11c6fdbd8SKent Overstreet // SPDX-License-Identifier: GPL-2.0 21c6fdbd8SKent Overstreet #ifndef NO_BCACHEFS_FS 31c6fdbd8SKent Overstreet 41c6fdbd8SKent Overstreet #include "bcachefs.h" 57b3f84eaSKent Overstreet #include "alloc_foreground.h" 61c6fdbd8SKent Overstreet #include "btree_update.h" 71c6fdbd8SKent Overstreet #include "buckets.h" 81c6fdbd8SKent Overstreet #include "clock.h" 91c6fdbd8SKent Overstreet #include "error.h" 10e2d9912cSKent Overstreet #include "extents.h" 111c6fdbd8SKent Overstreet #include "fs.h" 121c6fdbd8SKent Overstreet #include "fs-io.h" 131c6fdbd8SKent Overstreet #include "fsck.h" 141c6fdbd8SKent Overstreet #include "inode.h" 151c6fdbd8SKent Overstreet #include "journal.h" 161c6fdbd8SKent Overstreet #include "io.h" 171c6fdbd8SKent Overstreet #include "keylist.h" 181c6fdbd8SKent Overstreet #include "quota.h" 191c6fdbd8SKent Overstreet #include "trace.h" 201c6fdbd8SKent Overstreet 211c6fdbd8SKent Overstreet #include <linux/aio.h> 221c6fdbd8SKent Overstreet #include <linux/backing-dev.h> 231c6fdbd8SKent Overstreet #include <linux/falloc.h> 241c6fdbd8SKent Overstreet #include <linux/migrate.h> 251c6fdbd8SKent Overstreet #include <linux/mmu_context.h> 261c6fdbd8SKent Overstreet #include <linux/pagevec.h> 271c6fdbd8SKent Overstreet #include <linux/sched/signal.h> 281c6fdbd8SKent Overstreet #include <linux/task_io_accounting_ops.h> 291c6fdbd8SKent Overstreet #include <linux/uio.h> 301c6fdbd8SKent Overstreet #include <linux/writeback.h> 311c6fdbd8SKent Overstreet 321c6fdbd8SKent Overstreet #include <trace/events/writeback.h> 331c6fdbd8SKent Overstreet 341c6fdbd8SKent Overstreet struct quota_res { 351c6fdbd8SKent Overstreet u64 sectors; 361c6fdbd8SKent Overstreet }; 371c6fdbd8SKent Overstreet 381c6fdbd8SKent Overstreet struct bchfs_write_op { 391c6fdbd8SKent Overstreet struct bch_inode_info *inode; 401c6fdbd8SKent Overstreet s64 sectors_added; 411c6fdbd8SKent Overstreet bool is_dio; 421c6fdbd8SKent Overstreet bool unalloc; 431c6fdbd8SKent Overstreet u64 new_i_size; 441c6fdbd8SKent Overstreet 451c6fdbd8SKent Overstreet /* must be last: */ 461c6fdbd8SKent Overstreet struct bch_write_op op; 471c6fdbd8SKent Overstreet }; 481c6fdbd8SKent Overstreet 491c6fdbd8SKent Overstreet struct bch_writepage_io { 501c6fdbd8SKent Overstreet struct closure cl; 511c6fdbd8SKent Overstreet u64 new_sectors; 521c6fdbd8SKent Overstreet 531c6fdbd8SKent Overstreet /* must be last: */ 541c6fdbd8SKent Overstreet struct bchfs_write_op op; 551c6fdbd8SKent Overstreet }; 561c6fdbd8SKent Overstreet 571c6fdbd8SKent Overstreet struct dio_write { 581c6fdbd8SKent Overstreet struct closure cl; 591c6fdbd8SKent Overstreet struct kiocb *req; 60ed484030SKent Overstreet struct mm_struct *mm; 611c6fdbd8SKent Overstreet unsigned loop:1, 621c6fdbd8SKent Overstreet sync:1, 631c6fdbd8SKent Overstreet free_iov:1; 641c6fdbd8SKent Overstreet struct quota_res quota_res; 651c6fdbd8SKent Overstreet 661c6fdbd8SKent Overstreet struct iov_iter iter; 671c6fdbd8SKent Overstreet struct iovec inline_vecs[2]; 681c6fdbd8SKent Overstreet 691c6fdbd8SKent Overstreet /* must be last: */ 701c6fdbd8SKent Overstreet struct bchfs_write_op iop; 711c6fdbd8SKent Overstreet }; 721c6fdbd8SKent Overstreet 731c6fdbd8SKent Overstreet struct dio_read { 741c6fdbd8SKent Overstreet struct closure cl; 751c6fdbd8SKent Overstreet struct kiocb *req; 761c6fdbd8SKent Overstreet long ret; 771c6fdbd8SKent Overstreet struct bch_read_bio rbio; 781c6fdbd8SKent Overstreet }; 791c6fdbd8SKent Overstreet 801c6fdbd8SKent Overstreet /* pagecache_block must be held */ 811c6fdbd8SKent Overstreet static int write_invalidate_inode_pages_range(struct address_space *mapping, 821c6fdbd8SKent Overstreet loff_t start, loff_t end) 831c6fdbd8SKent Overstreet { 841c6fdbd8SKent Overstreet int ret; 851c6fdbd8SKent Overstreet 861c6fdbd8SKent Overstreet /* 871c6fdbd8SKent Overstreet * XXX: the way this is currently implemented, we can spin if a process 881c6fdbd8SKent Overstreet * is continually redirtying a specific page 891c6fdbd8SKent Overstreet */ 901c6fdbd8SKent Overstreet do { 911c6fdbd8SKent Overstreet if (!mapping->nrpages) 921c6fdbd8SKent Overstreet return 0; 931c6fdbd8SKent Overstreet 941c6fdbd8SKent Overstreet ret = filemap_write_and_wait_range(mapping, start, end); 951c6fdbd8SKent Overstreet if (ret) 961c6fdbd8SKent Overstreet break; 971c6fdbd8SKent Overstreet 981c6fdbd8SKent Overstreet if (!mapping->nrpages) 991c6fdbd8SKent Overstreet return 0; 1001c6fdbd8SKent Overstreet 1011c6fdbd8SKent Overstreet ret = invalidate_inode_pages2_range(mapping, 1021c6fdbd8SKent Overstreet start >> PAGE_SHIFT, 1031c6fdbd8SKent Overstreet end >> PAGE_SHIFT); 1041c6fdbd8SKent Overstreet } while (ret == -EBUSY); 1051c6fdbd8SKent Overstreet 1061c6fdbd8SKent Overstreet return ret; 1071c6fdbd8SKent Overstreet } 1081c6fdbd8SKent Overstreet 1091c6fdbd8SKent Overstreet /* quotas */ 1101c6fdbd8SKent Overstreet 1111c6fdbd8SKent Overstreet #ifdef CONFIG_BCACHEFS_QUOTA 1121c6fdbd8SKent Overstreet 1131c6fdbd8SKent Overstreet static void bch2_quota_reservation_put(struct bch_fs *c, 1141c6fdbd8SKent Overstreet struct bch_inode_info *inode, 1151c6fdbd8SKent Overstreet struct quota_res *res) 1161c6fdbd8SKent Overstreet { 1171c6fdbd8SKent Overstreet if (!res->sectors) 1181c6fdbd8SKent Overstreet return; 1191c6fdbd8SKent Overstreet 1201c6fdbd8SKent Overstreet mutex_lock(&inode->ei_quota_lock); 1211c6fdbd8SKent Overstreet BUG_ON(res->sectors > inode->ei_quota_reserved); 1221c6fdbd8SKent Overstreet 1231c6fdbd8SKent Overstreet bch2_quota_acct(c, inode->ei_qid, Q_SPC, 12426609b61SKent Overstreet -((s64) res->sectors), KEY_TYPE_QUOTA_PREALLOC); 1251c6fdbd8SKent Overstreet inode->ei_quota_reserved -= res->sectors; 1261c6fdbd8SKent Overstreet mutex_unlock(&inode->ei_quota_lock); 1271c6fdbd8SKent Overstreet 1281c6fdbd8SKent Overstreet res->sectors = 0; 1291c6fdbd8SKent Overstreet } 1301c6fdbd8SKent Overstreet 1311c6fdbd8SKent Overstreet static int bch2_quota_reservation_add(struct bch_fs *c, 1321c6fdbd8SKent Overstreet struct bch_inode_info *inode, 1331c6fdbd8SKent Overstreet struct quota_res *res, 1341c6fdbd8SKent Overstreet unsigned sectors, 1351c6fdbd8SKent Overstreet bool check_enospc) 1361c6fdbd8SKent Overstreet { 1371c6fdbd8SKent Overstreet int ret; 1381c6fdbd8SKent Overstreet 1391c6fdbd8SKent Overstreet mutex_lock(&inode->ei_quota_lock); 1401c6fdbd8SKent Overstreet ret = bch2_quota_acct(c, inode->ei_qid, Q_SPC, sectors, 14126609b61SKent Overstreet check_enospc ? KEY_TYPE_QUOTA_PREALLOC : KEY_TYPE_QUOTA_NOCHECK); 1421c6fdbd8SKent Overstreet if (likely(!ret)) { 1431c6fdbd8SKent Overstreet inode->ei_quota_reserved += sectors; 1441c6fdbd8SKent Overstreet res->sectors += sectors; 1451c6fdbd8SKent Overstreet } 1461c6fdbd8SKent Overstreet mutex_unlock(&inode->ei_quota_lock); 1471c6fdbd8SKent Overstreet 1481c6fdbd8SKent Overstreet return ret; 1491c6fdbd8SKent Overstreet } 1501c6fdbd8SKent Overstreet 1511c6fdbd8SKent Overstreet #else 1521c6fdbd8SKent Overstreet 1531c6fdbd8SKent Overstreet static void bch2_quota_reservation_put(struct bch_fs *c, 1541c6fdbd8SKent Overstreet struct bch_inode_info *inode, 1551c6fdbd8SKent Overstreet struct quota_res *res) 1561c6fdbd8SKent Overstreet { 1571c6fdbd8SKent Overstreet } 1581c6fdbd8SKent Overstreet 1591c6fdbd8SKent Overstreet static int bch2_quota_reservation_add(struct bch_fs *c, 1601c6fdbd8SKent Overstreet struct bch_inode_info *inode, 1611c6fdbd8SKent Overstreet struct quota_res *res, 1621c6fdbd8SKent Overstreet unsigned sectors, 1631c6fdbd8SKent Overstreet bool check_enospc) 1641c6fdbd8SKent Overstreet { 1651c6fdbd8SKent Overstreet return 0; 1661c6fdbd8SKent Overstreet } 1671c6fdbd8SKent Overstreet 1681c6fdbd8SKent Overstreet #endif 1691c6fdbd8SKent Overstreet 1701c6fdbd8SKent Overstreet /* i_size updates: */ 1711c6fdbd8SKent Overstreet 1722ea90048SKent Overstreet struct inode_new_size { 1732ea90048SKent Overstreet loff_t new_size; 1742ea90048SKent Overstreet u64 now; 1752ea90048SKent Overstreet unsigned fields; 1762ea90048SKent Overstreet }; 1772ea90048SKent Overstreet 1781c6fdbd8SKent Overstreet static int inode_set_size(struct bch_inode_info *inode, 1791c6fdbd8SKent Overstreet struct bch_inode_unpacked *bi, 1801c6fdbd8SKent Overstreet void *p) 1811c6fdbd8SKent Overstreet { 1822ea90048SKent Overstreet struct inode_new_size *s = p; 1831c6fdbd8SKent Overstreet 1842ea90048SKent Overstreet bi->bi_size = s->new_size; 1852ea90048SKent Overstreet if (s->fields & ATTR_ATIME) 1862ea90048SKent Overstreet bi->bi_atime = s->now; 1872ea90048SKent Overstreet if (s->fields & ATTR_MTIME) 1882ea90048SKent Overstreet bi->bi_mtime = s->now; 1892ea90048SKent Overstreet if (s->fields & ATTR_CTIME) 1902ea90048SKent Overstreet bi->bi_ctime = s->now; 1911c6fdbd8SKent Overstreet 1921c6fdbd8SKent Overstreet return 0; 1931c6fdbd8SKent Overstreet } 1941c6fdbd8SKent Overstreet 1951c6fdbd8SKent Overstreet static int __must_check bch2_write_inode_size(struct bch_fs *c, 1961c6fdbd8SKent Overstreet struct bch_inode_info *inode, 1972ea90048SKent Overstreet loff_t new_size, unsigned fields) 1981c6fdbd8SKent Overstreet { 1992ea90048SKent Overstreet struct inode_new_size s = { 2002ea90048SKent Overstreet .new_size = new_size, 2012ea90048SKent Overstreet .now = bch2_current_time(c), 2022ea90048SKent Overstreet .fields = fields, 2032ea90048SKent Overstreet }; 2042ea90048SKent Overstreet 2052ea90048SKent Overstreet return bch2_write_inode(c, inode, inode_set_size, &s, fields); 2061c6fdbd8SKent Overstreet } 2071c6fdbd8SKent Overstreet 2081c6fdbd8SKent Overstreet static void i_sectors_acct(struct bch_fs *c, struct bch_inode_info *inode, 209190fa7afSKent Overstreet struct quota_res *quota_res, s64 sectors) 2101c6fdbd8SKent Overstreet { 211190fa7afSKent Overstreet if (!sectors) 212190fa7afSKent Overstreet return; 213190fa7afSKent Overstreet 2141c6fdbd8SKent Overstreet mutex_lock(&inode->ei_quota_lock); 2151c6fdbd8SKent Overstreet #ifdef CONFIG_BCACHEFS_QUOTA 2161c6fdbd8SKent Overstreet if (quota_res && sectors > 0) { 2171c6fdbd8SKent Overstreet BUG_ON(sectors > quota_res->sectors); 2181c6fdbd8SKent Overstreet BUG_ON(sectors > inode->ei_quota_reserved); 2191c6fdbd8SKent Overstreet 2201c6fdbd8SKent Overstreet quota_res->sectors -= sectors; 2211c6fdbd8SKent Overstreet inode->ei_quota_reserved -= sectors; 2221c6fdbd8SKent Overstreet } else { 22326609b61SKent Overstreet bch2_quota_acct(c, inode->ei_qid, Q_SPC, sectors, KEY_TYPE_QUOTA_WARN); 2241c6fdbd8SKent Overstreet } 2251c6fdbd8SKent Overstreet #endif 2261c6fdbd8SKent Overstreet inode->v.i_blocks += sectors; 2271c6fdbd8SKent Overstreet mutex_unlock(&inode->ei_quota_lock); 2281c6fdbd8SKent Overstreet } 2291c6fdbd8SKent Overstreet 2301c6fdbd8SKent Overstreet /* normal i_size/i_sectors update machinery: */ 2311c6fdbd8SKent Overstreet 232446c562cSKent Overstreet static int sum_sector_overwrites(struct btree_trans *trans, 233446c562cSKent Overstreet struct btree_iter *extent_iter, 234446c562cSKent Overstreet struct bkey_i *new, bool *allocating, 235446c562cSKent Overstreet s64 *delta) 23608af47dfSKent Overstreet { 237446c562cSKent Overstreet struct btree_iter *iter; 23808af47dfSKent Overstreet struct bkey_s_c old; 2391c6fdbd8SKent Overstreet 240446c562cSKent Overstreet *delta = 0; 24108af47dfSKent Overstreet 242446c562cSKent Overstreet iter = bch2_trans_copy_iter(trans, extent_iter); 243446c562cSKent Overstreet if (IS_ERR(iter)) 244446c562cSKent Overstreet return PTR_ERR(iter); 24508af47dfSKent Overstreet 246446c562cSKent Overstreet old = bch2_btree_iter_peek_slot(iter); 24701a0108fSKent Overstreet 24801a0108fSKent Overstreet while (1) { 24901a0108fSKent Overstreet /* 25001a0108fSKent Overstreet * should not be possible to get an error here, since we're 25101a0108fSKent Overstreet * carefully not advancing past @new and thus whatever leaf node 25201a0108fSKent Overstreet * @_iter currently points to: 25301a0108fSKent Overstreet */ 2540f238367SKent Overstreet BUG_ON(bkey_err(old)); 25508af47dfSKent Overstreet 25608af47dfSKent Overstreet if (allocating && 2577ef2a73aSKent Overstreet !*allocating && 2587ef2a73aSKent Overstreet bch2_bkey_nr_ptrs_allocated(old) < 2597ef2a73aSKent Overstreet bch2_bkey_nr_dirty_ptrs(bkey_i_to_s_c(new))) 26008af47dfSKent Overstreet *allocating = true; 26108af47dfSKent Overstreet 262446c562cSKent Overstreet *delta += (min(new->k.p.offset, 26308af47dfSKent Overstreet old.k->p.offset) - 26408af47dfSKent Overstreet max(bkey_start_offset(&new->k), 26508af47dfSKent Overstreet bkey_start_offset(old.k))) * 26608af47dfSKent Overstreet (bkey_extent_is_allocation(&new->k) - 26708af47dfSKent Overstreet bkey_extent_is_allocation(old.k)); 26801a0108fSKent Overstreet 26901a0108fSKent Overstreet if (bkey_cmp(old.k->p, new->k.p) >= 0) 27001a0108fSKent Overstreet break; 27101a0108fSKent Overstreet 272446c562cSKent Overstreet old = bch2_btree_iter_next_slot(iter); 27308af47dfSKent Overstreet } 27408af47dfSKent Overstreet 275446c562cSKent Overstreet bch2_trans_iter_free(trans, iter); 276446c562cSKent Overstreet return 0; 27708af47dfSKent Overstreet } 27808af47dfSKent Overstreet 279b17657d0SKent Overstreet int bch2_extent_update(struct btree_trans *trans, 28008af47dfSKent Overstreet struct bch_inode_info *inode, 28108af47dfSKent Overstreet struct disk_reservation *disk_res, 28208af47dfSKent Overstreet struct quota_res *quota_res, 28308af47dfSKent Overstreet struct btree_iter *extent_iter, 28408af47dfSKent Overstreet struct bkey_i *k, 28508af47dfSKent Overstreet u64 new_i_size, 28608af47dfSKent Overstreet bool may_allocate, 28708af47dfSKent Overstreet bool direct, 28808af47dfSKent Overstreet s64 *total_delta) 28908af47dfSKent Overstreet { 29061f321fcSKent Overstreet struct bch_fs *c = trans->c; 29161f321fcSKent Overstreet struct btree_iter *inode_iter = NULL; 2921c6fdbd8SKent Overstreet struct bch_inode_unpacked inode_u; 2931c6fdbd8SKent Overstreet struct bkey_inode_buf inode_p; 29408af47dfSKent Overstreet bool allocating = false; 29508af47dfSKent Overstreet bool extended = false; 2965154704bSKent Overstreet bool inode_locked = false; 29708af47dfSKent Overstreet s64 i_sectors_delta; 29808af47dfSKent Overstreet int ret; 2991c6fdbd8SKent Overstreet 30008af47dfSKent Overstreet ret = bch2_btree_iter_traverse(extent_iter); 30108af47dfSKent Overstreet if (ret) 30208af47dfSKent Overstreet return ret; 3031c6fdbd8SKent Overstreet 30408af47dfSKent Overstreet bch2_extent_trim_atomic(k, extent_iter); 3051c6fdbd8SKent Overstreet 306446c562cSKent Overstreet ret = sum_sector_overwrites(trans, extent_iter, 307446c562cSKent Overstreet k, &allocating, 308446c562cSKent Overstreet &i_sectors_delta); 309446c562cSKent Overstreet if (ret) 310446c562cSKent Overstreet return ret; 311446c562cSKent Overstreet 31208af47dfSKent Overstreet if (!may_allocate && allocating) 31308af47dfSKent Overstreet return -ENOSPC; 31408af47dfSKent Overstreet 31508af47dfSKent Overstreet bch2_trans_update(trans, BTREE_INSERT_ENTRY(extent_iter, k)); 31608af47dfSKent Overstreet 31708af47dfSKent Overstreet new_i_size = min(k->k.p.offset << 9, new_i_size); 3181c6fdbd8SKent Overstreet 3191c6fdbd8SKent Overstreet /* XXX: inode->i_size locking */ 32008af47dfSKent Overstreet if (i_sectors_delta || 32108af47dfSKent Overstreet new_i_size > inode->ei_inode.bi_size) { 32261f321fcSKent Overstreet if (c->opts.new_inode_updates) { 32358fbf808SKent Overstreet bch2_trans_unlock(trans); 3245154704bSKent Overstreet mutex_lock(&inode->ei_update_lock); 32508af47dfSKent Overstreet 32658fbf808SKent Overstreet if (!bch2_trans_relock(trans)) { 3275154704bSKent Overstreet mutex_unlock(&inode->ei_update_lock); 3285154704bSKent Overstreet return -EINTR; 3295154704bSKent Overstreet } 3305154704bSKent Overstreet 3315154704bSKent Overstreet inode_locked = true; 3325154704bSKent Overstreet 3335154704bSKent Overstreet if (!inode->ei_inode_update) 3345154704bSKent Overstreet inode->ei_inode_update = 33561f321fcSKent Overstreet bch2_deferred_update_alloc(c, 3365154704bSKent Overstreet BTREE_ID_INODES, 64); 33708af47dfSKent Overstreet 33808af47dfSKent Overstreet inode_u = inode->ei_inode; 33908af47dfSKent Overstreet inode_u.bi_sectors += i_sectors_delta; 34008af47dfSKent Overstreet 34108af47dfSKent Overstreet /* XXX: this is slightly suspect */ 34208af47dfSKent Overstreet if (!(inode_u.bi_flags & BCH_INODE_I_SIZE_DIRTY) && 34308af47dfSKent Overstreet new_i_size > inode_u.bi_size) { 34408af47dfSKent Overstreet inode_u.bi_size = new_i_size; 34508af47dfSKent Overstreet extended = true; 3461c6fdbd8SKent Overstreet } 3471c6fdbd8SKent Overstreet 34808af47dfSKent Overstreet bch2_inode_pack(&inode_p, &inode_u); 34908af47dfSKent Overstreet bch2_trans_update(trans, 3505154704bSKent Overstreet BTREE_INSERT_DEFERRED(inode->ei_inode_update, 3515154704bSKent Overstreet &inode_p.inode.k_i)); 35261f321fcSKent Overstreet } else { 35361f321fcSKent Overstreet inode_iter = bch2_trans_get_iter(trans, 35461f321fcSKent Overstreet BTREE_ID_INODES, 35561f321fcSKent Overstreet POS(k->k.p.inode, 0), 35661f321fcSKent Overstreet BTREE_ITER_SLOTS|BTREE_ITER_INTENT); 35761f321fcSKent Overstreet if (IS_ERR(inode_iter)) 35861f321fcSKent Overstreet return PTR_ERR(inode_iter); 35961f321fcSKent Overstreet 36061f321fcSKent Overstreet ret = bch2_btree_iter_traverse(inode_iter); 36161f321fcSKent Overstreet if (ret) 36261f321fcSKent Overstreet goto err; 36361f321fcSKent Overstreet 36461f321fcSKent Overstreet inode_u = inode->ei_inode; 36561f321fcSKent Overstreet inode_u.bi_sectors += i_sectors_delta; 36661f321fcSKent Overstreet 36761f321fcSKent Overstreet /* XXX: this is slightly suspect */ 36861f321fcSKent Overstreet if (!