X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libbcachefs%2Ffs-io.c;h=98bd5babab193bec842dce20b0783e6c958ac5bf;hb=b90031efaa4500b8ce5ffc55e9d1ed716196b814;hp=0fbfa622586ecf14f36cfe559032ad28b4e1b892;hpb=55142cd0b5ef2a2150d4708dad0c3fd54a3ffd39;p=bcachefs-tools-debian diff --git a/libbcachefs/fs-io.c b/libbcachefs/fs-io.c index 0fbfa62..98bd5ba 100644 --- a/libbcachefs/fs-io.c +++ b/libbcachefs/fs-io.c @@ -12,13 +12,16 @@ #include "extent_update.h" #include "fs.h" #include "fs-io.h" +#include "fs-io-buffered.h" +#include "fs-io-pagecache.h" #include "fsck.h" #include "inode.h" #include "journal.h" -#include "io.h" +#include "io_misc.h" #include "keylist.h" #include "quota.h" #include "reflink.h" +#include "trace.h" #include #include @@ -30,155 +33,75 @@ #include #include #include -#include -#include #include -static inline struct address_space *faults_disabled_mapping(void) -{ - return (void *) (((unsigned long) current->faults_disabled_mapping) & ~1UL); -} - -static inline void set_fdm_dropped_locks(void) -{ - current->faults_disabled_mapping = - (void *) (((unsigned long) current->faults_disabled_mapping)|1); -} - -static inline bool fdm_dropped_locks(void) -{ - return ((unsigned long) current->faults_disabled_mapping) & 1; -} - -struct quota_res { - u64 sectors; -}; - -struct bch_writepage_io { - struct closure cl; - struct bch_inode_info *inode; - - /* must be last: */ - struct bch_write_op op; -}; - -struct dio_write { - struct completion done; - struct kiocb *req; - struct mm_struct *mm; - unsigned loop:1, - sync:1, - free_iov:1; - struct quota_res quota_res; - u64 written; - - struct iov_iter iter; - struct iovec inline_vecs[2]; - - /* must be last: */ - struct bch_write_op op; -}; - -struct dio_read { - struct closure cl; - struct kiocb *req; - long ret; - bool should_dirty; - struct bch_read_bio rbio; +struct nocow_flush { + struct closure *cl; + struct bch_dev *ca; + struct bio bio; }; -/* pagecache_block must be held */ -static int write_invalidate_inode_pages_range(struct address_space *mapping, - loff_t start, loff_t end) +static void nocow_flush_endio(struct bio *_bio) { - int ret; - - /* - * XXX: the way this is currently implemented, we can spin if a process - * is continually redirtying a specific page - */ - do { - if (!mapping->nrpages && - !mapping->nrexceptional) - return 0; - - ret = filemap_write_and_wait_range(mapping, start, end); - if (ret) - break; - - if (!mapping->nrpages) - return 0; - ret = invalidate_inode_pages2_range(mapping, - start >> PAGE_SHIFT, - end >> PAGE_SHIFT); - } while (ret == -EBUSY); + struct nocow_flush *bio = container_of(_bio, struct nocow_flush, bio); - return ret; + closure_put(bio->cl); + percpu_ref_put(&bio->ca->io_ref); + bio_put(&bio->bio); } -/* quotas */ - -#ifdef CONFIG_BCACHEFS_QUOTA - -static void bch2_quota_reservation_put(struct bch_fs *c, - struct bch_inode_info *inode, - struct quota_res *res) +void bch2_inode_flush_nocow_writes_async(struct bch_fs *c, + struct bch_inode_info *inode, + struct closure *cl) { - if (!res->sectors) - return; + struct nocow_flush *bio; + struct bch_dev *ca; + struct bch_devs_mask devs; + unsigned dev; - mutex_lock(&inode->ei_quota_lock); - BUG_ON(res->sectors > inode->ei_quota_reserved); + dev = find_first_bit(inode->ei_devs_need_flush.d, BCH_SB_MEMBERS_MAX); + if (dev == BCH_SB_MEMBERS_MAX) + return; - 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); + devs = inode->ei_devs_need_flush; + memset(&inode->ei_devs_need_flush, 0, sizeof(inode->ei_devs_need_flush)); - res->sectors = 0; -} + 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(); -static int bch2_quota_reservation_add(struct bch_fs *c, - struct bch_inode_info *inode, - struct quota_res *res, - unsigned sectors, - bool check_enospc) -{ - int ret; + if (!ca) + continue; - mutex_lock(&inode->ei_quota_lock); - ret = bch2_quota_acct(c, inode->ei_qid, Q_SPC, sectors, - check_enospc ? KEY_TYPE_QUOTA_PREALLOC : KEY_TYPE_QUOTA_NOCHECK); - if (likely(!ret)) { - inode->ei_quota_reserved += sectors; - res->sectors += sectors; + 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); } - mutex_unlock(&inode->ei_quota_lock); - - return ret; } -#else - -static void bch2_quota_reservation_put(struct bch_fs *c, - struct bch_inode_info *inode, - struct quota_res *res) +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); -static int bch2_quota_reservation_add(struct bch_fs *c, - struct bch_inode_info *inode, - struct quota_res *res, - unsigned sectors, - bool check_enospc) -{ return 0; } -#endif - /* i_size updates: */ struct inode_new_size { @@ -187,7 +110,8 @@ struct inode_new_size { unsigned fields; }; -static int inode_set_size(struct bch_inode_info *inode, +static int inode_set_size(struct btree_trans *trans, + struct bch_inode_info *inode, struct bch_inode_unpacked *bi, void *p) { @@ -217,15 +141,19 @@ 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, +void __bch2_i_sectors_acct(struct bch_fs *c, struct bch_inode_info *inode, struct quota_res *quota_res, s64 sectors) { - if (!sectors) - return; + 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; - mutex_lock(&inode->ei_quota_lock); #ifdef CONFIG_BCACHEFS_QUOTA - if (quota_res && sectors > 0) { + if (quota_res && + !test_bit(EI_INODE_SNAPSHOT, &inode->ei_flags) && + sectors > 0) { BUG_ON(sectors > quota_res->sectors); BUG_ON(sectors > inode->ei_quota_reserved); @@ -235,2026 +163,215 @@ 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 - inode->v.i_blocks += sectors; - mutex_unlock(&inode->ei_quota_lock); -} - -/* page state: */ - -/* stored in page->private: */ - -struct bch_page_sector { - /* Uncompressed, fully allocated replicas: */ - unsigned nr_replicas:3; - - /* Owns PAGE_SECTORS * replicas_reserved sized reservation: */ - unsigned replicas_reserved:3; - - /* i_sectors: */ - enum { - SECTOR_UNALLOCATED, - SECTOR_RESERVED, - SECTOR_DIRTY, - SECTOR_ALLOCATED, - } state:2; -}; - -struct bch_page_state { - spinlock_t lock; - atomic_t write_count; - struct bch_page_sector s[PAGE_SECTORS]; -}; - -static inline struct bch_page_state *__bch2_page_state(struct page *page) -{ - return page_has_private(page) - ? (struct bch_page_state *) page_private(page) - : NULL; -} - -static inline struct bch_page_state *bch2_page_state(struct page *page) -{ - EBUG_ON(!PageLocked(page)); - - return __bch2_page_state(page); -} - -/* for newly allocated pages: */ -static void __bch2_page_state_release(struct page *page) -{ - kfree(detach_page_private(page)); -} - -static void bch2_page_state_release(struct page *page) -{ - EBUG_ON(!PageLocked(page)); - __bch2_page_state_release(page); -} - -/* for newly allocated pages: */ -static struct bch_page_state *__bch2_page_state_create(struct page *page, - gfp_t gfp) -{ - struct bch_page_state *s; - - s = kzalloc(sizeof(*s), GFP_NOFS|gfp); - if (!s) - return NULL; - - spin_lock_init(&s->lock); - attach_page_private(page, s); - return s; -} - -static struct bch_page_state *bch2_page_state_create(struct page *page, - gfp_t gfp) -{ - return bch2_page_state(page) ?: __bch2_page_state_create(page, gfp); } -static inline unsigned inode_nr_replicas(struct bch_fs *c, struct bch_inode_info *inode) -{ - /* XXX: this should not be open coded */ - return inode->ei_inode.bi_data_replicas - ? inode->ei_inode.bi_data_replicas - 1 - : c->opts.data_replicas; -} - -static inline unsigned sectors_to_reserve(struct bch_page_sector *s, - unsigned nr_replicas) -{ - return max(0, (int) nr_replicas - - s->nr_replicas - - s->replicas_reserved); -} +/* fsync: */ -static int bch2_get_page_disk_reservation(struct bch_fs *c, - struct bch_inode_info *inode, - struct page *page, bool check_enospc) +/* + * 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, + struct bch_inode_info *inode) { - struct bch_page_state *s = bch2_page_state_create(page, 0); - unsigned nr_replicas = inode_nr_replicas(c, inode); - struct disk_reservation disk_res = { 0 }; - unsigned i, disk_res_sectors = 0; + struct bch_inode_unpacked u; int ret; - if (!s) - return -ENOMEM; - - for (i = 0; i < ARRAY_SIZE(s->s); i++) - disk_res_sectors += sectors_to_reserve(&s->s[i], nr_replicas); - - if (!disk_res_sectors) + if (c->opts.journal_flush_disabled) return 0; - ret = bch2_disk_reservation_get(c, &disk_res, - disk_res_sectors, 1, - !check_enospc - ? BCH_DISK_RESERVATION_NOFAIL - : 0); - if (unlikely(ret)) + ret = bch2_inode_find_by_inum(c, inode_inum(inode), &u); + if (ret) return ret; - for (i = 0; i < ARRAY_SIZE(s->s); i++) - s->s[i].replicas_reserved += - sectors_to_reserve(&s->s[i], nr_replicas); - - return 0; -} - -struct bch2_page_reservation { - struct disk_reservation disk; - struct quota_res quota; -}; - -static void bch2_page_reservation_init(struct bch_fs *c, - struct bch_inode_info *inode, - struct bch2_page_reservation *res) -{ - memset(res, 0, sizeof(*res)); - - res->disk.nr_replicas = inode_nr_replicas(c, inode); -} - -static void bch2_page_reservation_put(struct bch_fs *c, - struct bch_inode_info *inode, - struct bch2_page_reservation *res) -{ - bch2_disk_reservation_put(c, &res->disk); - bch2_quota_reservation_put(c, inode, &res->quota); -} - -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) -{ - struct bch_page_state *s = bch2_page_state_create(page, 0); - unsigned i, disk_sectors = 0, quota_sectors = 0; - int ret; - - if (!s) - return -ENOMEM; - - for (i = round_down(offset, block_bytes(c)) >> 9; - i < round_up(offset + len, block_bytes(c)) >> 9; - i++) { - disk_sectors += sectors_to_reserve(&s->s[i], - res->disk.nr_replicas); - quota_sectors += s->s[i].state == SECTOR_UNALLOCATED; - } - - if (disk_sectors) { - ret = bch2_disk_reservation_add(c, &res->disk, - disk_sectors, - !check_enospc - ? BCH_DISK_RESERVATION_NOFAIL - : 0); - if (unlikely(ret)) - return ret; - } - - if (quota_sectors) { - ret = bch2_quota_reservation_add(c, inode, &res->quota, - quota_sectors, - check_enospc); - if (unlikely(ret)) { - struct disk_reservation tmp = { - .sectors = disk_sectors - }; - - bch2_disk_reservation_put(c, &tmp); - res->disk.sectors -= disk_sectors; - return ret; - } - } - - return 0; -} - -static void bch2_clear_page_bits(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_page_state *s = bch2_page_state(page); - struct disk_reservation disk_res = { 0 }; - int i, dirty_sectors = 0; - - if (!s) - return; - - EBUG_ON(!PageLocked(page)); - EBUG_ON(PageWriteback(page)); - - for (i = 0; i < ARRAY_SIZE(s->s); i++) { - disk_res.sectors += s->s[i].replicas_reserved; - s->s[i].replicas_reserved = 0; - - if (s->s[i].state == SECTOR_DIRTY) { - dirty_sectors++; - s->s[i].state = SECTOR_UNALLOCATED; - } - } - - bch2_disk_reservation_put(c, &disk_res); - - if (dirty_sectors) - i_sectors_acct(c, inode, NULL, -dirty_sectors); - - bch2_page_state_release(page); -} - -static void bch2_set_page_dirty(struct bch_fs *c, - struct bch_inode_info *inode, struct page *page, - struct bch2_page_reservation *res, - unsigned offset, unsigned len) -{ - struct bch_page_state *s = bch2_page_state(page); - unsigned i, dirty_sectors = 0; - - WARN_ON((u64) page_offset(page) + offset + len > - round_up((u64) i_size_read(&inode->v), block_bytes(c))); - - spin_lock(&s->lock); - - for (i = round_down(offset, block_bytes(c)) >> 9; - i < round_up(offset + len, block_bytes(c)) >> 9; - i++) { - unsigned sectors = sectors_to_reserve(&s->s[i], - res->disk.nr_replicas); - - /* - * This can happen if we race with the error path in - * bch2_writepage_io_done(): - */ - sectors = min_t(unsigned, sectors, res->disk.sectors); - - s->s[i].replicas_reserved += sectors; - res->disk.sectors -= sectors; - - if (s->s[i].state == SECTOR_UNALLOCATED) - dirty_sectors++; - - s->s[i].state = max_t(unsigned, s->s[i].state, SECTOR_DIRTY); - } - - spin_unlock(&s->lock); - - if (dirty_sectors) - i_sectors_acct(c, inode, &res->quota, dirty_sectors); - - if (!PageDirty(page)) - __set_page_dirty_nobuffers(page); -} - -vm_fault_t bch2_page_fault(struct vm_fault *vmf) -{ - struct file *file = vmf->vma->vm_file; - struct address_space *mapping = file->f_mapping; - struct address_space *fdm = faults_disabled_mapping(); - struct bch_inode_info *inode = file_bch_inode(file); - int ret; - - if (fdm == mapping) - return VM_FAULT_SIGBUS; - - /* Lock ordering: */ - if (fdm > mapping) { - struct bch_inode_info *fdm_host = to_bch_ei(fdm->host); - - if (bch2_pagecache_add_tryget(&inode->ei_pagecache_lock)) - goto got_lock; - - bch2_pagecache_block_put(&fdm_host->ei_pagecache_lock); - - bch2_pagecache_add_get(&inode->ei_pagecache_lock); - bch2_pagecache_add_put(&inode->ei_pagecache_lock); - - bch2_pagecache_block_get(&fdm_host->ei_pagecache_lock); - - /* Signal that lock has been dropped: */ - set_fdm_dropped_locks(); - return VM_FAULT_SIGBUS; - } - - bch2_pagecache_add_get(&inode->ei_pagecache_lock); -got_lock: - ret = filemap_fault(vmf); - bch2_pagecache_add_put(&inode->ei_pagecache_lock); - - return ret; + return bch2_journal_flush_seq(&c->journal, u.bi_journal_seq) ?: + bch2_inode_flush_nocow_writes(c, inode); } -vm_fault_t bch2_page_mkwrite(struct vm_fault *vmf) +int bch2_fsync(struct file *file, loff_t start, loff_t end, int datasync) { - struct page *page = vmf->page; - struct file *file = vmf->vma->vm_file; struct bch_inode_info *inode = file_bch_inode(file); - struct address_space *mapping = file->f_mapping; struct bch_fs *c = inode->v.i_sb->s_fs_info; - struct bch2_page_reservation res; - unsigned len; - loff_t isize; - int ret = VM_FAULT_LOCKED; - - bch2_page_reservation_init(c, inode, &res); - - sb_start_pagefault(inode->v.i_sb); - file_update_time(file); - - /* - * Not strictly necessary, but helps avoid dio writes livelocking in - * write_invalidate_inode_pages_range() - can drop this if/when we get - * a write_invalidate_inode_pages_range() that works without dropping - * page lock before invalidating page - */ - bch2_pagecache_add_get(&inode->ei_pagecache_lock); - - lock_page(page); - isize = i_size_read(&inode->v); + int ret; - if (page->mapping != mapping || page_offset(page) >= isize) { - unlock_page(page); - ret = VM_FAULT_NOPAGE; + ret = file_write_and_wait_range(file, start, end); + if (ret) goto out; - } - - len = min_t(loff_t, PAGE_SIZE, isize - page_offset(page)); - - if (bch2_page_reservation_get(c, inode, page, &res, 0, len, true)) { - unlock_page(page); - ret = VM_FAULT_SIGBUS; + ret = sync_inode_metadata(&inode->v, 1); + if (ret) goto out; - } - - bch2_set_page_dirty(c, inode, page, &res, 0, len); - bch2_page_reservation_put(c, inode, &res); - - wait_for_stable_page(page); + ret = bch2_flush_inode(c, inode); out: - bch2_pagecache_add_put(&inode->ei_pagecache_lock); - sb_end_pagefault(inode->v.i_sb); - - return ret; -} - -void bch2_invalidatepage(struct page *page, unsigned int offset, - unsigned int length) -{ - if (offset || length < PAGE_SIZE) - return; - - bch2_clear_page_bits(page); + return bch2_err_class(ret); } -int bch2_releasepage(struct page *page, gfp_t gfp_mask) -{ - if (PageDirty(page)) - return 0; - - bch2_clear_page_bits(page); - return 1; -} +/* truncate: */ -#ifdef CONFIG_MIGRATION -int bch2_migrate_page(struct address_space *mapping, struct page *newpage, - struct page *page, enum migrate_mode mode) +static inline int range_has_data(struct bch_fs *c, u32 subvol, + struct bpos start, + struct bpos end) { - int ret; - - EBUG_ON(!PageLocked(page)); - EBUG_ON(!PageLocked(newpage)); + struct btree_trans *trans = bch2_trans_get(c); + struct btree_iter iter; + struct bkey_s_c k; + int ret = 0; +retry: + bch2_trans_begin(trans); - ret = migrate_page_move_mapping(mapping, newpage, page, 0); - if (ret != MIGRATEPAGE_SUCCESS) - return ret; + ret = bch2_subvolume_get_snapshot(trans, subvol, &start.snapshot); + if (ret) + goto err; - if (PagePrivate(page)) - attach_page_private(newpage, detach_page_private(page)); + for_each_btree_key_upto_norestart(trans, iter, BTREE_ID_extents, start, end, 0, k, ret) + if (bkey_extent_is_data(k.k) && !bkey_extent_is_unwritten(k)) { + ret = 1; + break; + } + start = iter.pos; + bch2_trans_iter_exit(trans, &iter); +err: + if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) + goto retry; - if (mode != MIGRATE_SYNC_NO_COPY) - migrate_page_copy(newpage, page); - else - migrate_page_states(newpage, page); - return MIGRATEPAGE_SUCCESS; + bch2_trans_put(trans); + return ret; } -#endif - -/* readpage(s): */ -static void bch2_readpages_end_io(struct bio *bio) +static int __bch2_truncate_folio(struct bch_inode_info *inode, + pgoff_t index, loff_t start, loff_t end) { - struct bvec_iter_all iter; - struct bio_vec *bv; + struct bch_fs *c = inode->v.i_sb->s_fs_info; + struct address_space *mapping = inode->v.i_mapping; + struct bch_folio *s; + unsigned start_offset; + unsigned end_offset; + unsigned i; + struct folio *folio; + s64 i_sectors_delta = 0; + int ret = 0; + u64 end_pos; - bio_for_each_segment_all(bv, bio, iter) { - struct page *page = bv->bv_page; + folio = filemap_lock_folio(mapping, index); + if (IS_ERR_OR_NULL(folio)) { + /* + * XXX: we're doing two index lookups when we end up reading the + * folio + */ + ret = range_has_data(c, inode->ei_subvol, + 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; - if (!bio->bi_status) { - SetPageUptodate(page); - } else { - ClearPageUptodate(page); - SetPageError(page); + folio = __filemap_get_folio(mapping, index, + FGP_LOCK|FGP_CREAT, GFP_KERNEL); + if (IS_ERR_OR_NULL(folio)) { + ret = -ENOMEM; + goto out; } - unlock_page(page); } - bio_put(bio); -} - -struct readpages_iter { - struct address_space *mapping; - struct page **pages; - unsigned nr_pages; - unsigned idx; - pgoff_t offset; -}; - -static int readpages_iter_init(struct readpages_iter *iter, - struct readahead_control *ractl) -{ - unsigned i, nr_pages = readahead_count(ractl); - - memset(iter, 0, sizeof(*iter)); - - iter->mapping = ractl->mapping; - iter->offset = readahead_index(ractl); - iter->nr_pages = nr_pages; + BUG_ON(start >= folio_end_pos(folio)); + BUG_ON(end <= folio_pos(folio)); - iter->pages = kmalloc_array(nr_pages, sizeof(struct page *), GFP_NOFS); - if (!iter->pages) - return -ENOMEM; + start_offset = max(start, folio_pos(folio)) - folio_pos(folio); + end_offset = min_t(u64, end, folio_end_pos(folio)) - folio_pos(folio); - 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]); + /* Folio boundary? Nothing to do */ + if (start_offset == 0 && + end_offset == folio_size(folio)) { + ret = 0; + goto unlock; } - return 0; -} - -static inline struct page *readpage_iter_next(struct readpages_iter *iter) -{ - if (iter->idx >= iter->nr_pages) - return NULL; - - EBUG_ON(iter->pages[iter->idx]->index != iter->offset + iter->idx); - - return iter->pages[iter->idx]; -} - -static void bch2_add_page_sectors(struct bio *bio, struct bkey_s_c k) -{ - struct bvec_iter iter; - struct bio_vec bv; - unsigned nr_ptrs = k.k->type == KEY_TYPE_reflink_v - ? 0 : bch2_bkey_nr_ptrs_fully_allocated(k); - unsigned state = k.k->type == KEY_TYPE_reservation - ? SECTOR_RESERVED - : SECTOR_ALLOCATED; - - bio_for_each_segment(bv, bio, iter) { - struct bch_page_state *s = bch2_page_state(bv.bv_page); - unsigned i; - - for (i = bv.bv_offset >> 9; - i < (bv.bv_offset + bv.bv_len) >> 9; - i++) { - s->s[i].nr_replicas = nr_ptrs; - s->s[i].state = state; - } + s = bch2_folio_create(folio, 0); + if (!s) { + ret = -ENOMEM; + goto unlock; } -} - -static bool extent_partial_reads_expensive(struct bkey_s_c k) -{ - struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k); - struct bch_extent_crc_unpacked crc; - const union bch_extent_entry *i; - - bkey_for_each_crc(k.k, ptrs, crc, i) - if (crc.csum_type || crc.compression_type) - return true; - return false; -} -static void readpage_bio_extend(struct readpages_iter *iter, - struct bio *bio, - unsigned sectors_this_extent, - bool get_more) -{ - while (bio_sectors(bio) < sectors_this_extent && - bio->bi_vcnt < bio->bi_max_vecs) { - pgoff_t page_offset = bio_end_sector(bio) >> PAGE_SECTOR_SHIFT; - struct page *page = readpage_iter_next(iter); - int ret; - - if (page) { - if (iter->offset + iter->idx != page_offset) - break; + if (!folio_test_uptodate(folio)) { + ret = bch2_read_single_folio(folio, mapping); + if (ret) + goto unlock; + } - iter->idx++; - } else { - if (!get_more) - break; - - 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)); - if (!page) - break; - - if (!