diff options
Diffstat (limited to 'fs/bcachefs/fs-io.c')
| -rw-r--r-- | fs/bcachefs/fs-io.c | 1474 |
1 files changed, 701 insertions, 773 deletions
diff --git a/fs/bcachefs/fs-io.c b/fs/bcachefs/fs-io.c index 8f0b2a745064..df2f317f5443 100644 --- a/fs/bcachefs/fs-io.c +++ b/fs/bcachefs/fs-io.c @@ -35,6 +35,81 @@ #include <trace/events/bcachefs.h> #include <trace/events/writeback.h> +struct nocow_flush { + struct closure *cl; + struct bch_dev *ca; + struct bio bio; +}; + +static void nocow_flush_endio(struct bio *_bio) +{ + + struct nocow_flush *bio = container_of(_bio, struct nocow_flush, bio); + + closure_put(bio->cl); + percpu_ref_put(&bio->ca->io_ref); + bio_put(&bio->bio); +} + +static void bch2_inode_flush_nocow_writes_async(struct bch_fs *c, + struct bch_inode_info *inode, + struct closure *cl) +{ + struct nocow_flush *bio; + struct bch_dev *ca; + struct bch_devs_mask devs; + unsigned dev; + + dev = find_first_bit(inode->ei_devs_need_flush.d, BCH_SB_MEMBERS_MAX); + if (dev == BCH_SB_MEMBERS_MAX) + return; + + devs = inode->ei_devs_need_flush; + memset(&inode->ei_devs_need_flush, 0, sizeof(inode->ei_devs_need_flush)); + + for_each_set_bit(dev, devs.d, BCH_SB_MEMBERS_MAX) { + rcu_read_lock(); + ca = rcu_dereference(c->devs[dev]); + if (ca && !percpu_ref_tryget(&ca->io_ref)) + ca = NULL; + rcu_read_unlock(); + + if (!ca) + continue; + + bio = container_of(bio_alloc_bioset(ca->disk_sb.bdev, 0, + REQ_OP_FLUSH, + GFP_KERNEL, + &c->nocow_flush_bioset), + struct nocow_flush, bio); + bio->cl = cl; + bio->ca = ca; + bio->bio.bi_end_io = nocow_flush_endio; + closure_bio_submit(&bio->bio, cl); + } +} + +static int bch2_inode_flush_nocow_writes(struct bch_fs *c, + struct bch_inode_info *inode) +{ + struct closure cl; + + closure_init_stack(&cl); + bch2_inode_flush_nocow_writes_async(c, inode, &cl); + closure_sync(&cl); + + return 0; +} + +static inline bool bio_full(struct bio *bio, unsigned len) +{ + if (bio->bi_vcnt >= bio->bi_max_vecs) + return true; + if (bio->bi_iter.bi_size > UINT_MAX - len) + return true; + return false; +} + static inline struct address_space *faults_disabled_mapping(void) { return (void *) (((unsigned long) current->faults_disabled_mapping) & ~1UL); @@ -56,7 +131,6 @@ struct quota_res { }; struct bch_writepage_io { - struct closure cl; struct bch_inode_info *inode; /* must be last: */ @@ -64,11 +138,14 @@ struct bch_writepage_io { }; struct dio_write { - struct completion done; struct kiocb *req; + struct address_space *mapping; + struct bch_inode_info *inode; struct mm_struct *mm; unsigned loop:1, + extending:1, sync:1, + flush:1, free_iov:1; struct quota_res quota_res; u64 written; @@ -89,7 +166,7 @@ struct dio_read { }; /* pagecache_block must be held */ -static int write_invalidate_inode_pages_range(struct address_space *mapping, +static noinline int write_invalidate_inode_pages_range(struct address_space *mapping, loff_t start, loff_t end) { int ret; @@ -121,28 +198,33 @@ static int write_invalidate_inode_pages_range(struct address_space *mapping, #ifdef CONFIG_BCACHEFS_QUOTA -static void bch2_quota_reservation_put(struct bch_fs *c, - struct bch_inode_info *inode, - struct quota_res *res) +static void __bch2_quota_reservation_put(struct bch_fs *c, + struct bch_inode_info *inode, + struct quota_res *res) { - if (!res->sectors) - return; - - mutex_lock(&inode->ei_quota_lock); BUG_ON(res->sectors > inode->ei_quota_reserved); bch2_quota_acct(c, inode->ei_qid, Q_SPC, -((s64) res->sectors), KEY_TYPE_QUOTA_PREALLOC); inode->ei_quota_reserved -= res->sectors; - mutex_unlock(&inode->ei_quota_lock); - res->sectors = 0; } +static void bch2_quota_reservation_put(struct bch_fs *c, + struct bch_inode_info *inode, + struct quota_res *res) +{ + if (res->sectors) { + mutex_lock(&inode->ei_quota_lock); + __bch2_quota_reservation_put(c, inode, res); + mutex_unlock(&inode->ei_quota_lock); + } +} + static int bch2_quota_reservation_add(struct bch_fs *c, struct bch_inode_info *inode, struct quota_res *res, - unsigned sectors, + u64 sectors, bool check_enospc) { int ret; @@ -161,11 +243,13 @@ static int bch2_quota_reservation_add(struct bch_fs *c, #else +static void __bch2_quota_reservation_put(struct bch_fs *c, + struct bch_inode_info *inode, + struct quota_res *res) {} + static void bch2_quota_reservation_put(struct bch_fs *c, struct bch_inode_info *inode, - struct quota_res *res) -{ -} + struct quota_res *res) {} static int bch2_quota_reservation_add(struct bch_fs *c, struct bch_inode_info *inode, @@ -216,14 +300,13 @@ int __must_check bch2_write_inode_size(struct bch_fs *c, return bch2_write_inode(c, inode, inode_set_size, &s, fields); } -static void i_sectors_acct(struct bch_fs *c, struct bch_inode_info *inode, +static void __i_sectors_acct(struct bch_fs *c, struct bch_inode_info *inode, struct quota_res *quota_res, s64 sectors) { - if (!sectors) - return; - - mutex_lock(&inode->ei_quota_lock); - BUG_ON((s64) inode->v.i_blocks + sectors < 0); + bch2_fs_inconsistent_on((s64) inode->v.i_blocks + sectors < 0, c, + "inode %lu i_blocks underflow: %llu + %lli < 0 (ondisk %lli)", + inode->v.i_ino, (u64) inode->v.i_blocks, sectors, + inode->ei_inode.bi_sectors); inode->v.i_blocks += sectors; #ifdef CONFIG_BCACHEFS_QUOTA @@ -237,7 +320,16 @@ static void i_sectors_acct(struct bch_fs *c, struct bch_inode_info *inode, bch2_quota_acct(c, inode->ei_qid, Q_SPC, sectors, KEY_TYPE_QUOTA_WARN); } #endif - mutex_unlock(&inode->ei_quota_lock); +} + +static void i_sectors_acct(struct bch_fs *c, struct bch_inode_info *inode, + struct quota_res *quota_res, s64 sectors) +{ + if (sectors) { + mutex_lock(&inode->ei_quota_lock); + __i_sectors_acct(c, inode, quota_res, sectors); + mutex_unlock(&inode->ei_quota_lock); + } } /* page state: */ @@ -285,28 +377,13 @@ static inline struct bch_page_state *bch2_page_state(struct page *page) /* for newly allocated pages: */ static void __bch2_page_state_release(struct page *page) { - struct bch_page_state *s = __bch2_page_state(page); - - if (!s) - return; - - ClearPagePrivate(page); - set_page_private(page, 0); - put_page(page); - kfree(s); + kfree(detach_page_private(page)); } static void bch2_page_state_release(struct page *page) { - struct bch_page_state *s = bch2_page_state(page); - - if (!s) - return; - - ClearPagePrivate(page); - set_page_private(page, 0); - put_page(page); - kfree(s); + EBUG_ON(!PageLocked(page)); + __bch2_page_state_release(page); } /* for newly allocated pages: */ @@ -320,13 +397,7 @@ static struct bch_page_state *__bch2_page_state_create(struct page *page, return NULL; spin_lock_init(&s->lock); - /* - * migrate_page_move_mapping() assumes that pages with private data - * have their count elevated by 1. - */ - get_page(page); - set_page_private(page, (unsigned long) s); - SetPagePrivate(page); + attach_page_private(page, s); return s; } @@ -336,11 +407,11 @@ static struct bch_page_state *bch2_page_state_create(struct page *page, return bch2_page_state(page) ?: __bch2_page_state_create(page, gfp); } -static unsigned bkey_to_sector_state(const struct bkey *k) +static unsigned bkey_to_sector_state(struct bkey_s_c k) { - if (k->type == KEY_TYPE_reservation) + if (bkey_extent_is_reservation(k)) return SECTOR_RESERVED; - if (bkey_extent_is_allocation(k)) + if (bkey_extent_is_allocation(k.k)) return SECTOR_ALLOCATED; return SECTOR_UNALLOCATED; } @@ -391,7 +462,7 @@ retry: SPOS(inum.inum, offset, snapshot), BTREE_ITER_SLOTS, k, ret) { unsigned nr_ptrs = bch2_bkey_nr_ptrs_fully_allocated(k); - unsigned state = bkey_to_sector_state(k.k); + unsigned state = bkey_to_sector_state(k); while (pg_idx < nr_pages) { struct page *page = pages[pg_idx]; @@ -418,7 +489,7 @@ retry: offset = iter.pos.offset; bch2_trans_iter_exit(&trans, &iter); err: - if (ret == -EINTR) + if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) goto retry; bch2_trans_exit(&trans); @@ -431,7 +502,7 @@ static void bch2_bio_page_state_set(struct bio *bio, struct bkey_s_c k) struct bio_vec bv; unsigned nr_ptrs = k.k->type == KEY_TYPE_reflink_v ? 