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); 234b33bf1bcSKent Overstreet bch2_fs_inconsistent_on((s64) inode->v.i_blocks + sectors < 0, c, 235b33bf1bcSKent Overstreet "inode %lu i_blocks underflow: %llu + %lli < 0 (ondisk %lli)", 236b33bf1bcSKent Overstreet inode->v.i_ino, (u64) inode->v.i_blocks, sectors, 237b33bf1bcSKent Overstreet inode->ei_inode.bi_sectors); 238b44a66a6SKent Overstreet inode->v.i_blocks += sectors; 239b44a66a6SKent Overstreet 2401c6fdbd8SKent Overstreet #ifdef CONFIG_BCACHEFS_QUOTA 2411c6fdbd8SKent Overstreet if (quota_res && sectors > 0) { 2421c6fdbd8SKent Overstreet BUG_ON(sectors > quota_res->sectors); 2431c6fdbd8SKent Overstreet BUG_ON(sectors > inode->ei_quota_reserved); 2441c6fdbd8SKent Overstreet 2451c6fdbd8SKent Overstreet quota_res->sectors -= sectors; 2461c6fdbd8SKent Overstreet inode->ei_quota_reserved -= sectors; 2471c6fdbd8SKent Overstreet } else { 24826609b61SKent Overstreet bch2_quota_acct(c, inode->ei_qid, Q_SPC, sectors, KEY_TYPE_QUOTA_WARN); 2491c6fdbd8SKent Overstreet } 2501c6fdbd8SKent Overstreet #endif 2511c6fdbd8SKent Overstreet mutex_unlock(&inode->ei_quota_lock); 2521c6fdbd8SKent Overstreet } 2531c6fdbd8SKent Overstreet 2541c6fdbd8SKent Overstreet /* page state: */ 2551c6fdbd8SKent Overstreet 2561c6fdbd8SKent Overstreet /* stored in page->private: */ 2571c6fdbd8SKent Overstreet 258f57a6a5dSKent Overstreet struct bch_page_sector { 259b44a66a6SKent Overstreet /* Uncompressed, fully allocated replicas (or on disk reservation): */ 260b44a66a6SKent Overstreet unsigned nr_replicas:4; 261f81b648dSKent Overstreet 262b44a66a6SKent Overstreet /* Owns PAGE_SECTORS * replicas_reserved sized in memory reservation: */ 263b44a66a6SKent Overstreet unsigned replicas_reserved:4; 2641c6fdbd8SKent Overstreet 265f57a6a5dSKent Overstreet /* i_sectors: */ 266f57a6a5dSKent Overstreet enum { 267f57a6a5dSKent Overstreet SECTOR_UNALLOCATED, 2682ba5d38bSKent Overstreet SECTOR_RESERVED, 269f57a6a5dSKent Overstreet SECTOR_DIRTY, 270b44a66a6SKent Overstreet SECTOR_DIRTY_RESERVED, 271f57a6a5dSKent Overstreet SECTOR_ALLOCATED, 272b44a66a6SKent Overstreet } state:8; 2731c6fdbd8SKent Overstreet }; 2741c6fdbd8SKent Overstreet 275f57a6a5dSKent Overstreet struct bch_page_state { 2763826ee0bSKent Overstreet spinlock_t lock; 2777f5e31e1SKent Overstreet atomic_t write_count; 278e6ec361fSKent Overstreet bool uptodate; 279f57a6a5dSKent Overstreet struct bch_page_sector s[PAGE_SECTORS]; 280f57a6a5dSKent Overstreet }; 281f57a6a5dSKent Overstreet 282f57a6a5dSKent Overstreet static inline struct bch_page_state *__bch2_page_state(struct page *page) 2831c6fdbd8SKent Overstreet { 284f57a6a5dSKent Overstreet return page_has_private(page) 285f57a6a5dSKent Overstreet ? (struct bch_page_state *) page_private(page) 286f57a6a5dSKent Overstreet : NULL; 287f57a6a5dSKent Overstreet } 2881c6fdbd8SKent Overstreet 289f57a6a5dSKent Overstreet static inline struct bch_page_state *bch2_page_state(struct page *page) 290f57a6a5dSKent Overstreet { 291e1036a2aSKent Overstreet EBUG_ON(!PageLocked(page)); 2921c6fdbd8SKent Overstreet 293f57a6a5dSKent Overstreet return __bch2_page_state(page); 294f57a6a5dSKent Overstreet } 295f57a6a5dSKent Overstreet 296f57a6a5dSKent Overstreet /* for newly allocated pages: */ 297f57a6a5dSKent Overstreet static void __bch2_page_state_release(struct page *page) 298f57a6a5dSKent Overstreet { 29900276f9fSMatthew Wilcox (Oracle) kfree(detach_page_private(page)); 300f57a6a5dSKent Overstreet } 301f57a6a5dSKent Overstreet 302f57a6a5dSKent Overstreet static void bch2_page_state_release(struct page *page) 303f57a6a5dSKent Overstreet { 30400276f9fSMatthew Wilcox (Oracle) EBUG_ON(!PageLocked(page)); 30500276f9fSMatthew Wilcox (Oracle) __bch2_page_state_release(page); 306f57a6a5dSKent Overstreet } 307f57a6a5dSKent Overstreet 308f57a6a5dSKent Overstreet /* for newly allocated pages: */ 309f57a6a5dSKent Overstreet static struct bch_page_state *__bch2_page_state_create(struct page *page, 310f57a6a5dSKent Overstreet gfp_t gfp) 311f57a6a5dSKent Overstreet { 312f57a6a5dSKent Overstreet struct bch_page_state *s; 313f57a6a5dSKent Overstreet 314f57a6a5dSKent Overstreet s = kzalloc(sizeof(*s), GFP_NOFS|gfp); 315f57a6a5dSKent Overstreet if (!s) 316f57a6a5dSKent Overstreet return NULL; 317f57a6a5dSKent Overstreet 3183826ee0bSKent Overstreet spin_lock_init(&s->lock); 31900276f9fSMatthew Wilcox (Oracle) attach_page_private(page, s); 3201c6fdbd8SKent Overstreet return s; 3211c6fdbd8SKent Overstreet } 3221c6fdbd8SKent Overstreet 323f57a6a5dSKent Overstreet static struct bch_page_state *bch2_page_state_create(struct page *page, 324f57a6a5dSKent Overstreet gfp_t gfp) 325f57a6a5dSKent Overstreet { 326f57a6a5dSKent Overstreet return bch2_page_state(page) ?: __bch2_page_state_create(page, gfp); 327f57a6a5dSKent Overstreet } 328f57a6a5dSKent Overstreet 329b44a66a6SKent Overstreet static unsigned bkey_to_sector_state(const struct bkey *k) 330b44a66a6SKent Overstreet { 331b44a66a6SKent Overstreet if (k->type == KEY_TYPE_reservation) 332b44a66a6SKent Overstreet return SECTOR_RESERVED; 333b44a66a6SKent Overstreet if (bkey_extent_is_allocation(k)) 334b44a66a6SKent Overstreet return SECTOR_ALLOCATED; 335b44a66a6SKent Overstreet return SECTOR_UNALLOCATED; 336b44a66a6SKent Overstreet } 337b44a66a6SKent Overstreet 338e6ec361fSKent Overstreet static void __bch2_page_state_set(struct page *page, 339e6ec361fSKent Overstreet unsigned pg_offset, unsigned pg_len, 340e6ec361fSKent Overstreet unsigned nr_ptrs, unsigned state) 341e6ec361fSKent Overstreet { 342e6ec361fSKent Overstreet struct bch_page_state *s = bch2_page_state_create(page, __GFP_NOFAIL); 343e6ec361fSKent Overstreet unsigned i; 344e6ec361fSKent Overstreet 345e6ec361fSKent Overstreet BUG_ON(pg_offset >= PAGE_SECTORS); 346e6ec361fSKent Overstreet BUG_ON(pg_offset + pg_len > PAGE_SECTORS); 347e6ec361fSKent Overstreet 348e6ec361fSKent Overstreet spin_lock(&s->lock); 349e6ec361fSKent Overstreet 350e6ec361fSKent Overstreet for (i = pg_offset; i < pg_offset + pg_len; i++) { 351e6ec361fSKent Overstreet s->s[i].nr_replicas = nr_ptrs; 352e6ec361fSKent Overstreet s->s[i].state = state; 353e6ec361fSKent Overstreet } 354e6ec361fSKent Overstreet 355e6ec361fSKent Overstreet if (i == PAGE_SECTORS) 356e6ec361fSKent Overstreet s->uptodate = true; 357e6ec361fSKent Overstreet 358e6ec361fSKent Overstreet spin_unlock(&s->lock); 359e6ec361fSKent Overstreet } 360e6ec361fSKent Overstreet 361e6ec361fSKent Overstreet static int bch2_page_state_set(struct bch_fs *c, subvol_inum inum, 362e6ec361fSKent Overstreet struct page **pages, unsigned nr_pages) 363e6ec361fSKent Overstreet { 364e6ec361fSKent Overstreet struct btree_trans trans; 365e6ec361fSKent Overstreet struct btree_iter iter; 366e6ec361fSKent Overstreet struct bkey_s_c k; 367e6ec361fSKent Overstreet u64 offset = pages[0]->index << PAGE_SECTORS_SHIFT; 368e6ec361fSKent Overstreet unsigned pg_idx = 0; 369e6ec361fSKent Overstreet u32 snapshot; 370e6ec361fSKent Overstreet int ret; 371e6ec361fSKent Overstreet 372e6ec361fSKent Overstreet bch2_trans_init(&trans, c, 0, 0); 373e6ec361fSKent Overstreet retry: 374e6ec361fSKent Overstreet bch2_trans_begin(&trans); 375e6ec361fSKent Overstreet 376e6ec361fSKent Overstreet ret = bch2_subvolume_get_snapshot(&trans, inum.subvol, &snapshot); 377e6ec361fSKent Overstreet if (ret) 378e6ec361fSKent Overstreet goto err; 379e6ec361fSKent Overstreet 380e6ec361fSKent Overstreet for_each_btree_key_norestart(&trans, iter, BTREE_ID_extents, 381e6ec361fSKent Overstreet SPOS(inum.inum, offset, snapshot), 382e6ec361fSKent Overstreet BTREE_ITER_SLOTS, k, ret) { 383e6ec361fSKent Overstreet unsigned nr_ptrs = bch2_bkey_nr_ptrs_fully_allocated(k); 384e6ec361fSKent Overstreet unsigned state = bkey_to_sector_state(k.k); 385e6ec361fSKent Overstreet 386e6ec361fSKent Overstreet while (pg_idx < nr_pages) { 387e6ec361fSKent Overstreet struct page *page = pages[pg_idx]; 388e6ec361fSKent Overstreet u64 pg_start = page->index << PAGE_SECTORS_SHIFT; 389e6ec361fSKent Overstreet u64 pg_end = (page->index + 1) << PAGE_SECTORS_SHIFT; 390e6ec361fSKent Overstreet unsigned pg_offset = max(bkey_start_offset(k.k), pg_start) - pg_start; 391e6ec361fSKent Overstreet unsigned pg_len = min(k.k->p.offset, pg_end) - pg_offset - pg_start; 392e6ec361fSKent Overstreet 393e6ec361fSKent Overstreet BUG_ON(k.k->p.offset < pg_start); 394e6ec361fSKent Overstreet BUG_ON(bkey_start_offset(k.k) > pg_end); 395e6ec361fSKent Overstreet 396e6ec361fSKent Overstreet if (!bch2_page_state_create(page, __GFP_NOFAIL)->uptodate) 397e6ec361fSKent Overstreet __bch2_page_state_set(page, pg_offset, pg_len, nr_ptrs, state); 398e6ec361fSKent Overstreet 399e6ec361fSKent Overstreet if (k.k->p.offset < pg_end) 400e6ec361fSKent Overstreet break; 401e6ec361fSKent Overstreet pg_idx++; 402e6ec361fSKent Overstreet } 403e6ec361fSKent Overstreet 404e6ec361fSKent Overstreet if (pg_idx == nr_pages) 405e6ec361fSKent Overstreet break; 406e6ec361fSKent Overstreet } 407e6ec361fSKent Overstreet 408e6ec361fSKent Overstreet offset = iter.pos.offset; 409e6ec361fSKent Overstreet bch2_trans_iter_exit(&trans, &iter); 410e6ec361fSKent Overstreet err: 411549d173cSKent Overstreet if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) 412e6ec361fSKent Overstreet goto retry; 413e6ec361fSKent Overstreet bch2_trans_exit(&trans); 414e6ec361fSKent Overstreet 415e6ec361fSKent Overstreet return ret; 416e6ec361fSKent Overstreet } 417e6ec361fSKent Overstreet 418b44a66a6SKent Overstreet static void bch2_bio_page_state_set(struct bio *bio, struct bkey_s_c k) 419b44a66a6SKent Overstreet { 420b44a66a6SKent Overstreet struct bvec_iter iter; 421b44a66a6SKent Overstreet struct bio_vec bv; 422b44a66a6SKent Overstreet unsigned nr_ptrs = k.k->type == KEY_TYPE_reflink_v 423b44a66a6SKent Overstreet ? 0 : bch2_bkey_nr_ptrs_fully_allocated(k); 424b44a66a6SKent Overstreet unsigned state = bkey_to_sector_state(k.k); 425b44a66a6SKent Overstreet 426e6ec361fSKent Overstreet bio_for_each_segment(bv, bio, iter) 427e6ec361fSKent Overstreet __bch2_page_state_set(bv.bv_page, bv.bv_offset >> 9, 428e6ec361fSKent Overstreet bv.bv_len >> 9, nr_ptrs, state); 429b44a66a6SKent Overstreet } 430b44a66a6SKent Overstreet 431dcfc593fSKent Overstreet static void mark_pagecache_unallocated(struct bch_inode_info *inode, 432dcfc593fSKent Overstreet u64 start, u64 end) 433dcfc593fSKent Overstreet { 434dcfc593fSKent Overstreet pgoff_t index = start >> PAGE_SECTORS_SHIFT; 435dcfc593fSKent Overstreet pgoff_t end_index = (end - 1) >> PAGE_SECTORS_SHIFT; 436dcfc593fSKent Overstreet struct folio_batch fbatch; 437dcfc593fSKent Overstreet unsigned i, j; 438dcfc593fSKent Overstreet 439dcfc593fSKent Overstreet if (end <= start) 440dcfc593fSKent Overstreet return; 441dcfc593fSKent Overstreet 442dcfc593fSKent Overstreet folio_batch_init(&fbatch); 443dcfc593fSKent Overstreet 444dcfc593fSKent Overstreet while (filemap_get_folios(inode->v.i_mapping, 445dcfc593fSKent Overstreet &index, end_index, &fbatch)) { 446dcfc593fSKent Overstreet for (i = 0; i < folio_batch_count(&fbatch); i++) { 447dcfc593fSKent Overstreet struct folio *folio = fbatch.folios[i]; 448dcfc593fSKent Overstreet u64 pg_start = folio->index << PAGE_SECTORS_SHIFT; 449dcfc593fSKent Overstreet u64 pg_end = (folio->index + 1) << PAGE_SECTORS_SHIFT; 450dcfc593fSKent Overstreet unsigned pg_offset = max(start, pg_start) - pg_start; 451dcfc593fSKent Overstreet unsigned pg_len = min(end, pg_end) - pg_offset - pg_start; 452dcfc593fSKent Overstreet struct bch_page_state *s; 453dcfc593fSKent Overstreet 454dcfc593fSKent Overstreet BUG_ON(end <= pg_start); 455dcfc593fSKent Overstreet BUG_ON(pg_offset >= PAGE_SECTORS); 456dcfc593fSKent Overstreet BUG_ON(pg_offset + pg_len > PAGE_SECTORS); 457dcfc593fSKent Overstreet 458dcfc593fSKent Overstreet folio_lock(folio); 459dcfc593fSKent Overstreet s = bch2_page_state(&folio->page); 460dcfc593fSKent Overstreet 461dcfc593fSKent Overstreet if (s) { 462dcfc593fSKent Overstreet spin_lock(&s->lock); 463dcfc593fSKent Overstreet for (j = pg_offset; j < pg_offset + pg_len; j++) 464dcfc593fSKent Overstreet s->s[j].nr_replicas = 0; 465dcfc593fSKent Overstreet spin_unlock(&s->lock); 466dcfc593fSKent Overstreet } 467dcfc593fSKent Overstreet 468dcfc593fSKent Overstreet folio_unlock(folio); 469dcfc593fSKent Overstreet } 470dcfc593fSKent Overstreet folio_batch_release(&fbatch); 471dcfc593fSKent Overstreet cond_resched(); 472dcfc593fSKent Overstreet } 473dcfc593fSKent Overstreet } 474dcfc593fSKent Overstreet 475dcfc593fSKent Overstreet static void mark_pagecache_reserved(struct bch_inode_info *inode, 476dcfc593fSKent Overstreet u64 start, u64 end) 477dcfc593fSKent Overstreet { 478dcfc593fSKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 479dcfc593fSKent Overstreet pgoff_t index = start >> PAGE_SECTORS_SHIFT; 480dcfc593fSKent Overstreet pgoff_t end_index = (end - 1) >> PAGE_SECTORS_SHIFT; 481dcfc593fSKent Overstreet struct folio_batch fbatch; 482dcfc593fSKent Overstreet s64 i_sectors_delta = 0; 483dcfc593fSKent Overstreet unsigned i, j; 484dcfc593fSKent Overstreet 485dcfc593fSKent Overstreet if (end <= start) 486dcfc593fSKent Overstreet return; 487dcfc593fSKent Overstreet 488dcfc593fSKent Overstreet folio_batch_init(&fbatch); 489dcfc593fSKent Overstreet 490dcfc593fSKent Overstreet while (filemap_get_folios(inode->v.i_mapping, 491dcfc593fSKent Overstreet &index, end_index, &fbatch)) { 492dcfc593fSKent Overstreet for (i = 0; i < folio_batch_count(&fbatch); i++) { 493dcfc593fSKent Overstreet struct folio *folio = fbatch.folios[i]; 494dcfc593fSKent Overstreet u64 pg_start = folio->index << PAGE_SECTORS_SHIFT; 495dcfc593fSKent Overstreet u64 pg_end = (folio->index + 1) << PAGE_SECTORS_SHIFT; 496dcfc593fSKent Overstreet unsigned pg_offset = max(start, pg_start) - pg_start; 497dcfc593fSKent Overstreet unsigned pg_len = min(end, pg_end) - pg_offset - pg_start; 498dcfc593fSKent Overstreet struct bch_page_state *s; 499dcfc593fSKent Overstreet 500dcfc593fSKent Overstreet BUG_ON(end <= pg_start); 501dcfc593fSKent Overstreet BUG_ON(pg_offset >= PAGE_SECTORS); 502dcfc593fSKent Overstreet BUG_ON(pg_offset + pg_len > PAGE_SECTORS); 503dcfc593fSKent Overstreet 504dcfc593fSKent Overstreet folio_lock(folio); 505dcfc593fSKent Overstreet s = bch2_page_state(&folio->page); 506dcfc593fSKent Overstreet 507dcfc593fSKent Overstreet if (s) { 508dcfc593fSKent Overstreet spin_lock(&s->lock); 509dcfc593fSKent Overstreet for (j = pg_offset; j < pg_offset + pg_len; j++) 510dcfc593fSKent Overstreet switch (s->s[j].state) { 511dcfc593fSKent Overstreet case SECTOR_UNALLOCATED: 512dcfc593fSKent Overstreet s->s[j].state = SECTOR_RESERVED; 513dcfc593fSKent Overstreet break; 514dcfc593fSKent Overstreet case SECTOR_DIRTY: 515dcfc593fSKent Overstreet s->s[j].state = SECTOR_DIRTY_RESERVED; 516dcfc593fSKent Overstreet i_sectors_delta--; 517dcfc593fSKent Overstreet break; 518dcfc593fSKent Overstreet default: 519dcfc593fSKent Overstreet break; 520dcfc593fSKent Overstreet } 521dcfc593fSKent Overstreet spin_unlock(&s->lock); 522dcfc593fSKent Overstreet } 523dcfc593fSKent Overstreet 524dcfc593fSKent Overstreet folio_unlock(folio); 525dcfc593fSKent Overstreet } 526dcfc593fSKent Overstreet folio_batch_release(&fbatch); 527dcfc593fSKent Overstreet cond_resched(); 528dcfc593fSKent Overstreet } 529dcfc593fSKent Overstreet 530dcfc593fSKent Overstreet i_sectors_acct(c, inode, NULL, i_sectors_delta); 531dcfc593fSKent Overstreet } 532dcfc593fSKent Overstreet 533e1036a2aSKent Overstreet static inline unsigned inode_nr_replicas(struct bch_fs *c, struct bch_inode_info *inode) 534e1036a2aSKent Overstreet { 535e1036a2aSKent Overstreet /* XXX: this should not be open coded */ 536e1036a2aSKent Overstreet return inode->ei_inode.bi_data_replicas 537e1036a2aSKent Overstreet ? inode->ei_inode.bi_data_replicas - 1 538e1036a2aSKent Overstreet : c->opts.data_replicas; 539e1036a2aSKent Overstreet } 540e1036a2aSKent Overstreet 541f57a6a5dSKent Overstreet static inline unsigned sectors_to_reserve(struct bch_page_sector *s, 542f57a6a5dSKent Overstreet unsigned nr_replicas) 543f57a6a5dSKent Overstreet { 544f57a6a5dSKent Overstreet return max(0, (int) nr_replicas - 545f57a6a5dSKent Overstreet s->nr_replicas - 546f57a6a5dSKent Overstreet s->replicas_reserved); 547f57a6a5dSKent Overstreet } 548f57a6a5dSKent Overstreet 549f57a6a5dSKent Overstreet static int bch2_get_page_disk_reservation(struct bch_fs *c, 550f57a6a5dSKent Overstreet struct bch_inode_info *inode, 5511c6fdbd8SKent Overstreet struct page *page, bool check_enospc) 5521c6fdbd8SKent Overstreet { 553f57a6a5dSKent Overstreet struct bch_page_state *s = bch2_page_state_create(page, 0); 554e1036a2aSKent Overstreet unsigned nr_replicas = inode_nr_replicas(c, inode); 555f57a6a5dSKent Overstreet struct disk_reservation disk_res = { 0 }; 556f57a6a5dSKent Overstreet unsigned i, disk_res_sectors = 0; 557f81b648dSKent Overstreet int ret; 5581c6fdbd8SKent Overstreet 559f57a6a5dSKent Overstreet if (!s) 560f57a6a5dSKent Overstreet return -ENOMEM; 5611c6fdbd8SKent Overstreet 562f57a6a5dSKent Overstreet for (i = 0; i < ARRAY_SIZE(s->s); i++) 563f57a6a5dSKent Overstreet disk_res_sectors += sectors_to_reserve(&s->s[i], nr_replicas); 564f57a6a5dSKent Overstreet 565f57a6a5dSKent Overstreet if (!disk_res_sectors) 566f57a6a5dSKent Overstreet return 0; 567f57a6a5dSKent Overstreet 568f57a6a5dSKent Overstreet ret = bch2_disk_reservation_get(c, &disk_res, 569f57a6a5dSKent Overstreet disk_res_sectors, 1, 570f57a6a5dSKent Overstreet !check_enospc 571f57a6a5dSKent Overstreet ? BCH_DISK_RESERVATION_NOFAIL 572f57a6a5dSKent Overstreet : 0); 5731c6fdbd8SKent Overstreet if (unlikely(ret)) 574f81b648dSKent Overstreet return ret; 575f81b648dSKent Overstreet 576f57a6a5dSKent Overstreet for (i = 0; i < ARRAY_SIZE(s->s); i++) 577f57a6a5dSKent Overstreet s->s[i].replicas_reserved += 578f57a6a5dSKent Overstreet sectors_to_reserve(&s->s[i], nr_replicas); 579f57a6a5dSKent Overstreet 580f57a6a5dSKent Overstreet return 0; 5811c6fdbd8SKent Overstreet } 5821c6fdbd8SKent Overstreet 583d1542e03SKent Overstreet struct bch2_page_reservation { 584d1542e03SKent Overstreet struct disk_reservation disk; 585d1542e03SKent Overstreet struct quota_res quota; 586d1542e03SKent Overstreet }; 587d1542e03SKent Overstreet 588d1542e03SKent Overstreet static void bch2_page_reservation_init(struct bch_fs *c, 589f57a6a5dSKent Overstreet struct bch_inode_info *inode, 590d1542e03SKent Overstreet struct bch2_page_reservation *res) 591d1542e03SKent Overstreet { 592d1542e03SKent Overstreet memset(res, 0, sizeof(*res)); 593d1542e03SKent Overstreet 594d1542e03SKent Overstreet res->disk.nr_replicas = inode_nr_replicas(c, inode); 595d1542e03SKent Overstreet } 596d1542e03SKent Overstreet 597d1542e03SKent Overstreet static void bch2_page_reservation_put(struct bch_fs *c, 598d1542e03SKent Overstreet struct bch_inode_info *inode, 599d1542e03SKent Overstreet struct bch2_page_reservation *res) 600d1542e03SKent Overstreet { 601d1542e03SKent Overstreet bch2_disk_reservation_put(c, &res->disk); 602d1542e03SKent Overstreet bch2_quota_reservation_put(c, inode, &res->quota); 603d1542e03SKent Overstreet } 604d1542e03SKent Overstreet 605d1542e03SKent Overstreet static int bch2_page_reservation_get(struct bch_fs *c, 606d1542e03SKent Overstreet struct bch_inode_info *inode, struct page *page, 607d1542e03SKent Overstreet struct bch2_page_reservation *res, 608d1542e03SKent Overstreet unsigned offset, unsigned len, bool check_enospc) 609f57a6a5dSKent Overstreet { 610f57a6a5dSKent Overstreet struct bch_page_state *s = bch2_page_state_create(page, 0); 611d1542e03SKent Overstreet unsigned i, disk_sectors = 0, quota_sectors = 0; 612f57a6a5dSKent Overstreet int ret; 613f57a6a5dSKent Overstreet 614f57a6a5dSKent Overstreet if (!s) 615f57a6a5dSKent Overstreet return -ENOMEM; 616f57a6a5dSKent Overstreet 617e6ec361fSKent Overstreet BUG_ON(!s->uptodate); 618e6ec361fSKent Overstreet 6194b0a66d5SKent Overstreet for (i = round_down(offset, block_bytes(c)) >> 9; 6204b0a66d5SKent Overstreet i < round_up(offset + len, block_bytes(c)) >> 9; 621d1542e03SKent Overstreet i++) { 622d1542e03SKent Overstreet disk_sectors += sectors_to_reserve(&s->s[i], 623d1542e03SKent Overstreet res->disk.nr_replicas); 624d1542e03SKent Overstreet quota_sectors += s->s[i].state == SECTOR_UNALLOCATED; 6251c6fdbd8SKent Overstreet } 6261c6fdbd8SKent Overstreet 627d1542e03SKent Overstreet if (disk_sectors) { 628d1542e03SKent Overstreet ret = bch2_disk_reservation_add(c, &res->disk, 629d1542e03SKent Overstreet disk_sectors, 630d1542e03SKent Overstreet !check_enospc 631d1542e03SKent Overstreet ? BCH_DISK_RESERVATION_NOFAIL 632d1542e03SKent Overstreet : 0); 633d1542e03SKent Overstreet if (unlikely(ret)) 634d1542e03SKent Overstreet return ret; 635d1542e03SKent Overstreet } 636d1542e03SKent Overstreet 637d1542e03SKent Overstreet if (quota_sectors) { 638d1542e03SKent Overstreet ret = bch2_quota_reservation_add(c, inode, &res->quota, 639d1542e03SKent Overstreet quota_sectors, 640d1542e03SKent Overstreet check_enospc); 641d1542e03SKent Overstreet if (unlikely(ret)) { 642d1542e03SKent Overstreet struct