1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * linux/fs/ext4/inode.c 4 * 5 * Copyright (C) 1992, 1993, 1994, 1995 6 * Remy Card (card@masi.ibp.fr) 7 * Laboratoire MASI - Institut Blaise Pascal 8 * Universite Pierre et Marie Curie (Paris VI) 9 * 10 * from 11 * 12 * linux/fs/minix/inode.c 13 * 14 * Copyright (C) 1991, 1992 Linus Torvalds 15 * 16 * 64-bit file support on 64-bit platforms by Jakub Jelinek 17 * (jj@sunsite.ms.mff.cuni.cz) 18 * 19 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000 20 */ 21 22 #include <linux/fs.h> 23 #include <linux/mount.h> 24 #include <linux/time.h> 25 #include <linux/highuid.h> 26 #include <linux/pagemap.h> 27 #include <linux/dax.h> 28 #include <linux/quotaops.h> 29 #include <linux/string.h> 30 #include <linux/buffer_head.h> 31 #include <linux/writeback.h> 32 #include <linux/pagevec.h> 33 #include <linux/mpage.h> 34 #include <linux/rmap.h> 35 #include <linux/namei.h> 36 #include <linux/uio.h> 37 #include <linux/bio.h> 38 #include <linux/workqueue.h> 39 #include <linux/kernel.h> 40 #include <linux/printk.h> 41 #include <linux/slab.h> 42 #include <linux/bitops.h> 43 #include <linux/iomap.h> 44 #include <linux/iversion.h> 45 46 #include "ext4_jbd2.h" 47 #include "xattr.h" 48 #include "acl.h" 49 #include "truncate.h" 50 51 #include <trace/events/ext4.h> 52 53 static void ext4_journalled_zero_new_buffers(handle_t *handle, 54 struct inode *inode, 55 struct folio *folio, 56 unsigned from, unsigned to); 57 58 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw, 59 struct ext4_inode_info *ei) 60 { 61 __u32 csum; 62 __u16 dummy_csum = 0; 63 int offset = offsetof(struct ext4_inode, i_checksum_lo); 64 unsigned int csum_size = sizeof(dummy_csum); 65 66 csum = ext4_chksum(ei->i_csum_seed, (__u8 *)raw, offset); 67 csum = ext4_chksum(csum, (__u8 *)&dummy_csum, csum_size); 68 offset += csum_size; 69 csum = ext4_chksum(csum, (__u8 *)raw + offset, 70 EXT4_GOOD_OLD_INODE_SIZE - offset); 71 72 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { 73 offset = offsetof(struct ext4_inode, i_checksum_hi); 74 csum = ext4_chksum(csum, (__u8 *)raw + EXT4_GOOD_OLD_INODE_SIZE, 75 offset - EXT4_GOOD_OLD_INODE_SIZE); 76 if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) { 77 csum = ext4_chksum(csum, (__u8 *)&dummy_csum, 78 csum_size); 79 offset += csum_size; 80 } 81 csum = ext4_chksum(csum, (__u8 *)raw + offset, 82 EXT4_INODE_SIZE(inode->i_sb) - offset); 83 } 84 85 return csum; 86 } 87 88 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw, 89 struct ext4_inode_info *ei) 90 { 91 __u32 provided, calculated; 92 93 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != 94 cpu_to_le32(EXT4_OS_LINUX) || 95 !ext4_has_feature_metadata_csum(inode->i_sb)) 96 return 1; 97 98 provided = le16_to_cpu(raw->i_checksum_lo); 99 calculated = ext4_inode_csum(inode, raw, ei); 100 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && 101 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) 102 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16; 103 else 104 calculated &= 0xFFFF; 105 106 return provided == calculated; 107 } 108 109 void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw, 110 struct ext4_inode_info *ei) 111 { 112 __u32 csum; 113 114 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != 115 cpu_to_le32(EXT4_OS_LINUX) || 116 !ext4_has_feature_metadata_csum(inode->i_sb)) 117 return; 118 119 csum = ext4_inode_csum(inode, raw, ei); 120 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF); 121 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && 122 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) 123 raw->i_checksum_hi = cpu_to_le16(csum >> 16); 124 } 125 126 static inline int ext4_begin_ordered_truncate(struct inode *inode, 127 loff_t new_size) 128 { 129 trace_ext4_begin_ordered_truncate(inode, new_size); 130 /* 131 * If jinode is zero, then we never opened the file for 132 * writing, so there's no need to call 133 * jbd2_journal_begin_ordered_truncate() since there's no 134 * outstanding writes we need to flush. 135 */ 136 if (!EXT4_I(inode)->jinode) 137 return 0; 138 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode), 139 EXT4_I(inode)->jinode, 140 new_size); 141 } 142 143 /* 144 * Test whether an inode is a fast symlink. 145 * A fast symlink has its symlink data stored in ext4_inode_info->i_data. 146 */ 147 int ext4_inode_is_fast_symlink(struct inode *inode) 148 { 149 if (!ext4_has_feature_ea_inode(inode->i_sb)) { 150 int ea_blocks = EXT4_I(inode)->i_file_acl ? 151 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0; 152 153 if (ext4_has_inline_data(inode)) 154 return 0; 155 156 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); 157 } 158 return S_ISLNK(inode->i_mode) && inode->i_size && 159 (inode->i_size < EXT4_N_BLOCKS * 4); 160 } 161 162 /* 163 * Called at the last iput() if i_nlink is zero. 164 */ 165 void ext4_evict_inode(struct inode *inode) 166 { 167 handle_t *handle; 168 int err; 169 /* 170 * Credits for final inode cleanup and freeing: 171 * sb + inode (ext4_orphan_del()), block bitmap, group descriptor 172 * (xattr block freeing), bitmap, group descriptor (inode freeing) 173 */ 174 int extra_credits = 6; 175 struct ext4_xattr_inode_array *ea_inode_array = NULL; 176 bool freeze_protected = false; 177 178 trace_ext4_evict_inode(inode); 179 180 dax_break_layout_final(inode); 181 182 if (EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL) 183 ext4_evict_ea_inode(inode); 184 if (inode->i_nlink) { 185 truncate_inode_pages_final(&inode->i_data); 186 187 goto no_delete; 188 } 189 190 if (is_bad_inode(inode)) 191 goto no_delete; 192 dquot_initialize(inode); 193 194 if (ext4_should_order_data(inode)) 195 ext4_begin_ordered_truncate(inode, 0); 196 truncate_inode_pages_final(&inode->i_data); 197 198 /* 199 * For inodes with journalled data, transaction commit could have 200 * dirtied the inode. And for inodes with dioread_nolock, unwritten 201 * extents converting worker could merge extents and also have dirtied 202 * the inode. Flush worker is ignoring it because of I_FREEING flag but 203 * we still need to remove the inode from the writeback lists. 204 */ 205 if (!list_empty_careful(&inode->i_io_list)) 206 inode_io_list_del(inode); 207 208 /* 209 * Protect us against freezing - iput() caller didn't have to have any 210 * protection against it. When we are in a running transaction though, 211 * we are already protected against freezing and we cannot grab further 212 * protection due to lock ordering constraints. 213 */ 214 if (!ext4_journal_current_handle()) { 215 sb_start_intwrite(inode->i_sb); 216 freeze_protected = true; 217 } 218 219 if (!IS_NOQUOTA(inode)) 220 extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb); 221 222 /* 223 * Block bitmap, group descriptor, and inode are accounted in both 224 * ext4_blocks_for_truncate() and extra_credits. So subtract 3. 225 */ 226 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, 227 ext4_blocks_for_truncate(inode) + extra_credits - 3); 228 if (IS_ERR(handle)) { 229 ext4_std_error(inode->i_sb, PTR_ERR(handle)); 230 /* 231 * If we're going to skip the normal cleanup, we still need to 232 * make sure that the in-core orphan linked list is properly 233 * cleaned up. 234 */ 235 ext4_orphan_del(NULL, inode); 236 if (freeze_protected) 237 sb_end_intwrite(inode->i_sb); 238 goto no_delete; 239 } 240 241 if (IS_SYNC(inode)) 242 ext4_handle_sync(handle); 243 244 /* 245 * Set inode->i_size to 0 before calling ext4_truncate(). We need 246 * special handling of symlinks here because i_size is used to 247 * determine whether ext4_inode_info->i_data contains symlink data or 248 * block mappings. Setting i_size to 0 will remove its fast symlink 249 * status. Erase i_data so that it becomes a valid empty block map. 250 */ 251 if (ext4_inode_is_fast_symlink(inode)) 252 memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data)); 253 inode->i_size = 0; 254 err = ext4_mark_inode_dirty(handle, inode); 255 if (err) { 256 ext4_warning(inode->i_sb, 257 "couldn't mark inode dirty (err %d)", err); 258 goto stop_handle; 259 } 260 if (inode->i_blocks) { 261 err = ext4_truncate(inode); 262 if (err) { 263 ext4_error_err(inode->i_sb, -err, 264 "couldn't truncate inode %lu (err %d)", 265 inode->i_ino, err); 266 goto stop_handle; 267 } 268 } 269 270 /* Remove xattr references. */ 271 err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array, 272 extra_credits); 273 if (err) { 274 ext4_warning(inode->i_sb, "xattr delete (err %d)", err); 275 stop_handle: 276 ext4_journal_stop(handle); 277 ext4_orphan_del(NULL, inode); 278 if (freeze_protected) 279 sb_end_intwrite(inode->i_sb); 280 ext4_xattr_inode_array_free(ea_inode_array); 281 goto no_delete; 282 } 283 284 /* 285 * Kill off the orphan record which ext4_truncate created. 286 * AKPM: I think this can be inside the above `if'. 287 * Note that ext4_orphan_del() has to be able to cope with the 288 * deletion of a non-existent orphan - this is because we don't 289 * know if ext4_truncate() actually created an orphan record. 290 * (Well, we could do this if we need to, but heck - it works) 291 */ 292 ext4_orphan_del(handle, inode); 293 EXT4_I(inode)->i_dtime = (__u32)ktime_get_real_seconds(); 294 295 /* 296 * One subtle ordering requirement: if anything has gone wrong 297 * (transaction abort, IO errors, whatever), then we can still 298 * do these next steps (the fs will already have been marked as 299 * having errors), but we can't free the inode if the mark_dirty 300 * fails. 301 */ 302 if (ext4_mark_inode_dirty(handle, inode)) 303 /* If that failed, just do the required in-core inode clear. */ 304 ext4_clear_inode(inode); 305 else 306 ext4_free_inode(handle, inode); 307 ext4_journal_stop(handle); 308 if (freeze_protected) 309 sb_end_intwrite(inode->i_sb); 310 ext4_xattr_inode_array_free(ea_inode_array); 311 return; 312 no_delete: 313 /* 314 * Check out some where else accidentally dirty the evicting inode, 315 * which may probably cause inode use-after-free issues later. 316 */ 317 WARN_ON_ONCE(!list_empty_careful(&inode->i_io_list)); 318 319 if (!list_empty(&EXT4_I(inode)->i_fc_list)) 320 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_NOMEM, NULL); 321 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */ 322 } 323 324 #ifdef CONFIG_QUOTA 325 qsize_t *ext4_get_reserved_space(struct inode *inode) 326 { 327 return &EXT4_I(inode)->i_reserved_quota; 328 } 329 #endif 330 331 /* 332 * Called with i_data_sem down, which is important since we can call 333 * ext4_discard_preallocations() from here. 334 */ 335 void ext4_da_update_reserve_space(struct inode *inode, 336 int used, int quota_claim) 337 { 338 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 339 struct ext4_inode_info *ei = EXT4_I(inode); 340 341 spin_lock(&ei->i_block_reservation_lock); 342 trace_ext4_da_update_reserve_space(inode, used, quota_claim); 343 if (unlikely(used > ei->i_reserved_data_blocks)) { 344 ext4_warning(inode->i_sb, "%s: ino %lu, used %d " 345 "with only %d reserved data blocks", 346 __func__, inode->i_ino, used, 347 ei->i_reserved_data_blocks); 348 WARN_ON(1); 349 used = ei->i_reserved_data_blocks; 350 } 351 352 /* Update per-inode reservations */ 353 ei->i_reserved_data_blocks -= used; 354 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used); 355 356 spin_unlock(&ei->i_block_reservation_lock); 357 358 /* Update quota subsystem for data blocks */ 359 if (quota_claim) 360 dquot_claim_block(inode, EXT4_C2B(sbi, used)); 361 else { 362 /* 363 * We did fallocate with an offset that is already delayed 364 * allocated. So on delayed allocated writeback we should 365 * not re-claim the quota for fallocated blocks. 366 */ 367 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used)); 368 } 369 370 /* 371 * If we have done all the pending block allocations and if 372 * there aren't any writers on the inode, we can discard the 373 * inode's preallocations. 374 */ 375 if ((ei->i_reserved_data_blocks == 0) && 376 !inode_is_open_for_write(inode)) 377 ext4_discard_preallocations(inode); 378 } 379 380 static int __check_block_validity(struct inode *inode, const char *func, 381 unsigned int line, 382 struct ext4_map_blocks *map) 383 { 384 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; 385 386 if (journal && inode == journal->j_inode) 387 return 0; 388 389 if (!ext4_inode_block_valid(inode, map->m_pblk, map->m_len)) { 390 ext4_error_inode(inode, func, line, map->m_pblk, 391 "lblock %lu mapped to illegal pblock %llu " 392 "(length %d)", (unsigned long) map->m_lblk, 393 map->m_pblk, map->m_len); 394 return -EFSCORRUPTED; 395 } 396 return 0; 397 } 398 399 int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk, 400 ext4_lblk_t len) 401 { 402 int ret; 403 404 if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode)) 405 return fscrypt_zeroout_range(inode, lblk, pblk, len); 406 407 ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS); 408 if (ret > 0) 409 ret = 0; 410 411 return ret; 412 } 413 414 /* 415 * For generic regular files, when updating the extent tree, Ext4 should 416 * hold the i_rwsem and invalidate_lock exclusively. This ensures 417 * exclusion against concurrent page faults, as well as reads and writes. 418 */ 419 #ifdef CONFIG_EXT4_DEBUG 420 void ext4_check_map_extents_env(struct inode *inode) 421 { 422 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) 423 return; 424 425 if (!S_ISREG(inode->i_mode) || 426 IS_NOQUOTA(inode) || IS_VERITY(inode) || 427 is_special_ino(inode->i_sb, inode->i_ino) || 428 (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW)) || 429 ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE) || 430 ext4_verity_in_progress(inode)) 431 return; 432 433 WARN_ON_ONCE(!inode_is_locked(inode) && 434 !rwsem_is_locked(&inode->i_mapping->invalidate_lock)); 435 } 436 #else 437 void ext4_check_map_extents_env(struct inode *inode) {} 438 #endif 439 440 #define check_block_validity(inode, map) \ 441 __check_block_validity((inode), __func__, __LINE__, (map)) 442 443 #ifdef ES_AGGRESSIVE_TEST 444 static void ext4_map_blocks_es_recheck(handle_t *handle, 445 struct inode *inode, 446 struct ext4_map_blocks *es_map, 447 struct ext4_map_blocks *map, 448 int flags) 449 { 450 int retval; 451 452 map->m_flags = 0; 453 /* 454 * There is a race window that the result is not the same. 455 * e.g. xfstests #223 when dioread_nolock enables. The reason 456 * is that we lookup a block mapping in extent status tree with 457 * out taking i_data_sem. So at the time the unwritten extent 458 * could be converted. 459 */ 460 down_read(&EXT4_I(inode)->i_data_sem); 461 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { 462 retval = ext4_ext_map_blocks(handle, inode, map, 0); 463 } else { 464 retval = ext4_ind_map_blocks(handle, inode, map, 0); 465 } 466 up_read((&EXT4_I(inode)->i_data_sem)); 467 468 /* 469 * We don't check m_len because extent will be collpased in status 470 * tree. So the m_len might not equal. 471 */ 472 if (es_map->m_lblk != map->m_lblk || 473 es_map->m_flags != map->m_flags || 474 es_map->m_pblk != map->m_pblk) { 475 printk("ES cache assertion failed for inode: %lu " 476 "es_cached ex [%d/%d/%llu/%x] != " 477 "found ex [%d/%d/%llu/%x] retval %d flags %x\n", 478 inode->i_ino, es_map->m_lblk, es_map->m_len, 479 es_map->m_pblk, es_map->m_flags, map->m_lblk, 480 map->m_len, map->m_pblk, map->m_flags, 481 retval, flags); 482 } 483 } 484 #endif /* ES_AGGRESSIVE_TEST */ 485 486 static int ext4_map_query_blocks_next_in_leaf(handle_t *handle, 487 struct inode *inode, struct ext4_map_blocks *map, 488 unsigned int orig_mlen) 489 { 490 struct ext4_map_blocks map2; 491 unsigned int status, status2; 492 int retval; 493 494 status = map->m_flags & EXT4_MAP_UNWRITTEN ? 495 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; 496 497 WARN_ON_ONCE(!(map->m_flags & EXT4_MAP_QUERY_LAST_IN_LEAF)); 498 WARN_ON_ONCE(orig_mlen <= map->m_len); 499 500 /* Prepare map2 for lookup in next leaf block */ 501 map2.m_lblk = map->m_lblk + map->m_len; 502 map2.m_len = orig_mlen - map->m_len; 503 map2.m_flags = 0; 504 retval = ext4_ext_map_blocks(handle, inode, &map2, 0); 505 506 if (retval <= 0) { 507 ext4_es_insert_extent(inode, map->m_lblk, map->m_len, 508 map->m_pblk, status, false); 509 return map->m_len; 510 } 511 512 if (unlikely(retval != map2.m_len)) { 513 ext4_warning(inode->i_sb, 514 "ES len assertion failed for inode " 515 "%lu: retval %d != map->m_len %d", 516 inode->i_ino, retval, map2.m_len); 517 WARN_ON(1); 518 } 519 520 status2 = map2.m_flags & EXT4_MAP_UNWRITTEN ? 521 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; 522 523 /* 524 * If map2 is contiguous with map, then let's insert it as a single 525 * extent in es cache and return the combined length of both the maps. 526 */ 527 if (map->m_pblk + map->m_len == map2.m_pblk && 528 status == status2) { 529 ext4_es_insert_extent(inode, map->m_lblk, 530 map->m_len + map2.m_len, map->m_pblk, 531 status, false); 532 map->m_len += map2.m_len; 533 } else { 534 ext4_es_insert_extent(inode, map->m_lblk, map->m_len, 535 map->m_pblk, status, false); 536 } 537 538 return map->m_len; 539 } 540 541 static int ext4_map_query_blocks(handle_t *handle, struct inode *inode, 542 struct ext4_map_blocks *map, int flags) 543 { 544 unsigned int status; 545 int retval; 546 unsigned int orig_mlen = map->m_len; 547 548 flags &= EXT4_EX_QUERY_FILTER; 549 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) 550 retval = ext4_ext_map_blocks(handle, inode, map, flags); 551 else 552 retval = ext4_ind_map_blocks(handle, inode, map, flags); 553 554 if (retval <= 0) 555 return retval; 556 557 if (unlikely(retval != map->m_len)) { 558 ext4_warning(inode->i_sb, 559 "ES len assertion failed for inode " 560 "%lu: retval %d != map->m_len %d", 561 inode->i_ino, retval, map->m_len); 562 WARN_ON(1); 563 } 564 565 /* 566 * No need to query next in leaf: 567 * - if returned extent is not last in leaf or 568 * - if the last in leaf is the full requested range 569 */ 570 if (!(map->m_flags & EXT4_MAP_QUERY_LAST_IN_LEAF) || 571 map->m_len == orig_mlen) { 572 status = map->m_flags & EXT4_MAP_UNWRITTEN ? 573 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; 574 ext4_es_insert_extent(inode, map->m_lblk, map->m_len, 575 map->m_pblk, status, false); 576 return retval; 577 } 578 579 return ext4_map_query_blocks_next_in_leaf(handle, inode, map, 580 orig_mlen); 581 } 582 583 static int ext4_map_create_blocks(handle_t *handle, struct inode *inode, 584 struct ext4_map_blocks *map, int flags) 585 { 586 struct extent_status es; 587 unsigned int status; 588 int err, retval = 0; 589 590 /* 591 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE 592 * indicates that the blocks and quotas has already been 593 * checked when the data was copied into the page cache. 594 */ 595 if (map->m_flags & EXT4_MAP_DELAYED) 596 flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; 597 598 /* 599 * Here we clear m_flags because after allocating an new extent, 600 * it will be set again. 601 */ 602 map->m_flags &= ~EXT4_MAP_FLAGS; 603 604 /* 605 * We need to check for EXT4 here because migrate could have 606 * changed the inode type in between. 607 */ 608 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { 609 retval = ext4_ext_map_blocks(handle, inode, map, flags); 610 } else { 611 retval = ext4_ind_map_blocks(handle, inode, map, flags); 612 613 /* 614 * We allocated new blocks which will result in i_data's 615 * format changing. Force the migrate to fail by clearing 616 * migrate flags. 617 */ 618 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) 619 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE); 620 } 621 if (retval <= 0) 622 return retval; 623 624 if (unlikely(retval != map->m_len)) { 625 ext4_warning(inode->i_sb, 626 "ES len assertion failed for inode %lu: " 627 "retval %d != map->m_len %d", 628 inode->i_ino, retval, map->m_len); 629 WARN_ON(1); 630 } 631 632 /* 633 * We have to zeroout blocks before inserting them into extent 634 * status tree. Otherwise someone could look them up there and 635 * use them before they are really zeroed. We also have to 636 * unmap metadata before zeroing as otherwise writeback can 637 * overwrite zeros with stale data from block device. 638 */ 639 if (flags & EXT4_GET_BLOCKS_ZERO && 640 map->m_flags & EXT4_MAP_MAPPED && map->m_flags & EXT4_MAP_NEW) { 641 err = ext4_issue_zeroout(inode, map->m_lblk, map->m_pblk, 642 map->m_len); 643 if (err) 644 return err; 645 } 646 647 /* 648 * If the extent has been zeroed out, we don't need to update 649 * extent status tree. 650 */ 651 if (flags & EXT4_GET_BLOCKS_PRE_IO && 652 ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) { 653 if (ext4_es_is_written(&es)) 654 return retval; 655 } 656 657 status = map->m_flags & EXT4_MAP_UNWRITTEN ? 658 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; 659 ext4_es_insert_extent(inode, map->m_lblk, map->m_len, map->m_pblk, 660 status, flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE); 661 662 return retval; 663 } 664 665 /* 666 * The ext4_map_blocks() function tries to look up the requested blocks, 667 * and returns if the blocks are already mapped. 668 * 669 * Otherwise it takes the write lock of the i_data_sem and allocate blocks 670 * and store the allocated blocks in the result buffer head and mark it 671 * mapped. 672 * 673 * If file type is extents based, it will call ext4_ext_map_blocks(), 674 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping 675 * based files 676 * 677 * On success, it returns the number of blocks being mapped or allocated. 678 * If flags doesn't contain EXT4_GET_BLOCKS_CREATE the blocks are 679 * pre-allocated and unwritten, the resulting @map is marked as unwritten. 680 * If the flags contain EXT4_GET_BLOCKS_CREATE, it will mark @map as mapped. 681 * 682 * It returns 0 if plain look up failed (blocks have not been allocated), in 683 * that case, @map is returned as unmapped but we still do fill map->m_len to 684 * indicate the length of a hole starting at map->m_lblk. 685 * 686 * It returns the error in case of allocation failure. 687 */ 688 int ext4_map_blocks(handle_t *handle, struct inode *inode, 689 struct ext4_map_blocks *map, int flags) 690 { 691 struct extent_status es; 692 int retval; 693 int ret = 0; 694 unsigned int orig_mlen = map->m_len; 695 #ifdef ES_AGGRESSIVE_TEST 696 struct ext4_map_blocks orig_map; 697 698 memcpy(&orig_map, map, sizeof(*map)); 699 #endif 700 701 map->m_flags = 0; 702 ext_debug(inode, "flag 0x%x, max_blocks %u, logical block %lu\n", 703 flags, map->m_len, (unsigned long) map->m_lblk); 704 705 /* 706 * ext4_map_blocks returns an int, and m_len is an unsigned int 707 */ 708 if (unlikely(map->m_len > INT_MAX)) 709 map->m_len = INT_MAX; 710 711 /* We can handle the block number less than EXT_MAX_BLOCKS */ 712 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS)) 713 return -EFSCORRUPTED; 714 715 /* 716 * Callers from the context of data submission are the only exceptions 717 * for regular files that do not hold the i_rwsem or invalidate_lock. 718 * However, caching unrelated ranges is not permitted. 719 */ 720 if (flags & EXT4_GET_BLOCKS_IO_SUBMIT) 721 WARN_ON_ONCE(!(flags & EXT4_EX_NOCACHE)); 722 else 723 ext4_check_map_extents_env(inode); 724 725 /* Lookup extent status tree firstly */ 726 if (ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) { 727 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) { 728 map->m_pblk = ext4_es_pblock(&es) + 729 map->m_lblk - es.es_lblk; 730 map->m_flags |= ext4_es_is_written(&es) ? 731 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN; 732 retval = es.es_len - (map->m_lblk - es.es_lblk); 733 if (retval > map->m_len) 734 retval = map->m_len; 735 map->m_len = retval; 736 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) { 737 map->m_pblk = 0; 738 map->m_flags |= ext4_es_is_delayed(&es) ? 739 EXT4_MAP_DELAYED : 0; 740 retval = es.es_len - (map->m_lblk - es.es_lblk); 741 if (retval > map->m_len) 742 retval = map->m_len; 743 map->m_len = retval; 744 retval = 0; 745 } else { 746 BUG(); 747 } 748 749 if (flags & EXT4_GET_BLOCKS_CACHED_NOWAIT) 750 return retval; 751 #ifdef ES_AGGRESSIVE_TEST 752 ext4_map_blocks_es_recheck(handle, inode, map, 753 &orig_map, flags); 754 #endif 755 if (!