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" 607a1006aSKent Overstreet #include "bkey_buf.h" 71c6fdbd8SKent Overstreet #include "btree_update.h" 81c6fdbd8SKent Overstreet #include "buckets.h" 91c6fdbd8SKent Overstreet #include "clock.h" 101c6fdbd8SKent Overstreet #include "error.h" 11e2d9912cSKent Overstreet #include "extents.h" 1208c07feaSKent Overstreet #include "extent_update.h" 131c6fdbd8SKent Overstreet #include "fs.h" 141c6fdbd8SKent Overstreet #include "fs-io.h" 151c6fdbd8SKent Overstreet #include "fsck.h" 161c6fdbd8SKent Overstreet #include "inode.h" 171c6fdbd8SKent Overstreet #include "journal.h" 181c6fdbd8SKent Overstreet #include "io.h" 191c6fdbd8SKent Overstreet #include "keylist.h" 201c6fdbd8SKent Overstreet #include "quota.h" 2176426098SKent Overstreet #include "reflink.h" 221c6fdbd8SKent Overstreet #include "trace.h" 231c6fdbd8SKent Overstreet 241c6fdbd8SKent Overstreet #include <linux/aio.h> 251c6fdbd8SKent Overstreet #include <linux/backing-dev.h> 261c6fdbd8SKent Overstreet #include <linux/falloc.h> 271c6fdbd8SKent Overstreet #include <linux/migrate.h> 281c6fdbd8SKent Overstreet #include <linux/mmu_context.h> 291c6fdbd8SKent Overstreet #include <linux/pagevec.h> 309ba2eb25SKent Overstreet #include <linux/rmap.h> 311c6fdbd8SKent Overstreet #include <linux/sched/signal.h> 321c6fdbd8SKent Overstreet #include <linux/task_io_accounting_ops.h> 331c6fdbd8SKent Overstreet #include <linux/uio.h> 341c6fdbd8SKent Overstreet #include <linux/writeback.h> 351c6fdbd8SKent Overstreet 361c6fdbd8SKent Overstreet #include <trace/events/writeback.h> 371c6fdbd8SKent Overstreet 387f5e31e1SKent Overstreet static inline bool bio_full(struct bio *bio, unsigned len) 397f5e31e1SKent Overstreet { 407f5e31e1SKent Overstreet if (bio->bi_vcnt >= bio->bi_max_vecs) 417f5e31e1SKent Overstreet return true; 427f5e31e1SKent Overstreet if (bio->bi_iter.bi_size > UINT_MAX - len) 437f5e31e1SKent Overstreet return true; 447f5e31e1SKent Overstreet return false; 457f5e31e1SKent Overstreet } 467f5e31e1SKent Overstreet 47eb8e6e9cSKent Overstreet static inline struct address_space *faults_disabled_mapping(void) 48eb8e6e9cSKent Overstreet { 49eb8e6e9cSKent Overstreet return (void *) (((unsigned long) current->faults_disabled_mapping) & ~1UL); 50eb8e6e9cSKent Overstreet } 51eb8e6e9cSKent Overstreet 52eb8e6e9cSKent Overstreet static inline void set_fdm_dropped_locks(void) 53eb8e6e9cSKent Overstreet { 54eb8e6e9cSKent Overstreet current->faults_disabled_mapping = 55eb8e6e9cSKent Overstreet (void *) (((unsigned long) current->faults_disabled_mapping)|1); 56eb8e6e9cSKent Overstreet } 57eb8e6e9cSKent Overstreet 58eb8e6e9cSKent Overstreet static inline bool fdm_dropped_locks(void) 59eb8e6e9cSKent Overstreet { 60eb8e6e9cSKent Overstreet return ((unsigned long) current->faults_disabled_mapping) & 1; 61eb8e6e9cSKent Overstreet } 62eb8e6e9cSKent Overstreet 631c6fdbd8SKent Overstreet struct quota_res { 641c6fdbd8SKent Overstreet u64 sectors; 651c6fdbd8SKent Overstreet }; 661c6fdbd8SKent Overstreet 679a3df993SKent Overstreet struct bch_writepage_io { 681c6fdbd8SKent Overstreet struct bch_inode_info *inode; 691c6fdbd8SKent Overstreet 701c6fdbd8SKent Overstreet /* must be last: */ 711c6fdbd8SKent Overstreet struct bch_write_op op; 721c6fdbd8SKent Overstreet }; 731c6fdbd8SKent Overstreet 741c6fdbd8SKent Overstreet struct dio_write { 75f8f30863SKent Overstreet struct completion done; 761c6fdbd8SKent Overstreet struct kiocb *req; 77ed484030SKent Overstreet struct mm_struct *mm; 781c6fdbd8SKent Overstreet unsigned loop:1, 791c6fdbd8SKent Overstreet sync:1, 801c6fdbd8SKent Overstreet free_iov:1; 811c6fdbd8SKent Overstreet struct quota_res quota_res; 82042a1f26SKent Overstreet u64 written; 831c6fdbd8SKent Overstreet 841c6fdbd8SKent Overstreet struct iov_iter iter; 851c6fdbd8SKent Overstreet struct iovec inline_vecs[2]; 861c6fdbd8SKent Overstreet 871c6fdbd8SKent Overstreet /* must be last: */ 889a3df993SKent Overstreet struct bch_write_op op; 891c6fdbd8SKent Overstreet }; 901c6fdbd8SKent Overstreet 911c6fdbd8SKent Overstreet struct dio_read { 921c6fdbd8SKent Overstreet struct closure cl; 931c6fdbd8SKent Overstreet struct kiocb *req; 941c6fdbd8SKent Overstreet long ret; 95b4725cc1SKent Overstreet bool should_dirty; 961c6fdbd8SKent Overstreet struct bch_read_bio rbio; 971c6fdbd8SKent Overstreet }; 981c6fdbd8SKent Overstreet 991c6fdbd8SKent Overstreet /* pagecache_block must be held */ 100a023127aSKent Overstreet static noinline int write_invalidate_inode_pages_range(struct address_space *mapping, 1011c6fdbd8SKent Overstreet loff_t start, loff_t end) 1021c6fdbd8SKent Overstreet { 1031c6fdbd8SKent Overstreet int ret; 1041c6fdbd8SKent Overstreet 1051c6fdbd8SKent Overstreet /* 1061c6fdbd8SKent Overstreet * XXX: the way this is currently implemented, we can spin if a process 1071c6fdbd8SKent Overstreet * is continually redirtying a specific page 1081c6fdbd8SKent Overstreet */ 1091c6fdbd8SKent Overstreet do { 1101c6fdbd8SKent Overstreet if (!mapping->nrpages) 1111c6fdbd8SKent Overstreet return 0; 1121c6fdbd8SKent Overstreet 1131c6fdbd8SKent Overstreet ret = filemap_write_and_wait_range(mapping, start, end); 1141c6fdbd8SKent Overstreet if (ret) 1151c6fdbd8SKent Overstreet break; 1161c6fdbd8SKent Overstreet 1171c6fdbd8SKent Overstreet if (!mapping->nrpages) 1181c6fdbd8SKent Overstreet return 0; 1191c6fdbd8SKent Overstreet 1201c6fdbd8SKent Overstreet ret = invalidate_inode_pages2_range(mapping, 1211c6fdbd8SKent Overstreet start >> PAGE_SHIFT, 1221c6fdbd8SKent Overstreet end >> PAGE_SHIFT); 1231c6fdbd8SKent Overstreet } while (ret == -EBUSY); 1241c6fdbd8SKent Overstreet 1251c6fdbd8SKent Overstreet return ret; 1261c6fdbd8SKent Overstreet } 1271c6fdbd8SKent Overstreet 1281c6fdbd8SKent Overstreet /* quotas */ 1291c6fdbd8SKent Overstreet 1301c6fdbd8SKent Overstreet #ifdef CONFIG_BCACHEFS_QUOTA 1311c6fdbd8SKent Overstreet 1321c6fdbd8SKent Overstreet static void bch2_quota_reservation_put(struct bch_fs *c, 1331c6fdbd8SKent Overstreet struct bch_inode_info *inode, 1341c6fdbd8SKent Overstreet struct quota_res *res) 1351c6fdbd8SKent Overstreet { 1361c6fdbd8SKent Overstreet if (!res->sectors) 1371c6fdbd8SKent Overstreet return; 1381c6fdbd8SKent Overstreet 1391c6fdbd8SKent Overstreet mutex_lock(&inode->ei_quota_lock); 1401c6fdbd8SKent Overstreet BUG_ON(res->sectors > inode->ei_quota_reserved); 1411c6fdbd8SKent Overstreet 1421c6fdbd8SKent Overstreet bch2_quota_acct(c, inode->ei_qid, Q_SPC, 14326609b61SKent Overstreet -((s64) res->sectors), KEY_TYPE_QUOTA_PREALLOC); 1441c6fdbd8SKent Overstreet inode->ei_quota_reserved -= res->sectors; 1451c6fdbd8SKent Overstreet mutex_unlock(&inode->ei_quota_lock); 1461c6fdbd8SKent Overstreet 1471c6fdbd8SKent Overstreet res->sectors = 0; 1481c6fdbd8SKent Overstreet } 1491c6fdbd8SKent Overstreet 1501c6fdbd8SKent Overstreet static int bch2_quota_reservation_add(struct bch_fs *c, 1511c6fdbd8SKent Overstreet struct bch_inode_info *inode, 1521c6fdbd8SKent Overstreet struct quota_res *res, 1531c6fdbd8SKent Overstreet unsigned sectors, 1541c6fdbd8SKent Overstreet bool check_enospc) 1551c6fdbd8SKent Overstreet { 1561c6fdbd8SKent Overstreet int ret; 1571c6fdbd8SKent Overstreet 1581c6fdbd8SKent Overstreet mutex_lock(&inode->ei_quota_lock); 1591c6fdbd8SKent Overstreet ret = bch2_quota_acct(c, inode->ei_qid, Q_SPC, sectors, 16026609b61SKent Overstreet check_enospc ? KEY_TYPE_QUOTA_PREALLOC : KEY_TYPE_QUOTA_NOCHECK); 1611c6fdbd8SKent Overstreet if (likely(!ret)) { 1621c6fdbd8SKent Overstreet inode->ei_quota_reserved += sectors; 1631c6fdbd8SKent Overstreet res->sectors += sectors; 1641c6fdbd8SKent Overstreet } 1651c6fdbd8SKent Overstreet mutex_unlock(&inode->ei_quota_lock); 1661c6fdbd8SKent Overstreet 1671c6fdbd8SKent Overstreet return ret; 1681c6fdbd8SKent Overstreet } 1691c6fdbd8SKent Overstreet 1701c6fdbd8SKent Overstreet #else 1711c6fdbd8SKent Overstreet 1721c6fdbd8SKent Overstreet static void bch2_quota_reservation_put(struct bch_fs *c, 1731c6fdbd8SKent Overstreet struct bch_inode_info *inode, 1741c6fdbd8SKent Overstreet struct quota_res *res) 1751c6fdbd8SKent Overstreet { 1761c6fdbd8SKent Overstreet } 1771c6fdbd8SKent Overstreet 1781c6fdbd8SKent Overstreet static int bch2_quota_reservation_add(struct bch_fs *c, 1791c6fdbd8SKent Overstreet struct bch_inode_info *inode, 1801c6fdbd8SKent Overstreet struct quota_res *res, 1811c6fdbd8SKent Overstreet unsigned sectors, 1821c6fdbd8SKent Overstreet bool check_enospc) 1831c6fdbd8SKent Overstreet { 1841c6fdbd8SKent Overstreet return 0; 1851c6fdbd8SKent Overstreet } 1861c6fdbd8SKent Overstreet 1871c6fdbd8SKent Overstreet #endif 1881c6fdbd8SKent Overstreet 1891c6fdbd8SKent Overstreet /* i_size updates: */ 1901c6fdbd8SKent Overstreet 1912ea90048SKent Overstreet struct inode_new_size { 1922ea90048SKent Overstreet loff_t new_size; 1932ea90048SKent Overstreet u64 now; 1942ea90048SKent Overstreet unsigned fields; 1952ea90048SKent Overstreet }; 1962ea90048SKent Overstreet 1971c6fdbd8SKent Overstreet static int inode_set_size(struct bch_inode_info *inode, 1981c6fdbd8SKent Overstreet struct bch_inode_unpacked *bi, 1991c6fdbd8SKent Overstreet void *p) 2001c6fdbd8SKent Overstreet { 2012ea90048SKent Overstreet struct inode_new_size *s = p; 2021c6fdbd8SKent Overstreet 2032ea90048SKent Overstreet bi->bi_size = s->new_size; 2042ea90048SKent Overstreet if (s->fields & ATTR_ATIME) 2052ea90048SKent Overstreet bi->bi_atime = s->now; 2062ea90048SKent Overstreet if (s->fields & ATTR_MTIME) 2072ea90048SKent Overstreet bi->bi_mtime = s->now; 2082ea90048SKent Overstreet if (s->fields & ATTR_CTIME) 2092ea90048SKent Overstreet bi->bi_ctime = s->now; 2101c6fdbd8SKent Overstreet 2111c6fdbd8SKent Overstreet return 0; 2121c6fdbd8SKent Overstreet } 2131c6fdbd8SKent Overstreet 21476426098SKent Overstreet int __must_check bch2_write_inode_size(struct bch_fs *c, 2151c6fdbd8SKent Overstreet struct bch_inode_info *inode, 2162ea90048SKent Overstreet loff_t new_size, unsigned fields) 2171c6fdbd8SKent Overstreet { 2182ea90048SKent Overstreet struct inode_new_size s = { 2192ea90048SKent Overstreet .new_size = new_size, 2202ea90048SKent Overstreet .now = bch2_current_time(c), 2212ea90048SKent Overstreet .fields = fields, 2222ea90048SKent Overstreet }; 2232ea90048SKent Overstreet 2242ea90048SKent Overstreet return bch2_write_inode(c, inode, inode_set_size, &s, fields); 2251c6fdbd8SKent Overstreet } 2261c6fdbd8SKent Overstreet 2271c6fdbd8SKent Overstreet static void i_sectors_acct(struct bch_fs *c, struct bch_inode_info *inode, 228190fa7afSKent Overstreet struct quota_res *quota_res, s64 sectors) 2291c6fdbd8SKent Overstreet { 230190fa7afSKent Overstreet if (!sectors) 231190fa7afSKent Overstreet return; 232190fa7afSKent Overstreet 2331c6fdbd8SKent Overstreet mutex_lock(&inode->ei_quota_lock); 2341c6fdbd8SKent Overstreet #ifdef CONFIG_BCACHEFS_QUOTA 2351c6fdbd8SKent Overstreet if (quota_res && sectors > 0) { 2361c6fdbd8SKent Overstreet BUG_ON(sectors > quota_res->sectors); 2371c6fdbd8SKent Overstreet BUG_ON(sectors > inode->ei_quota_reserved); 2381c6fdbd8SKent Overstreet 2391c6fdbd8SKent Overstreet quota_res->sectors -= sectors; 2401c6fdbd8SKent Overstreet inode->ei_quota_reserved -= sectors; 2411c6fdbd8SKent Overstreet } else { 24226609b61SKent Overstreet bch2_quota_acct(c, inode->ei_qid, Q_SPC, sectors, KEY_TYPE_QUOTA_WARN); 2431c6fdbd8SKent Overstreet } 2441c6fdbd8SKent Overstreet #endif 2451c6fdbd8SKent Overstreet inode->v.i_blocks += sectors; 2461c6fdbd8SKent Overstreet mutex_unlock(&inode->ei_quota_lock); 2471c6fdbd8SKent Overstreet } 2481c6fdbd8SKent Overstreet 2491c6fdbd8SKent Overstreet /* page state: */ 2501c6fdbd8SKent Overstreet 2511c6fdbd8SKent Overstreet /* stored in page->private: */ 2521c6fdbd8SKent Overstreet 253f57a6a5dSKent Overstreet struct bch_page_sector { 254f81b648dSKent Overstreet /* Uncompressed, fully allocated replicas: */ 255f57a6a5dSKent Overstreet unsigned nr_replicas:3; 256f81b648dSKent Overstreet 257f81b648dSKent Overstreet /* Owns PAGE_SECTORS * replicas_reserved sized reservation: */ 258f57a6a5dSKent Overstreet unsigned replicas_reserved:3; 2591c6fdbd8SKent Overstreet 260f57a6a5dSKent Overstreet /* i_sectors: */ 261f57a6a5dSKent Overstreet enum { 262f57a6a5dSKent Overstreet SECTOR_UNALLOCATED, 2632ba5d38bSKent Overstreet SECTOR_RESERVED, 264f57a6a5dSKent Overstreet SECTOR_DIRTY, 265f57a6a5dSKent Overstreet SECTOR_ALLOCATED, 266f57a6a5dSKent Overstreet } state:2; 2671c6fdbd8SKent Overstreet }; 2681c6fdbd8SKent Overstreet 269f57a6a5dSKent Overstreet struct bch_page_state { 2703826ee0bSKent Overstreet spinlock_t lock; 2717f5e31e1SKent Overstreet atomic_t write_count; 272f57a6a5dSKent Overstreet struct bch_page_sector s[PAGE_SECTORS]; 273f57a6a5dSKent Overstreet }; 274f57a6a5dSKent Overstreet 275f57a6a5dSKent Overstreet static inline struct bch_page_state *__bch2_page_state(struct page *page) 2761c6fdbd8SKent Overstreet { 277f57a6a5dSKent Overstreet return page_has_private(page) 278f57a6a5dSKent Overstreet ? (struct bch_page_state *) page_private(page) 279f57a6a5dSKent Overstreet : NULL; 280f57a6a5dSKent Overstreet } 2811c6fdbd8SKent Overstreet 282f57a6a5dSKent Overstreet static inline struct bch_page_state *bch2_page_state(struct page *page) 283f57a6a5dSKent Overstreet { 284e1036a2aSKent Overstreet EBUG_ON(!PageLocked(page)); 2851c6fdbd8SKent Overstreet 286f57a6a5dSKent Overstreet return __bch2_page_state(page); 287f57a6a5dSKent Overstreet } 288f57a6a5dSKent Overstreet 289f57a6a5dSKent Overstreet /* for newly allocated pages: */ 290f57a6a5dSKent Overstreet static void __bch2_page_state_release(struct page *page) 291f57a6a5dSKent Overstreet { 29200276f9fSMatthew Wilcox (Oracle) kfree(detach_page_private(page)); 293f57a6a5dSKent Overstreet } 294f57a6a5dSKent Overstreet 295f57a6a5dSKent Overstreet static void bch2_page_state_release(struct page *page) 296f57a6a5dSKent Overstreet { 29700276f9fSMatthew Wilcox (Oracle) EBUG_ON(!PageLocked(page)); 29800276f9fSMatthew Wilcox (Oracle) __bch2_page_state_release(page); 299f57a6a5dSKent Overstreet } 300f57a6a5dSKent Overstreet 301f57a6a5dSKent Overstreet /* for newly allocated pages: */ 302f57a6a5dSKent Overstreet static struct bch_page_state *__bch2_page_state_create(struct page *page, 303f57a6a5dSKent Overstreet gfp_t gfp) 304f57a6a5dSKent Overstreet { 305f57a6a5dSKent Overstreet struct bch_page_state *s; 306f57a6a5dSKent Overstreet 307f57a6a5dSKent Overstreet s = kzalloc(sizeof(*s), GFP_NOFS|gfp); 308f57a6a5dSKent Overstreet if (!s) 309f57a6a5dSKent Overstreet return NULL; 310f57a6a5dSKent Overstreet 3113826ee0bSKent Overstreet spin_lock_init(&s->lock); 31200276f9fSMatthew Wilcox (Oracle) attach_page_private(page, s); 3131c6fdbd8SKent Overstreet return s; 3141c6fdbd8SKent Overstreet } 3151c6fdbd8SKent Overstreet 316f57a6a5dSKent Overstreet static struct bch_page_state *bch2_page_state_create(struct page *page, 317f57a6a5dSKent Overstreet gfp_t gfp) 318f57a6a5dSKent Overstreet { 319f57a6a5dSKent Overstreet return bch2_page_state(page) ?: __bch2_page_state_create(page, gfp); 320f57a6a5dSKent Overstreet } 321f57a6a5dSKent Overstreet 322e1036a2aSKent Overstreet static inline unsigned inode_nr_replicas(struct bch_fs *c, struct bch_inode_info *inode) 323e1036a2aSKent Overstreet { 324e1036a2aSKent Overstreet /* XXX: this should not be open coded */ 325e1036a2aSKent Overstreet return inode->ei_inode.bi_data_replicas 326e1036a2aSKent Overstreet ? inode->ei_inode.bi_data_replicas - 1 327e1036a2aSKent Overstreet : c->opts.data_replicas; 328e1036a2aSKent Overstreet } 329e1036a2aSKent Overstreet 330f57a6a5dSKent Overstreet static inline unsigned sectors_to_reserve(struct bch_page_sector *s, 331f57a6a5dSKent Overstreet unsigned nr_replicas) 332f57a6a5dSKent Overstreet { 333f57a6a5dSKent Overstreet return max(0, (int) nr_replicas - 334f57a6a5dSKent Overstreet s->nr_replicas - 335f57a6a5dSKent Overstreet s->replicas_reserved); 336f57a6a5dSKent Overstreet } 337f57a6a5dSKent Overstreet 338f57a6a5dSKent Overstreet static int bch2_get_page_disk_reservation(struct bch_fs *c, 339f57a6a5dSKent Overstreet struct bch_inode_info *inode, 3401c6fdbd8SKent Overstreet struct page *page, bool check_enospc) 3411c6fdbd8SKent Overstreet { 342f57a6a5dSKent Overstreet struct bch_page_state *s = bch2_page_state_create(page, 0); 343e1036a2aSKent Overstreet unsigned nr_replicas = inode_nr_replicas(c, inode); 344f57a6a5dSKent Overstreet struct disk_reservation disk_res = { 0 }; 345f57a6a5dSKent Overstreet unsigned i, disk_res_sectors = 0; 346f81b648dSKent Overstreet int ret; 3471c6fdbd8SKent Overstreet 348f57a6a5dSKent Overstreet if (!s) 349f57a6a5dSKent Overstreet return -ENOMEM; 3501c6fdbd8SKent Overstreet 351f57a6a5dSKent Overstreet for (i = 0; i < ARRAY_SIZE(s->s); i++) 352f57a6a5dSKent Overstreet disk_res_sectors += sectors_to_reserve(&s->s[i], nr_replicas); 353f57a6a5dSKent Overstreet 354f57a6a5dSKent Overstreet if (!disk_res_sectors) 355f57a6a5dSKent Overstreet return 0; 356f57a6a5dSKent Overstreet 357f57a6a5dSKent Overstreet ret = bch2_disk_reservation_get(c, &disk_res, 358f57a6a5dSKent Overstreet disk_res_sectors, 1, 359f57a6a5dSKent Overstreet !check_enospc 360f57a6a5dSKent Overstreet ? BCH_DISK_RESERVATION_NOFAIL 361f57a6a5dSKent Overstreet : 0); 3621c6fdbd8SKent Overstreet if (unlikely(ret)) 363f81b648dSKent Overstreet return ret; 364f81b648dSKent Overstreet 365f57a6a5dSKent Overstreet for (i = 0; i < ARRAY_SIZE(s->s); i++) 366f57a6a5dSKent Overstreet s->s[i].replicas_reserved += 367f57a6a5dSKent Overstreet sectors_to_reserve(&s->s[i], nr_replicas); 368f57a6a5dSKent Overstreet 369f57a6a5dSKent Overstreet return 0; 3701c6fdbd8SKent Overstreet } 3711c6fdbd8SKent Overstreet 372d1542e03SKent Overstreet struct bch2_page_reservation { 373d1542e03SKent Overstreet struct disk_reservation disk; 374d1542e03SKent Overstreet struct quota_res quota; 375d1542e03SKent Overstreet }; 376d1542e03SKent Overstreet 377d1542e03SKent Overstreet static void bch2_page_reservation_init(struct bch_fs *c, 378f57a6a5dSKent Overstreet struct bch_inode_info *inode, 379d1542e03SKent Overstreet struct bch2_page_reservation *res) 380d1542e03SKent Overstreet { 381d1542e03SKent Overstreet memset(res, 0, sizeof(*res)); 382d1542e03SKent Overstreet 383d1542e03SKent Overstreet res->disk.nr_replicas = inode_nr_replicas(c, inode); 384d1542e03SKent Overstreet } 385d1542e03SKent Overstreet 386d1542e03SKent Overstreet static void bch2_page_reservation_put(struct bch_fs *c, 387d1542e03SKent Overstreet struct bch_inode_info *inode, 388d1542e03SKent Overstreet struct bch2_page_reservation *res) 389d1542e03SKent Overstreet { 390d1542e03SKent Overstreet bch2_disk_reservation_put(c, &res->disk); 391d1542e03SKent Overstreet bch2_quota_reservation_put(c, inode, &res->quota); 392d1542e03SKent Overstreet } 393d1542e03SKent Overstreet 394d1542e03SKent Overstreet static int bch2_page_reservation_get(struct bch_fs *c, 395d1542e03SKent Overstreet struct bch_inode_info *inode, struct page *page, 396d1542e03SKent Overstreet struct bch2_page_reservation *res, 397d1542e03SKent Overstreet unsigned offset, unsigned len, bool check_enospc) 398f57a6a5dSKent Overstreet { 399f57a6a5dSKent Overstreet struct bch_page_state *s = bch2_page_state_create(page, 0); 400d1542e03SKent Overstreet unsigned i, disk_sectors = 0, quota_sectors = 0; 401f57a6a5dSKent Overstreet int ret; 402f57a6a5dSKent Overstreet 403f57a6a5dSKent Overstreet if (!s) 404f57a6a5dSKent Overstreet return -ENOMEM; 405f57a6a5dSKent Overstreet 4064b0a66d5SKent Overstreet for (i = round_down(offset, block_bytes(c)) >> 9; 4074b0a66d5SKent Overstreet i < round_up(offset + len, block_bytes(c)) >> 9; 408d1542e03SKent Overstreet i++) { 409d1542e03SKent Overstreet disk_sectors += sectors_to_reserve(&s->s[i], 410d1542e03SKent Overstreet res->disk.nr_replicas); 411d1542e03SKent Overstreet quota_sectors += s->s[i].state == SECTOR_UNALLOCATED; 4121c6fdbd8SKent Overstreet } 4131c6fdbd8SKent Overstreet 414d1542e03SKent Overstreet if (disk_sectors) { 415d1542e03SKent Overstreet ret = bch2_disk_reservation_add(c, &res->disk, 416d1542e03SKent Overstreet disk_sectors, 417d1542e03SKent Overstreet !check_enospc 418d1542e03SKent Overstreet ? BCH_DISK_RESERVATION_NOFAIL 419d1542e03SKent Overstreet : 0); 420d1542e03SKent Overstreet if (unlikely(ret)) 421d1542e03SKent Overstreet return ret; 422d1542e03SKent Overstreet } 423d1542e03SKent Overstreet 424d1542e03SKent Overstreet if (quota_sectors) { 425d1542e03SKent Overstreet ret = bch2_quota_reservation_add(c, inode, &res->quota, 426d1542e03SKent Overstreet quota_sectors, 427d1542e03SKent Overstreet check_enospc); 428d1542e03SKent Overstreet if (unlikely(ret)) { 429d1542e03SKent Overstreet struct disk_reservation tmp = { 430d1542e03SKent Overstreet .sectors = disk_sectors 431d1542e03SKent Overstreet }; 432d1542e03SKent Overstreet 433d1542e03SKent Overstreet bch2_disk_reservation_put(c, &tmp); 434d1542e03SKent Overstreet res->disk.sectors -= disk_sectors; 435d1542e03SKent Overstreet return ret; 436d1542e03SKent Overstreet } 437d1542e03SKent Overstreet } 438d1542e03SKent Overstreet 439d1542e03SKent Overstreet return 0; 440f57a6a5dSKent Overstreet } 441f57a6a5dSKent Overstreet 4421c6fdbd8SKent Overstreet static void bch2_clear_page_bits(struct page *page) 4431c6fdbd8SKent Overstreet { 4441c6fdbd8SKent Overstreet struct bch_inode_info *inode = to_bch_ei(page->mapping->host); 4451c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 446f57a6a5dSKent Overstreet struct bch_page_state *s = bch2_page_state(page); 447d1542e03SKent Overstreet struct disk_reservation disk_res = { 0 }; 448f57a6a5dSKent Overstreet int i, dirty_sectors = 0; 4491c6fdbd8SKent Overstreet 450f57a6a5dSKent Overstreet if (!s) 4511c6fdbd8SKent Overstreet return; 4521c6fdbd8SKent Overstreet 4533826ee0bSKent Overstreet EBUG_ON(!PageLocked(page)); 4543826ee0bSKent Overstreet EBUG_ON(PageWriteback(page)); 4553826ee0bSKent Overstreet 456f57a6a5dSKent Overstreet for (i = 0; i < ARRAY_SIZE(s->s); i++) { 457d1542e03SKent Overstreet disk_res.sectors += s->s[i].replicas_reserved; 458d1542e03SKent Overstreet s->s[i].replicas_reserved = 0; 459d1542e03SKent Overstreet 460f57a6a5dSKent Overstreet if (s->s[i].state == SECTOR_DIRTY) { 461f57a6a5dSKent Overstreet dirty_sectors++; 462f57a6a5dSKent Overstreet s->s[i].state = SECTOR_UNALLOCATED; 463f57a6a5dSKent Overstreet } 464f57a6a5dSKent Overstreet } 465adfcfaf0SKent Overstreet 466d1542e03SKent Overstreet bch2_disk_reservation_put(c, &disk_res); 467d1542e03SKent Overstreet 468f57a6a5dSKent Overstreet if (dirty_sectors) 469f57a6a5dSKent Overstreet i_sectors_acct(c, inode, NULL, -dirty_sectors); 470adfcfaf0SKent Overstreet 471f57a6a5dSKent Overstreet bch2_page_state_release(page); 4721c6fdbd8SKent Overstreet } 4731c6fdbd8SKent Overstreet 474d1542e03SKent Overstreet static void bch2_set_page_dirty(struct bch_fs *c, 475d1542e03SKent Overstreet struct bch_inode_info *inode, struct page *page, 476d1542e03SKent Overstreet struct bch2_page_reservation *res, 477d1542e03SKent Overstreet unsigned offset, unsigned len) 4781c6fdbd8SKent Overstreet { 479d1542e03SKent Overstreet struct bch_page_state *s = bch2_page_state(page); 480f57a6a5dSKent Overstreet unsigned i, dirty_sectors = 0; 4811c6fdbd8SKent Overstreet 482877dfb34SKent Overstreet WARN_ON((u64) page_offset(page) + offset + len > 483877dfb34SKent Overstreet round_up((u64) i_size_read(&inode->v), block_bytes(c))); 484fb472ac5SKent Overstreet 4853826ee0bSKent Overstreet spin_lock(&s->lock); 4863826ee0bSKent Overstreet 4874b0a66d5SKent Overstreet for (i = round_down(offset, block_bytes(c)) >> 9; 4884b0a66d5SKent Overstreet i < round_up(offset + len, block_bytes(c)) >> 9; 489d1542e03SKent Overstreet i++) { 490d1542e03SKent Overstreet unsigned sectors = sectors_to_reserve(&s->s[i], 491d1542e03SKent Overstreet res->disk.nr_replicas); 4921c6fdbd8SKent Overstreet 493406d6d5aSKent Overstreet /* 494406d6d5aSKent Overstreet * This can happen if we race with the error path in 495406d6d5aSKent Overstreet * bch2_writepage_io_done(): 496406d6d5aSKent Overstreet */ 497406d6d5aSKent Overstreet sectors = min_t(unsigned, sectors, res->disk.sectors); 498406d6d5aSKent Overstreet 499d1542e03SKent Overstreet s->s[i].replicas_reserved += sectors; 500d1542e03SKent Overstreet res->disk.sectors -= sectors; 501adfcfaf0SKent Overstreet 5022ba5d38bSKent Overstreet if (s->s[i].state == SECTOR_UNALLOCATED) 503f57a6a5dSKent Overstreet dirty_sectors++; 5042ba5d38bSKent Overstreet 5052ba5d38bSKent Overstreet s->s[i].state = max_t(unsigned, s->s[i].state, SECTOR_DIRTY); 506f57a6a5dSKent Overstreet } 507f57a6a5dSKent Overstreet 5083826ee0bSKent Overstreet spin_unlock(&s->lock); 5093826ee0bSKent Overstreet 510f57a6a5dSKent Overstreet if (dirty_sectors) 511d1542e03SKent Overstreet i_sectors_acct(c, inode, &res->quota, dirty_sectors); 5121c6fdbd8SKent Overstreet 513d1542e03SKent Overstreet if (!PageDirty(page)) 514d1542e03SKent Overstreet filemap_dirty_folio(inode->v.i_mapping, page_folio(page)); 5151c6fdbd8SKent Overstreet } 5161c6fdbd8SKent Overstreet 5171c6fdbd8SKent Overstreet vm_fault_t bch2_page_fault(struct vm_fault *vmf) 5181c6fdbd8SKent Overstreet { 5191c6fdbd8SKent Overstreet struct file *file = vmf->vma->vm_file; 520eb8e6e9cSKent Overstreet struct address_space *mapping = file->f_mapping; 521eb8e6e9cSKent Overstreet struct address_space *fdm = faults_disabled_mapping(); 5221c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 5231c6fdbd8SKent Overstreet int ret; 5241c6fdbd8SKent Overstreet 525eb8e6e9cSKent Overstreet if (fdm == mapping) 526eb8e6e9cSKent Overstreet return VM_FAULT_SIGBUS; 527eb8e6e9cSKent Overstreet 528eb8e6e9cSKent Overstreet /* Lock ordering: */ 529eb8e6e9cSKent Overstreet if (fdm > mapping) { 530eb8e6e9cSKent Overstreet struct bch_inode_info *fdm_host = to_bch_ei(fdm->host); 531eb8e6e9cSKent Overstreet 532eb8e6e9cSKent Overstreet if (bch2_pagecache_add_tryget(&inode->ei_pagecache_lock)) 533eb8e6e9cSKent Overstreet goto got_lock; 534eb8e6e9cSKent Overstreet 535eb8e6e9cSKent Overstreet bch2_pagecache_block_put(&fdm_host->ei_pagecache_lock); 536eb8e6e9cSKent Overstreet 5371c6fdbd8SKent Overstreet bch2_pagecache_add_get(&inode->ei_pagecache_lock); 538eb8e6e9cSKent Overstreet bch2_pagecache_add_put(&inode->ei_pagecache_lock); 539eb8e6e9cSKent Overstreet 540eb8e6e9cSKent Overstreet bch2_pagecache_block_get(&fdm_host->ei_pagecache_lock); 541eb8e6e9cSKent Overstreet 542eb8e6e9cSKent Overstreet /* Signal that lock has been dropped: */ 543eb8e6e9cSKent Overstreet set_fdm_dropped_locks(); 544eb8e6e9cSKent Overstreet return VM_FAULT_SIGBUS; 545eb8e6e9cSKent Overstreet } 546eb8e6e9cSKent Overstreet 547eb8e6e9cSKent Overstreet bch2_pagecache_add_get(&inode->ei_pagecache_lock); 548eb8e6e9cSKent Overstreet got_lock: 5491c6fdbd8SKent Overstreet ret = filemap_fault(vmf); 5501c6fdbd8SKent Overstreet bch2_pagecache_add_put(&inode->ei_pagecache_lock); 5511c6fdbd8SKent Overstreet 5521c6fdbd8SKent Overstreet return ret; 5531c6fdbd8SKent Overstreet } 5541c6fdbd8SKent Overstreet 5551c6fdbd8SKent Overstreet vm_fault_t bch2_page_mkwrite(struct vm_fault *vmf) 5561c6fdbd8SKent Overstreet { 5571c6fdbd8SKent Overstreet struct page *page = vmf->page; 5581c6fdbd8SKent Overstreet struct file *file = vmf->vma->vm_file; 5591c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 5601c6fdbd8SKent Overstreet struct address_space *mapping = file->f_mapping; 5611c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 562d1542e03SKent Overstreet struct bch2_page_reservation res; 5636cc3535dSKent Overstreet unsigned len; 5646cc3535dSKent Overstreet loff_t isize; 5651c6fdbd8SKent Overstreet int ret = VM_FAULT_LOCKED; 5661c6fdbd8SKent Overstreet 567d1542e03SKent Overstreet bch2_page_reservation_init(c, inode, &res); 568d1542e03SKent Overstreet 5691c6fdbd8SKent Overstreet sb_start_pagefault(inode->v.i_sb); 5701c6fdbd8SKent Overstreet file_update_time(file); 5711c6fdbd8SKent Overstreet 5721c6fdbd8SKent Overstreet /* 5731c6fdbd8SKent Overstreet * Not strictly necessary, but helps avoid dio writes livelocking in 5741c6fdbd8SKent Overstreet * write_invalidate_inode_pages_range() - can drop this if/when we get 5751c6fdbd8SKent Overstreet * a write_invalidate_inode_pages_range() that works without dropping 5761c6fdbd8SKent Overstreet * page lock before invalidating page 5771c6fdbd8SKent Overstreet */ 5781c6fdbd8SKent Overstreet bch2_pagecache_add_get(&inode->ei_pagecache_lock); 5791c6fdbd8SKent Overstreet 5801c6fdbd8SKent Overstreet lock_page(page); 5816cc3535dSKent Overstreet isize = i_size_read(&inode->v); 5826cc3535dSKent Overstreet 5836cc3535dSKent Overstreet if (page->mapping != mapping || page_offset(page) >= isize) { 5841c6fdbd8SKent Overstreet unlock_page(page); 5851c6fdbd8SKent Overstreet ret = VM_FAULT_NOPAGE; 5861c6fdbd8SKent Overstreet goto out; 5871c6fdbd8SKent Overstreet } 5881c6fdbd8SKent Overstreet 589d55460bbSKent Overstreet len = min_t(loff_t, PAGE_SIZE, isize - page_offset(page)); 5906cc3535dSKent Overstreet 5916cc3535dSKent Overstreet if (bch2_page_reservation_get(c, inode, page, &res, 0, len, true)) { 5921c6fdbd8SKent Overstreet unlock_page(page); 5931c6fdbd8SKent Overstreet ret = VM_FAULT_SIGBUS; 5941c6fdbd8SKent Overstreet goto out; 5951c6fdbd8SKent Overstreet } 5961c6fdbd8SKent Overstreet 5976cc3535dSKent Overstreet bch2_set_page_dirty(c, inode, page, &res, 0, len); 5981b783a69SKent Overstreet bch2_page_reservation_put(c, inode, &res); 5991b783a69SKent Overstreet 6001c6fdbd8SKent Overstreet wait_for_stable_page(page); 6011c6fdbd8SKent Overstreet out: 6021c6fdbd8SKent Overstreet bch2_pagecache_add_put(&inode->ei_pagecache_lock); 6031c6fdbd8SKent Overstreet sb_end_pagefault(inode->v.i_sb); 604d1542e03SKent Overstreet 6051c6fdbd8SKent Overstreet return ret; 6061c6fdbd8SKent Overstreet } 6071c6fdbd8SKent Overstreet 6081c6fdbd8SKent Overstreet void bch2_invalidate_folio(struct folio *folio, size_t offset, size_t length) 6091c6fdbd8SKent Overstreet { 6101c6fdbd8SKent Overstreet if (offset || length < folio_size(folio)) 6111c6fdbd8SKent Overstreet return; 6121c6fdbd8SKent Overstreet 6131c6fdbd8SKent Overstreet bch2_clear_page_bits(&folio->page); 6141c6fdbd8SKent Overstreet } 6151c6fdbd8SKent Overstreet 6161c6fdbd8SKent Overstreet bool bch2_release_folio(struct folio *folio, gfp_t gfp_mask) 6171c6fdbd8SKent Overstreet { 618a6d90385SKent Overstreet if (folio_test_dirty(folio) || folio_test_writeback(folio)) 6191c6fdbd8SKent Overstreet return false; 6201c6fdbd8SKent Overstreet 6211c6fdbd8SKent Overstreet bch2_clear_page_bits(&folio->page); 6221c6fdbd8SKent Overstreet return true; 6231c6fdbd8SKent Overstreet } 6241c6fdbd8SKent Overstreet 6251c6fdbd8SKent Overstreet /* readpage(s): */ 6261c6fdbd8SKent Overstreet 6271c6fdbd8SKent Overstreet static void bch2_readpages_end_io(struct bio *bio) 6281c6fdbd8SKent Overstreet { 6291c6fdbd8SKent Overstreet struct bvec_iter_all iter; 6301c6fdbd8SKent Overstreet struct bio_vec *bv; 6311c6fdbd8SKent Overstreet 6321c6fdbd8SKent Overstreet bio_for_each_segment_all(bv, bio, iter) { 6331c6fdbd8SKent Overstreet struct page *page = bv->bv_page; 6341c6fdbd8SKent Overstreet 6351c6fdbd8SKent Overstreet if (!bio->bi_status) { 6361c6fdbd8SKent Overstreet SetPageUptodate(page); 6371c6fdbd8SKent Overstreet } else { 6381c6fdbd8SKent Overstreet ClearPageUptodate(page); 6391c6fdbd8SKent Overstreet SetPageError(page); 6401c6fdbd8SKent Overstreet } 6411c6fdbd8SKent Overstreet unlock_page(page); 6421c6fdbd8SKent Overstreet } 6431c6fdbd8SKent Overstreet 6441c6fdbd8SKent Overstreet bio_put(bio); 6451c6fdbd8SKent Overstreet } 6461c6fdbd8SKent Overstreet 6471c6fdbd8SKent Overstreet struct readpages_iter { 6481c6fdbd8SKent Overstreet struct address_space *mapping; 6491c6fdbd8SKent Overstreet struct page **pages; 6501c6fdbd8SKent Overstreet unsigned nr_pages; 6511c6fdbd8SKent Overstreet unsigned idx; 6521c6fdbd8SKent Overstreet pgoff_t offset; 6531c6fdbd8SKent Overstreet }; 6541c6fdbd8SKent Overstreet 6551c6fdbd8SKent Overstreet static int readpages_iter_init(struct readpages_iter *iter, 6561c6fdbd8SKent Overstreet struct readahead_control *ractl) 6571c6fdbd8SKent Overstreet { 6581c6fdbd8SKent Overstreet unsigned i, nr_pages = readahead_count(ractl); 6591c6fdbd8SKent Overstreet 6601c6fdbd8SKent Overstreet memset(iter, 0, sizeof(*iter)); 6611c6fdbd8SKent Overstreet 6621c6fdbd8SKent Overstreet iter->mapping = ractl->mapping; 6631c6fdbd8SKent Overstreet iter->offset = readahead_index(ractl); 6641c6fdbd8SKent Overstreet iter->nr_pages = nr_pages; 6651c6fdbd8SKent Overstreet 6661c6fdbd8SKent Overstreet iter->pages = kmalloc_array(nr_pages, sizeof(struct page *), GFP_NOFS); 6671c6fdbd8SKent Overstreet if (!iter->pages) 6681c6fdbd8SKent Overstreet return -ENOMEM; 6691c6fdbd8SKent Overstreet 67089931472SKent Overstreet nr_pages = __readahead_batch(ractl, iter->pages, nr_pages); 6711c6fdbd8SKent Overstreet for (i = 0; i < nr_pages; i++) { 672f57a6a5dSKent Overstreet __bch2_page_state_create(iter->pages[i], __GFP_NOFAIL); 6731c6fdbd8SKent Overstreet put_page(iter->pages[i]); 6741c6fdbd8SKent Overstreet } 6751c6fdbd8SKent Overstreet 6761c6fdbd8SKent Overstreet return 0; 6771c6fdbd8SKent Overstreet } 6781c6fdbd8SKent Overstreet 6791c6fdbd8SKent Overstreet static inline struct page *readpage_iter_next(struct readpages_iter *iter) 6801c6fdbd8SKent Overstreet { 6811c6fdbd8SKent Overstreet if (iter->idx >= iter->nr_pages) 6821c6fdbd8SKent Overstreet return NULL; 6831c6fdbd8SKent Overstreet 6841c6fdbd8SKent Overstreet EBUG_ON(iter->pages[iter->idx]->index != iter->offset + iter->idx); 6851c6fdbd8SKent Overstreet 6861c6fdbd8SKent Overstreet return iter->pages[iter->idx]; 6871c6fdbd8SKent Overstreet } 6881c6fdbd8SKent Overstreet 6891c6fdbd8SKent Overstreet static void bch2_add_page_sectors(struct bio *bio, struct bkey_s_c k) 6901c6fdbd8SKent Overstreet { 6911c6fdbd8SKent Overstreet struct bvec_iter iter; 6921c6fdbd8SKent Overstreet struct bio_vec bv; 69376426098SKent Overstreet unsigned nr_ptrs = k.k->type == KEY_TYPE_reflink_v 6944de77495SKent Overstreet ? 0 : bch2_bkey_nr_ptrs_fully_allocated(k); 6952ba5d38bSKent Overstreet unsigned state = k.k->type == KEY_TYPE_reservation 6962ba5d38bSKent Overstreet ? SECTOR_RESERVED 6972ba5d38bSKent Overstreet : SECTOR_ALLOCATED; 6981c6fdbd8SKent Overstreet 6991c6fdbd8SKent Overstreet bio_for_each_segment(bv, bio, iter) { 700f57a6a5dSKent Overstreet struct bch_page_state *s = bch2_page_state(bv.bv_page); 701f57a6a5dSKent Overstreet unsigned i; 702f81b648dSKent Overstreet 703f57a6a5dSKent Overstreet for (i = bv.bv_offset >> 9; 704f57a6a5dSKent Overstreet i < (bv.bv_offset + bv.bv_len) >> 9; 705f57a6a5dSKent Overstreet i++) { 706f57a6a5dSKent Overstreet s->s[i].nr_replicas = nr_ptrs; 7072ba5d38bSKent Overstreet s->s[i].state = state; 708f57a6a5dSKent Overstreet } 7091c6fdbd8SKent Overstreet } 7101c6fdbd8SKent Overstreet } 7111c6fdbd8SKent Overstreet 71235189e09SKent Overstreet static bool extent_partial_reads_expensive(struct bkey_s_c k) 71335189e09SKent Overstreet { 71435189e09SKent Overstreet struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k); 71535189e09SKent Overstreet struct bch_extent_crc_unpacked crc; 71635189e09SKent Overstreet const union bch_extent_entry *i; 71735189e09SKent Overstreet 71835189e09SKent Overstreet bkey_for_each_crc(k.k, ptrs, crc, i) 71935189e09SKent Overstreet if (crc.csum_type || crc.compression_type) 72035189e09SKent Overstreet return true; 72135189e09SKent Overstreet return false; 72235189e09SKent Overstreet } 72335189e09SKent Overstreet 7241c6fdbd8SKent Overstreet static void readpage_bio_extend(struct readpages_iter *iter, 72576426098SKent Overstreet struct bio *bio, 72676426098SKent Overstreet unsigned sectors_this_extent, 7271c6fdbd8SKent Overstreet bool get_more) 7281c6fdbd8SKent Overstreet { 72976426098SKent Overstreet while (bio_sectors(bio) < sectors_this_extent && 7301c6fdbd8SKent Overstreet bio->bi_vcnt < bio->bi_max_vecs) { 7311c6fdbd8SKent Overstreet pgoff_t page_offset = bio_end_sector(bio) >> PAGE_SECTOR_SHIFT; 7321c6fdbd8SKent Overstreet struct page *page = readpage_iter_next(iter); 7331c6fdbd8SKent Overstreet int ret; 7341c6fdbd8SKent Overstreet 7351c6fdbd8SKent Overstreet if (page) { 7361c6fdbd8SKent Overstreet if (iter->offset + iter->idx != page_offset) 7371c6fdbd8SKent Overstreet break; 7381c6fdbd8SKent Overstreet 7391c6fdbd8SKent Overstreet iter->idx++; 7401c6fdbd8SKent Overstreet } else { 7411c6fdbd8SKent Overstreet if (!get_more) 7421c6fdbd8SKent Overstreet break; 7431c6fdbd8SKent Overstreet 7441c6fdbd8SKent Overstreet page = xa_load(&iter->mapping->i_pages, page_offset); 7451c6fdbd8SKent Overstreet if (page && !xa_is_value(page)) 7461c6fdbd8SKent Overstreet break; 7471c6fdbd8SKent Overstreet 7481c6fdbd8SKent Overstreet page = __page_cache_alloc(readahead_gfp_mask(iter->mapping)); 7491c6fdbd8SKent Overstreet if (!page) 7501c6fdbd8SKent Overstreet break; 7511c6fdbd8SKent Overstreet 752f57a6a5dSKent Overstreet if (!__bch2_page_state_create(page, 0)) { 753f57a6a5dSKent Overstreet put_page(page); 754f57a6a5dSKent Overstreet break; 755f57a6a5dSKent Overstreet } 7561c6fdbd8SKent Overstreet 7571c6fdbd8SKent Overstreet ret = add_to_page_cache_lru(page, iter->mapping, 7581c6fdbd8SKent Overstreet page_offset, GFP_NOFS); 7591c6fdbd8SKent Overstreet if (ret) { 760f57a6a5dSKent Overstreet __bch2_page_state_release(page); 7611c6fdbd8SKent Overstreet put_page(page); 7621c6fdbd8SKent Overstreet break; 7631c6fdbd8SKent Overstreet } 7641c6fdbd8SKent Overstreet 7651c6fdbd8SKent Overstreet put_page(page); 7661c6fdbd8SKent Overstreet } 7671c6fdbd8SKent Overstreet 768885678f6SKent Overstreet BUG_ON(!bio_add_page(bio, page, PAGE_SIZE, 0)); 7691c6fdbd8SKent Overstreet } 7701c6fdbd8SKent Overstreet } 7711c6fdbd8SKent Overstreet 7728c6d298aSKent Overstreet static void bchfs_read(struct btree_trans *trans, 7738c6d298aSKent Overstreet struct bch_read_bio *rbio, 7748c6d298aSKent Overstreet subvol_inum inum, 7751c6fdbd8SKent Overstreet struct readpages_iter *readpages_iter) 7761c6fdbd8SKent Overstreet { 7770f238367SKent Overstreet struct bch_fs *c = trans->c; 7788c6d298aSKent Overstreet struct btree_iter iter; 77907a1006aSKent Overstreet struct bkey_buf sk; 7801c6fdbd8SKent Overstreet int flags = BCH_READ_RETRY_IF_STALE| 7811c6fdbd8SKent Overstreet BCH_READ_MAY_PROMOTE; 7828c6d298aSKent Overstreet u32 snapshot; 78376426098SKent Overstreet int ret = 0; 7841c6fdbd8SKent Overstreet 7851c6fdbd8SKent Overstreet rbio->c = c; 7861c6fdbd8SKent Overstreet rbio->start_time = local_clock(); 7878c6d298aSKent Overstreet rbio->subvol = inum.subvol; 78835189e09SKent Overstreet 78907a1006aSKent Overstreet bch2_bkey_buf_init(&sk); 79076426098SKent Overstreet retry: 791700c25b3SKent Overstreet bch2_trans_begin(trans); 7928c6d298aSKent Overstreet iter = (struct btree_iter) { NULL }; 793700c25b3SKent Overstreet 7948c6d298aSKent Overstreet ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot); 7958c6d298aSKent Overstreet if (ret) 7968c6d298aSKent Overstreet goto err; 7978c6d298aSKent Overstreet 7988c6d298aSKent Overstreet bch2_trans_iter_init(trans, &iter, BTREE_ID_extents, 7998c6d298aSKent Overstreet SPOS(inum.inum, rbio->bio.bi_iter.bi_sector, snapshot), 8008c6d298aSKent Overstreet BTREE_ITER_SLOTS|BTREE_ITER_FILTER_SNAPSHOTS); 8011c6fdbd8SKent Overstreet while (1) { 8021c6fdbd8SKent Overstreet struct bkey_s_c k; 80376426098SKent Overstreet unsigned bytes, sectors, offset_into_extent; 8045ff75ccbSKent Overstreet enum btree_id data_btree = BTREE_ID_extents; 8051c6fdbd8SKent Overstreet 8063737e0ddSKent Overstreet /* 8073737e0ddSKent Overstreet * read_extent -> io_time_reset may cause a transaction restart 8083737e0ddSKent Overstreet * without returning an error, we need to check for that here: 8093737e0ddSKent Overstreet */ 8103737e0ddSKent Overstreet if (!bch2_trans_relock(trans)) { 8113737e0ddSKent Overstreet ret = -EINTR; 8123737e0ddSKent Overstreet break; 8133737e0ddSKent Overstreet } 8143737e0ddSKent Overstreet 8158c6d298aSKent Overstreet bch2_btree_iter_set_pos(&iter, 8168c6d298aSKent Overstreet POS(inum.inum, rbio->bio.bi_iter.bi_sector)); 8171c6fdbd8SKent Overstreet 8188c6d298aSKent Overstreet k = bch2_btree_iter_peek_slot(&iter); 81976426098SKent Overstreet ret = bkey_err(k); 82076426098SKent Overstreet if (ret) 82176426098SKent Overstreet break; 8221c6fdbd8SKent Overstreet 8238c6d298aSKent Overstreet offset_into_extent = iter.pos.offset - 82406ed8558SKent Overstreet bkey_start_offset(k.k); 82576426098SKent Overstreet sectors = k.k->size - offset_into_extent; 82676426098SKent Overstreet 82707a1006aSKent Overstreet bch2_bkey_buf_reassemble(&sk, c, k); 82813dcd4abSKent Overstreet 8295ff75ccbSKent Overstreet ret = bch2_read_indirect_extent(trans, &data_btree, 83022d8a33dSYuxuan Shui &offset_into_extent, &sk); 83176426098SKent Overstreet if (ret) 83276426098SKent Overstreet break; 83376426098SKent Overstreet 83413dcd4abSKent Overstreet k = bkey_i_to_s_c(sk.k); 83513dcd4abSKent Overstreet 83676426098SKent Overstreet sectors = min(sectors, k.k->size - offset_into_extent); 83776426098SKent Overstreet 83876426098SKent Overstreet bch2_trans_unlock(trans); 83906ed8558SKent Overstreet 84035189e09SKent Overstreet if (readpages_iter) 84135189e09SKent Overstreet readpage_bio_extend(readpages_iter, &rbio->bio, sectors, 84235189e09SKent Overstreet extent_partial_reads_expensive(k)); 8431c6fdbd8SKent Overstreet 84476426098SKent Overstreet bytes = min(sectors, bio_sectors(&rbio->bio)) << 9; 84506ed8558SKent Overstreet swap(rbio->bio.bi_iter.bi_size, bytes); 8461c6fdbd8SKent Overstreet 84706ed8558SKent Overstreet if (rbio->bio.bi_iter.bi_size == bytes) 8481c6fdbd8SKent Overstreet flags |= BCH_READ_LAST_FRAGMENT; 8491c6fdbd8SKent Overstreet 8501c6fdbd8SKent Overstreet if (bkey_extent_is_allocation(k.k)) 85106ed8558SKent Overstreet bch2_add_page_sectors(&rbio->bio, k); 8521c6fdbd8SKent Overstreet 8538c6d298aSKent Overstreet bch2_read_extent(trans, rbio, iter.pos, 8545ff75ccbSKent Overstreet data_btree, k, offset_into_extent, flags); 8551c6fdbd8SKent Overstreet 8561c6fdbd8SKent Overstreet if (flags & BCH_READ_LAST_FRAGMENT) 85735189e09SKent Overstreet break; 8581c6fdbd8SKent Overstreet 85906ed8558SKent Overstreet swap(rbio->bio.bi_iter.bi_size, bytes); 86006ed8558SKent Overstreet