0 : bch2_bkey_nr_ptrs_fully_allocated(k); - unsigned state = bkey_to_sector_state(k.k); + unsigned state = bkey_to_sector_state(k); bio_for_each_segment(bv, bio, iter) __bch2_page_state_set(bv.bv_page, bv.bv_offset >> 9, @@ -443,22 +514,20 @@ static void mark_pagecache_unallocated(struct bch_inode_info *inode, { pgoff_t index = start >> PAGE_SECTORS_SHIFT; pgoff_t end_index = (end - 1) >> PAGE_SECTORS_SHIFT; - struct pagevec pvec; + struct folio_batch fbatch; + unsigned i, j; if (end <= start) return; - pagevec_init(&pvec); - - do { - unsigned nr_pages, i, j; + folio_batch_init(&fbatch); - nr_pages = pagevec_lookup_range(&pvec, inode->v.i_mapping, - &index, end_index); - for (i = 0; i < nr_pages; i++) { - struct page *page = pvec.pages[i]; - u64 pg_start = page->index << PAGE_SECTORS_SHIFT; - u64 pg_end = (page->index + 1) << PAGE_SECTORS_SHIFT; + while (filemap_get_folios(inode->v.i_mapping, + &index, end_index, &fbatch)) { + for (i = 0; i < folio_batch_count(&fbatch); i++) { + struct folio *folio = fbatch.folios[i]; + u64 pg_start = folio->index << PAGE_SECTORS_SHIFT; + u64 pg_end = (folio->index + 1) << PAGE_SECTORS_SHIFT; unsigned pg_offset = max(start, pg_start) - pg_start; unsigned pg_len = min(end, pg_end) - pg_offset - pg_start; struct bch_page_state *s; @@ -467,8 +536,8 @@ static void mark_pagecache_unallocated(struct bch_inode_info *inode, BUG_ON(pg_offset >= PAGE_SECTORS); BUG_ON(pg_offset + pg_len > PAGE_SECTORS); - lock_page(page); - s = bch2_page_state(page); + folio_lock(folio); + s = bch2_page_state(&folio->page); if (s) { spin_lock(&s->lock); @@ -477,10 +546,11 @@ static void mark_pagecache_unallocated(struct bch_inode_info *inode, spin_unlock(&s->lock); } - unlock_page(page); + folio_unlock(folio); } - pagevec_release(&pvec); - } while (index <= end_index); + folio_batch_release(&fbatch); + cond_resched(); + } } static void mark_pagecache_reserved(struct bch_inode_info *inode, @@ -489,23 +559,21 @@ static void mark_pagecache_reserved(struct bch_inode_info *inode, struct bch_fs *c = inode->v.i_sb->s_fs_info; pgoff_t index = start >> PAGE_SECTORS_SHIFT; pgoff_t end_index = (end - 1) >> PAGE_SECTORS_SHIFT; - struct pagevec pvec; + struct folio_batch fbatch; s64 i_sectors_delta = 0; + unsigned i, j; if (end <= start) return; - pagevec_init(&pvec); - - do { - unsigned nr_pages, i, j; + folio_batch_init(&fbatch); - nr_pages = pagevec_lookup_range(&pvec, inode->v.i_mapping, - &index, end_index); - for (i = 0; i < nr_pages; i++) { - struct page *page = pvec.pages[i]; - u64 pg_start = page->index << PAGE_SECTORS_SHIFT; - u64 pg_end = (page->index + 1) << PAGE_SECTORS_SHIFT; + while (filemap_get_folios(inode->v.i_mapping, + &index, end_index, &fbatch)) { + for (i = 0; i < folio_batch_count(&fbatch); i++) { + struct folio *folio = fbatch.folios[i]; + u64 pg_start = folio->index << PAGE_SECTORS_SHIFT; + u64 pg_end = (folio->index + 1) << PAGE_SECTORS_SHIFT; unsigned pg_offset = max(start, pg_start) - pg_start; unsigned pg_len = min(end, pg_end) - pg_offset - pg_start; struct bch_page_state *s; @@ -514,8 +582,8 @@ static void mark_pagecache_reserved(struct bch_inode_info *inode, BUG_ON(pg_offset >= PAGE_SECTORS); BUG_ON(pg_offset + pg_len > PAGE_SECTORS); - lock_page(page); - s = bch2_page_state(page); + folio_lock(folio); + s = bch2_page_state(&folio->page); if (s) { spin_lock(&s->lock); @@ -534,10 +602,11 @@ static void mark_pagecache_reserved(struct bch_inode_info *inode, spin_unlock(&s->lock); } - unlock_page(page); + folio_unlock(folio); } - pagevec_release(&pvec); - } while (index <= end_index); + folio_batch_release(&fbatch); + cond_resched(); + } i_sectors_acct(c, inode, NULL, i_sectors_delta); } @@ -617,7 +686,7 @@ static void bch2_page_reservation_put(struct bch_fs *c, static int bch2_page_reservation_get(struct bch_fs *c, struct bch_inode_info *inode, struct page *page, struct bch2_page_reservation *res, - unsigned offset, unsigned len, bool check_enospc) + unsigned offset, unsigned len) { struct bch_page_state *s = bch2_page_state_create(page, 0); unsigned i, disk_sectors = 0, quota_sectors = 0; @@ -637,19 +706,14 @@ static int bch2_page_reservation_get(struct bch_fs *c, } if (disk_sectors) { - ret = bch2_disk_reservation_add(c, &res->disk, - disk_sectors, - !check_enospc - ? BCH_DISK_RESERVATION_NOFAIL - : 0); + ret = bch2_disk_reservation_add(c, &res->disk, disk_sectors, 0); if (unlikely(ret)) return ret; } if (quota_sectors) { ret = bch2_quota_reservation_add(c, inode, &res->quota, - quota_sectors, - check_enospc); + quota_sectors, true); if (unlikely(ret)) { struct disk_reservation tmp = { .sectors = disk_sectors @@ -748,7 +812,7 @@ static void bch2_set_page_dirty(struct bch_fs *c, i_sectors_acct(c, inode, &res->quota, dirty_sectors); if (!PageDirty(page)) - __set_page_dirty_nobuffers(page); + filemap_dirty_folio(inode->v.i_mapping, page_folio(page)); } vm_fault_t bch2_page_fault(struct vm_fault *vmf) @@ -766,25 +830,25 @@ vm_fault_t bch2_page_fault(struct vm_fault *vmf) if (fdm > mapping) { struct bch_inode_info *fdm_host = to_bch_ei(fdm->host); - if (bch2_pagecache_add_tryget(&inode->ei_pagecache_lock)) + if (bch2_pagecache_add_tryget(inode)) goto got_lock; - bch2_pagecache_block_put(&fdm_host->ei_pagecache_lock); + bch2_pagecache_block_put(fdm_host); - bch2_pagecache_add_get(&inode->ei_pagecache_lock); - bch2_pagecache_add_put(&inode->ei_pagecache_lock); + bch2_pagecache_add_get(inode); + bch2_pagecache_add_put(inode); - bch2_pagecache_block_get(&fdm_host->ei_pagecache_lock); + bch2_pagecache_block_get(fdm_host); /* Signal that lock has been dropped: */ set_fdm_dropped_locks(); return VM_FAULT_SIGBUS; } - bch2_pagecache_add_get(&inode->ei_pagecache_lock); + bch2_pagecache_add_get(inode); got_lock: ret = filemap_fault(vmf); - bch2_pagecache_add_put(&inode->ei_pagecache_lock); + bch2_pagecache_add_put(inode); return ret; } @@ -812,7 +876,7 @@ vm_fault_t bch2_page_mkwrite(struct vm_fault *vmf) * a write_invalidate_inode_pages_range() that works without dropping * page lock before invalidating page */ - bch2_pagecache_add_get(&inode->ei_pagecache_lock); + bch2_pagecache_add_get(inode); lock_page(page); isize = i_size_read(&inode->v); @@ -833,7 +897,7 @@ vm_fault_t bch2_page_mkwrite(struct vm_fault *vmf) } } - if (bch2_page_reservation_get(c, inode, page, &res, 0, len, true)) { + if (bch2_page_reservation_get(c, inode, page, &res, 0, len)) { unlock_page(page); ret = VM_FAULT_SIGBUS; goto out; @@ -845,68 +909,37 @@ vm_fault_t bch2_page_mkwrite(struct vm_fault *vmf) wait_for_stable_page(page); ret = VM_FAULT_LOCKED; out: - bch2_pagecache_add_put(&inode->ei_pagecache_lock); + bch2_pagecache_add_put(inode); sb_end_pagefault(inode->v.i_sb); return ret; } -void bch2_invalidatepage(struct page *page, unsigned int offset, - unsigned int length) +void bch2_invalidate_folio(struct folio *folio, size_t offset, size_t length) { - if (offset || length < PAGE_SIZE) + if (offset || length < folio_size(folio)) return; - bch2_clear_page_bits(page); + bch2_clear_page_bits(&folio->page); } -int bch2_releasepage(struct page *page, gfp_t gfp_mask) +bool bch2_release_folio(struct folio *folio, gfp_t gfp_mask) { - if (PageDirty(page)) - return 0; + if (folio_test_dirty(folio) || folio_test_writeback(folio)) + return false; - bch2_clear_page_bits(page); - return 1; + bch2_clear_page_bits(&folio->page); + return true; } -#ifdef CONFIG_MIGRATION -int bch2_migrate_page(struct address_space *mapping, struct page *newpage, - struct page *page, enum migrate_mode mode) -{ - int ret; - - EBUG_ON(!PageLocked(page)); - EBUG_ON(!PageLocked(newpage)); - - ret = migrate_page_move_mapping(mapping, newpage, page, NULL, mode, 0); - if (ret != MIGRATEPAGE_SUCCESS) - return ret; - - if (PagePrivate(page)) { - ClearPagePrivate(page); - get_page(newpage); - set_page_private(newpage, page_private(page)); - set_page_private(page, 0); - put_page(page); - SetPagePrivate(newpage); - } - - if (mode != MIGRATE_SYNC_NO_COPY) - migrate_page_copy(newpage, page); - else - migrate_page_states(newpage, page); - return MIGRATEPAGE_SUCCESS; -} -#endif - /* readpage(s): */ static void bch2_readpages_end_io(struct bio *bio) { + struct bvec_iter_all iter; struct bio_vec *bv; - unsigned i; - bio_for_each_segment_all(bv, bio, i) { + bio_for_each_segment_all(bv, bio, iter) { struct page *page = bv->bv_page; if (!bio->bi_status) { @@ -925,31 +958,29 @@ struct readpages_iter { struct address_space *mapping; struct page **pages; unsigned nr_pages; - unsigned nr_added; unsigned idx; pgoff_t offset; }; static int readpages_iter_init(struct readpages_iter *iter, - struct address_space *mapping, - struct list_head *pages, unsigned nr_pages) + struct readahead_control *ractl) { + unsigned i, nr_pages = readahead_count(ractl); + memset(iter, 0, sizeof(*iter)); - iter->mapping = mapping; - iter->offset = list_last_entry(pages, struct page, lru)->index; + iter->mapping = ractl->mapping; + iter->offset = readahead_index(ractl); + iter->nr_pages = nr_pages; iter->pages = kmalloc_array(nr_pages, sizeof(struct page *), GFP_NOFS); if (!iter->pages) return -ENOMEM; - while (!list_empty(pages)) { - struct page *page = list_last_entry(pages, struct page, lru); - - __bch2_page_state_create(page, __GFP_NOFAIL); - - iter->pages[iter->nr_pages++] = page; - list_del(&page->lru); + nr_pages = __readahead_batch(ractl, iter->pages, nr_pages); + for (i = 0; i < nr_pages; i++) { + __bch2_page_state_create(iter->pages[i], __GFP_NOFAIL); + put_page(iter->pages[i]); } return 0; @@ -957,41 +988,9 @@ static int readpages_iter_init(struct readpages_iter *iter, static inline struct page *readpage_iter_next(struct readpages_iter *iter) { - struct page *page; - unsigned i; - int ret; - - BUG_ON(iter->idx > iter->nr_added); - BUG_ON(iter->nr_added > iter->nr_pages); - - if (iter->idx < iter->nr_added) - goto out; - - while (1) { - if (iter->idx == iter->nr_pages) - return NULL; - - ret = add_to_page_cache_lru_vec(iter->mapping, - iter->pages + iter->nr_added, - iter->nr_pages - iter->nr_added, - iter->offset + iter->nr_added, - GFP_NOFS); - if (ret > 0) - break; - - page = iter->pages[iter->nr_added]; - iter->idx++; - iter->nr_added++; - - __bch2_page_state_release(page); - put_page(page); - } - - iter->nr_added += ret; + if (iter->idx >= iter->nr_pages) + return NULL; - for (i = iter->idx; i < iter->nr_added; i++) - put_page(iter->pages[i]); -out: EBUG_ON(iter->pages[iter->idx]->index != iter->offset + iter->idx); return iter->pages[iter->idx]; @@ -1029,11 +1028,8 @@ static void readpage_bio_extend(struct readpages_iter *iter, if (!get_more) break; - rcu_read_lock(); - page = radix_tree_lookup(&iter->mapping->i_pages, page_offset); - rcu_read_unlock(); - - if (page && !radix_tree_exceptional_entry(page)) + page = xa_load(&iter->mapping->i_pages, page_offset); + if (page && !xa_is_value(page)) break; page = __page_cache_alloc(readahead_gfp_mask(iter->mapping)); @@ -1098,10 +1094,9 @@ retry: * read_extent -> io_time_reset may cause a transaction restart * without returning an error, we need to check for that here: */ - if (!bch2_trans_relock(trans)) { - ret = -EINTR; + ret = bch2_trans_relock(trans); + if (ret) break; - } bch2_btree_iter_set_pos(&iter, POS(inum.inum, rbio->bio.bi_iter.bi_sector)); @@ -1126,8 +1121,6 @@ retry: sectors = min(sectors, k.k->size - offset_into_extent); - bch2_trans_unlock(trans); - if (readpages_iter) readpage_bio_extend(readpages_iter, &rbio->bio, sectors, extent_partial_reads_expensive(k)); @@ -1156,11 +1149,13 @@ retry: err: bch2_trans_iter_exit(trans, &iter); - if (ret == -EINTR) + if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) goto retry; if (ret) { - bch_err_inum_ratelimited(c, inum.inum, + bch_err_inum_offset_ratelimited(c, + iter.pos.inode, + iter.pos.offset << 9, "read error %i from btree lookup", ret); rbio->bio.bi_status = BLK_STS_IOERR; bio_endio(&rbio->bio); @@ -1169,37 +1164,38 @@ err: bch2_bkey_buf_exit(&sk, c); } -int bch2_readpages(struct file *file, struct address_space *mapping, - struct list_head *pages, unsigned nr_pages) +void bch2_readahead(struct readahead_control *ractl) { - struct bch_inode_info *inode = to_bch_ei(mapping->host); + struct bch_inode_info *inode = to_bch_ei(ractl->mapping->host); struct bch_fs *c = inode->v.i_sb->s_fs_info; - struct bch_io_opts opts = io_opts(c, &inode->ei_inode); + struct bch_io_opts opts; struct btree_trans trans; struct page *page; struct readpages_iter readpages_iter; int ret; - ret = readpages_iter_init(&readpages_iter, mapping, pages, nr_pages); + bch2_inode_opts_get(&opts, c, &inode->ei_inode); + + ret = readpages_iter_init(&readpages_iter, ractl); BUG_ON(ret); bch2_trans_init(&trans, c, 0, 0); - bch2_pagecache_add_get(&inode->ei_pagecache_lock); + bch2_pagecache_add_get(inode); while ((page = readpage_iter_next(&readpages_iter))) { pgoff_t index = readpages_iter.offset + readpages_iter.idx; unsigned n = min_t(unsigned, readpages_iter.nr_pages - readpages_iter.idx, - BIO_MAX_PAGES); + BIO_MAX_VECS); struct bch_read_bio *rbio = - rbio_init(bio_alloc_bioset(GFP_NOFS, n, &c->bio_read), + rbio_init(bio_alloc_bioset(NULL, n, REQ_OP_READ, + GFP_NOFS, &c->bio_read), opts); readpages_iter.idx++; - bio_set_op_attrs(&rbio->bio, REQ_OP_READ, 0); rbio->bio.bi_iter.bi_sector = (sector_t) index << PAGE_SECTORS_SHIFT; rbio->bio.bi_end_io = bch2_readpages_end_io; BUG_ON(!bio_add_page(&rbio->bio, page, PAGE_SIZE, 0)); @@ -1208,12 +1204,10 @@ int bch2_readpages(struct file *file, struct address_space *mapping, &readpages_iter); } - bch2_pagecache_add_put(&inode->ei_pagecache_lock); + bch2_pagecache_add_put(inode); bch2_trans_exit(&trans); kfree(readpages_iter.pages); - - return 0; } static void __bchfs_readpage(struct bch_fs *c, struct bch_read_bio *rbio, @@ -1223,7 +1217,7 @@ static void __bchfs_readpage(struct bch_fs *c, struct bch_read_bio *rbio, bch2_page_state_create(page, __GFP_NOFAIL); - bio_set_op_attrs(&rbio->bio, REQ_OP_READ, REQ_SYNC); + rbio->bio.bi_opf = REQ_OP_READ|REQ_SYNC; rbio->bio.bi_iter.bi_sector = (sector_t) page->index << PAGE_SECTORS_SHIFT; BUG_ON(!bio_add_page(&rbio->bio, page, PAGE_SIZE, 0)); @@ -1233,20 +1227,6 @@ static void __bchfs_readpage(struct bch_fs *c, struct bch_read_bio *rbio, bch2_trans_exit(&trans); } -int bch2_readpage(struct file *file, struct page *page) -{ - struct bch_inode_info *inode = to_bch_ei(page->mapping->host); - struct bch_fs *c = inode->v.i_sb->s_fs_info; - struct bch_io_opts opts = io_opts(c, &inode->ei_inode); - struct bch_read_bio *rbio; - - rbio = rbio_init(bio_alloc_bioset(GFP_NOFS, 1, &c->bio_read), opts); - rbio->bio.bi_end_io = bch2_readpages_end_io; - - __bchfs_readpage(c, rbio, inode_inum(inode), page); - return 0; -} - static void bch2_read_single_page_end_io(struct bio *bio) { complete(bio->bi_private); @@ -1258,11 +1238,14 @@ static int bch2_read_single_page(struct page *page, struct bch_inode_info *inode = to_bch_ei(mapping->host); struct bch_fs *c = inode->v.i_sb->s_fs_info; struct bch_read_bio *rbio; + struct bch_io_opts opts; int ret; DECLARE_COMPLETION_ONSTACK(done); - rbio = rbio_init(bio_alloc_bioset(GFP_NOFS, 1, &c->bio_read), - io_opts(c, &inode->ei_inode)); + bch2_inode_opts_get(&opts, c, &inode->ei_inode); + + rbio = rbio_init(bio_alloc_bioset(NULL, 1, REQ_OP_READ, GFP_NOFS, &c->bio_read), + opts); rbio->bio.bi_private = &done; rbio->bio.bi_end_io = bch2_read_single_page_end_io; @@ -1279,6 +1262,16 @@ static int bch2_read_single_page(struct page *page, return 0; } +int bch2_read_folio(struct file *file, struct folio *folio) +{ + struct page *page = &folio->page; + int ret; + + ret = bch2_read_single_page(page, page->mapping); + folio_unlock(folio); + return bch2_err_class(ret); +} + /* writepages: */ struct bch_writepage_state { @@ -1289,55 +1282,47 @@ struct bch_writepage_state { static inline struct bch_writepage_state bch_writepage_state_init(struct bch_fs *c, struct bch_inode_info *inode) { - return (struct bch_writepage_state) { - .opts = io_opts(c, &inode->ei_inode) - }; -} + struct bch_writepage_state ret = { 0 }; -static void bch2_writepage_io_free(struct closure *cl) -{ - struct bch_writepage_io *io = container_of(cl, - struct bch_writepage_io, cl); - - bio_put(&io->op.wbio.bio); + bch2_inode_opts_get(&ret.opts, c, &inode->ei_inode); + return ret; } -static void bch2_writepage_io_done(struct closure *cl) +static void bch2_writepage_io_done(struct bch_write_op *op) { - struct bch_writepage_io *io = container_of(cl, - struct bch_writepage_io, cl); + struct bch_writepage_io *io = + container_of(op, struct bch_writepage_io, op); struct bch_fs *c = io->op.c; struct bio *bio = &io->op.wbio.bio; + struct bvec_iter_all iter; struct bio_vec *bvec; - unsigned i, j; - - up(&io->op.c->io_in_flight); + unsigned i; if (io->op.error) { set_bit(EI_INODE_ERROR, &io->inode->ei_flags); - bio_for_each_segment_all(bvec, bio, i) { + bio_for_each_segment_all(bvec, bio, iter) { struct bch_page_state *s; SetPageError(bvec->bv_page); - mapping_set_error(io->inode->v.i_mapping, -EIO); + mapping_set_error(bvec->bv_page->mapping, -EIO); s = __bch2_page_state(bvec->bv_page); spin_lock(&s->lock); - for (j = 0; j < PAGE_SECTORS; j++) - s->s[j].nr_replicas = 0; + for (i = 0; i < PAGE_SECTORS; i++) + s->s[i].nr_replicas = 0; spin_unlock(&s->lock); } } if (io->op.flags & BCH_WRITE_WROTE_DATA_INLINE) { - bio_for_each_segment_all(bvec, bio, i) { + bio_for_each_segment_all(bvec, bio, iter) { struct bch_page_state *s; s = __bch2_page_state(bvec->bv_page); spin_lock(&s->lock); - for (j = 0; j < PAGE_SECTORS; j++) - s->s[j].nr_replicas = 0; + for (i = 0; i < PAGE_SECTORS; i++) + s->s[i].nr_replicas = 0; spin_unlock(&s->lock); } } @@ -1346,7 +1331,7 @@ static void bch2_writepage_io_done(struct closure *cl) * racing with fallocate can cause us to add fewer sectors than * expected - but we shouldn't add more sectors than expected: */ - WARN_ON(io->op.i_sectors_delta > 0); + WARN_ON_ONCE(io->op.i_sectors_delta > 0); /* * (error (due to going RO) halfway through a page can screw that up @@ -1361,25 +1346,22 @@ static void bch2_writepage_io_done(struct closure *cl) */ i_sectors_acct(c, io->inode, NULL, io->op.i_sectors_delta); - bio_for_each_segment_all(bvec, bio, i) { + bio_for_each_segment_all(bvec, bio, iter) { struct bch_page_state *s = __bch2_page_state(bvec->bv_page); if (atomic_dec_and_test(&s->write_count)) end_page_writeback(bvec->bv_page); } - closure_return_with_destructor(&io->cl, bch2_writepage_io_free); + bio_put(&io->op.wbio.bio); } static void bch2_writepage_do_io(struct bch_writepage_state *w) { struct bch_writepage_io *io = w->io; - down(&io->op.c->io_in_flight); - w->io = NULL; - closure_call(&io->op.cl, bch2_write, NULL, &io->cl); - continue_at(&io->cl, bch2_writepage_io_done, NULL); + closure_call(&io->op.cl, bch2_write, NULL, NULL); } /* @@ -1395,13 +1377,13 @@ static void bch2_writepage_io_alloc(struct bch_fs *c, { struct bch_write_op *op; - w->io = container_of(bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, + w->io = container_of(bio_alloc_bioset(NULL, BIO_MAX_VECS, + REQ_OP_WRITE, + GFP_NOFS, &c->writepage_bioset), struct