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" 61809b8cbSKent 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" 15dbbfca9fSKent Overstreet #include "fs-io-buffered.h" 16dbbfca9fSKent Overstreet #include "fs-io-pagecache.h" 171c6fdbd8SKent Overstreet #include "fsck.h" 181c6fdbd8SKent Overstreet #include "inode.h" 191c6fdbd8SKent Overstreet #include "journal.h" 201809b8cbSKent Overstreet #include "io_misc.h" 211c6fdbd8SKent Overstreet #include "keylist.h" 221c6fdbd8SKent Overstreet #include "quota.h" 2376426098SKent Overstreet #include "reflink.h" 241c6fdbd8SKent Overstreet #include "trace.h" 251c6fdbd8SKent Overstreet 261c6fdbd8SKent Overstreet #include <linux/aio.h> 271c6fdbd8SKent Overstreet #include <linux/backing-dev.h> 281c6fdbd8SKent Overstreet #include <linux/falloc.h> 291c6fdbd8SKent Overstreet #include <linux/migrate.h> 301c6fdbd8SKent Overstreet #include <linux/mmu_context.h> 311c6fdbd8SKent Overstreet #include <linux/pagevec.h> 329ba2eb25SKent Overstreet #include <linux/rmap.h> 331c6fdbd8SKent Overstreet #include <linux/sched/signal.h> 341c6fdbd8SKent Overstreet #include <linux/task_io_accounting_ops.h> 351c6fdbd8SKent Overstreet #include <linux/uio.h> 361c6fdbd8SKent Overstreet 371c6fdbd8SKent Overstreet #include <trace/events/writeback.h> 381c6fdbd8SKent Overstreet 39a8b3a677SKent Overstreet struct nocow_flush { 40a8b3a677SKent Overstreet struct closure *cl; 41a8b3a677SKent Overstreet struct bch_dev *ca; 42a8b3a677SKent Overstreet struct bio bio; 43a8b3a677SKent Overstreet }; 44a8b3a677SKent Overstreet 45a8b3a677SKent Overstreet static void nocow_flush_endio(struct bio *_bio) 46a8b3a677SKent Overstreet { 47a8b3a677SKent Overstreet 48a8b3a677SKent Overstreet struct nocow_flush *bio = container_of(_bio, struct nocow_flush, bio); 49a8b3a677SKent Overstreet 50a8b3a677SKent Overstreet closure_put(bio->cl); 51a8b3a677SKent Overstreet percpu_ref_put(&bio->ca->io_ref); 52a8b3a677SKent Overstreet bio_put(&bio->bio); 53a8b3a677SKent Overstreet } 54a8b3a677SKent Overstreet 55dbbfca9fSKent Overstreet void bch2_inode_flush_nocow_writes_async(struct bch_fs *c, 56a8b3a677SKent Overstreet struct bch_inode_info *inode, 57a8b3a677SKent Overstreet struct closure *cl) 58a8b3a677SKent Overstreet { 59a8b3a677SKent Overstreet struct nocow_flush *bio; 60a8b3a677SKent Overstreet struct bch_dev *ca; 61a8b3a677SKent Overstreet struct bch_devs_mask devs; 62a8b3a677SKent Overstreet unsigned dev; 63a8b3a677SKent Overstreet 64a8b3a677SKent Overstreet dev = find_first_bit(inode->ei_devs_need_flush.d, BCH_SB_MEMBERS_MAX); 65a8b3a677SKent Overstreet if (dev == BCH_SB_MEMBERS_MAX) 66a8b3a677SKent Overstreet return; 67a8b3a677SKent Overstreet 68a8b3a677SKent Overstreet devs = inode->ei_devs_need_flush; 69a8b3a677SKent Overstreet memset(&inode->ei_devs_need_flush, 0, sizeof(inode->ei_devs_need_flush)); 70a8b3a677SKent Overstreet 71a8b3a677SKent Overstreet for_each_set_bit(dev, devs.d, BCH_SB_MEMBERS_MAX) { 72a8b3a677SKent Overstreet rcu_read_lock(); 73a8b3a677SKent Overstreet ca = rcu_dereference(c->devs[dev]); 74a8b3a677SKent Overstreet if (ca && !percpu_ref_tryget(&ca->io_ref)) 75a8b3a677SKent Overstreet ca = NULL; 76a8b3a677SKent Overstreet rcu_read_unlock(); 77a8b3a677SKent Overstreet 78a8b3a677SKent Overstreet if (!ca) 79a8b3a677SKent Overstreet continue; 80a8b3a677SKent Overstreet 81a8b3a677SKent Overstreet bio = container_of(bio_alloc_bioset(ca->disk_sb.bdev, 0, 82a8b3a677SKent Overstreet REQ_OP_FLUSH, 83a8b3a677SKent Overstreet GFP_KERNEL, 84a8b3a677SKent Overstreet &c->nocow_flush_bioset), 85a8b3a677SKent Overstreet struct nocow_flush, bio); 86a8b3a677SKent Overstreet bio->cl = cl; 87a8b3a677SKent Overstreet bio->ca = ca; 88a8b3a677SKent Overstreet bio->bio.bi_end_io = nocow_flush_endio; 89a8b3a677SKent Overstreet closure_bio_submit(&bio->bio, cl); 90a8b3a677SKent Overstreet } 91a8b3a677SKent Overstreet } 92a8b3a677SKent Overstreet 93a8b3a677SKent Overstreet static int bch2_inode_flush_nocow_writes(struct bch_fs *c, 94a8b3a677SKent Overstreet struct bch_inode_info *inode) 95a8b3a677SKent Overstreet { 96a8b3a677SKent Overstreet struct closure cl; 97a8b3a677SKent Overstreet 98a8b3a677SKent Overstreet closure_init_stack(&cl); 99a8b3a677SKent Overstreet bch2_inode_flush_nocow_writes_async(c, inode, &cl); 100a8b3a677SKent Overstreet closure_sync(&cl); 101a8b3a677SKent Overstreet 102a8b3a677SKent Overstreet return 0; 103a8b3a677SKent Overstreet } 104a8b3a677SKent Overstreet 1051c6fdbd8SKent Overstreet /* i_size updates: */ 1061c6fdbd8SKent Overstreet 1072ea90048SKent Overstreet struct inode_new_size { 1082ea90048SKent Overstreet loff_t new_size; 1092ea90048SKent Overstreet u64 now; 1102ea90048SKent Overstreet unsigned fields; 1112ea90048SKent Overstreet }; 1122ea90048SKent Overstreet 113791236b8SJoshua Ashton static int inode_set_size(struct btree_trans *trans, 114791236b8SJoshua Ashton struct bch_inode_info *inode, 1151c6fdbd8SKent Overstreet struct bch_inode_unpacked *bi, 1161c6fdbd8SKent Overstreet void *p) 1171c6fdbd8SKent Overstreet { 1182ea90048SKent Overstreet struct inode_new_size *s = p; 1191c6fdbd8SKent Overstreet 1202ea90048SKent Overstreet bi->bi_size = s->new_size; 1212ea90048SKent Overstreet if (s->fields & ATTR_ATIME) 1222ea90048SKent Overstreet bi->bi_atime = s->now; 1232ea90048SKent Overstreet if (s->fields & ATTR_MTIME) 1242ea90048SKent Overstreet bi->bi_mtime = s->now; 1252ea90048SKent Overstreet if (s->fields & ATTR_CTIME) 1262ea90048SKent Overstreet bi->bi_ctime = s->now; 1271c6fdbd8SKent Overstreet 1281c6fdbd8SKent Overstreet return 0; 1291c6fdbd8SKent Overstreet } 1301c6fdbd8SKent Overstreet 13176426098SKent Overstreet int __must_check bch2_write_inode_size(struct bch_fs *c, 1321c6fdbd8SKent Overstreet struct bch_inode_info *inode, 1332ea90048SKent Overstreet loff_t new_size, unsigned fields) 1341c6fdbd8SKent Overstreet { 1352ea90048SKent Overstreet struct inode_new_size s = { 1362ea90048SKent Overstreet .new_size = new_size, 1372ea90048SKent Overstreet .now = bch2_current_time(c), 1382ea90048SKent Overstreet .fields = fields, 1392ea90048SKent Overstreet }; 1402ea90048SKent Overstreet 1412ea90048SKent Overstreet return bch2_write_inode(c, inode, inode_set_size, &s, fields); 1421c6fdbd8SKent Overstreet } 1431c6fdbd8SKent Overstreet 144dbbfca9fSKent Overstreet void __bch2_i_sectors_acct(struct bch_fs *c, struct bch_inode_info *inode, 145190fa7afSKent Overstreet struct quota_res *quota_res, s64 sectors) 1461c6fdbd8SKent Overstreet { 147b33bf1bcSKent Overstreet bch2_fs_inconsistent_on((s64) inode->v.i_blocks + sectors < 0, c, 148b33bf1bcSKent Overstreet "inode %lu i_blocks underflow: %llu + %lli < 0 (ondisk %lli)", 149b33bf1bcSKent Overstreet inode->v.i_ino, (u64) inode->v.i_blocks, sectors, 150b33bf1bcSKent Overstreet inode->ei_inode.bi_sectors); 151b44a66a6SKent Overstreet inode->v.i_blocks += sectors; 