(flags & EXT4_GET_BLOCKS_QUERY_LAST_IN_LEAF) || 756 orig_mlen == map->m_len) 757 goto found; 758 759 map->m_len = orig_mlen; 760 } 761 /* 762 * In the query cache no-wait mode, nothing we can do more if we 763 * cannot find extent in the cache. 764 */ 765 if (flags & EXT4_GET_BLOCKS_CACHED_NOWAIT) 766 return 0; 767 768 /* 769 * Try to see if we can get the block without requesting a new 770 * file system block. 771 */ 772 down_read(&EXT4_I(inode)->i_data_sem); 773 retval = ext4_map_query_blocks(handle, inode, map, flags); 774 up_read((&EXT4_I(inode)->i_data_sem)); 775 776 found: 777 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { 778 ret = check_block_validity(inode, map); 779 if (ret != 0) 780 return ret; 781 } 782 783 /* If it is only a block(s) look up */ 784 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) 785 return retval; 786 787 /* 788 * Returns if the blocks have already allocated 789 * 790 * Note that if blocks have been preallocated 791 * ext4_ext_map_blocks() returns with buffer head unmapped 792 */ 793 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) 794 /* 795 * If we need to convert extent to unwritten 796 * we continue and do the actual work in 797 * ext4_ext_map_blocks() 798 */ 799 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) 800 return retval; 801 802 803 ext4_fc_track_inode(handle, inode); 804 /* 805 * New blocks allocate and/or writing to unwritten extent 806 * will possibly result in updating i_data, so we take 807 * the write lock of i_data_sem, and call get_block() 808 * with create == 1 flag. 809 */ 810 down_write(&EXT4_I(inode)->i_data_sem); 811 retval = ext4_map_create_blocks(handle, inode, map, flags); 812 up_write((&EXT4_I(inode)->i_data_sem)); 813 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { 814 ret = check_block_validity(inode, map); 815 if (ret != 0) 816 return ret; 817 818 /* 819 * Inodes with freshly allocated blocks where contents will be 820 * visible after transaction commit must be on transaction's 821 * ordered data list. 822 */ 823 if (map->m_flags & EXT4_MAP_NEW && 824 !(map->m_flags & EXT4_MAP_UNWRITTEN) && 825 !(flags & EXT4_GET_BLOCKS_ZERO) && 826 !ext4_is_quota_file(inode) && 827 ext4_should_order_data(inode)) { 828 loff_t start_byte = 829 (loff_t)map->m_lblk << inode->i_blkbits; 830 loff_t length = (loff_t)map->m_len << inode->i_blkbits; 831 832 if (flags & EXT4_GET_BLOCKS_IO_SUBMIT) 833 ret = ext4_jbd2_inode_add_wait(handle, inode, 834 start_byte, length); 835 else 836 ret = ext4_jbd2_inode_add_write(handle, inode, 837 start_byte, length); 838 if (ret) 839 return ret; 840 } 841 } 842 if (retval > 0 && (map->m_flags & EXT4_MAP_UNWRITTEN || 843 map->m_flags & EXT4_MAP_MAPPED)) 844 ext4_fc_track_range(handle, inode, map->m_lblk, 845 map->m_lblk + map->m_len - 1); 846 if (retval < 0) 847 ext_debug(inode, "failed with err %d\n", retval); 848 return retval; 849 } 850 851 /* 852 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages 853 * we have to be careful as someone else may be manipulating b_state as well. 854 */ 855 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags) 856 { 857 unsigned long old_state; 858 unsigned long new_state; 859 860 flags &= EXT4_MAP_FLAGS; 861 862 /* Dummy buffer_head? Set non-atomically. */ 863 if (!bh->b_folio) { 864 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags; 865 return; 866 } 867 /* 868 * Someone else may be modifying b_state. Be careful! This is ugly but 869 * once we get rid of using bh as a container for mapping information 870 * to pass to / from get_block functions, this can go away. 871 */ 872 old_state = READ_ONCE(bh->b_state); 873 do { 874 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags; 875 } while (unlikely(!try_cmpxchg(&bh->b_state, &old_state, new_state))); 876 } 877 878 /* 879 * Make sure that the current journal transaction has enough credits to map 880 * one extent. Return -EAGAIN if it cannot extend the current running 881 * transaction. 882 */ 883 static inline int ext4_journal_ensure_extent_credits(handle_t *handle, 884 struct inode *inode) 885 { 886 int credits; 887 int ret; 888 889 /* Called from ext4_da_write_begin() which has no handle started? */ 890 if (!handle) 891 return 0; 892 893 credits = ext4_chunk_trans_blocks(inode, 1); 894 ret = __ext4_journal_ensure_credits(handle, credits, credits, 0); 895 return ret <= 0 ? ret : -EAGAIN; 896 } 897 898 static int _ext4_get_block(struct inode *inode, sector_t iblock, 899 struct buffer_head *bh, int flags) 900 { 901 struct ext4_map_blocks map; 902 int ret = 0; 903 904 if (ext4_has_inline_data(inode)) 905 return -ERANGE; 906 907 map.m_lblk = iblock; 908 map.m_len = bh->b_size >> inode->i_blkbits; 909 910 ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map, 911 flags); 912 if (ret > 0) { 913 map_bh(bh, inode->i_sb, map.m_pblk); 914 ext4_update_bh_state(bh, map.m_flags); 915 bh->b_size = inode->i_sb->s_blocksize * map.m_len; 916 ret = 0; 917 } else if (ret == 0) { 918 /* hole case, need to fill in bh->b_size */ 919 bh->b_size = inode->i_sb->s_blocksize * map.m_len; 920 } 921 return ret; 922 } 923 924 int ext4_get_block(struct inode *inode, sector_t iblock, 925 struct buffer_head *bh, int create) 926 { 927 return _ext4_get_block(inode, iblock, bh, 928 create ? EXT4_GET_BLOCKS_CREATE : 0); 929 } 930 931 /* 932 * Get block function used when preparing for buffered write if we require 933 * creating an unwritten extent if blocks haven't been allocated. The extent 934 * will be converted to written after the IO is complete. 935 */ 936 int ext4_get_block_unwritten(struct inode *inode, sector_t iblock, 937 struct buffer_head *bh_result, int create) 938 { 939 int ret = 0; 940 941 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n", 942 inode->i_ino, create); 943 ret = _ext4_get_block(inode, iblock, bh_result, 944 EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT); 945 946 /* 947 * If the buffer is marked unwritten, mark it as new to make sure it is 948 * zeroed out correctly in case of partial writes. Otherwise, there is 949 * a chance of stale data getting exposed. 950 */ 951 if (ret == 0 && buffer_unwritten(bh_result)) 952 set_buffer_new(bh_result); 953 954 return ret; 955 } 956 957 /* Maximum number of blocks we map for direct IO at once. */ 958 #define DIO_MAX_BLOCKS 4096 959 960 /* 961 * `handle' can be NULL if create is zero 962 */ 963 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, 964 ext4_lblk_t block, int map_flags) 965 { 966 struct ext4_map_blocks map; 967 struct buffer_head *bh; 968 int create = map_flags & EXT4_GET_BLOCKS_CREATE; 969 bool nowait = map_flags & EXT4_GET_BLOCKS_CACHED_NOWAIT; 970 int err; 971 972 ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) 973 || handle != NULL || create == 0); 974 ASSERT(create == 0 || !nowait); 975 976 map.m_lblk = block; 977 map.m_len = 1; 978 err = ext4_map_blocks(handle, inode, &map, map_flags); 979 980 if (err == 0) 981 return create ? ERR_PTR(-ENOSPC) : NULL; 982 if (err < 0) 983 return ERR_PTR(err); 984 985 if (nowait) 986 return sb_find_get_block(inode->i_sb, map.m_pblk); 987 988 /* 989 * Since bh could introduce extra ref count such as referred by 990 * journal_head etc. Try to avoid using __GFP_MOVABLE here 991 * as it may fail the migration when journal_head remains. 992 */ 993 bh = getblk_unmovable(inode->i_sb->s_bdev, map.m_pblk, 994 inode->i_sb->s_blocksize); 995 996 if (unlikely(!bh)) 997 return ERR_PTR(-ENOMEM); 998 if (map.m_flags & EXT4_MAP_NEW) { 999 ASSERT(create != 0); 1000 ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) 1001 || (handle != NULL)); 1002 1003 /* 1004 * Now that we do not always journal data, we should 1005 * keep in mind whether this should always journal the 1006 * new buffer as metadata. For now, regular file 1007 * writes use ext4_get_block instead, so it's not a 1008 * problem. 1009 */ 1010 lock_buffer(bh); 1011 BUFFER_TRACE(bh, "call get_create_access"); 1012 err = ext4_journal_get_create_access(handle, inode->i_sb, bh, 1013 EXT4_JTR_NONE); 1014 if (unlikely(err)) { 1015 unlock_buffer(bh); 1016 goto errout; 1017 } 1018 if (!buffer_uptodate(bh)) { 1019 memset(bh->b_data, 0, inode->i_sb->s_blocksize); 1020 set_buffer_uptodate(bh); 1021 } 1022 unlock_buffer(bh); 1023 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); 1024 err = ext4_handle_dirty_metadata(handle, inode, bh); 1025 if (unlikely(err)) 1026 goto errout; 1027 } else 1028 BUFFER_TRACE(bh, "not a new buffer"); 1029 return bh; 1030 errout: 1031 brelse(bh); 1032 return ERR_PTR(err); 1033 } 1034 1035 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, 1036 ext4_lblk_t block, int map_flags) 1037 { 1038 struct buffer_head *bh; 1039 int ret; 1040 1041 bh = ext4_getblk(handle, inode, block, map_flags); 1042 if (IS_ERR(bh)) 1043 return bh; 1044 if (!bh || ext4_buffer_uptodate(bh)) 1045 return bh; 1046 1047 ret = ext4_read_bh_lock(bh, REQ_META | REQ_PRIO, true); 1048 if (ret) { 1049 put_bh(bh); 1050 return ERR_PTR(ret); 1051 } 1052 return bh; 1053 } 1054 1055 /* Read a contiguous batch of blocks. */ 1056 int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count, 1057 bool wait, struct buffer_head **bhs) 1058 { 1059 int i, err; 1060 1061 for (i = 0; i < bh_count; i++) { 1062 bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */); 1063 if (IS_ERR(bhs[i])) { 1064 err = PTR_ERR(bhs[i]); 1065 bh_count = i; 1066 goto out_brelse; 1067 } 1068 } 1069 1070 for (i = 0; i < bh_count; i++) 1071 /* Note that NULL bhs[i] is valid because of holes. */ 1072 if (bhs[i] && !ext4_buffer_uptodate(bhs[i])) 1073 ext4_read_bh_lock(bhs[i], REQ_META | REQ_PRIO, false); 1074 1075 if (!wait) 1076 return 0; 1077 1078 for (i = 0; i < bh_count; i++) 1079 if (bhs[i]) 1080 wait_on_buffer(bhs[i]); 1081 1082 for (i = 0; i < bh_count; i++) { 1083 if (bhs[i] && !buffer_uptodate(bhs[i])) { 1084 err = -EIO; 1085 goto out_brelse; 1086 } 1087 } 1088 return 0; 1089 1090 out_brelse: 1091 for (i = 0; i < bh_count; i++) { 1092 brelse(bhs[i]); 1093 bhs[i] = NULL; 1094 } 1095 return err; 1096 } 1097 1098 int ext4_walk_page_buffers(handle_t *handle, struct inode *inode, 1099 struct buffer_head *head, 1100 unsigned from, 1101 unsigned to, 1102 int *partial, 1103 int (*fn)(handle_t *handle, struct inode *inode, 1104 struct buffer_head *bh)) 1105 { 1106 struct buffer_head *bh; 1107 unsigned block_start, block_end; 1108 unsigned blocksize = head->b_size; 1109 int err, ret = 0; 1110 struct buffer_head *next; 1111 1112 for (bh = head, block_start = 0; 1113 ret == 0 && (bh != head || !block_start); 1114 block_start = block_end, bh = next) { 1115 next = bh->b_this_page; 1116 block_end = block_start + blocksize; 1117 if (block_end <= from || block_start >= to) { 1118 if (partial && !buffer_uptodate(bh)) 1119 *partial = 1; 1120 continue; 1121 } 1122 err = (*fn)(handle, inode, bh); 1123 if (!ret) 1124 ret = err; 1125 } 1126 return ret; 1127 } 1128 1129 /* 1130 * Helper for handling dirtying of journalled data. We also mark the folio as 1131 * dirty so that writeback code knows about this page (and inode) contains 1132 * dirty data. ext4_writepages() then commits appropriate transaction to 1133 * make data stable. 1134 */ 1135 static int ext4_dirty_journalled_data(handle_t *handle, struct buffer_head *bh) 1136 { 1137 struct folio *folio = bh->b_folio; 1138 struct inode *inode = folio->mapping->host; 1139 1140 /* only regular files have a_ops */ 1141 if (S_ISREG(inode->i_mode)) 1142 folio_mark_dirty(folio); 1143 return ext4_handle_dirty_metadata(handle, NULL, bh); 1144 } 1145 1146 int do_journal_get_write_access(handle_t *handle, struct inode *inode, 1147 struct buffer_head *bh) 1148 { 1149 if (!buffer_mapped(bh) || buffer_freed(bh)) 1150 return 0; 1151 BUFFER_TRACE(bh, "get write access"); 1152 return ext4_journal_get_write_access(handle, inode->i_sb, bh, 1153 EXT4_JTR_NONE); 1154 } 1155 1156 int ext4_block_write_begin(handle_t *handle, struct folio *folio, 1157 loff_t pos, unsigned len, 1158 get_block_t *get_block) 1159 { 1160 unsigned int from = offset_in_folio(folio, pos); 1161 unsigned to = from + len; 1162 struct inode *inode = folio->mapping->host; 1163 unsigned block_start, block_end; 1164 sector_t block; 1165 int err = 0; 1166 unsigned blocksize = inode->i_sb->s_blocksize; 1167 unsigned bbits; 1168 struct buffer_head *bh, *head, *wait[2]; 1169 int nr_wait = 0; 1170 int i; 1171 bool should_journal_data = ext4_should_journal_data(inode); 1172 1173 BUG_ON(!folio_test_locked(folio)); 1174 BUG_ON(to > folio_size(folio)); 1175 BUG_ON(from > to); 1176 1177 head = folio_buffers(folio); 1178 if (!head) 1179 head = create_empty_buffers(folio, blocksize, 0); 1180 bbits = ilog2(blocksize); 1181 block = (sector_t)folio->index << (PAGE_SHIFT - bbits); 1182 1183 for (bh = head, block_start = 0; bh != head || !block_start; 1184 block++, block_start = block_end, bh = bh->b_this_page) { 1185 block_end = block_start + blocksize; 1186 if (block_end <= from || block_start >= to) { 1187 if (folio_test_uptodate(folio)) { 1188 set_buffer_uptodate(bh); 1189 } 1190 continue; 1191 } 1192 if (WARN_ON_ONCE(buffer_new(bh))) 1193 clear_buffer_new(bh); 1194 if (!buffer_mapped(bh)) { 1195 WARN_ON(bh->b_size != blocksize); 1196 err = ext4_journal_ensure_extent_credits(handle, inode); 1197 if (!err) 1198 err = get_block(inode, block, bh, 1); 1199 if (err) 1200 break; 1201 if (buffer_new(bh)) { 1202 /* 1203 * We may be zeroing partial buffers or all new 1204 * buffers in case of failure. Prepare JBD2 for 1205 * that. 1206 */ 1207 if (should_journal_data) 1208 do_journal_get_write_access(handle, 1209 inode, bh); 1210 if (folio_test_uptodate(folio)) { 1211 /* 1212 * Unlike __block_write_begin() we leave 1213 * dirtying of new uptodate buffers to 1214 * ->write_end() time or 1215 * folio_zero_new_buffers(). 1216 */ 1217 set_buffer_uptodate(bh); 1218 continue; 1219 } 1220 if (block_end > to || block_start < from) 1221 folio_zero_segments(folio, to, 1222 block_end, 1223 block_start, from); 1224 continue; 1225 } 1226 } 1227 if (folio_test_uptodate(folio)) { 1228 set_buffer_uptodate(bh); 1229 continue; 1230 } 1231 if (!buffer_uptodate(bh) && !buffer_delay(bh) && 1232 !buffer_unwritten(bh) && 1233 (block_start < from || block_end > to)) { 1234 ext4_read_bh_lock(bh, 0, false); 1235 wait[nr_wait++] = bh; 1236 } 1237 } 1238 /* 1239 * If we issued read requests, let them complete. 1240 */ 1241 for (i = 0; i < nr_wait; i++) { 1242 wait_on_buffer(wait[i]); 1243 if (!buffer_uptodate(wait[i])) 1244 err = -EIO; 1245 } 1246 if (unlikely(err)) { 1247 if (should_journal_data) 1248 ext4_journalled_zero_new_buffers(handle, inode, folio, 1249 from, to); 1250 else 1251 folio_zero_new_buffers(folio, from, to); 1252 } else if (fscrypt_inode_uses_fs_layer_crypto(inode)) { 1253 for (i = 0; i < nr_wait; i++) { 1254 int err2; 1255 1256 err2 = fscrypt_decrypt_pagecache_blocks(folio, 1257 blocksize, bh_offset(wait[i])); 1258 if (err2) { 1259 clear_buffer_uptodate(wait[i]); 1260 err = err2; 1261 } 1262 } 1263 } 1264 1265 return err; 1266 } 1267 1268 /* 1269 * To preserve ordering, it is essential that the hole instantiation and 1270 * the data write be encapsulated in a single transaction. We cannot 1271 * close off a transaction and start a new one between the ext4_get_block() 1272 * and the ext4_write_end(). So doing the jbd2_journal_start at the start of 1273 * ext4_write_begin() is the right place. 1274 */ 1275 static int ext4_write_begin(const struct kiocb *iocb, 1276 struct address_space *mapping, 1277 loff_t pos, unsigned len, 1278 struct folio **foliop, void **fsdata) 1279 { 1280 struct inode *inode = mapping->host; 1281 int ret, needed_blocks; 1282 handle_t *handle; 1283 int retries = 0; 1284 struct folio *folio; 1285 pgoff_t index; 1286 unsigned from, to; 1287 1288 ret = ext4_emergency_state(inode->i_sb); 1289 if (unlikely(ret)) 1290 return ret; 1291 1292 trace_ext4_write_begin(inode, pos, len); 1293 /* 1294 * Reserve one block more for addition to orphan list in case 1295 * we allocate blocks but write fails for some reason 1296 */ 1297 needed_blocks = ext4_chunk_trans_extent(inode, 1298 ext4_journal_blocks_per_folio(inode)) + 1; 1299 index = pos >> PAGE_SHIFT; 1300 1301 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { 1302 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len, 1303 foliop); 1304 if (ret < 0) 1305 return ret; 1306 if (ret == 1) 1307 return 0; 1308 } 1309 1310 /* 1311 * write_begin_get_folio() can take a long time if the 1312 * system is thrashing due to memory pressure, or if the folio 1313 * is being written back. So grab it first before we start 1314 * the transaction handle. This also allows us to allocate 1315 * the folio (if needed) without using GFP_NOFS. 1316 */ 1317 retry_grab: 1318 folio = write_begin_get_folio(iocb, mapping, index, len); 1319 if (IS_ERR(folio)) 1320 return PTR_ERR(folio); 1321 1322 if (pos + len > folio_pos(folio) + folio_size(folio)) 1323 len = folio_pos(folio) + folio_size(folio) - pos; 1324 1325 from = offset_in_folio(folio, pos); 1326 to = from + len; 1327 1328 /* 1329 * The same as page allocation, we prealloc buffer heads before 1330 * starting the handle. 1331 */ 1332 if (!folio_buffers(folio)) 1333 create_empty_buffers(folio, inode->i_sb->s_blocksize, 0); 1334 1335 folio_unlock(folio); 1336 1337 retry_journal: 1338 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); 1339 if (IS_ERR(handle)) { 1340 folio_put(folio); 1341 return PTR_ERR(handle); 1342 } 1343 1344 folio_lock(folio); 1345 if (folio->mapping != mapping) { 1346 /* The folio got truncated from under us */ 1347 folio_unlock(folio); 1348 folio_put(folio); 1349 ext4_journal_stop(handle); 1350 goto retry_grab; 1351 } 1352 /* In case writeback began while the folio was unlocked */ 1353 folio_wait_stable(folio); 1354 1355 if (ext4_should_dioread_nolock(inode)) 1356 ret = ext4_block_write_begin(handle, folio, pos, len, 1357 ext4_get_block_unwritten); 1358 else 1359 ret = ext4_block_write_begin(handle, folio, pos, len, 1360 ext4_get_block); 1361 if (!ret && ext4_should_journal_data(inode)) { 1362 ret = ext4_walk_page_buffers(handle, inode, 1363 folio_buffers(folio), from, to, 1364 NULL, do_journal_get_write_access); 1365 } 1366 1367 if (ret) { 1368 bool extended = (pos + len > inode->i_size) && 1369 !ext4_verity_in_progress(inode); 1370 1371 folio_unlock(folio); 1372 /* 1373 * ext4_block_write_begin may have instantiated a few blocks 1374 * outside i_size. Trim these off again. Don't need 1375 * i_size_read because we hold i_rwsem. 1376 * 1377 * Add inode to orphan list in case we crash before 1378 * truncate finishes 1379 */ 1380 if (extended && ext4_can_truncate(inode)) 1381 ext4_orphan_add(handle, inode); 1382 1383 ext4_journal_stop(handle); 1384 if (extended) { 1385 ext4_truncate_failed_write(inode); 1386 /* 1387 * If truncate failed early the inode might 1388 * still be on the orphan list; we need to 1389 * make sure the inode is removed from the 1390 * orphan list in that case. 1391 */ 1392 if (inode->i_nlink) 1393 ext4_orphan_del(NULL, inode); 1394 } 1395 1396 if (ret == -EAGAIN || 1397 (ret == -ENOSPC && 1398 ext4_should_retry_alloc(inode->i_sb, &retries))) 1399 goto retry_journal; 1400 folio_put(folio); 1401 return ret; 1402 } 1403 *foliop = folio; 1404 return ret; 1405 } 1406 1407 /* For write_end() in data=journal mode */ 1408 static int write_end_fn(handle_t *handle, struct inode *inode, 1409 struct buffer_head *bh) 1410 { 1411 int ret; 1412 if (!buffer_mapped(bh) || buffer_freed(bh)) 1413 return 0; 1414 set_buffer_uptodate(bh); 1415 ret = ext4_dirty_journalled_data(handle, bh); 1416 clear_buffer_meta(bh); 1417 clear_buffer_prio(bh); 1418 clear_buffer_new(bh); 1419 return ret; 1420 } 1421 1422 /* 1423 * We need to pick up the new inode size which generic_commit_write gave us 1424 * `iocb` can be NULL - eg, when called from page_symlink(). 1425 * 1426 * ext4 never places buffers on inode->i_mapping->i_private_list. metadata 1427 * buffers are managed internally. 1428 */ 1429 static int ext4_write_end(const struct kiocb *iocb, 1430 struct address_space *mapping, 1431 loff_t pos, unsigned len, unsigned copied, 1432 struct folio *folio, void *fsdata) 1433 { 1434 handle_t *handle = ext4_journal_current_handle(); 1435 struct inode *inode = mapping->host; 1436 loff_t old_size = inode->i_size; 1437 int ret = 0, ret2; 1438 int i_size_changed = 0; 1439 bool verity = ext4_verity_in_progress(inode); 1440 1441 trace_ext4_write_end(inode, pos, len, copied); 1442 1443 if (ext4_has_inline_data(inode) && 1444 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) 1445 return ext4_write_inline_data_end(inode, pos, len, copied, 1446 folio); 1447 1448 copied = block_write_end(pos, len, copied, folio); 1449 /* 1450 * it's important to update i_size while still holding folio lock: 1451 * page writeout could otherwise come in and zero beyond i_size. 1452 * 1453 * If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree 1454 * blocks are being written past EOF, so skip the i_size update. 1455 */ 1456 if (!verity) 1457 i_size_changed = ext4_update_inode_size(inode, pos + copied); 1458 folio_unlock(folio); 1459 folio_put(folio); 1460 1461 if (old_size < pos && !verity) { 1462 pagecache_isize_extended(inode, old_size, pos); 1463 ext4_zero_partial_blocks(handle, inode, old_size, pos - old_size); 1464 } 1465 /* 1466 * Don't mark the inode dirty under folio lock. First, it unnecessarily 1467 * makes the holding time of folio lock longer. Second, it forces lock 1468 * ordering of folio lock and transaction start for journaling 1469 * filesystems. 1470 */ 1471 if (i_size_changed) 1472 ret = ext4_mark_inode_dirty(handle, inode); 1473 1474 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode)) 1475 /* if we have allocated more blocks and copied 1476 * less. We will have blocks allocated outside 1477 * inode->i_size. So truncate them 1478 */ 1479 ext4_orphan_add(handle, inode); 1480 1481 ret2 = ext4_journal_stop(handle); 1482 if (!ret) 1483 ret = ret2; 1484 1485 if (pos + len > inode->i_size && !verity) { 1486 ext4_truncate_failed_write(inode); 1487 /* 1488 * If truncate failed early the inode might still be 1489 * on the orphan list; we need to make sure the inode 1490 * is removed from the orphan list in that case. 1491 */ 1492 if (inode->i_nlink) 1493 ext4_orphan_del(NULL, inode); 1494 } 1495 1496 return ret ? ret : copied; 1497 } 1498 1499 /* 1500 * This is a private version of folio_zero_new_buffers() which doesn't 1501 * set the buffer to be dirty, since in data=journalled mode we need 1502 * to call ext4_dirty_journalled_data() instead. 