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; 601c6fdbd8SKent Overstreet struct task_struct *task; 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, 1241c6fdbd8SKent Overstreet -((s64) res->sectors), BCH_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, 1411c6fdbd8SKent Overstreet check_enospc ? BCH_QUOTA_PREALLOC : BCH_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 { 2231c6fdbd8SKent Overstreet bch2_quota_acct(c, inode->ei_qid, Q_SPC, sectors, BCH_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 23208af47dfSKent Overstreet static s64 sum_sector_overwrites(struct bkey_i *new, struct btree_iter *_iter, 23308af47dfSKent Overstreet bool *allocating) 23408af47dfSKent Overstreet { 23508af47dfSKent Overstreet struct btree_iter iter; 23608af47dfSKent Overstreet struct bkey_s_c old; 23708af47dfSKent Overstreet s64 delta = 0; 2381c6fdbd8SKent Overstreet 23908af47dfSKent Overstreet bch2_btree_iter_init(&iter, _iter->c, BTREE_ID_EXTENTS, POS_MIN, 24008af47dfSKent Overstreet BTREE_ITER_SLOTS); 24108af47dfSKent Overstreet 24208af47dfSKent Overstreet bch2_btree_iter_link(_iter, &iter); 24308af47dfSKent Overstreet bch2_btree_iter_copy(&iter, _iter); 24408af47dfSKent Overstreet 24508af47dfSKent Overstreet for_each_btree_key_continue(&iter, BTREE_ITER_SLOTS, old) { 24608af47dfSKent Overstreet if (bkey_cmp(new->k.p, bkey_start_pos(old.k)) <= 0) 24708af47dfSKent Overstreet break; 24808af47dfSKent Overstreet 24908af47dfSKent Overstreet if (allocating && 25008af47dfSKent Overstreet !bch2_extent_is_fully_allocated(old)) 25108af47dfSKent Overstreet *allocating = true; 25208af47dfSKent Overstreet 25308af47dfSKent Overstreet delta += (min(new->k.p.offset, 25408af47dfSKent Overstreet old.k->p.offset) - 25508af47dfSKent Overstreet max(bkey_start_offset(&new->k), 25608af47dfSKent Overstreet bkey_start_offset(old.k))) * 25708af47dfSKent Overstreet (bkey_extent_is_allocation(&new->k) - 25808af47dfSKent Overstreet bkey_extent_is_allocation(old.k)); 25908af47dfSKent Overstreet } 26008af47dfSKent Overstreet 26108af47dfSKent Overstreet bch2_btree_iter_unlink(&iter); 26208af47dfSKent Overstreet 26308af47dfSKent Overstreet return delta; 26408af47dfSKent Overstreet } 26508af47dfSKent Overstreet 26608af47dfSKent Overstreet static int bch2_extent_update(struct btree_trans *trans, 26708af47dfSKent Overstreet struct bch_inode_info *inode, 26808af47dfSKent Overstreet struct disk_reservation *disk_res, 26908af47dfSKent Overstreet struct quota_res *quota_res, 27008af47dfSKent Overstreet struct btree_iter *extent_iter, 27108af47dfSKent Overstreet struct bkey_i *k, 27208af47dfSKent Overstreet u64 new_i_size, 27308af47dfSKent Overstreet bool may_allocate, 27408af47dfSKent Overstreet bool direct, 27508af47dfSKent Overstreet s64 *total_delta) 27608af47dfSKent Overstreet { 27708af47dfSKent Overstreet struct btree_iter *inode_iter = NULL; 2781c6fdbd8SKent Overstreet struct bch_inode_unpacked inode_u; 2791c6fdbd8SKent Overstreet struct bkey_inode_buf inode_p; 28008af47dfSKent Overstreet bool allocating = false; 28108af47dfSKent Overstreet bool extended = false; 28208af47dfSKent Overstreet s64 i_sectors_delta; 28308af47dfSKent Overstreet int ret; 2841c6fdbd8SKent Overstreet 28508af47dfSKent Overstreet bch2_trans_begin_updates(trans); 2861c6fdbd8SKent Overstreet 28708af47dfSKent Overstreet ret = bch2_btree_iter_traverse(extent_iter); 28808af47dfSKent Overstreet if (ret) 28908af47dfSKent Overstreet return ret; 2901c6fdbd8SKent Overstreet 29108af47dfSKent Overstreet bch2_extent_trim_atomic(k, extent_iter); 2921c6fdbd8SKent Overstreet 29308af47dfSKent Overstreet i_sectors_delta = sum_sector_overwrites(k, extent_iter, &allocating); 29408af47dfSKent Overstreet if (!may_allocate && allocating) 29508af47dfSKent Overstreet return -ENOSPC; 29608af47dfSKent Overstreet 29708af47dfSKent Overstreet bch2_trans_update(trans, BTREE_INSERT_ENTRY(extent_iter, k)); 29808af47dfSKent Overstreet 29908af47dfSKent Overstreet new_i_size = min(k->k.p.offset << 9, new_i_size); 3001c6fdbd8SKent Overstreet 3011c6fdbd8SKent Overstreet /* XXX: inode->i_size locking */ 30208af47dfSKent Overstreet if (i_sectors_delta || 30308af47dfSKent Overstreet new_i_size > inode->ei_inode.bi_size) { 30408af47dfSKent Overstreet inode_iter = bch2_trans_get_iter(trans, 30508af47dfSKent Overstreet BTREE_ID_INODES, 30608af47dfSKent Overstreet POS(k->k.p.inode, 0), 30708af47dfSKent Overstreet BTREE_ITER_SLOTS|BTREE_ITER_INTENT); 30808af47dfSKent Overstreet if (IS_ERR(inode_iter)) 30908af47dfSKent Overstreet return PTR_ERR(inode_iter); 31008af47dfSKent Overstreet 31108af47dfSKent Overstreet ret = bch2_btree_iter_traverse(inode_iter); 31208af47dfSKent Overstreet if (ret) 31308af47dfSKent Overstreet goto err; 31408af47dfSKent Overstreet 31508af47dfSKent Overstreet inode_u = inode->ei_inode; 31608af47dfSKent Overstreet inode_u.bi_sectors += i_sectors_delta; 31708af47dfSKent Overstreet 31808af47dfSKent Overstreet /* XXX: this is slightly suspect */ 31908af47dfSKent Overstreet if (!(inode_u.bi_flags & BCH_INODE_I_SIZE_DIRTY) && 32008af47dfSKent Overstreet new_i_size > inode_u.bi_size) { 32108af47dfSKent Overstreet inode_u.bi_size = new_i_size; 32208af47dfSKent Overstreet extended = true; 3231c6fdbd8SKent Overstreet } 3241c6fdbd8SKent Overstreet 32508af47dfSKent Overstreet bch2_inode_pack(&inode_p, &inode_u); 32608af47dfSKent Overstreet bch2_trans_update(trans, 32708af47dfSKent Overstreet BTREE_INSERT_ENTRY(inode_iter, &inode_p.inode.k_i)); 32808af47dfSKent Overstreet } 3291c6fdbd8SKent Overstreet 330fc3268c1SKent Overstreet ret = bch2_trans_commit(trans, disk_res, 33108af47dfSKent Overstreet &inode->ei_journal_seq, 33208af47dfSKent Overstreet BTREE_INSERT_NOFAIL| 33308af47dfSKent Overstreet BTREE_INSERT_ATOMIC| 33408af47dfSKent Overstreet BTREE_INSERT_NOUNLOCK| 33508af47dfSKent Overstreet BTREE_INSERT_USE_RESERVE); 33608af47dfSKent Overstreet if (ret) 33708af47dfSKent Overstreet goto err; 3381c6fdbd8SKent Overstreet 33908af47dfSKent Overstreet inode->ei_inode.bi_sectors += i_sectors_delta; 34008af47dfSKent Overstreet 34108af47dfSKent Overstreet EBUG_ON(i_sectors_delta && 34208af47dfSKent Overstreet inode->ei_inode.bi_sectors != inode_u.bi_sectors); 34308af47dfSKent Overstreet 34408af47dfSKent Overstreet if (extended) { 34508af47dfSKent Overstreet inode->ei_inode.bi_size = new_i_size; 34608af47dfSKent Overstreet 34708af47dfSKent Overstreet if (direct) { 3481c6fdbd8SKent Overstreet spin_lock(&inode->v.i_lock); 34908af47dfSKent Overstreet if (new_i_size > inode->v.i_size) 35008af47dfSKent Overstreet i_size_write(&inode->v, new_i_size); 3511c6fdbd8SKent Overstreet spin_unlock(&inode->v.i_lock); 3521c6fdbd8SKent Overstreet } 3531c6fdbd8SKent Overstreet } 3541c6fdbd8SKent Overstreet 35508af47dfSKent Overstreet if (direct) 35608af47dfSKent Overstreet i_sectors_acct(trans->c, inode, quota_res, i_sectors_delta); 3571c6fdbd8SKent Overstreet 35808af47dfSKent Overstreet if (total_delta) 35908af47dfSKent Overstreet *total_delta += i_sectors_delta; 36008af47dfSKent Overstreet err: 36108af47dfSKent Overstreet if (!IS_ERR_OR_NULL(inode_iter)) 36208af47dfSKent Overstreet bch2_trans_iter_put(trans, inode_iter); 36308af47dfSKent Overstreet return ret; 3641c6fdbd8SKent Overstreet } 3651c6fdbd8SKent Overstreet 3661c6fdbd8SKent Overstreet static int bchfs_write_index_update(struct bch_write_op *wop) 3671c6fdbd8SKent Overstreet { 3681c6fdbd8SKent Overstreet struct bchfs_write_op *op = container_of(wop, 3691c6fdbd8SKent Overstreet struct bchfs_write_op, op); 37008af47dfSKent Overstreet struct quota_res *quota_res = op->is_dio 37108af47dfSKent Overstreet ? &container_of(op, struct dio_write, iop)->quota_res 37208af47dfSKent Overstreet : NULL; 37308af47dfSKent Overstreet struct bch_inode_info *inode = op->inode; 3741c6fdbd8SKent Overstreet struct keylist *keys = &op->op.insert_keys; 3751c6fdbd8SKent Overstreet struct bkey_i *k = bch2_keylist_front(keys); 37608af47dfSKent Overstreet struct btree_trans trans; 37708af47dfSKent Overstreet struct btree_iter *iter; 3781c6fdbd8SKent Overstreet int ret; 3791c6fdbd8SKent Overstreet 38008af47dfSKent Overstreet BUG_ON(k->k.p.inode != inode->v.i_ino); 3811c6fdbd8SKent Overstreet 382d69f41d6SKent Overstreet bch2_trans_init(&trans, wop->c); 38308af47dfSKent Overstreet bch2_trans_preload_iters(&trans); 384d69f41d6SKent Overstreet 38508af47dfSKent Overstreet iter = bch2_trans_get_iter(&trans, 386d69f41d6SKent Overstreet BTREE_ID_EXTENTS, 38708af47dfSKent Overstreet bkey_start_pos(&k->k), 3881c6fdbd8SKent Overstreet BTREE_ITER_INTENT); 3891c6fdbd8SKent Overstreet 3901c6fdbd8SKent Overstreet do { 391e2d9912cSKent Overstreet BKEY_PADDED(k) tmp; 3921c6fdbd8SKent Overstreet 393e2d9912cSKent Overstreet bkey_copy(&tmp.k, bch2_keylist_front(keys)); 394e2d9912cSKent Overstreet 39508af47dfSKent Overstreet ret = bch2_extent_update(&trans, inode, 39608af47dfSKent Overstreet &wop->res, quota_res, 39708af47dfSKent Overstreet iter, &tmp.k, 39808af47dfSKent Overstreet op->new_i_size, 39908af47dfSKent Overstreet !op->unalloc, 40008af47dfSKent Overstreet op->is_dio, 40108af47dfSKent Overstreet &op->sectors_added); 4021c6fdbd8SKent Overstreet if (ret == -EINTR) 4031c6fdbd8SKent Overstreet continue; 4041c6fdbd8SKent Overstreet if (ret) 4051c6fdbd8SKent Overstreet break; 4061c6fdbd8SKent Overstreet 40708af47dfSKent Overstreet if (bkey_cmp(iter->pos, bch2_keylist_front(keys)->k.p) < 0) 40808af47dfSKent Overstreet bch2_cut_front(iter->pos, bch2_keylist_front(keys)); 409e2d9912cSKent Overstreet else 4101c6fdbd8SKent Overstreet bch2_keylist_pop_front(keys); 4111c6fdbd8SKent Overstreet } while (!bch2_keylist_empty(keys)); 4121c6fdbd8SKent Overstreet 413d69f41d6SKent Overstreet bch2_trans_exit(&trans); 4141c6fdbd8SKent Overstreet 4151c6fdbd8SKent Overstreet return ret; 4161c6fdbd8SKent Overstreet } 4171c6fdbd8SKent Overstreet 4181c6fdbd8SKent Overstreet static inline void bch2_fswrite_op_init(struct bchfs_write_op *op, 4191c6fdbd8SKent Overstreet struct bch_fs *c, 4201c6fdbd8SKent Overstreet struct bch_inode_info *inode, 4211c6fdbd8SKent Overstreet struct bch_io_opts opts, 4221c6fdbd8SKent Overstreet bool is_dio) 4231c6fdbd8SKent Overstreet { 4241c6fdbd8SKent Overstreet op->inode = inode; 4251c6fdbd8SKent Overstreet op->sectors_added = 0; 4261c6fdbd8SKent Overstreet op->is_dio = is_dio; 4271c6fdbd8SKent Overstreet op->unalloc = false; 4281c6fdbd8SKent Overstreet op->new_i_size = U64_MAX; 4291c6fdbd8SKent Overstreet 4301c6fdbd8SKent Overstreet bch2_write_op_init(&op->op, c, opts); 4311c6fdbd8SKent Overstreet op->op.target = opts.foreground_target; 4321c6fdbd8SKent Overstreet op->op.index_update_fn = bchfs_write_index_update; 4331c6fdbd8SKent Overstreet op_journal_seq_set(&op->op, &inode->ei_journal_seq); 4341c6fdbd8SKent Overstreet } 4351c6fdbd8SKent Overstreet 4361c6fdbd8SKent Overstreet static inline struct bch_io_opts io_opts(struct bch_fs *c, struct bch_inode_info *inode) 4371c6fdbd8SKent Overstreet { 4381c6fdbd8SKent Overstreet struct bch_io_opts opts = bch2_opts_to_inode_opts(c->opts); 4391c6fdbd8SKent Overstreet 4401c6fdbd8SKent Overstreet bch2_io_opts_apply(&opts, bch2_inode_opts_get(&inode->ei_inode)); 4411c6fdbd8SKent Overstreet return opts; 4421c6fdbd8SKent Overstreet } 4431c6fdbd8SKent Overstreet 4441c6fdbd8SKent Overstreet /* page state: */ 4451c6fdbd8SKent Overstreet 4461c6fdbd8SKent Overstreet /* stored in page->private: */ 4471c6fdbd8SKent Overstreet 4481c6fdbd8SKent Overstreet /* 4491c6fdbd8SKent Overstreet * bch_page_state has to (unfortunately) be manipulated with cmpxchg - we could 4501c6fdbd8SKent Overstreet * almost protected it with the page lock, except that bch2_writepage_io_done has 4511c6fdbd8SKent Overstreet * to update the sector counts (and from interrupt/bottom half context). 4521c6fdbd8SKent Overstreet */ 4531c6fdbd8SKent Overstreet struct bch_page_state { 4541c6fdbd8SKent Overstreet union { struct { 4551c6fdbd8SKent Overstreet /* existing data: */ 4561c6fdbd8SKent Overstreet unsigned sectors:PAGE_SECTOR_SHIFT + 1; 4571c6fdbd8SKent Overstreet unsigned nr_replicas:4; 4581c6fdbd8SKent Overstreet unsigned compressed:1; 4591c6fdbd8SKent Overstreet 4601c6fdbd8SKent Overstreet /* Owns PAGE_SECTORS sized reservation: */ 4611c6fdbd8SKent Overstreet unsigned reserved:1; 4621c6fdbd8SKent Overstreet unsigned reservation_replicas:4; 4631c6fdbd8SKent Overstreet 4641c6fdbd8SKent Overstreet /* Owns PAGE_SECTORS sized quota reservation: */ 4651c6fdbd8SKent Overstreet unsigned quota_reserved:1; 4661c6fdbd8SKent Overstreet 4671c6fdbd8SKent Overstreet /* 4681c6fdbd8SKent Overstreet * Number of sectors on disk - for i_blocks 4691c6fdbd8SKent Overstreet * Uncompressed size, not compressed size: 4701c6fdbd8SKent Overstreet */ 4711c6fdbd8SKent Overstreet unsigned dirty_sectors:PAGE_SECTOR_SHIFT + 1; 4721c6fdbd8SKent Overstreet }; 4731c6fdbd8SKent Overstreet /* for cmpxchg: */ 4741c6fdbd8SKent Overstreet unsigned long v; 4751c6fdbd8SKent Overstreet }; 4761c6fdbd8SKent Overstreet }; 4771c6fdbd8SKent Overstreet 4781c6fdbd8SKent Overstreet #define page_state_cmpxchg(_ptr, _new, _expr) \ 4791c6fdbd8SKent Overstreet ({ \ 4801c6fdbd8SKent Overstreet unsigned long _v = READ_ONCE((_ptr)->v); \ 4811c6fdbd8SKent Overstreet struct bch_page_state _old; \ 4821c6fdbd8SKent Overstreet \ 4831c6fdbd8SKent Overstreet do { \ 4841c6fdbd8SKent Overstreet _old.v = _new.v = _v; \ 4851c6fdbd8SKent Overstreet _expr; \ 4861c6fdbd8SKent Overstreet \ 4871c6fdbd8SKent Overstreet EBUG_ON(_new.sectors + _new.dirty_sectors > PAGE_SECTORS);\ 4881c6fdbd8SKent Overstreet } while (_old.v != _new.v && \ 4891c6fdbd8SKent Overstreet (_v = cmpxchg(&(_ptr)->v, _old.v, _new.v)) != _old.v); \ 4901c6fdbd8SKent Overstreet \ 4911c6fdbd8SKent Overstreet _old; \ 4921c6fdbd8SKent Overstreet }) 4931c6fdbd8SKent Overstreet 4941c6fdbd8SKent Overstreet static inline struct bch_page_state *page_state(struct page *page) 4951c6fdbd8SKent Overstreet { 4961c6fdbd8SKent Overstreet struct bch_page_state *s = (void *) &page->private; 4971c6fdbd8SKent Overstreet 4981c6fdbd8SKent Overstreet BUILD_BUG_ON(sizeof(*s) > sizeof(page->private)); 4991c6fdbd8SKent Overstreet 5001c6fdbd8SKent Overstreet if (!PagePrivate(page)) 5011c6fdbd8SKent Overstreet SetPagePrivate(page); 5021c6fdbd8SKent Overstreet 5031c6fdbd8SKent Overstreet return s; 5041c6fdbd8SKent Overstreet } 5051c6fdbd8SKent Overstreet 5061c6fdbd8SKent Overstreet static inline unsigned page_res_sectors(struct bch_page_state s) 5071c6fdbd8SKent Overstreet { 5081c6fdbd8SKent Overstreet 5091c6fdbd8SKent Overstreet return s.reserved ? s.reservation_replicas * PAGE_SECTORS : 0; 5101c6fdbd8SKent Overstreet } 5111c6fdbd8SKent Overstreet 5121c6fdbd8SKent Overstreet static void __bch2_put_page_reservation(struct bch_fs *c, struct bch_inode_info *inode, 5131c6fdbd8SKent Overstreet struct bch_page_state s) 5141c6fdbd8SKent Overstreet { 5151c6fdbd8SKent Overstreet struct disk_reservation res = { .sectors = page_res_sectors(s) }; 5161c6fdbd8SKent Overstreet struct quota_res quota_res = { .sectors = s.quota_reserved ? PAGE_SECTORS : 0 }; 5171c6fdbd8SKent Overstreet 5181c6fdbd8SKent Overstreet bch2_quota_reservation_put(c, inode, "a_res); 5191c6fdbd8SKent Overstreet bch2_disk_reservation_put(c, &res); 5201c6fdbd8SKent Overstreet } 5211c6fdbd8SKent Overstreet 5221c6fdbd8SKent Overstreet static void bch2_put_page_reservation(struct bch_fs *c, struct bch_inode_info *inode, 5231c6fdbd8SKent Overstreet struct page *page) 5241c6fdbd8SKent Overstreet { 5251c6fdbd8SKent Overstreet struct bch_page_state s; 5261c6fdbd8SKent Overstreet 5271c6fdbd8SKent Overstreet s = page_state_cmpxchg(page_state(page), s, { 5281c6fdbd8SKent Overstreet s.reserved = 0; 5291c6fdbd8SKent Overstreet s.quota_reserved = 0; 5301c6fdbd8SKent Overstreet }); 5311c6fdbd8SKent Overstreet 5321c6fdbd8SKent Overstreet __bch2_put_page_reservation(c, inode, s); 5331c6fdbd8SKent Overstreet } 5341c6fdbd8SKent Overstreet 5351c6fdbd8SKent Overstreet static int bch2_get_page_reservation(struct bch_fs *c, struct bch_inode_info *inode, 5361c6fdbd8SKent Overstreet struct page *page, bool check_enospc) 5371c6fdbd8SKent Overstreet { 5381c6fdbd8SKent Overstreet struct bch_page_state *s = page_state(page), new, old; 5391c6fdbd8SKent Overstreet 5401c6fdbd8SKent Overstreet /* XXX: this should not be open coded */ 5411c6fdbd8SKent Overstreet unsigned nr_replicas = inode->ei_inode.bi_data_replicas 5421c6fdbd8SKent Overstreet ? inode->ei_inode.bi_data_replicas - 1 5431c6fdbd8SKent Overstreet : c->opts.data_replicas; 5441c6fdbd8SKent Overstreet 5451c6fdbd8SKent Overstreet struct disk_reservation disk_res = bch2_disk_reservation_init(c, 5461c6fdbd8SKent Overstreet nr_replicas); 5471c6fdbd8SKent Overstreet struct quota_res quota_res = { 0 }; 