(inode_u.bi_flags & BCH_INODE_I_SIZE_DIRTY) && 36961f321fcSKent Overstreet new_i_size > inode_u.bi_size) { 37061f321fcSKent Overstreet inode_u.bi_size = new_i_size; 37161f321fcSKent Overstreet extended = true; 37261f321fcSKent Overstreet } 37361f321fcSKent Overstreet 37461f321fcSKent Overstreet bch2_inode_pack(&inode_p, &inode_u); 37561f321fcSKent Overstreet bch2_trans_update(trans, 37661f321fcSKent Overstreet BTREE_INSERT_ENTRY(inode_iter, &inode_p.inode.k_i)); 37761f321fcSKent Overstreet } 37808af47dfSKent Overstreet } 3791c6fdbd8SKent Overstreet 380fc3268c1SKent Overstreet ret = bch2_trans_commit(trans, disk_res, 38108af47dfSKent Overstreet &inode->ei_journal_seq, 38208af47dfSKent Overstreet BTREE_INSERT_NOFAIL| 38308af47dfSKent Overstreet BTREE_INSERT_ATOMIC| 38408af47dfSKent Overstreet BTREE_INSERT_NOUNLOCK| 38508af47dfSKent Overstreet BTREE_INSERT_USE_RESERVE); 38608af47dfSKent Overstreet if (ret) 38708af47dfSKent Overstreet goto err; 3881c6fdbd8SKent Overstreet 38908af47dfSKent Overstreet inode->ei_inode.bi_sectors += i_sectors_delta; 39008af47dfSKent Overstreet 39108af47dfSKent Overstreet EBUG_ON(i_sectors_delta && 39208af47dfSKent Overstreet inode->ei_inode.bi_sectors != inode_u.bi_sectors); 39308af47dfSKent Overstreet 39408af47dfSKent Overstreet if (extended) { 39508af47dfSKent Overstreet inode->ei_inode.bi_size = new_i_size; 39608af47dfSKent Overstreet 39708af47dfSKent Overstreet if (direct) { 3981c6fdbd8SKent Overstreet spin_lock(&inode->v.i_lock); 39908af47dfSKent Overstreet if (new_i_size > inode->v.i_size) 40008af47dfSKent Overstreet i_size_write(&inode->v, new_i_size); 4011c6fdbd8SKent Overstreet spin_unlock(&inode->v.i_lock); 4021c6fdbd8SKent Overstreet } 4031c6fdbd8SKent Overstreet } 4041c6fdbd8SKent Overstreet 40508af47dfSKent Overstreet if (direct) 40661f321fcSKent Overstreet i_sectors_acct(c, inode, quota_res, i_sectors_delta); 4071c6fdbd8SKent Overstreet 40808af47dfSKent Overstreet if (total_delta) 40908af47dfSKent Overstreet *total_delta += i_sectors_delta; 41008af47dfSKent Overstreet err: 41161f321fcSKent Overstreet if (!IS_ERR_OR_NULL(inode_iter)) 41261f321fcSKent Overstreet bch2_trans_iter_put(trans, inode_iter); 4135154704bSKent Overstreet if (inode_locked) 4145154704bSKent Overstreet mutex_unlock(&inode->ei_update_lock); 4155154704bSKent Overstreet 41608af47dfSKent Overstreet return ret; 4171c6fdbd8SKent Overstreet } 4181c6fdbd8SKent Overstreet 4191c6fdbd8SKent Overstreet static int bchfs_write_index_update(struct bch_write_op *wop) 4201c6fdbd8SKent Overstreet { 4215154704bSKent Overstreet struct bch_fs *c = wop->c; 4221c6fdbd8SKent Overstreet struct bchfs_write_op *op = container_of(wop, 4231c6fdbd8SKent Overstreet struct bchfs_write_op, op); 42408af47dfSKent Overstreet struct quota_res *quota_res = op->is_dio 42508af47dfSKent Overstreet ? &container_of(op, struct dio_write, iop)->quota_res 42608af47dfSKent Overstreet : NULL; 42708af47dfSKent Overstreet struct bch_inode_info *inode = op->inode; 4281c6fdbd8SKent Overstreet struct keylist *keys = &op->op.insert_keys; 4291c6fdbd8SKent Overstreet struct bkey_i *k = bch2_keylist_front(keys); 43008af47dfSKent Overstreet struct btree_trans trans; 43108af47dfSKent Overstreet struct btree_iter *iter; 4321c6fdbd8SKent Overstreet int ret; 4331c6fdbd8SKent Overstreet 43408af47dfSKent Overstreet BUG_ON(k->k.p.inode != inode->v.i_ino); 4351c6fdbd8SKent Overstreet 43620bceecbSKent Overstreet bch2_trans_init(&trans, c, BTREE_ITER_MAX, 1024); 437d69f41d6SKent Overstreet 43808af47dfSKent Overstreet iter = bch2_trans_get_iter(&trans, 439d69f41d6SKent Overstreet BTREE_ID_EXTENTS, 44008af47dfSKent Overstreet bkey_start_pos(&k->k), 4411c6fdbd8SKent Overstreet BTREE_ITER_INTENT); 4421c6fdbd8SKent Overstreet 4431c6fdbd8SKent Overstreet do { 444e2d9912cSKent Overstreet BKEY_PADDED(k) tmp; 4451c6fdbd8SKent Overstreet 446e2d9912cSKent Overstreet bkey_copy(&tmp.k, bch2_keylist_front(keys)); 447e2d9912cSKent Overstreet 448b17657d0SKent Overstreet bch2_trans_begin_updates(&trans); 449b17657d0SKent Overstreet 45008af47dfSKent Overstreet ret = bch2_extent_update(&trans, inode, 45108af47dfSKent Overstreet &wop->res, quota_res, 45208af47dfSKent Overstreet iter, &tmp.k, 45308af47dfSKent Overstreet op->new_i_size, 45408af47dfSKent Overstreet !op->unalloc, 45508af47dfSKent Overstreet op->is_dio, 45608af47dfSKent Overstreet &op->sectors_added); 4571c6fdbd8SKent Overstreet if (ret == -EINTR) 4581c6fdbd8SKent Overstreet continue; 4591c6fdbd8SKent Overstreet if (ret) 4601c6fdbd8SKent Overstreet break; 4611c6fdbd8SKent Overstreet 46208af47dfSKent Overstreet if (bkey_cmp(iter->pos, bch2_keylist_front(keys)->k.p) < 0) 46308af47dfSKent Overstreet bch2_cut_front(iter->pos, bch2_keylist_front(keys)); 464e2d9912cSKent Overstreet else 4651c6fdbd8SKent Overstreet bch2_keylist_pop_front(keys); 4661c6fdbd8SKent Overstreet } while (!bch2_keylist_empty(keys)); 4671c6fdbd8SKent Overstreet 468d69f41d6SKent Overstreet bch2_trans_exit(&trans); 4691c6fdbd8SKent Overstreet 4701c6fdbd8SKent Overstreet return ret; 4711c6fdbd8SKent Overstreet } 4721c6fdbd8SKent Overstreet 4731c6fdbd8SKent Overstreet static inline void bch2_fswrite_op_init(struct bchfs_write_op *op, 4741c6fdbd8SKent Overstreet struct bch_fs *c, 4751c6fdbd8SKent Overstreet struct bch_inode_info *inode, 4761c6fdbd8SKent Overstreet struct bch_io_opts opts, 4771c6fdbd8SKent Overstreet bool is_dio) 4781c6fdbd8SKent Overstreet { 4791c6fdbd8SKent Overstreet op->inode = inode; 4801c6fdbd8SKent Overstreet op->sectors_added = 0; 4811c6fdbd8SKent Overstreet op->is_dio = is_dio; 4821c6fdbd8SKent Overstreet op->unalloc = false; 4831c6fdbd8SKent Overstreet op->new_i_size = U64_MAX; 4841c6fdbd8SKent Overstreet 4851c6fdbd8SKent Overstreet bch2_write_op_init(&op->op, c, opts); 4861c6fdbd8SKent Overstreet op->op.target = opts.foreground_target; 4871c6fdbd8SKent Overstreet op->op.index_update_fn = bchfs_write_index_update; 4881c6fdbd8SKent Overstreet op_journal_seq_set(&op->op, &inode->ei_journal_seq); 4891c6fdbd8SKent Overstreet } 4901c6fdbd8SKent Overstreet 4911c6fdbd8SKent Overstreet static inline struct bch_io_opts io_opts(struct bch_fs *c, struct bch_inode_info *inode) 4921c6fdbd8SKent Overstreet { 4931c6fdbd8SKent Overstreet struct bch_io_opts opts = bch2_opts_to_inode_opts(c->opts); 4941c6fdbd8SKent Overstreet 4951c6fdbd8SKent Overstreet bch2_io_opts_apply(&opts, bch2_inode_opts_get(&inode->ei_inode)); 4961c6fdbd8SKent Overstreet return opts; 4971c6fdbd8SKent Overstreet } 4981c6fdbd8SKent Overstreet 4991c6fdbd8SKent Overstreet /* page state: */ 5001c6fdbd8SKent Overstreet 5011c6fdbd8SKent Overstreet /* stored in page->private: */ 5021c6fdbd8SKent Overstreet 503f57a6a5dSKent Overstreet struct bch_page_sector { 504f81b648dSKent Overstreet /* Uncompressed, fully allocated replicas: */ 505f57a6a5dSKent Overstreet unsigned nr_replicas:3; 506f81b648dSKent Overstreet 507f81b648dSKent Overstreet /* Owns PAGE_SECTORS * replicas_reserved sized reservation: */ 508f57a6a5dSKent Overstreet unsigned replicas_reserved:3; 5091c6fdbd8SKent Overstreet 510f57a6a5dSKent Overstreet /* i_sectors: */ 511f57a6a5dSKent Overstreet enum { 512f57a6a5dSKent Overstreet SECTOR_UNALLOCATED, 513f57a6a5dSKent Overstreet SECTOR_QUOTA_RESERVED, 514f57a6a5dSKent Overstreet SECTOR_DIRTY, 515f57a6a5dSKent Overstreet SECTOR_ALLOCATED, 516f57a6a5dSKent Overstreet } state:2; 5171c6fdbd8SKent Overstreet }; 5181c6fdbd8SKent Overstreet 519f57a6a5dSKent Overstreet struct bch_page_state { 520f57a6a5dSKent Overstreet struct bch_page_sector s[PAGE_SECTORS]; 521f57a6a5dSKent Overstreet }; 522f57a6a5dSKent Overstreet 523f57a6a5dSKent Overstreet static inline struct bch_page_state *__bch2_page_state(struct page *page) 5241c6fdbd8SKent Overstreet { 525f57a6a5dSKent Overstreet return page_has_private(page) 526f57a6a5dSKent Overstreet ? (struct bch_page_state *) page_private(page) 527f57a6a5dSKent Overstreet : NULL; 528f57a6a5dSKent Overstreet } 5291c6fdbd8SKent Overstreet 530f57a6a5dSKent Overstreet static inline struct bch_page_state *bch2_page_state(struct page *page) 531f57a6a5dSKent Overstreet { 532e1036a2aSKent Overstreet EBUG_ON(!PageLocked(page)); 5331c6fdbd8SKent Overstreet 534f57a6a5dSKent Overstreet return __bch2_page_state(page); 535f57a6a5dSKent Overstreet } 536f57a6a5dSKent Overstreet 537f57a6a5dSKent Overstreet /* for newly allocated pages: */ 538f57a6a5dSKent Overstreet static void __bch2_page_state_release(struct page *page) 539f57a6a5dSKent Overstreet { 540f57a6a5dSKent Overstreet struct bch_page_state *s = __bch2_page_state(page); 541f57a6a5dSKent Overstreet 542f57a6a5dSKent Overstreet if (!s) 543f57a6a5dSKent Overstreet return; 544f57a6a5dSKent Overstreet 545f57a6a5dSKent Overstreet ClearPagePrivate(page); 546f57a6a5dSKent Overstreet set_page_private(page, 0); 547f57a6a5dSKent Overstreet put_page(page); 548f57a6a5dSKent Overstreet kfree(s); 549f57a6a5dSKent Overstreet } 550f57a6a5dSKent Overstreet 551f57a6a5dSKent Overstreet static void bch2_page_state_release(struct page *page) 552f57a6a5dSKent Overstreet { 553f57a6a5dSKent Overstreet struct bch_page_state *s = bch2_page_state(page); 554f57a6a5dSKent Overstreet 555f57a6a5dSKent Overstreet if (!s) 556f57a6a5dSKent Overstreet return; 557f57a6a5dSKent Overstreet 558f57a6a5dSKent Overstreet ClearPagePrivate(page); 559f57a6a5dSKent Overstreet set_page_private(page, 0); 560f57a6a5dSKent Overstreet put_page(page); 561f57a6a5dSKent Overstreet kfree(s); 562f57a6a5dSKent Overstreet } 563f57a6a5dSKent Overstreet 564f57a6a5dSKent Overstreet /* for newly allocated pages: */ 565f57a6a5dSKent Overstreet static struct bch_page_state *__bch2_page_state_create(struct page *page, 566f57a6a5dSKent Overstreet gfp_t gfp) 567f57a6a5dSKent Overstreet { 568f57a6a5dSKent Overstreet struct bch_page_state *s; 569f57a6a5dSKent Overstreet 570f57a6a5dSKent Overstreet s = kzalloc(sizeof(*s), GFP_NOFS|gfp); 571f57a6a5dSKent Overstreet if (!s) 572f57a6a5dSKent Overstreet return NULL; 573f57a6a5dSKent Overstreet 574f57a6a5dSKent Overstreet /* 575f57a6a5dSKent Overstreet * migrate_page_move_mapping() assumes that pages with private data 576f57a6a5dSKent Overstreet * have their count elevated by 1. 577f57a6a5dSKent Overstreet */ 578f57a6a5dSKent Overstreet get_page(page); 579f57a6a5dSKent Overstreet set_page_private(page, (unsigned long) s); 5801c6fdbd8SKent Overstreet SetPagePrivate(page); 5811c6fdbd8SKent Overstreet return s; 5821c6fdbd8SKent Overstreet } 5831c6fdbd8SKent Overstreet 584f57a6a5dSKent Overstreet static struct bch_page_state *bch2_page_state_create(struct page *page, 585f57a6a5dSKent Overstreet gfp_t gfp) 586f57a6a5dSKent Overstreet { 587f57a6a5dSKent Overstreet return bch2_page_state(page) ?: __bch2_page_state_create(page, gfp); 588f57a6a5dSKent Overstreet } 589f57a6a5dSKent Overstreet 5901c6fdbd8SKent Overstreet static void bch2_put_page_reservation(struct bch_fs *c, struct bch_inode_info *inode, 5911c6fdbd8SKent Overstreet struct page *page) 5921c6fdbd8SKent Overstreet { 593f57a6a5dSKent Overstreet struct bch_page_state *s = bch2_page_state(page); 594f57a6a5dSKent Overstreet struct disk_reservation disk_res = { 0 }; 595f57a6a5dSKent Overstreet struct quota_res quota_res = { 0 }; 596f57a6a5dSKent Overstreet unsigned i; 5971c6fdbd8SKent Overstreet 598f57a6a5dSKent Overstreet if (!s) 599f57a6a5dSKent Overstreet return; 600f57a6a5dSKent Overstreet 601f57a6a5dSKent Overstreet for (i = 0; i < ARRAY_SIZE(s->s); i++) { 602f57a6a5dSKent Overstreet disk_res.sectors += s->s[i].replicas_reserved; 603f57a6a5dSKent Overstreet s->s[i].replicas_reserved = 0; 604f57a6a5dSKent Overstreet 605f57a6a5dSKent Overstreet if (s->s[i].state == SECTOR_QUOTA_RESERVED) { 606f57a6a5dSKent Overstreet quota_res.sectors++; 607f57a6a5dSKent Overstreet s->s[i].state = SECTOR_UNALLOCATED; 608f57a6a5dSKent Overstreet } 609f57a6a5dSKent Overstreet } 610f81b648dSKent Overstreet 611adfcfaf0SKent Overstreet bch2_quota_reservation_put(c, inode, "a_res); 612adfcfaf0SKent Overstreet bch2_disk_reservation_put(c, &disk_res); 6131c6fdbd8SKent Overstreet } 6141c6fdbd8SKent Overstreet 615e1036a2aSKent Overstreet static inline unsigned inode_nr_replicas(struct bch_fs *c, struct bch_inode_info *inode) 616e1036a2aSKent Overstreet { 617e1036a2aSKent Overstreet /* XXX: this should not be open coded */ 618e1036a2aSKent Overstreet return inode->ei_inode.bi_data_replicas 619e1036a2aSKent Overstreet ? inode->ei_inode.bi_data_replicas - 1 620e1036a2aSKent Overstreet : c->opts.data_replicas; 621e1036a2aSKent Overstreet } 622e1036a2aSKent Overstreet 623f57a6a5dSKent Overstreet static inline unsigned sectors_to_reserve(struct bch_page_sector *s, 624f57a6a5dSKent Overstreet unsigned nr_replicas) 625f57a6a5dSKent Overstreet { 626f57a6a5dSKent Overstreet return max(0, (int) nr_replicas - 627f57a6a5dSKent Overstreet s->nr_replicas - 628f57a6a5dSKent Overstreet s->replicas_reserved); 629f57a6a5dSKent Overstreet } 630f57a6a5dSKent Overstreet 631f57a6a5dSKent Overstreet static int bch2_get_page_disk_reservation(struct bch_fs *c, 632f57a6a5dSKent Overstreet struct bch_inode_info *inode, 6331c6fdbd8SKent Overstreet struct page *page, bool check_enospc) 6341c6fdbd8SKent Overstreet { 635f57a6a5dSKent Overstreet struct bch_page_state *s = bch2_page_state_create(page, 0); 636e1036a2aSKent Overstreet unsigned nr_replicas = inode_nr_replicas(c, inode); 637f57a6a5dSKent Overstreet struct disk_reservation disk_res = { 0 }; 638f57a6a5dSKent Overstreet unsigned i, disk_res_sectors = 0; 639f81b648dSKent Overstreet int ret; 6401c6fdbd8SKent Overstreet 641f57a6a5dSKent Overstreet if (!s) 642f57a6a5dSKent Overstreet return -ENOMEM; 6431c6fdbd8SKent Overstreet 644f57a6a5dSKent Overstreet for (i = 0; i < ARRAY_SIZE(s->s); i++) 645f57a6a5dSKent Overstreet disk_res_sectors += sectors_to_reserve(&s->s[i], nr_replicas); 646f57a6a5dSKent Overstreet 647f57a6a5dSKent Overstreet if (!disk_res_sectors) 648f57a6a5dSKent Overstreet return 0; 649f57a6a5dSKent Overstreet 650f57a6a5dSKent Overstreet ret = bch2_disk_reservation_get(c, &disk_res, 651f57a6a5dSKent Overstreet disk_res_sectors, 1, 652f57a6a5dSKent Overstreet !check_enospc 653f57a6a5dSKent Overstreet ? BCH_DISK_RESERVATION_NOFAIL 654f57a6a5dSKent Overstreet : 0); 6551c6fdbd8SKent Overstreet if (unlikely(ret)) 656f81b648dSKent Overstreet return ret; 657f81b648dSKent Overstreet 658f57a6a5dSKent Overstreet for (i = 0; i < ARRAY_SIZE(s->s); i++) 659f57a6a5dSKent Overstreet s->s[i].replicas_reserved += 660f57a6a5dSKent Overstreet sectors_to_reserve(&s->s[i], nr_replicas); 661f57a6a5dSKent Overstreet 662f57a6a5dSKent Overstreet return 0; 6631c6fdbd8SKent Overstreet } 6641c6fdbd8SKent Overstreet 665f57a6a5dSKent Overstreet static int bch2_get_page_quota_reservation(struct bch_fs *c, 666f57a6a5dSKent Overstreet struct bch_inode_info *inode, 667f57a6a5dSKent Overstreet struct page *page, bool check_enospc) 668f57a6a5dSKent Overstreet { 669f57a6a5dSKent Overstreet struct bch_page_state *s = bch2_page_state_create(page, 0); 670f57a6a5dSKent Overstreet struct quota_res quota_res = { 0 }; 671f57a6a5dSKent Overstreet unsigned i, quota_res_sectors = 0; 672f57a6a5dSKent Overstreet int ret; 673f57a6a5dSKent Overstreet 674f57a6a5dSKent Overstreet if (!s) 675f57a6a5dSKent Overstreet return -ENOMEM; 676f57a6a5dSKent Overstreet 677f57a6a5dSKent Overstreet for (i = 0; i < ARRAY_SIZE(s->s); i++) 678f57a6a5dSKent Overstreet quota_res_sectors += s->s[i].state == SECTOR_UNALLOCATED; 679f57a6a5dSKent Overstreet 680f57a6a5dSKent Overstreet if (!quota_res_sectors) 681f57a6a5dSKent Overstreet return 0; 682f57a6a5dSKent Overstreet 6831c6fdbd8SKent Overstreet ret = bch2_quota_reservation_add(c, inode, "a_res, 684f57a6a5dSKent Overstreet quota_res_sectors, 6851c6fdbd8SKent Overstreet check_enospc); 6861c6fdbd8SKent Overstreet if (unlikely(ret)) 687f81b648dSKent Overstreet return ret; 6881c6fdbd8SKent Overstreet 689f57a6a5dSKent Overstreet for (i = 0; i < ARRAY_SIZE(s->s); i++) 690f57a6a5dSKent Overstreet if (s->s[i].state == SECTOR_UNALLOCATED) 691f57a6a5dSKent Overstreet s->s[i].state = SECTOR_QUOTA_RESERVED; 6921c6fdbd8SKent Overstreet 693adfcfaf0SKent Overstreet return 0; 6941c6fdbd8SKent Overstreet } 6951c6fdbd8SKent Overstreet 696f57a6a5dSKent Overstreet static int bch2_get_page_reservation(struct bch_fs *c, struct bch_inode_info *inode, 697f57a6a5dSKent Overstreet struct page *page, bool check_enospc) 698f57a6a5dSKent Overstreet { 699f57a6a5dSKent Overstreet return bch2_get_page_disk_reservation(c, inode, page, check_enospc) ?: 700f57a6a5dSKent Overstreet bch2_get_page_quota_reservation(c, inode, page, check_enospc); 701f57a6a5dSKent Overstreet } 702f57a6a5dSKent Overstreet 7031c6fdbd8SKent Overstreet static void bch2_clear_page_bits(struct page *page) 7041c6fdbd8SKent Overstreet { 7051c6fdbd8SKent Overstreet struct bch_inode_info *inode = to_bch_ei(page->mapping->host); 7061c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 707f57a6a5dSKent Overstreet struct bch_page_state *s = bch2_page_state(page); 708f57a6a5dSKent Overstreet int i, dirty_sectors = 0; 7091c6fdbd8SKent Overstreet 710f57a6a5dSKent Overstreet if (!s) 7111c6fdbd8SKent Overstreet return; 7121c6fdbd8SKent Overstreet 713f57a6a5dSKent Overstreet for (i = 0; i < ARRAY_SIZE(s->s); i++) { 714f57a6a5dSKent Overstreet if (s->s[i].state == SECTOR_DIRTY) { 715f57a6a5dSKent Overstreet dirty_sectors++; 716f57a6a5dSKent Overstreet s->s[i].state = SECTOR_UNALLOCATED; 717f57a6a5dSKent Overstreet } 718f57a6a5dSKent Overstreet } 719adfcfaf0SKent Overstreet 720f57a6a5dSKent Overstreet if (dirty_sectors) 721f57a6a5dSKent Overstreet i_sectors_acct(c, inode, NULL, -dirty_sectors); 722adfcfaf0SKent Overstreet bch2_put_page_reservation(c, inode, page); 723adfcfaf0SKent Overstreet 724f57a6a5dSKent Overstreet bch2_page_state_release(page); 7251c6fdbd8SKent Overstreet } 7261c6fdbd8SKent Overstreet 727e1036a2aSKent Overstreet static void __bch2_set_page_dirty(struct address_space *mapping, struct folio *folio) 7281c6fdbd8SKent Overstreet { 7291c6fdbd8SKent Overstreet struct bch_inode_info *inode = to_bch_ei(mapping->host); 7301c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 731f57a6a5dSKent Overstreet struct bch_page_state *s = bch2_page_state(&folio->page); 732f57a6a5dSKent Overstreet struct quota_res quota_res = { 0 }; 733f57a6a5dSKent Overstreet unsigned i, dirty_sectors = 0; 7341c6fdbd8SKent Overstreet 735f57a6a5dSKent Overstreet BUG_ON(!s); 7361c6fdbd8SKent Overstreet 737f57a6a5dSKent Overstreet for (i = 0; i < ARRAY_SIZE(s->s); i++) { 738f57a6a5dSKent Overstreet if (s->s[i].state == SECTOR_QUOTA_RESERVED) 739f57a6a5dSKent Overstreet quota_res.sectors++; 740adfcfaf0SKent Overstreet 741f57a6a5dSKent Overstreet if (s->s[i].state == SECTOR_UNALLOCATED || 742f57a6a5dSKent Overstreet s->s[i].state == SECTOR_QUOTA_RESERVED) { 743f57a6a5dSKent Overstreet s->s[i].state = SECTOR_DIRTY; 744f57a6a5dSKent Overstreet dirty_sectors++; 745f57a6a5dSKent Overstreet } 746f57a6a5dSKent Overstreet } 747f57a6a5dSKent Overstreet 748f57a6a5dSKent Overstreet if (dirty_sectors) 749f57a6a5dSKent Overstreet i_sectors_acct(c, inode, "a_res, dirty_sectors); 7501c6fdbd8SKent Overstreet bch2_quota_reservation_put(c, inode, "a_res); 751e1036a2aSKent Overstreet } 7521c6fdbd8SKent Overstreet 753e1036a2aSKent Overstreet static void bch2_set_page_dirty(struct address_space *mapping, struct page *page) 754e1036a2aSKent Overstreet { 755e1036a2aSKent Overstreet struct folio *folio = page_folio(page); 756e1036a2aSKent Overstreet 757e1036a2aSKent Overstreet __bch2_set_page_dirty(mapping, folio); 758e1036a2aSKent Overstreet filemap_dirty_folio(mapping, folio); 7591c6fdbd8SKent Overstreet } 7601c6fdbd8SKent Overstreet 7611c6fdbd8SKent Overstreet vm_fault_t bch2_page_fault(struct vm_fault *vmf) 7621c6fdbd8SKent Overstreet { 7631c6fdbd8SKent Overstreet struct file *file = vmf->vma->vm_file; 7641c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 7651c6fdbd8SKent Overstreet int ret; 7661c6fdbd8SKent Overstreet 7671c6fdbd8SKent Overstreet bch2_pagecache_add_get(&inode->ei_pagecache_lock); 7681c6fdbd8SKent Overstreet ret = filemap_fault(vmf); 7691c6fdbd8SKent Overstreet bch2_pagecache_add_put(&inode->ei_pagecache_lock); 7701c6fdbd8SKent Overstreet 7711c6fdbd8SKent Overstreet return ret; 7721c6fdbd8SKent Overstreet } 7731c6fdbd8SKent Overstreet 7741c6fdbd8SKent Overstreet vm_fault_t bch2_page_mkwrite(struct vm_fault *vmf) 7751c6fdbd8SKent Overstreet { 7761c6fdbd8SKent Overstreet