1503 */ 1504 static void ext4_journalled_zero_new_buffers(handle_t *handle, 1505 struct inode *inode, 1506 struct folio *folio, 1507 unsigned from, unsigned to) 1508 { 1509 unsigned int block_start = 0, block_end; 1510 struct buffer_head *head, *bh; 1511 1512 bh = head = folio_buffers(folio); 1513 do { 1514 block_end = block_start + bh->b_size; 1515 if (buffer_new(bh)) { 1516 if (block_end > from && block_start < to) { 1517 if (!folio_test_uptodate(folio)) { 1518 unsigned start, size; 1519 1520 start = max(from, block_start); 1521 size = min(to, block_end) - start; 1522 1523 folio_zero_range(folio, start, size); 1524 } 1525 clear_buffer_new(bh); 1526 write_end_fn(handle, inode, bh); 1527 } 1528 } 1529 block_start = block_end; 1530 bh = bh->b_this_page; 1531 } while (bh != head); 1532 } 1533 1534 static int ext4_journalled_write_end(const struct kiocb *iocb, 1535 struct address_space *mapping, 1536 loff_t pos, unsigned len, unsigned copied, 1537 struct folio *folio, void *fsdata) 1538 { 1539 handle_t *handle = ext4_journal_current_handle(); 1540 struct inode *inode = mapping->host; 1541 loff_t old_size = inode->i_size; 1542 int ret = 0, ret2; 1543 int partial = 0; 1544 unsigned from, to; 1545 int size_changed = 0; 1546 bool verity = ext4_verity_in_progress(inode); 1547 1548 trace_ext4_journalled_write_end(inode, pos, len, copied); 1549 from = pos & (PAGE_SIZE - 1); 1550 to = from + len; 1551 1552 BUG_ON(!ext4_handle_valid(handle)); 1553 1554 if (ext4_has_inline_data(inode)) 1555 return ext4_write_inline_data_end(inode, pos, len, copied, 1556 folio); 1557 1558 if (unlikely(copied < len) && !folio_test_uptodate(folio)) { 1559 copied = 0; 1560 ext4_journalled_zero_new_buffers(handle, inode, folio, 1561 from, to); 1562 } else { 1563 if (unlikely(copied < len)) 1564 ext4_journalled_zero_new_buffers(handle, inode, folio, 1565 from + copied, to); 1566 ret = ext4_walk_page_buffers(handle, inode, 1567 folio_buffers(folio), 1568 from, from + copied, &partial, 1569 write_end_fn); 1570 if (!partial) 1571 folio_mark_uptodate(folio); 1572 } 1573 if (!verity) 1574 size_changed = ext4_update_inode_size(inode, pos + copied); 1575 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; 1576 folio_unlock(folio); 1577 folio_put(folio); 1578 1579 if (old_size < pos && !verity) { 1580 pagecache_isize_extended(inode, old_size, pos); 1581 ext4_zero_partial_blocks(handle, inode, old_size, pos - old_size); 1582 } 1583 1584 if (size_changed) { 1585 ret2 = ext4_mark_inode_dirty(handle, inode); 1586 if (!ret) 1587 ret = ret2; 1588 } 1589 1590 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode)) 1591 /* if we have allocated more blocks and copied 1592 * less. We will have blocks allocated outside 1593 * inode->i_size. So truncate them 1594 */ 1595 ext4_orphan_add(handle, inode); 1596 1597 ret2 = ext4_journal_stop(handle); 1598 if (!ret) 1599 ret = ret2; 1600 if (pos + len > inode->i_size && !verity) { 1601 ext4_truncate_failed_write(inode); 1602 /* 1603 * If truncate failed early the inode might still be 1604 * on the orphan list; we need to make sure the inode 1605 * is removed from the orphan list in that case. 1606 */ 1607 if (inode->i_nlink) 1608 ext4_orphan_del(NULL, inode); 1609 } 1610 1611 return ret ? ret : copied; 1612 } 1613 1614 /* 1615 * Reserve space for 'nr_resv' clusters 1616 */ 1617 static int ext4_da_reserve_space(struct inode *inode, int nr_resv) 1618 { 1619 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 1620 struct ext4_inode_info *ei = EXT4_I(inode); 1621 int ret; 1622 1623 /* 1624 * We will charge metadata quota at writeout time; this saves 1625 * us from metadata over-estimation, though we may go over by 1626 * a small amount in the end. Here we just reserve for data. 1627 */ 1628 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, nr_resv)); 1629 if (ret) 1630 return ret; 1631 1632 spin_lock(&ei->i_block_reservation_lock); 1633 if (ext4_claim_free_clusters(sbi, nr_resv, 0)) { 1634 spin_unlock(&ei->i_block_reservation_lock); 1635 dquot_release_reservation_block(inode, EXT4_C2B(sbi, nr_resv)); 1636 return -ENOSPC; 1637 } 1638 ei->i_reserved_data_blocks += nr_resv; 1639 trace_ext4_da_reserve_space(inode, nr_resv); 1640 spin_unlock(&ei->i_block_reservation_lock); 1641 1642 return 0; /* success */ 1643 } 1644 1645 void ext4_da_release_space(struct inode *inode, int to_free) 1646 { 1647 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 1648 struct ext4_inode_info *ei = EXT4_I(inode); 1649 1650 if (!to_free) 1651 return; /* Nothing to release, exit */ 1652 1653 spin_lock(&EXT4_I(inode)->i_block_reservation_lock); 1654 1655 trace_ext4_da_release_space(inode, to_free); 1656 if (unlikely(to_free > ei->i_reserved_data_blocks)) { 1657 /* 1658 * if there aren't enough reserved blocks, then the 1659 * counter is messed up somewhere. Since this 1660 * function is called from invalidate page, it's 1661 * harmless to return without any action. 1662 */ 1663 ext4_warning(inode->i_sb, "ext4_da_release_space: " 1664 "ino %lu, to_free %d with only %d reserved " 1665 "data blocks", inode->i_ino, to_free, 1666 ei->i_reserved_data_blocks); 1667 WARN_ON(1); 1668 to_free = ei->i_reserved_data_blocks; 1669 } 1670 ei->i_reserved_data_blocks -= to_free; 1671 1672 /* update fs dirty data blocks counter */ 1673 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free); 1674 1675 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); 1676 1677 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free)); 1678 } 1679 1680 /* 1681 * Delayed allocation stuff 1682 */ 1683 1684 struct mpage_da_data { 1685 /* These are input fields for ext4_do_writepages() */ 1686 struct inode *inode; 1687 struct writeback_control *wbc; 1688 unsigned int can_map:1; /* Can writepages call map blocks? */ 1689 1690 /* These are internal state of ext4_do_writepages() */ 1691 loff_t start_pos; /* The start pos to write */ 1692 loff_t next_pos; /* Current pos to examine */ 1693 loff_t end_pos; /* Last pos to examine */ 1694 1695 /* 1696 * Extent to map - this can be after start_pos because that can be 1697 * fully mapped. We somewhat abuse m_flags to store whether the extent 1698 * is delalloc or unwritten. 1699 */ 1700 struct ext4_map_blocks map; 1701 struct ext4_io_submit io_submit; /* IO submission data */ 1702 unsigned int do_map:1; 1703 unsigned int scanned_until_end:1; 1704 unsigned int journalled_more_data:1; 1705 }; 1706 1707 static void mpage_release_unused_pages(struct mpage_da_data *mpd, 1708 bool invalidate) 1709 { 1710 unsigned nr, i; 1711 pgoff_t index, end; 1712 struct folio_batch fbatch; 1713 struct inode *inode = mpd->inode; 1714 struct address_space *mapping = inode->i_mapping; 1715 1716 /* This is necessary when next_pos == 0. */ 1717 if (mpd->start_pos >= mpd->next_pos) 1718 return; 1719 1720 mpd->scanned_until_end = 0; 1721 if (invalidate) { 1722 ext4_lblk_t start, last; 1723 start = EXT4_B_TO_LBLK(inode, mpd->start_pos); 1724 last = mpd->next_pos >> inode->i_blkbits; 1725 1726 /* 1727 * avoid racing with extent status tree scans made by 1728 * ext4_insert_delayed_block() 1729 */ 1730 down_write(&EXT4_I(inode)->i_data_sem); 1731 ext4_es_remove_extent(inode, start, last - start); 1732 up_write(&EXT4_I(inode)->i_data_sem); 1733 } 1734 1735 folio_batch_init(&fbatch); 1736 index = mpd->start_pos >> PAGE_SHIFT; 1737 end = mpd->next_pos >> PAGE_SHIFT; 1738 while (index < end) { 1739 nr = filemap_get_folios(mapping, &index, end - 1, &fbatch); 1740 if (nr == 0) 1741 break; 1742 for (i = 0; i < nr; i++) { 1743 struct folio *folio = fbatch.folios[i]; 1744 1745 if (folio_pos(folio) < mpd->start_pos) 1746 continue; 1747 if (folio_next_index(folio) > end) 1748 continue; 1749 BUG_ON(!folio_test_locked(folio)); 1750 BUG_ON(folio_test_writeback(folio)); 1751 if (invalidate) { 1752 if (folio_mapped(folio)) 1753 folio_clear_dirty_for_io(folio); 1754 block_invalidate_folio(folio, 0, 1755 folio_size(folio)); 1756 folio_clear_uptodate(folio); 1757 } 1758 folio_unlock(folio); 1759 } 1760 folio_batch_release(&fbatch); 1761 } 1762 } 1763 1764 static void ext4_print_free_blocks(struct inode *inode) 1765 { 1766 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 1767 struct super_block *sb = inode->i_sb; 1768 struct ext4_inode_info *ei = EXT4_I(inode); 1769 1770 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld", 1771 EXT4_C2B(EXT4_SB(inode->i_sb), 1772 ext4_count_free_clusters(sb))); 1773 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details"); 1774 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld", 1775 (long long) EXT4_C2B(EXT4_SB(sb), 1776 percpu_counter_sum(&sbi->s_freeclusters_counter))); 1777 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld", 1778 (long long) EXT4_C2B(EXT4_SB(sb), 1779 percpu_counter_sum(&sbi->s_dirtyclusters_counter))); 1780 ext4_msg(sb, KERN_CRIT, "Block reservation details"); 1781 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u", 1782 ei->i_reserved_data_blocks); 1783 return; 1784 } 1785 1786 /* 1787 * Check whether the cluster containing lblk has been allocated or has 1788 * delalloc reservation. 1789 * 1790 * Returns 0 if the cluster doesn't have either, 1 if it has delalloc 1791 * reservation, 2 if it's already been allocated, negative error code on 1792 * failure. 1793 */ 1794 static int ext4_clu_alloc_state(struct inode *inode, ext4_lblk_t lblk) 1795 { 1796 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 1797 int ret; 1798 1799 /* Has delalloc reservation? */ 1800 if (ext4_es_scan_clu(inode, &ext4_es_is_delayed, lblk)) 1801 return 1; 1802 1803 /* Already been allocated? */ 1804 if (ext4_es_scan_clu(inode, &ext4_es_is_mapped, lblk)) 1805 return 2; 1806 ret = ext4_clu_mapped(inode, EXT4_B2C(sbi, lblk)); 1807 if (ret < 0) 1808 return ret; 1809 if (ret > 0) 1810 return 2; 1811 1812 return 0; 1813 } 1814 1815 /* 1816 * ext4_insert_delayed_blocks - adds a multiple delayed blocks to the extents 1817 * status tree, incrementing the reserved 1818 * cluster/block count or making pending 1819 * reservations where needed 1820 * 1821 * @inode - file containing the newly added block 1822 * @lblk - start logical block to be added 1823 * @len - length of blocks to be added 1824 * 1825 * Returns 0 on success, negative error code on failure. 1826 */ 1827 static int ext4_insert_delayed_blocks(struct inode *inode, ext4_lblk_t lblk, 1828 ext4_lblk_t len) 1829 { 1830 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 1831 int ret; 1832 bool lclu_allocated = false; 1833 bool end_allocated = false; 1834 ext4_lblk_t resv_clu; 1835 ext4_lblk_t end = lblk + len - 1; 1836 1837 /* 1838 * If the cluster containing lblk or end is shared with a delayed, 1839 * written, or unwritten extent in a bigalloc file system, it's 1840 * already been accounted for and does not need to be reserved. 1841 * A pending reservation must be made for the cluster if it's 1842 * shared with a written or unwritten extent and doesn't already 1843 * have one. Written and unwritten extents can be purged from the 1844 * extents status tree if the system is under memory pressure, so 1845 * it's necessary to examine the extent tree if a search of the 1846 * extents status tree doesn't get a match. 1847 */ 1848 if (sbi->s_cluster_ratio == 1) { 1849 ret = ext4_da_reserve_space(inode, len); 1850 if (ret != 0) /* ENOSPC */ 1851 return ret; 1852 } else { /* bigalloc */ 1853 resv_clu = EXT4_B2C(sbi, end) - EXT4_B2C(sbi, lblk) + 1; 1854 1855 ret = ext4_clu_alloc_state(inode, lblk); 1856 if (ret < 0) 1857 return ret; 1858 if (ret > 0) { 1859 resv_clu--; 1860 lclu_allocated = (ret == 2); 1861 } 1862 1863 if (EXT4_B2C(sbi, lblk) != EXT4_B2C(sbi, end)) { 1864 ret = ext4_clu_alloc_state(inode, end); 1865 if (ret < 0) 1866 return ret; 1867 if (ret > 0) { 1868 resv_clu--; 1869 end_allocated = (ret == 2); 1870 } 1871 } 1872 1873 if (resv_clu) { 1874 ret = ext4_da_reserve_space(inode, resv_clu); 1875 if (ret != 0) /* ENOSPC */ 1876 return ret; 1877 } 1878 } 1879 1880 ext4_es_insert_delayed_extent(inode, lblk, len, lclu_allocated, 1881 end_allocated); 1882 return 0; 1883 } 1884 1885 /* 1886 * Looks up the requested blocks and sets the delalloc extent map. 1887 * First try to look up for the extent entry that contains the requested 1888 * blocks in the extent status tree without i_data_sem, then try to look 1889 * up for the ondisk extent mapping with i_data_sem in read mode, 1890 * finally hold i_data_sem in write mode, looks up again and add a 1891 * delalloc extent entry if it still couldn't find any extent. Pass out 1892 * the mapped extent through @map and return 0 on success. 1893 */ 1894 static int ext4_da_map_blocks(struct inode *inode, struct ext4_map_blocks *map) 1895 { 1896 struct extent_status es; 1897 int retval; 1898 #ifdef ES_AGGRESSIVE_TEST 1899 struct ext4_map_blocks orig_map; 1900 1901 memcpy(&orig_map, map, sizeof(*map)); 1902 #endif 1903 1904 map->m_flags = 0; 1905 ext_debug(inode, "max_blocks %u, logical block %lu\n", map->m_len, 1906 (unsigned long) map->m_lblk); 1907 1908 ext4_check_map_extents_env(inode); 1909 1910 /* Lookup extent status tree firstly */ 1911 if (ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) { 1912 map->m_len = min_t(unsigned int, map->m_len, 1913 es.es_len - (map->m_lblk - es.es_lblk)); 1914 1915 if (ext4_es_is_hole(&es)) 1916 goto add_delayed; 1917 1918 found: 1919 /* 1920 * Delayed extent could be allocated by fallocate. 1921 * So we need to check it. 1922 */ 1923 if (ext4_es_is_delayed(&es)) { 1924 map->m_flags |= EXT4_MAP_DELAYED; 1925 return 0; 1926 } 1927 1928 map->m_pblk = ext4_es_pblock(&es) + map->m_lblk - es.es_lblk; 1929 if (ext4_es_is_written(&es)) 1930 map->m_flags |= EXT4_MAP_MAPPED; 1931 else if (ext4_es_is_unwritten(&es)) 1932 map->m_flags |= EXT4_MAP_UNWRITTEN; 1933 else 1934 BUG(); 1935 1936 #ifdef ES_AGGRESSIVE_TEST 1937 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0); 1938 #endif 1939 return 0; 1940 } 1941 1942 /* 1943 * Try to see if we can get the block without requesting a new 1944 * file system block. 1945 */ 1946 down_read(&EXT4_I(inode)->i_data_sem); 1947 if (ext4_has_inline_data(inode)) 1948 retval = 0; 1949 else 1950 retval = ext4_map_query_blocks(NULL, inode, map, 0); 1951 up_read(&EXT4_I(inode)->i_data_sem); 1952 if (retval) 1953 return retval < 0 ? retval : 0; 1954 1955 add_delayed: 1956 down_write(&EXT4_I(inode)->i_data_sem); 1957 /* 1958 * Page fault path (ext4_page_mkwrite does not take i_rwsem) 1959 * and fallocate path (no folio lock) can race. Make sure we 1960 * lookup the extent status tree here again while i_data_sem 1961 * is held in write mode, before inserting a new da entry in 1962 * the extent status tree. 1963 */ 1964 if (ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) { 1965 map->m_len = min_t(unsigned int, map->m_len, 1966 es.es_len - (map->m_lblk - es.es_lblk)); 1967 1968 if (!ext4_es_is_hole(&es)) { 1969 up_write(&EXT4_I(inode)->i_data_sem); 1970 goto found; 1971 } 1972 } else if (!ext4_has_inline_data(inode)) { 1973 retval = ext4_map_query_blocks(NULL, inode, map, 0); 1974 if (retval) { 1975 up_write(&EXT4_I(inode)->i_data_sem); 1976 return retval < 0 ? retval : 0; 1977 } 1978 } 1979 1980 map->m_flags |= EXT4_MAP_DELAYED; 1981 retval = ext4_insert_delayed_blocks(inode, map->m_lblk, map->m_len); 1982 up_write(&EXT4_I(inode)->i_data_sem); 1983 1984 return retval; 1985 } 1986 1987 /* 1988 * This is a special get_block_t callback which is used by 1989 * ext4_da_write_begin(). It will either return mapped block or 1990 * reserve space for a single block. 1991 * 1992 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. 1993 * We also have b_blocknr = -1 and b_bdev initialized properly 1994 * 1995 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. 1996 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev 1997 * initialized properly. 1998 */ 1999 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, 2000 struct buffer_head *bh, int create) 2001 { 2002 struct ext4_map_blocks map; 2003 sector_t invalid_block = ~((sector_t) 0xffff); 2004 int ret = 0; 2005 2006 BUG_ON(create == 0); 2007 BUG_ON(bh->b_size != inode->i_sb->s_blocksize); 2008 2009 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) 2010 invalid_block = ~0; 2011 2012 map.m_lblk = iblock; 2013 map.m_len = 1; 2014 2015 /* 2016 * first, we need to know whether the block is allocated already 2017 * preallocated blocks are unmapped but should treated 2018 * the same as allocated blocks. 2019 */ 2020 ret = ext4_da_map_blocks(inode, &map); 2021 if (ret < 0) 2022 return ret; 2023 2024 if (map.m_flags & EXT4_MAP_DELAYED) { 2025 map_bh(bh, inode->i_sb, invalid_block); 2026 set_buffer_new(bh); 2027 set_buffer_delay(bh); 2028 return 0; 2029 } 2030 2031 map_bh(bh, inode->i_sb, map.m_pblk); 2032 ext4_update_bh_state(bh, map.m_flags); 2033 2034 if (buffer_unwritten(bh)) { 2035 /* A delayed write to unwritten bh should be marked 2036 * new and mapped. Mapped ensures that we don't do 2037 * get_block multiple times when we write to the same 2038 * offset and new ensures that we do proper zero out 2039 * for partial write. 2040 */ 2041 set_buffer_new(bh); 2042 set_buffer_mapped(bh); 2043 } 2044 return 0; 2045 } 2046 2047 static void mpage_folio_done(struct mpage_da_data *mpd, struct folio *folio) 2048 { 2049 mpd->start_pos += folio_size(folio); 2050 mpd->wbc->nr_to_write -= folio_nr_pages(folio); 2051 folio_unlock(folio); 2052 } 2053 2054 static int mpage_submit_folio(struct mpage_da_data *mpd, struct folio *folio) 2055 { 2056 size_t len; 2057 loff_t size; 2058 int err; 2059 2060 WARN_ON_ONCE(folio_pos(folio) != mpd->start_pos); 2061 folio_clear_dirty_for_io(folio); 2062 /* 2063 * We have to be very careful here! Nothing protects writeback path 2064 * against i_size changes and the page can be writeably mapped into 2065 * page tables. So an application can be growing i_size and writing 2066 * data through mmap while writeback runs. folio_clear_dirty_for_io() 2067 * write-protects our page in page tables and the page cannot get 2068 * written to again until we release folio lock. So only after 2069 * folio_clear_dirty_for_io() we are safe to sample i_size for 2070 * ext4_bio_write_folio() to zero-out tail of the written page. We rely 2071 * on the barrier provided by folio_test_clear_dirty() in 2072 * folio_clear_dirty_for_io() to make sure i_size is really sampled only 2073 * after page tables are updated. 2074 */ 2075 size = i_size_read(mpd->inode); 2076 len = folio_size(folio); 2077 if (folio_pos(folio) + len > size && 2078 !ext4_verity_in_progress(mpd->inode)) 2079 len = size & (len - 1); 2080 err = ext4_bio_write_folio(&mpd->io_submit, folio, len); 2081 2082 return err; 2083 } 2084 2085 #define BH_FLAGS (BIT(BH_Unwritten) | BIT(BH_Delay)) 2086 2087 /* 2088 * mballoc gives us at most this number of blocks... 2089 * XXX: That seems to be only a limitation of ext4_mb_normalize_request(). 2090 * The rest of mballoc seems to handle chunks up to full group size. 2091 */ 2092 #define MAX_WRITEPAGES_EXTENT_LEN 2048 2093 2094 /* 2095 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map 2096 * 2097 * @mpd - extent of blocks 2098 * @lblk - logical number of the block in the file 2099 * @bh - buffer head we want to add to the extent 2100 * 2101 * The function is used to collect contig. blocks in the same state. If the 2102 * buffer doesn't require mapping for writeback and we haven't started the 2103 * extent of buffers to map yet, the function returns 'true' immediately - the 2104 * caller can write the buffer right away. Otherwise the function returns true 2105 * if the block has been added to the extent, false if the block couldn't be 2106 * added. 2107 */ 2108 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk, 2109 struct buffer_head *bh) 2110 { 2111 struct ext4_map_blocks *map = &mpd->map; 2112 2113 /* Buffer that doesn't need mapping for writeback? */ 2114 if (!buffer_dirty(bh) || !buffer_mapped(bh) || 2115 (!buffer_delay(bh) && !buffer_unwritten(bh))) { 2116 /* So far no extent to map => we write the buffer right away */ 2117 if (map->m_len == 0) 2118 return true; 2119 return false; 2120 } 2121 2122 /* First block in the extent? */ 2123 if (map->m_len == 0) { 2124 /* We cannot map unless handle is started... */ 2125 if (!mpd->do_map) 2126 return false; 2127 map->m_lblk = lblk; 2128 map->m_len = 1; 2129 map->m_flags = bh->b_state & BH_FLAGS; 2130 return true; 2131 } 2132 2133 /* Don't go larger than mballoc is willing to allocate */ 2134 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN) 2135 return false; 2136 2137 /* Can we merge the block to our big extent? */ 2138 if (lblk == map->m_lblk + map->m_len && 2139 (bh->b_state & BH_FLAGS) == map->m_flags) { 2140 map->m_len++; 2141 return true; 2142 } 2143 return false; 2144 } 2145 2146 /* 2147 * mpage_process_page_bufs - submit page buffers for IO or add them to extent 2148 * 2149 * @mpd - extent of blocks for mapping 2150 * @head - the first buffer in the page 2151 * @bh - buffer we should start processing from 2152 * @lblk - logical number of the block in the file corresponding to @bh 2153 * 2154 * Walk through page buffers from @bh upto @head (exclusive) and either submit 2155 * the page for IO if all buffers in this page were mapped and there's no 2156 * accumulated extent of buffers to map or add buffers in the page to the 2157 * extent of buffers to map. The function returns 1 if the caller can continue 2158 * by processing the next page, 0 if it should stop adding buffers to the 2159 * extent to map because we cannot extend it anymore. It can also return value 2160 * < 0 in case of error during IO submission. 2161 */ 2162 static int mpage_process_page_bufs(struct mpage_da_data *mpd, 2163 struct buffer_head *head, 2164 struct buffer_head *bh, 2165 ext4_lblk_t lblk) 2166 { 2167 struct inode *inode = mpd->inode; 2168 int err; 2169 ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1) 2170 >> inode->i_blkbits; 2171 2172 if (ext4_verity_in_progress(inode)) 2173 blocks = EXT_MAX_BLOCKS; 2174 2175 do { 2176 BUG_ON(buffer_locked(bh)); 2177 2178 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) { 2179 /* Found extent to map? */ 2180 if (mpd->map.m_len) 2181 return 0; 2182 /* Buffer needs mapping and handle is not started? */ 2183 if (!mpd->do_map) 2184 return 0; 2185 /* Everything mapped so far and we hit EOF */ 2186 break; 2187 } 2188 } while (lblk++, (bh = bh->b_this_page) != head); 2189 /* So far everything mapped? Submit the page for IO. */ 2190 if (mpd->map.m_len == 0) { 2191 err = mpage_submit_folio(mpd, head->b_folio); 2192 if (err < 0) 2193 return err; 2194 mpage_folio_done(mpd, head->b_folio); 2195 } 2196 if (lblk >= blocks) { 2197 mpd->scanned_until_end = 1; 2198 return 0; 2199 } 2200 return 1; 2201 } 2202 2203 /* 2204 * mpage_process_folio - update folio buffers corresponding to changed extent 2205 * and may submit fully mapped page for IO 2206 * @mpd: description of extent to map, on return next extent to map 2207 * @folio: Contains these buffers. 2208 * @m_lblk: logical block mapping. 2209 * @m_pblk: corresponding physical mapping. 2210 * @map_bh: determines on return whether this page requires any further 2211 * mapping or not. 2212 * 2213 * Scan given folio buffers corresponding to changed extent and update buffer 2214 * state according to new extent state. 2215 * We map delalloc buffers to their physical location, clear unwritten bits. 2216 * If the given folio is not fully mapped, we update @mpd to the next extent in 2217 * the given folio that needs mapping & return @map_bh as true. 2218 */ 2219 static int mpage_process_folio(struct mpage_da_data *mpd, struct folio *folio, 2220 ext4_lblk_t *m_lblk, ext4_fsblk_t *m_pblk, 2221 bool *map_bh) 2222 { 2223 struct buffer_head *head, *bh; 2224 ext4_io_end_t *io_end = mpd->io_submit.io_end; 2225 ext4_lblk_t lblk = *m_lblk; 2226 ext4_fsblk_t pblock = *m_pblk; 2227 int err = 0; 2228 int blkbits = mpd->inode->i_blkbits; 2229 ssize_t io_end_size = 0; 2230 struct ext4_io_end_vec *io_end_vec = ext4_last_io_end_vec(io_end); 2231 2232 bh = head = folio_buffers(folio); 2233 do { 2234 if (lblk < mpd->map.m_lblk) 2235 continue; 2236 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) { 2237 /* 2238 * Buffer after end of mapped extent. 2239 * Find next buffer in the folio to map. 2240 */ 2241 mpd->map.m_len = 0; 2242 mpd->map.m_flags = 0; 2243 io_end_vec->size += io_end_size; 2244 2245 err = mpage_process_page_bufs(mpd, head, bh, lblk); 2246 if (err > 0) 2247 err = 0; 2248 if (!err && mpd->map.m_len && mpd->map.m_lblk > lblk) { 2249 io_end_vec = ext4_alloc_io_end_vec(io_end); 2250 if (IS_ERR(io_end_vec)) { 2251 err = PTR_ERR(io_end_vec); 2252 goto out; 2253 } 2254 io_end_vec->offset = (loff_t)mpd->map.m_lblk << blkbits; 2255 } 2256 *map_bh = true; 2257 goto out; 2258 } 2259 if (buffer_delay(bh)) { 2260 clear_buffer_delay(bh); 2261 bh->b_blocknr = pblock++; 2262 } 2263 clear_buffer_unwritten(bh); 2264 io_end_size += (1 << blkbits); 2265 } while (lblk++, (bh = bh->b_this_page) != head); 2266 2267 io_end_vec->size += io_end_size; 2268 *map_bh = false; 2269 out: 2270 *m_lblk = lblk; 2271 *m_pblk = pblock; 2272 return err; 2273 } 2274 2275 /* 2276 * mpage_map_buffers - update buffers corresponding to changed extent and 2277 * submit fully mapped pages for IO 2278 * 2279 * @mpd - description of extent to map, on return next extent to map 2280 * 2281 * Scan buffers corresponding to changed extent (we expect corresponding pages 2282 * to be already locked) and update buffer state according to new extent state. 