5481c6fdbd8SKent Overstreet int ret = 0; 5491c6fdbd8SKent Overstreet 5501c6fdbd8SKent Overstreet /* 5511c6fdbd8SKent Overstreet * XXX: this could likely be quite a bit simpler, page reservations 5521c6fdbd8SKent Overstreet * _should_ only be manipulated with page locked: 5531c6fdbd8SKent Overstreet */ 5541c6fdbd8SKent Overstreet 5551c6fdbd8SKent Overstreet old = page_state_cmpxchg(s, new, { 5561c6fdbd8SKent Overstreet if (new.reserved 5571c6fdbd8SKent Overstreet ? (new.reservation_replicas < disk_res.nr_replicas) 5581c6fdbd8SKent Overstreet : (new.sectors < PAGE_SECTORS || 5591c6fdbd8SKent Overstreet new.nr_replicas < disk_res.nr_replicas || 5601c6fdbd8SKent Overstreet new.compressed)) { 5611c6fdbd8SKent Overstreet int sectors = (disk_res.nr_replicas * PAGE_SECTORS - 5621c6fdbd8SKent Overstreet page_res_sectors(new) - 5631c6fdbd8SKent Overstreet disk_res.sectors); 5641c6fdbd8SKent Overstreet 5651c6fdbd8SKent Overstreet if (sectors > 0) { 5661c6fdbd8SKent Overstreet ret = bch2_disk_reservation_add(c, &disk_res, sectors, 5671c6fdbd8SKent Overstreet !check_enospc 5681c6fdbd8SKent Overstreet ? BCH_DISK_RESERVATION_NOFAIL : 0); 5691c6fdbd8SKent Overstreet if (unlikely(ret)) 5701c6fdbd8SKent Overstreet goto err; 5711c6fdbd8SKent Overstreet } 5721c6fdbd8SKent Overstreet 5731c6fdbd8SKent Overstreet new.reserved = 1; 5741c6fdbd8SKent Overstreet new.reservation_replicas = disk_res.nr_replicas; 5751c6fdbd8SKent Overstreet } 5761c6fdbd8SKent Overstreet 5771c6fdbd8SKent Overstreet if (!new.quota_reserved && 5781c6fdbd8SKent Overstreet new.sectors + new.dirty_sectors < PAGE_SECTORS) { 5791c6fdbd8SKent Overstreet ret = bch2_quota_reservation_add(c, inode, "a_res, 5801c6fdbd8SKent Overstreet PAGE_SECTORS - quota_res.sectors, 5811c6fdbd8SKent Overstreet check_enospc); 5821c6fdbd8SKent Overstreet if (unlikely(ret)) 5831c6fdbd8SKent Overstreet goto err; 5841c6fdbd8SKent Overstreet 5851c6fdbd8SKent Overstreet new.quota_reserved = 1; 5861c6fdbd8SKent Overstreet } 5871c6fdbd8SKent Overstreet }); 5881c6fdbd8SKent Overstreet 5891c6fdbd8SKent Overstreet quota_res.sectors -= (new.quota_reserved - old.quota_reserved) * PAGE_SECTORS; 5901c6fdbd8SKent Overstreet disk_res.sectors -= page_res_sectors(new) - page_res_sectors(old); 5911c6fdbd8SKent Overstreet err: 5921c6fdbd8SKent Overstreet bch2_quota_reservation_put(c, inode, "a_res); 5931c6fdbd8SKent Overstreet bch2_disk_reservation_put(c, &disk_res); 5941c6fdbd8SKent Overstreet return ret; 5951c6fdbd8SKent Overstreet } 5961c6fdbd8SKent Overstreet 5971c6fdbd8SKent Overstreet static void bch2_clear_page_bits(struct page *page) 5981c6fdbd8SKent Overstreet { 5991c6fdbd8SKent Overstreet struct bch_inode_info *inode = to_bch_ei(page->mapping->host); 6001c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 6011c6fdbd8SKent Overstreet struct bch_page_state s; 6021c6fdbd8SKent Overstreet 6031c6fdbd8SKent Overstreet if (!PagePrivate(page)) 6041c6fdbd8SKent Overstreet return; 6051c6fdbd8SKent Overstreet 6061c6fdbd8SKent Overstreet s.v = xchg(&page_state(page)->v, 0); 6071c6fdbd8SKent Overstreet ClearPagePrivate(page); 6081c6fdbd8SKent Overstreet 6091c6fdbd8SKent Overstreet if (s.dirty_sectors) 6101c6fdbd8SKent Overstreet i_sectors_acct(c, inode, NULL, -s.dirty_sectors); 6111c6fdbd8SKent Overstreet 6121c6fdbd8SKent Overstreet __bch2_put_page_reservation(c, inode, s); 6131c6fdbd8SKent Overstreet } 6141c6fdbd8SKent Overstreet 6151c6fdbd8SKent Overstreet bool bch2_dirty_folio(struct address_space *mapping, struct folio *folio) 6161c6fdbd8SKent Overstreet { 6171c6fdbd8SKent Overstreet struct bch_inode_info *inode = to_bch_ei(mapping->host); 6181c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 6191c6fdbd8SKent Overstreet struct quota_res quota_res = { 0 }; 6201c6fdbd8SKent Overstreet struct bch_page_state old, new; 6211c6fdbd8SKent Overstreet 6221c6fdbd8SKent Overstreet old = page_state_cmpxchg(page_state(&folio->page), new, 6231c6fdbd8SKent Overstreet new.dirty_sectors = PAGE_SECTORS - new.sectors; 6241c6fdbd8SKent Overstreet new.quota_reserved = 0; 6251c6fdbd8SKent Overstreet ); 6261c6fdbd8SKent Overstreet 6271c6fdbd8SKent Overstreet quota_res.sectors += old.quota_reserved * PAGE_SECTORS; 6281c6fdbd8SKent Overstreet 6291c6fdbd8SKent Overstreet if (old.dirty_sectors != new.dirty_sectors) 6301c6fdbd8SKent Overstreet i_sectors_acct(c, inode, "a_res, 6311c6fdbd8SKent Overstreet new.dirty_sectors - old.dirty_sectors); 6321c6fdbd8SKent Overstreet bch2_quota_reservation_put(c, inode, "a_res); 6331c6fdbd8SKent Overstreet 6341c6fdbd8SKent Overstreet return filemap_dirty_folio(mapping, folio); 6351c6fdbd8SKent Overstreet } 6361c6fdbd8SKent Overstreet 6371c6fdbd8SKent Overstreet vm_fault_t bch2_page_fault(struct vm_fault *vmf) 6381c6fdbd8SKent Overstreet { 6391c6fdbd8SKent Overstreet struct file *file = vmf->vma->vm_file; 6401c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 6411c6fdbd8SKent Overstreet int ret; 6421c6fdbd8SKent Overstreet 6431c6fdbd8SKent Overstreet bch2_pagecache_add_get(&inode->ei_pagecache_lock); 6441c6fdbd8SKent Overstreet ret = filemap_fault(vmf); 6451c6fdbd8SKent Overstreet bch2_pagecache_add_put(&inode->ei_pagecache_lock); 6461c6fdbd8SKent Overstreet 6471c6fdbd8SKent Overstreet return ret; 6481c6fdbd8SKent Overstreet } 6491c6fdbd8SKent Overstreet 6501c6fdbd8SKent Overstreet vm_fault_t bch2_page_mkwrite(struct vm_fault *vmf) 6511c6fdbd8SKent Overstreet { 6521c6fdbd8SKent Overstreet struct page *page = vmf->page; 6531c6fdbd8SKent Overstreet struct file *file = vmf->vma->vm_file; 6541c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 6551c6fdbd8SKent Overstreet struct address_space *mapping = file->f_mapping; 6561c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 6571c6fdbd8SKent Overstreet int ret = VM_FAULT_LOCKED; 6581c6fdbd8SKent Overstreet 6591c6fdbd8SKent Overstreet sb_start_pagefault(inode->v.i_sb); 6601c6fdbd8SKent Overstreet file_update_time(file); 6611c6fdbd8SKent Overstreet 6621c6fdbd8SKent Overstreet /* 6631c6fdbd8SKent Overstreet * Not strictly necessary, but helps avoid dio writes livelocking in 6641c6fdbd8SKent Overstreet * write_invalidate_inode_pages_range() - can drop this if/when we get 6651c6fdbd8SKent Overstreet * a write_invalidate_inode_pages_range() that works without dropping 6661c6fdbd8SKent Overstreet * page lock before invalidating page 6671c6fdbd8SKent Overstreet */ 6681c6fdbd8SKent Overstreet bch2_pagecache_add_get(&inode->ei_pagecache_lock); 6691c6fdbd8SKent Overstreet 6701c6fdbd8SKent Overstreet lock_page(page); 6711c6fdbd8SKent Overstreet if (page->mapping != mapping || 6721c6fdbd8SKent Overstreet page_offset(page) > i_size_read(&inode->v)) { 6731c6fdbd8SKent Overstreet unlock_page(page); 6741c6fdbd8SKent Overstreet ret = VM_FAULT_NOPAGE; 6751c6fdbd8SKent Overstreet goto out; 6761c6fdbd8SKent Overstreet } 6771c6fdbd8SKent Overstreet 6781c6fdbd8SKent Overstreet if (bch2_get_page_reservation(c, inode, page, true)) { 6791c6fdbd8SKent Overstreet unlock_page(page); 6801c6fdbd8SKent Overstreet ret = VM_FAULT_SIGBUS; 6811c6fdbd8SKent Overstreet goto out; 6821c6fdbd8SKent Overstreet } 6831c6fdbd8SKent Overstreet 6841c6fdbd8SKent Overstreet if (!PageDirty(page)) 6851c6fdbd8SKent Overstreet set_page_dirty(page); 6861c6fdbd8SKent Overstreet wait_for_stable_page(page); 6871c6fdbd8SKent Overstreet out: 6881c6fdbd8SKent Overstreet bch2_pagecache_add_put(&inode->ei_pagecache_lock); 6891c6fdbd8SKent Overstreet sb_end_pagefault(inode->v.i_sb); 6901c6fdbd8SKent Overstreet return ret; 6911c6fdbd8SKent Overstreet } 6921c6fdbd8SKent Overstreet 6931c6fdbd8SKent Overstreet void bch2_invalidate_folio(struct folio *folio, size_t offset, size_t length) 6941c6fdbd8SKent Overstreet { 6951c6fdbd8SKent Overstreet EBUG_ON(!PageLocked(&folio->page)); 6961c6fdbd8SKent Overstreet EBUG_ON(folio_test_writeback(folio)); 6971c6fdbd8SKent Overstreet 6981c6fdbd8SKent Overstreet if (offset || length < folio_size(folio)) 6991c6fdbd8SKent Overstreet return; 7001c6fdbd8SKent Overstreet 7011c6fdbd8SKent Overstreet bch2_clear_page_bits(&folio->page); 7021c6fdbd8SKent Overstreet } 7031c6fdbd8SKent Overstreet 7041c6fdbd8SKent Overstreet bool bch2_release_folio(struct folio *folio, gfp_t gfp_mask) 7051c6fdbd8SKent Overstreet { 7061c6fdbd8SKent Overstreet /* XXX: this can't take locks that are held while we allocate memory */ 7071c6fdbd8SKent Overstreet EBUG_ON(!PageLocked(&folio->page)); 7081c6fdbd8SKent Overstreet EBUG_ON(folio_test_writeback(folio)); 7091c6fdbd8SKent Overstreet 7101c6fdbd8SKent Overstreet if (folio_test_dirty(folio)) 7111c6fdbd8SKent Overstreet return false; 7121c6fdbd8SKent Overstreet 7131c6fdbd8SKent Overstreet bch2_clear_page_bits(&folio->page); 7141c6fdbd8SKent Overstreet return true; 7151c6fdbd8SKent Overstreet } 7161c6fdbd8SKent Overstreet 7171c6fdbd8SKent Overstreet /* readpages/writepages: */ 7181c6fdbd8SKent Overstreet 7191c6fdbd8SKent Overstreet static bool bio_can_add_page_contig(struct bio *bio, struct page *page) 7201c6fdbd8SKent Overstreet { 7211c6fdbd8SKent Overstreet sector_t offset = (sector_t) page->index << PAGE_SECTOR_SHIFT; 7221c6fdbd8SKent Overstreet 7231c6fdbd8SKent Overstreet return bio->bi_vcnt < bio->bi_max_vecs && 7241c6fdbd8SKent Overstreet bio_end_sector(bio) == offset; 7251c6fdbd8SKent Overstreet } 7261c6fdbd8SKent Overstreet 7271c6fdbd8SKent Overstreet static int bio_add_page_contig(struct bio *bio, struct page *page) 7281c6fdbd8SKent Overstreet { 7291c6fdbd8SKent Overstreet sector_t offset = (sector_t) page->index << PAGE_SECTOR_SHIFT; 7301c6fdbd8SKent Overstreet 7311c6fdbd8SKent Overstreet EBUG_ON(!bio->bi_max_vecs); 7321c6fdbd8SKent Overstreet 7331c6fdbd8SKent Overstreet if (!bio->bi_vcnt) 7341c6fdbd8SKent Overstreet bio->bi_iter.bi_sector = offset; 7351c6fdbd8SKent Overstreet else if (!bio_can_add_page_contig(bio, page)) 7361c6fdbd8SKent Overstreet return -1; 7371c6fdbd8SKent Overstreet 7381c6fdbd8SKent Overstreet __bio_add_page(bio, page, PAGE_SIZE, 0); 7391c6fdbd8SKent Overstreet return 0; 7401c6fdbd8SKent Overstreet } 7411c6fdbd8SKent Overstreet 7421c6fdbd8SKent Overstreet /* readpage(s): */ 7431c6fdbd8SKent Overstreet 7441c6fdbd8SKent Overstreet static void bch2_readpages_end_io(struct bio *bio) 7451c6fdbd8SKent Overstreet { 7461c6fdbd8SKent Overstreet struct bvec_iter_all iter; 7471c6fdbd8SKent Overstreet struct bio_vec *bv; 7481c6fdbd8SKent Overstreet 7491c6fdbd8SKent Overstreet bio_for_each_segment_all(bv, bio, iter) { 7501c6fdbd8SKent Overstreet struct page *page = bv->bv_page; 7511c6fdbd8SKent Overstreet 7521c6fdbd8SKent Overstreet if (!bio->bi_status) { 7531c6fdbd8SKent Overstreet SetPageUptodate(page); 7541c6fdbd8SKent Overstreet } else { 7551c6fdbd8SKent Overstreet ClearPageUptodate(page); 7561c6fdbd8SKent Overstreet SetPageError(page); 7571c6fdbd8SKent Overstreet } 7581c6fdbd8SKent Overstreet unlock_page(page); 7591c6fdbd8SKent Overstreet } 7601c6fdbd8SKent Overstreet 7611c6fdbd8SKent Overstreet bio_put(bio); 7621c6fdbd8SKent Overstreet } 7631c6fdbd8SKent Overstreet 7641c6fdbd8SKent Overstreet static inline void page_state_init_for_read(struct page *page) 7651c6fdbd8SKent Overstreet { 7661c6fdbd8SKent Overstreet struct bch_page_state *s = page_state(page); 7671c6fdbd8SKent Overstreet 7681c6fdbd8SKent Overstreet BUG_ON(s->reserved); 7691c6fdbd8SKent Overstreet s->sectors = 0; 7701c6fdbd8SKent Overstreet s->compressed = 0; 7711c6fdbd8SKent Overstreet } 7721c6fdbd8SKent Overstreet 7731c6fdbd8SKent Overstreet struct readpages_iter { 7741c6fdbd8SKent Overstreet struct address_space *mapping; 7751c6fdbd8SKent Overstreet struct page **pages; 7761c6fdbd8SKent Overstreet unsigned nr_pages; 7771c6fdbd8SKent Overstreet unsigned idx; 7781c6fdbd8SKent Overstreet pgoff_t offset; 7791c6fdbd8SKent Overstreet }; 7801c6fdbd8SKent Overstreet 7811c6fdbd8SKent Overstreet static int readpages_iter_init(struct readpages_iter *iter, 7821c6fdbd8SKent Overstreet struct readahead_control *ractl) 7831c6fdbd8SKent Overstreet { 7841c6fdbd8SKent Overstreet unsigned i, nr_pages = readahead_count(ractl); 7851c6fdbd8SKent Overstreet 7861c6fdbd8SKent Overstreet memset(iter, 0, sizeof(*iter)); 7871c6fdbd8SKent Overstreet 7881c6fdbd8SKent Overstreet iter->mapping = ractl->mapping; 7891c6fdbd8SKent Overstreet iter->offset = readahead_index(ractl); 7901c6fdbd8SKent Overstreet iter->nr_pages = nr_pages; 7911c6fdbd8SKent Overstreet 7921c6fdbd8SKent Overstreet iter->pages = kmalloc_array(nr_pages, sizeof(struct page *), GFP_NOFS); 7931c6fdbd8SKent Overstreet if (!iter->pages) 7941c6fdbd8SKent Overstreet return -ENOMEM; 7951c6fdbd8SKent Overstreet 7961c6fdbd8SKent Overstreet __readahead_batch(ractl, iter->pages, nr_pages); 7971c6fdbd8SKent Overstreet for (i = 0; i < nr_pages; i++) { 7981c6fdbd8SKent Overstreet put_page(iter->pages[i]); 7991c6fdbd8SKent Overstreet } 8001c6fdbd8SKent Overstreet 8011c6fdbd8SKent Overstreet return 0; 8021c6fdbd8SKent Overstreet } 8031c6fdbd8SKent Overstreet 8041c6fdbd8SKent Overstreet static inline struct page *readpage_iter_next(struct readpages_iter *iter) 8051c6fdbd8SKent Overstreet { 8061c6fdbd8SKent Overstreet if (iter->idx >= iter->nr_pages) 8071c6fdbd8SKent Overstreet return NULL; 8081c6fdbd8SKent Overstreet 8091c6fdbd8SKent Overstreet EBUG_ON(iter->pages[iter->idx]->index != iter->offset + iter->idx); 8101c6fdbd8SKent Overstreet 8111c6fdbd8SKent Overstreet page_state_init_for_read(iter->pages[iter->idx]); 8121c6fdbd8SKent Overstreet return iter->pages[iter->idx]; 8131c6fdbd8SKent Overstreet } 8141c6fdbd8SKent Overstreet 8151c6fdbd8SKent Overstreet static void bch2_add_page_sectors(struct bio *bio, struct bkey_s_c k) 8161c6fdbd8SKent Overstreet { 8171c6fdbd8SKent Overstreet struct bvec_iter iter; 8181c6fdbd8SKent Overstreet struct bio_vec bv; 8191c6fdbd8SKent Overstreet bool compressed = bch2_extent_is_compressed(k); 8201c6fdbd8SKent Overstreet unsigned nr_ptrs = bch2_extent_nr_dirty_ptrs(k); 8211c6fdbd8SKent Overstreet 8221c6fdbd8SKent Overstreet bio_for_each_segment(bv, bio, iter) { 8231c6fdbd8SKent Overstreet struct bch_page_state *s = page_state(bv.bv_page); 8241c6fdbd8SKent Overstreet 8251c6fdbd8SKent Overstreet /* sectors in @k from the start of this page: */ 8261c6fdbd8SKent Overstreet unsigned k_sectors = k.k->size - (iter.bi_sector - k.k->p.offset); 8271c6fdbd8SKent Overstreet 8281c6fdbd8SKent Overstreet unsigned page_sectors = min(bv.bv_len >> 9, k_sectors); 8291c6fdbd8SKent Overstreet 8301c6fdbd8SKent Overstreet s->nr_replicas = !s->sectors 8311c6fdbd8SKent Overstreet ? nr_ptrs 8321c6fdbd8SKent Overstreet : min_t(unsigned, s->nr_replicas, nr_ptrs); 8331c6fdbd8SKent Overstreet 8341c6fdbd8SKent Overstreet BUG_ON(s->sectors + page_sectors > PAGE_SECTORS); 8351c6fdbd8SKent Overstreet s->sectors += page_sectors; 8361c6fdbd8SKent Overstreet 8371c6fdbd8SKent Overstreet s->compressed |= compressed; 8381c6fdbd8SKent Overstreet } 8391c6fdbd8SKent Overstreet } 8401c6fdbd8SKent Overstreet 8411c6fdbd8SKent Overstreet static void readpage_bio_extend(struct readpages_iter *iter, 8421c6fdbd8SKent Overstreet struct bio *bio, u64 offset, 8431c6fdbd8SKent Overstreet bool get_more) 8441c6fdbd8SKent Overstreet { 8451c6fdbd8SKent Overstreet while (bio_end_sector(bio) < offset && 8461c6fdbd8SKent Overstreet bio->bi_vcnt < bio->bi_max_vecs) { 8471c6fdbd8SKent Overstreet pgoff_t page_offset = bio_end_sector(bio) >> PAGE_SECTOR_SHIFT; 8481c6fdbd8SKent Overstreet struct page *page = readpage_iter_next(iter); 8491c6fdbd8SKent Overstreet int ret; 8501c6fdbd8SKent Overstreet 8511c6fdbd8SKent Overstreet if (page) { 8521c6fdbd8SKent Overstreet if (iter->offset + iter->idx != page_offset) 8531c6fdbd8SKent Overstreet break; 8541c6fdbd8SKent Overstreet 8551c6fdbd8SKent Overstreet iter->idx++; 8561c6fdbd8SKent Overstreet } else { 8571c6fdbd8SKent Overstreet if (!get_more) 8581c6fdbd8SKent Overstreet break; 8591c6fdbd8SKent Overstreet 8601c6fdbd8SKent Overstreet page = xa_load(&iter->mapping->i_pages, page_offset); 8611c6fdbd8SKent Overstreet if (page && !xa_is_value(page)) 8621c6fdbd8SKent Overstreet break; 8631c6fdbd8SKent Overstreet 8641c6fdbd8SKent Overstreet page = __page_cache_alloc(readahead_gfp_mask(iter->mapping)); 8651c6fdbd8SKent Overstreet if (!page) 8661c6fdbd8SKent Overstreet break; 8671c6fdbd8SKent Overstreet 8681c6fdbd8SKent Overstreet page_state_init_for_read(page); 8691c6fdbd8SKent Overstreet 8701c6fdbd8SKent Overstreet ret = add_to_page_cache_lru(page, iter->mapping, 8711c6fdbd8SKent Overstreet page_offset, GFP_NOFS); 8721c6fdbd8SKent Overstreet if (ret) { 8731c6fdbd8SKent Overstreet ClearPagePrivate(page); 8741c6fdbd8SKent Overstreet put_page(page); 8751c6fdbd8SKent Overstreet break; 8761c6fdbd8SKent Overstreet } 8771c6fdbd8SKent Overstreet 8781c6fdbd8SKent Overstreet put_page(page); 