bio_advance(&rbio->bio, bytes); 8611c6fdbd8SKent Overstreet } 8628c6d298aSKent Overstreet err: 8638c6d298aSKent Overstreet bch2_trans_iter_exit(trans, &iter); 86476426098SKent Overstreet 86576426098SKent Overstreet if (ret == -EINTR) 86676426098SKent Overstreet goto retry; 86776426098SKent Overstreet 86835189e09SKent Overstreet if (ret) { 8698c6d298aSKent Overstreet bch_err_inum_ratelimited(c, inum.inum, 8700fefe8d8SKent Overstreet "read error %i from btree lookup", ret); 8710fefe8d8SKent Overstreet rbio->bio.bi_status = BLK_STS_IOERR; 87276426098SKent Overstreet bio_endio(&rbio->bio); 8731c6fdbd8SKent Overstreet } 8741c6fdbd8SKent Overstreet 87507a1006aSKent Overstreet bch2_bkey_buf_exit(&sk, c); 87635189e09SKent Overstreet } 87735189e09SKent Overstreet 8781c6fdbd8SKent Overstreet void bch2_readahead(struct readahead_control *ractl) 8791c6fdbd8SKent Overstreet { 8801c6fdbd8SKent Overstreet struct bch_inode_info *inode = to_bch_ei(ractl->mapping->host); 8811c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 8829a3df993SKent Overstreet struct bch_io_opts opts = io_opts(c, &inode->ei_inode); 883424eb881SKent Overstreet struct btree_trans trans; 8841c6fdbd8SKent Overstreet struct page *page; 8851c6fdbd8SKent Overstreet struct readpages_iter readpages_iter; 8861c6fdbd8SKent Overstreet int ret; 8871c6fdbd8SKent Overstreet 8881c6fdbd8SKent Overstreet ret = readpages_iter_init(&readpages_iter, ractl); 8891c6fdbd8SKent Overstreet BUG_ON(ret); 8901c6fdbd8SKent Overstreet 89120bceecbSKent Overstreet bch2_trans_init(&trans, c, 0, 0); 8921c6fdbd8SKent Overstreet 8931c6fdbd8SKent Overstreet bch2_pagecache_add_get(&inode->ei_pagecache_lock); 8941c6fdbd8SKent Overstreet 8951c6fdbd8SKent Overstreet while ((page = readpage_iter_next(&readpages_iter))) { 8961c6fdbd8SKent Overstreet pgoff_t index = readpages_iter.offset + readpages_iter.idx; 8971c6fdbd8SKent Overstreet unsigned n = min_t(unsigned, 8981c6fdbd8SKent Overstreet readpages_iter.nr_pages - 8991c6fdbd8SKent Overstreet readpages_iter.idx, 9001c6fdbd8SKent Overstreet BIO_MAX_VECS); 9011c6fdbd8SKent Overstreet struct bch_read_bio *rbio = 9021c6fdbd8SKent Overstreet rbio_init(bio_alloc_bioset(NULL, n, REQ_OP_READ, 9031c6fdbd8SKent Overstreet GFP_NOFS, &c->bio_read), 9041c6fdbd8SKent Overstreet opts); 9051c6fdbd8SKent Overstreet 9061c6fdbd8SKent Overstreet readpages_iter.idx++; 9071c6fdbd8SKent Overstreet 9081c6fdbd8SKent Overstreet rbio->bio.bi_iter.bi_sector = (sector_t) index << PAGE_SECTOR_SHIFT; 9091c6fdbd8SKent Overstreet rbio->bio.bi_end_io = bch2_readpages_end_io; 910885678f6SKent Overstreet BUG_ON(!bio_add_page(&rbio->bio, page, PAGE_SIZE, 0)); 9111c6fdbd8SKent Overstreet 9128c6d298aSKent Overstreet bchfs_read(&trans, rbio, inode_inum(inode), 9130f238367SKent Overstreet &readpages_iter); 9141c6fdbd8SKent Overstreet } 9151c6fdbd8SKent Overstreet 9161c6fdbd8SKent Overstreet bch2_pagecache_add_put(&inode->ei_pagecache_lock); 917424eb881SKent Overstreet 918424eb881SKent Overstreet bch2_trans_exit(&trans); 9191c6fdbd8SKent Overstreet kfree(readpages_iter.pages); 9201c6fdbd8SKent Overstreet } 9211c6fdbd8SKent Overstreet 9221c6fdbd8SKent Overstreet static void __bchfs_readpage(struct bch_fs *c, struct bch_read_bio *rbio, 9238c6d298aSKent Overstreet subvol_inum inum, struct page *page) 9241c6fdbd8SKent Overstreet { 925424eb881SKent Overstreet struct btree_trans trans; 9261c6fdbd8SKent Overstreet 927f57a6a5dSKent Overstreet bch2_page_state_create(page, __GFP_NOFAIL); 9281c6fdbd8SKent Overstreet 9291c6fdbd8SKent Overstreet rbio->bio.bi_opf = REQ_OP_READ|REQ_SYNC; 9307f5e31e1SKent Overstreet rbio->bio.bi_iter.bi_sector = 9317f5e31e1SKent Overstreet (sector_t) page->index << PAGE_SECTOR_SHIFT; 9327f5e31e1SKent Overstreet BUG_ON(!bio_add_page(&rbio->bio, page, PAGE_SIZE, 0)); 9331c6fdbd8SKent Overstreet 93420bceecbSKent Overstreet bch2_trans_init(&trans, c, 0, 0); 9358c6d298aSKent Overstreet bchfs_read(&trans, rbio, inum, NULL); 936424eb881SKent Overstreet bch2_trans_exit(&trans); 9371c6fdbd8SKent Overstreet } 9381c6fdbd8SKent Overstreet 9391c6fdbd8SKent Overstreet static void bch2_read_single_page_end_io(struct bio *bio) 9401c6fdbd8SKent Overstreet { 9411c6fdbd8SKent Overstreet complete(bio->bi_private); 9421c6fdbd8SKent Overstreet } 9431c6fdbd8SKent Overstreet 9441c6fdbd8SKent Overstreet static int bch2_read_single_page(struct page *page, 9451c6fdbd8SKent Overstreet struct address_space *mapping) 9461c6fdbd8SKent Overstreet { 9471c6fdbd8SKent Overstreet struct bch_inode_info *inode = to_bch_ei(mapping->host); 9481c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 9491c6fdbd8SKent Overstreet struct bch_read_bio *rbio; 9501c6fdbd8SKent Overstreet int ret; 9511c6fdbd8SKent Overstreet DECLARE_COMPLETION_ONSTACK(done); 9521c6fdbd8SKent Overstreet 9531c6fdbd8SKent Overstreet rbio = rbio_init(bio_alloc_bioset(NULL, 1, REQ_OP_READ, GFP_NOFS, &c->bio_read), 9549a3df993SKent Overstreet io_opts(c, &inode->ei_inode)); 9551c6fdbd8SKent Overstreet rbio->bio.bi_private = &done; 9561c6fdbd8SKent Overstreet rbio->bio.bi_end_io = bch2_read_single_page_end_io; 9571c6fdbd8SKent Overstreet 9588c6d298aSKent Overstreet __bchfs_readpage(c, rbio, inode_inum(inode), page); 9591c6fdbd8SKent Overstreet wait_for_completion(&done); 9601c6fdbd8SKent Overstreet 9611c6fdbd8SKent Overstreet ret = blk_status_to_errno(rbio->bio.bi_status); 9621c6fdbd8SKent Overstreet bio_put(&rbio->bio); 9631c6fdbd8SKent Overstreet 9641c6fdbd8SKent Overstreet if (ret < 0) 9651c6fdbd8SKent Overstreet return ret; 9661c6fdbd8SKent Overstreet 9671c6fdbd8SKent Overstreet SetPageUptodate(page); 9681c6fdbd8SKent Overstreet return 0; 9691c6fdbd8SKent Overstreet } 9701c6fdbd8SKent Overstreet 9711c6fdbd8SKent Overstreet int bch2_read_folio(struct file *file, struct folio *folio) 9721c6fdbd8SKent Overstreet { 9731c6fdbd8SKent Overstreet struct page *page = &folio->page; 9741c6fdbd8SKent Overstreet int ret; 9751c6fdbd8SKent Overstreet 9761c6fdbd8SKent Overstreet ret = bch2_read_single_page(page, page->mapping); 9771c6fdbd8SKent Overstreet folio_unlock(folio); 9781c6fdbd8SKent Overstreet return ret; 9791c6fdbd8SKent Overstreet } 9801c6fdbd8SKent Overstreet 9811c6fdbd8SKent Overstreet /* writepages: */ 9821c6fdbd8SKent Overstreet 9831c6fdbd8SKent Overstreet struct bch_writepage_state { 9841c6fdbd8SKent Overstreet struct bch_writepage_io *io; 9851c6fdbd8SKent Overstreet struct bch_io_opts opts; 9861c6fdbd8SKent Overstreet }; 9871c6fdbd8SKent Overstreet 9881c6fdbd8SKent Overstreet static inline struct bch_writepage_state bch_writepage_state_init(struct bch_fs *c, 9891c6fdbd8SKent Overstreet struct bch_inode_info *inode) 9901c6fdbd8SKent Overstreet { 9919a3df993SKent Overstreet return (struct bch_writepage_state) { 9929a3df993SKent Overstreet .opts = io_opts(c, &inode->ei_inode) 9939a3df993SKent Overstreet }; 9941c6fdbd8SKent Overstreet } 9951c6fdbd8SKent Overstreet 9969f311f21SKent Overstreet static void bch2_writepage_io_done(struct bch_write_op *op) 9971c6fdbd8SKent Overstreet { 9989f311f21SKent Overstreet struct bch_writepage_io *io = 9999f311f21SKent Overstreet container_of(op, struct bch_writepage_io, op); 10009a3df993SKent Overstreet struct bch_fs *c = io->op.c; 10019a3df993SKent Overstreet struct bio *bio = &io->op.wbio.bio; 10021c6fdbd8SKent Overstreet struct bvec_iter_all iter; 10031c6fdbd8SKent Overstreet struct bio_vec *bvec; 1004b3fce09cSKent Overstreet unsigned i; 10051c6fdbd8SKent Overstreet 1006ef1b2092SKent Overstreet up(&io->op.c->io_in_flight); 1007ef1b2092SKent Overstreet 10089a3df993SKent Overstreet if (io->op.error) { 100933c74e41SKent Overstreet set_bit(EI_INODE_ERROR, &io->inode->ei_flags); 101033c74e41SKent Overstreet 101175812e70SKent Overstreet bio_for_each_segment_all(bvec, bio, iter) { 1012b3fce09cSKent Overstreet struct bch_page_state *s; 1013b3fce09cSKent Overstreet 10141c6fdbd8SKent Overstreet SetPageError(bvec->bv_page); 101575812e70SKent Overstreet mapping_set_error(bvec->bv_page->mapping, -EIO); 1016b3fce09cSKent Overstreet 10173826ee0bSKent Overstreet s = __bch2_page_state(bvec->bv_page); 10183826ee0bSKent Overstreet spin_lock(&s->lock); 1019b3fce09cSKent Overstreet for (i = 0; i < PAGE_SECTORS; i++) 1020b3fce09cSKent Overstreet s->s[i].nr_replicas = 0; 10213826ee0bSKent Overstreet spin_unlock(&s->lock); 102275812e70SKent Overstreet } 10231c6fdbd8SKent Overstreet } 10241c6fdbd8SKent Overstreet 10254be1a412SKent Overstreet if (io->op.flags & BCH_WRITE_WROTE_DATA_INLINE) { 10264be1a412SKent Overstreet bio_for_each_segment_all(bvec, bio, iter) { 10274be1a412SKent Overstreet struct bch_page_state *s; 10284be1a412SKent Overstreet 10294be1a412SKent Overstreet s = __bch2_page_state(bvec->bv_page); 10304be1a412SKent Overstreet spin_lock(&s->lock); 10314be1a412SKent Overstreet for (i = 0; i < PAGE_SECTORS; i++) 10324be1a412SKent Overstreet s->s[i].nr_replicas = 0; 10334be1a412SKent Overstreet spin_unlock(&s->lock); 10344be1a412SKent Overstreet } 10354be1a412SKent Overstreet } 10364be1a412SKent Overstreet 10371c6fdbd8SKent Overstreet /* 10381c6fdbd8SKent Overstreet * racing with fallocate can cause us to add fewer sectors than 10391c6fdbd8SKent Overstreet * expected - but we shouldn't add more sectors than expected: 10401c6fdbd8SKent Overstreet */ 10419a3df993SKent Overstreet BUG_ON(io->op.i_sectors_delta > 0); 10421c6fdbd8SKent Overstreet 10431c6fdbd8SKent Overstreet /* 10441c6fdbd8SKent Overstreet * (error (due to going RO) halfway through a page can screw that up 10451c6fdbd8SKent Overstreet * slightly) 10461c6fdbd8SKent Overstreet * XXX wtf? 10479a3df993SKent Overstreet BUG_ON(io->op.op.i_sectors_delta >= PAGE_SECTORS); 10481c6fdbd8SKent Overstreet */ 10491c6fdbd8SKent Overstreet 10501c6fdbd8SKent Overstreet /* 10511c6fdbd8SKent Overstreet * PageWriteback is effectively our ref on the inode - fixup i_blocks 10521c6fdbd8SKent Overstreet * before calling end_page_writeback: 10531c6fdbd8SKent Overstreet */ 10549a3df993SKent Overstreet i_sectors_acct(c, io->inode, NULL, io->op.i_sectors_delta); 10551c6fdbd8SKent Overstreet 10567f5e31e1SKent Overstreet bio_for_each_segment_all(bvec, bio, iter) { 10577f5e31e1SKent Overstreet struct bch_page_state *s = __bch2_page_state(bvec->bv_page); 10587f5e31e1SKent Overstreet 10597f5e31e1SKent Overstreet if (atomic_dec_and_test(&s->write_count)) 10601c6fdbd8SKent Overstreet end_page_writeback(bvec->bv_page); 10617f5e31e1SKent Overstreet } 10621c6fdbd8SKent Overstreet 10639f311f21SKent Overstreet bio_put(&io->op.wbio.bio); 10641c6fdbd8SKent Overstreet } 10651c6fdbd8SKent Overstreet 10661c6fdbd8SKent Overstreet static void bch2_writepage_do_io(struct bch_writepage_state *w) 10671c6fdbd8SKent Overstreet { 10681c6fdbd8SKent Overstreet struct bch_writepage_io *io = w->io; 10691c6fdbd8SKent Overstreet 1070ef1b2092SKent Overstreet down(&io->op.c->io_in_flight); 1071ef1b2092SKent Overstreet 10721c6fdbd8SKent Overstreet w->io = NULL; 10739f311f21SKent Overstreet closure_call(&io->op.cl, bch2_write, NULL, NULL); 10741c6fdbd8SKent Overstreet } 10751c6fdbd8SKent Overstreet 10761c6fdbd8SKent Overstreet /* 10771c6fdbd8SKent Overstreet * Get a bch_writepage_io and add @page to it - appending to an existing one if 10781c6fdbd8SKent Overstreet * possible, else allocating a new one: 10791c6fdbd8SKent Overstreet */ 10801c6fdbd8SKent Overstreet static void bch2_writepage_io_alloc(struct bch_fs *c, 108150fe5bd6SKent Overstreet struct writeback_control *wbc, 10821c6fdbd8SKent Overstreet struct bch_writepage_state *w, 10831c6fdbd8SKent Overstreet struct bch_inode_info *inode, 10847f5e31e1SKent Overstreet u64 sector, 10851c6fdbd8SKent Overstreet unsigned nr_replicas) 10861c6fdbd8SKent Overstreet { 10871c6fdbd8SKent Overstreet struct bch_write_op *op; 10881c6fdbd8SKent Overstreet 10891c6fdbd8SKent Overstreet w->io = container_of(bio_alloc_bioset(NULL, BIO_MAX_VECS, 10901c6fdbd8SKent Overstreet REQ_OP_WRITE, 10911c6fdbd8SKent Overstreet GFP_NOFS, 10921c6fdbd8SKent Overstreet &c->writepage_bioset), 10939a3df993SKent Overstreet struct bch_writepage_io, op.wbio.bio); 10941c6fdbd8SKent Overstreet 10959a3df993SKent Overstreet w->io->inode = inode; 10969a3df993SKent Overstreet op = &w->io->op; 10979a3df993SKent Overstreet bch2_write_op_init(op, c, w->opts); 10989a3df993SKent Overstreet op->target = w->opts.foreground_target; 10991c6fdbd8SKent Overstreet op->nr_replicas = nr_replicas; 11001c6fdbd8SKent Overstreet op->res.nr_replicas = nr_replicas; 11011c6fdbd8SKent Overstreet op->write_point = writepoint_hashed(inode->ei_last_dirtied); 11028c6d298aSKent Overstreet op->subvol = inode->ei_subvol; 11037f5e31e1SKent Overstreet op->pos = POS(inode->v.i_ino, sector); 11049f311f21SKent Overstreet op->end_io = bch2_writepage_io_done; 11057f5e31e1SKent Overstreet op->wbio.bio.bi_iter.bi_sector = sector; 110650fe5bd6SKent Overstreet op->wbio.bio.bi_opf = wbc_to_write_flags(wbc); 11071c6fdbd8SKent Overstreet } 11081c6fdbd8SKent Overstreet 11091c6fdbd8SKent Overstreet static int __bch2_writepage(struct folio *folio, 11101c6fdbd8SKent Overstreet struct writeback_control *wbc, 11111c6fdbd8SKent Overstreet void *data) 11121c6fdbd8SKent Overstreet { 11131c6fdbd8SKent Overstreet struct page *page = &folio->page; 11141c6fdbd8SKent Overstreet struct bch_inode_info *inode = to_bch_ei(page->mapping->host); 11151c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 11161c6fdbd8SKent Overstreet struct bch_writepage_state *w = data; 11177f5e31e1SKent Overstreet struct bch_page_state *s, orig; 11187f5e31e1SKent Overstreet unsigned i, offset, nr_replicas_this_write = U32_MAX; 11191c6fdbd8SKent Overstreet loff_t i_size = i_size_read(&inode->v); 11201c6fdbd8SKent Overstreet pgoff_t end_index = i_size >> PAGE_SHIFT; 1121e1036a2aSKent Overstreet int ret; 11221c6fdbd8SKent Overstreet 11231c6fdbd8SKent Overstreet EBUG_ON(!PageUptodate(page)); 11241c6fdbd8SKent Overstreet 11251c6fdbd8SKent Overstreet /* Is the page fully inside i_size? */ 11261c6fdbd8SKent Overstreet if (page->index < end_index) 11271c6fdbd8SKent Overstreet goto do_io; 11281c6fdbd8SKent Overstreet 11291c6fdbd8SKent Overstreet /* Is the page fully outside i_size? (truncate in progress) */ 11301c6fdbd8SKent Overstreet offset = i_size & (PAGE_SIZE - 1); 11311c6fdbd8SKent Overstreet if (page->index > end_index || !offset) { 11321c6fdbd8SKent Overstreet unlock_page(page); 11331c6fdbd8SKent Overstreet return 0; 11341c6fdbd8SKent Overstreet } 11351c6fdbd8SKent Overstreet 11361c6fdbd8SKent Overstreet /* 11371c6fdbd8SKent Overstreet * The page straddles i_size. It must be zeroed out on each and every 11381c6fdbd8SKent Overstreet * writepage invocation because it may be mmapped. "A file is mapped 11391c6fdbd8SKent Overstreet * in multiples of the page size. For a file that is not a multiple of 11401c6fdbd8SKent Overstreet * the page size, the remaining memory is zeroed when mapped, and 11411c6fdbd8SKent Overstreet * writes to that region are not written out to the file." 11421c6fdbd8SKent Overstreet */ 11431c6fdbd8SKent Overstreet zero_user_segment(page, offset, PAGE_SIZE); 11441c6fdbd8SKent Overstreet do_io: 1145f57a6a5dSKent Overstreet s = bch2_page_state_create(page, __GFP_NOFAIL); 1146f81b648dSKent Overstreet 1147f57a6a5dSKent Overstreet ret = bch2_get_page_disk_reservation(c, inode, page, true); 1148e1036a2aSKent Overstreet if (ret) { 1149e1036a2aSKent Overstreet SetPageError(page); 1150e1036a2aSKent Overstreet mapping_set_error(page->mapping, ret); 1151e1036a2aSKent Overstreet unlock_page(page); 1152e1036a2aSKent Overstreet return 0; 1153e1036a2aSKent Overstreet } 1154e1036a2aSKent Overstreet 11557f5e31e1SKent Overstreet /* Before unlocking the page, get copy of reservations: */ 11567f5e31e1SKent Overstreet orig = *s; 11577f5e31e1SKent Overstreet 11587f5e31e1SKent Overstreet for (i = 0; i < PAGE_SECTORS; i++) { 11592ba5d38bSKent Overstreet if (s->s[i].state < SECTOR_DIRTY) 11607f5e31e1SKent Overstreet continue; 11617f5e31e1SKent Overstreet 1162f81b648dSKent Overstreet nr_replicas_this_write = 1163f57a6a5dSKent Overstreet min_t(unsigned, nr_replicas_this_write, 1164f57a6a5dSKent Overstreet s->s[i].nr_replicas + 1165f57a6a5dSKent Overstreet s->s[i].replicas_reserved); 11667f5e31e1SKent Overstreet } 11671c6fdbd8SKent Overstreet 1168f57a6a5dSKent Overstreet for (i = 0; i < PAGE_SECTORS; i++) { 11692ba5d38bSKent Overstreet if (s->s[i].state < SECTOR_DIRTY) 11707f5e31e1SKent Overstreet continue; 11717f5e31e1SKent Overstreet 1172f57a6a5dSKent Overstreet s->s[i].nr_replicas = w->opts.compression 1173f57a6a5dSKent Overstreet ? 0 : nr_replicas_this_write; 1174e1036a2aSKent Overstreet 1175f57a6a5dSKent Overstreet s->s[i].replicas_reserved = 0; 1176f57a6a5dSKent Overstreet s->s[i].state = SECTOR_ALLOCATED; 1177f57a6a5dSKent Overstreet } 11781c6fdbd8SKent Overstreet 11797f5e31e1SKent Overstreet BUG_ON(atomic_read(&s->write_count)); 11807f5e31e1SKent Overstreet atomic_set(&s->write_count, 1); 11817f5e31e1SKent Overstreet 11821c6fdbd8SKent Overstreet BUG_ON(PageWriteback(page)); 11831c6fdbd8SKent Overstreet set_page_writeback(page); 11847f5e31e1SKent Overstreet 11851c6fdbd8SKent Overstreet unlock_page(page); 11861c6fdbd8SKent Overstreet 11877f5e31e1SKent Overstreet offset = 0; 11887f5e31e1SKent Overstreet while (1) { 11897f5e31e1SKent Overstreet unsigned sectors = 1, dirty_sectors = 0, reserved_sectors = 0; 11907f5e31e1SKent Overstreet u64 sector; 11917f5e31e1SKent Overstreet 11927f5e31e1SKent Overstreet while (offset < PAGE_SECTORS && 11932ba5d38bSKent Overstreet orig.s[offset].state < SECTOR_DIRTY) 11947f5e31e1SKent Overstreet offset++; 11957f5e31e1SKent Overstreet 11967f5e31e1SKent Overstreet if (offset == PAGE_SECTORS) 11977f5e31e1SKent Overstreet break; 11987f5e31e1SKent Overstreet 11997f5e31e1SKent Overstreet sector = ((u64) page->index << PAGE_SECTOR_SHIFT) + offset; 12007f5e31e1SKent Overstreet 12017f5e31e1SKent Overstreet while (offset + sectors < PAGE_SECTORS && 12022ba5d38bSKent Overstreet orig.s[offset + sectors].state >= SECTOR_DIRTY) 12037f5e31e1SKent Overstreet sectors++; 12047f5e31e1SKent Overstreet 12057f5e31e1SKent Overstreet for (i = offset; i < offset + sectors; i++) { 12067f5e31e1SKent Overstreet reserved_sectors += orig.s[i].replicas_reserved; 12077f5e31e1SKent Overstreet dirty_sectors += orig.s[i].state == SECTOR_DIRTY; 12087f5e31e1SKent Overstreet } 12097f5e31e1SKent Overstreet 12101c6fdbd8SKent Overstreet if (w->io && 12119a3df993SKent Overstreet (w->io->op.res.nr_replicas != nr_replicas_this_write || 12129a3df993SKent Overstreet bio_full(&w->io->op.wbio.bio, PAGE_SIZE) || 1213f59b3464SKent Overstreet w->io->op.wbio.bio.bi_iter.bi_size + (sectors << 9) >= 1214f59b3464SKent Overstreet (BIO_MAX_VECS * PAGE_SIZE) || 12159a3df993SKent Overstreet bio_end_sector(&w->io->op.wbio.bio) != sector)) 12161c6fdbd8SKent Overstreet bch2_writepage_do_io(w); 12171c6fdbd8SKent Overstreet 12181c6fdbd8SKent Overstreet if (!w->io) 121950fe5bd6SKent Overstreet bch2_writepage_io_alloc(c, wbc, w, inode, sector, 1220f81b648dSKent Overstreet nr_replicas_this_write); 12211c6fdbd8SKent Overstreet 12227f5e31e1SKent Overstreet atomic_inc(&s->write_count); 12237f5e31e1SKent Overstreet 12249a3df993SKent Overstreet BUG_ON(inode != w->io->inode); 12259a3df993SKent Overstreet BUG_ON(!bio_add_page(&w->io->op.wbio.bio, page, 12267f5e31e1SKent Overstreet sectors << 9, offset << 9)); 12271c6fdbd8SKent Overstreet 12286cc3535dSKent Overstreet /* Check for writing past i_size: */ 12299a3df993SKent Overstreet WARN_ON((bio_end_sector(&w->io->op.wbio.bio) << 9) > 12306cc3535dSKent Overstreet round_up(i_size, block_bytes(c))); 12316cc3535dSKent Overstreet 12329a3df993SKent Overstreet w->io->op.res.sectors += reserved_sectors; 12339a3df993SKent Overstreet w->io->op.i_sectors_delta -= dirty_sectors; 12341c6fdbd8SKent Overstreet w->io->op.new_i_size = i_size; 12351c6fdbd8SKent Overstreet 12367f5e31e1SKent Overstreet offset += sectors; 12377f5e31e1SKent Overstreet } 12387f5e31e1SKent Overstreet 12397f5e31e1SKent Overstreet if (atomic_dec_and_test(&s->write_count)) 12407f5e31e1SKent Overstreet end_page_writeback(page); 12417f5e31e1SKent Overstreet 12421c6fdbd8SKent Overstreet return 0; 12431c6fdbd8SKent Overstreet } 12441c6fdbd8SKent Overstreet 12451c6fdbd8SKent Overstreet int bch2_writepages(struct address_space *mapping, struct writeback_control *wbc) 12461c6fdbd8SKent Overstreet { 12471c6fdbd8SKent Overstreet struct bch_fs *c = mapping->host->i_sb->s_fs_info; 12481c6fdbd8SKent Overstreet struct bch_writepage_state w = 12491c6fdbd8SKent Overstreet bch_writepage_state_init(c, to_bch_ei(mapping->host)); 12501c6fdbd8SKent Overstreet struct blk_plug plug; 12511c6fdbd8SKent Overstreet int ret; 12521c6fdbd8SKent Overstreet 12531c6fdbd8SKent Overstreet blk_start_plug(&plug); 12541c6fdbd8SKent Overstreet ret = write_cache_pages(mapping, wbc, __bch2_writepage, &w); 12551c6fdbd8SKent Overstreet if (w.io) 12561c6fdbd8SKent Overstreet bch2_writepage_do_io(&w); 12571c6fdbd8SKent Overstreet blk_finish_plug(&plug); 12581c6fdbd8SKent Overstreet return ret; 12591c6fdbd8SKent Overstreet } 12601c6fdbd8SKent Overstreet 12611c6fdbd8SKent Overstreet int bch2_writepage(struct page *page, struct writeback_control *wbc) 12621c6fdbd8SKent Overstreet { 12631c6fdbd8SKent Overstreet struct bch_fs *c = page->mapping->host->i_sb->s_fs_info; 12641c6fdbd8SKent Overstreet struct bch_writepage_state w = 12651c6fdbd8SKent Overstreet bch_writepage_state_init(c, to_bch_ei(page->mapping->host)); 12661c6fdbd8SKent Overstreet int ret; 12671c6fdbd8SKent Overstreet 12681c6fdbd8SKent Overstreet ret = __bch2_writepage(page_folio(page), wbc, &w); 12691c6fdbd8SKent Overstreet if (w.io) 12701c6fdbd8SKent Overstreet bch2_writepage_do_io(&w); 12711c6fdbd8SKent Overstreet 12721c6fdbd8SKent Overstreet return ret; 12731c6fdbd8SKent Overstreet } 12741c6fdbd8SKent Overstreet 12751c6fdbd8SKent Overstreet /* buffered writes: */ 12761c6fdbd8SKent Overstreet 12771c6fdbd8SKent Overstreet int bch2_write_begin(struct file *file, struct address_space *mapping, 12781c6fdbd8SKent Overstreet loff_t pos, unsigned len, 12791c6fdbd8SKent Overstreet struct page **pagep, void **fsdata) 12801c6fdbd8SKent Overstreet { 12811c6fdbd8SKent Overstreet struct bch_inode_info *inode = to_bch_ei(mapping->host); 12821c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 1283d1542e03SKent Overstreet struct bch2_page_reservation *res; 12841c6fdbd8SKent Overstreet pgoff_t index = pos >> PAGE_SHIFT; 12851c6fdbd8SKent Overstreet unsigned offset = pos & (PAGE_SIZE - 1); 12861c6fdbd8SKent Overstreet struct page *page; 12871c6fdbd8SKent Overstreet int ret = -ENOMEM; 12881c6fdbd8SKent Overstreet 1289d1542e03SKent Overstreet res = kmalloc(sizeof(*res), GFP_KERNEL); 1290d1542e03SKent Overstreet if (!res) 1291d1542e03SKent Overstreet return -ENOMEM; 1292d1542e03SKent Overstreet 1293d1542e03SKent Overstreet bch2_page_reservation_init(c, inode, res); 1294d1542e03SKent Overstreet *fsdata = res; 12951c6fdbd8SKent Overstreet 12961c6fdbd8SKent Overstreet bch2_pagecache_add_get(&inode->ei_pagecache_lock); 12971c6fdbd8SKent Overstreet 12981c6fdbd8SKent Overstreet page = grab_cache_page_write_begin(mapping, index); 12991c6fdbd8SKent Overstreet if (!page) 13001c6fdbd8SKent Overstreet goto err_unlock; 13011c6fdbd8SKent Overstreet 13021c6fdbd8SKent Overstreet if (PageUptodate(page)) 13031c6fdbd8SKent Overstreet goto out; 13041c6fdbd8SKent Overstreet 13051c6fdbd8SKent Overstreet /* If we're writing entire page, don't need to read it in first: */ 13061c6fdbd8SKent Overstreet if (len == PAGE_SIZE) 13071c6fdbd8SKent Overstreet goto out; 13081c6fdbd8SKent Overstreet 13091c6fdbd8SKent Overstreet if (!offset && pos + len >= inode->v.i_size) { 13101c6fdbd8SKent Overstreet zero_user_segment(page, len, PAGE_SIZE); 13111c6fdbd8SKent Overstreet flush_dcache_page(page); 13121c6fdbd8SKent Overstreet goto out; 13131c6fdbd8SKent Overstreet } 13141c6fdbd8SKent Overstreet 13151c6fdbd8SKent Overstreet if (index > inode->v.i_size >> PAGE_SHIFT) { 13161c6fdbd8SKent Overstreet zero_user_segments(page, 0, offset, offset + len, PAGE_SIZE); 13171c6fdbd8SKent Overstreet flush_dcache_page(page); 13181c6fdbd8SKent Overstreet goto out; 13191c6fdbd8SKent Overstreet } 13201c6fdbd8SKent Overstreet readpage: 13211c6fdbd8SKent Overstreet ret = bch2_read_single_page(page, mapping); 13221c6fdbd8SKent Overstreet if (ret) 13231c6fdbd8SKent Overstreet goto err; 13241c6fdbd8SKent Overstreet out: 1325d1542e03SKent Overstreet ret = bch2_page_reservation_get(c, inode, page, res, 1326d1542e03SKent Overstreet offset, len, true); 13271c6fdbd8SKent Overstreet if (ret) { 13281c6fdbd8SKent Overstreet if (!PageUptodate(page)) { 13291c6fdbd8SKent Overstreet /* 13301c6fdbd8SKent Overstreet * If the page hasn't been read in, we won't know if we 13311c6fdbd8SKent Overstreet * actually need a reservation - we don't actually need 13321c6fdbd8SKent Overstreet * to read here, we just need to check if the page is 13331c6fdbd8SKent Overstreet * fully backed by uncompressed data: 13341c6fdbd8SKent Overstreet */ 13351c6fdbd8SKent Overstreet goto readpage; 13361c6fdbd8SKent Overstreet } 13371c6fdbd8SKent Overstreet 13381c6fdbd8SKent Overstreet goto err; 13391c6fdbd8SKent Overstreet } 13401c6fdbd8SKent Overstreet 13411c6fdbd8SKent Overstreet *pagep = page; 13421c6fdbd8SKent Overstreet return 0; 13431c6fdbd8SKent Overstreet err: 13441c6fdbd8SKent Overstreet unlock_page(page); 13451c6fdbd8SKent Overstreet put_page(page); 13461c6fdbd8SKent Overstreet *pagep = NULL; 13471c6fdbd8SKent Overstreet err_unlock: 13481c6fdbd8SKent Overstreet bch2_pagecache_add_put(&inode->ei_pagecache_lock); 1349d1542e03SKent Overstreet kfree(res); 1350d1542e03SKent Overstreet *fsdata = NULL; 13511c6fdbd8SKent Overstreet return ret; 13521c6fdbd8SKent Overstreet } 13531c6fdbd8SKent Overstreet 13541c6fdbd8SKent Overstreet int bch2_write_end(struct file *file, struct address_space *mapping, 13551c6fdbd8SKent Overstreet loff_t pos, unsigned len, unsigned copied, 13561c6fdbd8SKent Overstreet struct page *page, void *fsdata) 13571c6fdbd8SKent Overstreet { 13581c6fdbd8SKent Overstreet struct bch_inode_info *inode = to_bch_ei(mapping->host); 13591c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 1360d1542e03SKent Overstreet struct bch2_page_reservation *res = fsdata; 1361d1542e03SKent Overstreet unsigned offset = pos & (PAGE_SIZE - 1); 13621c6fdbd8SKent Overstreet 13631c6fdbd8SKent Overstreet lockdep_assert_held(&inode->v.i_rwsem); 13641c6fdbd8SKent Overstreet 13651c6fdbd8SKent Overstreet if (unlikely(copied < len && !PageUptodate(page))) { 13661c6fdbd8SKent Overstreet /* 13671c6fdbd8SKent Overstreet * The page needs to be read in, but that would destroy 13681c6fdbd8SKent Overstreet * our partial write - simplest thing is to just force 13691c6fdbd8SKent Overstreet * userspace to redo the write: 13701c6fdbd8SKent Overstreet */ 13711c6fdbd8SKent Overstreet zero_user(page, 0, PAGE_SIZE); 13721c6fdbd8SKent Overstreet flush_dcache_page(page); 13731c6fdbd8SKent Overstreet copied = 0; 13741c6fdbd8SKent Overstreet } 13751c6fdbd8SKent Overstreet 13761c6fdbd8SKent Overstreet spin_lock(&inode->v.i_lock); 13771c6fdbd8SKent Overstreet if (pos + copied > inode->v.i_size) 13781c6fdbd8SKent Overstreet i_size_write(&inode->v, pos + copied); 13791c6fdbd8SKent Overstreet spin_unlock(&inode->v.i_lock); 13801c6fdbd8SKent Overstreet 13811c6fdbd8SKent Overstreet if (copied) { 13821c6fdbd8SKent Overstreet if (!PageUptodate(page)) 13831c6fdbd8SKent Overstreet SetPageUptodate(page); 1384d1542e03SKent Overstreet 1385d1542e03SKent Overstreet bch2_set_page_dirty(c, inode, page, res, offset, copied); 13861c6fdbd8SKent Overstreet 13871c6fdbd8SKent Overstreet inode->ei_last_dirtied = (unsigned long) current; 13881c6fdbd8SKent Overstreet } 13891c6fdbd8SKent Overstreet 13901c6fdbd8SKent Overstreet unlock_page(page); 13911c6fdbd8SKent Overstreet put_page(page); 13921c6fdbd8SKent Overstreet bch2_pagecache_add_put(&inode->ei_pagecache_lock); 13931c6fdbd8SKent Overstreet 1394d1542e03SKent Overstreet bch2_page_reservation_put(c, inode, res); 1395d1542e03SKent Overstreet kfree(res); 1396d1542e03SKent Overstreet 13971c6fdbd8SKent Overstreet return copied; 13981c6fdbd8SKent Overstreet } 13991c6fdbd8SKent Overstreet 14001c6fdbd8SKent Overstreet #define WRITE_BATCH_PAGES 32 14011c6fdbd8SKent Overstreet 14021c6fdbd8SKent Overstreet static int __bch2_buffered_write(struct bch_inode_info *inode, 14031c6fdbd8SKent Overstreet struct address_space *mapping, 14041c6fdbd8SKent Overstreet struct iov_iter *iter, 14051c6fdbd8SKent Overstreet loff_t pos, unsigned len) 14061c6fdbd8SKent Overstreet { 14071c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 14081c6fdbd8SKent Overstreet struct page *pages[WRITE_BATCH_PAGES]; 1409d1542e03SKent Overstreet struct bch2_page_reservation res; 14101c6fdbd8SKent Overstreet unsigned long index = pos >> PAGE_SHIFT; 14111c6fdbd8SKent Overstreet unsigned offset = pos & (PAGE_SIZE - 1); 14121c6fdbd8SKent Overstreet unsigned nr_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE); 1413d1542e03SKent Overstreet unsigned i, reserved = 0, set_dirty = 0; 1414d1542e03SKent Overstreet unsigned copied = 0, nr_pages_copied = 0; 14151c6fdbd8SKent Overstreet int ret = 0; 14161c6fdbd8SKent Overstreet 14171c6fdbd8SKent Overstreet BUG_ON(!len); 14181c6fdbd8SKent Overstreet BUG_ON(nr_pages > ARRAY_SIZE(pages)); 14191c6fdbd8SKent Overstreet 1420d1542e03SKent Overstreet bch2_page_reservation_init(c, inode, &res); 1421d1542e03SKent Overstreet 14221c6fdbd8SKent Overstreet for (i = 0; i < nr_pages; i++) { 14231c6fdbd8SKent Overstreet pages[i] = grab_cache_page_write_begin(mapping, index + i); 14241c6fdbd8SKent Overstreet if (!pages[i]) { 14251c6fdbd8SKent Overstreet nr_pages = i; 14268de819f8SKent Overstreet if (!i) { 14271c6fdbd8SKent Overstreet ret = -ENOMEM; 14281c6fdbd8SKent Overstreet goto out; 14291c6fdbd8SKent Overstreet } 14308de819f8SKent Overstreet len = min_t(unsigned, len, 14318de819f8SKent Overstreet nr_pages * PAGE_SIZE - offset); 14328de819f8SKent Overstreet break; 14338de819f8SKent Overstreet } 14341c6fdbd8SKent Overstreet } 14351c6fdbd8SKent Overstreet 14361c6fdbd8SKent Overstreet if (offset && !PageUptodate(pages[0])) { 14371c6fdbd8SKent Overstreet ret = bch2_read_single_page(pages[0], mapping); 14381c6fdbd8SKent Overstreet if (ret) 14391c6fdbd8SKent Overstreet goto out; 14401c6fdbd8SKent Overstreet } 14411c6fdbd8SKent Overstreet 14421c6fdbd8SKent Overstreet if ((pos + len) & (PAGE_SIZE - 1) && 14431c6fdbd8SKent Overstreet !PageUptodate(pages[nr_pages - 1])) { 14441c6fdbd8SKent Overstreet if ((index + nr_pages - 1) << PAGE_SHIFT >= inode->v.i_size) { 14451c6fdbd8SKent Overstreet zero_user(pages[nr_pages - 1], 0, PAGE_SIZE); 14461c6fdbd8SKent Overstreet } else { 14471c6fdbd8SKent Overstreet ret = bch2_read_single_page(pages[nr_pages - 1], mapping); 14481c6fdbd8SKent Overstreet if (ret) 14491c6fdbd8SKent Overstreet goto out; 14501c6fdbd8SKent Overstreet } 14511c6fdbd8SKent Overstreet } 14521c6fdbd8SKent Overstreet 1453d1542e03SKent Overstreet while (reserved < len) { 1454d1542e03SKent Overstreet struct page *page = pages[(offset + reserved) >> PAGE_SHIFT]; 1455d1542e03SKent Overstreet unsigned pg_offset = (offset + reserved) & (PAGE_SIZE - 1); 1456d1542e03SKent Overstreet unsigned pg_len = min_t(unsigned, len - reserved, 1457d1542e03SKent Overstreet PAGE_SIZE - pg_offset); 1458d1542e03SKent Overstreet retry_reservation: 1459d1542e03SKent Overstreet ret = bch2_page_reservation_get(c, inode, page, &res, 1460d1542e03SKent Overstreet pg_offset, pg_len, true); 14611c6fdbd8SKent Overstreet 1462d1542e03SKent Overstreet if (ret && !PageUptodate(page)) { 1463d1542e03SKent Overstreet ret = bch2_read_single_page(page, mapping); 1464d1542e03SKent Overstreet if (!ret) 1465d1542e03SKent Overstreet goto retry_reservation; 14661c6fdbd8SKent Overstreet } 14671c6fdbd8SKent Overstreet 14681c6fdbd8SKent Overstreet if (ret) 14691c6fdbd8SKent Overstreet goto out; 1470d1542e03SKent Overstreet 1471d1542e03SKent Overstreet reserved += pg_len; 14721c6fdbd8SKent Overstreet } 14731c6fdbd8SKent Overstreet 14741c6fdbd8SKent Overstreet if (mapping_writably_mapped(mapping)) 14751c6fdbd8SKent Overstreet for (i = 0; i < nr_pages; i++) 14761c6fdbd8SKent Overstreet flush_dcache_page(pages[i]); 14771c6fdbd8SKent Overstreet 14781c6fdbd8SKent Overstreet while (copied < len) { 14791c6fdbd8SKent Overstreet struct page *page = pages[(offset + copied) >> PAGE_SHIFT]; 14801c6fdbd8SKent Overstreet unsigned pg_offset = (offset + copied) & (PAGE_SIZE - 1); 1481d1542e03SKent Overstreet unsigned pg_len = min_t(unsigned, len - copied, 14821c6fdbd8SKent Overstreet PAGE_SIZE - pg_offset); 14831c6fdbd8SKent Overstreet unsigned pg_copied = copy_page_from_iter_atomic(page, 1484d1542e03SKent Overstreet pg_offset, pg_len, iter); 1485d1542e03SKent Overstreet 1486d1542e03SKent Overstreet if (!pg_copied) 1487d1542e03SKent Overstreet break; 14881c6fdbd8SKent Overstreet 1489912bdf17SKent Overstreet if (!PageUptodate(page) && 1490912bdf17SKent Overstreet pg_copied != PAGE_SIZE && 1491912bdf17SKent Overstreet pos + copied + pg_copied < inode->v.i_size) { 1492912bdf17SKent Overstreet zero_user(page, 0, PAGE_SIZE); 1493912bdf17SKent Overstreet break; 1494912bdf17SKent Overstreet } 1495912bdf17SKent Overstreet 14961c6fdbd8SKent Overstreet flush_dcache_page(page); 14971c6fdbd8SKent Overstreet copied += pg_copied; 1498912bdf17SKent Overstreet 1499912bdf17SKent Overstreet if (pg_copied != pg_len) 1500912bdf17SKent Overstreet break; 15011c6fdbd8SKent Overstreet } 15021c6fdbd8SKent Overstreet 15031c6fdbd8SKent Overstreet if (!copied) 15041c6fdbd8SKent Overstreet goto out; 15051c6fdbd8SKent Overstreet 1506877dfb34SKent Overstreet spin_lock(&inode->v.i_lock); 1507877dfb34SKent Overstreet if (pos + copied > inode->v.i_size) 1508877dfb34SKent Overstreet i_size_write(&inode->v, pos + copied); 1509877dfb34SKent Overstreet spin_unlock(&inode->v.i_lock); 1510877dfb34SKent Overstreet 1511d1542e03SKent Overstreet while (set_dirty < copied) { 1512d1542e03SKent Overstreet struct page *page = pages[(offset + set_dirty) >> PAGE_SHIFT]; 1513d1542e03SKent Overstreet unsigned pg_offset = (offset + set_dirty) & (PAGE_SIZE - 1); 1514d1542e03SKent Overstreet unsigned pg_len = min_t(unsigned, copied - set_dirty, 1515d1542e03SKent Overstreet PAGE_SIZE - pg_offset); 1516d1542e03SKent Overstreet 1517d1542e03SKent Overstreet if (!PageUptodate(page)) 1518d1542e03SKent Overstreet SetPageUptodate(page); 1519d1542e03SKent Overstreet 1520d1542e03SKent Overstreet bch2_set_page_dirty(c, inode, page, &res, pg_offset, pg_len); 1521d1542e03SKent Overstreet unlock_page(page); 1522d1542e03SKent Overstreet put_page(page); 1523d1542e03SKent Overstreet 1524d1542e03SKent Overstreet set_dirty += pg_len; 1525d1542e03SKent Overstreet } 1526877dfb34SKent Overstreet 1527877dfb34SKent Overstreet nr_pages_copied = DIV_ROUND_UP(offset + copied, PAGE_SIZE); 1528877dfb34SKent Overstreet inode->ei_last_dirtied = (unsigned long) current; 15291c6fdbd8SKent Overstreet out: 1530d1542e03SKent Overstreet for (i = nr_pages_copied; i < nr_pages; i++) { 15311c6fdbd8SKent Overstreet unlock_page(pages[i]); 15321c6fdbd8SKent Overstreet put_page(pages[i]); 15331c6fdbd8SKent Overstreet } 15341c6fdbd8SKent Overstreet 1535d1542e03SKent Overstreet bch2_page_reservation_put(c, inode, &res); 15361c6fdbd8SKent Overstreet 15371c6fdbd8SKent Overstreet return copied ?: ret; 15381c6fdbd8SKent Overstreet } 15391c6fdbd8SKent Overstreet 15401c6fdbd8SKent Overstreet static ssize_t bch2_buffered_write(struct kiocb *iocb, struct iov_iter *iter) 15411c6fdbd8SKent Overstreet { 15421c6fdbd8SKent Overstreet struct file *file = iocb->ki_filp; 15431c6fdbd8SKent Overstreet struct address_space *mapping = file->f_mapping; 15441c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 15451c6fdbd8SKent Overstreet loff_t pos = iocb->ki_pos; 15461c6fdbd8SKent Overstreet ssize_t written = 0; 15471c6fdbd8SKent Overstreet int ret = 0; 15481c6fdbd8SKent Overstreet 15491c6fdbd8SKent Overstreet bch2_pagecache_add_get(&inode->ei_pagecache_lock); 15501c6fdbd8SKent Overstreet 15511c6fdbd8SKent Overstreet do { 15521c6fdbd8SKent Overstreet unsigned offset = pos & (PAGE_SIZE - 1); 15531c6fdbd8SKent Overstreet unsigned bytes = min_t(unsigned long, iov_iter_count(iter), 15541c6fdbd8SKent Overstreet PAGE_SIZE * WRITE_BATCH_PAGES - offset); 15551c6fdbd8SKent Overstreet again: 15561c6fdbd8SKent Overstreet /* 15571c6fdbd8SKent Overstreet * Bring in the user page that we will copy from _first_. 15581c6fdbd8SKent Overstreet * Otherwise there's a nasty deadlock on copying from the 15591c6fdbd8SKent Overstreet * same page as we're writing to, without it being marked 15601c6fdbd8SKent Overstreet * up-to-date. 15611c6fdbd8SKent Overstreet * 15621c6fdbd8SKent Overstreet * Not only is this an optimisation, but it is also required 15631c6fdbd8SKent Overstreet * to check that the address is actually valid, when atomic 15641c6fdbd8SKent Overstreet * usercopies are used, below. 15651c6fdbd8SKent Overstreet */ 15661c6fdbd8SKent Overstreet if (unlikely(fault_in_iov_iter_readable(iter, bytes))) { 15671c6fdbd8SKent Overstreet bytes = min_t(unsigned long, iov_iter_count(iter), 15681c6fdbd8SKent Overstreet PAGE_SIZE - offset); 15691c6fdbd8SKent Overstreet 15701c6fdbd8SKent Overstreet if (unlikely(fault_in_iov_iter_readable(iter, bytes))) { 15711c6fdbd8SKent Overstreet ret = -EFAULT; 15721c6fdbd8SKent Overstreet break; 15731c6fdbd8SKent Overstreet } 15741c6fdbd8SKent Overstreet } 15751c6fdbd8SKent Overstreet 15761c6fdbd8SKent Overstreet if (unlikely(fatal_signal_pending(current))) { 15771c6fdbd8SKent Overstreet ret = -EINTR; 15781c6fdbd8SKent Overstreet break; 15791c6fdbd8SKent Overstreet } 15801c6fdbd8SKent Overstreet 15811c6fdbd8SKent Overstreet ret = __bch2_buffered_write(inode, mapping, iter, pos, bytes); 15821c6fdbd8SKent Overstreet if (unlikely(ret < 0)) 15831c6fdbd8SKent Overstreet break; 15841c6fdbd8SKent Overstreet 15851c6fdbd8SKent Overstreet cond_resched(); 15861c6fdbd8SKent Overstreet 15871c6fdbd8SKent Overstreet if (unlikely(ret == 0)) { 15881c6fdbd8SKent Overstreet /* 15891c6fdbd8SKent Overstreet * If we were unable to copy any data at all, we must 15901c6fdbd8SKent Overstreet * fall back to a single segment length write. 15911c6fdbd8SKent Overstreet * 15921c6fdbd8SKent Overstreet * If we didn't fallback here, we could livelock 15931c6fdbd8SKent Overstreet * because not all segments in the iov can be copied at 15941c6fdbd8SKent Overstreet * once without a pagefault. 15951c6fdbd8SKent Overstreet */ 15961c6fdbd8SKent Overstreet bytes = min_t(unsigned long, PAGE_SIZE - offset, 15971c6fdbd8SKent Overstreet iov_iter_single_seg_count(iter)); 15981c6fdbd8SKent Overstreet goto again; 15991c6fdbd8SKent Overstreet } 16001c6fdbd8SKent Overstreet pos += ret; 16011c6fdbd8SKent Overstreet written += ret; 1602912bdf17SKent Overstreet ret = 0; 16031c6fdbd8SKent Overstreet 16041c6fdbd8SKent Overstreet balance_dirty_pages_ratelimited(mapping); 16051c6fdbd8SKent Overstreet } while (iov_iter_count(iter)); 16061c6fdbd8SKent Overstreet 16071c6fdbd8SKent Overstreet bch2_pagecache_add_put(&inode->ei_pagecache_lock); 16081c6fdbd8SKent Overstreet 16091c6fdbd8SKent Overstreet return written ? written : ret; 16101c6fdbd8SKent Overstreet } 16111c6fdbd8SKent Overstreet 16121c6fdbd8SKent Overstreet /* O_DIRECT reads */ 16131c6fdbd8SKent Overstreet 1614b4725cc1SKent Overstreet static void bio_check_or_release(struct bio *bio, bool check_dirty) 1615b4725cc1SKent Overstreet { 1616b4725cc1SKent Overstreet if (check_dirty) { 1617b4725cc1SKent Overstreet bio_check_pages_dirty(bio); 1618b4725cc1SKent Overstreet } else { 1619b4725cc1SKent Overstreet bio_release_pages(bio, false); 1620b4725cc1SKent Overstreet bio_put(bio); 1621b4725cc1SKent Overstreet } 1622b4725cc1SKent Overstreet } 1623b4725cc1SKent Overstreet 16241c6fdbd8SKent Overstreet static void bch2_dio_read_complete(struct closure *cl) 16251c6fdbd8SKent Overstreet { 16261c6fdbd8SKent Overstreet struct dio_read *dio = container_of(cl, struct dio_read, cl); 16271c6fdbd8SKent Overstreet 16281c6fdbd8SKent Overstreet dio->req->ki_complete(dio->req, dio->ret); 1629b4725cc1SKent Overstreet bio_check_or_release(&dio->rbio.bio, dio->should_dirty); 16301c6fdbd8SKent Overstreet } 16311c6fdbd8SKent Overstreet 16321c6fdbd8SKent Overstreet static void bch2_direct_IO_read_endio(struct bio *bio) 16331c6fdbd8SKent Overstreet { 16341c6fdbd8SKent Overstreet struct dio_read *dio = bio->bi_private; 16351c6fdbd8SKent Overstreet 16361c6fdbd8SKent Overstreet if (bio->bi_status) 16371c6fdbd8SKent Overstreet dio->ret = blk_status_to_errno(bio->bi_status); 16381c6fdbd8SKent Overstreet 16391c6fdbd8SKent Overstreet closure_put(&dio->cl); 16401c6fdbd8SKent Overstreet } 16411c6fdbd8SKent Overstreet 16421c6fdbd8SKent Overstreet static void bch2_direct_IO_read_split_endio(struct bio *bio) 16431c6fdbd8SKent Overstreet { 1644b4725cc1SKent Overstreet struct dio_read *dio = bio->bi_private; 1645b4725cc1SKent Overstreet bool should_dirty = dio->should_dirty; 1646b4725cc1SKent Overstreet 16471c6fdbd8SKent Overstreet bch2_direct_IO_read_endio(bio); 1648b4725cc1SKent Overstreet bio_check_or_release(bio, should_dirty); 16491c6fdbd8SKent Overstreet } 16501c6fdbd8SKent Overstreet 16511c6fdbd8SKent Overstreet static int bch2_direct_IO_read(struct kiocb *req, struct iov_iter *iter) 16521c6fdbd8SKent Overstreet { 16531c6fdbd8SKent Overstreet struct file *file = req->ki_filp; 16541c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 16551c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 16569a3df993SKent Overstreet struct bch_io_opts opts = io_opts(c, &inode->ei_inode); 16571c6fdbd8SKent Overstreet struct dio_read *dio; 16581c6fdbd8SKent Overstreet struct bio *bio; 16591c6fdbd8SKent Overstreet loff_t offset = req->ki_pos; 16601c6fdbd8SKent Overstreet bool sync = is_sync_kiocb(req); 16611c6fdbd8SKent Overstreet size_t shorten; 16621c6fdbd8SKent Overstreet ssize_t ret; 16631c6fdbd8SKent Overstreet 16641c6fdbd8SKent Overstreet if ((offset|iter->count) & (block_bytes(c) - 1)) 16651c6fdbd8SKent Overstreet return -EINVAL; 16661c6fdbd8SKent Overstreet 16671c6fdbd8SKent Overstreet ret = min_t(loff_t, iter->count, 16681c6fdbd8SKent Overstreet max_t(loff_t, 