On 9/12/25 07:53, Jaegeuk Kim wrote: > Given the performance data and implementation overhead, I'm also questioning > whether we really need to support this for writes or not. Can we get some > common > sense of usage models?
Seems uncached write implementation affects the performance a lot, I don't see a good reason to merge this for now. I think we can try to enable uncached read functionality and return -EOPNOTSUPP for uncached write first, meanwhile, let's see if there is anything good usecase for uncached write. Thanks, > > On 08/28, Qi Han wrote: >> In the link [1], we adapted uncached buffer I/O read support in f2fs. >> Now, let's move forward to enabling uncached buffer I/O write support >> in f2fs. >> >> In f2fs_write_end_io, a separate asynchronous workqueue is created to >> perform the page drop operation for bios that contain pages of type >> FGP_DONTCACHE. >> >> The following patch is developed and tested based on the v6.17-rc3 branch. >> My local testing results are as follows, along with some issues observed: >> 1) Write performance degradation. Uncached buffer I/O write is slower than >> normal buffered write because uncached I/O triggers a sync operation for >> each I/O after data is written to memory, in order to drop pages promptly >> at end_io. I assume this impact might be less visible on high-performance >> storage devices such as PCIe 6.0 SSDs. >> - f2fs_file_write_iter >> - f2fs_buffered_write_iter >> - generic_write_sync >> - filemap_fdatawrite_range_kick >> 2) As expected, page cache usage does not significantly increase during >> writes. >> 3) The kswapd0 memory reclaim thread remains mostly idle, but additional >> asynchronous work overhead is introduced, e.g: >> PID USER PR NI VIRT RES SHR S[%CPU] %MEM TIME+ ARGS >> 19650 root 0 -20 0 0 0 I 7.0 0.0 0:00.21 >> [kworker/u33:3-f2fs_post_write_wq] >> 95 root 0 -20 0 0 0 I 6.6 0.0 0:02.08 >> [kworker/u33:0-f2fs_post_write_wq] >> 19653 root 0 -20 0 0 0 I 4.6 0.0 0:01.25 >> [kworker/u33:6-f2fs_post_write_wq] >> 19652 root 0 -20 0 0 0 I 4.3 0.0 0:00.92 >> [kworker/u33:5-f2fs_post_write_wq] >> 19613 root 0 -20 0 0 0 I 4.3 0.0 0:00.99 >> [kworker/u33:1-f2fs_post_write_wq] >> 19651 root 0 -20 0 0 0 I 3.6 0.0 0:00.98 >> [kworker/u33:4-f2fs_post_write_wq] >> 19654 root 0 -20 0 0 0 I 3.0 0.0 0:01.05 >> [kworker/u33:7-f2fs_post_write_wq] >> 19655 root 0 -20 0 0 0 I 2.3 0.0 0:01.18 >> [kworker/u33:8-f2fs_post_write_wq] >> >> >From these results on my test device, introducing uncached buffer I/O write >> >on >> f2fs seems to bring more drawbacks than benefits. Do we really need to >> support >> uncached buffer I/O write in f2fs? >> >> Write test data without using uncached buffer I/O: >> Starting 1 threads >> pid: 17609 >> writing bs 8192, uncached 0 >> 1s: 753MB/sec, MB=753 >> 2s: 792MB/sec, MB=1546 >> 3s: 430MB/sec, MB=1978 >> 4s: 661MB/sec, MB=2636 >> 5s: 900MB/sec, MB=3542 >> 6s: 769MB/sec, MB=4308 >> 7s: 808MB/sec, MB=5113 >> 8s: 766MB/sec, MB=5884 >> 9s: 654MB/sec, MB=6539 >> 10s: 456MB/sec, MB=6995 >> 11s: 797MB/sec, MB=7793 >> 12s: 770MB/sec, MB=8563 >> 13s: 778MB/sec, MB=9341 >> 14s: 726MB/sec, MB=10077 >> 15s: 736MB/sec, MB=10803 >> 16s: 798MB/sec, MB=11602 >> 17s: 728MB/sec, MB=12330 >> 18s: 749MB/sec, MB=13080 >> 19s: 777MB/sec, MB=13857 >> 20s: 688MB/sec, MB=14395 >> >> 19:29:34 UID PID %usr %system %guest %wait %CPU CPU >> Command >> 19:29:35 0 94 0.00 0.00 