8791c6fdbd8SKent Overstreet } 8801c6fdbd8SKent Overstreet 8811c6fdbd8SKent Overstreet __bio_add_page(bio, page, PAGE_SIZE, 0); 8821c6fdbd8SKent Overstreet } 8831c6fdbd8SKent Overstreet } 8841c6fdbd8SKent Overstreet 8851c6fdbd8SKent Overstreet static void bchfs_read(struct bch_fs *c, struct btree_iter *iter, 8861c6fdbd8SKent Overstreet struct bch_read_bio *rbio, u64 inum, 8871c6fdbd8SKent Overstreet struct readpages_iter *readpages_iter) 8881c6fdbd8SKent Overstreet { 8891c6fdbd8SKent Overstreet struct bio *bio = &rbio->bio; 8901c6fdbd8SKent Overstreet int flags = BCH_READ_RETRY_IF_STALE| 8911c6fdbd8SKent Overstreet BCH_READ_MAY_PROMOTE; 8921c6fdbd8SKent Overstreet 8931c6fdbd8SKent Overstreet rbio->c = c; 8941c6fdbd8SKent Overstreet rbio->start_time = local_clock(); 8951c6fdbd8SKent Overstreet 8961c6fdbd8SKent Overstreet while (1) { 8971c6fdbd8SKent Overstreet BKEY_PADDED(k) tmp; 8981c6fdbd8SKent Overstreet struct bkey_s_c k; 8991c6fdbd8SKent Overstreet unsigned bytes; 9001c6fdbd8SKent Overstreet 9011c6fdbd8SKent Overstreet bch2_btree_iter_set_pos(iter, POS(inum, bio->bi_iter.bi_sector)); 9021c6fdbd8SKent Overstreet 9031c6fdbd8SKent Overstreet k = bch2_btree_iter_peek_slot(iter); 9041c6fdbd8SKent Overstreet BUG_ON(!k.k); 9051c6fdbd8SKent Overstreet 9061c6fdbd8SKent Overstreet if (IS_ERR(k.k)) { 9071c6fdbd8SKent Overstreet int ret = bch2_btree_iter_unlock(iter); 9081c6fdbd8SKent Overstreet BUG_ON(!ret); 9091c6fdbd8SKent Overstreet bcache_io_error(c, bio, "btree IO error %i", ret); 9101c6fdbd8SKent Overstreet bio_endio(bio); 9111c6fdbd8SKent Overstreet return; 9121c6fdbd8SKent Overstreet } 9131c6fdbd8SKent Overstreet 9141c6fdbd8SKent Overstreet bkey_reassemble(&tmp.k, k); 9151c6fdbd8SKent Overstreet bch2_btree_iter_unlock(iter); 9161c6fdbd8SKent Overstreet k = bkey_i_to_s_c(&tmp.k); 9171c6fdbd8SKent Overstreet 9181c6fdbd8SKent Overstreet if (readpages_iter) { 9191c6fdbd8SKent Overstreet bool want_full_extent = false; 9201c6fdbd8SKent Overstreet 9211c6fdbd8SKent Overstreet if (bkey_extent_is_data(k.k)) { 9221c6fdbd8SKent Overstreet struct bkey_s_c_extent e = bkey_s_c_to_extent(k); 9231c6fdbd8SKent Overstreet const union bch_extent_entry *i; 9241742237bSKent Overstreet struct extent_ptr_decoded p; 9251c6fdbd8SKent Overstreet 9261742237bSKent Overstreet extent_for_each_ptr_decode(e, p, i) 9271742237bSKent Overstreet want_full_extent |= ((p.crc.csum_type != 0) | 9281742237bSKent Overstreet (p.crc.compression_type != 0)); 9291c6fdbd8SKent Overstreet } 9301c6fdbd8SKent Overstreet 9311c6fdbd8SKent Overstreet readpage_bio_extend(readpages_iter, 9321c6fdbd8SKent Overstreet bio, k.k->p.offset, 9331c6fdbd8SKent Overstreet want_full_extent); 9341c6fdbd8SKent Overstreet } 9351c6fdbd8SKent Overstreet 9361c6fdbd8SKent Overstreet bytes = (min_t(u64, k.k->p.offset, bio_end_sector(bio)) - 9371c6fdbd8SKent Overstreet bio->bi_iter.bi_sector) << 9; 9381c6fdbd8SKent Overstreet swap(bio->bi_iter.bi_size, bytes); 9391c6fdbd8SKent Overstreet 9401c6fdbd8SKent Overstreet if (bytes == bio->bi_iter.bi_size) 9411c6fdbd8SKent Overstreet flags |= BCH_READ_LAST_FRAGMENT; 9421c6fdbd8SKent Overstreet 9431c6fdbd8SKent Overstreet if (bkey_extent_is_allocation(k.k)) 9441c6fdbd8SKent Overstreet bch2_add_page_sectors(bio, k); 9451c6fdbd8SKent Overstreet 9461c6fdbd8SKent Overstreet bch2_read_extent(c, rbio, k, flags); 9471c6fdbd8SKent Overstreet 9481c6fdbd8SKent Overstreet if (flags & BCH_READ_LAST_FRAGMENT) 9491c6fdbd8SKent Overstreet return; 9501c6fdbd8SKent Overstreet 9511c6fdbd8SKent Overstreet swap(bio->bi_iter.bi_size, bytes); 9521c6fdbd8SKent Overstreet bio_advance(bio, bytes); 9531c6fdbd8SKent Overstreet } 9541c6fdbd8SKent Overstreet } 9551c6fdbd8SKent Overstreet 9561c6fdbd8SKent Overstreet void bch2_readahead(struct readahead_control *ractl) 9571c6fdbd8SKent Overstreet { 9581c6fdbd8SKent Overstreet struct bch_inode_info *inode = to_bch_ei(ractl->mapping->host); 9591c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 9601c6fdbd8SKent Overstreet struct bch_io_opts opts = io_opts(c, inode); 9611c6fdbd8SKent Overstreet struct btree_iter iter; 9621c6fdbd8SKent Overstreet struct page *page; 9631c6fdbd8SKent Overstreet struct readpages_iter readpages_iter; 9641c6fdbd8SKent Overstreet int ret; 9651c6fdbd8SKent Overstreet 9661c6fdbd8SKent Overstreet ret = readpages_iter_init(&readpages_iter, ractl); 9671c6fdbd8SKent Overstreet BUG_ON(ret); 9681c6fdbd8SKent Overstreet 9691c6fdbd8SKent Overstreet bch2_btree_iter_init(&iter, c, BTREE_ID_EXTENTS, POS_MIN, 9701c6fdbd8SKent Overstreet BTREE_ITER_SLOTS); 9711c6fdbd8SKent Overstreet 9721c6fdbd8SKent Overstreet bch2_pagecache_add_get(&inode->ei_pagecache_lock); 9731c6fdbd8SKent Overstreet 9741c6fdbd8SKent Overstreet while ((page = readpage_iter_next(&readpages_iter))) { 9751c6fdbd8SKent Overstreet pgoff_t index = readpages_iter.offset + readpages_iter.idx; 9761c6fdbd8SKent Overstreet unsigned n = min_t(unsigned, 9771c6fdbd8SKent Overstreet readpages_iter.nr_pages - 9781c6fdbd8SKent Overstreet readpages_iter.idx, 9791c6fdbd8SKent Overstreet BIO_MAX_VECS); 9801c6fdbd8SKent Overstreet struct bch_read_bio *rbio = 9811c6fdbd8SKent Overstreet rbio_init(bio_alloc_bioset(NULL, n, REQ_OP_READ, 9821c6fdbd8SKent Overstreet GFP_NOFS, &c->bio_read), 9831c6fdbd8SKent Overstreet opts); 9841c6fdbd8SKent Overstreet 9851c6fdbd8SKent Overstreet readpages_iter.idx++; 9861c6fdbd8SKent Overstreet 9871c6fdbd8SKent Overstreet rbio->bio.bi_iter.bi_sector = (sector_t) index << PAGE_SECTOR_SHIFT; 9881c6fdbd8SKent Overstreet rbio->bio.bi_end_io = bch2_readpages_end_io; 9891c6fdbd8SKent Overstreet __bio_add_page(&rbio->bio, page, PAGE_SIZE, 0); 9901c6fdbd8SKent Overstreet 9911c6fdbd8SKent Overstreet bchfs_read(c, &iter, rbio, inode->v.i_ino, &readpages_iter); 9921c6fdbd8SKent Overstreet } 9931c6fdbd8SKent Overstreet 9941c6fdbd8SKent Overstreet bch2_pagecache_add_put(&inode->ei_pagecache_lock); 9951c6fdbd8SKent Overstreet kfree(readpages_iter.pages); 9961c6fdbd8SKent Overstreet } 9971c6fdbd8SKent Overstreet 9981c6fdbd8SKent Overstreet static void __bchfs_readpage(struct bch_fs *c, struct bch_read_bio *rbio, 9991c6fdbd8SKent Overstreet u64 inum, struct page *page) 10001c6fdbd8SKent Overstreet { 10011c6fdbd8SKent Overstreet struct btree_iter iter; 10021c6fdbd8SKent Overstreet 10031c6fdbd8SKent Overstreet page_state_init_for_read(page); 10041c6fdbd8SKent Overstreet 10051c6fdbd8SKent Overstreet rbio->bio.bi_opf = REQ_OP_READ|REQ_SYNC; 10061c6fdbd8SKent Overstreet bio_add_page_contig(&rbio->bio, page); 10071c6fdbd8SKent Overstreet 10081c6fdbd8SKent Overstreet bch2_btree_iter_init(&iter, c, BTREE_ID_EXTENTS, POS_MIN, 10091c6fdbd8SKent Overstreet BTREE_ITER_SLOTS); 10101c6fdbd8SKent Overstreet bchfs_read(c, &iter, rbio, inum, NULL); 10111c6fdbd8SKent Overstreet } 10121c6fdbd8SKent Overstreet 10131c6fdbd8SKent Overstreet static void bch2_read_single_page_end_io(struct bio *bio) 10141c6fdbd8SKent Overstreet { 10151c6fdbd8SKent Overstreet complete(bio->bi_private); 10161c6fdbd8SKent Overstreet } 10171c6fdbd8SKent Overstreet 10181c6fdbd8SKent Overstreet static int bch2_read_single_page(struct page *page, 10191c6fdbd8SKent Overstreet struct address_space *mapping) 10201c6fdbd8SKent Overstreet { 10211c6fdbd8SKent Overstreet struct bch_inode_info *inode = to_bch_ei(mapping->host); 10221c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 10231c6fdbd8SKent Overstreet struct bch_read_bio *rbio; 10241c6fdbd8SKent Overstreet int ret; 10251c6fdbd8SKent Overstreet DECLARE_COMPLETION_ONSTACK(done); 10261c6fdbd8SKent Overstreet 10271c6fdbd8SKent Overstreet rbio = rbio_init(bio_alloc_bioset(NULL, 1, REQ_OP_READ, GFP_NOFS, &c->bio_read), 10281c6fdbd8SKent Overstreet io_opts(c, inode)); 10291c6fdbd8SKent Overstreet rbio->bio.bi_private = &done; 10301c6fdbd8SKent Overstreet rbio->bio.bi_end_io = bch2_read_single_page_end_io; 10311c6fdbd8SKent Overstreet 10321c6fdbd8SKent Overstreet __bchfs_readpage(c, rbio, inode->v.i_ino, page); 10331c6fdbd8SKent Overstreet wait_for_completion(&done); 10341c6fdbd8SKent Overstreet 10351c6fdbd8SKent Overstreet ret = blk_status_to_errno(rbio->bio.bi_status); 10361c6fdbd8SKent Overstreet bio_put(&rbio->bio); 10371c6fdbd8SKent Overstreet 10381c6fdbd8SKent Overstreet if (ret < 0) 10391c6fdbd8SKent Overstreet return ret; 10401c6fdbd8SKent Overstreet 10411c6fdbd8SKent Overstreet SetPageUptodate(page); 10421c6fdbd8SKent Overstreet return 0; 10431c6fdbd8SKent Overstreet } 10441c6fdbd8SKent Overstreet 10451c6fdbd8SKent Overstreet int bch2_read_folio(struct file *file, struct folio *folio) 10461c6fdbd8SKent Overstreet { 10471c6fdbd8SKent Overstreet struct page *page = &folio->page; 10481c6fdbd8SKent Overstreet int ret; 10491c6fdbd8SKent Overstreet 10501c6fdbd8SKent Overstreet ret = bch2_read_single_page(page, page->mapping); 10511c6fdbd8SKent Overstreet folio_unlock(folio); 10521c6fdbd8SKent Overstreet return ret; 10531c6fdbd8SKent Overstreet } 10541c6fdbd8SKent Overstreet 10551c6fdbd8SKent Overstreet /* writepages: */ 10561c6fdbd8SKent Overstreet 10571c6fdbd8SKent Overstreet struct bch_writepage_state { 10581c6fdbd8SKent Overstreet struct bch_writepage_io *io; 10591c6fdbd8SKent Overstreet struct bch_io_opts opts; 10601c6fdbd8SKent Overstreet }; 10611c6fdbd8SKent Overstreet 10621c6fdbd8SKent Overstreet static inline struct bch_writepage_state bch_writepage_state_init(struct bch_fs *c, 10631c6fdbd8SKent Overstreet struct bch_inode_info *inode) 10641c6fdbd8SKent Overstreet { 10651c6fdbd8SKent Overstreet return (struct bch_writepage_state) { .opts = io_opts(c, inode) }; 10661c6fdbd8SKent Overstreet } 10671c6fdbd8SKent Overstreet 10681c6fdbd8SKent Overstreet static void bch2_writepage_io_free(struct closure *cl) 10691c6fdbd8SKent Overstreet { 10701c6fdbd8SKent Overstreet struct bch_writepage_io *io = container_of(cl, 10711c6fdbd8SKent Overstreet struct bch_writepage_io, cl); 10721c6fdbd8SKent Overstreet 10731c6fdbd8SKent Overstreet bio_put(&io->op.op.wbio.bio); 10741c6fdbd8SKent Overstreet } 10751c6fdbd8SKent Overstreet 10761c6fdbd8SKent Overstreet static void bch2_writepage_io_done(struct closure *cl) 10771c6fdbd8SKent Overstreet { 10781c6fdbd8SKent Overstreet struct bch_writepage_io *io = container_of(cl, 10791c6fdbd8SKent Overstreet struct bch_writepage_io, cl); 10801c6fdbd8SKent Overstreet struct bch_fs *c = io->op.op.c; 10811c6fdbd8SKent Overstreet struct bio *bio = &io->op.op.wbio.bio; 10821c6fdbd8SKent Overstreet struct bvec_iter_all iter; 10831c6fdbd8SKent Overstreet struct bio_vec *bvec; 10841c6fdbd8SKent Overstreet 10851c6fdbd8SKent Overstreet if (io->op.op.error) { 10861c6fdbd8SKent Overstreet bio_for_each_segment_all(bvec, bio, iter) 10871c6fdbd8SKent Overstreet SetPageError(bvec->bv_page); 10881c6fdbd8SKent Overstreet set_bit(AS_EIO, &io->op.inode->v.i_mapping->flags); 10891c6fdbd8SKent Overstreet } 10901c6fdbd8SKent Overstreet 10911c6fdbd8SKent Overstreet /* 10921c6fdbd8SKent Overstreet * racing with fallocate can cause us to add fewer sectors than 10931c6fdbd8SKent Overstreet * expected - but we shouldn't add more sectors than expected: 10941c6fdbd8SKent Overstreet */ 10951c6fdbd8SKent Overstreet BUG_ON(io->op.sectors_added > (s64) io->new_sectors); 10961c6fdbd8SKent Overstreet 10971c6fdbd8SKent Overstreet /* 10981c6fdbd8SKent Overstreet * (error (due to going RO) halfway through a page can screw that up 10991c6fdbd8SKent Overstreet * slightly) 11001c6fdbd8SKent Overstreet * XXX wtf? 11011c6fdbd8SKent Overstreet BUG_ON(io->op.sectors_added - io->new_sectors >= (s64) PAGE_SECTORS); 11021c6fdbd8SKent Overstreet */ 11031c6fdbd8SKent Overstreet 11041c6fdbd8SKent Overstreet /* 11051c6fdbd8SKent Overstreet * PageWriteback is effectively our ref on the inode - fixup i_blocks 11061c6fdbd8SKent Overstreet * before calling end_page_writeback: 11071c6fdbd8SKent Overstreet */ 11081c6fdbd8SKent Overstreet if (io->op.sectors_added != io->new_sectors) 11091c6fdbd8SKent Overstreet i_sectors_acct(c, io->op.inode, NULL, 11101c6fdbd8SKent Overstreet io->op.sectors_added - (s64) io->new_sectors); 11111c6fdbd8SKent Overstreet 11121c6fdbd8SKent Overstreet bio_for_each_segment_all(bvec, bio, iter) 11131c6fdbd8SKent Overstreet end_page_writeback(bvec->bv_page); 11141c6fdbd8SKent Overstreet 11151c6fdbd8SKent Overstreet closure_return_with_destructor(&io->cl, bch2_writepage_io_free); 11161c6fdbd8SKent Overstreet } 11171c6fdbd8SKent Overstreet 11181c6fdbd8SKent Overstreet static void bch2_writepage_do_io(struct bch_writepage_state *w) 11191c6fdbd8SKent Overstreet { 11201c6fdbd8SKent Overstreet struct bch_writepage_io *io = w->io; 11211c6fdbd8SKent Overstreet 11221c6fdbd8SKent Overstreet w->io = NULL; 11231c6fdbd8SKent Overstreet closure_call(&io->op.op.cl, bch2_write, NULL, &io->cl); 11241c6fdbd8SKent Overstreet continue_at(&io->cl, bch2_writepage_io_done, NULL); 11251c6fdbd8SKent Overstreet } 11261c6fdbd8SKent Overstreet 11271c6fdbd8SKent Overstreet /* 11281c6fdbd8SKent Overstreet * Get a bch_writepage_io and add @page to it - appending to an existing one if 11291c6fdbd8SKent Overstreet * possible, else allocating a new one: 11301c6fdbd8SKent Overstreet */ 11311c6fdbd8SKent Overstreet static void bch2_writepage_io_alloc(struct bch_fs *c, 11321c6fdbd8SKent Overstreet struct bch_writepage_state *w, 11331c6fdbd8SKent Overstreet struct bch_inode_info *inode, 11341c6fdbd8SKent Overstreet struct page *page, 11351c6fdbd8SKent Overstreet unsigned nr_replicas) 11361c6fdbd8SKent Overstreet { 11371c6fdbd8SKent Overstreet struct bch_write_op *op; 11381c6fdbd8SKent Overstreet u64 offset = (u64) page->index << PAGE_SECTOR_SHIFT; 11391c6fdbd8SKent Overstreet 11401c6fdbd8SKent Overstreet w->io = container_of(bio_alloc_bioset(NULL, BIO_MAX_VECS, 11411c6fdbd8SKent Overstreet REQ_OP_WRITE, 11421c6fdbd8SKent Overstreet GFP_NOFS, 11431c6fdbd8SKent Overstreet &c->writepage_bioset), 11441c6fdbd8SKent Overstreet struct bch_writepage_io, op.op.wbio.bio); 11451c6fdbd8SKent Overstreet 11461c6fdbd8SKent Overstreet closure_init(&w->io->cl, NULL); 11471c6fdbd8SKent Overstreet w->io->new_sectors = 0; 11481c6fdbd8SKent Overstreet bch2_fswrite_op_init(&w->io->op, c, inode, w->opts, false); 11491c6fdbd8SKent Overstreet op = &w->io->op.op; 11501c6fdbd8SKent Overstreet op->nr_replicas = nr_replicas; 11511c6fdbd8SKent Overstreet op->res.nr_replicas = nr_replicas; 11521c6fdbd8SKent Overstreet op->write_point = writepoint_hashed(inode->ei_last_dirtied); 11531c6fdbd8SKent Overstreet op->pos = POS(inode->v.i_ino, offset); 11541c6fdbd8SKent Overstreet op->wbio.bio.bi_iter.bi_sector = offset; 11551c6fdbd8SKent Overstreet } 11561c6fdbd8SKent Overstreet 11571c6fdbd8SKent Overstreet static int __bch2_writepage(struct folio *folio, 11581c6fdbd8SKent Overstreet struct writeback_control *wbc, 11591c6fdbd8SKent Overstreet void *data) 11601c6fdbd8SKent Overstreet { 11611c6fdbd8SKent Overstreet struct page *page = &folio->page; 11621c6fdbd8SKent Overstreet struct bch_inode_info *inode = to_bch_ei(page->mapping->host); 11631c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 11641c6fdbd8SKent Overstreet struct bch_writepage_state *w = data; 11651c6fdbd8SKent Overstreet struct bch_page_state new, old; 11661c6fdbd8SKent Overstreet unsigned offset; 11671c6fdbd8SKent Overstreet loff_t i_size = i_size_read(&inode->v); 11681c6fdbd8SKent Overstreet pgoff_t end_index = i_size >> PAGE_SHIFT; 11691c6fdbd8SKent Overstreet 11701c6fdbd8SKent Overstreet EBUG_ON(!PageUptodate(page)); 11711c6fdbd8SKent Overstreet 11721c6fdbd8SKent Overstreet /* Is the page fully inside i_size? */ 11731c6fdbd8SKent Overstreet if (page->index < end_index) 11741c6fdbd8SKent Overstreet goto do_io; 11751c6fdbd8SKent Overstreet 11761c6fdbd8SKent Overstreet /* Is the page fully outside i_size? (truncate in progress) */ 11771c6fdbd8SKent Overstreet offset = i_size & (PAGE_SIZE - 1); 11781c6fdbd8SKent Overstreet if (page->index > end_index || !offset) { 11791c6fdbd8SKent Overstreet unlock_page(page); 11801c6fdbd8SKent Overstreet return 0; 11811c6fdbd8SKent Overstreet } 11821c6fdbd8SKent Overstreet 11831c6fdbd8SKent Overstreet /* 11841c6fdbd8SKent Overstreet * The page straddles i_size. It must be zeroed out on each and every 11851c6fdbd8SKent Overstreet * writepage invocation because it may be mmapped. "A file is mapped 11861c6fdbd8SKent Overstreet * in multiples of the page size. For a file that is not a multiple of 11871c6fdbd8SKent Overstreet * the page size, the remaining memory is zeroed when mapped, and 11881c6fdbd8SKent Overstreet * writes to that region are not written out to the file." 11891c6fdbd8SKent Overstreet */ 11901c6fdbd8SKent Overstreet zero_user_segment(page, offset, PAGE_SIZE); 11911c6fdbd8SKent Overstreet do_io: 11921c6fdbd8SKent Overstreet /* Before unlocking the page, transfer reservation