio_lock -> inode_lock_shared (exclusive=false)
__fuse_direct_IO(.., false)
fuse_direct_io() write to server
fuse_write_update_attr() commit i_size (on success)
no rollback
exclusive (append / caching / !parallel):
fuse_direct_write_iter
fuse_dio_lock -> inode_lock (exclusive=true)
__fuse_direct_IO(.., true)
fuse_direct_io() write to server
fuse_write_update_attr() commit i_size (on success)
ret<0 & extend -> fuse_do_truncate() rollback
exclusive (caching-mode O_DIRECT):
fuse_cache_write_iter -> inode_lock (exclusive)
generic_file_direct_write -> fuse_direct_IO
__fuse_direct_IO(.., true)
fuse_direct_io() write to server
fuse_write_update_attr() commit i_size (on success)
ret<0 & extend -> fuse_do_truncate() rollback
Signed-off-by: Bernd Schubert <bschubert@ddn.com>
hbirth
force-pushed
the
redfs-ubuntu-hwe-multiple-open-DIO-handling
branch
from
2493f33 to
609aa83
Compare
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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.