2283 * We map delalloc buffers to their physical location, clear unwritten bits, 2284 * and mark buffers as uninit when we perform writes to unwritten extents 2285 * and do extent conversion after IO is finished. If the last page is not fully 2286 * mapped, we update @map to the next extent in the last page that needs 2287 * mapping. Otherwise we submit the page for IO. 2288 */ 2289 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd) 2290 { 2291 struct folio_batch fbatch; 2292 unsigned nr, i; 2293 struct inode *inode = mpd->inode; 2294 int bpp_bits = PAGE_SHIFT - inode->i_blkbits; 2295 pgoff_t start, end; 2296 ext4_lblk_t lblk; 2297 ext4_fsblk_t pblock; 2298 int err; 2299 bool map_bh = false; 2300 2301 start = mpd->map.m_lblk >> bpp_bits; 2302 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits; 2303 pblock = mpd->map.m_pblk; 2304 2305 folio_batch_init(&fbatch); 2306 while (start <= end) { 2307 nr = filemap_get_folios(inode->i_mapping, &start, end, &fbatch); 2308 if (nr == 0) 2309 break; 2310 for (i = 0; i < nr; i++) { 2311 struct folio *folio = fbatch.folios[i]; 2312 2313 lblk = folio->index << bpp_bits; 2314 err = mpage_process_folio(mpd, folio, &lblk, &pblock, 2315 &map_bh); 2316 /* 2317 * If map_bh is true, means page may require further bh 2318 * mapping, or maybe the page was submitted for IO. 2319 * So we return to call further extent mapping. 2320 */ 2321 if (err < 0 || map_bh) 2322 goto out; 2323 /* Page fully mapped - let IO run! */ 2324 err = mpage_submit_folio(mpd, folio); 2325 if (err < 0) 2326 goto out; 2327 mpage_folio_done(mpd, folio); 2328 } 2329 folio_batch_release(&fbatch); 2330 } 2331 /* Extent fully mapped and matches with page boundary. We are done. */ 2332 mpd->map.m_len = 0; 2333 mpd->map.m_flags = 0; 2334 return 0; 2335 out: 2336 folio_batch_release(&fbatch); 2337 return err; 2338 } 2339 2340 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd) 2341 { 2342 struct inode *inode = mpd->inode; 2343 struct ext4_map_blocks *map = &mpd->map; 2344 int get_blocks_flags; 2345 int err, dioread_nolock; 2346 2347 /* Make sure transaction has enough credits for this extent */ 2348 err = ext4_journal_ensure_extent_credits(handle, inode); 2349 if (err < 0) 2350 return err; 2351 2352 trace_ext4_da_write_pages_extent(inode, map); 2353 /* 2354 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or 2355 * to convert an unwritten extent to be initialized (in the case 2356 * where we have written into one or more preallocated blocks). It is 2357 * possible that we're going to need more metadata blocks than 2358 * previously reserved. However we must not fail because we're in 2359 * writeback and there is nothing we can do about it so it might result 2360 * in data loss. So use reserved blocks to allocate metadata if 2361 * possible. In addition, do not cache any unrelated extents, as it 2362 * only holds the folio lock but does not hold the i_rwsem or 2363 * invalidate_lock, which could corrupt the extent status tree. 2364 */ 2365 get_blocks_flags = EXT4_GET_BLOCKS_CREATE | 2366 EXT4_GET_BLOCKS_METADATA_NOFAIL | 2367 EXT4_GET_BLOCKS_IO_SUBMIT | 2368 EXT4_EX_NOCACHE; 2369 2370 dioread_nolock = ext4_should_dioread_nolock(inode); 2371 if (dioread_nolock) 2372 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; 2373 2374 err = ext4_map_blocks(handle, inode, map, get_blocks_flags); 2375 if (err < 0) 2376 return err; 2377 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) { 2378 if (!mpd->io_submit.io_end->handle && 2379 ext4_handle_valid(handle)) { 2380 mpd->io_submit.io_end->handle = handle->h_rsv_handle; 2381 handle->h_rsv_handle = NULL; 2382 } 2383 ext4_set_io_unwritten_flag(mpd->io_submit.io_end); 2384 } 2385 2386 BUG_ON(map->m_len == 0); 2387 return 0; 2388 } 2389 2390 /* 2391 * This is used to submit mapped buffers in a single folio that is not fully 2392 * mapped for various reasons, such as insufficient space or journal credits. 2393 */ 2394 static int mpage_submit_partial_folio(struct mpage_da_data *mpd) 2395 { 2396 struct inode *inode = mpd->inode; 2397 struct folio *folio; 2398 loff_t pos; 2399 int ret; 2400 2401 folio = filemap_get_folio(inode->i_mapping, 2402 mpd->start_pos >> PAGE_SHIFT); 2403 if (IS_ERR(folio)) 2404 return PTR_ERR(folio); 2405 /* 2406 * The mapped position should be within the current processing folio 2407 * but must not be the folio start position. 2408 */ 2409 pos = ((loff_t)mpd->map.m_lblk) << inode->i_blkbits; 2410 if (WARN_ON_ONCE((folio_pos(folio) == pos) || 2411 !folio_contains(folio, pos >> PAGE_SHIFT))) 2412 return -EINVAL; 2413 2414 ret = mpage_submit_folio(mpd, folio); 2415 if (ret) 2416 goto out; 2417 /* 2418 * Update start_pos to prevent this folio from being released in 2419 * mpage_release_unused_pages(), it will be reset to the aligned folio 2420 * pos when this folio is written again in the next round. Additionally, 2421 * do not update wbc->nr_to_write here, as it will be updated once the 2422 * entire folio has finished processing. 2423 */ 2424 mpd->start_pos = pos; 2425 out: 2426 folio_unlock(folio); 2427 folio_put(folio); 2428 return ret; 2429 } 2430 2431 /* 2432 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length 2433 * mpd->len and submit pages underlying it for IO 2434 * 2435 * @handle - handle for journal operations 2436 * @mpd - extent to map 2437 * @give_up_on_write - we set this to true iff there is a fatal error and there 2438 * is no hope of writing the data. The caller should discard 2439 * dirty pages to avoid infinite loops. 2440 * 2441 * The function maps extent starting at mpd->lblk of length mpd->len. If it is 2442 * delayed, blocks are allocated, if it is unwritten, we may need to convert 2443 * them to initialized or split the described range from larger unwritten 2444 * extent. Note that we need not map all the described range since allocation 2445 * can return less blocks or the range is covered by more unwritten extents. We 2446 * cannot map more because we are limited by reserved transaction credits. On 2447 * the other hand we always make sure that the last touched page is fully 2448 * mapped so that it can be written out (and thus forward progress is 2449 * guaranteed). After mapping we submit all mapped pages for IO. 2450 */ 2451 static int mpage_map_and_submit_extent(handle_t *handle, 2452 struct mpage_da_data *mpd, 2453 bool *give_up_on_write) 2454 { 2455 struct inode *inode = mpd->inode; 2456 struct ext4_map_blocks *map = &mpd->map; 2457 int err; 2458 loff_t disksize; 2459 int progress = 0; 2460 ext4_io_end_t *io_end = mpd->io_submit.io_end; 2461 struct ext4_io_end_vec *io_end_vec; 2462 2463 io_end_vec = ext4_alloc_io_end_vec(io_end); 2464 if (IS_ERR(io_end_vec)) 2465 return PTR_ERR(io_end_vec); 2466 io_end_vec->offset = ((loff_t)map->m_lblk) << inode->i_blkbits; 2467 do { 2468 err = mpage_map_one_extent(handle, mpd); 2469 if (err < 0) { 2470 struct super_block *sb = inode->i_sb; 2471 2472 if (ext4_emergency_state(sb)) 2473 goto invalidate_dirty_pages; 2474 /* 2475 * Let the uper layers retry transient errors. 2476 * In the case of ENOSPC, if ext4_count_free_blocks() 2477 * is non-zero, a commit should free up blocks. 2478 */ 2479 if ((err == -ENOMEM) || (err == -EAGAIN) || 2480 (err == -ENOSPC && ext4_count_free_clusters(sb))) { 2481 /* 2482 * We may have already allocated extents for 2483 * some bhs inside the folio, issue the 2484 * corresponding data to prevent stale data. 2485 */ 2486 if (progress) { 2487 if (mpage_submit_partial_folio(mpd)) 2488 goto invalidate_dirty_pages; 2489 goto update_disksize; 2490 } 2491 return err; 2492 } 2493 ext4_msg(sb, KERN_CRIT, 2494 "Delayed block allocation failed for " 2495 "inode %lu at logical offset %llu with" 2496 " max blocks %u with error %d", 2497 inode->i_ino, 2498 (unsigned long long)map->m_lblk, 2499 (unsigned)map->m_len, -err); 2500 ext4_msg(sb, KERN_CRIT, 2501 "This should not happen!! Data will " 2502 "be lost\n"); 2503 if (err == -ENOSPC) 2504 ext4_print_free_blocks(inode); 2505 invalidate_dirty_pages: 2506 *give_up_on_write = true; 2507 return err; 2508 } 2509 progress = 1; 2510 /* 2511 * Update buffer state, submit mapped pages, and get us new 2512 * extent to map 2513 */ 2514 err = mpage_map_and_submit_buffers(mpd); 2515 if (err < 0) 2516 goto update_disksize; 2517 } while (map->m_len); 2518 2519 update_disksize: 2520 /* 2521 * Update on-disk size after IO is submitted. Races with 2522 * truncate are avoided by checking i_size under i_data_sem. 2523 */ 2524 disksize = mpd->start_pos; 2525 if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) { 2526 int err2; 2527 loff_t i_size; 2528 2529 down_write(&EXT4_I(inode)->i_data_sem); 2530 i_size = i_size_read(inode); 2531 if (disksize > i_size) 2532 disksize = i_size; 2533 if (disksize > EXT4_I(inode)->i_disksize) 2534 EXT4_I(inode)->i_disksize = disksize; 2535 up_write(&EXT4_I(inode)->i_data_sem); 2536 err2 = ext4_mark_inode_dirty(handle, inode); 2537 if (err2) { 2538 ext4_error_err(inode->i_sb, -err2, 2539 "Failed to mark inode %lu dirty", 2540 inode->i_ino); 2541 } 2542 if (!err) 2543 err = err2; 2544 } 2545 return err; 2546 } 2547 2548 static int ext4_journal_folio_buffers(handle_t *handle, struct folio *folio, 2549 size_t len) 2550 { 2551 struct buffer_head *page_bufs = folio_buffers(folio); 2552 struct inode *inode = folio->mapping->host; 2553 int ret, err; 2554 2555 ret = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len, 2556 NULL, do_journal_get_write_access); 2557 err = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len, 2558 NULL, write_end_fn); 2559 if (ret == 0) 2560 ret = err; 2561 err = ext4_jbd2_inode_add_write(handle, inode, folio_pos(folio), len); 2562 if (ret == 0) 2563 ret = err; 2564 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; 2565 2566 return ret; 2567 } 2568 2569 static int mpage_journal_page_buffers(handle_t *handle, 2570 struct mpage_da_data *mpd, 2571 struct folio *folio) 2572 { 2573 struct inode *inode = mpd->inode; 2574 loff_t size = i_size_read(inode); 2575 size_t len = folio_size(folio); 2576 2577 folio_clear_checked(folio); 2578 mpd->wbc->nr_to_write -= folio_nr_pages(folio); 2579 2580 if (folio_pos(folio) + len > size && 2581 !ext4_verity_in_progress(inode)) 2582 len = size & (len - 1); 2583 2584 return ext4_journal_folio_buffers(handle, folio, len); 2585 } 2586 2587 /* 2588 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages 2589 * needing mapping, submit mapped pages 2590 * 2591 * @mpd - where to look for pages 2592 * 2593 * Walk dirty pages in the mapping. If they are fully mapped, submit them for 2594 * IO immediately. If we cannot map blocks, we submit just already mapped 2595 * buffers in the page for IO and keep page dirty. When we can map blocks and 2596 * we find a page which isn't mapped we start accumulating extent of buffers 2597 * underlying these pages that needs mapping (formed by either delayed or 2598 * unwritten buffers). We also lock the pages containing these buffers. The 2599 * extent found is returned in @mpd structure (starting at mpd->lblk with 2600 * length mpd->len blocks). 2601 * 2602 * Note that this function can attach bios to one io_end structure which are 2603 * neither logically nor physically contiguous. Although it may seem as an 2604 * unnecessary complication, it is actually inevitable in blocksize < pagesize 2605 * case as we need to track IO to all buffers underlying a page in one io_end. 2606 */ 2607 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd) 2608 { 2609 struct address_space *mapping = mpd->inode->i_mapping; 2610 struct folio_batch fbatch; 2611 unsigned int nr_folios; 2612 pgoff_t index = mpd->start_pos >> PAGE_SHIFT; 2613 pgoff_t end = mpd->end_pos >> PAGE_SHIFT; 2614 xa_mark_t tag; 2615 int i, err = 0; 2616 int blkbits = mpd->inode->i_blkbits; 2617 ext4_lblk_t lblk; 2618 struct buffer_head *head; 2619 handle_t *handle = NULL; 2620 int bpp = ext4_journal_blocks_per_folio(mpd->inode); 2621 2622 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages) 2623 tag = PAGECACHE_TAG_TOWRITE; 2624 else 2625 tag = PAGECACHE_TAG_DIRTY; 2626 2627 mpd->map.m_len = 0; 2628 mpd->next_pos = mpd->start_pos; 2629 if (ext4_should_journal_data(mpd->inode)) { 2630 handle = ext4_journal_start(mpd->inode, EXT4_HT_WRITE_PAGE, 2631 bpp); 2632 if (IS_ERR(handle)) 2633 return PTR_ERR(handle); 2634 } 2635 folio_batch_init(&fbatch); 2636 while (index <= end) { 2637 nr_folios = filemap_get_folios_tag(mapping, &index, end, 2638 tag, &fbatch); 2639 if (nr_folios == 0) 2640 break; 2641 2642 for (i = 0; i < nr_folios; i++) { 2643 struct folio *folio = fbatch.folios[i]; 2644 2645 /* 2646 * Accumulated enough dirty pages? This doesn't apply 2647 * to WB_SYNC_ALL mode. For integrity sync we have to 2648 * keep going because someone may be concurrently 2649 * dirtying pages, and we might have synced a lot of 2650 * newly appeared dirty pages, but have not synced all 2651 * of the old dirty pages. 2652 */ 2653 if (mpd->wbc->sync_mode == WB_SYNC_NONE && 2654 mpd->wbc->nr_to_write <= 2655 mpd->map.m_len >> (PAGE_SHIFT - blkbits)) 2656 goto out; 2657 2658 /* If we can't merge this page, we are done. */ 2659 if (mpd->map.m_len > 0 && 2660 mpd->next_pos != folio_pos(folio)) 2661 goto out; 2662 2663 if (handle) { 2664 err = ext4_journal_ensure_credits(handle, bpp, 2665 0); 2666 if (err < 0) 2667 goto out; 2668 } 2669 2670 folio_lock(folio); 2671 /* 2672 * If the page is no longer dirty, or its mapping no 2673 * longer corresponds to inode we are writing (which 2674 * means it has been truncated or invalidated), or the 2675 * page is already under writeback and we are not doing 2676 * a data integrity writeback, skip the page 2677 */ 2678 if (!folio_test_dirty(folio) || 2679 (folio_test_writeback(folio) && 2680 (mpd->wbc->sync_mode == WB_SYNC_NONE)) || 2681 unlikely(folio->mapping != mapping)) { 2682 folio_unlock(folio); 2683 continue; 2684 } 2685 2686 folio_wait_writeback(folio); 2687 BUG_ON(folio_test_writeback(folio)); 2688 2689 /* 2690 * Should never happen but for buggy code in 2691 * other subsystems that call 2692 * set_page_dirty() without properly warning 2693 * the file system first. See [1] for more 2694 * information. 2695 * 2696 * [1] https://lore.kernel.org/linux-mm/20180103100430.GE4911@quack2.suse.cz 2697 */ 2698 if (!folio_buffers(folio)) { 2699 ext4_warning_inode(mpd->inode, "page %lu does not have buffers attached", folio->index); 2700 folio_clear_dirty(folio); 2701 folio_unlock(folio); 2702 continue; 2703 } 2704 2705 if (mpd->map.m_len == 0) 2706 mpd->start_pos = folio_pos(folio); 2707 mpd->next_pos = folio_pos(folio) + folio_size(folio); 2708 /* 2709 * Writeout when we cannot modify metadata is simple. 2710 * Just submit the page. For data=journal mode we 2711 * first handle writeout of the page for checkpoint and 2712 * only after that handle delayed page dirtying. This 2713 * makes sure current data is checkpointed to the final 2714 * location before possibly journalling it again which 2715 * is desirable when the page is frequently dirtied 2716 * through a pin. 2717 */ 2718 if (!mpd->can_map) { 2719 err = mpage_submit_folio(mpd, folio); 2720 if (err < 0) 2721 goto out; 2722 /* Pending dirtying of journalled data? */ 2723 if (folio_test_checked(folio)) { 2724 err = mpage_journal_page_buffers(handle, 2725 mpd, folio); 2726 if (err < 0) 2727 goto out; 2728 mpd->journalled_more_data = 1; 2729 } 2730 mpage_folio_done(mpd, folio); 2731 } else { 2732 /* Add all dirty buffers to mpd */ 2733 lblk = ((ext4_lblk_t)folio->index) << 2734 (PAGE_SHIFT - blkbits); 2735 head = folio_buffers(folio); 2736 err = mpage_process_page_bufs(mpd, head, head, 2737 lblk); 2738 if (err <= 0) 2739 goto out; 2740 err = 0; 2741 } 2742 } 2743 folio_batch_release(&fbatch); 2744 cond_resched(); 2745 } 2746 mpd->scanned_until_end = 1; 2747 if (handle) 2748 ext4_journal_stop(handle); 2749 return 0; 2750 out: 2751 folio_batch_release(&fbatch); 2752 if (handle) 2753 ext4_journal_stop(handle); 2754 return err; 2755 } 2756 2757 static int ext4_do_writepages(struct mpage_da_data *mpd) 2758 { 2759 struct writeback_control *wbc = mpd->wbc; 2760 pgoff_t writeback_index = 0; 2761 long nr_to_write = wbc->nr_to_write; 2762 int range_whole = 0; 2763 int cycled = 1; 2764 handle_t *handle = NULL; 2765 struct inode *inode = mpd->inode; 2766 struct address_space *mapping = inode->i_mapping; 2767 int needed_blocks, rsv_blocks = 0, ret = 0; 2768 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); 2769 struct blk_plug plug; 2770 bool give_up_on_write = false; 2771 2772 trace_ext4_writepages(inode, wbc); 2773 2774 /* 2775 * No pages to write? This is mainly a kludge to avoid starting 2776 * a transaction for special inodes like journal inode on last iput() 2777 * because that could violate lock ordering on umount 2778 */ 2779 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 2780 goto out_writepages; 2781 2782 /* 2783 * If the filesystem has aborted, it is read-only, so return 2784 * right away instead of dumping stack traces later on that 2785 * will obscure the real source of the problem. We test 2786 * fs shutdown state instead of sb->s_flag's SB_RDONLY because 2787 * the latter could be true if the filesystem is mounted 2788 * read-only, and in that case, ext4_writepages should 2789 * *never* be called, so if that ever happens, we would want 2790 * the stack trace. 2791 */ 2792 ret = ext4_emergency_state(mapping->host->i_sb); 2793 if (unlikely(ret)) 2794 goto out_writepages; 2795 2796 /* 2797 * If we have inline data and arrive here, it means that 2798 * we will soon create the block for the 1st page, so 2799 * we'd better clear the inline data here. 2800 */ 2801 if (ext4_has_inline_data(inode)) { 2802 /* Just inode will be modified... */ 2803 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); 2804 if (IS_ERR(handle)) { 2805 ret = PTR_ERR(handle); 2806 goto out_writepages; 2807 } 2808 BUG_ON(ext4_test_inode_state(inode, 2809 EXT4_STATE_MAY_INLINE_DATA)); 2810 ext4_destroy_inline_data(handle, inode); 2811 ext4_journal_stop(handle); 2812 } 2813 2814 /* 2815 * data=journal mode does not do delalloc so we just need to writeout / 2816 * journal already mapped buffers. On the other hand we need to commit 2817 * transaction to make data stable. We expect all the data to be 2818 * already in the journal (the only exception are DMA pinned pages 2819 * dirtied behind our back) so we commit transaction here and run the 2820 * writeback loop to checkpoint them. The checkpointing is not actually 2821 * necessary to make data persistent *but* quite a few places (extent 2822 * shifting operations, fsverity, ...) depend on being able to drop 2823 * pagecache pages after calling filemap_write_and_wait() and for that 2824 * checkpointing needs to happen. 2825 */ 2826 if (ext4_should_journal_data(inode)) { 2827 mpd->can_map = 0; 2828 if (wbc->sync_mode == WB_SYNC_ALL) 2829 ext4_fc_commit(sbi->s_journal, 2830 EXT4_I(inode)->i_datasync_tid); 2831 } 2832 mpd->journalled_more_data = 0; 2833 2834 if (ext4_should_dioread_nolock(inode)) { 2835 int bpf = ext4_journal_blocks_per_folio(inode); 2836 /* 2837 * We may need to convert up to one extent per block in 2838 * the folio and we may dirty the inode. 2839 */ 2840 rsv_blocks = 1 + ext4_ext_index_trans_blocks(inode, bpf); 2841 } 2842 2843 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) 2844 range_whole = 1; 2845 2846 if (wbc->range_cyclic) { 2847 writeback_index = mapping->writeback_index; 2848 if (writeback_index) 2849 cycled = 0; 2850 mpd->start_pos = writeback_index << PAGE_SHIFT; 2851 mpd->end_pos = LLONG_MAX; 2852 } else { 2853 mpd->start_pos = wbc->range_start; 2854 mpd->end_pos = wbc->range_end; 2855 } 2856 2857 ext4_io_submit_init(&mpd->io_submit, wbc); 2858 retry: 2859 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) 2860 tag_pages_for_writeback(mapping, mpd->start_pos >> PAGE_SHIFT, 2861 mpd->end_pos >> PAGE_SHIFT); 2862 blk_start_plug(&plug); 2863 2864 /* 2865 * First writeback pages that don't need mapping - we can avoid 2866 * starting a transaction unnecessarily and also avoid being blocked 2867 * in the block layer on device congestion while having transaction 2868 * started. 2869 */ 2870 mpd->do_map = 0; 2871 mpd->scanned_until_end = 0; 2872 mpd->io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); 2873 if (!mpd->io_submit.io_end) { 2874 ret = -ENOMEM; 2875 goto unplug; 2876 } 2877 ret = mpage_prepare_extent_to_map(mpd); 2878 /* Unlock pages we didn't use */ 2879 mpage_release_unused_pages(mpd, false); 2880 /* Submit prepared bio */ 2881 ext4_io_submit(&mpd->io_submit); 2882 ext4_put_io_end_defer(mpd->io_submit.io_end); 2883 mpd->io_submit.io_end = NULL; 2884 if (ret < 0) 2885 goto unplug; 2886 2887 while (!mpd->scanned_until_end && wbc->nr_to_write > 0) { 2888 /* For each extent of pages we use new io_end */ 2889 mpd->io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); 2890 if (!mpd->io_submit.io_end) { 2891 ret = -ENOMEM; 2892 break; 2893 } 2894 2895 WARN_ON_ONCE(!mpd->can_map); 2896 /* 2897 * We have two constraints: We find one extent to map and we 2898 * must always write out whole page (makes a difference when 2899 * blocksize < pagesize) so that we don't block on IO when we 2900 * try to write out the rest of the page. Journalled mode is 2901 * not supported by delalloc. 2902 */ 2903 BUG_ON(ext4_should_journal_data(inode)); 2904 /* 2905 * Calculate the number of credits needed to reserve for one 2906 * extent of up to MAX_WRITEPAGES_EXTENT_LEN blocks. It will 2907 * attempt to extend the transaction or start a new iteration 2908 * if the reserved credits are insufficient. 2909 */ 2910 needed_blocks = ext4_chunk_trans_blocks(inode, 2911 MAX_WRITEPAGES_EXTENT_LEN); 2912 /* start a new transaction */ 2913 handle = ext4_journal_start_with_reserve(inode, 2914 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks); 2915 if (IS_ERR(handle)) { 2916 ret = PTR_ERR(handle); 2917 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " 2918 "%ld pages, ino %lu; err %d", __func__, 2919 wbc->nr_to_write, inode->i_ino, ret); 2920 /* Release allocated io_end */ 2921 ext4_put_io_end(mpd->io_submit.io_end); 2922 mpd->io_submit.io_end = NULL; 2923 break; 2924 } 2925 mpd->do_map = 1; 2926 2927 trace_ext4_da_write_folios_start(inode, mpd->start_pos, 2928 mpd->next_pos, wbc); 2929 ret = mpage_prepare_extent_to_map(mpd); 2930 if (!ret && mpd->map.m_len) 2931 ret = mpage_map_and_submit_extent(handle, mpd, 2932 &give_up_on_write); 2933 /* 2934 * Caution: If the handle is synchronous, 2935 * ext4_journal_stop() can wait for transaction commit 2936 * to finish which may depend on writeback of pages to 2937 * complete or on page lock to be released. In that 2938 * case, we have to wait until after we have 2939 * submitted all the IO, released page locks we hold, 2940 * and dropped io_end reference (for extent conversion 2941 * to be able to complete) before stopping the handle. 2942 */ 2943 if (!ext4_handle_valid(handle) || handle->h_sync == 0) { 2944 ext4_journal_stop(handle); 2945 handle = NULL; 2946 mpd->do_map = 0; 2947 } 2948 /* Unlock pages we didn't use */ 2949 mpage_release_unused_pages(mpd, give_up_on_write); 2950 /* Submit prepared bio */ 2951 ext4_io_submit(&mpd->io_submit); 2952 2953 /* 2954 * Drop our io_end reference we got from init. We have 2955 * to be careful and use deferred io_end finishing if 2956 * we are still holding the transaction as we can 2957 * release the last reference to io_end which may end 2958 * up doing unwritten extent conversion. 