| Message ID | 20241115224459.427610-6-joannelkoong@gmail.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | fuse: remove temp page copies in writeback | expand |
On 11/16/24 6:44 AM, Joanne Koong wrote: > @@ -1838,7 +1748,7 @@ static void fuse_writepage_finish_stat(struct inode *inode, struct folio *folio) > struct backing_dev_info *bdi = inode_to_bdi(inode); > > dec_wb_stat(&bdi->wb, WB_WRITEBACK); > - node_stat_sub_folio(folio, NR_WRITEBACK_TEMP); > + node_stat_sub_folio(folio, NR_WRITEBACK); Now fuse_writepage_finish_stat() has only one caller and we could make it embedded into its only caller. > static void fuse_writepage_args_page_fill(struct fuse_writepage_args *wpa, struct folio *folio, > - struct folio *tmp_folio, uint32_t folio_index) > + uint32_t folio_index) > { > struct inode *inode = folio->mapping->host; > struct fuse_args_pages *ap = &wpa->ia.ap; > > - folio_copy(tmp_folio, folio); > - > - ap->folios[folio_index] = tmp_folio; > + ap->folios[folio_index] = folio; > ap->descs[folio_index].offset = 0; > ap->descs[folio_index].length = PAGE_SIZE; > > inc_wb_stat(&inode_to_bdi(inode)->wb, WB_WRITEBACK); > - node_stat_add_folio(tmp_folio, NR_WRITEBACK_TEMP); > + node_stat_add_folio(folio, NR_WRITEBACK); This inc NR_WRITEBACK counter along with the corresponding dec NR_WRITEBACK counter in fuse_writepage_finish_stat() seems unnecessary, as folio_start_writeback() will increase the NR_WRITEBACK counter, while folio_end_writeback() will decrease the NR_WRITEBACK counter.
On 11/16/24 6:44 AM, Joanne Koong wrote: > In the current FUSE writeback design (see commit 3be5a52b30aa > ("fuse: support writable mmap")), a temp page is allocated for every > dirty page to be written back, the contents of the dirty page are copied over > to the temp page, and the temp page gets handed to the server to write back. > > This is done so that writeback may be immediately cleared on the dirty page, > and this in turn is done for two reasons: > a) in order to mitigate the following deadlock scenario that may arise > if reclaim waits on writeback on the dirty page to complete: > * single-threaded FUSE server is in the middle of handling a request > that needs a memory allocation > * memory allocation triggers direct reclaim > * direct reclaim waits on a folio under writeback > * the FUSE server can't write back the folio since it's stuck in > direct reclaim > b) in order to unblock internal (eg sync, page compaction) waits on > writeback without needing the server to complete writing back to disk, > which may take an indeterminate amount of time. > > With a recent change that added AS_WRITEBACK_INDETERMINATE and mitigates > the situations described above, FUSE writeback does not need to use > temp pages if it sets AS_WRITEBACK_INDETERMINATE on its inode mappings. > > This commit sets AS_WRITEBACK_INDETERMINATE on the inode mappings > and removes the temporary pages + extra copying and the internal rb > tree. > > fio benchmarks -- > (using averages observed from 10 runs, throwing away outliers) > > Setup: > sudo mount -t tmpfs -o size=30G tmpfs ~/tmp_mount > ./libfuse/build/example/passthrough_ll -o writeback -o max_threads=4 -o source=~/tmp_mount ~/fuse_mount > > fio --name=writeback --ioengine=sync --rw=write --bs={1k,4k,1M} --size=2G > --numjobs=2 --ramp_time=30 --group_reporting=1 --directory=/root/fuse_mount > > bs = 1k 4k 1M > Before 351 MiB/s 1818 MiB/s 1851 MiB/s > After 341 MiB/s 2246 MiB/s 2685 MiB/s > % diff -3% 23% 45% > > Signed-off-by: Joanne Koong <joannelkoong@gmail.com> > --- > fs/fuse/file.c | 339 +++---------------------------------------------- > 1 file changed, 20 insertions(+), 319 deletions(-) > > diff --git a/fs/fuse/file.c b/fs/fuse/file.c > index 88d0946b5bc9..56289ac58596 100644 > --- a/fs/fuse/file.c > +++ b/fs/fuse/file.c > @@ -415,89 +415,11 @@ u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id) > > struct fuse_writepage_args { > struct fuse_io_args ia; > - struct rb_node writepages_entry; > struct list_head queue_entry; > - struct fuse_writepage_args *next; > struct inode *inode; > struct fuse_sync_bucket *bucket; > }; > > -static struct fuse_writepage_args *fuse_find_writeback(struct fuse_inode *fi, > - pgoff_t idx_from, pgoff_t idx_to) > -{ > - struct rb_node *n; > - > - n = fi->writepages.rb_node; > - > - while (n) { > - struct fuse_writepage_args *wpa; > - pgoff_t curr_index; > - > - wpa = rb_entry(n, struct fuse_writepage_args, writepages_entry); > - WARN_ON(get_fuse_inode(wpa->inode) != fi); > - curr_index = wpa->ia.write.in.offset >> PAGE_SHIFT; > - if (idx_from >= curr_index + wpa->ia.ap.num_folios) > - n = n->rb_right; > - else if (idx_to < curr_index) > - n = n->rb_left; > - else > - return wpa; > - } > - return NULL; > -} > - > -/* > - * Check if any page in a range is under writeback > - */ > -static bool fuse_range_is_writeback(struct inode *inode, pgoff_t idx_from, > - pgoff_t idx_to) > -{ > - struct fuse_inode *fi = get_fuse_inode(inode); > - bool found; > - > - if (RB_EMPTY_ROOT(&fi->writepages)) > - return false; > - > - spin_lock(&fi->lock); > - found = fuse_find_writeback(fi, idx_from, idx_to); > - spin_unlock(&fi->lock); > - > - return found; > -} > - > -static inline bool fuse_page_is_writeback(struct inode *inode, pgoff_t index) > -{ > - return fuse_range_is_writeback(inode, index, index); > -} > - > -/* > - * Wait for page writeback to be completed. > - * > - * Since fuse doesn't rely on the VM writeback tracking, this has to > - * use some other means. > - */ > -static void fuse_wait_on_page_writeback(struct inode *inode, pgoff_t index) > -{ > - struct fuse_inode *fi = get_fuse_inode(inode); > - > - wait_event(fi->page_waitq, !fuse_page_is_writeback(inode, index)); > -} > - > -static inline bool fuse_folio_is_writeback(struct inode *inode, > - struct folio *folio) > -{ > - pgoff_t last = folio_next_index(folio) - 1; > - return fuse_range_is_writeback(inode, folio_index(folio), last); > -} > - > -static void fuse_wait_on_folio_writeback(struct inode *inode, > - struct folio *folio) > -{ > - struct fuse_inode *fi = get_fuse_inode(inode); > - > - wait_event(fi->page_waitq, !fuse_folio_is_writeback(inode, folio)); > -} > - > /* > * Wait for all pending writepages on the inode to finish. > * > @@ -886,13 +808,6 @@ static int fuse_do_readfolio(struct file *file, struct folio *folio) > ssize_t res; > u64 attr_ver; > > - /* > - * With the temporary pages that are used to complete writeback, we can > - * have writeback that extends beyond the lifetime of the folio. So > - * make sure we read a properly synced folio. > - */ > - fuse_wait_on_folio_writeback(inode, folio); > - > attr_ver = fuse_get_attr_version(fm->fc); > > /* Don't overflow end offset */ > @@ -1003,17 +918,12 @@ static void fuse_send_readpages(struct fuse_io_args *ia, struct file *file) > static void fuse_readahead(struct readahead_control *rac) > { > struct inode *inode = rac->mapping->host; > - struct fuse_inode *fi = get_fuse_inode(inode); > struct fuse_conn *fc = get_fuse_conn(inode); > unsigned int max_pages, nr_pages; > - pgoff_t first = readahead_index(rac); > - pgoff_t last = first + readahead_count(rac) - 1; > > if (fuse_is_bad(inode)) > return; > > - wait_event(fi->page_waitq, !fuse_range_is_writeback(inode, first, last)); > - > max_pages = min_t(unsigned int, fc->max_pages, > fc->max_read / PAGE_SIZE); > > @@ -1172,7 +1082,7 @@ static ssize_t fuse_send_write_pages(struct fuse_io_args *ia, > int err; > > for (i = 0; i < ap->num_folios; i++) > - fuse_wait_on_folio_writeback(inode, ap->folios[i]); > + folio_wait_writeback(ap->folios[i]); > > fuse_write_args_fill(ia, ff, pos, count); > ia->write.in.flags = fuse_write_flags(iocb); > @@ -1622,7 +1532,7 @@ ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter, > return res; > } > } > - if (!cuse && fuse_range_is_writeback(inode, idx_from, idx_to)) { > + if (!cuse && filemap_range_has_writeback(mapping, pos, (pos + count - 1))) { > if (!write) > inode_lock(inode); > fuse_sync_writes(inode); > @@ -1825,7 +1735,7 @@ static void fuse_writepage_free(struct fuse_writepage_args *wpa) > fuse_sync_bucket_dec(wpa->bucket); > > for (i = 0; i < ap->num_folios; i++) > - folio_put(ap->folios[i]); > + folio_end_writeback(ap->folios[i]); I noticed that if we folio_end_writeback() in fuse_writepage_finish() (rather than fuse_writepage_free()), there's ~50% buffer write bandwridth performance gain (5500MB -> 8500MB)[*] The fuse server is generally implemented in multi-thread style, and multi (fuse server) worker threads could fetch and process FUSE_WRITE requests of one fuse inode. Then there's serious lock contention for the xarray lock (of the address space) when these multi worker threads call fuse_writepage_end->folio_end_writeback when they are sending replies of FUSE_WRITE requests. The lock contention is greatly alleviated when folio_end_writeback() is serialized with fi->lock. IOWs in the current implementation (folio_end_writeback() in fuse_writepage_free()), each worker thread needs to compete for the xarray lock for 256 times (one fuse request can contain at most 256 pages if FUSE_MAX_MAX_PAGES is 256) when completing a FUSE_WRITE request. After moving folio_end_writeback() to fuse_writepage_finish(), each worker thread needs to compete for fi->lock only once. IOWs the locking granularity is larger now. > @@ -2367,54 +2111,23 @@ static int fuse_writepages_fill(struct folio *folio, > data->wpa = NULL; > } > > - err = -ENOMEM; > - tmp_folio = folio_alloc(GFP_NOFS | __GFP_HIGHMEM, 0); > - if (!tmp_folio) > - goto out_unlock; > - > - /* > - * The page must not be redirtied until the writeout is completed > - * (i.e. userspace has sent a reply to the write request). Otherwise > - * there could be more than one temporary page instance for each real > - * page. > - * > - * This is ensured by holding the page lock in page_mkwrite() while > - * checking fuse_page_is_writeback(). We already hold the page lock > - * since clear_page_dirty_for_io() and keep it held until we add the > - * request to the fi->writepages list and increment ap->num_folios. > - * After this fuse_page_is_writeback() will indicate that the page is > - * under writeback, so we can release the page lock. > - */ > if (data->wpa == NULL) { > err = -ENOMEM; > wpa = fuse_writepage_args_setup(folio, data->ff); > - if (!wpa) { > - folio_put(tmp_folio); > + if (!wpa) > goto out_unlock; > - } > fuse_file_get(wpa->ia.ff); > data->max_folios = 1; > ap = &wpa->ia.ap; > } > folio_start_writeback(folio); There's also a lock contention for the xarray lock when calling folio_start_writeback(). I also noticed a strange thing that, if we lock fi->lock and unlock immediately, the write bandwidth improves by 5% (8500MB -> 9000MB). The palce where to insert the "locking fi->lock and unlocking" actually doesn't matter. "perf lock contention" shows the lock contention for the xarray lock is greatly alleviated, though I can't understand how it is done quite well... As the performance gain is not significant (~5%), I think we can leave this stange phenomenon aside for now. [*] test case: ./passthrough_hp --bypass-rw 2 /tmp /mnt (testbench mode in https://github.com/libfuse/libfuse/pull/807/commits/e83789cc6e83ca42ccc9899c4f7f8c69f31cbff9 bypass the buffer copy along with the persistence procedure) fio -fallocate=0 -numjobs=32 -iodepth=1 -ioengine=sync -sync=0 --direct=0 -rw=write -bs=1M -size=100G --time_based --runtime=300 -directory=/mnt/ --group_reporting --name=Fio
On Mon, Nov 18, 2024 at 11:59 PM Jingbo Xu <jefflexu@linux.alibaba.com> wrote: > > On 11/16/24 6:44 AM, Joanne Koong wrote: > > > @@ -1838,7 +1748,7 @@ static void fuse_writepage_finish_stat(struct inode *inode, struct folio *folio) > > struct backing_dev_info *bdi = inode_to_bdi(inode); > > > > dec_wb_stat(&bdi->wb, WB_WRITEBACK); > > - node_stat_sub_folio(folio, NR_WRITEBACK_TEMP); > > + node_stat_sub_folio(folio, NR_WRITEBACK); > > Now fuse_writepage_finish_stat() has only one caller and we could make > it embedded into its only caller. I'll make this change in v6. > > > > static void fuse_writepage_args_page_fill(struct fuse_writepage_args *wpa, struct folio *folio, > > - struct folio *tmp_folio, uint32_t folio_index) > > + uint32_t folio_index) > > { > > struct inode *inode = folio->mapping->host; > > struct fuse_args_pages *ap = &wpa->ia.ap; > > > > - folio_copy(tmp_folio, folio); > > - > > - ap->folios[folio_index] = tmp_folio; > > + ap->folios[folio_index] = folio; > > ap->descs[folio_index].offset = 0; > > ap->descs[folio_index].length = PAGE_SIZE; > > > > inc_wb_stat(&inode_to_bdi(inode)->wb, WB_WRITEBACK); > > - node_stat_add_folio(tmp_folio, NR_WRITEBACK_TEMP); > > + node_stat_add_folio(folio, NR_WRITEBACK); > > This inc NR_WRITEBACK counter along with the corresponding dec > NR_WRITEBACK counter in fuse_writepage_finish_stat() seems unnecessary, > as folio_start_writeback() will increase the NR_WRITEBACK counter, while > folio_end_writeback() will decrease the NR_WRITEBACK counter. > Nice find, I'll make this change in v6. Thanks, Joanne > > -- > Thanks, > Jingbo
On Wed, Nov 20, 2024 at 1:56 AM Jingbo Xu <jefflexu@linux.alibaba.com> wrote: > > On 11/16/24 6:44 AM, Joanne Koong wrote: > > In the current FUSE writeback design (see commit 3be5a52b30aa > > ("fuse: support writable mmap")), a temp page is allocated for every > > dirty page to be written back, the contents of the dirty page are copied over > > to the temp page, and the temp page gets handed to the server to write back. > > > > This is done so that writeback may be immediately cleared on the dirty page, > > and this in turn is done for two reasons: > > a) in order to mitigate the following deadlock scenario that may arise > > if reclaim waits on writeback on the dirty page to complete: > > * single-threaded FUSE server is in the middle of handling a request > > that needs a memory allocation > > * memory allocation triggers direct reclaim > > * direct reclaim waits on a folio under writeback > > * the FUSE server can't write back the folio since it's stuck in > > direct reclaim > > b) in order to unblock internal (eg sync, page compaction) waits on > > writeback without needing the server to complete writing back to disk, > > which may take an indeterminate amount of time. > > > > With a recent change that added AS_WRITEBACK_INDETERMINATE and mitigates > > the situations described above, FUSE writeback does not need to use > > temp pages if it sets AS_WRITEBACK_INDETERMINATE on its inode mappings. > > > > This commit sets AS_WRITEBACK_INDETERMINATE on the inode mappings > > and removes the temporary pages + extra copying and the internal rb > > tree. > > > > fio benchmarks -- > > (using averages observed from 10 runs, throwing away outliers) > > > > Setup: > > sudo mount -t tmpfs -o size=30G tmpfs ~/tmp_mount > > ./libfuse/build/example/passthrough_ll -o writeback -o max_threads=4 -o source=~/tmp_mount ~/fuse_mount > > > > fio --name=writeback --ioengine=sync --rw=write --bs={1k,4k,1M} --size=2G > > --numjobs=2 --ramp_time=30 --group_reporting=1 --directory=/root/fuse_mount > > > > bs = 1k 4k 1M > > Before 351 MiB/s 1818 MiB/s 1851 MiB/s > > After 341 MiB/s 2246 MiB/s 2685 MiB/s > > % diff -3% 23% 45% > > > > Signed-off-by: Joanne Koong <joannelkoong@gmail.com> > > --- > > fs/fuse/file.c | 339 +++---------------------------------------------- > > 1 file changed, 20 insertions(+), 319 deletions(-) > > > > diff --git a/fs/fuse/file.c b/fs/fuse/file.c > > index 88d0946b5bc9..56289ac58596 100644 > > --- a/fs/fuse/file.c > > +++ b/fs/fuse/file.c > > @@ -1172,7 +1082,7 @@ static ssize_t fuse_send_write_pages(struct fuse_io_args *ia, > > int err; > > > > for (i = 0; i < ap->num_folios; i++) > > - fuse_wait_on_folio_writeback(inode, ap->folios[i]); > > + folio_wait_writeback(ap->folios[i]); > > > > fuse_write_args_fill(ia, ff, pos, count); > > ia->write.in.flags = fuse_write_flags(iocb); > > @@ -1622,7 +1532,7 @@ ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter, > > return res; > > } > > } > > - if (!cuse && fuse_range_is_writeback(inode, idx_from, idx_to)) { > > + if (!cuse && filemap_range_has_writeback(mapping, pos, (pos + count - 1))) { > > if (!write) > > inode_lock(inode); > > fuse_sync_writes(inode); > > @@ -1825,7 +1735,7 @@ static void fuse_writepage_free(struct fuse_writepage_args *wpa) > > fuse_sync_bucket_dec(wpa->bucket); > > > > for (i = 0; i < ap->num_folios; i++) > > - folio_put(ap->folios[i]); > > + folio_end_writeback(ap->folios[i]); > > I noticed that if we folio_end_writeback() in fuse_writepage_finish() > (rather than fuse_writepage_free()), there's ~50% buffer write > bandwridth performance gain (5500MB -> 8500MB)[*] > > The fuse server is generally implemented in multi-thread style, and > multi (fuse server) worker threads could fetch and process FUSE_WRITE > requests of one fuse inode. Then there's serious lock contention for > the xarray lock (of the address space) when these multi worker threads > call fuse_writepage_end->folio_end_writeback when they are sending > replies of FUSE_WRITE requests. > > The lock contention is greatly alleviated when folio_end_writeback() is > serialized with fi->lock. IOWs in the current implementation > (folio_end_writeback() in fuse_writepage_free()), each worker thread > needs to compete for the xarray lock for 256 times (one fuse request can > contain at most 256 pages if FUSE_MAX_MAX_PAGES is 256) when completing > a FUSE_WRITE request. > > After moving folio_end_writeback() to fuse_writepage_finish(), each > worker thread needs to compete for fi->lock only once. IOWs the locking > granularity is larger now. > Interesting! Thanks for sharing. Are you able to consistently repro these results and on different machines? When I run it locally on my machine using the commands you shared, I'm seeing roughly the same throughput: Current implementation (folio_end_writeback() in fuse_writepage_free()): WRITE: bw=385MiB/s (404MB/s), 385MiB/s-385MiB/s (404MB/s-404MB/s), io=113GiB (121GB), run=300177-300177msec WRITE: bw=384MiB/s (403MB/s), 384MiB/s-384MiB/s (403MB/s-403MB/s), io=113GiB (121GB), run=300178-300178msec fuse_end_writeback() in fuse_writepage_finish(): WRITE: bw=387MiB/s (406MB/s), 387MiB/s-387MiB/s (406MB/s-406MB/s), io=113GiB (122GB), run=300165-300165msec WRITE: bw=381MiB/s (399MB/s), 381MiB/s-381MiB/s (399MB/s-399MB/s), io=112GiB (120GB), run=300143-300143msec I wonder if it's because your machine is so much faster that lock contention makes a difference for you whereas on my machine there's other things that slow it down before lock contention comes into play. I see your point about why it would make sense that having folio_end_writeback() in fuse_writepage_finish() inside the scope of the fi->lock could make it faster, but I also could see how having it outside the lock could make it faster as well. I'm thinking about the scenario where if there's 8 threads all executing fuse_send_writepage() at the same time, calling folio_end_writeback() outside the fi->lock would unblock other threads trying to get the fi->lock and that other thread could execute while folio_end_writeback() gets executed. Looking at it some more, it seems like it'd be useful if there was some equivalent api to folio_end_writeback() that takes in an array of folios and would only need to grab the xarray lock once to clear writeback on all the folios in the array. When fuse supports large folios [*] this will help lock contention on the xarray lock as well because there'll be less folio_end_writeback() calls. I'm happy to move the fuse_end_writeback() call to fuse_writepage_finish() considering what you're seeing. 5500 Mb -> 8800 Mb is a huge perf improvement! [*] https://lore.kernel.org/linux-fsdevel/20241109001258.2216604-1-joannelkoong@gmail.com/ > > > > @@ -2367,54 +2111,23 @@ static int fuse_writepages_fill(struct folio *folio, > > data->wpa = NULL; > > } > > > > - err = -ENOMEM; > > - tmp_folio = folio_alloc(GFP_NOFS | __GFP_HIGHMEM, 0); > > - if (!tmp_folio) > > - goto out_unlock; > > - > > - /* > > - * The page must not be redirtied until the writeout is completed > > - * (i.e. userspace has sent a reply to the write request). Otherwise > > - * there could be more than one temporary page instance for each real > > - * page. > > - * > > - * This is ensured by holding the page lock in page_mkwrite() while > > - * checking fuse_page_is_writeback(). We already hold the page lock > > - * since clear_page_dirty_for_io() and keep it held until we add the > > - * request to the fi->writepages list and increment ap->num_folios. > > - * After this fuse_page_is_writeback() will indicate that the page is > > - * under writeback, so we can release the page lock. > > - */ > > if (data->wpa == NULL) { > > err = -ENOMEM; > > wpa = fuse_writepage_args_setup(folio, data->ff); > > - if (!wpa) { > > - folio_put(tmp_folio); > > + if (!wpa) > > goto out_unlock; > > - } > > fuse_file_get(wpa->ia.ff); > > data->max_folios = 1; > > ap = &wpa->ia.ap; > > } > > folio_start_writeback(folio); > > There's also a lock contention for the xarray lock when calling > folio_start_writeback(). > > I also noticed a strange thing that, if we lock fi->lock and unlock > immediately, the write bandwidth improves by 5% (8500MB -> 9000MB). The Interesting! By lock fi->lock and unlock immediately, do you mean locking it, then unlocking it, then calling folio_start_writeback() or locking it, calling folio_start_writeback() and then unlocking it? Thanks, Joanne > palce where to insert the "locking fi->lock and unlocking" actually > doesn't matter. "perf lock contention" shows the lock contention for > the xarray lock is greatly alleviated, though I can't understand how it > is done quite well... > > As the performance gain is not significant (~5%), I think we can leave > this stange phenomenon aside for now. > > > > [*] test case: > ./passthrough_hp --bypass-rw 2 /tmp /mnt > (testbench mode in > https://github.com/libfuse/libfuse/pull/807/commits/e83789cc6e83ca42ccc9899c4f7f8c69f31cbff9 > bypass the buffer copy along with the persistence procedure) > > fio -fallocate=0 -numjobs=32 -iodepth=1 -ioengine=sync -sync=0 > --direct=0 -rw=write -bs=1M -size=100G --time_based --runtime=300 > -directory=/mnt/ --group_reporting --name=Fio > -- > Thanks, > Jingbo
On 11/21/24 5:53 AM, Joanne Koong wrote: > On Wed, Nov 20, 2024 at 1:56 AM Jingbo Xu <jefflexu@linux.alibaba.com> wrote: >> >> On 11/16/24 6:44 AM, Joanne Koong wrote: >>> In the current FUSE writeback design (see commit 3be5a52b30aa >>> ("fuse: support writable mmap")), a temp page is allocated for every >>> dirty page to be written back, the contents of the dirty page are copied over >>> to the temp page, and the temp page gets handed to the server to write back. >>> >>> This is done so that writeback may be immediately cleared on the dirty page, >>> and this in turn is done for two reasons: >>> a) in order to mitigate the following deadlock scenario that may arise >>> if reclaim waits on writeback on the dirty page to complete: >>> * single-threaded FUSE server is in the middle of handling a request >>> that needs a memory allocation >>> * memory allocation triggers direct reclaim >>> * direct reclaim waits on a folio under writeback >>> * the FUSE server can't write back the folio since it's stuck in >>> direct reclaim >>> b) in order to unblock internal (eg sync, page compaction) waits on >>> writeback without needing the server to complete writing back to disk, >>> which may take an indeterminate amount of time. >>> >>> With a recent change that added AS_WRITEBACK_INDETERMINATE and mitigates >>> the situations described above, FUSE writeback does not need to use >>> temp pages if it sets