__bch2_page_state_create(page, 0)) { - put_page(page); - break; - } - - ret = add_to_page_cache_lru(page, iter->mapping, - page_offset, GFP_NOFS); - if (ret) { - __bch2_page_state_release(page); - put_page(page); - break; - } - - put_page(page); - } - - BUG_ON(!bio_add_page(bio, page, PAGE_SIZE, 0)); - } -} - -static void bchfs_read(struct btree_trans *trans, struct btree_iter *iter, - struct bch_read_bio *rbio, u64 inum, - struct readpages_iter *readpages_iter) -{ - struct bch_fs *c = trans->c; - struct bkey_buf sk; - int flags = BCH_READ_RETRY_IF_STALE| - BCH_READ_MAY_PROMOTE; - int ret = 0; - - rbio->c = c; - rbio->start_time = local_clock(); - - bch2_bkey_buf_init(&sk); -retry: - while (1) { - struct bkey_s_c k; - unsigned bytes, sectors, offset_into_extent; - enum btree_id data_btree = BTREE_ID_extents; - - bch2_btree_iter_set_pos(iter, - POS(inum, rbio->bio.bi_iter.bi_sector)); - - k = bch2_btree_iter_peek_slot(iter); - ret = bkey_err(k); - if (ret) - break; - - offset_into_extent = iter->pos.offset - - bkey_start_offset(k.k); - sectors = k.k->size - offset_into_extent; - - bch2_bkey_buf_reassemble(&sk, c, k); - - ret = bch2_read_indirect_extent(trans, &data_btree, - &offset_into_extent, &sk); - if (ret) - break; - - k = bkey_i_to_s_c(sk.k); - - 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)); - - bytes = min(sectors, bio_sectors(&rbio->bio)) << 9; - swap(rbio->bio.bi_iter.bi_size, bytes); - - if (rbio->bio.bi_iter.bi_size == bytes) - flags |= BCH_READ_LAST_FRAGMENT; - - if (bkey_extent_is_allocation(k.k)) - bch2_add_page_sectors(&rbio->bio, k); - - bch2_read_extent(trans, rbio, iter->pos, - data_btree, k, offset_into_extent, flags); - - if (flags & BCH_READ_LAST_FRAGMENT) - break; - - swap(rbio->bio.bi_iter.bi_size, bytes); - bio_advance(&rbio->bio, bytes); - } - - if (ret == -EINTR) - goto retry; - - if (ret) { - bch_err_inum_ratelimited(c, inum, - "read error %i from btree lookup", ret); - rbio->bio.bi_status = BLK_STS_IOERR; - bio_endio(&rbio->bio); - } - - bch2_bkey_buf_exit(&sk, c); -} - -void bch2_readahead(struct readahead_control *ractl) -{ - 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 btree_trans trans; - struct btree_iter *iter; - struct page *page; - struct readpages_iter readpages_iter; - int ret; - - ret = readpages_iter_init(&readpages_iter, ractl); - BUG_ON(ret); - - bch2_trans_init(&trans, c, 0, 0); - iter = bch2_trans_get_iter(&trans, BTREE_ID_extents, POS_MIN, - BTREE_ITER_SLOTS); - - bch2_pagecache_add_get(&inode->ei_pagecache_lock); - - 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_VECS); - struct bch_read_bio *rbio = - rbio_init(bio_alloc_bioset(GFP_NOFS, n, &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_SECTOR_SHIFT; - rbio->bio.bi_end_io = bch2_readpages_end_io; - BUG_ON(!bio_add_page(&rbio->bio, page, PAGE_SIZE, 0)); - - bchfs_read(&trans, iter, rbio, inode->v.i_ino, - &readpages_iter); - } - - bch2_pagecache_add_put(&inode->ei_pagecache_lock); - - bch2_trans_iter_put(&trans, iter); - bch2_trans_exit(&trans); - kfree(readpages_iter.pages); -} - -static void __bchfs_readpage(struct bch_fs *c, struct bch_read_bio *rbio, - u64 inum, struct page *page) -{ - struct btree_trans trans; - struct btree_iter *iter; - - bch2_page_state_create(page, __GFP_NOFAIL); - - bio_set_op_attrs(&rbio->bio, REQ_OP_READ, REQ_SYNC); - rbio->bio.bi_iter.bi_sector = - (sector_t) page->index << PAGE_SECTOR_SHIFT; - BUG_ON(!bio_add_page(&rbio->bio, page, PAGE_SIZE, 0)); - - bch2_trans_init(&trans, c, 0, 0); - iter = bch2_trans_get_iter(&trans, BTREE_ID_extents, POS_MIN, - BTREE_ITER_SLOTS); - - bchfs_read(&trans, iter, rbio, inum, NULL); - - bch2_trans_iter_put(&trans, iter); - 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->v.i_ino, page); - return 0; -} - -static void bch2_read_single_page_end_io(struct bio *bio) -{ - complete(bio->bi_private); -} - -static int bch2_read_single_page(struct page *page, - struct address_space *mapping) -{ - 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; - int ret; - DECLARE_COMPLETION_ONSTACK(done); - - rbio = rbio_init(bio_alloc_bioset(GFP_NOFS, 1, &c->bio_read), - io_opts(c, &inode->ei_inode)); - rbio->bio.bi_private = &done; - rbio->bio.bi_end_io = bch2_read_single_page_end_io; - - __bchfs_readpage(c, rbio, inode->v.i_ino, page); - wait_for_completion(&done); - - ret = blk_status_to_errno(rbio->bio.bi_status); - bio_put(&rbio->bio); - - if (ret < 0) - return ret; - - SetPageUptodate(page); - return 0; -} - -/* writepages: */ - -struct bch_writepage_state { - struct bch_writepage_io *io; - struct bch_io_opts opts; -}; - -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) - }; -} - -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); -} - -static void bch2_writepage_io_done(struct closure *cl) -{ - struct bch_writepage_io *io = container_of(cl, - struct bch_writepage_io, cl); - 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; - - up(&io->op.c->io_in_flight); - - if (io->op.error) { - set_bit(EI_INODE_ERROR, &io->inode->ei_flags); - - bio_for_each_segment_all(bvec, bio, iter) { - struct bch_page_state *s; - - SetPageError(bvec->bv_page); - mapping_set_error(bvec->bv_page->mapping, -EIO); - - s = __bch2_page_state(bvec->bv_page); - spin_lock(&s->lock); - 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, iter) { - struct bch_page_state *s; - - s = __bch2_page_state(bvec->bv_page); - spin_lock(&s->lock); - for (i = 0; i < PAGE_SECTORS; i++) - s->s[i].nr_replicas = 0; - spin_unlock(&s->lock); - } - } - - /* - * racing with fallocate can cause us to add fewer sectors than - * expected - but we shouldn't add more sectors than expected: - */ - BUG_ON(io->op.i_sectors_delta > 0); - - /* - * (error (due to going RO) halfway through a page can screw that up - * slightly) - * XXX wtf? - BUG_ON(io->op.op.i_sectors_delta >= PAGE_SECTORS); - */ - - /* - * PageWriteback is effectively our ref on the inode - fixup i_blocks - * before calling end_page_writeback: - */ - i_sectors_acct(c, io->inode, NULL, io->op.i_sectors_delta); - - 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); -} - -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); -} - -/* - * Get a bch_writepage_io and add @page to it - appending to an existing one if - * possible, else allocating a new one: - */ -static void bch2_writepage_io_alloc(struct bch_fs *c, - struct writeback_control *wbc, - struct bch_writepage_state *w, - struct bch_inode_info *inode, - u64 sector, - unsigned nr_replicas) -{ - struct bch_write_op *op; - - w->io = container_of(bio_alloc_bioset(GFP_NOFS, BIO_MAX_VECS, - &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; - op_journal_seq_set(op, &inode->ei_journal_seq); - op->nr_replicas = nr_replicas; - op->res.nr_replicas = nr_replicas; - op->write_point = writepoint_hashed(inode->ei_last_dirtied); - op->pos = POS(inode->v.i_ino, sector); - op->wbio.bio.bi_iter.bi_sector = sector; - op->wbio.bio.bi_opf = wbc_to_write_flags(wbc); -} - -static int __bch2_writepage(struct page *page, - struct writeback_control *wbc, - void *data) -{ - struct bch_inode_info *inode = to_bch_ei(page->mapping->host); - struct bch_fs *c = inode->v.i_sb->s_fs_info; - struct bch_writepage_state *w = data; - struct bch_page_state *s, orig; - unsigned i, offset, nr_replicas_this_write = U32_MAX; - loff_t i_size = i_size_read(&inode->v); - pgoff_t end_index = i_size >> PAGE_SHIFT; - int ret; - - EBUG_ON(!PageUptodate(page)); - - /* Is the page fully inside i_size? */ - if (page->index < end_index) - goto do_io; - - /* Is the page fully outside i_size? (truncate in progress) */ - offset = i_size & (PAGE_SIZE - 1); - if (page->index > end_index || !offset) { - unlock_page(page); - return 0; - } - - /* - * The page straddles i_size. It must be zeroed out on each and every - * writepage invocation because it may be mmapped. "A file is mapped - * in multiples of the page size. For a file that is not a multiple of - * the page size, the remaining memory is zeroed when mapped, and - * writes to that region are not written out to the file." - */ - zero_user_segment(page, offset, PAGE_SIZE); -do_io: - s = bch2_page_state_create(page, __GFP_NOFAIL); - - ret = bch2_get_page_disk_reservation(c, inode, page, true); - if (ret) { - SetPageError(page); - mapping_set_error(page->mapping, ret); - unlock_page(page); - return 0; - } - - /* Before unlocking the page, get copy of reservations: */ - orig = *s; - - for (i = 0; i < PAGE_SECTORS; i++) { - if (s->s[i].state < SECTOR_DIRTY) - continue; - - nr_replicas_this_write = - min_t(unsigned, nr_replicas_this_write, - s->s[i].nr_replicas + - s->s[i].replicas_reserved); - } - - for (i = 0; i < PAGE_SECTORS; i++) { - if (s->s[i].state < SECTOR_DIRTY) - continue; - - s->s[i].nr_replicas = w->opts.compression - ? 0 : nr_replicas_this_write; - - s->s[i].replicas_reserved = 0; - s->s[i].state = SECTOR_ALLOCATED; - } - - BUG_ON(atomic_read(&s->write_count)); - atomic_set(&s->write_count, 1); - - BUG_ON(PageWriteback(page)); - set_page_writeback(page); - - unlock_page(page); - - offset = 0; - while (1) { - unsigned sectors = 1, dirty_sectors = 0, reserved_sectors = 0; - u64 sector; - - while (offset < PAGE_SECTORS && - orig.s[offset].state < SECTOR_DIRTY) - offset++; - - if (offset == PAGE_SECTORS) - break; - - sector = ((u64) page->index << PAGE_SECTOR_SHIFT) + offset; - - while (offset + sectors < PAGE_SECTORS && - orig.s[offset + sectors].state >= SECTOR_DIRTY) - sectors++; - - for (i = offset; i < offset + sectors; i++) { - reserved_sectors += orig.s[i].replicas_reserved; - dirty_sectors += orig.s[i].state == SECTOR_DIRTY; - } - - if (w->io && - (w->io->op.res.nr_replicas != nr_replicas_this_write || - bio_full(&w->io->op.wbio.bio, PAGE_SIZE) || - w->io->op.wbio.bio.bi_iter.bi_size + (sectors << 9) >= - (BIO_MAX_VECS * PAGE_SIZE) || - bio_end_sector(&w->io->op.wbio.bio) != sector)) - bch2_writepage_do_io(w); - - if (!w->io) - bch2_writepage_io_alloc(c, wbc, w, inode, sector, - nr_replicas_this_write); - - atomic_inc(&s->write_count); - - BUG_ON(inode != w->io->inode); - BUG_ON(!bio_add_page(&w->io->op.wbio.bio, page, - 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))); - - w->io->op.res.sectors += reserved_sectors; - w->io->op.i_sectors_delta -= dirty_sectors; - w->io->op.new_i_size = i_size; - - offset += sectors; - } - - if (atomic_dec_and_test(&s->write_count)) - end_page_writeback(page); - - return 0; -} - -int bch2_writepages(struct address_space *mapping, struct writeback_control *wbc) -{ - struct bch_fs *c = mapping->host->i_sb->s_fs_info; - struct