to w->io: */ 11931c6fdbd8SKent Overstreet old = page_state_cmpxchg(page_state(page), new, { 11941c6fdbd8SKent Overstreet EBUG_ON(!new.reserved && 11951c6fdbd8SKent Overstreet (new.sectors != PAGE_SECTORS || 11961c6fdbd8SKent Overstreet new.compressed)); 11971c6fdbd8SKent Overstreet 11981c6fdbd8SKent Overstreet if (new.reserved) 11991c6fdbd8SKent Overstreet new.nr_replicas = new.reservation_replicas; 12001c6fdbd8SKent Overstreet new.reserved = 0; 12011c6fdbd8SKent Overstreet 12021c6fdbd8SKent Overstreet new.compressed |= w->opts.compression != 0; 12031c6fdbd8SKent Overstreet 12041c6fdbd8SKent Overstreet new.sectors += new.dirty_sectors; 12051c6fdbd8SKent Overstreet new.dirty_sectors = 0; 12061c6fdbd8SKent Overstreet }); 12071c6fdbd8SKent Overstreet 12081c6fdbd8SKent Overstreet BUG_ON(PageWriteback(page)); 12091c6fdbd8SKent Overstreet set_page_writeback(page); 12101c6fdbd8SKent Overstreet unlock_page(page); 12111c6fdbd8SKent Overstreet 12121c6fdbd8SKent Overstreet if (w->io && 12131c6fdbd8SKent Overstreet (w->io->op.op.res.nr_replicas != new.nr_replicas || 12141c6fdbd8SKent Overstreet !bio_can_add_page_contig(&w->io->op.op.wbio.bio, page))) 12151c6fdbd8SKent Overstreet bch2_writepage_do_io(w); 12161c6fdbd8SKent Overstreet 12171c6fdbd8SKent Overstreet if (!w->io) 12181c6fdbd8SKent Overstreet bch2_writepage_io_alloc(c, w, inode, page, new.nr_replicas); 12191c6fdbd8SKent Overstreet 12201c6fdbd8SKent Overstreet w->io->new_sectors += new.sectors - old.sectors; 12211c6fdbd8SKent Overstreet 12221c6fdbd8SKent Overstreet BUG_ON(inode != w->io->op.inode); 12231c6fdbd8SKent Overstreet BUG_ON(bio_add_page_contig(&w->io->op.op.wbio.bio, page)); 12241c6fdbd8SKent Overstreet 12251c6fdbd8SKent Overstreet if (old.reserved) 12261c6fdbd8SKent Overstreet w->io->op.op.res.sectors += old.reservation_replicas * PAGE_SECTORS; 12271c6fdbd8SKent Overstreet 12281c6fdbd8SKent Overstreet w->io->op.new_i_size = i_size; 12291c6fdbd8SKent Overstreet 12301c6fdbd8SKent Overstreet if (wbc->sync_mode == WB_SYNC_ALL) 12311c6fdbd8SKent Overstreet w->io->op.op.wbio.bio.bi_opf |= REQ_SYNC; 12321c6fdbd8SKent Overstreet 12331c6fdbd8SKent Overstreet return 0; 12341c6fdbd8SKent Overstreet } 12351c6fdbd8SKent Overstreet 12361c6fdbd8SKent Overstreet int bch2_writepages(struct address_space *mapping, struct writeback_control *wbc) 12371c6fdbd8SKent Overstreet { 12381c6fdbd8SKent Overstreet struct bch_fs *c = mapping->host->i_sb->s_fs_info; 12391c6fdbd8SKent Overstreet struct bch_writepage_state w = 12401c6fdbd8SKent Overstreet bch_writepage_state_init(c, to_bch_ei(mapping->host)); 12411c6fdbd8SKent Overstreet struct blk_plug plug; 12421c6fdbd8SKent Overstreet int ret; 12431c6fdbd8SKent Overstreet 12441c6fdbd8SKent Overstreet blk_start_plug(&plug); 12451c6fdbd8SKent Overstreet ret = write_cache_pages(mapping, wbc, __bch2_writepage, &w); 12461c6fdbd8SKent Overstreet if (w.io) 12471c6fdbd8SKent Overstreet bch2_writepage_do_io(&w); 12481c6fdbd8SKent Overstreet blk_finish_plug(&plug); 12491c6fdbd8SKent Overstreet return ret; 12501c6fdbd8SKent Overstreet } 12511c6fdbd8SKent Overstreet 12521c6fdbd8SKent Overstreet int bch2_writepage(struct page *page, struct writeback_control *wbc) 12531c6fdbd8SKent Overstreet { 12541c6fdbd8SKent Overstreet struct bch_fs *c = page->mapping->host->i_sb->s_fs_info; 12551c6fdbd8SKent Overstreet struct bch_writepage_state w = 12561c6fdbd8SKent Overstreet bch_writepage_state_init(c, to_bch_ei(page->mapping->host)); 12571c6fdbd8SKent Overstreet int ret; 12581c6fdbd8SKent Overstreet 12591c6fdbd8SKent Overstreet ret = __bch2_writepage(page_folio(page), wbc, &w); 12601c6fdbd8SKent Overstreet if (w.io) 12611c6fdbd8SKent Overstreet bch2_writepage_do_io(&w); 12621c6fdbd8SKent Overstreet 12631c6fdbd8SKent Overstreet return ret; 12641c6fdbd8SKent Overstreet } 12651c6fdbd8SKent Overstreet 12661c6fdbd8SKent Overstreet /* buffered writes: */ 12671c6fdbd8SKent Overstreet 12681c6fdbd8SKent Overstreet int bch2_write_begin(struct file *file, struct address_space *mapping, 12691c6fdbd8SKent Overstreet loff_t pos, unsigned len, 12701c6fdbd8SKent Overstreet struct page **pagep, void **fsdata) 12711c6fdbd8SKent Overstreet { 12721c6fdbd8SKent Overstreet struct bch_inode_info *inode = to_bch_ei(mapping->host); 12731c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 12741c6fdbd8SKent Overstreet pgoff_t index = pos >> PAGE_SHIFT; 12751c6fdbd8SKent Overstreet unsigned offset = pos & (PAGE_SIZE - 1); 12761c6fdbd8SKent Overstreet struct page *page; 12771c6fdbd8SKent Overstreet int ret = -ENOMEM; 12781c6fdbd8SKent Overstreet 12791c6fdbd8SKent Overstreet BUG_ON(inode_unhashed(&inode->v)); 12801c6fdbd8SKent Overstreet 12811c6fdbd8SKent Overstreet bch2_pagecache_add_get(&inode->ei_pagecache_lock); 12821c6fdbd8SKent Overstreet 12831c6fdbd8SKent Overstreet page = grab_cache_page_write_begin(mapping, index); 12841c6fdbd8SKent Overstreet if (!page) 12851c6fdbd8SKent Overstreet goto err_unlock; 12861c6fdbd8SKent Overstreet 12871c6fdbd8SKent Overstreet if (PageUptodate(page)) 12881c6fdbd8SKent Overstreet goto out; 12891c6fdbd8SKent Overstreet 12901c6fdbd8SKent Overstreet /* If we're writing entire page, don't need to read it in first: */ 12911c6fdbd8SKent Overstreet if (len == PAGE_SIZE) 12921c6fdbd8SKent Overstreet goto out; 12931c6fdbd8SKent Overstreet 12941c6fdbd8SKent Overstreet if (!offset && pos + len >= inode->v.i_size) { 12951c6fdbd8SKent Overstreet zero_user_segment(page, len, PAGE_SIZE); 12961c6fdbd8SKent Overstreet flush_dcache_page(page); 12971c6fdbd8SKent Overstreet goto out; 12981c6fdbd8SKent Overstreet } 12991c6fdbd8SKent Overstreet 13001c6fdbd8SKent Overstreet if (index > inode->v.i_size >> PAGE_SHIFT) { 13011c6fdbd8SKent Overstreet zero_user_segments(page, 0, offset, offset + len, PAGE_SIZE); 13021c6fdbd8SKent Overstreet flush_dcache_page(page); 13031c6fdbd8SKent Overstreet goto out; 13041c6fdbd8SKent Overstreet } 13051c6fdbd8SKent Overstreet readpage: 13061c6fdbd8SKent Overstreet ret = bch2_read_single_page(page, mapping); 13071c6fdbd8SKent Overstreet if (ret) 13081c6fdbd8SKent Overstreet goto err; 13091c6fdbd8SKent Overstreet out: 13101c6fdbd8SKent Overstreet ret = bch2_get_page_reservation(c, inode, page, true); 13111c6fdbd8SKent Overstreet if (ret) { 13121c6fdbd8SKent Overstreet if (!PageUptodate(page)) { 13131c6fdbd8SKent Overstreet /* 13141c6fdbd8SKent Overstreet * If the page hasn't been read in, we won't know if we 13151c6fdbd8SKent Overstreet * actually need a reservation - we don't actually need 13161c6fdbd8SKent Overstreet * to read here, we just need to check if the page is 13171c6fdbd8SKent Overstreet * fully backed by uncompressed data: 13181c6fdbd8SKent Overstreet */ 13191c6fdbd8SKent Overstreet goto readpage; 13201c6fdbd8SKent Overstreet } 13211c6fdbd8SKent Overstreet 13221c6fdbd8SKent Overstreet goto err; 13231c6fdbd8SKent Overstreet } 13241c6fdbd8SKent Overstreet 13251c6fdbd8SKent Overstreet *pagep = page; 13261c6fdbd8SKent Overstreet return 0; 13271c6fdbd8SKent Overstreet err: 13281c6fdbd8SKent Overstreet unlock_page(page); 13291c6fdbd8SKent Overstreet put_page(page); 13301c6fdbd8SKent Overstreet *pagep = NULL; 13311c6fdbd8SKent Overstreet err_unlock: 13321c6fdbd8SKent Overstreet bch2_pagecache_add_put(&inode->ei_pagecache_lock); 13331c6fdbd8SKent Overstreet return ret; 13341c6fdbd8SKent Overstreet } 13351c6fdbd8SKent Overstreet 13361c6fdbd8SKent Overstreet int bch2_write_end(struct file *file, struct address_space *mapping, 13371c6fdbd8SKent Overstreet loff_t pos, unsigned len, unsigned copied, 13381c6fdbd8SKent Overstreet struct page *page, void *fsdata) 13391c6fdbd8SKent Overstreet { 13401c6fdbd8SKent Overstreet struct bch_inode_info *inode = to_bch_ei(mapping->host); 13411c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 13421c6fdbd8SKent Overstreet 13431c6fdbd8SKent Overstreet lockdep_assert_held(&inode->v.i_rwsem); 13441c6fdbd8SKent Overstreet 13451c6fdbd8SKent Overstreet if (unlikely(copied < len && !PageUptodate(page))) { 13461c6fdbd8SKent Overstreet /* 13471c6fdbd8SKent Overstreet * The page needs to be read in, but that would destroy 13481c6fdbd8SKent Overstreet * our partial write - simplest thing is to just force 13491c6fdbd8SKent Overstreet * userspace to redo the write: 13501c6fdbd8SKent Overstreet */ 13511c6fdbd8SKent Overstreet zero_user(page, 0, PAGE_SIZE); 13521c6fdbd8SKent Overstreet flush_dcache_page(page); 13531c6fdbd8SKent Overstreet copied = 0; 13541c6fdbd8SKent Overstreet } 13551c6fdbd8SKent Overstreet 13561c6fdbd8SKent Overstreet spin_lock(&inode->v.i_lock); 13571c6fdbd8SKent Overstreet if (pos + copied > inode->v.i_size) 13581c6fdbd8SKent Overstreet i_size_write(&inode->v, pos + copied); 13591c6fdbd8SKent Overstreet spin_unlock(&inode->v.i_lock); 13601c6fdbd8SKent Overstreet 13611c6fdbd8SKent Overstreet if (copied) { 13621c6fdbd8SKent Overstreet if (!PageUptodate(page)) 13631c6fdbd8SKent Overstreet SetPageUptodate(page); 13641c6fdbd8SKent Overstreet if (!PageDirty(page)) 13651c6fdbd8SKent Overstreet set_page_dirty(page); 13661c6fdbd8SKent Overstreet 13671c6fdbd8SKent Overstreet inode->ei_last_dirtied = (unsigned long) current; 13681c6fdbd8SKent Overstreet } else { 13691c6fdbd8SKent Overstreet bch2_put_page_reservation(c, inode, page); 13701c6fdbd8SKent Overstreet } 13711c6fdbd8SKent Overstreet 13721c6fdbd8SKent Overstreet unlock_page(page); 13731c6fdbd8SKent Overstreet put_page(page); 13741c6fdbd8SKent Overstreet bch2_pagecache_add_put(&inode->ei_pagecache_lock); 13751c6fdbd8SKent Overstreet 13761c6fdbd8SKent Overstreet return copied; 13771c6fdbd8SKent Overstreet } 13781c6fdbd8SKent Overstreet 13791c6fdbd8SKent Overstreet #define WRITE_BATCH_PAGES 32 13801c6fdbd8SKent Overstreet 13811c6fdbd8SKent Overstreet static int __bch2_buffered_write(struct bch_inode_info *inode, 13821c6fdbd8SKent Overstreet struct address_space *mapping, 13831c6fdbd8SKent Overstreet struct iov_iter *iter, 13841c6fdbd8SKent Overstreet loff_t pos, unsigned len) 13851c6fdbd8SKent Overstreet { 13861c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 13871c6fdbd8SKent Overstreet struct page *pages[WRITE_BATCH_PAGES]; 13881c6fdbd8SKent Overstreet unsigned long index = pos >> PAGE_SHIFT; 13891c6fdbd8SKent Overstreet unsigned offset = pos & (PAGE_SIZE - 1); 13901c6fdbd8SKent Overstreet unsigned nr_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE); 13911c6fdbd8SKent Overstreet unsigned i, copied = 0, nr_pages_copied = 0; 13921c6fdbd8SKent Overstreet int ret = 0; 13931c6fdbd8SKent Overstreet 13941c6fdbd8SKent Overstreet BUG_ON(!len); 13951c6fdbd8SKent Overstreet BUG_ON(nr_pages > ARRAY_SIZE(pages)); 13961c6fdbd8SKent Overstreet 13971c6fdbd8SKent Overstreet for (i = 0; i < nr_pages; i++) { 13981c6fdbd8SKent Overstreet pages[i] = grab_cache_page_write_begin(mapping, index + i); 13991c6fdbd8SKent Overstreet if (!pages[i]) { 14001c6fdbd8SKent Overstreet nr_pages = i; 14011c6fdbd8SKent Overstreet ret = -ENOMEM; 14021c6fdbd8SKent Overstreet goto out; 14031c6fdbd8SKent Overstreet } 14041c6fdbd8SKent Overstreet } 14051c6fdbd8SKent Overstreet 14061c6fdbd8SKent Overstreet if (offset && !PageUptodate(pages[0])) { 14071c6fdbd8SKent Overstreet ret = bch2_read_single_page(pages[0], mapping); 14081c6fdbd8SKent Overstreet if (ret) 14091c6fdbd8SKent Overstreet goto out; 14101c6fdbd8SKent Overstreet } 14111c6fdbd8SKent Overstreet 14121c6fdbd8SKent Overstreet if ((pos + len) & (PAGE_SIZE - 1) && 14131c6fdbd8SKent Overstreet !PageUptodate(pages[nr_pages - 1])) { 14141c6fdbd8SKent Overstreet if ((index + nr_pages - 1) << PAGE_SHIFT >= inode->v.i_size) { 14151c6fdbd8SKent Overstreet zero_user(pages[nr_pages - 1], 0, PAGE_SIZE); 14161c6fdbd8SKent Overstreet } else { 14171c6fdbd8SKent Overstreet ret = bch2_read_single_page(pages[nr_pages - 1], mapping); 14181c6fdbd8SKent Overstreet if (ret) 14191c6fdbd8SKent Overstreet goto out; 14201c6fdbd8SKent Overstreet } 14211c6fdbd8SKent Overstreet } 14221c6fdbd8SKent Overstreet 14231c6fdbd8SKent Overstreet for (i = 0; i < nr_pages; i++) { 14241c6fdbd8SKent Overstreet ret = bch2_get_page_reservation(c, inode, pages[i], true); 14251c6fdbd8SKent Overstreet 14261c6fdbd8SKent Overstreet if (ret && !PageUptodate(pages[i])) { 14271c6fdbd8SKent Overstreet ret = bch2_read_single_page(pages[i], mapping); 14281c6fdbd8SKent Overstreet if (ret) 14291c6fdbd8SKent Overstreet goto out; 14301c6fdbd8SKent Overstreet 14311c6fdbd8SKent Overstreet ret = bch2_get_page_reservation(c, inode, pages[i], true); 14321c6fdbd8SKent Overstreet } 14331c6fdbd8SKent Overstreet 14341c6fdbd8SKent Overstreet if (ret) 14351c6fdbd8SKent Overstreet goto out; 14361c6fdbd8SKent Overstreet } 14371c6fdbd8SKent Overstreet 14381c6fdbd8SKent Overstreet if (mapping_writably_mapped(mapping)) 14391c6fdbd8SKent Overstreet for (i = 0; i < nr_pages; i++) 14401c6fdbd8SKent Overstreet flush_dcache_page(pages[i]); 14411c6fdbd8SKent Overstreet 14421c6fdbd8SKent Overstreet while (copied < len) { 14431c6fdbd8SKent Overstreet struct page *page = pages[(offset + copied) >> PAGE_SHIFT]; 14441c6fdbd8SKent Overstreet unsigned pg_offset = (offset + copied) & (PAGE_SIZE - 1); 14451c6fdbd8SKent Overstreet unsigned pg_bytes = min_t(unsigned, len - copied, 14461c6fdbd8SKent Overstreet PAGE_SIZE - pg_offset); 14471c6fdbd8SKent Overstreet unsigned pg_copied = copy_page_from_iter_atomic(page, 14481c6fdbd8SKent Overstreet pg_offset, pg_bytes, iter); 14491c6fdbd8SKent Overstreet 14501c6fdbd8SKent Overstreet flush_dcache_page(page); 14511c6fdbd8SKent Overstreet copied += pg_copied; 14521c6fdbd8SKent Overstreet 14531c6fdbd8SKent Overstreet if (pg_copied != pg_bytes) 14541c6fdbd8SKent Overstreet break; 14551c6fdbd8SKent Overstreet } 14561c6fdbd8SKent Overstreet 14571c6fdbd8SKent Overstreet if (!copied) 14581c6fdbd8SKent Overstreet goto out; 14591c6fdbd8SKent Overstreet 14601c6fdbd8SKent Overstreet nr_pages_copied = DIV_ROUND_UP(offset + copied, PAGE_SIZE); 14611c6fdbd8SKent Overstreet inode->ei_last_dirtied = (unsigned long) current; 14621c6fdbd8SKent Overstreet 14631c6fdbd8SKent Overstreet spin_lock(&inode->v.i_lock); 14641c6fdbd8SKent Overstreet if (pos + copied > inode->v.i_size) 14651c6fdbd8SKent Overstreet i_size_write(&inode->v, pos + copied); 14661c6fdbd8SKent Overstreet spin_unlock(&inode->v.i_lock); 14671c6fdbd8SKent Overstreet 14681c6fdbd8SKent Overstreet if (copied < len && 14691c6fdbd8SKent Overstreet ((offset + copied) & (PAGE_SIZE - 1))) { 14701c6fdbd8SKent Overstreet struct page *page = pages[(offset + copied) >> PAGE_SHIFT]; 14711c6fdbd8SKent Overstreet 14721c6fdbd8SKent Overstreet if (!PageUptodate(page)) { 14731c6fdbd8SKent Overstreet zero_user(page, 0, PAGE_SIZE); 14741c6fdbd8SKent Overstreet copied -= (offset + copied) & (PAGE_SIZE - 1); 14751c6fdbd8SKent Overstreet } 14761c6fdbd8SKent Overstreet } 14771c6fdbd8SKent Overstreet out: 14781c6fdbd8SKent Overstreet for (i = 0; i < nr_pages_copied; i++) { 14791c6fdbd8SKent Overstreet if (!PageUptodate(pages[i])) 14801c6fdbd8SKent Overstreet SetPageUptodate(pages[i]); 14811c6fdbd8SKent Overstreet if (!PageDirty(pages[i])) 14821c6fdbd8SKent Overstreet set_page_dirty(pages[i]); 14831c6fdbd8SKent Overstreet unlock_page(pages[i]); 14841c6fdbd8SKent Overstreet put_page(pages[i]); 14851c6fdbd8SKent Overstreet } 14861c6fdbd8SKent Overstreet 14871c6fdbd8SKent Overstreet for (i = nr_pages_copied; i < nr_pages; i++) { 14881c6fdbd8SKent Overstreet if (!PageDirty(pages[i])) 14891c6fdbd8SKent Overstreet bch2_put_page_reservation(c, inode, pages[i]); 14901c6fdbd8SKent Overstreet unlock_page(pages[i]); 14911c6fdbd8SKent Overstreet put_page(pages[i]); 14921c6fdbd8SKent Overstreet } 14931c6fdbd8SKent Overstreet 14941c6fdbd8SKent Overstreet return copied ?: ret; 14951c6fdbd8SKent Overstreet } 14961c6fdbd8SKent Overstreet 14971c6fdbd8SKent Overstreet static ssize_t bch2_buffered_write(struct kiocb *iocb, struct iov_iter *iter) 14981c6fdbd8SKent Overstreet { 14991c6fdbd8SKent Overstreet struct file *file = iocb->ki_filp; 15001c6fdbd8SKent Overstreet struct address_space *mapping = file->f_mapping; 15011c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 15021c6fdbd8SKent Overstreet loff_t pos = iocb->ki_pos; 15031c6fdbd8SKent Overstreet ssize_t written = 0; 15041c6fdbd8SKent Overstreet int ret = 0; 15051c6fdbd8SKent Overstreet 15061c6fdbd8SKent Overstreet bch2_pagecache_add_get(&inode->ei_pagecache_lock); 15071c6fdbd8SKent Overstreet 15081c6fdbd8SKent Overstreet do { 15091c6fdbd8SKent Overstreet unsigned offset = pos & (PAGE_SIZE - 1); 15101c6fdbd8SKent Overstreet unsigned bytes = min_t(unsigned long, iov_iter_count(iter), 15111c6fdbd8SKent Overstreet PAGE_SIZE * WRITE_BATCH_PAGES - offset); 15121c6fdbd8SKent Overstreet again: 15131c6fdbd8SKent Overstreet /* 15141c6fdbd8SKent Overstreet * Bring in the user page that we will copy from _first_. 