152b44a66a6SKent Overstreet 1531c6fdbd8SKent Overstreet #ifdef CONFIG_BCACHEFS_QUOTA 154cb1b479dSKent Overstreet if (quota_res && 155cb1b479dSKent Overstreet !test_bit(EI_INODE_SNAPSHOT, &inode->ei_flags) && 156cb1b479dSKent Overstreet sectors > 0) { 1571c6fdbd8SKent Overstreet BUG_ON(sectors > quota_res->sectors); 1581c6fdbd8SKent Overstreet BUG_ON(sectors > inode->ei_quota_reserved); 1591c6fdbd8SKent Overstreet 1601c6fdbd8SKent Overstreet quota_res->sectors -= sectors; 1611c6fdbd8SKent Overstreet inode->ei_quota_reserved -= sectors; 1621c6fdbd8SKent Overstreet } else { 16326609b61SKent Overstreet bch2_quota_acct(c, inode->ei_qid, Q_SPC, sectors, KEY_TYPE_QUOTA_WARN); 1641c6fdbd8SKent Overstreet } 1651c6fdbd8SKent Overstreet #endif 1666b1b186aSKent Overstreet } 1676b1b186aSKent Overstreet 1681c6fdbd8SKent Overstreet /* fsync: */ 1691c6fdbd8SKent Overstreet 17068a2054dSKent Overstreet /* 17168a2054dSKent Overstreet * inode->ei_inode.bi_journal_seq won't be up to date since it's set in an 17268a2054dSKent Overstreet * insert trigger: look up the btree inode instead 17368a2054dSKent Overstreet */ 174a8b3a677SKent Overstreet static int bch2_flush_inode(struct bch_fs *c, 175a8b3a677SKent Overstreet struct bch_inode_info *inode) 17668a2054dSKent Overstreet { 177a8b3a677SKent Overstreet struct bch_inode_unpacked u; 17868a2054dSKent Overstreet int ret; 17968a2054dSKent Overstreet 18068a2054dSKent Overstreet if (c->opts.journal_flush_disabled) 18168a2054dSKent Overstreet return 0; 18268a2054dSKent Overstreet 183a8b3a677SKent Overstreet ret = bch2_inode_find_by_inum(c, inode_inum(inode), &u); 18468a2054dSKent Overstreet if (ret) 18568a2054dSKent Overstreet return ret; 18668a2054dSKent Overstreet 187a8b3a677SKent Overstreet return bch2_journal_flush_seq(&c->journal, u.bi_journal_seq) ?: 188a8b3a677SKent Overstreet bch2_inode_flush_nocow_writes(c, inode); 18968a2054dSKent Overstreet } 19068a2054dSKent Overstreet 1911c6fdbd8SKent Overstreet int bch2_fsync(struct file *file, loff_t start, loff_t end, int datasync) 1921c6fdbd8SKent Overstreet { 1931c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 1941c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 19568a2054dSKent Overstreet int ret, ret2, ret3; 1961c6fdbd8SKent Overstreet 1971c6fdbd8SKent Overstreet ret = file_write_and_wait_range(file, start, end); 19868a2054dSKent Overstreet ret2 = sync_inode_metadata(&inode->v, 1); 199a8b3a677SKent Overstreet ret3 = bch2_flush_inode(c, inode); 2001c6fdbd8SKent Overstreet 2015c1ef830SKent Overstreet return bch2_err_class(ret ?: ret2 ?: ret3); 2021c6fdbd8SKent Overstreet } 2031c6fdbd8SKent Overstreet 2041c6fdbd8SKent Overstreet /* truncate: */ 2051c6fdbd8SKent Overstreet 2066fed42bbSKent Overstreet static inline int range_has_data(struct bch_fs *c, u32 subvol, 2071c6fdbd8SKent Overstreet struct bpos start, 2081c6fdbd8SKent Overstreet struct bpos end) 2091c6fdbd8SKent Overstreet { 210*6bd68ec2SKent Overstreet struct btree_trans *trans = bch2_trans_get(c); 21167e0dd8fSKent Overstreet struct btree_iter iter; 2121c6fdbd8SKent Overstreet struct bkey_s_c k; 2131c6fdbd8SKent Overstreet int ret = 0; 2146fed42bbSKent Overstreet retry: 215*6bd68ec2SKent Overstreet bch2_trans_begin(trans); 2166fed42bbSKent Overstreet 217*6bd68ec2SKent Overstreet ret = bch2_subvolume_get_snapshot(trans, subvol, &start.snapshot); 2186fed42bbSKent Overstreet if (ret) 2196fed42bbSKent Overstreet goto err; 220424eb881SKent Overstreet 221*6bd68ec2SKent Overstreet for_each_btree_key_upto_norestart(trans, iter, BTREE_ID_extents, start, end, 0, k, ret) 2224ad6aa46SBrian Foster if (bkey_extent_is_data(k.k) && !bkey_extent_is_unwritten(k)) { 2231c6fdbd8SKent Overstreet ret = 1; 2241c6fdbd8SKent Overstreet break; 2251c6fdbd8SKent Overstreet } 2266fed42bbSKent Overstreet start = iter.pos; 227*6bd68ec2SKent Overstreet bch2_trans_iter_exit(trans, &iter); 2286fed42bbSKent Overstreet err: 229549d173cSKent Overstreet if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) 2306fed42bbSKent Overstreet goto retry; 2311c6fdbd8SKent Overstreet 232*6bd68ec2SKent Overstreet bch2_trans_put(trans); 2339a796fdbSKent Overstreet return ret; 2341c6fdbd8SKent Overstreet } 2351c6fdbd8SKent Overstreet 236959f7368SKent Overstreet static int __bch2_truncate_folio(struct bch_inode_info *inode, 2371c6fdbd8SKent Overstreet pgoff_t index, loff_t start, loff_t end) 2381c6fdbd8SKent Overstreet { 2391c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 2401c6fdbd8SKent Overstreet struct address_space *mapping = inode->v.i_mapping; 2413342ac13SKent Overstreet struct bch_folio *s; 242c04cbc0dSColin Ian King unsigned start_offset; 243c04cbc0dSColin Ian King unsigned end_offset; 244a99b1cafSKent Overstreet unsigned i; 24530bff594SKent Overstreet struct folio *folio; 246b19d307dSKent Overstreet s64 i_sectors_delta = 0; 2471c6fdbd8SKent Overstreet int ret = 0; 2486b9857b2SBrian Foster u64 end_pos; 2491c6fdbd8SKent Overstreet 25030bff594SKent Overstreet folio = filemap_lock_folio(mapping, index); 251b6898917SKent Overstreet if (IS_ERR_OR_NULL(folio)) { 2521c6fdbd8SKent Overstreet /* 2531c6fdbd8SKent Overstreet * XXX: we're doing two index lookups when we end up reading the 25430bff594SKent Overstreet * folio 2551c6fdbd8SKent Overstreet */ 2566fed42bbSKent Overstreet ret = range_has_data(c, inode->ei_subvol, 257c72f687aSKent Overstreet POS(inode->v.i_ino, (index << PAGE_SECTORS_SHIFT)), 258c72f687aSKent Overstreet POS(inode->v.i_ino, (index << PAGE_SECTORS_SHIFT) + PAGE_SECTORS)); 2591c6fdbd8SKent Overstreet if (ret <= 0) 2601c6fdbd8SKent Overstreet return ret; 2611c6fdbd8SKent Overstreet 26230bff594SKent Overstreet folio = __filemap_get_folio(mapping, index, 26330bff594SKent Overstreet FGP_LOCK|FGP_CREAT, GFP_KERNEL); 2641e81f89bSKent Overstreet if (IS_ERR_OR_NULL(folio)) { 2651c6fdbd8SKent Overstreet ret = -ENOMEM; 2661c6fdbd8SKent Overstreet goto out; 2671c6fdbd8SKent Overstreet } 2681c6fdbd8SKent Overstreet } 2691c6fdbd8SKent Overstreet 270959f7368SKent Overstreet BUG_ON(start >= folio_end_pos(folio)); 271959f7368SKent Overstreet BUG_ON(end <= folio_pos(folio)); 272959f7368SKent Overstreet 273959f7368SKent Overstreet start_offset = max(start, folio_pos(folio)) - folio_pos(folio); 2746b9857b2SBrian Foster end_offset = min_t(u64, end, folio_end_pos(folio)) - folio_pos(folio); 275959f7368SKent Overstreet 276959f7368SKent Overstreet /* Folio boundary? Nothing to do */ 277959f7368SKent Overstreet if (start_offset == 0 && 278959f7368SKent Overstreet end_offset == folio_size(folio)) { 279959f7368SKent Overstreet ret = 0; 280959f7368SKent Overstreet goto unlock; 281959f7368SKent Overstreet } 282959f7368SKent Overstreet 28330bff594SKent Overstreet s = bch2_folio_create(folio, 0); 284a99b1cafSKent Overstreet if (!s) { 285a99b1cafSKent Overstreet ret = -ENOMEM; 286a99b1cafSKent Overstreet goto unlock; 287a99b1cafSKent