bch_writepage_state w = - bch_writepage_state_init(c, to_bch_ei(mapping->host)); - struct blk_plug plug; - int ret; - - blk_start_plug(&plug); - ret = write_cache_pages(mapping, wbc, __bch2_writepage, &w); - 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; -} - -/* buffered writes: */ - -int bch2_write_begin(struct file *file, struct address_space *mapping, - loff_t pos, unsigned len, unsigned flags, - struct page **pagep, void **fsdata) -{ - struct bch_inode_info *inode = to_bch_ei(mapping->host); - struct bch_fs *c = inode->v.i_sb->s_fs_info; - struct bch2_page_reservation *res; - pgoff_t index = pos >> PAGE_SHIFT; - unsigned offset = pos & (PAGE_SIZE - 1); - struct page *page; - int ret = -ENOMEM; - - res = kmalloc(sizeof(*res), GFP_KERNEL); - if (!res) - return -ENOMEM; - - bch2_page_reservation_init(c, inode, res); - *fsdata = res; - - bch2_pagecache_add_get(&inode->ei_pagecache_lock); - - page = grab_cache_page_write_begin(mapping, index, flags); - if (!page) - goto err_unlock; - - if (PageUptodate(page)) - goto out; - - /* If we're writing entire page, don't need to read it in first: */ - if (len == PAGE_SIZE) - goto out; - - if (!offset && pos + len >= inode->v.i_size) { - zero_user_segment(page, len, PAGE_SIZE); - flush_dcache_page(page); - goto out; - } - - if (index > inode->v.i_size >> PAGE_SHIFT) { - zero_user_segments(page, 0, offset, offset + len, PAGE_SIZE); - flush_dcache_page(page); - goto out; - } -readpage: - ret = bch2_read_single_page(page, mapping); - if (ret) - goto err; -out: - ret = bch2_page_reservation_get(c, inode, page, res, - offset, len, true); - if (ret) { - if (!PageUptodate(page)) { - /* - * If the page hasn't been read in, we won't know if we - * actually need a reservation - we don't actually need - * to read here, we just need to check if the page is - * fully backed by uncompressed data: - */ - goto readpage; - } - - goto err; - } - - *pagep = page; - return 0; -err: - unlock_page(page); - put_page(page); - *pagep = NULL; -err_unlock: - bch2_pagecache_add_put(&inode->ei_pagecache_lock); - kfree(res); - *fsdata = NULL; - return ret; -} - -int bch2_write_end(struct file *file, struct address_space *mapping, - loff_t pos, unsigned len, unsigned copied, - struct page *page, void *fsdata) -{ - struct bch_inode_info *inode = to_bch_ei(mapping->host); - struct bch_fs *c = inode->v.i_sb->s_fs_info; - struct bch2_page_reservation *res = fsdata; - unsigned offset = pos & (PAGE_SIZE - 1); - - lockdep_assert_held(&inode->v.i_rwsem); - - if (unlikely(copied < len && !PageUptodate(page))) { - /* - * The page needs to be read in, but that would destroy - * our partial write - simplest thing is to just force - * userspace to redo the write: - */ - zero_user(page, 0, PAGE_SIZE); - flush_dcache_page(page); - copied = 0; - } - - spin_lock(&inode->v.i_lock); - if (pos + copied > inode->v.i_size) - i_size_write(&inode->v, pos + copied); - spin_unlock(&inode->v.i_lock); - - if (copied) { - if (!PageUptodate(page)) - SetPageUptodate(page); - - bch2_set_page_dirty(c, inode, page, res, offset, copied); - - inode->ei_last_dirtied = (unsigned long) current; - } - - unlock_page(page); - put_page(page); - bch2_pagecache_add_put(&inode->ei_pagecache_lock); - - bch2_page_reservation_put(c, inode, res); - kfree(res); - - return copied; -} - -#define WRITE_BATCH_PAGES 32 - -static int __bch2_buffered_write(struct bch_inode_info *inode, - struct address_space *mapping, - struct iov_iter *iter, - loff_t pos, unsigned len) -{ - struct bch_fs *c = inode->v.i_sb->s_fs_info; - struct page *pages[WRITE_BATCH_PAGES]; - struct bch2_page_reservation res; - unsigned long index = pos >> PAGE_SHIFT; - unsigned offset = pos & (PAGE_SIZE - 1); - unsigned nr_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE); - unsigned i, reserved = 0, set_dirty = 0; - unsigned copied = 0, nr_pages_copied = 0; - int ret = 0; - - BUG_ON(!len); - BUG_ON(nr_pages > ARRAY_SIZE(pages)); - - 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); - if (!pages[i]) { - nr_pages = i; - if (!i) { - ret = -ENOMEM; - goto out; - } - len = min_t(unsigned, len, - nr_pages * PAGE_SIZE - offset); - break; - } - } - - if (offset && !PageUptodate(pages[0])) { - ret = bch2_read_single_page(pages[0], mapping); - if (ret) - goto out; - } - - if ((pos + len) & (PAGE_SIZE - 1) && - !PageUptodate(pages[nr_pages - 1])) { - if ((index + nr_pages - 1) << PAGE_SHIFT >= inode->v.i_size) { - zero_user(pages[nr_pages - 1], 0, PAGE_SIZE); - } else { - ret = bch2_read_single_page(pages[nr_pages - 1], mapping); - if (ret) - goto out; - } - } - - while (reserved < len) { - struct page *page = pages[(offset + reserved) >> PAGE_SHIFT]; - unsigned pg_offset = (offset + reserved) & (PAGE_SIZE - 1); - unsigned pg_len = min_t(unsigned, len - reserved, - PAGE_SIZE - pg_offset); -retry_reservation: - ret = bch2_page_reservation_get(c, inode, page, &res, - pg_offset, pg_len, true); - - if (ret && !PageUptodate(page)) { - ret = bch2_read_single_page(page, mapping); - if (!ret) - goto retry_reservation; - } - - if (ret) - goto out; - - reserved += pg_len; - } - - if (mapping_writably_mapped(mapping)) - for (i = 0; i < nr_pages; i++) - flush_dcache_page(pages[i]); - - while (copied < len) { - struct page *page = pages[(offset + copied) >> PAGE_SHIFT]; - unsigned pg_offset = (offset + copied) & (PAGE_SIZE - 1); - unsigned pg_len = min_t(unsigned, len - copied, - PAGE_SIZE - pg_offset); - unsigned pg_copied = iov_iter_copy_from_user_atomic(page, - iter, pg_offset, pg_len); - - if (!pg_copied) - break; - - if (!PageUptodate(page) && - pg_copied != PAGE_SIZE && - pos + copied + pg_copied < inode->v.i_size) { - zero_user(page, 0, PAGE_SIZE); - break; - } - - flush_dcache_page(page); - iov_iter_advance(iter, pg_copied); - copied += pg_copied; - - if (pg_copied != pg_len) - break; - } - - if (!copied) - goto out; - - spin_lock(&inode->v.i_lock); - if (pos + copied > inode->v.i_size) - i_size_write(&inode->v, pos + copied); - spin_unlock(&inode->v.i_lock); - - while (set_dirty < copied) { - struct page *page = pages[(offset + set_dirty) >> PAGE_SHIFT]; - unsigned pg_offset = (offset + set_dirty) & (PAGE_SIZE - 1); - unsigned pg_len = min_t(unsigned, copied - set_dirty, - PAGE_SIZE - pg_offset); - - if (!PageUptodate(page)) - SetPageUptodate(page); - - bch2_set_page_dirty(c, inode, page, &res, pg_offset, pg_len); - unlock_page(page); - put_page(page); - - set_dirty += pg_len; - } - - nr_pages_copied = DIV_ROUND_UP(offset + copied, PAGE_SIZE); - inode->ei_last_dirtied = (unsigned long) current; -out: - for (i = nr_pages_copied; i < nr_pages; i++) { - unlock_page(pages[i]); - put_page(pages[i]); - } - - bch2_page_reservation_put(c, inode, &res); - - return copied ?: ret; -} - -static ssize_t bch2_buffered_write(struct kiocb *iocb, struct iov_iter *iter) -{ - struct file *file = iocb->ki_filp; - struct address_space *mapping = file->f_mapping; - struct bch_inode_info *inode = file_bch_inode(file); - loff_t pos = iocb->ki_pos; - ssize_t written = 0; - int ret = 0; - - bch2_pagecache_add_get(&inode->ei_pagecache_lock); - - do { - unsigned offset = pos & (PAGE_SIZE - 1); - unsigned bytes = min_t(unsigned long, iov_iter_count(iter), - PAGE_SIZE * WRITE_BATCH_PAGES - offset); -again: - /* - * Bring in the user page that we will copy from _first_. - * Otherwise there's a nasty deadlock on copying from the - * same page as we're writing to, without it being marked - * up-to-date. - * - * Not only is this an optimisation, but it is also required - * to check that the address is actually valid, when atomic - * usercopies are used, below. - */ - if (unlikely(iov_iter_fault_in_readable(iter, bytes))) { - bytes = min_t(unsigned long, iov_iter_count(iter), - PAGE_SIZE - offset); - - if (unlikely(iov_iter_fault_in_readable(iter, bytes))) { - ret = -EFAULT; - break; - } - } - - if (unlikely(fatal_signal_pending(current))) { - ret = -EINTR; - break; - } - - ret = __bch2_buffered_write(inode, mapping, iter, pos, bytes); - if (unlikely(ret < 0)) - break; - - cond_resched(); - - if (unlikely(ret == 0)) { - /* - * If we were unable to copy any data at all, we must - * fall back to a single segment length write. - * - * If we didn't fallback here, we could livelock - * because not all segments in the iov can be copied at - * once without a pagefault. - */ - bytes = min_t(unsigned long, PAGE_SIZE - offset, - iov_iter_single_seg_count(iter)); - goto again; - } - pos += ret; - written += ret; - ret = 0; - - balance_dirty_pages_ratelimited(mapping); - } while (iov_iter_count(iter)); - - bch2_pagecache_add_put(&inode->ei_pagecache_lock); - - return written ? written : ret; -} - -/* O_DIRECT reads */ - -static void bio_check_or_release(struct bio *bio, bool check_dirty) -{ - if (check_dirty) { - bio_check_pages_dirty(bio); - } else { - bio_release_pages(bio, false); - bio_put(bio); - } -} - -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); - bio_check_or_release(&dio->rbio.bio, dio->should_dirty); -} - -static void bch2_direct_IO_read_endio(struct bio *bio) -{ - struct dio_read *dio = bio->bi_private; - - if (bio->bi_status) - dio->ret = blk_status_to_errno(bio->bi_status); - - closure_put(&dio->cl); -} - -static void bch2_direct_IO_read_split_endio(struct bio *bio) -{ - struct dio_read *dio = bio->bi_private; - bool should_dirty = dio->should_dirty; - - bch2_direct_IO_read_endio(bio); - bio_check_or_release(bio, should_dirty); -} - -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 dio_read *dio; - struct bio *bio; - loff_t offset = req->ki_pos; - bool sync = is_sync_kiocb(req); - size_t shorten; - ssize_t ret; - - if ((offset|iter->count) & (block_bytes(c) - 1)) - return -EINVAL; - - ret = min_t(loff_t, iter->count, - max_t(loff_t, 0, i_size_read(&inode->v) - offset)); - - if (!ret) - return ret; - - 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_VECS), - &c->dio_read_bioset); - - bio->bi_end_io = bch2_direct_IO_read_endio; - - dio = container_of(bio, struct dio_read, rbio.bio); - closure_init(&dio->cl, NULL); - - /* - * this is a _really_ horrible hack just to avoid an atomic sub at the - * end: - */ - if (!sync) { - set_closure_fn(&dio->cl, bch2_dio_read_complete, NULL); - atomic_set(&dio->cl.remaining, - CLOSURE_REMAINING_INITIALIZER - - CLOSURE_RUNNING + - CLOSURE_DESTRUCTOR); - } else { - atomic_set(&dio->cl.remaining, - CLOSURE_REMAINING_INITIALIZER + 1); - } - - dio->req = req; - dio->ret = ret; - /* - * This is one of the sketchier things I've encountered: we have to skip - * the dirtying of requests that are internal from the kernel (i.e. from - * loopback), because we'll deadlock on page_lock. - */ - dio->should_dirty = iter_is_iovec(iter); - - goto start; - while (iter->count) { - bio = bio_alloc_bioset(GFP_KERNEL, - iov_iter_npages(iter, BIO_MAX_VECS), - &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_iter.bi_sector = offset >> 9; - bio->bi_private = dio; - - ret = bio_iov_iter_get_pages(bio, iter); - if (ret < 0) { - /* XXX: fault inject this path */ - bio->bi_status = BLK_STS_RESOURCE; - bio_endio(bio); - break; - } - - offset += bio->bi_iter.bi_size; - - if (dio->should_dirty) - bio_set_pages_dirty(bio); - - if (iter->count) - closure_get(&dio->cl); - - bch2_read(c, rbio_init(bio, opts), inode->v.i_ino); - } - - iter->count += shorten; - - if (sync) { - closure_sync(&dio->cl); - closure_debug_destroy(&dio->cl); - ret = dio->ret; - bio_check_or_release(&dio->rbio.bio, dio->should_dirty); - return ret; - } else { - return -EIOCBQUEUED; - } -} - -ssize_t bch2_read_iter(struct kiocb *iocb, struct iov_iter *iter) -{ - struct file *file = iocb->ki_filp; - struct bch_inode_info *inode = file_bch_inode(file); - struct address_space *mapping = file->f_mapping; - size_t count = iov_iter_count(iter); - ssize_t ret; - - if (!count) - return 0; /* skip atime */ - - 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; - - file_accessed(file); - - blk_start_plug(&plug); - ret = bch2_direct_IO_read(iocb, iter); - blk_finish_plug(&plug); - - if (ret >= 0) - iocb->ki_pos += ret; - } else { - bch2_pagecache_add_get(&inode->ei_pagecache_lock); - ret = generic_file_read_iter(iocb, iter); - bch2_pagecache_add_put(&inode->ei_pagecache_lock); - } - - return ret; -} - -/* O_DIRECT writes */ - -static void bch2_dio_write_loop_async(struct bch_write_op *); - -static long bch2_dio_write_loop(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 bio *bio = &dio->op.wbio.bio; - struct bvec_iter_all iter; - struct bio_vec *bv; - unsigned unaligned, iter_count; - bool sync = dio->sync, dropped_locks; - long ret; - - if (dio->loop) - goto loop; - - down(&c->io_in_flight); - - while (1) { - iter_count = dio->iter.count; - - if (kthread) - kthread_use_mm(dio->mm); - BUG_ON(current->faults_disabled_mapping); - current->faults_disabled_mapping = mapping; - - ret = bio_iov_iter_get_pages(bio, &dio->iter); - - dropped_locks = fdm_dropped_locks(); - - current->faults_disabled_mapping = NULL; - if (kthread) - kthread_unuse_mm(dio->mm); - - /* - * If the fault handler returned an error but also signalled - * that it dropped & retook ei_pagecache_lock, we just need to - * re-shoot down the page cache and retry: - */ - if (dropped_locks && ret) - ret = 0; - - if (unlikely(ret < 0)) - goto err; - - if (unlikely(dropped_locks)) { - ret = write_invalidate_inode_pages_range(mapping, - req->ki_pos, - req->ki_pos + iter_count - 1); - if (unlikely(ret)) - goto err; - - if (!bio->bi_iter.bi_size) - continue; - } - - unaligned = bio->bi_iter.bi_size & (block_bytes(c) - 1); - bio->bi_iter.bi_size -= unaligned; - iov_iter_revert(&dio->iter, unaligned); - - if (!bio->bi_iter.bi_size) { - /* - * bio_iov_iter_get_pages was only able to get < - * blocksize worth of pages: - */ - bio_for_each_segment_all(bv, bio, iter) - put_page(bv->bv_page); - ret = -EFAULT; - goto err; - } - - bch2_write_op_init(&dio->op, c, io_opts(c, &inode->ei_inode)); - dio->op.end_io = bch2_dio_write_loop_async; - dio->op.target = dio->op.opts.foreground_target; - op_journal_seq_set(&dio->op, &inode->ei_journal_seq); - dio->op.write_point = writepoint_hashed((unsigned long) current); - dio->op.nr_replicas = dio->op.opts.data_replicas; - dio->op.pos = POS(inode->v.i_ino, (u64) req->ki_pos >> 9); - - if ((req->ki_flags & IOCB_DSYNC) && - !c->opts.journal_flush_disabled) - dio->op.flags |= BCH_WRITE_FLUSH; - dio->op.flags |= BCH_WRITE_CHECK_ENOSPC; - - 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, dio->op.pos, - bio_sectors(bio), - dio->op.opts.data_replicas, - dio->op.opts.compression != 0)) - 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; - } - - 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 - 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); - - if (likely(!bio_flagged(bio, BIO_NO_PAGE_REF))) - bio_for_each_segment_all(bv, bio, iter) - put_page(bv->bv_page); - - if (dio->op.error) { - set_bit(EI_INODE_ERROR, &inode->ei_flags); - break; - } - - if (!dio->iter.count) - break; - - bio_reset(bio); - reinit_completion(&dio->done); - } - - 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); - - if (dio->free_iov) - kfree(dio->iter.iov); - - bio_put(bio); - - /* inode->i_dio_count is our ref on inode and thus bch_fs */ - inode_dio_end(&inode->v); - - if (!sync) { - req->ki_complete(req, ret, 0); - ret = -EIOCBQUEUED; - } - return ret; -} - -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); - else - bch2_dio_write_loop(dio); -} - -static noinline -ssize_t bch2_direct_write(struct kiocb *req, struct iov_iter *iter) -{ - struct file *file = req->ki_filp; - struct address_space *mapping = file->f_mapping; - struct bch_inode_info *inode = file_bch_inode(file); - struct bch_fs *c = inode->v.i_sb->s_fs_info; - struct dio_write *dio; - struct bio *bio; - bool locked = true, extending; - ssize_t ret; - - prefetch(&c->opts); - prefetch((void *) &c->opts + 64); - prefetch(&inode->ei_inode); - prefetch((void *) &inode->ei_inode + 64); - - inode_lock(&inode->v); - - ret = generic_write_checks(req, iter); - if (unlikely(ret <= 0)) - goto err; - - ret = file_remove_privs(file); - if (unlikely(ret)) - goto err; - - ret = file_update_time(file); - if (unlikely(ret)) - goto err; - - if (unlikely((req->ki_pos|iter->count) & (block_bytes(c) - 1))) - goto err; - - inode_dio_begin(&inode->v); - bch2_pagecache_block_get(&inode->ei_pagecache_lock); - - extending = req->ki_pos + iter->count > inode->v.i_size; - if (!extending) { - inode_unlock(&inode->v); - locked = false; - } - - bio = bio_alloc_bioset(GFP_KERNEL, - iov_iter_is_bvec(iter) - ? 0 - : iov_iter_npages(iter, BIO_MAX_VECS), - &c->dio_write_bioset); - dio = container_of(bio, struct dio_write, op.wbio.bio); - init_completion(&dio->done); - dio->req = req; - dio->mm = current->mm; - dio->loop = false; - dio->sync = is_sync_kiocb(req) || extending; - dio->free_iov = false; - dio->quota_res.sectors = 0; - dio->written = 0; - dio->iter = *iter; - - 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; - - ret = bch2_dio_write_loop(dio); -err: - if (locked) - 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); - bio_put(bio); - inode_dio_end(&inode->v); - goto err; -} - -ssize_t bch2_write_iter(struct kiocb *iocb, struct iov_iter *from) -{ - struct file *file = iocb->ki_filp; - struct bch_inode_info *inode = file_bch_inode(file); - ssize_t ret; - - if (iocb->ki_flags & IOCB_DIRECT) - return bch2_direct_write(iocb, from); - - /* We can write back this queue in page reclaim */ - current->backing_dev_info = inode_to_bdi(&inode->v); - inode_lock(&inode->v); - - ret = generic_write_checks(iocb, from); - if (ret <= 0) - goto unlock; - - ret = file_remove_privs(file); - if (ret) - goto unlock; - - ret = file_update_time(file); - if (ret) - goto unlock; - - ret = bch2_buffered_write(iocb, from); - if (likely(ret > 0)) - iocb->ki_pos += ret; -unlock: - inode_unlock(&inode->v); - current->backing_dev_info = NULL; - - if (ret > 0) - ret = generic_write_sync(iocb, ret); - - return ret; -} - -/* fsync: */ - -int bch2_fsync(struct file *file, loff_t start, loff_t end, int datasync) -{ - struct bch_inode_info *inode = file_bch_inode(file); - struct bch_fs *c = inode->v.i_sb->s_fs_info; - int ret, ret2; - - ret = file_write_and_wait_range(file, start, end); - if (ret) - return ret; - - if (datasync && !(inode->v.i_state & I_DIRTY_DATASYNC)) - goto out; - - ret = sync_inode_metadata(&inode->v, 1); + ret = bch2_folio_set(c, inode_inum(inode), &folio, 1); if (ret) - return ret; -out: - if (!c->opts.journal_flush_disabled) - ret = bch2_journal_flush_seq(&c->journal, - inode->ei_journal_seq); - ret2 = file_check_and_advance_wb_err(file); - - return ret ?: ret2; -} - -/* truncate: */ - -static inline int range_has_data(struct bch_fs *c, - struct bpos start, - struct bpos end) -{ - struct btree_trans trans; - struct btree_iter *iter; - struct bkey_s_c k; - int ret = 0; - - bch2_trans_init(&trans, c, 0, 0); - - for_each_btree_key(&trans, iter, BTREE_ID_extents, start, 0, k, ret) { - if (bkey_cmp(bkey_start_pos(k.k), end) >= 0) - break; - - if (bkey_extent_is_data(k.k)) { - ret = 1; - break; - } - } - bch2_trans_iter_put(&trans, iter); - - return bch2_trans_exit(&trans) ?: ret; -} - -static int __bch2_truncate_page(struct bch_inode_info *inode, - pgoff_t index, loff_t start, loff_t end) -{ - struct bch_fs *c = inode->v.i_sb->s_fs_info; - struct address_space *mapping = inode->v.i_mapping; - struct bch_page_state *s; - unsigned start_offset = start & (PAGE_SIZE - 1); - unsigned end_offset = ((end - 1) & (PAGE_SIZE - 1)) + 1; - unsigned i; - struct page *page; - int ret = 0; - - /* Page boundary? Nothing to do */ - if (!