disk_reservation tmp = { 643d1542e03SKent Overstreet .sectors = disk_sectors 644d1542e03SKent Overstreet }; 645d1542e03SKent Overstreet 646d1542e03SKent Overstreet bch2_disk_reservation_put(c, &tmp); 647d1542e03SKent Overstreet res->disk.sectors -= disk_sectors; 648d1542e03SKent Overstreet return ret; 649d1542e03SKent Overstreet } 650d1542e03SKent Overstreet } 651d1542e03SKent Overstreet 652d1542e03SKent Overstreet return 0; 653f57a6a5dSKent Overstreet } 654f57a6a5dSKent Overstreet 6551c6fdbd8SKent Overstreet static void bch2_clear_page_bits(struct page *page) 6561c6fdbd8SKent Overstreet { 6571c6fdbd8SKent Overstreet struct bch_inode_info *inode = to_bch_ei(page->mapping->host); 6581c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 659f57a6a5dSKent Overstreet struct bch_page_state *s = bch2_page_state(page); 660d1542e03SKent Overstreet struct disk_reservation disk_res = { 0 }; 661f57a6a5dSKent Overstreet int i, dirty_sectors = 0; 6621c6fdbd8SKent Overstreet 663f57a6a5dSKent Overstreet if (!s) 6641c6fdbd8SKent Overstreet return; 6651c6fdbd8SKent Overstreet 6663826ee0bSKent Overstreet EBUG_ON(!PageLocked(page)); 6673826ee0bSKent Overstreet EBUG_ON(PageWriteback(page)); 6683826ee0bSKent Overstreet 669f57a6a5dSKent Overstreet for (i = 0; i < ARRAY_SIZE(s->s); i++) { 670d1542e03SKent Overstreet disk_res.sectors += s->s[i].replicas_reserved; 671d1542e03SKent Overstreet s->s[i].replicas_reserved = 0; 672d1542e03SKent Overstreet 673b44a66a6SKent Overstreet switch (s->s[i].state) { 674b44a66a6SKent Overstreet case SECTOR_DIRTY: 675f57a6a5dSKent Overstreet s->s[i].state = SECTOR_UNALLOCATED; 676b44a66a6SKent Overstreet --dirty_sectors; 677b44a66a6SKent Overstreet break; 678b44a66a6SKent Overstreet case SECTOR_DIRTY_RESERVED: 679b44a66a6SKent Overstreet s->s[i].state = SECTOR_RESERVED; 680b44a66a6SKent Overstreet break; 681b44a66a6SKent Overstreet default: 682b44a66a6SKent Overstreet break; 683f57a6a5dSKent Overstreet } 684f57a6a5dSKent Overstreet } 685adfcfaf0SKent Overstreet 686d1542e03SKent Overstreet bch2_disk_reservation_put(c, &disk_res); 687d1542e03SKent Overstreet 688b44a66a6SKent Overstreet i_sectors_acct(c, inode, NULL, dirty_sectors); 689adfcfaf0SKent Overstreet 690f57a6a5dSKent Overstreet bch2_page_state_release(page); 6911c6fdbd8SKent Overstreet } 6921c6fdbd8SKent Overstreet 693d1542e03SKent Overstreet static void bch2_set_page_dirty(struct bch_fs *c, 694d1542e03SKent Overstreet struct bch_inode_info *inode, struct page *page, 695d1542e03SKent Overstreet struct bch2_page_reservation *res, 696d1542e03SKent Overstreet unsigned offset, unsigned len) 6971c6fdbd8SKent Overstreet { 698d1542e03SKent Overstreet struct bch_page_state *s = bch2_page_state(page); 699f57a6a5dSKent Overstreet unsigned i, dirty_sectors = 0; 7001c6fdbd8SKent Overstreet 701877dfb34SKent Overstreet WARN_ON((u64) page_offset(page) + offset + len > 702877dfb34SKent Overstreet round_up((u64) i_size_read(&inode->v), block_bytes(c))); 703fb472ac5SKent Overstreet 7043826ee0bSKent Overstreet spin_lock(&s->lock); 7053826ee0bSKent Overstreet 7064b0a66d5SKent Overstreet for (i = round_down(offset, block_bytes(c)) >> 9; 7074b0a66d5SKent Overstreet i < round_up(offset + len, block_bytes(c)) >> 9; 708d1542e03SKent Overstreet i++) { 709d1542e03SKent Overstreet unsigned sectors = sectors_to_reserve(&s->s[i], 710d1542e03SKent Overstreet res->disk.nr_replicas); 7111c6fdbd8SKent Overstreet 712406d6d5aSKent Overstreet /* 713406d6d5aSKent Overstreet * This can happen if we race with the error path in 714406d6d5aSKent Overstreet * bch2_writepage_io_done(): 715406d6d5aSKent Overstreet */ 716406d6d5aSKent Overstreet sectors = min_t(unsigned, sectors, res->disk.sectors); 717406d6d5aSKent Overstreet 718d1542e03SKent Overstreet s->s[i].replicas_reserved += sectors; 719d1542e03SKent Overstreet res->disk.sectors -= sectors; 720adfcfaf0SKent Overstreet 721b44a66a6SKent Overstreet switch (s->s[i].state) { 722b44a66a6SKent Overstreet case SECTOR_UNALLOCATED: 723b44a66a6SKent Overstreet s->s[i].state = SECTOR_DIRTY; 724f57a6a5dSKent Overstreet dirty_sectors++; 725b44a66a6SKent Overstreet break; 726b44a66a6SKent Overstreet case SECTOR_RESERVED: 727b44a66a6SKent Overstreet s->s[i].state = SECTOR_DIRTY_RESERVED; 728b44a66a6SKent Overstreet break; 729b44a66a6SKent Overstreet default: 730b44a66a6SKent Overstreet break; 731b44a66a6SKent Overstreet } 732f57a6a5dSKent Overstreet } 733f57a6a5dSKent Overstreet 7343826ee0bSKent Overstreet spin_unlock(&s->lock); 7353826ee0bSKent Overstreet 736d1542e03SKent Overstreet i_sectors_acct(c, inode, &res->quota, dirty_sectors); 7371c6fdbd8SKent Overstreet 738d1542e03SKent Overstreet if (!PageDirty(page)) 739d1542e03SKent Overstreet filemap_dirty_folio(inode->v.i_mapping, page_folio(page)); 7401c6fdbd8SKent Overstreet } 7411c6fdbd8SKent Overstreet 7421c6fdbd8SKent Overstreet vm_fault_t bch2_page_fault(struct vm_fault *vmf) 7431c6fdbd8SKent Overstreet { 7441c6fdbd8SKent Overstreet struct file *file = vmf->vma->vm_file; 745eb8e6e9cSKent Overstreet struct address_space *mapping = file->f_mapping; 746eb8e6e9cSKent Overstreet struct address_space *fdm = faults_disabled_mapping(); 7471c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 7481c6fdbd8SKent Overstreet int ret; 7491c6fdbd8SKent Overstreet 750eb8e6e9cSKent Overstreet if (fdm == mapping) 751eb8e6e9cSKent Overstreet return VM_FAULT_SIGBUS; 752eb8e6e9cSKent Overstreet 753eb8e6e9cSKent Overstreet /* Lock ordering: */ 754eb8e6e9cSKent Overstreet if (fdm > mapping) { 755eb8e6e9cSKent Overstreet struct bch_inode_info *fdm_host = to_bch_ei(fdm->host); 756eb8e6e9cSKent Overstreet 757eb8e6e9cSKent Overstreet if (bch2_pagecache_add_tryget(&inode->ei_pagecache_lock)) 758eb8e6e9cSKent Overstreet goto got_lock; 759eb8e6e9cSKent Overstreet 760eb8e6e9cSKent Overstreet bch2_pagecache_block_put(&fdm_host->ei_pagecache_lock); 761eb8e6e9cSKent Overstreet 7621c6fdbd8SKent Overstreet bch2_pagecache_add_get(&inode->ei_pagecache_lock); 763eb8e6e9cSKent Overstreet bch2_pagecache_add_put(&inode->ei_pagecache_lock); 764eb8e6e9cSKent Overstreet 765eb8e6e9cSKent Overstreet bch2_pagecache_block_get(&fdm_host->ei_pagecache_lock); 766eb8e6e9cSKent Overstreet 767eb8e6e9cSKent Overstreet /* Signal that lock has been dropped: */ 768eb8e6e9cSKent Overstreet set_fdm_dropped_locks(); 769eb8e6e9cSKent Overstreet return VM_FAULT_SIGBUS; 770eb8e6e9cSKent Overstreet } 771eb8e6e9cSKent Overstreet 772eb8e6e9cSKent Overstreet bch2_pagecache_add_get(&inode->ei_pagecache_lock); 773eb8e6e9cSKent Overstreet got_lock: 7741c6fdbd8SKent Overstreet ret = filemap_fault(vmf); 7751c6fdbd8SKent Overstreet bch2_pagecache_add_put(&inode->ei_pagecache_lock); 7761c6fdbd8SKent Overstreet 7771c6fdbd8SKent Overstreet return ret; 7781c6fdbd8SKent Overstreet } 7791c6fdbd8SKent Overstreet 7801c6fdbd8SKent Overstreet vm_fault_t bch2_page_mkwrite(struct vm_fault *vmf) 7811c6fdbd8SKent Overstreet { 7821c6fdbd8SKent Overstreet struct page *page = vmf->page; 7831c6fdbd8SKent Overstreet struct file *file = vmf->vma->vm_file; 7841c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 7851c6fdbd8SKent Overstreet struct address_space *mapping = file->f_mapping; 7861c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 787d1542e03SKent Overstreet struct bch2_page_reservation res; 7886cc3535dSKent Overstreet unsigned len; 7896cc3535dSKent Overstreet loff_t isize; 790e6ec361fSKent Overstreet int ret; 7911c6fdbd8SKent Overstreet 792d1542e03SKent Overstreet bch2_page_reservation_init(c, inode, &res); 793d1542e03SKent Overstreet 7941c6fdbd8SKent Overstreet sb_start_pagefault(inode->v.i_sb); 7951c6fdbd8SKent Overstreet file_update_time(file); 7961c6fdbd8SKent Overstreet 7971c6fdbd8SKent Overstreet /* 7981c6fdbd8SKent Overstreet * Not strictly necessary, but helps avoid dio writes livelocking in 7991c6fdbd8SKent Overstreet * write_invalidate_inode_pages_range() - can drop this if/when we get 8001c6fdbd8SKent Overstreet * a write_invalidate_inode_pages_range() that works without dropping 8011c6fdbd8SKent Overstreet * page lock before invalidating page 8021c6fdbd8SKent Overstreet */ 8031c6fdbd8SKent Overstreet bch2_pagecache_add_get(&inode->ei_pagecache_lock); 8041c6fdbd8SKent Overstreet 8051c6fdbd8SKent Overstreet lock_page(page); 8066cc3535dSKent Overstreet isize = i_size_read(&inode->v); 8076cc3535dSKent Overstreet 8086cc3535dSKent Overstreet if (page->mapping != mapping || page_offset(page) >= isize) { 8091c6fdbd8SKent Overstreet unlock_page(page); 8101c6fdbd8SKent Overstreet ret = VM_FAULT_NOPAGE; 8111c6fdbd8SKent Overstreet goto out; 8121c6fdbd8SKent Overstreet } 8131c6fdbd8SKent Overstreet 814d55460bbSKent Overstreet len = min_t(loff_t, PAGE_SIZE, isize - page_offset(page)); 8156cc3535dSKent Overstreet 816e6ec361fSKent Overstreet if (!bch2_page_state_create(page, __GFP_NOFAIL)->uptodate) { 817e6ec361fSKent Overstreet if (bch2_page_state_set(c, inode_inum(inode), &page, 1)) { 818e6ec361fSKent Overstreet unlock_page(page); 819e6ec361fSKent Overstreet ret = VM_FAULT_SIGBUS; 820e6ec361fSKent Overstreet goto out; 821e6ec361fSKent Overstreet } 822e6ec361fSKent Overstreet } 823e6ec361fSKent Overstreet 8246cc3535dSKent Overstreet if (bch2_page_reservation_get(c, inode, page, &res, 0, len, true)) { 8251c6fdbd8SKent Overstreet unlock_page(page); 8261c6fdbd8SKent Overstreet ret = VM_FAULT_SIGBUS; 8271c6fdbd8SKent Overstreet goto out; 8281c6fdbd8SKent Overstreet } 8291c6fdbd8SKent Overstreet 8306cc3535dSKent Overstreet bch2_set_page_dirty(c, inode, page, &res, 0, len); 8311b783a69SKent Overstreet bch2_page_reservation_put(c, inode, &res); 8321b783a69SKent Overstreet 8331c6fdbd8SKent Overstreet wait_for_stable_page(page); 834e6ec361fSKent Overstreet ret = VM_FAULT_LOCKED; 8351c6fdbd8SKent Overstreet out: 8361c6fdbd8SKent Overstreet bch2_pagecache_add_put(&inode->ei_pagecache_lock); 8371c6fdbd8SKent Overstreet sb_end_pagefault(inode->v.i_sb); 838d1542e03SKent Overstreet 8391c6fdbd8SKent Overstreet return ret; 8401c6fdbd8SKent Overstreet } 8411c6fdbd8SKent Overstreet 8421c6fdbd8SKent Overstreet void bch2_invalidate_folio(struct folio *folio, size_t offset, size_t length) 8431c6fdbd8SKent Overstreet { 8441c6fdbd8SKent Overstreet if (offset || length < folio_size(folio)) 8451c6fdbd8SKent Overstreet return; 8461c6fdbd8SKent Overstreet 8471c6fdbd8SKent Overstreet bch2_clear_page_bits(&folio->page); 8481c6fdbd8SKent Overstreet } 8491c6fdbd8SKent Overstreet 8501c6fdbd8SKent Overstreet bool bch2_release_folio(struct folio *folio, gfp_t gfp_mask) 8511c6fdbd8SKent Overstreet { 852a6d90385SKent Overstreet if (folio_test_dirty(folio) || folio_test_writeback(folio)) 8531c6fdbd8SKent Overstreet return false; 8541c6fdbd8SKent Overstreet 8551c6fdbd8SKent Overstreet bch2_clear_page_bits(&folio->page); 8561c6fdbd8SKent Overstreet return true; 8571c6fdbd8SKent Overstreet } 8581c6fdbd8SKent Overstreet 8591c6fdbd8SKent Overstreet /* readpage(s): */ 8601c6fdbd8SKent Overstreet 8611c6fdbd8SKent Overstreet static void bch2_readpages_end_io(struct bio *bio) 8621c6fdbd8SKent Overstreet { 8631c6fdbd8SKent Overstreet struct bvec_iter_all iter; 8641c6fdbd8SKent Overstreet struct bio_vec *bv; 8651c6fdbd8SKent Overstreet 8661c6fdbd8SKent Overstreet bio_for_each_segment_all(bv, bio, iter) { 8671c6fdbd8SKent Overstreet struct page *page = bv->bv_page; 8681c6fdbd8SKent Overstreet 8691c6fdbd8SKent Overstreet if (!bio->bi_status) { 8701c6fdbd8SKent Overstreet SetPageUptodate(page); 8711c6fdbd8SKent Overstreet } else { 8721c6fdbd8SKent Overstreet ClearPageUptodate(page); 8731c6fdbd8SKent Overstreet SetPageError(page); 8741c6fdbd8SKent Overstreet } 8751c6fdbd8SKent Overstreet unlock_page(page); 8761c6fdbd8SKent Overstreet } 8771c6fdbd8SKent Overstreet 8781c6fdbd8SKent Overstreet bio_put(bio); 8791c6fdbd8SKent Overstreet } 8801c6fdbd8SKent Overstreet 8811c6fdbd8SKent Overstreet struct readpages_iter { 8821c6fdbd8SKent Overstreet struct address_space *mapping; 8831c6fdbd8SKent Overstreet struct page **pages; 8841c6fdbd8SKent Overstreet unsigned nr_pages; 8851c6fdbd8SKent Overstreet unsigned idx; 8861c6fdbd8SKent Overstreet pgoff_t offset; 8871c6fdbd8SKent Overstreet }; 8881c6fdbd8SKent Overstreet 8891c6fdbd8SKent Overstreet static int readpages_iter_init(struct readpages_iter *iter, 8901c6fdbd8SKent Overstreet struct readahead_control *ractl) 8911c6fdbd8SKent Overstreet { 8921c6fdbd8SKent Overstreet unsigned i, nr_pages = readahead_count(ractl); 8931c6fdbd8SKent Overstreet 8941c6fdbd8SKent Overstreet memset(iter, 0, sizeof(*iter)); 8951c6fdbd8SKent Overstreet 8961c6fdbd8SKent Overstreet iter->mapping = ractl->mapping; 8971c6fdbd8SKent Overstreet iter->offset = readahead_index(ractl); 8981c6fdbd8SKent Overstreet iter->nr_pages = nr_pages; 8991c6fdbd8SKent Overstreet 9001c6fdbd8SKent Overstreet iter->pages = kmalloc_array(nr_pages, sizeof(struct page *), GFP_NOFS); 9011c6fdbd8SKent Overstreet if (!iter->pages) 9021c6fdbd8SKent Overstreet return -ENOMEM; 9031c6fdbd8SKent Overstreet 90489931472SKent Overstreet nr_pages = __readahead_batch(ractl, iter->pages, nr_pages); 9051c6fdbd8SKent Overstreet for (i = 0; i < nr_pages; i++) { 906f57a6a5dSKent Overstreet __bch2_page_state_create(iter->pages[i], __GFP_NOFAIL); 9071c6fdbd8SKent Overstreet put_page(iter->pages[i]); 9081c6fdbd8SKent Overstreet } 9091c6fdbd8SKent Overstreet 9101c6fdbd8SKent Overstreet return 0; 9111c6fdbd8SKent Overstreet } 9121c6fdbd8SKent Overstreet 9131c6fdbd8SKent Overstreet static inline struct page *readpage_iter_next(struct readpages_iter *iter) 9141c6fdbd8SKent Overstreet { 9151c6fdbd8SKent Overstreet if (iter->idx >= iter->nr_pages) 9161c6fdbd8SKent Overstreet return NULL; 9171c6fdbd8SKent Overstreet 9181c6fdbd8SKent Overstreet EBUG_ON(iter->pages[iter->idx]->index != iter->offset + iter->idx); 9191c6fdbd8SKent Overstreet 9201c6fdbd8SKent Overstreet return iter->pages[iter->idx]; 9211c6fdbd8SKent Overstreet } 9221c6fdbd8SKent Overstreet 92335189e09SKent Overstreet static bool extent_partial_reads_expensive(struct bkey_s_c k) 92435189e09SKent Overstreet { 92535189e09SKent Overstreet struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k); 92635189e09SKent Overstreet struct bch_extent_crc_unpacked crc; 92735189e09SKent Overstreet const union bch_extent_entry *i; 92835189e09SKent Overstreet 92935189e09SKent Overstreet bkey_for_each_crc(k.k, ptrs, crc, i) 93035189e09SKent Overstreet if (crc.csum_type || crc.compression_type) 93135189e09SKent Overstreet return true; 93235189e09SKent Overstreet return false; 93335189e09SKent Overstreet } 93435189e09SKent Overstreet 9351c6fdbd8SKent Overstreet static void readpage_bio_extend(struct readpages_iter *iter, 93676426098SKent Overstreet struct bio *bio, 93776426098SKent Overstreet unsigned sectors_this_extent, 9381c6fdbd8SKent Overstreet bool get_more) 9391c6fdbd8SKent Overstreet { 94076426098SKent Overstreet while (bio_sectors(bio) < sectors_this_extent && 9411c6fdbd8SKent Overstreet bio->bi_vcnt < bio->bi_max_vecs) { 9427279c1a2SKent Overstreet pgoff_t page_offset = bio_end_sector(bio) >> PAGE_SECTORS_SHIFT; 9431c6fdbd8SKent Overstreet struct page *page = readpage_iter_next(iter); 9441c6fdbd8SKent Overstreet int ret; 9451c6fdbd8SKent Overstreet 9461c6fdbd8SKent Overstreet if (page) { 9471c6fdbd8SKent Overstreet if (iter->offset + iter->idx != page_offset) 9481c6fdbd8SKent Overstreet break; 9491c6fdbd8SKent Overstreet 9501c6fdbd8SKent Overstreet iter->idx++; 9511c6fdbd8SKent Overstreet } else { 9521c6fdbd8SKent Overstreet if (!get_more) 9531c6fdbd8SKent Overstreet break; 9541c6fdbd8SKent Overstreet 9551c6fdbd8SKent Overstreet page = xa_load(&iter->mapping->i_pages, page_offset); 9561c6fdbd8SKent Overstreet if (page && !xa_is_value(page)) 9571c6fdbd8SKent Overstreet break; 9581c6fdbd8SKent Overstreet 9591c6fdbd8SKent Overstreet page = __page_cache_alloc(readahead_gfp_mask(iter->mapping)); 9601c6fdbd8SKent Overstreet if (!page) 9611c6fdbd8SKent Overstreet break; 9621c6fdbd8SKent Overstreet 963f57a6a5dSKent Overstreet if (!__bch2_page_state_create(page, 0)) { 964f57a6a5dSKent Overstreet put_page(page); 965f57a6a5dSKent Overstreet break; 966f57a6a5dSKent Overstreet } 9671c6fdbd8SKent Overstreet 9681c6fdbd8SKent Overstreet ret = add_to_page_cache_lru(page, iter->mapping, 9691c6fdbd8SKent Overstreet page_offset, GFP_NOFS); 9701c6fdbd8SKent Overstreet if (ret) { 971f57a6a5dSKent Overstreet __bch2_page_state_release(page); 9721c6fdbd8SKent Overstreet put_page(page); 9731c6fdbd8SKent Overstreet break; 9741c6fdbd8SKent Overstreet } 9751c6fdbd8SKent Overstreet 9761c6fdbd8SKent Overstreet put_page(page); 9771c6fdbd8SKent Overstreet } 9781c6fdbd8SKent Overstreet 979885678f6SKent Overstreet BUG_ON(!bio_add_page(bio, page, PAGE_SIZE, 0)); 9801c6fdbd8SKent Overstreet } 9811c6fdbd8SKent Overstreet } 9821c6fdbd8SKent Overstreet 9838c6d298aSKent Overstreet static void bchfs_read(struct btree_trans *trans, 9848c6d298aSKent Overstreet struct bch_read_bio *rbio, 9858c6d298aSKent Overstreet subvol_inum inum, 9861c6fdbd8SKent Overstreet struct readpages_iter *readpages_iter) 9871c6fdbd8SKent Overstreet { 9880f238367SKent Overstreet struct bch_fs *c = trans->c; 9898c6d298aSKent Overstreet struct btree_iter iter; 99007a1006aSKent Overstreet struct bkey_buf sk; 9911c6fdbd8SKent Overstreet int flags = BCH_READ_RETRY_IF_STALE| 9921c6fdbd8SKent Overstreet BCH_READ_MAY_PROMOTE; 9938c6d298aSKent Overstreet u32 snapshot; 99476426098SKent Overstreet int ret = 0; 9951c6fdbd8SKent Overstreet 9961c6fdbd8SKent Overstreet rbio->c = c; 9971c6fdbd8SKent Overstreet rbio->start_time = local_clock(); 9988c6d298aSKent Overstreet rbio->subvol = inum.subvol; 99935189e09SKent Overstreet 100007a1006aSKent Overstreet bch2_bkey_buf_init(&sk); 100176426098SKent Overstreet retry: 1002700c25b3SKent Overstreet bch2_trans_begin(trans); 10038c6d298aSKent Overstreet iter = (struct btree_iter) { NULL }; 1004700c25b3SKent Overstreet 10058c6d298aSKent Overstreet ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot); 10068c6d298aSKent Overstreet if (ret) 10078c6d298aSKent Overstreet goto err; 10088c6d298aSKent Overstreet 10098c6d298aSKent Overstreet bch2_trans_iter_init(trans, &iter, BTREE_ID_extents, 10108c6d298aSKent Overstreet SPOS(inum.inum, rbio->bio.bi_iter.bi_sector, snapshot), 101157cfdd8bSKent Overstreet BTREE_ITER_SLOTS); 10121c6fdbd8SKent Overstreet while (1) { 10131c6fdbd8SKent Overstreet struct bkey_s_c k; 101476426098SKent Overstreet unsigned bytes, sectors, offset_into_extent; 10155ff75ccbSKent Overstreet enum btree_id data_btree = BTREE_ID_extents; 10161c6fdbd8SKent Overstreet 10173737e0ddSKent Overstreet /* 10183737e0ddSKent Overstreet * read_extent -> io_time_reset may cause a transaction restart 10193737e0ddSKent Overstreet * without returning an error, we need to check for that here: 10203737e0ddSKent Overstreet */ 1021549d173cSKent Overstreet ret = bch2_trans_relock(trans); 1022549d173cSKent Overstreet if (ret) 10233737e0ddSKent Overstreet break; 10243737e0ddSKent Overstreet 10258c6d298aSKent Overstreet bch2_btree_iter_set_pos(&iter, 10268c6d298aSKent Overstreet POS(inum.inum, rbio->bio.bi_iter.bi_sector)); 10271c6fdbd8SKent Overstreet 10288c6d298aSKent Overstreet k = bch2_btree_iter_peek_slot(&iter); 102976426098SKent Overstreet ret = bkey_err(k); 103076426098SKent Overstreet if (ret) 103176426098SKent Overstreet break; 10321c6fdbd8SKent Overstreet 10338c6d298aSKent Overstreet offset_into_extent = iter.pos.offset - 103406ed8558SKent Overstreet bkey_start_offset(k.k); 103576426098SKent Overstreet sectors = k.k->size - offset_into_extent; 103676426098SKent Overstreet 103707a1006aSKent Overstreet bch2_bkey_buf_reassemble(&sk, c, k); 103813dcd4abSKent Overstreet 10395ff75ccbSKent Overstreet ret = bch2_read_indirect_extent(trans, &data_btree, 104022d8a33dSYuxuan Shui &offset_into_extent, &sk); 104176426098SKent Overstreet if (ret) 104276426098SKent Overstreet break; 104376426098SKent Overstreet 104413dcd4abSKent Overstreet k = bkey_i_to_s_c(sk.k); 104513dcd4abSKent Overstreet 104676426098SKent Overstreet sectors = min(sectors, k.k->size - offset_into_extent); 104776426098SKent Overstreet 104835189e09SKent Overstreet if (readpages_iter) 104935189e09SKent Overstreet readpage_bio_extend(readpages_iter, &rbio->bio, sectors, 105035189e09SKent Overstreet extent_partial_reads_expensive(k)); 10511c6fdbd8SKent Overstreet 105276426098SKent Overstreet bytes = min(sectors, bio_sectors(&rbio->bio)) << 9; 105306ed8558SKent Overstreet swap(rbio->bio.bi_iter.bi_size, bytes); 10541c6fdbd8SKent Overstreet 105506ed8558SKent Overstreet if (rbio->bio.bi_iter.bi_size == bytes) 10561c6fdbd8SKent Overstreet flags |= BCH_READ_LAST_FRAGMENT; 10571c6fdbd8SKent Overstreet 1058b44a66a6SKent Overstreet bch2_bio_page_state_set(&rbio->bio, k); 10591c6fdbd8SKent Overstreet 10608c6d298aSKent Overstreet bch2_read_extent(trans, rbio, iter.pos, 10615ff75ccbSKent Overstreet data_btree, k, offset_into_extent, flags); 10621c6fdbd8SKent Overstreet 10631c6fdbd8SKent Overstreet if (flags & BCH_READ_LAST_FRAGMENT) 106435189e09SKent Overstreet break; 10651c6fdbd8SKent Overstreet 106606ed8558SKent Overstreet swap(rbio->bio.bi_iter.bi_size, bytes); 