15151c6fdbd8SKent Overstreet * Otherwise there's a nasty deadlock on copying from the 15161c6fdbd8SKent Overstreet * same page as we're writing to, without it being marked 15171c6fdbd8SKent Overstreet * up-to-date. 15181c6fdbd8SKent Overstreet * 15191c6fdbd8SKent Overstreet * Not only is this an optimisation, but it is also required 15201c6fdbd8SKent Overstreet * to check that the address is actually valid, when atomic 15211c6fdbd8SKent Overstreet * usercopies are used, below. 15221c6fdbd8SKent Overstreet */ 15231c6fdbd8SKent Overstreet if (unlikely(fault_in_iov_iter_readable(iter, bytes))) { 15241c6fdbd8SKent Overstreet bytes = min_t(unsigned long, iov_iter_count(iter), 15251c6fdbd8SKent Overstreet PAGE_SIZE - offset); 15261c6fdbd8SKent Overstreet 15271c6fdbd8SKent Overstreet if (unlikely(fault_in_iov_iter_readable(iter, bytes))) { 15281c6fdbd8SKent Overstreet ret = -EFAULT; 15291c6fdbd8SKent Overstreet break; 15301c6fdbd8SKent Overstreet } 15311c6fdbd8SKent Overstreet } 15321c6fdbd8SKent Overstreet 15331c6fdbd8SKent Overstreet if (unlikely(fatal_signal_pending(current))) { 15341c6fdbd8SKent Overstreet ret = -EINTR; 15351c6fdbd8SKent Overstreet break; 15361c6fdbd8SKent Overstreet } 15371c6fdbd8SKent Overstreet 15381c6fdbd8SKent Overstreet ret = __bch2_buffered_write(inode, mapping, iter, pos, bytes); 15391c6fdbd8SKent Overstreet if (unlikely(ret < 0)) 15401c6fdbd8SKent Overstreet break; 15411c6fdbd8SKent Overstreet 15421c6fdbd8SKent Overstreet cond_resched(); 15431c6fdbd8SKent Overstreet 15441c6fdbd8SKent Overstreet if (unlikely(ret == 0)) { 15451c6fdbd8SKent Overstreet /* 15461c6fdbd8SKent Overstreet * If we were unable to copy any data at all, we must 15471c6fdbd8SKent Overstreet * fall back to a single segment length write. 15481c6fdbd8SKent Overstreet * 15491c6fdbd8SKent Overstreet * If we didn't fallback here, we could livelock 15501c6fdbd8SKent Overstreet * because not all segments in the iov can be copied at 15511c6fdbd8SKent Overstreet * once without a pagefault. 15521c6fdbd8SKent Overstreet */ 15531c6fdbd8SKent Overstreet bytes = min_t(unsigned long, PAGE_SIZE - offset, 15541c6fdbd8SKent Overstreet iov_iter_single_seg_count(iter)); 15551c6fdbd8SKent Overstreet goto again; 15561c6fdbd8SKent Overstreet } 15571c6fdbd8SKent Overstreet pos += ret; 15581c6fdbd8SKent Overstreet written += ret; 15591c6fdbd8SKent Overstreet 15601c6fdbd8SKent Overstreet balance_dirty_pages_ratelimited(mapping); 15611c6fdbd8SKent Overstreet } while (iov_iter_count(iter)); 15621c6fdbd8SKent Overstreet 15631c6fdbd8SKent Overstreet bch2_pagecache_add_put(&inode->ei_pagecache_lock); 15641c6fdbd8SKent Overstreet 15651c6fdbd8SKent Overstreet return written ? written : ret; 15661c6fdbd8SKent Overstreet } 15671c6fdbd8SKent Overstreet 15681c6fdbd8SKent Overstreet /* O_DIRECT reads */ 15691c6fdbd8SKent Overstreet 15701c6fdbd8SKent Overstreet static void bch2_dio_read_complete(struct closure *cl) 15711c6fdbd8SKent Overstreet { 15721c6fdbd8SKent Overstreet struct dio_read *dio = container_of(cl, struct dio_read, cl); 15731c6fdbd8SKent Overstreet 15741c6fdbd8SKent Overstreet dio->req->ki_complete(dio->req, dio->ret); 15751c6fdbd8SKent Overstreet bio_check_pages_dirty(&dio->rbio.bio); /* transfers ownership */ 15761c6fdbd8SKent Overstreet } 15771c6fdbd8SKent Overstreet 15781c6fdbd8SKent Overstreet static void bch2_direct_IO_read_endio(struct bio *bio) 15791c6fdbd8SKent Overstreet { 15801c6fdbd8SKent Overstreet struct dio_read *dio = bio->bi_private; 15811c6fdbd8SKent Overstreet 15821c6fdbd8SKent Overstreet if (bio->bi_status) 15831c6fdbd8SKent Overstreet dio->ret = blk_status_to_errno(bio->bi_status); 15841c6fdbd8SKent Overstreet 15851c6fdbd8SKent Overstreet closure_put(&dio->cl); 15861c6fdbd8SKent Overstreet } 15871c6fdbd8SKent Overstreet 15881c6fdbd8SKent Overstreet static void bch2_direct_IO_read_split_endio(struct bio *bio) 15891c6fdbd8SKent Overstreet { 15901c6fdbd8SKent Overstreet bch2_direct_IO_read_endio(bio); 15911c6fdbd8SKent Overstreet bio_check_pages_dirty(bio); /* transfers ownership */ 15921c6fdbd8SKent Overstreet } 15931c6fdbd8SKent Overstreet 15941c6fdbd8SKent Overstreet static int bch2_direct_IO_read(struct kiocb *req, struct iov_iter *iter) 15951c6fdbd8SKent Overstreet { 15961c6fdbd8SKent Overstreet struct file *file = req->ki_filp; 15971c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 15981c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 15991c6fdbd8SKent Overstreet struct bch_io_opts opts = io_opts(c, inode); 16001c6fdbd8SKent Overstreet struct dio_read *dio; 16011c6fdbd8SKent Overstreet struct bio *bio; 16021c6fdbd8SKent Overstreet loff_t offset = req->ki_pos; 16031c6fdbd8SKent Overstreet bool sync = is_sync_kiocb(req); 16041c6fdbd8SKent Overstreet size_t shorten; 16051c6fdbd8SKent Overstreet ssize_t ret; 16061c6fdbd8SKent Overstreet 16071c6fdbd8SKent Overstreet if ((offset|iter->count) & (block_bytes(c) - 1)) 16081c6fdbd8SKent Overstreet return -EINVAL; 16091c6fdbd8SKent Overstreet 16101c6fdbd8SKent Overstreet ret = min_t(loff_t, iter->count, 16111c6fdbd8SKent Overstreet max_t(loff_t, 0, i_size_read(&inode->v) - offset)); 16121c6fdbd8SKent Overstreet 16131c6fdbd8SKent Overstreet if (!ret) 16141c6fdbd8SKent Overstreet return ret; 16151c6fdbd8SKent Overstreet 16161c6fdbd8SKent Overstreet shorten = iov_iter_count(iter) - round_up(ret, block_bytes(c)); 16171c6fdbd8SKent Overstreet iter->count -= shorten; 16181c6fdbd8SKent Overstreet 16191c6fdbd8SKent Overstreet bio = bio_alloc_bioset(NULL, 16201c6fdbd8SKent Overstreet iov_iter_npages(iter, BIO_MAX_VECS), 16211c6fdbd8SKent Overstreet REQ_OP_READ, 16221c6fdbd8SKent Overstreet GFP_KERNEL, 16231c6fdbd8SKent Overstreet &c->dio_read_bioset); 16241c6fdbd8SKent Overstreet 16251c6fdbd8SKent Overstreet bio->bi_end_io = bch2_direct_IO_read_endio; 16261c6fdbd8SKent Overstreet 16271c6fdbd8SKent Overstreet dio = container_of(bio, struct dio_read, rbio.bio); 16281c6fdbd8SKent Overstreet closure_init(&dio->cl, NULL); 16291c6fdbd8SKent Overstreet 16301c6fdbd8SKent Overstreet /* 16311c6fdbd8SKent Overstreet * this is a _really_ horrible hack just to avoid an atomic sub at the 16321c6fdbd8SKent Overstreet * end: 16331c6fdbd8SKent Overstreet */ 16341c6fdbd8SKent Overstreet if (!sync) { 16351c6fdbd8SKent Overstreet set_closure_fn(&dio->cl, bch2_dio_read_complete, NULL); 16361c6fdbd8SKent Overstreet atomic_set(&dio->cl.remaining, 16371c6fdbd8SKent Overstreet CLOSURE_REMAINING_INITIALIZER - 16381c6fdbd8SKent Overstreet CLOSURE_RUNNING + 16391c6fdbd8SKent Overstreet CLOSURE_DESTRUCTOR); 16401c6fdbd8SKent Overstreet } else { 16411c6fdbd8SKent Overstreet atomic_set(&dio->cl.remaining, 16421c6fdbd8SKent Overstreet CLOSURE_REMAINING_INITIALIZER + 1); 16431c6fdbd8SKent Overstreet } 16441c6fdbd8SKent Overstreet 16451c6fdbd8SKent Overstreet dio->req = req; 16461c6fdbd8SKent Overstreet dio->ret = ret; 16471c6fdbd8SKent Overstreet 16481c6fdbd8SKent Overstreet goto start; 16491c6fdbd8SKent Overstreet while (iter->count) { 16501c6fdbd8SKent Overstreet bio = bio_alloc_bioset(NULL, 16511c6fdbd8SKent Overstreet iov_iter_npages(iter, BIO_MAX_VECS), 16521c6fdbd8SKent Overstreet REQ_OP_READ, 16531c6fdbd8SKent Overstreet GFP_KERNEL, 16541c6fdbd8SKent Overstreet &c->bio_read); 16551c6fdbd8SKent Overstreet bio->bi_end_io = bch2_direct_IO_read_split_endio; 16561c6fdbd8SKent Overstreet start: 16571c6fdbd8SKent Overstreet bio->bi_opf = REQ_OP_READ|REQ_SYNC; 16581c6fdbd8SKent Overstreet bio->bi_iter.bi_sector = offset >> 9; 16591c6fdbd8SKent Overstreet bio->bi_private = dio; 16601c6fdbd8SKent Overstreet 16611c6fdbd8SKent Overstreet ret = bio_iov_iter_get_pages(bio, iter); 16621c6fdbd8SKent Overstreet if (ret < 0) { 16631c6fdbd8SKent Overstreet /* XXX: fault inject this path */ 16641c6fdbd8SKent Overstreet bio->bi_status = BLK_STS_RESOURCE; 16651c6fdbd8SKent Overstreet bio_endio(bio); 16661c6fdbd8SKent Overstreet break; 16671c6fdbd8SKent Overstreet } 16681c6fdbd8SKent Overstreet 16691c6fdbd8SKent Overstreet offset += bio->bi_iter.bi_size; 16701c6fdbd8SKent Overstreet bio_set_pages_dirty(bio); 16711c6fdbd8SKent Overstreet 16721c6fdbd8SKent Overstreet if (iter->count) 16731c6fdbd8SKent Overstreet closure_get(&dio->cl); 16741c6fdbd8SKent Overstreet 16751c6fdbd8SKent Overstreet bch2_read(c, rbio_init(bio, opts), inode->v.i_ino); 16761c6fdbd8SKent Overstreet } 16771c6fdbd8SKent Overstreet 16781c6fdbd8SKent Overstreet iter->count += shorten; 16791c6fdbd8SKent Overstreet 16801c6fdbd8SKent Overstreet if (sync) { 16811c6fdbd8SKent Overstreet closure_sync(&dio->cl); 16821c6fdbd8SKent Overstreet closure_debug_destroy(&dio->cl); 16831c6fdbd8SKent Overstreet ret = dio->ret; 16841c6fdbd8SKent Overstreet bio_check_pages_dirty(&dio->rbio.bio); /* transfers ownership */ 16851c6fdbd8SKent Overstreet return ret; 16861c6fdbd8SKent Overstreet } else { 16871c6fdbd8SKent Overstreet return -EIOCBQUEUED; 16881c6fdbd8SKent Overstreet } 16891c6fdbd8SKent Overstreet } 16901c6fdbd8SKent Overstreet 16911c6fdbd8SKent Overstreet ssize_t bch2_read_iter(struct kiocb *iocb, struct iov_iter *iter) 16921c6fdbd8SKent Overstreet { 16931c6fdbd8SKent Overstreet struct file *file = iocb->ki_filp; 16941c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 16951c6fdbd8SKent Overstreet struct address_space *mapping = file->f_mapping; 16961c6fdbd8SKent Overstreet size_t count = iov_iter_count(iter); 16971c6fdbd8SKent Overstreet ssize_t ret; 16981c6fdbd8SKent Overstreet 16991c6fdbd8SKent Overstreet if (!count) 17001c6fdbd8SKent Overstreet return 0; /* skip atime */ 17011c6fdbd8SKent Overstreet 17021c6fdbd8SKent Overstreet if (iocb->ki_flags & IOCB_DIRECT) { 17031c6fdbd8SKent Overstreet struct blk_plug plug; 17041c6fdbd8SKent Overstreet 17051c6fdbd8SKent Overstreet ret = filemap_write_and_wait_range(mapping, 17061c6fdbd8SKent Overstreet iocb->ki_pos, 17071c6fdbd8SKent Overstreet iocb->ki_pos + count - 1); 17081c6fdbd8SKent Overstreet if (ret < 0) 17091c6fdbd8SKent Overstreet return ret; 17101c6fdbd8SKent Overstreet 17111c6fdbd8SKent Overstreet file_accessed(file); 17121c6fdbd8SKent Overstreet 17131c6fdbd8SKent Overstreet blk_start_plug(&plug); 17141c6fdbd8SKent Overstreet ret = bch2_direct_IO_read(iocb, iter); 17151c6fdbd8SKent Overstreet blk_finish_plug(&plug); 17161c6fdbd8SKent Overstreet 17171c6fdbd8SKent Overstreet if (ret >= 0) 17181c6fdbd8SKent Overstreet iocb->ki_pos += ret; 17191c6fdbd8SKent Overstreet } else { 17201c6fdbd8SKent Overstreet bch2_pagecache_add_get(&inode->ei_pagecache_lock); 17211c6fdbd8SKent Overstreet ret = generic_file_read_iter(iocb, iter); 17221c6fdbd8SKent Overstreet bch2_pagecache_add_put(&inode->ei_pagecache_lock); 17231c6fdbd8SKent Overstreet } 17241c6fdbd8SKent Overstreet 17251c6fdbd8SKent Overstreet return ret; 17261c6fdbd8SKent Overstreet } 17271c6fdbd8SKent Overstreet 17281c6fdbd8SKent Overstreet /* O_DIRECT writes */ 17291c6fdbd8SKent Overstreet 17301c6fdbd8SKent Overstreet /* 17311c6fdbd8SKent Overstreet * We're going to return -EIOCBQUEUED, but we haven't finished consuming the 17321c6fdbd8SKent Overstreet * iov_iter yet, so we need to stash a copy of the iovec: it might be on the 17331c6fdbd8SKent Overstreet * caller's stack, we're not guaranteed that it will live for the duration of 17341c6fdbd8SKent Overstreet * the IO: 17351c6fdbd8SKent Overstreet */ 17361c6fdbd8SKent Overstreet static noinline int bch2_dio_write_copy_iov(struct dio_write *dio) 17371c6fdbd8SKent Overstreet { 17381c6fdbd8SKent Overstreet struct iovec *iov = dio->inline_vecs; 17391c6fdbd8SKent Overstreet 17401c6fdbd8SKent Overstreet /* 17411c6fdbd8SKent Overstreet * iov_iter has a single embedded iovec - nothing to do: 17421c6fdbd8SKent Overstreet */ 17431c6fdbd8SKent Overstreet if (iter_is_ubuf(&dio->iter)) 17441c6fdbd8SKent Overstreet return 0; 17451c6fdbd8SKent Overstreet 17461c6fdbd8SKent Overstreet /* 17471c6fdbd8SKent Overstreet * We don't currently handle non-iovec iov_iters here - return an error, 17481c6fdbd8SKent Overstreet * and we'll fall back to doing the IO synchronously: 17491c6fdbd8SKent Overstreet */ 17501c6fdbd8SKent Overstreet if (!iter_is_iovec(&dio->iter)) 17511c6fdbd8SKent Overstreet return -1; 17521c6fdbd8SKent Overstreet 17531c6fdbd8SKent Overstreet if (dio->iter.nr_segs > ARRAY_SIZE(dio->inline_vecs)) { 17541c6fdbd8SKent Overstreet iov = kmalloc_array(dio->iter.nr_segs, sizeof(*iov), 17551c6fdbd8SKent Overstreet GFP_KERNEL); 17561c6fdbd8SKent Overstreet if (unlikely(!iov)) 17571c6fdbd8SKent Overstreet return -ENOMEM; 17581c6fdbd8SKent Overstreet 17591c6fdbd8SKent Overstreet dio->free_iov = true; 17601c6fdbd8SKent Overstreet } 17611c6fdbd8SKent Overstreet 17621c6fdbd8SKent Overstreet memcpy(iov, dio->iter.__iov, dio->iter.nr_segs * sizeof(*iov)); 17631c6fdbd8SKent Overstreet dio->iter.__iov = iov; 17641c6fdbd8SKent Overstreet return 0; 17651c6fdbd8SKent Overstreet } 17661c6fdbd8SKent Overstreet 17671c6fdbd8SKent Overstreet static void bch2_dio_write_loop_async(struct closure *); 17681c6fdbd8SKent Overstreet 17691c6fdbd8SKent Overstreet static long bch2_dio_write_loop(struct dio_write *dio) 17701c6fdbd8SKent Overstreet { 17711c6fdbd8SKent Overstreet struct kiocb *req = dio->req; 17721c6fdbd8SKent Overstreet struct address_space *mapping = req->ki_filp->f_mapping; 17731c6fdbd8SKent Overstreet struct bch_inode_info *inode = dio->iop.inode; 17741c6fdbd8SKent Overstreet struct bio *bio = &dio->iop.op.wbio.bio; 17751c6fdbd8SKent Overstreet struct bvec_iter_all iter; 17761c6fdbd8SKent Overstreet struct bio_vec *bv; 17771c6fdbd8SKent Overstreet bool sync; 17781c6fdbd8SKent Overstreet long ret; 17791c6fdbd8SKent Overstreet 17801c6fdbd8SKent Overstreet if (dio->loop) 17811c6fdbd8SKent Overstreet goto loop; 17821c6fdbd8SKent Overstreet 17831c6fdbd8SKent Overstreet inode_dio_begin(&inode->v); 17841c6fdbd8SKent Overstreet bch2_pagecache_block_get(&inode->ei_pagecache_lock); 17851c6fdbd8SKent Overstreet 17861c6fdbd8SKent Overstreet /* Write and invalidate pagecache range that we're writing to: */ 17871c6fdbd8SKent Overstreet ret = write_invalidate_inode_pages_range(mapping, req->ki_pos, 17881c6fdbd8SKent Overstreet req->ki_pos + iov_iter_count(&dio->iter) - 1); 17891c6fdbd8SKent Overstreet if (unlikely(ret)) 17901c6fdbd8SKent Overstreet goto err; 17911c6fdbd8SKent Overstreet 17921c6fdbd8SKent Overstreet while (1) { 17931c6fdbd8SKent Overstreet if (current != dio->task) 17941c6fdbd8SKent Overstreet kthread_use_mm(dio->task->mm); 17951c6fdbd8SKent Overstreet BUG_ON(current->faults_disabled_mapping); 17961c6fdbd8SKent Overstreet current->faults_disabled_mapping = mapping; 17971c6fdbd8SKent Overstreet 17981c6fdbd8SKent Overstreet ret = bio_iov_iter_get_pages(bio, &dio->iter); 17991c6fdbd8SKent Overstreet 18001c6fdbd8SKent Overstreet current->faults_disabled_mapping = NULL; 18011c6fdbd8SKent Overstreet if (current != dio->task) 18021c6fdbd8SKent Overstreet kthread_unuse_mm(dio->task->mm); 18031c6fdbd8SKent Overstreet 18041c6fdbd8SKent Overstreet if (unlikely(ret < 0)) 18051c6fdbd8SKent Overstreet goto err; 18061c6fdbd8SKent Overstreet 18071c6fdbd8SKent Overstreet /* gup might have faulted pages back in: */ 18081c6fdbd8SKent Overstreet ret = write_invalidate_inode_pages_range(mapping, 18091c6fdbd8SKent Overstreet req->ki_pos + (dio->iop.op.written << 9), 18101c6fdbd8SKent Overstreet req->ki_pos + iov_iter_count(&dio->iter) - 1); 18111c6fdbd8SKent Overstreet if (unlikely(ret)) 18121c6fdbd8SKent Overstreet goto err; 18131c6fdbd8SKent Overstreet 18141c6fdbd8SKent Overstreet dio->iop.op.pos = POS(inode->v.i_ino, 18151c6fdbd8SKent Overstreet (req->ki_pos >> 9) + dio->iop.op.written); 18161c6fdbd8SKent Overstreet 18171c6fdbd8SKent Overstreet task_io_account_write(bio->bi_iter.bi_size); 18181c6fdbd8SKent Overstreet 18191c6fdbd8SKent Overstreet closure_call(&dio->iop.op.cl, bch2_write, NULL, &dio->cl); 18201c6fdbd8SKent Overstreet 18211c6fdbd8SKent Overstreet if (!dio->sync && !dio->loop && dio->iter.count) { 18221c6fdbd8SKent Overstreet if (bch2_dio_write_copy_iov(dio)) { 18231c6fdbd8SKent Overstreet dio->iop.op.error = -ENOMEM; 18241c6fdbd8SKent Overstreet goto err_wait_io; 18251c6fdbd8SKent Overstreet } 18261c6fdbd8SKent Overstreet } 18271c6fdbd8SKent Overstreet err_wait_io: 18281c6fdbd8SKent Overstreet dio->loop = true; 18291c6fdbd8SKent Overstreet 18301c6fdbd8SKent Overstreet if (!dio->sync) { 18311c6fdbd8SKent Overstreet continue_at(&dio->cl, bch2_dio_write_loop_async, NULL); 18321c6fdbd8SKent Overstreet return -EIOCBQUEUED; 18331c6fdbd8SKent Overstreet } 18341c6fdbd8SKent Overstreet 18351c6fdbd8SKent Overstreet closure_sync(&dio->cl); 18361c6fdbd8SKent Overstreet loop: 18371c6fdbd8SKent Overstreet bio_for_each_segment_all(bv, bio, iter) 18381c6fdbd8SKent Overstreet put_page(bv->bv_page); 18391c6fdbd8SKent Overstreet if (!dio->iter.count || dio->iop.op.error) 18401c6fdbd8SKent Overstreet break; 18411c6fdbd8SKent Overstreet bio_reset(bio, NULL, REQ_OP_WRITE); 18421c6fdbd8SKent Overstreet } 18431c6fdbd8SKent Overstreet 18441c6fdbd8SKent Overstreet ret = dio->iop.op.error ?: ((long) dio->iop.op.written << 9); 18451c6fdbd8SKent Overstreet err: 18461c6fdbd8SKent Overstreet bch2_pagecache_block_put(&inode->ei_pagecache_lock); 18471c6fdbd8SKent Overstreet bch2_disk_reservation_put(dio->iop.op.c, &dio->iop.op.res); 18481c6fdbd8SKent Overstreet bch2_quota_reservation_put(dio->iop.op.c, inode, &dio->quota_res); 18491c6fdbd8SKent Overstreet 18501c6fdbd8SKent Overstreet if (dio->free_iov) 18511c6fdbd8SKent Overstreet kfree(dio->iter.