((index == start >> PAGE_SHIFT && start_offset) || - (index == end >> PAGE_SHIFT && end_offset != PAGE_SIZE))) - return 0; - - /* Above i_size? */ - if (index << PAGE_SHIFT >= inode->v.i_size) - return 0; - - page = find_lock_page(mapping, index); - if (!page) { - /* - * XXX: we're doing two index lookups when we end up reading the - * page - */ - ret = range_has_data(c, - POS(inode->v.i_ino, index << PAGE_SECTOR_SHIFT), - POS(inode->v.i_ino, (index + 1) << PAGE_SECTOR_SHIFT)); - if (ret <= 0) - return ret; - - page = find_or_create_page(mapping, index, GFP_KERNEL); - if (unlikely(!page)) { - ret = -ENOMEM; - goto out; - } - } - - s = bch2_page_state_create(page, 0); - if (!s) { - ret = -ENOMEM; - goto unlock; - } - - if (!PageUptodate(page)) { - ret = bch2_read_single_page(page, mapping); - if (ret) - goto unlock; - } - - if (index != start >> PAGE_SHIFT) - start_offset = 0; - if (index != end >> PAGE_SHIFT) - end_offset = PAGE_SIZE; + goto unlock; for (i = round_up(start_offset, block_bytes(c)) >> 9; i < round_down(end_offset, block_bytes(c)) >> 9; i++) { s->s[i].nr_replicas = 0; - s->s[i].state = SECTOR_UNALLOCATED; + + i_sectors_delta -= s->s[i].state == SECTOR_dirty; + bch2_folio_sector_set(folio, s, i, SECTOR_unallocated); } - zero_user_segment(page, start_offset, end_offset); + bch2_i_sectors_acct(c, inode, NULL, i_sectors_delta); + + /* + * Caller needs to know whether this folio will be written out by + * writeback - doing an i_size update if necessary - or whether it will + * be responsible for the i_size update. + * + * Note that we shouldn't ever see a folio beyond EOF, but check and + * warn if so. This has been observed by failure to clean up folios + * after a short write and there's still a chance reclaim will fix + * things up. + */ + WARN_ON_ONCE(folio_pos(folio) >= inode->v.i_size); + end_pos = folio_end_pos(folio); + if (inode->v.i_size > folio_pos(folio)) + end_pos = min_t(u64, inode->v.i_size, end_pos); + ret = s->s[folio_pos_to_s(folio, end_pos - 1)].state >= SECTOR_dirty; + + folio_zero_segment(folio, start_offset, end_offset); /* * Bit of a hack - we don't want truncate to fail due to -ENOSPC. * - * XXX: because we aren't currently tracking whether the page has actual + * XXX: because we aren't currently tracking whether the folio has actual * data in it (vs. just 0s, or only partially written) this wrong. ick. */ - ret = bch2_get_page_disk_reservation(c, inode, page, false); - BUG_ON(ret); + BUG_ON(bch2_get_folio_disk_reservation(c, inode, folio, false)); /* * This removes any writeable userspace mappings; we need to force * .page_mkwrite to be called again before any mmapped writes, to * redirty the full page: */ - page_mkclean(page); - __set_page_dirty_nobuffers(page); + folio_mkclean(folio); + filemap_dirty_folio(mapping, folio); unlock: - unlock_page(page); - put_page(page); + folio_unlock(folio); + folio_put(folio); out: return ret; } -static int bch2_truncate_page(struct bch_inode_info *inode, loff_t from) +static int bch2_truncate_folio(struct bch_inode_info *inode, loff_t from) +{ + return __bch2_truncate_folio(inode, from >> PAGE_SHIFT, + from, ANYSINT_MAX(loff_t)); +} + +static int bch2_truncate_folios(struct bch_inode_info *inode, + loff_t start, loff_t end) { - return __bch2_truncate_page(inode, from >> PAGE_SHIFT, - from, round_up(from, PAGE_SIZE)); + int ret = __bch2_truncate_folio(inode, start >> PAGE_SHIFT, + start, end); + + if (ret >= 0 && + start >> PAGE_SHIFT != end >> PAGE_SHIFT) + ret = __bch2_truncate_folio(inode, + (end - 1) >> PAGE_SHIFT, + start, end); + return ret; } -static int bch2_extend(struct user_namespace *mnt_userns, +static int bch2_extend(struct mnt_idmap *idmap, struct bch_inode_info *inode, struct bch_inode_unpacked *inode_u, struct iattr *iattr) @@ -2273,64 +390,35 @@ static int bch2_extend(struct user_namespace *mnt_userns, truncate_setsize(&inode->v, iattr->ia_size); - return bch2_setattr_nonsize(mnt_userns, inode, iattr); -} - -static int bch2_truncate_finish_fn(struct bch_inode_info *inode, - struct bch_inode_unpacked *bi, - void *p) -{ - bi->bi_flags &= ~BCH_INODE_I_SIZE_DIRTY; - return 0; -} - -static int bch2_truncate_start_fn(struct bch_inode_info *inode, - struct bch_inode_unpacked *bi, void *p) -{ - u64 *new_i_size = p; - - bi->bi_flags |= BCH_INODE_I_SIZE_DIRTY; - bi->bi_size = *new_i_size; - return 0; + return bch2_setattr_nonsize(idmap, inode, iattr); } -int bch2_truncate(struct user_namespace *mnt_userns, +int bchfs_truncate(struct mnt_idmap *idmap, struct bch_inode_info *inode, struct iattr *iattr) { struct bch_fs *c = inode->v.i_sb->s_fs_info; struct address_space *mapping = inode->v.i_mapping; struct bch_inode_unpacked inode_u; - struct btree_trans trans; - struct btree_iter *iter; - u64 new_i_size = iattr->ia_size; s64 i_sectors_delta = 0; int ret = 0; /* - * Don't update timestamps if we're not doing anything: + * If the truncate call with change the size of the file, the + * cmtimes should be updated. If the size will not change, we + * do not need to update the cmtimes. */ - if (iattr->ia_size == inode->v.i_size) - return 0; - - if (!(iattr->ia_valid & ATTR_MTIME)) - ktime_get_coarse_real_ts64(&iattr->ia_mtime); - if (!(iattr->ia_valid & ATTR_CTIME)) - ktime_get_coarse_real_ts64(&iattr->ia_ctime); - iattr->ia_valid |= ATTR_MTIME|ATTR_CTIME; + if (iattr->ia_size != inode->v.i_size) { + if (!(iattr->ia_valid & ATTR_MTIME)) + ktime_get_coarse_real_ts64(&iattr->ia_mtime); + if (!(iattr->ia_valid & ATTR_CTIME)) + ktime_get_coarse_real_ts64(&iattr->ia_ctime); + iattr->ia_valid |= ATTR_MTIME|ATTR_CTIME; + } inode_dio_wait(&inode->v); - bch2_pagecache_block_get(&inode->ei_pagecache_lock); - - /* - * fetch current on disk i_size: inode is locked, i_size can only - * increase underneath us: - */ - bch2_trans_init(&trans, c, 0, 0); - iter = bch2_inode_peek(&trans, &inode_u, inode->v.i_ino, 0); - ret = PTR_ERR_OR_ZERO(iter); - bch2_trans_iter_put(&trans, iter); - bch2_trans_exit(&trans); + bch2_pagecache_block_get(inode); + ret = bch2_inode_find_by_inum(c, inode_inum(inode), &inode_u); if (ret) goto err; @@ -2343,20 +431,24 @@ 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); + ret = bch2_extend(idmap, inode, &inode_u, iattr); goto err; } iattr->ia_valid &= ~ATTR_SIZE; - ret = bch2_truncate_page(inode, iattr->ia_size); - if (unlikely(ret)) + ret = bch2_truncate_folio(inode, iattr->ia_size); + if (unlikely(ret < 0)) goto err; + truncate_setsize(&inode->v, iattr->ia_size); + /* * When extending, we're going to write the new i_size to disk * immediately so we need to flush anything above the current on disk @@ -2378,37 +470,33 @@ int bch2_truncate(struct user_namespace *mnt_userns, if (ret) goto err; - mutex_lock(&inode->ei_update_lock); - ret = bch2_write_inode(c, inode, bch2_truncate_start_fn, - &new_i_size, 0); - mutex_unlock(&inode->ei_update_lock); - - if (unlikely(ret)) - goto err; - - truncate_setsize(&inode->v, iattr->ia_size); - - ret = bch2_fpunch(c, inode->v.i_ino, - round_up(iattr->ia_size, block_bytes(c)) >> 9, - U64_MAX, &inode->ei_journal_seq, &i_sectors_delta); - i_sectors_acct(c, inode, NULL, i_sectors_delta); + ret = bch2_truncate(c, inode_inum(inode), iattr->ia_size, &i_sectors_delta); + bch2_i_sectors_acct(c, inode, NULL, i_sectors_delta); - if (unlikely(ret)) + if (unlikely(ret)) { + /* + * If we error here, VFS caches are now inconsistent with btree + */ + set_bit(EI_INODE_ERROR, &inode->ei_flags); goto err; + } - mutex_lock(&inode->ei_update_lock); - ret = bch2_write_inode(c, inode, bch2_truncate_finish_fn, NULL, 0); - mutex_unlock(&inode->ei_update_lock); + 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); - ret = bch2_setattr_nonsize(mnt_userns, inode, iattr); + ret = bch2_setattr_nonsize(idmap, inode, iattr); err: - bch2_pagecache_block_put(&inode->ei_pagecache_lock); - return ret; + bch2_pagecache_block_put(inode); + return bch2_err_class(ret); } /* fallocate: */ -static int inode_update_times_fn(struct bch_inode_info *inode, +static int inode_update_times_fn(struct btree_trans *trans, + struct bch_inode_info *inode, struct bch_inode_unpacked *bi, void *p) { struct bch_fs *c = inode->v.i_sb->s_fs_info; @@ -2420,49 +508,39 @@ static int inode_update_times_fn(struct bch_inode_info *inode, static long bchfs_fpunch(struct bch_inode_info *inode, loff_t offset, loff_t len) { struct bch_fs *c = inode->v.i_sb->s_fs_info; - u64 discard_start = round_up(offset, block_bytes(c)) >> 9; - u64 discard_end = round_down(offset + len, block_bytes(c)) >> 9; + u64 end = offset + len; + u64 block_start = round_up(offset, block_bytes(c)); + u64 block_end = round_down(end, block_bytes(c)); + bool truncated_last_page; int ret = 0; - inode_lock(&inode->v); - inode_dio_wait(&inode->v); - bch2_pagecache_block_get(&inode->ei_pagecache_lock); - - ret = __bch2_truncate_page(inode, - offset >> PAGE_SHIFT, - offset, offset + len); - if (unlikely(ret)) + ret = bch2_truncate_folios(inode, offset, end); + if (unlikely(ret < 0)) goto err; - if (offset >> PAGE_SHIFT != - (offset + len) >> PAGE_SHIFT) { - ret = __bch2_truncate_page(inode, - (offset + len) >> PAGE_SHIFT, - offset, offset + len); - if (unlikely(ret)) - goto err; - } + truncated_last_page = ret; - truncate_pagecache_range(&inode->v, offset, offset + len - 1); + truncate_pagecache_range(&inode->v, offset, end - 1); - if (discard_start < discard_end) { + if (block_start < block_end) { s64 i_sectors_delta = 0; - ret = bch2_fpunch(c, inode->v.i_ino, - discard_start, discard_end, - &inode->ei_journal_seq, + ret = bch2_fpunch(c, inode_inum(inode), + block_start >> 9, block_end >> 9, &i_sectors_delta); - i_sectors_acct(c, inode, NULL, i_sectors_delta); + bch2_i_sectors_acct(c, inode, NULL, i_sectors_delta); } mutex_lock(&inode->ei_update_lock); - ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL, - ATTR_MTIME|ATTR_CTIME) ?: ret; + if (end >= inode->v.i_size && !truncated_last_page) { + ret = bch2_write_inode_size(c, inode, inode->v.i_size, + ATTR_MTIME|ATTR_CTIME); + } else { + ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL, + ATTR_MTIME|ATTR_CTIME); + } mutex_unlock(&inode->ei_update_lock); err: - bch2_pagecache_block_put(&inode->ei_pagecache_lock); - inode_unlock(&inode->v); - return ret; } @@ -2472,180 +550,33 @@ static long bchfs_fcollapse_finsert(struct bch_inode_info *inode, { struct bch_fs *c = inode->v.i_sb->s_fs_info; struct address_space *mapping = inode->v.i_mapping; - struct bkey_buf copy; - struct btree_trans trans; - struct btree_iter *src, *dst, *del; - loff_t shift, new_size; - u64 src_start; + s64 i_sectors_delta = 0; int ret = 0; if ((offset | len) & (block_bytes(c) - 1)) return -EINVAL; - /* - * We need i_mutex to keep the page cache consistent with the extents - * btree, and the btree consistent with i_size - we don't need outside - * locking for the extents btree itself, because we're using linked - * iterators - */ - inode_lock(&inode->v); - inode_dio_wait(&inode->v); - bch2_pagecache_block_get(&inode->ei_pagecache_lock); - if (insert) { - ret = -EFBIG; - if (inode->v.i_sb->s_maxbytes - inode->v.i_size < len) - goto err; - - ret = -EINVAL; if (offset >= inode->v.i_size) - goto err; - - src_start = U64_MAX; - shift = len; + return -EINVAL; } else { - ret = -EINVAL; if (offset + len >= inode->v.i_size) - goto err; - - src_start = offset + len; - shift = -len; + return -EINVAL; } - new_size = inode->v.i_size + shift; - - ret = write_invalidate_inode_pages_range(mapping, offset, LLONG_MAX); + ret = bch2_write_invalidate_inode_pages_range(mapping, offset, LLONG_MAX); if (ret) - goto err; - - if (insert) { - i_size_write(&inode->v, new_size); - mutex_lock(&inode->ei_update_lock); - ret = bch2_write_inode_size(c, inode, new_size, - ATTR_MTIME|ATTR_CTIME); - mutex_unlock(&inode->ei_update_lock); - } else { - s64 i_sectors_delta = 0; - - ret = bch2_fpunch(c, inode->v.i_ino, - offset >> 9, (offset + len) >> 9, - &inode->ei_journal_seq, - &i_sectors_delta); - i_sectors_acct(c, inode, NULL, i_sectors_delta); - - if (ret) - goto err; - } - - bch2_bkey_buf_init(©); - bch2_trans_init(&trans, c, BTREE_ITER_MAX, 1024); - src = bch2_trans_get_iter(&trans, BTREE_ID_extents, - POS(inode->v.i_ino, src_start >> 9), - BTREE_ITER_INTENT); - dst = bch2_trans_copy_iter(&trans, src); - del = bch2_trans_copy_iter(&trans, src); - - while (ret == 0 || ret == -EINTR) { - struct disk_reservation disk_res = - bch2_disk_reservation_init(c, 0); - struct bkey_i delete; - struct bkey_s_c k; - struct bpos next_pos; - struct bpos move_pos = POS(inode->v.i_ino, offset >> 9); - struct bpos atomic_end; - unsigned trigger_flags = 0; - - k = insert - ? bch2_btree_iter_peek_prev(src) - : bch2_btree_iter_peek(src); - if ((ret = bkey_err(k))) - continue; - - if (!k.k || k.k->p.inode != inode->v.i_ino) - break; - - if (insert && - bkey_cmp(k.k->p, POS(inode->v.i_ino, offset >> 9)) <= 0) - break; -reassemble: - bch2_bkey_buf_reassemble(©, c, k); - - if (insert && - bkey_cmp(bkey_start_pos(k.k), move_pos) < 0) - bch2_cut_front(move_pos, copy.k); - - copy.k->k.p.offset += shift >> 9; - bch2_btree_iter_set_pos(dst, bkey_start_pos(©.k->k)); - - ret = bch2_extent_atomic_end(dst, copy.k, &atomic_end); - if (ret) - continue; - - if (bkey_cmp(atomic_end, copy.k->k.p)) { - if (insert) { - move_pos = atomic_end; - move_pos.offset -= shift >> 9; - goto reassemble; - } else { - bch2_cut_back(atomic_end, copy.k); - } - } - - bkey_init(&delete.k); - delete.k.p = copy.k->k.p; - delete.k.size = copy.k->k.size; - delete.k.p.offset -= shift >> 9; - bch2_btree_iter_set_pos(del, bkey_start_pos(&delete.k)); - - 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 { - /* We might end up splitting compressed extents: */ - unsigned nr_ptrs = - bch2_bkey_nr_ptrs_allocated(bkey_i_to_s_c(copy.k)); - - ret = bch2_disk_reservation_get(c, &disk_res, - copy.k->k.size, nr_ptrs, - BCH_DISK_RESERVATION_NOFAIL); - BUG_ON(ret); - } - - ret = bch2_btree_iter_traverse(del) ?: - bch2_trans_update(&trans, del, &delete, trigger_flags) ?: - bch2_trans_update(&trans, dst, copy.k, trigger_flags) ?: - bch2_trans_commit(&trans, &disk_res, - &inode->ei_journal_seq, - BTREE_INSERT_NOFAIL); - bch2_disk_reservation_put(c, &disk_res); + return ret; - if (!ret) - bch2_btree_iter_set_pos(src, next_pos); - } - bch2_trans_iter_put(&trans, del); - bch2_trans_iter_put(&trans, dst); - bch2_trans_iter_put(&trans, src); - bch2_trans_exit(&trans); - bch2_bkey_buf_exit(©, c); + if (insert) + i_size_write(&inode->v, inode->v.i_size + len); - if (ret) - goto err; + ret = bch2_fcollapse_finsert(c, inode_inum(inode), offset >> 9, len >> 9, + insert, &i_sectors_delta); + if (!ret && !insert) + i_size_write(&inode->v, inode->v.i_size - len); + bch2_i_sectors_acct(c, inode, NULL, i_sectors_delta); - if (!insert) { - i_size_write(&inode->v, new_size); - mutex_lock(&inode->ei_update_lock); - ret = bch2_write_inode_size(c, inode, new_size, - ATTR_MTIME|ATTR_CTIME); - mutex_unlock(&inode->ei_update_lock); - } -err: - bch2_pagecache_block_put(&inode->ei_pagecache_lock); - inode_unlock(&inode->v); return ret; } @@ -2653,163 +584,175 @@ static int __bchfs_fallocate(struct bch_inode_info *inode, int mode, u64 start_sector, u64 end_sector) { struct bch_fs *c = inode->v.i_sb->s_fs_info; - struct btree_trans trans; - struct btree_iter *iter; + struct btree_trans *trans = bch2_trans_get(c); + 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_trans_init(&trans, c, BTREE_ITER_MAX, 512); + bch2_inode_opts_get(&opts, c, &inode->ei_inode); - iter = bch2_trans_get_iter(&trans, BTREE_ID_extents, + 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; + bool is_allocation; + u64 hole_start, hole_end; + u32 snapshot; + + bch2_trans_begin(trans); + + ret = bch2_subvolume_get_snapshot(trans, + inode->ei_subvol, &snapshot); + if (ret) + goto bkey_err; - bch2_trans_begin(&trans); + bch2_btree_iter_set_snapshot(&iter, snapshot); - k = bch2_btree_iter_peek_slot(iter); + k = bch2_btree_iter_peek_slot(&iter); if ((ret = bkey_err(k))) goto bkey_err; + hole_start = iter.pos.offset; + hole_end = bpos_min(k.k->p, end_pos).offset; + is_allocation = bkey_extent_is_allocation(k.k); + /* already reserved */ - if (k.k->type == KEY_TYPE_reservation && - bkey_s_c_to_reservation(k).v->nr_replicas >= replicas) { - bch2_btree_iter_next_slot(iter); + if (bkey_extent_is_reservation(k) && + bch2_bkey_nr_ptrs_fully_allocated(k) >= opts.data_replicas) { + bch2_btree_iter_advance(&iter); continue; } if (bkey_extent_is_data(k.k) && !(mode & FALLOC_FL_ZERO_RANGE)) { - bch2_btree_iter_next_slot(iter); + bch2_btree_iter_advance(&iter); 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; + if (!(mode & FALLOC_FL_ZERO_RANGE)) { + /* + * Lock ordering - can't be holding btree locks while + * blocking on a folio lock: + */ + if (bch2_clamp_data_hole(&inode->v, + &hole_start, + &hole_end, + opts.data_replicas, true)) + ret = drop_locks_do(trans, + (bch2_clamp_data_hole(&inode->v, + &hole_start, + &hole_end, + opts.data_replicas, false), 0)); + bch2_btree_iter_set_pos(&iter, POS(iter.pos.inode, hole_start)); + + if (ret) + goto bkey_err; - bch2_cut_front(iter->pos, &reservation.k_i); - bch2_cut_back(end_pos, &reservation.k_i); + if (hole_start == hole_end) + continue; + } - sectors = reservation.k.size; - reservation.v.nr_replicas = bch2_bkey_nr_ptrs_allocated(k); + sectors = hole_end - hole_start; - if (!bkey_extent_is_allocation(k.k)) { + if (!is_allocation) { ret = bch2_quota_reservation_add(c, inode, - "a_res, - sectors, true); + "a_res, sectors, true); if (unlikely(ret)) 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; + ret = bch2_extent_fallocate(trans, inode_inum(inode), &iter, + sectors, opts, &i_sectors_delta, + writepoint_hashed((unsigned long) current)); + if (ret) + goto bkey_err; - reservation.v.nr_replicas = disk_res.nr_replicas; - } + bch2_i_sectors_acct(c, inode, "a_res, i_sectors_delta); - ret = bch2_extent_update(&trans, iter, &reservation.k_i, - &disk_res, &inode->ei_journal_seq, - 0, &i_sectors_delta, true); - i_sectors_acct(c, inode, "a_res, i_sectors_delta); + drop_locks_do(trans, + (bch2_mark_pagecache_reserved(inode, hole_start, iter.pos.offset), 0)); 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_iter_put(&trans, iter); - bch2_trans_exit(&trans); + + if (bch2_err_matches(ret, ENOSPC) && (mode & FALLOC_FL_ZERO_RANGE)) { + struct quota_res quota_res = { 0 }; + s64 i_sectors_delta = 0; + + bch2_fpunch_at(trans, &iter, inode_inum(inode), + end_sector, &i_sectors_delta); + bch2_i_sectors_acct(c, inode, "a_res, i_sectors_delta); + bch2_quota_reservation_put(c, inode, "a_res); + } + + bch2_trans_iter_exit(trans, &iter); + bch2_trans_put(trans); return ret; } static long bchfs_fallocate(struct bch_inode_info *inode, int mode, loff_t offset, loff_t len) { - struct address_space *mapping = inode->v.i_mapping; struct bch_fs *c = inode->v.i_sb->s_fs_info; - loff_t end = offset + len; - loff_t block_start = round_down(offset, block_bytes(c)); - loff_t block_end = round_up(end, block_bytes(c)); - int ret; - - inode_lock(&inode->v); - inode_dio_wait(&inode->v); - bch2_pagecache_block_get(&inode->ei_pagecache_lock); + u64 end = offset + len; + u64 block_start = round_down(offset, block_bytes(c)); + u64 block_end = round_up(end, block_bytes(c)); + bool truncated_last_page = false; + int ret, ret2 = 0; if (!(mode & FALLOC_FL_KEEP_SIZE) && end > inode->v.i_size) { ret = inode_newsize_ok(&inode->v, end); if (ret) - goto err; + return ret; } if (mode & FALLOC_FL_ZERO_RANGE) { - ret = __bch2_truncate_page(inode, - offset >> PAGE_SHIFT, - offset, end); - - if (!ret && - offset >> PAGE_SHIFT != end >> PAGE_SHIFT) - ret = __bch2_truncate_page(inode, - end >> PAGE_SHIFT, - offset, end); + ret = bch2_truncate_folios(inode, offset, end); + if (unlikely(ret < 0)) + return ret; - if (unlikely(ret)) - goto err; + truncated_last_page = ret; truncate_pagecache_range(&inode->v, offset, end - 1); + + block_start = round_up(offset, block_bytes(c)); + block_end = round_down(end, block_bytes(c)); } ret = __bchfs_fallocate(inode, mode, block_start >> 9, block_end >> 9); - if (ret) - goto err; /* - * Do we need to extend the file? - * - * If we zeroed up to the end of the file, we dropped whatever writes - * were going to write out the current i_size, so we have to extend - * manually even if FL_KEEP_SIZE was set: + * On -ENOSPC in ZERO_RANGE mode, we still want to do the inode update, + * so that the VFS cache i_size is consistent with the btree i_size: */ - if (end >= inode->v.i_size && - (!(mode & FALLOC_FL_KEEP_SIZE) || - (mode & FALLOC_FL_ZERO_RANGE))) { + if (ret && + !(bch2_err_matches(ret, ENOSPC) && (mode & FALLOC_FL_ZERO_RANGE))) + return ret; - /* - * Sync existing appends before extending i_size, - * as in bch2_extend(): - */ - ret = filemap_write_and_wait_range(mapping, - inode->ei_inode.bi_size, S64_MAX); - if (ret) - goto err; + if (mode & FALLOC_FL_KEEP_SIZE && end > inode->v.i_size) + end = inode->v.i_size; - if (mode & FALLOC_FL_KEEP_SIZE) - end = inode->v.i_size; - else - i_size_write(&inode->v, end); + if (end >= inode->v.i_size && + (((mode & FALLOC_FL_ZERO_RANGE) && !truncated_last_page) || + !