2959 */ 2960 if (handle) { 2961 ext4_put_io_end_defer(mpd->io_submit.io_end); 2962 ext4_journal_stop(handle); 2963 } else 2964 ext4_put_io_end(mpd->io_submit.io_end); 2965 mpd->io_submit.io_end = NULL; 2966 trace_ext4_da_write_folios_end(inode, mpd->start_pos, 2967 mpd->next_pos, wbc, ret); 2968 2969 if (ret == -ENOSPC && sbi->s_journal) { 2970 /* 2971 * Commit the transaction which would 2972 * free blocks released in the transaction 2973 * and try again 2974 */ 2975 jbd2_journal_force_commit_nested(sbi->s_journal); 2976 ret = 0; 2977 continue; 2978 } 2979 if (ret == -EAGAIN) 2980 ret = 0; 2981 /* Fatal error - ENOMEM, EIO... */ 2982 if (ret) 2983 break; 2984 } 2985 unplug: 2986 blk_finish_plug(&plug); 2987 if (!ret && !cycled && wbc->nr_to_write > 0) { 2988 cycled = 1; 2989 mpd->end_pos = (writeback_index << PAGE_SHIFT) - 1; 2990 mpd->start_pos = 0; 2991 goto retry; 2992 } 2993 2994 /* Update index */ 2995 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) 2996 /* 2997 * Set the writeback_index so that range_cyclic 2998 * mode will write it back later 2999 */ 3000 mapping->writeback_index = mpd->start_pos >> PAGE_SHIFT; 3001 3002 out_writepages: 3003 trace_ext4_writepages_result(inode, wbc, ret, 3004 nr_to_write - wbc->nr_to_write); 3005 return ret; 3006 } 3007 3008 static int ext4_writepages(struct address_space *mapping, 3009 struct writeback_control *wbc) 3010 { 3011 struct super_block *sb = mapping->host->i_sb; 3012 struct mpage_da_data mpd = { 3013 .inode = mapping->host, 3014 .wbc = wbc, 3015 .can_map = 1, 3016 }; 3017 int ret; 3018 int alloc_ctx; 3019 3020 ret = ext4_emergency_state(sb); 3021 if (unlikely(ret)) 3022 return ret; 3023 3024 alloc_ctx = ext4_writepages_down_read(sb); 3025 ret = ext4_do_writepages(&mpd); 3026 /* 3027 * For data=journal writeback we could have come across pages marked 3028 * for delayed dirtying (PageChecked) which were just added to the 3029 * running transaction. Try once more to get them to stable storage. 3030 */ 3031 if (!ret && mpd.journalled_more_data) 3032 ret = ext4_do_writepages(&mpd); 3033 ext4_writepages_up_read(sb, alloc_ctx); 3034 3035 return ret; 3036 } 3037 3038 int ext4_normal_submit_inode_data_buffers(struct jbd2_inode *jinode) 3039 { 3040 struct writeback_control wbc = { 3041 .sync_mode = WB_SYNC_ALL, 3042 .nr_to_write = LONG_MAX, 3043 .range_start = jinode->i_dirty_start, 3044 .range_end = jinode->i_dirty_end, 3045 }; 3046 struct mpage_da_data mpd = { 3047 .inode = jinode->i_vfs_inode, 3048 .wbc = &wbc, 3049 .can_map = 0, 3050 }; 3051 return ext4_do_writepages(&mpd); 3052 } 3053 3054 static int ext4_dax_writepages(struct address_space *mapping, 3055 struct writeback_control *wbc) 3056 { 3057 int ret; 3058 long nr_to_write = wbc->nr_to_write; 3059 struct inode *inode = mapping->host; 3060 int alloc_ctx; 3061 3062 ret = ext4_emergency_state(inode->i_sb); 3063 if (unlikely(ret)) 3064 return ret; 3065 3066 alloc_ctx = ext4_writepages_down_read(inode->i_sb); 3067 trace_ext4_writepages(inode, wbc); 3068 3069 ret = dax_writeback_mapping_range(mapping, 3070 EXT4_SB(inode->i_sb)->s_daxdev, wbc); 3071 trace_ext4_writepages_result(inode, wbc, ret, 3072 nr_to_write - wbc->nr_to_write); 3073 ext4_writepages_up_read(inode->i_sb, alloc_ctx); 3074 return ret; 3075 } 3076 3077 static int ext4_nonda_switch(struct super_block *sb) 3078 { 3079 s64 free_clusters, dirty_clusters; 3080 struct ext4_sb_info *sbi = EXT4_SB(sb); 3081 3082 /* 3083 * switch to non delalloc mode if we are running low 3084 * on free block. The free block accounting via percpu 3085 * counters can get slightly wrong with percpu_counter_batch getting 3086 * accumulated on each CPU without updating global counters 3087 * Delalloc need an accurate free block accounting. So switch 3088 * to non delalloc when we are near to error range. 3089 */ 3090 free_clusters = 3091 percpu_counter_read_positive(&sbi->s_freeclusters_counter); 3092 dirty_clusters = 3093 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter); 3094 /* 3095 * Start pushing delalloc when 1/2 of free blocks are dirty. 3096 */ 3097 if (dirty_clusters && (free_clusters < 2 * dirty_clusters)) 3098 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE); 3099 3100 if (2 * free_clusters < 3 * dirty_clusters || 3101 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) { 3102 /* 3103 * free block count is less than 150% of dirty blocks 3104 * or free blocks is less than watermark 3105 */ 3106 return 1; 3107 } 3108 return 0; 3109 } 3110 3111 static int ext4_da_write_begin(const struct kiocb *iocb, 3112 struct address_space *mapping, 3113 loff_t pos, unsigned len, 3114 struct folio **foliop, void **fsdata) 3115 { 3116 int ret, retries = 0; 3117 struct folio *folio; 3118 pgoff_t index; 3119 struct inode *inode = mapping->host; 3120 3121 ret = ext4_emergency_state(inode->i_sb); 3122 if (unlikely(ret)) 3123 return ret; 3124 3125 index = pos >> PAGE_SHIFT; 3126 3127 if (ext4_nonda_switch(inode->i_sb) || ext4_verity_in_progress(inode)) { 3128 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; 3129 return ext4_write_begin(iocb, mapping, pos, 3130 len, foliop, fsdata); 3131 } 3132 *fsdata = (void *)0; 3133 trace_ext4_da_write_begin(inode, pos, len); 3134 3135 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { 3136 ret = ext4_generic_write_inline_data(mapping, inode, pos, len, 3137 foliop, fsdata, true); 3138 if (ret < 0) 3139 return ret; 3140 if (ret == 1) 3141 return 0; 3142 } 3143 3144 retry: 3145 folio = write_begin_get_folio(iocb, mapping, index, len); 3146 if (IS_ERR(folio)) 3147 return PTR_ERR(folio); 3148 3149 if (pos + len > folio_pos(folio) + folio_size(folio)) 3150 len = folio_pos(folio) + folio_size(folio) - pos; 3151 3152 ret = ext4_block_write_begin(NULL, folio, pos, len, 3153 ext4_da_get_block_prep); 3154 if (ret < 0) { 3155 folio_unlock(folio); 3156 folio_put(folio); 3157 /* 3158 * ext4_block_write_begin may have instantiated a few blocks 3159 * outside i_size. Trim these off again. Don't need 3160 * i_size_read because we hold inode lock. 3161 */ 3162 if (pos + len > inode->i_size) 3163 ext4_truncate_failed_write(inode); 3164 3165 if (ret == -ENOSPC && 3166 ext4_should_retry_alloc(inode->i_sb, &retries)) 3167 goto retry; 3168 return ret; 3169 } 3170 3171 *foliop = folio; 3172 return ret; 3173 } 3174 3175 /* 3176 * Check if we should update i_disksize 3177 * when write to the end of file but not require block allocation 3178 */ 3179 static int ext4_da_should_update_i_disksize(struct folio *folio, 3180 unsigned long offset) 3181 { 3182 struct buffer_head *bh; 3183 struct inode *inode = folio->mapping->host; 3184 unsigned int idx; 3185 int i; 3186 3187 bh = folio_buffers(folio); 3188 idx = offset >> inode->i_blkbits; 3189 3190 for (i = 0; i < idx; i++) 3191 bh = bh->b_this_page; 3192 3193 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) 3194 return 0; 3195 return 1; 3196 } 3197 3198 static int ext4_da_do_write_end(struct address_space *mapping, 3199 loff_t pos, unsigned len, unsigned copied, 3200 struct folio *folio) 3201 { 3202 struct inode *inode = mapping->host; 3203 loff_t old_size = inode->i_size; 3204 bool disksize_changed = false; 3205 loff_t new_i_size, zero_len = 0; 3206 handle_t *handle; 3207 3208 if (unlikely(!folio_buffers(folio))) { 3209 folio_unlock(folio); 3210 folio_put(folio); 3211 return -EIO; 3212 } 3213 /* 3214 * block_write_end() will mark the inode as dirty with I_DIRTY_PAGES 3215 * flag, which all that's needed to trigger page writeback. 3216 */ 3217 copied = block_write_end(pos, len, copied, folio); 3218 new_i_size = pos + copied; 3219 3220 /* 3221 * It's important to update i_size while still holding folio lock, 3222 * because folio writeout could otherwise come in and zero beyond 3223 * i_size. 3224 * 3225 * Since we are holding inode lock, we are sure i_disksize <= 3226 * i_size. We also know that if i_disksize < i_size, there are 3227 * delalloc writes pending in the range up to i_size. If the end of 3228 * the current write is <= i_size, there's no need to touch 3229 * i_disksize since writeback will push i_disksize up to i_size 3230 * eventually. If the end of the current write is > i_size and 3231 * inside an allocated block which ext4_da_should_update_i_disksize() 3232 * checked, we need to update i_disksize here as certain 3233 * ext4_writepages() paths not allocating blocks and update i_disksize. 3234 */ 3235 if (new_i_size > inode->i_size) { 3236 unsigned long end; 3237 3238 i_size_write(inode, new_i_size); 3239 end = offset_in_folio(folio, new_i_size - 1); 3240 if (copied && ext4_da_should_update_i_disksize(folio, end)) { 3241 ext4_update_i_disksize(inode, new_i_size); 3242 disksize_changed = true; 3243 } 3244 } 3245 3246 folio_unlock(folio); 3247 folio_put(folio); 3248 3249 if (pos > old_size) { 3250 pagecache_isize_extended(inode, old_size, pos); 3251 zero_len = pos - old_size; 3252 } 3253 3254 if (!disksize_changed && !zero_len) 3255 return copied; 3256 3257 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); 3258 if (IS_ERR(handle)) 3259 return PTR_ERR(handle); 3260 if (zero_len) 3261 ext4_zero_partial_blocks(handle, inode, old_size, zero_len); 3262 ext4_mark_inode_dirty(handle, inode); 3263 ext4_journal_stop(handle); 3264 3265 return copied; 3266 } 3267 3268 static int ext4_da_write_end(const struct kiocb *iocb, 3269 struct address_space *mapping, 3270 loff_t pos, unsigned len, unsigned copied, 3271 struct folio *folio, void *fsdata) 3272 { 3273 struct inode *inode = mapping->host; 3274 int write_mode = (int)(unsigned long)fsdata; 3275 3276 if (write_mode == FALL_BACK_TO_NONDELALLOC) 3277 return ext4_write_end(iocb, mapping, pos, 3278 len, copied, folio, fsdata); 3279 3280 trace_ext4_da_write_end(inode, pos, len, copied); 3281 3282 if (write_mode != CONVERT_INLINE_DATA && 3283 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) && 3284 ext4_has_inline_data(inode)) 3285 return ext4_write_inline_data_end(inode, pos, len, copied, 3286 folio); 3287 3288 if (unlikely(copied < len) && !folio_test_uptodate(folio)) 3289 copied = 0; 3290 3291 return ext4_da_do_write_end(mapping, pos, len, copied, folio); 3292 } 3293 3294 /* 3295 * Force all delayed allocation blocks to be allocated for a given inode. 3296 */ 3297 int ext4_alloc_da_blocks(struct inode *inode) 3298 { 3299 trace_ext4_alloc_da_blocks(inode); 3300 3301 if (!EXT4_I(inode)->i_reserved_data_blocks) 3302 return 0; 3303 3304 /* 3305 * We do something simple for now. The filemap_flush() will 3306 * also start triggering a write of the data blocks, which is 3307 * not strictly speaking necessary (and for users of 3308 * laptop_mode, not even desirable). However, to do otherwise 3309 * would require replicating code paths in: 3310 * 3311 * ext4_writepages() -> 3312 * write_cache_pages() ---> (via passed in callback function) 3313 * __mpage_da_writepage() --> 3314 * mpage_add_bh_to_extent() 3315 * mpage_da_map_blocks() 3316 * 3317 * The problem is that write_cache_pages(), located in 3318 * mm/page-writeback.c, marks pages clean in preparation for 3319 * doing I/O, which is not desirable if we're not planning on 3320 * doing I/O at all. 3321 * 3322 * We could call write_cache_pages(), and then redirty all of 3323 * the pages by calling redirty_page_for_writepage() but that 3324 * would be ugly in the extreme. So instead we would need to 3325 * replicate parts of the code in the above functions, 3326 * simplifying them because we wouldn't actually intend to 3327 * write out the pages, but rather only collect contiguous 3328 * logical block extents, call the multi-block allocator, and 3329 * then update the buffer heads with the block allocations. 3330 * 3331 * For now, though, we'll cheat by calling filemap_flush(), 3332 * which will map the blocks, and start the I/O, but not 3333 * actually wait for the I/O to complete. 3334 */ 3335 return filemap_flush(inode->i_mapping); 3336 } 3337 3338 /* 3339 * bmap() is special. It gets used by applications such as lilo and by 3340 * the swapper to find the on-disk block of a specific piece of data. 3341 * 3342 * Naturally, this is dangerous if the block concerned is still in the 3343 * journal. If somebody makes a swapfile on an ext4 data-journaling 3344 * filesystem and enables swap, then they may get a nasty shock when the 3345 * data getting swapped to that swapfile suddenly gets overwritten by 3346 * the original zero's written out previously to the journal and 3347 * awaiting writeback in the kernel's buffer cache. 3348 * 3349 * So, if we see any bmap calls here on a modified, data-journaled file, 3350 * take extra steps to flush any blocks which might be in the cache. 3351 */ 3352 static sector_t ext4_bmap(struct address_space *mapping, sector_t block) 3353 { 3354 struct inode *inode = mapping->host; 3355 sector_t ret = 0; 3356 3357 inode_lock_shared(inode); 3358 /* 3359 * We can get here for an inline file via the FIBMAP ioctl 3360 */ 3361 if (ext4_has_inline_data(inode)) 3362 goto out; 3363 3364 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && 3365 (test_opt(inode->i_sb, DELALLOC) || 3366 ext4_should_journal_data(inode))) { 3367 /* 3368 * With delalloc or journalled data we want to sync the file so 3369 * that we can make sure we allocate blocks for file and data 3370 * is in place for the user to see it 3371 */ 3372 filemap_write_and_wait(mapping); 3373 } 3374 3375 ret = iomap_bmap(mapping, block, &ext4_iomap_ops); 3376 3377 out: 3378 inode_unlock_shared(inode); 3379 return ret; 3380 } 3381 3382 static int ext4_read_folio(struct file *file, struct folio *folio) 3383 { 3384 int ret = -EAGAIN; 3385 struct inode *inode = folio->mapping->host; 3386 3387 trace_ext4_read_folio(inode, folio); 3388 3389 if (ext4_has_inline_data(inode)) 3390 ret = ext4_readpage_inline(inode, folio); 3391 3392 if (ret == -EAGAIN) 3393 return ext4_mpage_readpages(inode, NULL, folio); 3394 3395 return ret; 3396 } 3397 3398 static void ext4_readahead(struct readahead_control *rac) 3399 { 3400 struct inode *inode = rac->mapping->host; 3401 3402 /* If the file has inline data, no need to do readahead. */ 3403 if (ext4_has_inline_data(inode)) 3404 return; 3405 3406 ext4_mpage_readpages(inode, rac, NULL); 3407 } 3408 3409 static void ext4_invalidate_folio(struct folio *folio, size_t offset, 3410 size_t length) 3411 { 3412 trace_ext4_invalidate_folio(folio, offset, length); 3413 3414 /* No journalling happens on data buffers when this function is used */ 3415 WARN_ON(folio_buffers(folio) && buffer_jbd(folio_buffers(folio))); 3416 3417 block_invalidate_folio(folio, offset, length); 3418 } 3419 3420 static int __ext4_journalled_invalidate_folio(struct folio *folio, 3421 size_t offset, size_t length) 3422 { 3423 journal_t *journal = EXT4_JOURNAL(folio->mapping->host); 3424 3425 trace_ext4_journalled_invalidate_folio(folio, offset, length); 3426 3427 /* 3428 * If it's a full truncate we just forget about the pending dirtying 3429 */ 3430 if (offset == 0 && length == folio_size(folio)) 3431 folio_clear_checked(folio); 3432 3433 return jbd2_journal_invalidate_folio(journal, folio, offset, length); 3434 } 3435 3436 /* Wrapper for aops... */ 3437 static void ext4_journalled_invalidate_folio(struct folio *folio, 3438 size_t offset, 3439 size_t length) 3440 { 3441 WARN_ON(__ext4_journalled_invalidate_folio(folio, offset, length) < 0); 3442 } 3443 3444 static bool ext4_release_folio(struct folio *folio, gfp_t wait) 3445 { 3446 struct inode *inode = folio->mapping->host; 3447 journal_t *journal = EXT4_JOURNAL(inode); 3448 3449 trace_ext4_release_folio(inode, folio); 3450 3451 /* Page has dirty journalled data -> cannot release */ 3452 if (folio_test_checked(folio)) 3453 return false; 3454 if (journal) 3455 return jbd2_journal_try_to_free_buffers(journal, folio); 3456 else 3457 return try_to_free_buffers(folio); 3458 } 3459 3460 static bool ext4_inode_datasync_dirty(struct inode *inode) 3461 { 3462 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; 3463 3464 if (journal) { 3465 if (jbd2_transaction_committed(journal, 3466 EXT4_I(inode)->i_datasync_tid)) 3467 return false; 3468 if (test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT)) 3469 return !list_empty(&EXT4_I(inode)->i_fc_list); 3470 return true; 3471 } 3472 3473 /* Any metadata buffers to write? */ 3474 if (!list_empty(&inode->i_mapping->i_private_list)) 3475 return true; 3476 return inode->i_state & I_DIRTY_DATASYNC; 3477 } 3478 3479 static void ext4_set_iomap(struct inode *inode, struct iomap *iomap, 3480 struct ext4_map_blocks *map, loff_t offset, 3481 loff_t length, unsigned int flags) 3482 { 3483 u8 blkbits = inode->i_blkbits; 3484 3485 /* 3486 * Writes that span EOF might trigger an I/O size update on completion, 3487 * so consider them to be dirty for the purpose of O_DSYNC, even if 3488 * there is no other metadata changes being made or are pending. 3489 */ 3490 iomap->flags = 0; 3491 if (ext4_inode_datasync_dirty(inode) || 3492 offset + length > i_size_read(inode)) 3493 iomap->flags |= IOMAP_F_DIRTY; 3494 3495 if (map->m_flags & EXT4_MAP_NEW) 3496 iomap->flags |= IOMAP_F_NEW; 3497 3498 /* HW-offload atomics are always used */ 3499 if (flags & IOMAP_ATOMIC) 3500 iomap->flags |= IOMAP_F_ATOMIC_BIO; 3501 3502 if (flags & IOMAP_DAX) 3503 iomap->dax_dev = EXT4_SB(inode->i_sb)->s_daxdev; 3504 else 3505 iomap->bdev = inode->i_sb->s_bdev; 3506 iomap->offset = (u64) map->m_lblk << blkbits; 3507 iomap->length = (u64) map->m_len << blkbits; 3508 3509 if ((map->m_flags & EXT4_MAP_MAPPED) && 3510 !ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) 3511 iomap->flags |= IOMAP_F_MERGED; 3512 3513 /* 3514 * Flags passed to ext4_map_blocks() for direct I/O writes can result 3515 * in m_flags having both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN bits 3516 * set. In order for any allocated unwritten extents to be converted 3517 * into written extents correctly within the ->end_io() handler, we 3518 * need to ensure that the iomap->type is set appropriately. Hence, the 3519 * reason why we need to check whether the EXT4_MAP_UNWRITTEN bit has 3520 * been set first. 3521 */ 3522 if (map->m_flags & EXT4_MAP_UNWRITTEN) { 3523 iomap->type = IOMAP_UNWRITTEN; 3524 iomap->addr = (u64) map->m_pblk << blkbits; 3525 if (flags & IOMAP_DAX) 3526 iomap->addr += EXT4_SB(inode->i_sb)->s_dax_part_off; 3527 } else if (map->m_flags & EXT4_MAP_MAPPED) { 3528 iomap->type = IOMAP_MAPPED; 3529 iomap->addr = (u64) map->m_pblk << blkbits; 3530 if (flags & IOMAP_DAX) 3531 iomap->addr += EXT4_SB(inode->i_sb)->s_dax_part_off; 3532 } else if (map->m_flags & EXT4_MAP_DELAYED) { 3533 iomap->type = IOMAP_DELALLOC; 3534 iomap->addr = IOMAP_NULL_ADDR; 3535 } else { 3536 iomap->type = IOMAP_HOLE; 3537 iomap->addr = IOMAP_NULL_ADDR; 3538 } 3539 } 3540 3541 static int ext4_map_blocks_atomic_write_slow(handle_t *handle, 3542 struct inode *inode, struct ext4_map_blocks *map) 3543 { 3544 ext4_lblk_t m_lblk = map->m_lblk; 3545 unsigned int m_len = map->m_len; 3546 unsigned int mapped_len = 0, m_flags = 0; 3547 ext4_fsblk_t next_pblk; 3548 bool check_next_pblk = false; 3549 int ret = 0; 3550 3551 WARN_ON_ONCE(!ext4_has_feature_bigalloc(inode->i_sb)); 3552 3553 /* 3554 * This is a slow path in case of mixed mapping. We use 3555 * EXT4_GET_BLOCKS_CREATE_ZERO flag here to make sure we get a single 3556 * contiguous mapped mapping. This will ensure any unwritten or hole 3557 * regions within the requested range is zeroed out and we return 3558 * a single contiguous mapped extent. 3559 */ 3560 m_flags = EXT4_GET_BLOCKS_CREATE_ZERO; 3561 3562 do { 3563 ret = ext4_map_blocks(handle, inode, map, m_flags); 3564 if (ret < 0 && ret != -ENOSPC) 3565 goto out_err; 3566 /* 3567 * This should never happen, but let's return an error code to 3568 * avoid an infinite loop in here. 3569 */ 3570 if (ret == 0) { 3571 ret = -EFSCORRUPTED; 3572 ext4_warning_inode(inode, 3573 "ext4_map_blocks() couldn't allocate blocks m_flags: 0x%x, ret:%d", 3574 m_flags, ret); 3575 goto out_err; 3576 } 3577 /* 3578 * With bigalloc we should never get ENOSPC nor discontiguous 3579 * physical extents. 3580 */ 3581 if ((check_next_pblk && next_pblk != map->m_pblk) || 3582 ret == -ENOSPC) { 3583 ext4_warning_inode(inode, 3584 "Non-contiguous allocation detected: expected %llu, got %llu, " 3585 "or ext4_map_blocks() returned out of space ret: %d", 3586 next_pblk, map->m_pblk, ret); 3587 ret = -EFSCORRUPTED; 3588 goto out_err; 3589 } 3590 next_pblk = map->m_pblk + map->m_len; 3591 check_next_pblk = true; 3592 3593 mapped_len += map->m_len; 3594 map->m_lblk += map->m_len; 3595 map->m_len = m_len - mapped_len; 3596 } while (mapped_len < m_len); 3597 3598 /* 3599 * We might have done some work in above loop, so we need to query the 3600 * start of the physical extent, based on the origin m_lblk and m_len. 3601 * Let's also ensure we were able to allocate the required range for 3602 * mixed mapping case. 3603 */ 3604 map->m_lblk = m_lblk; 3605 map->m_len = m_len; 3606 map->m_flags = 0; 3607 3608 ret = ext4_map_blocks(handle, inode, map, 3609 EXT4_GET_BLOCKS_QUERY_LAST_IN_LEAF); 3610 if (ret != m_len) { 3611 ext4_warning_inode(inode, 3612 "allocation failed for atomic write request m_lblk:%u, m_len:%u, ret:%d\n", 3613 m_lblk, m_len, ret); 3614 ret = -EINVAL; 3615 } 3616 return ret; 3617 3618 out_err: 3619 /* reset map before returning an error */ 3620 map->m_lblk = m_lblk; 3621 map->m_len = m_len; 3622 map->m_flags = 0; 3623 return ret; 3624 } 3625 3626 /* 3627 * ext4_map_blocks_atomic: Helper routine to ensure the entire requested 3628 * range in @map [lblk, lblk + len) is one single contiguous extent with no 3629 * mixed mappings. 3630 * 3631 * We first use m_flags passed to us by our caller (ext4_iomap_alloc()). 3632 * We only call EXT4_GET_BLOCKS_ZERO in the slow path, when the underlying 3633 * physical extent for the requested range does not have a single contiguous 3634 * mapping type i.e. (Hole, Mapped, or Unwritten) throughout. 3635 * In that case we will loop over the requested range to allocate and zero out 3636 * the unwritten / holes in between, to get a single mapped extent from 3637 * [m_lblk, m_lblk + m_len). Note that this is only possible because we know 3638 * this can be called only with bigalloc enabled filesystem where the underlying 3639 * cluster is already allocated. This avoids allocating discontiguous extents 3640 * in the slow path due to multiple calls to ext4_map_blocks(). 3641 * The slow path is mostly non-performance critical path, so it should be ok to 3642 * loop using ext4_map_blocks() with appropriate flags to allocate & zero the 3643 * underlying short holes/unwritten extents within the requested range. 3644 */ 3645 static int ext4_map_blocks_atomic_write(handle_t *handle, struct inode *inode, 3646 struct ext4_map_blocks *map, int m_flags, 3647 bool *force_commit) 3648 { 3649 ext4_lblk_t m_lblk = map->m_lblk; 3650 unsigned int m_len = map->m_len; 3651 int ret = 0; 3652 3653 WARN_ON_ONCE(m_len > 1 && !ext4_has_feature_bigalloc(inode->i_sb)); 3654 3655 ret = ext4_map_blocks(handle, inode, map, m_flags); 3656 if (ret < 0 || ret == m_len) 3657 goto out; 3658 /* 3659 * This is a mixed mapping case where we were not able to allocate 3660 * a single contiguous extent. In that case let's reset requested 3661 * mapping and call the slow path. 3662 */ 3663 map->m_lblk = m_lblk; 3664 map->m_len = m_len; 3665 map->m_flags = 0; 3666 3667 /* 3668 * slow path means we have mixed mapping, that means we will need 3669 * to force txn commit. 3670 */ 3671 *force_commit = true; 3672 return ext4_map_blocks_atomic_write_slow(handle, inode, map); 3673 out: 3674 return ret; 3675 } 3676 3677 static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map, 3678 unsigned int flags) 3679 { 3680 handle_t *handle; 3681 u8 blkbits = inode->i_blkbits; 3682 int ret, dio_credits, m_flags = 0, retries = 0; 3683 bool force_commit = false; 3684 3685 /* 3686 * Trim the mapping request to the maximum value that we can map at 3687 * once for direct I/O. 3688 */ 3689 if (map->m_len > DIO_MAX_BLOCKS) 3690 map->m_len = DIO_MAX_BLOCKS; 3691 3692 /* 3693 * journal credits estimation for atomic writes. We call 3694 * ext4_map_blocks(), to find if there could be a mixed mapping. If yes, 3695 * then let's assume the no. of pextents required can be m_len i.e. 3696 * every alternate block can be unwritten and hole. 3697 */ 3698 if (flags & IOMAP_ATOMIC) { 3699 unsigned int orig_mlen = map->m_len; 3700 3701 ret = ext4_map_blocks(NULL, inode, map, 0); 3702 if (ret < 0) 3703 return ret; 3704 if (map->m_len < orig_mlen) { 3705 map->m_len = orig_mlen; 3706 dio_credits = ext4_meta_trans_blocks(inode, orig_mlen, 3707 map->m_len); 3708 } else { 3709 dio_credits = ext4_chunk_trans_blocks(inode, 3710 map->m_len); 3711 } 3712 } else { 3713 dio_credits = ext4_chunk_trans_blocks(inode, map->m_len); 3714 } 3715 3716 retry: 3717 /* 3718 * Either we allocate blocks and then don't get an unwritten extent, so 3719 * in that case we have reserved enough credits. Or, the blocks are 3720 * already allocated and unwritten. In that case, the extent conversion 3721 * fits into the credits as well. 3722 */ 3723 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits); 3724 if (IS_ERR(handle)) 3725 return PTR_ERR(handle); 3726 3727 /* 3728 * DAX and direct I/O are the only two operations that are currently 3729 * supported with IOMAP_WRITE. 