0, i_size_read(&inode->v) - offset)); 16691c6fdbd8SKent Overstreet 16701c6fdbd8SKent Overstreet if (!ret) 16711c6fdbd8SKent Overstreet return ret; 16721c6fdbd8SKent Overstreet 16731c6fdbd8SKent Overstreet shorten = iov_iter_count(iter) - round_up(ret, block_bytes(c)); 16741c6fdbd8SKent Overstreet iter->count -= shorten; 16751c6fdbd8SKent Overstreet 16761c6fdbd8SKent Overstreet bio = bio_alloc_bioset(NULL, 16771c6fdbd8SKent Overstreet iov_iter_npages(iter, BIO_MAX_VECS), 16781c6fdbd8SKent Overstreet REQ_OP_READ, 16791c6fdbd8SKent Overstreet GFP_KERNEL, 16801c6fdbd8SKent Overstreet &c->dio_read_bioset); 16811c6fdbd8SKent Overstreet 16821c6fdbd8SKent Overstreet bio->bi_end_io = bch2_direct_IO_read_endio; 16831c6fdbd8SKent Overstreet 16841c6fdbd8SKent Overstreet dio = container_of(bio, struct dio_read, rbio.bio); 16851c6fdbd8SKent Overstreet closure_init(&dio->cl, NULL); 16861c6fdbd8SKent Overstreet 16871c6fdbd8SKent Overstreet /* 16881c6fdbd8SKent Overstreet * this is a _really_ horrible hack just to avoid an atomic sub at the 16891c6fdbd8SKent Overstreet * end: 16901c6fdbd8SKent Overstreet */ 16911c6fdbd8SKent Overstreet if (!sync) { 16921c6fdbd8SKent Overstreet set_closure_fn(&dio->cl, bch2_dio_read_complete, NULL); 16931c6fdbd8SKent Overstreet atomic_set(&dio->cl.remaining, 16941c6fdbd8SKent Overstreet CLOSURE_REMAINING_INITIALIZER - 16951c6fdbd8SKent Overstreet CLOSURE_RUNNING + 16961c6fdbd8SKent Overstreet CLOSURE_DESTRUCTOR); 16971c6fdbd8SKent Overstreet } else { 16981c6fdbd8SKent Overstreet atomic_set(&dio->cl.remaining, 16991c6fdbd8SKent Overstreet CLOSURE_REMAINING_INITIALIZER + 1); 17001c6fdbd8SKent Overstreet } 17011c6fdbd8SKent Overstreet 17021c6fdbd8SKent Overstreet dio->req = req; 17031c6fdbd8SKent Overstreet dio->ret = ret; 1704b4725cc1SKent Overstreet /* 1705b4725cc1SKent Overstreet * This is one of the sketchier things I've encountered: we have to skip 1706b4725cc1SKent Overstreet * the dirtying of requests that are internal from the kernel (i.e. from 1707b4725cc1SKent Overstreet * loopback), because we'll deadlock on page_lock. 1708b4725cc1SKent Overstreet */ 1709b4725cc1SKent Overstreet dio->should_dirty = iter_is_iovec(iter); 17101c6fdbd8SKent Overstreet 17111c6fdbd8SKent Overstreet goto start; 17121c6fdbd8SKent Overstreet while (iter->count) { 17131c6fdbd8SKent Overstreet bio = bio_alloc_bioset(NULL, 17141c6fdbd8SKent Overstreet iov_iter_npages(iter, BIO_MAX_VECS), 17151c6fdbd8SKent Overstreet REQ_OP_READ, 17161c6fdbd8SKent Overstreet GFP_KERNEL, 17171c6fdbd8SKent Overstreet &c->bio_read); 17181c6fdbd8SKent Overstreet bio->bi_end_io = bch2_direct_IO_read_split_endio; 17191c6fdbd8SKent Overstreet start: 17201c6fdbd8SKent Overstreet bio->bi_opf = REQ_OP_READ|REQ_SYNC; 17211c6fdbd8SKent Overstreet bio->bi_iter.bi_sector = offset >> 9; 17221c6fdbd8SKent Overstreet bio->bi_private = dio; 17231c6fdbd8SKent Overstreet 17241c6fdbd8SKent Overstreet ret = bio_iov_iter_get_pages(bio, iter); 17251c6fdbd8SKent Overstreet if (ret < 0) { 17261c6fdbd8SKent Overstreet /* XXX: fault inject this path */ 17271c6fdbd8SKent Overstreet bio->bi_status = BLK_STS_RESOURCE; 17281c6fdbd8SKent Overstreet bio_endio(bio); 17291c6fdbd8SKent Overstreet break; 17301c6fdbd8SKent Overstreet } 17311c6fdbd8SKent Overstreet 17321c6fdbd8SKent Overstreet offset += bio->bi_iter.bi_size; 1733b4725cc1SKent Overstreet 1734b4725cc1SKent Overstreet if (dio->should_dirty) 17351c6fdbd8SKent Overstreet bio_set_pages_dirty(bio); 17361c6fdbd8SKent Overstreet 17371c6fdbd8SKent Overstreet if (iter->count) 17381c6fdbd8SKent Overstreet closure_get(&dio->cl); 17391c6fdbd8SKent Overstreet 17408c6d298aSKent Overstreet bch2_read(c, rbio_init(bio, opts), inode_inum(inode)); 17411c6fdbd8SKent Overstreet } 17421c6fdbd8SKent Overstreet 17431c6fdbd8SKent Overstreet iter->count += shorten; 17441c6fdbd8SKent Overstreet 17451c6fdbd8SKent Overstreet if (sync) { 17461c6fdbd8SKent Overstreet closure_sync(&dio->cl); 17471c6fdbd8SKent Overstreet closure_debug_destroy(&dio->cl); 17481c6fdbd8SKent Overstreet ret = dio->ret; 1749b4725cc1SKent Overstreet bio_check_or_release(&dio->rbio.bio, dio->should_dirty); 17501c6fdbd8SKent Overstreet return ret; 17511c6fdbd8SKent Overstreet } else { 17521c6fdbd8SKent Overstreet return -EIOCBQUEUED; 17531c6fdbd8SKent Overstreet } 17541c6fdbd8SKent Overstreet } 17551c6fdbd8SKent Overstreet 17561c6fdbd8SKent Overstreet ssize_t bch2_read_iter(struct kiocb *iocb, struct iov_iter *iter) 17571c6fdbd8SKent Overstreet { 17581c6fdbd8SKent Overstreet struct file *file = iocb->ki_filp; 17591c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 17601c6fdbd8SKent Overstreet struct address_space *mapping = file->f_mapping; 17611c6fdbd8SKent Overstreet size_t count = iov_iter_count(iter); 17621c6fdbd8SKent Overstreet ssize_t ret; 17631c6fdbd8SKent Overstreet 17641c6fdbd8SKent Overstreet if (!count) 17651c6fdbd8SKent Overstreet return 0; /* skip atime */ 17661c6fdbd8SKent Overstreet 17671c6fdbd8SKent Overstreet if (iocb->ki_flags & IOCB_DIRECT) { 17681c6fdbd8SKent Overstreet struct blk_plug plug; 17691c6fdbd8SKent Overstreet 1770a023127aSKent Overstreet if (unlikely(mapping->nrpages)) { 17711c6fdbd8SKent Overstreet ret = filemap_write_and_wait_range(mapping, 17721c6fdbd8SKent Overstreet iocb->ki_pos, 17731c6fdbd8SKent Overstreet iocb->ki_pos + count - 1); 17741c6fdbd8SKent Overstreet if (ret < 0) 17751c6fdbd8SKent Overstreet return ret; 1776a023127aSKent Overstreet } 17771c6fdbd8SKent Overstreet 17781c6fdbd8SKent Overstreet file_accessed(file); 17791c6fdbd8SKent Overstreet 17801c6fdbd8SKent Overstreet blk_start_plug(&plug); 17811c6fdbd8SKent Overstreet ret = bch2_direct_IO_read(iocb, iter); 17821c6fdbd8SKent Overstreet blk_finish_plug(&plug); 17831c6fdbd8SKent Overstreet 17841c6fdbd8SKent Overstreet if (ret >= 0) 17851c6fdbd8SKent Overstreet iocb->ki_pos += ret; 17861c6fdbd8SKent Overstreet } else { 17871c6fdbd8SKent Overstreet bch2_pagecache_add_get(&inode->ei_pagecache_lock); 17881c6fdbd8SKent Overstreet ret = generic_file_read_iter(iocb, iter); 17891c6fdbd8SKent Overstreet bch2_pagecache_add_put(&inode->ei_pagecache_lock); 17901c6fdbd8SKent Overstreet } 17911c6fdbd8SKent Overstreet 17921c6fdbd8SKent Overstreet return ret; 17931c6fdbd8SKent Overstreet } 17941c6fdbd8SKent Overstreet 17951c6fdbd8SKent Overstreet /* O_DIRECT writes */ 17961c6fdbd8SKent Overstreet 17976fed42bbSKent Overstreet static bool bch2_check_range_allocated(struct bch_fs *c, subvol_inum inum, 17986fed42bbSKent Overstreet u64 offset, u64 size, 17996fed42bbSKent Overstreet unsigned nr_replicas, bool compressed) 18006fed42bbSKent Overstreet { 18016fed42bbSKent Overstreet struct btree_trans trans; 18026fed42bbSKent Overstreet struct btree_iter iter; 18036fed42bbSKent Overstreet struct bkey_s_c k; 18046fed42bbSKent Overstreet u64 end = offset + size; 18056fed42bbSKent Overstreet u32 snapshot; 18066fed42bbSKent Overstreet bool ret = true; 18076fed42bbSKent Overstreet int err; 18086fed42bbSKent Overstreet 18096fed42bbSKent Overstreet bch2_trans_init(&trans, c, 0, 0); 18106fed42bbSKent Overstreet retry: 18116fed42bbSKent Overstreet bch2_trans_begin(&trans); 18126fed42bbSKent Overstreet 18136fed42bbSKent Overstreet err = bch2_subvolume_get_snapshot(&trans, inum.subvol, &snapshot); 18146fed42bbSKent Overstreet if (err) 18156fed42bbSKent Overstreet goto err; 18166fed42bbSKent Overstreet 1817e5fa91d7SKent Overstreet for_each_btree_key_norestart(&trans, iter, BTREE_ID_extents, 18186fed42bbSKent Overstreet SPOS(inum.inum, offset, snapshot), 18196fed42bbSKent Overstreet BTREE_ITER_SLOTS, k, err) { 18206fed42bbSKent Overstreet if (bkey_cmp(bkey_start_pos(k.k), POS(inum.inum, end)) >= 0) 18216fed42bbSKent Overstreet break; 18226fed42bbSKent Overstreet 18238c6d298aSKent Overstreet if (k.k->p.snapshot != snapshot || 18248c6d298aSKent Overstreet nr_replicas > bch2_bkey_replicas(c, k) || 18256fed42bbSKent Overstreet (!compressed && bch2_bkey_sectors_compressed(k))) { 18266fed42bbSKent Overstreet ret = false; 18276fed42bbSKent Overstreet break; 18286fed42bbSKent Overstreet } 18296fed42bbSKent Overstreet } 18306fed42bbSKent Overstreet 18316fed42bbSKent Overstreet offset = iter.pos.offset; 18326fed42bbSKent Overstreet bch2_trans_iter_exit(&trans, &iter); 18336fed42bbSKent Overstreet err: 18346fed42bbSKent Overstreet if (err == -EINTR) 18356fed42bbSKent Overstreet goto retry; 18366fed42bbSKent Overstreet bch2_trans_exit(&trans); 18376fed42bbSKent Overstreet 18386fed42bbSKent Overstreet return err ? false : ret; 18396fed42bbSKent Overstreet } 18406fed42bbSKent Overstreet 18411c6fdbd8SKent Overstreet /* 18421c6fdbd8SKent Overstreet * We're going to return -EIOCBQUEUED, but we haven't finished consuming the 18431c6fdbd8SKent Overstreet * iov_iter yet, so we need to stash a copy of the iovec: it might be on the 18441c6fdbd8SKent Overstreet * caller's stack, we're not guaranteed that it will live for the duration of 18451c6fdbd8SKent Overstreet * the IO: 18461c6fdbd8SKent Overstreet */ 18471c6fdbd8SKent Overstreet static noinline int bch2_dio_write_copy_iov(struct dio_write *dio) 18481c6fdbd8SKent Overstreet { 18491c6fdbd8SKent Overstreet struct iovec *iov = dio->inline_vecs; 18501c6fdbd8SKent Overstreet 18511c6fdbd8SKent Overstreet /* 18521c6fdbd8SKent Overstreet * iov_iter has a single embedded iovec - nothing to do: 18531c6fdbd8SKent Overstreet */ 18541c6fdbd8SKent Overstreet if (iter_is_ubuf(&dio->iter)) 18551c6fdbd8SKent Overstreet return 0; 18561c6fdbd8SKent Overstreet 18571c6fdbd8SKent Overstreet /* 18581c6fdbd8SKent Overstreet * We don't currently handle non-iovec iov_iters here - return an error, 18591c6fdbd8SKent Overstreet * and we'll fall back to doing the IO synchronously: 18601c6fdbd8SKent Overstreet */ 18611c6fdbd8SKent Overstreet if (!iter_is_iovec(&dio->iter)) 18621c6fdbd8SKent Overstreet return -1; 18631c6fdbd8SKent Overstreet 18641c6fdbd8SKent Overstreet if (dio->iter.nr_segs > ARRAY_SIZE(dio->inline_vecs)) { 18651c6fdbd8SKent Overstreet iov = kmalloc_array(dio->iter.nr_segs, sizeof(*iov), 18661c6fdbd8SKent Overstreet GFP_KERNEL); 18671c6fdbd8SKent Overstreet if (unlikely(!iov)) 18681c6fdbd8SKent Overstreet return -ENOMEM; 18691c6fdbd8SKent Overstreet 18701c6fdbd8SKent Overstreet dio->free_iov = true; 18711c6fdbd8SKent Overstreet } 18721c6fdbd8SKent Overstreet 18731c6fdbd8SKent Overstreet memcpy(iov, dio->iter.__iov, dio->iter.nr_segs * sizeof(*iov)); 18741c6fdbd8SKent Overstreet dio->iter.__iov = iov; 18751c6fdbd8SKent Overstreet return 0; 18761c6fdbd8SKent Overstreet } 18771c6fdbd8SKent Overstreet 1878042a1f26SKent Overstreet static void bch2_dio_write_loop_async(struct bch_write_op *); 1879042a1f26SKent Overstreet 18801c6fdbd8SKent Overstreet static long bch2_dio_write_loop(struct dio_write *dio) 18811c6fdbd8SKent Overstreet { 1882ed484030SKent Overstreet bool kthread = (current->flags & PF_KTHREAD) != 0; 18831c6fdbd8SKent Overstreet struct kiocb *req = dio->req; 18841c6fdbd8SKent Overstreet struct address_space *mapping = req->ki_filp->f_mapping; 18859a3df993SKent Overstreet struct bch_inode_info *inode = file_bch_inode(req->ki_filp); 1886042a1f26SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 18879a3df993SKent Overstreet struct bio *bio = &dio->op.wbio.bio; 1888eb8e6e9cSKent Overstreet unsigned unaligned, iter_count; 1889eb8e6e9cSKent Overstreet bool sync = dio->sync, dropped_locks; 18901c6fdbd8SKent Overstreet long ret; 18911c6fdbd8SKent Overstreet 18921c6fdbd8SKent Overstreet if (dio->loop) 18931c6fdbd8SKent Overstreet goto loop; 18941c6fdbd8SKent Overstreet 18951c6fdbd8SKent Overstreet while (1) { 1896eb8e6e9cSKent Overstreet iter_count = dio->iter.count; 1897eb8e6e9cSKent Overstreet 1898ed484030SKent Overstreet if (kthread) 1899ed484030SKent Overstreet kthread_use_mm(dio->mm); 19001c6fdbd8SKent Overstreet BUG_ON(current->faults_disabled_mapping); 19011c6fdbd8SKent Overstreet current->faults_disabled_mapping = mapping; 19021c6fdbd8SKent Overstreet 19031c6fdbd8SKent Overstreet ret = bio_iov_iter_get_pages(bio, &dio->iter); 19041c6fdbd8SKent Overstreet 1905eb8e6e9cSKent Overstreet dropped_locks = fdm_dropped_locks(); 1906eb8e6e9cSKent Overstreet 19071c6fdbd8SKent Overstreet current->faults_disabled_mapping = NULL; 1908ed484030SKent Overstreet if (kthread) 1909ed484030SKent Overstreet kthread_unuse_mm(dio->mm); 19101c6fdbd8SKent Overstreet 1911eb8e6e9cSKent Overstreet /* 1912eb8e6e9cSKent Overstreet * If the fault handler returned an error but also signalled 1913eb8e6e9cSKent Overstreet * that it dropped & retook ei_pagecache_lock, we just need to 1914eb8e6e9cSKent Overstreet * re-shoot down the page cache and retry: 1915eb8e6e9cSKent Overstreet */ 1916eb8e6e9cSKent Overstreet if (dropped_locks && ret) 1917eb8e6e9cSKent Overstreet ret = 0; 1918eb8e6e9cSKent Overstreet 19191c6fdbd8SKent Overstreet if (unlikely(ret < 0)) 19201c6fdbd8SKent Overstreet goto err; 19211c6fdbd8SKent Overstreet 1922eb8e6e9cSKent Overstreet if (unlikely(dropped_locks)) { 1923eb8e6e9cSKent Overstreet ret = write_invalidate_inode_pages_range(mapping, 1924eb8e6e9cSKent Overstreet req->ki_pos, 1925eb8e6e9cSKent Overstreet req->ki_pos + iter_count - 1); 1926eb8e6e9cSKent Overstreet if (unlikely(ret)) 1927eb8e6e9cSKent Overstreet goto err; 1928eb8e6e9cSKent Overstreet 1929eb8e6e9cSKent Overstreet if (!bio->bi_iter.bi_size) 1930eb8e6e9cSKent Overstreet continue; 1931eb8e6e9cSKent Overstreet } 1932eb8e6e9cSKent Overstreet 19330a426c32SKent Overstreet unaligned = bio->bi_iter.bi_size & (block_bytes(c) - 1); 19340a426c32SKent Overstreet bio->bi_iter.bi_size -= unaligned; 19350a426c32SKent Overstreet iov_iter_revert(&dio->iter, unaligned); 19360a426c32SKent Overstreet 19370a426c32SKent Overstreet if (!bio->bi_iter.bi_size) { 19380a426c32SKent Overstreet /* 19390a426c32SKent Overstreet * bio_iov_iter_get_pages was only able to get < 19400a426c32SKent Overstreet * blocksize worth of pages: 19410a426c32SKent Overstreet */ 19420a426c32SKent Overstreet ret = -EFAULT; 19430a426c32SKent Overstreet goto err; 19440a426c32SKent Overstreet } 19450a426c32SKent Overstreet 1946042a1f26SKent Overstreet bch2_write_op_init(&dio->op, c, io_opts(c, &inode->ei_inode)); 1947042a1f26SKent Overstreet dio->op.end_io = bch2_dio_write_loop_async; 1948042a1f26SKent Overstreet dio->op.target = dio->op.opts.foreground_target; 1949042a1f26SKent Overstreet dio->op.write_point = writepoint_hashed((unsigned long) current); 1950042a1f26SKent Overstreet dio->op.nr_replicas = dio->op.opts.data_replicas; 19518c6d298aSKent Overstreet dio->op.subvol = inode->ei_subvol; 1952042a1f26SKent Overstreet dio->op.pos = POS(inode->v.i_ino, (u64) req->ki_pos >> 9); 1953042a1f26SKent Overstreet 1954042a1f26SKent Overstreet if ((req->ki_flags & IOCB_DSYNC) && 1955042a1f26SKent Overstreet !c->opts.journal_flush_disabled) 1956042a1f26SKent Overstreet dio->op.flags |= BCH_WRITE_FLUSH; 1957a6336910SKent Overstreet dio->op.flags |= BCH_WRITE_CHECK_ENOSPC; 1958042a1f26SKent Overstreet 1959042a1f26SKent Overstreet ret = bch2_disk_reservation_get(c, &dio->op.res, bio_sectors(bio), 1960042a1f26SKent Overstreet dio->op.opts.data_replicas, 0); 1961042a1f26SKent Overstreet if (unlikely(ret) && 19626fed42bbSKent Overstreet !bch2_check_range_allocated(c, inode_inum(inode), 19636fed42bbSKent Overstreet dio->op.pos.offset, bio_sectors(bio), 196435a067b4SKent Overstreet dio->op.opts.data_replicas, 196535a067b4SKent Overstreet dio->op.opts.compression != 0)) 1966042a1f26SKent Overstreet goto err; 19671c6fdbd8SKent Overstreet 19681c6fdbd8SKent Overstreet task_io_account_write(bio->bi_iter.bi_size); 19691c6fdbd8SKent Overstreet 19701c6fdbd8SKent Overstreet if (!dio->sync && !dio->loop && dio->iter.count) { 19711c6fdbd8SKent Overstreet if (bch2_dio_write_copy_iov(dio)) { 1972286d8ad0SKent Overstreet dio->sync = sync = true; 1973f8f30863SKent Overstreet goto do_io; 19741c6fdbd8SKent Overstreet } 19751c6fdbd8SKent Overstreet } 1976f8f30863SKent Overstreet do_io: 19771c6fdbd8SKent Overstreet dio->loop = true; 1978f8f30863SKent Overstreet closure_call(&dio->op.cl, bch2_write, NULL, NULL); 19791c6fdbd8SKent Overstreet 1980286d8ad0SKent Overstreet if (sync) 1981f8f30863SKent Overstreet wait_for_completion(&dio->done); 1982f8f30863SKent Overstreet else 19831c6fdbd8SKent Overstreet return -EIOCBQUEUED; 19841c6fdbd8SKent Overstreet loop: 19859a3df993SKent Overstreet i_sectors_acct(c, inode, &dio->quota_res, 19869a3df993SKent Overstreet dio->op.i_sectors_delta); 1987042a1f26SKent Overstreet req->ki_pos += (u64) dio->op.written << 9; 1988042a1f26SKent Overstreet dio->written += dio->op.written; 19899a3df993SKent Overstreet 19909a3df993SKent Overstreet spin_lock(&inode->v.i_lock); 1991042a1f26SKent Overstreet if (req->ki_pos > inode->v.i_size) 1992042a1f26SKent Overstreet i_size_write(&inode->v, req->ki_pos); 19939a3df993SKent Overstreet spin_unlock(&inode->v.i_lock); 19949a3df993SKent Overstreet 1995032ac32cSKent Overstreet bio_release_pages(bio, false); 19965468f119SKent Overstreet bio->bi_vcnt = 0; 199733c74e41SKent Overstreet 199833c74e41SKent Overstreet if (dio->op.error) { 199933c74e41SKent Overstreet set_bit(EI_INODE_ERROR, &inode->ei_flags); 200033c74e41SKent Overstreet break; 200133c74e41SKent Overstreet } 200233c74e41SKent Overstreet 200333c74e41SKent Overstreet if (!dio->iter.count) 20041c6fdbd8SKent Overstreet break; 2005f8f30863SKent Overstreet 20061c6fdbd8SKent Overstreet bio_reset(bio, NULL, REQ_OP_WRITE); 2007f8f30863SKent Overstreet reinit_completion(&dio->done); 20081c6fdbd8SKent Overstreet } 20091c6fdbd8SKent Overstreet 2010042a1f26SKent Overstreet ret = dio->op.error ?: ((long) dio->written << 9); 20111c6fdbd8SKent Overstreet err: 20121c6fdbd8SKent Overstreet bch2_pagecache_block_put(&inode->ei_pagecache_lock); 20130a426c32SKent Overstreet bch2_quota_reservation_put(c, inode, &dio->quota_res); 20141c6fdbd8SKent Overstreet 20151c6fdbd8SKent Overstreet if (dio->free_iov) 20161c6fdbd8SKent Overstreet kfree(dio->iter.__iov); 20171c6fdbd8SKent Overstreet 20185468f119SKent Overstreet bio_release_pages(bio, false); 20191c6fdbd8SKent Overstreet bio_put(bio); 20201c6fdbd8SKent Overstreet 20211c6fdbd8SKent Overstreet /* inode->i_dio_count is our ref on inode and thus bch_fs */ 20221c6fdbd8SKent Overstreet inode_dio_end(&inode->v); 20231c6fdbd8SKent Overstreet 20241c6fdbd8SKent Overstreet if (!sync) { 20251c6fdbd8SKent Overstreet req->ki_complete(req, ret); 20261c6fdbd8SKent Overstreet ret = -EIOCBQUEUED; 20271c6fdbd8SKent Overstreet } 20281c6fdbd8SKent Overstreet return ret; 20291c6fdbd8SKent Overstreet } 20301c6fdbd8SKent Overstreet 2031f8f30863SKent Overstreet static void bch2_dio_write_loop_async(struct bch_write_op *op) 20321c6fdbd8SKent Overstreet { 2033f8f30863SKent Overstreet struct dio_write *dio = container_of(op, struct dio_write, op); 20341c6fdbd8SKent Overstreet 2035f8f30863SKent Overstreet if (dio->sync) 2036f8f30863SKent Overstreet complete(&dio->done); 2037f8f30863SKent Overstreet else 20381c6fdbd8SKent Overstreet bch2_dio_write_loop(dio); 20391c6fdbd8SKent Overstreet } 20401c6fdbd8SKent Overstreet 20411c6fdbd8SKent Overstreet static noinline 20421c6fdbd8SKent Overstreet ssize_t bch2_direct_write(struct kiocb *req, struct iov_iter *iter) 20431c6fdbd8SKent Overstreet { 20441c6fdbd8SKent Overstreet struct file *file = req->ki_filp; 204554847d25SKent Overstreet struct address_space *mapping = file->f_mapping; 20461c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 20471c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 20481c6fdbd8SKent Overstreet struct dio_write *dio; 20491c6fdbd8SKent Overstreet struct bio *bio; 20507edcfbfeSKent Overstreet bool locked = true, extending; 20511c6fdbd8SKent Overstreet ssize_t ret; 20521c6fdbd8SKent Overstreet 20537edcfbfeSKent Overstreet prefetch(&c->opts); 20547edcfbfeSKent Overstreet prefetch((void *) &c->opts + 64); 20557edcfbfeSKent Overstreet prefetch(&inode->ei_inode); 20567edcfbfeSKent Overstreet prefetch((void *) &inode->ei_inode + 64); 20571c6fdbd8SKent Overstreet 20587edcfbfeSKent Overstreet inode_lock(&inode->v); 20597edcfbfeSKent Overstreet 20607edcfbfeSKent Overstreet ret = generic_write_checks(req, iter); 20617edcfbfeSKent Overstreet if (unlikely(ret <= 0)) 20627edcfbfeSKent Overstreet goto err; 20637edcfbfeSKent Overstreet 20647edcfbfeSKent Overstreet ret = file_remove_privs(file); 20657edcfbfeSKent Overstreet if (unlikely(ret)) 20667edcfbfeSKent Overstreet goto err; 20677edcfbfeSKent Overstreet 20687edcfbfeSKent Overstreet ret = file_update_time(file); 20697edcfbfeSKent Overstreet if (unlikely(ret)) 20707edcfbfeSKent Overstreet goto err; 20711c6fdbd8SKent Overstreet 2072919dbbd1SKent Overstreet if (unlikely((req->ki_pos|iter->count) & (block_bytes(c) - 1))) 20737edcfbfeSKent Overstreet goto err; 20747edcfbfeSKent Overstreet 20757edcfbfeSKent Overstreet inode_dio_begin(&inode->v); 20767edcfbfeSKent Overstreet bch2_pagecache_block_get(&inode->ei_pagecache_lock); 20777edcfbfeSKent Overstreet 20787edcfbfeSKent Overstreet extending = req->ki_pos + iter->count > inode->v.i_size; 20797edcfbfeSKent Overstreet if (!extending) { 20807edcfbfeSKent Overstreet inode_unlock(&inode->v); 20817edcfbfeSKent Overstreet locked = false; 20827edcfbfeSKent Overstreet } 20831c6fdbd8SKent Overstreet 20841c6fdbd8SKent Overstreet bio = bio_alloc_bioset(NULL, 20852ed5cd50SKent Overstreet iov_iter_is_bvec(iter) 20862ed5cd50SKent Overstreet ? 