__iov); 18521c6fdbd8SKent Overstreet 18531c6fdbd8SKent Overstreet closure_debug_destroy(&dio->cl); 18541c6fdbd8SKent Overstreet 18551c6fdbd8SKent Overstreet sync = dio->sync; 18561c6fdbd8SKent Overstreet bio_put(bio); 18571c6fdbd8SKent Overstreet 18581c6fdbd8SKent Overstreet /* inode->i_dio_count is our ref on inode and thus bch_fs */ 18591c6fdbd8SKent Overstreet inode_dio_end(&inode->v); 18601c6fdbd8SKent Overstreet 18611c6fdbd8SKent Overstreet if (!sync) { 18621c6fdbd8SKent Overstreet req->ki_complete(req, ret); 18631c6fdbd8SKent Overstreet ret = -EIOCBQUEUED; 18641c6fdbd8SKent Overstreet } 18651c6fdbd8SKent Overstreet return ret; 18661c6fdbd8SKent Overstreet } 18671c6fdbd8SKent Overstreet 18681c6fdbd8SKent Overstreet static void bch2_dio_write_loop_async(struct closure *cl) 18691c6fdbd8SKent Overstreet { 18701c6fdbd8SKent Overstreet struct dio_write *dio = container_of(cl, struct dio_write, cl); 18711c6fdbd8SKent Overstreet 18721c6fdbd8SKent Overstreet bch2_dio_write_loop(dio); 18731c6fdbd8SKent Overstreet } 18741c6fdbd8SKent Overstreet 18751c6fdbd8SKent Overstreet static noinline 18761c6fdbd8SKent Overstreet ssize_t bch2_direct_write(struct kiocb *req, struct iov_iter *iter) 18771c6fdbd8SKent Overstreet { 18781c6fdbd8SKent Overstreet struct file *file = req->ki_filp; 18791c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 18801c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 18811c6fdbd8SKent Overstreet struct dio_write *dio; 18821c6fdbd8SKent Overstreet struct bio *bio; 18831c6fdbd8SKent Overstreet loff_t offset = req->ki_pos; 18841c6fdbd8SKent Overstreet ssize_t ret; 18851c6fdbd8SKent Overstreet 18861c6fdbd8SKent Overstreet lockdep_assert_held(&inode->v.i_rwsem); 18871c6fdbd8SKent Overstreet 18881c6fdbd8SKent Overstreet if (unlikely(!iter->count)) 18891c6fdbd8SKent Overstreet return 0; 18901c6fdbd8SKent Overstreet 18911c6fdbd8SKent Overstreet if (unlikely((offset|iter->count) & (block_bytes(c) - 1))) 18921c6fdbd8SKent Overstreet return -EINVAL; 18931c6fdbd8SKent Overstreet 18941c6fdbd8SKent Overstreet bio = bio_alloc_bioset(NULL, 18951c6fdbd8SKent Overstreet iov_iter_npages(iter, BIO_MAX_VECS), 18961c6fdbd8SKent Overstreet REQ_OP_WRITE, 18971c6fdbd8SKent Overstreet GFP_KERNEL, 18981c6fdbd8SKent Overstreet &c->dio_write_bioset); 18991c6fdbd8SKent Overstreet dio = container_of(bio, struct dio_write, iop.op.wbio.bio); 19001c6fdbd8SKent Overstreet closure_init(&dio->cl, NULL); 19011c6fdbd8SKent Overstreet dio->req = req; 19021c6fdbd8SKent Overstreet dio->task = current; 19031c6fdbd8SKent Overstreet dio->loop = false; 19041c6fdbd8SKent Overstreet dio->sync = is_sync_kiocb(req) || 19051c6fdbd8SKent Overstreet offset + iter->count > inode->v.i_size; 19061c6fdbd8SKent Overstreet dio->free_iov = false; 19071c6fdbd8SKent Overstreet dio->quota_res.sectors = 0; 19081c6fdbd8SKent Overstreet dio->iter = *iter; 19091c6fdbd8SKent Overstreet bch2_fswrite_op_init(&dio->iop, c, inode, io_opts(c, inode), true); 19101c6fdbd8SKent Overstreet dio->iop.op.write_point = writepoint_hashed((unsigned long) dio->task); 19111c6fdbd8SKent Overstreet dio->iop.op.flags |= BCH_WRITE_NOPUT_RESERVATION; 19121c6fdbd8SKent Overstreet 19131c6fdbd8SKent Overstreet if ((req->ki_flags & IOCB_DSYNC) && 19141c6fdbd8SKent Overstreet !c->opts.journal_flush_disabled) 19151c6fdbd8SKent Overstreet dio->iop.op.flags |= BCH_WRITE_FLUSH; 19161c6fdbd8SKent Overstreet 19171c6fdbd8SKent Overstreet ret = bch2_quota_reservation_add(c, inode, &dio->quota_res, 19181c6fdbd8SKent Overstreet iter->count >> 9, true); 19191c6fdbd8SKent Overstreet if (unlikely(ret)) 19201c6fdbd8SKent Overstreet goto err; 19211c6fdbd8SKent Overstreet 19221c6fdbd8SKent Overstreet ret = bch2_disk_reservation_get(c, &dio->iop.op.res, iter->count >> 9, 19231c6fdbd8SKent Overstreet dio->iop.op.opts.data_replicas, 0); 19241c6fdbd8SKent Overstreet if (unlikely(ret)) { 19251c6fdbd8SKent Overstreet if (bch2_check_range_allocated(c, POS(inode->v.i_ino, 19261c6fdbd8SKent Overstreet offset >> 9), 19271c6fdbd8SKent Overstreet iter->count >> 9)) 19281c6fdbd8SKent Overstreet goto err; 19291c6fdbd8SKent Overstreet 19301c6fdbd8SKent Overstreet dio->iop.unalloc = true; 19311c6fdbd8SKent Overstreet } 19321c6fdbd8SKent Overstreet 19331c6fdbd8SKent Overstreet dio->iop.op.nr_replicas = dio->iop.op.res.nr_replicas; 19341c6fdbd8SKent Overstreet 19351c6fdbd8SKent Overstreet return bch2_dio_write_loop(dio); 19361c6fdbd8SKent Overstreet err: 19371c6fdbd8SKent Overstreet bch2_disk_reservation_put(c, &dio->iop.op.res); 19381c6fdbd8SKent Overstreet bch2_quota_reservation_put(c, inode, &dio->quota_res); 19391c6fdbd8SKent Overstreet closure_debug_destroy(&dio->cl); 19401c6fdbd8SKent Overstreet bio_put(bio); 19411c6fdbd8SKent Overstreet return ret; 19421c6fdbd8SKent Overstreet } 19431c6fdbd8SKent Overstreet 19441c6fdbd8SKent Overstreet static ssize_t __bch2_write_iter(struct kiocb *iocb, struct iov_iter *from) 19451c6fdbd8SKent Overstreet { 19461c6fdbd8SKent Overstreet struct file *file = iocb->ki_filp; 19471c6fdbd8SKent Overstreet ssize_t ret; 19481c6fdbd8SKent Overstreet 19491c6fdbd8SKent Overstreet if (iocb->ki_flags & IOCB_DIRECT) 19501c6fdbd8SKent Overstreet return bch2_direct_write(iocb, from); 19511c6fdbd8SKent Overstreet 19521c6fdbd8SKent Overstreet ret = file_remove_privs(file); 19531c6fdbd8SKent Overstreet if (ret) 19541c6fdbd8SKent Overstreet return ret; 19551c6fdbd8SKent Overstreet 19561c6fdbd8SKent Overstreet ret = file_update_time(file); 19571c6fdbd8SKent Overstreet if (ret) 19581c6fdbd8SKent Overstreet return ret; 19591c6fdbd8SKent Overstreet 19601c6fdbd8SKent Overstreet ret = iocb->ki_flags & IOCB_DIRECT 19611c6fdbd8SKent Overstreet ? bch2_direct_write(iocb, from) 19621c6fdbd8SKent Overstreet : bch2_buffered_write(iocb, from); 19631c6fdbd8SKent Overstreet 19641c6fdbd8SKent Overstreet if (likely(ret > 0)) 19651c6fdbd8SKent Overstreet iocb->ki_pos += ret; 19661c6fdbd8SKent Overstreet 19671c6fdbd8SKent Overstreet return ret; 19681c6fdbd8SKent Overstreet } 19691c6fdbd8SKent Overstreet 19701c6fdbd8SKent Overstreet ssize_t bch2_write_iter(struct kiocb *iocb, struct iov_iter *from) 19711c6fdbd8SKent Overstreet { 19721c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(iocb->ki_filp); 19731c6fdbd8SKent Overstreet bool direct = iocb->ki_flags & IOCB_DIRECT; 19741c6fdbd8SKent Overstreet ssize_t ret; 19751c6fdbd8SKent Overstreet 19761c6fdbd8SKent Overstreet inode_lock(&inode->v); 19771c6fdbd8SKent Overstreet ret = generic_write_checks(iocb, from); 19781c6fdbd8SKent Overstreet if (ret > 0) 19791c6fdbd8SKent Overstreet ret = __bch2_write_iter(iocb, from); 19801c6fdbd8SKent Overstreet inode_unlock(&inode->v); 19811c6fdbd8SKent Overstreet 19821c6fdbd8SKent Overstreet if (ret > 0 && !direct) 19831c6fdbd8SKent Overstreet ret = generic_write_sync(iocb, ret); 19841c6fdbd8SKent Overstreet 19851c6fdbd8SKent Overstreet return ret; 19861c6fdbd8SKent Overstreet } 19871c6fdbd8SKent Overstreet 19881c6fdbd8SKent Overstreet /* fsync: */ 19891c6fdbd8SKent Overstreet 19901c6fdbd8SKent Overstreet int bch2_fsync(struct file *file, loff_t start, loff_t end, int datasync) 19911c6fdbd8SKent Overstreet { 19921c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 19931c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 1994bb1b3658SKent Overstreet int ret, ret2; 19951c6fdbd8SKent Overstreet 19961c6fdbd8SKent Overstreet ret = file_write_and_wait_range(file, start, end); 19971c6fdbd8SKent Overstreet if (ret) 19981c6fdbd8SKent Overstreet return ret; 19991c6fdbd8SKent Overstreet 20001c6fdbd8SKent Overstreet if (datasync && !(inode->v.i_state & I_DIRTY_DATASYNC)) 20011c6fdbd8SKent Overstreet goto out; 20021c6fdbd8SKent Overstreet 20031c6fdbd8SKent Overstreet ret = sync_inode_metadata(&inode->v, 1); 20041c6fdbd8SKent Overstreet if (ret) 20051c6fdbd8SKent Overstreet return ret; 20061c6fdbd8SKent Overstreet out: 20071c6fdbd8SKent Overstreet if (c->opts.journal_flush_disabled) 20081c6fdbd8SKent Overstreet return 0; 20091c6fdbd8SKent Overstreet 2010bb1b3658SKent Overstreet ret = bch2_journal_flush_seq(&c->journal, inode->ei_journal_seq); 2011bb1b3658SKent Overstreet ret2 = file_check_and_advance_wb_err(file); 2012bb1b3658SKent Overstreet 2013bb1b3658SKent Overstreet return ret ?: ret2; 20141c6fdbd8SKent Overstreet } 20151c6fdbd8SKent Overstreet 20161c6fdbd8SKent Overstreet /* truncate: */ 20171c6fdbd8SKent Overstreet 201854e2264eSKent Overstreet static int __bch2_fpunch(struct bch_fs *c, struct bch_inode_info *inode, 201954e2264eSKent Overstreet u64 start_offset, u64 end_offset, u64 *journal_seq) 202054e2264eSKent Overstreet { 202154e2264eSKent Overstreet struct bpos start = POS(inode->v.i_ino, start_offset); 202254e2264eSKent Overstreet struct bpos end = POS(inode->v.i_ino, end_offset); 202354e2264eSKent Overstreet unsigned max_sectors = KEY_SIZE_MAX & (~0 << c->block_bits); 202454e2264eSKent Overstreet struct btree_trans trans; 202554e2264eSKent Overstreet struct btree_iter *iter; 202654e2264eSKent Overstreet struct bkey_s_c k; 202754e2264eSKent Overstreet int ret = 0; 202854e2264eSKent Overstreet 202954e2264eSKent Overstreet bch2_trans_init(&trans, c); 203054e2264eSKent Overstreet bch2_trans_preload_iters(&trans); 203154e2264eSKent Overstreet 203254e2264eSKent Overstreet iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS, start, 203354e2264eSKent Overstreet BTREE_ITER_INTENT); 203454e2264eSKent Overstreet 203554e2264eSKent Overstreet while ((k = bch2_btree_iter_peek(iter)).k && 203654e2264eSKent Overstreet !(ret = btree_iter_err(k)) && 203754e2264eSKent Overstreet bkey_cmp(iter->pos, end) < 0) { 203854e2264eSKent Overstreet struct disk_reservation disk_res = 203954e2264eSKent Overstreet bch2_disk_reservation_init(c, 0); 204054e2264eSKent Overstreet struct bkey_i delete; 204154e2264eSKent Overstreet 204254e2264eSKent Overstreet bkey_init(&delete.k); 204354e2264eSKent Overstreet delete.k.p = iter->pos; 204454e2264eSKent Overstreet 204554e2264eSKent Overstreet /* create the biggest key we can */ 204654e2264eSKent Overstreet bch2_key_resize(&delete.k, max_sectors); 204754e2264eSKent Overstreet bch2_cut_back(end, &delete.k); 204854e2264eSKent Overstreet 204954e2264eSKent Overstreet ret = bch2_extent_update(&trans, inode, 205054e2264eSKent Overstreet &disk_res, NULL, iter, &delete, 205154e2264eSKent Overstreet 0, true, true, NULL); 205254e2264eSKent Overstreet bch2_disk_reservation_put(c, &disk_res); 205354e2264eSKent Overstreet 205454e2264eSKent Overstreet if (ret == -EINTR) 205554e2264eSKent Overstreet ret = 0; 205654e2264eSKent Overstreet if (ret) 205754e2264eSKent Overstreet break; 205854e2264eSKent Overstreet 205954e2264eSKent Overstreet bch2_btree_iter_cond_resched(iter); 206054e2264eSKent Overstreet } 206154e2264eSKent Overstreet 206254e2264eSKent Overstreet bch2_trans_exit(&trans); 206354e2264eSKent Overstreet 206454e2264eSKent Overstreet return ret; 206554e2264eSKent Overstreet } 206654e2264eSKent Overstreet 20671c6fdbd8SKent Overstreet static inline int range_has_data(struct bch_fs *c, 20681c6fdbd8SKent Overstreet struct bpos start, 20691c6fdbd8SKent Overstreet struct bpos end) 20701c6fdbd8SKent Overstreet { 20711c6fdbd8SKent Overstreet 20721c6fdbd8SKent Overstreet struct btree_iter iter; 20731c6fdbd8SKent Overstreet struct bkey_s_c k; 20741c6fdbd8SKent Overstreet int ret = 0; 20751c6fdbd8SKent Overstreet 20761c6fdbd8SKent Overstreet for_each_btree_key(&iter, c, BTREE_ID_EXTENTS, 20771c6fdbd8SKent Overstreet start, 0, k) { 20781c6fdbd8SKent Overstreet if (bkey_cmp(bkey_start_pos(k.k), end) >= 0) 20791c6fdbd8SKent Overstreet break; 20801c6fdbd8SKent Overstreet 20811c6fdbd8SKent Overstreet if (bkey_extent_is_data(k.k)) { 20821c6fdbd8SKent Overstreet ret = 1; 20831c6fdbd8SKent Overstreet break; 20841c6fdbd8SKent Overstreet } 20851c6fdbd8SKent Overstreet } 20861c6fdbd8SKent Overstreet 20871c6fdbd8SKent Overstreet return bch2_btree_iter_unlock(&iter) ?: ret; 20881c6fdbd8SKent Overstreet } 20891c6fdbd8SKent Overstreet 20901c6fdbd8SKent Overstreet static int __bch2_truncate_page(struct bch_inode_info *inode, 20911c6fdbd8SKent Overstreet pgoff_t index, loff_t start, loff_t end) 20921c6fdbd8SKent Overstreet { 20931c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 20941c6fdbd8SKent Overstreet struct address_space *mapping = inode->v.i_mapping; 20951c6fdbd8SKent Overstreet unsigned start_offset = start & (PAGE_SIZE - 1); 20961c6fdbd8SKent Overstreet unsigned end_offset = ((end - 1) & (PAGE_SIZE - 1)) + 1; 20971c6fdbd8SKent Overstreet struct page *page; 20981c6fdbd8SKent Overstreet int ret = 0; 20991c6fdbd8SKent Overstreet 21001c6fdbd8SKent Overstreet /* Page boundary? Nothing to do */ 21011c6fdbd8SKent Overstreet if (!((index == start >> PAGE_SHIFT && start_offset) || 21021c6fdbd8SKent Overstreet (index == end >> PAGE_SHIFT && end_offset != PAGE_SIZE))) 21031c6fdbd8SKent Overstreet return 0; 21041c6fdbd8SKent Overstreet 21051c6fdbd8SKent Overstreet /* Above i_size? */ 21061c6fdbd8SKent Overstreet if (index << PAGE_SHIFT >= inode->v.i_size) 21071c6fdbd8SKent Overstreet return 0; 21081c6fdbd8SKent Overstreet 21091c6fdbd8SKent Overstreet page = find_lock_page(mapping, index); 21101c6fdbd8SKent Overstreet if (!page) { 21111c6fdbd8SKent Overstreet /* 21121c6fdbd8SKent Overstreet * XXX: we're doing two index lookups when we end up reading the 21131c6fdbd8SKent Overstreet * page 21141c6fdbd8SKent Overstreet */ 21151c6fdbd8SKent Overstreet ret = range_has_data(c, 21161c6fdbd8SKent Overstreet POS(inode->v.i_ino, index << PAGE_SECTOR_SHIFT), 21171c6fdbd8SKent Overstreet POS(inode->v.i_ino, (index + 1) << PAGE_SECTOR_SHIFT)); 21181c6fdbd8SKent Overstreet if (ret <= 0) 21191c6fdbd8SKent Overstreet return ret; 21201c6fdbd8SKent Overstreet 21211c6fdbd8SKent Overstreet page = find_or_create_page(mapping, index, GFP_KERNEL); 21221c6fdbd8SKent Overstreet if (unlikely(!page)) { 21231c6fdbd8SKent Overstreet ret = -ENOMEM; 21241c6fdbd8SKent Overstreet goto out; 21251c6fdbd8SKent Overstreet } 21261c6fdbd8SKent Overstreet } 21271c6fdbd8SKent Overstreet 21281c6fdbd8SKent Overstreet if (!PageUptodate(page)) { 21291c6fdbd8SKent Overstreet ret = bch2_read_single_page(page, mapping); 21301c6fdbd8SKent Overstreet if (ret) 21311c6fdbd8SKent Overstreet goto unlock; 21321c6fdbd8SKent Overstreet } 21331c6fdbd8SKent Overstreet 21341c6fdbd8SKent Overstreet /* 21351c6fdbd8SKent Overstreet * Bit of a hack - we don't want truncate to fail due to -ENOSPC. 