3730 */ 3731 WARN_ON(!(flags & (IOMAP_DAX | IOMAP_DIRECT))); 3732 if (flags & IOMAP_DAX) 3733 m_flags = EXT4_GET_BLOCKS_CREATE_ZERO; 3734 /* 3735 * We use i_size instead of i_disksize here because delalloc writeback 3736 * can complete at any point during the I/O and subsequently push the 3737 * i_disksize out to i_size. This could be beyond where direct I/O is 3738 * happening and thus expose allocated blocks to direct I/O reads. 3739 */ 3740 else if (((loff_t)map->m_lblk << blkbits) >= i_size_read(inode)) 3741 m_flags = EXT4_GET_BLOCKS_CREATE; 3742 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) 3743 m_flags = EXT4_GET_BLOCKS_IO_CREATE_EXT; 3744 3745 if (flags & IOMAP_ATOMIC) 3746 ret = ext4_map_blocks_atomic_write(handle, inode, map, m_flags, 3747 &force_commit); 3748 else 3749 ret = ext4_map_blocks(handle, inode, map, m_flags); 3750 3751 /* 3752 * We cannot fill holes in indirect tree based inodes as that could 3753 * expose stale data in the case of a crash. Use the magic error code 3754 * to fallback to buffered I/O. 3755 */ 3756 if (!m_flags && !ret) 3757 ret = -ENOTBLK; 3758 3759 ext4_journal_stop(handle); 3760 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) 3761 goto retry; 3762 3763 /* 3764 * Force commit the current transaction if the allocation spans a mixed 3765 * mapping range. This ensures any pending metadata updates (like 3766 * unwritten to written extents conversion) in this range are in 3767 * consistent state with the file data blocks, before performing the 3768 * actual write I/O. If the commit fails, the whole I/O must be aborted 3769 * to prevent any possible torn writes. 3770 */ 3771 if (ret > 0 && force_commit) { 3772 int ret2; 3773 3774 ret2 = ext4_force_commit(inode->i_sb); 3775 if (ret2) 3776 return ret2; 3777 } 3778 3779 return ret; 3780 } 3781 3782 3783 static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length, 3784 unsigned flags, struct iomap *iomap, struct iomap *srcmap) 3785 { 3786 int ret; 3787 struct ext4_map_blocks map; 3788 u8 blkbits = inode->i_blkbits; 3789 unsigned int orig_mlen; 3790 3791 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK) 3792 return -EINVAL; 3793 3794 if (WARN_ON_ONCE(ext4_has_inline_data(inode))) 3795 return -ERANGE; 3796 3797 /* 3798 * Calculate the first and last logical blocks respectively. 3799 */ 3800 map.m_lblk = offset >> blkbits; 3801 map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits, 3802 EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1; 3803 orig_mlen = map.m_len; 3804 3805 if (flags & IOMAP_WRITE) { 3806 /* 3807 * We check here if the blocks are already allocated, then we 3808 * don't need to start a journal txn and we can directly return 3809 * the mapping information. This could boost performance 3810 * especially in multi-threaded overwrite requests. 3811 */ 3812 if (offset + length <= i_size_read(inode)) { 3813 ret = ext4_map_blocks(NULL, inode, &map, 0); 3814 /* 3815 * For atomic writes the entire requested length should 3816 * be mapped. 3817 */ 3818 if (map.m_flags & EXT4_MAP_MAPPED) { 3819 if ((!(flags & IOMAP_ATOMIC) && ret > 0) || 3820 (flags & IOMAP_ATOMIC && ret >= orig_mlen)) 3821 goto out; 3822 } 3823 map.m_len = orig_mlen; 3824 } 3825 ret = ext4_iomap_alloc(inode, &map, flags); 3826 } else { 3827 /* 3828 * This can be called for overwrites path from 3829 * ext4_iomap_overwrite_begin(). 3830 */ 3831 ret = ext4_map_blocks(NULL, inode, &map, 0); 3832 } 3833 3834 if (ret < 0) 3835 return ret; 3836 out: 3837 /* 3838 * When inline encryption is enabled, sometimes I/O to an encrypted file 3839 * has to be broken up to guarantee DUN contiguity. Handle this by 3840 * limiting the length of the mapping returned. 3841 */ 3842 map.m_len = fscrypt_limit_io_blocks(inode, map.m_lblk, map.m_len); 3843 3844 /* 3845 * Before returning to iomap, let's ensure the allocated mapping 3846 * covers the entire requested length for atomic writes. 3847 */ 3848 if (flags & IOMAP_ATOMIC) { 3849 if (map.m_len < (length >> blkbits)) { 3850 WARN_ON_ONCE(1); 3851 return -EINVAL; 3852 } 3853 } 3854 ext4_set_iomap(inode, iomap, &map, offset, length, flags); 3855 3856 return 0; 3857 } 3858 3859 static int ext4_iomap_overwrite_begin(struct inode *inode, loff_t offset, 3860 loff_t length, unsigned flags, struct iomap *iomap, 3861 struct iomap *srcmap) 3862 { 3863 int ret; 3864 3865 /* 3866 * Even for writes we don't need to allocate blocks, so just pretend 3867 * we are reading to save overhead of starting a transaction. 3868 */ 3869 flags &= ~IOMAP_WRITE; 3870 ret = ext4_iomap_begin(inode, offset, length, flags, iomap, srcmap); 3871 WARN_ON_ONCE(!ret && iomap->type != IOMAP_MAPPED); 3872 return ret; 3873 } 3874 3875 const struct iomap_ops ext4_iomap_ops = { 3876 .iomap_begin = ext4_iomap_begin, 3877 }; 3878 3879 const struct iomap_ops ext4_iomap_overwrite_ops = { 3880 .iomap_begin = ext4_iomap_overwrite_begin, 3881 }; 3882 3883 static int ext4_iomap_begin_report(struct inode *inode, loff_t offset, 3884 loff_t length, unsigned int flags, 3885 struct iomap *iomap, struct iomap *srcmap) 3886 { 3887 int ret; 3888 struct ext4_map_blocks map; 3889 u8 blkbits = inode->i_blkbits; 3890 3891 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK) 3892 return -EINVAL; 3893 3894 if (ext4_has_inline_data(inode)) { 3895 ret = ext4_inline_data_iomap(inode, iomap); 3896 if (ret != -EAGAIN) { 3897 if (ret == 0 && offset >= iomap->length) 3898 ret = -ENOENT; 3899 return ret; 3900 } 3901 } 3902 3903 /* 3904 * Calculate the first and last logical block respectively. 3905 */ 3906 map.m_lblk = offset >> blkbits; 3907 map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits, 3908 EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1; 3909 3910 /* 3911 * Fiemap callers may call for offset beyond s_bitmap_maxbytes. 3912 * So handle it here itself instead of querying ext4_map_blocks(). 3913 * Since ext4_map_blocks() will warn about it and will return 3914 * -EIO error. 3915 */ 3916 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { 3917 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 3918 3919 if (offset >= sbi->s_bitmap_maxbytes) { 3920 map.m_flags = 0; 3921 goto set_iomap; 3922 } 3923 } 3924 3925 ret = ext4_map_blocks(NULL, inode, &map, 0); 3926 if (ret < 0) 3927 return ret; 3928 set_iomap: 3929 ext4_set_iomap(inode, iomap, &map, offset, length, flags); 3930 3931 return 0; 3932 } 3933 3934 const struct iomap_ops ext4_iomap_report_ops = { 3935 .iomap_begin = ext4_iomap_begin_report, 3936 }; 3937 3938 /* 3939 * For data=journal mode, folio should be marked dirty only when it was 3940 * writeably mapped. When that happens, it was already attached to the 3941 * transaction and marked as jbddirty (we take care of this in 3942 * ext4_page_mkwrite()). On transaction commit, we writeprotect page mappings 3943 * so we should have nothing to do here, except for the case when someone 3944 * had the page pinned and dirtied the page through this pin (e.g. by doing 3945 * direct IO to it). In that case we'd need to attach buffers here to the 3946 * transaction but we cannot due to lock ordering. We cannot just dirty the 3947 * folio and leave attached buffers clean, because the buffers' dirty state is 3948 * "definitive". We cannot just set the buffers dirty or jbddirty because all 3949 * the journalling code will explode. So what we do is to mark the folio 3950 * "pending dirty" and next time ext4_writepages() is called, attach buffers 3951 * to the transaction appropriately. 3952 */ 3953 static bool ext4_journalled_dirty_folio(struct address_space *mapping, 3954 struct folio *folio) 3955 { 3956 WARN_ON_ONCE(!folio_buffers(folio)); 3957 if (folio_maybe_dma_pinned(folio)) 3958 folio_set_checked(folio); 3959 return filemap_dirty_folio(mapping, folio); 3960 } 3961 3962 static bool ext4_dirty_folio(struct address_space *mapping, struct folio *folio) 3963 { 3964 WARN_ON_ONCE(!folio_test_locked(folio) && !folio_test_dirty(folio)); 3965 WARN_ON_ONCE(!folio_buffers(folio)); 3966 return block_dirty_folio(mapping, folio); 3967 } 3968 3969 static int ext4_iomap_swap_activate(struct swap_info_struct *sis, 3970 struct file *file, sector_t *span) 3971 { 3972 return iomap_swapfile_activate(sis, file, span, 3973 &ext4_iomap_report_ops); 3974 } 3975 3976 static const struct address_space_operations ext4_aops = { 3977 .read_folio = ext4_read_folio, 3978 .readahead = ext4_readahead, 3979 .writepages = ext4_writepages, 3980 .write_begin = ext4_write_begin, 3981 .write_end = ext4_write_end, 3982 .dirty_folio = ext4_dirty_folio, 3983 .bmap = ext4_bmap, 3984 .invalidate_folio = ext4_invalidate_folio, 3985 .release_folio = ext4_release_folio, 3986 .migrate_folio = buffer_migrate_folio, 3987 .is_partially_uptodate = block_is_partially_uptodate, 3988 .error_remove_folio = generic_error_remove_folio, 3989 .swap_activate = ext4_iomap_swap_activate, 3990 }; 3991 3992 static const struct address_space_operations ext4_journalled_aops = { 3993 .read_folio = ext4_read_folio, 3994 .readahead = ext4_readahead, 3995 .writepages = ext4_writepages, 3996 .write_begin = ext4_write_begin, 3997 .write_end = ext4_journalled_write_end, 3998 .dirty_folio = ext4_journalled_dirty_folio, 3999 .bmap = ext4_bmap, 4000 .invalidate_folio = ext4_journalled_invalidate_folio, 4001 .release_folio = ext4_release_folio, 4002 .migrate_folio = buffer_migrate_folio_norefs, 4003 .is_partially_uptodate = block_is_partially_uptodate, 4004 .error_remove_folio = generic_error_remove_folio, 4005 .swap_activate = ext4_iomap_swap_activate, 4006 }; 4007 4008 static const struct address_space_operations ext4_da_aops = { 4009 .read_folio = ext4_read_folio, 4010 .readahead = ext4_readahead, 4011 .writepages = ext4_writepages, 4012 .write_begin = ext4_da_write_begin, 4013 .write_end = ext4_da_write_end, 4014 .dirty_folio = ext4_dirty_folio, 4015 .bmap = ext4_bmap, 4016 .invalidate_folio = ext4_invalidate_folio, 4017 .release_folio = ext4_release_folio, 4018 .migrate_folio = buffer_migrate_folio, 4019 .is_partially_uptodate = block_is_partially_uptodate, 4020 .error_remove_folio = generic_error_remove_folio, 4021 .swap_activate = ext4_iomap_swap_activate, 4022 }; 4023 4024 static const struct address_space_operations ext4_dax_aops = { 4025 .writepages = ext4_dax_writepages, 4026 .dirty_folio = noop_dirty_folio, 4027 .bmap = ext4_bmap, 4028 .swap_activate = ext4_iomap_swap_activate, 4029 }; 4030 4031 void ext4_set_aops(struct inode *inode) 4032 { 4033 switch (ext4_inode_journal_mode(inode)) { 4034 case EXT4_INODE_ORDERED_DATA_MODE: 4035 case EXT4_INODE_WRITEBACK_DATA_MODE: 4036 break; 4037 case EXT4_INODE_JOURNAL_DATA_MODE: 4038 inode->i_mapping->a_ops = &ext4_journalled_aops; 4039 return; 4040 default: 4041 BUG(); 4042 } 4043 if (IS_DAX(inode)) 4044 inode->i_mapping->a_ops = &ext4_dax_aops; 4045 else if (test_opt(inode->i_sb, DELALLOC)) 4046 inode->i_mapping->a_ops = &ext4_da_aops; 4047 else 4048 inode->i_mapping->a_ops = &ext4_aops; 4049 } 4050 4051 /* 4052 * Here we can't skip an unwritten buffer even though it usually reads zero 4053 * because it might have data in pagecache (eg, if called from ext4_zero_range, 4054 * ext4_punch_hole, etc) which needs to be properly zeroed out. Otherwise a 4055 * racing writeback can come later and flush the stale pagecache to disk. 4056 */ 4057 static int __ext4_block_zero_page_range(handle_t *handle, 4058 struct address_space *mapping, loff_t from, loff_t length) 4059 { 4060 unsigned int offset, blocksize, pos; 4061 ext4_lblk_t iblock; 4062 struct inode *inode = mapping->host; 4063 struct buffer_head *bh; 4064 struct folio *folio; 4065 int err = 0; 4066 4067 folio = __filemap_get_folio(mapping, from >> PAGE_SHIFT, 4068 FGP_LOCK | FGP_ACCESSED | FGP_CREAT, 4069 mapping_gfp_constraint(mapping, ~__GFP_FS)); 4070 if (IS_ERR(folio)) 4071 return PTR_ERR(folio); 4072 4073 blocksize = inode->i_sb->s_blocksize; 4074 4075 iblock = folio->index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits); 4076 4077 bh = folio_buffers(folio); 4078 if (!bh) 4079 bh = create_empty_buffers(folio, blocksize, 0); 4080 4081 /* Find the buffer that contains "offset" */ 4082 offset = offset_in_folio(folio, from); 4083 pos = blocksize; 4084 while (offset >= pos) { 4085 bh = bh->b_this_page; 4086 iblock++; 4087 pos += blocksize; 4088 } 4089 if (buffer_freed(bh)) { 4090 BUFFER_TRACE(bh, "freed: skip"); 4091 goto unlock; 4092 } 4093 if (!buffer_mapped(bh)) { 4094 BUFFER_TRACE(bh, "unmapped"); 4095 ext4_get_block(inode, iblock, bh, 0); 4096 /* unmapped? It's a hole - nothing to do */ 4097 if (!buffer_mapped(bh)) { 4098 BUFFER_TRACE(bh, "still unmapped"); 4099 goto unlock; 4100 } 4101 } 4102 4103 /* Ok, it's mapped. Make sure it's up-to-date */ 4104 if (folio_test_uptodate(folio)) 4105 set_buffer_uptodate(bh); 4106 4107 if (!buffer_uptodate(bh)) { 4108 err = ext4_read_bh_lock(bh, 0, true); 4109 if (err) 4110 goto unlock; 4111 if (fscrypt_inode_uses_fs_layer_crypto(inode)) { 4112 /* We expect the key to be set. */ 4113 BUG_ON(!fscrypt_has_encryption_key(inode)); 4114 err = fscrypt_decrypt_pagecache_blocks(folio, 4115 blocksize, 4116 bh_offset(bh)); 4117 if (err) { 4118 clear_buffer_uptodate(bh); 4119 goto unlock; 4120 } 4121 } 4122 } 4123 if (ext4_should_journal_data(inode)) { 4124 BUFFER_TRACE(bh, "get write access"); 4125 err = ext4_journal_get_write_access(handle, inode->i_sb, bh, 4126 EXT4_JTR_NONE); 4127 if (err) 4128 goto unlock; 4129 } 4130 folio_zero_range(folio, offset, length); 4131 BUFFER_TRACE(bh, "zeroed end of block"); 4132 4133 if (ext4_should_journal_data(inode)) { 4134 err = ext4_dirty_journalled_data(handle, bh); 4135 } else { 4136 err = 0; 4137 mark_buffer_dirty(bh); 4138 if (ext4_should_order_data(inode)) 4139 err = ext4_jbd2_inode_add_write(handle, inode, from, 4140 length); 4141 } 4142 4143 unlock: 4144 folio_unlock(folio); 4145 folio_put(folio); 4146 return err; 4147 } 4148 4149 /* 4150 * ext4_block_zero_page_range() zeros out a mapping of length 'length' 4151 * starting from file offset 'from'. The range to be zero'd must 4152 * be contained with in one block. If the specified range exceeds 4153 * the end of the block it will be shortened to end of the block 4154 * that corresponds to 'from' 4155 */ 4156 static int ext4_block_zero_page_range(handle_t *handle, 4157 struct address_space *mapping, loff_t from, loff_t length) 4158 { 4159 struct inode *inode = mapping->host; 4160 unsigned offset = from & (PAGE_SIZE-1); 4161 unsigned blocksize = inode->i_sb->s_blocksize; 4162 unsigned max = blocksize - (offset & (blocksize - 1)); 4163 4164 /* 4165 * correct length if it does not fall between 4166 * 'from' and the end of the block 4167 */ 4168 if (length > max || length < 0) 4169 length = max; 4170 4171 if (IS_DAX(inode)) { 4172 return dax_zero_range(inode, from, length, NULL, 4173 &ext4_iomap_ops); 4174 } 4175 return __ext4_block_zero_page_range(handle, mapping, from, length); 4176 } 4177 4178 /* 4179 * ext4_block_truncate_page() zeroes out a mapping from file offset `from' 4180 * up to the end of the block which corresponds to `from'. 4181 * This required during truncate. We need to physically zero the tail end 4182 * of that block so it doesn't yield old data if the file is later grown. 4183 */ 4184 static int ext4_block_truncate_page(handle_t *handle, 4185 struct address_space *mapping, loff_t from) 4186 { 4187 unsigned offset = from & (PAGE_SIZE-1); 4188 unsigned length; 4189 unsigned blocksize; 4190 struct inode *inode = mapping->host; 4191 4192 /* If we are processing an encrypted inode during orphan list handling */ 4193 if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode)) 4194 return 0; 4195 4196 blocksize = inode->i_sb->s_blocksize; 4197 length = blocksize - (offset & (blocksize - 1)); 4198 4199 return ext4_block_zero_page_range(handle, mapping, from, length); 4200 } 4201 4202 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode, 4203 loff_t lstart, loff_t length) 4204 { 4205 struct super_block *sb = inode->i_sb; 4206 struct address_space *mapping = inode->i_mapping; 4207 unsigned partial_start, partial_end; 4208 ext4_fsblk_t start, end; 4209 loff_t byte_end = (lstart + length - 1); 4210 int err = 0; 4211 4212 partial_start = lstart & (sb->s_blocksize - 1); 4213 partial_end = byte_end & (sb->s_blocksize - 1); 4214 4215 start = lstart >> sb->s_blocksize_bits; 4216 end = byte_end >> sb->s_blocksize_bits; 4217 4218 /* Handle partial zero within the single block */ 4219 if (start == end && 4220 (partial_start || (partial_end != sb->s_blocksize - 1))) { 4221 err = ext4_block_zero_page_range(handle, mapping, 4222 lstart, length); 4223 return err; 4224 } 4225 /* Handle partial zero out on the start of the range */ 4226 if (partial_start) { 4227 err = ext4_block_zero_page_range(handle, mapping, 4228 lstart, sb->s_blocksize); 4229 if (err) 4230 return err; 4231 } 4232 /* Handle partial zero out on the end of the range */ 4233 if (partial_end != sb->s_blocksize - 1) 4234 err = ext4_block_zero_page_range(handle, mapping, 4235 byte_end - partial_end, 4236 partial_end + 1); 4237 return err; 4238 } 4239 4240 int ext4_can_truncate(struct inode *inode) 4241 { 4242 if (S_ISREG(inode->i_mode)) 4243 return 1; 4244 if (S_ISDIR(inode->i_mode)) 4245 return 1; 4246 if (S_ISLNK(inode->i_mode)) 4247 return !ext4_inode_is_fast_symlink(inode); 4248 return 0; 4249 } 4250 4251 /* 4252 * We have to make sure i_disksize gets properly updated before we truncate 4253 * page cache due to hole punching or zero range. Otherwise i_disksize update 4254 * can get lost as it may have been postponed to submission of writeback but 4255 * that will never happen if we remove the folio containing i_size from the 4256 * page cache. Also if we punch hole within i_size but above i_disksize, 4257 * following ext4_page_mkwrite() may mistakenly allocate written blocks over 4258 * the hole and thus introduce allocated blocks beyond i_disksize which is 4259 * not allowed (e2fsck would complain in case of crash). 4260 */ 4261 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset, 4262 loff_t len) 4263 { 4264 handle_t *handle; 4265 int ret; 4266 4267 loff_t size = i_size_read(inode); 4268 4269 WARN_ON(!inode_is_locked(inode)); 4270 if (offset > size) 4271 return 0; 4272 4273 if (offset + len < size) 4274 size = offset + len; 4275 if (EXT4_I(inode)->i_disksize >= size) 4276 return 0; 4277 4278 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1); 4279 if (IS_ERR(handle)) 4280 return PTR_ERR(handle); 4281 ext4_update_i_disksize(inode, size); 4282 ret = ext4_mark_inode_dirty(handle, inode); 4283 ext4_journal_stop(handle); 4284 4285 return ret; 4286 } 4287 4288 static inline void ext4_truncate_folio(struct inode *inode, 4289 loff_t start, loff_t end) 4290 { 4291 unsigned long blocksize = i_blocksize(inode); 4292 struct folio *folio; 4293 4294 /* Nothing to be done if no complete block needs to be truncated. */ 4295 if (round_up(start, blocksize) >= round_down(end, blocksize)) 4296 return; 4297 4298 folio = filemap_lock_folio(inode->i_mapping, start >> PAGE_SHIFT); 4299 if (IS_ERR(folio)) 4300 return; 4301 4302 if (folio_mkclean(folio)) 4303 folio_mark_dirty(folio); 4304 folio_unlock(folio); 4305 folio_put(folio); 4306 } 4307 4308 int ext4_truncate_page_cache_block_range(struct inode *inode, 4309 loff_t start, loff_t end) 4310 { 4311 unsigned long blocksize = i_blocksize(inode); 4312 int ret; 4313 4314 /* 4315 * For journalled data we need to write (and checkpoint) pages 4316 * before discarding page cache to avoid inconsitent data on disk 4317 * in case of crash before freeing or unwritten converting trans 4318 * is committed. 4319 */ 4320 if (ext4_should_journal_data(inode)) { 4321 ret = filemap_write_and_wait_range(inode->i_mapping, start, 4322 end - 1); 4323 if (ret) 4324 return ret; 4325 goto truncate_pagecache; 4326 } 4327 4328 /* 4329 * If the block size is less than the page size, the file's mapped 4330 * blocks within one page could be freed or converted to unwritten. 4331 * So it's necessary to remove writable userspace mappings, and then 4332 * ext4_page_mkwrite() can be called during subsequent write access 4333 * to these partial folios. 4334 */ 4335 if (!IS_ALIGNED(start | end, PAGE_SIZE) && 4336 blocksize < PAGE_SIZE && start < inode->i_size) { 4337 loff_t page_boundary = round_up(start, PAGE_SIZE); 4338 4339 ext4_truncate_folio(inode, start, min(page_boundary, end)); 4340 if (end > page_boundary) 4341 ext4_truncate_folio(inode, 4342 round_down(end, PAGE_SIZE), end); 4343 } 4344 4345 truncate_pagecache: 4346 truncate_pagecache_range(inode, start, end - 1); 4347 return 0; 4348 } 4349 4350 static void ext4_wait_dax_page(struct inode *inode) 4351 { 4352 filemap_invalidate_unlock(inode->i_mapping); 4353 schedule(); 4354 filemap_invalidate_lock(inode->i_mapping); 4355 } 4356 4357 int ext4_break_layouts(struct inode *inode) 4358 { 4359 if (WARN_ON_ONCE(!rwsem_is_locked(&inode->i_mapping->invalidate_lock))) 4360 return -EINVAL; 4361 4362 return dax_break_layout_inode(inode, ext4_wait_dax_page); 4363 } 4364 4365 /* 4366 * ext4_punch_hole: punches a hole in a file by releasing the blocks 4367 * associated with the given offset and length 4368 * 4369 * @inode: File inode 4370 * @offset: The offset where the hole will begin 4371 * @len: The length of the hole 4372 * 4373 * Returns: 0 on success or negative on failure 4374 */ 4375 4376 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length) 4377 { 4378 struct inode *inode = file_inode(file); 4379 struct super_block *sb = inode->i_sb; 4380 ext4_lblk_t start_lblk, end_lblk; 4381 loff_t max_end = sb->s_maxbytes; 4382 loff_t end = offset + length; 4383 handle_t *handle; 4384 unsigned int credits; 4385 int ret; 4386 4387 trace_ext4_punch_hole(inode, offset, length, 0); 4388 WARN_ON_ONCE(!inode_is_locked(inode)); 4389 4390 /* 4391 * For indirect-block based inodes, make sure that the hole within 4392 * one block before last range. 4393 */ 4394 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) 4395 max_end = EXT4_SB(sb)->s_bitmap_maxbytes - sb->s_blocksize; 4396 4397 /* No need to punch hole beyond i_size */ 4398 if (offset >= inode->i_size || offset >= max_end) 4399 return 0; 4400 4401 /* 4402 * If the hole extends beyond i_size, set the hole to end after 4403 * the page that contains i_size. 4404 */ 4405 if (end > inode->i_size) 4406 end = round_up(inode->i_size, PAGE_SIZE); 4407 if (end > max_end) 4408 end = max_end; 4409 length = end - offset; 4410 4411 /* 4412 * Attach jinode to inode for jbd2 if we do any zeroing of partial 4413 * block. 