0 20872ed5cd50SKent Overstreet : iov_iter_npages(iter, BIO_MAX_VECS), 20881c6fdbd8SKent Overstreet REQ_OP_WRITE, 20891c6fdbd8SKent Overstreet GFP_KERNEL, 20901c6fdbd8SKent Overstreet &c->dio_write_bioset); 20919a3df993SKent Overstreet dio = container_of(bio, struct dio_write, op.wbio.bio); 2092f8f30863SKent Overstreet init_completion(&dio->done); 20931c6fdbd8SKent Overstreet dio->req = req; 2094ed484030SKent Overstreet dio->mm = current->mm; 20951c6fdbd8SKent Overstreet dio->loop = false; 20967edcfbfeSKent Overstreet dio->sync = is_sync_kiocb(req) || extending; 20971c6fdbd8SKent Overstreet dio->free_iov = false; 20981c6fdbd8SKent Overstreet dio->quota_res.sectors = 0; 2099042a1f26SKent Overstreet dio->written = 0; 21001c6fdbd8SKent Overstreet dio->iter = *iter; 21019a3df993SKent Overstreet 21021c6fdbd8SKent Overstreet ret = bch2_quota_reservation_add(c, inode, &dio->quota_res, 21031c6fdbd8SKent Overstreet iter->count >> 9, true); 21041c6fdbd8SKent Overstreet if (unlikely(ret)) 21057edcfbfeSKent Overstreet goto err_put_bio; 21061c6fdbd8SKent Overstreet 2107a023127aSKent Overstreet if (unlikely(mapping->nrpages)) { 210854847d25SKent Overstreet ret = write_invalidate_inode_pages_range(mapping, 210954847d25SKent Overstreet req->ki_pos, 211054847d25SKent Overstreet req->ki_pos + iter->count - 1); 211154847d25SKent Overstreet if (unlikely(ret)) 211254847d25SKent Overstreet goto err_put_bio; 2113a023127aSKent Overstreet } 211454847d25SKent Overstreet 21157edcfbfeSKent Overstreet ret = bch2_dio_write_loop(dio); 21161c6fdbd8SKent Overstreet err: 21177edcfbfeSKent Overstreet if (locked) 21187edcfbfeSKent Overstreet inode_unlock(&inode->v); 21197edcfbfeSKent Overstreet return ret; 21207edcfbfeSKent Overstreet err_put_bio: 21217edcfbfeSKent Overstreet bch2_pagecache_block_put(&inode->ei_pagecache_lock); 21221c6fdbd8SKent Overstreet bch2_quota_reservation_put(c, inode, &dio->quota_res); 21231c6fdbd8SKent Overstreet bio_put(bio); 21247edcfbfeSKent Overstreet inode_dio_end(&inode->v); 21257edcfbfeSKent Overstreet goto err; 21261c6fdbd8SKent Overstreet } 21271c6fdbd8SKent Overstreet 21287edcfbfeSKent Overstreet ssize_t bch2_write_iter(struct kiocb *iocb, struct iov_iter *from) 21291c6fdbd8SKent Overstreet { 21301c6fdbd8SKent Overstreet struct file *file = iocb->ki_filp; 21317edcfbfeSKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 21321c6fdbd8SKent Overstreet ssize_t ret; 21331c6fdbd8SKent Overstreet 21341c6fdbd8SKent Overstreet if (iocb->ki_flags & IOCB_DIRECT) 21351c6fdbd8SKent Overstreet return bch2_direct_write(iocb, from); 21361c6fdbd8SKent Overstreet 21377edcfbfeSKent Overstreet inode_lock(&inode->v); 21387edcfbfeSKent Overstreet 21397edcfbfeSKent Overstreet ret = generic_write_checks(iocb, from); 21407edcfbfeSKent Overstreet if (ret <= 0) 21417edcfbfeSKent Overstreet goto unlock; 21427edcfbfeSKent Overstreet 21431c6fdbd8SKent Overstreet ret = file_remove_privs(file); 21441c6fdbd8SKent Overstreet if (ret) 21457edcfbfeSKent Overstreet goto unlock; 21461c6fdbd8SKent Overstreet 21471c6fdbd8SKent Overstreet ret = file_update_time(file); 21481c6fdbd8SKent Overstreet if (ret) 21497edcfbfeSKent Overstreet goto unlock; 21501c6fdbd8SKent Overstreet 21517edcfbfeSKent Overstreet ret = bch2_buffered_write(iocb, from); 21521c6fdbd8SKent Overstreet if (likely(ret > 0)) 21531c6fdbd8SKent Overstreet iocb->ki_pos += ret; 21547edcfbfeSKent Overstreet unlock: 21551c6fdbd8SKent Overstreet inode_unlock(&inode->v); 21561c6fdbd8SKent Overstreet 21577edcfbfeSKent Overstreet if (ret > 0) 21581c6fdbd8SKent Overstreet ret = generic_write_sync(iocb, ret); 21591c6fdbd8SKent Overstreet 21601c6fdbd8SKent Overstreet return ret; 21611c6fdbd8SKent Overstreet } 21621c6fdbd8SKent Overstreet 21631c6fdbd8SKent Overstreet /* fsync: */ 21641c6fdbd8SKent Overstreet 216568a2054dSKent Overstreet /* 216668a2054dSKent Overstreet * inode->ei_inode.bi_journal_seq won't be up to date since it's set in an 216768a2054dSKent Overstreet * insert trigger: look up the btree inode instead 216868a2054dSKent Overstreet */ 216968a2054dSKent Overstreet static int bch2_flush_inode(struct bch_fs *c, subvol_inum inum) 217068a2054dSKent Overstreet { 217168a2054dSKent Overstreet struct bch_inode_unpacked inode; 217268a2054dSKent Overstreet int ret; 217368a2054dSKent Overstreet 217468a2054dSKent Overstreet if (c->opts.journal_flush_disabled) 217568a2054dSKent Overstreet return 0; 217668a2054dSKent Overstreet 217768a2054dSKent Overstreet ret = bch2_inode_find_by_inum(c, inum, &inode); 217868a2054dSKent Overstreet if (ret) 217968a2054dSKent Overstreet return ret; 218068a2054dSKent Overstreet 218168a2054dSKent Overstreet return bch2_journal_flush_seq(&c->journal, inode.bi_journal_seq); 218268a2054dSKent Overstreet } 218368a2054dSKent Overstreet 21841c6fdbd8SKent Overstreet int bch2_fsync(struct file *file, loff_t start, loff_t end, int datasync) 21851c6fdbd8SKent Overstreet { 21861c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 21871c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 218868a2054dSKent Overstreet int ret, ret2, ret3; 21891c6fdbd8SKent Overstreet 21901c6fdbd8SKent Overstreet ret = file_write_and_wait_range(file, start, end); 219168a2054dSKent Overstreet ret2 = sync_inode_metadata(&inode->v, 1); 219268a2054dSKent Overstreet ret3 = bch2_flush_inode(c, inode_inum(inode)); 21931c6fdbd8SKent Overstreet 219468a2054dSKent Overstreet return ret ?: ret2 ?: ret3; 21951c6fdbd8SKent Overstreet } 21961c6fdbd8SKent Overstreet 21971c6fdbd8SKent Overstreet /* truncate: */ 21981c6fdbd8SKent Overstreet 21996fed42bbSKent Overstreet static inline int range_has_data(struct bch_fs *c, u32 subvol, 22001c6fdbd8SKent Overstreet struct bpos start, 22011c6fdbd8SKent Overstreet struct bpos end) 22021c6fdbd8SKent Overstreet { 2203424eb881SKent Overstreet struct btree_trans trans; 220467e0dd8fSKent Overstreet struct btree_iter iter; 22051c6fdbd8SKent Overstreet struct bkey_s_c k; 22061c6fdbd8SKent Overstreet int ret = 0; 22071c6fdbd8SKent Overstreet 220820bceecbSKent Overstreet bch2_trans_init(&trans, c, 0, 0); 22096fed42bbSKent Overstreet retry: 22106fed42bbSKent Overstreet bch2_trans_begin(&trans); 22116fed42bbSKent Overstreet 22126fed42bbSKent Overstreet ret = bch2_subvolume_get_snapshot(&trans, subvol, &start.snapshot); 22136fed42bbSKent Overstreet if (ret) 22146fed42bbSKent Overstreet goto err; 2215424eb881SKent Overstreet 2216e5fa91d7SKent Overstreet for_each_btree_key_norestart(&trans, iter, BTREE_ID_extents, start, 0, k, ret) { 22171c6fdbd8SKent Overstreet if (bkey_cmp(bkey_start_pos(k.k), end) >= 0) 22181c6fdbd8SKent Overstreet break; 22191c6fdbd8SKent Overstreet 22201c6fdbd8SKent Overstreet if (bkey_extent_is_data(k.k)) { 22211c6fdbd8SKent Overstreet ret = 1; 22221c6fdbd8SKent Overstreet break; 22231c6fdbd8SKent Overstreet } 22241c6fdbd8SKent Overstreet } 22256fed42bbSKent Overstreet start = iter.pos; 222667e0dd8fSKent Overstreet bch2_trans_iter_exit(&trans, &iter); 22276fed42bbSKent Overstreet err: 22286fed42bbSKent Overstreet if (ret == -EINTR) 22296fed42bbSKent Overstreet goto retry; 22301c6fdbd8SKent Overstreet 22319a796fdbSKent Overstreet bch2_trans_exit(&trans); 22329a796fdbSKent Overstreet return ret; 22331c6fdbd8SKent Overstreet } 22341c6fdbd8SKent Overstreet 22351c6fdbd8SKent Overstreet static int __bch2_truncate_page(struct bch_inode_info *inode, 22361c6fdbd8SKent Overstreet pgoff_t index, loff_t start, loff_t end) 22371c6fdbd8SKent Overstreet { 22381c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 22391c6fdbd8SKent Overstreet struct address_space *mapping = inode->v.i_mapping; 2240a99b1cafSKent Overstreet struct bch_page_state *s; 22411c6fdbd8SKent Overstreet unsigned start_offset = start & (PAGE_SIZE - 1); 22421c6fdbd8SKent Overstreet unsigned end_offset = ((end - 1) & (PAGE_SIZE - 1)) + 1; 2243a99b1cafSKent Overstreet unsigned i; 22441c6fdbd8SKent Overstreet struct page *page; 22451c6fdbd8SKent Overstreet int ret = 0; 22461c6fdbd8SKent Overstreet 22471c6fdbd8SKent Overstreet /* Page boundary? Nothing to do */ 22481c6fdbd8SKent Overstreet if (!((index == start >> PAGE_SHIFT && start_offset) || 22491c6fdbd8SKent Overstreet (index == end >> PAGE_SHIFT && end_offset != PAGE_SIZE))) 22501c6fdbd8SKent Overstreet return 0; 22511c6fdbd8SKent Overstreet 22521c6fdbd8SKent Overstreet /* Above i_size? */ 22531c6fdbd8SKent Overstreet if (index << PAGE_SHIFT >= inode->v.i_size) 22541c6fdbd8SKent Overstreet return 0; 22551c6fdbd8SKent Overstreet 22561c6fdbd8SKent Overstreet page = find_lock_page(mapping, index); 22571c6fdbd8SKent Overstreet if (!page) { 22581c6fdbd8SKent Overstreet /* 22591c6fdbd8SKent Overstreet * XXX: we're doing two index lookups when we end up reading the 22601c6fdbd8SKent Overstreet * page 22611c6fdbd8SKent Overstreet */ 22626fed42bbSKent Overstreet ret = range_has_data(c, inode->ei_subvol, 22631c6fdbd8SKent Overstreet POS(inode->v.i_ino, index << PAGE_SECTOR_SHIFT), 22641c6fdbd8SKent Overstreet POS(inode->v.i_ino, (index + 1) << PAGE_SECTOR_SHIFT)); 22651c6fdbd8SKent Overstreet if (ret <= 0) 22661c6fdbd8SKent Overstreet return ret; 22671c6fdbd8SKent Overstreet 22681c6fdbd8SKent Overstreet page = find_or_create_page(mapping, index, GFP_KERNEL); 22691c6fdbd8SKent Overstreet if (unlikely(!page)) { 22701c6fdbd8SKent Overstreet ret = -ENOMEM; 22711c6fdbd8SKent Overstreet goto out; 22721c6fdbd8SKent Overstreet } 22731c6fdbd8SKent Overstreet } 22741c6fdbd8SKent Overstreet 2275a99b1cafSKent Overstreet s = bch2_page_state_create(page, 0); 2276a99b1cafSKent Overstreet if (!s) { 2277a99b1cafSKent Overstreet ret = -ENOMEM; 2278a99b1cafSKent Overstreet goto unlock; 2279a99b1cafSKent Overstreet } 2280a99b1cafSKent Overstreet 22811c6fdbd8SKent Overstreet if (!PageUptodate(page)) { 22821c6fdbd8SKent Overstreet ret = bch2_read_single_page(page, mapping); 22831c6fdbd8SKent Overstreet if (ret) 22841c6fdbd8SKent Overstreet goto unlock; 22851c6fdbd8SKent Overstreet } 22861c6fdbd8SKent Overstreet 2287a99b1cafSKent Overstreet if (index != start >> PAGE_SHIFT) 2288a99b1cafSKent Overstreet start_offset = 0; 2289a99b1cafSKent Overstreet if (index != end >> PAGE_SHIFT) 2290a99b1cafSKent Overstreet end_offset = PAGE_SIZE; 2291a99b1cafSKent Overstreet 2292a99b1cafSKent Overstreet for (i = round_up(start_offset, block_bytes(c)) >> 9; 2293a99b1cafSKent Overstreet i < round_down(end_offset, block_bytes(c)) >> 9; 2294a99b1cafSKent Overstreet i++) { 2295a99b1cafSKent Overstreet s->s[i].nr_replicas = 0; 2296a99b1cafSKent Overstreet s->s[i].state = SECTOR_UNALLOCATED; 2297a99b1cafSKent Overstreet } 2298a99b1cafSKent Overstreet 2299*74163da7SKent Overstreet /* 2300*74163da7SKent Overstreet * Caller needs to know whether this page will be written out by 2301*74163da7SKent Overstreet * writeback - doing an i_size update if necessary - or whether it will 2302*74163da7SKent Overstreet * be responsible for the i_size update: 2303*74163da7SKent Overstreet */ 2304*74163da7SKent Overstreet ret = s->s[(min_t(u64, inode->v.i_size - (index << PAGE_SHIFT), 2305*74163da7SKent Overstreet PAGE_SIZE) - 1) >> 9].state >= SECTOR_DIRTY; 2306*74163da7SKent Overstreet 2307a99b1cafSKent Overstreet zero_user_segment(page, start_offset, end_offset); 2308a99b1cafSKent Overstreet 23091c6fdbd8SKent Overstreet /* 23101c6fdbd8SKent Overstreet * Bit of a hack - we don't want truncate to fail due to -ENOSPC. 23111c6fdbd8SKent Overstreet * 23121c6fdbd8SKent Overstreet * XXX: because we aren't currently tracking whether the page has actual 23131c6fdbd8SKent Overstreet * data in it (vs. just 0s, or only partially written) this wrong. ick. 23141c6fdbd8SKent Overstreet */ 2315*74163da7SKent Overstreet BUG_ON(bch2_get_page_disk_reservation(c, inode, page, false)); 23161c6fdbd8SKent Overstreet 23179ba2eb25SKent Overstreet /* 23189ba2eb25SKent Overstreet * This removes any writeable userspace mappings; we need to force 23199ba2eb25SKent Overstreet * .page_mkwrite to be called again before any mmapped writes, to 23209ba2eb25SKent Overstreet * redirty the full page: 23219ba2eb25SKent Overstreet */ 23229ba2eb25SKent Overstreet page_mkclean(page); 2323d1542e03SKent Overstreet filemap_dirty_folio(mapping, page_folio(page)); 23241c6fdbd8SKent Overstreet unlock: 23251c6fdbd8SKent Overstreet unlock_page(page); 23261c6fdbd8SKent Overstreet put_page(page); 23271c6fdbd8SKent Overstreet out: 23281c6fdbd8SKent Overstreet return ret; 23291c6fdbd8SKent Overstreet } 23301c6fdbd8SKent Overstreet 23311c6fdbd8SKent Overstreet static int bch2_truncate_page(struct bch_inode_info *inode, loff_t from) 23321c6fdbd8SKent Overstreet { 23331c6fdbd8SKent Overstreet return __bch2_truncate_page(inode, from >> PAGE_SHIFT, 2334a99b1cafSKent Overstreet from, round_up(from, PAGE_SIZE)); 23351c6fdbd8SKent Overstreet } 23361c6fdbd8SKent Overstreet 2337*74163da7SKent Overstreet static int bch2_truncate_pages(struct bch_inode_info *inode, 2338*74163da7SKent Overstreet loff_t start, loff_t end) 2339*74163da7SKent Overstreet { 2340*74163da7SKent Overstreet int ret = __bch2_truncate_page(inode, start >> PAGE_SHIFT, 2341*74163da7SKent Overstreet start, end); 2342*74163da7SKent Overstreet 2343*74163da7SKent Overstreet if (ret >= 0 && 2344*74163da7SKent Overstreet start >> PAGE_SHIFT != end >> PAGE_SHIFT) 2345*74163da7SKent Overstreet ret = __bch2_truncate_page(inode, 2346*74163da7SKent Overstreet end >> PAGE_SHIFT, 2347*74163da7SKent Overstreet start, end); 2348*74163da7SKent Overstreet return ret; 2349*74163da7SKent Overstreet } 2350*74163da7SKent Overstreet 235168a507a2SKent Overstreet static int bch2_extend(struct mnt_idmap *idmap, 235268a507a2SKent Overstreet struct bch_inode_info *inode, 2353e0541a93SKent Overstreet struct bch_inode_unpacked *inode_u, 2354e0541a93SKent Overstreet struct iattr *iattr) 23551c6fdbd8SKent Overstreet { 23561c6fdbd8SKent Overstreet struct address_space *mapping = inode->v.i_mapping; 23571c6fdbd8SKent Overstreet int ret; 23581c6fdbd8SKent Overstreet 2359e0541a93SKent Overstreet /* 2360e0541a93SKent Overstreet * sync appends: 23612925fc49SKent Overstreet * 23622925fc49SKent Overstreet * this has to be done _before_ extending i_size: 2363e0541a93SKent Overstreet */ 2364e0541a93SKent Overstreet ret = filemap_write_and_wait_range(mapping, inode_u->bi_size, S64_MAX); 23651c6fdbd8SKent Overstreet if (ret) 23661c6fdbd8SKent Overstreet return ret; 23671c6fdbd8SKent Overstreet 23681c6fdbd8SKent Overstreet truncate_setsize(&inode->v, iattr->ia_size); 23691c6fdbd8SKent Overstreet 237068a507a2SKent Overstreet return bch2_setattr_nonsize(idmap, inode, iattr); 23711c6fdbd8SKent Overstreet } 23721c6fdbd8SKent Overstreet 237354e2264eSKent Overstreet static int bch2_truncate_finish_fn(struct bch_inode_info *inode, 237454e2264eSKent Overstreet struct bch_inode_unpacked *bi, 237554e2264eSKent Overstreet void *p) 237654e2264eSKent Overstreet { 237754e2264eSKent Overstreet bi->bi_flags &= ~BCH_INODE_I_SIZE_DIRTY; 237854e2264eSKent Overstreet return 0; 237954e2264eSKent Overstreet } 238054e2264eSKent Overstreet 238154e2264eSKent Overstreet static int bch2_truncate_start_fn(struct bch_inode_info *inode, 238254e2264eSKent Overstreet struct bch_inode_unpacked *bi, void *p) 238354e2264eSKent Overstreet { 238454e2264eSKent Overstreet u64 *new_i_size = p; 238554e2264eSKent Overstreet 238654e2264eSKent Overstreet bi->bi_flags |= BCH_INODE_I_SIZE_DIRTY; 238754e2264eSKent Overstreet bi->bi_size = *new_i_size; 238854e2264eSKent Overstreet return 0; 238954e2264eSKent Overstreet } 239054e2264eSKent Overstreet 239168a507a2SKent Overstreet int bch2_truncate(struct mnt_idmap *idmap, 239268a507a2SKent Overstreet struct bch_inode_info *inode, struct iattr *iattr) 23931c6fdbd8SKent Overstreet { 23941c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 23951c6fdbd8SKent Overstreet struct address_space *mapping = inode->v.i_mapping; 2396e0541a93SKent Overstreet struct bch_inode_unpacked inode_u; 239754e2264eSKent Overstreet u64 new_i_size = iattr->ia_size; 23982e87eae1SKent Overstreet s64 i_sectors_delta = 0; 23991c6fdbd8SKent Overstreet int ret = 0; 24001c6fdbd8SKent Overstreet 240168a507a2SKent Overstreet /* 240278d66ab1SDan Robertson * If the truncate call with change the size of the file, the 240378d66ab1SDan Robertson * cmtimes should be updated. If the size will not change, we 240478d66ab1SDan Robertson * do not need to update the cmtimes. 240568a507a2SKent Overstreet */ 240678d66ab1SDan Robertson if (iattr->ia_size != inode->v.i_size) { 240768a507a2SKent Overstreet if (!(iattr->ia_valid & ATTR_MTIME)) 240868a507a2SKent Overstreet ktime_get_coarse_real_ts64(&iattr->ia_mtime); 240968a507a2SKent Overstreet if (!(iattr->ia_valid & ATTR_CTIME)) 241068a507a2SKent Overstreet ktime_get_coarse_real_ts64(&iattr->ia_ctime); 241168a507a2SKent Overstreet iattr->ia_valid |= ATTR_MTIME|ATTR_CTIME; 241278d66ab1SDan Robertson } 241368a507a2SKent Overstreet 24141c6fdbd8SKent Overstreet inode_dio_wait(&inode->v); 24151c6fdbd8SKent Overstreet bch2_pagecache_block_get(&inode->ei_pagecache_lock); 24161c6fdbd8SKent Overstreet 24176fed42bbSKent Overstreet ret = bch2_inode_find_by_inum(c, inode_inum(inode), &inode_u); 2418e0541a93SKent Overstreet if (ret) 2419e0541a93SKent Overstreet goto err; 24201c6fdbd8SKent Overstreet 2421c45d473dSKent Overstreet /* 2422c45d473dSKent Overstreet * check this before next assertion; on filesystem error our normal 2423c45d473dSKent Overstreet * invariants are a bit broken (truncate has to truncate the page cache 2424c45d473dSKent Overstreet * before the inode). 2425c45d473dSKent Overstreet */ 2426c45d473dSKent Overstreet ret = bch2_journal_error(&c->journal); 2427c45d473dSKent Overstreet if (ret) 2428c45d473dSKent Overstreet goto err; 2429c45d473dSKent Overstreet 243033c74e41SKent Overstreet WARN_ON(!test_bit(EI_INODE_ERROR, &inode->ei_flags) && 243133c74e41SKent Overstreet inode->v.i_size < inode_u.bi_size); 2432e0541a93SKent Overstreet 2433e0541a93SKent Overstreet if (iattr->ia_size > inode->v.i_size) { 243468a507a2SKent Overstreet ret = bch2_extend(idmap, inode, &inode_u, iattr); 243554e2264eSKent Overstreet goto err; 24361c6fdbd8SKent Overstreet } 24371c6fdbd8SKent Overstreet 243868a507a2SKent Overstreet iattr->ia_valid &= ~ATTR_SIZE; 243968a507a2SKent Overstreet 24401c6fdbd8SKent Overstreet ret = bch2_truncate_page(inode, iattr->ia_size); 2441*74163da7SKent Overstreet if (unlikely(ret < 0)) 244254e2264eSKent Overstreet goto err; 24431c6fdbd8SKent Overstreet 24446cc3535dSKent Overstreet /* 24456cc3535dSKent Overstreet * When extending, we're going to write the new i_size to disk 24466cc3535dSKent Overstreet * immediately so we need to flush anything above the current on disk 24476cc3535dSKent Overstreet * i_size first: 24486cc3535dSKent Overstreet * 24496cc3535dSKent Overstreet * Also, when extending we need to flush the page that i_size currently 24506cc3535dSKent Overstreet * straddles - if it's mapped to userspace, we need to ensure that 24516cc3535dSKent Overstreet * userspace has to redirty it and call .mkwrite -> set_page_dirty 24526cc3535dSKent Overstreet * again to allocate the part of the page that was extended. 