106706ed8558SKent Overstreet bio_advance(&rbio->bio, bytes); 1068084d42bbSKent Overstreet 1069084d42bbSKent Overstreet ret = btree_trans_too_many_iters(trans); 1070084d42bbSKent Overstreet if (ret) 1071084d42bbSKent Overstreet break; 10721c6fdbd8SKent Overstreet } 10738c6d298aSKent Overstreet err: 10748c6d298aSKent Overstreet bch2_trans_iter_exit(trans, &iter); 107576426098SKent Overstreet 1076549d173cSKent Overstreet if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) 107776426098SKent Overstreet goto retry; 107876426098SKent Overstreet 107935189e09SKent Overstreet if (ret) { 10808c6d298aSKent Overstreet bch_err_inum_ratelimited(c, inum.inum, 10810fefe8d8SKent Overstreet "read error %i from btree lookup", ret); 10820fefe8d8SKent Overstreet rbio->bio.bi_status = BLK_STS_IOERR; 108376426098SKent Overstreet bio_endio(&rbio->bio); 10841c6fdbd8SKent Overstreet } 10851c6fdbd8SKent Overstreet 108607a1006aSKent Overstreet bch2_bkey_buf_exit(&sk, c); 108735189e09SKent Overstreet } 108835189e09SKent Overstreet 10891c6fdbd8SKent Overstreet void bch2_readahead(struct readahead_control *ractl) 10901c6fdbd8SKent Overstreet { 10911c6fdbd8SKent Overstreet struct bch_inode_info *inode = to_bch_ei(ractl->mapping->host); 10921c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 10939a3df993SKent Overstreet struct bch_io_opts opts = io_opts(c, &inode->ei_inode); 1094424eb881SKent Overstreet struct btree_trans trans; 10951c6fdbd8SKent Overstreet struct page *page; 10961c6fdbd8SKent Overstreet struct readpages_iter readpages_iter; 10971c6fdbd8SKent Overstreet int ret; 10981c6fdbd8SKent Overstreet 10991c6fdbd8SKent Overstreet ret = readpages_iter_init(&readpages_iter, ractl); 11001c6fdbd8SKent Overstreet BUG_ON(ret); 11011c6fdbd8SKent Overstreet 110220bceecbSKent Overstreet bch2_trans_init(&trans, c, 0, 0); 11031c6fdbd8SKent Overstreet 11041c6fdbd8SKent Overstreet bch2_pagecache_add_get(&inode->ei_pagecache_lock); 11051c6fdbd8SKent Overstreet 11061c6fdbd8SKent Overstreet while ((page = readpage_iter_next(&readpages_iter))) { 11071c6fdbd8SKent Overstreet pgoff_t index = readpages_iter.offset + readpages_iter.idx; 11081c6fdbd8SKent Overstreet unsigned n = min_t(unsigned, 11091c6fdbd8SKent Overstreet readpages_iter.nr_pages - 11101c6fdbd8SKent Overstreet readpages_iter.idx, 11111c6fdbd8SKent Overstreet BIO_MAX_VECS); 11121c6fdbd8SKent Overstreet struct bch_read_bio *rbio = 11131c6fdbd8SKent Overstreet rbio_init(bio_alloc_bioset(NULL, n, REQ_OP_READ, 11141c6fdbd8SKent Overstreet GFP_NOFS, &c->bio_read), 11151c6fdbd8SKent Overstreet opts); 11161c6fdbd8SKent Overstreet 11171c6fdbd8SKent Overstreet readpages_iter.idx++; 11181c6fdbd8SKent Overstreet 11197279c1a2SKent Overstreet rbio->bio.bi_iter.bi_sector = (sector_t) index << PAGE_SECTORS_SHIFT; 11201c6fdbd8SKent Overstreet rbio->bio.bi_end_io = bch2_readpages_end_io; 1121885678f6SKent Overstreet BUG_ON(!bio_add_page(&rbio->bio, page, PAGE_SIZE, 0)); 11221c6fdbd8SKent Overstreet 11238c6d298aSKent Overstreet bchfs_read(&trans, rbio, inode_inum(inode), 11240f238367SKent Overstreet &readpages_iter); 11251c6fdbd8SKent Overstreet } 11261c6fdbd8SKent Overstreet 11271c6fdbd8SKent Overstreet bch2_pagecache_add_put(&inode->ei_pagecache_lock); 1128424eb881SKent Overstreet 1129424eb881SKent Overstreet bch2_trans_exit(&trans); 11301c6fdbd8SKent Overstreet kfree(readpages_iter.pages); 11311c6fdbd8SKent Overstreet } 11321c6fdbd8SKent Overstreet 11331c6fdbd8SKent Overstreet static void __bchfs_readpage(struct bch_fs *c, struct bch_read_bio *rbio, 11348c6d298aSKent Overstreet subvol_inum inum, struct page *page) 11351c6fdbd8SKent Overstreet { 1136424eb881SKent Overstreet struct btree_trans trans; 11371c6fdbd8SKent Overstreet 1138f57a6a5dSKent Overstreet bch2_page_state_create(page, __GFP_NOFAIL); 11391c6fdbd8SKent Overstreet 11401c6fdbd8SKent Overstreet rbio->bio.bi_opf = REQ_OP_READ|REQ_SYNC; 11417f5e31e1SKent Overstreet rbio->bio.bi_iter.bi_sector = 11427279c1a2SKent Overstreet (sector_t) page->index << PAGE_SECTORS_SHIFT; 11437f5e31e1SKent Overstreet BUG_ON(!bio_add_page(&rbio->bio, page, PAGE_SIZE, 0)); 11441c6fdbd8SKent Overstreet 114520bceecbSKent Overstreet bch2_trans_init(&trans, c, 0, 0); 11468c6d298aSKent Overstreet bchfs_read(&trans, rbio, inum, NULL); 1147424eb881SKent Overstreet bch2_trans_exit(&trans); 11481c6fdbd8SKent Overstreet } 11491c6fdbd8SKent Overstreet 11501c6fdbd8SKent Overstreet static void bch2_read_single_page_end_io(struct bio *bio) 11511c6fdbd8SKent Overstreet { 11521c6fdbd8SKent Overstreet complete(bio->bi_private); 11531c6fdbd8SKent Overstreet } 11541c6fdbd8SKent Overstreet 11551c6fdbd8SKent Overstreet static int bch2_read_single_page(struct page *page, 11561c6fdbd8SKent Overstreet struct address_space *mapping) 11571c6fdbd8SKent Overstreet { 11581c6fdbd8SKent Overstreet struct bch_inode_info *inode = to_bch_ei(mapping->host); 11591c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 11601c6fdbd8SKent Overstreet struct bch_read_bio *rbio; 11611c6fdbd8SKent Overstreet int ret; 11621c6fdbd8SKent Overstreet DECLARE_COMPLETION_ONSTACK(done); 11631c6fdbd8SKent Overstreet 11641c6fdbd8SKent Overstreet rbio = rbio_init(bio_alloc_bioset(NULL, 1, REQ_OP_READ, GFP_NOFS, &c->bio_read), 11659a3df993SKent Overstreet io_opts(c, &inode->ei_inode)); 11661c6fdbd8SKent Overstreet rbio->bio.bi_private = &done; 11671c6fdbd8SKent Overstreet rbio->bio.bi_end_io = bch2_read_single_page_end_io; 11681c6fdbd8SKent Overstreet 11698c6d298aSKent Overstreet __bchfs_readpage(c, rbio, inode_inum(inode), page); 11701c6fdbd8SKent Overstreet wait_for_completion(&done); 11711c6fdbd8SKent Overstreet 11721c6fdbd8SKent Overstreet ret = blk_status_to_errno(rbio->bio.bi_status); 11731c6fdbd8SKent Overstreet bio_put(&rbio->bio); 11741c6fdbd8SKent Overstreet 11751c6fdbd8SKent Overstreet if (ret < 0) 11761c6fdbd8SKent Overstreet return ret; 11771c6fdbd8SKent Overstreet 11781c6fdbd8SKent Overstreet SetPageUptodate(page); 11791c6fdbd8SKent Overstreet return 0; 11801c6fdbd8SKent Overstreet } 11811c6fdbd8SKent Overstreet 11821c6fdbd8SKent Overstreet int bch2_read_folio(struct file *file, struct folio *folio) 11831c6fdbd8SKent Overstreet { 11841c6fdbd8SKent Overstreet struct page *page = &folio->page; 11851c6fdbd8SKent Overstreet int ret; 11861c6fdbd8SKent Overstreet 11871c6fdbd8SKent Overstreet ret = bch2_read_single_page(page, page->mapping); 11881c6fdbd8SKent Overstreet folio_unlock(folio); 11895c1ef830SKent Overstreet return bch2_err_class(ret); 11901c6fdbd8SKent Overstreet } 11911c6fdbd8SKent Overstreet 11921c6fdbd8SKent Overstreet /* writepages: */ 11931c6fdbd8SKent Overstreet 11941c6fdbd8SKent Overstreet struct bch_writepage_state { 11951c6fdbd8SKent Overstreet struct bch_writepage_io *io; 11961c6fdbd8SKent Overstreet struct bch_io_opts opts; 11971c6fdbd8SKent Overstreet }; 11981c6fdbd8SKent Overstreet 11991c6fdbd8SKent Overstreet static inline struct bch_writepage_state bch_writepage_state_init(struct bch_fs *c, 12001c6fdbd8SKent Overstreet struct bch_inode_info *inode) 12011c6fdbd8SKent Overstreet { 12029a3df993SKent Overstreet return (struct bch_writepage_state) { 12039a3df993SKent Overstreet .opts = io_opts(c, &inode->ei_inode) 12049a3df993SKent Overstreet }; 12051c6fdbd8SKent Overstreet } 12061c6fdbd8SKent Overstreet 12079f311f21SKent Overstreet static void bch2_writepage_io_done(struct bch_write_op *op) 12081c6fdbd8SKent Overstreet { 12099f311f21SKent Overstreet struct bch_writepage_io *io = 12109f311f21SKent Overstreet container_of(op, struct bch_writepage_io, op); 12119a3df993SKent Overstreet struct bch_fs *c = io->op.c; 12129a3df993SKent Overstreet struct bio *bio = &io->op.wbio.bio; 12131c6fdbd8SKent Overstreet struct bvec_iter_all iter; 12141c6fdbd8SKent Overstreet struct bio_vec *bvec; 1215b3fce09cSKent Overstreet unsigned i; 12161c6fdbd8SKent Overstreet 1217ef1b2092SKent Overstreet up(&io->op.c->io_in_flight); 1218ef1b2092SKent Overstreet 12199a3df993SKent Overstreet if (io->op.error) { 122033c74e41SKent Overstreet set_bit(EI_INODE_ERROR, &io->inode->ei_flags); 122133c74e41SKent Overstreet 122275812e70SKent Overstreet bio_for_each_segment_all(bvec, bio, iter) { 1223b3fce09cSKent Overstreet struct bch_page_state *s; 1224b3fce09cSKent Overstreet 12251c6fdbd8SKent Overstreet SetPageError(bvec->bv_page); 122675812e70SKent Overstreet mapping_set_error(bvec->bv_page->mapping, -EIO); 1227b3fce09cSKent Overstreet 12283826ee0bSKent Overstreet s = __bch2_page_state(bvec->bv_page); 12293826ee0bSKent Overstreet spin_lock(&s->lock); 1230b3fce09cSKent Overstreet for (i = 0; i < PAGE_SECTORS; i++) 1231b3fce09cSKent Overstreet s->s[i].nr_replicas = 0; 12323826ee0bSKent Overstreet spin_unlock(&s->lock); 123375812e70SKent Overstreet } 12341c6fdbd8SKent Overstreet } 12351c6fdbd8SKent Overstreet 12364be1a412SKent Overstreet if (io->op.flags & BCH_WRITE_WROTE_DATA_INLINE) { 12374be1a412SKent Overstreet bio_for_each_segment_all(bvec, bio, iter) { 12384be1a412SKent Overstreet struct bch_page_state *s; 12394be1a412SKent Overstreet 12404be1a412SKent Overstreet s = __bch2_page_state(bvec->bv_page); 12414be1a412SKent Overstreet spin_lock(&s->lock); 12424be1a412SKent Overstreet for (i = 0; i < PAGE_SECTORS; i++) 12434be1a412SKent Overstreet s->s[i].nr_replicas = 0; 12444be1a412SKent Overstreet spin_unlock(&s->lock); 12454be1a412SKent Overstreet } 12464be1a412SKent Overstreet } 12474be1a412SKent Overstreet 12481c6fdbd8SKent Overstreet /* 12491c6fdbd8SKent Overstreet * racing with fallocate can cause us to add fewer sectors than 12501c6fdbd8SKent Overstreet * expected - but we shouldn't add more sectors than expected: 12511c6fdbd8SKent Overstreet */ 1252f8494d25SKent Overstreet WARN_ON_ONCE(io->op.i_sectors_delta > 0); 12531c6fdbd8SKent Overstreet 12541c6fdbd8SKent Overstreet /* 12551c6fdbd8SKent Overstreet * (error (due to going RO) halfway through a page can screw that up 12561c6fdbd8SKent Overstreet * slightly) 12571c6fdbd8SKent Overstreet * XXX wtf? 12589a3df993SKent Overstreet BUG_ON(io->op.op.i_sectors_delta >= PAGE_SECTORS); 12591c6fdbd8SKent Overstreet */ 12601c6fdbd8SKent Overstreet 12611c6fdbd8SKent Overstreet /* 12621c6fdbd8SKent Overstreet * PageWriteback is effectively our ref on the inode - fixup i_blocks 12631c6fdbd8SKent Overstreet * before calling end_page_writeback: 12641c6fdbd8SKent Overstreet */ 12659a3df993SKent Overstreet i_sectors_acct(c, io->inode, NULL, io->op.i_sectors_delta); 12661c6fdbd8SKent Overstreet 12677f5e31e1SKent Overstreet bio_for_each_segment_all(bvec, bio, iter) { 12687f5e31e1SKent Overstreet struct bch_page_state *s = __bch2_page_state(bvec->bv_page); 12697f5e31e1SKent Overstreet 12707f5e31e1SKent Overstreet if (atomic_dec_and_test(&s->write_count)) 12711c6fdbd8SKent Overstreet end_page_writeback(bvec->bv_page); 12727f5e31e1SKent Overstreet } 12731c6fdbd8SKent Overstreet 12749f311f21SKent Overstreet bio_put(&io->op.wbio.bio); 12751c6fdbd8SKent Overstreet } 12761c6fdbd8SKent Overstreet 12771c6fdbd8SKent Overstreet static void bch2_writepage_do_io(struct bch_writepage_state *w) 12781c6fdbd8SKent Overstreet { 12791c6fdbd8SKent Overstreet struct bch_writepage_io *io = w->io; 12801c6fdbd8SKent Overstreet 1281ef1b2092SKent Overstreet down(&io->op.c->io_in_flight); 1282ef1b2092SKent Overstreet 12831c6fdbd8SKent Overstreet w->io = NULL; 12849f311f21SKent Overstreet closure_call(&io->op.cl, bch2_write, NULL, NULL); 12851c6fdbd8SKent Overstreet } 12861c6fdbd8SKent Overstreet 12871c6fdbd8SKent Overstreet /* 12881c6fdbd8SKent Overstreet * Get a bch_writepage_io and add @page to it - appending to an existing one if 12891c6fdbd8SKent Overstreet * possible, else allocating a new one: 12901c6fdbd8SKent Overstreet */ 12911c6fdbd8SKent Overstreet static void bch2_writepage_io_alloc(struct bch_fs *c, 129250fe5bd6SKent Overstreet struct writeback_control *wbc, 12931c6fdbd8SKent Overstreet struct bch_writepage_state *w, 12941c6fdbd8SKent Overstreet struct bch_inode_info *inode, 12957f5e31e1SKent Overstreet u64 sector, 12961c6fdbd8SKent Overstreet unsigned nr_replicas) 12971c6fdbd8SKent Overstreet { 12981c6fdbd8SKent Overstreet struct bch_write_op *op; 12991c6fdbd8SKent Overstreet 13001c6fdbd8SKent Overstreet w->io = container_of(bio_alloc_bioset(NULL, BIO_MAX_VECS, 13011c6fdbd8SKent Overstreet REQ_OP_WRITE, 13021c6fdbd8SKent Overstreet GFP_NOFS, 13031c6fdbd8SKent Overstreet &c->writepage_bioset), 13049a3df993SKent Overstreet struct bch_writepage_io, op.wbio.bio); 13051c6fdbd8SKent Overstreet 13069a3df993SKent Overstreet w->io->inode = inode; 13079a3df993SKent Overstreet op = &w->io->op; 13089a3df993SKent Overstreet bch2_write_op_init(op, c, w->opts); 13099a3df993SKent Overstreet op->target = w->opts.foreground_target; 13101c6fdbd8SKent Overstreet op->nr_replicas = nr_replicas; 13111c6fdbd8SKent Overstreet op->res.nr_replicas = nr_replicas; 13121c6fdbd8SKent Overstreet op->write_point = writepoint_hashed(inode->ei_last_dirtied); 13138c6d298aSKent Overstreet op->subvol = inode->ei_subvol; 13147f5e31e1SKent Overstreet op->pos = POS(inode->v.i_ino, sector); 13159f311f21SKent Overstreet op->end_io = bch2_writepage_io_done; 13167f5e31e1SKent Overstreet op->wbio.bio.bi_iter.bi_sector = sector; 131750fe5bd6SKent Overstreet op->wbio.bio.bi_opf = wbc_to_write_flags(wbc); 13181c6fdbd8SKent Overstreet } 13191c6fdbd8SKent Overstreet 13201c6fdbd8SKent Overstreet static int __bch2_writepage(struct folio *folio, 13211c6fdbd8SKent Overstreet struct writeback_control *wbc, 13221c6fdbd8SKent Overstreet void *data) 13231c6fdbd8SKent Overstreet { 13241c6fdbd8SKent Overstreet struct page *page = &folio->page; 13251c6fdbd8SKent Overstreet struct bch_inode_info *inode = to_bch_ei(page->mapping->host); 13261c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 13271c6fdbd8SKent Overstreet struct bch_writepage_state *w = data; 13287f5e31e1SKent Overstreet struct bch_page_state *s, orig; 13297f5e31e1SKent Overstreet unsigned i, offset, nr_replicas_this_write = U32_MAX; 13301c6fdbd8SKent Overstreet loff_t i_size = i_size_read(&inode->v); 13311c6fdbd8SKent Overstreet pgoff_t end_index = i_size >> PAGE_SHIFT; 1332e1036a2aSKent Overstreet int ret; 13331c6fdbd8SKent Overstreet 13341c6fdbd8SKent Overstreet EBUG_ON(!PageUptodate(page)); 13351c6fdbd8SKent Overstreet 13361c6fdbd8SKent Overstreet /* Is the page fully inside i_size? */ 13371c6fdbd8SKent Overstreet if (page->index < end_index) 13381c6fdbd8SKent Overstreet goto do_io; 13391c6fdbd8SKent Overstreet 13401c6fdbd8SKent Overstreet /* Is the page fully outside i_size? (truncate in progress) */ 13411c6fdbd8SKent Overstreet offset = i_size & (PAGE_SIZE - 1); 13421c6fdbd8SKent Overstreet if (page->index > end_index || !offset) { 13431c6fdbd8SKent Overstreet unlock_page(page); 13441c6fdbd8SKent Overstreet return 0; 13451c6fdbd8SKent Overstreet } 13461c6fdbd8SKent Overstreet 13471c6fdbd8SKent Overstreet /* 13481c6fdbd8SKent Overstreet * The page straddles i_size. It must be zeroed out on each and every 13491c6fdbd8SKent Overstreet * writepage invocation because it may be mmapped. "A file is mapped 13501c6fdbd8SKent Overstreet * in multiples of the page size. For a file that is not a multiple of 13511c6fdbd8SKent Overstreet * the page size, the remaining memory is zeroed when mapped, and 13521c6fdbd8SKent Overstreet * writes to that region are not written out to the file." 13531c6fdbd8SKent Overstreet */ 13541c6fdbd8SKent Overstreet zero_user_segment(page, offset, PAGE_SIZE); 13551c6fdbd8SKent Overstreet do_io: 1356f57a6a5dSKent Overstreet s = bch2_page_state_create(page, __GFP_NOFAIL); 1357f81b648dSKent Overstreet 1358f74a5051SKent Overstreet /* 1359f74a5051SKent Overstreet * Things get really hairy with errors during writeback: 1360f74a5051SKent Overstreet */ 1361f74a5051SKent Overstreet ret = bch2_get_page_disk_reservation(c, inode, page, false); 1362f74a5051SKent Overstreet BUG_ON(ret); 1363e1036a2aSKent Overstreet 13647f5e31e1SKent Overstreet /* Before unlocking the page, get copy of reservations: */ 1365f74a5051SKent Overstreet spin_lock(&s->lock); 13667f5e31e1SKent Overstreet orig = *s; 1367f74a5051SKent Overstreet spin_unlock(&s->lock); 13687f5e31e1SKent Overstreet 13697f5e31e1SKent Overstreet for (i = 0; i < PAGE_SECTORS; i++) { 13702ba5d38bSKent Overstreet if (s->s[i].state < SECTOR_DIRTY) 13717f5e31e1SKent Overstreet continue; 13727f5e31e1SKent Overstreet 1373f81b648dSKent Overstreet nr_replicas_this_write = 1374f57a6a5dSKent Overstreet min_t(unsigned, nr_replicas_this_write, 1375f57a6a5dSKent Overstreet s->s[i].nr_replicas + 1376f57a6a5dSKent Overstreet s->s[i].replicas_reserved); 13777f5e31e1SKent Overstreet } 13781c6fdbd8SKent Overstreet 1379f57a6a5dSKent Overstreet for (i = 0; i < PAGE_SECTORS; i++) { 13802ba5d38bSKent Overstreet if (s->s[i].state < SECTOR_DIRTY) 13817f5e31e1SKent Overstreet continue; 13827f5e31e1SKent Overstreet 1383f57a6a5dSKent Overstreet s->s[i].nr_replicas = w->opts.compression 1384f57a6a5dSKent Overstreet ? 0 : nr_replicas_this_write; 1385e1036a2aSKent Overstreet 1386f57a6a5dSKent Overstreet s->s[i].replicas_reserved = 0; 1387f57a6a5dSKent Overstreet s->s[i].state = SECTOR_ALLOCATED; 1388f57a6a5dSKent Overstreet } 13891c6fdbd8SKent Overstreet 13907f5e31e1SKent Overstreet BUG_ON(atomic_read(&s->write_count)); 13917f5e31e1SKent Overstreet atomic_set(&s->write_count, 1); 13927f5e31e1SKent Overstreet 13931c6fdbd8SKent Overstreet BUG_ON(PageWriteback(page)); 13941c6fdbd8SKent Overstreet set_page_writeback(page); 13957f5e31e1SKent Overstreet 13961c6fdbd8SKent Overstreet unlock_page(page); 13971c6fdbd8SKent Overstreet 13987f5e31e1SKent Overstreet offset = 0; 13997f5e31e1SKent Overstreet while (1) { 1400f74a5051SKent Overstreet unsigned sectors = 0, dirty_sectors = 0, reserved_sectors = 0; 14017f5e31e1SKent Overstreet u64 sector; 14027f5e31e1SKent Overstreet 14037f5e31e1SKent Overstreet while (offset < PAGE_SECTORS && 14042ba5d38bSKent Overstreet orig.s[offset].state < SECTOR_DIRTY) 14057f5e31e1SKent Overstreet offset++; 14067f5e31e1SKent Overstreet 14077f5e31e1SKent Overstreet if (offset == PAGE_SECTORS) 14087f5e31e1SKent Overstreet break; 14097f5e31e1SKent Overstreet 14107f5e31e1SKent Overstreet while (offset + sectors < PAGE_SECTORS && 1411f74a5051SKent Overstreet orig.s[offset + sectors].state >= SECTOR_DIRTY) { 1412f74a5051SKent Overstreet reserved_sectors += orig.s[offset + sectors].replicas_reserved; 1413f74a5051SKent Overstreet dirty_sectors += orig.s[offset + sectors].state == SECTOR_DIRTY; 14147f5e31e1SKent Overstreet sectors++; 14157f5e31e1SKent Overstreet } 1416f74a5051SKent Overstreet BUG_ON(!sectors); 1417f74a5051SKent Overstreet 14187279c1a2SKent Overstreet sector = ((u64) page->index << PAGE_SECTORS_SHIFT) + offset; 14197f5e31e1SKent Overstreet 14201c6fdbd8SKent Overstreet if (w->io && 14219a3df993SKent Overstreet (w->io->op.res.nr_replicas != nr_replicas_this_write || 14229a3df993SKent Overstreet bio_full(&w->io->op.wbio.bio, PAGE_SIZE) || 1423f59b3464SKent Overstreet w->io->op.wbio.bio.bi_iter.bi_size + (sectors << 9) >= 1424f59b3464SKent Overstreet (BIO_MAX_VECS * PAGE_SIZE) || 14259a3df993SKent Overstreet bio_end_sector(&w->io->op.wbio.bio) != sector)) 14261c6fdbd8SKent Overstreet bch2_writepage_do_io(w); 14271c6fdbd8SKent Overstreet 14281c6fdbd8SKent Overstreet if (!w->io) 142950fe5bd6SKent Overstreet bch2_writepage_io_alloc(c, wbc, w, inode, sector, 1430f81b648dSKent Overstreet nr_replicas_this_write); 14311c6fdbd8SKent Overstreet 14327f5e31e1SKent Overstreet atomic_inc(&s->write_count); 14337f5e31e1SKent Overstreet 14349a3df993SKent Overstreet BUG_ON(inode != w->io->inode); 14359a3df993SKent Overstreet BUG_ON(!bio_add_page(&w->io->op.wbio.bio, page, 14367f5e31e1SKent Overstreet sectors << 9, offset << 9)); 14371c6fdbd8SKent Overstreet 14386cc3535dSKent Overstreet /* Check for writing past i_size: */ 1439f8494d25SKent Overstreet WARN_ON_ONCE((bio_end_sector(&w->io->op.wbio.bio) << 9) > 14406cc3535dSKent Overstreet round_up(i_size, block_bytes(c))); 14416cc3535dSKent Overstreet 14429a3df993SKent Overstreet w->io->op.res.sectors += reserved_sectors; 14439a3df993SKent Overstreet w->io->op.i_sectors_delta -= dirty_sectors; 14441c6fdbd8SKent Overstreet w->io->op.new_i_size = i_size; 14451c6fdbd8SKent Overstreet 14467f5e31e1SKent Overstreet offset += sectors; 14477f5e31e1SKent Overstreet } 14487f5e31e1SKent Overstreet 14497f5e31e1SKent Overstreet if (atomic_dec_and_test(&s->write_count)) 14507f5e31e1SKent Overstreet end_page_writeback(page); 14517f5e31e1SKent Overstreet 14521c6fdbd8SKent Overstreet return 0; 14531c6fdbd8SKent Overstreet } 14541c6fdbd8SKent Overstreet 14551c6fdbd8SKent Overstreet int bch2_writepages(struct address_space *mapping, struct writeback_control *wbc) 14561c6fdbd8SKent Overstreet { 14571c6fdbd8SKent Overstreet struct bch_fs *c = mapping->host->i_sb->s_fs_info; 14581c6fdbd8SKent Overstreet struct bch_writepage_state w = 14591c6fdbd8SKent Overstreet bch_writepage_state_init(c, to_bch_ei(mapping->host)); 14601c6fdbd8SKent Overstreet struct blk_plug plug; 14611c6fdbd8SKent Overstreet int ret; 14621c6fdbd8SKent Overstreet 14631c6fdbd8SKent Overstreet blk_start_plug(&plug); 