struct page *page = vmf->page; 7771c6fdbd8SKent Overstreet struct file *file = vmf->vma->vm_file; 7781c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 7791c6fdbd8SKent Overstreet struct address_space *mapping = file->f_mapping; 7801c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 7811c6fdbd8SKent Overstreet int ret = VM_FAULT_LOCKED; 7821c6fdbd8SKent Overstreet 7831c6fdbd8SKent Overstreet sb_start_pagefault(inode->v.i_sb); 7841c6fdbd8SKent Overstreet file_update_time(file); 7851c6fdbd8SKent Overstreet 7861c6fdbd8SKent Overstreet /* 7871c6fdbd8SKent Overstreet * Not strictly necessary, but helps avoid dio writes livelocking in 7881c6fdbd8SKent Overstreet * write_invalidate_inode_pages_range() - can drop this if/when we get 7891c6fdbd8SKent Overstreet * a write_invalidate_inode_pages_range() that works without dropping 7901c6fdbd8SKent Overstreet * page lock before invalidating page 7911c6fdbd8SKent Overstreet */ 7921c6fdbd8SKent Overstreet bch2_pagecache_add_get(&inode->ei_pagecache_lock); 7931c6fdbd8SKent Overstreet 7941c6fdbd8SKent Overstreet lock_page(page); 7951c6fdbd8SKent Overstreet if (page->mapping != mapping || 7961c6fdbd8SKent Overstreet page_offset(page) > i_size_read(&inode->v)) { 7971c6fdbd8SKent Overstreet unlock_page(page); 7981c6fdbd8SKent Overstreet ret = VM_FAULT_NOPAGE; 7991c6fdbd8SKent Overstreet goto out; 8001c6fdbd8SKent Overstreet } 8011c6fdbd8SKent Overstreet 8021c6fdbd8SKent Overstreet if (bch2_get_page_reservation(c, inode, page, true)) { 8031c6fdbd8SKent Overstreet unlock_page(page); 8041c6fdbd8SKent Overstreet ret = VM_FAULT_SIGBUS; 8051c6fdbd8SKent Overstreet goto out; 8061c6fdbd8SKent Overstreet } 8071c6fdbd8SKent Overstreet 8081c6fdbd8SKent Overstreet if (!PageDirty(page)) 809e1036a2aSKent Overstreet bch2_set_page_dirty(mapping, page); 8101c6fdbd8SKent Overstreet wait_for_stable_page(page); 8111c6fdbd8SKent Overstreet out: 8121c6fdbd8SKent Overstreet bch2_pagecache_add_put(&inode->ei_pagecache_lock); 8131c6fdbd8SKent Overstreet sb_end_pagefault(inode->v.i_sb); 8141c6fdbd8SKent Overstreet return ret; 8151c6fdbd8SKent Overstreet } 8161c6fdbd8SKent Overstreet 8171c6fdbd8SKent Overstreet void bch2_invalidate_folio(struct folio *folio, size_t offset, size_t length) 8181c6fdbd8SKent Overstreet { 8191c6fdbd8SKent Overstreet if (offset || length < folio_size(folio)) 8201c6fdbd8SKent Overstreet return; 8211c6fdbd8SKent Overstreet 8221c6fdbd8SKent Overstreet bch2_clear_page_bits(&folio->page); 8231c6fdbd8SKent Overstreet } 8241c6fdbd8SKent Overstreet 8251c6fdbd8SKent Overstreet bool bch2_release_folio(struct folio *folio, gfp_t gfp_mask) 8261c6fdbd8SKent Overstreet { 8271c6fdbd8SKent Overstreet /* XXX: this can't take locks that are held while we allocate memory */ 8281c6fdbd8SKent Overstreet EBUG_ON(!PageLocked(&folio->page)); 8291c6fdbd8SKent Overstreet EBUG_ON(folio_test_writeback(folio)); 8301c6fdbd8SKent Overstreet 831a6d90385SKent Overstreet if (folio_test_dirty(folio) || folio_test_writeback(folio)) 8321c6fdbd8SKent Overstreet return false; 8331c6fdbd8SKent Overstreet 8341c6fdbd8SKent Overstreet bch2_clear_page_bits(&folio->page); 8351c6fdbd8SKent Overstreet return true; 8361c6fdbd8SKent Overstreet } 8371c6fdbd8SKent Overstreet 8381c6fdbd8SKent Overstreet /* readpages/writepages: */ 8391c6fdbd8SKent Overstreet 8401c6fdbd8SKent Overstreet static bool bio_can_add_page_contig(struct bio *bio, struct page *page) 8411c6fdbd8SKent Overstreet { 8421c6fdbd8SKent Overstreet sector_t offset = (sector_t) page->index << PAGE_SECTOR_SHIFT; 8431c6fdbd8SKent Overstreet 8441c6fdbd8SKent Overstreet return bio->bi_vcnt < bio->bi_max_vecs && 8451c6fdbd8SKent Overstreet bio_end_sector(bio) == offset; 8461c6fdbd8SKent Overstreet } 8471c6fdbd8SKent Overstreet 8481c6fdbd8SKent Overstreet static int bio_add_page_contig(struct bio *bio, struct page *page) 8491c6fdbd8SKent Overstreet { 8501c6fdbd8SKent Overstreet sector_t offset = (sector_t) page->index << PAGE_SECTOR_SHIFT; 8511c6fdbd8SKent Overstreet 8521c6fdbd8SKent Overstreet EBUG_ON(!bio->bi_max_vecs); 8531c6fdbd8SKent Overstreet 8541c6fdbd8SKent Overstreet if (!bio->bi_vcnt) 8551c6fdbd8SKent Overstreet bio->bi_iter.bi_sector = offset; 8561c6fdbd8SKent Overstreet else if (!bio_can_add_page_contig(bio, page)) 8571c6fdbd8SKent Overstreet return -1; 8581c6fdbd8SKent Overstreet 859885678f6SKent Overstreet BUG_ON(!bio_add_page(bio, page, PAGE_SIZE, 0)); 8601c6fdbd8SKent Overstreet return 0; 8611c6fdbd8SKent Overstreet } 8621c6fdbd8SKent Overstreet 8631c6fdbd8SKent Overstreet /* readpage(s): */ 8641c6fdbd8SKent Overstreet 8651c6fdbd8SKent Overstreet static void bch2_readpages_end_io(struct bio *bio) 8661c6fdbd8SKent Overstreet { 8671c6fdbd8SKent Overstreet struct bvec_iter_all iter; 8681c6fdbd8SKent Overstreet struct bio_vec *bv; 8691c6fdbd8SKent Overstreet 8701c6fdbd8SKent Overstreet bio_for_each_segment_all(bv, bio, iter) { 8711c6fdbd8SKent Overstreet struct page *page = bv->bv_page; 8721c6fdbd8SKent Overstreet 8731c6fdbd8SKent Overstreet if (!bio->bi_status) { 8741c6fdbd8SKent Overstreet SetPageUptodate(page); 8751c6fdbd8SKent Overstreet } else { 8761c6fdbd8SKent Overstreet ClearPageUptodate(page); 8771c6fdbd8SKent Overstreet SetPageError(page); 8781c6fdbd8SKent Overstreet } 8791c6fdbd8SKent Overstreet unlock_page(page); 8801c6fdbd8SKent Overstreet } 8811c6fdbd8SKent Overstreet 8821c6fdbd8SKent Overstreet bio_put(bio); 8831c6fdbd8SKent Overstreet } 8841c6fdbd8SKent Overstreet 8851c6fdbd8SKent Overstreet static inline void page_state_init_for_read(struct page *page) 8861c6fdbd8SKent Overstreet { 887f81b648dSKent Overstreet SetPagePrivate(page); 888f81b648dSKent Overstreet page->private = 0; 8891c6fdbd8SKent Overstreet } 8901c6fdbd8SKent Overstreet 8911c6fdbd8SKent Overstreet struct readpages_iter { 8921c6fdbd8SKent Overstreet struct address_space *mapping; 8931c6fdbd8SKent Overstreet struct page **pages; 8941c6fdbd8SKent Overstreet unsigned nr_pages; 8951c6fdbd8SKent Overstreet unsigned idx; 8961c6fdbd8SKent Overstreet pgoff_t offset; 8971c6fdbd8SKent Overstreet }; 8981c6fdbd8SKent Overstreet 8991c6fdbd8SKent Overstreet static int readpages_iter_init(struct readpages_iter *iter, 9001c6fdbd8SKent Overstreet struct readahead_control *ractl) 9011c6fdbd8SKent Overstreet { 9021c6fdbd8SKent Overstreet unsigned i, nr_pages = readahead_count(ractl); 9031c6fdbd8SKent Overstreet 9041c6fdbd8SKent Overstreet memset(iter, 0, sizeof(*iter)); 9051c6fdbd8SKent Overstreet 9061c6fdbd8SKent Overstreet iter->mapping = ractl->mapping; 9071c6fdbd8SKent Overstreet iter->offset = readahead_index(ractl); 9081c6fdbd8SKent Overstreet iter->nr_pages = nr_pages; 9091c6fdbd8SKent Overstreet 9101c6fdbd8SKent Overstreet iter->pages = kmalloc_array(nr_pages, sizeof(struct page *), GFP_NOFS); 9111c6fdbd8SKent Overstreet if (!iter->pages) 9121c6fdbd8SKent Overstreet return -ENOMEM; 9131c6fdbd8SKent Overstreet 9141c6fdbd8SKent Overstreet __readahead_batch(ractl, iter->pages, nr_pages); 9151c6fdbd8SKent Overstreet for (i = 0; i < nr_pages; i++) { 916f57a6a5dSKent Overstreet __bch2_page_state_create(iter->pages[i], __GFP_NOFAIL); 9171c6fdbd8SKent Overstreet put_page(iter->pages[i]); 9181c6fdbd8SKent Overstreet } 9191c6fdbd8SKent Overstreet 9201c6fdbd8SKent Overstreet return 0; 9211c6fdbd8SKent Overstreet } 9221c6fdbd8SKent Overstreet 9231c6fdbd8SKent Overstreet static inline struct page *readpage_iter_next(struct readpages_iter *iter) 9241c6fdbd8SKent Overstreet { 9251c6fdbd8SKent Overstreet if (iter->idx >= iter->nr_pages) 9261c6fdbd8SKent Overstreet return NULL; 9271c6fdbd8SKent Overstreet 9281c6fdbd8SKent Overstreet EBUG_ON(iter->pages[iter->idx]->index != iter->offset + iter->idx); 9291c6fdbd8SKent Overstreet 9301c6fdbd8SKent Overstreet return iter->pages[iter->idx]; 9311c6fdbd8SKent Overstreet } 9321c6fdbd8SKent Overstreet 9331c6fdbd8SKent Overstreet static void bch2_add_page_sectors(struct bio *bio, struct bkey_s_c k) 9341c6fdbd8SKent Overstreet { 9351c6fdbd8SKent Overstreet struct bvec_iter iter; 9361c6fdbd8SKent Overstreet struct bio_vec bv; 9377ef2a73aSKent Overstreet unsigned nr_ptrs = bch2_bkey_nr_ptrs_allocated(k); 9381c6fdbd8SKent Overstreet 939f57a6a5dSKent Overstreet BUG_ON(bio->bi_iter.bi_sector < bkey_start_offset(k.k)); 940f57a6a5dSKent Overstreet BUG_ON(bio_end_sector(bio) > k.k->p.offset); 941f57a6a5dSKent Overstreet 942f57a6a5dSKent Overstreet 9431c6fdbd8SKent Overstreet bio_for_each_segment(bv, bio, iter) { 944f57a6a5dSKent Overstreet struct bch_page_state *s = bch2_page_state(bv.bv_page); 945f57a6a5dSKent Overstreet unsigned i; 946f81b648dSKent Overstreet 947f57a6a5dSKent Overstreet for (i = bv.bv_offset >> 9; 948f57a6a5dSKent Overstreet i < (bv.bv_offset + bv.bv_len) >> 9; 949f57a6a5dSKent Overstreet i++) { 950f57a6a5dSKent Overstreet s->s[i].nr_replicas = nr_ptrs; 951f57a6a5dSKent Overstreet s->s[i].state = SECTOR_ALLOCATED; 952f57a6a5dSKent Overstreet } 9531c6fdbd8SKent Overstreet } 9541c6fdbd8SKent Overstreet } 9551c6fdbd8SKent Overstreet 9561c6fdbd8SKent Overstreet static void readpage_bio_extend(struct readpages_iter *iter, 9571c6fdbd8SKent Overstreet struct bio *bio, u64 offset, 9581c6fdbd8SKent Overstreet bool get_more) 9591c6fdbd8SKent Overstreet { 9601c6fdbd8SKent Overstreet while (bio_end_sector(bio) < offset && 9611c6fdbd8SKent Overstreet bio->bi_vcnt < bio->bi_max_vecs) { 9621c6fdbd8SKent Overstreet pgoff_t page_offset = bio_end_sector(bio) >> PAGE_SECTOR_SHIFT; 9631c6fdbd8SKent Overstreet struct page *page = readpage_iter_next(iter); 9641c6fdbd8SKent Overstreet int ret; 9651c6fdbd8SKent Overstreet 9661c6fdbd8SKent Overstreet if (page) { 9671c6fdbd8SKent Overstreet if (iter->offset + iter->idx != page_offset) 9681c6fdbd8SKent Overstreet break; 9691c6fdbd8SKent Overstreet 9701c6fdbd8SKent Overstreet iter->idx++; 9711c6fdbd8SKent Overstreet } else { 9721c6fdbd8SKent Overstreet if (!get_more) 9731c6fdbd8SKent Overstreet break; 9741c6fdbd8SKent Overstreet 9751c6fdbd8SKent Overstreet page = xa_load(&iter->mapping->i_pages, page_offset); 9761c6fdbd8SKent Overstreet if (page && !xa_is_value(page)) 9771c6fdbd8SKent Overstreet break; 9781c6fdbd8SKent Overstreet 9791c6fdbd8SKent Overstreet page = __page_cache_alloc(readahead_gfp_mask(iter->mapping)); 9801c6fdbd8SKent Overstreet if (!page) 9811c6fdbd8SKent Overstreet break; 9821c6fdbd8SKent Overstreet 983f57a6a5dSKent Overstreet if (!__bch2_page_state_create(page, 0)) { 984f57a6a5dSKent Overstreet put_page(page); 985f57a6a5dSKent Overstreet break; 986f57a6a5dSKent Overstreet } 9871c6fdbd8SKent Overstreet 9881c6fdbd8SKent Overstreet ret = add_to_page_cache_lru(page, iter->mapping, 9891c6fdbd8SKent Overstreet page_offset, GFP_NOFS); 9901c6fdbd8SKent Overstreet if (ret) { 991f57a6a5dSKent Overstreet __bch2_page_state_release(page); 9921c6fdbd8SKent Overstreet put_page(page); 9931c6fdbd8SKent Overstreet break; 9941c6fdbd8SKent Overstreet } 9951c6fdbd8SKent Overstreet 9961c6fdbd8SKent Overstreet put_page(page); 9971c6fdbd8SKent Overstreet } 9981c6fdbd8SKent Overstreet 999885678f6SKent Overstreet BUG_ON(!bio_add_page(bio, page, PAGE_SIZE, 0)); 10001c6fdbd8SKent Overstreet } 10011c6fdbd8SKent Overstreet } 10021c6fdbd8SKent Overstreet 10030f238367SKent Overstreet static void bchfs_read(struct btree_trans *trans, struct btree_iter *iter, 10041c6fdbd8SKent Overstreet struct bch_read_bio *rbio, u64 inum, 10051c6fdbd8SKent Overstreet struct readpages_iter *readpages_iter) 10061c6fdbd8SKent Overstreet { 10070f238367SKent Overstreet struct bch_fs *c = trans->c; 10081c6fdbd8SKent Overstreet int flags = BCH_READ_RETRY_IF_STALE| 10091c6fdbd8SKent Overstreet BCH_READ_MAY_PROMOTE; 10101c6fdbd8SKent Overstreet 10111c6fdbd8SKent Overstreet rbio->c = c; 10121c6fdbd8SKent Overstreet rbio->start_time = local_clock(); 10131c6fdbd8SKent Overstreet 10141c6fdbd8SKent Overstreet while (1) { 10151c6fdbd8SKent Overstreet BKEY_PADDED(k) tmp; 10161c6fdbd8SKent Overstreet struct bkey_s_c k; 101706ed8558SKent Overstreet unsigned bytes, offset_into_extent; 10181c6fdbd8SKent Overstreet 101906ed8558SKent Overstreet bch2_btree_iter_set_pos(iter, 102006ed8558SKent Overstreet POS(inum, rbio->bio.bi_iter.bi_sector)); 10211c6fdbd8SKent Overstreet 10221c6fdbd8SKent Overstreet k = bch2_btree_iter_peek_slot(iter); 10231c6fdbd8SKent Overstreet BUG_ON(!k.k); 10241c6fdbd8SKent Overstreet 10251c6fdbd8SKent Overstreet if (IS_ERR(k.k)) { 10260f238367SKent Overstreet int ret = btree_iter_err(iter); 10271c6fdbd8SKent Overstreet BUG_ON(!ret); 102806ed8558SKent Overstreet bcache_io_error(c, &rbio->bio, "btree IO error %i", ret); 102906ed8558SKent Overstreet bio_endio(&rbio->bio); 10301c6fdbd8SKent Overstreet return; 10311c6fdbd8SKent Overstreet } 10321c6fdbd8SKent Overstreet 10331c6fdbd8SKent Overstreet bkey_reassemble(&tmp.k, k); 103458fbf808SKent Overstreet bch2_trans_unlock(trans); 10351c6fdbd8SKent Overstreet k = bkey_i_to_s_c(&tmp.k); 10361c6fdbd8SKent Overstreet 103706ed8558SKent Overstreet offset_into_extent = iter->pos.offset - 103806ed8558SKent Overstreet bkey_start_offset(k.k); 103906ed8558SKent Overstreet 10401c6fdbd8SKent Overstreet if (readpages_iter) { 10411c6fdbd8SKent Overstreet bool want_full_extent = false; 10421c6fdbd8SKent Overstreet 10431c6fdbd8SKent Overstreet if (bkey_extent_is_data(k.k)) { 104499aaf570SKent Overstreet struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k); 10451c6fdbd8SKent Overstreet const union bch_extent_entry *i; 10461742237bSKent Overstreet struct extent_ptr_decoded p; 10471c6fdbd8SKent Overstreet 104899aaf570SKent Overstreet bkey_for_each_ptr_decode(k.k, ptrs, p, i) 10491742237bSKent Overstreet want_full_extent |= ((p.crc.csum_type != 0) | 10501742237bSKent Overstreet (p.crc.compression_type != 0)); 10511c6fdbd8SKent Overstreet } 10521c6fdbd8SKent Overstreet 10531c6fdbd8SKent Overstreet readpage_bio_extend(readpages_iter, 105406ed8558SKent Overstreet &rbio->bio, k.k->p.offset, 10551c6fdbd8SKent Overstreet want_full_extent); 10561c6fdbd8SKent Overstreet } 10571c6fdbd8SKent Overstreet 105806ed8558SKent Overstreet bytes = min_t(unsigned, bio_sectors(&rbio->bio), 105906ed8558SKent Overstreet (k.k->size - offset_into_extent)) << 9; 106006ed8558SKent Overstreet swap(rbio->bio.bi_iter.bi_size, bytes); 10611c6fdbd8SKent Overstreet 106206ed8558SKent Overstreet if (rbio->bio.bi_iter.bi_size == bytes) 10631c6fdbd8SKent Overstreet flags |= BCH_READ_LAST_FRAGMENT; 10641c6fdbd8SKent Overstreet 10651c6fdbd8SKent Overstreet if (bkey_extent_is_allocation(k.k)) 106606ed8558SKent Overstreet bch2_add_page_sectors(&rbio->bio, k); 10671c6fdbd8SKent Overstreet 106806ed8558SKent Overstreet bch2_read_extent(c, rbio, k, offset_into_extent, flags); 10691c6fdbd8SKent Overstreet 10701c6fdbd8SKent Overstreet if (flags & BCH_READ_LAST_FRAGMENT) 10711c6fdbd8SKent Overstreet return; 10721c6fdbd8SKent Overstreet 107306ed8558SKent Overstreet swap(rbio->bio.bi_iter.bi_size, bytes); 107406ed8558SKent Overstreet bio_advance(&rbio->bio, bytes); 10751c6fdbd8SKent Overstreet } 10761c6fdbd8SKent Overstreet } 10771c6fdbd8SKent Overstreet 10781c6fdbd8SKent Overstreet void bch2_readahead(struct readahead_control *ractl) 10791c6fdbd8SKent Overstreet { 10801c6fdbd8SKent Overstreet struct bch_inode_info *inode = to_bch_ei(ractl->mapping->host); 10811c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 10821c6fdbd8SKent Overstreet struct bch_io_opts opts = io_opts(c, inode); 1083424eb881SKent Overstreet struct btree_trans trans; 1084424eb881SKent Overstreet struct btree_iter *iter; 10851c6fdbd8SKent Overstreet struct page *page; 10861c6fdbd8SKent Overstreet struct readpages_iter readpages_iter; 10871c6fdbd8SKent Overstreet int ret; 10881c6fdbd8SKent Overstreet 10891c6fdbd8SKent Overstreet ret = readpages_iter_init(&readpages_iter, ractl); 10901c6fdbd8SKent Overstreet BUG_ON(ret); 10911c6fdbd8SKent Overstreet 109220bceecbSKent Overstreet bch2_trans_init(&trans, c, 0, 0); 1093424eb881SKent Overstreet 1094424eb881SKent Overstreet iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS, POS_MIN, 10951c6fdbd8SKent Overstreet BTREE_ITER_SLOTS); 10961c6fdbd8SKent Overstreet 10971c6fdbd8SKent Overstreet bch2_pagecache_add_get(&inode->ei_pagecache_lock); 10981c6fdbd8SKent Overstreet 10991c6fdbd8SKent Overstreet while ((page = readpage_iter_next(&readpages_iter))) { 11001c6fdbd8SKent Overstreet pgoff_t index = readpages_iter.offset + readpages_iter.idx; 11011c6fdbd8SKent Overstreet unsigned n = min_t(unsigned, 11021c6fdbd8SKent Overstreet readpages_iter.nr_pages - 11031c6fdbd8SKent Overstreet readpages_iter.idx, 11041c6fdbd8SKent Overstreet BIO_MAX_VECS); 11051c6fdbd8SKent Overstreet struct bch_read_bio *rbio = 11061c6fdbd8SKent Overstreet rbio_init(bio_alloc_bioset(NULL, n, REQ_OP_READ, 11071c6fdbd8SKent Overstreet GFP_NOFS, &c->bio_read), 11081c6fdbd8SKent Overstreet opts); 11091c6fdbd8SKent Overstreet 11101c6fdbd8SKent Overstreet readpages_iter.idx++; 11111c6fdbd8SKent Overstreet 11121c6fdbd8SKent Overstreet rbio->bio.bi_iter.bi_sector = (sector_t) index << PAGE_SECTOR_SHIFT; 11131c6fdbd8SKent Overstreet rbio->bio.bi_end_io = bch2_readpages_end_io; 1114885678f6SKent Overstreet BUG_ON(!bio_add_page(&rbio->bio, page, PAGE_SIZE, 0)); 11151c6fdbd8SKent Overstreet 11160f238367SKent Overstreet bchfs_read(&trans, iter, rbio, inode->v.i_ino, 11170f238367SKent Overstreet &readpages_iter); 11181c6fdbd8SKent Overstreet } 11191c6fdbd8SKent Overstreet 11201c6fdbd8SKent Overstreet bch2_pagecache_add_put(&inode->ei_pagecache_lock); 1121424eb881SKent Overstreet 1122424eb881SKent Overstreet bch2_trans_exit(&trans); 11231c6fdbd8SKent Overstreet kfree(readpages_iter.pages); 11241c6fdbd8SKent Overstreet } 11251c6fdbd8SKent Overstreet 11261c6fdbd8SKent Overstreet static void __bchfs_readpage(struct bch_fs *c, struct bch_read_bio *rbio, 11271c6fdbd8SKent Overstreet u64 inum, struct page *page) 11281c6fdbd8SKent Overstreet { 1129424eb881SKent Overstreet struct btree_trans trans; 1130424eb881SKent Overstreet struct btree_iter *iter; 11311c6fdbd8SKent Overstreet 1132f57a6a5dSKent Overstreet bch2_page_state_create(page, __GFP_NOFAIL); 11331c6fdbd8SKent Overstreet 11341c6fdbd8SKent Overstreet rbio->bio.bi_opf = REQ_OP_READ|REQ_SYNC; 11351c6fdbd8SKent Overstreet bio_add_page_contig(&rbio->bio, page); 11361c6fdbd8SKent Overstreet 113720bceecbSKent Overstreet bch2_trans_init(&trans, c, 0, 0); 1138424eb881SKent Overstreet iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS, POS_MIN, 11391c6fdbd8SKent Overstreet BTREE_ITER_SLOTS); 1140424eb881SKent Overstreet 11410f238367SKent Overstreet bchfs_read(&trans, iter, rbio, inum, NULL); 1142424eb881SKent Overstreet 1143424eb881SKent Overstreet bch2_trans_exit(&trans); 11441c6fdbd8SKent Overstreet } 11451c6fdbd8SKent Overstreet 11461c6fdbd8SKent Overstreet static void bch2_read_single_page_end_io(struct bio *bio) 11471c6fdbd8SKent Overstreet { 11481c6fdbd8SKent Overstreet complete(bio->bi_private); 11491c6fdbd8SKent Overstreet } 11501c6fdbd8SKent Overstreet 11511c6fdbd8SKent Overstreet static int bch2_read_single_page(struct page *page, 