0.00 0.00 0.00 4 >> kswapd0 >> 19:29:36 0 94 0.00 0.00 0.00 0.00 0.00 4 >> kswapd0 >> 19:29:37 0 94 0.00 0.00 0.00 0.00 0.00 4 >> kswapd0 >> 19:29:38 0 94 0.00 0.00 0.00 0.00 0.00 4 >> kswapd0 >> 19:29:39 0 94 0.00 0.00 0.00 0.00 0.00 4 >> kswapd0 >> 19:29:40 0 94 0.00 0.00 0.00 0.00 0.00 4 >> kswapd0 >> 19:29:41 0 94 0.00 2.00 0.00 0.00 2.00 0 >> kswapd0 >> 19:29:42 0 94 0.00 59.00 0.00 0.00 59.00 7 >> kswapd0 >> 19:29:43 0 94 0.00 45.00 0.00 0.00 45.00 7 >> kswapd0 >> 19:29:44 0 94 0.00 36.00 0.00 0.00 36.00 0 >> kswapd0 >> 19:29:45 0 94 0.00 27.00 0.00 1.00 27.00 0 >> kswapd0 >> 19:29:46 0 94 0.00 26.00 0.00 0.00 26.00 2 >> kswapd0 >> 19:29:47 0 94 0.00 57.00 0.00 0.00 57.00 7 >> kswapd0 >> 19:29:48 0 94 0.00 41.00 0.00 0.00 41.00 7 >> kswapd0 >> 19:29:49 0 94 0.00 38.00 0.00 0.00 38.00 7 >> kswapd0 >> 19:29:50 0 94 0.00 47.00 0.00 0.00 47.00 7 >> kswapd0 >> 19:29:51 0 94 0.00 43.00 0.00 1.00 43.00 7 >> kswapd0 >> 19:29:52 0 94 0.00 36.00 0.00 0.00 36.00 7 >> kswapd0 >> 19:29:53 0 94 0.00 39.00 0.00 0.00 39.00 2 >> kswapd0 >> 19:29:54 0 94 0.00 46.00 0.00 0.00 46.00 7 >> kswapd0 >> 19:29:55 0 94 0.00 43.00 0.00 0.00 43.00 7 >> kswapd0 >> 19:29:56 0 94 0.00 39.00 0.00 0.00 39.00 7 >> kswapd0 >> 19:29:57 0 94 0.00 29.00 0.00 1.00 29.00 1 >> kswapd0 >> 19:29:58 0 94 0.00 17.00 0.00 0.00 17.00 4 >> kswapd0 >> >> 19:29:33 kbmemfree kbavail kbmemused %memused kbbuffers kbcached >> kbcommit %commit kbactive kbinact kbdirty >> 19:29:34 4464588 6742648 4420876 38.12 6156 2032600 >> 179730872 743.27 1863412 1822544 4 >> 19:29:35 4462572 6740784 4422752 38.13 6156 2032752 >> 179739004 743.30 1863460 1823584 16 >> 19:29:36 4381512 6740856 4422420 38.13 6156 2114144 >> 179746508 743.33 1863476 1905384 81404 >> 19:29:37 3619456 6741840 4421588 38.12 6156 2877032 >> 179746652 743.33 1863536 2668896 592584 >> 19:29:38 2848184 6740720 4422472 38.13 6164 3646188 >> 179746652 743.33 1863600 3438520 815692 >> 19:29:39 2436336 6739452 4423720 38.14 6164 4056772 >> 179746652 743.33 1863604 3849164 357096 >> 19:29:40 1712660 6737700 4425140 38.15 6164 4779020 >> 179746604 743.33 1863612 4571124 343716 >> 19:29:41 810664 6738020 4425004 38.15 6164 5681152 >> 179746604 743.33 1863612 5473444 297268 >> 19:29:42 673756 6779120 4373200 37.71 5656 5869928 >> 179746604 743.33 1902852 5589452 269032 >> 19:29:43 688480 6782024 4371012 37.69 5648 5856940 >> 179750048 743.34 1926336 5542004 279344 >> 19:29:44 688956 6789028 4364260 37.63 5584 5863272 >> 179750048 743.34 1941608 5518808 300096 >> 19:29:45 740768 6804560 4348772 37.49 5524 5827248 >> 179750000 743.34 1954084 5452844 123120 >> 19:29:46 697936 6810612 4342768 37.44 5524 5876048 >> 179750048 743.34 1962020 5483944 274908 >> 19:29:47 734504 6818716 4334156 37.37 5512 5849188 >> 179750000 743.34 1978120 5426796 274504 >> 19:29:48 771696 6828316 4324180 37.28 5504 5820948 >> 179762260 743.39 2006732 5354152 305388 >> 19:29:49 691944 6838812 4313108 37.19 5476 5912444 >> 179749952 743.34 2021720 5418996 296852 >> 19:29:50 679392 6844496 4306892 37.13 5452 5931356 >> 179749952 743.34 1982772 5463040 233600 >> 19:29:51 768528 6868080 4284224 36.94 5412 5865704 >> 176317452 729.15 1990220 5359012 343160 >> 19:29:52 717880 6893940 4259968 36.73 5400 5942368 >> 176317404 729.15 1965624 5444140 304856 >> 19:29:53 712408 6902660 4251268 36.65 5372 5956584 >> 176318376 729.15 1969192 5442132 290224 >> 19:29:54 