bch_writepage_io, op.wbio.bio); - closure_init(&w->io->cl, NULL); w->io->inode = inode; - op = &w->io->op; bch2_write_op_init(op, c, w->opts); op->target = w->opts.foreground_target; @@ -1410,6 +1392,8 @@ static void bch2_writepage_io_alloc(struct bch_fs *c, op->write_point = writepoint_hashed(inode->ei_last_dirtied); op->subvol = inode->ei_subvol; op->pos = POS(inode->v.i_ino, sector); + op->end_io = bch2_writepage_io_done; + op->devs_need_flush = &inode->ei_devs_need_flush; op->wbio.bio.bi_iter.bi_sector = sector; op->wbio.bio.bi_opf = wbc_to_write_flags(wbc); } @@ -1515,9 +1499,9 @@ do_io: if (w->io && (w->io->op.res.nr_replicas != nr_replicas_this_write || - bio_full(&w->io->op.wbio.bio) || + bio_full(&w->io->op.wbio.bio, PAGE_SIZE) || w->io->op.wbio.bio.bi_iter.bi_size + (sectors << 9) >= - (BIO_MAX_PAGES * PAGE_SIZE) || + (BIO_MAX_VECS * PAGE_SIZE) || bio_end_sector(&w->io->op.wbio.bio) != sector)) bch2_writepage_do_io(w); @@ -1532,8 +1516,13 @@ do_io: sectors << 9, offset << 9)); /* Check for writing past i_size: */ - WARN_ON((bio_end_sector(&w->io->op.wbio.bio) << 9) > - round_up(i_size, block_bytes(c))); + WARN_ONCE((bio_end_sector(&w->io->op.wbio.bio) << 9) > + round_up(i_size, block_bytes(c)) && + !test_bit(BCH_FS_EMERGENCY_RO, &c->flags), + "writing past i_size: %llu > %llu (unrounded %llu)\n", + bio_end_sector(&w->io->op.wbio.bio) << 9, + round_up(i_size, block_bytes(c)), + i_size); w->io->op.res.sectors += reserved_sectors; w->io->op.i_sectors_delta -= dirty_sectors; @@ -1561,27 +1550,13 @@ int bch2_writepages(struct address_space *mapping, struct writeback_control *wbc if (w.io) bch2_writepage_do_io(&w); blk_finish_plug(&plug); - return ret; -} - -int bch2_writepage(struct page *page, struct writeback_control *wbc) -{ - struct bch_fs *c = page->mapping->host->i_sb->s_fs_info; - struct bch_writepage_state w = - bch_writepage_state_init(c, to_bch_ei(page->mapping->host)); - int ret; - - ret = __bch2_writepage(page, wbc, &w); - if (w.io) - bch2_writepage_do_io(&w); - - return ret; + return bch2_err_class(ret); } /* buffered writes: */ int bch2_write_begin(struct file *file, struct address_space *mapping, - loff_t pos, unsigned len, unsigned flags, + loff_t pos, unsigned len, struct page **pagep, void **fsdata) { struct bch_inode_info *inode = to_bch_ei(mapping->host); @@ -1599,9 +1574,9 @@ int bch2_write_begin(struct file *file, struct address_space *mapping, bch2_page_reservation_init(c, inode, res); *fsdata = res; - bch2_pagecache_add_get(&inode->ei_pagecache_lock); + bch2_pagecache_add_get(inode); - page = grab_cache_page_write_begin(mapping, index, flags); + page = grab_cache_page_write_begin(mapping, index); if (!page) goto err_unlock; @@ -1631,11 +1606,10 @@ out: if (!bch2_page_state_create(page, __GFP_NOFAIL)->uptodate) { ret = bch2_page_state_set(c, inode_inum(inode), &page, 1); if (ret) - goto out; + goto err; } - ret = bch2_page_reservation_get(c, inode, page, res, - offset, len, true); + ret = bch2_page_reservation_get(c, inode, page, res, offset, len); if (ret) { if (!PageUptodate(page)) { /* @@ -1657,10 +1631,10 @@ err: put_page(page); *pagep = NULL; err_unlock: - bch2_pagecache_add_put(&inode->ei_pagecache_lock); + bch2_pagecache_add_put(inode); kfree(res); *fsdata = NULL; - return ret; + return bch2_err_class(ret); } int bch2_write_end(struct file *file, struct address_space *mapping, @@ -1701,7 +1675,7 @@ int bch2_write_end(struct file *file, struct address_space *mapping, unlock_page(page); put_page(page); - bch2_pagecache_add_put(&inode->ei_pagecache_lock); + bch2_pagecache_add_put(inode); bch2_page_reservation_put(c, inode, res); kfree(res); @@ -1732,7 +1706,7 @@ static int __bch2_buffered_write(struct bch_inode_info *inode, bch2_page_reservation_init(c, inode, &res); for (i = 0; i < nr_pages; i++) { - pages[i] = grab_cache_page_write_begin(mapping, index + i, 0); + pages[i] = grab_cache_page_write_begin(mapping, index + i); if (!pages[i]) { nr_pages = i; if (!i) { @@ -1776,10 +1750,21 @@ static int __bch2_buffered_write(struct bch_inode_info *inode, goto out; } + /* + * XXX: per POSIX and fstests generic/275, on -ENOSPC we're + * supposed to write as much as we have disk space for. + * + * On failure here we should still write out a partial page if + * we aren't completely out of disk space - we don't do that + * yet: + */ ret = bch2_page_reservation_get(c, inode, page, &res, - pg_offset, pg_len, true); - if (ret) - goto out; + pg_offset, pg_len); + if (unlikely(ret)) { + if (!reserved) + goto out; + break; + } reserved += pg_len; } @@ -1788,13 +1773,13 @@ static int __bch2_buffered_write(struct bch_inode_info *inode, for (i = 0; i < nr_pages; i++) flush_dcache_page(pages[i]); - while (copied < len) { + while (copied < reserved) { struct page *page = pages[(offset + copied) >> PAGE_SHIFT]; unsigned pg_offset = (offset + copied) & (PAGE_SIZE - 1); - unsigned pg_len = min_t(unsigned, len - copied, + unsigned pg_len = min_t(unsigned, reserved - copied, PAGE_SIZE - pg_offset); - unsigned pg_copied = iov_iter_copy_from_user_atomic(page, - iter, pg_offset, pg_len); + unsigned pg_copied = copy_page_from_iter_atomic(page, + pg_offset, pg_len, iter); if (!pg_copied) break; @@ -1807,7 +1792,6 @@ static int __bch2_buffered_write(struct bch_inode_info *inode, } flush_dcache_page(page); - iov_iter_advance(iter, pg_copied); copied += pg_copied; if (pg_copied != pg_len) @@ -1860,7 +1844,7 @@ static ssize_t bch2_buffered_write(struct kiocb *iocb, struct iov_iter *iter) ssize_t written = 0; int ret = 0; - bch2_pagecache_add_get(&inode->ei_pagecache_lock); + bch2_pagecache_add_get(inode); do { unsigned offset = pos & (PAGE_SIZE - 1); @@ -1877,11 +1861,11 @@ again: * to check that the address is actually valid, when atomic * usercopies are used, below. */ - if (unlikely(iov_iter_fault_in_readable(iter, bytes))) { + if (unlikely(fault_in_iov_iter_readable(iter, bytes))) { bytes = min_t(unsigned long, iov_iter_count(iter), PAGE_SIZE - offset); - if (unlikely(iov_iter_fault_in_readable(iter, bytes))) { + if (unlikely(fault_in_iov_iter_readable(iter, bytes))) { ret = -EFAULT; break; } @@ -1918,25 +1902,13 @@ again: balance_dirty_pages_ratelimited(mapping); } while (iov_iter_count(iter)); - bch2_pagecache_add_put(&inode->ei_pagecache_lock); + bch2_pagecache_add_put(inode); return written ? written : ret; } /* O_DIRECT reads */ -static void bio_release_pages(struct bio *bio, bool mark_dirty) -{ - struct bio_vec *bvec; - unsigned i; - - bio_for_each_segment_all(bvec, bio, i) { - if (mark_dirty && !PageCompound(bvec->bv_page)) - set_page_dirty_lock(bvec->bv_page); - put_page(bvec->bv_page); - } -} - static void bio_check_or_release(struct bio *bio, bool check_dirty) { if (check_dirty) { @@ -1951,7 +1923,7 @@ static void bch2_dio_read_complete(struct closure *cl) { struct dio_read *dio = container_of(cl, struct dio_read, cl); - dio->req->ki_complete(dio->req, dio->ret, 0); + dio->req->ki_complete(dio->req, dio->ret); bio_check_or_release(&dio->rbio.bio, dio->should_dirty); } @@ -1979,7 +1951,7 @@ static int bch2_direct_IO_read(struct kiocb *req, struct iov_iter *iter) struct file *file = req->ki_filp; struct bch_inode_info *inode = file_bch_inode(file); struct bch_fs *c = inode->v.i_sb->s_fs_info; - struct bch_io_opts opts = io_opts(c, &inode->ei_inode); + struct bch_io_opts opts; struct dio_read *dio; struct bio *bio; loff_t offset = req->ki_pos; @@ -1987,6 +1959,8 @@ static int bch2_direct_IO_read(struct kiocb *req, struct iov_iter *iter) size_t shorten; ssize_t ret; + bch2_inode_opts_get(&opts, c, &inode->ei_inode); + if ((offset|iter->count) & (block_bytes(c) - 1)) return -EINVAL; @@ -1999,8 +1973,10 @@ static int bch2_direct_IO_read(struct kiocb *req, struct iov_iter *iter) shorten = iov_iter_count(iter) - round_up(ret, block_bytes(c)); iter->count -= shorten; - bio = bio_alloc_bioset(GFP_KERNEL, - iov_iter_npages(iter, BIO_MAX_PAGES), + bio = bio_alloc_bioset(NULL, + bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS), + REQ_OP_READ, + GFP_KERNEL, &c->dio_read_bioset); bio->bi_end_io = bch2_direct_IO_read_endio; @@ -2034,12 +2010,14 @@ static int bch2_direct_IO_read(struct kiocb *req, struct iov_iter *iter) goto start; while (iter->count) { - bio = bio_alloc_bioset(GFP_KERNEL, - iov_iter_npages(iter, BIO_MAX_PAGES), + bio = bio_alloc_bioset(NULL, + bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS), + REQ_OP_READ, + GFP_KERNEL, &c->bio_read); bio->bi_end_io = bch2_direct_IO_read_split_endio; start: - bio_set_op_attrs(bio, REQ_OP_READ, REQ_SYNC); + bio->bi_opf = REQ_OP_READ|REQ_SYNC; bio->bi_iter.bi_sector = offset >> 9; bio->bi_private = dio; @@ -2089,11 +2067,13 @@ ssize_t bch2_read_iter(struct kiocb *iocb, struct iov_iter *iter) if (iocb->ki_flags & IOCB_DIRECT) { struct blk_plug plug; - ret = filemap_write_and_wait_range(mapping, - iocb->ki_pos, - iocb->ki_pos + count - 1); - if (ret < 0) - return ret; + if (unlikely(mapping->nrpages)) { + ret = filemap_write_and_wait_range(mapping, + iocb->ki_pos, + iocb->ki_pos + count - 1); + if (ret < 0) + goto