14641c6fdbd8SKent Overstreet ret = write_cache_pages(mapping, wbc, __bch2_writepage, &w); 14651c6fdbd8SKent Overstreet if (w.io) 14661c6fdbd8SKent Overstreet bch2_writepage_do_io(&w); 14671c6fdbd8SKent Overstreet blk_finish_plug(&plug); 14685c1ef830SKent Overstreet return bch2_err_class(ret); 14691c6fdbd8SKent Overstreet } 14701c6fdbd8SKent Overstreet 14711c6fdbd8SKent Overstreet /* buffered writes: */ 14721c6fdbd8SKent Overstreet 14731c6fdbd8SKent Overstreet int bch2_write_begin(struct file *file, struct address_space *mapping, 14741c6fdbd8SKent Overstreet loff_t pos, unsigned len, 14751c6fdbd8SKent Overstreet struct page **pagep, void **fsdata) 14761c6fdbd8SKent Overstreet { 14771c6fdbd8SKent Overstreet struct bch_inode_info *inode = to_bch_ei(mapping->host); 14781c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 1479d1542e03SKent Overstreet struct bch2_page_reservation *res; 14801c6fdbd8SKent Overstreet pgoff_t index = pos >> PAGE_SHIFT; 14811c6fdbd8SKent Overstreet unsigned offset = pos & (PAGE_SIZE - 1); 14821c6fdbd8SKent Overstreet struct page *page; 14831c6fdbd8SKent Overstreet int ret = -ENOMEM; 14841c6fdbd8SKent Overstreet 1485d1542e03SKent Overstreet res = kmalloc(sizeof(*res), GFP_KERNEL); 1486d1542e03SKent Overstreet if (!res) 1487d1542e03SKent Overstreet return -ENOMEM; 1488d1542e03SKent Overstreet 1489d1542e03SKent Overstreet bch2_page_reservation_init(c, inode, res); 1490d1542e03SKent Overstreet *fsdata = res; 14911c6fdbd8SKent Overstreet 14921c6fdbd8SKent Overstreet bch2_pagecache_add_get(&inode->ei_pagecache_lock); 14931c6fdbd8SKent Overstreet 14941c6fdbd8SKent Overstreet page = grab_cache_page_write_begin(mapping, index); 14951c6fdbd8SKent Overstreet if (!page) 14961c6fdbd8SKent Overstreet goto err_unlock; 14971c6fdbd8SKent Overstreet 14981c6fdbd8SKent Overstreet if (PageUptodate(page)) 14991c6fdbd8SKent Overstreet goto out; 15001c6fdbd8SKent Overstreet 15011c6fdbd8SKent Overstreet /* If we're writing entire page, don't need to read it in first: */ 15021c6fdbd8SKent Overstreet if (len == PAGE_SIZE) 15031c6fdbd8SKent Overstreet goto out; 15041c6fdbd8SKent Overstreet 15051c6fdbd8SKent Overstreet if (!offset && pos + len >= inode->v.i_size) { 15061c6fdbd8SKent Overstreet zero_user_segment(page, len, PAGE_SIZE); 15071c6fdbd8SKent Overstreet flush_dcache_page(page); 15081c6fdbd8SKent Overstreet goto out; 15091c6fdbd8SKent Overstreet } 15101c6fdbd8SKent Overstreet 15111c6fdbd8SKent Overstreet if (index > inode->v.i_size >> PAGE_SHIFT) { 15121c6fdbd8SKent Overstreet zero_user_segments(page, 0, offset, offset + len, PAGE_SIZE); 15131c6fdbd8SKent Overstreet flush_dcache_page(page); 15141c6fdbd8SKent Overstreet goto out; 15151c6fdbd8SKent Overstreet } 15161c6fdbd8SKent Overstreet readpage: 15171c6fdbd8SKent Overstreet ret = bch2_read_single_page(page, mapping); 15181c6fdbd8SKent Overstreet if (ret) 15191c6fdbd8SKent Overstreet goto err; 15201c6fdbd8SKent Overstreet out: 1521e6ec361fSKent Overstreet if (!bch2_page_state_create(page, __GFP_NOFAIL)->uptodate) { 1522e6ec361fSKent Overstreet ret = bch2_page_state_set(c, inode_inum(inode), &page, 1); 1523e6ec361fSKent Overstreet if (ret) 1524e6ec361fSKent Overstreet goto out; 1525e6ec361fSKent Overstreet } 1526e6ec361fSKent Overstreet 1527d1542e03SKent Overstreet ret = bch2_page_reservation_get(c, inode, page, res, 1528d1542e03SKent Overstreet offset, len, true); 15291c6fdbd8SKent Overstreet if (ret) { 15301c6fdbd8SKent Overstreet if (!PageUptodate(page)) { 15311c6fdbd8SKent Overstreet /* 15321c6fdbd8SKent Overstreet * If the page hasn't been read in, we won't know if we 15331c6fdbd8SKent Overstreet * actually need a reservation - we don't actually need 15341c6fdbd8SKent Overstreet * to read here, we just need to check if the page is 15351c6fdbd8SKent Overstreet * fully backed by uncompressed data: 15361c6fdbd8SKent Overstreet */ 15371c6fdbd8SKent Overstreet goto readpage; 15381c6fdbd8SKent Overstreet } 15391c6fdbd8SKent Overstreet 15401c6fdbd8SKent Overstreet goto err; 15411c6fdbd8SKent Overstreet } 15421c6fdbd8SKent Overstreet 15431c6fdbd8SKent Overstreet *pagep = page; 15441c6fdbd8SKent Overstreet return 0; 15451c6fdbd8SKent Overstreet err: 15461c6fdbd8SKent Overstreet unlock_page(page); 15471c6fdbd8SKent Overstreet put_page(page); 15481c6fdbd8SKent Overstreet *pagep = NULL; 15491c6fdbd8SKent Overstreet err_unlock: 15501c6fdbd8SKent Overstreet bch2_pagecache_add_put(&inode->ei_pagecache_lock); 1551d1542e03SKent Overstreet kfree(res); 1552d1542e03SKent Overstreet *fsdata = NULL; 15535c1ef830SKent Overstreet return bch2_err_class(ret); 15541c6fdbd8SKent Overstreet } 15551c6fdbd8SKent Overstreet 15561c6fdbd8SKent Overstreet int bch2_write_end(struct file *file, struct address_space *mapping, 15571c6fdbd8SKent Overstreet loff_t pos, unsigned len, unsigned copied, 15581c6fdbd8SKent Overstreet struct page *page, void *fsdata) 15591c6fdbd8SKent Overstreet { 15601c6fdbd8SKent Overstreet struct bch_inode_info *inode = to_bch_ei(mapping->host); 15611c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 1562d1542e03SKent Overstreet struct bch2_page_reservation *res = fsdata; 1563d1542e03SKent Overstreet unsigned offset = pos & (PAGE_SIZE - 1); 15641c6fdbd8SKent Overstreet 15651c6fdbd8SKent Overstreet lockdep_assert_held(&inode->v.i_rwsem); 15661c6fdbd8SKent Overstreet 15671c6fdbd8SKent Overstreet if (unlikely(copied < len && !PageUptodate(page))) { 15681c6fdbd8SKent Overstreet /* 15691c6fdbd8SKent Overstreet * The page needs to be read in, but that would destroy 15701c6fdbd8SKent Overstreet * our partial write - simplest thing is to just force 15711c6fdbd8SKent Overstreet * userspace to redo the write: 15721c6fdbd8SKent Overstreet */ 15731c6fdbd8SKent Overstreet zero_user(page, 0, PAGE_SIZE); 15741c6fdbd8SKent Overstreet flush_dcache_page(page); 15751c6fdbd8SKent Overstreet copied = 0; 15761c6fdbd8SKent Overstreet } 15771c6fdbd8SKent Overstreet 15781c6fdbd8SKent Overstreet spin_lock(&inode->v.i_lock); 15791c6fdbd8SKent Overstreet if (pos + copied > inode->v.i_size) 15801c6fdbd8SKent Overstreet i_size_write(&inode->v, pos + copied); 15811c6fdbd8SKent Overstreet spin_unlock(&inode->v.i_lock); 15821c6fdbd8SKent Overstreet 15831c6fdbd8SKent Overstreet if (copied) { 15841c6fdbd8SKent Overstreet if (!PageUptodate(page)) 15851c6fdbd8SKent Overstreet SetPageUptodate(page); 1586d1542e03SKent Overstreet 1587d1542e03SKent Overstreet bch2_set_page_dirty(c, inode, page, res, offset, copied); 15881c6fdbd8SKent Overstreet 15891c6fdbd8SKent Overstreet inode->ei_last_dirtied = (unsigned long) current; 15901c6fdbd8SKent Overstreet } 15911c6fdbd8SKent Overstreet 15921c6fdbd8SKent Overstreet unlock_page(page); 15931c6fdbd8SKent Overstreet put_page(page); 15941c6fdbd8SKent Overstreet bch2_pagecache_add_put(&inode->ei_pagecache_lock); 15951c6fdbd8SKent Overstreet 1596d1542e03SKent Overstreet bch2_page_reservation_put(c, inode, res); 1597d1542e03SKent Overstreet kfree(res); 1598d1542e03SKent Overstreet 15991c6fdbd8SKent Overstreet return copied; 16001c6fdbd8SKent Overstreet } 16011c6fdbd8SKent Overstreet 16021c6fdbd8SKent Overstreet #define WRITE_BATCH_PAGES 32 16031c6fdbd8SKent Overstreet 16041c6fdbd8SKent Overstreet static int __bch2_buffered_write(struct bch_inode_info *inode, 16051c6fdbd8SKent Overstreet struct address_space *mapping, 16061c6fdbd8SKent Overstreet struct iov_iter *iter, 16071c6fdbd8SKent Overstreet loff_t pos, unsigned len) 16081c6fdbd8SKent Overstreet { 16091c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 16101c6fdbd8SKent Overstreet struct page *pages[WRITE_BATCH_PAGES]; 1611d1542e03SKent Overstreet struct bch2_page_reservation res; 16121c6fdbd8SKent Overstreet unsigned long index = pos >> PAGE_SHIFT; 16131c6fdbd8SKent Overstreet unsigned offset = pos & (PAGE_SIZE - 1); 16141c6fdbd8SKent Overstreet unsigned nr_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE); 1615d1542e03SKent Overstreet unsigned i, reserved = 0, set_dirty = 0; 1616d1542e03SKent Overstreet unsigned copied = 0, nr_pages_copied = 0; 16171c6fdbd8SKent Overstreet int ret = 0; 16181c6fdbd8SKent Overstreet 16191c6fdbd8SKent Overstreet BUG_ON(!len); 16201c6fdbd8SKent Overstreet BUG_ON(nr_pages > ARRAY_SIZE(pages)); 16211c6fdbd8SKent Overstreet 1622d1542e03SKent Overstreet bch2_page_reservation_init(c, inode, &res); 1623d1542e03SKent Overstreet 16241c6fdbd8SKent Overstreet for (i = 0; i < nr_pages; i++) { 16251c6fdbd8SKent Overstreet pages[i] = grab_cache_page_write_begin(mapping, index + i); 16261c6fdbd8SKent Overstreet if (!pages[i]) { 16271c6fdbd8SKent Overstreet nr_pages = i; 16288de819f8SKent Overstreet if (!i) { 16291c6fdbd8SKent Overstreet ret = -ENOMEM; 16301c6fdbd8SKent Overstreet goto out; 16311c6fdbd8SKent Overstreet } 16328de819f8SKent Overstreet len = min_t(unsigned, len, 16338de819f8SKent Overstreet nr_pages * PAGE_SIZE - offset); 16348de819f8SKent Overstreet break; 16358de819f8SKent Overstreet } 16361c6fdbd8SKent Overstreet } 16371c6fdbd8SKent Overstreet 16381c6fdbd8SKent Overstreet if (offset && !PageUptodate(pages[0])) { 16391c6fdbd8SKent Overstreet ret = bch2_read_single_page(pages[0], mapping); 16401c6fdbd8SKent Overstreet if (ret) 16411c6fdbd8SKent Overstreet goto out; 16421c6fdbd8SKent Overstreet } 16431c6fdbd8SKent Overstreet 16441c6fdbd8SKent Overstreet if ((pos + len) & (PAGE_SIZE - 1) && 16451c6fdbd8SKent Overstreet !PageUptodate(pages[nr_pages - 1])) { 16461c6fdbd8SKent Overstreet if ((index + nr_pages - 1) << PAGE_SHIFT >= inode->v.i_size) { 16471c6fdbd8SKent Overstreet zero_user(pages[nr_pages - 1], 0, PAGE_SIZE); 16481c6fdbd8SKent Overstreet } else { 16491c6fdbd8SKent Overstreet ret = bch2_read_single_page(pages[nr_pages - 1], mapping); 16501c6fdbd8SKent Overstreet if (ret) 16511c6fdbd8SKent Overstreet goto out; 16521c6fdbd8SKent Overstreet } 16531c6fdbd8SKent Overstreet } 16541c6fdbd8SKent Overstreet 1655d1542e03SKent Overstreet while (reserved < len) { 1656e6ec361fSKent Overstreet unsigned i = (offset + reserved) >> PAGE_SHIFT; 1657e6ec361fSKent Overstreet struct page *page = pages[i]; 1658d1542e03SKent Overstreet unsigned pg_offset = (offset + reserved) & (PAGE_SIZE - 1); 1659d1542e03SKent Overstreet unsigned pg_len = min_t(unsigned, len - reserved, 1660d1542e03SKent Overstreet PAGE_SIZE - pg_offset); 16611c6fdbd8SKent Overstreet 1662e6ec361fSKent Overstreet if (!bch2_page_state_create(page, __GFP_NOFAIL)->uptodate) { 1663e6ec361fSKent Overstreet ret = bch2_page_state_set(c, inode_inum(inode), 1664e6ec361fSKent Overstreet pages + i, nr_pages - i); 1665e6ec361fSKent Overstreet if (ret) 1666e6ec361fSKent Overstreet goto out; 16671c6fdbd8SKent Overstreet } 16681c6fdbd8SKent Overstreet 1669e6ec361fSKent Overstreet ret = bch2_page_reservation_get(c, inode, page, &res, 1670e6ec361fSKent Overstreet pg_offset, pg_len, true); 16711c6fdbd8SKent Overstreet if (ret) 16721c6fdbd8SKent Overstreet goto out; 1673d1542e03SKent Overstreet 1674d1542e03SKent Overstreet reserved += pg_len; 16751c6fdbd8SKent Overstreet } 16761c6fdbd8SKent Overstreet 16771c6fdbd8SKent Overstreet if (mapping_writably_mapped(mapping)) 16781c6fdbd8SKent Overstreet for (i = 0; i < nr_pages; i++) 16791c6fdbd8SKent Overstreet flush_dcache_page(pages[i]); 16801c6fdbd8SKent Overstreet 16811c6fdbd8SKent Overstreet while (copied < len) { 16821c6fdbd8SKent Overstreet struct page *page = pages[(offset + copied) >> PAGE_SHIFT]; 16831c6fdbd8SKent Overstreet unsigned pg_offset = (offset + copied) & (PAGE_SIZE - 1); 1684d1542e03SKent Overstreet unsigned pg_len = min_t(unsigned, len - copied, 16851c6fdbd8SKent Overstreet PAGE_SIZE - pg_offset); 16861c6fdbd8SKent Overstreet unsigned pg_copied = copy_page_from_iter_atomic(page, 1687d1542e03SKent Overstreet pg_offset, pg_len, iter); 1688d1542e03SKent Overstreet 1689d1542e03SKent Overstreet if (!pg_copied) 1690d1542e03SKent Overstreet break; 16911c6fdbd8SKent Overstreet 1692912bdf17SKent Overstreet if (!PageUptodate(page) && 1693912bdf17SKent Overstreet pg_copied != PAGE_SIZE && 1694912bdf17SKent Overstreet pos + copied + pg_copied < inode->v.i_size) { 1695912bdf17SKent Overstreet zero_user(page, 0, PAGE_SIZE); 1696912bdf17SKent Overstreet break; 1697912bdf17SKent Overstreet } 1698912bdf17SKent Overstreet 16991c6fdbd8SKent Overstreet flush_dcache_page(page); 17001c6fdbd8SKent Overstreet copied += pg_copied; 1701912bdf17SKent Overstreet 1702912bdf17SKent Overstreet if (pg_copied != pg_len) 1703912bdf17SKent Overstreet break; 17041c6fdbd8SKent Overstreet } 17051c6fdbd8SKent Overstreet 17061c6fdbd8SKent Overstreet if (!copied) 17071c6fdbd8SKent Overstreet goto out; 17081c6fdbd8SKent Overstreet 1709877dfb34SKent Overstreet spin_lock(&inode->v.i_lock); 1710877dfb34SKent Overstreet if (pos + copied > inode->v.i_size) 1711877dfb34SKent Overstreet i_size_write(&inode->v, pos + copied); 1712877dfb34SKent Overstreet spin_unlock(&inode->v.i_lock); 1713877dfb34SKent Overstreet 1714d1542e03SKent Overstreet while (set_dirty < copied) { 1715d1542e03SKent Overstreet struct page *page = pages[(offset + set_dirty) >> PAGE_SHIFT]; 1716d1542e03SKent Overstreet unsigned pg_offset = (offset + set_dirty) & (PAGE_SIZE - 1); 1717d1542e03SKent Overstreet unsigned pg_len = min_t(unsigned, copied - set_dirty, 1718d1542e03SKent Overstreet PAGE_SIZE - pg_offset); 1719d1542e03SKent Overstreet 1720d1542e03SKent Overstreet if (!PageUptodate(page)) 1721d1542e03SKent Overstreet SetPageUptodate(page); 1722d1542e03SKent Overstreet 1723d1542e03SKent Overstreet bch2_set_page_dirty(c, inode, page, &res, pg_offset, pg_len); 1724d1542e03SKent Overstreet unlock_page(page); 1725d1542e03SKent Overstreet put_page(page); 1726d1542e03SKent Overstreet 1727d1542e03SKent Overstreet set_dirty += pg_len; 1728d1542e03SKent Overstreet } 1729877dfb34SKent Overstreet 1730877dfb34SKent Overstreet nr_pages_copied = DIV_ROUND_UP(offset + copied, PAGE_SIZE); 1731877dfb34SKent Overstreet inode->ei_last_dirtied = (unsigned long) current; 17321c6fdbd8SKent Overstreet out: 1733d1542e03SKent Overstreet for (i = nr_pages_copied; i < nr_pages; i++) { 17341c6fdbd8SKent Overstreet unlock_page(pages[i]); 17351c6fdbd8SKent Overstreet put_page(pages[i]); 17361c6fdbd8SKent Overstreet } 17371c6fdbd8SKent Overstreet 1738d1542e03SKent Overstreet bch2_page_reservation_put(c, inode, &res); 17391c6fdbd8SKent Overstreet 17401c6fdbd8SKent Overstreet return copied ?: ret; 17411c6fdbd8SKent Overstreet } 17421c6fdbd8SKent Overstreet 17431c6fdbd8SKent Overstreet static ssize_t bch2_buffered_write(struct kiocb *iocb, struct iov_iter *iter) 17441c6fdbd8SKent Overstreet { 17451c6fdbd8SKent Overstreet struct file *file = iocb->ki_filp; 17461c6fdbd8SKent Overstreet struct address_space *mapping = file->f_mapping; 17471c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 17481c6fdbd8SKent Overstreet loff_t pos = iocb->ki_pos; 17491c6fdbd8SKent Overstreet ssize_t written = 0; 17501c6fdbd8SKent Overstreet int ret = 0; 17511c6fdbd8SKent Overstreet 17521c6fdbd8SKent Overstreet bch2_pagecache_add_get(&inode->ei_pagecache_lock); 17531c6fdbd8SKent Overstreet 17541c6fdbd8SKent Overstreet do { 17551c6fdbd8SKent Overstreet unsigned offset = pos & (PAGE_SIZE - 1); 17561c6fdbd8SKent Overstreet unsigned bytes = min_t(unsigned long, iov_iter_count(iter), 17571c6fdbd8SKent Overstreet PAGE_SIZE * WRITE_BATCH_PAGES - offset); 17581c6fdbd8SKent Overstreet again: 17591c6fdbd8SKent Overstreet /* 17601c6fdbd8SKent Overstreet * Bring in the user page that we will copy from _first_. 17611c6fdbd8SKent Overstreet * Otherwise there's a nasty deadlock on copying from the 17621c6fdbd8SKent Overstreet * same page as we're writing to, without it being marked 17631c6fdbd8SKent Overstreet * up-to-date. 17641c6fdbd8SKent Overstreet * 17651c6fdbd8SKent Overstreet * Not only is this an optimisation, but it is also required 17661c6fdbd8SKent Overstreet * to check that the address is actually valid, when atomic 17671c6fdbd8SKent Overstreet * usercopies are used, below. 17681c6fdbd8SKent Overstreet */ 17691c6fdbd8SKent Overstreet if (unlikely(fault_in_iov_iter_readable(iter, bytes))) { 17701c6fdbd8SKent Overstreet bytes = min_t(unsigned long, iov_iter_count(iter), 17711c6fdbd8SKent Overstreet PAGE_SIZE - offset); 17721c6fdbd8SKent Overstreet 17731c6fdbd8SKent Overstreet if (unlikely(fault_in_iov_iter_readable(iter, bytes))) { 17741c6fdbd8SKent Overstreet ret = -EFAULT; 17751c6fdbd8SKent Overstreet break; 17761c6fdbd8SKent Overstreet } 17771c6fdbd8SKent Overstreet } 17781c6fdbd8SKent Overstreet 17791c6fdbd8SKent Overstreet if (unlikely(fatal_signal_pending(current))) { 17801c6fdbd8SKent Overstreet ret = -EINTR; 17811c6fdbd8SKent Overstreet break; 17821c6fdbd8SKent Overstreet } 17831c6fdbd8SKent Overstreet 17841c6fdbd8SKent Overstreet ret = __bch2_buffered_write(inode, mapping, iter, pos, bytes); 17851c6fdbd8SKent Overstreet if (unlikely(ret < 0)) 17861c6fdbd8SKent Overstreet break; 17871c6fdbd8SKent Overstreet 17881c6fdbd8SKent Overstreet cond_resched(); 17891c6fdbd8SKent Overstreet 17901c6fdbd8SKent Overstreet if (unlikely(ret == 0)) { 17911c6fdbd8SKent Overstreet /* 17921c6fdbd8SKent Overstreet * If we were unable to copy any data at all, we must 17931c6fdbd8SKent Overstreet * fall back to a single segment length write. 17941c6fdbd8SKent Overstreet * 17951c6fdbd8SKent Overstreet * If we didn't fallback here, we could livelock 17961c6fdbd8SKent Overstreet * because not all segments in the iov can be copied at 17971c6fdbd8SKent Overstreet * once without a pagefault. 17981c6fdbd8SKent Overstreet */ 17991c6fdbd8SKent Overstreet bytes = min_t(unsigned long, PAGE_SIZE - offset, 18001c6fdbd8SKent Overstreet iov_iter_single_seg_count(iter)); 18011c6fdbd8SKent Overstreet goto again; 18021c6fdbd8SKent Overstreet } 18031c6fdbd8SKent Overstreet pos += ret; 18041c6fdbd8SKent Overstreet written += ret; 1805912bdf17SKent Overstreet ret = 0; 18061c6fdbd8SKent Overstreet 18071c6fdbd8SKent Overstreet balance_dirty_pages_ratelimited(mapping); 18081c6fdbd8SKent Overstreet } while (iov_iter_count(iter)); 18091c6fdbd8SKent Overstreet 18101c6fdbd8SKent Overstreet bch2_pagecache_add_put(&inode->ei_pagecache_lock); 18111c6fdbd8SKent Overstreet 18121c6fdbd8SKent Overstreet return written ? written : ret; 18131c6fdbd8SKent Overstreet } 18141c6fdbd8SKent Overstreet 18151c6fdbd8SKent Overstreet /* O_DIRECT reads */ 18161c6fdbd8SKent Overstreet 1817b4725cc1SKent Overstreet static void bio_check_or_release(struct bio *bio, bool check_dirty) 1818b4725cc1SKent Overstreet { 1819b4725cc1SKent Overstreet if (check_dirty) { 1820b4725cc1SKent Overstreet bio_check_pages_dirty(bio); 1821b4725cc1SKent Overstreet } else { 1822b4725cc1SKent Overstreet bio_release_pages(bio, false); 1823b4725cc1SKent Overstreet bio_put(bio); 1824b4725cc1SKent Overstreet } 1825b4725cc1SKent Overstreet } 1826b4725cc1SKent Overstreet 18271c6fdbd8SKent Overstreet static void bch2_dio_read_complete(struct closure *cl) 18281c6fdbd8SKent Overstreet { 18291c6fdbd8SKent Overstreet struct dio_read *dio = container_of(cl, struct dio_read, cl); 18301c6fdbd8SKent Overstreet 18311c6fdbd8SKent Overstreet dio->req->ki_complete(dio->req, dio->ret); 1832b4725cc1SKent Overstreet bio_check_or_release(&dio->rbio.bio, dio->should_dirty); 18331c6fdbd8SKent Overstreet } 18341c6fdbd8SKent Overstreet 18351c6fdbd8SKent Overstreet static void bch2_direct_IO_read_endio(struct bio *bio) 18361c6fdbd8SKent Overstreet { 18371c6fdbd8SKent Overstreet struct dio_read *dio = bio->bi_private; 18381c6fdbd8SKent Overstreet 18391c6fdbd8SKent Overstreet if (bio->bi_status) 18401c6fdbd8SKent Overstreet dio->ret = blk_status_to_errno(bio->bi_status); 18411c6fdbd8SKent Overstreet 18421c6fdbd8SKent Overstreet closure_put(&dio->cl); 18431c6fdbd8SKent Overstreet } 18441c6fdbd8SKent Overstreet 18451c6fdbd8SKent Overstreet static void bch2_direct_IO_read_split_endio(struct bio *bio) 18461c6fdbd8SKent Overstreet { 1847b4725cc1SKent Overstreet struct dio_read *dio = bio->bi_private; 1848b4725cc1SKent Overstreet bool should_dirty = dio->should_dirty; 1849b4725cc1SKent Overstreet 18501c6fdbd8SKent Overstreet bch2_direct_IO_read_endio(bio); 1851b4725cc1SKent Overstreet bio_check_or_release(bio, should_dirty); 18521c6fdbd8SKent Overstreet } 18531c6fdbd8SKent Overstreet 18541c6fdbd8SKent Overstreet static int bch2_direct_IO_read(struct kiocb *req, struct iov_iter *iter) 18551c6fdbd8SKent Overstreet { 18561c6fdbd8SKent Overstreet struct file *file = req->ki_filp; 18571c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 18581c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 18599a3df993SKent Overstreet struct bch_io_opts opts = io_opts(c, &inode->ei_inode); 18601c6fdbd8SKent Overstreet struct dio_read *dio; 18611c6fdbd8SKent Overstreet struct bio *bio; 18621c6fdbd8SKent Overstreet loff_t offset = req->ki_pos; 18631c6fdbd8SKent Overstreet bool sync = is_sync_kiocb(req); 18641c6fdbd8SKent Overstreet size_t shorten; 18651c6fdbd8SKent Overstreet ssize_t ret; 18661c6fdbd8SKent Overstreet 18671c6fdbd8SKent Overstreet if ((offset|iter->count) & (block_bytes(c) - 1)) 18681c6fdbd8SKent Overstreet return -EINVAL; 18691c6fdbd8SKent Overstreet 