707184 6917512 4236160 36.52 5344 5976944 >> 176318568 729.15 1968716 5443620 336948 >> 19:29:55 703172 6921608 4232332 36.49 5292 5984836 >> 176318568 729.15 1979788 5429484 328716 >> 19:29:56 733256 6933020 4220864 36.39 5212 5966340 >> 176318568 729.15 1983636 5396256 300008 >> 19:29:57 723308 6936340 4217280 36.36 5120 5979816 >> 176318568 729.15 1987088 5394360 508792 >> 19:29:58 732148 6942972 4210680 36.30 5108 5977656 >> 176311064 729.12 1990400 5379884 214936 >> >> Write test data after using uncached buffer I/O: >> Starting 1 threads >> pid: 17742 >> writing bs 8192, uncached 1 >> 1s: 433MB/sec, MB=433 >> 2s: 195MB/sec, MB=628 >> 3s: 209MB/sec, MB=836 >> 4s: 54MB/sec, MB=883 >> 5s: 277MB/sec, MB=1169 >> 6s: 141MB/sec, MB=1311 >> 7s: 185MB/sec, MB=1495 >> 8s: 134MB/sec, MB=1631 >> 9s: 201MB/sec, MB=1834 >> 10s: 283MB/sec, MB=2114 >> 11s: 223MB/sec, MB=2339 >> 12s: 164MB/sec, MB=2506 >> 13s: 155MB/sec, MB=2657 >> 14s: 132MB/sec, MB=2792 >> 15s: 186MB/sec, MB=2965 >> 16s: 218MB/sec, MB=3198 >> 17s: 220MB/sec, MB=3412 >> 18s: 191MB/sec, MB=3606 >> 19s: 214MB/sec, MB=3828 >> 20s: 257MB/sec, MB=4085 >> >> 19:31:31 UID PID %usr %system %guest %wait %CPU CPU >> Command >> 19:31:32 0 94 0.00 0.00 0.00 0.00 0.00 4 >> kswapd0 >> 19:31:33 0 94 0.00 0.00 0.00 0.00 0.00 4 >> kswapd0 >> 19:31:34 0 94 0.00 0.00 0.00 0.00 0.00 4 >> kswapd0 >> 19:31:35 0 94 0.00 0.00 0.00 0.00 0.00 4 >> kswapd0 >> 19:31:36 0 94 0.00 0.00 0.00 0.00 0.00 4 >> kswapd0 >> 19:31:37 0 94 0.00 0.00 0.00 0.00 0.00 4 >> kswapd0 >> 19:31:38 0 94 0.00 0.00 0.00 0.00 0.00 4 >> kswapd0 >> 19:31:39 0 94 0.00 0.00 0.00 0.00 0.00 4 >> kswapd0 >> 19:31:40 0 94 0.00 0.00 0.00 0.00 0.00 4 >> kswapd0 >> 19:31:41 0 94 0.00 0.00 0.00 0.00 0.00 4 >> kswapd0 >> 19:31:42 0 94 0.00 0.00 0.00 0.00 0.00 4 >> kswapd0 >> 19:31:43 0 94 0.00 0.00 0.00 0.00 0.00 4 >> kswapd0 >> 19:31:44 0 94 0.00 0.00 0.00 0.00 0.00 4 >> kswapd0 >> 19:31:45 0 94 0.00 0.00 0.00 0.00 0.00 4 >> kswapd0 >> 19:31:46 0 94 0.00 0.00 0.00 0.00 0.00 4 >> kswapd0 >> 19:31:47 0 94 0.00 0.00 0.00 0.00 0.00 4 >> kswapd0 >> 19:31:48 0 94 0.00 0.00 0.00 0.00 0.00 4 >> kswapd0 >> 19:31:49 0 94 0.00 0.00 0.00 0.00 0.00 4 >> kswapd0 >> >> 19:31:31 kbmemfree kbavail kbmemused %memused kbbuffers kbcached >> kbcommit %commit kbactive kbinact kbdirty >> 19:31:32 4816812 6928788 4225812 36.43 5148 1879676 >> 176322636 729.17 1920900 1336548 285748 >> 19:31:33 4781880 6889428 4265592 36.78 5148 1874860 >> 176322636 729.17 1920920 1332268 279028 >> 19:31:34 4758972 6822588 4332376 37.35 5148 1830984 >> 176322636 729.17 1920920 1288976 233040 >> 19:31:35 4850248 6766480 4387840 37.83 5148 1684244 >> 176322636 729.17 1920920 1142408 90508 >> 19:31:36 4644176 6741676 4413256 38.05 5148 1864900 >> 176322636 729.17 1920920 1323452 269380 >> 19:31:37 4637900 6681480 4473436 38.57 5148 1810996 >> 176322588 729.17 1920920 1269612 217632 >> 19:31:38 4502108 6595508 4559500 39.31 5148 1860724 >> 176322492 729.17 1920920 1319588 267760 >> 19:31:39 4498844 6551068 4603928 39.69 5148 1819528 >> 176322492 729.17 1920920 1278440 226496 >> 19:31:40 4498812 6587396 4567340 39.38 5148 1856116 >> 176322492 729.17 1920920 1314800 263292 >> 19:31:41 4656784 6706252 4448372 38.35 5148 1817112 >> 176322492 729.17 1920920 1275704 224600 >> 19:31:42 4635032 6673328 4481436 38.64 5148 1805816 >> 176322492 729.17 1920920 1264548 213436 >> 19:31:43 4636852 