out; + } file_accessed(file); @@ -2104,12 +2084,12 @@ ssize_t bch2_read_iter(struct kiocb *iocb, struct iov_iter *iter) if (ret >= 0) iocb->ki_pos += ret; } else { - bch2_pagecache_add_get(&inode->ei_pagecache_lock); + bch2_pagecache_add_get(inode); ret = generic_file_read_iter(iocb, iter); - bch2_pagecache_add_put(&inode->ei_pagecache_lock); + bch2_pagecache_add_put(inode); } - - return ret; +out: + return bch2_err_class(ret); } /* O_DIRECT writes */ @@ -2137,7 +2117,7 @@ retry: for_each_btree_key_norestart(&trans, iter, BTREE_ID_extents, SPOS(inum.inum, offset, snapshot), BTREE_ITER_SLOTS, k, err) { - if (bkey_cmp(bkey_start_pos(k.k), POS(inum.inum, end)) >= 0) + if (bkey_ge(bkey_start_pos(k.k), POS(inum.inum, end))) break; if (k.k->p.snapshot != snapshot || @@ -2151,39 +2131,172 @@ retry: offset = iter.pos.offset; bch2_trans_iter_exit(&trans, &iter); err: - if (err == -EINTR) + if (bch2_err_matches(err, BCH_ERR_transaction_restart)) goto retry; bch2_trans_exit(&trans); return err ? false : ret; } +static noinline bool bch2_dio_write_check_allocated(struct dio_write *dio) +{ + struct bch_fs *c = dio->op.c; + struct bch_inode_info *inode = dio->inode; + struct bio *bio = &dio->op.wbio.bio; + + return bch2_check_range_allocated(c, inode_inum(inode), + dio->op.pos.offset, bio_sectors(bio), + dio->op.opts.data_replicas, + dio->op.opts.compression != 0); +} + static void bch2_dio_write_loop_async(struct bch_write_op *); +static __always_inline long bch2_dio_write_done(struct dio_write *dio); + +static noinline int bch2_dio_write_copy_iov(struct dio_write *dio) +{ + struct iovec *iov = dio->inline_vecs; + + if (dio->iter.nr_segs > ARRAY_SIZE(dio->inline_vecs)) { + iov = kmalloc_array(dio->iter.nr_segs, sizeof(*iov), + GFP_KERNEL); + if (unlikely(!iov)) + return -ENOMEM; + + dio->free_iov = true; + } -static long bch2_dio_write_loop(struct dio_write *dio) + memcpy(iov, dio->iter.iov, dio->iter.nr_segs * sizeof(*iov)); + dio->iter.iov = iov; + return 0; +} + +static void bch2_dio_write_flush_done(struct closure *cl) +{ + struct dio_write *dio = container_of(cl, struct dio_write, op.cl); + struct bch_fs *c = dio->op.c; + + closure_debug_destroy(cl); + + dio->op.error = bch2_journal_error(&c->journal); + + bch2_dio_write_done(dio); +} + +static noinline void bch2_dio_write_flush(struct dio_write *dio) +{ + struct bch_fs *c = dio->op.c; + struct bch_inode_unpacked inode; + int ret; + + dio->flush = 0; + + closure_init(&dio->op.cl, NULL); + + if (!dio->op.error) { + ret = bch2_inode_find_by_inum(c, inode_inum(dio->inode), &inode); + if (ret) { + dio->op.error = ret; + } else { + bch2_journal_flush_seq_async(&c->journal, inode.bi_journal_seq, &dio->op.cl); + bch2_inode_flush_nocow_writes_async(c, dio->inode, &dio->op.cl); + } + } + + if (dio->sync) { + closure_sync(&dio->op.cl); + closure_debug_destroy(&dio->op.cl); + } else { + continue_at(&dio->op.cl, bch2_dio_write_flush_done, NULL); + } +} + +static __always_inline long bch2_dio_write_done(struct dio_write *dio) { - bool kthread = (current->flags & PF_KTHREAD) != 0; struct kiocb *req = dio->req; - struct address_space *mapping = req->ki_filp->f_mapping; - struct bch_inode_info *inode = file_bch_inode(req->ki_filp); - struct bch_fs *c = inode->v.i_sb->s_fs_info; + struct bch_inode_info *inode = dio->inode; + bool sync = dio->sync; + long ret; + + if (unlikely(dio->flush)) { + bch2_dio_write_flush(dio); + if (!sync) + return -EIOCBQUEUED; + } + + bch2_pagecache_block_put(inode); + + if (dio->free_iov) + kfree(dio->iter.iov); + + ret = dio->op.error ?: ((long) dio->written << 9); + bio_put(&dio->op.wbio.bio); + + /* inode->i_dio_count is our ref on inode and thus bch_fs */ + inode_dio_end(&inode->v); + + if (ret < 0) + ret = bch2_err_class(ret); + + if (!sync) { + req->ki_complete(req, ret); + ret = -EIOCBQUEUED; + } + return ret; +} + +static __always_inline void bch2_dio_write_end(struct dio_write *dio) +{ + struct bch_fs *c = dio->op.c; + struct kiocb *req = dio->req; + struct bch_inode_info *inode = dio->inode; struct bio *bio = &dio->op.wbio.bio; + struct bvec_iter_all iter; struct bio_vec *bv; - unsigned i, unaligned, iter_count; + + req->ki_pos += (u64) dio->op.written << 9; + dio->written += dio->op.written; + + if (dio->extending) { + spin_lock(&inode->v.i_lock); + if (req->ki_pos > inode->v.i_size) + i_size_write(&inode->v, req->ki_pos); + spin_unlock(&inode->v.i_lock); + } + + if (dio->op.i_sectors_delta || dio->quota_res.sectors) { + mutex_lock(&inode->ei_quota_lock); + __i_sectors_acct(c, inode, &dio->quota_res, dio->op.i_sectors_delta); + __bch2_quota_reservation_put(c, inode, &dio->quota_res); + mutex_unlock(&inode->ei_quota_lock); + } + + if (likely(!bio_flagged(bio, BIO_NO_PAGE_REF))) + bio_for_each_segment_all(bv, bio, iter) + put_page(bv->bv_page); + + if (unlikely(dio->op.error)) + set_bit(EI_INODE_ERROR, &inode->ei_flags); +} + +static __always_inline long bch2_dio_write_loop(struct dio_write *dio) +{ + struct bch_fs *c = dio->op.c; + struct kiocb *req = dio->req; + struct address_space *mapping = dio->mapping; + struct bch_inode_info *inode = dio->inode; + struct bch_io_opts opts; + struct bio *bio = &dio->op.wbio.bio; + unsigned unaligned, iter_count; bool sync = dio->sync, dropped_locks; long ret; - if (dio->loop) - goto loop; - - down(&c->io_in_flight); + bch2_inode_opts_get(&opts, c, &inode->ei_inode); while (1) { iter_count = dio->iter.count; - if (kthread) - use_mm(dio->mm); - BUG_ON(current->faults_disabled_mapping); + EBUG_ON(current->faults_disabled_mapping); current->faults_disabled_mapping = mapping; ret = bio_iov_iter_get_pages(bio, &dio->iter); @@ -2191,8 +2304,6 @@ static long bch2_dio_write_loop(struct dio_write *dio) dropped_locks = fdm_dropped_locks(); current->faults_disabled_mapping = NULL; - if (kthread) - unuse_mm(dio->mm); /* * If the fault handler returned an error but also signalled @@ -2229,113 +2340,93 @@ static long bch2_dio_write_loop(struct dio_write *dio) goto err; } - bch2_write_op_init(&dio->op, c, io_opts(c, &inode->ei_inode)); - dio->op.end_io = bch2_dio_write_loop_async; + bch2_write_op_init(&dio->op, c, opts); + dio->op.end_io = sync + ? NULL + : bch2_dio_write_loop_async; dio->op.target = dio->op.opts.foreground_target; dio->op.write_point = writepoint_hashed((unsigned long) current); dio->op.nr_replicas = dio->op.opts.data_replicas; dio->op.subvol = inode->ei_subvol; dio->op.pos = POS(inode->v.i_ino, (u64) req->ki_pos >> 9); + dio->op.devs_need_flush = &inode->ei_devs_need_flush; - if ((req->ki_flags & IOCB_DSYNC) && - !c->opts.journal_flush_disabled) - dio->op.flags |= BCH_WRITE_FLUSH; + if (sync) + dio->op.flags |= BCH_WRITE_SYNC; dio->op.flags |= BCH_WRITE_CHECK_ENOSPC; + ret = bch2_quota_reservation_add(c, inode, &dio->quota_res, + bio_sectors(bio), true); + if (unlikely(ret)) + goto err; + ret = bch2_disk_reservation_get(c, &dio->op.res, bio_sectors(bio), dio->op.opts.data_replicas, 0); if (unlikely(ret) && - !bch2_check_range_allocated(c, inode_inum(inode), - dio->op.pos.offset, bio_sectors(bio), - dio->op.opts.data_replicas, - dio->op.opts.compression != 0)) + !bch2_dio_write_check_allocated(dio)) goto err; task_io_account_write(bio->bi_iter.bi_size); - if (!dio->sync && !dio->loop && dio->iter.count) { - struct iovec *iov = dio->inline_vecs; - - if (dio->iter.nr_segs > ARRAY_SIZE(dio->inline_vecs)) { - iov = kmalloc(dio->iter.nr_segs * sizeof(*iov), - GFP_KERNEL); - if (unlikely(!iov)) { - dio->sync = sync = true; - goto do_io; - } - - dio->free_iov = true; - } + if (unlikely(dio->iter.count) && + !dio->sync && + !dio->loop && + bch2_dio_write_copy_iov(dio)) + dio->sync = sync = true; - memcpy(iov, dio->iter.iov, dio->iter.nr_segs * sizeof(*iov)); - dio->iter.iov = iov; - } -do_io: dio->loop = true; closure_call(&dio->op.cl, bch2_write, NULL, NULL); - if (sync) - wait_for_completion(&dio->done); - else + if (!sync) return -EIOCBQUEUED; -loop: - i_sectors_acct(c, inode, &dio->quota_res, - dio->op.i_sectors_delta); - req->ki_pos += (u64) dio->op.written << 9; - dio->written += dio->op.written; - spin_lock(&inode->v.i_lock); - if (req->ki_pos > inode->v.i_size) - i_size_write(&inode->v, req->ki_pos); - spin_unlock(&inode->v.i_lock); + bch2_dio_write_end(dio); - bio_for_each_segment_all(bv, bio, i) - put_page(bv->bv_page); - bio->bi_vcnt = 0; - - if (dio->op.error) { - set_bit(EI_INODE_ERROR, &inode->ei_flags); + if (likely(!dio->iter.count) || dio->op.error) break; - } - if (!dio->iter.count) - break; + bio_reset(bio, NULL, REQ_OP_WRITE); + } +out: + return bch2_dio_write_done(dio); +err: + dio->op.error = ret; - bio_reset(bio); - reinit_completion(&dio->done); + if (!bio_flagged(bio, BIO_NO_PAGE_REF)) { + struct bvec_iter_all iter; + struct bio_vec *bv; + + bio_for_each_segment_all(bv, bio, iter) + put_page(bv->bv_page); } - ret = dio->op.error ?: ((long) dio->written << 9); -err: - up(&c->io_in_flight); - bch2_pagecache_block_put(&inode->ei_pagecache_lock); bch2_quota_reservation_put(c, inode, &dio->quota_res); + goto out; +} - if (dio->free_iov) - kfree(dio->iter.iov); - - bio_for_each_segment_all(bv, bio, i) - put_page(bv->bv_page); - bio_put(bio); +static noinline __cold void bch2_dio_write_continue(struct dio_write *dio) +{ + struct mm_struct *mm = dio->mm; - /* inode->i_dio_count is our ref on inode and thus bch_fs */ - inode_dio_end(&inode->v); + bio_reset(&dio->op.wbio.bio, NULL, REQ_OP_WRITE); - if (!sync) { - req->ki_complete(req, ret, 0); - ret = -EIOCBQUEUED; - } - return ret; + if (mm) + kthread_use_mm(mm); + bch2_dio_write_loop(dio); + if (mm) + kthread_unuse_mm(mm); } static void bch2_dio_write_loop_async(struct bch_write_op *op) { struct dio_write *dio = container_of(op, struct dio_write, op); - if (dio->sync) - complete(&dio->done); + bch2_dio_write_end(dio); + + if (likely(!dio->iter.count) || dio->op.error) + bch2_dio_write_done(dio); else - bch2_dio_write_loop(dio); + bch2_dio_write_continue(dio); } static noinline @@ -2373,7 +2464,7 @@ ssize_t bch2_direct_write(struct kiocb *req, struct iov_iter *iter) goto err; inode_dio_begin(&inode->v); - bch2_pagecache_block_get(&inode->ei_pagecache_lock); + bch2_pagecache_block_get(inode); extending = req->ki_pos + iter->count > inode->v.i_size; if (!extending) { @@ -2381,30 +2472,33 @@ ssize_t bch2_direct_write(struct kiocb *req, struct iov_iter *iter) locked = false; } - bio = bio_alloc_bioset(GFP_KERNEL, - iov_iter_npages(iter, BIO_MAX_PAGES), + bio = bio_alloc_bioset(NULL, + bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS), + REQ_OP_WRITE, + GFP_KERNEL, &c->dio_write_bioset); dio = container_of(bio, struct dio_write, op.wbio.bio); - init_completion(&dio->done); dio->req = req; + dio->mapping = mapping; + dio->inode = inode; dio->mm = current->mm; dio->loop = false; + dio->extending = extending; dio->sync = is_sync_kiocb(req) || extending; + dio->flush = iocb_is_dsync(req) && !c->opts.journal_flush_disabled; dio->free_iov = false; dio->quota_res.sectors = 0; dio->written = 0; dio->iter = *iter; + dio->op.c = c; - ret = bch2_quota_reservation_add(c, inode, &dio->quota_res, - iter->count >> 9, true); - if (unlikely(ret)) - goto err_put_bio; - - ret = write_invalidate_inode_pages_range(mapping, - req->ki_pos, - req->ki_pos + iter->count - 1); - if (unlikely(ret)) - goto err_put_bio; + if (unlikely(mapping->nrpages)) { + ret = write_invalidate_inode_pages_range(mapping, + req->ki_pos, + req->ki_pos + iter->count - 1); + if (unlikely(ret)) + goto err_put_bio; + } ret = bch2_dio_write_loop(dio); err: @@ -2412,8 +2506,7 @@ err: inode_unlock(&inode->v); return ret; err_put_bio: - bch2_pagecache_block_put(&inode->ei_pagecache_lock); - bch2_quota_reservation_put(c, inode, &dio->quota_res); + bch2_pagecache_block_put(inode); bio_put(bio); inode_dio_end(&inode->v); goto err; @@ -2425,8 +2518,10 @@ ssize_t bch2_write_iter(struct kiocb *iocb, struct iov_iter *from) struct bch_inode_info *inode = file_bch_inode(file); ssize_t ret; - if (iocb->ki_flags & IOCB_DIRECT) - return bch2_direct_write(iocb, from); + if (iocb->ki_flags & IOCB_DIRECT) { + ret = bch2_direct_write(iocb, from); + goto out; + } /* We can write back this queue in page reclaim */ current->backing_dev_info = inode_to_bdi(&inode->v); @@ -2453,8 +2548,8 @@ unlock: if (ret > 0) ret = generic_write_sync(iocb, ret); - - return ret; +out: + return bch2_err_class(ret); } /* fsync: */ @@ -2463,19 +2558,21 @@ unlock: * inode->ei_inode.bi_journal_seq won't be up to date since it's set in an * insert trigger: look up the btree inode instead */ -static int bch2_flush_inode(struct bch_fs *c, subvol_inum inum) +static int bch2_flush_inode(struct bch_fs *c, + struct bch_inode_info *inode) { - struct bch_inode_unpacked inode; + struct bch_inode_unpacked u; int ret; if (c->opts.journal_flush_disabled) return 0; - ret = bch2_inode_find_by_inum(c, inum, &inode); + ret = bch2_inode_find_by_inum(c, inode_inum(inode), &u); if (ret) return ret; - return bch2_journal_flush_seq(&c->journal, inode.bi_journal_seq); + return bch2_journal_flush_seq(&c->journal, u.bi_journal_seq) ?: + bch2_inode_flush_nocow_writes(c, inode); } int bch2_fsync(struct file *file, loff_t start, loff_t end, int datasync) @@ -2486,9 +2583,9 @@ int bch2_fsync(struct file *file, loff_t start, loff_t end, int datasync) ret = file_write_and_wait_range(file, start, end); ret2 = sync_inode_metadata(&inode->v, 1); - ret3 = bch2_flush_inode(c, inode_inum(inode)); + ret3 = bch2_flush_inode(c, inode); - return ret ?: ret2 ?: ret3; + return bch2_err_class(ret ?: ret2 ?: ret3); } /* truncate: */ @@ -2510,19 +2607,15 @@ retry: if (ret) goto err; - for_each_btree_key_norestart(&trans, iter, BTREE_ID_extents, start, 0, k, ret) { - if (bkey_cmp(bkey_start_pos(k.k), end) >= 0) - break; - + for_each_btree_key_upto_norestart(&trans, iter, BTREE_ID_extents, start, end, 0, k, ret) if (bkey_extent_is_data(k.k)) { ret = 1; break; } - } start = iter.pos; bch2_trans_iter_exit(&trans, &iter); err: - if (ret == -EINTR) + if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) goto retry; bch2_trans_exit(&trans); @@ -2558,8 +2651,8 @@ static int __bch2_truncate_page(struct bch_inode_info *inode, * page */ ret = range_has_data(c, inode->ei_subvol, - POS(inode->v.i_ino, index << PAGE_SECTORS_SHIFT), - POS(inode->v.i_ino, (index + 1) << PAGE_SECTORS_SHIFT)); + POS(inode->v.i_ino, (index << PAGE_SECTORS_SHIFT)), + POS(inode->v.i_ino, (index << PAGE_SECTORS_SHIFT) + PAGE_SECTORS)); if (ret <= 0) return ret; @@ -2622,7 +2715,7 @@ static int __bch2_truncate_page(struct bch_inode_info *inode, * redirty the full page: */ page_mkclean(page); - __set_page_dirty_nobuffers(page); + filemap_dirty_folio(mapping, page_folio(page)); unlock: unlock_page(page); put_page(page); @@ -2669,7 +2762,7 @@ static int bch2_extend(struct user_namespace *mnt_userns, truncate_setsize(&inode->v, iattr->ia_size); - return bch2_setattr_nonsize(inode, iattr); + return bch2_setattr_nonsize(mnt_userns, inode, iattr); } static int bch2_truncate_finish_fn(struct bch_inode_info *inode, @@ -2714,7 +2807,7 @@ int bch2_truncate(struct user_namespace *mnt_userns, } inode_dio_wait(&inode->v); - bch2_pagecache_block_get(&inode->ei_pagecache_lock); + bch2_pagecache_block_get(inode); ret = bch2_inode_find_by_inum(c, inode_inum(inode), &inode_u); if (ret) @@ -2729,8 +2822,10 @@ int bch2_truncate(struct user_namespace *mnt_userns, if (ret) goto err; - WARN_ON(!test_bit(EI_INODE_ERROR, &inode->ei_flags) && - inode->v.i_size < inode_u.bi_size); + WARN_ONCE(!test_bit(EI_INODE_ERROR, &inode->ei_flags) && + inode->v.i_size < inode_u.bi_size, + "truncate spotted in mem i_size < btree i_size: %llu < %llu\n", + (u64) inode->v.i_size, inode_u.bi_size); if (iattr->ia_size > inode->v.i_size) { ret = bch2_extend(mnt_userns, inode, &inode_u, iattr); @@ -2779,9 +2874,11 @@ int bch2_truncate(struct user_namespace *mnt_userns, U64_MAX, &i_sectors_delta); i_sectors_acct(c, inode, NULL, i_sectors_delta); - WARN_ON(!inode->v.i_size && inode->v.i_blocks && - !bch2_journal_error(&c->journal)); - + bch2_fs_inconsistent_on(!inode->v.i_size && inode->v.i_blocks && + !bch2_journal_error(&c->journal), c, + "inode %lu truncated to 0 but i_blocks %llu (ondisk %lli)", + inode->v.i_ino, (u64) inode->v.i_blocks, + inode->ei_inode.bi_sectors); if (unlikely(ret)) goto err; @@ -2789,10 +2886,10 @@ int bch2_truncate(struct user_namespace *mnt_userns, ret = bch2_write_inode(c, inode, bch2_truncate_finish_fn, NULL, 0); mutex_unlock(&inode->ei_update_lock); - ret = bch2_setattr_nonsize(inode, iattr); + ret = bch2_setattr_nonsize(mnt_userns, inode, iattr); err: - bch2_pagecache_block_put(&inode->ei_pagecache_lock); - return ret; + bch2_pagecache_block_put(inode); + return bch2_err_class(ret); } /* fallocate: */ @@ -2823,7 +2920,7 @@ static long bchfs_fpunch(struct bch_inode_info *inode, loff_t offset, loff_t len truncate_pagecache_range(&inode->v, offset, end - 1); - if (block_start < block_end ) { + if (block_start < block_end) { s64 i_sectors_delta = 0; ret = bch2_fpunch(c, inode_inum(inode), @@ -2910,7 +3007,8 @@ static long bchfs_fcollapse_finsert(struct bch_inode_info *inode, bch2_trans_copy_iter(&dst, &src); bch2_trans_copy_iter(&del, &src); - while (ret == 0 || ret == -EINTR) { + while (ret == 0 || + bch2_err_matches(ret, BCH_ERR_transaction_restart)) { struct disk_reservation disk_res = bch2_disk_reservation_init(c, 0); struct bkey_i delete; @@ -2936,7 +3034,7 @@ static long bchfs_fcollapse_finsert(struct bch_inode_info *inode, k = insert ? bch2_btree_iter_peek_prev(&src) - : bch2_btree_iter_peek(&src); + : bch2_btree_iter_peek_upto(&src, POS(inode->v.i_ino, U64_MAX)); if ((ret = bkey_err(k))) continue; @@ -2944,13 +3042,13 @@ static long bchfs_fcollapse_finsert(struct bch_inode_info *inode, break; if (insert && - bkey_cmp(k.k->p, POS(inode->v.i_ino, offset >> 9)) <= 0) + bkey_le(k.k->p, POS(inode->v.i_ino, offset >> 9))) break; reassemble: bch2_bkey_buf_reassemble(©, c, k); if (insert && - bkey_cmp(bkey_start_pos(k.k), move_pos) < 0) + bkey_lt(bkey_start_pos(k.k), move_pos)) bch2_cut_front(move_pos, copy.k); copy.k->k.p.offset += shift >> 9; @@ -2960,7 +3058,7 @@ reassemble: if (ret) continue; - if (bkey_cmp(atomic_end, copy.k->k.p)) { + if (!bkey_eq(atomic_end, copy.k->k.p)) { if (insert) { move_pos = atomic_end; move_pos.offset -= shift >> 9; @@ -2978,13 +3076,7 @@ reassemble: next_pos = insert ? bkey_start_pos(&delete.k) : delete.k.p; - if (copy.k->k.size == k.k->size) { - /* - * If we're moving the entire extent, we can skip - * running triggers: - */ - trigger_flags |= BTREE_TRIGGER_NORUN; - } else { + if (copy.k->k.size != k.k->size) { /* We might end up splitting compressed extents: */ unsigned nr_ptrs = bch2_bkey_nr_ptrs_allocated(bkey_i_to_s_c(copy.k)); @@ -3035,20 +3127,19 @@ static int __bchfs_fallocate(struct bch_inode_info *inode, int mode, struct btree_trans trans; struct btree_iter iter; struct bpos end_pos = POS(inode->v.i_ino, end_sector); - unsigned replicas = io_opts(c, &inode->ei_inode).data_replicas; + struct bch_io_opts opts; int ret = 0; + bch2_inode_opts_get(&opts, c, &inode->ei_inode); bch2_trans_init(&trans, c, BTREE_ITER_MAX, 512); bch2_trans_iter_init(&trans, &iter, BTREE_ID_extents, POS(inode->v.i_ino, start_sector), BTREE_ITER_SLOTS|BTREE_ITER_INTENT); - while (!ret && bkey_cmp(iter.pos, end_pos) < 0) { + while (!ret && bkey_lt(iter.pos, end_pos)) { s64 i_sectors_delta = 0; - struct disk_reservation disk_res = { 0 }; struct quota_res quota_res = { 0 }; - struct bkey_i_reservation reservation; struct bkey_s_c k; unsigned sectors; u32 snapshot; @@ -3067,8 +3158,8 @@ static int __bchfs_fallocate(struct bch_inode_info *inode, int mode, goto bkey_err; /* already reserved */ - if (k.k->type == KEY_TYPE_reservation && - bkey_s_c_to_reservation(k).v->nr_replicas >= replicas) { + if (bkey_extent_is_reservation(k) && + bch2_bkey_nr_ptrs_fully_allocated(k) >= opts.data_replicas) { bch2_btree_iter_advance(&iter); continue; } @@ -3079,16 +3170,12 @@ static int __bchfs_fallocate(struct bch_inode_info *inode, int mode, continue; } - bkey_reservation_init(&reservation.k_i); - reservation.k.type = KEY_TYPE_reservation; - reservation.k.p = k.k->p; - reservation.k.size = k.k->size; - - bch2_cut_front(iter.pos, &reservation.k_i); - bch2_cut_back(end_pos, &reservation.k_i); + /* + * XXX: for nocow mode, we should promote shared extents to + * unshared here + */ - sectors = reservation.k.size; - reservation.v.nr_replicas = bch2_bkey_nr_ptrs_allocated(k); + sectors = bpos_min(k.k->p, end_pos).offset - iter.pos.offset; if (!bkey_extent_is_allocation(k.k)) { ret = bch2_quota_reservation_add(c, inode, @@ -3098,34 +3185,23 @@ static int __bchfs_fallocate(struct bch_inode_info *inode, int mode, goto bkey_err; } - if (reservation.v.nr_replicas < replicas || - bch2_bkey_sectors_compressed(k)) { - ret = bch2_disk_reservation_get(c, &disk_res, sectors, - replicas, 0); - if (unlikely(ret)) - goto bkey_err; - - reservation.v.nr_replicas = disk_res.nr_replicas; - } - - ret = bch2_extent_update(&trans, inode_inum(inode), &iter, - &reservation.k_i, - &disk_res, NULL, - 0, &i_sectors_delta, true); + ret = bch2_extent_fallocate(&trans, inode_inum(inode), &iter, + sectors, opts, &i_sectors_delta, + writepoint_hashed((unsigned long) current)); if (ret) goto bkey_err; + i_sectors_acct(c, inode, "a_res, i_sectors_delta); bkey_err: bch2_quota_reservation_put(c, inode, "a_res); - bch2_disk_reservation_put(c, &disk_res); - if (ret == -EINTR) + if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) ret = 0; } bch2_trans_unlock(&trans); /* lock ordering, before taking pagecache locks: */ mark_pagecache_reserved(inode, start_sector, iter.pos.offset); - if (ret == -ENOSPC && (mode & FALLOC_FL_ZERO_RANGE)) { + if (bch2_err_matches(ret, ENOSPC) && (mode & FALLOC_FL_ZERO_RANGE)) { struct quota_res quota_res = { 0 }; s64 i_sectors_delta = 0; @@ -3176,7 +3252,7 @@ static long bchfs_fallocate(struct bch_inode_info *inode, int mode, * so that the VFS cache i_size is consistent with the btree i_size: */ if (ret && - !(ret == -ENOSPC && (mode & FALLOC_FL_ZERO_RANGE))) + !(bch2_err_matches(ret, ENOSPC) && (mode & FALLOC_FL_ZERO_RANGE))) return ret; if (mode & FALLOC_FL_KEEP_SIZE && end > inode->v.i_size) @@ -3204,12 +3280,16 @@ long bch2_fallocate_dispatch(struct file *file, int mode, struct bch_fs *c = inode->v.i_sb->s_fs_info; long ret; - if (!percpu_ref_tryget(&c->writes)) + if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_fallocate)) return -EROFS; inode_lock(&inode->v); inode_dio_wait(&inode->v); - bch2_pagecache_block_get(&inode->ei_pagecache_lock); + bch2_pagecache_block_get(inode); + + ret = file_modified(file); + if (ret) + goto err; if (!(mode & ~(FALLOC_FL_KEEP_SIZE|FALLOC_FL_ZERO_RANGE))) ret = bchfs_fallocate(inode, mode, offset, len); @@ -3221,242 +3301,65 @@ long bch2_fallocate_dispatch(struct file *file, int mode, ret = bchfs_fcollapse_finsert(inode, offset, len, false); else ret = -EOPNOTSUPP; - - - bch2_pagecache_block_put(&inode->ei_pagecache_lock); +err: + bch2_pagecache_block_put(inode); inode_unlock(&inode->v); - percpu_ref_put(&c->writes); + bch2_write_ref_put(c, BCH_WRITE_REF_fallocate); - return ret; + return bch2_err_class(ret); } -static int generic_access_check_limits(struct file *file, loff_t pos, - loff_t *count) -{ - struct inode *inode = file->f_mapping->host; - loff_t max_size = inode->i_sb->s_maxbytes; - - if (!(file->f_flags & O_LARGEFILE)) - max_size = MAX_NON_LFS; - - if (unlikely(pos >= max_size)) - return -EFBIG; - *count = min(*count, max_size - pos); - return 0; -} - -static int generic_write_check_limits(struct file *file, loff_t pos, - loff_t *count) -{ - loff_t limit = rlimit(RLIMIT_FSIZE); - - if (limit != RLIM_INFINITY) { - if (pos >= limit) { - send_sig(SIGXFSZ, current, 0); - return -EFBIG; - } - *count = min(*count, limit - pos); - } - - return generic_access_check_limits(file, pos, count); -} - -static int generic_remap_checks(struct file *file_in, loff_t pos_in, - struct file *file_out, loff_t pos_out, - loff_t *req_count, unsigned int remap_flags) +/* + * Take a quota reservation for unallocated blocks in a given file range + * Does not check pagecache + */ +static int quota_reserve_range(struct bch_inode_info *inode, + struct quota_res *res, + u64 start, u64 end) { - struct inode *inode_in = file_in->f_mapping->host; - struct inode *inode_out = file_out->f_mapping->host; - uint64_t count = *req_count; - uint64_t bcount; - loff_t size_in, size_out; - loff_t bs = inode_out->i_sb->s_blocksize; + struct bch_fs *c = inode->v.i_sb->s_fs_info; + struct btree_trans trans; + struct btree_iter iter; + struct bkey_s_c k; + u32 snapshot; + u64 sectors = end - start; + u64 pos = start; int ret; - /* The start of both ranges must be aligned to an fs block. */ - if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_out, bs)) - return -EINVAL; - - /* Ensure offsets don't wrap. */ - if (pos_in + count < pos_in || pos_out + count < pos_out) - return -EINVAL; - - size_in = i_size_read(inode_in); - size_out = i_size_read(inode_out); - - /* Dedupe requires both ranges to be within EOF. */ - if ((remap_flags & REMAP_FILE_DEDUP) && - (pos_in >= size_in || pos_in + count > size_in || - pos_out >= size_out || pos_out + count > size_out)) - return -EINVAL; - - /* Ensure the infile range is within the infile. */ - if (pos_in >= size_in) - return -EINVAL; - count = min(count, size_in - (uint64_t)pos_in); - - ret = generic_access_check_limits(file_in, pos_in, &count); - if (ret) - return ret; + bch2_trans_init(&trans, c, 0, 0); +retry: + bch2_trans_begin(&trans); - ret = generic_write_check_limits(file_out, pos_out, &count); + ret = bch2_subvolume_get_snapshot(&trans, inode->ei_subvol, &snapshot); if (ret) - return ret; - - /* - * If the user wanted us to link to the infile's EOF, round up to the - * next block boundary for this check. - * - * Otherwise, make sure the count is also block-aligned, having - * already confirmed the starting offsets' block alignment. - */ - if (pos_in + count == size_in) { - bcount = ALIGN(size_in, bs) - pos_in; - } else { - if (!IS_ALIGNED(count, bs)) - count = ALIGN_DOWN(count, bs); - bcount = count; - } - - /* Don't allow overlapped cloning within the same file. */ - if (inode_in == inode_out && - pos_out + bcount > pos_in && - pos_out < pos_in + bcount) - return -EINVAL; - - /* - * We shortened the request but the caller can't deal with that, so - * bounce the request back to userspace. - */ - if (*req_count != count && !