18701c6fdbd8SKent Overstreet ret = min_t(loff_t, iter->count, 18711c6fdbd8SKent Overstreet max_t(loff_t, 0, i_size_read(&inode->v) - offset)); 18721c6fdbd8SKent Overstreet 18731c6fdbd8SKent Overstreet if (!ret) 18741c6fdbd8SKent Overstreet return ret; 18751c6fdbd8SKent Overstreet 18761c6fdbd8SKent Overstreet shorten = iov_iter_count(iter) - round_up(ret, block_bytes(c)); 18771c6fdbd8SKent Overstreet iter->count -= shorten; 18781c6fdbd8SKent Overstreet 18791c6fdbd8SKent Overstreet bio = bio_alloc_bioset(NULL, 18804d126dc8SKent Overstreet bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS), 18811c6fdbd8SKent Overstreet REQ_OP_READ, 18821c6fdbd8SKent Overstreet GFP_KERNEL, 18831c6fdbd8SKent Overstreet &c->dio_read_bioset); 18841c6fdbd8SKent Overstreet 18851c6fdbd8SKent Overstreet bio->bi_end_io = bch2_direct_IO_read_endio; 18861c6fdbd8SKent Overstreet 18871c6fdbd8SKent Overstreet dio = container_of(bio, struct dio_read, rbio.bio); 18881c6fdbd8SKent Overstreet closure_init(&dio->cl, NULL); 18891c6fdbd8SKent Overstreet 18901c6fdbd8SKent Overstreet /* 18911c6fdbd8SKent Overstreet * this is a _really_ horrible hack just to avoid an atomic sub at the 18921c6fdbd8SKent Overstreet * end: 18931c6fdbd8SKent Overstreet */ 18941c6fdbd8SKent Overstreet if (!sync) { 18951c6fdbd8SKent Overstreet set_closure_fn(&dio->cl, bch2_dio_read_complete, NULL); 18961c6fdbd8SKent Overstreet atomic_set(&dio->cl.remaining, 18971c6fdbd8SKent Overstreet CLOSURE_REMAINING_INITIALIZER - 18981c6fdbd8SKent Overstreet CLOSURE_RUNNING + 18991c6fdbd8SKent Overstreet CLOSURE_DESTRUCTOR); 19001c6fdbd8SKent Overstreet } else { 19011c6fdbd8SKent Overstreet atomic_set(&dio->cl.remaining, 19021c6fdbd8SKent Overstreet CLOSURE_REMAINING_INITIALIZER + 1); 19031c6fdbd8SKent Overstreet } 19041c6fdbd8SKent Overstreet 19051c6fdbd8SKent Overstreet dio->req = req; 19061c6fdbd8SKent Overstreet dio->ret = ret; 1907b4725cc1SKent Overstreet /* 1908b4725cc1SKent Overstreet * This is one of the sketchier things I've encountered: we have to skip 1909b4725cc1SKent Overstreet * the dirtying of requests that are internal from the kernel (i.e. from 1910b4725cc1SKent Overstreet * loopback), because we'll deadlock on page_lock. 1911b4725cc1SKent Overstreet */ 1912b4725cc1SKent Overstreet dio->should_dirty = iter_is_iovec(iter); 19131c6fdbd8SKent Overstreet 19141c6fdbd8SKent Overstreet goto start; 19151c6fdbd8SKent Overstreet while (iter->count) { 19161c6fdbd8SKent Overstreet bio = bio_alloc_bioset(NULL, 19174d126dc8SKent Overstreet bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS), 19181c6fdbd8SKent Overstreet REQ_OP_READ, 19191c6fdbd8SKent Overstreet GFP_KERNEL, 19201c6fdbd8SKent Overstreet &c->bio_read); 19211c6fdbd8SKent Overstreet bio->bi_end_io = bch2_direct_IO_read_split_endio; 19221c6fdbd8SKent Overstreet start: 19231c6fdbd8SKent Overstreet bio->bi_opf = REQ_OP_READ|REQ_SYNC; 19241c6fdbd8SKent Overstreet bio->bi_iter.bi_sector = offset >> 9; 19251c6fdbd8SKent Overstreet bio->bi_private = dio; 19261c6fdbd8SKent Overstreet 19271c6fdbd8SKent Overstreet ret = bio_iov_iter_get_pages(bio, iter); 19281c6fdbd8SKent Overstreet if (ret < 0) { 19291c6fdbd8SKent Overstreet /* XXX: fault inject this path */ 19301c6fdbd8SKent Overstreet bio->bi_status = BLK_STS_RESOURCE; 19311c6fdbd8SKent Overstreet bio_endio(bio); 19321c6fdbd8SKent Overstreet break; 19331c6fdbd8SKent Overstreet } 19341c6fdbd8SKent Overstreet 19351c6fdbd8SKent Overstreet offset += bio->bi_iter.bi_size; 1936b4725cc1SKent Overstreet 1937b4725cc1SKent Overstreet if (dio->should_dirty) 19381c6fdbd8SKent Overstreet bio_set_pages_dirty(bio); 19391c6fdbd8SKent Overstreet 19401c6fdbd8SKent Overstreet if (iter->count) 19411c6fdbd8SKent Overstreet closure_get(&dio->cl); 19421c6fdbd8SKent Overstreet 19438c6d298aSKent Overstreet bch2_read(c, rbio_init(bio, opts), inode_inum(inode)); 19441c6fdbd8SKent Overstreet } 19451c6fdbd8SKent Overstreet 19461c6fdbd8SKent Overstreet iter->count += shorten; 19471c6fdbd8SKent Overstreet 19481c6fdbd8SKent Overstreet if (sync) { 19491c6fdbd8SKent Overstreet closure_sync(&dio->cl); 19501c6fdbd8SKent Overstreet closure_debug_destroy(&dio->cl); 19511c6fdbd8SKent Overstreet ret = dio->ret; 1952b4725cc1SKent Overstreet bio_check_or_release(&dio->rbio.bio, dio->should_dirty); 19531c6fdbd8SKent Overstreet return ret; 19541c6fdbd8SKent Overstreet } else { 19551c6fdbd8SKent Overstreet return -EIOCBQUEUED; 19561c6fdbd8SKent Overstreet } 19571c6fdbd8SKent Overstreet } 19581c6fdbd8SKent Overstreet 19591c6fdbd8SKent Overstreet ssize_t bch2_read_iter(struct kiocb *iocb, struct iov_iter *iter) 19601c6fdbd8SKent Overstreet { 19611c6fdbd8SKent Overstreet struct file *file = iocb->ki_filp; 19621c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 19631c6fdbd8SKent Overstreet struct address_space *mapping = file->f_mapping; 19641c6fdbd8SKent Overstreet size_t count = iov_iter_count(iter); 19651c6fdbd8SKent Overstreet ssize_t ret; 19661c6fdbd8SKent Overstreet 19671c6fdbd8SKent Overstreet if (!count) 19681c6fdbd8SKent Overstreet return 0; /* skip atime */ 19691c6fdbd8SKent Overstreet 19701c6fdbd8SKent Overstreet if (iocb->ki_flags & IOCB_DIRECT) { 19711c6fdbd8SKent Overstreet struct blk_plug plug; 19721c6fdbd8SKent Overstreet 1973a023127aSKent Overstreet if (unlikely(mapping->nrpages)) { 19741c6fdbd8SKent Overstreet ret = filemap_write_and_wait_range(mapping, 19751c6fdbd8SKent Overstreet iocb->ki_pos, 19761c6fdbd8SKent Overstreet iocb->ki_pos + count - 1); 19771c6fdbd8SKent Overstreet if (ret < 0) 19785c1ef830SKent Overstreet goto out; 1979a023127aSKent Overstreet } 19801c6fdbd8SKent Overstreet 19811c6fdbd8SKent Overstreet file_accessed(file); 19821c6fdbd8SKent Overstreet 19831c6fdbd8SKent Overstreet blk_start_plug(&plug); 19841c6fdbd8SKent Overstreet ret = bch2_direct_IO_read(iocb, iter); 19851c6fdbd8SKent Overstreet blk_finish_plug(&plug); 19861c6fdbd8SKent Overstreet 19871c6fdbd8SKent Overstreet if (ret >= 0) 19881c6fdbd8SKent Overstreet iocb->ki_pos += ret; 19891c6fdbd8SKent Overstreet } else { 19901c6fdbd8SKent Overstreet bch2_pagecache_add_get(&inode->ei_pagecache_lock); 19911c6fdbd8SKent Overstreet ret = generic_file_read_iter(iocb, iter); 19921c6fdbd8SKent Overstreet bch2_pagecache_add_put(&inode->ei_pagecache_lock); 19931c6fdbd8SKent Overstreet } 19945c1ef830SKent Overstreet out: 19955c1ef830SKent Overstreet return bch2_err_class(ret); 19961c6fdbd8SKent Overstreet } 19971c6fdbd8SKent Overstreet 19981c6fdbd8SKent Overstreet /* O_DIRECT writes */ 19991c6fdbd8SKent Overstreet 20006fed42bbSKent Overstreet static bool bch2_check_range_allocated(struct bch_fs *c, subvol_inum inum, 20016fed42bbSKent Overstreet u64 offset, u64 size, 20026fed42bbSKent Overstreet unsigned nr_replicas, bool compressed) 20036fed42bbSKent Overstreet { 20046fed42bbSKent Overstreet struct btree_trans trans; 20056fed42bbSKent Overstreet struct btree_iter iter; 20066fed42bbSKent Overstreet struct bkey_s_c k; 20076fed42bbSKent Overstreet u64 end = offset + size; 20086fed42bbSKent Overstreet u32 snapshot; 20096fed42bbSKent Overstreet bool ret = true; 20106fed42bbSKent Overstreet int err; 20116fed42bbSKent Overstreet 20126fed42bbSKent Overstreet bch2_trans_init(&trans, c, 0, 0); 20136fed42bbSKent Overstreet retry: 20146fed42bbSKent Overstreet bch2_trans_begin(&trans); 20156fed42bbSKent Overstreet 20166fed42bbSKent Overstreet err = bch2_subvolume_get_snapshot(&trans, inum.subvol, &snapshot); 20176fed42bbSKent Overstreet if (err) 20186fed42bbSKent Overstreet goto err; 20196fed42bbSKent Overstreet 2020e5fa91d7SKent Overstreet for_each_btree_key_norestart(&trans, iter, BTREE_ID_extents, 20216fed42bbSKent Overstreet SPOS(inum.inum, offset, snapshot), 20226fed42bbSKent Overstreet BTREE_ITER_SLOTS, k, err) { 20236fed42bbSKent Overstreet if (bkey_cmp(bkey_start_pos(k.k), POS(inum.inum, end)) >= 0) 20246fed42bbSKent Overstreet break; 20256fed42bbSKent Overstreet 20268c6d298aSKent Overstreet if (k.k->p.snapshot != snapshot || 20278c6d298aSKent Overstreet nr_replicas > bch2_bkey_replicas(c, k) || 20286fed42bbSKent Overstreet (!compressed && bch2_bkey_sectors_compressed(k))) { 20296fed42bbSKent Overstreet ret = false; 20306fed42bbSKent Overstreet break; 20316fed42bbSKent Overstreet } 20326fed42bbSKent Overstreet } 20336fed42bbSKent Overstreet 20346fed42bbSKent Overstreet offset = iter.pos.offset; 20356fed42bbSKent Overstreet bch2_trans_iter_exit(&trans, &iter); 20366fed42bbSKent Overstreet err: 2037549d173cSKent Overstreet if (bch2_err_matches(err, BCH_ERR_transaction_restart)) 20386fed42bbSKent Overstreet goto retry; 20396fed42bbSKent Overstreet bch2_trans_exit(&trans); 20406fed42bbSKent Overstreet 20416fed42bbSKent Overstreet return err ? false : ret; 20426fed42bbSKent Overstreet } 20436fed42bbSKent Overstreet 20441c6fdbd8SKent Overstreet /* 20451c6fdbd8SKent Overstreet * We're going to return -EIOCBQUEUED, but we haven't finished consuming the 20461c6fdbd8SKent Overstreet * iov_iter yet, so we need to stash a copy of the iovec: it might be on the 20471c6fdbd8SKent Overstreet * caller's stack, we're not guaranteed that it will live for the duration of 20481c6fdbd8SKent Overstreet * the IO: 20491c6fdbd8SKent Overstreet */ 20501c6fdbd8SKent Overstreet static noinline int bch2_dio_write_copy_iov(struct dio_write *dio) 20511c6fdbd8SKent Overstreet { 20521c6fdbd8SKent Overstreet struct iovec *iov = dio->inline_vecs; 20531c6fdbd8SKent Overstreet 20541c6fdbd8SKent Overstreet /* 20551c6fdbd8SKent Overstreet * iov_iter has a single embedded iovec - nothing to do: 20561c6fdbd8SKent Overstreet */ 20571c6fdbd8SKent Overstreet if (iter_is_ubuf(&dio->iter)) 20581c6fdbd8SKent Overstreet return 0; 20591c6fdbd8SKent Overstreet 20601c6fdbd8SKent Overstreet /* 20611c6fdbd8SKent Overstreet * We don't currently handle non-iovec iov_iters here - return an error, 20621c6fdbd8SKent Overstreet * and we'll fall back to doing the IO synchronously: 20631c6fdbd8SKent Overstreet */ 20641c6fdbd8SKent Overstreet if (!iter_is_iovec(&dio->iter)) 20651c6fdbd8SKent Overstreet return -1; 20661c6fdbd8SKent Overstreet 20671c6fdbd8SKent Overstreet if (dio->iter.nr_segs > ARRAY_SIZE(dio->inline_vecs)) { 20681c6fdbd8SKent Overstreet iov = kmalloc_array(dio->iter.nr_segs, sizeof(*iov), 20691c6fdbd8SKent Overstreet GFP_KERNEL); 20701c6fdbd8SKent Overstreet if (unlikely(!iov)) 20711c6fdbd8SKent Overstreet return -ENOMEM; 20721c6fdbd8SKent Overstreet 20731c6fdbd8SKent Overstreet dio->free_iov = true; 20741c6fdbd8SKent Overstreet } 20751c6fdbd8SKent Overstreet 20761c6fdbd8SKent Overstreet memcpy(iov, dio->iter.__iov, dio->iter.nr_segs * sizeof(*iov)); 20771c6fdbd8SKent Overstreet dio->iter.__iov = iov; 20781c6fdbd8SKent Overstreet return 0; 20791c6fdbd8SKent Overstreet } 20801c6fdbd8SKent Overstreet 2081042a1f26SKent Overstreet static void bch2_dio_write_loop_async(struct bch_write_op *); 2082042a1f26SKent Overstreet 20831c6fdbd8SKent Overstreet static long bch2_dio_write_loop(struct dio_write *dio) 20841c6fdbd8SKent Overstreet { 2085ed484030SKent Overstreet bool kthread = (current->flags & PF_KTHREAD) != 0; 20861c6fdbd8SKent Overstreet struct kiocb *req = dio->req; 20871c6fdbd8SKent Overstreet struct address_space *mapping = req->ki_filp->f_mapping; 20889a3df993SKent Overstreet struct bch_inode_info *inode = file_bch_inode(req->ki_filp); 2089042a1f26SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 20909a3df993SKent Overstreet struct bio *bio = &dio->op.wbio.bio; 2091eb8e6e9cSKent Overstreet unsigned unaligned, iter_count; 2092eb8e6e9cSKent Overstreet bool sync = dio->sync, dropped_locks; 20931c6fdbd8SKent Overstreet long ret; 20941c6fdbd8SKent Overstreet 20951c6fdbd8SKent Overstreet if (dio->loop) 20961c6fdbd8SKent Overstreet goto loop; 20971c6fdbd8SKent Overstreet 20981c6fdbd8SKent Overstreet while (1) { 2099eb8e6e9cSKent Overstreet iter_count = dio->iter.count; 2100eb8e6e9cSKent Overstreet 21019552e19fSKent Overstreet if (kthread && dio->mm) 2102ed484030SKent Overstreet kthread_use_mm(dio->mm); 21031c6fdbd8SKent Overstreet BUG_ON(current->faults_disabled_mapping); 21041c6fdbd8SKent Overstreet current->faults_disabled_mapping = mapping; 21051c6fdbd8SKent Overstreet 21061c6fdbd8SKent Overstreet ret = bio_iov_iter_get_pages(bio, &dio->iter); 21071c6fdbd8SKent Overstreet 2108eb8e6e9cSKent Overstreet dropped_locks = fdm_dropped_locks(); 2109eb8e6e9cSKent Overstreet 21101c6fdbd8SKent Overstreet current->faults_disabled_mapping = NULL; 21119552e19fSKent Overstreet if (kthread && dio->mm) 2112ed484030SKent Overstreet kthread_unuse_mm(dio->mm); 21131c6fdbd8SKent Overstreet 2114eb8e6e9cSKent Overstreet /* 2115eb8e6e9cSKent Overstreet * If the fault handler returned an error but also signalled 2116eb8e6e9cSKent Overstreet * that it dropped & retook ei_pagecache_lock, we just need to 2117eb8e6e9cSKent Overstreet * re-shoot down the page cache and retry: 2118eb8e6e9cSKent Overstreet */ 2119eb8e6e9cSKent Overstreet if (dropped_locks && ret) 2120eb8e6e9cSKent Overstreet ret = 0; 2121eb8e6e9cSKent Overstreet 21221c6fdbd8SKent Overstreet if (unlikely(ret < 0)) 21231c6fdbd8SKent Overstreet goto err; 21241c6fdbd8SKent Overstreet 2125eb8e6e9cSKent Overstreet if (unlikely(dropped_locks)) { 2126eb8e6e9cSKent Overstreet ret = write_invalidate_inode_pages_range(mapping, 2127eb8e6e9cSKent Overstreet req->ki_pos, 2128eb8e6e9cSKent Overstreet req->ki_pos + iter_count - 1); 2129eb8e6e9cSKent Overstreet if (unlikely(ret)) 2130eb8e6e9cSKent Overstreet goto err; 2131eb8e6e9cSKent Overstreet 2132eb8e6e9cSKent Overstreet if (!bio->bi_iter.bi_size) 2133eb8e6e9cSKent Overstreet continue; 2134eb8e6e9cSKent Overstreet } 2135eb8e6e9cSKent Overstreet 21360a426c32SKent Overstreet unaligned = bio->bi_iter.bi_size & (block_bytes(c) - 1); 21370a426c32SKent Overstreet bio->bi_iter.bi_size -= unaligned; 21380a426c32SKent Overstreet iov_iter_revert(&dio->iter, unaligned); 21390a426c32SKent Overstreet 21400a426c32SKent Overstreet if (!bio->bi_iter.bi_size) { 21410a426c32SKent Overstreet /* 21420a426c32SKent Overstreet * bio_iov_iter_get_pages was only able to get < 21430a426c32SKent Overstreet * blocksize worth of pages: 21440a426c32SKent Overstreet */ 21450a426c32SKent Overstreet ret = -EFAULT; 21460a426c32SKent Overstreet goto err; 21470a426c32SKent Overstreet } 21480a426c32SKent Overstreet 2149042a1f26SKent Overstreet bch2_write_op_init(&dio->op, c, io_opts(c, &inode->ei_inode)); 2150042a1f26SKent Overstreet dio->op.end_io = bch2_dio_write_loop_async; 2151042a1f26SKent Overstreet dio->op.target = dio->op.opts.foreground_target; 2152042a1f26SKent Overstreet dio->op.write_point = writepoint_hashed((unsigned long) current); 2153042a1f26SKent Overstreet dio->op.nr_replicas = dio->op.opts.data_replicas; 21548c6d298aSKent Overstreet dio->op.subvol = inode->ei_subvol; 2155042a1f26SKent Overstreet dio->op.pos = POS(inode->v.i_ino, (u64) req->ki_pos >> 9); 2156042a1f26SKent Overstreet 2157042a1f26SKent Overstreet if ((req->ki_flags & IOCB_DSYNC) && 2158042a1f26SKent Overstreet !c->opts.journal_flush_disabled) 2159042a1f26SKent Overstreet dio->op.flags |= BCH_WRITE_FLUSH; 2160a6336910SKent Overstreet dio->op.flags |= BCH_WRITE_CHECK_ENOSPC; 2161042a1f26SKent Overstreet 2162042a1f26SKent Overstreet ret = bch2_disk_reservation_get(c, &dio->op.res, bio_sectors(bio), 2163042a1f26SKent Overstreet dio->op.opts.data_replicas, 0); 2164042a1f26SKent Overstreet if (unlikely(ret) && 21656fed42bbSKent Overstreet !bch2_check_range_allocated(c, inode_inum(inode), 21666fed42bbSKent Overstreet dio->op.pos.offset, bio_sectors(bio), 216735a067b4SKent Overstreet dio->op.opts.data_replicas, 216835a067b4SKent Overstreet dio->op.opts.compression != 0)) 2169042a1f26SKent Overstreet goto err; 21701c6fdbd8SKent Overstreet 21711c6fdbd8SKent Overstreet task_io_account_write(bio->bi_iter.bi_size); 21721c6fdbd8SKent Overstreet 21731c6fdbd8SKent Overstreet if (!dio->sync && !dio->loop && dio->iter.count) { 21741c6fdbd8SKent Overstreet if (bch2_dio_write_copy_iov(dio)) { 2175286d8ad0SKent Overstreet dio->sync = sync = true; 2176f8f30863SKent Overstreet goto do_io; 21771c6fdbd8SKent Overstreet } 21781c6fdbd8SKent Overstreet } 2179f8f30863SKent Overstreet do_io: 21801c6fdbd8SKent Overstreet dio->loop = true; 2181f8f30863SKent Overstreet closure_call(&dio->op.cl, bch2_write, NULL, NULL); 21821c6fdbd8SKent Overstreet 2183286d8ad0SKent Overstreet if (sync) 2184f8f30863SKent Overstreet wait_for_completion(&dio->done); 2185f8f30863SKent Overstreet else 21861c6fdbd8SKent Overstreet return -EIOCBQUEUED; 21871c6fdbd8SKent Overstreet loop: 21889a3df993SKent Overstreet i_sectors_acct(c, inode, &dio->quota_res, 21899a3df993SKent Overstreet dio->op.i_sectors_delta); 2190042a1f26SKent Overstreet req->ki_pos += (u64) dio->op.written << 9; 2191042a1f26SKent Overstreet dio->written += dio->op.written; 21929a3df993SKent Overstreet 21939a3df993SKent Overstreet spin_lock(&inode->v.i_lock); 2194042a1f26SKent Overstreet if (req->ki_pos > inode->v.i_size) 2195042a1f26SKent Overstreet i_size_write(&inode->v, req->ki_pos); 21969a3df993SKent Overstreet spin_unlock(&inode->v.i_lock); 21979a3df993SKent Overstreet 2198032ac32cSKent Overstreet bio_release_pages(bio, false); 21995468f119SKent Overstreet bio->bi_vcnt = 0; 220033c74e41SKent Overstreet 220133c74e41SKent Overstreet if (dio->op.error) { 220233c74e41SKent Overstreet set_bit(EI_INODE_ERROR, &inode->ei_flags); 220333c74e41SKent Overstreet break; 220433c74e41SKent Overstreet } 220533c74e41SKent Overstreet 220633c74e41SKent Overstreet if (!dio->iter.count) 22071c6fdbd8SKent Overstreet break; 2208f8f30863SKent Overstreet 22091c6fdbd8SKent Overstreet bio_reset(bio, NULL, REQ_OP_WRITE); 2210f8f30863SKent Overstreet reinit_completion(&dio->done); 22111c6fdbd8SKent Overstreet } 22121c6fdbd8SKent Overstreet 2213042a1f26SKent Overstreet ret = dio->op.error ?: ((long) dio->written << 9); 22141c6fdbd8SKent Overstreet err: 22151c6fdbd8SKent Overstreet bch2_pagecache_block_put(&inode->ei_pagecache_lock); 22160a426c32SKent Overstreet bch2_quota_reservation_put(c, inode, &dio->quota_res); 22171c6fdbd8SKent Overstreet 22181c6fdbd8SKent Overstreet if (dio->free_iov) 22191c6fdbd8SKent Overstreet kfree(dio->iter.