6679736 4474884 38.58 5148 1810548 >> 176322492 729.17 1920932 1269796 218276 >> 19:31:44 4654740 6669104 4485544 38.67 5148 1782000 >> 176322444 729.17 1920932 1241552 189880 >> 19:31:45 4821604 6693156 4461848 38.47 5148 1638864 >> 176322444 729.17 1920932 1098784 31076 >> 19:31:46 4707548 6728796 4426400 38.16 5148 1788368 >> 176322444 729.17 1920932 1248936 196596 >> 19:31:47 4683996 6747632 4407348 38.00 5148 1830968 >> 176322444 729.17 1920932 1291396 239636 >> 19:31:48 4694648 6773808 4381320 37.78 5148 1846376 >> 176322624 729.17 1920944 1307576 254800 >> 19:31:49 4663784 6730212 4424776 38.15 5148 1833784 >> 176322772 729.17 1920948 1295156 242200 >> >> [1] >> https://lore.kernel.org/lkml/20250725075310.1614614-1-ha...@vivo.com/ >> >> Signed-off-by: Qi Han <ha...@vivo.com> >> --- >> fs/f2fs/data.c | 178 ++++++++++++++++++++++++++++++++++------------ >> fs/f2fs/f2fs.h | 5 ++ >> fs/f2fs/file.c | 2 +- >> fs/f2fs/iostat.c | 8 ++- >> fs/f2fs/iostat.h | 4 +- >> fs/f2fs/segment.c | 2 +- >> fs/f2fs/super.c | 16 ++++- >> 7 files changed, 161 insertions(+), 54 deletions(-) >> >> diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c >> index 7961e0ddfca3..4eeb2b36473d 100644 >> --- a/fs/f2fs/data.c >> +++ b/fs/f2fs/data.c >> @@ -30,8 +30,10 @@ >> #define NUM_PREALLOC_POST_READ_CTXS 128 >> >> static struct kmem_cache *bio_post_read_ctx_cache; >> +static struct kmem_cache *bio_post_write_ctx_cache; >> static struct kmem_cache *bio_entry_slab; >> static mempool_t *bio_post_read_ctx_pool; >> +static mempool_t *bio_post_write_ctx_pool; >> static struct bio_set f2fs_bioset; >> >> #define F2FS_BIO_POOL_SIZE NR_CURSEG_TYPE >> @@ -120,6 +122,12 @@ struct bio_post_read_ctx { >> block_t fs_blkaddr; >> }; >> >> +struct bio_post_write_ctx { >> + struct bio *bio; >> + struct f2fs_sb_info *sbi; >> + struct work_struct work; >> +}; >> + >> /* >> * Update and unlock a bio's pages, and free the bio. >> * >> @@ -159,6 +167,56 @@ static void f2fs_finish_read_bio(struct bio *bio, bool >> in_task) >> bio_put(bio); >> } >> >> +static void f2fs_finish_write_bio(struct f2fs_sb_info *sbi, struct bio *bio) >> +{ >> + struct folio_iter fi; >> + struct bio_post_write_ctx *write_ctx = (struct bio_post_write_ctx >> *)bio->bi_private; >> + >> + bio_for_each_folio_all(fi, bio) { >> + struct folio *folio = fi.folio; >> + enum count_type type; >> + >> + if (fscrypt_is_bounce_folio(folio)) { >> + struct folio *io_folio = folio; >> + >> + folio = fscrypt_pagecache_folio(io_folio); >> + fscrypt_free_bounce_page(&io_folio->page); >> + } >> + >> +#ifdef CONFIG_F2FS_FS_COMPRESSION >> + if (f2fs_is_compressed_page(folio)) { >> + f2fs_compress_write_end_io(bio, folio); >> + continue; >> + } >> +#endif >> + >> + type = WB_DATA_TYPE(folio, false); >> + >> + if (unlikely(bio->bi_status != BLK_STS_OK)) { >> + mapping_set_error(folio->mapping, -EIO); >> + if (type == F2FS_WB_CP_DATA) >> + f2fs_stop_checkpoint(sbi, true, >> + STOP_CP_REASON_WRITE_FAIL); >> + } >> + >> + f2fs_bug_on(sbi, is_node_folio(folio) && >> + folio->index != nid_of_node(folio)); >> + >> + dec_page_count(sbi, type); >> + if (f2fs_in_warm_node_list(sbi, folio)) >> + f2fs_del_fsync_node_entry(sbi, folio); >> + folio_clear_f2fs_gcing(folio); >> + folio_end_writeback(folio); >> + } >> + if (!get_pages(sbi, F2FS_WB_CP_DATA) && >> + wq_has_sleeper(&sbi->cp_wait)) >> + wake_up(&sbi->cp_wait); >> + >> + if (write_ctx) >> + mempool_free(write_ctx, bio_post_write_ctx_pool); >> + bio_put(bio); >> +} >> + >> static void f2fs_verify_bio(struct work_struct *work) >> { >> struct bio_post_read_ctx *ctx = >> @@ -314,58 +372,32 @@ static void f2fs_read_end_io(struct bio *bio) >> f2fs_verify_and_finish_bio(bio, intask); >> } >> >> +static void f2fs_finish_write_bio_async_work(struct work_struct *work) >> +{ >> + struct bio_post_write_ctx *ctx = >> + container_of(work, struct bio_post_write_ctx, work); >> + >> + f2fs_finish_write_bio(ctx->sbi, ctx->bio); >> +} >> + >> static void f2fs_write_end_io(struct bio *bio) >> { >> - struct f2fs_sb_info *sbi; >> - struct folio_iter fi; >> + struct f2fs_sb_info *sbi = F2FS_F_SB(bio_first_folio_all(bio)); >> + struct bio_post_write_ctx *write_ctx; >> >> iostat_update_and_unbind_ctx(bio); >> - sbi = bio->bi_private; >> >> if (time_to_inject(sbi, FAULT_WRITE_IO)) >> bio->bi_status = BLK_STS_IOERR; >> >> - bio_for_each_folio_all(fi, bio) { >> - struct folio *folio = fi.folio; >> - enum count_type type; >> - >> - if (fscrypt_is_bounce_folio(folio)) { >> - struct folio *io_folio = folio; >> - >> - folio = fscrypt_pagecache_folio(io_folio); >> - fscrypt_free_bounce_page(&io_folio->page); >> - } >> - >> -#ifdef CONFIG_F2FS_FS_COMPRESSION >> - if (f2fs_is_compressed_page(folio)) { >> - f2fs_compress_write_end_io(bio, folio); >> - continue; >> - } >> -#endif >> - >> - type = WB_DATA_TYPE(folio, false); >> - >> - if (unlikely(bio->bi_status != BLK_STS_OK)) { >> - mapping_set_error(folio->mapping, -EIO); >> - if (type == F2FS_WB_CP_DATA) >> - f2fs_stop_checkpoint(sbi, true, >> - STOP_CP_REASON_WRITE_FAIL); >> - } >> - >> - f2fs_bug_on(sbi, is_node_folio(folio) && >> - folio->index != nid_of_node(folio)); >> - >> - dec_page_count(sbi, type); >> - if (f2fs_in_warm_node_list(sbi, folio)) >> - f2fs_del_fsync_node_entry(sbi, folio); >> - folio_clear_f2fs_gcing(folio); >> - folio_end_writeback(folio); >> + write_ctx = (struct bio_post_write_ctx *)bio->bi_private; >> + if (write_ctx) { >> + INIT_WORK(&write_ctx->work, f2fs_finish_write_bio_async_work); >> + queue_work(write_ctx->sbi->post_write_wq, &write_ctx->work); >> + return; >> } >> - if (!get_pages(sbi, F2FS_WB_CP_DATA) && >> - wq_has_sleeper(&sbi->cp_wait)) >> - wake_up(&sbi->cp_wait); >> >> - bio_put(bio); >> + f2fs_finish_write_bio(sbi, bio); >> } >> >> #ifdef CONFIG_BLK_DEV_ZONED >> @@ -467,11 +499,10 @@ static struct bio *__bio_alloc(struct f2fs_io_info >> *fio, int npages) >> bio->bi_private = NULL; >> } else { >> bio->bi_end_io = f2fs_write_end_io; >> - bio->bi_private = sbi; >> + bio->bi_private = NULL; >> bio->bi_write_hint = f2fs_io_type_to_rw_hint(sbi, >> fio->type, fio->temp); >> } >> - iostat_alloc_and_bind_ctx(sbi, bio, NULL); >> >> if (fio->io_wbc) >> wbc_init_bio(fio->io_wbc, bio); >> @@ -701,6 +732,7 @@ int f2fs_submit_page_bio(struct f2fs_io_info *fio) >> >> /* Allocate a new bio */ >> bio = __bio_alloc(fio, 1); >> + iostat_alloc_and_bind_ctx(fio->sbi, bio, NULL, NULL); >> >> f2fs_set_bio_crypt_ctx(bio, fio_folio->mapping->host, >> fio_folio->index, fio, GFP_NOIO); >> @@ -899,6 +931,8 @@ int f2fs_merge_page_bio(struct f2fs_io_info *fio) >> alloc_new: >> if (!bio) { >> bio = __bio_alloc(fio, BIO_MAX_VECS); >> + iostat_alloc_and_bind_ctx(fio->sbi, bio, NULL, NULL); >> + >> f2fs_set_bio_crypt_ctx(bio, folio->mapping->host, >> folio->index, fio, GFP_NOIO); >> >> @@ -948,6 +982,7 @@ void f2fs_submit_page_write(struct f2fs_io_info *fio) >> struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp; >> struct folio *bio_folio; >> enum count_type type; >> + struct bio_post_write_ctx *write_ctx = NULL; >> >> f2fs_bug_on(sbi, is_read_io(fio->op)); >> >> @@ -1001,6 +1036,13 @@ void f2fs_submit_page_write(struct f2fs_io_info *fio) >> f2fs_set_bio_crypt_ctx(io->bio, fio_inode(fio), >> bio_folio->index, fio, GFP_NOIO); >> io->fio = *fio; >> + >> + if (folio_test_dropbehind(bio_folio)) { >> + write_ctx = mempool_alloc(bio_post_write_ctx_pool, >> GFP_NOFS); >> + write_ctx->bio = io->bio; >> + write_ctx->sbi = sbi; >> + } >> + iostat_alloc_and_bind_ctx(fio->sbi, io->bio, NULL, write_ctx); >> } >> >> if (!bio_add_folio(io->bio, bio_folio, folio_size(bio_folio), 0)) { >> @@ -1077,7 +1119,7 @@ static struct bio *f2fs_grab_read_bio(struct inode >> *inode, block_t blkaddr, >> ctx->decompression_attempted = false; >> bio->bi_private = ctx; >> } >> - iostat_alloc_and_bind_ctx(sbi, bio, ctx); >> + iostat_alloc_and_bind_ctx(sbi, bio, ctx, NULL); >> >> return bio; >> } >> @@ -3540,6 +3582,7 @@ static int f2fs_write_begin(const struct kiocb *iocb, >> bool use_cow = false; >> block_t blkaddr = NULL_ADDR; >> int err = 0; >> + fgf_t fgp = FGP_LOCK | FGP_WRITE | FGP_CREAT; >> >> trace_f2fs_write_begin(inode, pos, len); >> >> @@ -3582,12 +3625,13 @@ static int f2fs_write_begin(const struct kiocb *iocb, >> #endif >> >> repeat: >> + if (iocb && iocb->ki_flags & IOCB_DONTCACHE) >> + fgp |= FGP_DONTCACHE; >> /* >> * Do not use FGP_STABLE to avoid deadlock. >> * Will wait that below with our IO control. >> */ >> - folio = __filemap_get_folio(mapping, index, >> - FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS); >> + folio = __filemap_get_folio(mapping, index, fgp, GFP_NOFS); >> if (IS_ERR(folio)) { >> err = PTR_ERR(folio); >> goto fail; >> @@ -4127,12 +4171,38 @@ int __init f2fs_init_post_read_processing(void) >> return -ENOMEM; >> } >> >> +int __init f2fs_init_post_write_processing(void) >> +{ >> + bio_post_write_ctx_cache = >> + kmem_cache_create("f2fs_bio_post_write_ctx", >> + sizeof(struct bio_post_write_ctx), 0, 0, NULL); >> + if (!bio_post_write_ctx_cache) >> + goto fail; >> + bio_post_write_ctx_pool = >> + mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS, >> + bio_post_write_ctx_cache); >> + if (!bio_post_write_ctx_pool) >> + goto fail_free_cache; >> + return 0; >> + >> +fail_free_cache: >> + kmem_cache_destroy(bio_post_write_ctx_cache); >> +fail: >> + return -ENOMEM; >> +} >> + >> void f2fs_destroy_post_read_processing(void) >> { >> mempool_destroy(bio_post_read_ctx_pool); >> kmem_cache_destroy(bio_post_read_ctx_cache); >> } >> >> +void f2fs_destroy_post_write_processing(void) >> +{ >> + mempool_destroy(bio_post_write_ctx_pool); >> + kmem_cache_destroy(bio_post_write_ctx_cache); >> +} >> + >> int f2fs_init_post_read_wq(struct f2fs_sb_info *sbi) >> { >> if (!f2fs_sb_has_encrypt(sbi) && >> @@ -4146,12 +4216,26 @@ int f2fs_init_post_read_wq(struct f2fs_sb_info *sbi) >> return sbi->post_read_wq ? 0 : -ENOMEM; >> } >> >> +int f2fs_init_post_write_wq(struct f2fs_sb_info *sbi) >> +{ >> + sbi->post_write_wq = alloc_workqueue("f2fs_post_write_wq", >> + WQ_UNBOUND | WQ_HIGHPRI, >> + num_online_cpus()); >> + return sbi->post_write_wq ? 