(remap_flags & REMAP_FILE_CAN_SHORTEN)) - return -EINVAL; - - *req_count = count; - return 0; -} - -static int generic_remap_check_len(struct inode *inode_in, - struct inode *inode_out, - loff_t pos_out, - loff_t *len, - unsigned int remap_flags) -{ - u64 blkmask = i_blocksize(inode_in) - 1; - loff_t new_len = *len; - - if ((*len & blkmask) == 0) - return 0; - - if ((remap_flags & REMAP_FILE_DEDUP) || - pos_out + *len < i_size_read(inode_out)) - new_len &= ~blkmask; - - if (new_len == *len) - return 0; - - if (remap_flags & REMAP_FILE_CAN_SHORTEN) { - *len = new_len; - return 0; - } - - return (remap_flags & REMAP_FILE_DEDUP) ? -EBADE : -EINVAL; -} - -static int generic_remap_file_range_prep(struct file *file_in, loff_t pos_in, - struct file *file_out, loff_t pos_out, - loff_t *len, unsigned int remap_flags) -{ - struct inode *inode_in = file_inode(file_in); - struct inode *inode_out = file_inode(file_out); - bool same_inode = (inode_in == inode_out); - int ret; - - /* Don't touch certain kinds of inodes */ - if (IS_IMMUTABLE(inode_out)) - return -EPERM; - - if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out)) - return -ETXTBSY; - - /* Don't reflink dirs, pipes, sockets... */ - if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode)) - return -EISDIR; - if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode)) - return -EINVAL; - - /* Zero length dedupe exits immediately; reflink goes to EOF. */ - if (*len == 0) { - loff_t isize = i_size_read(inode_in); + goto err; - if ((remap_flags & REMAP_FILE_DEDUP) || pos_in == isize) - return 0; - if (pos_in > isize) - return -EINVAL; - *len = isize - pos_in; - if (*len == 0) - return 0; + bch2_trans_iter_init(&trans, &iter, BTREE_ID_extents, + SPOS(inode->v.i_ino, pos, snapshot), 0); + + while (!(ret = btree_trans_too_many_iters(&trans)) && + (k = bch2_btree_iter_peek_upto(&iter, POS(inode->v.i_ino, end - 1))).k && + !(ret = bkey_err(k))) { + if (bkey_extent_is_allocation(k.k)) { + u64 s = min(end, k.k->p.offset) - + max(start, bkey_start_offset(k.k)); + BUG_ON(s > sectors); + sectors -= s; + } + bch2_btree_iter_advance(&iter); } + pos = iter.pos.offset; + bch2_trans_iter_exit(&trans, &iter); +err: + if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) + goto retry; - /* Check that we don't violate system file offset limits. */ - ret = generic_remap_checks(file_in, pos_in, file_out, pos_out, len, - remap_flags); - if (ret) - return ret; - - /* Wait for the completion of any pending IOs on both files */ - inode_dio_wait(inode_in); - if (!same_inode) - inode_dio_wait(inode_out); - - ret = filemap_write_and_wait_range(inode_in->i_mapping, - pos_in, pos_in + *len - 1); - if (ret) - return ret; - - ret = filemap_write_and_wait_range(inode_out->i_mapping, - pos_out, pos_out + *len - 1); - if (ret) - return ret; - - /* - * Check that the extents are the same. - */ - if (remap_flags & REMAP_FILE_DEDUP) { - bool is_same = false; - - ret = vfs_dedupe_file_range_compare(inode_in, pos_in, - inode_out, pos_out, *len, &is_same); - if (ret) - return ret; - if (!is_same) - return -EBADE; - } + bch2_trans_exit(&trans); - ret = generic_remap_check_len(inode_in, inode_out, pos_out, len, - remap_flags); if (ret) return ret; - /* If can't alter the file contents, we're done. */ - if (!(remap_flags & REMAP_FILE_DEDUP)) { - /* Update the timestamps, since we can alter file contents. */ - if (!(file_out->f_mode & FMODE_NOCMTIME)) { - ret = file_update_time(file_out); - if (ret) - return ret; - } - - /* - * Clear the security bits if the process is not being run by - * root. This keeps people from modifying setuid and setgid - * binaries. - */ - ret = file_remove_privs(file_out); - if (ret) - return ret; - } - - return 0; + return bch2_quota_reservation_add(c, inode, res, sectors, true); } loff_t bch2_remap_file_range(struct file *file_src, loff_t pos_src, @@ -3466,6 +3369,7 @@ loff_t bch2_remap_file_range(struct file *file_src, loff_t pos_src, struct bch_inode_info *src = file_bch_inode(file_src); struct bch_inode_info *dst = file_bch_inode(file_dst); struct bch_fs *c = src->v.i_sb->s_fs_info; + struct quota_res quota_res = { 0 }; s64 i_sectors_delta = 0; u64 aligned_len; loff_t ret = 0; @@ -3486,8 +3390,6 @@ loff_t bch2_remap_file_range(struct file *file_src, loff_t pos_src, bch2_lock_inodes(INODE_LOCK|INODE_PAGECACHE_BLOCK, src, dst); - file_update_time(file_dst); - inode_dio_wait(&src->v); inode_dio_wait(&dst->v); @@ -3504,6 +3406,13 @@ loff_t bch2_remap_file_range(struct file *file_src, loff_t pos_src, if (ret) goto err; + ret = quota_reserve_range(dst, "a_res, pos_dst >> 9, + (pos_dst + aligned_len) >> 9); + if (ret) + goto err; + + file_update_time(file_dst); + mark_pagecache_unallocated(src, pos_src >> 9, (pos_src + aligned_len) >> 9); @@ -3520,8 +3429,7 @@ loff_t bch2_remap_file_range(struct file *file_src, loff_t pos_src, */ ret = min((u64) ret << 9, (u64) len); - /* XXX get a quota reservation */ - i_sectors_acct(c, dst, NULL, i_sectors_delta); + i_sectors_acct(c, dst, "a_res, i_sectors_delta); spin_lock(&dst->v.i_lock); if (pos_dst + ret > dst->v.i_size) @@ -3530,18 +3438,19 @@ loff_t bch2_remap_file_range(struct file *file_src, loff_t pos_src, if ((file_dst->f_flags & (__O_SYNC | O_DSYNC)) || IS_SYNC(file_inode(file_dst))) - ret = bch2_flush_inode(c, inode_inum(dst)); + ret = bch2_flush_inode(c, dst); err: + bch2_quota_reservation_put(c, dst, "a_res); bch2_unlock_inodes(INODE_LOCK|INODE_PAGECACHE_BLOCK, src, dst); - return ret; + return bch2_err_class(ret); } /* fseek: */ -static int page_data_offset(struct page *page, unsigned offset) +static int folio_data_offset(struct folio *folio, unsigned offset) { - struct bch_page_state *s = bch2_page_state(page); + struct bch_page_state *s = bch2_page_state(&folio->page); unsigned i; if (s) @@ -3556,36 +3465,38 @@ static loff_t bch2_seek_pagecache_data(struct inode *vinode, loff_t start_offset, loff_t end_offset) { - struct address_space *mapping = vinode->i_mapping; - struct page *page; + struct folio_batch fbatch; pgoff_t start_index = start_offset >> PAGE_SHIFT; pgoff_t end_index = end_offset >> PAGE_SHIFT; pgoff_t index = start_index; + unsigned i; loff_t ret; int offset; - while (index <= end_index) { - if (find_get_pages_range(mapping, &index, end_index, 1, &page)) { - lock_page(page); + folio_batch_init(&fbatch); - offset = page_data_offset(page, - page->index == start_index + while (filemap_get_folios(vinode->i_mapping, + &index, end_index, &fbatch)) { + for (i = 0; i < folio_batch_count(&fbatch); i++) { + struct folio *folio = fbatch.folios[i]; + + folio_lock(folio); + offset = folio_data_offset(folio, + folio->index == start_index ? start_offset & (PAGE_SIZE - 1) : 0); if (offset >= 0) { - ret = clamp(((loff_t) page->index << PAGE_SHIFT) + + ret = clamp(((loff_t) folio->index << PAGE_SHIFT) + offset, start_offset, end_offset); - unlock_page(page); - put_page(page); + folio_unlock(folio); + folio_batch_release(&fbatch); return ret; } - - unlock_page(page); - put_page(page); - } else { - break; + folio_unlock(folio); } + folio_batch_release(&fbatch); + cond_resched(); } return end_offset; @@ -3615,11 +3526,11 @@ retry: if (ret) goto err; - for_each_btree_key_norestart(&trans, iter, BTREE_ID_extents, - SPOS(inode->v.i_ino, offset >> 9, snapshot), 0, k, ret) { - if (k.k->p.inode != inode->v.i_ino) { - break; - } else if (bkey_extent_is_data(k.k)) { + for_each_btree_key_upto_norestart(&trans, iter, BTREE_ID_extents, + SPOS(inode->v.i_ino, offset >> 9, snapshot), + POS(inode->v.i_ino, U64_MAX), + 0, k, ret) { + if (bkey_extent_is_data(k.k)) { next_data = max(offset, bkey_start_offset(k.k) << 9); break; } else if (k.k->p.offset >> 9 > isize) @@ -3627,7 +3538,7 @@ retry: } bch2_trans_iter_exit(&trans, &iter); err: - if (ret == -EINTR) + if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) goto retry; bch2_trans_exit(&trans); @@ -3742,7 +3653,7 @@ retry: } bch2_trans_iter_exit(&trans, &iter); err: - if (ret == -EINTR) + if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) goto retry; bch2_trans_exit(&trans); @@ -3757,22 +3668,31 @@ err: loff_t bch2_llseek(struct file *file, loff_t offset, int whence) { + loff_t ret; + switch (whence) { case SEEK_SET: case SEEK_CUR: case SEEK_END: - return generic_file_llseek(file, offset, whence); + ret = generic_file_llseek(file, offset, whence); + break; case SEEK_DATA: - return bch2_seek_data(file, offset); + ret = bch2_seek_data(file, offset); + break; case SEEK_HOLE: - return bch2_seek_hole(file, offset); + ret = bch2_seek_hole(file, offset); + break; + default: + ret = -EINVAL; + break; } - return -EINVAL; + return bch2_err_class(ret); } void bch2_fs_fsio_exit(struct bch_fs *c) { + bioset_exit(&c->nocow_flush_bioset); bioset_exit(&c->dio_write_bioset); bioset_exit(&c->dio_read_bioset); bioset_exit(&c->writepage_bioset); @@ -3786,14 +3706,22 @@ int bch2_fs_fsio_init(struct bch_fs *c) if (bioset_init(&c->writepage_bioset, 4, offsetof(struct bch_writepage_io, op.wbio.bio), - BIOSET_NEED_BVECS) || - bioset_init(&c->dio_read_bioset, + BIOSET_NEED_BVECS)) + return -BCH_ERR_ENOMEM_writepage_bioset_init; + + if (bioset_init(&c->dio_read_bioset, 4, offsetof(struct dio_read, rbio.bio), - BIOSET_NEED_BVECS) || - bioset_init(&c->dio_write_bioset, + BIOSET_NEED_BVECS)) + return -BCH_ERR_ENOMEM_dio_read_bioset_init; + + if (bioset_init(&c->dio_write_bioset, 4, offsetof(struct dio_write, op.wbio.bio), BIOSET_NEED_BVECS)) - ret = -ENOMEM; + return -BCH_ERR_ENOMEM_dio_write_bioset_init; + + if (bioset_init(&c->nocow_flush_bioset, + 1, offsetof(struct nocow_flush, bio), 0)) + return -BCH_ERR_ENOMEM_nocow_flush_bioset_init; pr_verbose_init(c->opts, "ret %i", ret); return ret; |