AS_WRITEBACK_INDETERMINATE on its inode mappings. >>> >>> This commit sets AS_WRITEBACK_INDETERMINATE on the inode mappings >>> and removes the temporary pages + extra copying and the internal rb >>> tree. >>> >>> fio benchmarks -- >>> (using averages observed from 10 runs, throwing away outliers) >>> >>> Setup: >>> sudo mount -t tmpfs -o size=30G tmpfs ~/tmp_mount >>> ./libfuse/build/example/passthrough_ll -o writeback -o max_threads=4 -o source=~/tmp_mount ~/fuse_mount >>> >>> fio --name=writeback --ioengine=sync --rw=write --bs={1k,4k,1M} --size=2G >>> --numjobs=2 --ramp_time=30 --group_reporting=1 --directory=/root/fuse_mount >>> >>> bs = 1k 4k 1M >>> Before 351 MiB/s 1818 MiB/s 1851 MiB/s >>> After 341 MiB/s 2246 MiB/s 2685 MiB/s >>> % diff -3% 23% 45% >>> >>> Signed-off-by: Joanne Koong <joannelkoong@gmail.com> >>> --- >>> fs/fuse/file.c | 339 +++---------------------------------------------- >>> 1 file changed, 20 insertions(+), 319 deletions(-) >>> >>> diff --git a/fs/fuse/file.c b/fs/fuse/file.c >>> index 88d0946b5bc9..56289ac58596 100644 >>> --- a/fs/fuse/file.c >>> +++ b/fs/fuse/file.c >>> @@ -1172,7 +1082,7 @@ static ssize_t fuse_send_write_pages(struct fuse_io_args *ia, >>> int err; >>> >>> for (i = 0; i < ap->num_folios; i++) >>> - fuse_wait_on_folio_writeback(inode, ap->folios[i]); >>> + folio_wait_writeback(ap->folios[i]); >>> >>> fuse_write_args_fill(ia, ff, pos, count); >>> ia->write.in.flags = fuse_write_flags(iocb); >>> @@ -1622,7 +1532,7 @@ ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter, >>> return res; >>> } >>> } >>> - if (!cuse && fuse_range_is_writeback(inode, idx_from, idx_to)) { >>> + if (!cuse && filemap_range_has_writeback(mapping, pos, (pos + count - 1))) { >>> if (!write) >>> inode_lock(inode); >>> fuse_sync_writes(inode); >>> @@ -1825,7 +1735,7 @@ static void fuse_writepage_free(struct fuse_writepage_args *wpa) >>> fuse_sync_bucket_dec(wpa->bucket); >>> >>> for (i = 0; i < ap->num_folios; i++) >>> - folio_put(ap->folios[i]); >>> + folio_end_writeback(ap->folios[i]); >> >> I noticed that if we folio_end_writeback() in fuse_writepage_finish() >> (rather than fuse_writepage_free()), there's ~50% buffer write >> bandwridth performance gain (5500MB -> 8500MB)[*] >> >> The fuse server is generally implemented in multi-thread style, and >> multi (fuse server) worker threads could fetch and process FUSE_WRITE >> requests of one fuse inode. Then there's serious lock contention for >> the xarray lock (of the address space) when these multi worker threads >> call fuse_writepage_end->folio_end_writeback when they are sending >> replies of FUSE_WRITE requests. >> >> The lock contention is greatly alleviated when folio_end_writeback() is >> serialized with fi->lock. IOWs in the current implementation >> (folio_end_writeback() in fuse_writepage_free()), each worker thread >> needs to compete for the xarray lock for 256 times (one fuse request can >> contain at most 256 pages if FUSE_MAX_MAX_PAGES is 256) when completing >> a FUSE_WRITE request. >> >> After moving folio_end_writeback() to fuse_writepage_finish(), each >> worker thread needs to compete for fi->lock only once. IOWs the locking >> granularity is larger now. >> > > Interesting! Thanks for sharing. Are you able to consistently repro > these results and on different machines? When I run it locally on my > machine using the commands you shared, I'm seeing roughly the same > throughput: > > Current implementation (folio_end_writeback() in fuse_writepage_free()): > WRITE: bw=385MiB/s (404MB/s), 385MiB/s-385MiB/s (404MB/s-404MB/s), > io=113GiB (121GB), run=300177-300177msec > WRITE: bw=384MiB/s (403MB/s), 384MiB/s-384MiB/s (403MB/s-403MB/s), > io=113GiB (121GB), run=300178-300178msec > > fuse_end_writeback() in fuse_writepage_finish(): > WRITE: bw=387MiB/s (406MB/s), 387MiB/s-387MiB/s (406MB/s-406MB/s), > io=113GiB (122GB), run=300165-300165msec > WRITE: bw=381MiB/s (399MB/s), 381MiB/s-381MiB/s (399MB/s-399MB/s), > io=112GiB (120GB), run=300143-300143msec > > I wonder if it's because your machine is so much faster that lock > contention makes a difference for you whereas on my machine there's > other things that slow it down before lock contention comes into play. Yeah, I agree that the lock contention matters only when the writeback kworker consumes 100% CPU, i.e. when the writeback kworker is the bottleneck. To expose that, the passthrough_hp daemon works in benchmark[*] mode (I noticed that passthrough_hp can be the bottleneck when disabling "--bypass-rw" mode). [*] https://github.com/libfuse/libfuse/pull/807/commits/e83789cc6e83ca42ccc9899c4f7f8c69f31cbff9 > > I see your point about why it would make sense that having > folio_end_writeback() in fuse_writepage_finish() inside the scope of > the fi->lock could make it faster, but I also could see how having it > outside the lock could make it faster as well. I'm thinking about the > scenario where if there's 8 threads all executing > fuse_send_writepage() at the same time, calling folio_end_writeback() > outside the fi->lock would unblock other threads trying to get the > fi->lock and that other thread could execute while > folio_end_writeback() gets executed. > > Looking at it some more, it seems like it'd be useful if there was > some equivalent api to folio_end_writeback() that takes in an array of > folios and would only need to grab the xarray lock once to clear > writeback on all the folios in the array. Yes it's exactly what we need. > > When fuse supports large folios [*] this will help lock contention on > the xarray lock as well because there'll be less folio_end_writeback() > calls. Cool, it definitely helps. > > I'm happy to move the fuse_end_writeback() call to > fuse_writepage_finish() considering what you're seeing. 5500 Mb -> > 8800 Mb is a huge perf improvement! This statistics is tested in benchmark ("--bypass-rw") mode. When disabling "--bypass-rw" mode and testing fuse passthrough_hp over a ext4 over nvme, the performance gain is ~10% (4009MB/s ->4428MB/s). >>> @@ -2367,54 +2111,23 @@ static int fuse_writepages_fill(struct folio *folio, >>> data->wpa = NULL; >>> } >>> >>> - err = -ENOMEM; >>> - tmp_folio = folio_alloc(GFP_NOFS | __GFP_HIGHMEM, 0); >>> - if (!tmp_folio) >>> - goto out_unlock; >>> - >>> - /* >>> - * The page must not be redirtied until the writeout is completed >>> - * (i.e. userspace has sent a reply to the write request). Otherwise >>> - * there could be more than one temporary page instance for each real >>> - * page. >>> - * >>> - * This is ensured by holding the page lock in page_mkwrite() while >>> - * checking fuse_page_is_writeback(). We already hold the page lock >>> - * since clear_page_dirty_for_io() and keep it held until we add the >>> - * request to the fi->writepages list and increment ap->num_folios. >>> - * After this fuse_page_is_writeback() will indicate that the page is >>> - * under writeback, so we can release the page lock. >>> - */ >>> if (data->wpa == NULL) { >>> err = -ENOMEM; >>> wpa = fuse_writepage_args_setup(folio, data->ff); >>> - if (!wpa) { >>> - folio_put(tmp_folio); >>> + if (!wpa) >>> goto out_unlock; >>> - } >>> fuse_file_get(wpa->ia.ff); >>> data->max_folios = 1; >>> ap = &wpa->ia.ap; >>> } >>> folio_start_writeback(folio); >> >> There's also a lock contention for the xarray lock when calling >> folio_start_writeback(). >> >> I also noticed a strange thing that, if we lock fi->lock and unlock >> immediately, the write bandwidth improves by 5% (8500MB -> 9000MB). The > > Interesting! By lock fi->lock and unlock immediately, do you mean > locking it, then unlocking it, then calling folio_start_writeback() or > locking it, calling folio_start_writeback() and then unlocking it? Either way works, as long as we lock/unlock fi->lock in fuse_writepages_fill()... The lock contention is further alleviated when folio_start_writeback() is inside the critical area of fi->lock.