11521c6fdbd8SKent Overstreet struct address_space *mapping) 11531c6fdbd8SKent Overstreet { 11541c6fdbd8SKent Overstreet struct bch_inode_info *inode = to_bch_ei(mapping->host); 11551c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 11561c6fdbd8SKent Overstreet struct bch_read_bio *rbio; 11571c6fdbd8SKent Overstreet int ret; 11581c6fdbd8SKent Overstreet DECLARE_COMPLETION_ONSTACK(done); 11591c6fdbd8SKent Overstreet 11601c6fdbd8SKent Overstreet rbio = rbio_init(bio_alloc_bioset(NULL, 1, REQ_OP_READ, GFP_NOFS, &c->bio_read), 11611c6fdbd8SKent Overstreet io_opts(c, inode)); 11621c6fdbd8SKent Overstreet rbio->bio.bi_private = &done; 11631c6fdbd8SKent Overstreet rbio->bio.bi_end_io = bch2_read_single_page_end_io; 11641c6fdbd8SKent Overstreet 11651c6fdbd8SKent Overstreet __bchfs_readpage(c, rbio, inode->v.i_ino, page); 11661c6fdbd8SKent Overstreet wait_for_completion(&done); 11671c6fdbd8SKent Overstreet 11681c6fdbd8SKent Overstreet ret = blk_status_to_errno(rbio->bio.bi_status); 11691c6fdbd8SKent Overstreet bio_put(&rbio->bio); 11701c6fdbd8SKent Overstreet 11711c6fdbd8SKent Overstreet if (ret < 0) 11721c6fdbd8SKent Overstreet return ret; 11731c6fdbd8SKent Overstreet 11741c6fdbd8SKent Overstreet SetPageUptodate(page); 11751c6fdbd8SKent Overstreet return 0; 11761c6fdbd8SKent Overstreet } 11771c6fdbd8SKent Overstreet 11781c6fdbd8SKent Overstreet int bch2_read_folio(struct file *file, struct folio *folio) 11791c6fdbd8SKent Overstreet { 11801c6fdbd8SKent Overstreet struct page *page = &folio->page; 11811c6fdbd8SKent Overstreet int ret; 11821c6fdbd8SKent Overstreet 11831c6fdbd8SKent Overstreet ret = bch2_read_single_page(page, page->mapping); 11841c6fdbd8SKent Overstreet folio_unlock(folio); 11851c6fdbd8SKent Overstreet return ret; 11861c6fdbd8SKent Overstreet } 11871c6fdbd8SKent Overstreet 11881c6fdbd8SKent Overstreet /* writepages: */ 11891c6fdbd8SKent Overstreet 11901c6fdbd8SKent Overstreet struct bch_writepage_state { 11911c6fdbd8SKent Overstreet struct bch_writepage_io *io; 11921c6fdbd8SKent Overstreet struct bch_io_opts opts; 11931c6fdbd8SKent Overstreet }; 11941c6fdbd8SKent Overstreet 11951c6fdbd8SKent Overstreet static inline struct bch_writepage_state bch_writepage_state_init(struct bch_fs *c, 11961c6fdbd8SKent Overstreet struct bch_inode_info *inode) 11971c6fdbd8SKent Overstreet { 11981c6fdbd8SKent Overstreet return (struct bch_writepage_state) { .opts = io_opts(c, inode) }; 11991c6fdbd8SKent Overstreet } 12001c6fdbd8SKent Overstreet 12011c6fdbd8SKent Overstreet static void bch2_writepage_io_free(struct closure *cl) 12021c6fdbd8SKent Overstreet { 12031c6fdbd8SKent Overstreet struct bch_writepage_io *io = container_of(cl, 12041c6fdbd8SKent Overstreet struct bch_writepage_io, cl); 12051c6fdbd8SKent Overstreet 12061c6fdbd8SKent Overstreet bio_put(&io->op.op.wbio.bio); 12071c6fdbd8SKent Overstreet } 12081c6fdbd8SKent Overstreet 12091c6fdbd8SKent Overstreet static void bch2_writepage_io_done(struct closure *cl) 12101c6fdbd8SKent Overstreet { 12111c6fdbd8SKent Overstreet struct bch_writepage_io *io = container_of(cl, 12121c6fdbd8SKent Overstreet struct bch_writepage_io, cl); 12131c6fdbd8SKent Overstreet struct bch_fs *c = io->op.op.c; 12141c6fdbd8SKent Overstreet struct bio *bio = &io->op.op.wbio.bio; 12151c6fdbd8SKent Overstreet struct bvec_iter_all iter; 12161c6fdbd8SKent Overstreet struct bio_vec *bvec; 12171c6fdbd8SKent Overstreet 12181c6fdbd8SKent Overstreet if (io->op.op.error) { 121975812e70SKent Overstreet bio_for_each_segment_all(bvec, bio, iter) { 12201c6fdbd8SKent Overstreet SetPageError(bvec->bv_page); 122175812e70SKent Overstreet mapping_set_error(bvec->bv_page->mapping, -EIO); 122275812e70SKent Overstreet } 12231c6fdbd8SKent Overstreet } 12241c6fdbd8SKent Overstreet 12251c6fdbd8SKent Overstreet /* 12261c6fdbd8SKent Overstreet * racing with fallocate can cause us to add fewer sectors than 12271c6fdbd8SKent Overstreet * expected - but we shouldn't add more sectors than expected: 12281c6fdbd8SKent Overstreet */ 12291c6fdbd8SKent Overstreet BUG_ON(io->op.sectors_added > (s64) io->new_sectors); 12301c6fdbd8SKent Overstreet 12311c6fdbd8SKent Overstreet /* 12321c6fdbd8SKent Overstreet * (error (due to going RO) halfway through a page can screw that up 12331c6fdbd8SKent Overstreet * slightly) 12341c6fdbd8SKent Overstreet * XXX wtf? 12351c6fdbd8SKent Overstreet BUG_ON(io->op.sectors_added - io->new_sectors >= (s64) PAGE_SECTORS); 12361c6fdbd8SKent Overstreet */ 12371c6fdbd8SKent Overstreet 12381c6fdbd8SKent Overstreet /* 12391c6fdbd8SKent Overstreet * PageWriteback is effectively our ref on the inode - fixup i_blocks 12401c6fdbd8SKent Overstreet * before calling end_page_writeback: 12411c6fdbd8SKent Overstreet */ 12421c6fdbd8SKent Overstreet if (io->op.sectors_added != io->new_sectors) 12431c6fdbd8SKent Overstreet i_sectors_acct(c, io->op.inode, NULL, 12441c6fdbd8SKent Overstreet io->op.sectors_added - (s64) io->new_sectors); 12451c6fdbd8SKent Overstreet 12461c6fdbd8SKent Overstreet bio_for_each_segment_all(bvec, bio, iter) 12471c6fdbd8SKent Overstreet end_page_writeback(bvec->bv_page); 12481c6fdbd8SKent Overstreet 12491c6fdbd8SKent Overstreet closure_return_with_destructor(&io->cl, bch2_writepage_io_free); 12501c6fdbd8SKent Overstreet } 12511c6fdbd8SKent Overstreet 12521c6fdbd8SKent Overstreet static void bch2_writepage_do_io(struct bch_writepage_state *w) 12531c6fdbd8SKent Overstreet { 12541c6fdbd8SKent Overstreet struct bch_writepage_io *io = w->io; 12551c6fdbd8SKent Overstreet 12561c6fdbd8SKent Overstreet w->io = NULL; 12571c6fdbd8SKent Overstreet closure_call(&io->op.op.cl, bch2_write, NULL, &io->cl); 12581c6fdbd8SKent Overstreet continue_at(&io->cl, bch2_writepage_io_done, NULL); 12591c6fdbd8SKent Overstreet } 12601c6fdbd8SKent Overstreet 12611c6fdbd8SKent Overstreet /* 12621c6fdbd8SKent Overstreet * Get a bch_writepage_io and add @page to it - appending to an existing one if 12631c6fdbd8SKent Overstreet * possible, else allocating a new one: 12641c6fdbd8SKent Overstreet */ 12651c6fdbd8SKent Overstreet static void bch2_writepage_io_alloc(struct bch_fs *c, 12661c6fdbd8SKent Overstreet struct bch_writepage_state *w, 12671c6fdbd8SKent Overstreet struct bch_inode_info *inode, 12681c6fdbd8SKent Overstreet struct page *page, 12691c6fdbd8SKent Overstreet unsigned nr_replicas) 12701c6fdbd8SKent Overstreet { 12711c6fdbd8SKent Overstreet struct bch_write_op *op; 12721c6fdbd8SKent Overstreet u64 offset = (u64) page->index << PAGE_SECTOR_SHIFT; 12731c6fdbd8SKent Overstreet 12741c6fdbd8SKent Overstreet w->io = container_of(bio_alloc_bioset(NULL, BIO_MAX_VECS, 12751c6fdbd8SKent Overstreet REQ_OP_WRITE, 12761c6fdbd8SKent Overstreet GFP_NOFS, 12771c6fdbd8SKent Overstreet &c->writepage_bioset), 12781c6fdbd8SKent Overstreet struct bch_writepage_io, op.op.wbio.bio); 12791c6fdbd8SKent Overstreet 12801c6fdbd8SKent Overstreet closure_init(&w->io->cl, NULL); 12811c6fdbd8SKent Overstreet w->io->new_sectors = 0; 12821c6fdbd8SKent Overstreet bch2_fswrite_op_init(&w->io->op, c, inode, w->opts, false); 12831c6fdbd8SKent Overstreet op = &w->io->op.op; 12841c6fdbd8SKent Overstreet op->nr_replicas = nr_replicas; 12851c6fdbd8SKent Overstreet op->res.nr_replicas = nr_replicas; 12861c6fdbd8SKent Overstreet op->write_point = writepoint_hashed(inode->ei_last_dirtied); 12871c6fdbd8SKent Overstreet op->pos = POS(inode->v.i_ino, offset); 12881c6fdbd8SKent Overstreet op->wbio.bio.bi_iter.bi_sector = offset; 12891c6fdbd8SKent Overstreet } 12901c6fdbd8SKent Overstreet 12911c6fdbd8SKent Overstreet static int __bch2_writepage(struct folio *folio, 12921c6fdbd8SKent Overstreet struct writeback_control *wbc, 12931c6fdbd8SKent Overstreet void *data) 12941c6fdbd8SKent Overstreet { 12951c6fdbd8SKent Overstreet struct page *page = &folio->page; 12961c6fdbd8SKent Overstreet struct bch_inode_info *inode = to_bch_ei(page->mapping->host); 12971c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 12981c6fdbd8SKent Overstreet struct bch_writepage_state *w = data; 1299e1036a2aSKent Overstreet struct bch_page_state *s; 1300f57a6a5dSKent Overstreet unsigned offset, nr_replicas_this_write = U32_MAX; 1301f57a6a5dSKent Overstreet unsigned dirty_sectors = 0, reserved_sectors = 0; 13021c6fdbd8SKent Overstreet loff_t i_size = i_size_read(&inode->v); 13031c6fdbd8SKent Overstreet pgoff_t end_index = i_size >> PAGE_SHIFT; 1304f57a6a5dSKent Overstreet unsigned i; 1305e1036a2aSKent Overstreet int ret; 13061c6fdbd8SKent Overstreet 13071c6fdbd8SKent Overstreet EBUG_ON(!PageUptodate(page)); 13081c6fdbd8SKent Overstreet 13091c6fdbd8SKent Overstreet /* Is the page fully inside i_size? */ 13101c6fdbd8SKent Overstreet if (page->index < end_index) 13111c6fdbd8SKent Overstreet goto do_io; 13121c6fdbd8SKent Overstreet 13131c6fdbd8SKent Overstreet /* Is the page fully outside i_size? (truncate in progress) */ 13141c6fdbd8SKent Overstreet offset = i_size & (PAGE_SIZE - 1); 13151c6fdbd8SKent Overstreet if (page->index > end_index || !offset) { 13161c6fdbd8SKent Overstreet unlock_page(page); 13171c6fdbd8SKent Overstreet return 0; 13181c6fdbd8SKent Overstreet } 13191c6fdbd8SKent Overstreet 13201c6fdbd8SKent Overstreet /* 13211c6fdbd8SKent Overstreet * The page straddles i_size. It must be zeroed out on each and every 13221c6fdbd8SKent Overstreet * writepage invocation because it may be mmapped. "A file is mapped 13231c6fdbd8SKent Overstreet * in multiples of the page size. For a file that is not a multiple of 13241c6fdbd8SKent Overstreet * the page size, the remaining memory is zeroed when mapped, and 13251c6fdbd8SKent Overstreet * writes to that region are not written out to the file." 13261c6fdbd8SKent Overstreet */ 13271c6fdbd8SKent Overstreet zero_user_segment(page, offset, PAGE_SIZE); 13281c6fdbd8SKent Overstreet do_io: 1329f57a6a5dSKent Overstreet s = bch2_page_state_create(page, __GFP_NOFAIL); 1330f81b648dSKent Overstreet 1331f57a6a5dSKent Overstreet ret = bch2_get_page_disk_reservation(c, inode, page, true); 1332e1036a2aSKent Overstreet if (ret) { 1333e1036a2aSKent Overstreet SetPageError(page); 1334e1036a2aSKent Overstreet mapping_set_error(page->mapping, ret); 1335e1036a2aSKent Overstreet unlock_page(page); 1336e1036a2aSKent Overstreet return 0; 1337e1036a2aSKent Overstreet } 1338e1036a2aSKent Overstreet 1339f57a6a5dSKent Overstreet for (i = 0; i < PAGE_SECTORS; i++) 1340f81b648dSKent Overstreet nr_replicas_this_write = 1341f57a6a5dSKent Overstreet min_t(unsigned, nr_replicas_this_write, 1342f57a6a5dSKent Overstreet s->s[i].nr_replicas + 1343f57a6a5dSKent Overstreet s->s[i].replicas_reserved); 1344f81b648dSKent Overstreet 1345e1036a2aSKent Overstreet /* Before unlocking the page, transfer reservation to w->io: */ 13461c6fdbd8SKent Overstreet 1347f57a6a5dSKent Overstreet for (i = 0; i < PAGE_SECTORS; i++) { 1348f57a6a5dSKent Overstreet s->s[i].nr_replicas = w->opts.compression 1349f57a6a5dSKent Overstreet ? 0 : nr_replicas_this_write; 1350e1036a2aSKent Overstreet 1351f57a6a5dSKent Overstreet reserved_sectors += s->s[i].replicas_reserved; 1352f57a6a5dSKent Overstreet s->s[i].replicas_reserved = 0; 1353f57a6a5dSKent Overstreet 1354f57a6a5dSKent Overstreet dirty_sectors += s->s[i].state == SECTOR_DIRTY; 1355f57a6a5dSKent Overstreet s->s[i].state = SECTOR_ALLOCATED; 1356f57a6a5dSKent Overstreet } 13571c6fdbd8SKent Overstreet 13581c6fdbd8SKent Overstreet BUG_ON(PageWriteback(page)); 13591c6fdbd8SKent Overstreet set_page_writeback(page); 13601c6fdbd8SKent Overstreet unlock_page(page); 13611c6fdbd8SKent Overstreet 13621c6fdbd8SKent Overstreet if (w->io && 1363f81b648dSKent Overstreet (w->io->op.op.res.nr_replicas != nr_replicas_this_write || 13641c6fdbd8SKent Overstreet !bio_can_add_page_contig(&w->io->op.op.wbio.bio, page))) 13651c6fdbd8SKent Overstreet bch2_writepage_do_io(w); 13661c6fdbd8SKent Overstreet 13671c6fdbd8SKent Overstreet if (!w->io) 1368f81b648dSKent Overstreet bch2_writepage_io_alloc(c, w, inode, page, 1369f81b648dSKent Overstreet nr_replicas_this_write); 13701c6fdbd8SKent Overstreet 1371e1036a2aSKent Overstreet w->io->new_sectors += dirty_sectors; 13721c6fdbd8SKent Overstreet 13731c6fdbd8SKent Overstreet BUG_ON(inode != w->io->op.inode); 13741c6fdbd8SKent Overstreet BUG_ON(bio_add_page_contig(&w->io->op.op.wbio.bio, page)); 13751c6fdbd8SKent Overstreet 1376f57a6a5dSKent Overstreet w->io->op.op.res.sectors += reserved_sectors; 13771c6fdbd8SKent Overstreet w->io->op.new_i_size = i_size; 13781c6fdbd8SKent Overstreet 13791c6fdbd8SKent Overstreet if (wbc->sync_mode == WB_SYNC_ALL) 13801c6fdbd8SKent Overstreet w->io->op.op.wbio.bio.bi_opf |= REQ_SYNC; 13811c6fdbd8SKent Overstreet 13821c6fdbd8SKent Overstreet return 0; 13831c6fdbd8SKent Overstreet } 13841c6fdbd8SKent Overstreet 13851c6fdbd8SKent Overstreet int bch2_writepages(struct address_space *mapping, struct writeback_control *wbc) 13861c6fdbd8SKent Overstreet { 13871c6fdbd8SKent Overstreet struct bch_fs *c = mapping->host->i_sb->s_fs_info; 13881c6fdbd8SKent Overstreet struct bch_writepage_state w = 13891c6fdbd8SKent Overstreet bch_writepage_state_init(c, to_bch_ei(mapping->host)); 13901c6fdbd8SKent Overstreet struct blk_plug plug; 13911c6fdbd8SKent Overstreet int ret; 13921c6fdbd8SKent Overstreet 13931c6fdbd8SKent Overstreet blk_start_plug(&plug); 13941c6fdbd8SKent Overstreet ret = write_cache_pages(mapping, wbc, __bch2_writepage, &w); 13951c6fdbd8SKent Overstreet if (w.io) 13961c6fdbd8SKent Overstreet bch2_writepage_do_io(&w); 13971c6fdbd8SKent Overstreet blk_finish_plug(&plug); 13981c6fdbd8SKent Overstreet return ret; 13991c6fdbd8SKent Overstreet } 14001c6fdbd8SKent Overstreet 14011c6fdbd8SKent Overstreet int bch2_writepage(struct page *page, struct writeback_control *wbc) 14021c6fdbd8SKent Overstreet { 14031c6fdbd8SKent Overstreet struct bch_fs *c = page->mapping->host->i_sb->s_fs_info; 14041c6fdbd8SKent Overstreet struct bch_writepage_state w = 14051c6fdbd8SKent Overstreet bch_writepage_state_init(c, to_bch_ei(page->mapping->host)); 14061c6fdbd8SKent Overstreet int ret; 14071c6fdbd8SKent Overstreet 14081c6fdbd8SKent Overstreet ret = __bch2_writepage(page_folio(page), wbc, &w); 14091c6fdbd8SKent Overstreet if (w.io) 14101c6fdbd8SKent Overstreet bch2_writepage_do_io(&w); 14111c6fdbd8SKent Overstreet 14121c6fdbd8SKent Overstreet return ret; 14131c6fdbd8SKent Overstreet } 14141c6fdbd8SKent Overstreet 14151c6fdbd8SKent Overstreet /* buffered writes: */ 14161c6fdbd8SKent Overstreet 14171c6fdbd8SKent Overstreet int bch2_write_begin(struct file *file, struct address_space *mapping, 14181c6fdbd8SKent Overstreet loff_t pos, unsigned len, 14191c6fdbd8SKent Overstreet struct page **pagep, void **fsdata) 14201c6fdbd8SKent Overstreet { 14211c6fdbd8SKent Overstreet struct bch_inode_info *inode = to_bch_ei(mapping->host); 14221c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 14231c6fdbd8SKent Overstreet pgoff_t index = pos >> PAGE_SHIFT; 14241c6fdbd8SKent Overstreet unsigned offset = pos & (PAGE_SIZE - 1); 14251c6fdbd8SKent Overstreet struct page *page; 14261c6fdbd8SKent Overstreet int ret = -ENOMEM; 14271c6fdbd8SKent Overstreet 14281c6fdbd8SKent Overstreet BUG_ON(inode_unhashed(&inode->v)); 14291c6fdbd8SKent Overstreet 14301c6fdbd8SKent Overstreet bch2_pagecache_add_get(&inode->ei_pagecache_lock); 14311c6fdbd8SKent Overstreet 14321c6fdbd8SKent Overstreet page = grab_cache_page_write_begin(mapping, index); 14331c6fdbd8SKent Overstreet if (!page) 14341c6fdbd8SKent Overstreet goto err_unlock; 14351c6fdbd8SKent Overstreet 14361c6fdbd8SKent Overstreet if (PageUptodate(page)) 14371c6fdbd8SKent Overstreet goto out; 14381c6fdbd8SKent Overstreet 14391c6fdbd8SKent Overstreet /* If we're writing entire page, don't need to read it in first: */ 14401c6fdbd8SKent Overstreet if (len == PAGE_SIZE) 14411c6fdbd8SKent Overstreet goto out; 14421c6fdbd8SKent Overstreet 14431c6fdbd8SKent Overstreet if (!offset && pos + len >= inode->v.i_size) { 14441c6fdbd8SKent Overstreet zero_user_segment(page, len, PAGE_SIZE); 14451c6fdbd8SKent Overstreet flush_dcache_page(page); 14461c6fdbd8SKent Overstreet goto out; 14471c6fdbd8SKent Overstreet } 14481c6fdbd8SKent Overstreet 14491c6fdbd8SKent Overstreet if (index > inode->v.i_size >> PAGE_SHIFT) { 14501c6fdbd8SKent Overstreet zero_user_segments(page, 0, offset, offset + len, PAGE_SIZE); 14511c6fdbd8SKent Overstreet flush_dcache_page(page); 14521c6fdbd8SKent Overstreet goto out; 14531c6fdbd8SKent Overstreet } 14541c6fdbd8SKent Overstreet readpage: 14551c6fdbd8SKent Overstreet ret = bch2_read_single_page(page, mapping); 14561c6fdbd8SKent Overstreet if (ret) 14571c6fdbd8SKent Overstreet goto err; 14581c6fdbd8SKent Overstreet out: 14591c6fdbd8SKent Overstreet ret = bch2_get_page_reservation(c, inode, page, true); 14601c6fdbd8SKent Overstreet if (ret) { 14611c6fdbd8SKent Overstreet if (!PageUptodate(page)) { 14621c6fdbd8SKent Overstreet /* 14631c6fdbd8SKent Overstreet * If the page hasn't been read in, we won't know if we 14641c6fdbd8SKent Overstreet * actually need a reservation - we don't actually need 14651c6fdbd8SKent Overstreet * to read here, we just need to check if the page is 14661c6fdbd8SKent Overstreet * fully backed by uncompressed data: 14671c6fdbd8SKent Overstreet */ 14681c6fdbd8SKent Overstreet goto readpage; 14691c6fdbd8SKent Overstreet } 14701c6fdbd8SKent Overstreet 14711c6fdbd8SKent Overstreet goto err; 14721c6fdbd8SKent Overstreet } 14731c6fdbd8SKent Overstreet 14741c6fdbd8SKent Overstreet *pagep = page; 14751c6fdbd8SKent Overstreet return 0; 14761c6fdbd8SKent Overstreet err: 14771c6fdbd8SKent Overstreet unlock_page(page); 14781c6fdbd8SKent Overstreet put_page(page); 14791c6fdbd8SKent Overstreet *pagep = NULL; 14801c6fdbd8SKent Overstreet err_unlock: 14811c6fdbd8SKent Overstreet bch2_pagecache_add_put(&inode->ei_pagecache_lock); 14821c6fdbd8SKent Overstreet return ret; 14831c6fdbd8SKent Overstreet } 14841c6fdbd8SKent Overstreet 14851c6fdbd8SKent Overstreet int bch2_write_end(struct file *file, struct address_space *mapping, 14861c6fdbd8SKent Overstreet loff_t pos, unsigned len, unsigned copied, 14871c6fdbd8SKent Overstreet struct page *page, void *fsdata) 14881c6fdbd8SKent Overstreet { 14891c6fdbd8SKent Overstreet struct bch_inode_info *inode = to_bch_ei(mapping->host); 14901c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 14911c6fdbd8SKent Overstreet 14921c6fdbd8SKent Overstreet lockdep_assert_held(&inode->v.i_rwsem); 14931c6fdbd8SKent Overstreet 14941c6fdbd8SKent Overstreet if (unlikely(copied < len && !PageUptodate(page))) { 14951c6fdbd8SKent Overstreet /* 14961c6fdbd8SKent Overstreet * The page needs to be read in, but that would destroy 14971c6fdbd8SKent Overstreet * our partial write - simplest thing is to just force 14981c6fdbd8SKent Overstreet * userspace to redo the write: 14991c6fdbd8SKent Overstreet */ 15001c6fdbd8SKent Overstreet zero_user(page, 0, PAGE_SIZE); 15011c6fdbd8SKent Overstreet flush_dcache_page(page); 15021c6fdbd8SKent Overstreet copied = 0; 15031c6fdbd8SKent Overstreet } 15041c6fdbd8SKent Overstreet 15051c6fdbd8SKent Overstreet spin_lock(&inode->v.i_lock); 