Overstreet } 288a99b1cafSKent Overstreet 28930bff594SKent Overstreet if (!folio_test_uptodate(folio)) { 29030bff594SKent Overstreet ret = bch2_read_single_folio(folio, mapping); 2911c6fdbd8SKent Overstreet if (ret) 2921c6fdbd8SKent Overstreet goto unlock; 2931c6fdbd8SKent Overstreet } 2941c6fdbd8SKent Overstreet 29534fdcf06SKent Overstreet ret = bch2_folio_set(c, inode_inum(inode), &folio, 1); 29634fdcf06SKent Overstreet if (ret) 29734fdcf06SKent Overstreet goto unlock; 298c437e153SKent Overstreet 299a99b1cafSKent Overstreet for (i = round_up(start_offset, block_bytes(c)) >> 9; 300a99b1cafSKent Overstreet i < round_down(end_offset, block_bytes(c)) >> 9; 301a99b1cafSKent Overstreet i++) { 302a99b1cafSKent Overstreet s->s[i].nr_replicas = 0; 303a1774a05SKent Overstreet 304a1774a05SKent Overstreet i_sectors_delta -= s->s[i].state == SECTOR_dirty; 305dbbfca9fSKent Overstreet bch2_folio_sector_set(folio, s, i, SECTOR_unallocated); 306a99b1cafSKent Overstreet } 307a99b1cafSKent Overstreet 308dbbfca9fSKent Overstreet bch2_i_sectors_acct(c, inode, NULL, i_sectors_delta); 309b19d307dSKent Overstreet 31074163da7SKent Overstreet /* 31130bff594SKent Overstreet * Caller needs to know whether this folio will be written out by 31274163da7SKent Overstreet * writeback - doing an i_size update if necessary - or whether it will 3134ad6aa46SBrian Foster * be responsible for the i_size update. 3144ad6aa46SBrian Foster * 3154ad6aa46SBrian Foster * Note that we shouldn't ever see a folio beyond EOF, but check and 3164ad6aa46SBrian Foster * warn if so. This has been observed by failure to clean up folios 3174ad6aa46SBrian Foster * after a short write and there's still a chance reclaim will fix 3184ad6aa46SBrian Foster * things up. 31974163da7SKent Overstreet */ 3204ad6aa46SBrian Foster WARN_ON_ONCE(folio_pos(folio) >= inode->v.i_size); 3214ad6aa46SBrian Foster end_pos = folio_end_pos(folio); 3224ad6aa46SBrian Foster if (inode->v.i_size > folio_pos(folio)) 3236b9857b2SBrian Foster end_pos = min_t(u64, inode->v.i_size, end_pos); 324bf98ee10SBrian Foster ret = s->s[folio_pos_to_s(folio, end_pos - 1)].state >= SECTOR_dirty; 32574163da7SKent Overstreet 32630bff594SKent Overstreet folio_zero_segment(folio, start_offset, end_offset); 327a99b1cafSKent Overstreet 3281c6fdbd8SKent Overstreet /* 3291c6fdbd8SKent Overstreet * Bit of a hack - we don't want truncate to fail due to -ENOSPC. 3301c6fdbd8SKent Overstreet * 33130bff594SKent Overstreet * XXX: because we aren't currently tracking whether the folio has actual 3321c6fdbd8SKent Overstreet * data in it (vs. just 0s, or only partially written) this wrong. ick. 3331c6fdbd8SKent Overstreet */ 33430bff594SKent Overstreet BUG_ON(bch2_get_folio_disk_reservation(c, inode, folio, false)); 3351c6fdbd8SKent Overstreet 3369ba2eb25SKent Overstreet /* 3379ba2eb25SKent Overstreet * This removes any writeable userspace mappings; we need to force 3389ba2eb25SKent Overstreet * .page_mkwrite to be called again before any mmapped writes, to 3399ba2eb25SKent Overstreet * redirty the full page: 3409ba2eb25SKent Overstreet */ 34130bff594SKent Overstreet folio_mkclean(folio); 34230bff594SKent Overstreet filemap_dirty_folio(mapping, folio); 3431c6fdbd8SKent Overstreet unlock: 34430bff594SKent Overstreet folio_unlock(folio); 34530bff594SKent Overstreet folio_put(folio); 3461c6fdbd8SKent Overstreet out: 3471c6fdbd8SKent Overstreet return ret; 3481c6fdbd8SKent Overstreet } 3491c6fdbd8SKent Overstreet 350959f7368SKent Overstreet static int bch2_truncate_folio(struct bch_inode_info *inode, loff_t from) 3511c6fdbd8SKent Overstreet { 352959f7368SKent Overstreet return __bch2_truncate_folio(inode, from >> PAGE_SHIFT, 353959f7368SKent Overstreet from, ANYSINT_MAX(loff_t)); 3541c6fdbd8SKent Overstreet } 3551c6fdbd8SKent Overstreet 356959f7368SKent Overstreet static int bch2_truncate_folios(struct bch_inode_info *inode, 35774163da7SKent Overstreet loff_t start, loff_t end) 35874163da7SKent Overstreet { 359959f7368SKent Overstreet int ret = __bch2_truncate_folio(inode, start >> PAGE_SHIFT, 36074163da7SKent Overstreet start, end); 36174163da7SKent Overstreet 36274163da7SKent Overstreet if (ret >= 0 && 36374163da7SKent Overstreet start >> PAGE_SHIFT != end >> PAGE_SHIFT) 364959f7368SKent Overstreet ret = __bch2_truncate_folio(inode, 365959f7368SKent Overstreet (end - 1) >> PAGE_SHIFT, 36674163da7SKent Overstreet start, end); 36774163da7SKent Overstreet return ret; 36874163da7SKent Overstreet } 36974163da7SKent Overstreet 37068a507a2SKent Overstreet static int bch2_extend(struct mnt_idmap *idmap, 37168a507a2SKent Overstreet struct bch_inode_info *inode, 372e0541a93SKent Overstreet struct bch_inode_unpacked *inode_u, 373e0541a93SKent Overstreet struct iattr *iattr) 3741c6fdbd8SKent Overstreet { 3751c6fdbd8SKent Overstreet struct address_space *mapping = inode->v.i_mapping; 3761c6fdbd8SKent Overstreet int ret; 3771c6fdbd8SKent Overstreet 378e0541a93SKent Overstreet /* 379e0541a93SKent Overstreet * sync appends: 3802925fc49SKent Overstreet * 3812925fc49SKent Overstreet * this has to be done _before_ extending i_size: 382e0541a93SKent Overstreet */ 383e0541a93SKent Overstreet ret = filemap_write_and_wait_range(mapping, inode_u->bi_size, S64_MAX); 3841c6fdbd8SKent Overstreet if (ret) 3851c6fdbd8SKent Overstreet return ret; 3861c6fdbd8SKent Overstreet 3871c6fdbd8SKent Overstreet truncate_setsize(&inode->v, iattr->ia_size); 3881c6fdbd8SKent Overstreet 38968a507a2SKent Overstreet return bch2_setattr_nonsize(idmap, inode, iattr); 3901c6fdbd8SKent Overstreet } 3911c6fdbd8SKent Overstreet 3925902cc28SKent Overstreet int bchfs_truncate(struct mnt_idmap *idmap, 39368a507a2SKent Overstreet struct bch_inode_info *inode, struct iattr *iattr) 3941c6fdbd8SKent Overstreet { 3951c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 3961c6fdbd8SKent Overstreet struct address_space *mapping = inode->v.i_mapping; 397e0541a93SKent Overstreet struct bch_inode_unpacked inode_u; 3982e87eae1SKent Overstreet s64 i_sectors_delta = 0; 3991c6fdbd8SKent Overstreet int ret = 0; 4001c6fdbd8SKent Overstreet 40168a507a2SKent Overstreet /* 40278d66ab1SDan Robertson * If the truncate call with change the size of the file, the 40378d66ab1SDan Robertson * cmtimes should be updated. If the size will not change, we 40478d66ab1SDan Robertson * do not need to update the cmtimes. 40568a507a2SKent Overstreet */ 40678d66ab1SDan Robertson if (iattr->ia_size != inode->v.i_size) { 40768a507a2SKent Overstreet if (!(iattr->ia_valid & ATTR_MTIME)) 40868a507a2SKent Overstreet ktime_get_coarse_real_ts64(&iattr->ia_mtime); 40968a507a2SKent Overstreet if (!