On 11/21/24 04:08, Jingbo Xu wrote: > > > On 11/21/24 5:53 AM, Joanne Koong wrote: >> On Wed, Nov 20, 2024 at 1:56 AM Jingbo Xu <jefflexu@linux.alibaba.com> wrote: >>> >>> On 11/16/24 6:44 AM, Joanne Koong wrote: >>>> In the current FUSE writeback design (see commit 3be5a52b30aa >>>> ("fuse: support writable mmap")), a temp page is allocated for every >>>> dirty page to be written back, the contents of the dirty page are copied over >>>> to the temp page, and the temp page gets handed to the server to write back. >>>> >>>> This is done so that writeback may be immediately cleared on the dirty page, >>>> and this in turn is done for two reasons: >>>> a) in order to mitigate the following deadlock scenario that may arise >>>> if reclaim waits on writeback on the dirty page to complete: >>>> * single-threaded FUSE server is in the middle of handling a request >>>> that needs a memory allocation >>>> * memory allocation triggers direct reclaim >>>> * direct reclaim waits on a folio under writeback >>>> * the FUSE server can't write back the folio since it's stuck in >>>> direct reclaim >>>> b) in order to unblock internal (eg sync, page compaction) waits on >>>> writeback without needing the server to complete writing back to disk, >>>> which may take an indeterminate amount of time. >>>> >>>> With a recent change that added AS_WRITEBACK_INDETERMINATE and mitigates >>>> the situations described above, FUSE writeback does not need to use >>>> temp pages if it sets AS_WRITEBACK_INDETERMINATE on its inode mappings. >>>> >>>> This commit sets AS_WRITEBACK_INDETERMINATE on the inode mappings >>>> and removes the temporary pages + extra copying and the internal rb >>>> tree. >>>> >>>> fio benchmarks -- >>>> (using averages observed from 10 runs, throwing away outliers) >>>> >>>> Setup: >>>> sudo mount -t tmpfs -o size=30G tmpfs ~/tmp_mount >>>> ./libfuse/build/example/passthrough_ll -o writeback -o max_threads=4 -o source=~/tmp_mount ~/fuse_mount >>>> >>>> fio --name=writeback --ioengine=sync --rw=write --bs={1k,4k,1M} --size=2G >>>> --numjobs=2 --ramp_time=30 --group_reporting=1 --directory=/root/fuse_mount >>>> >>>> bs = 1k 4k 1M >>>> Before 351 MiB/s 1818 MiB/s 1851 MiB/s >>>> After 341 MiB/s 2246 MiB/s 2685 MiB/s >>>> % diff -3% 23% 45% >>>> >>>> Signed-off-by: Joanne Koong <joannelkoong@gmail.com> >>>> --- >>>> fs/fuse/file.c | 339 +++---------------------------------------------- >>>> 1 file changed, 20 insertions(+), 319 deletions(-) >>>> >>>> diff --git a/fs/fuse/file.c b/fs/fuse/file.c >>>> index 88d0946b5bc9..56289ac58596 100644 >>>> --- a/fs/fuse/file.c >>>> +++ b/fs/fuse/file.c >>>> @@ -1172,7 +1082,7 @@ static ssize_t fuse_send_write_pages(struct fuse_io_args *ia, >>>> int err; >>>> >>>> for (i = 0; i < ap->num_folios; i++) >>>> - fuse_wait_on_folio_writeback(inode, ap->folios[i]); >>>> + folio_wait_writeback(ap->folios[i]); >>>> >>>> fuse_write_args_fill(ia, ff, pos, count); >>>> ia->write.in.flags = fuse_write_flags(iocb); >>>> @@ -1622,7 +1532,7 @@ ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter, >>>> return res; >>>> } >>>> } >>>> - if (!cuse && fuse_range_is_writeback(inode, idx_from, idx_to)) { >>>> + if (!cuse && filemap_range_has_writeback(mapping, pos, (pos + count - 1))) { >>>> if (!write) >>>> inode_lock(inode); >>>> fuse_sync_writes(inode); >>>> @@ -1825,7 +1735,7 @@ static void fuse_writepage_free(struct fuse_writepage_args *wpa) >>>> fuse_sync_bucket_dec(wpa->bucket); >>>> >>>> for (i = 0; i < ap->num_folios; i++) >>>> - folio_put(ap->folios[i]); >>>> + folio_end_writeback(ap->folios[i]); >>> >>> I noticed that if we folio_end_writeback() in fuse_writepage_finish() >>> (rather than fuse_writepage_free()), there's ~50% buffer write >>> bandwridth performance gain (5500MB -> 8500MB)[*] >>> >>> The fuse server is generally implemented in multi-thread style, and >>> multi (fuse server) worker threads could fetch and process FUSE_WRITE >>> requests of one fuse inode. Then there's serious lock contention for >>> the xarray lock (of the address space) when these multi worker threads >>> call fuse_writepage_end->folio_end_writeback when they are sending >>> replies of FUSE_WRITE requests. >>> >>> The lock contention is greatly alleviated when folio_end_writeback() is >>> serialized with fi->lock. IOWs in the current implementation >>> (folio_end_writeback() in fuse_writepage_free()), each worker thread >>> needs to compete for the xarray lock for 256 times (one fuse request can >>> contain at most 256 pages if FUSE_MAX_MAX_PAGES is 256) when completing >>> a FUSE_WRITE request. >>> >>> After moving folio_end_writeback() to fuse_writepage_finish(), each >>> worker thread needs to compete for fi->lock only once. IOWs the locking >>> granularity is larger now. >>> >> >> Interesting! Thanks for sharing. Are you able to consistently repro >> these results and on different machines? When I run it locally on my >> machine using the commands you shared, I'm seeing roughly the same >> throughput: >> >> Current implementation (folio_end_writeback() in fuse_writepage_free()): >> WRITE: bw=385MiB/s (404MB/s), 385MiB/s-385MiB/s (404MB/s-404MB/s), >> io=113GiB (121GB), run=300177-300177msec >> WRITE: bw=384MiB/s (403MB/s), 384MiB/s-384MiB/s (403MB/s-403MB/s), >> io=113GiB (121GB), run=300178-300178msec >> >> fuse_end_writeback() in fuse_writepage_finish(): >> WRITE: bw=387MiB/s (406MB/s), 387MiB/s-387MiB/s (406MB/s-406MB/s), >> io=113GiB (122GB), run=300165-300165msec >> WRITE: bw=381MiB/s (399MB/s), 381MiB/s-381MiB/s (399MB/s-399MB/s), >> io=112GiB (120GB), run=300143-300143msec >> >> I wonder if it's because your machine is so much faster that lock >> contention makes a difference for you whereas on my machine there's >> other things that slow it down before lock contention comes into play. > > Yeah, I agree that the lock contention matters only when the writeback > kworker consumes 100% CPU, i.e. when the writeback kworker is the > bottleneck. To expose that, the passthrough_hp daemon works in > benchmark[*] mode (I noticed that passthrough_hp can be the bottleneck > when disabling "--bypass-rw" mode). > > [*] > https://github.com/libfuse/libfuse/pull/807/commits/e83789cc6e83ca42ccc9899c4f7f8c69f31cbff9 > > >> >> I see your point about why it would make sense that having >> folio_end_writeback() in fuse_writepage_finish() inside the scope of >> the fi->lock could make it faster, but I also could see how having it >> outside the lock could make it faster as well. I'm thinking about the >> scenario where if there's 8 threads all executing >> fuse_send_writepage() at the same time, calling folio_end_writeback() >> outside the fi->lock would unblock other threads trying to get the >> fi->lock and that other thread could execute while >> folio_end_writeback() gets executed. >> >> Looking at it some more, it seems like it'd be useful if there was >> some equivalent api to folio_end_writeback() that takes in an array of >> folios and would only need to grab the xarray lock once to clear >> writeback on all the folios in the array. > > Yes it's exactly what we need. > > >> >> When fuse supports large folios [*] this will help lock contention on >> the xarray lock as well because there'll be less folio_end_writeback() >> calls. > > Cool, it definitely helps. > > >> >> I'm happy to move the fuse_end_writeback() call to >> fuse_writepage_finish() considering what you're seeing. 