15061c6fdbd8SKent Overstreet if (pos + copied > inode->v.i_size) 15071c6fdbd8SKent Overstreet i_size_write(&inode->v, pos + copied); 15081c6fdbd8SKent Overstreet spin_unlock(&inode->v.i_lock); 15091c6fdbd8SKent Overstreet 15101c6fdbd8SKent Overstreet if (copied) { 15111c6fdbd8SKent Overstreet if (!PageUptodate(page)) 15121c6fdbd8SKent Overstreet SetPageUptodate(page); 15131c6fdbd8SKent Overstreet if (!PageDirty(page)) 1514e1036a2aSKent Overstreet bch2_set_page_dirty(mapping, page); 15151c6fdbd8SKent Overstreet 15161c6fdbd8SKent Overstreet inode->ei_last_dirtied = (unsigned long) current; 15171c6fdbd8SKent Overstreet } else { 15181c6fdbd8SKent Overstreet bch2_put_page_reservation(c, inode, page); 15191c6fdbd8SKent Overstreet } 15201c6fdbd8SKent Overstreet 15211c6fdbd8SKent Overstreet unlock_page(page); 15221c6fdbd8SKent Overstreet put_page(page); 15231c6fdbd8SKent Overstreet bch2_pagecache_add_put(&inode->ei_pagecache_lock); 15241c6fdbd8SKent Overstreet 15251c6fdbd8SKent Overstreet return copied; 15261c6fdbd8SKent Overstreet } 15271c6fdbd8SKent Overstreet 15281c6fdbd8SKent Overstreet #define WRITE_BATCH_PAGES 32 15291c6fdbd8SKent Overstreet 15301c6fdbd8SKent Overstreet static int __bch2_buffered_write(struct bch_inode_info *inode, 15311c6fdbd8SKent Overstreet struct address_space *mapping, 15321c6fdbd8SKent Overstreet struct iov_iter *iter, 15331c6fdbd8SKent Overstreet loff_t pos, unsigned len) 15341c6fdbd8SKent Overstreet { 15351c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 15361c6fdbd8SKent Overstreet struct page *pages[WRITE_BATCH_PAGES]; 15371c6fdbd8SKent Overstreet unsigned long index = pos >> PAGE_SHIFT; 15381c6fdbd8SKent Overstreet unsigned offset = pos & (PAGE_SIZE - 1); 15391c6fdbd8SKent Overstreet unsigned nr_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE); 15401c6fdbd8SKent Overstreet unsigned i, copied = 0, nr_pages_copied = 0; 15411c6fdbd8SKent Overstreet int ret = 0; 15421c6fdbd8SKent Overstreet 15431c6fdbd8SKent Overstreet BUG_ON(!len); 15441c6fdbd8SKent Overstreet BUG_ON(nr_pages > ARRAY_SIZE(pages)); 15451c6fdbd8SKent Overstreet 15461c6fdbd8SKent Overstreet for (i = 0; i < nr_pages; i++) { 15471c6fdbd8SKent Overstreet pages[i] = grab_cache_page_write_begin(mapping, index + i); 15481c6fdbd8SKent Overstreet if (!pages[i]) { 15491c6fdbd8SKent Overstreet nr_pages = i; 15501c6fdbd8SKent Overstreet ret = -ENOMEM; 15511c6fdbd8SKent Overstreet goto out; 15521c6fdbd8SKent Overstreet } 15531c6fdbd8SKent Overstreet } 15541c6fdbd8SKent Overstreet 15551c6fdbd8SKent Overstreet if (offset && !PageUptodate(pages[0])) { 15561c6fdbd8SKent Overstreet ret = bch2_read_single_page(pages[0], mapping); 15571c6fdbd8SKent Overstreet if (ret) 15581c6fdbd8SKent Overstreet goto out; 15591c6fdbd8SKent Overstreet } 15601c6fdbd8SKent Overstreet 15611c6fdbd8SKent Overstreet if ((pos + len) & (PAGE_SIZE - 1) && 15621c6fdbd8SKent Overstreet !PageUptodate(pages[nr_pages - 1])) { 15631c6fdbd8SKent Overstreet if ((index + nr_pages - 1) << PAGE_SHIFT >= inode->v.i_size) { 15641c6fdbd8SKent Overstreet zero_user(pages[nr_pages - 1], 0, PAGE_SIZE); 15651c6fdbd8SKent Overstreet } else { 15661c6fdbd8SKent Overstreet ret = bch2_read_single_page(pages[nr_pages - 1], mapping); 15671c6fdbd8SKent Overstreet if (ret) 15681c6fdbd8SKent Overstreet goto out; 15691c6fdbd8SKent Overstreet } 15701c6fdbd8SKent Overstreet } 15711c6fdbd8SKent Overstreet 15721c6fdbd8SKent Overstreet for (i = 0; i < nr_pages; i++) { 15731c6fdbd8SKent Overstreet ret = bch2_get_page_reservation(c, inode, pages[i], true); 15741c6fdbd8SKent Overstreet 15751c6fdbd8SKent Overstreet if (ret && !PageUptodate(pages[i])) { 15761c6fdbd8SKent Overstreet ret = bch2_read_single_page(pages[i], mapping); 15771c6fdbd8SKent Overstreet if (ret) 15781c6fdbd8SKent Overstreet goto out; 15791c6fdbd8SKent Overstreet 15801c6fdbd8SKent Overstreet ret = bch2_get_page_reservation(c, inode, pages[i], true); 15811c6fdbd8SKent Overstreet } 15821c6fdbd8SKent Overstreet 15831c6fdbd8SKent Overstreet if (ret) 15841c6fdbd8SKent Overstreet goto out; 15851c6fdbd8SKent Overstreet } 15861c6fdbd8SKent Overstreet 15871c6fdbd8SKent Overstreet if (mapping_writably_mapped(mapping)) 15881c6fdbd8SKent Overstreet for (i = 0; i < nr_pages; i++) 15891c6fdbd8SKent Overstreet flush_dcache_page(pages[i]); 15901c6fdbd8SKent Overstreet 15911c6fdbd8SKent Overstreet while (copied < len) { 15921c6fdbd8SKent Overstreet struct page *page = pages[(offset + copied) >> PAGE_SHIFT]; 15931c6fdbd8SKent Overstreet unsigned pg_offset = (offset + copied) & (PAGE_SIZE - 1); 15941c6fdbd8SKent Overstreet unsigned pg_bytes = min_t(unsigned, len - copied, 15951c6fdbd8SKent Overstreet PAGE_SIZE - pg_offset); 15961c6fdbd8SKent Overstreet unsigned pg_copied = copy_page_from_iter_atomic(page, 15971c6fdbd8SKent Overstreet pg_offset, pg_bytes, iter); 15981c6fdbd8SKent Overstreet 15991c6fdbd8SKent Overstreet flush_dcache_page(page); 16001c6fdbd8SKent Overstreet copied += pg_copied; 16011c6fdbd8SKent Overstreet 16021c6fdbd8SKent Overstreet if (pg_copied != pg_bytes) 16031c6fdbd8SKent Overstreet break; 16041c6fdbd8SKent Overstreet } 16051c6fdbd8SKent Overstreet 16061c6fdbd8SKent Overstreet if (!copied) 16071c6fdbd8SKent Overstreet goto out; 16081c6fdbd8SKent Overstreet 16091c6fdbd8SKent Overstreet nr_pages_copied = DIV_ROUND_UP(offset + copied, PAGE_SIZE); 16101c6fdbd8SKent Overstreet inode->ei_last_dirtied = (unsigned long) current; 16111c6fdbd8SKent Overstreet 16121c6fdbd8SKent Overstreet spin_lock(&inode->v.i_lock); 16131c6fdbd8SKent Overstreet if (pos + copied > inode->v.i_size) 16141c6fdbd8SKent Overstreet i_size_write(&inode->v, pos + copied); 16151c6fdbd8SKent Overstreet spin_unlock(&inode->v.i_lock); 16161c6fdbd8SKent Overstreet 16171c6fdbd8SKent Overstreet if (copied < len && 16181c6fdbd8SKent Overstreet ((offset + copied) & (PAGE_SIZE - 1))) { 16191c6fdbd8SKent Overstreet struct page *page = pages[(offset + copied) >> PAGE_SHIFT]; 16201c6fdbd8SKent Overstreet 16211c6fdbd8SKent Overstreet if (!PageUptodate(page)) { 16221c6fdbd8SKent Overstreet zero_user(page, 0, PAGE_SIZE); 16231c6fdbd8SKent Overstreet copied -= (offset + copied) & (PAGE_SIZE - 1); 16241c6fdbd8SKent Overstreet } 16251c6fdbd8SKent Overstreet } 16261c6fdbd8SKent Overstreet out: 16271c6fdbd8SKent Overstreet for (i = 0; i < nr_pages_copied; i++) { 16281c6fdbd8SKent Overstreet if (!PageUptodate(pages[i])) 16291c6fdbd8SKent Overstreet SetPageUptodate(pages[i]); 16301c6fdbd8SKent Overstreet if (!PageDirty(pages[i])) 1631e1036a2aSKent Overstreet bch2_set_page_dirty(mapping, pages[i]); 16321c6fdbd8SKent Overstreet unlock_page(pages[i]); 16331c6fdbd8SKent Overstreet put_page(pages[i]); 16341c6fdbd8SKent Overstreet } 16351c6fdbd8SKent Overstreet 16361c6fdbd8SKent Overstreet for (i = nr_pages_copied; i < nr_pages; i++) { 16371c6fdbd8SKent Overstreet if (!PageDirty(pages[i])) 16381c6fdbd8SKent Overstreet bch2_put_page_reservation(c, inode, pages[i]); 16391c6fdbd8SKent Overstreet unlock_page(pages[i]); 16401c6fdbd8SKent Overstreet put_page(pages[i]); 16411c6fdbd8SKent Overstreet } 16421c6fdbd8SKent Overstreet 16431c6fdbd8SKent Overstreet return copied ?: ret; 16441c6fdbd8SKent Overstreet } 16451c6fdbd8SKent Overstreet 16461c6fdbd8SKent Overstreet static ssize_t bch2_buffered_write(struct kiocb *iocb, struct iov_iter *iter) 16471c6fdbd8SKent Overstreet { 16481c6fdbd8SKent Overstreet struct file *file = iocb->ki_filp; 16491c6fdbd8SKent Overstreet struct address_space *mapping = file->f_mapping; 16501c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 16511c6fdbd8SKent Overstreet loff_t pos = iocb->ki_pos; 16521c6fdbd8SKent Overstreet ssize_t written = 0; 16531c6fdbd8SKent Overstreet int ret = 0; 16541c6fdbd8SKent Overstreet 16551c6fdbd8SKent Overstreet bch2_pagecache_add_get(&inode->ei_pagecache_lock); 16561c6fdbd8SKent Overstreet 16571c6fdbd8SKent Overstreet do { 16581c6fdbd8SKent Overstreet unsigned offset = pos & (PAGE_SIZE - 1); 16591c6fdbd8SKent Overstreet unsigned bytes = min_t(unsigned long, iov_iter_count(iter), 16601c6fdbd8SKent Overstreet PAGE_SIZE * WRITE_BATCH_PAGES - offset); 16611c6fdbd8SKent Overstreet again: 16621c6fdbd8SKent Overstreet /* 16631c6fdbd8SKent Overstreet * Bring in the user page that we will copy from _first_. 16641c6fdbd8SKent Overstreet * Otherwise there's a nasty deadlock on copying from the 16651c6fdbd8SKent Overstreet * same page as we're writing to, without it being marked 16661c6fdbd8SKent Overstreet * up-to-date. 16671c6fdbd8SKent Overstreet * 16681c6fdbd8SKent Overstreet * Not only is this an optimisation, but it is also required 16691c6fdbd8SKent Overstreet * to check that the address is actually valid, when atomic 16701c6fdbd8SKent Overstreet * usercopies are used, below. 16711c6fdbd8SKent Overstreet */ 16721c6fdbd8SKent Overstreet if (unlikely(fault_in_iov_iter_readable(iter, bytes))) { 16731c6fdbd8SKent Overstreet bytes = min_t(unsigned long, iov_iter_count(iter), 16741c6fdbd8SKent Overstreet PAGE_SIZE - offset); 16751c6fdbd8SKent Overstreet 16761c6fdbd8SKent Overstreet if (unlikely(fault_in_iov_iter_readable(iter, bytes))) { 16771c6fdbd8SKent Overstreet ret = -EFAULT; 16781c6fdbd8SKent Overstreet break; 16791c6fdbd8SKent Overstreet } 16801c6fdbd8SKent Overstreet } 16811c6fdbd8SKent Overstreet 16821c6fdbd8SKent Overstreet if (unlikely(fatal_signal_pending(current))) { 16831c6fdbd8SKent Overstreet ret = -EINTR; 16841c6fdbd8SKent Overstreet break; 16851c6fdbd8SKent Overstreet } 16861c6fdbd8SKent Overstreet 16871c6fdbd8SKent Overstreet ret = __bch2_buffered_write(inode, mapping, iter, pos, bytes); 16881c6fdbd8SKent Overstreet if (unlikely(ret < 0)) 16891c6fdbd8SKent Overstreet break; 16901c6fdbd8SKent Overstreet 16911c6fdbd8SKent Overstreet cond_resched(); 16921c6fdbd8SKent Overstreet 16931c6fdbd8SKent Overstreet if (unlikely(ret == 0)) { 16941c6fdbd8SKent Overstreet /* 16951c6fdbd8SKent Overstreet * If we were unable to copy any data at all, we must 16961c6fdbd8SKent Overstreet * fall back to a single segment length write. 16971c6fdbd8SKent Overstreet * 16981c6fdbd8SKent Overstreet * If we didn't fallback here, we could livelock 16991c6fdbd8SKent Overstreet * because not all segments in the iov can be copied at 17001c6fdbd8SKent Overstreet * once without a pagefault. 17011c6fdbd8SKent Overstreet */ 17021c6fdbd8SKent Overstreet bytes = min_t(unsigned long, PAGE_SIZE - offset, 17031c6fdbd8SKent Overstreet iov_iter_single_seg_count(iter)); 17041c6fdbd8SKent Overstreet goto again; 17051c6fdbd8SKent Overstreet } 17061c6fdbd8SKent Overstreet pos += ret; 17071c6fdbd8SKent Overstreet written += ret; 17081c6fdbd8SKent Overstreet 17091c6fdbd8SKent Overstreet balance_dirty_pages_ratelimited(mapping); 17101c6fdbd8SKent Overstreet } while (iov_iter_count(iter)); 17111c6fdbd8SKent Overstreet 17121c6fdbd8SKent Overstreet bch2_pagecache_add_put(&inode->ei_pagecache_lock); 17131c6fdbd8SKent Overstreet 17141c6fdbd8SKent Overstreet return written ? written : ret; 17151c6fdbd8SKent Overstreet } 17161c6fdbd8SKent Overstreet 17171c6fdbd8SKent Overstreet /* O_DIRECT reads */ 17181c6fdbd8SKent Overstreet 17191c6fdbd8SKent Overstreet static void bch2_dio_read_complete(struct closure *cl) 17201c6fdbd8SKent Overstreet { 17211c6fdbd8SKent Overstreet struct dio_read *dio = container_of(cl, struct dio_read, cl); 17221c6fdbd8SKent Overstreet 17231c6fdbd8SKent Overstreet dio->req->ki_complete(dio->req, dio->ret); 17241c6fdbd8SKent Overstreet bio_check_pages_dirty(&dio->rbio.bio); /* transfers ownership */ 17251c6fdbd8SKent Overstreet } 17261c6fdbd8SKent Overstreet 17271c6fdbd8SKent Overstreet static void bch2_direct_IO_read_endio(struct bio *bio) 17281c6fdbd8SKent Overstreet { 17291c6fdbd8SKent Overstreet struct dio_read *dio = bio->bi_private; 17301c6fdbd8SKent Overstreet 17311c6fdbd8SKent Overstreet if (bio->bi_status) 17321c6fdbd8SKent Overstreet dio->ret = blk_status_to_errno(bio->bi_status); 17331c6fdbd8SKent Overstreet 17341c6fdbd8SKent Overstreet closure_put(&dio->cl); 17351c6fdbd8SKent Overstreet } 17361c6fdbd8SKent Overstreet 17371c6fdbd8SKent Overstreet static void bch2_direct_IO_read_split_endio(struct bio *bio) 17381c6fdbd8SKent Overstreet { 17391c6fdbd8SKent Overstreet bch2_direct_IO_read_endio(bio); 17401c6fdbd8SKent Overstreet bio_check_pages_dirty(bio); /* transfers ownership */ 17411c6fdbd8SKent Overstreet } 17421c6fdbd8SKent Overstreet 17431c6fdbd8SKent Overstreet static int bch2_direct_IO_read(struct kiocb *req, struct iov_iter *iter) 17441c6fdbd8SKent Overstreet { 17451c6fdbd8SKent Overstreet struct file *file = req->ki_filp; 17461c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 17471c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 17481c6fdbd8SKent Overstreet struct bch_io_opts opts = io_opts(c, inode); 17491c6fdbd8SKent Overstreet struct dio_read *dio; 17501c6fdbd8SKent Overstreet struct bio *bio; 17511c6fdbd8SKent Overstreet loff_t offset = req->ki_pos; 17521c6fdbd8SKent Overstreet bool sync = is_sync_kiocb(req); 17531c6fdbd8SKent Overstreet size_t shorten; 17541c6fdbd8SKent Overstreet ssize_t ret; 17551c6fdbd8SKent Overstreet 17561c6fdbd8SKent Overstreet if ((offset|iter->count) & (block_bytes(c) - 1)) 17571c6fdbd8SKent Overstreet return -EINVAL; 17581c6fdbd8SKent Overstreet 17591c6fdbd8SKent Overstreet ret = min_t(loff_t, iter->count, 17601c6fdbd8SKent Overstreet max_t(loff_t, 0, i_size_read(&inode->v) - offset)); 17611c6fdbd8SKent Overstreet 17621c6fdbd8SKent Overstreet if (!ret) 17631c6fdbd8SKent Overstreet return ret; 17641c6fdbd8SKent Overstreet 17651c6fdbd8SKent Overstreet shorten = iov_iter_count(iter) - round_up(ret, block_bytes(c)); 17661c6fdbd8SKent Overstreet iter->count -= shorten; 17671c6fdbd8SKent Overstreet 17681c6fdbd8SKent Overstreet bio = bio_alloc_bioset(NULL, 17691c6fdbd8SKent Overstreet iov_iter_npages(iter, BIO_MAX_VECS), 17701c6fdbd8SKent Overstreet REQ_OP_READ, 17711c6fdbd8SKent Overstreet GFP_KERNEL, 17721c6fdbd8SKent Overstreet &c->dio_read_bioset); 17731c6fdbd8SKent Overstreet 17741c6fdbd8SKent Overstreet bio->bi_end_io = bch2_direct_IO_read_endio; 17751c6fdbd8SKent Overstreet 17761c6fdbd8SKent Overstreet dio = container_of(bio, struct dio_read, rbio.bio); 17771c6fdbd8SKent Overstreet closure_init(&dio->cl, NULL); 17781c6fdbd8SKent Overstreet 17791c6fdbd8SKent Overstreet /* 17801c6fdbd8SKent Overstreet * this is a _really_ horrible hack just to avoid an atomic sub at the 17811c6fdbd8SKent Overstreet * end: 17821c6fdbd8SKent Overstreet */ 17831c6fdbd8SKent Overstreet if (!sync) { 17841c6fdbd8SKent Overstreet set_closure_fn(&dio->cl, bch2_dio_read_complete, NULL); 17851c6fdbd8SKent Overstreet atomic_set(&dio->cl.remaining, 17861c6fdbd8SKent Overstreet CLOSURE_REMAINING_INITIALIZER - 17871c6fdbd8SKent Overstreet CLOSURE_RUNNING + 17881c6fdbd8SKent Overstreet CLOSURE_DESTRUCTOR); 17891c6fdbd8SKent Overstreet } else { 17901c6fdbd8SKent Overstreet atomic_set(&dio->cl.remaining, 17911c6fdbd8SKent Overstreet CLOSURE_REMAINING_INITIALIZER + 1); 17921c6fdbd8SKent Overstreet } 17931c6fdbd8SKent Overstreet 17941c6fdbd8SKent Overstreet dio->req = req; 17951c6fdbd8SKent Overstreet dio->ret = ret; 17961c6fdbd8SKent Overstreet 17971c6fdbd8SKent Overstreet goto start; 17981c6fdbd8SKent Overstreet while (iter->count) { 17991c6fdbd8SKent Overstreet bio = bio_alloc_bioset(NULL, 18001c6fdbd8SKent Overstreet iov_iter_npages(iter, BIO_MAX_VECS), 18011c6fdbd8SKent Overstreet REQ_OP_READ, 18021c6fdbd8SKent Overstreet GFP_KERNEL, 18031c6fdbd8SKent Overstreet &c->bio_read); 18041c6fdbd8SKent Overstreet bio->bi_end_io = bch2_direct_IO_read_split_endio; 18051c6fdbd8SKent Overstreet start: 18061c6fdbd8SKent Overstreet bio->bi_opf = REQ_OP_READ|REQ_SYNC; 18071c6fdbd8SKent Overstreet bio->bi_iter.bi_sector = offset >> 9; 18081c6fdbd8SKent Overstreet bio->bi_private = dio; 18091c6fdbd8SKent Overstreet 18101c6fdbd8SKent Overstreet ret = bio_iov_iter_get_pages(bio, iter); 18111c6fdbd8SKent Overstreet if (ret < 0) { 18121c6fdbd8SKent Overstreet /* XXX: fault inject this path */ 18131c6fdbd8SKent Overstreet bio->bi_status = BLK_STS_RESOURCE; 18141c6fdbd8SKent Overstreet bio_endio(bio); 18151c6fdbd8SKent Overstreet break; 18161c6fdbd8SKent Overstreet } 18171c6fdbd8SKent Overstreet 18181c6fdbd8SKent Overstreet offset += bio->bi_iter.bi_size; 18191c6fdbd8SKent Overstreet bio_set_pages_dirty(bio); 18201c6fdbd8SKent Overstreet 18211c6fdbd8SKent Overstreet if (iter->count) 18221c6fdbd8SKent Overstreet closure_get(&dio->cl); 18231c6fdbd8SKent Overstreet 18241c6fdbd8SKent Overstreet bch2_read(c, rbio_init(bio, opts), inode->v.i_ino); 18251c6fdbd8SKent Overstreet } 18261c6fdbd8SKent Overstreet 18271c6fdbd8SKent Overstreet iter->count += shorten; 18281c6fdbd8SKent Overstreet 18291c6fdbd8SKent Overstreet if (sync) { 18301c6fdbd8SKent Overstreet closure_sync(&dio->cl); 18311c6fdbd8SKent Overstreet closure_debug_destroy(&dio->cl); 18321c6fdbd8SKent Overstreet ret = dio->ret; 18331c6fdbd8SKent Overstreet bio_check_pages_dirty(&dio->rbio.bio); /* transfers ownership */ 18341c6fdbd8SKent Overstreet return ret; 18351c6fdbd8SKent Overstreet } else { 18361c6fdbd8SKent Overstreet return -EIOCBQUEUED; 18371c6fdbd8SKent Overstreet } 18381c6fdbd8SKent Overstreet } 18391c6fdbd8SKent Overstreet 18401c6fdbd8SKent Overstreet ssize_t bch2_read_iter(struct kiocb *iocb, struct iov_iter *iter) 18411c6fdbd8SKent Overstreet { 18421c6fdbd8SKent Overstreet struct file *file = iocb->ki_filp; 18431c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 18441c6fdbd8SKent Overstreet struct address_space *mapping = file->f_mapping; 18451c6fdbd8SKent Overstreet size_t count = iov_iter_count(iter); 18461c6fdbd8SKent Overstreet ssize_t ret; 18471c6fdbd8SKent Overstreet 18481c6fdbd8SKent Overstreet if (!count) 18491c6fdbd8SKent Overstreet return 0; /* skip atime */ 18501c6fdbd8SKent Overstreet 18511c6fdbd8SKent Overstreet if (iocb->ki_flags & IOCB_DIRECT) { 18521c6fdbd8SKent Overstreet struct blk_plug plug; 18531c6fdbd8SKent Overstreet 18541c6fdbd8SKent Overstreet ret = filemap_write_and_wait_range(mapping, 18551c6fdbd8SKent Overstreet iocb->ki_pos, 18561c6fdbd8SKent Overstreet iocb->ki_pos + count - 1); 18571c6fdbd8SKent Overstreet if (ret < 0) 18581c6fdbd8SKent Overstreet return ret; 18591c6fdbd8SKent Overstreet 18601c6fdbd8SKent Overstreet file_accessed(file); 18611c6fdbd8SKent Overstreet 18621c6fdbd8SKent Overstreet blk_start_plug(&plug); 18631c6fdbd8SKent Overstreet ret = bch2_direct_IO_read(iocb, iter); 18641c6fdbd8SKent Overstreet blk_finish_plug(&plug); 18651c6fdbd8SKent Overstreet 18661c6fdbd8SKent Overstreet if (ret >= 0) 18671c6fdbd8SKent Overstreet iocb->ki_pos += ret; 18681c6fdbd8SKent Overstreet } else { 18691c6fdbd8SKent Overstreet bch2_pagecache_add_get(&inode->ei_pagecache_lock); 18701c6fdbd8SKent Overstreet ret = generic_file_read_iter(iocb, iter); 18711c6fdbd8SKent Overstreet bch2_pagecache_add_put(&inode->ei_pagecache_lock); 18721c6fdbd8SKent Overstreet } 18731c6fdbd8SKent Overstreet 18741c6fdbd8SKent Overstreet return ret; 18751c6fdbd8SKent Overstreet } 18761c6fdbd8SKent Overstreet 18771c6fdbd8SKent Overstreet /* O_DIRECT writes */ 18781c6fdbd8SKent Overstreet 18791c6fdbd8SKent Overstreet /* 18801c6fdbd8SKent Overstreet * We're going to return -EIOCBQUEUED, but we haven't finished consuming the 18811c6fdbd8SKent Overstreet * iov_iter yet, so we need to stash a copy of the iovec: it might be on the 18821c6fdbd8SKent Overstreet * caller's stack, we're not guaranteed that it will live for the duration of 18831c6fdbd8SKent Overstreet * the IO: 18841c6fdbd8SKent Overstreet */ 18851c6fdbd8SKent Overstreet static noinline int bch2_dio_write_copy_iov(struct dio_write *dio) 18861c6fdbd8SKent Overstreet { 18871c6fdbd8SKent Overstreet struct iovec *iov = dio->inline_vecs; 18881c6fdbd8SKent Overstreet 18891c6fdbd8SKent Overstreet /* 18901c6fdbd8SKent Overstreet * iov_iter has a single embedded iovec - nothing to do: 18911c6fdbd8SKent Overstreet */ 18921c6fdbd8SKent Overstreet if (iter_is_ubuf(&dio->iter)) 18931c6fdbd8SKent Overstreet return 0; 18941c6fdbd8SKent Overstreet 18951c6fdbd8SKent Overstreet /* 18961c6fdbd8SKent Overstreet * We don't currently handle non-iovec iov_iters here - return an error, 18971c6fdbd8SKent Overstreet * and we'll fall back to doing the IO synchronously: 18981c6fdbd8SKent Overstreet */ 18991c6fdbd8SKent Overstreet if (!iter_is_iovec(&dio->iter)) 19001c6fdbd8SKent Overstreet return -1; 19011c6fdbd8SKent Overstreet 19021c6fdbd8SKent Overstreet if (dio->iter.nr_segs > ARRAY_SIZE(dio->inline_vecs)) { 19031c6fdbd8SKent Overstreet iov = kmalloc_array(dio->iter.nr_segs, sizeof(*iov), 19041c6fdbd8SKent Overstreet GFP_KERNEL); 19051c6fdbd8SKent Overstreet if (unlikely(!iov)) 19061c6fdbd8SKent Overstreet return -ENOMEM; 19071c6fdbd8SKent Overstreet 19081c6fdbd8SKent Overstreet dio->free_iov = true; 19091c6fdbd8SKent Overstreet } 19101c6fdbd8SKent Overstreet 19111c6fdbd8SKent Overstreet memcpy(iov, dio->iter.