(iattr->ia_valid & ATTR_CTIME)) 41068a507a2SKent Overstreet ktime_get_coarse_real_ts64(&iattr->ia_ctime); 41168a507a2SKent Overstreet iattr->ia_valid |= ATTR_MTIME|ATTR_CTIME; 41278d66ab1SDan Robertson } 41368a507a2SKent Overstreet 4141c6fdbd8SKent Overstreet inode_dio_wait(&inode->v); 415a7ecd30cSKent Overstreet bch2_pagecache_block_get(inode); 4161c6fdbd8SKent Overstreet 4176fed42bbSKent Overstreet ret = bch2_inode_find_by_inum(c, inode_inum(inode), &inode_u); 418e0541a93SKent Overstreet if (ret) 419e0541a93SKent Overstreet goto err; 4201c6fdbd8SKent Overstreet 421c45d473dSKent Overstreet /* 422c45d473dSKent Overstreet * check this before next assertion; on filesystem error our normal 423c45d473dSKent Overstreet * invariants are a bit broken (truncate has to truncate the page cache 424c45d473dSKent Overstreet * before the inode). 425c45d473dSKent Overstreet */ 426c45d473dSKent Overstreet ret = bch2_journal_error(&c->journal); 427c45d473dSKent Overstreet if (ret) 428c45d473dSKent Overstreet goto err; 429c45d473dSKent Overstreet 4308eb71e9eSKent Overstreet WARN_ONCE(!test_bit(EI_INODE_ERROR, &inode->ei_flags) && 4318eb71e9eSKent Overstreet inode->v.i_size < inode_u.bi_size, 4328eb71e9eSKent Overstreet "truncate spotted in mem i_size < btree i_size: %llu < %llu\n", 4338eb71e9eSKent Overstreet (u64) inode->v.i_size, inode_u.bi_size); 434e0541a93SKent Overstreet 435e0541a93SKent Overstreet if (iattr->ia_size > inode->v.i_size) { 43668a507a2SKent Overstreet ret = bch2_extend(idmap, inode, &inode_u, iattr); 43754e2264eSKent Overstreet goto err; 4381c6fdbd8SKent Overstreet } 4391c6fdbd8SKent Overstreet 44068a507a2SKent Overstreet iattr->ia_valid &= ~ATTR_SIZE; 44168a507a2SKent Overstreet 442959f7368SKent Overstreet ret = bch2_truncate_folio(inode, iattr->ia_size); 44374163da7SKent Overstreet if (unlikely(ret < 0)) 44454e2264eSKent Overstreet goto err; 4451c6fdbd8SKent Overstreet 4465902cc28SKent Overstreet truncate_setsize(&inode->v, iattr->ia_size); 4475902cc28SKent Overstreet 4486cc3535dSKent Overstreet /* 4496cc3535dSKent Overstreet * When extending, we're going to write the new i_size to disk 4506cc3535dSKent Overstreet * immediately so we need to flush anything above the current on disk 4516cc3535dSKent Overstreet * i_size first: 4526cc3535dSKent Overstreet * 4536cc3535dSKent Overstreet * Also, when extending we need to flush the page that i_size currently 4546cc3535dSKent Overstreet * straddles - if it's mapped to userspace, we need to ensure that 4556cc3535dSKent Overstreet * userspace has to redirty it and call .mkwrite -> set_page_dirty 4566cc3535dSKent Overstreet * again to allocate the part of the page that was extended. 4576cc3535dSKent Overstreet */ 458e0541a93SKent Overstreet if (iattr->ia_size > inode_u.bi_size) 4591c6fdbd8SKent Overstreet ret = filemap_write_and_wait_range(mapping, 460e0541a93SKent Overstreet inode_u.bi_size, 4611c6fdbd8SKent Overstreet iattr->ia_size - 1); 4621c6fdbd8SKent Overstreet else if (iattr->ia_size & (PAGE_SIZE - 1)) 4631c6fdbd8SKent Overstreet ret = filemap_write_and_wait_range(mapping, 4641c6fdbd8SKent Overstreet round_down(iattr->ia_size, PAGE_SIZE), 4651c6fdbd8SKent Overstreet iattr->ia_size - 1); 4661c6fdbd8SKent Overstreet if (ret) 46754e2264eSKent Overstreet goto err; 4681c6fdbd8SKent Overstreet 4695902cc28SKent Overstreet ret = bch2_truncate(c, inode_inum(inode), iattr->ia_size, &i_sectors_delta); 470dbbfca9fSKent Overstreet bch2_i_sectors_acct(c, inode, NULL, i_sectors_delta); 4712e87eae1SKent Overstreet 4725902cc28SKent Overstreet if (unlikely(ret)) { 4735902cc28SKent Overstreet /* 4745902cc28SKent Overstreet * If we error here, VFS caches are now inconsistent with btree 4755902cc28SKent Overstreet */ 4765902cc28SKent Overstreet set_bit(EI_INODE_ERROR, &inode->ei_flags); 4775902cc28SKent Overstreet goto err; 4785902cc28SKent Overstreet } 4795902cc28SKent Overstreet 480b33bf1bcSKent Overstreet bch2_fs_inconsistent_on(!inode->v.i_size && inode->v.i_blocks && 481b33bf1bcSKent Overstreet !bch2_journal_error(&c->journal), c, 482b33bf1bcSKent Overstreet "inode %lu truncated to 0 but i_blocks %llu (ondisk %lli)", 483b33bf1bcSKent Overstreet inode->v.i_ino, (u64) inode->v.i_blocks, 484b33bf1bcSKent Overstreet inode->ei_inode.bi_sectors); 48568a507a2SKent Overstreet 48668a507a2SKent Overstreet ret = bch2_setattr_nonsize(idmap, inode, iattr); 48754e2264eSKent Overstreet err: 488a7ecd30cSKent Overstreet bch2_pagecache_block_put(inode); 4895c1ef830SKent Overstreet return bch2_err_class(ret); 4901c6fdbd8SKent Overstreet } 4911c6fdbd8SKent Overstreet 4921c6fdbd8SKent Overstreet /* fallocate: */ 4931c6fdbd8SKent Overstreet 494791236b8SJoshua Ashton static int inode_update_times_fn(struct btree_trans *trans, 495791236b8SJoshua Ashton struct bch_inode_info *inode, 496050197b1SKent Overstreet struct bch_inode_unpacked *bi, void *p) 497050197b1SKent Overstreet { 498050197b1SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 499050197b1SKent Overstreet 500050197b1SKent Overstreet bi->bi_mtime = bi->bi_ctime = bch2_current_time(c); 501050197b1SKent Overstreet return 0; 502050197b1SKent Overstreet } 503050197b1SKent Overstreet 5042e87eae1SKent Overstreet static long bchfs_fpunch(struct bch_inode_info *inode, loff_t offset, loff_t len) 5051c6fdbd8SKent Overstreet { 5061c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 50774163da7SKent Overstreet u64 end = offset + len; 50874163da7SKent Overstreet u64 block_start = round_up(offset, block_bytes(c)); 50974163da7SKent Overstreet u64 block_end = round_down(end, block_bytes(c)); 51074163da7SKent Overstreet bool truncated_last_page; 5111c6fdbd8SKent Overstreet int ret = 0; 5121c6fdbd8SKent Overstreet 513959f7368SKent Overstreet ret = bch2_truncate_folios(inode, offset, end); 51474163da7SKent Overstreet if (unlikely(ret < 0)) 5151c6fdbd8SKent Overstreet goto err; 5161c6fdbd8SKent Overstreet 51774163da7SKent Overstreet truncated_last_page = ret; 5181c6fdbd8SKent Overstreet 51974163da7SKent Overstreet truncate_pagecache_range(&inode->v, offset, end - 1); 5201c6fdbd8SKent Overstreet 52174163da7SKent Overstreet if (block_start < block_end) { 5222e87eae1SKent Overstreet s64 i_sectors_delta = 0; 5232e87eae1SKent Overstreet 5248c6d298aSKent Overstreet ret = bch2_fpunch(c, inode_inum(inode), 52574163da7SKent Overstreet block_start >> 9, block_end >> 9, 5262e87eae1SKent Overstreet &i_sectors_delta); 527dbbfca9fSKent Overstreet bch2_i_sectors_acct(c, inode, NULL, i_sectors_delta); 5282e87eae1SKent Overstreet } 529050197b1SKent Overstreet 530050197b1SKent Overstreet mutex_lock(&inode->ei_update_lock); 53174163da7SKent Overstreet if (end >= inode->v.i_size && !truncated_last_page) { 53274163da7SKent Overstreet ret = bch2_write_inode_size(c, inode, inode->v.i_size, 53374163da7SKent Overstreet ATTR_MTIME|ATTR_CTIME); 53474163da7SKent Overstreet } else { 535050197b1SKent Overstreet ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL, 53674163da7SKent Overstreet ATTR_MTIME|ATTR_CTIME); 53774163da7SKent Overstreet } 538050197b1SKent Overstreet mutex_unlock(&inode->ei_update_lock); 5391c6fdbd8SKent