5500 Mb -> >> 8800 Mb is a huge perf improvement! > > This statistics is tested in benchmark ("--bypass-rw") mode. When > disabling "--bypass-rw" mode and testing fuse passthrough_hp over a ext4 > over nvme, the performance gain is ~10% (4009MB/s ->4428MB/s). Thanks for Jingbo and Joanne for looking into this! In typical HPC case one expects currently >15GB/s for a single client - Infiniband RDMA and possibly multi rail. The --bypass-rw is rather close to that. Thanks, Bernd
diff --git a/fs/fuse/file.c b/fs/fuse/file.c index 88d0946b5bc9..56289ac58596 100644 --- a/fs/fuse/file.c +++ b/fs/fuse/file.c @@ -415,89 +415,11 @@ u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id) struct fuse_writepage_args { struct fuse_io_args ia; - struct rb_node writepages_entry; struct list_head queue_entry; - struct fuse_writepage_args *next; struct inode *inode; struct fuse_sync_bucket *bucket; }; -static struct fuse_writepage_args *fuse_find_writeback(struct fuse_inode *fi, - pgoff_t idx_from, pgoff_t idx_to) -{ - struct rb_node *n; - - n = fi->writepages.rb_node; - - while (n) { - struct fuse_writepage_args *wpa; - pgoff_t curr_index; - - wpa = rb_entry(n, struct fuse_writepage_args, writepages_entry); - WARN_ON(get_fuse_inode(wpa->inode) != fi); - curr_index = wpa->ia.write.in.offset >> PAGE_SHIFT; - if (idx_from >= curr_index + wpa->ia.ap.num_folios) - n = n->rb_right; - else if (idx_to < curr_index) - n = n->rb_left; - else - return wpa; - } - return NULL; -} - -/* - * Check if any page in a range is under writeback - */ -static bool fuse_range_is_writeback(struct inode *inode, pgoff_t idx_from, - pgoff_t idx_to) -{ - struct fuse_inode *fi = get_fuse_inode(inode); - bool found; - - if (RB_EMPTY_ROOT(&fi->writepages)) - return false; - - spin_lock(&fi->lock); - found = fuse_find_writeback(fi, idx_from, idx_to); - spin_unlock(&fi->lock); - - return found; -} - -static inline bool fuse_page_is_writeback(struct inode *inode, pgoff_t index) -{ - return fuse_range_is_writeback(inode, index, index); -} - -/* - * Wait for page writeback to be completed. - * - * Since fuse doesn't rely on the VM writeback tracking, this has to - * use some other means. - */ -static void fuse_wait_on_page_writeback(struct inode *inode, pgoff_t index) -{ - struct fuse_inode *fi = get_fuse_inode(inode); - - wait_event(fi->page_waitq, !fuse_page_is_writeback(inode, index)); -} - -static inline bool fuse_folio_is_writeback(struct inode *inode, - struct folio *folio) -{ - pgoff_t last = folio_next_index(folio) - 1; - return fuse_range_is_writeback(inode, folio_index(folio), last); -} - -static void fuse_wait_on_folio_writeback(struct inode *inode, - struct folio *folio) -{ - struct fuse_inode *fi = get_fuse_inode(inode); - - wait_event(fi->page_waitq, !fuse_folio_is_writeback(inode, folio)); -} - /* * Wait for all pending writepages on the inode to finish. * @@ -886,13 +808,6 @@ static int fuse_do_readfolio(struct file *file, struct folio *folio) ssize_t res; u64 attr_ver; - /* - * With the temporary pages that are used to complete writeback, we can - * have writeback that extends beyond the lifetime of the folio. So - * make sure we read a properly synced folio. - */ - fuse_wait_on_folio_writeback(inode, folio); - attr_ver = fuse_get_attr_version(fm->fc); /* Don't overflow end offset */ @@ -1003,17 +918,12 @@ static void fuse_send_readpages(struct fuse_io_args *ia, struct file *file) static void fuse_readahead(struct readahead_control *rac) { struct inode *inode = rac->mapping->host; - struct fuse_inode *fi = get_fuse_inode(inode); struct fuse_conn *fc = get_fuse_conn(inode); unsigned int max_pages, nr_pages; - pgoff_t first = readahead_index(rac); - pgoff_t last = first + readahead_count(rac) - 1; if (fuse_is_bad(inode)) return; - wait_event(fi->page_waitq, !fuse_range_is_writeback(inode, first, last)); - max_pages = min_t(unsigned int, fc->max_pages, fc->max_read / PAGE_SIZE); @@ -1172,7 +1082,7 @@ static ssize_t fuse_send_write_pages(struct fuse_io_args *ia, int err; for (i = 0; i < ap->num_folios; i++) - fuse_wait_on_folio_writeback(inode, ap->folios[i]); + folio_wait_writeback(ap->folios[i]); fuse_write_args_fill(ia, ff, pos, count); ia->write.in.flags = fuse_write_flags(iocb); @@ -1622,7 +1532,7 @@ ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter, return res; } } - if (!cuse && fuse_range_is_writeback(inode, idx_from, idx_to)) { + if (!cuse && filemap_range_has_writeback(mapping, pos, (pos + count - 1))) { if (!write) inode_lock(inode); fuse_sync_writes(inode); @@ -1825,7 +1735,7 @@ static void fuse_writepage_free(struct fuse_writepage_args *wpa) fuse_sync_bucket_dec(wpa->bucket); for (i = 0; i < ap->num_folios; i++) - folio_put(ap->folios[i]); + folio_end_writeback(ap->folios[i]); fuse_file_put(wpa->ia.ff, false); @@ -1838,7 +1748,7 @@ static void fuse_writepage_finish_stat(struct inode *inode, struct folio *folio) struct backing_dev_info *bdi = inode_to_bdi(inode); dec_wb_stat(&bdi->wb, WB_WRITEBACK); - node_stat_sub_folio(folio, NR_WRITEBACK_TEMP); + node_stat_sub_folio(folio, NR_WRITEBACK); wb_writeout_inc(&bdi->wb); } @@ -1861,7 +1771,6 @@ static void fuse_send_writepage(struct fuse_mount *fm, __releases(fi->lock) __acquires(fi->lock) { - struct fuse_writepage_args *aux, *next; struct fuse_inode *fi = get_fuse_inode(wpa->inode); struct fuse_write_in *inarg = &wpa->ia.write.in; struct fuse_args *args = &wpa->ia.ap.args; @@ -1898,19 +1807,8 @@ __acquires(fi->lock) out_free: fi->writectr--; - rb_erase(&wpa->writepages_entry, &fi->writepages); fuse_writepage_finish(wpa); spin_unlock(&fi->lock); - - /* After rb_erase() aux request list is private */ - for (aux = wpa->next; aux; aux = next) { - next = aux->next; - aux->next = NULL; - fuse_writepage_finish_stat(aux->inode, - aux->ia.ap.folios[0]); - fuse_writepage_free(aux); - } - fuse_writepage_free(wpa); spin_lock(&fi->lock); } @@ -1938,43 +1836,6 @@ __acquires(fi->lock) } } -static struct fuse_writepage_args *fuse_insert_writeback(struct rb_root *root, - struct fuse_writepage_args *wpa) -{ - pgoff_t idx_from = wpa->ia.write.in.offset >> PAGE_SHIFT; - pgoff_t idx_to = idx_from + wpa->ia.ap.num_folios - 1; - struct rb_node **p = &root->rb_node; - struct rb_node *parent = NULL; - - WARN_ON(!wpa->ia.ap.num_folios); - while (*p) { - struct fuse_writepage_args *curr; - pgoff_t curr_index; - - parent = *p; - curr = rb_entry(parent, struct fuse_writepage_args, - writepages_entry); - WARN_ON(curr->inode != wpa->inode); - curr_index = curr->ia.write.in.offset >> PAGE_SHIFT; - - if (idx_from >= curr_index + curr->ia.ap.num_folios) - p = &(*p)->rb_right; - else if (idx_to < curr_index) - p = &(*p)->rb_left; - else - return curr; - } - - rb_link_node(&wpa->writepages_entry, parent, p); - rb_insert_color(&wpa->writepages_entry, root); - return NULL; -} - -static void tree_insert(struct rb_root *root, struct fuse_writepage_args *wpa) -{ - WARN_ON(fuse_insert_writeback(root, wpa)); -} - static void fuse_writepage_end(struct fuse_mount *fm, struct fuse_args *args, int error) { @@ -1994,41 +1855,6 @@ static void fuse_writepage_end(struct fuse_mount *fm, struct fuse_args *args, if (!fc->writeback_cache) fuse_invalidate_attr_mask(inode, FUSE_STATX_MODIFY); spin_lock(&fi->lock); - rb_erase(&wpa->writepages_entry, &fi->writepages); - while (wpa->next) { - struct fuse_mount *fm = get_fuse_mount(inode); - struct fuse_write_in *inarg = &wpa->ia.write.in; - struct fuse_writepage_args *next = wpa->next; - - wpa->next = next->next; - next->next = NULL; - tree_insert(&fi->writepages, next); - - /* - * Skip fuse_flush_writepages() to make it easy to crop requests - * based on primary request size. - * - * 1st case (trivial): there are no concurrent activities using - * fuse_set/release_nowrite. Then we're on safe side because - * fuse_flush_writepages() would call fuse_send_writepage() - * anyway. - * - * 2nd case: someone called fuse_set_nowrite and it is waiting - * now for completion of all in-flight requests. This happens - * rarely and no more than once per page, so this should be - * okay. - * - * 3rd case: someone (e.g. fuse_do_setattr()) is in the middle - * of fuse_set_nowrite..fuse_release_nowrite section. The fact - * that fuse_set_nowrite returned implies that all in-flight - * requests were completed along with all of their secondary - * requests. Further primary requests are blocked by negative - * writectr. Hence there cannot be any in-flight requests and - * no invocations of fuse_writepage_end() while we're in - * fuse_set_nowrite..fuse_release_nowrite section. - */ - fuse_send_writepage(fm, next, inarg->offset + inarg->size); - } fi->writectr--; fuse_writepage_finish(wpa); spin_unlock(&fi->lock); @@ -2115,19 +1941,17 @@ static void fuse_writepage_add_to_bucket(struct fuse_conn *fc, } static void fuse_writepage_args_page_fill(struct fuse_writepage_args *wpa, struct folio *folio, - struct folio *tmp_folio, uint32_t folio_index) + uint32_t folio_index) { struct inode *inode = folio->mapping->host; struct fuse_args_pages *ap = &wpa->ia.ap; - folio_copy(tmp_folio, folio); - - ap->folios[folio_index] = tmp_folio; + ap->folios[folio_index] = folio; ap->descs[folio_index].offset = 0; ap->descs[folio_index].length = PAGE_SIZE; inc_wb_stat(&inode_to_bdi(inode)->wb, WB_WRITEBACK); - node_stat_add_folio(tmp_folio, NR_WRITEBACK_TEMP); + node_stat_add_folio(folio, NR_WRITEBACK); } static struct fuse_writepage_args *fuse_writepage_args_setup(struct folio *folio, @@ -2162,18 +1986,12 @@ static int fuse_writepage_locked(struct folio *folio) struct fuse_inode *fi = get_fuse_inode(inode); struct fuse_writepage_args *wpa; struct fuse_args_pages *ap; - struct folio *tmp_folio; struct fuse_file *ff; - int error = -ENOMEM; - - tmp_folio = folio_alloc(GFP_NOFS | __GFP_HIGHMEM, 0); - if (!tmp_folio) - goto err; + int error = -EIO; - error = -EIO; ff = fuse_write_file_get(fi); if (!ff) - goto err_nofile; + goto err; wpa = fuse_writepage_args_setup(folio, ff); error = -ENOMEM; @@ -2184,22 +2002,17 @@ static int fuse_writepage_locked(struct folio *folio) ap->num_folios = 1; folio_start_writeback(folio); - fuse_writepage_args_page_fill(wpa, folio, tmp_folio, 0); + fuse_writepage_args_page_fill(wpa, folio, 0); spin_lock(&fi->lock); - tree_insert(&fi->writepages, wpa); list_add_tail(&wpa->queue_entry, &fi->queued_writes); fuse_flush_writepages(inode); spin_unlock(&fi->lock); - folio_end_writeback(folio); - return 0; err_writepage_args: fuse_file_put(ff, false); -err_nofile: - folio_put(tmp_folio); err: mapping_set_error(folio->mapping, error); return error; @@ -2209,7 +2022,6 @@ struct fuse_fill_wb_data { struct fuse_writepage_args *wpa; struct fuse_file *ff; struct inode *inode; - struct folio **orig_folios; unsigned int max_folios; }; @@ -2244,69 +2056,11 @@ static void fuse_writepages_send(struct fuse_fill_wb_data *data) struct fuse_writepage_args *wpa = data->wpa; struct inode *inode = data->inode; struct fuse_inode *fi = get_fuse_inode(inode); - int num_folios = wpa->ia.ap.num_folios; - int i; spin_lock(&fi->lock); list_add_tail(&wpa->queue_entry, &fi->queued_writes); fuse_flush_writepages(inode); spin_unlock(&fi->lock); - - for (i = 0; i < num_folios; i++) - folio_end_writeback(data->orig_folios[i]); -} - -/* - * Check under fi->lock if the page is under writeback, and insert it onto the - * rb_tree if not. Otherwise iterate auxiliary write requests, to see if there's - * one already added for a page at this offset. If there's none, then insert - * this new request onto the auxiliary list, otherwise reuse the existing one by - * swapping the new temp page with the old one. - */ -static bool fuse_writepage_add(struct fuse_writepage_args *new_wpa, - struct folio *folio) -{ - struct fuse_inode *fi = get_fuse_inode(new_wpa->inode); - struct fuse_writepage_args *tmp; - struct fuse_writepage_args *old_wpa; - struct fuse_args_pages *new_ap = &new_wpa->ia.ap; - - WARN_ON(new_ap->num_folios != 0); - new_ap->num_folios = 1; - - spin_lock(&fi->lock); - old_wpa = fuse_insert_writeback(&fi->writepages, new_wpa); - if (!old_wpa) { - spin_unlock(&fi->lock); - return true; - } - - for (tmp = old_wpa->next; tmp; tmp = tmp->next) { - pgoff_t curr_index; - - WARN_ON(tmp->inode != new_wpa->inode); - curr_index = tmp->ia.write.in.offset >> PAGE_SHIFT; - if (curr_index == folio->index) { - WARN_ON(tmp->ia.ap.num_folios != 1); - swap(tmp->ia.ap.folios[0], new_ap->folios[0]); - break; - } - } - - if (!tmp) { - new_wpa->next = old_wpa->next; - old_wpa->next = new_wpa; - } - - spin_unlock(&fi->lock); - - if (tmp) { - fuse_writepage_finish_stat(new_wpa->inode, - folio); - fuse_writepage_free(new_wpa); - } - - return false; } static bool fuse_writepage_need_send(struct fuse_conn *fc, struct folio *folio, @@ -2315,15 +2069,6 @@ static bool fuse_writepage_need_send(struct fuse_conn *fc, struct folio *folio, { WARN_ON(!ap->num_folios); - /* - * Being under writeback is unlikely but possible. For example direct - * read to an mmaped fuse file will set