__iov, dio->iter.nr_segs * sizeof(*iov)); 19121c6fdbd8SKent Overstreet dio->iter.__iov = iov; 19131c6fdbd8SKent Overstreet return 0; 19141c6fdbd8SKent Overstreet } 19151c6fdbd8SKent Overstreet 19161c6fdbd8SKent Overstreet static void bch2_dio_write_loop_async(struct closure *); 19171c6fdbd8SKent Overstreet 19181c6fdbd8SKent Overstreet static long bch2_dio_write_loop(struct dio_write *dio) 19191c6fdbd8SKent Overstreet { 1920ed484030SKent Overstreet bool kthread = (current->flags & PF_KTHREAD) != 0; 19211c6fdbd8SKent Overstreet struct kiocb *req = dio->req; 19221c6fdbd8SKent Overstreet struct address_space *mapping = req->ki_filp->f_mapping; 19231c6fdbd8SKent Overstreet struct bch_inode_info *inode = dio->iop.inode; 19241c6fdbd8SKent Overstreet struct bio *bio = &dio->iop.op.wbio.bio; 19251c6fdbd8SKent Overstreet struct bvec_iter_all iter; 19261c6fdbd8SKent Overstreet struct bio_vec *bv; 1927919dbbd1SKent Overstreet loff_t offset; 19281c6fdbd8SKent Overstreet bool sync; 19291c6fdbd8SKent Overstreet long ret; 19301c6fdbd8SKent Overstreet 19311c6fdbd8SKent Overstreet if (dio->loop) 19321c6fdbd8SKent Overstreet goto loop; 19331c6fdbd8SKent Overstreet 19341c6fdbd8SKent Overstreet inode_dio_begin(&inode->v); 19351c6fdbd8SKent Overstreet bch2_pagecache_block_get(&inode->ei_pagecache_lock); 19361c6fdbd8SKent Overstreet 19371c6fdbd8SKent Overstreet /* Write and invalidate pagecache range that we're writing to: */ 1938919dbbd1SKent Overstreet offset = req->ki_pos + (dio->iop.op.written << 9); 1939919dbbd1SKent Overstreet ret = write_invalidate_inode_pages_range(mapping, 1940919dbbd1SKent Overstreet offset, 1941919dbbd1SKent Overstreet offset + iov_iter_count(&dio->iter) - 1); 19421c6fdbd8SKent Overstreet if (unlikely(ret)) 19431c6fdbd8SKent Overstreet goto err; 19441c6fdbd8SKent Overstreet 19451c6fdbd8SKent Overstreet while (1) { 1946919dbbd1SKent Overstreet offset = req->ki_pos + (dio->iop.op.written << 9); 1947919dbbd1SKent Overstreet 1948ed484030SKent Overstreet if (kthread) 1949ed484030SKent Overstreet kthread_use_mm(dio->mm); 19501c6fdbd8SKent Overstreet BUG_ON(current->faults_disabled_mapping); 19511c6fdbd8SKent Overstreet current->faults_disabled_mapping = mapping; 19521c6fdbd8SKent Overstreet 19531c6fdbd8SKent Overstreet ret = bio_iov_iter_get_pages(bio, &dio->iter); 19541c6fdbd8SKent Overstreet 19551c6fdbd8SKent Overstreet current->faults_disabled_mapping = NULL; 1956ed484030SKent Overstreet if (kthread) 1957ed484030SKent Overstreet kthread_unuse_mm(dio->mm); 19581c6fdbd8SKent Overstreet 19591c6fdbd8SKent Overstreet if (unlikely(ret < 0)) 19601c6fdbd8SKent Overstreet goto err; 19611c6fdbd8SKent Overstreet 19621c6fdbd8SKent Overstreet /* gup might have faulted pages back in: */ 19631c6fdbd8SKent Overstreet ret = write_invalidate_inode_pages_range(mapping, 1964919dbbd1SKent Overstreet offset, 1965919dbbd1SKent Overstreet offset + bio->bi_iter.bi_size - 1); 19661c6fdbd8SKent Overstreet if (unlikely(ret)) 19671c6fdbd8SKent Overstreet goto err; 19681c6fdbd8SKent Overstreet 1969919dbbd1SKent Overstreet dio->iop.op.pos = POS(inode->v.i_ino, offset >> 9); 19701c6fdbd8SKent Overstreet 19711c6fdbd8SKent Overstreet task_io_account_write(bio->bi_iter.bi_size); 19721c6fdbd8SKent Overstreet 19731c6fdbd8SKent Overstreet closure_call(&dio->iop.op.cl, bch2_write, NULL, &dio->cl); 19741c6fdbd8SKent Overstreet 19751c6fdbd8SKent Overstreet if (!dio->sync && !dio->loop && dio->iter.count) { 19761c6fdbd8SKent Overstreet if (bch2_dio_write_copy_iov(dio)) { 19771c6fdbd8SKent Overstreet dio->iop.op.error = -ENOMEM; 19781c6fdbd8SKent Overstreet goto err_wait_io; 19791c6fdbd8SKent Overstreet } 19801c6fdbd8SKent Overstreet } 19811c6fdbd8SKent Overstreet err_wait_io: 19821c6fdbd8SKent Overstreet dio->loop = true; 19831c6fdbd8SKent Overstreet 19841c6fdbd8SKent Overstreet if (!dio->sync) { 19851c6fdbd8SKent Overstreet continue_at(&dio->cl, bch2_dio_write_loop_async, NULL); 19861c6fdbd8SKent Overstreet return -EIOCBQUEUED; 19871c6fdbd8SKent Overstreet } 19881c6fdbd8SKent Overstreet 19891c6fdbd8SKent Overstreet closure_sync(&dio->cl); 19901c6fdbd8SKent Overstreet loop: 19911c6fdbd8SKent Overstreet bio_for_each_segment_all(bv, bio, iter) 19921c6fdbd8SKent Overstreet put_page(bv->bv_page); 19931c6fdbd8SKent Overstreet if (!dio->iter.count || dio->iop.op.error) 19941c6fdbd8SKent Overstreet break; 19951c6fdbd8SKent Overstreet bio_reset(bio, NULL, REQ_OP_WRITE); 19961c6fdbd8SKent Overstreet } 19971c6fdbd8SKent Overstreet 19981c6fdbd8SKent Overstreet ret = dio->iop.op.error ?: ((long) dio->iop.op.written << 9); 19991c6fdbd8SKent Overstreet err: 20001c6fdbd8SKent Overstreet bch2_pagecache_block_put(&inode->ei_pagecache_lock); 20011c6fdbd8SKent Overstreet bch2_disk_reservation_put(dio->iop.op.c, &dio->iop.op.res); 20021c6fdbd8SKent Overstreet bch2_quota_reservation_put(dio->iop.op.c, inode, &dio->quota_res); 20031c6fdbd8SKent Overstreet 20041c6fdbd8SKent Overstreet if (dio->free_iov) 20051c6fdbd8SKent Overstreet kfree(dio->iter.__iov); 20061c6fdbd8SKent Overstreet 20071c6fdbd8SKent Overstreet closure_debug_destroy(&dio->cl); 20081c6fdbd8SKent Overstreet 20091c6fdbd8SKent Overstreet sync = dio->sync; 20101c6fdbd8SKent Overstreet bio_put(bio); 20111c6fdbd8SKent Overstreet 20121c6fdbd8SKent Overstreet /* inode->i_dio_count is our ref on inode and thus bch_fs */ 20131c6fdbd8SKent Overstreet inode_dio_end(&inode->v); 20141c6fdbd8SKent Overstreet 20151c6fdbd8SKent Overstreet if (!sync) { 20161c6fdbd8SKent Overstreet req->ki_complete(req, ret); 20171c6fdbd8SKent Overstreet ret = -EIOCBQUEUED; 20181c6fdbd8SKent Overstreet } 20191c6fdbd8SKent Overstreet return ret; 20201c6fdbd8SKent Overstreet } 20211c6fdbd8SKent Overstreet 20221c6fdbd8SKent Overstreet static void bch2_dio_write_loop_async(struct closure *cl) 20231c6fdbd8SKent Overstreet { 20241c6fdbd8SKent Overstreet struct dio_write *dio = container_of(cl, struct dio_write, cl); 20251c6fdbd8SKent Overstreet 20261c6fdbd8SKent Overstreet bch2_dio_write_loop(dio); 20271c6fdbd8SKent Overstreet } 20281c6fdbd8SKent Overstreet 20291c6fdbd8SKent Overstreet static noinline 20301c6fdbd8SKent Overstreet ssize_t bch2_direct_write(struct kiocb *req, struct iov_iter *iter) 20311c6fdbd8SKent Overstreet { 20321c6fdbd8SKent Overstreet struct file *file = req->ki_filp; 20331c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 20341c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 20351c6fdbd8SKent Overstreet struct dio_write *dio; 20361c6fdbd8SKent Overstreet struct bio *bio; 20371c6fdbd8SKent Overstreet ssize_t ret; 20381c6fdbd8SKent Overstreet 20391c6fdbd8SKent Overstreet lockdep_assert_held(&inode->v.i_rwsem); 20401c6fdbd8SKent Overstreet 20411c6fdbd8SKent Overstreet if (unlikely(!iter->count)) 20421c6fdbd8SKent Overstreet return 0; 20431c6fdbd8SKent Overstreet 2044919dbbd1SKent Overstreet if (unlikely((req->ki_pos|iter->count) & (block_bytes(c) - 1))) 20451c6fdbd8SKent Overstreet return -EINVAL; 20461c6fdbd8SKent Overstreet 20471c6fdbd8SKent Overstreet bio = bio_alloc_bioset(NULL, 20481c6fdbd8SKent Overstreet iov_iter_npages(iter, BIO_MAX_VECS), 20491c6fdbd8SKent Overstreet REQ_OP_WRITE, 20501c6fdbd8SKent Overstreet GFP_KERNEL, 20511c6fdbd8SKent Overstreet &c->dio_write_bioset); 20521c6fdbd8SKent Overstreet dio = container_of(bio, struct dio_write, iop.op.wbio.bio); 20531c6fdbd8SKent Overstreet closure_init(&dio->cl, NULL); 20541c6fdbd8SKent Overstreet dio->req = req; 2055ed484030SKent Overstreet dio->mm = current->mm; 20561c6fdbd8SKent Overstreet dio->loop = false; 20571c6fdbd8SKent Overstreet dio->sync = is_sync_kiocb(req) || 2058919dbbd1SKent Overstreet req->ki_pos + iter->count > inode->v.i_size; 20591c6fdbd8SKent Overstreet dio->free_iov = false; 20601c6fdbd8SKent Overstreet dio->quota_res.sectors = 0; 20611c6fdbd8SKent Overstreet dio->iter = *iter; 20621c6fdbd8SKent Overstreet bch2_fswrite_op_init(&dio->iop, c, inode, io_opts(c, inode), true); 2063ed484030SKent Overstreet dio->iop.op.write_point = writepoint_hashed((unsigned long) current); 20641c6fdbd8SKent Overstreet dio->iop.op.flags |= BCH_WRITE_NOPUT_RESERVATION; 20651c6fdbd8SKent Overstreet 20661c6fdbd8SKent Overstreet if ((req->ki_flags & IOCB_DSYNC) && 20671c6fdbd8SKent Overstreet !c->opts.journal_flush_disabled) 20681c6fdbd8SKent Overstreet dio->iop.op.flags |= BCH_WRITE_FLUSH; 20691c6fdbd8SKent Overstreet 20701c6fdbd8SKent Overstreet ret = bch2_quota_reservation_add(c, inode, &dio->quota_res, 20711c6fdbd8SKent Overstreet iter->count >> 9, true); 20721c6fdbd8SKent Overstreet if (unlikely(ret)) 20731c6fdbd8SKent Overstreet goto err; 20741c6fdbd8SKent Overstreet 20757ef2a73aSKent Overstreet dio->iop.op.nr_replicas = dio->iop.op.opts.data_replicas; 20767ef2a73aSKent Overstreet 20771c6fdbd8SKent Overstreet ret = bch2_disk_reservation_get(c, &dio->iop.op.res, iter->count >> 9, 20781c6fdbd8SKent Overstreet dio->iop.op.opts.data_replicas, 0); 20791c6fdbd8SKent Overstreet if (unlikely(ret)) { 20807ef2a73aSKent Overstreet if (!bch2_check_range_allocated(c, POS(inode->v.i_ino, 2081919dbbd1SKent Overstreet req->ki_pos >> 9), 20827ef2a73aSKent Overstreet iter->count >> 9, 20837ef2a73aSKent Overstreet dio->iop.op.opts.data_replicas)) 20841c6fdbd8SKent Overstreet goto err; 20851c6fdbd8SKent Overstreet 20861c6fdbd8SKent Overstreet dio->iop.unalloc = true; 20871c6fdbd8SKent Overstreet } 20881c6fdbd8SKent Overstreet 20891c6fdbd8SKent Overstreet return bch2_dio_write_loop(dio); 20901c6fdbd8SKent Overstreet err: 20911c6fdbd8SKent Overstreet bch2_disk_reservation_put(c, &dio->iop.op.res); 20921c6fdbd8SKent Overstreet bch2_quota_reservation_put(c, inode, &dio->quota_res); 20931c6fdbd8SKent Overstreet closure_debug_destroy(&dio->cl); 20941c6fdbd8SKent Overstreet bio_put(bio); 20951c6fdbd8SKent Overstreet return ret; 20961c6fdbd8SKent Overstreet } 20971c6fdbd8SKent Overstreet 20981c6fdbd8SKent Overstreet static ssize_t __bch2_write_iter(struct kiocb *iocb, struct iov_iter *from) 20991c6fdbd8SKent Overstreet { 21001c6fdbd8SKent Overstreet struct file *file = iocb->ki_filp; 21011c6fdbd8SKent Overstreet ssize_t ret; 21021c6fdbd8SKent Overstreet 21031c6fdbd8SKent Overstreet if (iocb->ki_flags & IOCB_DIRECT) 21041c6fdbd8SKent Overstreet return bch2_direct_write(iocb, from); 21051c6fdbd8SKent Overstreet 21061c6fdbd8SKent Overstreet ret = file_remove_privs(file); 21071c6fdbd8SKent Overstreet if (ret) 21081c6fdbd8SKent Overstreet return ret; 21091c6fdbd8SKent Overstreet 21101c6fdbd8SKent Overstreet ret = file_update_time(file); 21111c6fdbd8SKent Overstreet if (ret) 21121c6fdbd8SKent Overstreet return ret; 21131c6fdbd8SKent Overstreet 21141c6fdbd8SKent Overstreet ret = iocb->ki_flags & IOCB_DIRECT 21151c6fdbd8SKent Overstreet ? bch2_direct_write(iocb, from) 21161c6fdbd8SKent Overstreet : bch2_buffered_write(iocb, from); 21171c6fdbd8SKent Overstreet 21181c6fdbd8SKent Overstreet if (likely(ret > 0)) 21191c6fdbd8SKent Overstreet iocb->ki_pos += ret; 21201c6fdbd8SKent Overstreet 21211c6fdbd8SKent Overstreet return ret; 21221c6fdbd8SKent Overstreet } 21231c6fdbd8SKent Overstreet 21241c6fdbd8SKent Overstreet ssize_t bch2_write_iter(struct kiocb *iocb, struct iov_iter *from) 21251c6fdbd8SKent Overstreet { 21261c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(iocb->ki_filp); 21271c6fdbd8SKent Overstreet bool direct = iocb->ki_flags & IOCB_DIRECT; 21281c6fdbd8SKent Overstreet ssize_t ret; 21291c6fdbd8SKent Overstreet 21301c6fdbd8SKent Overstreet inode_lock(&inode->v); 21311c6fdbd8SKent Overstreet ret = generic_write_checks(iocb, from); 21321c6fdbd8SKent Overstreet if (ret > 0) 21331c6fdbd8SKent Overstreet ret = __bch2_write_iter(iocb, from); 21341c6fdbd8SKent Overstreet inode_unlock(&inode->v); 21351c6fdbd8SKent Overstreet 21361c6fdbd8SKent Overstreet if (ret > 0 && !direct) 21371c6fdbd8SKent Overstreet ret = generic_write_sync(iocb, ret); 21381c6fdbd8SKent Overstreet 21391c6fdbd8SKent Overstreet return ret; 21401c6fdbd8SKent Overstreet } 21411c6fdbd8SKent Overstreet 21421c6fdbd8SKent Overstreet /* fsync: */ 21431c6fdbd8SKent Overstreet 21441c6fdbd8SKent Overstreet int bch2_fsync(struct file *file, loff_t start, loff_t end, int datasync) 21451c6fdbd8SKent Overstreet { 21461c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 21471c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 2148bb1b3658SKent Overstreet int ret, ret2; 21491c6fdbd8SKent Overstreet 21501c6fdbd8SKent Overstreet ret = file_write_and_wait_range(file, start, end); 21511c6fdbd8SKent Overstreet if (ret) 21521c6fdbd8SKent Overstreet return ret; 21531c6fdbd8SKent Overstreet 21541c6fdbd8SKent Overstreet if (datasync && !(inode->v.i_state & I_DIRTY_DATASYNC)) 21551c6fdbd8SKent Overstreet goto out; 21561c6fdbd8SKent Overstreet 21571c6fdbd8SKent Overstreet ret = sync_inode_metadata(&inode->v, 1); 21581c6fdbd8SKent Overstreet if (ret) 21591c6fdbd8SKent Overstreet return ret; 21601c6fdbd8SKent Overstreet out: 216175812e70SKent Overstreet if (!c->opts.journal_flush_disabled) 216275812e70SKent Overstreet ret = bch2_journal_flush_seq(&c->journal, 216375812e70SKent Overstreet inode->ei_journal_seq); 2164bb1b3658SKent Overstreet ret2 = file_check_and_advance_wb_err(file); 2165bb1b3658SKent Overstreet 2166bb1b3658SKent Overstreet return ret ?: ret2; 21671c6fdbd8SKent Overstreet } 21681c6fdbd8SKent Overstreet 21691c6fdbd8SKent Overstreet /* truncate: */ 21701c6fdbd8SKent Overstreet 217154e2264eSKent Overstreet static int __bch2_fpunch(struct bch_fs *c, struct bch_inode_info *inode, 217254e2264eSKent Overstreet u64 start_offset, u64 end_offset, u64 *journal_seq) 217354e2264eSKent Overstreet { 217454e2264eSKent Overstreet struct bpos start = POS(inode->v.i_ino, start_offset); 217554e2264eSKent Overstreet struct bpos end = POS(inode->v.i_ino, end_offset); 217654e2264eSKent Overstreet unsigned max_sectors = KEY_SIZE_MAX & (~0 << c->block_bits); 217754e2264eSKent Overstreet struct btree_trans trans; 217854e2264eSKent Overstreet struct btree_iter *iter; 217954e2264eSKent Overstreet struct bkey_s_c k; 218054e2264eSKent Overstreet int ret = 0; 218154e2264eSKent Overstreet 218220bceecbSKent Overstreet bch2_trans_init(&trans, c, BTREE_ITER_MAX, 1024); 218354e2264eSKent Overstreet 218454e2264eSKent Overstreet iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS, start, 218554e2264eSKent Overstreet BTREE_ITER_INTENT); 218654e2264eSKent Overstreet 218754e2264eSKent Overstreet while ((k = bch2_btree_iter_peek(iter)).k && 21880f238367SKent Overstreet !(ret = bkey_err(k)) && 218954e2264eSKent Overstreet bkey_cmp(iter->pos, end) < 0) { 219054e2264eSKent Overstreet struct disk_reservation disk_res = 219154e2264eSKent Overstreet bch2_disk_reservation_init(c, 0); 219254e2264eSKent Overstreet struct bkey_i delete; 219354e2264eSKent Overstreet 219454e2264eSKent Overstreet bkey_init(&delete.k); 219554e2264eSKent Overstreet delete.k.p = iter->pos; 219654e2264eSKent Overstreet 219754e2264eSKent Overstreet /* create the biggest key we can */ 219854e2264eSKent Overstreet bch2_key_resize(&delete.k, max_sectors); 219954e2264eSKent Overstreet bch2_cut_back(end, &delete.k); 220054e2264eSKent Overstreet 2201b17657d0SKent Overstreet bch2_trans_begin_updates(&trans); 2202b17657d0SKent Overstreet 220354e2264eSKent Overstreet ret = bch2_extent_update(&trans, inode, 220454e2264eSKent Overstreet &disk_res, NULL, iter, &delete, 220554e2264eSKent Overstreet 0, true, true, NULL); 220654e2264eSKent Overstreet bch2_disk_reservation_put(c, &disk_res); 220754e2264eSKent Overstreet 220854e2264eSKent Overstreet if (ret == -EINTR) 220954e2264eSKent Overstreet ret = 0; 221054e2264eSKent Overstreet if (ret) 221154e2264eSKent Overstreet break; 221254e2264eSKent Overstreet 2213424eb881SKent Overstreet bch2_trans_cond_resched(&trans); 221454e2264eSKent Overstreet } 221554e2264eSKent Overstreet 221654e2264eSKent Overstreet bch2_trans_exit(&trans); 221754e2264eSKent Overstreet 221854e2264eSKent Overstreet return ret; 221954e2264eSKent Overstreet } 222054e2264eSKent Overstreet 22211c6fdbd8SKent Overstreet static inline int range_has_data(struct bch_fs *c, 22221c6fdbd8SKent Overstreet struct bpos start, 22231c6fdbd8SKent Overstreet struct bpos end) 22241c6fdbd8SKent Overstreet { 2225424eb881SKent Overstreet struct btree_trans trans; 2226424eb881SKent Overstreet struct btree_iter *iter; 22271c6fdbd8SKent Overstreet struct bkey_s_c k; 22281c6fdbd8SKent Overstreet int ret = 0; 22291c6fdbd8SKent Overstreet 223020bceecbSKent Overstreet bch2_trans_init(&trans, c, 0, 0); 2231424eb881SKent Overstreet 223294f651e2SKent Overstreet for_each_btree_key(&trans, iter, BTREE_ID_EXTENTS, start, 0, k, ret) { 22331c6fdbd8SKent Overstreet if (bkey_cmp(bkey_start_pos(k.k), end) >= 0) 22341c6fdbd8SKent Overstreet break; 22351c6fdbd8SKent Overstreet 22361c6fdbd8SKent Overstreet if (bkey_extent_is_data(k.k)) { 22371c6fdbd8SKent Overstreet ret = 1; 22381c6fdbd8SKent Overstreet break; 22391c6fdbd8SKent Overstreet } 22401c6fdbd8SKent Overstreet } 22411c6fdbd8SKent Overstreet 2242424eb881SKent Overstreet return bch2_trans_exit(&trans) ?: ret; 22431c6fdbd8SKent Overstreet } 22441c6fdbd8SKent Overstreet 22451c6fdbd8SKent Overstreet static int __bch2_truncate_page(struct bch_inode_info *inode, 22461c6fdbd8SKent Overstreet pgoff_t index, loff_t start, loff_t end) 22471c6fdbd8SKent Overstreet { 22481c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 22491c6fdbd8SKent Overstreet struct address_space *mapping = inode->v.i_mapping; 22501c6fdbd8SKent Overstreet unsigned start_offset = start & (PAGE_SIZE - 1); 22511c6fdbd8SKent Overstreet unsigned end_offset = ((end - 1) & (PAGE_SIZE - 1)) + 1; 22521c6fdbd8SKent Overstreet struct page *page; 22531c6fdbd8SKent Overstreet int ret = 0; 22541c6fdbd8SKent Overstreet 22551c6fdbd8SKent Overstreet /* Page boundary? Nothing to do */ 22561c6fdbd8SKent Overstreet if (!