__iov); 22201c6fdbd8SKent Overstreet 22215468f119SKent Overstreet bio_release_pages(bio, false); 22221c6fdbd8SKent Overstreet bio_put(bio); 22231c6fdbd8SKent Overstreet 22241c6fdbd8SKent Overstreet /* inode->i_dio_count is our ref on inode and thus bch_fs */ 22251c6fdbd8SKent Overstreet inode_dio_end(&inode->v); 22261c6fdbd8SKent Overstreet 22275c1ef830SKent Overstreet if (ret < 0) 22285c1ef830SKent Overstreet ret = bch2_err_class(ret); 22295c1ef830SKent Overstreet 22301c6fdbd8SKent Overstreet if (!sync) { 22311c6fdbd8SKent Overstreet req->ki_complete(req, ret); 22321c6fdbd8SKent Overstreet ret = -EIOCBQUEUED; 22331c6fdbd8SKent Overstreet } 22341c6fdbd8SKent Overstreet return ret; 22351c6fdbd8SKent Overstreet } 22361c6fdbd8SKent Overstreet 2237f8f30863SKent Overstreet static void bch2_dio_write_loop_async(struct bch_write_op *op) 22381c6fdbd8SKent Overstreet { 2239f8f30863SKent Overstreet struct dio_write *dio = container_of(op, struct dio_write, op); 22401c6fdbd8SKent Overstreet 2241f8f30863SKent Overstreet if (dio->sync) 2242f8f30863SKent Overstreet complete(&dio->done); 2243f8f30863SKent Overstreet else 22441c6fdbd8SKent Overstreet bch2_dio_write_loop(dio); 22451c6fdbd8SKent Overstreet } 22461c6fdbd8SKent Overstreet 22471c6fdbd8SKent Overstreet static noinline 22481c6fdbd8SKent Overstreet ssize_t bch2_direct_write(struct kiocb *req, struct iov_iter *iter) 22491c6fdbd8SKent Overstreet { 22501c6fdbd8SKent Overstreet struct file *file = req->ki_filp; 225154847d25SKent Overstreet struct address_space *mapping = file->f_mapping; 22521c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 22531c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 22541c6fdbd8SKent Overstreet struct dio_write *dio; 22551c6fdbd8SKent Overstreet struct bio *bio; 22567edcfbfeSKent Overstreet bool locked = true, extending; 22571c6fdbd8SKent Overstreet ssize_t ret; 22581c6fdbd8SKent Overstreet 22597edcfbfeSKent Overstreet prefetch(&c->opts); 22607edcfbfeSKent Overstreet prefetch((void *) &c->opts + 64); 22617edcfbfeSKent Overstreet prefetch(&inode->ei_inode); 22627edcfbfeSKent Overstreet prefetch((void *) &inode->ei_inode + 64); 22631c6fdbd8SKent Overstreet 22647edcfbfeSKent Overstreet inode_lock(&inode->v); 22657edcfbfeSKent Overstreet 22667edcfbfeSKent Overstreet ret = generic_write_checks(req, iter); 22677edcfbfeSKent Overstreet if (unlikely(ret <= 0)) 22687edcfbfeSKent Overstreet goto err; 22697edcfbfeSKent Overstreet 22707edcfbfeSKent Overstreet ret = file_remove_privs(file); 22717edcfbfeSKent Overstreet if (unlikely(ret)) 22727edcfbfeSKent Overstreet goto err; 22737edcfbfeSKent Overstreet 22747edcfbfeSKent Overstreet ret = file_update_time(file); 22757edcfbfeSKent Overstreet if (unlikely(ret)) 22767edcfbfeSKent Overstreet goto err; 22771c6fdbd8SKent Overstreet 2278919dbbd1SKent Overstreet if (unlikely((req->ki_pos|iter->count) & (block_bytes(c) - 1))) 22797edcfbfeSKent Overstreet goto err; 22807edcfbfeSKent Overstreet 22817edcfbfeSKent Overstreet inode_dio_begin(&inode->v); 22827edcfbfeSKent Overstreet bch2_pagecache_block_get(&inode->ei_pagecache_lock); 22837edcfbfeSKent Overstreet 22847edcfbfeSKent Overstreet extending = req->ki_pos + iter->count > inode->v.i_size; 22857edcfbfeSKent Overstreet if (!extending) { 22867edcfbfeSKent Overstreet inode_unlock(&inode->v); 22877edcfbfeSKent Overstreet locked = false; 22887edcfbfeSKent Overstreet } 22891c6fdbd8SKent Overstreet 22901c6fdbd8SKent Overstreet bio = bio_alloc_bioset(NULL, 22914d126dc8SKent Overstreet bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS), 22921c6fdbd8SKent Overstreet REQ_OP_WRITE, 22931c6fdbd8SKent Overstreet GFP_KERNEL, 22941c6fdbd8SKent Overstreet &c->dio_write_bioset); 22959a3df993SKent Overstreet dio = container_of(bio, struct dio_write, op.wbio.bio); 2296f8f30863SKent Overstreet init_completion(&dio->done); 22971c6fdbd8SKent Overstreet dio->req = req; 2298ed484030SKent Overstreet dio->mm = current->mm; 22991c6fdbd8SKent Overstreet dio->loop = false; 23007edcfbfeSKent Overstreet dio->sync = is_sync_kiocb(req) || extending; 23011c6fdbd8SKent Overstreet dio->free_iov = false; 23021c6fdbd8SKent Overstreet dio->quota_res.sectors = 0; 2303042a1f26SKent Overstreet dio->written = 0; 23041c6fdbd8SKent Overstreet dio->iter = *iter; 23059a3df993SKent Overstreet 23061c6fdbd8SKent Overstreet ret = bch2_quota_reservation_add(c, inode, &dio->quota_res, 23071c6fdbd8SKent Overstreet iter->count >> 9, true); 23081c6fdbd8SKent Overstreet if (unlikely(ret)) 23097edcfbfeSKent Overstreet goto err_put_bio; 23101c6fdbd8SKent Overstreet 2311a023127aSKent Overstreet if (unlikely(mapping->nrpages)) { 231254847d25SKent Overstreet ret = write_invalidate_inode_pages_range(mapping, 231354847d25SKent Overstreet req->ki_pos, 231454847d25SKent Overstreet req->ki_pos + iter->count - 1); 231554847d25SKent Overstreet if (unlikely(ret)) 231654847d25SKent Overstreet goto err_put_bio; 2317a023127aSKent Overstreet } 231854847d25SKent Overstreet 23197edcfbfeSKent Overstreet ret = bch2_dio_write_loop(dio); 23201c6fdbd8SKent Overstreet err: 23217edcfbfeSKent Overstreet if (locked) 23227edcfbfeSKent Overstreet inode_unlock(&inode->v); 23237edcfbfeSKent Overstreet return ret; 23247edcfbfeSKent Overstreet err_put_bio: 23257edcfbfeSKent Overstreet bch2_pagecache_block_put(&inode->ei_pagecache_lock); 23261c6fdbd8SKent Overstreet bch2_quota_reservation_put(c, inode, &dio->quota_res); 23271c6fdbd8SKent Overstreet bio_put(bio); 23287edcfbfeSKent Overstreet inode_dio_end(&inode->v); 23297edcfbfeSKent Overstreet goto err; 23301c6fdbd8SKent Overstreet } 23311c6fdbd8SKent Overstreet 23327edcfbfeSKent Overstreet ssize_t bch2_write_iter(struct kiocb *iocb, struct iov_iter *from) 23331c6fdbd8SKent Overstreet { 23341c6fdbd8SKent Overstreet struct file *file = iocb->ki_filp; 23357edcfbfeSKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 23361c6fdbd8SKent Overstreet ssize_t ret; 23371c6fdbd8SKent Overstreet 23385c1ef830SKent Overstreet if (iocb->ki_flags & IOCB_DIRECT) { 23395c1ef830SKent Overstreet ret = bch2_direct_write(iocb, from); 23405c1ef830SKent Overstreet goto out; 23415c1ef830SKent Overstreet } 23421c6fdbd8SKent Overstreet 23437edcfbfeSKent Overstreet inode_lock(&inode->v); 23447edcfbfeSKent Overstreet 23457edcfbfeSKent Overstreet ret = generic_write_checks(iocb, from); 23467edcfbfeSKent Overstreet if (ret <= 0) 23477edcfbfeSKent Overstreet goto unlock; 23487edcfbfeSKent Overstreet 23491c6fdbd8SKent Overstreet ret = file_remove_privs(file); 23501c6fdbd8SKent Overstreet if (ret) 23517edcfbfeSKent Overstreet goto unlock; 23521c6fdbd8SKent Overstreet 23531c6fdbd8SKent Overstreet ret = file_update_time(file); 23541c6fdbd8SKent Overstreet if (ret) 23557edcfbfeSKent Overstreet goto unlock; 23561c6fdbd8SKent Overstreet 23577edcfbfeSKent Overstreet ret = bch2_buffered_write(iocb, from); 23581c6fdbd8SKent Overstreet if (likely(ret > 0)) 23591c6fdbd8SKent Overstreet iocb->ki_pos += ret; 23607edcfbfeSKent Overstreet unlock: 23611c6fdbd8SKent Overstreet inode_unlock(&inode->v); 23621c6fdbd8SKent Overstreet 23637edcfbfeSKent Overstreet if (ret > 0) 23641c6fdbd8SKent Overstreet ret = generic_write_sync(iocb, ret); 23655c1ef830SKent Overstreet out: 23665c1ef830SKent Overstreet return bch2_err_class(ret); 23671c6fdbd8SKent Overstreet } 23681c6fdbd8SKent Overstreet 23691c6fdbd8SKent Overstreet /* fsync: */ 23701c6fdbd8SKent Overstreet 237168a2054dSKent Overstreet /* 237268a2054dSKent Overstreet * inode->ei_inode.bi_journal_seq won't be up to date since it's set in an 237368a2054dSKent Overstreet * insert trigger: look up the btree inode instead 237468a2054dSKent Overstreet */ 237568a2054dSKent Overstreet static int bch2_flush_inode(struct bch_fs *c, subvol_inum inum) 237668a2054dSKent Overstreet { 237768a2054dSKent Overstreet struct bch_inode_unpacked inode; 237868a2054dSKent Overstreet int ret; 237968a2054dSKent Overstreet 238068a2054dSKent Overstreet if (c->opts.journal_flush_disabled) 238168a2054dSKent Overstreet return 0; 238268a2054dSKent Overstreet 238368a2054dSKent Overstreet ret = bch2_inode_find_by_inum(c, inum, &inode); 238468a2054dSKent Overstreet if (ret) 238568a2054dSKent Overstreet return ret; 238668a2054dSKent Overstreet 238768a2054dSKent Overstreet return bch2_journal_flush_seq(&c->journal, inode.bi_journal_seq); 238868a2054dSKent Overstreet } 238968a2054dSKent Overstreet 23901c6fdbd8SKent Overstreet int bch2_fsync(struct file *file, loff_t start, loff_t end, int datasync) 23911c6fdbd8SKent Overstreet { 23921c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 23931c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 239468a2054dSKent Overstreet int ret, ret2, ret3; 23951c6fdbd8SKent Overstreet 23961c6fdbd8SKent Overstreet ret = file_write_and_wait_range(file, start, end); 239768a2054dSKent Overstreet ret2 = sync_inode_metadata(&inode->v, 1); 239868a2054dSKent Overstreet ret3 = bch2_flush_inode(c, inode_inum(inode)); 23991c6fdbd8SKent Overstreet 24005c1ef830SKent Overstreet return bch2_err_class(ret ?: ret2 ?: ret3); 24011c6fdbd8SKent Overstreet } 24021c6fdbd8SKent Overstreet 24031c6fdbd8SKent Overstreet /* truncate: */ 24041c6fdbd8SKent Overstreet 24056fed42bbSKent Overstreet static inline int range_has_data(struct bch_fs *c, u32 subvol, 24061c6fdbd8SKent Overstreet struct bpos start, 24071c6fdbd8SKent Overstreet struct bpos end) 24081c6fdbd8SKent Overstreet { 2409424eb881SKent Overstreet struct btree_trans trans; 241067e0dd8fSKent Overstreet struct btree_iter iter; 24111c6fdbd8SKent Overstreet struct bkey_s_c k; 24121c6fdbd8SKent Overstreet int ret = 0; 24131c6fdbd8SKent Overstreet 241420bceecbSKent Overstreet bch2_trans_init(&trans, c, 0, 0); 24156fed42bbSKent Overstreet retry: 24166fed42bbSKent Overstreet bch2_trans_begin(&trans); 24176fed42bbSKent Overstreet 24186fed42bbSKent Overstreet ret = bch2_subvolume_get_snapshot(&trans, subvol, &start.snapshot); 24196fed42bbSKent Overstreet if (ret) 24206fed42bbSKent Overstreet goto err; 2421424eb881SKent Overstreet 2422e5fa91d7SKent Overstreet for_each_btree_key_norestart(&trans, iter, BTREE_ID_extents, start, 0, k, ret) { 24231c6fdbd8SKent Overstreet if (bkey_cmp(bkey_start_pos(k.k), end) >= 0) 24241c6fdbd8SKent Overstreet break; 24251c6fdbd8SKent Overstreet 24261c6fdbd8SKent Overstreet if (bkey_extent_is_data(k.k)) { 24271c6fdbd8SKent Overstreet ret = 1; 24281c6fdbd8SKent Overstreet break; 24291c6fdbd8SKent Overstreet } 24301c6fdbd8SKent Overstreet } 24316fed42bbSKent Overstreet start = iter.pos; 243267e0dd8fSKent Overstreet bch2_trans_iter_exit(&trans, &iter); 24336fed42bbSKent Overstreet err: 2434549d173cSKent Overstreet if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) 24356fed42bbSKent Overstreet goto retry; 24361c6fdbd8SKent Overstreet 24379a796fdbSKent Overstreet bch2_trans_exit(&trans); 24389a796fdbSKent Overstreet return ret; 24391c6fdbd8SKent Overstreet } 24401c6fdbd8SKent Overstreet 24411c6fdbd8SKent Overstreet static int __bch2_truncate_page(struct bch_inode_info *inode, 24421c6fdbd8SKent Overstreet pgoff_t index, loff_t start, loff_t end) 24431c6fdbd8SKent Overstreet { 24441c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 24451c6fdbd8SKent Overstreet struct address_space *mapping = inode->v.i_mapping; 2446a99b1cafSKent Overstreet struct bch_page_state *s; 24471c6fdbd8SKent Overstreet unsigned start_offset = start & (PAGE_SIZE - 1); 24481c6fdbd8SKent Overstreet unsigned end_offset = ((end - 1) & (PAGE_SIZE - 1)) + 1; 2449a99b1cafSKent Overstreet unsigned i; 24501c6fdbd8SKent Overstreet struct page *page; 2451b19d307dSKent Overstreet s64 i_sectors_delta = 0; 24521c6fdbd8SKent Overstreet int ret = 0; 24531c6fdbd8SKent Overstreet 24541c6fdbd8SKent Overstreet /* Page boundary? Nothing to do */ 24551c6fdbd8SKent Overstreet if (!((index == start >> PAGE_SHIFT && start_offset) || 24561c6fdbd8SKent Overstreet (index == end >> PAGE_SHIFT && end_offset != PAGE_SIZE))) 24571c6fdbd8SKent Overstreet return 0; 24581c6fdbd8SKent Overstreet 24591c6fdbd8SKent Overstreet /* Above i_size? */ 24601c6fdbd8SKent Overstreet if (index << PAGE_SHIFT >= inode->v.i_size) 24611c6fdbd8SKent Overstreet return 0; 24621c6fdbd8SKent Overstreet 24631c6fdbd8SKent Overstreet page = find_lock_page(mapping, index); 24641c6fdbd8SKent Overstreet if (!page) { 24651c6fdbd8SKent Overstreet /* 24661c6fdbd8SKent Overstreet * XXX: we're doing two index lookups when we end up reading the 24671c6fdbd8SKent Overstreet * page 24681c6fdbd8SKent Overstreet */ 24696fed42bbSKent Overstreet ret = range_has_data(c, inode->ei_subvol, 24707279c1a2SKent Overstreet POS(inode->v.i_ino, index << PAGE_SECTORS_SHIFT), 24717279c1a2SKent Overstreet POS(inode->v.i_ino, (index + 1) << PAGE_SECTORS_SHIFT)); 24721c6fdbd8SKent Overstreet if (ret <= 0) 24731c6fdbd8SKent Overstreet return ret; 24741c6fdbd8SKent Overstreet 24751c6fdbd8SKent Overstreet page = find_or_create_page(mapping, index, GFP_KERNEL); 24761c6fdbd8SKent Overstreet if (unlikely(!page)) { 24771c6fdbd8SKent Overstreet ret = -ENOMEM; 24781c6fdbd8SKent Overstreet goto out; 24791c6fdbd8SKent Overstreet } 24801c6fdbd8SKent Overstreet } 24811c6fdbd8SKent Overstreet 2482a99b1cafSKent Overstreet s = bch2_page_state_create(page, 0); 2483a99b1cafSKent Overstreet if (!s) { 2484a99b1cafSKent Overstreet ret = -ENOMEM; 2485a99b1cafSKent Overstreet goto unlock; 2486a99b1cafSKent Overstreet } 2487a99b1cafSKent Overstreet 24881c6fdbd8SKent Overstreet if (!PageUptodate(page)) { 24891c6fdbd8SKent Overstreet ret = bch2_read_single_page(page, mapping); 24901c6fdbd8SKent Overstreet if (ret) 24911c6fdbd8SKent Overstreet goto unlock; 24921c6fdbd8SKent Overstreet } 24931c6fdbd8SKent Overstreet 2494a99b1cafSKent Overstreet if (index != start >> PAGE_SHIFT) 2495a99b1cafSKent Overstreet start_offset = 0; 2496a99b1cafSKent Overstreet if (index != end >> PAGE_SHIFT) 2497a99b1cafSKent Overstreet end_offset = PAGE_SIZE; 2498a99b1cafSKent Overstreet 2499a99b1cafSKent Overstreet for (i = round_up(start_offset, block_bytes(c)) >> 9; 2500a99b1cafSKent Overstreet i < round_down(end_offset, block_bytes(c)) >> 9; 2501a99b1cafSKent Overstreet i++) { 2502a99b1cafSKent Overstreet s->s[i].nr_replicas = 0; 2503b19d307dSKent Overstreet if (s->s[i].state == SECTOR_DIRTY) 2504b19d307dSKent Overstreet i_sectors_delta--; 2505a99b1cafSKent Overstreet s->s[i].state = SECTOR_UNALLOCATED; 2506a99b1cafSKent Overstreet } 2507a99b1cafSKent Overstreet 2508b19d307dSKent Overstreet i_sectors_acct(c, inode, NULL, i_sectors_delta); 2509b19d307dSKent Overstreet 251074163da7SKent Overstreet /* 251174163da7SKent Overstreet * Caller needs to know whether this page will be written out by 251274163da7SKent Overstreet * writeback - doing an i_size update if necessary - or whether it will 251374163da7SKent Overstreet * be responsible for the i_size update: 251474163da7SKent Overstreet */ 251574163da7SKent Overstreet ret = s->s[(min_t(u64, inode->v.i_size - (index << PAGE_SHIFT), 251674163da7SKent Overstreet PAGE_SIZE) - 1) >> 9].state >= SECTOR_DIRTY; 251774163da7SKent Overstreet 2518a99b1cafSKent Overstreet zero_user_segment(page, start_offset, end_offset); 2519a99b1cafSKent Overstreet 25201c6fdbd8SKent Overstreet /* 25211c6fdbd8SKent Overstreet * Bit of a hack - we don't want truncate to fail due to -ENOSPC. 25221c6fdbd8SKent Overstreet * 25231c6fdbd8SKent Overstreet * XXX: because we aren't currently tracking whether the page has actual 25241c6fdbd8SKent Overstreet * data in it (vs. just 0s, or only partially written) this wrong. ick. 25251c6fdbd8SKent Overstreet */ 252674163da7SKent Overstreet BUG_ON(bch2_get_page_disk_reservation(c, inode, page, false)); 25271c6fdbd8SKent Overstreet 25289ba2eb25SKent Overstreet /* 25299ba2eb25SKent Overstreet * This removes any writeable userspace mappings; we need to force 25309ba2eb25SKent Overstreet * .page_mkwrite to be called again before any mmapped writes, to 25319ba2eb25SKent Overstreet * redirty the full page: 25329ba2eb25SKent Overstreet */ 25339ba2eb25SKent Overstreet page_mkclean(page); 2534d1542e03SKent Overstreet filemap_dirty_folio(mapping, page_folio(page)); 25351c6fdbd8SKent Overstreet unlock: 25361c6fdbd8SKent Overstreet unlock_page(page); 25371c6fdbd8SKent Overstreet put_page(page); 25381c6fdbd8SKent Overstreet out: 25391c6fdbd8SKent Overstreet return ret; 25401c6fdbd8SKent Overstreet } 25411c6fdbd8SKent Overstreet 25421c6fdbd8SKent Overstreet static int bch2_truncate_page(struct bch_inode_info *inode, loff_t from) 25431c6fdbd8SKent Overstreet { 25441c6fdbd8SKent Overstreet return __bch2_truncate_page(inode, from >> PAGE_SHIFT, 2545a99b1cafSKent Overstreet from, round_up(from, PAGE_SIZE)); 25461c6fdbd8SKent Overstreet } 25471c6fdbd8SKent Overstreet 254874163da7SKent Overstreet static int bch2_truncate_pages(struct bch_inode_info *inode, 254974163da7SKent Overstreet loff_t start, loff_t end) 255074163da7SKent Overstreet { 255174163da7SKent Overstreet int ret = __bch2_truncate_page(inode, start >> PAGE_SHIFT, 255274163da7SKent Overstreet start, end); 255374163da7SKent Overstreet 255474163da7SKent Overstreet if (ret >= 0 && 255574163da7SKent Overstreet start >> PAGE_SHIFT != end >> PAGE_SHIFT) 255674163da7SKent Overstreet ret = __bch2_truncate_page(inode, 255774163da7SKent Overstreet end >> PAGE_SHIFT, 255874163da7SKent Overstreet start, end); 255974163da7SKent Overstreet return ret; 256074163da7SKent Overstreet } 256174163da7SKent Overstreet 256268a507a2SKent Overstreet static int bch2_extend(struct mnt_idmap *idmap, 256368a507a2SKent Overstreet struct bch_inode_info *inode, 2564e0541a93SKent Overstreet struct bch_inode_unpacked *inode_u, 2565e0541a93SKent Overstreet struct iattr *iattr) 25661c6fdbd8SKent Overstreet { 25671c6fdbd8SKent Overstreet struct address_space *mapping = inode->v.i_mapping; 25681c6fdbd8SKent Overstreet int ret; 25691c6fdbd8SKent Overstreet 2570e0541a93SKent Overstreet /* 2571e0541a93SKent Overstreet * sync appends: 25722925fc49SKent Overstreet * 25732925fc49SKent Overstreet * this has to be done _before_ extending i_size: 2574e0541a93SKent Overstreet */ 2575e0541a93SKent Overstreet ret = filemap_write_and_wait_range(mapping, inode_u->bi_size, S64_MAX); 25761c6fdbd8SKent Overstreet if (ret) 25771c6fdbd8SKent Overstreet return ret; 25781c6fdbd8SKent Overstreet 25791c6fdbd8SKent Overstreet truncate_setsize(&inode->v, iattr->ia_size); 25801c6fdbd8SKent Overstreet 258168a507a2SKent Overstreet return bch2_setattr_nonsize(idmap, inode, iattr); 25821c6fdbd8SKent Overstreet } 25831c6fdbd8SKent Overstreet 258454e2264eSKent Overstreet static int bch2_truncate_finish_fn(struct bch_inode_info *inode, 258554e2264eSKent Overstreet struct bch_inode_unpacked *bi, 258654e2264eSKent Overstreet void *p) 258754e2264eSKent Overstreet { 258854e2264eSKent Overstreet bi->bi_flags &= ~BCH_INODE_I_SIZE_DIRTY; 258954e2264eSKent Overstreet return 0; 259054e2264eSKent Overstreet } 259154e2264eSKent Overstreet 259254e2264eSKent Overstreet static int bch2_truncate_start_fn(struct bch_inode_info *inode, 259354e2264eSKent Overstreet struct bch_inode_unpacked *bi, void *p) 259454e2264eSKent Overstreet { 259554e2264eSKent Overstreet u64 *new_i_size = p; 259654e2264eSKent Overstreet 259754e2264eSKent Overstreet bi->bi_flags |= BCH_INODE_I_SIZE_DIRTY; 259854e2264eSKent Overstreet bi->bi_size = *new_i_size; 259954e2264eSKent Overstreet return 0; 260054e2264eSKent Overstreet } 260154e2264eSKent Overstreet 260268a507a2SKent Overstreet int bch2_truncate(struct mnt_idmap *idmap, 260368a507a2SKent Overstreet struct bch_inode_info *inode, struct iattr *iattr) 26041c6fdbd8SKent Overstreet { 26051c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 26061c6fdbd8SKent Overstreet struct address_space *mapping = inode->v.i_mapping; 2607e0541a93SKent Overstreet struct bch_inode_unpacked inode_u; 260854e2264eSKent Overstreet u64 new_i_size = iattr->ia_size; 26092e87eae1SKent Overstreet s64 i_sectors_delta = 0; 26101c6fdbd8SKent Overstreet int ret = 0; 26111c6fdbd8SKent Overstreet 261268a507a2SKent Overstreet /* 261378d66ab1SDan Robertson * If the truncate call with change the size of the file, the 261478d66ab1SDan Robertson * cmtimes should be updated. If the size will not change, we 261578d66ab1SDan Robertson * do not need to update the cmtimes. 261668a507a2SKent Overstreet */ 261778d66ab1SDan Robertson if (iattr->ia_size != inode->v.i_size) { 261868a507a2SKent Overstreet if (!(iattr->ia_valid & ATTR_MTIME)) 261968a507a2SKent Overstreet ktime_get_coarse_real_ts64(&iattr->ia_mtime); 262068a507a2SKent Overstreet if (!(iattr->ia_valid & ATTR_CTIME)) 262168a507a2SKent Overstreet ktime_get_coarse_real_ts64(&iattr->ia_ctime); 262268a507a2SKent Overstreet iattr->ia_valid |= ATTR_MTIME|ATTR_CTIME; 262378d66ab1SDan Robertson } 262468a507a2SKent Overstreet 26251c6fdbd8SKent Overstreet inode_dio_wait(&inode->v); 26261c6fdbd8SKent Overstreet bch2_pagecache_block_get(&inode->ei_pagecache_lock); 26271c6fdbd8SKent Overstreet 26286fed42bbSKent Overstreet ret = bch2_inode_find_by_inum(c, inode_inum(inode), &inode_u); 2629e0541a93SKent Overstreet if (ret) 2630e0541a93SKent Overstreet goto err; 26311c6fdbd8SKent Overstreet 2632c45d473dSKent Overstreet /* 2633c45d473dSKent Overstreet * check this before next assertion; on filesystem error our normal 2634c45d473dSKent Overstreet * invariants are a bit broken (truncate has to truncate the page cache 2635c45d473dSKent Overstreet * before the inode). 2636c45d473dSKent Overstreet */ 2637c45d473dSKent Overstreet ret = bch2_journal_error(&c->journal); 2638c45d473dSKent Overstreet if (ret) 2639c45d473dSKent Overstreet goto err; 2640c45d473dSKent Overstreet 264133c74e41SKent Overstreet WARN_ON(!test_bit(EI_INODE_ERROR, &inode->ei_flags) && 264233c74e41SKent Overstreet inode->v.i_size < inode_u.bi_size); 2643e0541a93SKent Overstreet 2644e0541a93SKent Overstreet if (iattr->ia_size > inode->v.i_size) { 264568a507a2SKent Overstreet ret = bch2_extend(idmap, inode, &inode_u, iattr); 264654e2264eSKent Overstreet goto err; 26471c6fdbd8SKent Overstreet } 26481c6fdbd8SKent Overstreet 264968a507a2SKent Overstreet iattr->ia_valid &= ~ATTR_SIZE; 265068a507a2SKent Overstreet 26511c6fdbd8SKent Overstreet ret = bch2_truncate_page(inode, iattr->ia_size); 265274163da7SKent Overstreet if (unlikely(ret < 0)) 265354e2264eSKent Overstreet goto err; 26541c6fdbd8SKent Overstreet 26556cc3535dSKent Overstreet /* 26566cc3535dSKent Overstreet * When extending, we're going to write the new i_size to disk 26576cc3535dSKent Overstreet * immediately so we need to flush anything above the current on disk 26586cc3535dSKent Overstreet * i_size first: 26596cc3535dSKent Overstreet * 26606cc3535dSKent Overstreet * Also, when extending we need to flush the page that i_size currently 26616cc3535dSKent Overstreet * straddles - if it's mapped to userspace, we need to ensure that 26626cc3535dSKent Overstreet * userspace has to redirty it and call .mkwrite -> set_page_dirty 26636cc3535dSKent Overstreet * again to allocate the part of the page that was extended. 