the page dirty twice; once when - * the pages are faulted with get_user_pages(), and then after the read - * completed. - */ - if (fuse_folio_is_writeback(data->inode, folio)) - return true; - /* Reached max pages */ if (ap->num_folios == fc->max_pages) return true; @@ -2333,7 +2078,7 @@ static bool fuse_writepage_need_send(struct fuse_conn *fc, struct folio *folio, return true; /* Discontinuity */ - if (data->orig_folios[ap->num_folios - 1]->index + 1 != folio_index(folio)) + if (ap->folios[ap->num_folios - 1]->index + 1 != folio_index(folio)) return true; /* Need to grow the pages array? If so, did the expansion fail? */ @@ -2352,7 +2097,6 @@ static int fuse_writepages_fill(struct folio *folio, struct inode *inode = data->inode; struct fuse_inode *fi = get_fuse_inode(inode); struct fuse_conn *fc = get_fuse_conn(inode); - struct folio *tmp_folio; int err; if (!data->ff) { @@ -2367,54 +2111,23 @@ static int fuse_writepages_fill(struct folio *folio, data->wpa = NULL; } - err = -ENOMEM; - tmp_folio = folio_alloc(GFP_NOFS | __GFP_HIGHMEM, 0); - if (!tmp_folio) - goto out_unlock; - - /* - * The page must not be redirtied until the writeout is completed - * (i.e. userspace has sent a reply to the write request). Otherwise - * there could be more than one temporary page instance for each real - * page. - * - * This is ensured by holding the page lock in page_mkwrite() while - * checking fuse_page_is_writeback(). We already hold the page lock - * since clear_page_dirty_for_io() and keep it held until we add the - * request to the fi->writepages list and increment ap->num_folios. - * After this fuse_page_is_writeback() will indicate that the page is - * under writeback, so we can release the page lock. - */ if (data->wpa == NULL) { err = -ENOMEM; wpa = fuse_writepage_args_setup(folio, data->ff); - if (!wpa) { - folio_put(tmp_folio); + if (!wpa) goto out_unlock; - } fuse_file_get(wpa->ia.ff); data->max_folios = 1; ap = &wpa->ia.ap; } folio_start_writeback(folio); - fuse_writepage_args_page_fill(wpa, folio, tmp_folio, ap->num_folios); - data->orig_folios[ap->num_folios] = folio; + fuse_writepage_args_page_fill(wpa, folio, ap->num_folios); err = 0; - if (data->wpa) { - /* - * Protected by fi->lock against concurrent access by - * fuse_page_is_writeback(). - */ - spin_lock(&fi->lock); - ap->num_folios++; - spin_unlock(&fi->lock); - } else if (fuse_writepage_add(wpa, folio)) { + ap->num_folios++; + if (!data->wpa) data->wpa = wpa; - } else { - folio_end_writeback(folio); - } out_unlock: folio_unlock(folio); @@ -2441,13 +2154,6 @@ static int fuse_writepages(struct address_space *mapping, data.wpa = NULL; data.ff = NULL; - err = -ENOMEM; - data.orig_folios = kcalloc(fc->max_pages, - sizeof(struct folio *), - GFP_NOFS); - if (!data.orig_folios) - goto out; - err = write_cache_pages(mapping, wbc, fuse_writepages_fill, &data); if (data.wpa) { WARN_ON(!data.wpa->ia.ap.num_folios); @@ -2456,7 +2162,6 @@ static int fuse_writepages(struct address_space *mapping, if (data.ff) fuse_file_put(data.ff, false); - kfree(data.orig_folios); out: return err; } @@ -2481,8 +2186,6 @@ static int fuse_write_begin(struct file *file, struct address_space *mapping, if (IS_ERR(folio)) goto error; - fuse_wait_on_page_writeback(mapping->host, folio->index); - if (folio_test_uptodate(folio) || len >= folio_size(folio)) goto success; /* @@ -2545,13 +2248,9 @@ static int fuse_launder_folio(struct folio *folio) { int err = 0; if (folio_clear_dirty_for_io(folio)) { - struct inode *inode = folio->mapping->host; - - /* Serialize with pending writeback for the same page */ - fuse_wait_on_page_writeback(inode, folio->index); err = fuse_writepage_locked(folio); if (!err) - fuse_wait_on_page_writeback(inode, folio->index); + folio_wait_writeback(folio); } return err; } @@ -2595,7 +2294,7 @@ static vm_fault_t fuse_page_mkwrite(struct vm_fault *vmf) return VM_FAULT_NOPAGE; } - fuse_wait_on_folio_writeback(inode, folio); + folio_wait_writeback(folio); return VM_FAULT_LOCKED; } @@ -3413,9 +3112,12 @@ static const struct address_space_operations fuse_file_aops = { void fuse_init_file_inode(struct inode *inode, unsigned int flags) { struct fuse_inode *fi = get_fuse_inode(inode); + struct fuse_conn *fc = get_fuse_conn(inode); inode->i_fop = &fuse_file_operations; inode->i_data.a_ops = &fuse_file_aops; + if (fc->writeback_cache) + mapping_set_writeback_indeterminate(&inode->i_data); INIT_LIST_HEAD(&fi->write_files); INIT_LIST_HEAD(&fi->queued_writes); @@ -3423,7 +3125,6 @@ void fuse_init_file_inode(struct inode *inode, unsigned int flags) fi->iocachectr = 0; init_waitqueue_head(&fi->page_waitq); init_waitqueue_head(&fi->direct_io_waitq); - fi->writepages = RB_ROOT; if (IS_ENABLED(CONFIG_FUSE_DAX)) fuse_dax_inode_init(inode, flags);
In the current FUSE writeback design (see commit 3be5a52b30aa ("fuse: support writable mmap")), a temp page is allocated for every dirty page to be written back, the contents of the dirty page are copied over to the temp page, and the temp page gets handed to the server to write back. This is done so that writeback may be immediately cleared on the dirty page, and this in turn is done for two reasons: a) in order to mitigate the following deadlock scenario that may arise if reclaim waits on writeback on the dirty page to complete: * single-threaded FUSE server is in the middle of handling a request that needs a memory allocation * memory allocation triggers direct reclaim * direct reclaim waits on a folio under writeback * the FUSE server can't write back the folio since it's stuck in direct reclaim b) in order to unblock internal (eg sync, page compaction) waits on writeback without needing the server to complete writing back to disk, which may take an indeterminate amount of time. With a recent change that added AS_WRITEBACK_INDETERMINATE and mitigates the situations described above, FUSE writeback does not need to use temp pages if it sets AS_WRITEBACK_INDETERMINATE on its inode mappings. This commit sets AS_WRITEBACK_INDETERMINATE on the inode mappings and removes the temporary pages + extra copying and the internal rb tree. fio benchmarks -- (using averages observed from 10 runs, throwing away outliers) Setup: sudo mount -t tmpfs -o size=30G tmpfs ~/tmp_mount ./libfuse/build/example/passthrough_ll -o writeback -o max_threads=4 -o source=~/tmp_mount ~/fuse_mount fio --name=writeback --ioengine=sync --rw=write --bs={1k,4k,1M} --size=2G --numjobs=2 --ramp_time=30 --group_reporting=1 --directory=/root/fuse_mount bs = 1k 4k 1M Before 351 MiB/s 1818 MiB/s 1851 MiB/s After 341 MiB/s 2246 MiB/s 2685 MiB/s % diff -3% 23% 45% Signed-off-by: Joanne Koong <joannelkoong@gmail.com> --- fs/fuse/file.c | 339 +++---------------------------------------------- 1 file changed, 20 insertions(+), 319 deletions(-)