((index == start >> PAGE_SHIFT && start_offset) || 22571c6fdbd8SKent Overstreet (index == end >> PAGE_SHIFT && end_offset != PAGE_SIZE))) 22581c6fdbd8SKent Overstreet return 0; 22591c6fdbd8SKent Overstreet 22601c6fdbd8SKent Overstreet /* Above i_size? */ 22611c6fdbd8SKent Overstreet if (index << PAGE_SHIFT >= inode->v.i_size) 22621c6fdbd8SKent Overstreet return 0; 22631c6fdbd8SKent Overstreet 22641c6fdbd8SKent Overstreet page = find_lock_page(mapping, index); 22651c6fdbd8SKent Overstreet if (!page) { 22661c6fdbd8SKent Overstreet /* 22671c6fdbd8SKent Overstreet * XXX: we're doing two index lookups when we end up reading the 22681c6fdbd8SKent Overstreet * page 22691c6fdbd8SKent Overstreet */ 22701c6fdbd8SKent Overstreet ret = range_has_data(c, 22711c6fdbd8SKent Overstreet POS(inode->v.i_ino, index << PAGE_SECTOR_SHIFT), 22721c6fdbd8SKent Overstreet POS(inode->v.i_ino, (index + 1) << PAGE_SECTOR_SHIFT)); 22731c6fdbd8SKent Overstreet if (ret <= 0) 22741c6fdbd8SKent Overstreet return ret; 22751c6fdbd8SKent Overstreet 22761c6fdbd8SKent Overstreet page = find_or_create_page(mapping, index, GFP_KERNEL); 22771c6fdbd8SKent Overstreet if (unlikely(!page)) { 22781c6fdbd8SKent Overstreet ret = -ENOMEM; 22791c6fdbd8SKent Overstreet goto out; 22801c6fdbd8SKent Overstreet } 22811c6fdbd8SKent Overstreet } 22821c6fdbd8SKent Overstreet 22831c6fdbd8SKent Overstreet if (!PageUptodate(page)) { 22841c6fdbd8SKent Overstreet ret = bch2_read_single_page(page, mapping); 22851c6fdbd8SKent Overstreet if (ret) 22861c6fdbd8SKent Overstreet goto unlock; 22871c6fdbd8SKent Overstreet } 22881c6fdbd8SKent Overstreet 22891c6fdbd8SKent Overstreet /* 22901c6fdbd8SKent Overstreet * Bit of a hack - we don't want truncate to fail due to -ENOSPC. 22911c6fdbd8SKent Overstreet * 22921c6fdbd8SKent Overstreet * XXX: because we aren't currently tracking whether the page has actual 22931c6fdbd8SKent Overstreet * data in it (vs. just 0s, or only partially written) this wrong. ick. 22941c6fdbd8SKent Overstreet */ 22951c6fdbd8SKent Overstreet ret = bch2_get_page_reservation(c, inode, page, false); 22961c6fdbd8SKent Overstreet BUG_ON(ret); 22971c6fdbd8SKent Overstreet 22981c6fdbd8SKent Overstreet if (index == start >> PAGE_SHIFT && 22991c6fdbd8SKent Overstreet index == end >> PAGE_SHIFT) 23001c6fdbd8SKent Overstreet zero_user_segment(page, start_offset, end_offset); 23011c6fdbd8SKent Overstreet else if (index == start >> PAGE_SHIFT) 23021c6fdbd8SKent Overstreet zero_user_segment(page, start_offset, PAGE_SIZE); 23031c6fdbd8SKent Overstreet else if (index == end >> PAGE_SHIFT) 23041c6fdbd8SKent Overstreet zero_user_segment(page, 0, end_offset); 23051c6fdbd8SKent Overstreet 23061c6fdbd8SKent Overstreet if (!PageDirty(page)) 2307e1036a2aSKent Overstreet bch2_set_page_dirty(mapping, page); 23081c6fdbd8SKent Overstreet unlock: 23091c6fdbd8SKent Overstreet unlock_page(page); 23101c6fdbd8SKent Overstreet put_page(page); 23111c6fdbd8SKent Overstreet out: 23121c6fdbd8SKent Overstreet return ret; 23131c6fdbd8SKent Overstreet } 23141c6fdbd8SKent Overstreet 23151c6fdbd8SKent Overstreet static int bch2_truncate_page(struct bch_inode_info *inode, loff_t from) 23161c6fdbd8SKent Overstreet { 23171c6fdbd8SKent Overstreet return __bch2_truncate_page(inode, from >> PAGE_SHIFT, 23181c6fdbd8SKent Overstreet from, from + PAGE_SIZE); 23191c6fdbd8SKent Overstreet } 23201c6fdbd8SKent Overstreet 23211c6fdbd8SKent Overstreet static int bch2_extend(struct bch_inode_info *inode, struct iattr *iattr) 23221c6fdbd8SKent Overstreet { 23231c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 23241c6fdbd8SKent Overstreet struct address_space *mapping = inode->v.i_mapping; 23251c6fdbd8SKent Overstreet int ret; 23261c6fdbd8SKent Overstreet 23271c6fdbd8SKent Overstreet ret = filemap_write_and_wait_range(mapping, 23281c6fdbd8SKent Overstreet inode->ei_inode.bi_size, S64_MAX); 23291c6fdbd8SKent Overstreet if (ret) 23301c6fdbd8SKent Overstreet return ret; 23311c6fdbd8SKent Overstreet 23321c6fdbd8SKent Overstreet truncate_setsize(&inode->v, iattr->ia_size); 23331c6fdbd8SKent Overstreet /* ATTR_MODE will never be set here, ns argument isn't needed: */ 23341c6fdbd8SKent Overstreet setattr_copy(NULL, &inode->v, iattr); 23351c6fdbd8SKent Overstreet 23361c6fdbd8SKent Overstreet mutex_lock(&inode->ei_update_lock); 23372ea90048SKent Overstreet ret = bch2_write_inode_size(c, inode, inode->v.i_size, 23382ea90048SKent Overstreet ATTR_MTIME|ATTR_CTIME); 23391c6fdbd8SKent Overstreet mutex_unlock(&inode->ei_update_lock); 23401c6fdbd8SKent Overstreet 23411c6fdbd8SKent Overstreet return ret; 23421c6fdbd8SKent Overstreet } 23431c6fdbd8SKent Overstreet 234454e2264eSKent Overstreet static int bch2_truncate_finish_fn(struct bch_inode_info *inode, 234554e2264eSKent Overstreet struct bch_inode_unpacked *bi, 234654e2264eSKent Overstreet void *p) 234754e2264eSKent Overstreet { 234854e2264eSKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 234954e2264eSKent Overstreet 235054e2264eSKent Overstreet bi->bi_flags &= ~BCH_INODE_I_SIZE_DIRTY; 235154e2264eSKent Overstreet bi->bi_mtime = bi->bi_ctime = bch2_current_time(c); 235254e2264eSKent Overstreet return 0; 235354e2264eSKent Overstreet } 235454e2264eSKent Overstreet 235554e2264eSKent Overstreet static int bch2_truncate_start_fn(struct bch_inode_info *inode, 235654e2264eSKent Overstreet struct bch_inode_unpacked *bi, void *p) 235754e2264eSKent Overstreet { 235854e2264eSKent Overstreet u64 *new_i_size = p; 235954e2264eSKent Overstreet 236054e2264eSKent Overstreet bi->bi_flags |= BCH_INODE_I_SIZE_DIRTY; 236154e2264eSKent Overstreet bi->bi_size = *new_i_size; 236254e2264eSKent Overstreet return 0; 236354e2264eSKent Overstreet } 236454e2264eSKent Overstreet 23651c6fdbd8SKent Overstreet int bch2_truncate(struct bch_inode_info *inode, struct iattr *iattr) 23661c6fdbd8SKent Overstreet { 23671c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 23681c6fdbd8SKent Overstreet struct address_space *mapping = inode->v.i_mapping; 236954e2264eSKent Overstreet u64 new_i_size = iattr->ia_size; 23701c6fdbd8SKent Overstreet bool shrink; 23711c6fdbd8SKent Overstreet int ret = 0; 23721c6fdbd8SKent Overstreet 23731c6fdbd8SKent Overstreet inode_dio_wait(&inode->v); 23741c6fdbd8SKent Overstreet bch2_pagecache_block_get(&inode->ei_pagecache_lock); 23751c6fdbd8SKent Overstreet 23761c6fdbd8SKent Overstreet BUG_ON(inode->v.i_size < inode->ei_inode.bi_size); 23771c6fdbd8SKent Overstreet 23781c6fdbd8SKent Overstreet shrink = iattr->ia_size <= inode->v.i_size; 23791c6fdbd8SKent Overstreet 23801c6fdbd8SKent Overstreet if (!shrink) { 23811c6fdbd8SKent Overstreet ret = bch2_extend(inode, iattr); 238254e2264eSKent Overstreet goto err; 23831c6fdbd8SKent Overstreet } 23841c6fdbd8SKent Overstreet 23851c6fdbd8SKent Overstreet ret = bch2_truncate_page(inode, iattr->ia_size); 23861c6fdbd8SKent Overstreet if (unlikely(ret)) 238754e2264eSKent Overstreet goto err; 23881c6fdbd8SKent Overstreet 23891c6fdbd8SKent Overstreet if (iattr->ia_size > inode->ei_inode.bi_size) 23901c6fdbd8SKent Overstreet ret = filemap_write_and_wait_range(mapping, 23911c6fdbd8SKent Overstreet inode->ei_inode.bi_size, 23921c6fdbd8SKent Overstreet iattr->ia_size - 1); 23931c6fdbd8SKent Overstreet else if (iattr->ia_size & (PAGE_SIZE - 1)) 23941c6fdbd8SKent Overstreet ret = filemap_write_and_wait_range(mapping, 23951c6fdbd8SKent Overstreet round_down(iattr->ia_size, PAGE_SIZE), 23961c6fdbd8SKent Overstreet iattr->ia_size - 1); 23971c6fdbd8SKent Overstreet if (ret) 239854e2264eSKent Overstreet goto err; 23991c6fdbd8SKent Overstreet 240054e2264eSKent Overstreet mutex_lock(&inode->ei_update_lock); 240154e2264eSKent Overstreet ret = bch2_write_inode(c, inode, bch2_truncate_start_fn, 240254e2264eSKent Overstreet &new_i_size, 0); 240354e2264eSKent Overstreet mutex_unlock(&inode->ei_update_lock); 24041c6fdbd8SKent Overstreet 24051c6fdbd8SKent Overstreet if (unlikely(ret)) 240654e2264eSKent Overstreet goto err; 24071c6fdbd8SKent Overstreet 24081c6fdbd8SKent Overstreet truncate_setsize(&inode->v, iattr->ia_size); 24091c6fdbd8SKent Overstreet 241054e2264eSKent Overstreet /* 241154e2264eSKent Overstreet * XXX: need a comment explaining why PAGE_SIZE and not block_bytes() 241254e2264eSKent Overstreet * here: 241354e2264eSKent Overstreet */ 241454e2264eSKent Overstreet ret = __bch2_fpunch(c, inode, 24151c6fdbd8SKent Overstreet round_up(iattr->ia_size, PAGE_SIZE) >> 9, 241654e2264eSKent Overstreet U64_MAX, &inode->ei_journal_seq); 24171c6fdbd8SKent Overstreet if (unlikely(ret)) 241854e2264eSKent Overstreet goto err; 24191c6fdbd8SKent Overstreet 24201c6fdbd8SKent Overstreet /* ATTR_MODE will never be set here, ns argument isn't needed: */ 24211c6fdbd8SKent Overstreet setattr_copy(NULL, &inode->v, iattr); 242254e2264eSKent Overstreet 242354e2264eSKent Overstreet mutex_lock(&inode->ei_update_lock); 242454e2264eSKent Overstreet ret = bch2_write_inode(c, inode, bch2_truncate_finish_fn, NULL, 242554e2264eSKent Overstreet ATTR_MTIME|ATTR_CTIME); 242654e2264eSKent Overstreet mutex_unlock(&inode->ei_update_lock); 242754e2264eSKent Overstreet err: 24281c6fdbd8SKent Overstreet bch2_pagecache_block_put(&inode->ei_pagecache_lock); 24291c6fdbd8SKent Overstreet return ret; 24301c6fdbd8SKent Overstreet } 24311c6fdbd8SKent Overstreet 24321c6fdbd8SKent Overstreet /* fallocate: */ 24331c6fdbd8SKent Overstreet 24341c6fdbd8SKent Overstreet static long bch2_fpunch(struct bch_inode_info *inode, loff_t offset, loff_t len) 24351c6fdbd8SKent Overstreet { 24361c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 24371c6fdbd8SKent Overstreet u64 discard_start = round_up(offset, PAGE_SIZE) >> 9; 24381c6fdbd8SKent Overstreet u64 discard_end = round_down(offset + len, PAGE_SIZE) >> 9; 24391c6fdbd8SKent Overstreet int ret = 0; 24401c6fdbd8SKent Overstreet 24411c6fdbd8SKent Overstreet inode_lock(&inode->v); 24421c6fdbd8SKent Overstreet inode_dio_wait(&inode->v); 24431c6fdbd8SKent Overstreet bch2_pagecache_block_get(&inode->ei_pagecache_lock); 24441c6fdbd8SKent Overstreet 24451c6fdbd8SKent Overstreet ret = __bch2_truncate_page(inode, 24461c6fdbd8SKent Overstreet offset >> PAGE_SHIFT, 24471c6fdbd8SKent Overstreet offset, offset + len); 24481c6fdbd8SKent Overstreet if (unlikely(ret)) 24491c6fdbd8SKent Overstreet goto err; 24501c6fdbd8SKent Overstreet 24511c6fdbd8SKent Overstreet if (offset >> PAGE_SHIFT != 24521c6fdbd8SKent Overstreet (offset + len) >> PAGE_SHIFT) { 24531c6fdbd8SKent Overstreet ret = __bch2_truncate_page(inode, 24541c6fdbd8SKent Overstreet (offset + len) >> PAGE_SHIFT, 24551c6fdbd8SKent Overstreet offset, offset + len); 24561c6fdbd8SKent Overstreet if (unlikely(ret)) 24571c6fdbd8SKent Overstreet goto err; 24581c6fdbd8SKent Overstreet } 24591c6fdbd8SKent Overstreet 24601c6fdbd8SKent Overstreet truncate_pagecache_range(&inode->v, offset, offset + len - 1); 24611c6fdbd8SKent Overstreet 24625f461e01SKent Overstreet if (discard_start < discard_end) 24635f461e01SKent Overstreet ret = __bch2_fpunch(c, inode, discard_start, discard_end, 24641c6fdbd8SKent Overstreet &inode->ei_journal_seq); 24651c6fdbd8SKent Overstreet err: 24661c6fdbd8SKent Overstreet bch2_pagecache_block_put(&inode->ei_pagecache_lock); 24671c6fdbd8SKent Overstreet inode_unlock(&inode->v); 24681c6fdbd8SKent Overstreet 24691c6fdbd8SKent Overstreet return ret; 24701c6fdbd8SKent Overstreet } 24711c6fdbd8SKent Overstreet 24721c6fdbd8SKent Overstreet static long bch2_fcollapse(struct bch_inode_info *inode, 24731c6fdbd8SKent Overstreet loff_t offset, loff_t len) 24741c6fdbd8SKent Overstreet { 24751c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 24761c6fdbd8SKent Overstreet struct address_space *mapping = inode->v.i_mapping; 2477d69f41d6SKent Overstreet struct btree_trans trans; 2478d69f41d6SKent Overstreet struct btree_iter *src, *dst; 24791c6fdbd8SKent Overstreet BKEY_PADDED(k) copy; 24801c6fdbd8SKent Overstreet struct bkey_s_c k; 24811c6fdbd8SKent Overstreet loff_t new_size; 24821c6fdbd8SKent Overstreet int ret; 24831c6fdbd8SKent Overstreet 24841c6fdbd8SKent Overstreet if ((offset | len) & (block_bytes(c) - 1)) 24851c6fdbd8SKent Overstreet return -EINVAL; 24861c6fdbd8SKent Overstreet 248720bceecbSKent Overstreet bch2_trans_init(&trans, c, BTREE_ITER_MAX, 256); 24881c6fdbd8SKent Overstreet 24891c6fdbd8SKent Overstreet /* 24901c6fdbd8SKent Overstreet * We need i_mutex to keep the page cache consistent with the extents 24911c6fdbd8SKent Overstreet * btree, and the btree consistent with i_size - we don't need outside 24921c6fdbd8SKent Overstreet * locking for the extents btree itself, because we're using linked 24931c6fdbd8SKent Overstreet * iterators 24941c6fdbd8SKent Overstreet */ 24951c6fdbd8SKent Overstreet inode_lock(&inode->v); 24961c6fdbd8SKent Overstreet inode_dio_wait(&inode->v); 24971c6fdbd8SKent Overstreet bch2_pagecache_block_get(&inode->ei_pagecache_lock); 24981c6fdbd8SKent Overstreet 24991c6fdbd8SKent Overstreet ret = -EINVAL; 25001c6fdbd8SKent Overstreet if (offset + len >= inode->v.i_size) 25011c6fdbd8SKent Overstreet goto err; 25021c6fdbd8SKent Overstreet 25031c6fdbd8SKent Overstreet if (inode->v.i_size < len) 25041c6fdbd8SKent Overstreet goto err; 25051c6fdbd8SKent Overstreet 25061c6fdbd8SKent Overstreet new_size = inode->v.i_size - len; 25071c6fdbd8SKent Overstreet 25081c6fdbd8SKent Overstreet ret = write_invalidate_inode_pages_range(mapping, offset, LLONG_MAX); 25091c6fdbd8SKent Overstreet if (ret) 25101c6fdbd8SKent Overstreet goto err; 25111c6fdbd8SKent Overstreet 25128ef231bdSKent Overstreet dst = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS, 25138ef231bdSKent Overstreet POS(inode->v.i_ino, offset >> 9), 25148ef231bdSKent Overstreet BTREE_ITER_SLOTS|BTREE_ITER_INTENT); 25158ef231bdSKent Overstreet BUG_ON(IS_ERR_OR_NULL(dst)); 25168ef231bdSKent Overstreet 25178ef231bdSKent Overstreet src = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS, 25188ef231bdSKent Overstreet POS_MIN, BTREE_ITER_SLOTS); 25198ef231bdSKent Overstreet BUG_ON(IS_ERR_OR_NULL(src)); 25201c6fdbd8SKent Overstreet 2521d69f41d6SKent Overstreet while (bkey_cmp(dst->pos, 25221c6fdbd8SKent Overstreet POS(inode->v.i_ino, 25231c6fdbd8SKent Overstreet round_up(new_size, PAGE_SIZE) >> 9)) < 0) { 25241c6fdbd8SKent Overstreet struct disk_reservation disk_res; 25251c6fdbd8SKent Overstreet 25268ef231bdSKent Overstreet ret = bch2_btree_iter_traverse(dst); 25278ef231bdSKent Overstreet if (ret) 25280f238367SKent Overstreet goto bkey_err; 25298ef231bdSKent Overstreet 2530d69f41d6SKent Overstreet bch2_btree_iter_set_pos(src, 2531d69f41d6SKent Overstreet POS(dst->pos.inode, dst->pos.offset + (len >> 9))); 25321c6fdbd8SKent Overstreet 2533d69f41d6SKent Overstreet k = bch2_btree_iter_peek_slot(src); 25340f238367SKent Overstreet if ((ret = bkey_err(k))) 25350f238367SKent Overstreet goto bkey_err; 25361c6fdbd8SKent Overstreet 25371c6fdbd8SKent Overstreet bkey_reassemble(©.k, k); 25381c6fdbd8SKent Overstreet 2539d69f41d6SKent Overstreet bch2_cut_front(src->pos, ©.k); 25401c6fdbd8SKent Overstreet copy.k.k.p.offset -= len >> 9; 25411c6fdbd8SKent Overstreet 2542e2d9912cSKent Overstreet bch2_extent_trim_atomic(©.k, dst); 2543e2d9912cSKent Overstreet 2544d69f41d6SKent Overstreet BUG_ON(bkey_cmp(dst->pos, bkey_start_pos(©.k.k))); 25451c6fdbd8SKent Overstreet 25461c6fdbd8SKent Overstreet ret = bch2_disk_reservation_get(c, &disk_res, copy.k.k.size, 254726609b61SKent Overstreet bch2_bkey_nr_dirty_ptrs(bkey_i_to_s_c(©.k)), 25481c6fdbd8SKent Overstreet BCH_DISK_RESERVATION_NOFAIL); 25491c6fdbd8SKent Overstreet BUG_ON(ret); 25501c6fdbd8SKent Overstreet 2551b17657d0SKent Overstreet bch2_trans_begin_updates(&trans); 2552b17657d0SKent Overstreet 25538ef231bdSKent Overstreet ret = bch2_extent_update(&trans, inode, 25548ef231bdSKent Overstreet &disk_res, NULL, 25558ef231bdSKent Overstreet dst, ©.k, 25568ef231bdSKent Overstreet 0, true, true, NULL); 25571c6fdbd8SKent Overstreet bch2_disk_reservation_put(c, &disk_res); 25580f238367SKent Overstreet bkey_err: 25591c6fdbd8SKent Overstreet if (ret == -EINTR) 25601c6fdbd8SKent Overstreet ret = 0; 25618ef231bdSKent Overstreet if (ret) 25628ef231bdSKent Overstreet goto err; 25631c6fdbd8SKent Overstreet /* 25641c6fdbd8SKent Overstreet * XXX: if we error here we've left data with multiple 25651c6fdbd8SKent Overstreet * pointers... which isn't a _super_ serious problem... 25661c6fdbd8SKent Overstreet */ 25671c6fdbd8SKent Overstreet 2568424eb881SKent Overstreet bch2_trans_cond_resched(&trans); 25691c6fdbd8SKent Overstreet } 25708ef231bdSKent Overstreet bch2_trans_unlock(&trans); 25711c6fdbd8SKent Overstreet 25728ef231bdSKent Overstreet ret = __bch2_fpunch(c, inode, 25731c6fdbd8SKent Overstreet round_up(new_size, block_bytes(c)) >> 9, 25748ef231bdSKent Overstreet U64_MAX, &inode->ei_journal_seq); 25751c6fdbd8SKent Overstreet if (ret) 25768ef231bdSKent Overstreet goto err; 25771c6fdbd8SKent Overstreet 25788ef231bdSKent Overstreet i_size_write(&inode->v, new_size); 25798ef231bdSKent Overstreet mutex_lock(&inode->ei_update_lock); 25808ef231bdSKent Overstreet ret = bch2_write_inode_size(c, inode, new_size, 25818ef231bdSKent Overstreet ATTR_MTIME|ATTR_CTIME); 25828ef231bdSKent Overstreet mutex_unlock(&inode->ei_update_lock); 25831c6fdbd8SKent Overstreet err: 25848ef231bdSKent Overstreet bch2_trans_exit(&trans); 25851c6fdbd8SKent Overstreet bch2_pagecache_block_put(&inode->ei_pagecache_lock); 25861c6fdbd8SKent Overstreet inode_unlock(&inode->v); 25871c6fdbd8SKent Overstreet return ret; 25881c6fdbd8SKent Overstreet } 25891c6fdbd8SKent Overstreet 25901c6fdbd8SKent Overstreet static long bch2_fallocate(struct bch_inode_info *inode, int mode, 25911c6fdbd8SKent Overstreet loff_t offset, loff_t len) 25921c6fdbd8SKent Overstreet { 25931c6fdbd8SKent Overstreet struct address_space *mapping = inode->v.i_mapping; 25941c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 2595190fa7afSKent Overstreet struct btree_trans trans; 2596190fa7afSKent Overstreet struct btree_iter *iter; 25971c6fdbd8SKent Overstreet struct bpos end_pos; 25981c6fdbd8SKent Overstreet loff_t block_start, block_end; 25991c6fdbd8SKent Overstreet loff_t end = offset + len; 26001c6fdbd8SKent Overstreet unsigned sectors; 26011c6fdbd8SKent Overstreet unsigned replicas = io_opts(c, inode).data_replicas; 26021c6fdbd8SKent Overstreet int ret; 26031c6fdbd8SKent Overstreet 260420bceecbSKent Overstreet bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0); 26051c6fdbd8SKent Overstreet 26061c6fdbd8SKent Overstreet inode_lock(&inode->v); 26071c6fdbd8SKent Overstreet inode_dio_wait(&inode->v); 26081c6fdbd8SKent Overstreet bch2_pagecache_block_get(&inode->ei_pagecache_lock); 26091c6fdbd8SKent Overstreet 26101c6fdbd8SKent Overstreet if (!