21361c6fdbd8SKent Overstreet * 21371c6fdbd8SKent Overstreet * XXX: because we aren't currently tracking whether the page has actual 21381c6fdbd8SKent Overstreet * data in it (vs. just 0s, or only partially written) this wrong. ick. 21391c6fdbd8SKent Overstreet */ 21401c6fdbd8SKent Overstreet ret = bch2_get_page_reservation(c, inode, page, false); 21411c6fdbd8SKent Overstreet BUG_ON(ret); 21421c6fdbd8SKent Overstreet 21431c6fdbd8SKent Overstreet if (index == start >> PAGE_SHIFT && 21441c6fdbd8SKent Overstreet index == end >> PAGE_SHIFT) 21451c6fdbd8SKent Overstreet zero_user_segment(page, start_offset, end_offset); 21461c6fdbd8SKent Overstreet else if (index == start >> PAGE_SHIFT) 21471c6fdbd8SKent Overstreet zero_user_segment(page, start_offset, PAGE_SIZE); 21481c6fdbd8SKent Overstreet else if (index == end >> PAGE_SHIFT) 21491c6fdbd8SKent Overstreet zero_user_segment(page, 0, end_offset); 21501c6fdbd8SKent Overstreet 21511c6fdbd8SKent Overstreet if (!PageDirty(page)) 21521c6fdbd8SKent Overstreet set_page_dirty(page); 21531c6fdbd8SKent Overstreet unlock: 21541c6fdbd8SKent Overstreet unlock_page(page); 21551c6fdbd8SKent Overstreet put_page(page); 21561c6fdbd8SKent Overstreet out: 21571c6fdbd8SKent Overstreet return ret; 21581c6fdbd8SKent Overstreet } 21591c6fdbd8SKent Overstreet 21601c6fdbd8SKent Overstreet static int bch2_truncate_page(struct bch_inode_info *inode, loff_t from) 21611c6fdbd8SKent Overstreet { 21621c6fdbd8SKent Overstreet return __bch2_truncate_page(inode, from >> PAGE_SHIFT, 21631c6fdbd8SKent Overstreet from, from + PAGE_SIZE); 21641c6fdbd8SKent Overstreet } 21651c6fdbd8SKent Overstreet 21661c6fdbd8SKent Overstreet static int bch2_extend(struct bch_inode_info *inode, struct iattr *iattr) 21671c6fdbd8SKent Overstreet { 21681c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 21691c6fdbd8SKent Overstreet struct address_space *mapping = inode->v.i_mapping; 21701c6fdbd8SKent Overstreet int ret; 21711c6fdbd8SKent Overstreet 21721c6fdbd8SKent Overstreet ret = filemap_write_and_wait_range(mapping, 21731c6fdbd8SKent Overstreet inode->ei_inode.bi_size, S64_MAX); 21741c6fdbd8SKent Overstreet if (ret) 21751c6fdbd8SKent Overstreet return ret; 21761c6fdbd8SKent Overstreet 21771c6fdbd8SKent Overstreet truncate_setsize(&inode->v, iattr->ia_size); 21781c6fdbd8SKent Overstreet /* ATTR_MODE will never be set here, ns argument isn't needed: */ 21791c6fdbd8SKent Overstreet setattr_copy(NULL, &inode->v, iattr); 21801c6fdbd8SKent Overstreet 21811c6fdbd8SKent Overstreet mutex_lock(&inode->ei_update_lock); 21822ea90048SKent Overstreet ret = bch2_write_inode_size(c, inode, inode->v.i_size, 21832ea90048SKent Overstreet ATTR_MTIME|ATTR_CTIME); 21841c6fdbd8SKent Overstreet mutex_unlock(&inode->ei_update_lock); 21851c6fdbd8SKent Overstreet 21861c6fdbd8SKent Overstreet return ret; 21871c6fdbd8SKent Overstreet } 21881c6fdbd8SKent Overstreet 218954e2264eSKent Overstreet static int bch2_truncate_finish_fn(struct bch_inode_info *inode, 219054e2264eSKent Overstreet struct bch_inode_unpacked *bi, 219154e2264eSKent Overstreet void *p) 219254e2264eSKent Overstreet { 219354e2264eSKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 219454e2264eSKent Overstreet 219554e2264eSKent Overstreet bi->bi_flags &= ~BCH_INODE_I_SIZE_DIRTY; 219654e2264eSKent Overstreet bi->bi_mtime = bi->bi_ctime = bch2_current_time(c); 219754e2264eSKent Overstreet return 0; 219854e2264eSKent Overstreet } 219954e2264eSKent Overstreet 220054e2264eSKent Overstreet static int bch2_truncate_start_fn(struct bch_inode_info *inode, 220154e2264eSKent Overstreet struct bch_inode_unpacked *bi, void *p) 220254e2264eSKent Overstreet { 220354e2264eSKent Overstreet u64 *new_i_size = p; 220454e2264eSKent Overstreet 220554e2264eSKent Overstreet bi->bi_flags |= BCH_INODE_I_SIZE_DIRTY; 220654e2264eSKent Overstreet bi->bi_size = *new_i_size; 220754e2264eSKent Overstreet return 0; 220854e2264eSKent Overstreet } 220954e2264eSKent Overstreet 22101c6fdbd8SKent Overstreet int bch2_truncate(struct bch_inode_info *inode, struct iattr *iattr) 22111c6fdbd8SKent Overstreet { 22121c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 22131c6fdbd8SKent Overstreet struct address_space *mapping = inode->v.i_mapping; 221454e2264eSKent Overstreet u64 new_i_size = iattr->ia_size; 22151c6fdbd8SKent Overstreet bool shrink; 22161c6fdbd8SKent Overstreet int ret = 0; 22171c6fdbd8SKent Overstreet 22181c6fdbd8SKent Overstreet inode_dio_wait(&inode->v); 22191c6fdbd8SKent Overstreet bch2_pagecache_block_get(&inode->ei_pagecache_lock); 22201c6fdbd8SKent Overstreet 22211c6fdbd8SKent Overstreet BUG_ON(inode->v.i_size < inode->ei_inode.bi_size); 22221c6fdbd8SKent Overstreet 22231c6fdbd8SKent Overstreet shrink = iattr->ia_size <= inode->v.i_size; 22241c6fdbd8SKent Overstreet 22251c6fdbd8SKent Overstreet if (!shrink) { 22261c6fdbd8SKent Overstreet ret = bch2_extend(inode, iattr); 222754e2264eSKent Overstreet goto err; 22281c6fdbd8SKent Overstreet } 22291c6fdbd8SKent Overstreet 22301c6fdbd8SKent Overstreet ret = bch2_truncate_page(inode, iattr->ia_size); 22311c6fdbd8SKent Overstreet if (unlikely(ret)) 223254e2264eSKent Overstreet goto err; 22331c6fdbd8SKent Overstreet 22341c6fdbd8SKent Overstreet if (iattr->ia_size > inode->ei_inode.bi_size) 22351c6fdbd8SKent Overstreet ret = filemap_write_and_wait_range(mapping, 22361c6fdbd8SKent Overstreet inode->ei_inode.bi_size, 22371c6fdbd8SKent Overstreet iattr->ia_size - 1); 22381c6fdbd8SKent Overstreet else if (iattr->ia_size & (PAGE_SIZE - 1)) 22391c6fdbd8SKent Overstreet ret = filemap_write_and_wait_range(mapping, 22401c6fdbd8SKent Overstreet round_down(iattr->ia_size, PAGE_SIZE), 22411c6fdbd8SKent Overstreet iattr->ia_size - 1); 22421c6fdbd8SKent Overstreet if (ret) 224354e2264eSKent Overstreet goto err; 22441c6fdbd8SKent Overstreet 224554e2264eSKent Overstreet mutex_lock(&inode->ei_update_lock); 224654e2264eSKent Overstreet ret = bch2_write_inode(c, inode, bch2_truncate_start_fn, 224754e2264eSKent Overstreet &new_i_size, 0); 224854e2264eSKent Overstreet mutex_unlock(&inode->ei_update_lock); 22491c6fdbd8SKent Overstreet 22501c6fdbd8SKent Overstreet if (unlikely(ret)) 225154e2264eSKent Overstreet goto err; 22521c6fdbd8SKent Overstreet 22531c6fdbd8SKent Overstreet truncate_setsize(&inode->v, iattr->ia_size); 22541c6fdbd8SKent Overstreet 225554e2264eSKent Overstreet /* 225654e2264eSKent Overstreet * XXX: need a comment explaining why PAGE_SIZE and not block_bytes() 225754e2264eSKent Overstreet * here: 225854e2264eSKent Overstreet */ 225954e2264eSKent Overstreet ret = __bch2_fpunch(c, inode, 22601c6fdbd8SKent Overstreet round_up(iattr->ia_size, PAGE_SIZE) >> 9, 226154e2264eSKent Overstreet U64_MAX, &inode->ei_journal_seq); 22621c6fdbd8SKent Overstreet if (unlikely(ret)) 226354e2264eSKent Overstreet goto err; 22641c6fdbd8SKent Overstreet 22651c6fdbd8SKent Overstreet /* ATTR_MODE will never be set here, ns argument isn't needed: */ 22661c6fdbd8SKent Overstreet setattr_copy(NULL, &inode->v, iattr); 226754e2264eSKent Overstreet 226854e2264eSKent Overstreet mutex_lock(&inode->ei_update_lock); 226954e2264eSKent Overstreet ret = bch2_write_inode(c, inode, bch2_truncate_finish_fn, NULL, 227054e2264eSKent Overstreet ATTR_MTIME|ATTR_CTIME); 227154e2264eSKent Overstreet mutex_unlock(&inode->ei_update_lock); 227254e2264eSKent Overstreet err: 22731c6fdbd8SKent Overstreet bch2_pagecache_block_put(&inode->ei_pagecache_lock); 22741c6fdbd8SKent Overstreet return ret; 22751c6fdbd8SKent Overstreet } 22761c6fdbd8SKent Overstreet 22771c6fdbd8SKent Overstreet /* fallocate: */ 22781c6fdbd8SKent Overstreet 22791c6fdbd8SKent Overstreet static long bch2_fpunch(struct bch_inode_info *inode, loff_t offset, loff_t len) 22801c6fdbd8SKent Overstreet { 22811c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 22821c6fdbd8SKent Overstreet u64 discard_start = round_up(offset, PAGE_SIZE) >> 9; 22831c6fdbd8SKent Overstreet u64 discard_end = round_down(offset + len, PAGE_SIZE) >> 9; 22841c6fdbd8SKent Overstreet int ret = 0; 22851c6fdbd8SKent Overstreet 22861c6fdbd8SKent Overstreet inode_lock(&inode->v); 22871c6fdbd8SKent Overstreet inode_dio_wait(&inode->v); 22881c6fdbd8SKent Overstreet bch2_pagecache_block_get(&inode->ei_pagecache_lock); 22891c6fdbd8SKent Overstreet 22901c6fdbd8SKent Overstreet ret = __bch2_truncate_page(inode, 22911c6fdbd8SKent Overstreet offset >> PAGE_SHIFT, 22921c6fdbd8SKent Overstreet offset, offset + len); 22931c6fdbd8SKent Overstreet if (unlikely(ret)) 22941c6fdbd8SKent Overstreet goto err; 22951c6fdbd8SKent Overstreet 22961c6fdbd8SKent Overstreet if (offset >> PAGE_SHIFT != 22971c6fdbd8SKent Overstreet (offset + len) >> PAGE_SHIFT) { 22981c6fdbd8SKent Overstreet ret = __bch2_truncate_page(inode, 22991c6fdbd8SKent Overstreet (offset + len) >> PAGE_SHIFT, 23001c6fdbd8SKent Overstreet offset, offset + len); 23011c6fdbd8SKent Overstreet if (unlikely(ret)) 23021c6fdbd8SKent Overstreet goto err; 23031c6fdbd8SKent Overstreet } 23041c6fdbd8SKent Overstreet 23051c6fdbd8SKent Overstreet truncate_pagecache_range(&inode->v, offset, offset + len - 1); 23061c6fdbd8SKent Overstreet 23075f461e01SKent Overstreet if (discard_start < discard_end) 23085f461e01SKent Overstreet ret = __bch2_fpunch(c, inode, discard_start, discard_end, 23091c6fdbd8SKent Overstreet &inode->ei_journal_seq); 23101c6fdbd8SKent Overstreet err: 23111c6fdbd8SKent Overstreet bch2_pagecache_block_put(&inode->ei_pagecache_lock); 23121c6fdbd8SKent Overstreet inode_unlock(&inode->v); 23131c6fdbd8SKent Overstreet 23141c6fdbd8SKent Overstreet return ret; 23151c6fdbd8SKent Overstreet } 23161c6fdbd8SKent Overstreet 23171c6fdbd8SKent Overstreet static long bch2_fcollapse(struct bch_inode_info *inode, 23181c6fdbd8SKent Overstreet loff_t offset, loff_t len) 23191c6fdbd8SKent Overstreet { 23201c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 23211c6fdbd8SKent Overstreet struct address_space *mapping = inode->v.i_mapping; 2322d69f41d6SKent Overstreet struct btree_trans trans; 2323d69f41d6SKent Overstreet struct btree_iter *src, *dst; 23241c6fdbd8SKent Overstreet BKEY_PADDED(k) copy; 23251c6fdbd8SKent Overstreet struct bkey_s_c k; 23261c6fdbd8SKent Overstreet loff_t new_size; 23271c6fdbd8SKent Overstreet int ret; 23281c6fdbd8SKent Overstreet 23291c6fdbd8SKent Overstreet if ((offset | len) & (block_bytes(c) - 1)) 23301c6fdbd8SKent Overstreet return -EINVAL; 23311c6fdbd8SKent Overstreet 2332d69f41d6SKent Overstreet bch2_trans_init(&trans, c); 23338ef231bdSKent Overstreet bch2_trans_preload_iters(&trans); 23341c6fdbd8SKent Overstreet 23351c6fdbd8SKent Overstreet /* 23361c6fdbd8SKent Overstreet * We need i_mutex to keep the page cache consistent with the extents 23371c6fdbd8SKent Overstreet * btree, and the btree consistent with i_size - we don't need outside 23381c6fdbd8SKent Overstreet * locking for the extents btree itself, because we're using linked 23391c6fdbd8SKent Overstreet * iterators 23401c6fdbd8SKent Overstreet */ 23411c6fdbd8SKent Overstreet inode_lock(&inode->v); 23421c6fdbd8SKent Overstreet inode_dio_wait(&inode->v); 23431c6fdbd8SKent Overstreet bch2_pagecache_block_get(&inode->ei_pagecache_lock); 23441c6fdbd8SKent Overstreet 23451c6fdbd8SKent Overstreet ret = -EINVAL; 23461c6fdbd8SKent Overstreet if (offset + len >= inode->v.i_size) 23471c6fdbd8SKent Overstreet goto err; 23481c6fdbd8SKent Overstreet 23491c6fdbd8SKent Overstreet if (inode->v.i_size < len) 23501c6fdbd8SKent Overstreet goto err; 23511c6fdbd8SKent Overstreet 23521c6fdbd8SKent Overstreet new_size = inode->v.i_size - len; 23531c6fdbd8SKent Overstreet 23541c6fdbd8SKent Overstreet ret = write_invalidate_inode_pages_range(mapping, offset, LLONG_MAX); 23551c6fdbd8SKent Overstreet if (ret) 23561c6fdbd8SKent Overstreet goto err; 23571c6fdbd8SKent Overstreet 23588ef231bdSKent Overstreet dst = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS, 23598ef231bdSKent Overstreet POS(inode->v.i_ino, offset >> 9), 23608ef231bdSKent Overstreet BTREE_ITER_SLOTS|BTREE_ITER_INTENT); 23618ef231bdSKent Overstreet BUG_ON(IS_ERR_OR_NULL(dst)); 23628ef231bdSKent Overstreet 23638ef231bdSKent Overstreet src = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS, 23648ef231bdSKent Overstreet POS_MIN, BTREE_ITER_SLOTS); 23658ef231bdSKent Overstreet BUG_ON(IS_ERR_OR_NULL(src)); 23661c6fdbd8SKent Overstreet 2367d69f41d6SKent Overstreet while (bkey_cmp(dst->pos, 23681c6fdbd8SKent Overstreet POS(inode->v.i_ino, 23691c6fdbd8SKent Overstreet round_up(new_size, PAGE_SIZE) >> 9)) < 0) { 23701c6fdbd8SKent Overstreet struct disk_reservation disk_res; 23711c6fdbd8SKent Overstreet 23728ef231bdSKent Overstreet ret = bch2_btree_iter_traverse(dst); 23738ef231bdSKent Overstreet if (ret) 23748ef231bdSKent Overstreet goto btree_iter_err; 23758ef231bdSKent Overstreet 2376d69f41d6SKent Overstreet bch2_btree_iter_set_pos(src, 2377d69f41d6SKent Overstreet POS(dst->pos.inode, dst->pos.offset + (len >> 9))); 23781c6fdbd8SKent Overstreet 2379d69f41d6SKent Overstreet k = bch2_btree_iter_peek_slot(src); 23801c6fdbd8SKent Overstreet if ((ret = btree_iter_err(k))) 23811c6fdbd8SKent Overstreet goto btree_iter_err; 23821c6fdbd8SKent Overstreet 23831c6fdbd8SKent Overstreet bkey_reassemble(©.k, k); 23841c6fdbd8SKent Overstreet 2385d69f41d6SKent Overstreet bch2_cut_front(src->pos, ©.k); 23861c6fdbd8SKent Overstreet copy.k.k.p.offset -= len >> 9; 23871c6fdbd8SKent Overstreet 2388e2d9912cSKent Overstreet bch2_extent_trim_atomic(©.k, dst); 2389e2d9912cSKent Overstreet 2390d69f41d6SKent Overstreet BUG_ON(bkey_cmp(dst->pos, bkey_start_pos(©.k.k))); 23911c6fdbd8SKent Overstreet 23921c6fdbd8SKent Overstreet ret = bch2_disk_reservation_get(c, &disk_res, copy.k.k.size, 23931c6fdbd8SKent Overstreet bch2_extent_nr_dirty_ptrs(bkey_i_to_s_c(©.k)), 23941c6fdbd8SKent Overstreet BCH_DISK_RESERVATION_NOFAIL); 23951c6fdbd8SKent Overstreet BUG_ON(ret); 23961c6fdbd8SKent Overstreet 23978ef231bdSKent Overstreet ret = bch2_extent_update(&trans, inode, 23988ef231bdSKent Overstreet &disk_res, NULL, 23998ef231bdSKent Overstreet dst, ©.k, 24008ef231bdSKent Overstreet 0, true, true, NULL); 24011c6fdbd8SKent Overstreet bch2_disk_reservation_put(c, &disk_res); 24021c6fdbd8SKent Overstreet btree_iter_err: 24031c6fdbd8SKent Overstreet if (ret == -EINTR) 24041c6fdbd8SKent Overstreet ret = 0; 24058ef231bdSKent Overstreet if (ret) 24068ef231bdSKent Overstreet goto err; 24071c6fdbd8SKent Overstreet /* 24081c6fdbd8SKent Overstreet * XXX: if we error here we've left data with multiple 24091c6fdbd8SKent Overstreet * pointers... which isn't a _super_ serious problem... 24101c6fdbd8SKent Overstreet */ 24111c6fdbd8SKent Overstreet 2412d69f41d6SKent Overstreet bch2_btree_iter_cond_resched(src); 24131c6fdbd8SKent Overstreet } 24148ef231bdSKent Overstreet bch2_trans_unlock(&trans); 24151c6fdbd8SKent Overstreet 24168ef231bdSKent Overstreet ret = __bch2_fpunch(c, inode, 24171c6fdbd8SKent Overstreet round_up(new_size, block_bytes(c)) >> 9, 24188ef231bdSKent Overstreet U64_MAX, &inode->ei_journal_seq); 24191c6fdbd8SKent Overstreet if (ret) 24208ef231bdSKent Overstreet goto err; 24211c6fdbd8SKent Overstreet 24228ef231bdSKent Overstreet i_size_write(&inode->v, new_size); 24238ef231bdSKent Overstreet mutex_lock(&inode->ei_update_lock); 24248ef231bdSKent Overstreet ret = bch2_write_inode_size(c, inode, new_size, 24258ef231bdSKent Overstreet ATTR_MTIME|ATTR_CTIME); 24268ef231bdSKent Overstreet mutex_unlock(&inode->ei_update_lock); 24271c6fdbd8SKent Overstreet err: 24288ef231bdSKent Overstreet bch2_trans_exit(&trans); 24291c6fdbd8SKent Overstreet bch2_pagecache_block_put(&inode->ei_pagecache_lock); 24301c6fdbd8SKent Overstreet inode_unlock(&inode->v); 24311c6fdbd8SKent Overstreet return ret; 24321c6fdbd8SKent Overstreet } 24331c6fdbd8SKent Overstreet 24341c6fdbd8SKent Overstreet static long bch2_fallocate(struct bch_inode_info *inode, int mode, 24351c6fdbd8SKent Overstreet loff_t offset, loff_t len) 24361c6fdbd8SKent Overstreet { 24371c6fdbd8SKent Overstreet struct address_space *mapping = inode->v.i_mapping; 24381c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 2439190fa7afSKent Overstreet struct btree_trans trans; 2440190fa7afSKent Overstreet struct btree_iter *iter; 24411c6fdbd8SKent Overstreet struct bpos end_pos; 24421c6fdbd8SKent Overstreet loff_t block_start, block_end; 24431c6fdbd8SKent Overstreet loff_t end = offset + len; 24441c6fdbd8SKent Overstreet unsigned sectors; 24451c6fdbd8SKent Overstreet unsigned replicas = io_opts(c, inode).data_replicas; 24461c6fdbd8SKent Overstreet int ret; 24471c6fdbd8SKent Overstreet 2448190fa7afSKent Overstreet bch2_trans_init(&trans, c); 2449190fa7afSKent Overstreet bch2_trans_preload_iters(&trans); 24501c6fdbd8SKent Overstreet 24511c6fdbd8SKent Overstreet inode_lock(&inode->v); 24521c6fdbd8SKent Overstreet inode_dio_wait(&inode->v); 24531c6fdbd8SKent Overstreet bch2_pagecache_block_get(&inode->ei_pagecache_lock); 24541c6fdbd8SKent Overstreet 24551c6fdbd8SKent Overstreet if (!