Overstreet err: 5401c6fdbd8SKent Overstreet return ret; 5411c6fdbd8SKent Overstreet } 5421c6fdbd8SKent Overstreet 5432e87eae1SKent Overstreet static long bchfs_fcollapse_finsert(struct bch_inode_info *inode, 5445f786787SKent Overstreet loff_t offset, loff_t len, 5455f786787SKent Overstreet bool insert) 5461c6fdbd8SKent Overstreet { 5471c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 5481c6fdbd8SKent Overstreet struct address_space *mapping = inode->v.i_mapping; 5495902cc28SKent Overstreet s64 i_sectors_delta = 0; 55050dc0f69SKent Overstreet int ret = 0; 5511c6fdbd8SKent Overstreet 5521c6fdbd8SKent Overstreet if ((offset | len) & (block_bytes(c) - 1)) 5531c6fdbd8SKent Overstreet return -EINVAL; 5541c6fdbd8SKent Overstreet 5555f786787SKent Overstreet if (insert) { 5565f786787SKent Overstreet if (offset >= inode->v.i_size) 55774163da7SKent Overstreet return -EINVAL; 5585f786787SKent Overstreet } else { 5591c6fdbd8SKent Overstreet if (offset + len >= inode->v.i_size) 56074163da7SKent Overstreet return -EINVAL; 5615f786787SKent Overstreet } 5621c6fdbd8SKent Overstreet 563dbbfca9fSKent Overstreet ret = bch2_write_invalidate_inode_pages_range(mapping, offset, LLONG_MAX); 5641c6fdbd8SKent Overstreet if (ret) 56574163da7SKent Overstreet return ret; 5661c6fdbd8SKent Overstreet 5675902cc28SKent Overstreet if (insert) 5685902cc28SKent Overstreet i_size_write(&inode->v, inode->v.i_size + len); 5692e87eae1SKent Overstreet 5705902cc28SKent Overstreet ret = bch2_fcollapse_finsert(c, inode_inum(inode), offset >> 9, len >> 9, 5715902cc28SKent Overstreet insert, &i_sectors_delta); 5725902cc28SKent Overstreet if (!ret && !insert) 5735902cc28SKent Overstreet i_size_write(&inode->v, inode->v.i_size - len); 574dbbfca9fSKent Overstreet bch2_i_sectors_acct(c, inode, NULL, i_sectors_delta); 5752e87eae1SKent Overstreet 5761c6fdbd8SKent Overstreet return ret; 5771c6fdbd8SKent Overstreet } 5781c6fdbd8SKent Overstreet 579694015c2SKent Overstreet static int __bchfs_fallocate(struct bch_inode_info *inode, int mode, 580694015c2SKent Overstreet u64 start_sector, u64 end_sector) 5811c6fdbd8SKent Overstreet { 5821c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 583*6bd68ec2SKent Overstreet struct btree_trans *trans = bch2_trans_get(c); 58467e0dd8fSKent Overstreet struct btree_iter iter; 585694015c2SKent Overstreet struct bpos end_pos = POS(inode->v.i_ino, end_sector); 58601ad6737SKent Overstreet struct bch_io_opts opts; 587694015c2SKent Overstreet int ret = 0; 5881c6fdbd8SKent Overstreet 58901ad6737SKent Overstreet bch2_inode_opts_get(&opts, c, &inode->ei_inode); 5901c6fdbd8SKent Overstreet 591*6bd68ec2SKent Overstreet bch2_trans_iter_init(trans, &iter, BTREE_ID_extents, 592694015c2SKent Overstreet POS(inode->v.i_ino, start_sector), 593190fa7afSKent Overstreet BTREE_ITER_SLOTS|BTREE_ITER_INTENT); 5941c6fdbd8SKent Overstreet 595e88a75ebSKent Overstreet while (!ret && bkey_lt(iter.pos, end_pos)) { 5962e87eae1SKent Overstreet s64 i_sectors_delta = 0; 597190fa7afSKent Overstreet struct quota_res quota_res = { 0 }; 5981c6fdbd8SKent Overstreet struct bkey_s_c k; 599694015c2SKent Overstreet unsigned sectors; 600a09818c7SKent Overstreet bool is_allocation; 601a09818c7SKent Overstreet u64 hole_start, hole_end; 6026fed42bbSKent Overstreet u32 snapshot; 6031c6fdbd8SKent Overstreet 604*6bd68ec2SKent Overstreet bch2_trans_begin(trans); 605a8abd3a7SKent Overstreet 606*6bd68ec2SKent Overstreet ret = bch2_subvolume_get_snapshot(trans, 6076fed42bbSKent Overstreet inode->ei_subvol, &snapshot); 6086fed42bbSKent Overstreet if (ret) 6096fed42bbSKent Overstreet goto bkey_err; 6106fed42bbSKent Overstreet 6116fed42bbSKent Overstreet bch2_btree_iter_set_snapshot(&iter, snapshot); 6126fed42bbSKent Overstreet 61367e0dd8fSKent Overstreet k = bch2_btree_iter_peek_slot(&iter); 6140f238367SKent Overstreet if ((ret = bkey_err(k))) 6150f238367SKent Overstreet goto bkey_err; 6161c6fdbd8SKent Overstreet 617a09818c7SKent Overstreet hole_start = iter.pos.offset; 618a09818c7SKent Overstreet hole_end = bpos_min(k.k->p, end_pos).offset; 619a09818c7SKent Overstreet is_allocation = bkey_extent_is_allocation(k.k); 620a09818c7SKent Overstreet 6211c6fdbd8SKent Overstreet /* already reserved */ 62279203111SKent Overstreet if (bkey_extent_is_reservation(k) && 62379203111SKent Overstreet bch2_bkey_nr_ptrs_fully_allocated(k) >= opts.data_replicas) { 62467e0dd8fSKent Overstreet bch2_btree_iter_advance(&iter); 6251c6fdbd8SKent Overstreet continue; 6261c6fdbd8SKent Overstreet } 6271c6fdbd8SKent Overstreet 628190fa7afSKent Overstreet if (bkey_extent_is_data(k.k) && 629190fa7afSKent Overstreet !(mode & FALLOC_FL_ZERO_RANGE)) { 63067e0dd8fSKent Overstreet bch2_btree_iter_advance(&iter); 6311c6fdbd8SKent Overstreet continue; 6321c6fdbd8SKent Overstreet } 6331c6fdbd8SKent Overstreet 634a09818c7SKent Overstreet if (!(mode & FALLOC_FL_ZERO_RANGE)) { 6354198bf03SKent Overstreet /* 6364198bf03SKent Overstreet * Lock ordering - can't be holding btree locks while 6374198bf03SKent Overstreet * blocking on a folio lock: 6384198bf03SKent Overstreet */ 6394198bf03SKent Overstreet if (bch2_clamp_data_hole(&inode->v, 6404198bf03SKent Overstreet &hole_start, 6414198bf03SKent Overstreet &hole_end, 6424198bf03SKent Overstreet opts.data_replicas, true)) 643*6bd68ec2SKent Overstreet ret = drop_locks_do(trans, 644a09818c7SKent Overstreet (bch2_clamp_data_hole(&inode->v, 645a09818c7SKent Overstreet &hole_start, 646a09818c7SKent Overstreet &hole_end, 6474198bf03SKent Overstreet opts.data_replicas, false), 0)); 648a09818c7SKent Overstreet bch2_btree_iter_set_pos(&iter, POS(iter.pos.inode, hole_start)); 649a8b3a677SKent Overstreet 650a09818c7SKent Overstreet if (ret) 651a09818c7SKent Overstreet goto bkey_err; 6521c6fdbd8SKent Overstreet 653a09818c7SKent Overstreet if (hole_start == hole_end) 654a09818c7SKent Overstreet continue; 655a09818c7SKent Overstreet } 656a09818c7SKent Overstreet 657a09818c7SKent Overstreet sectors = hole_end - hole_start; 658a09818c7SKent Overstreet 659a09818c7SKent Overstreet if (!is_allocation) { 6601c6fdbd8SKent Overstreet ret = bch2_quota_reservation_add(c, inode, 661a09818c7SKent Overstreet "a_res, sectors, true); 6621c6fdbd8SKent Overstreet if (unlikely(ret)) 6630f238367SKent Overstreet goto bkey_err; 6641c6fdbd8SKent Overstreet } 6651c6fdbd8SKent Overstreet 666*6bd68ec2SKent Overstreet ret = bch2_extent_fallocate(trans, inode_inum(inode), &iter, 66770de7a47SKent Overstreet sectors, opts, &i_sectors_delta, 66870de7a47SKent Overstreet writepoint_hashed((unsigned long) current)); 6698810386fSKent Overstreet if (ret) 6708810386fSKent Overstreet goto bkey_err; 67170de7a47SKent Overstreet 672dbbfca9fSKent Overstreet bch2_i_sectors_acct(c, inode, "a_res, i_sectors_delta); 673a09818c7SKent Overstreet 674*6bd68ec2SKent Overstreet drop_locks_do(trans, 675dbbfca9fSKent Overstreet (bch2_mark_pagecache_reserved(inode, hole_start, iter.pos.offset), 0)); 6760f238367SKent Overstreet bkey_err: 677190fa7afSKent Overstreet bch2_quota_reservation_put(c, inode, "a_res); 678549d173cSKent Overstreet if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) 6791c6fdbd8SKent Overstreet ret = 0; 68050dc0f69SKent Overstreet } 68174163da7SKent Overstreet 682098ef98dSKent Overstreet if (bch2_err_matches(ret, ENOSPC) && (mode & FALLOC_FL_ZERO_RANGE)) { 68374163da7SKent Overstreet struct quota_res quota_res = { 0 }; 68474163da7SKent Overstreet s64 i_sectors_delta = 0; 68574163da7SKent Overstreet 686*6bd68ec2SKent Overstreet bch2_fpunch_at(trans, &iter, inode_inum(inode), 68774163da7SKent Overstreet end_sector, &i_sectors_delta); 688dbbfca9fSKent Overstreet bch2_i_sectors_acct(c, inode, "a_res, i_sectors_delta); 68974163da7SKent Overstreet bch2_quota_reservation_put(c, inode, "a_res); 69074163da7SKent Overstreet } 69174163da7SKent Overstreet 692*6bd68ec2SKent Overstreet bch2_trans_iter_exit(trans, &iter); 693*6bd68ec2SKent Overstreet bch2_trans_put(trans); 694694015c2SKent Overstreet return ret; 695694015c2SKent Overstreet } 69650dc0f69SKent Overstreet 697694015c2SKent Overstreet static long bchfs_fallocate(struct bch_inode_info *inode, int mode, 698694015c2SKent Overstreet loff_t offset, loff_t len) 699694015c2SKent Overstreet { 700694015c2SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 70174163da7SKent Overstreet u64 end = offset + len; 70274163da7SKent Overstreet u64 block_start = round_down(offset, block_bytes(c)); 70374163da7SKent Overstreet u64 block_end = round_up(end, block_bytes(c)); 70474163da7SKent Overstreet bool truncated_last_page = false; 70574163da7SKent Overstreet int ret, ret2 = 0; 706694015c2SKent Overstreet 707694015c2SKent Overstreet if (!(mode & FALLOC_FL_KEEP_SIZE) && end > inode->v.i_size) { 708694015c2SKent Overstreet ret = inode_newsize_ok(&inode->v, end); 709694015c2SKent Overstreet if (ret) 71074163da7SKent Overstreet return ret; 711694015c2SKent Overstreet } 712694015c2SKent Overstreet 713694015c2SKent Overstreet if (mode & FALLOC_FL_ZERO_RANGE) { 714959f7368SKent Overstreet ret = bch2_truncate_folios(inode, offset, end); 71574163da7SKent Overstreet if (unlikely(ret < 0)) 71674163da7SKent Overstreet return ret; 717694015c2SKent Overstreet 71874163da7SKent Overstreet truncated_last_page = ret; 719694015c2SKent Overstreet 720694015c2SKent Overstreet truncate_pagecache_range(&inode->v, offset, end - 1); 72174163da7SKent Overstreet 72274163da7SKent Overstreet block_start = round_up(offset, block_bytes(c)); 72374163da7SKent Overstreet block_end = round_down(end, block_bytes(c)); 724694015c2SKent Overstreet } 725694015c2SKent Overstreet 726694015c2SKent Overstreet ret = __bchfs_fallocate(inode, mode, block_start >> 9, block_end >> 9); 727e0541a93SKent Overstreet 728e0541a93SKent Overstreet /* 72974163da7SKent Overstreet * On -ENOSPC in ZERO_RANGE mode, we still want to do the inode update, 73074163da7SKent Overstreet * so that the VFS cache i_size is consistent with the btree i_size: 731e0541a93SKent Overstreet */ 73274163da7SKent Overstreet if (ret && 733098ef98dSKent Overstreet !(bch2_err_matches(ret, ENOSPC) && (mode & FALLOC_FL_ZERO_RANGE))) 73474163da7SKent Overstreet return ret; 7351c6fdbd8SKent Overstreet 73674163da7SKent Overstreet if (mode & FALLOC_FL_KEEP_SIZE && end > inode->v.i_size) 737e0541a93SKent Overstreet end = inode->v.i_size; 73874163da7SKent Overstreet 73974163da7SKent Overstreet if (end >= inode->v.i_size && 74074163da7SKent Overstreet (((mode & FALLOC_FL_ZERO_RANGE) && !truncated_last_page) || 74174163da7SKent Overstreet !(mode & FALLOC_FL_KEEP_SIZE))) { 74274163da7SKent Overstreet spin_lock(&inode->v.i_lock); 743e0541a93SKent Overstreet i_size_write(&inode->v, end); 74474163da7SKent Overstreet spin_unlock(&inode->v.i_lock); 745e0541a93SKent Overstreet 7461c6fdbd8SKent Overstreet mutex_lock(&inode->ei_update_lock); 74774163da7SKent Overstreet ret2 = bch2_write_inode_size(c, inode, end, 0); 7481c6fdbd8SKent Overstreet mutex_unlock(&inode->ei_update_lock); 7491c6fdbd8SKent Overstreet } 75074163da7SKent Overstreet 75174163da7SKent Overstreet return ret ?: ret2; 7521c6fdbd8SKent Overstreet } 7531c6fdbd8SKent Overstreet 7541c6fdbd8SKent Overstreet long bch2_fallocate_dispatch(struct file *file, int mode, 7551c6fdbd8SKent Overstreet loff_t offset, loff_t len) 7561c6fdbd8SKent Overstreet { 7571c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 7582a9101a9SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 7592a9101a9SKent Overstreet long ret; 7602a9101a9SKent Overstreet 761d94189adSKent Overstreet if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_fallocate)) 7622a9101a9SKent Overstreet return -EROFS; 7631c6fdbd8SKent Overstreet 76474163da7SKent Overstreet inode_lock(&inode->v); 76574163da7SKent Overstreet inode_dio_wait(&inode->v); 766a7ecd30cSKent Overstreet bch2_pagecache_block_get(inode); 76774163da7SKent Overstreet 76807bfcc0bSKent Overstreet ret = file_modified(file); 76907bfcc0bSKent Overstreet if (ret) 77007bfcc0bSKent Overstreet goto err; 77107bfcc0bSKent Overstreet 7721c6fdbd8SKent Overstreet if (!(mode & ~(FALLOC_FL_KEEP_SIZE|FALLOC_FL_ZERO_RANGE))) 7732a9101a9SKent Overstreet ret = bchfs_fallocate(inode, mode, offset, len); 7742a9101a9SKent Overstreet else if (mode == (FALLOC_FL_PUNCH_HOLE|FALLOC_FL_KEEP_SIZE)) 7752a9101a9SKent Overstreet ret = bchfs_fpunch(inode, offset, len); 7762a9101a9SKent Overstreet else if (mode == FALLOC_FL_INSERT_RANGE) 7772a9101a9SKent Overstreet ret = bchfs_fcollapse_finsert(inode, offset, len, true); 7782a9101a9SKent Overstreet else if (mode == FALLOC_FL_COLLAPSE_RANGE) 7792a9101a9SKent Overstreet ret = bchfs_fcollapse_finsert(inode, offset, len, false); 7802a9101a9SKent Overstreet else 7812a9101a9SKent Overstreet ret = -EOPNOTSUPP; 78207bfcc0bSKent Overstreet err: 783a7ecd30cSKent Overstreet bch2_pagecache_block_put(inode); 78474163da7SKent Overstreet inode_unlock(&inode->v); 785d94189adSKent Overstreet bch2_write_ref_put(c, BCH_WRITE_REF_fallocate); 7861c6fdbd8SKent Overstreet 7875c1ef830SKent Overstreet return bch2_err_class(ret); 7881c6fdbd8SKent Overstreet } 7891c6fdbd8SKent Overstreet 790c72f687aSKent Overstreet /* 791c72f687aSKent Overstreet * Take a quota reservation for unallocated blocks in a given file range 792c72f687aSKent Overstreet * Does not check pagecache 793c72f687aSKent Overstreet */ 794e8540e56SKent Overstreet static int quota_reserve_range(struct bch_inode_info *inode, 795e8540e56SKent Overstreet struct quota_res *res, 796e8540e56SKent Overstreet u64 start, u64 end) 797e8540e56SKent Overstreet { 798e8540e56SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 799*6bd68ec2SKent Overstreet struct btree_trans *trans = bch2_trans_get(c); 800e8540e56SKent Overstreet struct btree_iter iter; 801e8540e56SKent Overstreet struct bkey_s_c k; 802e8540e56SKent Overstreet u32 snapshot; 803e8540e56SKent Overstreet u64 sectors = end - start; 804e8540e56SKent Overstreet u64 pos = start; 805e8540e56SKent Overstreet int ret; 806e8540e56SKent Overstreet retry: 807*6bd68ec2SKent Overstreet bch2_trans_begin(trans); 808e8540e56SKent Overstreet 809*6bd68ec2SKent Overstreet ret = bch2_subvolume_get_snapshot(trans, inode->ei_subvol, &snapshot); 810e8540e56SKent Overstreet if (ret) 811e8540e56SKent Overstreet goto err; 812e8540e56SKent Overstreet 813*6bd68ec2SKent Overstreet bch2_trans_iter_init(trans, &iter, BTREE_ID_extents, 814e8540e56SKent Overstreet SPOS(inode->v.i_ino, pos, snapshot), 0); 815e8540e56SKent Overstreet 816*6bd68ec2SKent Overstreet while (!