4414 */ 4415 if (!IS_ALIGNED(offset | end, sb->s_blocksize)) { 4416 ret = ext4_inode_attach_jinode(inode); 4417 if (ret < 0) 4418 return ret; 4419 } 4420 4421 4422 ret = ext4_update_disksize_before_punch(inode, offset, length); 4423 if (ret) 4424 return ret; 4425 4426 /* Now release the pages and zero block aligned part of pages*/ 4427 ret = ext4_truncate_page_cache_block_range(inode, offset, end); 4428 if (ret) 4429 return ret; 4430 4431 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) 4432 credits = ext4_chunk_trans_extent(inode, 2); 4433 else 4434 credits = ext4_blocks_for_truncate(inode); 4435 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); 4436 if (IS_ERR(handle)) { 4437 ret = PTR_ERR(handle); 4438 ext4_std_error(sb, ret); 4439 return ret; 4440 } 4441 4442 ret = ext4_zero_partial_blocks(handle, inode, offset, length); 4443 if (ret) 4444 goto out_handle; 4445 4446 /* If there are blocks to remove, do it */ 4447 start_lblk = EXT4_B_TO_LBLK(inode, offset); 4448 end_lblk = end >> inode->i_blkbits; 4449 4450 if (end_lblk > start_lblk) { 4451 ext4_lblk_t hole_len = end_lblk - start_lblk; 4452 4453 ext4_fc_track_inode(handle, inode); 4454 ext4_check_map_extents_env(inode); 4455 down_write(&EXT4_I(inode)->i_data_sem); 4456 ext4_discard_preallocations(inode); 4457 4458 ext4_es_remove_extent(inode, start_lblk, hole_len); 4459 4460 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) 4461 ret = ext4_ext_remove_space(inode, start_lblk, 4462 end_lblk - 1); 4463 else 4464 ret = ext4_ind_remove_space(handle, inode, start_lblk, 4465 end_lblk); 4466 if (ret) { 4467 up_write(&EXT4_I(inode)->i_data_sem); 4468 goto out_handle; 4469 } 4470 4471 ext4_es_insert_extent(inode, start_lblk, hole_len, ~0, 4472 EXTENT_STATUS_HOLE, 0); 4473 up_write(&EXT4_I(inode)->i_data_sem); 4474 } 4475 ext4_fc_track_range(handle, inode, start_lblk, end_lblk); 4476 4477 ret = ext4_mark_inode_dirty(handle, inode); 4478 if (unlikely(ret)) 4479 goto out_handle; 4480 4481 ext4_update_inode_fsync_trans(handle, inode, 1); 4482 if (IS_SYNC(inode)) 4483 ext4_handle_sync(handle); 4484 out_handle: 4485 ext4_journal_stop(handle); 4486 return ret; 4487 } 4488 4489 int ext4_inode_attach_jinode(struct inode *inode) 4490 { 4491 struct ext4_inode_info *ei = EXT4_I(inode); 4492 struct jbd2_inode *jinode; 4493 4494 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal) 4495 return 0; 4496 4497 jinode = jbd2_alloc_inode(GFP_KERNEL); 4498 spin_lock(&inode->i_lock); 4499 if (!ei->jinode) { 4500 if (!jinode) { 4501 spin_unlock(&inode->i_lock); 4502 return -ENOMEM; 4503 } 4504 ei->jinode = jinode; 4505 jbd2_journal_init_jbd_inode(ei->jinode, inode); 4506 jinode = NULL; 4507 } 4508 spin_unlock(&inode->i_lock); 4509 if (unlikely(jinode != NULL)) 4510 jbd2_free_inode(jinode); 4511 return 0; 4512 } 4513 4514 /* 4515 * ext4_truncate() 4516 * 4517 * We block out ext4_get_block() block instantiations across the entire 4518 * transaction, and VFS/VM ensures that ext4_truncate() cannot run 4519 * simultaneously on behalf of the same inode. 4520 * 4521 * As we work through the truncate and commit bits of it to the journal there 4522 * is one core, guiding principle: the file's tree must always be consistent on 4523 * disk. We must be able to restart the truncate after a crash. 4524 * 4525 * The file's tree may be transiently inconsistent in memory (although it 4526 * probably isn't), but whenever we close off and commit a journal transaction, 4527 * the contents of (the filesystem + the journal) must be consistent and 4528 * restartable. It's pretty simple, really: bottom up, right to left (although 4529 * left-to-right works OK too). 4530 * 4531 * Note that at recovery time, journal replay occurs *before* the restart of 4532 * truncate against the orphan inode list. 4533 * 4534 * The committed inode has the new, desired i_size (which is the same as 4535 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see 4536 * that this inode's truncate did not complete and it will again call 4537 * ext4_truncate() to have another go. So there will be instantiated blocks 4538 * to the right of the truncation point in a crashed ext4 filesystem. But 4539 * that's fine - as long as they are linked from the inode, the post-crash 4540 * ext4_truncate() run will find them and release them. 4541 */ 4542 int ext4_truncate(struct inode *inode) 4543 { 4544 struct ext4_inode_info *ei = EXT4_I(inode); 4545 unsigned int credits; 4546 int err = 0, err2; 4547 handle_t *handle; 4548 struct address_space *mapping = inode->i_mapping; 4549 4550 /* 4551 * There is a possibility that we're either freeing the inode 4552 * or it's a completely new inode. In those cases we might not 4553 * have i_rwsem locked because it's not necessary. 4554 */ 4555 if (!(inode->i_state & (I_NEW|I_FREEING))) 4556 WARN_ON(!inode_is_locked(inode)); 4557 trace_ext4_truncate_enter(inode); 4558 4559 if (!ext4_can_truncate(inode)) 4560 goto out_trace; 4561 4562 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) 4563 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); 4564 4565 if (ext4_has_inline_data(inode)) { 4566 int has_inline = 1; 4567 4568 err = ext4_inline_data_truncate(inode, &has_inline); 4569 if (err || has_inline) 4570 goto out_trace; 4571 } 4572 4573 /* If we zero-out tail of the page, we have to create jinode for jbd2 */ 4574 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) { 4575 err = ext4_inode_attach_jinode(inode); 4576 if (err) 4577 goto out_trace; 4578 } 4579 4580 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) 4581 credits = ext4_chunk_trans_extent(inode, 1); 4582 else 4583 credits = ext4_blocks_for_truncate(inode); 4584 4585 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); 4586 if (IS_ERR(handle)) { 4587 err = PTR_ERR(handle); 4588 goto out_trace; 4589 } 4590 4591 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) 4592 ext4_block_truncate_page(handle, mapping, inode->i_size); 4593 4594 /* 4595 * We add the inode to the orphan list, so that if this 4596 * truncate spans multiple transactions, and we crash, we will 4597 * resume the truncate when the filesystem recovers. It also 4598 * marks the inode dirty, to catch the new size. 4599 * 4600 * Implication: the file must always be in a sane, consistent 4601 * truncatable state while each transaction commits. 4602 */ 4603 err = ext4_orphan_add(handle, inode); 4604 if (err) 4605 goto out_stop; 4606 4607 ext4_fc_track_inode(handle, inode); 4608 ext4_check_map_extents_env(inode); 4609 4610 down_write(&EXT4_I(inode)->i_data_sem); 4611 ext4_discard_preallocations(inode); 4612 4613 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) 4614 err = ext4_ext_truncate(handle, inode); 4615 else 4616 ext4_ind_truncate(handle, inode); 4617 4618 up_write(&ei->i_data_sem); 4619 if (err) 4620 goto out_stop; 4621 4622 if (IS_SYNC(inode)) 4623 ext4_handle_sync(handle); 4624 4625 out_stop: 4626 /* 4627 * If this was a simple ftruncate() and the file will remain alive, 4628 * then we need to clear up the orphan record which we created above. 4629 * However, if this was a real unlink then we were called by 4630 * ext4_evict_inode(), and we allow that function to clean up the 4631 * orphan info for us. 4632 */ 4633 if (inode->i_nlink) 4634 ext4_orphan_del(handle, inode); 4635 4636 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode)); 4637 err2 = ext4_mark_inode_dirty(handle, inode); 4638 if (unlikely(err2 && !err)) 4639 err = err2; 4640 ext4_journal_stop(handle); 4641 4642 out_trace: 4643 trace_ext4_truncate_exit(inode); 4644 return err; 4645 } 4646 4647 static inline u64 ext4_inode_peek_iversion(const struct inode *inode) 4648 { 4649 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) 4650 return inode_peek_iversion_raw(inode); 4651 else 4652 return inode_peek_iversion(inode); 4653 } 4654 4655 static int ext4_inode_blocks_set(struct ext4_inode *raw_inode, 4656 struct ext4_inode_info *ei) 4657 { 4658 struct inode *inode = &(ei->vfs_inode); 4659 u64 i_blocks = READ_ONCE(inode->i_blocks); 4660 struct super_block *sb = inode->i_sb; 4661 4662 if (i_blocks <= ~0U) { 4663 /* 4664 * i_blocks can be represented in a 32 bit variable 4665 * as multiple of 512 bytes 4666 */ 4667 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); 4668 raw_inode->i_blocks_high = 0; 4669 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); 4670 return 0; 4671 } 4672 4673 /* 4674 * This should never happen since sb->s_maxbytes should not have 4675 * allowed this, sb->s_maxbytes was set according to the huge_file 4676 * feature in ext4_fill_super(). 4677 */ 4678 if (!ext4_has_feature_huge_file(sb)) 4679 return -EFSCORRUPTED; 4680 4681 if (i_blocks <= 0xffffffffffffULL) { 4682 /* 4683 * i_blocks can be represented in a 48 bit variable 4684 * as multiple of 512 bytes 4685 */ 4686 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); 4687 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); 4688 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); 4689 } else { 4690 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); 4691 /* i_block is stored in file system block size */ 4692 i_blocks = i_blocks >> (inode->i_blkbits - 9); 4693 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); 4694 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); 4695 } 4696 return 0; 4697 } 4698 4699 static int ext4_fill_raw_inode(struct inode *inode, struct ext4_inode *raw_inode) 4700 { 4701 struct ext4_inode_info *ei = EXT4_I(inode); 4702 uid_t i_uid; 4703 gid_t i_gid; 4704 projid_t i_projid; 4705 int block; 4706 int err; 4707 4708 err = ext4_inode_blocks_set(raw_inode, ei); 4709 4710 raw_inode->i_mode = cpu_to_le16(inode->i_mode); 4711 i_uid = i_uid_read(inode); 4712 i_gid = i_gid_read(inode); 4713 i_projid = from_kprojid(&init_user_ns, ei->i_projid); 4714 if (!(test_opt(inode->i_sb, NO_UID32))) { 4715 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid)); 4716 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid)); 4717 /* 4718 * Fix up interoperability with old kernels. Otherwise, 4719 * old inodes get re-used with the upper 16 bits of the 4720 * uid/gid intact. 4721 */ 4722 if (ei->i_dtime && !ext4_inode_orphan_tracked(inode)) { 4723 raw_inode->i_uid_high = 0; 4724 raw_inode->i_gid_high = 0; 4725 } else { 4726 raw_inode->i_uid_high = 4727 cpu_to_le16(high_16_bits(i_uid)); 4728 raw_inode->i_gid_high = 4729 cpu_to_le16(high_16_bits(i_gid)); 4730 } 4731 } else { 4732 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid)); 4733 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid)); 4734 raw_inode->i_uid_high = 0; 4735 raw_inode->i_gid_high = 0; 4736 } 4737 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); 4738 4739 EXT4_INODE_SET_CTIME(inode, raw_inode); 4740 EXT4_INODE_SET_MTIME(inode, raw_inode); 4741 EXT4_INODE_SET_ATIME(inode, raw_inode); 4742 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); 4743 4744 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); 4745 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF); 4746 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) 4747 raw_inode->i_file_acl_high = 4748 cpu_to_le16(ei->i_file_acl >> 32); 4749 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); 4750 ext4_isize_set(raw_inode, ei->i_disksize); 4751 4752 raw_inode->i_generation = cpu_to_le32(inode->i_generation); 4753 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { 4754 if (old_valid_dev(inode->i_rdev)) { 4755 raw_inode->i_block[0] = 4756 cpu_to_le32(old_encode_dev(inode->i_rdev)); 4757 raw_inode->i_block[1] = 0; 4758 } else { 4759 raw_inode->i_block[0] = 0; 4760 raw_inode->i_block[1] = 4761 cpu_to_le32(new_encode_dev(inode->i_rdev)); 4762 raw_inode->i_block[2] = 0; 4763 } 4764 } else if (!ext4_has_inline_data(inode)) { 4765 for (block = 0; block < EXT4_N_BLOCKS; block++) 4766 raw_inode->i_block[block] = ei->i_data[block]; 4767 } 4768 4769 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { 4770 u64 ivers = ext4_inode_peek_iversion(inode); 4771 4772 raw_inode->i_disk_version = cpu_to_le32(ivers); 4773 if (ei->i_extra_isize) { 4774 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) 4775 raw_inode->i_version_hi = 4776 cpu_to_le32(ivers >> 32); 4777 raw_inode->i_extra_isize = 4778 cpu_to_le16(ei->i_extra_isize); 4779 } 4780 } 4781 4782 if (i_projid != EXT4_DEF_PROJID && 4783 !ext4_has_feature_project(inode->i_sb)) 4784 err = err ?: -EFSCORRUPTED; 4785 4786 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && 4787 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid)) 4788 raw_inode->i_projid = cpu_to_le32(i_projid); 4789 4790 ext4_inode_csum_set(inode, raw_inode, ei); 4791 return err; 4792 } 4793 4794 /* 4795 * ext4_get_inode_loc returns with an extra refcount against the inode's 4796 * underlying buffer_head on success. If we pass 'inode' and it does not 4797 * have in-inode xattr, we have all inode data in memory that is needed 4798 * to recreate the on-disk version of this inode. 4799 */ 4800 static int __ext4_get_inode_loc(struct super_block *sb, unsigned long ino, 4801 struct inode *inode, struct ext4_iloc *iloc, 4802 ext4_fsblk_t *ret_block) 4803 { 4804 struct ext4_group_desc *gdp; 4805 struct buffer_head *bh; 4806 ext4_fsblk_t block; 4807 struct blk_plug plug; 4808 int inodes_per_block, inode_offset; 4809 4810 iloc->bh = NULL; 4811 if (ino < EXT4_ROOT_INO || 4812 ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)) 4813 return -EFSCORRUPTED; 4814 4815 iloc->block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); 4816 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); 4817 if (!gdp) 4818 return -EIO; 4819 4820 /* 4821 * Figure out the offset within the block group inode table 4822 */ 4823 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; 4824 inode_offset = ((ino - 1) % 4825 EXT4_INODES_PER_GROUP(sb)); 4826 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); 4827 4828 block = ext4_inode_table(sb, gdp); 4829 if ((block <= le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) || 4830 (block >= ext4_blocks_count(EXT4_SB(sb)->s_es))) { 4831 ext4_error(sb, "Invalid inode table block %llu in " 4832 "block_group %u", block, iloc->block_group); 4833 return -EFSCORRUPTED; 4834 } 4835 block += (inode_offset / inodes_per_block); 4836 4837 bh = sb_getblk(sb, block); 4838 if (unlikely(!bh)) 4839 return -ENOMEM; 4840 if (ext4_buffer_uptodate(bh)) 4841 goto has_buffer; 4842 4843 lock_buffer(bh); 4844 if (ext4_buffer_uptodate(bh)) { 4845 /* Someone brought it uptodate while we waited */ 4846 unlock_buffer(bh); 4847 goto has_buffer; 4848 } 4849 4850 /* 4851 * If we have all information of the inode in memory and this 4852 * is the only valid inode in the block, we need not read the 4853 * block. 4854 */ 4855 if (inode && !ext4_test_inode_state(inode, EXT4_STATE_XATTR)) { 4856 struct buffer_head *bitmap_bh; 4857 int i, start; 4858 4859 start = inode_offset & ~(inodes_per_block - 1); 4860 4861 /* Is the inode bitmap in cache? */ 4862 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); 4863 if (unlikely(!bitmap_bh)) 4864 goto make_io; 4865 4866 /* 4867 * If the inode bitmap isn't in cache then the 4868 * optimisation may end up performing two reads instead 4869 * of one, so skip it. 4870 */ 4871 if (!buffer_uptodate(bitmap_bh)) { 4872 brelse(bitmap_bh); 4873 goto make_io; 4874 } 4875 for (i = start; i < start + inodes_per_block; i++) { 4876 if (i == inode_offset) 4877 continue; 4878 if (ext4_test_bit(i, bitmap_bh->b_data)) 4879 break; 4880 } 4881 brelse(bitmap_bh); 4882 if (i == start + inodes_per_block) { 4883 struct ext4_inode *raw_inode = 4884 (struct ext4_inode *) (bh->b_data + iloc->offset); 4885 4886 /* all other inodes are free, so skip I/O */ 4887 memset(bh->b_data, 0, bh->b_size); 4888 if (!ext4_test_inode_state(inode, EXT4_STATE_NEW)) 4889 ext4_fill_raw_inode(inode, raw_inode); 4890 set_buffer_uptodate(bh); 4891 unlock_buffer(bh); 4892 goto has_buffer; 4893 } 4894 } 4895 4896 make_io: 4897 /* 4898 * If we need to do any I/O, try to pre-readahead extra 4899 * blocks from the inode table. 4900 */ 4901 blk_start_plug(&plug); 4902 if (EXT4_SB(sb)->s_inode_readahead_blks) { 4903 ext4_fsblk_t b, end, table; 4904 unsigned num; 4905 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks; 4906 4907 table = ext4_inode_table(sb, gdp); 4908 /* s_inode_readahead_blks is always a power of 2 */ 4909 b = block & ~((ext4_fsblk_t) ra_blks - 1); 4910 if (table > b) 4911 b = table; 4912 end = b + ra_blks; 4913 num = EXT4_INODES_PER_GROUP(sb); 4914 if (ext4_has_group_desc_csum(sb)) 4915 num -= ext4_itable_unused_count(sb, gdp); 4916 table += num / inodes_per_block; 4917 if (end > table) 4918 end = table; 4919 while (b <= end) 4920 ext4_sb_breadahead_unmovable(sb, b++); 4921 } 4922 4923 /* 4924 * There are other valid inodes in the buffer, this inode 4925 * has in-inode xattrs, or we don't have this inode in memory. 4926 * Read the block from disk. 4927 */ 4928 trace_ext4_load_inode(sb, ino); 4929 ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO, NULL, 4930 ext4_simulate_fail(sb, EXT4_SIM_INODE_EIO)); 4931 blk_finish_plug(&plug); 4932 wait_on_buffer(bh); 4933 if (!buffer_uptodate(bh)) { 4934 if (ret_block) 4935 *ret_block = block; 4936 brelse(bh); 4937 return -EIO; 4938 } 4939 has_buffer: 4940 iloc->bh = bh; 4941 return 0; 4942 } 4943 4944 static int __ext4_get_inode_loc_noinmem(struct inode *inode, 4945 struct ext4_iloc *iloc) 4946 { 4947 ext4_fsblk_t err_blk = 0; 4948 int ret; 4949 4950 ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, NULL, iloc, 4951 &err_blk); 4952 4953 if (ret == -EIO) 4954 ext4_error_inode_block(inode, err_blk, EIO, 4955 "unable to read itable block"); 4956 4957 return ret; 4958 } 4959 4960 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) 4961 { 4962 ext4_fsblk_t err_blk = 0; 4963 int ret; 4964 4965 ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, inode, iloc, 4966 &err_blk); 4967 4968 if (ret == -EIO) 4969 ext4_error_inode_block(inode, err_blk, EIO, 4970 "unable to read itable block"); 4971 4972 return ret; 4973 } 4974 4975 4976 int ext4_get_fc_inode_loc(struct super_block *sb, unsigned long ino, 4977 struct ext4_iloc *iloc) 4978 { 4979 return __ext4_get_inode_loc(sb, ino, NULL, iloc, NULL); 4980 } 4981 4982 static bool ext4_should_enable_dax(struct inode *inode) 4983 { 4984 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 4985 4986 if (test_opt2(inode->i_sb, DAX_NEVER)) 4987 return false; 4988 if (!S_ISREG(inode->i_mode)) 4989 return false; 4990 if (ext4_should_journal_data(inode)) 4991 return false; 4992 if (ext4_has_inline_data(inode)) 4993 return false; 4994 if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT)) 4995 return false; 4996 if (ext4_test_inode_flag(inode, EXT4_INODE_VERITY)) 4997 return false; 4998 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) 4999 return false; 5000 if (test_opt(inode->i_sb, DAX_ALWAYS)) 5001 return true; 5002 5003 return ext4_test_inode_flag(inode, EXT4_INODE_DAX); 5004 } 5005 5006 void ext4_set_inode_flags(struct inode *inode, bool init) 5007 { 5008 unsigned int flags = EXT4_I(inode)->i_flags; 5009 unsigned int new_fl = 0; 5010 5011 WARN_ON_ONCE(IS_DAX(inode) && init); 5012 5013 if (flags & EXT4_SYNC_FL) 5014 new_fl |= S_SYNC; 5015 if (flags & EXT4_APPEND_FL) 5016 new_fl |= S_APPEND; 5017 if (flags & EXT4_IMMUTABLE_FL) 5018 new_fl |= S_IMMUTABLE; 5019 if (flags & EXT4_NOATIME_FL) 5020 new_fl |= S_NOATIME; 5021 if (flags & EXT4_DIRSYNC_FL) 5022 new_fl |= S_DIRSYNC; 5023 5024 /* Because of the way inode_set_flags() works we must preserve S_DAX 5025 * here if already set. */ 5026 new_fl |= (inode->i_flags & S_DAX); 5027 if (init && ext4_should_enable_dax(inode)) 5028 new_fl |= S_DAX; 5029 5030 if (flags & EXT4_ENCRYPT_FL) 5031 new_fl |= S_ENCRYPTED; 5032 if (flags & EXT4_CASEFOLD_FL) 5033 new_fl |= S_CASEFOLD; 5034 if (flags & EXT4_VERITY_FL) 5035 new_fl |= S_VERITY; 5036 inode_set_flags(inode, new_fl, 5037 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX| 5038 S_ENCRYPTED|S_CASEFOLD|S_VERITY); 5039 } 5040 5041 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, 5042 struct ext4_inode_info *ei) 5043 { 5044 blkcnt_t i_blocks ; 5045 struct inode *inode = &(ei->vfs_inode); 5046 struct super_block *sb = inode->i_sb; 5047 5048 if (ext4_has_feature_huge_file(sb)) { 5049 /* we are using combined 48 bit field */ 5050 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | 5051 le32_to_cpu(raw_inode->i_blocks_lo); 5052 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { 5053 /* i_blocks represent file system block size */ 5054 return i_blocks << (inode->i_blkbits - 9); 5055 } else { 5056 return i_blocks; 5057 } 5058 } else { 5059 return le32_to_cpu(raw_inode->i_blocks_lo); 5060 } 5061 } 5062 5063 static inline int ext4_iget_extra_inode(struct inode *inode, 5064 struct ext4_inode *raw_inode, 5065 struct ext4_inode_info *ei) 5066 { 5067 __le32 *magic = (void *)raw_inode + 5068 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize; 5069 5070 if (EXT4_INODE_HAS_XATTR_SPACE(inode) && 5071 *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) { 5072 int err; 5073 5074 err = xattr_check_inode(inode, IHDR(inode, raw_inode), 5075 ITAIL(inode, raw_inode)); 5076 if (err) 5077 return err; 5078 5079 ext4_set_inode_state(inode, EXT4_STATE_XATTR); 5080 err = ext4_find_inline_data_nolock(inode); 5081 if (!err && ext4_has_inline_data(inode)) 5082 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); 5083 return err; 5084 } else 5085 EXT4_I(inode)->i_inline_off = 0; 5086 return 0; 5087 } 5088 5089 int ext4_get_projid(struct inode *inode, kprojid_t *projid) 5090 { 5091 if (!ext4_has_feature_project(inode->i_sb)) 5092 return -EOPNOTSUPP; 5093 *projid = EXT4_I(inode)->i_projid; 5094 return 0; 5095 } 5096 5097 /* 5098 * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of 5099 * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag 5100 * set. 5101 */ 5102 static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val) 5103 { 5104 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) 5105 inode_set_iversion_raw(inode, val); 5106 else 5107 inode_set_iversion_queried(inode, val); 5108 } 5109 5110 static int check_igot_inode(struct inode *inode, ext4_iget_flags flags, 5111 const char *function, unsigned int line) 5112 { 5113 const char *err_str; 5114 5115 if (flags & EXT4_IGET_EA_INODE) { 5116 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) { 5117 err_str = "missing EA_INODE flag"; 5118 goto error; 5119 } 5120 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR) || 5121 EXT4_I(inode)->i_file_acl) { 5122 err_str = "ea_inode with extended attributes"; 5123 goto error; 5124 } 5125 } else { 5126 if ((EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) { 5127 /* 5128 * open_by_handle_at() could provide an old inode number 5129 * that has since been reused for an ea_inode; this does 5130 * not indicate filesystem corruption 5131 */ 5132 if (flags & EXT4_IGET_HANDLE) 5133 return -ESTALE; 5134 err_str = "unexpected EA_INODE flag"; 5135 goto error; 5136 } 5137 } 5138 if (is_bad_inode(inode) && !(flags & EXT4_IGET_BAD)) { 5139 err_str = "unexpected bad inode w/o EXT4_IGET_BAD"; 5140 goto error; 5141 } 5142 return 0; 5143 5144 error: 5145 ext4_error_inode(inode, function, line, 0, "%s", err_str); 5146 return -EFSCORRUPTED; 5147 } 5148 5149 static bool ext4_should_enable_large_folio(struct inode *inode) 5150 { 5151 struct super_block *sb = inode->i_sb; 5152 5153 if (!S_ISREG(inode->i_mode)) 5154 return false; 5155 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA || 5156 ext4_test_inode_flag(inode, EXT4_INODE_JOURNAL_DATA)) 5157 return false; 5158 if (ext4_has_feature_verity(sb)) 5159 return false; 5160 if (ext4_has_feature_encrypt(sb)) 5161 return false; 5162 5163 return true; 5164 } 5165 5166 /* 5167 * Limit the maximum folio order to 2048 blocks to prevent overestimation 5168 * of reserve handle credits during the folio writeback in environments 5169 * where the PAGE_SIZE exceeds 4KB. 5170 */ 5171 #define EXT4_MAX_PAGECACHE_ORDER(i) \ 5172 umin(MAX_PAGECACHE_ORDER, (11 + (i)->i_blkbits - PAGE_SHIFT)) 5173 void ext4_set_inode_mapping_order(struct inode *inode) 5174 { 5175 if (!ext4_should_enable_large_folio(inode)) 5176 return; 5177 5178 mapping_set_folio_order_range(inode->i_mapping, 0, 5179 EXT4_MAX_PAGECACHE_ORDER(inode)); 5180 } 5181 5182 struct inode *__ext4_iget(struct super_block *sb, unsigned long ino, 5183 ext4_iget_flags flags, const char *function, 5184 unsigned int line) 5185 { 5186 struct ext4_iloc iloc; 5187 struct ext4_inode *raw_inode; 5188 struct ext4_inode_info *ei; 5189 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 5190 struct inode *inode; 5191 journal_t *journal = EXT4_SB(sb)->s_journal; 5192 long ret; 5193 loff_t size; 5194 int block; 5195 uid_t i_uid; 5196 gid_t i_gid; 5197 projid_t i_projid; 5198 5199 if ((!(flags & EXT4_IGET_SPECIAL) && is_special_ino(sb, ino)) || 5200 (ino < EXT4_ROOT_INO) || 5201 (ino > le32_to_cpu(es->s_inodes_count))) { 5202 if (flags & EXT4_IGET_HANDLE) 5203 return ERR_PTR(-ESTALE); 5204 __ext4_error(sb, function, line, false, EFSCORRUPTED, 0, 5205 "inode #%lu: comm %s: iget: illegal inode #", 5206 ino, current->comm); 5207 return ERR_PTR(-EFSCORRUPTED); 5208 } 5209 5210 inode = iget_locked(sb, ino); 5211 if (!inode) 5212 return ERR_PTR(-ENOMEM); 5213 if (!