24536cc3535dSKent Overstreet */ 2454e0541a93SKent Overstreet if (iattr->ia_size > inode_u.bi_size) 24551c6fdbd8SKent Overstreet ret = filemap_write_and_wait_range(mapping, 2456e0541a93SKent Overstreet inode_u.bi_size, 24571c6fdbd8SKent Overstreet iattr->ia_size - 1); 24581c6fdbd8SKent Overstreet else if (iattr->ia_size & (PAGE_SIZE - 1)) 24591c6fdbd8SKent Overstreet ret = filemap_write_and_wait_range(mapping, 24601c6fdbd8SKent Overstreet round_down(iattr->ia_size, PAGE_SIZE), 24611c6fdbd8SKent Overstreet iattr->ia_size - 1); 24621c6fdbd8SKent Overstreet if (ret) 246354e2264eSKent Overstreet goto err; 24641c6fdbd8SKent Overstreet 246554e2264eSKent Overstreet mutex_lock(&inode->ei_update_lock); 246654e2264eSKent Overstreet ret = bch2_write_inode(c, inode, bch2_truncate_start_fn, 246754e2264eSKent Overstreet &new_i_size, 0); 246854e2264eSKent Overstreet mutex_unlock(&inode->ei_update_lock); 24691c6fdbd8SKent Overstreet 24701c6fdbd8SKent Overstreet if (unlikely(ret)) 247154e2264eSKent Overstreet goto err; 24721c6fdbd8SKent Overstreet 24731c6fdbd8SKent Overstreet truncate_setsize(&inode->v, iattr->ia_size); 24741c6fdbd8SKent Overstreet 24758c6d298aSKent Overstreet ret = bch2_fpunch(c, inode_inum(inode), 2476a99b1cafSKent Overstreet round_up(iattr->ia_size, block_bytes(c)) >> 9, 247768a2054dSKent Overstreet U64_MAX, &i_sectors_delta); 24782e87eae1SKent Overstreet i_sectors_acct(c, inode, NULL, i_sectors_delta); 24792e87eae1SKent Overstreet 24801c6fdbd8SKent Overstreet if (unlikely(ret)) 248154e2264eSKent Overstreet goto err; 24821c6fdbd8SKent Overstreet 248354e2264eSKent Overstreet mutex_lock(&inode->ei_update_lock); 248468a507a2SKent Overstreet ret = bch2_write_inode(c, inode, bch2_truncate_finish_fn, NULL, 0); 248554e2264eSKent Overstreet mutex_unlock(&inode->ei_update_lock); 248668a507a2SKent Overstreet 248768a507a2SKent Overstreet ret = bch2_setattr_nonsize(idmap, inode, iattr); 248854e2264eSKent Overstreet err: 24891c6fdbd8SKent Overstreet bch2_pagecache_block_put(&inode->ei_pagecache_lock); 24901c6fdbd8SKent Overstreet return ret; 24911c6fdbd8SKent Overstreet } 24921c6fdbd8SKent Overstreet 24931c6fdbd8SKent Overstreet /* fallocate: */ 24941c6fdbd8SKent Overstreet 2495050197b1SKent Overstreet static int inode_update_times_fn(struct bch_inode_info *inode, 2496050197b1SKent Overstreet struct bch_inode_unpacked *bi, void *p) 2497050197b1SKent Overstreet { 2498050197b1SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 2499050197b1SKent Overstreet 2500050197b1SKent Overstreet bi->bi_mtime = bi->bi_ctime = bch2_current_time(c); 2501050197b1SKent Overstreet return 0; 2502050197b1SKent Overstreet } 2503050197b1SKent Overstreet 25042e87eae1SKent Overstreet static long bchfs_fpunch(struct bch_inode_info *inode, loff_t offset, loff_t len) 25051c6fdbd8SKent Overstreet { 25061c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 2507*74163da7SKent Overstreet u64 end = offset + len; 2508*74163da7SKent Overstreet u64 block_start = round_up(offset, block_bytes(c)); 2509*74163da7SKent Overstreet u64 block_end = round_down(end, block_bytes(c)); 2510*74163da7SKent Overstreet bool truncated_last_page; 25111c6fdbd8SKent Overstreet int ret = 0; 25121c6fdbd8SKent Overstreet 2513*74163da7SKent Overstreet ret = bch2_truncate_pages(inode, offset, end); 2514*74163da7SKent Overstreet if (unlikely(ret < 0)) 25151c6fdbd8SKent Overstreet goto err; 25161c6fdbd8SKent Overstreet 2517*74163da7SKent Overstreet truncated_last_page = ret; 25181c6fdbd8SKent Overstreet 2519*74163da7SKent Overstreet truncate_pagecache_range(&inode->v, offset, end - 1); 25201c6fdbd8SKent Overstreet 2521*74163da7SKent Overstreet if (block_start < block_end ) { 25222e87eae1SKent Overstreet s64 i_sectors_delta = 0; 25232e87eae1SKent Overstreet 25248c6d298aSKent Overstreet ret = bch2_fpunch(c, inode_inum(inode), 2525*74163da7SKent Overstreet block_start >> 9, block_end >> 9, 25262e87eae1SKent Overstreet &i_sectors_delta); 25272e87eae1SKent Overstreet i_sectors_acct(c, inode, NULL, i_sectors_delta); 25282e87eae1SKent Overstreet } 2529050197b1SKent Overstreet 2530050197b1SKent Overstreet mutex_lock(&inode->ei_update_lock); 2531*74163da7SKent Overstreet if (end >= inode->v.i_size && !truncated_last_page) { 2532*74163da7SKent Overstreet ret = bch2_write_inode_size(c, inode, inode->v.i_size, 2533*74163da7SKent Overstreet ATTR_MTIME|ATTR_CTIME); 2534*74163da7SKent Overstreet } else { 2535050197b1SKent Overstreet ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL, 2536*74163da7SKent Overstreet ATTR_MTIME|ATTR_CTIME); 2537*74163da7SKent Overstreet } 2538050197b1SKent Overstreet mutex_unlock(&inode->ei_update_lock); 25391c6fdbd8SKent Overstreet err: 25401c6fdbd8SKent Overstreet return ret; 25411c6fdbd8SKent Overstreet } 25421c6fdbd8SKent Overstreet 25432e87eae1SKent Overstreet static long bchfs_fcollapse_finsert(struct bch_inode_info *inode, 25445f786787SKent Overstreet loff_t offset, loff_t len, 25455f786787SKent Overstreet bool insert) 25461c6fdbd8SKent Overstreet { 25471c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 25481c6fdbd8SKent Overstreet struct address_space *mapping = inode->v.i_mapping; 254907a1006aSKent Overstreet struct bkey_buf copy; 2550d69f41d6SKent Overstreet struct btree_trans trans; 255167e0dd8fSKent Overstreet struct btree_iter src, dst, del; 25525f786787SKent Overstreet loff_t shift, new_size; 25535f786787SKent Overstreet u64 src_start; 255450dc0f69SKent Overstreet int ret = 0; 25551c6fdbd8SKent Overstreet 25561c6fdbd8SKent Overstreet if ((offset | len) & (block_bytes(c) - 1)) 25571c6fdbd8SKent Overstreet return -EINVAL; 25581c6fdbd8SKent Overstreet 25595f786787SKent Overstreet if (insert) { 25605f786787SKent Overstreet if (inode->v.i_sb->s_maxbytes - inode->v.i_size < len) 2561*74163da7SKent Overstreet return -EFBIG; 25625f786787SKent Overstreet 25635f786787SKent Overstreet if (offset >= inode->v.i_size) 2564*74163da7SKent Overstreet return -EINVAL; 25655f786787SKent Overstreet 25665f786787SKent Overstreet src_start = U64_MAX; 25675f786787SKent Overstreet shift = len; 25685f786787SKent Overstreet } else { 25691c6fdbd8SKent Overstreet if (offset + len >= inode->v.i_size) 2570*74163da7SKent Overstreet return -EINVAL; 25711c6fdbd8SKent Overstreet 25725f786787SKent Overstreet src_start = offset + len; 25735f786787SKent Overstreet shift = -len; 25745f786787SKent Overstreet } 25751c6fdbd8SKent Overstreet 25765f786787SKent Overstreet new_size = inode->v.i_size + shift; 25771c6fdbd8SKent Overstreet 25781c6fdbd8SKent Overstreet ret = write_invalidate_inode_pages_range(mapping, offset, LLONG_MAX); 25791c6fdbd8SKent Overstreet if (ret) 2580*74163da7SKent Overstreet return ret; 25811c6fdbd8SKent Overstreet 25825f786787SKent Overstreet if (insert) { 25835f786787SKent Overstreet i_size_write(&inode->v, new_size); 25845f786787SKent Overstreet mutex_lock(&inode->ei_update_lock); 25855f786787SKent Overstreet ret = bch2_write_inode_size(c, inode, new_size, 25865f786787SKent Overstreet ATTR_MTIME|ATTR_CTIME); 25875f786787SKent Overstreet mutex_unlock(&inode->ei_update_lock); 25885f786787SKent Overstreet } else { 25892e87eae1SKent Overstreet s64 i_sectors_delta = 0; 25902e87eae1SKent Overstreet 25918c6d298aSKent Overstreet ret = bch2_fpunch(c, inode_inum(inode), 25922e87eae1SKent Overstreet offset >> 9, (offset + len) >> 9, 25932e87eae1SKent Overstreet &i_sectors_delta); 25942e87eae1SKent Overstreet i_sectors_acct(c, inode, NULL, i_sectors_delta); 25952e87eae1SKent Overstreet 259663095894SKent Overstreet if (ret) 2597*74163da7SKent Overstreet return ret; 25985f786787SKent Overstreet } 25998ef231bdSKent Overstreet 260050dc0f69SKent Overstreet bch2_bkey_buf_init(©); 2601f7beb4caSKent Overstreet bch2_trans_init(&trans, c, BTREE_ITER_MAX, 1024); 260267e0dd8fSKent Overstreet bch2_trans_iter_init(&trans, &src, BTREE_ID_extents, 26035f786787SKent Overstreet POS(inode->v.i_ino, src_start >> 9), 260463095894SKent Overstreet BTREE_ITER_INTENT); 260567e0dd8fSKent Overstreet bch2_trans_copy_iter(&dst, &src); 260667e0dd8fSKent Overstreet bch2_trans_copy_iter(&del, &src); 26075f786787SKent Overstreet 260850dc0f69SKent Overstreet while (ret == 0 || ret == -EINTR) { 260963095894SKent Overstreet struct disk_reservation disk_res = 261063095894SKent Overstreet bch2_disk_reservation_init(c, 0); 261163095894SKent Overstreet struct bkey_i delete; 261263095894SKent Overstreet struct bkey_s_c k; 261363095894SKent Overstreet struct bpos next_pos; 26145f786787SKent Overstreet struct bpos move_pos = POS(inode->v.i_ino, offset >> 9); 26155f786787SKent Overstreet struct bpos atomic_end; 26162d594dfbSKent Overstreet unsigned trigger_flags = 0; 26176fed42bbSKent Overstreet u32 snapshot; 26186fed42bbSKent Overstreet 26196fed42bbSKent Overstreet bch2_trans_begin(&trans); 26206fed42bbSKent Overstreet 26216fed42bbSKent Overstreet ret = bch2_subvolume_get_snapshot(&trans, 26226fed42bbSKent Overstreet inode->ei_subvol, &snapshot); 26236fed42bbSKent Overstreet if (ret) 26246fed42bbSKent Overstreet continue; 26256fed42bbSKent Overstreet 26266fed42bbSKent Overstreet bch2_btree_iter_set_snapshot(&src, snapshot); 26276fed42bbSKent Overstreet bch2_btree_iter_set_snapshot(&dst, snapshot); 26286fed42bbSKent Overstreet bch2_btree_iter_set_snapshot(&del, snapshot); 262963095894SKent Overstreet 2630700c25b3SKent Overstreet bch2_trans_begin(&trans); 2631700c25b3SKent Overstreet 26325f786787SKent Overstreet k = insert 263367e0dd8fSKent Overstreet ? bch2_btree_iter_peek_prev(&src) 263467e0dd8fSKent Overstreet : bch2_btree_iter_peek(&src); 263563095894SKent Overstreet if ((ret = bkey_err(k))) 263650dc0f69SKent Overstreet continue; 263763095894SKent Overstreet 263863095894SKent Overstreet if (!k.k || k.k->p.inode != inode->v.i_ino) 263963095894SKent Overstreet break; 264063095894SKent Overstreet 26415f786787SKent Overstreet if (insert && 26425f786787SKent Overstreet bkey_cmp(k.k->p, POS(inode->v.i_ino, offset >> 9)) <= 0) 26435f786787SKent Overstreet break; 26445f786787SKent Overstreet reassemble: 264507a1006aSKent Overstreet bch2_bkey_buf_reassemble(©, c, k); 26465f786787SKent Overstreet 26475f786787SKent Overstreet if (insert && 2648283eda57SKent Overstreet bkey_cmp(bkey_start_pos(k.k), move_pos) < 0) 264935189e09SKent Overstreet bch2_cut_front(move_pos, copy.k); 26505f786787SKent Overstreet 265135189e09SKent Overstreet copy.k->k.p.offset += shift >> 9; 265267e0dd8fSKent Overstreet bch2_btree_iter_set_pos(&dst, bkey_start_pos(©.k->k)); 26531c6fdbd8SKent Overstreet 265467e0dd8fSKent Overstreet ret = bch2_extent_atomic_end(&trans, &dst, copy.k, &atomic_end); 26553c7f3b7aSKent Overstreet if (ret) 265650dc0f69SKent Overstreet continue; 2657e2d9912cSKent Overstreet 265835189e09SKent Overstreet if (bkey_cmp(atomic_end, copy.k->k.p)) { 26595f786787SKent Overstreet if (insert) { 26605f786787SKent Overstreet move_pos = atomic_end; 26615f786787SKent Overstreet move_pos.offset -= shift >> 9; 26625f786787SKent Overstreet goto reassemble; 26635f786787SKent Overstreet } else { 2664085ab693SKent Overstreet bch2_cut_back(atomic_end, copy.k); 26655f786787SKent Overstreet } 26665f786787SKent Overstreet } 26675f786787SKent Overstreet 266863095894SKent Overstreet bkey_init(&delete.k); 2669283eda57SKent Overstreet delete.k.p = copy.k->k.p; 2670283eda57SKent Overstreet delete.k.size = copy.k->k.size; 2671283eda57SKent Overstreet delete.k.p.offset -= shift >> 9; 267267e0dd8fSKent Overstreet bch2_btree_iter_set_pos(&del, bkey_start_pos(&delete.k)); 26731c6fdbd8SKent Overstreet 26745f786787SKent Overstreet next_pos = insert ? bkey_start_pos(&delete.k) : delete.k.p; 267563095894SKent Overstreet 267635189e09SKent Overstreet if (copy.k->k.size == k.k->size) { 267763095894SKent Overstreet /* 267863095894SKent Overstreet * If we're moving the entire extent, we can skip 267963095894SKent Overstreet * running triggers: 268063095894SKent Overstreet */ 26812d594dfbSKent Overstreet trigger_flags |= BTREE_TRIGGER_NORUN; 268263095894SKent Overstreet } else { 268363095894SKent Overstreet /* We might end up splitting compressed extents: */ 268463095894SKent Overstreet unsigned nr_ptrs = 26854de77495SKent Overstreet bch2_bkey_nr_ptrs_allocated(bkey_i_to_s_c(copy.k)); 268663095894SKent Overstreet 268763095894SKent Overstreet ret = bch2_disk_reservation_get(c, &disk_res, 268835189e09SKent Overstreet copy.k->k.size, nr_ptrs, 26891c6fdbd8SKent Overstreet BCH_DISK_RESERVATION_NOFAIL); 26901c6fdbd8SKent Overstreet BUG_ON(ret); 269163095894SKent Overstreet } 26921c6fdbd8SKent Overstreet 269367e0dd8fSKent Overstreet ret = bch2_btree_iter_traverse(&del) ?: 269467e0dd8fSKent Overstreet bch2_trans_update(&trans, &del, &delete, trigger_flags) ?: 269567e0dd8fSKent Overstreet bch2_trans_update(&trans, &dst, copy.k, trigger_flags) ?: 269668a2054dSKent Overstreet bch2_trans_commit(&trans, &disk_res, NULL, 26972d594dfbSKent Overstreet BTREE_INSERT_NOFAIL); 26981c6fdbd8SKent Overstreet bch2_disk_reservation_put(c, &disk_res); 269950dc0f69SKent Overstreet 270063095894SKent Overstreet if (!ret) 270167e0dd8fSKent Overstreet bch2_btree_iter_set_pos(&src, next_pos); 270250dc0f69SKent Overstreet } 270367e0dd8fSKent Overstreet bch2_trans_iter_exit(&trans, &del); 270467e0dd8fSKent Overstreet bch2_trans_iter_exit(&trans, &dst); 270567e0dd8fSKent Overstreet bch2_trans_iter_exit(&trans, &src); 270650dc0f69SKent Overstreet bch2_trans_exit(&trans); 270750dc0f69SKent Overstreet bch2_bkey_buf_exit(©, c); 270863095894SKent Overstreet 27098ef231bdSKent Overstreet if (ret) 2710*74163da7SKent Overstreet return ret; 27111c6fdbd8SKent Overstreet 2712*74163da7SKent Overstreet mutex_lock(&inode->ei_update_lock); 27135f786787SKent Overstreet if (!insert) { 27148ef231bdSKent Overstreet i_size_write(&inode->v, new_size); 27158ef231bdSKent Overstreet ret = bch2_write_inode_size(c, inode, new_size, 27168ef231bdSKent Overstreet ATTR_MTIME|ATTR_CTIME); 2717*74163da7SKent Overstreet } else { 2718*74163da7SKent Overstreet /* We need an inode update to update bi_journal_seq for fsync: */ 2719*74163da7SKent Overstreet ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL, 2720*74163da7SKent Overstreet ATTR_MTIME|ATTR_CTIME); 27215f786787SKent Overstreet } 2722*74163da7SKent Overstreet mutex_unlock(&inode->ei_update_lock); 27231c6fdbd8SKent Overstreet return ret; 27241c6fdbd8SKent Overstreet } 27251c6fdbd8SKent Overstreet 2726694015c2SKent Overstreet static int __bchfs_fallocate(struct bch_inode_info *inode, int mode, 2727694015c2SKent Overstreet u64 start_sector, u64 end_sector) 27281c6fdbd8SKent Overstreet { 27291c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 2730190fa7afSKent Overstreet struct btree_trans trans; 273167e0dd8fSKent Overstreet struct btree_iter iter; 2732694015c2SKent Overstreet struct bpos end_pos = POS(inode->v.i_ino, end_sector); 27339a3df993SKent Overstreet unsigned replicas = io_opts(c, &inode->ei_inode).data_replicas; 2734694015c2SKent Overstreet int ret = 0; 27351c6fdbd8SKent Overstreet 2736f7beb4caSKent Overstreet bch2_trans_init(&trans, c, BTREE_ITER_MAX, 512); 27371c6fdbd8SKent Overstreet 273867e0dd8fSKent Overstreet bch2_trans_iter_init(&trans, &iter, BTREE_ID_extents, 2739694015c2SKent Overstreet POS(inode->v.i_ino, start_sector), 2740190fa7afSKent Overstreet BTREE_ITER_SLOTS|BTREE_ITER_INTENT); 27411c6fdbd8SKent Overstreet 274267e0dd8fSKent Overstreet while (!ret && bkey_cmp(iter.pos, end_pos) < 0) { 27432e87eae1SKent Overstreet s64 i_sectors_delta = 0; 27441c6fdbd8SKent Overstreet struct disk_reservation disk_res = { 0 }; 2745190fa7afSKent Overstreet struct quota_res quota_res = { 0 }; 27461c6fdbd8SKent Overstreet struct bkey_i_reservation reservation; 27471c6fdbd8SKent Overstreet struct bkey_s_c k; 2748694015c2SKent Overstreet unsigned sectors; 27496fed42bbSKent Overstreet u32 snapshot; 27501c6fdbd8SKent Overstreet 2751163e885aSKent Overstreet bch2_trans_begin(&trans); 2752a8abd3a7SKent Overstreet 27536fed42bbSKent Overstreet ret = bch2_subvolume_get_snapshot(&trans, 27546fed42bbSKent Overstreet inode->ei_subvol, &snapshot); 27556fed42bbSKent Overstreet if (ret) 27566fed42bbSKent Overstreet goto bkey_err; 27576fed42bbSKent Overstreet 27586fed42bbSKent Overstreet bch2_btree_iter_set_snapshot(&iter, snapshot); 27596fed42bbSKent Overstreet 276067e0dd8fSKent Overstreet k = bch2_btree_iter_peek_slot(&iter); 27610f238367SKent Overstreet if ((ret = bkey_err(k))) 27620f238367SKent Overstreet goto bkey_err; 27631c6fdbd8SKent Overstreet 27641c6fdbd8SKent Overstreet /* already reserved */ 276526609b61SKent Overstreet if (k.k->type == KEY_TYPE_reservation && 27661c6fdbd8SKent Overstreet bkey_s_c_to_reservation(k).v->nr_replicas >= replicas) { 276767e0dd8fSKent Overstreet bch2_btree_iter_advance(&iter); 27681c6fdbd8SKent Overstreet continue; 27691c6fdbd8SKent Overstreet } 27701c6fdbd8SKent Overstreet 2771190fa7afSKent Overstreet if (bkey_extent_is_data(k.k) && 2772190fa7afSKent Overstreet !(mode & FALLOC_FL_ZERO_RANGE)) { 277367e0dd8fSKent Overstreet bch2_btree_iter_advance(&iter); 27741c6fdbd8SKent Overstreet continue; 27751c6fdbd8SKent Overstreet } 27761c6fdbd8SKent Overstreet 27771c6fdbd8SKent Overstreet bkey_reservation_init(&reservation.k_i); 277826609b61SKent Overstreet reservation.k.type = KEY_TYPE_reservation; 27791c6fdbd8SKent Overstreet reservation.k.p = k.k->p; 27801c6fdbd8SKent Overstreet reservation.k.size = k.k->size; 27811c6fdbd8SKent Overstreet 278267e0dd8fSKent Overstreet bch2_cut_front(iter.pos, &reservation.k_i); 2783085ab693SKent Overstreet bch2_cut_back(end_pos, &reservation.k_i); 27841c6fdbd8SKent Overstreet 27851c6fdbd8SKent Overstreet sectors = reservation.k.size; 27864de77495SKent Overstreet reservation.v.nr_replicas = bch2_bkey_nr_ptrs_allocated(k); 27871c6fdbd8SKent Overstreet 27881c6fdbd8SKent Overstreet if (!bkey_extent_is_allocation(k.k)) { 27891c6fdbd8SKent Overstreet ret = bch2_quota_reservation_add(c, inode, 2790190fa7afSKent Overstreet "a_res, 27911c6fdbd8SKent Overstreet sectors, true); 27921c6fdbd8SKent Overstreet if (unlikely(ret)) 27930f238367SKent Overstreet goto bkey_err; 27941c6fdbd8SKent Overstreet } 27951c6fdbd8SKent Overstreet 27961c6fdbd8SKent Overstreet if (reservation.v.nr_replicas < replicas || 27974de77495SKent Overstreet bch2_bkey_sectors_compressed(k)) { 27981c6fdbd8SKent Overstreet ret = bch2_disk_reservation_get(c, &disk_res, sectors, 27991c6fdbd8SKent Overstreet replicas, 0); 28001c6fdbd8SKent Overstreet if (unlikely(ret)) 28010f238367SKent Overstreet goto bkey_err; 28021c6fdbd8SKent Overstreet 28031c6fdbd8SKent Overstreet reservation.v.nr_replicas = disk_res.nr_replicas; 28041c6fdbd8SKent Overstreet } 28051c6fdbd8SKent Overstreet 28068c6d298aSKent Overstreet ret = bch2_extent_update(&trans, inode_inum(inode), &iter, 28078c6d298aSKent Overstreet &reservation.k_i, 280868a2054dSKent Overstreet &disk_res, NULL, 2809a6336910SKent Overstreet 0, &i_sectors_delta, true); 28102e87eae1SKent Overstreet i_sectors_acct(c, inode, "a_res, i_sectors_delta); 28110f238367SKent Overstreet bkey_err: 2812190fa7afSKent Overstreet bch2_quota_reservation_put(c, inode, "a_res); 28131c6fdbd8SKent Overstreet bch2_disk_reservation_put(c, &disk_res); 28141c6fdbd8SKent Overstreet if (ret == -EINTR) 28151c6fdbd8SKent Overstreet ret = 0; 281650dc0f69SKent Overstreet } 2817*74163da7SKent Overstreet 2818*74163da7SKent Overstreet if (ret == -ENOSPC && (mode & FALLOC_FL_ZERO_RANGE)) { 2819*74163da7SKent Overstreet struct quota_res quota_res = { 0 }; 2820*74163da7SKent Overstreet s64 i_sectors_delta = 0; 2821*74163da7SKent Overstreet 2822*74163da7SKent Overstreet bch2_fpunch_at(&trans, &iter, inode_inum(inode), 2823*74163da7SKent Overstreet end_sector, &i_sectors_delta); 2824*74163da7SKent Overstreet i_sectors_acct(c, inode, "a_res, i_sectors_delta); 2825*74163da7SKent Overstreet bch2_quota_reservation_put(c, inode, "a_res); 2826*74163da7SKent Overstreet } 2827*74163da7SKent Overstreet 282867e0dd8fSKent Overstreet bch2_trans_iter_exit(&trans, &iter); 2829694015c2SKent Overstreet bch2_trans_exit(&trans); 2830694015c2SKent Overstreet return ret; 2831694015c2SKent Overstreet } 283250dc0f69SKent Overstreet 2833694015c2SKent Overstreet static long bchfs_fallocate(struct bch_inode_info *inode, int mode, 2834694015c2SKent Overstreet loff_t offset, loff_t len) 2835694015c2SKent Overstreet { 2836694015c2SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 2837*74163da7SKent Overstreet u64 end = offset + len; 2838*74163da7SKent Overstreet u64 block_start = round_down(offset, block_bytes(c)); 2839*74163da7SKent Overstreet u64 block_end = round_up(end, block_bytes(c)); 2840*74163da7SKent Overstreet bool truncated_last_page = false; 2841*74163da7SKent Overstreet int ret, ret2 = 0; 2842694015c2SKent Overstreet 2843694015c2SKent Overstreet if (!(mode & FALLOC_FL_KEEP_SIZE) && end > inode->v.i_size) { 2844694015c2SKent Overstreet ret = inode_newsize_ok(&inode->v, end); 2845694015c2SKent Overstreet if (ret) 2846*74163da7SKent Overstreet return ret; 2847694015c2SKent Overstreet } 2848694015c2SKent Overstreet 2849694015c2SKent Overstreet if (mode & FALLOC_FL_ZERO_RANGE) { 2850*74163da7SKent Overstreet ret = bch2_truncate_pages(inode, offset, end); 2851*74163da7SKent Overstreet if (unlikely(ret < 0)) 2852*74163da7SKent Overstreet return ret; 2853694015c2SKent Overstreet 2854*74163da7SKent Overstreet truncated_last_page = ret; 2855694015c2SKent Overstreet 2856694015c2SKent Overstreet truncate_pagecache_range(&inode->v, offset, end - 1); 2857*74163da7SKent Overstreet 2858*74163da7SKent Overstreet block_start = round_up(offset, block_bytes(c)); 2859*74163da7SKent Overstreet block_end = round_down(end, block_bytes(c)); 2860694015c2SKent Overstreet } 2861694015c2SKent Overstreet 2862694015c2SKent Overstreet ret = __bchfs_fallocate(inode, mode, block_start >> 9, block_end >> 9); 2863e0541a93SKent Overstreet 2864e0541a93SKent Overstreet /* 2865*74163da7SKent Overstreet * On -ENOSPC in ZERO_RANGE mode, we still want to do the inode update, 2866*74163da7SKent Overstreet * so that the VFS cache i_size is consistent with the btree i_size: 2867e0541a93SKent Overstreet */ 2868*74163da7SKent Overstreet if (ret && 2869*74163da7SKent Overstreet !(ret == -ENOSPC && (mode & FALLOC_FL_ZERO_RANGE))) 2870*74163da7SKent Overstreet return ret; 28711c6fdbd8SKent Overstreet 2872*74163da7SKent Overstreet if (mode & FALLOC_FL_KEEP_SIZE && end > inode->v.i_size) 2873e0541a93SKent Overstreet end = inode->v.i_size; 2874*74163da7SKent Overstreet 2875*74163da7SKent Overstreet if (end >= inode->v.i_size && 2876*74163da7SKent Overstreet (((mode & FALLOC_FL_ZERO_RANGE) && !truncated_last_page) || 2877*74163da7SKent Overstreet !