26646cc3535dSKent Overstreet */ 2665e0541a93SKent Overstreet if (iattr->ia_size > inode_u.bi_size) 26661c6fdbd8SKent Overstreet ret = filemap_write_and_wait_range(mapping, 2667e0541a93SKent Overstreet inode_u.bi_size, 26681c6fdbd8SKent Overstreet iattr->ia_size - 1); 26691c6fdbd8SKent Overstreet else if (iattr->ia_size & (PAGE_SIZE - 1)) 26701c6fdbd8SKent Overstreet ret = filemap_write_and_wait_range(mapping, 26711c6fdbd8SKent Overstreet round_down(iattr->ia_size, PAGE_SIZE), 26721c6fdbd8SKent Overstreet iattr->ia_size - 1); 26731c6fdbd8SKent Overstreet if (ret) 267454e2264eSKent Overstreet goto err; 26751c6fdbd8SKent Overstreet 267654e2264eSKent Overstreet mutex_lock(&inode->ei_update_lock); 267754e2264eSKent Overstreet ret = bch2_write_inode(c, inode, bch2_truncate_start_fn, 267854e2264eSKent Overstreet &new_i_size, 0); 267954e2264eSKent Overstreet mutex_unlock(&inode->ei_update_lock); 26801c6fdbd8SKent Overstreet 26811c6fdbd8SKent Overstreet if (unlikely(ret)) 268254e2264eSKent Overstreet goto err; 26831c6fdbd8SKent Overstreet 26841c6fdbd8SKent Overstreet truncate_setsize(&inode->v, iattr->ia_size); 26851c6fdbd8SKent Overstreet 26868c6d298aSKent Overstreet ret = bch2_fpunch(c, inode_inum(inode), 2687a99b1cafSKent Overstreet round_up(iattr->ia_size, block_bytes(c)) >> 9, 268868a2054dSKent Overstreet U64_MAX, &i_sectors_delta); 26892e87eae1SKent Overstreet i_sectors_acct(c, inode, NULL, i_sectors_delta); 26902e87eae1SKent Overstreet 2691b33bf1bcSKent Overstreet bch2_fs_inconsistent_on(!inode->v.i_size && inode->v.i_blocks && 2692b33bf1bcSKent Overstreet !bch2_journal_error(&c->journal), c, 2693b33bf1bcSKent Overstreet "inode %lu truncated to 0 but i_blocks %llu (ondisk %lli)", 2694b33bf1bcSKent Overstreet inode->v.i_ino, (u64) inode->v.i_blocks, 2695b33bf1bcSKent Overstreet inode->ei_inode.bi_sectors); 26961c6fdbd8SKent Overstreet if (unlikely(ret)) 269754e2264eSKent Overstreet goto err; 26981c6fdbd8SKent Overstreet 269954e2264eSKent Overstreet mutex_lock(&inode->ei_update_lock); 270068a507a2SKent Overstreet ret = bch2_write_inode(c, inode, bch2_truncate_finish_fn, NULL, 0); 270154e2264eSKent Overstreet mutex_unlock(&inode->ei_update_lock); 270268a507a2SKent Overstreet 270368a507a2SKent Overstreet ret = bch2_setattr_nonsize(idmap, inode, iattr); 270454e2264eSKent Overstreet err: 27051c6fdbd8SKent Overstreet bch2_pagecache_block_put(&inode->ei_pagecache_lock); 27065c1ef830SKent Overstreet return bch2_err_class(ret); 27071c6fdbd8SKent Overstreet } 27081c6fdbd8SKent Overstreet 27091c6fdbd8SKent Overstreet /* fallocate: */ 27101c6fdbd8SKent Overstreet 2711050197b1SKent Overstreet static int inode_update_times_fn(struct bch_inode_info *inode, 2712050197b1SKent Overstreet struct bch_inode_unpacked *bi, void *p) 2713050197b1SKent Overstreet { 2714050197b1SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 2715050197b1SKent Overstreet 2716050197b1SKent Overstreet bi->bi_mtime = bi->bi_ctime = bch2_current_time(c); 2717050197b1SKent Overstreet return 0; 2718050197b1SKent Overstreet } 2719050197b1SKent Overstreet 27202e87eae1SKent Overstreet static long bchfs_fpunch(struct bch_inode_info *inode, loff_t offset, loff_t len) 27211c6fdbd8SKent Overstreet { 27221c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 272374163da7SKent Overstreet u64 end = offset + len; 272474163da7SKent Overstreet u64 block_start = round_up(offset, block_bytes(c)); 272574163da7SKent Overstreet u64 block_end = round_down(end, block_bytes(c)); 272674163da7SKent Overstreet bool truncated_last_page; 27271c6fdbd8SKent Overstreet int ret = 0; 27281c6fdbd8SKent Overstreet 272974163da7SKent Overstreet ret = bch2_truncate_pages(inode, offset, end); 273074163da7SKent Overstreet if (unlikely(ret < 0)) 27311c6fdbd8SKent Overstreet goto err; 27321c6fdbd8SKent Overstreet 273374163da7SKent Overstreet truncated_last_page = ret; 27341c6fdbd8SKent Overstreet 273574163da7SKent Overstreet truncate_pagecache_range(&inode->v, offset, end - 1); 27361c6fdbd8SKent Overstreet 273774163da7SKent Overstreet if (block_start < block_end ) { 27382e87eae1SKent Overstreet s64 i_sectors_delta = 0; 27392e87eae1SKent Overstreet 27408c6d298aSKent Overstreet ret = bch2_fpunch(c, inode_inum(inode), 274174163da7SKent Overstreet block_start >> 9, block_end >> 9, 27422e87eae1SKent Overstreet &i_sectors_delta); 27432e87eae1SKent Overstreet i_sectors_acct(c, inode, NULL, i_sectors_delta); 27442e87eae1SKent Overstreet } 2745050197b1SKent Overstreet 2746050197b1SKent Overstreet mutex_lock(&inode->ei_update_lock); 274774163da7SKent Overstreet if (end >= inode->v.i_size && !truncated_last_page) { 274874163da7SKent Overstreet ret = bch2_write_inode_size(c, inode, inode->v.i_size, 274974163da7SKent Overstreet ATTR_MTIME|ATTR_CTIME); 275074163da7SKent Overstreet } else { 2751050197b1SKent Overstreet ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL, 275274163da7SKent Overstreet ATTR_MTIME|ATTR_CTIME); 275374163da7SKent Overstreet } 2754050197b1SKent Overstreet mutex_unlock(&inode->ei_update_lock); 27551c6fdbd8SKent Overstreet err: 27561c6fdbd8SKent Overstreet return ret; 27571c6fdbd8SKent Overstreet } 27581c6fdbd8SKent Overstreet 27592e87eae1SKent Overstreet static long bchfs_fcollapse_finsert(struct bch_inode_info *inode, 27605f786787SKent Overstreet loff_t offset, loff_t len, 27615f786787SKent Overstreet bool insert) 27621c6fdbd8SKent Overstreet { 27631c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 27641c6fdbd8SKent Overstreet struct address_space *mapping = inode->v.i_mapping; 276507a1006aSKent Overstreet struct bkey_buf copy; 2766d69f41d6SKent Overstreet struct btree_trans trans; 276767e0dd8fSKent Overstreet struct btree_iter src, dst, del; 27685f786787SKent Overstreet loff_t shift, new_size; 27695f786787SKent Overstreet u64 src_start; 277050dc0f69SKent Overstreet int ret = 0; 27711c6fdbd8SKent Overstreet 27721c6fdbd8SKent Overstreet if ((offset | len) & (block_bytes(c) - 1)) 27731c6fdbd8SKent Overstreet return -EINVAL; 27741c6fdbd8SKent Overstreet 27755f786787SKent Overstreet if (insert) { 27765f786787SKent Overstreet if (inode->v.i_sb->s_maxbytes - inode->v.i_size < len) 277774163da7SKent Overstreet return -EFBIG; 27785f786787SKent Overstreet 27795f786787SKent Overstreet if (offset >= inode->v.i_size) 278074163da7SKent Overstreet return -EINVAL; 27815f786787SKent Overstreet 27825f786787SKent Overstreet src_start = U64_MAX; 27835f786787SKent Overstreet shift = len; 27845f786787SKent Overstreet } else { 27851c6fdbd8SKent Overstreet if (offset + len >= inode->v.i_size) 278674163da7SKent Overstreet return -EINVAL; 27871c6fdbd8SKent Overstreet 27885f786787SKent Overstreet src_start = offset + len; 27895f786787SKent Overstreet shift = -len; 27905f786787SKent Overstreet } 27911c6fdbd8SKent Overstreet 27925f786787SKent Overstreet new_size = inode->v.i_size + shift; 27931c6fdbd8SKent Overstreet 27941c6fdbd8SKent Overstreet ret = write_invalidate_inode_pages_range(mapping, offset, LLONG_MAX); 27951c6fdbd8SKent Overstreet if (ret) 279674163da7SKent Overstreet return ret; 27971c6fdbd8SKent Overstreet 27985f786787SKent Overstreet if (insert) { 27995f786787SKent Overstreet i_size_write(&inode->v, new_size); 28005f786787SKent Overstreet mutex_lock(&inode->ei_update_lock); 28015f786787SKent Overstreet ret = bch2_write_inode_size(c, inode, new_size, 28025f786787SKent Overstreet ATTR_MTIME|ATTR_CTIME); 28035f786787SKent Overstreet mutex_unlock(&inode->ei_update_lock); 28045f786787SKent Overstreet } else { 28052e87eae1SKent Overstreet s64 i_sectors_delta = 0; 28062e87eae1SKent Overstreet 28078c6d298aSKent Overstreet ret = bch2_fpunch(c, inode_inum(inode), 28082e87eae1SKent Overstreet offset >> 9, (offset + len) >> 9, 28092e87eae1SKent Overstreet &i_sectors_delta); 28102e87eae1SKent Overstreet i_sectors_acct(c, inode, NULL, i_sectors_delta); 28112e87eae1SKent Overstreet 281263095894SKent Overstreet if (ret) 281374163da7SKent Overstreet return ret; 28145f786787SKent Overstreet } 28158ef231bdSKent Overstreet 281650dc0f69SKent Overstreet bch2_bkey_buf_init(©); 2817f7beb4caSKent Overstreet bch2_trans_init(&trans, c, BTREE_ITER_MAX, 1024); 281867e0dd8fSKent Overstreet bch2_trans_iter_init(&trans, &src, BTREE_ID_extents, 28195f786787SKent Overstreet POS(inode->v.i_ino, src_start >> 9), 282063095894SKent Overstreet BTREE_ITER_INTENT); 282167e0dd8fSKent Overstreet bch2_trans_copy_iter(&dst, &src); 282267e0dd8fSKent Overstreet bch2_trans_copy_iter(&del, &src); 28235f786787SKent Overstreet 2824549d173cSKent Overstreet while (ret == 0 || 2825549d173cSKent Overstreet bch2_err_matches(ret, BCH_ERR_transaction_restart)) { 282663095894SKent Overstreet struct disk_reservation disk_res = 282763095894SKent Overstreet bch2_disk_reservation_init(c, 0); 282863095894SKent Overstreet struct bkey_i delete; 282963095894SKent Overstreet struct bkey_s_c k; 283063095894SKent Overstreet struct bpos next_pos; 28315f786787SKent Overstreet struct bpos move_pos = POS(inode->v.i_ino, offset >> 9); 28325f786787SKent Overstreet struct bpos atomic_end; 28332d594dfbSKent Overstreet unsigned trigger_flags = 0; 28346fed42bbSKent Overstreet u32 snapshot; 28356fed42bbSKent Overstreet 28366fed42bbSKent Overstreet bch2_trans_begin(&trans); 28376fed42bbSKent Overstreet 28386fed42bbSKent Overstreet ret = bch2_subvolume_get_snapshot(&trans, 28396fed42bbSKent Overstreet inode->ei_subvol, &snapshot); 28406fed42bbSKent Overstreet if (ret) 28416fed42bbSKent Overstreet continue; 28426fed42bbSKent Overstreet 28436fed42bbSKent Overstreet bch2_btree_iter_set_snapshot(&src, snapshot); 28446fed42bbSKent Overstreet bch2_btree_iter_set_snapshot(&dst, snapshot); 28456fed42bbSKent Overstreet bch2_btree_iter_set_snapshot(&del, snapshot); 284663095894SKent Overstreet 2847700c25b3SKent Overstreet bch2_trans_begin(&trans); 2848700c25b3SKent Overstreet 28495f786787SKent Overstreet k = insert 285067e0dd8fSKent Overstreet ? bch2_btree_iter_peek_prev(&src) 285167e0dd8fSKent Overstreet : bch2_btree_iter_peek(&src); 285263095894SKent Overstreet if ((ret = bkey_err(k))) 285350dc0f69SKent Overstreet continue; 285463095894SKent Overstreet 285563095894SKent Overstreet if (!k.k || k.k->p.inode != inode->v.i_ino) 285663095894SKent Overstreet break; 285763095894SKent Overstreet 28585f786787SKent Overstreet if (insert && 28595f786787SKent Overstreet bkey_cmp(k.k->p, POS(inode->v.i_ino, offset >> 9)) <= 0) 28605f786787SKent Overstreet break; 28615f786787SKent Overstreet reassemble: 286207a1006aSKent Overstreet bch2_bkey_buf_reassemble(©, c, k); 28635f786787SKent Overstreet 28645f786787SKent Overstreet if (insert && 2865283eda57SKent Overstreet bkey_cmp(bkey_start_pos(k.k), move_pos) < 0) 286635189e09SKent Overstreet bch2_cut_front(move_pos, copy.k); 28675f786787SKent Overstreet 286835189e09SKent Overstreet copy.k->k.p.offset += shift >> 9; 286967e0dd8fSKent Overstreet bch2_btree_iter_set_pos(&dst, bkey_start_pos(©.k->k)); 28701c6fdbd8SKent Overstreet 287167e0dd8fSKent Overstreet ret = bch2_extent_atomic_end(&trans, &dst, copy.k, &atomic_end); 28723c7f3b7aSKent Overstreet if (ret) 287350dc0f69SKent Overstreet continue; 2874e2d9912cSKent Overstreet 287535189e09SKent Overstreet if (bkey_cmp(atomic_end, copy.k->k.p)) { 28765f786787SKent Overstreet if (insert) { 28775f786787SKent Overstreet move_pos = atomic_end; 28785f786787SKent Overstreet move_pos.offset -= shift >> 9; 28795f786787SKent Overstreet goto reassemble; 28805f786787SKent Overstreet } else { 2881085ab693SKent Overstreet bch2_cut_back(atomic_end, copy.k); 28825f786787SKent Overstreet } 28835f786787SKent Overstreet } 28845f786787SKent Overstreet 288563095894SKent Overstreet bkey_init(&delete.k); 2886283eda57SKent Overstreet delete.k.p = copy.k->k.p; 2887283eda57SKent Overstreet delete.k.size = copy.k->k.size; 2888283eda57SKent Overstreet delete.k.p.offset -= shift >> 9; 288967e0dd8fSKent Overstreet bch2_btree_iter_set_pos(&del, bkey_start_pos(&delete.k)); 28901c6fdbd8SKent Overstreet 28915f786787SKent Overstreet next_pos = insert ? bkey_start_pos(&delete.k) : delete.k.p; 289263095894SKent Overstreet 28937c4ca54aSKent Overstreet if (copy.k->k.size != k.k->size) { 289463095894SKent Overstreet /* We might end up splitting compressed extents: */ 289563095894SKent Overstreet unsigned nr_ptrs = 28964de77495SKent Overstreet bch2_bkey_nr_ptrs_allocated(bkey_i_to_s_c(copy.k)); 289763095894SKent Overstreet 289863095894SKent Overstreet ret = bch2_disk_reservation_get(c, &disk_res, 289935189e09SKent Overstreet copy.k->k.size, nr_ptrs, 29001c6fdbd8SKent Overstreet BCH_DISK_RESERVATION_NOFAIL); 29011c6fdbd8SKent Overstreet BUG_ON(ret); 290263095894SKent Overstreet } 29031c6fdbd8SKent Overstreet 290467e0dd8fSKent Overstreet ret = bch2_btree_iter_traverse(&del) ?: 290567e0dd8fSKent Overstreet bch2_trans_update(&trans, &del, &delete, trigger_flags) ?: 290667e0dd8fSKent Overstreet bch2_trans_update(&trans, &dst, copy.k, trigger_flags) ?: 290768a2054dSKent Overstreet bch2_trans_commit(&trans, &disk_res, NULL, 29082d594dfbSKent Overstreet BTREE_INSERT_NOFAIL); 29091c6fdbd8SKent Overstreet bch2_disk_reservation_put(c, &disk_res); 291050dc0f69SKent Overstreet 291163095894SKent Overstreet if (!ret) 291267e0dd8fSKent Overstreet bch2_btree_iter_set_pos(&src, next_pos); 291350dc0f69SKent Overstreet } 291467e0dd8fSKent Overstreet bch2_trans_iter_exit(&trans, &del); 291567e0dd8fSKent Overstreet bch2_trans_iter_exit(&trans, &dst); 291667e0dd8fSKent Overstreet bch2_trans_iter_exit(&trans, &src); 291750dc0f69SKent Overstreet bch2_trans_exit(&trans); 291850dc0f69SKent Overstreet bch2_bkey_buf_exit(©, c); 291963095894SKent Overstreet 29208ef231bdSKent Overstreet if (ret) 292174163da7SKent Overstreet return ret; 29221c6fdbd8SKent Overstreet 292374163da7SKent Overstreet mutex_lock(&inode->ei_update_lock); 29245f786787SKent Overstreet if (!insert) { 29258ef231bdSKent Overstreet i_size_write(&inode->v, new_size); 29268ef231bdSKent Overstreet ret = bch2_write_inode_size(c, inode, new_size, 29278ef231bdSKent Overstreet ATTR_MTIME|ATTR_CTIME); 292874163da7SKent Overstreet } else { 292974163da7SKent Overstreet /* We need an inode update to update bi_journal_seq for fsync: */ 293074163da7SKent Overstreet ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL, 293174163da7SKent Overstreet ATTR_MTIME|ATTR_CTIME); 29325f786787SKent Overstreet } 293374163da7SKent Overstreet mutex_unlock(&inode->ei_update_lock); 29341c6fdbd8SKent Overstreet return ret; 29351c6fdbd8SKent Overstreet } 29361c6fdbd8SKent Overstreet 2937694015c2SKent Overstreet static int __bchfs_fallocate(struct bch_inode_info *inode, int mode, 2938694015c2SKent Overstreet u64 start_sector, u64 end_sector) 29391c6fdbd8SKent Overstreet { 29401c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 2941190fa7afSKent Overstreet struct btree_trans trans; 294267e0dd8fSKent Overstreet struct btree_iter iter; 2943694015c2SKent Overstreet struct bpos end_pos = POS(inode->v.i_ino, end_sector); 29449a3df993SKent Overstreet unsigned replicas = io_opts(c, &inode->ei_inode).data_replicas; 2945694015c2SKent Overstreet int ret = 0; 29461c6fdbd8SKent Overstreet 2947f7beb4caSKent Overstreet bch2_trans_init(&trans, c, BTREE_ITER_MAX, 512); 29481c6fdbd8SKent Overstreet 294967e0dd8fSKent Overstreet bch2_trans_iter_init(&trans, &iter, BTREE_ID_extents, 2950694015c2SKent Overstreet POS(inode->v.i_ino, start_sector), 2951190fa7afSKent Overstreet BTREE_ITER_SLOTS|BTREE_ITER_INTENT); 29521c6fdbd8SKent Overstreet 295367e0dd8fSKent Overstreet while (!ret && bkey_cmp(iter.pos, end_pos) < 0) { 29542e87eae1SKent Overstreet s64 i_sectors_delta = 0; 29551c6fdbd8SKent Overstreet struct disk_reservation disk_res = { 0 }; 2956190fa7afSKent Overstreet struct quota_res quota_res = { 0 }; 29571c6fdbd8SKent Overstreet struct bkey_i_reservation reservation; 29581c6fdbd8SKent Overstreet struct bkey_s_c k; 2959694015c2SKent Overstreet unsigned sectors; 29606fed42bbSKent Overstreet u32 snapshot; 29611c6fdbd8SKent Overstreet 2962163e885aSKent Overstreet bch2_trans_begin(&trans); 2963a8abd3a7SKent Overstreet 29646fed42bbSKent Overstreet ret = bch2_subvolume_get_snapshot(&trans, 29656fed42bbSKent Overstreet inode->ei_subvol, &snapshot); 29666fed42bbSKent Overstreet if (ret) 29676fed42bbSKent Overstreet goto bkey_err; 29686fed42bbSKent Overstreet 29696fed42bbSKent Overstreet bch2_btree_iter_set_snapshot(&iter, snapshot); 29706fed42bbSKent Overstreet 297167e0dd8fSKent Overstreet k = bch2_btree_iter_peek_slot(&iter); 29720f238367SKent Overstreet if ((ret = bkey_err(k))) 29730f238367SKent Overstreet goto bkey_err; 29741c6fdbd8SKent Overstreet 29751c6fdbd8SKent Overstreet /* already reserved */ 297626609b61SKent Overstreet if (k.k->type == KEY_TYPE_reservation && 29771c6fdbd8SKent Overstreet bkey_s_c_to_reservation(k).v->nr_replicas >= replicas) { 297867e0dd8fSKent Overstreet bch2_btree_iter_advance(&iter); 29791c6fdbd8SKent Overstreet continue; 29801c6fdbd8SKent Overstreet } 29811c6fdbd8SKent Overstreet 2982190fa7afSKent Overstreet if (bkey_extent_is_data(k.k) && 2983190fa7afSKent Overstreet !(mode & FALLOC_FL_ZERO_RANGE)) { 298467e0dd8fSKent Overstreet bch2_btree_iter_advance(&iter); 29851c6fdbd8SKent Overstreet continue; 29861c6fdbd8SKent Overstreet } 29871c6fdbd8SKent Overstreet 29881c6fdbd8SKent Overstreet bkey_reservation_init(&reservation.k_i); 298926609b61SKent Overstreet reservation.k.type = KEY_TYPE_reservation; 29901c6fdbd8SKent Overstreet reservation.k.p = k.k->p; 29911c6fdbd8SKent Overstreet reservation.k.size = k.k->size; 29921c6fdbd8SKent Overstreet 299367e0dd8fSKent Overstreet bch2_cut_front(iter.pos, &reservation.k_i); 2994085ab693SKent Overstreet bch2_cut_back(end_pos, &reservation.k_i); 29951c6fdbd8SKent Overstreet 29961c6fdbd8SKent Overstreet sectors = reservation.k.size; 29974de77495SKent Overstreet reservation.v.nr_replicas = bch2_bkey_nr_ptrs_allocated(k); 29981c6fdbd8SKent Overstreet 29991c6fdbd8SKent Overstreet if (!bkey_extent_is_allocation(k.k)) { 30001c6fdbd8SKent Overstreet ret = bch2_quota_reservation_add(c, inode, 3001190fa7afSKent Overstreet "a_res, 30021c6fdbd8SKent Overstreet sectors, true); 30031c6fdbd8SKent Overstreet if (unlikely(ret)) 30040f238367SKent Overstreet goto bkey_err; 30051c6fdbd8SKent Overstreet } 30061c6fdbd8SKent Overstreet 30071c6fdbd8SKent Overstreet if (reservation.v.nr_replicas < replicas || 30084de77495SKent Overstreet bch2_bkey_sectors_compressed(k)) { 30091c6fdbd8SKent Overstreet ret = bch2_disk_reservation_get(c, &disk_res, sectors, 30101c6fdbd8SKent Overstreet replicas, 0); 30111c6fdbd8SKent Overstreet if (unlikely(ret)) 30120f238367SKent Overstreet goto bkey_err; 30131c6fdbd8SKent Overstreet 30141c6fdbd8SKent Overstreet reservation.v.nr_replicas = disk_res.nr_replicas; 30151c6fdbd8SKent Overstreet } 30161c6fdbd8SKent Overstreet 30178c6d298aSKent Overstreet ret = bch2_extent_update(&trans, inode_inum(inode), &iter, 30188c6d298aSKent Overstreet &reservation.k_i, 301968a2054dSKent Overstreet &disk_res, NULL, 3020a6336910SKent Overstreet 0, &i_sectors_delta, true); 30218810386fSKent Overstreet if (ret) 30228810386fSKent Overstreet goto bkey_err; 30232e87eae1SKent Overstreet i_sectors_acct(c, inode, "a_res, i_sectors_delta); 30240f238367SKent Overstreet bkey_err: 3025190fa7afSKent Overstreet bch2_quota_reservation_put(c, inode, "a_res); 30261c6fdbd8SKent Overstreet bch2_disk_reservation_put(c, &disk_res); 3027549d173cSKent Overstreet if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) 30281c6fdbd8SKent Overstreet ret = 0; 302950dc0f69SKent Overstreet } 303074163da7SKent Overstreet 3031dcfc593fSKent Overstreet bch2_trans_unlock(&trans); /* lock ordering, before taking pagecache locks: */ 3032dcfc593fSKent Overstreet mark_pagecache_reserved(inode, start_sector, iter.pos.offset); 3033dcfc593fSKent Overstreet 3034*098ef98dSKent Overstreet if (bch2_err_matches(ret, ENOSPC) && (mode & FALLOC_FL_ZERO_RANGE)) { 303574163da7SKent Overstreet struct quota_res quota_res = { 0 }; 303674163da7SKent Overstreet s64 i_sectors_delta = 0; 303774163da7SKent Overstreet 303874163da7SKent Overstreet bch2_fpunch_at(&trans, &iter, inode_inum(inode), 303974163da7SKent Overstreet end_sector, &i_sectors_delta); 304074163da7SKent Overstreet i_sectors_acct(c, inode, "a_res, i_sectors_delta); 304174163da7SKent Overstreet bch2_quota_reservation_put(c, inode, "a_res); 304274163da7SKent Overstreet } 304374163da7SKent Overstreet 304467e0dd8fSKent Overstreet bch2_trans_iter_exit(&trans, &iter); 3045694015c2SKent Overstreet bch2_trans_exit(&trans); 3046694015c2SKent Overstreet return ret; 3047694015c2SKent Overstreet } 304850dc0f69SKent Overstreet 3049694015c2SKent Overstreet static long bchfs_fallocate(struct bch_inode_info *inode, int mode, 3050694015c2SKent Overstreet loff_t offset, loff_t len) 3051694015c2SKent Overstreet { 3052694015c2SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 305374163da7SKent Overstreet u64 end = offset + len; 305474163da7SKent Overstreet u64 block_start = round_down(offset, block_bytes(c)); 305574163da7SKent Overstreet u64 block_end = round_up(end, block_bytes(c)); 305674163da7SKent Overstreet bool truncated_last_page = false; 305774163da7SKent Overstreet int ret, ret2 = 0; 3058694015c2SKent Overstreet 3059694015c2SKent Overstreet if (!(mode & FALLOC_FL_KEEP_SIZE) && end > inode->v.i_size) { 3060694015c2SKent Overstreet ret = inode_newsize_ok(&inode->v, end); 3061694015c2SKent Overstreet if (ret) 306274163da7SKent Overstreet return ret; 3063694015c2SKent Overstreet } 3064694015c2SKent Overstreet 3065694015c2SKent Overstreet if (mode & FALLOC_FL_ZERO_RANGE) { 306674163da7SKent Overstreet ret = bch2_truncate_pages(inode, offset, end); 306774163da7SKent Overstreet if (unlikely(ret < 0)) 306874163da7SKent Overstreet return ret; 3069694015c2SKent Overstreet 307074163da7SKent Overstreet truncated_last_page = ret; 3071694015c2SKent Overstreet 3072694015c2SKent Overstreet truncate_pagecache_range(&inode->v, offset, end - 1); 307374163da7SKent Overstreet 307474163da7SKent Overstreet block_start = round_up(offset, block_bytes(c)); 307574163da7SKent Overstreet block_end = round_down(end, block_bytes(c)); 3076694015c2SKent Overstreet } 3077694015c2SKent Overstreet 3078694015c2SKent Overstreet ret = __bchfs_fallocate(inode, mode, block_start >> 9, block_end >> 9); 3079e0541a93SKent Overstreet 3080e0541a93SKent Overstreet /* 308174163da7SKent Overstreet * On -ENOSPC in ZERO_RANGE mode, we still want to do the inode update, 308274163da7SKent Overstreet * so that the VFS cache i_size is consistent with the btree i_size: 3083e0541a93SKent Overstreet */ 308474163da7SKent Overstreet if (ret && 3085*098ef98dSKent Overstreet !