(mode & FALLOC_FL_KEEP_SIZE) && end > inode->v.i_size) { 26111c6fdbd8SKent Overstreet ret = inode_newsize_ok(&inode->v, end); 26121c6fdbd8SKent Overstreet if (ret) 26131c6fdbd8SKent Overstreet goto err; 26141c6fdbd8SKent Overstreet } 26151c6fdbd8SKent Overstreet 26161c6fdbd8SKent Overstreet if (mode & FALLOC_FL_ZERO_RANGE) { 26171c6fdbd8SKent Overstreet ret = __bch2_truncate_page(inode, 26181c6fdbd8SKent Overstreet offset >> PAGE_SHIFT, 26191c6fdbd8SKent Overstreet offset, end); 26201c6fdbd8SKent Overstreet 26211c6fdbd8SKent Overstreet if (!ret && 26221c6fdbd8SKent Overstreet offset >> PAGE_SHIFT != end >> PAGE_SHIFT) 26231c6fdbd8SKent Overstreet ret = __bch2_truncate_page(inode, 26241c6fdbd8SKent Overstreet end >> PAGE_SHIFT, 26251c6fdbd8SKent Overstreet offset, end); 26261c6fdbd8SKent Overstreet 26271c6fdbd8SKent Overstreet if (unlikely(ret)) 26281c6fdbd8SKent Overstreet goto err; 26291c6fdbd8SKent Overstreet 26301c6fdbd8SKent Overstreet truncate_pagecache_range(&inode->v, offset, end - 1); 26311c6fdbd8SKent Overstreet 26321c6fdbd8SKent Overstreet block_start = round_up(offset, PAGE_SIZE); 26331c6fdbd8SKent Overstreet block_end = round_down(end, PAGE_SIZE); 26341c6fdbd8SKent Overstreet } else { 26351c6fdbd8SKent Overstreet block_start = round_down(offset, PAGE_SIZE); 26361c6fdbd8SKent Overstreet block_end = round_up(end, PAGE_SIZE); 26371c6fdbd8SKent Overstreet } 26381c6fdbd8SKent Overstreet 2639190fa7afSKent Overstreet iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS, 2640190fa7afSKent Overstreet POS(inode->v.i_ino, block_start >> 9), 2641190fa7afSKent Overstreet BTREE_ITER_SLOTS|BTREE_ITER_INTENT); 26421c6fdbd8SKent Overstreet end_pos = POS(inode->v.i_ino, block_end >> 9); 26431c6fdbd8SKent Overstreet 2644190fa7afSKent Overstreet while (bkey_cmp(iter->pos, end_pos) < 0) { 26451c6fdbd8SKent Overstreet struct disk_reservation disk_res = { 0 }; 2646190fa7afSKent Overstreet struct quota_res quota_res = { 0 }; 26471c6fdbd8SKent Overstreet struct bkey_i_reservation reservation; 26481c6fdbd8SKent Overstreet struct bkey_s_c k; 26491c6fdbd8SKent Overstreet 2650190fa7afSKent Overstreet k = bch2_btree_iter_peek_slot(iter); 26510f238367SKent Overstreet if ((ret = bkey_err(k))) 26520f238367SKent Overstreet goto bkey_err; 26531c6fdbd8SKent Overstreet 26541c6fdbd8SKent Overstreet /* already reserved */ 265526609b61SKent Overstreet if (k.k->type == KEY_TYPE_reservation && 26561c6fdbd8SKent Overstreet bkey_s_c_to_reservation(k).v->nr_replicas >= replicas) { 2657190fa7afSKent Overstreet bch2_btree_iter_next_slot(iter); 26581c6fdbd8SKent Overstreet continue; 26591c6fdbd8SKent Overstreet } 26601c6fdbd8SKent Overstreet 2661190fa7afSKent Overstreet if (bkey_extent_is_data(k.k) && 2662190fa7afSKent Overstreet !(mode & FALLOC_FL_ZERO_RANGE)) { 2663190fa7afSKent Overstreet bch2_btree_iter_next_slot(iter); 26641c6fdbd8SKent Overstreet continue; 26651c6fdbd8SKent Overstreet } 26661c6fdbd8SKent Overstreet 26671c6fdbd8SKent Overstreet bkey_reservation_init(&reservation.k_i); 266826609b61SKent Overstreet reservation.k.type = KEY_TYPE_reservation; 26691c6fdbd8SKent Overstreet reservation.k.p = k.k->p; 26701c6fdbd8SKent Overstreet reservation.k.size = k.k->size; 26711c6fdbd8SKent Overstreet 2672190fa7afSKent Overstreet bch2_cut_front(iter->pos, &reservation.k_i); 26731c6fdbd8SKent Overstreet bch2_cut_back(end_pos, &reservation.k); 26741c6fdbd8SKent Overstreet 26751c6fdbd8SKent Overstreet sectors = reservation.k.size; 267626609b61SKent Overstreet reservation.v.nr_replicas = bch2_bkey_nr_dirty_ptrs(k); 26771c6fdbd8SKent Overstreet 26781c6fdbd8SKent Overstreet if (!bkey_extent_is_allocation(k.k)) { 26791c6fdbd8SKent Overstreet ret = bch2_quota_reservation_add(c, inode, 2680190fa7afSKent Overstreet "a_res, 26811c6fdbd8SKent Overstreet sectors, true); 26821c6fdbd8SKent Overstreet if (unlikely(ret)) 26830f238367SKent Overstreet goto bkey_err; 26841c6fdbd8SKent Overstreet } 26851c6fdbd8SKent Overstreet 26861c6fdbd8SKent Overstreet if (reservation.v.nr_replicas < replicas || 26871c6fdbd8SKent Overstreet bch2_extent_is_compressed(k)) { 26881c6fdbd8SKent Overstreet ret = bch2_disk_reservation_get(c, &disk_res, sectors, 26891c6fdbd8SKent Overstreet replicas, 0); 26901c6fdbd8SKent Overstreet if (unlikely(ret)) 26910f238367SKent Overstreet goto bkey_err; 26921c6fdbd8SKent Overstreet 26931c6fdbd8SKent Overstreet reservation.v.nr_replicas = disk_res.nr_replicas; 26941c6fdbd8SKent Overstreet } 26951c6fdbd8SKent Overstreet 2696b17657d0SKent Overstreet bch2_trans_begin_updates(&trans); 2697b17657d0SKent Overstreet 2698190fa7afSKent Overstreet ret = bch2_extent_update(&trans, inode, 2699190fa7afSKent Overstreet &disk_res, "a_res, 2700190fa7afSKent Overstreet iter, &reservation.k_i, 2701190fa7afSKent Overstreet 0, true, true, NULL); 27020f238367SKent Overstreet bkey_err: 2703190fa7afSKent Overstreet bch2_quota_reservation_put(c, inode, "a_res); 27041c6fdbd8SKent Overstreet bch2_disk_reservation_put(c, &disk_res); 27051c6fdbd8SKent Overstreet if (ret == -EINTR) 27061c6fdbd8SKent Overstreet ret = 0; 2707190fa7afSKent Overstreet if (ret) 2708190fa7afSKent Overstreet goto err; 27091c6fdbd8SKent Overstreet } 2710190fa7afSKent Overstreet bch2_trans_unlock(&trans); 27111c6fdbd8SKent Overstreet 27121c6fdbd8SKent Overstreet if (!(mode & FALLOC_FL_KEEP_SIZE) && 27131c6fdbd8SKent Overstreet end > inode->v.i_size) { 27141c6fdbd8SKent Overstreet i_size_write(&inode->v, end); 27151c6fdbd8SKent Overstreet 27161c6fdbd8SKent Overstreet mutex_lock(&inode->ei_update_lock); 27172ea90048SKent Overstreet ret = bch2_write_inode_size(c, inode, inode->v.i_size, 0); 27181c6fdbd8SKent Overstreet mutex_unlock(&inode->ei_update_lock); 27191c6fdbd8SKent Overstreet } 27201c6fdbd8SKent Overstreet 27211c6fdbd8SKent Overstreet /* blech */ 27221c6fdbd8SKent Overstreet if ((mode & FALLOC_FL_KEEP_SIZE) && 27231c6fdbd8SKent Overstreet (mode & FALLOC_FL_ZERO_RANGE) && 27241c6fdbd8SKent Overstreet inode->ei_inode.bi_size != inode->v.i_size) { 27251c6fdbd8SKent Overstreet /* sync appends.. */ 27261c6fdbd8SKent Overstreet ret = filemap_write_and_wait_range(mapping, 27271c6fdbd8SKent Overstreet inode->ei_inode.bi_size, S64_MAX); 27281c6fdbd8SKent Overstreet if (ret) 27291c6fdbd8SKent Overstreet goto err; 27301c6fdbd8SKent Overstreet 27311c6fdbd8SKent Overstreet if (inode->ei_inode.bi_size != inode->v.i_size) { 27321c6fdbd8SKent Overstreet mutex_lock(&inode->ei_update_lock); 27332ea90048SKent Overstreet ret = bch2_write_inode_size(c, inode, 27342ea90048SKent Overstreet inode->v.i_size, 0); 27351c6fdbd8SKent Overstreet mutex_unlock(&inode->ei_update_lock); 27361c6fdbd8SKent Overstreet } 27371c6fdbd8SKent Overstreet } 27381c6fdbd8SKent Overstreet err: 2739190fa7afSKent Overstreet bch2_trans_exit(&trans); 27401c6fdbd8SKent Overstreet bch2_pagecache_block_put(&inode->ei_pagecache_lock); 27411c6fdbd8SKent Overstreet inode_unlock(&inode->v); 27421c6fdbd8SKent Overstreet return ret; 27431c6fdbd8SKent Overstreet } 27441c6fdbd8SKent Overstreet 27451c6fdbd8SKent Overstreet long bch2_fallocate_dispatch(struct file *file, int mode, 27461c6fdbd8SKent Overstreet loff_t offset, loff_t len) 27471c6fdbd8SKent Overstreet { 27481c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 27491c6fdbd8SKent Overstreet 27501c6fdbd8SKent Overstreet if (!(mode & ~(FALLOC_FL_KEEP_SIZE|FALLOC_FL_ZERO_RANGE))) 27511c6fdbd8SKent Overstreet return bch2_fallocate(inode, mode, offset, len); 27521c6fdbd8SKent Overstreet 27531c6fdbd8SKent Overstreet if (mode == (FALLOC_FL_PUNCH_HOLE|FALLOC_FL_KEEP_SIZE)) 27541c6fdbd8SKent Overstreet return bch2_fpunch(inode, offset, len); 27551c6fdbd8SKent Overstreet 27561c6fdbd8SKent Overstreet if (mode == FALLOC_FL_COLLAPSE_RANGE) 27571c6fdbd8SKent Overstreet return bch2_fcollapse(inode, offset, len); 27581c6fdbd8SKent Overstreet 27591c6fdbd8SKent Overstreet return -EOPNOTSUPP; 27601c6fdbd8SKent Overstreet } 27611c6fdbd8SKent Overstreet 27621c6fdbd8SKent Overstreet /* fseek: */ 27631c6fdbd8SKent Overstreet 27641c6fdbd8SKent Overstreet static bool folio_is_data(struct folio *folio) 27651c6fdbd8SKent Overstreet { 2766f57a6a5dSKent Overstreet struct bch_page_state *s = bch2_page_state(&folio->page); 2767f57a6a5dSKent Overstreet unsigned i; 2768f81b648dSKent Overstreet 2769f57a6a5dSKent Overstreet if (!s) 2770f57a6a5dSKent Overstreet return false; 2771f57a6a5dSKent Overstreet 2772f57a6a5dSKent Overstreet for (i = 0; i < PAGE_SECTORS; i++) 2773f57a6a5dSKent Overstreet if (s->s[i].state >= SECTOR_DIRTY) 2774f57a6a5dSKent Overstreet return true; 2775f57a6a5dSKent Overstreet 2776f57a6a5dSKent Overstreet return false; 27771c6fdbd8SKent Overstreet } 27781c6fdbd8SKent Overstreet 27791c6fdbd8SKent Overstreet static loff_t bch2_next_pagecache_data(struct inode *vinode, 27801c6fdbd8SKent Overstreet loff_t start_offset, 27811c6fdbd8SKent Overstreet loff_t end_offset) 27821c6fdbd8SKent Overstreet { 27831c6fdbd8SKent Overstreet struct folio_batch fbatch; 27841c6fdbd8SKent Overstreet pgoff_t start_index = start_offset >> PAGE_SHIFT; 27851c6fdbd8SKent Overstreet pgoff_t end_index = end_offset >> PAGE_SHIFT; 27861c6fdbd8SKent Overstreet pgoff_t index = start_index; 27871c6fdbd8SKent Overstreet unsigned i; 27881c6fdbd8SKent Overstreet 27891c6fdbd8SKent Overstreet folio_batch_init(&fbatch); 27901c6fdbd8SKent Overstreet 27911c6fdbd8SKent Overstreet while (filemap_get_folios(vinode->i_mapping, 27921c6fdbd8SKent Overstreet &index, end_index, &fbatch)) { 27931c6fdbd8SKent Overstreet for (i = 0; i < folio_batch_count(&fbatch); i++) { 27941c6fdbd8SKent Overstreet struct folio *folio = fbatch.folios[i]; 27951c6fdbd8SKent Overstreet 27961c6fdbd8SKent Overstreet folio_lock(folio); 27971c6fdbd8SKent Overstreet if (folio_is_data(folio)) { 27981c6fdbd8SKent Overstreet end_offset = 27991c6fdbd8SKent Overstreet min(end_offset, 28001c6fdbd8SKent Overstreet max(start_offset, 2801*e10d3094SKent Overstreet ((loff_t) folio->index) << PAGE_SHIFT)); 28021c6fdbd8SKent Overstreet folio_unlock(folio); 28031c6fdbd8SKent Overstreet folio_batch_release(&fbatch); 28041c6fdbd8SKent Overstreet return end_offset; 28051c6fdbd8SKent Overstreet } 28061c6fdbd8SKent Overstreet folio_unlock(folio); 28071c6fdbd8SKent Overstreet } 28081c6fdbd8SKent Overstreet folio_batch_release(&fbatch); 28091c6fdbd8SKent Overstreet cond_resched(); 28101c6fdbd8SKent Overstreet } 28111c6fdbd8SKent Overstreet 28121c6fdbd8SKent Overstreet return end_offset; 28131c6fdbd8SKent Overstreet } 28141c6fdbd8SKent Overstreet 28151c6fdbd8SKent Overstreet static loff_t bch2_seek_data(struct file *file, u64 offset) 28161c6fdbd8SKent Overstreet { 28171c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 28181c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 2819424eb881SKent Overstreet struct btree_trans trans; 2820424eb881SKent Overstreet struct btree_iter *iter; 28211c6fdbd8SKent Overstreet struct bkey_s_c k; 28221c6fdbd8SKent Overstreet u64 isize, next_data = MAX_LFS_FILESIZE; 28231c6fdbd8SKent Overstreet int ret; 28241c6fdbd8SKent Overstreet 28251c6fdbd8SKent Overstreet isize = i_size_read(&inode->v); 28261c6fdbd8SKent Overstreet if (offset >= isize) 28271c6fdbd8SKent Overstreet return -ENXIO; 28281c6fdbd8SKent Overstreet 282920bceecbSKent Overstreet bch2_trans_init(&trans, c, 0, 0); 2830424eb881SKent Overstreet 2831424eb881SKent Overstreet for_each_btree_key(&trans, iter, BTREE_ID_EXTENTS, 283294f651e2SKent Overstreet POS(inode->v.i_ino, offset >> 9), 0, k, ret) { 28331c6fdbd8SKent Overstreet if (k.k->p.inode != inode->v.i_ino) { 28341c6fdbd8SKent Overstreet break; 28351c6fdbd8SKent Overstreet } else if (bkey_extent_is_data(k.k)) { 28361c6fdbd8SKent Overstreet next_data = max(offset, bkey_start_offset(k.k) << 9); 28371c6fdbd8SKent Overstreet break; 28381c6fdbd8SKent Overstreet } else if (k.k->p.offset >> 9 > isize) 28391c6fdbd8SKent Overstreet break; 28401c6fdbd8SKent Overstreet } 28411c6fdbd8SKent Overstreet 284294f651e2SKent Overstreet ret = bch2_trans_exit(&trans) ?: ret; 28431c6fdbd8SKent Overstreet if (ret) 28441c6fdbd8SKent Overstreet return ret; 28451c6fdbd8SKent Overstreet 28461c6fdbd8SKent Overstreet if (next_data > offset) 28471c6fdbd8SKent Overstreet next_data = bch2_next_pagecache_data(&inode->v, 28481c6fdbd8SKent Overstreet offset, next_data); 28491c6fdbd8SKent Overstreet 2850*e10d3094SKent Overstreet if (next_data >= isize) 28511c6fdbd8SKent Overstreet return -ENXIO; 28521c6fdbd8SKent Overstreet 28531c6fdbd8SKent Overstreet return vfs_setpos(file, next_data, MAX_LFS_FILESIZE); 28541c6fdbd8SKent Overstreet } 28551c6fdbd8SKent Overstreet 28561c6fdbd8SKent Overstreet static bool page_slot_is_data(struct address_space *mapping, pgoff_t index) 28571c6fdbd8SKent Overstreet { 28581c6fdbd8SKent Overstreet struct page *page; 28591c6fdbd8SKent Overstreet bool ret; 28601c6fdbd8SKent Overstreet 28611c6fdbd8SKent Overstreet page = find_lock_page(mapping, index); 28621c6fdbd8SKent Overstreet if (!page) 28631c6fdbd8SKent Overstreet return false; 28641c6fdbd8SKent Overstreet 28651c6fdbd8SKent Overstreet ret = folio_is_data(page_folio(page)); 28661c6fdbd8SKent Overstreet unlock_page(page); 28671c6fdbd8SKent Overstreet 28681c6fdbd8SKent Overstreet return ret; 28691c6fdbd8SKent Overstreet } 28701c6fdbd8SKent Overstreet 28711c6fdbd8SKent Overstreet static loff_t bch2_next_pagecache_hole(struct inode *vinode, 28721c6fdbd8SKent Overstreet loff_t start_offset, 28731c6fdbd8SKent Overstreet loff_t end_offset) 28741c6fdbd8SKent Overstreet { 28751c6fdbd8SKent Overstreet struct address_space *mapping = vinode->i_mapping; 28761c6fdbd8SKent Overstreet pgoff_t index; 28771c6fdbd8SKent Overstreet 28781c6fdbd8SKent Overstreet for (index = start_offset >> PAGE_SHIFT; 28791c6fdbd8SKent Overstreet index < end_offset >> PAGE_SHIFT; 28801c6fdbd8SKent Overstreet index++) 28811c6fdbd8SKent Overstreet if (!page_slot_is_data(mapping, index)) 28821c6fdbd8SKent Overstreet end_offset = max(start_offset, 28831c6fdbd8SKent Overstreet ((loff_t) index) << PAGE_SHIFT); 28841c6fdbd8SKent Overstreet 28851c6fdbd8SKent Overstreet return end_offset; 28861c6fdbd8SKent Overstreet } 28871c6fdbd8SKent Overstreet 28881c6fdbd8SKent Overstreet static loff_t bch2_seek_hole(struct file *file, u64 offset) 28891c6fdbd8SKent Overstreet { 28901c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 28911c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 2892424eb881SKent Overstreet struct btree_trans trans; 2893424eb881SKent Overstreet struct btree_iter *iter; 28941c6fdbd8SKent Overstreet struct bkey_s_c k; 28951c6fdbd8SKent Overstreet u64 isize, next_hole = MAX_LFS_FILESIZE; 28961c6fdbd8SKent Overstreet int ret; 28971c6fdbd8SKent Overstreet 28981c6fdbd8SKent Overstreet isize = i_size_read(&inode->v); 28991c6fdbd8SKent Overstreet if (offset >= isize) 29001c6fdbd8SKent Overstreet return -ENXIO; 29011c6fdbd8SKent Overstreet 290220bceecbSKent Overstreet bch2_trans_init(&trans, c, 0, 0); 2903424eb881SKent Overstreet 2904424eb881SKent Overstreet for_each_btree_key(&trans, iter, BTREE_ID_EXTENTS, 29051c6fdbd8SKent Overstreet POS(inode->v.i_ino, offset >> 9), 290694f651e2SKent Overstreet BTREE_ITER_SLOTS, k, ret) { 29071c6fdbd8SKent Overstreet if (k.k->p.inode != inode->v.i_ino) { 29081c6fdbd8SKent Overstreet next_hole = bch2_next_pagecache_hole(&inode->v, 29091c6fdbd8SKent Overstreet offset, MAX_LFS_FILESIZE); 29101c6fdbd8SKent Overstreet break; 29111c6fdbd8SKent Overstreet } else if (!bkey_extent_is_data(k.k)) { 29121c6fdbd8SKent Overstreet next_hole = bch2_next_pagecache_hole(&inode->v, 29131c6fdbd8SKent Overstreet max(offset, bkey_start_offset(k.k) << 9), 29141c6fdbd8SKent Overstreet k.k->p.offset << 9); 29151c6fdbd8SKent Overstreet 29161c6fdbd8SKent Overstreet if (next_hole < k.k->p.offset << 9) 29171c6fdbd8SKent Overstreet break; 29181c6fdbd8SKent Overstreet } else { 29191c6fdbd8SKent Overstreet offset = max(offset, bkey_start_offset(k.k) << 9); 29201c6fdbd8SKent Overstreet } 29211c6fdbd8SKent Overstreet } 29221c6fdbd8SKent Overstreet 292394f651e2SKent Overstreet ret = bch2_trans_exit(&trans) ?: ret; 29241c6fdbd8SKent Overstreet if (ret) 29251c6fdbd8SKent Overstreet return ret; 29261c6fdbd8SKent Overstreet 29271c6fdbd8SKent Overstreet if (next_hole > isize) 29281c6fdbd8SKent Overstreet next_hole = isize; 29291c6fdbd8SKent Overstreet 29301c6fdbd8SKent Overstreet return vfs_setpos(file, next_hole, MAX_LFS_FILESIZE); 29311c6fdbd8SKent Overstreet } 29321c6fdbd8SKent Overstreet 29331c6fdbd8SKent Overstreet loff_t bch2_llseek(struct file *file, loff_t offset, int whence) 29341c6fdbd8SKent Overstreet { 29351c6fdbd8SKent Overstreet switch (whence) { 29361c6fdbd8SKent Overstreet case SEEK_SET: 29371c6fdbd8SKent Overstreet case SEEK_CUR: 29381c6fdbd8SKent Overstreet case SEEK_END: 29391c6fdbd8SKent Overstreet return generic_file_llseek(file, offset, whence); 29401c6fdbd8SKent Overstreet case SEEK_DATA: 29411c6fdbd8SKent Overstreet return bch2_seek_data(file, offset); 29421c6fdbd8SKent Overstreet case SEEK_HOLE: 29431c6fdbd8SKent Overstreet return bch2_seek_hole(file, offset); 29441c6fdbd8SKent Overstreet } 29451c6fdbd8SKent Overstreet 29461c6fdbd8SKent Overstreet return -EINVAL; 29471c6fdbd8SKent Overstreet } 29481c6fdbd8SKent Overstreet 29491c6fdbd8SKent Overstreet void bch2_fs_fsio_exit(struct bch_fs *c) 29501c6fdbd8SKent Overstreet { 29511c6fdbd8SKent Overstreet bioset_exit(&c->dio_write_bioset); 29521c6fdbd8SKent Overstreet bioset_exit(&c->dio_read_bioset); 29531c6fdbd8SKent Overstreet bioset_exit(&c->writepage_bioset); 29541c6fdbd8SKent Overstreet } 29551c6fdbd8SKent Overstreet 29561c6fdbd8SKent Overstreet int bch2_fs_fsio_init(struct bch_fs *c) 29571c6fdbd8SKent Overstreet { 29581c6fdbd8SKent Overstreet int ret = 0; 29591c6fdbd8SKent Overstreet 29601c6fdbd8SKent Overstreet pr_verbose_init(c->opts, ""); 29611c6fdbd8SKent Overstreet 29621c6fdbd8SKent Overstreet if (bioset_init(&c->writepage_bioset, 29631c6fdbd8SKent Overstreet 4, offsetof(struct bch_writepage_io, op.op.wbio.bio), 29641c6fdbd8SKent Overstreet BIOSET_NEED_BVECS) || 29651c6fdbd8SKent Overstreet bioset_init(&c->dio_read_bioset, 29661c6fdbd8SKent Overstreet 4, offsetof(struct dio_read, rbio.bio), 29671c6fdbd8SKent Overstreet BIOSET_NEED_BVECS) || 29681c6fdbd8SKent Overstreet bioset_init(&c->dio_write_bioset, 29691c6fdbd8SKent Overstreet 4, offsetof(struct dio_write, iop.op.wbio.bio), 29701c6fdbd8SKent Overstreet BIOSET_NEED_BVECS)) 29711c6fdbd8SKent Overstreet ret = -ENOMEM; 29721c6fdbd8SKent Overstreet 29731c6fdbd8SKent Overstreet pr_verbose_init(c->opts, "ret %i", ret); 29741c6fdbd8SKent Overstreet return ret; 29751c6fdbd8SKent Overstreet } 29761c6fdbd8SKent Overstreet 29771c6fdbd8SKent Overstreet #endif /* NO_BCACHEFS_FS */ 2978