(mode & FALLOC_FL_KEEP_SIZE))) { + spin_lock(&inode->v.i_lock); + i_size_write(&inode->v, end); + spin_unlock(&inode->v.i_lock); mutex_lock(&inode->ei_update_lock); - ret = bch2_write_inode_size(c, inode, end, 0); + ret2 = bch2_write_inode_size(c, inode, end, 0); mutex_unlock(&inode->ei_update_lock); } -err: - bch2_pagecache_block_put(&inode->ei_pagecache_lock); - inode_unlock(&inode->v); - return ret; + + return ret ?: ret2; } long bch2_fallocate_dispatch(struct file *file, int mode, @@ -2819,9 +762,17 @@ 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); + + 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); else if (mode == (FALLOC_FL_PUNCH_HOLE|FALLOC_FL_KEEP_SIZE)) @@ -2832,47 +783,60 @@ long bch2_fallocate_dispatch(struct file *file, int mode, ret = bchfs_fcollapse_finsert(inode, offset, len, false); else ret = -EOPNOTSUPP; +err: + bch2_pagecache_block_put(inode); + inode_unlock(&inode->v); + bch2_write_ref_put(c, BCH_WRITE_REF_fallocate); - percpu_ref_put(&c->writes); - - return ret; + return bch2_err_class(ret); } -static void mark_range_unallocated(struct bch_inode_info *inode, - loff_t start, loff_t end) +/* + * 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) { - pgoff_t index = start >> PAGE_SHIFT; - pgoff_t end_index = (end - 1) >> PAGE_SHIFT; - struct pagevec pvec; - - pagevec_init(&pvec); - - do { - unsigned nr_pages, i, j; - - nr_pages = pagevec_lookup_range(&pvec, inode->v.i_mapping, - &index, end_index); - if (nr_pages == 0) - break; + struct bch_fs *c = inode->v.i_sb->s_fs_info; + struct btree_trans *trans = bch2_trans_get(c); + struct btree_iter iter; + struct bkey_s_c k; + u32 snapshot; + u64 sectors = end - start; + u64 pos = start; + int ret; +retry: + bch2_trans_begin(trans); - for (i = 0; i < nr_pages; i++) { - struct page *page = pvec.pages[i]; - struct bch_page_state *s; + ret = bch2_subvolume_get_snapshot(trans, inode->ei_subvol, &snapshot); + if (ret) + goto err; - lock_page(page); - s = bch2_page_state(page); + 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; - if (s) { - spin_lock(&s->lock); - for (j = 0; j < PAGE_SECTORS; j++) - s->s[j].nr_replicas = 0; - spin_unlock(&s->lock); - } + bch2_trans_put(trans); - unlock_page(page); - } - pagevec_release(&pvec); - } while (index <= end_index); + return ret ?: bch2_quota_reservation_add(c, inode, res, sectors, true); } loff_t bch2_remap_file_range(struct file *file_src, loff_t pos_src, @@ -2882,6 +846,7 @@ 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; @@ -2900,9 +865,8 @@ loff_t bch2_remap_file_range(struct file *file_src, loff_t pos_src, abs(pos_src - pos_dst) < len) return -EINVAL; - bch2_lock_inodes(INODE_LOCK|INODE_PAGECACHE_BLOCK, src, dst); - - file_update_time(file_dst); + lock_two_nondirectories(&src->v, &dst->v); + bch2_lock_inodes(INODE_PAGECACHE_BLOCK, src, dst); inode_dio_wait(&src->v); inode_dio_wait(&dst->v); @@ -2915,18 +879,25 @@ loff_t bch2_remap_file_range(struct file *file_src, loff_t pos_src, aligned_len = round_up((u64) len, block_bytes(c)); - ret = write_invalidate_inode_pages_range(dst->v.i_mapping, + ret = bch2_write_invalidate_inode_pages_range(dst->v.i_mapping, pos_dst, pos_dst + len - 1); if (ret) goto err; - mark_range_unallocated(src, pos_src, pos_src + aligned_len); + ret = quota_reserve_range(dst, "a_res, pos_dst >> 9, + (pos_dst + aligned_len) >> 9); + if (ret) + goto err; + + file_update_time(file_dst); + + bch2_mark_pagecache_unallocated(src, pos_src >> 9, + (pos_src + aligned_len) >> 9); ret = bch2_remap_range(c, - POS(dst->v.i_ino, pos_dst >> 9), - POS(src->v.i_ino, pos_src >> 9), + inode_inum(dst), pos_dst >> 9, + inode_inum(src), pos_src >> 9, aligned_len >> 9, - &dst->ei_journal_seq, pos_dst + len, &i_sectors_delta); if (ret < 0) goto err; @@ -2936,113 +907,72 @@ 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); + bch2_i_sectors_acct(c, dst, "a_res, i_sectors_delta); spin_lock(&dst->v.i_lock); if (pos_dst + ret > dst->v.i_size) i_size_write(&dst->v, pos_dst + ret); spin_unlock(&dst->v.i_lock); - if (((file_dst->f_flags & (__O_SYNC | O_DSYNC)) || - IS_SYNC(file_inode(file_dst))) && - !c->opts.journal_flush_disabled) - ret = bch2_journal_flush_seq(&c->journal, dst->ei_journal_seq); + if ((file_dst->f_flags & (__O_SYNC | O_DSYNC)) || + IS_SYNC(file_inode(file_dst))) + ret = bch2_flush_inode(c, dst); err: - bch2_unlock_inodes(INODE_LOCK|INODE_PAGECACHE_BLOCK, src, dst); + bch2_quota_reservation_put(c, dst, "a_res); + bch2_unlock_inodes(INODE_PAGECACHE_BLOCK, src, dst); + unlock_two_nondirectories(&src->v, &dst->v); - return ret; + return bch2_err_class(ret); } /* fseek: */ -static int page_data_offset(struct page *page, unsigned offset) -{ - struct bch_page_state *s = bch2_page_state(page); - unsigned i; - - if (s) - for (i = offset >> 9; i < PAGE_SECTORS; i++) - if (s->s[i].state >= SECTOR_DIRTY) - return i << 9; - - return -1; -} - -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; - pgoff_t start_index = start_offset >> PAGE_SHIFT; - pgoff_t end_index = end_offset >> PAGE_SHIFT; - pgoff_t index = start_index; - loff_t ret; - int offset; - - while (index <= end_index) { - if (find_get_pages_range(mapping, &index, end_index, 1, &page)) { - lock_page(page); - - offset = page_data_offset(page, - page->index == start_index - ? start_offset & (PAGE_SIZE - 1) - : 0); - if (offset >= 0) { - ret = clamp(((loff_t) page->index << PAGE_SHIFT) + - offset, - start_offset, end_offset); - unlock_page(page); - put_page(page); - return ret; - } - - unlock_page(page); - put_page(page); - } else { - break; - } - } - - return end_offset; -} - static loff_t bch2_seek_data(struct file *file, u64 offset) { struct bch_inode_info *inode = file_bch_inode(file); struct bch_fs *c = inode->v.i_sb->s_fs_info; - struct btree_trans trans; - struct btree_iter *iter; + struct btree_trans *trans; + struct btree_iter iter; struct bkey_s_c k; + subvol_inum inum = inode_inum(inode); u64 isize, next_data = MAX_LFS_FILESIZE; + u32 snapshot; int ret; isize = i_size_read(&inode->v); if (offset >= isize) return -ENXIO; - bch2_trans_init(&trans, c, 0, 0); + trans = bch2_trans_get(c); +retry: + bch2_trans_begin(trans); - for_each_btree_key(&trans, iter, BTREE_ID_extents, - POS(inode->v.i_ino, offset >> 9), 0, k, ret) { - if (k.k->p.inode != inode->v.i_ino) { - break; - } else if (bkey_extent_is_data(k.k)) { + ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot); + if (ret) + goto err; + + 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) break; } - bch2_trans_iter_put(&trans, iter); + bch2_trans_iter_exit(trans, &iter); +err: + if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) + goto retry; - ret = bch2_trans_exit(&trans) ?: ret; + bch2_trans_put(trans); if (ret) return ret; if (next_data > offset) next_data = bch2_seek_pagecache_data(&inode->v, - offset, next_data); + offset, next_data, 0, false); if (next_data >= isize) return -ENXIO; @@ -3050,87 +980,41 @@ static loff_t bch2_seek_data(struct file *file, u64 offset) return vfs_setpos(file, next_data, MAX_LFS_FILESIZE); } -static int __page_hole_offset(struct page *page, unsigned offset) -{ - struct bch_page_state *s = bch2_page_state(page); - unsigned i; - - if (!s) - return 0; - - for (i = offset >> 9; i < PAGE_SECTORS; i++) - if (s->s[i].state < SECTOR_DIRTY) - return i << 9; - - return -1; -} - -static loff_t page_hole_offset(struct address_space *mapping, loff_t offset) -{ - pgoff_t index = offset >> PAGE_SHIFT; - struct page *page; - int pg_offset; - loff_t ret = -1; - - page = find_lock_page(mapping, index); - if (!page) - return offset; - - pg_offset = __page_hole_offset(page, offset & (PAGE_SIZE - 1)); - if (pg_offset >= 0) - ret = ((loff_t) index << PAGE_SHIFT) + pg_offset; - - unlock_page(page); - - return ret; -} - -static loff_t bch2_seek_pagecache_hole(struct inode *vinode, - loff_t start_offset, - loff_t end_offset) -{ - struct address_space *mapping = vinode->i_mapping; - loff_t offset = start_offset, hole; - - while (offset < end_offset) { - hole = page_hole_offset(mapping, offset); - if (hole >= 0 && hole <= end_offset) - return max(start_offset, hole); - - offset += PAGE_SIZE; - offset &= PAGE_MASK; - } - - return end_offset; -} - static loff_t bch2_seek_hole(struct file *file, u64 offset) { struct bch_inode_info *inode = file_bch_inode(file); struct bch_fs *c = inode->v.i_sb->s_fs_info; - struct btree_trans trans; - struct btree_iter *iter; + struct btree_trans *trans; + struct btree_iter iter; struct bkey_s_c k; + subvol_inum inum = inode_inum(inode); u64 isize, next_hole = MAX_LFS_FILESIZE; + u32 snapshot; int ret; isize = i_size_read(&inode->v); if (offset >= isize) return -ENXIO; - bch2_trans_init(&trans, c, 0, 0); + trans = bch2_trans_get(c); +retry: + bch2_trans_begin(trans); + + ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot); + if (ret) + goto err; - for_each_btree_key(&trans, iter, BTREE_ID_extents, - POS(inode->v.i_ino, offset >> 9), + for_each_btree_key_norestart(trans, iter, BTREE_ID_extents, + SPOS(inode->v.i_ino, offset >> 9, snapshot), BTREE_ITER_SLOTS, k, ret) { if (k.k->p.inode != inode->v.i_ino) { next_hole = bch2_seek_pagecache_hole(&inode->v, - offset, MAX_LFS_FILESIZE); + offset, MAX_LFS_FILESIZE, 0, false); break; } else if (!bkey_extent_is_data(k.k)) { next_hole = bch2_seek_pagecache_hole(&inode->v, max(offset, bkey_start_offset(k.k) << 9), - k.k->p.offset << 9); + k.k->p.offset << 9, 0, false); if (next_hole < k.k->p.offset << 9) break; @@ -3138,9 +1022,12 @@ static loff_t bch2_seek_hole(struct file *file, u64 offset) offset = max(offset, bkey_start_offset(k.k) << 9); } } - bch2_trans_iter_put(&trans, iter); + bch2_trans_iter_exit(trans, &iter); +err: + if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) + goto retry; - ret = bch2_trans_exit(&trans) ?: ret; + bch2_trans_put(trans); if (ret) return ret; @@ -3152,46 +1039,40 @@ static loff_t bch2_seek_hole(struct file *file, u64 offset) 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->dio_write_bioset); - bioset_exit(&c->dio_read_bioset); - bioset_exit(&c->writepage_bioset); + bioset_exit(&c->nocow_flush_bioset); } int bch2_fs_fsio_init(struct bch_fs *c) { - int ret = 0; + 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, ""); - - if (bioset_init(&c->writepage_bioset, - 4, offsetof(struct bch_writepage_io, op.wbio.bio), - BIOSET_NEED_BVECS) || - bioset_init(&c->dio_read_bioset, - 4, offsetof(struct dio_read, rbio.bio), - BIOSET_NEED_BVECS) || - bioset_init(&c->dio_write_bioset, - 4, offsetof(struct dio_write, op.wbio.bio), - BIOSET_NEED_BVECS)) - ret = -ENOMEM; - - pr_verbose_init(c->opts, "ret %i", ret); - return ret; + return 0; } #endif /* NO_BCACHEFS_FS */