(ret = btree_trans_too_many_iters(trans)) && 817e8540e56SKent Overstreet (k = bch2_btree_iter_peek_upto(&iter, POS(inode->v.i_ino, end - 1))).k && 818e8540e56SKent Overstreet !(ret = bkey_err(k))) { 819e8540e56SKent Overstreet if (bkey_extent_is_allocation(k.k)) { 820e8540e56SKent Overstreet u64 s = min(end, k.k->p.offset) - 821e8540e56SKent Overstreet max(start, bkey_start_offset(k.k)); 822e8540e56SKent Overstreet BUG_ON(s > sectors); 823e8540e56SKent Overstreet sectors -= s; 824e8540e56SKent Overstreet } 825e8540e56SKent Overstreet bch2_btree_iter_advance(&iter); 826e8540e56SKent Overstreet } 827e8540e56SKent Overstreet pos = iter.pos.offset; 828*6bd68ec2SKent Overstreet bch2_trans_iter_exit(trans, &iter); 829e8540e56SKent Overstreet err: 830e8540e56SKent Overstreet if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) 831e8540e56SKent Overstreet goto retry; 832e8540e56SKent Overstreet 833*6bd68ec2SKent Overstreet bch2_trans_put(trans); 834e8540e56SKent Overstreet 835*6bd68ec2SKent Overstreet return ret ?: bch2_quota_reservation_add(c, inode, res, sectors, true); 836e8540e56SKent Overstreet } 837e8540e56SKent Overstreet 83876426098SKent Overstreet loff_t bch2_remap_file_range(struct file *file_src, loff_t pos_src, 83976426098SKent Overstreet struct file *file_dst, loff_t pos_dst, 84076426098SKent Overstreet loff_t len, unsigned remap_flags) 84176426098SKent Overstreet { 84276426098SKent Overstreet struct bch_inode_info *src = file_bch_inode(file_src); 84376426098SKent Overstreet struct bch_inode_info *dst = file_bch_inode(file_dst); 84476426098SKent Overstreet struct bch_fs *c = src->v.i_sb->s_fs_info; 845e8540e56SKent Overstreet struct quota_res quota_res = { 0 }; 8462e87eae1SKent Overstreet s64 i_sectors_delta = 0; 847677fc056SKent Overstreet u64 aligned_len; 84876426098SKent Overstreet loff_t ret = 0; 84976426098SKent Overstreet 85076426098SKent Overstreet if (remap_flags & ~(REMAP_FILE_DEDUP|REMAP_FILE_ADVISORY)) 85176426098SKent Overstreet return -EINVAL; 85276426098SKent Overstreet 85376426098SKent Overstreet if (remap_flags & REMAP_FILE_DEDUP) 85476426098SKent Overstreet return -EOPNOTSUPP; 85576426098SKent Overstreet 85676426098SKent Overstreet if ((pos_src & (block_bytes(c) - 1)) || 85776426098SKent Overstreet (pos_dst & (block_bytes(c) - 1))) 85876426098SKent Overstreet return -EINVAL; 85976426098SKent Overstreet 86076426098SKent Overstreet if (src == dst && 86176426098SKent Overstreet abs(pos_src - pos_dst) < len) 86276426098SKent Overstreet return -EINVAL; 86376426098SKent Overstreet 86476426098SKent Overstreet bch2_lock_inodes(INODE_LOCK|INODE_PAGECACHE_BLOCK, src, dst); 86576426098SKent Overstreet 86676426098SKent Overstreet inode_dio_wait(&src->v); 86776426098SKent Overstreet inode_dio_wait(&dst->v); 86876426098SKent Overstreet 86976426098SKent Overstreet ret = generic_remap_file_range_prep(file_src, pos_src, 87076426098SKent Overstreet file_dst, pos_dst, 87176426098SKent Overstreet &len, remap_flags); 87276426098SKent Overstreet if (ret < 0 || len == 0) 8732e87eae1SKent Overstreet goto err; 87476426098SKent Overstreet 875677fc056SKent Overstreet aligned_len = round_up((u64) len, block_bytes(c)); 87676426098SKent Overstreet 877dbbfca9fSKent Overstreet ret = bch2_write_invalidate_inode_pages_range(dst->v.i_mapping, 878677fc056SKent Overstreet pos_dst, pos_dst + len - 1); 87976426098SKent Overstreet if (ret) 8802e87eae1SKent Overstreet goto err; 88176426098SKent Overstreet 882e8540e56SKent Overstreet ret = quota_reserve_range(dst, "a_res, pos_dst >> 9, 883e8540e56SKent Overstreet (pos_dst + aligned_len) >> 9); 884e8540e56SKent Overstreet if (ret) 885e8540e56SKent Overstreet goto err; 886e8540e56SKent Overstreet 887e8540e56SKent Overstreet file_update_time(file_dst); 888e8540e56SKent Overstreet 889dbbfca9fSKent Overstreet bch2_mark_pagecache_unallocated(src, pos_src >> 9, 890dcfc593fSKent Overstreet (pos_src + aligned_len) >> 9); 89176426098SKent Overstreet 8922e87eae1SKent Overstreet ret = bch2_remap_range(c, 8936fed42bbSKent Overstreet inode_inum(dst), pos_dst >> 9, 8946fed42bbSKent Overstreet inode_inum(src), pos_src >> 9, 89576426098SKent Overstreet aligned_len >> 9, 8962e87eae1SKent Overstreet pos_dst + len, &i_sectors_delta); 8972e87eae1SKent Overstreet if (ret < 0) 8982e87eae1SKent Overstreet goto err; 89976426098SKent Overstreet 9002e87eae1SKent Overstreet /* 9012e87eae1SKent Overstreet * due to alignment, we might have remapped slightly more than requsted 9022e87eae1SKent Overstreet */ 903677fc056SKent Overstreet ret = min((u64) ret << 9, (u64) len); 9042e87eae1SKent Overstreet 905dbbfca9fSKent Overstreet bch2_i_sectors_acct(c, dst, "a_res, i_sectors_delta); 9062e87eae1SKent Overstreet 9072e87eae1SKent Overstreet spin_lock(&dst->v.i_lock); 908677fc056SKent Overstreet if (pos_dst + ret > dst->v.i_size) 909677fc056SKent Overstreet i_size_write(&dst->v, pos_dst + ret); 9102e87eae1SKent Overstreet spin_unlock(&dst->v.i_lock); 911e7084c9cSKent Overstreet 91268a2054dSKent Overstreet if ((file_dst->f_flags & (__O_SYNC | O_DSYNC)) || 91368a2054dSKent Overstreet IS_SYNC(file_inode(file_dst))) 914a8b3a677SKent Overstreet ret = bch2_flush_inode(c, dst); 9152e87eae1SKent Overstreet err: 916e8540e56SKent Overstreet bch2_quota_reservation_put(c, dst, "a_res); 91776426098SKent Overstreet bch2_unlock_inodes(INODE_LOCK|INODE_PAGECACHE_BLOCK, src, dst); 91876426098SKent Overstreet 9195c1ef830SKent Overstreet return bch2_err_class(ret); 92076426098SKent Overstreet } 92176426098SKent Overstreet 9221c6fdbd8SKent Overstreet /* fseek: */ 9231c6fdbd8SKent Overstreet 9241c6fdbd8SKent Overstreet static loff_t bch2_seek_data(struct file *file, u64 offset) 9251c6fdbd8SKent Overstreet { 9261c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 9271c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 928*6bd68ec2SKent Overstreet struct btree_trans *trans; 92967e0dd8fSKent Overstreet struct btree_iter iter; 9301c6fdbd8SKent Overstreet struct bkey_s_c k; 9316fed42bbSKent Overstreet subvol_inum inum = inode_inum(inode); 9321c6fdbd8SKent Overstreet u64 isize, next_data = MAX_LFS_FILESIZE; 9336fed42bbSKent Overstreet