(bch2_err_matches(ret, ENOSPC) && (mode & FALLOC_FL_ZERO_RANGE))) 308674163da7SKent Overstreet return ret; 30871c6fdbd8SKent Overstreet 308874163da7SKent Overstreet if (mode & FALLOC_FL_KEEP_SIZE && end > inode->v.i_size) 3089e0541a93SKent Overstreet end = inode->v.i_size; 309074163da7SKent Overstreet 309174163da7SKent Overstreet if (end >= inode->v.i_size && 309274163da7SKent Overstreet (((mode & FALLOC_FL_ZERO_RANGE) && !truncated_last_page) || 309374163da7SKent Overstreet !(mode & FALLOC_FL_KEEP_SIZE))) { 309474163da7SKent Overstreet spin_lock(&inode->v.i_lock); 3095e0541a93SKent Overstreet i_size_write(&inode->v, end); 309674163da7SKent Overstreet spin_unlock(&inode->v.i_lock); 3097e0541a93SKent Overstreet 30981c6fdbd8SKent Overstreet mutex_lock(&inode->ei_update_lock); 309974163da7SKent Overstreet ret2 = bch2_write_inode_size(c, inode, end, 0); 31001c6fdbd8SKent Overstreet mutex_unlock(&inode->ei_update_lock); 31011c6fdbd8SKent Overstreet } 310274163da7SKent Overstreet 310374163da7SKent Overstreet return ret ?: ret2; 31041c6fdbd8SKent Overstreet } 31051c6fdbd8SKent Overstreet 31061c6fdbd8SKent Overstreet long bch2_fallocate_dispatch(struct file *file, int mode, 31071c6fdbd8SKent Overstreet loff_t offset, loff_t len) 31081c6fdbd8SKent Overstreet { 31091c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 31102a9101a9SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 31112a9101a9SKent Overstreet long ret; 31122a9101a9SKent Overstreet 3113a3d7afa5SKent Overstreet if (!percpu_ref_tryget_live(&c->writes)) 31142a9101a9SKent Overstreet return -EROFS; 31151c6fdbd8SKent Overstreet 311674163da7SKent Overstreet inode_lock(&inode->v); 311774163da7SKent Overstreet inode_dio_wait(&inode->v); 311874163da7SKent Overstreet bch2_pagecache_block_get(&inode->ei_pagecache_lock); 311974163da7SKent Overstreet 31201c6fdbd8SKent Overstreet if (!(mode & ~(FALLOC_FL_KEEP_SIZE|FALLOC_FL_ZERO_RANGE))) 31212a9101a9SKent Overstreet ret = bchfs_fallocate(inode, mode, offset, len); 31222a9101a9SKent Overstreet else if (mode == (FALLOC_FL_PUNCH_HOLE|FALLOC_FL_KEEP_SIZE)) 31232a9101a9SKent Overstreet ret = bchfs_fpunch(inode, offset, len); 31242a9101a9SKent Overstreet else if (mode == FALLOC_FL_INSERT_RANGE) 31252a9101a9SKent Overstreet ret = bchfs_fcollapse_finsert(inode, offset, len, true); 31262a9101a9SKent Overstreet else if (mode == FALLOC_FL_COLLAPSE_RANGE) 31272a9101a9SKent Overstreet ret = bchfs_fcollapse_finsert(inode, offset, len, false); 31282a9101a9SKent Overstreet else 31292a9101a9SKent Overstreet ret = -EOPNOTSUPP; 31301c6fdbd8SKent Overstreet 313174163da7SKent Overstreet 313274163da7SKent Overstreet bch2_pagecache_block_put(&inode->ei_pagecache_lock); 313374163da7SKent Overstreet inode_unlock(&inode->v); 31342a9101a9SKent Overstreet percpu_ref_put(&c->writes); 31351c6fdbd8SKent Overstreet 31365c1ef830SKent Overstreet return bch2_err_class(ret); 31371c6fdbd8SKent Overstreet } 31381c6fdbd8SKent Overstreet 313976426098SKent Overstreet loff_t bch2_remap_file_range(struct file *file_src, loff_t pos_src, 314076426098SKent Overstreet struct file *file_dst, loff_t pos_dst, 314176426098SKent Overstreet loff_t len, unsigned remap_flags) 314276426098SKent Overstreet { 314376426098SKent Overstreet struct bch_inode_info *src = file_bch_inode(file_src); 314476426098SKent Overstreet struct bch_inode_info *dst = file_bch_inode(file_dst); 314576426098SKent Overstreet struct bch_fs *c = src->v.i_sb->s_fs_info; 31462e87eae1SKent Overstreet s64 i_sectors_delta = 0; 3147677fc056SKent Overstreet u64 aligned_len; 314876426098SKent Overstreet loff_t ret = 0; 314976426098SKent Overstreet 315076426098SKent Overstreet if (remap_flags & ~(REMAP_FILE_DEDUP|REMAP_FILE_ADVISORY)) 315176426098SKent Overstreet return -EINVAL; 315276426098SKent Overstreet 315376426098SKent Overstreet if (remap_flags & REMAP_FILE_DEDUP) 315476426098SKent Overstreet return -EOPNOTSUPP; 315576426098SKent Overstreet 315676426098SKent Overstreet if ((pos_src & (block_bytes(c) - 1)) || 315776426098SKent Overstreet (pos_dst & (block_bytes(c) - 1))) 315876426098SKent Overstreet return -EINVAL; 315976426098SKent Overstreet 316076426098SKent Overstreet if (src == dst && 316176426098SKent Overstreet abs(pos_src - pos_dst) < len) 316276426098SKent Overstreet return -EINVAL; 316376426098SKent Overstreet 316476426098SKent Overstreet bch2_lock_inodes(INODE_LOCK|INODE_PAGECACHE_BLOCK, src, dst); 316576426098SKent Overstreet 31662e87eae1SKent Overstreet file_update_time(file_dst); 31672e87eae1SKent Overstreet 316876426098SKent Overstreet inode_dio_wait(&src->v); 316976426098SKent Overstreet inode_dio_wait(&dst->v); 317076426098SKent Overstreet 317176426098SKent Overstreet ret = generic_remap_file_range_prep(file_src, pos_src, 317276426098SKent Overstreet file_dst, pos_dst, 317376426098SKent Overstreet &len, remap_flags); 317476426098SKent Overstreet if (ret < 0 || len == 0) 31752e87eae1SKent Overstreet goto err; 317676426098SKent Overstreet 3177677fc056SKent Overstreet aligned_len = round_up((u64) len, block_bytes(c)); 317876426098SKent Overstreet 317976426098SKent Overstreet ret = write_invalidate_inode_pages_range(dst->v.i_mapping, 3180677fc056SKent Overstreet pos_dst, pos_dst + len - 1); 318176426098SKent Overstreet if (ret) 31822e87eae1SKent Overstreet goto err; 318376426098SKent Overstreet 3184dcfc593fSKent Overstreet mark_pagecache_unallocated(src, pos_src >> 9, 3185dcfc593fSKent Overstreet (pos_src + aligned_len) >> 9); 318676426098SKent Overstreet 31872e87eae1SKent Overstreet ret = bch2_remap_range(c, 31886fed42bbSKent Overstreet inode_inum(dst), pos_dst >> 9, 31896fed42bbSKent Overstreet inode_inum(src), pos_src >> 9, 319076426098SKent Overstreet aligned_len >> 9, 31912e87eae1SKent Overstreet pos_dst + len, &i_sectors_delta); 31922e87eae1SKent Overstreet if (ret < 0) 31932e87eae1SKent Overstreet goto err; 319476426098SKent Overstreet 31952e87eae1SKent Overstreet /* 31962e87eae1SKent Overstreet * due to alignment, we might have remapped slightly more than requsted 31972e87eae1SKent Overstreet */ 3198677fc056SKent Overstreet ret = min((u64) ret << 9, (u64) len); 31992e87eae1SKent Overstreet 32002e87eae1SKent Overstreet /* XXX get a quota reservation */ 32012e87eae1SKent Overstreet i_sectors_acct(c, dst, NULL, i_sectors_delta); 32022e87eae1SKent Overstreet 32032e87eae1SKent Overstreet spin_lock(&dst->v.i_lock); 3204677fc056SKent Overstreet if (pos_dst + ret > dst->v.i_size) 3205677fc056SKent Overstreet i_size_write(&dst->v, pos_dst + ret); 32062e87eae1SKent Overstreet spin_unlock(&dst->v.i_lock); 3207e7084c9cSKent Overstreet 320868a2054dSKent Overstreet if ((file_dst->f_flags & (__O_SYNC | O_DSYNC)) || 320968a2054dSKent Overstreet IS_SYNC(file_inode(file_dst))) 321068a2054dSKent Overstreet ret = bch2_flush_inode(c, inode_inum(dst)); 32112e87eae1SKent Overstreet err: 321276426098SKent Overstreet bch2_unlock_inodes(INODE_LOCK|INODE_PAGECACHE_BLOCK, src, dst); 321376426098SKent Overstreet 32145c1ef830SKent Overstreet return bch2_err_class(ret); 321576426098SKent Overstreet } 321676426098SKent Overstreet 32171c6fdbd8SKent Overstreet /* fseek: */ 32181c6fdbd8SKent Overstreet 3219543ef2ebSKent Overstreet static int folio_data_offset(struct folio *folio, unsigned offset) 32201c6fdbd8SKent Overstreet { 3221f57a6a5dSKent Overstreet struct bch_page_state *s = bch2_page_state(&folio->page); 3222f57a6a5dSKent Overstreet unsigned i; 3223f81b648dSKent Overstreet 3224543ef2ebSKent Overstreet if (s) 3225543ef2ebSKent Overstreet for (i = offset >> 9; i < PAGE_SECTORS; i++) 3226f57a6a5dSKent Overstreet if (s->s[i].state >= SECTOR_DIRTY) 3227543ef2ebSKent Overstreet return i << 9; 3228f57a6a5dSKent Overstreet 3229543ef2ebSKent Overstreet return -1; 32301c6fdbd8SKent Overstreet } 32311c6fdbd8SKent Overstreet 3232543ef2ebSKent Overstreet static loff_t bch2_seek_pagecache_data(struct inode *vinode, 32331c6fdbd8SKent Overstreet loff_t start_offset, 32341c6fdbd8SKent Overstreet loff_t end_offset) 32351c6fdbd8SKent Overstreet { 32361c6fdbd8SKent Overstreet struct folio_batch fbatch; 32371c6fdbd8SKent Overstreet pgoff_t start_index = start_offset >> PAGE_SHIFT; 32381c6fdbd8SKent Overstreet pgoff_t end_index = end_offset >> PAGE_SHIFT; 32391c6fdbd8SKent Overstreet pgoff_t index = start_index; 32401c6fdbd8SKent Overstreet unsigned i; 3241543ef2ebSKent Overstreet loff_t ret; 3242543ef2ebSKent Overstreet int offset; 32431c6fdbd8SKent Overstreet 32441c6fdbd8SKent Overstreet folio_batch_init(&fbatch); 32451c6fdbd8SKent Overstreet 32461c6fdbd8SKent Overstreet while (filemap_get_folios(vinode->i_mapping, 32471c6fdbd8SKent Overstreet &index, end_index, &fbatch)) { 32481c6fdbd8SKent Overstreet for (i = 0; i < folio_batch_count(&fbatch); i++) { 32491c6fdbd8SKent Overstreet struct folio *folio = fbatch.folios[i]; 32501c6fdbd8SKent Overstreet 32511c6fdbd8SKent Overstreet folio_lock(folio); 3252543ef2ebSKent Overstreet offset = folio_data_offset(folio, 3253543ef2ebSKent Overstreet folio->index == start_index 3254543ef2ebSKent Overstreet ? start_offset & (PAGE_SIZE - 1) 3255543ef2ebSKent Overstreet : 0); 3256543ef2ebSKent Overstreet if (offset >= 0) { 3257543ef2ebSKent Overstreet ret = clamp(((loff_t) folio->index << PAGE_SHIFT) + 3258543ef2ebSKent Overstreet offset, 3259543ef2ebSKent Overstreet start_offset, end_offset); 32601c6fdbd8SKent Overstreet folio_unlock(folio); 32611c6fdbd8SKent Overstreet folio_batch_release(&fbatch); 3262543ef2ebSKent Overstreet return ret; 32631c6fdbd8SKent Overstreet } 32641c6fdbd8SKent Overstreet folio_unlock(folio); 32651c6fdbd8SKent Overstreet } 32661c6fdbd8SKent Overstreet folio_batch_release(&fbatch); 32671c6fdbd8SKent Overstreet cond_resched(); 32681c6fdbd8SKent Overstreet } 32691c6fdbd8SKent Overstreet 32701c6fdbd8SKent Overstreet return end_offset; 32711c6fdbd8SKent Overstreet } 32721c6fdbd8SKent Overstreet 32731c6fdbd8SKent Overstreet static loff_t bch2_seek_data(struct file *file, u64 offset) 32741c6fdbd8SKent Overstreet { 32751c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 32761c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 3277424eb881SKent Overstreet struct btree_trans trans; 327867e0dd8fSKent Overstreet struct btree_iter iter; 32791c6fdbd8SKent Overstreet struct bkey_s_c k; 32806fed42bbSKent Overstreet subvol_inum inum = inode_inum(inode); 32811c6fdbd8SKent Overstreet u64 isize, next_data = MAX_LFS_FILESIZE; 32826fed42bbSKent Overstreet u32 snapshot; 32831c6fdbd8SKent Overstreet int ret; 32841c6fdbd8SKent Overstreet 32851c6fdbd8SKent Overstreet isize = i_size_read(&inode->v); 32861c6fdbd8SKent Overstreet if (offset >= isize) 32871c6fdbd8SKent Overstreet return -ENXIO; 32881c6fdbd8SKent Overstreet 328920bceecbSKent Overstreet bch2_trans_init(&trans, c, 0, 0); 32906fed42bbSKent Overstreet retry: 32916fed42bbSKent Overstreet bch2_trans_begin(&trans); 32926fed42bbSKent Overstreet 32936fed42bbSKent Overstreet ret = bch2_subvolume_get_snapshot(&trans, inum.subvol, &snapshot); 32946fed42bbSKent Overstreet if (ret) 32956fed42bbSKent Overstreet goto err; 3296424eb881SKent Overstreet 3297e5fa91d7SKent Overstreet for_each_btree_key_norestart(&trans, iter, BTREE_ID_extents, 32986fed42bbSKent Overstreet SPOS(inode->v.i_ino, offset >> 9, snapshot), 0, k, ret) { 32991c6fdbd8SKent Overstreet if (k.k->p.inode != inode->v.i_ino) { 33001c6fdbd8SKent Overstreet break; 33011c6fdbd8SKent Overstreet } else if (bkey_extent_is_data(k.k)) { 33021c6fdbd8SKent Overstreet next_data = max(offset, bkey_start_offset(k.k) << 9); 33031c6fdbd8SKent Overstreet break; 33041c6fdbd8SKent Overstreet } else if (k.k->p.offset >> 9 > isize) 33051c6fdbd8SKent Overstreet break; 33061c6fdbd8SKent Overstreet } 330767e0dd8fSKent Overstreet bch2_trans_iter_exit(&trans, &iter); 33086fed42bbSKent Overstreet err: 3309549d173cSKent Overstreet if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) 33106fed42bbSKent Overstreet goto retry; 33111c6fdbd8SKent Overstreet 33129a796fdbSKent Overstreet bch2_trans_exit(&trans); 33131c6fdbd8SKent Overstreet if (ret) 33141c6fdbd8SKent Overstreet return ret; 33151c6fdbd8SKent Overstreet 33161c6fdbd8SKent Overstreet if (next_data > offset) 3317543ef2ebSKent Overstreet next_data = bch2_seek_pagecache_data(&inode->v, 33181c6fdbd8SKent Overstreet offset, next_data); 33191c6fdbd8SKent Overstreet 3320e10d3094SKent Overstreet if (next_data >= isize) 33211c6fdbd8SKent Overstreet return -ENXIO; 33221c6fdbd8SKent Overstreet 33231c6fdbd8SKent Overstreet return vfs_setpos(file, next_data, MAX_LFS_FILESIZE); 33241c6fdbd8SKent Overstreet } 33251c6fdbd8SKent Overstreet 3326543ef2ebSKent Overstreet static int __page_hole_offset(struct page *page, unsigned offset) 33271c6fdbd8SKent Overstreet { 3328543ef2ebSKent Overstreet struct bch_page_state *s = bch2_page_state(page); 3329543ef2ebSKent Overstreet unsigned i; 3330543ef2ebSKent Overstreet 3331543ef2ebSKent Overstreet if (!s) 3332543ef2ebSKent Overstreet return 0; 3333543ef2ebSKent Overstreet 3334543ef2ebSKent Overstreet for (i = offset >> 9; i < PAGE_SECTORS; i++) 3335543ef2ebSKent Overstreet if (s->s[i].state < SECTOR_DIRTY) 3336543ef2ebSKent Overstreet return i << 9; 3337543ef2ebSKent Overstreet 3338543ef2ebSKent Overstreet return -1; 3339543ef2ebSKent Overstreet } 3340543ef2ebSKent Overstreet 3341543ef2ebSKent Overstreet static loff_t page_hole_offset(struct address_space *mapping, loff_t offset) 3342543ef2ebSKent Overstreet { 3343543ef2ebSKent Overstreet pgoff_t index = offset >> PAGE_SHIFT; 33441c6fdbd8SKent Overstreet struct page *page; 3345543ef2ebSKent Overstreet int pg_offset; 3346543ef2ebSKent Overstreet loff_t ret = -1; 33471c6fdbd8SKent Overstreet 33481c6fdbd8SKent Overstreet page = find_lock_page(mapping, index); 33491c6fdbd8SKent Overstreet if (!page) 3350543ef2ebSKent Overstreet return offset; 33511c6fdbd8SKent Overstreet 3352543ef2ebSKent Overstreet pg_offset = __page_hole_offset(page, offset & (PAGE_SIZE - 1)); 3353543ef2ebSKent Overstreet if (pg_offset >= 0) 3354543ef2ebSKent Overstreet ret = ((loff_t) index << PAGE_SHIFT) + pg_offset; 3355543ef2ebSKent Overstreet 33561c6fdbd8SKent Overstreet unlock_page(page); 33571c6fdbd8SKent Overstreet 33581c6fdbd8SKent Overstreet return ret; 33591c6fdbd8SKent Overstreet } 33601c6fdbd8SKent Overstreet 3361543ef2ebSKent Overstreet static loff_t bch2_seek_pagecache_hole(struct inode *vinode, 33621c6fdbd8SKent Overstreet loff_t start_offset, 33631c6fdbd8SKent Overstreet loff_t end_offset) 33641c6fdbd8SKent Overstreet { 33651c6fdbd8SKent Overstreet struct address_space *mapping = vinode->i_mapping; 3366543ef2ebSKent Overstreet loff_t offset = start_offset, hole; 33671c6fdbd8SKent Overstreet 3368543ef2ebSKent Overstreet while (offset < end_offset) { 3369543ef2ebSKent Overstreet hole = page_hole_offset(mapping, offset); 3370543ef2ebSKent Overstreet if (hole >= 0 && hole <= end_offset) 3371543ef2ebSKent Overstreet return max(start_offset, hole); 3372543ef2ebSKent Overstreet 3373543ef2ebSKent Overstreet offset += PAGE_SIZE; 3374543ef2ebSKent Overstreet offset &= PAGE_MASK; 3375543ef2ebSKent Overstreet } 33761c6fdbd8SKent Overstreet 33771c6fdbd8SKent Overstreet return end_offset; 33781c6fdbd8SKent Overstreet } 33791c6fdbd8SKent Overstreet 33801c6fdbd8SKent Overstreet static loff_t bch2_seek_hole(struct file *file, u64 offset) 33811c6fdbd8SKent Overstreet { 33821c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 33831c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 3384424eb881SKent Overstreet struct btree_trans trans; 338567e0dd8fSKent Overstreet struct btree_iter iter; 33861c6fdbd8SKent Overstreet struct bkey_s_c k; 33876fed42bbSKent Overstreet subvol_inum inum = inode_inum(inode); 33881c6fdbd8SKent Overstreet u64 isize, next_hole = MAX_LFS_FILESIZE; 33896fed42bbSKent Overstreet u32 snapshot; 33901c6fdbd8SKent Overstreet int ret; 33911c6fdbd8SKent Overstreet 33921c6fdbd8SKent Overstreet isize = i_size_read(&inode->v); 33931c6fdbd8SKent Overstreet if (offset >= isize) 33941c6fdbd8SKent Overstreet return -ENXIO; 33951c6fdbd8SKent Overstreet 339620bceecbSKent Overstreet bch2_trans_init(&trans, c, 0, 0); 33976fed42bbSKent Overstreet retry: 33986fed42bbSKent Overstreet bch2_trans_begin(&trans); 33996fed42bbSKent Overstreet 34006fed42bbSKent Overstreet ret = bch2_subvolume_get_snapshot(&trans, inum.subvol, &snapshot); 34016fed42bbSKent Overstreet if (ret) 34026fed42bbSKent Overstreet goto err; 3403424eb881SKent Overstreet 3404e5fa91d7SKent Overstreet for_each_btree_key_norestart(&trans, iter, BTREE_ID_extents, 34056fed42bbSKent Overstreet SPOS(inode->v.i_ino, offset >> 9, snapshot), 340694f651e2SKent Overstreet BTREE_ITER_SLOTS, k, ret) { 34071c6fdbd8SKent Overstreet if (k.k->p.inode != inode->v.i_ino) { 3408543ef2ebSKent Overstreet next_hole = bch2_seek_pagecache_hole(&inode->v, 34091c6fdbd8SKent Overstreet offset, MAX_LFS_FILESIZE); 34101c6fdbd8SKent Overstreet break; 34111c6fdbd8SKent Overstreet } else if (!bkey_extent_is_data(k.k)) { 3412543ef2ebSKent Overstreet next_hole = bch2_seek_pagecache_hole(&inode->v, 34131c6fdbd8SKent Overstreet max(offset, bkey_start_offset(k.k) << 9), 34141c6fdbd8SKent Overstreet k.k->p.offset << 9); 34151c6fdbd8SKent Overstreet 34161c6fdbd8SKent Overstreet if (next_hole < k.k->p.offset << 9) 34171c6fdbd8SKent Overstreet break; 34181c6fdbd8SKent Overstreet } else { 34191c6fdbd8SKent Overstreet offset = max(offset, bkey_start_offset(k.k) << 9); 34201c6fdbd8SKent Overstreet } 34211c6fdbd8SKent Overstreet } 342267e0dd8fSKent Overstreet bch2_trans_iter_exit(&trans, &iter); 34236fed42bbSKent Overstreet err: 3424549d173cSKent Overstreet if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) 34256fed42bbSKent Overstreet goto retry; 34261c6fdbd8SKent Overstreet 34279a796fdbSKent Overstreet bch2_trans_exit(&trans); 34281c6fdbd8SKent Overstreet if (ret) 34291c6fdbd8SKent Overstreet return ret; 34301c6fdbd8SKent Overstreet 34311c6fdbd8SKent Overstreet if (next_hole > isize) 34321c6fdbd8SKent Overstreet next_hole = isize; 34331c6fdbd8SKent Overstreet 34341c6fdbd8SKent Overstreet return vfs_setpos(file, next_hole, MAX_LFS_FILESIZE); 34351c6fdbd8SKent Overstreet } 34361c6fdbd8SKent Overstreet 34371c6fdbd8SKent Overstreet loff_t bch2_llseek(struct file *file, loff_t offset, int whence) 34381c6fdbd8SKent Overstreet { 34395c1ef830SKent Overstreet loff_t ret; 34405c1ef830SKent Overstreet 34411c6fdbd8SKent Overstreet switch (whence) { 34421c6fdbd8SKent Overstreet case SEEK_SET: 34431c6fdbd8SKent Overstreet case SEEK_CUR: 34441c6fdbd8SKent Overstreet case SEEK_END: 34455c1ef830SKent Overstreet ret = generic_file_llseek(file, offset, whence); 34465c1ef830SKent Overstreet break; 34471c6fdbd8SKent Overstreet case SEEK_DATA: 34485c1ef830SKent Overstreet ret = bch2_seek_data(file, offset); 34495c1ef830SKent Overstreet break; 34501c6fdbd8SKent Overstreet case SEEK_HOLE: 34515c1ef830SKent Overstreet ret = bch2_seek_hole(file, offset); 34525c1ef830SKent Overstreet break; 34535c1ef830SKent Overstreet default: 34545c1ef830SKent Overstreet ret = -EINVAL; 34555c1ef830SKent Overstreet break; 34561c6fdbd8SKent Overstreet } 34571c6fdbd8SKent Overstreet 34585c1ef830SKent Overstreet return bch2_err_class(ret); 34591c6fdbd8SKent Overstreet } 34601c6fdbd8SKent Overstreet 34611c6fdbd8SKent Overstreet void bch2_fs_fsio_exit(struct bch_fs *c) 34621c6fdbd8SKent Overstreet { 34631c6fdbd8SKent Overstreet bioset_exit(&c->dio_write_bioset); 34641c6fdbd8SKent Overstreet bioset_exit(&c->dio_read_bioset); 34651c6fdbd8SKent Overstreet bioset_exit(&c->writepage_bioset); 34661c6fdbd8SKent Overstreet } 34671c6fdbd8SKent Overstreet 34681c6fdbd8SKent Overstreet int bch2_fs_fsio_init(struct bch_fs *c) 34691c6fdbd8SKent Overstreet { 34701c6fdbd8SKent Overstreet int ret = 0; 34711c6fdbd8SKent Overstreet 34721c6fdbd8SKent Overstreet pr_verbose_init(c->opts, ""); 34731c6fdbd8SKent Overstreet 34741c6fdbd8SKent Overstreet if (bioset_init(&c->writepage_bioset, 34759a3df993SKent Overstreet 4, offsetof(struct bch_writepage_io, op.wbio.bio), 34761c6fdbd8SKent Overstreet BIOSET_NEED_BVECS) || 34771c6fdbd8SKent Overstreet bioset_init(&c->dio_read_bioset, 34781c6fdbd8SKent Overstreet 4, offsetof(struct dio_read, rbio.bio), 34791c6fdbd8SKent Overstreet BIOSET_NEED_BVECS) || 34801c6fdbd8SKent Overstreet bioset_init(&c->dio_write_bioset, 34819a3df993SKent Overstreet 4, offsetof(struct dio_write, op.wbio.bio), 34821c6fdbd8SKent Overstreet BIOSET_NEED_BVECS)) 34831c6fdbd8SKent Overstreet ret = -ENOMEM; 34841c6fdbd8SKent Overstreet 34851c6fdbd8SKent Overstreet pr_verbose_init(c->opts, "ret %i", ret); 34861c6fdbd8SKent Overstreet return ret; 34871c6fdbd8SKent Overstreet } 34881c6fdbd8SKent Overstreet 34891c6fdbd8SKent Overstreet #endif /* NO_BCACHEFS_FS */ 3490