0 : -ENOMEM; >> +} >> + >> void f2fs_destroy_post_read_wq(struct f2fs_sb_info *sbi) >> { >> if (sbi->post_read_wq) >> destroy_workqueue(sbi->post_read_wq); >> } >> >> +void f2fs_destroy_post_write_wq(struct f2fs_sb_info *sbi) >> +{ >> + if (sbi->post_write_wq) >> + destroy_workqueue(sbi->post_write_wq); >> +} >> + >> int __init f2fs_init_bio_entry_cache(void) >> { >> bio_entry_slab = f2fs_kmem_cache_create("f2fs_bio_entry_slab", >> diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h >> index 46be7560548c..fe3f81876b23 100644 >> --- a/fs/f2fs/f2fs.h >> +++ b/fs/f2fs/f2fs.h >> @@ -1812,6 +1812,7 @@ struct f2fs_sb_info { >> /* Precomputed FS UUID checksum for seeding other checksums */ >> __u32 s_chksum_seed; >> >> + struct workqueue_struct *post_write_wq; >> struct workqueue_struct *post_read_wq; /* post read workqueue */ >> >> /* >> @@ -4023,9 +4024,13 @@ bool f2fs_release_folio(struct folio *folio, gfp_t >> wait); >> bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len); >> void f2fs_clear_page_cache_dirty_tag(struct folio *folio); >> int f2fs_init_post_read_processing(void); >> +int f2fs_init_post_write_processing(void); >> void f2fs_destroy_post_read_processing(void); >> +void f2fs_destroy_post_write_processing(void); >> int f2fs_init_post_read_wq(struct f2fs_sb_info *sbi); >> +int f2fs_init_post_write_wq(struct f2fs_sb_info *sbi); >> void f2fs_destroy_post_read_wq(struct f2fs_sb_info *sbi); >> +void f2fs_destroy_post_write_wq(struct f2fs_sb_info *sbi); >> extern const struct iomap_ops f2fs_iomap_ops; >> >> /* >> diff --git a/fs/f2fs/file.c b/fs/f2fs/file.c >> index 42faaed6a02d..8aa6a4fd52e8 100644 >> --- a/fs/f2fs/file.c >> +++ b/fs/f2fs/file.c >> @@ -5443,5 +5443,5 @@ const struct file_operations f2fs_file_operations = { >> .splice_read = f2fs_file_splice_read, >> .splice_write = iter_file_splice_write, >> .fadvise = f2fs_file_fadvise, >> - .fop_flags = FOP_BUFFER_RASYNC, >> + .fop_flags = FOP_BUFFER_RASYNC | FOP_DONTCACHE, >> }; >> diff --git a/fs/f2fs/iostat.c b/fs/f2fs/iostat.c >> index f8703038e1d8..b2e6ce80c68d 100644 >> --- a/fs/f2fs/iostat.c >> +++ b/fs/f2fs/iostat.c >> @@ -245,7 +245,7 @@ void iostat_update_and_unbind_ctx(struct bio *bio) >> if (op_is_write(bio_op(bio))) { >> lat_type = bio->bi_opf & REQ_SYNC ? >> WRITE_SYNC_IO : WRITE_ASYNC_IO; >> - bio->bi_private = iostat_ctx->sbi; >> + bio->bi_private = iostat_ctx->post_write_ctx; >> } else { >> lat_type = READ_IO; >> bio->bi_private = iostat_ctx->post_read_ctx; >> @@ -256,7 +256,8 @@ void iostat_update_and_unbind_ctx(struct bio *bio) >> } >> >> void iostat_alloc_and_bind_ctx(struct f2fs_sb_info *sbi, >> - struct bio *bio, struct bio_post_read_ctx *ctx) >> + struct bio *bio, struct bio_post_read_ctx *read_ctx, >> + struct bio_post_write_ctx *write_ctx) >> { >> struct bio_iostat_ctx *iostat_ctx; >> /* Due to the mempool, this never fails. */ >> @@ -264,7 +265,8 @@ void iostat_alloc_and_bind_ctx(struct f2fs_sb_info *sbi, >> iostat_ctx->sbi = sbi; >> iostat_ctx->submit_ts = 0; >> iostat_ctx->type = 0; >> - iostat_ctx->post_read_ctx = ctx; >> + iostat_ctx->post_read_ctx = read_ctx; >> + iostat_ctx->post_write_ctx = write_ctx; >> bio->bi_private = iostat_ctx; >> } >> >> diff --git a/fs/f2fs/iostat.h b/fs/f2fs/iostat.h >> index eb99d05cf272..a358909bb5e8 100644 >> --- a/fs/f2fs/iostat.h >> +++ b/fs/f2fs/iostat.h >> @@ -40,6 +40,7 @@ struct bio_iostat_ctx { >> unsigned long submit_ts; >> enum page_type type; >> struct bio_post_read_ctx *post_read_ctx; >> + struct bio_post_write_ctx *post_write_ctx; >> }; >> >> static inline void iostat_update_submit_ctx(struct bio *bio, >> @@ -60,7 +61,8 @@ static inline struct bio_post_read_ctx >> *get_post_read_ctx(struct bio *bio) >> >> extern void iostat_update_and_unbind_ctx(struct bio *bio); >> extern void iostat_alloc_and_bind_ctx(struct f2fs_sb_info *sbi, >> - struct bio *bio, struct bio_post_read_ctx *ctx); >> + struct bio *bio, struct bio_post_read_ctx *read_ctx, >> + struct bio_post_write_ctx *write_ctx); >> extern int f2fs_init_iostat_processing(void); >> extern void f2fs_destroy_iostat_processing(void); >> extern int f2fs_init_iostat(struct f2fs_sb_info *sbi); >> diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c >> index cc82d42ef14c..8501008e42b2 100644 >> --- a/fs/f2fs/segment.c >> +++ b/fs/f2fs/segment.c >> @@ -3856,7 +3856,7 @@ int f2fs_allocate_data_block(struct f2fs_sb_info *sbi, >> struct folio *folio, >> f2fs_inode_chksum_set(sbi, folio); >> } >> >> - if (fio) { >> + if (fio && !folio_test_dropbehind(folio)) { >> struct f2fs_bio_info *io; >> >> INIT_LIST_HEAD(&fio->list); >> diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c >> index e16c4e2830c2..110dfe073aee 100644 >> --- a/fs/f2fs/super.c >> +++ b/fs/f2fs/super.c >> @@ -1963,6 +1963,7 @@ static void f2fs_put_super(struct super_block *sb) >> flush_work(&sbi->s_error_work); >> >> f2fs_destroy_post_read_wq(sbi); >> + f2fs_destroy_post_write_wq(sbi); >> >> kvfree(sbi->ckpt); >> >> @@ -4959,6 +4960,12 @@ static int f2fs_fill_super(struct super_block *sb, >> struct fs_context *fc) >> goto free_devices; >> } >> >> + err = f2fs_init_post_write_wq(sbi); >> + if (err) { >> + f2fs_err(sbi, "Failed to initialize post write workqueue"); >> + goto free_devices; >> + } >> + >> sbi->total_valid_node_count = >> le32_to_cpu(sbi->ckpt->valid_node_count); >> percpu_counter_set(&sbi->total_valid_inode_count, >> @@ -5240,6 +5247,7 @@ static int f2fs_fill_super(struct super_block *sb, >> struct fs_context *fc) >> /* flush s_error_work before sbi destroy */ >> flush_work(&sbi->s_error_work); >> f2fs_destroy_post_read_wq(sbi); >> + f2fs_destroy_post_write_wq(sbi); >> free_devices: >> destroy_device_list(sbi); >> kvfree(sbi->ckpt); >> @@ -5435,9 +5443,12 @@ static int __init init_f2fs_fs(void) >> err = f2fs_init_post_read_processing(); >> if (err) >> goto free_root_stats; >> - err = f2fs_init_iostat_processing(); >> + err = f2fs_init_post_write_processing(); >> if (err) >> goto free_post_read; >> + err = f2fs_init_iostat_processing(); >> + if (err) >> + goto free_post_write; >> err = f2fs_init_bio_entry_cache(); >> if (err) >> goto free_iostat; >> @@ -5469,6 +5480,8 @@ static int __init init_f2fs_fs(void) >> f2fs_destroy_bio_entry_cache(); >> free_iostat: >> f2fs_destroy_iostat_processing(); >> +free_post_write: >> + f2fs_destroy_post_write_processing(); >> free_post_read: >> f2fs_destroy_post_read_processing(); >> free_root_stats: >> @@ -5504,6 +5517,7 @@ static void __exit exit_f2fs_fs(void) >> f2fs_destroy_bio_entry_cache(); >> f2fs_destroy_iostat_processing(); >> f2fs_destroy_post_read_processing(); >> + f2fs_destroy_post_write_processing(); >> f2fs_destroy_root_stats(); >> f2fs_exit_shrinker(); >> f2fs_exit_sysfs(); >> -- >> 2.50.0 _______________________________________________ Linux-f2fs-devel mailing list Linux-f2fs-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