u32 snapshot; 9341c6fdbd8SKent Overstreet int ret; 9351c6fdbd8SKent Overstreet 9361c6fdbd8SKent Overstreet isize = i_size_read(&inode->v); 9371c6fdbd8SKent Overstreet if (offset >= isize) 9381c6fdbd8SKent Overstreet return -ENXIO; 9391c6fdbd8SKent Overstreet 940*6bd68ec2SKent Overstreet trans = bch2_trans_get(c); 9416fed42bbSKent Overstreet retry: 942*6bd68ec2SKent Overstreet bch2_trans_begin(trans); 9436fed42bbSKent Overstreet 944*6bd68ec2SKent Overstreet ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot); 9456fed42bbSKent Overstreet if (ret) 9466fed42bbSKent Overstreet goto err; 947424eb881SKent Overstreet 948*6bd68ec2SKent Overstreet for_each_btree_key_upto_norestart(trans, iter, BTREE_ID_extents, 949c72f687aSKent Overstreet SPOS(inode->v.i_ino, offset >> 9, snapshot), 950c72f687aSKent Overstreet POS(inode->v.i_ino, U64_MAX), 951c72f687aSKent Overstreet 0, k, ret) { 952c72f687aSKent Overstreet if (bkey_extent_is_data(k.k)) { 9531c6fdbd8SKent Overstreet next_data = max(offset, bkey_start_offset(k.k) << 9); 9541c6fdbd8SKent Overstreet break; 9551c6fdbd8SKent Overstreet } else if (k.k->p.offset >> 9 > isize) 9561c6fdbd8SKent Overstreet break; 9571c6fdbd8SKent Overstreet } 958*6bd68ec2SKent Overstreet bch2_trans_iter_exit(trans, &iter); 9596fed42bbSKent Overstreet err: 960549d173cSKent Overstreet if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) 9616fed42bbSKent Overstreet goto retry; 9621c6fdbd8SKent Overstreet 963*6bd68ec2SKent Overstreet bch2_trans_put(trans); 9641c6fdbd8SKent Overstreet if (ret) 9651c6fdbd8SKent Overstreet return ret; 9661c6fdbd8SKent Overstreet 9671c6fdbd8SKent Overstreet if (next_data > offset) 968543ef2ebSKent Overstreet next_data = bch2_seek_pagecache_data(&inode->v, 9694198bf03SKent Overstreet offset, next_data, 0, false); 9701c6fdbd8SKent Overstreet 971e10d3094SKent Overstreet if (next_data >= isize) 9721c6fdbd8SKent Overstreet return -ENXIO; 9731c6fdbd8SKent Overstreet 9741c6fdbd8SKent Overstreet return vfs_setpos(file, next_data, MAX_LFS_FILESIZE); 9751c6fdbd8SKent Overstreet } 9761c6fdbd8SKent Overstreet 9771c6fdbd8SKent Overstreet static loff_t bch2_seek_hole(struct file *file, u64 offset) 9781c6fdbd8SKent Overstreet { 9791c6fdbd8SKent Overstreet struct bch_inode_info *inode = file_bch_inode(file); 9801c6fdbd8SKent Overstreet struct bch_fs *c = inode->v.i_sb->s_fs_info; 981*6bd68ec2SKent Overstreet struct btree_trans *trans; 98267e0dd8fSKent Overstreet struct btree_iter iter; 9831c6fdbd8SKent Overstreet struct bkey_s_c k; 9846fed42bbSKent Overstreet subvol_inum inum = inode_inum(inode); 9851c6fdbd8SKent Overstreet u64 isize, next_hole = MAX_LFS_FILESIZE; 9866fed42bbSKent Overstreet u32 snapshot; 9871c6fdbd8SKent Overstreet int ret; 9881c6fdbd8SKent Overstreet 9891c6fdbd8SKent Overstreet isize = i_size_read(&inode->v); 9901c6fdbd8SKent Overstreet if (offset >= isize) 9911c6fdbd8SKent Overstreet return -ENXIO; 9921c6fdbd8SKent Overstreet 993*6bd68ec2SKent Overstreet trans = bch2_trans_get(c); 9946fed42bbSKent Overstreet retry: 995*6bd68ec2SKent Overstreet bch2_trans_begin(trans); 9966fed42bbSKent Overstreet 997*6bd68ec2SKent Overstreet ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot); 9986fed42bbSKent Overstreet if (ret) 9996fed42bbSKent Overstreet goto err; 1000424eb881SKent Overstreet 1001*6bd68ec2SKent Overstreet for_each_btree_key_norestart(trans, iter, BTREE_ID_extents, 10026fed42bbSKent Overstreet SPOS(inode->v.i_ino, offset >> 9, snapshot), 100394f651e2SKent Overstreet BTREE_ITER_SLOTS, k, ret) { 10041c6fdbd8SKent Overstreet if (k.k->p.inode != inode->v.i_ino) { 1005543ef2ebSKent Overstreet next_hole = bch2_seek_pagecache_hole(&inode->v, 10064198bf03SKent Overstreet offset, MAX_LFS_FILESIZE, 0, false); 10071c6fdbd8SKent Overstreet break; 10081c6fdbd8SKent Overstreet } else if (!bkey_extent_is_data(k.k)) { 1009543ef2ebSKent Overstreet next_hole = bch2_seek_pagecache_hole(&inode->v, 10101c6fdbd8SKent Overstreet max(offset, bkey_start_offset(k.k) << 9), 10114198bf03SKent Overstreet k.k->p.offset << 9, 0, false); 10121c6fdbd8SKent Overstreet 10131c6fdbd8SKent Overstreet if (next_hole < k.k->p.offset << 9) 10141c6fdbd8SKent Overstreet break; 10151c6fdbd8SKent Overstreet } else { 10161c6fdbd8SKent Overstreet offset = max(offset, bkey_start_offset(k.k) << 9); 10171c6fdbd8SKent Overstreet } 10181c6fdbd8SKent Overstreet } 1019*6bd68ec2SKent Overstreet bch2_trans_iter_exit(trans, &iter); 10206fed42bbSKent Overstreet err: 1021549d173cSKent Overstreet if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) 10226fed42bbSKent Overstreet goto retry; 10231c6fdbd8SKent Overstreet 1024*6bd68ec2SKent Overstreet bch2_trans_put(trans); 10251c6fdbd8SKent Overstreet if (ret) 10261c6fdbd8SKent Overstreet return ret; 10271c6fdbd8SKent Overstreet 10281c6fdbd8SKent Overstreet if (next_hole > isize) 10291c6fdbd8SKent Overstreet next_hole = isize; 10301c6fdbd8SKent Overstreet 10311c6fdbd8SKent Overstreet return vfs_setpos(file, next_hole, MAX_LFS_FILESIZE); 10321c6fdbd8SKent Overstreet } 10331c6fdbd8SKent Overstreet 10341c6fdbd8SKent Overstreet loff_t bch2_llseek(struct file *file, loff_t offset, int whence) 10351c6fdbd8SKent Overstreet { 10365c1ef830SKent Overstreet loff_t ret; 10375c1ef830SKent Overstreet 10381c6fdbd8SKent Overstreet switch (whence) { 10391c6fdbd8SKent Overstreet case SEEK_SET: 10401c6fdbd8SKent Overstreet case SEEK_CUR: 10411c6fdbd8SKent Overstreet case SEEK_END: 10425c1ef830SKent Overstreet ret = generic_file_llseek(file, offset, whence); 10435c1ef830SKent Overstreet break; 10441c6fdbd8SKent Overstreet case SEEK_DATA: 10455c1ef830SKent Overstreet ret = bch2_seek_data(file, offset); 10465c1ef830SKent Overstreet break; 10471c6fdbd8SKent Overstreet case SEEK_HOLE: 10485c1ef830SKent Overstreet ret = bch2_seek_hole(file, offset); 10495c1ef830SKent Overstreet break; 10505c1ef830SKent Overstreet default: 10515c1ef830SKent Overstreet ret = -EINVAL; 10525c1ef830SKent Overstreet break; 10531c6fdbd8SKent Overstreet } 10541c6fdbd8SKent Overstreet 10555c1ef830SKent Overstreet return bch2_err_class(ret); 10561c6fdbd8SKent Overstreet } 10571c6fdbd8SKent Overstreet 10581c6fdbd8SKent Overstreet void bch2_fs_fsio_exit(struct bch_fs *c) 10591c6fdbd8SKent Overstreet { 1060a8b3a677SKent Overstreet bioset_exit(&c->nocow_flush_bioset); 10611c6fdbd8SKent Overstreet } 10621c6fdbd8SKent Overstreet 10631c6fdbd8SKent Overstreet int bch2_fs_fsio_init(struct bch_fs *c) 10641c6fdbd8SKent Overstreet { 106565d48e35SKent Overstreet if (bioset_init(&c->nocow_flush_bioset, 1066a8b3a677SKent Overstreet 1, offsetof(struct nocow_flush, bio), 0)) 106765d48e35SKent Overstreet return -BCH_ERR_ENOMEM_nocow_flush_bioset_init; 10681c6fdbd8SKent Overstreet 1069c8b4534dSKent Overstreet return 0; 10701c6fdbd8SKent Overstreet } 10711c6fdbd8SKent Overstreet 10721c6fdbd8SKent Overstreet #endif /* NO_BCACHEFS_FS */ 1073