(mode & FALLOC_FL_KEEP_SIZE))) { 2878*74163da7SKent Overstreet spin_lock(&inode->v.i_lock); 2879e0541a93SKent Overstreet i_size_write(&inode->v, end); 2880*74163da7SKent Overstreet spin_unlock(&inode->v.i_lock); 2881e0541a93SKent Overstreet 28821c6fdbd8SKent Overstreet mutex_lock(&inode->ei_update_lock); 2883*74163da7SKent Overstreet ret2 = bch2_write_inode_size(c, inode, end, 0); 28841c6fdbd8SKent Overstreet mutex_unlock(&inode->ei_update_lock); 28851c6fdbd8SKent Overstreet } 2886*74163da7SKent Overstreet 2887*74163da7SKent Overstreet return ret ?: ret2; 28881c6fdbd8SKent Overstreet } 28891c6fdbd8SKent Overstreet 28901c6fdbd8SKent Overstreet long bch2_fallocate_dispatch(struct file *file, int mode, 28911c6fdbd8SKent Overstreet loff_t offset, loff_t len) 28921c6fdbd8SKent Overstreet { 28931c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 28942a9101a9SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 28952a9101a9SKent Overstreet long ret; 28962a9101a9SKent Overstreet 28972a9101a9SKent Overstreet if (!percpu_ref_tryget(&c->writes)) 28982a9101a9SKent Overstreet return -EROFS; 28991c6fdbd8SKent Overstreet 2900*74163da7SKent Overstreet inode_lock(&inode->v); 2901*74163da7SKent Overstreet inode_dio_wait(&inode->v); 2902*74163da7SKent Overstreet bch2_pagecache_block_get(&inode->ei_pagecache_lock); 2903*74163da7SKent Overstreet 29041c6fdbd8SKent Overstreet if (!(mode & ~(FALLOC_FL_KEEP_SIZE|FALLOC_FL_ZERO_RANGE))) 29052a9101a9SKent Overstreet ret = bchfs_fallocate(inode, mode, offset, len); 29062a9101a9SKent Overstreet else if (mode == (FALLOC_FL_PUNCH_HOLE|FALLOC_FL_KEEP_SIZE)) 29072a9101a9SKent Overstreet ret = bchfs_fpunch(inode, offset, len); 29082a9101a9SKent Overstreet else if (mode == FALLOC_FL_INSERT_RANGE) 29092a9101a9SKent Overstreet ret = bchfs_fcollapse_finsert(inode, offset, len, true); 29102a9101a9SKent Overstreet else if (mode == FALLOC_FL_COLLAPSE_RANGE) 29112a9101a9SKent Overstreet ret = bchfs_fcollapse_finsert(inode, offset, len, false); 29122a9101a9SKent Overstreet else 29132a9101a9SKent Overstreet ret = -EOPNOTSUPP; 29141c6fdbd8SKent Overstreet 2915*74163da7SKent Overstreet 2916*74163da7SKent Overstreet bch2_pagecache_block_put(&inode->ei_pagecache_lock); 2917*74163da7SKent Overstreet inode_unlock(&inode->v); 29182a9101a9SKent Overstreet percpu_ref_put(&c->writes); 29191c6fdbd8SKent Overstreet 29202a9101a9SKent Overstreet return ret; 29211c6fdbd8SKent Overstreet } 29221c6fdbd8SKent Overstreet 292376426098SKent Overstreet static void mark_range_unallocated(struct bch_inode_info *inode, 292476426098SKent Overstreet loff_t start, loff_t end) 292576426098SKent Overstreet { 292676426098SKent Overstreet pgoff_t index = start >> PAGE_SHIFT; 292776426098SKent Overstreet pgoff_t end_index = (end - 1) >> PAGE_SHIFT; 292876426098SKent Overstreet struct folio_batch fbatch; 292976426098SKent Overstreet unsigned i, j; 293076426098SKent Overstreet 293176426098SKent Overstreet folio_batch_init(&fbatch); 293276426098SKent Overstreet 293376426098SKent Overstreet while (filemap_get_folios(inode->v.i_mapping, 293476426098SKent Overstreet &index, end_index, &fbatch)) { 293576426098SKent Overstreet for (i = 0; i < folio_batch_count(&fbatch); i++) { 293676426098SKent Overstreet struct folio *folio = fbatch.folios[i]; 293776426098SKent Overstreet struct bch_page_state *s; 293876426098SKent Overstreet 293976426098SKent Overstreet folio_lock(folio); 294076426098SKent Overstreet s = bch2_page_state(&folio->page); 294176426098SKent Overstreet 29423826ee0bSKent Overstreet if (s) { 29433826ee0bSKent Overstreet spin_lock(&s->lock); 294476426098SKent Overstreet for (j = 0; j < PAGE_SECTORS; j++) 294576426098SKent Overstreet s->s[j].nr_replicas = 0; 29463826ee0bSKent Overstreet spin_unlock(&s->lock); 29473826ee0bSKent Overstreet } 294876426098SKent Overstreet 294976426098SKent Overstreet folio_unlock(folio); 295076426098SKent Overstreet } 295176426098SKent Overstreet folio_batch_release(&fbatch); 295276426098SKent Overstreet cond_resched(); 295376426098SKent Overstreet } 295476426098SKent Overstreet } 295576426098SKent Overstreet 295676426098SKent Overstreet loff_t bch2_remap_file_range(struct file *file_src, loff_t pos_src, 295776426098SKent Overstreet struct file *file_dst, loff_t pos_dst, 295876426098SKent Overstreet loff_t len, unsigned remap_flags) 295976426098SKent Overstreet { 296076426098SKent Overstreet struct bch_inode_info *src = file_bch_inode(file_src); 296176426098SKent Overstreet struct bch_inode_info *dst = file_bch_inode(file_dst); 296276426098SKent Overstreet struct bch_fs *c = src->v.i_sb->s_fs_info; 29632e87eae1SKent Overstreet s64 i_sectors_delta = 0; 2964677fc056SKent Overstreet u64 aligned_len; 296576426098SKent Overstreet loff_t ret = 0; 296676426098SKent Overstreet 296776426098SKent Overstreet if (remap_flags & ~(REMAP_FILE_DEDUP|REMAP_FILE_ADVISORY)) 296876426098SKent Overstreet return -EINVAL; 296976426098SKent Overstreet 297076426098SKent Overstreet if (remap_flags & REMAP_FILE_DEDUP) 297176426098SKent Overstreet return -EOPNOTSUPP; 297276426098SKent Overstreet 297376426098SKent Overstreet if ((pos_src & (block_bytes(c) - 1)) || 297476426098SKent Overstreet (pos_dst & (block_bytes(c) - 1))) 297576426098SKent Overstreet return -EINVAL; 297676426098SKent Overstreet 297776426098SKent Overstreet if (src == dst && 297876426098SKent Overstreet abs(pos_src - pos_dst) < len) 297976426098SKent Overstreet return -EINVAL; 298076426098SKent Overstreet 298176426098SKent Overstreet bch2_lock_inodes(INODE_LOCK|INODE_PAGECACHE_BLOCK, src, dst); 298276426098SKent Overstreet 29832e87eae1SKent Overstreet file_update_time(file_dst); 29842e87eae1SKent Overstreet 298576426098SKent Overstreet inode_dio_wait(&src->v); 298676426098SKent Overstreet inode_dio_wait(&dst->v); 298776426098SKent Overstreet 298876426098SKent Overstreet ret = generic_remap_file_range_prep(file_src, pos_src, 298976426098SKent Overstreet file_dst, pos_dst, 299076426098SKent Overstreet &len, remap_flags); 299176426098SKent Overstreet if (ret < 0 || len == 0) 29922e87eae1SKent Overstreet goto err; 299376426098SKent Overstreet 2994677fc056SKent Overstreet aligned_len = round_up((u64) len, block_bytes(c)); 299576426098SKent Overstreet 299676426098SKent Overstreet ret = write_invalidate_inode_pages_range(dst->v.i_mapping, 2997677fc056SKent Overstreet pos_dst, pos_dst + len - 1); 299876426098SKent Overstreet if (ret) 29992e87eae1SKent Overstreet goto err; 300076426098SKent Overstreet 300176426098SKent Overstreet mark_range_unallocated(src, pos_src, pos_src + aligned_len); 300276426098SKent Overstreet 30032e87eae1SKent Overstreet ret = bch2_remap_range(c, 30046fed42bbSKent Overstreet inode_inum(dst), pos_dst >> 9, 30056fed42bbSKent Overstreet inode_inum(src), pos_src >> 9, 300676426098SKent Overstreet aligned_len >> 9, 30072e87eae1SKent Overstreet pos_dst + len, &i_sectors_delta); 30082e87eae1SKent Overstreet if (ret < 0) 30092e87eae1SKent Overstreet goto err; 301076426098SKent Overstreet 30112e87eae1SKent Overstreet /* 30122e87eae1SKent Overstreet * due to alignment, we might have remapped slightly more than requsted 30132e87eae1SKent Overstreet */ 3014677fc056SKent Overstreet ret = min((u64) ret << 9, (u64) len); 30152e87eae1SKent Overstreet 30162e87eae1SKent Overstreet /* XXX get a quota reservation */ 30172e87eae1SKent Overstreet i_sectors_acct(c, dst, NULL, i_sectors_delta); 30182e87eae1SKent Overstreet 30192e87eae1SKent Overstreet spin_lock(&dst->v.i_lock); 3020677fc056SKent Overstreet if (pos_dst + ret > dst->v.i_size) 3021677fc056SKent Overstreet i_size_write(&dst->v, pos_dst + ret); 30222e87eae1SKent Overstreet spin_unlock(&dst->v.i_lock); 3023e7084c9cSKent Overstreet 302468a2054dSKent Overstreet if ((file_dst->f_flags & (__O_SYNC | O_DSYNC)) || 302568a2054dSKent Overstreet IS_SYNC(file_inode(file_dst))) 302668a2054dSKent Overstreet ret = bch2_flush_inode(c, inode_inum(dst)); 30272e87eae1SKent Overstreet err: 302876426098SKent Overstreet bch2_unlock_inodes(INODE_LOCK|INODE_PAGECACHE_BLOCK, src, dst); 302976426098SKent Overstreet 303076426098SKent Overstreet return ret; 303176426098SKent Overstreet } 303276426098SKent Overstreet 30331c6fdbd8SKent Overstreet /* fseek: */ 30341c6fdbd8SKent Overstreet 3035543ef2ebSKent Overstreet static int folio_data_offset(struct folio *folio, unsigned offset) 30361c6fdbd8SKent Overstreet { 3037f57a6a5dSKent Overstreet struct bch_page_state *s = bch2_page_state(&folio->page); 3038f57a6a5dSKent Overstreet unsigned i; 3039f81b648dSKent Overstreet 3040543ef2ebSKent Overstreet if (s) 3041543ef2ebSKent Overstreet for (i = offset >> 9; i < PAGE_SECTORS; i++) 3042f57a6a5dSKent Overstreet if (s->s[i].state >= SECTOR_DIRTY) 3043543ef2ebSKent Overstreet return i << 9; 3044f57a6a5dSKent Overstreet 3045543ef2ebSKent Overstreet return -1; 30461c6fdbd8SKent Overstreet } 30471c6fdbd8SKent Overstreet 3048543ef2ebSKent Overstreet static loff_t bch2_seek_pagecache_data(struct inode *vinode, 30491c6fdbd8SKent Overstreet loff_t start_offset, 30501c6fdbd8SKent Overstreet loff_t end_offset) 30511c6fdbd8SKent Overstreet { 30521c6fdbd8SKent Overstreet struct folio_batch fbatch; 30531c6fdbd8SKent Overstreet pgoff_t start_index = start_offset >> PAGE_SHIFT; 30541c6fdbd8SKent Overstreet pgoff_t end_index = end_offset >> PAGE_SHIFT; 30551c6fdbd8SKent Overstreet pgoff_t index = start_index; 30561c6fdbd8SKent Overstreet unsigned i; 3057543ef2ebSKent Overstreet loff_t ret; 3058543ef2ebSKent Overstreet int offset; 30591c6fdbd8SKent Overstreet 30601c6fdbd8SKent Overstreet folio_batch_init(&fbatch); 30611c6fdbd8SKent Overstreet 30621c6fdbd8SKent Overstreet while (filemap_get_folios(vinode->i_mapping, 30631c6fdbd8SKent Overstreet &index, end_index, &fbatch)) { 30641c6fdbd8SKent Overstreet for (i = 0; i < folio_batch_count(&fbatch); i++) { 30651c6fdbd8SKent Overstreet struct folio *folio = fbatch.folios[i]; 30661c6fdbd8SKent Overstreet 30671c6fdbd8SKent Overstreet folio_lock(folio); 3068543ef2ebSKent Overstreet offset = folio_data_offset(folio, 3069543ef2ebSKent Overstreet folio->index == start_index 3070543ef2ebSKent Overstreet ? start_offset & (PAGE_SIZE - 1) 3071543ef2ebSKent Overstreet : 0); 3072543ef2ebSKent Overstreet if (offset >= 0) { 3073543ef2ebSKent Overstreet ret = clamp(((loff_t) folio->index << PAGE_SHIFT) + 3074543ef2ebSKent Overstreet offset, 3075543ef2ebSKent Overstreet start_offset, end_offset); 30761c6fdbd8SKent Overstreet folio_unlock(folio); 30771c6fdbd8SKent Overstreet folio_batch_release(&fbatch); 3078543ef2ebSKent Overstreet return ret; 30791c6fdbd8SKent Overstreet } 30801c6fdbd8SKent Overstreet folio_unlock(folio); 30811c6fdbd8SKent Overstreet } 30821c6fdbd8SKent Overstreet folio_batch_release(&fbatch); 30831c6fdbd8SKent Overstreet cond_resched(); 30841c6fdbd8SKent Overstreet } 30851c6fdbd8SKent Overstreet 30861c6fdbd8SKent Overstreet return end_offset; 30871c6fdbd8SKent Overstreet } 30881c6fdbd8SKent Overstreet 30891c6fdbd8SKent Overstreet static loff_t bch2_seek_data(struct file *file, u64 offset) 30901c6fdbd8SKent Overstreet { 30911c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 30921c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 3093424eb881SKent Overstreet struct btree_trans trans; 309467e0dd8fSKent Overstreet struct btree_iter iter; 30951c6fdbd8SKent Overstreet struct bkey_s_c k; 30966fed42bbSKent Overstreet subvol_inum inum = inode_inum(inode); 30971c6fdbd8SKent Overstreet u64 isize, next_data = MAX_LFS_FILESIZE; 30986fed42bbSKent Overstreet u32 snapshot; 30991c6fdbd8SKent Overstreet int ret; 31001c6fdbd8SKent Overstreet 31011c6fdbd8SKent Overstreet isize = i_size_read(&inode->v); 31021c6fdbd8SKent Overstreet if (offset >= isize) 31031c6fdbd8SKent Overstreet return -ENXIO; 31041c6fdbd8SKent Overstreet 310520bceecbSKent Overstreet bch2_trans_init(&trans, c, 0, 0); 31066fed42bbSKent Overstreet retry: 31076fed42bbSKent Overstreet bch2_trans_begin(&trans); 31086fed42bbSKent Overstreet 31096fed42bbSKent Overstreet ret = bch2_subvolume_get_snapshot(&trans, inum.subvol, &snapshot); 31106fed42bbSKent Overstreet if (ret) 31116fed42bbSKent Overstreet goto err; 3112424eb881SKent Overstreet 3113e5fa91d7SKent Overstreet for_each_btree_key_norestart(&trans, iter, BTREE_ID_extents, 31146fed42bbSKent Overstreet SPOS(inode->v.i_ino, offset >> 9, snapshot), 0, k, ret) { 31151c6fdbd8SKent Overstreet if (k.k->p.inode != inode->v.i_ino) { 31161c6fdbd8SKent Overstreet break; 31171c6fdbd8SKent Overstreet } else if (bkey_extent_is_data(k.k)) { 31181c6fdbd8SKent Overstreet next_data = max(offset, bkey_start_offset(k.k) << 9); 31191c6fdbd8SKent Overstreet break; 31201c6fdbd8SKent Overstreet } else if (k.k->p.offset >> 9 > isize) 31211c6fdbd8SKent Overstreet break; 31221c6fdbd8SKent Overstreet } 312367e0dd8fSKent Overstreet bch2_trans_iter_exit(&trans, &iter); 31246fed42bbSKent Overstreet err: 31256fed42bbSKent Overstreet if (ret == -EINTR) 31266fed42bbSKent Overstreet goto retry; 31271c6fdbd8SKent Overstreet 31289a796fdbSKent Overstreet bch2_trans_exit(&trans); 31291c6fdbd8SKent Overstreet if (ret) 31301c6fdbd8SKent Overstreet return ret; 31311c6fdbd8SKent Overstreet 31321c6fdbd8SKent Overstreet if (next_data > offset) 3133543ef2ebSKent Overstreet next_data = bch2_seek_pagecache_data(&inode->v, 31341c6fdbd8SKent Overstreet offset, next_data); 31351c6fdbd8SKent Overstreet 3136e10d3094SKent Overstreet if (next_data >= isize) 31371c6fdbd8SKent Overstreet return -ENXIO; 31381c6fdbd8SKent Overstreet 31391c6fdbd8SKent Overstreet return vfs_setpos(file, next_data, MAX_LFS_FILESIZE); 31401c6fdbd8SKent Overstreet } 31411c6fdbd8SKent Overstreet 3142543ef2ebSKent Overstreet static int __page_hole_offset(struct page *page, unsigned offset) 31431c6fdbd8SKent Overstreet { 3144543ef2ebSKent Overstreet struct bch_page_state *s = bch2_page_state(page); 3145543ef2ebSKent Overstreet unsigned i; 3146543ef2ebSKent Overstreet 3147543ef2ebSKent Overstreet if (!s) 3148543ef2ebSKent Overstreet return 0; 3149543ef2ebSKent Overstreet 3150543ef2ebSKent Overstreet for (i = offset >> 9; i < PAGE_SECTORS; i++) 3151543ef2ebSKent Overstreet if (s->s[i].state < SECTOR_DIRTY) 3152543ef2ebSKent Overstreet return i << 9; 3153543ef2ebSKent Overstreet 3154543ef2ebSKent Overstreet return -1; 3155543ef2ebSKent Overstreet } 3156543ef2ebSKent Overstreet 3157543ef2ebSKent Overstreet static loff_t page_hole_offset(struct address_space *mapping, loff_t offset) 3158543ef2ebSKent Overstreet { 3159543ef2ebSKent Overstreet pgoff_t index = offset >> PAGE_SHIFT; 31601c6fdbd8SKent Overstreet struct page *page; 3161543ef2ebSKent Overstreet int pg_offset; 3162543ef2ebSKent Overstreet loff_t ret = -1; 31631c6fdbd8SKent Overstreet 31641c6fdbd8SKent Overstreet page = find_lock_page(mapping, index); 31651c6fdbd8SKent Overstreet if (!page) 3166543ef2ebSKent Overstreet return offset; 31671c6fdbd8SKent Overstreet 3168543ef2ebSKent Overstreet pg_offset = __page_hole_offset(page, offset & (PAGE_SIZE - 1)); 3169543ef2ebSKent Overstreet if (pg_offset >= 0) 3170543ef2ebSKent Overstreet ret = ((loff_t) index << PAGE_SHIFT) + pg_offset; 3171543ef2ebSKent Overstreet 31721c6fdbd8SKent Overstreet unlock_page(page); 31731c6fdbd8SKent Overstreet 31741c6fdbd8SKent Overstreet return ret; 31751c6fdbd8SKent Overstreet } 31761c6fdbd8SKent Overstreet 3177543ef2ebSKent Overstreet static loff_t bch2_seek_pagecache_hole(struct inode *vinode, 31781c6fdbd8SKent Overstreet loff_t start_offset, 31791c6fdbd8SKent Overstreet loff_t end_offset) 31801c6fdbd8SKent Overstreet { 31811c6fdbd8SKent Overstreet struct address_space *mapping = vinode->i_mapping; 3182543ef2ebSKent Overstreet loff_t offset = start_offset, hole; 31831c6fdbd8SKent Overstreet 3184543ef2ebSKent Overstreet while (offset < end_offset) { 3185543ef2ebSKent Overstreet hole = page_hole_offset(mapping, offset); 3186543ef2ebSKent Overstreet if (hole >= 0 && hole <= end_offset) 3187543ef2ebSKent Overstreet return max(start_offset, hole); 3188543ef2ebSKent Overstreet 3189543ef2ebSKent Overstreet offset += PAGE_SIZE; 3190543ef2ebSKent Overstreet offset &= PAGE_MASK; 3191543ef2ebSKent Overstreet } 31921c6fdbd8SKent Overstreet 31931c6fdbd8SKent Overstreet return end_offset; 31941c6fdbd8SKent Overstreet } 31951c6fdbd8SKent Overstreet 31961c6fdbd8SKent Overstreet static loff_t bch2_seek_hole(struct file *file, u64 offset) 31971c6fdbd8SKent Overstreet { 31981c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 31991c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 3200424eb881SKent Overstreet struct btree_trans trans; 320167e0dd8fSKent Overstreet struct btree_iter iter; 32021c6fdbd8SKent Overstreet struct bkey_s_c k; 32036fed42bbSKent Overstreet subvol_inum inum = inode_inum(inode); 32041c6fdbd8SKent Overstreet u64 isize, next_hole = MAX_LFS_FILESIZE; 32056fed42bbSKent Overstreet u32 snapshot; 32061c6fdbd8SKent Overstreet int ret; 32071c6fdbd8SKent Overstreet 32081c6fdbd8SKent Overstreet isize = i_size_read(&inode->v); 32091c6fdbd8SKent Overstreet if (offset >= isize) 32101c6fdbd8SKent Overstreet return -ENXIO; 32111c6fdbd8SKent Overstreet 321220bceecbSKent Overstreet bch2_trans_init(&trans, c, 0, 0); 32136fed42bbSKent Overstreet retry: 32146fed42bbSKent Overstreet bch2_trans_begin(&trans); 32156fed42bbSKent Overstreet 32166fed42bbSKent Overstreet ret = bch2_subvolume_get_snapshot(&trans, inum.subvol, &snapshot); 32176fed42bbSKent Overstreet if (ret) 32186fed42bbSKent Overstreet goto err; 3219424eb881SKent Overstreet 3220e5fa91d7SKent Overstreet for_each_btree_key_norestart(&trans, iter, BTREE_ID_extents, 32216fed42bbSKent Overstreet SPOS(inode->v.i_ino, offset >> 9, snapshot), 322294f651e2SKent Overstreet BTREE_ITER_SLOTS, k, ret) { 32231c6fdbd8SKent Overstreet if (k.k->p.inode != inode->v.i_ino) { 3224543ef2ebSKent Overstreet next_hole = bch2_seek_pagecache_hole(&inode->v, 32251c6fdbd8SKent Overstreet offset, MAX_LFS_FILESIZE); 32261c6fdbd8SKent Overstreet break; 32271c6fdbd8SKent Overstreet } else if (!bkey_extent_is_data(k.k)) { 3228543ef2ebSKent Overstreet next_hole = bch2_seek_pagecache_hole(&inode->v, 32291c6fdbd8SKent Overstreet max(offset, bkey_start_offset(k.k) << 9), 32301c6fdbd8SKent Overstreet k.k->p.offset << 9); 32311c6fdbd8SKent Overstreet 32321c6fdbd8SKent Overstreet if (next_hole < k.k->p.offset << 9) 32331c6fdbd8SKent Overstreet break; 32341c6fdbd8SKent Overstreet } else { 32351c6fdbd8SKent Overstreet offset = max(offset, bkey_start_offset(k.k) << 9); 32361c6fdbd8SKent Overstreet } 32371c6fdbd8SKent Overstreet } 323867e0dd8fSKent Overstreet bch2_trans_iter_exit(&trans, &iter); 32396fed42bbSKent Overstreet err: 32406fed42bbSKent Overstreet if (ret == -EINTR) 32416fed42bbSKent Overstreet goto retry; 32421c6fdbd8SKent Overstreet 32439a796fdbSKent Overstreet bch2_trans_exit(&trans); 32441c6fdbd8SKent Overstreet if (ret) 32451c6fdbd8SKent Overstreet return ret; 32461c6fdbd8SKent Overstreet 32471c6fdbd8SKent Overstreet if (next_hole > isize) 32481c6fdbd8SKent Overstreet next_hole = isize; 32491c6fdbd8SKent Overstreet 32501c6fdbd8SKent Overstreet return vfs_setpos(file, next_hole, MAX_LFS_FILESIZE); 32511c6fdbd8SKent Overstreet } 32521c6fdbd8SKent Overstreet 32531c6fdbd8SKent Overstreet loff_t bch2_llseek(struct file *file, loff_t offset, int whence) 32541c6fdbd8SKent Overstreet { 32551c6fdbd8SKent Overstreet switch (whence) { 32561c6fdbd8SKent Overstreet case SEEK_SET: 32571c6fdbd8SKent Overstreet case SEEK_CUR: 32581c6fdbd8SKent Overstreet case SEEK_END: 32591c6fdbd8SKent Overstreet return generic_file_llseek(file, offset, whence); 32601c6fdbd8SKent Overstreet case SEEK_DATA: 32611c6fdbd8SKent Overstreet return bch2_seek_data(file, offset); 32621c6fdbd8SKent Overstreet case SEEK_HOLE: 32631c6fdbd8SKent Overstreet return bch2_seek_hole(file, offset); 32641c6fdbd8SKent Overstreet } 32651c6fdbd8SKent Overstreet 32661c6fdbd8SKent Overstreet return -EINVAL; 32671c6fdbd8SKent Overstreet } 32681c6fdbd8SKent Overstreet 32691c6fdbd8SKent Overstreet void bch2_fs_fsio_exit(struct bch_fs *c) 32701c6fdbd8SKent Overstreet { 32711c6fdbd8SKent Overstreet bioset_exit(&c->dio_write_bioset); 32721c6fdbd8SKent Overstreet bioset_exit(&c->dio_read_bioset); 32731c6fdbd8SKent Overstreet bioset_exit(&c->writepage_bioset); 32741c6fdbd8SKent Overstreet } 32751c6fdbd8SKent Overstreet 32761c6fdbd8SKent Overstreet int bch2_fs_fsio_init(struct bch_fs *c) 32771c6fdbd8SKent Overstreet { 32781c6fdbd8SKent Overstreet int ret = 0; 32791c6fdbd8SKent Overstreet 32801c6fdbd8SKent Overstreet pr_verbose_init(c->opts, ""); 32811c6fdbd8SKent Overstreet 32821c6fdbd8SKent Overstreet if (bioset_init(&c->writepage_bioset, 32839a3df993SKent Overstreet 4, offsetof(struct bch_writepage_io, op.wbio.bio), 32841c6fdbd8SKent Overstreet BIOSET_NEED_BVECS) || 32851c6fdbd8SKent Overstreet bioset_init(&c->dio_read_bioset, 32861c6fdbd8SKent Overstreet 4, offsetof(struct dio_read, rbio.bio), 32871c6fdbd8SKent Overstreet BIOSET_NEED_BVECS) || 32881c6fdbd8SKent Overstreet bioset_init(&c->dio_write_bioset, 32899a3df993SKent Overstreet 4, offsetof(struct dio_write, op.wbio.bio), 32901c6fdbd8SKent Overstreet BIOSET_NEED_BVECS)) 32911c6fdbd8SKent Overstreet ret = -ENOMEM; 32921c6fdbd8SKent Overstreet 32931c6fdbd8SKent Overstreet pr_verbose_init(c->opts, "ret %i", ret); 32941c6fdbd8SKent Overstreet return ret; 32951c6fdbd8SKent Overstreet } 32961c6fdbd8SKent Overstreet 32971c6fdbd8SKent Overstreet #endif /* NO_BCACHEFS_FS */ 3298