(mode & FALLOC_FL_KEEP_SIZE) && end > inode->v.i_size) { 24561c6fdbd8SKent Overstreet ret = inode_newsize_ok(&inode->v, end); 24571c6fdbd8SKent Overstreet if (ret) 24581c6fdbd8SKent Overstreet goto err; 24591c6fdbd8SKent Overstreet } 24601c6fdbd8SKent Overstreet 24611c6fdbd8SKent Overstreet if (mode & FALLOC_FL_ZERO_RANGE) { 24621c6fdbd8SKent Overstreet ret = __bch2_truncate_page(inode, 24631c6fdbd8SKent Overstreet offset >> PAGE_SHIFT, 24641c6fdbd8SKent Overstreet offset, end); 24651c6fdbd8SKent Overstreet 24661c6fdbd8SKent Overstreet if (!ret && 24671c6fdbd8SKent Overstreet offset >> PAGE_SHIFT != end >> PAGE_SHIFT) 24681c6fdbd8SKent Overstreet ret = __bch2_truncate_page(inode, 24691c6fdbd8SKent Overstreet end >> PAGE_SHIFT, 24701c6fdbd8SKent Overstreet offset, end); 24711c6fdbd8SKent Overstreet 24721c6fdbd8SKent Overstreet if (unlikely(ret)) 24731c6fdbd8SKent Overstreet goto err; 24741c6fdbd8SKent Overstreet 24751c6fdbd8SKent Overstreet truncate_pagecache_range(&inode->v, offset, end - 1); 24761c6fdbd8SKent Overstreet 24771c6fdbd8SKent Overstreet block_start = round_up(offset, PAGE_SIZE); 24781c6fdbd8SKent Overstreet block_end = round_down(end, PAGE_SIZE); 24791c6fdbd8SKent Overstreet } else { 24801c6fdbd8SKent Overstreet block_start = round_down(offset, PAGE_SIZE); 24811c6fdbd8SKent Overstreet block_end = round_up(end, PAGE_SIZE); 24821c6fdbd8SKent Overstreet } 24831c6fdbd8SKent Overstreet 2484190fa7afSKent Overstreet iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS, 2485190fa7afSKent Overstreet POS(inode->v.i_ino, block_start >> 9), 2486190fa7afSKent Overstreet BTREE_ITER_SLOTS|BTREE_ITER_INTENT); 24871c6fdbd8SKent Overstreet end_pos = POS(inode->v.i_ino, block_end >> 9); 24881c6fdbd8SKent Overstreet 2489190fa7afSKent Overstreet while (bkey_cmp(iter->pos, end_pos) < 0) { 24901c6fdbd8SKent Overstreet struct disk_reservation disk_res = { 0 }; 2491190fa7afSKent Overstreet struct quota_res quota_res = { 0 }; 24921c6fdbd8SKent Overstreet struct bkey_i_reservation reservation; 24931c6fdbd8SKent Overstreet struct bkey_s_c k; 24941c6fdbd8SKent Overstreet 2495190fa7afSKent Overstreet k = bch2_btree_iter_peek_slot(iter); 24961c6fdbd8SKent Overstreet if ((ret = btree_iter_err(k))) 24971c6fdbd8SKent Overstreet goto btree_iter_err; 24981c6fdbd8SKent Overstreet 24991c6fdbd8SKent Overstreet /* already reserved */ 25001c6fdbd8SKent Overstreet if (k.k->type == BCH_RESERVATION && 25011c6fdbd8SKent Overstreet bkey_s_c_to_reservation(k).v->nr_replicas >= replicas) { 2502190fa7afSKent Overstreet bch2_btree_iter_next_slot(iter); 25031c6fdbd8SKent Overstreet continue; 25041c6fdbd8SKent Overstreet } 25051c6fdbd8SKent Overstreet 2506190fa7afSKent Overstreet if (bkey_extent_is_data(k.k) && 2507190fa7afSKent Overstreet !(mode & FALLOC_FL_ZERO_RANGE)) { 2508190fa7afSKent Overstreet bch2_btree_iter_next_slot(iter); 25091c6fdbd8SKent Overstreet continue; 25101c6fdbd8SKent Overstreet } 25111c6fdbd8SKent Overstreet 25121c6fdbd8SKent Overstreet bkey_reservation_init(&reservation.k_i); 25131c6fdbd8SKent Overstreet reservation.k.type = BCH_RESERVATION; 25141c6fdbd8SKent Overstreet reservation.k.p = k.k->p; 25151c6fdbd8SKent Overstreet reservation.k.size = k.k->size; 25161c6fdbd8SKent Overstreet 2517190fa7afSKent Overstreet bch2_cut_front(iter->pos, &reservation.k_i); 25181c6fdbd8SKent Overstreet bch2_cut_back(end_pos, &reservation.k); 25191c6fdbd8SKent Overstreet 25201c6fdbd8SKent Overstreet sectors = reservation.k.size; 25211c6fdbd8SKent Overstreet reservation.v.nr_replicas = bch2_extent_nr_dirty_ptrs(k); 25221c6fdbd8SKent Overstreet 25231c6fdbd8SKent Overstreet if (!bkey_extent_is_allocation(k.k)) { 25241c6fdbd8SKent Overstreet ret = bch2_quota_reservation_add(c, inode, 2525190fa7afSKent Overstreet "a_res, 25261c6fdbd8SKent Overstreet sectors, true); 25271c6fdbd8SKent Overstreet if (unlikely(ret)) 25281c6fdbd8SKent Overstreet goto btree_iter_err; 25291c6fdbd8SKent Overstreet } 25301c6fdbd8SKent Overstreet 25311c6fdbd8SKent Overstreet if (reservation.v.nr_replicas < replicas || 25321c6fdbd8SKent Overstreet bch2_extent_is_compressed(k)) { 25331c6fdbd8SKent Overstreet ret = bch2_disk_reservation_get(c, &disk_res, sectors, 25341c6fdbd8SKent Overstreet replicas, 0); 25351c6fdbd8SKent Overstreet if (unlikely(ret)) 25361c6fdbd8SKent Overstreet goto btree_iter_err; 25371c6fdbd8SKent Overstreet 25381c6fdbd8SKent Overstreet reservation.v.nr_replicas = disk_res.nr_replicas; 25391c6fdbd8SKent Overstreet } 25401c6fdbd8SKent Overstreet 2541190fa7afSKent Overstreet ret = bch2_extent_update(&trans, inode, 2542190fa7afSKent Overstreet &disk_res, "a_res, 2543190fa7afSKent Overstreet iter, &reservation.k_i, 2544190fa7afSKent Overstreet 0, true, true, NULL); 2545*b1ba2359SKent Overstreet btree_iter_err: 2546190fa7afSKent Overstreet bch2_quota_reservation_put(c, inode, "a_res); 25471c6fdbd8SKent Overstreet bch2_disk_reservation_put(c, &disk_res); 25481c6fdbd8SKent Overstreet if (ret == -EINTR) 25491c6fdbd8SKent Overstreet ret = 0; 2550190fa7afSKent Overstreet if (ret) 2551190fa7afSKent Overstreet goto err; 25521c6fdbd8SKent Overstreet } 2553190fa7afSKent Overstreet bch2_trans_unlock(&trans); 25541c6fdbd8SKent Overstreet 25551c6fdbd8SKent Overstreet if (!(mode & FALLOC_FL_KEEP_SIZE) && 25561c6fdbd8SKent Overstreet end > inode->v.i_size) { 25571c6fdbd8SKent Overstreet i_size_write(&inode->v, end); 25581c6fdbd8SKent Overstreet 25591c6fdbd8SKent Overstreet mutex_lock(&inode->ei_update_lock); 25602ea90048SKent Overstreet ret = bch2_write_inode_size(c, inode, inode->v.i_size, 0); 25611c6fdbd8SKent Overstreet mutex_unlock(&inode->ei_update_lock); 25621c6fdbd8SKent Overstreet } 25631c6fdbd8SKent Overstreet 25641c6fdbd8SKent Overstreet /* blech */ 25651c6fdbd8SKent Overstreet if ((mode & FALLOC_FL_KEEP_SIZE) && 25661c6fdbd8SKent Overstreet (mode & FALLOC_FL_ZERO_RANGE) && 25671c6fdbd8SKent Overstreet inode->ei_inode.bi_size != inode->v.i_size) { 25681c6fdbd8SKent Overstreet /* sync appends.. */ 25691c6fdbd8SKent Overstreet ret = filemap_write_and_wait_range(mapping, 25701c6fdbd8SKent Overstreet inode->ei_inode.bi_size, S64_MAX); 25711c6fdbd8SKent Overstreet if (ret) 25721c6fdbd8SKent Overstreet goto err; 25731c6fdbd8SKent Overstreet 25741c6fdbd8SKent Overstreet if (inode->ei_inode.bi_size != inode->v.i_size) { 25751c6fdbd8SKent Overstreet mutex_lock(&inode->ei_update_lock); 25762ea90048SKent Overstreet ret = bch2_write_inode_size(c, inode, 25772ea90048SKent Overstreet inode->v.i_size, 0); 25781c6fdbd8SKent Overstreet mutex_unlock(&inode->ei_update_lock); 25791c6fdbd8SKent Overstreet } 25801c6fdbd8SKent Overstreet } 25811c6fdbd8SKent Overstreet err: 2582190fa7afSKent Overstreet bch2_trans_exit(&trans); 25831c6fdbd8SKent Overstreet bch2_pagecache_block_put(&inode->ei_pagecache_lock); 25841c6fdbd8SKent Overstreet inode_unlock(&inode->v); 25851c6fdbd8SKent Overstreet return ret; 25861c6fdbd8SKent Overstreet } 25871c6fdbd8SKent Overstreet 25881c6fdbd8SKent Overstreet long bch2_fallocate_dispatch(struct file *file, int mode, 25891c6fdbd8SKent Overstreet loff_t offset, loff_t len) 25901c6fdbd8SKent Overstreet { 25911c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 25921c6fdbd8SKent Overstreet 25931c6fdbd8SKent Overstreet if (!(mode & ~(FALLOC_FL_KEEP_SIZE|FALLOC_FL_ZERO_RANGE))) 25941c6fdbd8SKent Overstreet return bch2_fallocate(inode, mode, offset, len); 25951c6fdbd8SKent Overstreet 25961c6fdbd8SKent Overstreet if (mode == (FALLOC_FL_PUNCH_HOLE|FALLOC_FL_KEEP_SIZE)) 25971c6fdbd8SKent Overstreet return bch2_fpunch(inode, offset, len); 25981c6fdbd8SKent Overstreet 25991c6fdbd8SKent Overstreet if (mode == FALLOC_FL_COLLAPSE_RANGE) 26001c6fdbd8SKent Overstreet return bch2_fcollapse(inode, offset, len); 26011c6fdbd8SKent Overstreet 26021c6fdbd8SKent Overstreet return -EOPNOTSUPP; 26031c6fdbd8SKent Overstreet } 26041c6fdbd8SKent Overstreet 26051c6fdbd8SKent Overstreet /* fseek: */ 26061c6fdbd8SKent Overstreet 26071c6fdbd8SKent Overstreet static bool folio_is_data(struct folio *folio) 26081c6fdbd8SKent Overstreet { 26091c6fdbd8SKent Overstreet /* XXX: should only have to check PageDirty */ 26101c6fdbd8SKent Overstreet return folio_test_private(folio) && 26111c6fdbd8SKent Overstreet (page_state(&folio->page)->sectors || 26121c6fdbd8SKent Overstreet page_state(&folio->page)->dirty_sectors); 26131c6fdbd8SKent Overstreet } 26141c6fdbd8SKent Overstreet 26151c6fdbd8SKent Overstreet static loff_t bch2_next_pagecache_data(struct inode *vinode, 26161c6fdbd8SKent Overstreet loff_t start_offset, 26171c6fdbd8SKent Overstreet loff_t end_offset) 26181c6fdbd8SKent Overstreet { 26191c6fdbd8SKent Overstreet struct folio_batch fbatch; 26201c6fdbd8SKent Overstreet pgoff_t start_index = start_offset >> PAGE_SHIFT; 26211c6fdbd8SKent Overstreet pgoff_t end_index = end_offset >> PAGE_SHIFT; 26221c6fdbd8SKent Overstreet pgoff_t index = start_index; 26231c6fdbd8SKent Overstreet unsigned i; 26241c6fdbd8SKent Overstreet 26251c6fdbd8SKent Overstreet folio_batch_init(&fbatch); 26261c6fdbd8SKent Overstreet 26271c6fdbd8SKent Overstreet while (filemap_get_folios(vinode->i_mapping, 26281c6fdbd8SKent Overstreet &index, end_index, &fbatch)) { 26291c6fdbd8SKent Overstreet for (i = 0; i < folio_batch_count(&fbatch); i++) { 26301c6fdbd8SKent Overstreet struct folio *folio = fbatch.folios[i]; 26311c6fdbd8SKent Overstreet 26321c6fdbd8SKent Overstreet folio_lock(folio); 26331c6fdbd8SKent Overstreet if (folio_is_data(folio)) { 26341c6fdbd8SKent Overstreet end_offset = 26351c6fdbd8SKent Overstreet min(end_offset, 26361c6fdbd8SKent Overstreet max(start_offset, 26371c6fdbd8SKent Overstreet ((loff_t) index) << PAGE_SHIFT)); 26381c6fdbd8SKent Overstreet folio_unlock(folio); 26391c6fdbd8SKent Overstreet folio_batch_release(&fbatch); 26401c6fdbd8SKent Overstreet return end_offset; 26411c6fdbd8SKent Overstreet } 26421c6fdbd8SKent Overstreet folio_unlock(folio); 26431c6fdbd8SKent Overstreet } 26441c6fdbd8SKent Overstreet folio_batch_release(&fbatch); 26451c6fdbd8SKent Overstreet cond_resched(); 26461c6fdbd8SKent Overstreet } 26471c6fdbd8SKent Overstreet 26481c6fdbd8SKent Overstreet return end_offset; 26491c6fdbd8SKent Overstreet } 26501c6fdbd8SKent Overstreet 26511c6fdbd8SKent Overstreet static loff_t bch2_seek_data(struct file *file, u64 offset) 26521c6fdbd8SKent Overstreet { 26531c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 26541c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 26551c6fdbd8SKent Overstreet struct btree_iter iter; 26561c6fdbd8SKent Overstreet struct bkey_s_c k; 26571c6fdbd8SKent Overstreet u64 isize, next_data = MAX_LFS_FILESIZE; 26581c6fdbd8SKent Overstreet int ret; 26591c6fdbd8SKent Overstreet 26601c6fdbd8SKent Overstreet isize = i_size_read(&inode->v); 26611c6fdbd8SKent Overstreet if (offset >= isize) 26621c6fdbd8SKent Overstreet return -ENXIO; 26631c6fdbd8SKent Overstreet 26641c6fdbd8SKent Overstreet for_each_btree_key(&iter, c, BTREE_ID_EXTENTS, 26651c6fdbd8SKent Overstreet POS(inode->v.i_ino, offset >> 9), 0, k) { 26661c6fdbd8SKent Overstreet if (k.k->p.inode != inode->v.i_ino) { 26671c6fdbd8SKent Overstreet break; 26681c6fdbd8SKent Overstreet } else if (bkey_extent_is_data(k.k)) { 26691c6fdbd8SKent Overstreet next_data = max(offset, bkey_start_offset(k.k) << 9); 26701c6fdbd8SKent Overstreet break; 26711c6fdbd8SKent Overstreet } else if (k.k->p.offset >> 9 > isize) 26721c6fdbd8SKent Overstreet break; 26731c6fdbd8SKent Overstreet } 26741c6fdbd8SKent Overstreet 26751c6fdbd8SKent Overstreet ret = bch2_btree_iter_unlock(&iter); 26761c6fdbd8SKent Overstreet if (ret) 26771c6fdbd8SKent Overstreet return ret; 26781c6fdbd8SKent Overstreet 26791c6fdbd8SKent Overstreet if (next_data > offset) 26801c6fdbd8SKent Overstreet next_data = bch2_next_pagecache_data(&inode->v, 26811c6fdbd8SKent Overstreet offset, next_data); 26821c6fdbd8SKent Overstreet 26831c6fdbd8SKent Overstreet if (next_data > isize) 26841c6fdbd8SKent Overstreet return -ENXIO; 26851c6fdbd8SKent Overstreet 26861c6fdbd8SKent Overstreet return vfs_setpos(file, next_data, MAX_LFS_FILESIZE); 26871c6fdbd8SKent Overstreet } 26881c6fdbd8SKent Overstreet 26891c6fdbd8SKent Overstreet static bool page_slot_is_data(struct address_space *mapping, pgoff_t index) 26901c6fdbd8SKent Overstreet { 26911c6fdbd8SKent Overstreet struct page *page; 26921c6fdbd8SKent Overstreet bool ret; 26931c6fdbd8SKent Overstreet 26941c6fdbd8SKent Overstreet page = find_lock_page(mapping, index); 26951c6fdbd8SKent Overstreet if (!page) 26961c6fdbd8SKent Overstreet return false; 26971c6fdbd8SKent Overstreet 26981c6fdbd8SKent Overstreet ret = folio_is_data(page_folio(page)); 26991c6fdbd8SKent Overstreet unlock_page(page); 27001c6fdbd8SKent Overstreet 27011c6fdbd8SKent Overstreet return ret; 27021c6fdbd8SKent Overstreet } 27031c6fdbd8SKent Overstreet 27041c6fdbd8SKent Overstreet static loff_t bch2_next_pagecache_hole(struct inode *vinode, 27051c6fdbd8SKent Overstreet loff_t start_offset, 27061c6fdbd8SKent Overstreet loff_t end_offset) 27071c6fdbd8SKent Overstreet { 27081c6fdbd8SKent Overstreet struct address_space *mapping = vinode->i_mapping; 27091c6fdbd8SKent Overstreet pgoff_t index; 27101c6fdbd8SKent Overstreet 27111c6fdbd8SKent Overstreet for (index = start_offset >> PAGE_SHIFT; 27121c6fdbd8SKent Overstreet index < end_offset >> PAGE_SHIFT; 27131c6fdbd8SKent Overstreet index++) 27141c6fdbd8SKent Overstreet if (!page_slot_is_data(mapping, index)) 27151c6fdbd8SKent Overstreet end_offset = max(start_offset, 27161c6fdbd8SKent Overstreet ((loff_t) index) << PAGE_SHIFT); 27171c6fdbd8SKent Overstreet 27181c6fdbd8SKent Overstreet return end_offset; 27191c6fdbd8SKent Overstreet } 27201c6fdbd8SKent Overstreet 27211c6fdbd8SKent Overstreet static loff_t bch2_seek_hole(struct file *file, u64 offset) 27221c6fdbd8SKent Overstreet { 27231c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 27241c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 27251c6fdbd8SKent Overstreet struct btree_iter iter; 27261c6fdbd8SKent Overstreet struct bkey_s_c k; 27271c6fdbd8SKent Overstreet u64 isize, next_hole = MAX_LFS_FILESIZE; 27281c6fdbd8SKent Overstreet int ret; 27291c6fdbd8SKent Overstreet 27301c6fdbd8SKent Overstreet isize = i_size_read(&inode->v); 27311c6fdbd8SKent Overstreet if (offset >= isize) 27321c6fdbd8SKent Overstreet return -ENXIO; 27331c6fdbd8SKent Overstreet 27341c6fdbd8SKent Overstreet for_each_btree_key(&iter, c, BTREE_ID_EXTENTS, 27351c6fdbd8SKent Overstreet POS(inode->v.i_ino, offset >> 9), 27361c6fdbd8SKent Overstreet BTREE_ITER_SLOTS, k) { 27371c6fdbd8SKent Overstreet if (k.k->p.inode != inode->v.i_ino) { 27381c6fdbd8SKent Overstreet next_hole = bch2_next_pagecache_hole(&inode->v, 27391c6fdbd8SKent Overstreet offset, MAX_LFS_FILESIZE); 27401c6fdbd8SKent Overstreet break; 27411c6fdbd8SKent Overstreet } else if (!bkey_extent_is_data(k.k)) { 27421c6fdbd8SKent Overstreet next_hole = bch2_next_pagecache_hole(&inode->v, 27431c6fdbd8SKent Overstreet max(offset, bkey_start_offset(k.k) << 9), 27441c6fdbd8SKent Overstreet k.k->p.offset << 9); 27451c6fdbd8SKent Overstreet 27461c6fdbd8SKent Overstreet if (next_hole < k.k->p.offset << 9) 27471c6fdbd8SKent Overstreet break; 27481c6fdbd8SKent Overstreet } else { 27491c6fdbd8SKent Overstreet offset = max(offset, bkey_start_offset(k.k) << 9); 27501c6fdbd8SKent Overstreet } 27511c6fdbd8SKent Overstreet } 27521c6fdbd8SKent Overstreet 27531c6fdbd8SKent Overstreet ret = bch2_btree_iter_unlock(&iter); 27541c6fdbd8SKent Overstreet if (ret) 27551c6fdbd8SKent Overstreet return ret; 27561c6fdbd8SKent Overstreet 27571c6fdbd8SKent Overstreet if (next_hole > isize) 27581c6fdbd8SKent Overstreet next_hole = isize; 27591c6fdbd8SKent Overstreet 27601c6fdbd8SKent Overstreet return vfs_setpos(file, next_hole, MAX_LFS_FILESIZE); 27611c6fdbd8SKent Overstreet } 27621c6fdbd8SKent Overstreet 27631c6fdbd8SKent Overstreet loff_t bch2_llseek(struct file *file, loff_t offset, int whence) 27641c6fdbd8SKent Overstreet { 27651c6fdbd8SKent Overstreet switch (whence) { 27661c6fdbd8SKent Overstreet case SEEK_SET: 27671c6fdbd8SKent Overstreet case SEEK_CUR: 27681c6fdbd8SKent Overstreet case SEEK_END: 27691c6fdbd8SKent Overstreet return generic_file_llseek(file, offset, whence); 27701c6fdbd8SKent Overstreet case SEEK_DATA: 27711c6fdbd8SKent Overstreet return bch2_seek_data(file, offset); 27721c6fdbd8SKent Overstreet case SEEK_HOLE: 27731c6fdbd8SKent Overstreet return bch2_seek_hole(file, offset); 27741c6fdbd8SKent Overstreet } 27751c6fdbd8SKent Overstreet 27761c6fdbd8SKent Overstreet return -EINVAL; 27771c6fdbd8SKent Overstreet } 27781c6fdbd8SKent Overstreet 27791c6fdbd8SKent Overstreet void bch2_fs_fsio_exit(struct bch_fs *c) 27801c6fdbd8SKent Overstreet { 27811c6fdbd8SKent Overstreet bioset_exit(&c->dio_write_bioset); 27821c6fdbd8SKent Overstreet bioset_exit(&c->dio_read_bioset); 27831c6fdbd8SKent Overstreet bioset_exit(&c->writepage_bioset); 27841c6fdbd8SKent Overstreet } 27851c6fdbd8SKent Overstreet 27861c6fdbd8SKent Overstreet int bch2_fs_fsio_init(struct bch_fs *c) 27871c6fdbd8SKent Overstreet { 27881c6fdbd8SKent Overstreet int ret = 0; 27891c6fdbd8SKent Overstreet 27901c6fdbd8SKent Overstreet pr_verbose_init(c->opts, ""); 27911c6fdbd8SKent Overstreet 27921c6fdbd8SKent Overstreet if (bioset_init(&c->writepage_bioset, 27931c6fdbd8SKent Overstreet 4, offsetof(struct bch_writepage_io, op.op.wbio.bio), 27941c6fdbd8SKent Overstreet BIOSET_NEED_BVECS) || 27951c6fdbd8SKent Overstreet bioset_init(&c->dio_read_bioset, 27961c6fdbd8SKent Overstreet 4, offsetof(struct dio_read, rbio.bio), 27971c6fdbd8SKent Overstreet BIOSET_NEED_BVECS) || 27981c6fdbd8SKent Overstreet bioset_init(&c->dio_write_bioset, 27991c6fdbd8SKent Overstreet 4, offsetof(struct dio_write, iop.op.wbio.bio), 28001c6fdbd8SKent Overstreet BIOSET_NEED_BVECS)) 28011c6fdbd8SKent Overstreet ret = -ENOMEM; 28021c6fdbd8SKent Overstreet 28031c6fdbd8SKent Overstreet pr_verbose_init(c->opts, "ret %i", ret); 28041c6fdbd8SKent Overstreet return ret; 28051c6fdbd8SKent Overstreet } 28061c6fdbd8SKent Overstreet 28071c6fdbd8SKent Overstreet #endif /* NO_BCACHEFS_FS */ 2808