(inode->i_state & I_NEW)) { 5214 ret = check_igot_inode(inode, flags, function, line); 5215 if (ret) { 5216 iput(inode); 5217 return ERR_PTR(ret); 5218 } 5219 return inode; 5220 } 5221 5222 ei = EXT4_I(inode); 5223 iloc.bh = NULL; 5224 5225 ret = __ext4_get_inode_loc_noinmem(inode, &iloc); 5226 if (ret < 0) 5227 goto bad_inode; 5228 raw_inode = ext4_raw_inode(&iloc); 5229 5230 if ((flags & EXT4_IGET_HANDLE) && 5231 (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) { 5232 ret = -ESTALE; 5233 goto bad_inode; 5234 } 5235 5236 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { 5237 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); 5238 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > 5239 EXT4_INODE_SIZE(inode->i_sb) || 5240 (ei->i_extra_isize & 3)) { 5241 ext4_error_inode(inode, function, line, 0, 5242 "iget: bad extra_isize %u " 5243 "(inode size %u)", 5244 ei->i_extra_isize, 5245 EXT4_INODE_SIZE(inode->i_sb)); 5246 ret = -EFSCORRUPTED; 5247 goto bad_inode; 5248 } 5249 } else 5250 ei->i_extra_isize = 0; 5251 5252 /* Precompute checksum seed for inode metadata */ 5253 if (ext4_has_feature_metadata_csum(sb)) { 5254 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 5255 __u32 csum; 5256 __le32 inum = cpu_to_le32(inode->i_ino); 5257 __le32 gen = raw_inode->i_generation; 5258 csum = ext4_chksum(sbi->s_csum_seed, (__u8 *)&inum, 5259 sizeof(inum)); 5260 ei->i_csum_seed = ext4_chksum(csum, (__u8 *)&gen, sizeof(gen)); 5261 } 5262 5263 if ((!ext4_inode_csum_verify(inode, raw_inode, ei) || 5264 ext4_simulate_fail(sb, EXT4_SIM_INODE_CRC)) && 5265 (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))) { 5266 ext4_error_inode_err(inode, function, line, 0, 5267 EFSBADCRC, "iget: checksum invalid"); 5268 ret = -EFSBADCRC; 5269 goto bad_inode; 5270 } 5271 5272 inode->i_mode = le16_to_cpu(raw_inode->i_mode); 5273 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); 5274 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); 5275 if (ext4_has_feature_project(sb) && 5276 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE && 5277 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid)) 5278 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid); 5279 else 5280 i_projid = EXT4_DEF_PROJID; 5281 5282 if (!(test_opt(inode->i_sb, NO_UID32))) { 5283 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; 5284 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; 5285 } 5286 i_uid_write(inode, i_uid); 5287 i_gid_write(inode, i_gid); 5288 ei->i_projid = make_kprojid(&init_user_ns, i_projid); 5289 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count)); 5290 5291 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ 5292 ei->i_inline_off = 0; 5293 ei->i_dir_start_lookup = 0; 5294 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); 5295 /* We now have enough fields to check if the inode was active or not. 5296 * This is needed because nfsd might try to access dead inodes 5297 * the test is that same one that e2fsck uses 5298 * NeilBrown 1999oct15 5299 */ 5300 if (inode->i_nlink == 0) { 5301 if ((inode->i_mode == 0 || flags & EXT4_IGET_SPECIAL || 5302 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) && 5303 ino != EXT4_BOOT_LOADER_INO) { 5304 /* this inode is deleted or unallocated */ 5305 if (flags & EXT4_IGET_SPECIAL) { 5306 ext4_error_inode(inode, function, line, 0, 5307 "iget: special inode unallocated"); 5308 ret = -EFSCORRUPTED; 5309 } else 5310 ret = -ESTALE; 5311 goto bad_inode; 5312 } 5313 /* The only unlinked inodes we let through here have 5314 * valid i_mode and are being read by the orphan 5315 * recovery code: that's fine, we're about to complete 5316 * the process of deleting those. 5317 * OR it is the EXT4_BOOT_LOADER_INO which is 5318 * not initialized on a new filesystem. */ 5319 } 5320 ei->i_flags = le32_to_cpu(raw_inode->i_flags); 5321 ext4_set_inode_flags(inode, true); 5322 inode->i_blocks = ext4_inode_blocks(raw_inode, ei); 5323 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); 5324 if (ext4_has_feature_64bit(sb)) 5325 ei->i_file_acl |= 5326 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; 5327 inode->i_size = ext4_isize(sb, raw_inode); 5328 size = i_size_read(inode); 5329 if (size < 0 || size > ext4_get_maxbytes(inode)) { 5330 ext4_error_inode(inode, function, line, 0, 5331 "iget: bad i_size value: %lld", size); 5332 ret = -EFSCORRUPTED; 5333 goto bad_inode; 5334 } 5335 /* 5336 * If dir_index is not enabled but there's dir with INDEX flag set, 5337 * we'd normally treat htree data as empty space. But with metadata 5338 * checksumming that corrupts checksums so forbid that. 5339 */ 5340 if (!ext4_has_feature_dir_index(sb) && 5341 ext4_has_feature_metadata_csum(sb) && 5342 ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) { 5343 ext4_error_inode(inode, function, line, 0, 5344 "iget: Dir with htree data on filesystem without dir_index feature."); 5345 ret = -EFSCORRUPTED; 5346 goto bad_inode; 5347 } 5348 ei->i_disksize = inode->i_size; 5349 #ifdef CONFIG_QUOTA 5350 ei->i_reserved_quota = 0; 5351 #endif 5352 inode->i_generation = le32_to_cpu(raw_inode->i_generation); 5353 ei->i_block_group = iloc.block_group; 5354 ei->i_last_alloc_group = ~0; 5355 /* 5356 * NOTE! The in-memory inode i_data array is in little-endian order 5357 * even on big-endian machines: we do NOT byteswap the block numbers! 5358 */ 5359 for (block = 0; block < EXT4_N_BLOCKS; block++) 5360 ei->i_data[block] = raw_inode->i_block[block]; 5361 INIT_LIST_HEAD(&ei->i_orphan); 5362 ext4_fc_init_inode(&ei->vfs_inode); 5363 5364 /* 5365 * Set transaction id's of transactions that have to be committed 5366 * to finish f[data]sync. We set them to currently running transaction 5367 * as we cannot be sure that the inode or some of its metadata isn't 5368 * part of the transaction - the inode could have been reclaimed and 5369 * now it is reread from disk. 5370 */ 5371 if (journal) { 5372 transaction_t *transaction; 5373 tid_t tid; 5374 5375 read_lock(&journal->j_state_lock); 5376 if (journal->j_running_transaction) 5377 transaction = journal->j_running_transaction; 5378 else 5379 transaction = journal->j_committing_transaction; 5380 if (transaction) 5381 tid = transaction->t_tid; 5382 else 5383 tid = journal->j_commit_sequence; 5384 read_unlock(&journal->j_state_lock); 5385 ei->i_sync_tid = tid; 5386 ei->i_datasync_tid = tid; 5387 } 5388 5389 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { 5390 if (ei->i_extra_isize == 0) { 5391 /* The extra space is currently unused. Use it. */ 5392 BUILD_BUG_ON(sizeof(struct ext4_inode) & 3); 5393 ei->i_extra_isize = sizeof(struct ext4_inode) - 5394 EXT4_GOOD_OLD_INODE_SIZE; 5395 } else { 5396 ret = ext4_iget_extra_inode(inode, raw_inode, ei); 5397 if (ret) 5398 goto bad_inode; 5399 } 5400 } 5401 5402 EXT4_INODE_GET_CTIME(inode, raw_inode); 5403 EXT4_INODE_GET_ATIME(inode, raw_inode); 5404 EXT4_INODE_GET_MTIME(inode, raw_inode); 5405 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); 5406 5407 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { 5408 u64 ivers = le32_to_cpu(raw_inode->i_disk_version); 5409 5410 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { 5411 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) 5412 ivers |= 5413 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; 5414 } 5415 ext4_inode_set_iversion_queried(inode, ivers); 5416 } 5417 5418 ret = 0; 5419 if (ei->i_file_acl && 5420 !ext4_inode_block_valid(inode, ei->i_file_acl, 1)) { 5421 ext4_error_inode(inode, function, line, 0, 5422 "iget: bad extended attribute block %llu", 5423 ei->i_file_acl); 5424 ret = -EFSCORRUPTED; 5425 goto bad_inode; 5426 } else if (!ext4_has_inline_data(inode)) { 5427 /* validate the block references in the inode */ 5428 if (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) && 5429 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 5430 (S_ISLNK(inode->i_mode) && 5431 !ext4_inode_is_fast_symlink(inode)))) { 5432 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) 5433 ret = ext4_ext_check_inode(inode); 5434 else 5435 ret = ext4_ind_check_inode(inode); 5436 } 5437 } 5438 if (ret) 5439 goto bad_inode; 5440 5441 if (S_ISREG(inode->i_mode)) { 5442 inode->i_op = &ext4_file_inode_operations; 5443 inode->i_fop = &ext4_file_operations; 5444 ext4_set_aops(inode); 5445 } else if (S_ISDIR(inode->i_mode)) { 5446 inode->i_op = &ext4_dir_inode_operations; 5447 inode->i_fop = &ext4_dir_operations; 5448 } else if (S_ISLNK(inode->i_mode)) { 5449 /* VFS does not allow setting these so must be corruption */ 5450 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) { 5451 ext4_error_inode(inode, function, line, 0, 5452 "iget: immutable or append flags " 5453 "not allowed on symlinks"); 5454 ret = -EFSCORRUPTED; 5455 goto bad_inode; 5456 } 5457 if (IS_ENCRYPTED(inode)) { 5458 inode->i_op = &ext4_encrypted_symlink_inode_operations; 5459 } else if (ext4_inode_is_fast_symlink(inode)) { 5460 inode->i_op = &ext4_fast_symlink_inode_operations; 5461 if (inode->i_size == 0 || 5462 inode->i_size >= sizeof(ei->i_data) || 5463 strnlen((char *)ei->i_data, inode->i_size + 1) != 5464 inode->i_size) { 5465 ext4_error_inode(inode, function, line, 0, 5466 "invalid fast symlink length %llu", 5467 (unsigned long long)inode->i_size); 5468 ret = -EFSCORRUPTED; 5469 goto bad_inode; 5470 } 5471 inode_set_cached_link(inode, (char *)ei->i_data, 5472 inode->i_size); 5473 } else { 5474 inode->i_op = &ext4_symlink_inode_operations; 5475 } 5476 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || 5477 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { 5478 inode->i_op = &ext4_special_inode_operations; 5479 if (raw_inode->i_block[0]) 5480 init_special_inode(inode, inode->i_mode, 5481 old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); 5482 else 5483 init_special_inode(inode, inode->i_mode, 5484 new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); 5485 } else if (ino == EXT4_BOOT_LOADER_INO) { 5486 make_bad_inode(inode); 5487 } else { 5488 ret = -EFSCORRUPTED; 5489 ext4_error_inode(inode, function, line, 0, 5490 "iget: bogus i_mode (%o)", inode->i_mode); 5491 goto bad_inode; 5492 } 5493 if (IS_CASEFOLDED(inode) && !ext4_has_feature_casefold(inode->i_sb)) { 5494 ext4_error_inode(inode, function, line, 0, 5495 "casefold flag without casefold feature"); 5496 ret = -EFSCORRUPTED; 5497 goto bad_inode; 5498 } 5499 5500 ext4_set_inode_mapping_order(inode); 5501 5502 ret = check_igot_inode(inode, flags, function, line); 5503 /* 5504 * -ESTALE here means there is nothing inherently wrong with the inode, 5505 * it's just not an inode we can return for an fhandle lookup. 5506 */ 5507 if (ret == -ESTALE) { 5508 brelse(iloc.bh); 5509 unlock_new_inode(inode); 5510 iput(inode); 5511 return ERR_PTR(-ESTALE); 5512 } 5513 if (ret) 5514 goto bad_inode; 5515 brelse(iloc.bh); 5516 5517 unlock_new_inode(inode); 5518 return inode; 5519 5520 bad_inode: 5521 brelse(iloc.bh); 5522 iget_failed(inode); 5523 return ERR_PTR(ret); 5524 } 5525 5526 static void __ext4_update_other_inode_time(struct super_block *sb, 5527 unsigned long orig_ino, 5528 unsigned long ino, 5529 struct ext4_inode *raw_inode) 5530 { 5531 struct inode *inode; 5532 5533 inode = find_inode_by_ino_rcu(sb, ino); 5534 if (!inode) 5535 return; 5536 5537 if (!inode_is_dirtytime_only(inode)) 5538 return; 5539 5540 spin_lock(&inode->i_lock); 5541 if (inode_is_dirtytime_only(inode)) { 5542 struct ext4_inode_info *ei = EXT4_I(inode); 5543 5544 inode->i_state &= ~I_DIRTY_TIME; 5545 spin_unlock(&inode->i_lock); 5546 5547 spin_lock(&ei->i_raw_lock); 5548 EXT4_INODE_SET_CTIME(inode, raw_inode); 5549 EXT4_INODE_SET_MTIME(inode, raw_inode); 5550 EXT4_INODE_SET_ATIME(inode, raw_inode); 5551 ext4_inode_csum_set(inode, raw_inode, ei); 5552 spin_unlock(&ei->i_raw_lock); 5553 trace_ext4_other_inode_update_time(inode, orig_ino); 5554 return; 5555 } 5556 spin_unlock(&inode->i_lock); 5557 } 5558 5559 /* 5560 * Opportunistically update the other time fields for other inodes in 5561 * the same inode table block. 5562 */ 5563 static void ext4_update_other_inodes_time(struct super_block *sb, 5564 unsigned long orig_ino, char *buf) 5565 { 5566 unsigned long ino; 5567 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; 5568 int inode_size = EXT4_INODE_SIZE(sb); 5569 5570 /* 5571 * Calculate the first inode in the inode table block. Inode 5572 * numbers are one-based. That is, the first inode in a block 5573 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1). 5574 */ 5575 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1; 5576 rcu_read_lock(); 5577 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) { 5578 if (ino == orig_ino) 5579 continue; 5580 __ext4_update_other_inode_time(sb, orig_ino, ino, 5581 (struct ext4_inode *)buf); 5582 } 5583 rcu_read_unlock(); 5584 } 5585 5586 /* 5587 * Post the struct inode info into an on-disk inode location in the 5588 * buffer-cache. This gobbles the caller's reference to the 5589 * buffer_head in the inode location struct. 5590 * 5591 * The caller must have write access to iloc->bh. 5592 */ 5593 static int ext4_do_update_inode(handle_t *handle, 5594 struct inode *inode, 5595 struct ext4_iloc *iloc) 5596 { 5597 struct ext4_inode *raw_inode = ext4_raw_inode(iloc); 5598 struct ext4_inode_info *ei = EXT4_I(inode); 5599 struct buffer_head *bh = iloc->bh; 5600 struct super_block *sb = inode->i_sb; 5601 int err; 5602 int need_datasync = 0, set_large_file = 0; 5603 5604 spin_lock(&ei->i_raw_lock); 5605 5606 /* 5607 * For fields not tracked in the in-memory inode, initialise them 5608 * to zero for new inodes. 5609 */ 5610 if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) 5611 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); 5612 5613 if (READ_ONCE(ei->i_disksize) != ext4_isize(inode->i_sb, raw_inode)) 5614 need_datasync = 1; 5615 if (ei->i_disksize > 0x7fffffffULL) { 5616 if (!ext4_has_feature_large_file(sb) || 5617 EXT4_SB(sb)->s_es->s_rev_level == cpu_to_le32(EXT4_GOOD_OLD_REV)) 5618 set_large_file = 1; 5619 } 5620 5621 err = ext4_fill_raw_inode(inode, raw_inode); 5622 spin_unlock(&ei->i_raw_lock); 5623 if (err) { 5624 EXT4_ERROR_INODE(inode, "corrupted inode contents"); 5625 goto out_brelse; 5626 } 5627 5628 if (inode->i_sb->s_flags & SB_LAZYTIME) 5629 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino, 5630 bh->b_data); 5631 5632 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); 5633 err = ext4_handle_dirty_metadata(handle, NULL, bh); 5634 if (err) 5635 goto out_error; 5636 ext4_clear_inode_state(inode, EXT4_STATE_NEW); 5637 if (set_large_file) { 5638 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access"); 5639 err = ext4_journal_get_write_access(handle, sb, 5640 EXT4_SB(sb)->s_sbh, 5641 EXT4_JTR_NONE); 5642 if (err) 5643 goto out_error; 5644 lock_buffer(EXT4_SB(sb)->s_sbh); 5645 ext4_set_feature_large_file(sb); 5646 ext4_superblock_csum_set(sb); 5647 unlock_buffer(EXT4_SB(sb)->s_sbh); 5648 ext4_handle_sync(handle); 5649 err = ext4_handle_dirty_metadata(handle, NULL, 5650 EXT4_SB(sb)->s_sbh); 5651 } 5652 ext4_update_inode_fsync_trans(handle, inode, need_datasync); 5653 out_error: 5654 ext4_std_error(inode->i_sb, err); 5655 out_brelse: 5656 brelse(bh); 5657 return err; 5658 } 5659 5660 /* 5661 * ext4_write_inode() 5662 * 5663 * We are called from a few places: 5664 * 5665 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files. 5666 * Here, there will be no transaction running. We wait for any running 5667 * transaction to commit. 5668 * 5669 * - Within flush work (sys_sync(), kupdate and such). 5670 * We wait on commit, if told to. 5671 * 5672 * - Within iput_final() -> write_inode_now() 5673 * We wait on commit, if told to. 5674 * 5675 * In all cases it is actually safe for us to return without doing anything, 5676 * because the inode has been copied into a raw inode buffer in 5677 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL 5678 * writeback. 5679 * 5680 * Note that we are absolutely dependent upon all inode dirtiers doing the 5681 * right thing: they *must* call mark_inode_dirty() after dirtying info in 5682 * which we are interested. 5683 * 5684 * It would be a bug for them to not do this. The code: 5685 * 5686 * mark_inode_dirty(inode) 5687 * stuff(); 5688 * inode->i_size = expr; 5689 * 5690 * is in error because write_inode() could occur while `stuff()' is running, 5691 * and the new i_size will be lost. Plus the inode will no longer be on the 5692 * superblock's dirty inode list. 5693 */ 5694 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) 5695 { 5696 int err; 5697 5698 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC)) 5699 return 0; 5700 5701 err = ext4_emergency_state(inode->i_sb); 5702 if (unlikely(err)) 5703 return err; 5704 5705 if (EXT4_SB(inode->i_sb)->s_journal) { 5706 if (ext4_journal_current_handle()) { 5707 ext4_debug("called recursively, non-PF_MEMALLOC!\n"); 5708 dump_stack(); 5709 return -EIO; 5710 } 5711 5712 /* 5713 * No need to force transaction in WB_SYNC_NONE mode. Also 5714 * ext4_sync_fs() will force the commit after everything is 5715 * written. 5716 */ 5717 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync) 5718 return 0; 5719 5720 err = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal, 5721 EXT4_I(inode)->i_sync_tid); 5722 } else { 5723 struct ext4_iloc iloc; 5724 5725 err = __ext4_get_inode_loc_noinmem(inode, &iloc); 5726 if (err) 5727 return err; 5728 /* 5729 * sync(2) will flush the whole buffer cache. No need to do 5730 * it here separately for each inode. 5731 */ 5732 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) 5733 sync_dirty_buffer(iloc.bh); 5734 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { 5735 ext4_error_inode_block(inode, iloc.bh->b_blocknr, EIO, 5736 "IO error syncing inode"); 5737 err = -EIO; 5738 } 5739 brelse(iloc.bh); 5740 } 5741 return err; 5742 } 5743 5744 /* 5745 * In data=journal mode ext4_journalled_invalidate_folio() may fail to invalidate 5746 * buffers that are attached to a folio straddling i_size and are undergoing 5747 * commit. In that case we have to wait for commit to finish and try again. 5748 */ 5749 static void ext4_wait_for_tail_page_commit(struct inode *inode) 5750 { 5751 unsigned offset; 5752 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; 5753 tid_t commit_tid; 5754 int ret; 5755 bool has_transaction; 5756 5757 offset = inode->i_size & (PAGE_SIZE - 1); 5758 /* 5759 * If the folio is fully truncated, we don't need to wait for any commit 5760 * (and we even should not as __ext4_journalled_invalidate_folio() may 5761 * strip all buffers from the folio but keep the folio dirty which can then 5762 * confuse e.g. concurrent ext4_writepages() seeing dirty folio without 5763 * buffers). Also we don't need to wait for any commit if all buffers in 5764 * the folio remain valid. This is most beneficial for the common case of 5765 * blocksize == PAGESIZE. 5766 */ 5767 if (!offset || offset > (PAGE_SIZE - i_blocksize(inode))) 5768 return; 5769 while (1) { 5770 struct folio *folio = filemap_lock_folio(inode->i_mapping, 5771 inode->i_size >> PAGE_SHIFT); 5772 if (IS_ERR(folio)) 5773 return; 5774 ret = __ext4_journalled_invalidate_folio(folio, offset, 5775 folio_size(folio) - offset); 5776 folio_unlock(folio); 5777 folio_put(folio); 5778 if (ret != -EBUSY) 5779 return; 5780 has_transaction = false; 5781 read_lock(&journal->j_state_lock); 5782 if (journal->j_committing_transaction) { 5783 commit_tid = journal->j_committing_transaction->t_tid; 5784 has_transaction = true; 5785 } 5786 read_unlock(&journal->j_state_lock); 5787 if (has_transaction) 5788 jbd2_log_wait_commit(journal, commit_tid); 5789 } 5790 } 5791 5792 /* 5793 * ext4_setattr() 5794 * 5795 * Called from notify_change. 5796 * 5797 * We want to trap VFS attempts to truncate the file as soon as 5798 * possible. In particular, we want to make sure that when the VFS 5799 * shrinks i_size, we put the inode on the orphan list and modify 5800 * i_disksize immediately, so that during the subsequent flushing of 5801 * dirty pages and freeing of disk blocks, we can guarantee that any 5802 * commit will leave the blocks being flushed in an unused state on 5803 * disk. (On recovery, the inode will get truncated and the blocks will 5804 * be freed, so we have a strong guarantee that no future commit will 5805 * leave these blocks visible to the user.) 5806 * 5807 * Another thing we have to assure is that if we are in ordered mode 5808 * and inode is still attached to the committing transaction, we must 5809 * we start writeout of all the dirty pages which are being truncated. 5810 * This way we are sure that all the data written in the previous 5811 * transaction are already on disk (truncate waits for pages under 5812 * writeback). 5813 * 5814 * Called with inode->i_rwsem down. 5815 */ 5816 int ext4_setattr(struct mnt_idmap *idmap, struct dentry *dentry, 5817 struct iattr *attr) 5818 { 5819 struct inode *inode = d_inode(dentry); 5820 int error, rc = 0; 5821 int orphan = 0; 5822 const unsigned int ia_valid = attr->ia_valid; 5823 bool inc_ivers = true; 5824 5825 error = ext4_emergency_state(inode->i_sb); 5826 if (unlikely(error)) 5827 return error; 5828 5829 if (unlikely(IS_IMMUTABLE(inode))) 5830 return -EPERM; 5831 5832 if (unlikely(IS_APPEND(inode) && 5833 (ia_valid & (ATTR_MODE | ATTR_UID | 5834 ATTR_GID | ATTR_TIMES_SET)))) 5835 return -EPERM; 5836 5837 error = setattr_prepare(idmap, dentry, attr); 5838 if (error) 5839 return error; 5840 5841 error = fscrypt_prepare_setattr(dentry, attr); 5842 if (error) 5843 return error; 5844 5845 error = fsverity_prepare_setattr(dentry, attr); 5846 if (error) 5847 return error; 5848 5849 if (is_quota_modification(idmap, inode, attr)) { 5850 error = dquot_initialize(inode); 5851 if (error) 5852 return error; 5853 } 5854 5855 if (i_uid_needs_update(idmap, attr, inode) || 5856 i_gid_needs_update(idmap, attr, inode)) { 5857 handle_t *handle; 5858 5859 /* (user+group)*(old+new) structure, inode write (sb, 5860 * inode block, ? - but truncate inode update has it) */ 5861 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 5862 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) + 5863 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3); 5864 if (IS_ERR(handle)) { 5865 error = PTR_ERR(handle); 5866 goto err_out; 5867 } 5868 5869 /* dquot_transfer() calls back ext4_get_inode_usage() which 5870 * counts xattr inode references. 5871 */ 5872 down_read(&EXT4_I(inode)->xattr_sem); 5873 error = dquot_transfer(idmap, inode, attr); 5874 up_read(&EXT4_I(inode)->xattr_sem); 5875 5876 if (error) { 5877 ext4_journal_stop(handle); 5878 return error; 5879 } 5880 /* Update corresponding info in inode so that everything is in 5881 * one transaction */ 5882 i_uid_update(idmap, attr, inode); 5883 i_gid_update(idmap, attr, inode); 5884 error = ext4_mark_inode_dirty(handle, inode); 5885 ext4_journal_stop(handle); 5886 if (unlikely(error)) { 5887 return error; 5888 } 5889 } 5890 5891 if (attr->ia_valid & ATTR_SIZE) { 5892 handle_t *handle; 5893 loff_t oldsize = inode->i_size; 5894 loff_t old_disksize; 5895 int shrink = (attr->ia_size < inode->i_size); 5896 5897 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { 5898 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 5899 5900 if (attr->ia_size > sbi->s_bitmap_maxbytes) { 5901 return -EFBIG; 5902 } 5903 } 5904 if (!S_ISREG(inode->i_mode)) { 5905 return -EINVAL; 5906 } 5907 5908 if (attr->ia_size == inode->i_size) 5909 inc_ivers = false; 5910 5911 if (shrink) { 5912 if (ext4_should_order_data(inode)) { 5913 error = ext4_begin_ordered_truncate(inode, 5914 attr->ia_size); 5915 if (error) 5916 goto err_out; 5917 } 5918 /* 5919 * Blocks are going to be removed from the inode. Wait 5920 * for dio in flight. 5921 */ 5922 inode_dio_wait(inode); 5923 } 5924 5925 filemap_invalidate_lock(inode->i_mapping); 5926 5927 rc = ext4_break_layouts(inode); 5928 if (rc) { 5929 filemap_invalidate_unlock(inode->i_mapping); 5930 goto err_out; 5931 } 5932 5933 if (attr->ia_size != inode->i_size) { 5934 /* attach jbd2 jinode for EOF folio tail zeroing */ 5935 if (attr->ia_size & (inode->i_sb->s_blocksize - 1) || 5936 oldsize & (inode->i_sb->s_blocksize - 1)) { 5937 error = ext4_inode_attach_jinode(inode); 5938 if (error) 5939 goto out_mmap_sem; 5940 } 5941 5942 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3); 5943 if (IS_ERR(handle)) { 5944 error = PTR_ERR(handle); 5945 goto out_mmap_sem; 5946 } 5947 if (ext4_handle_valid(handle) && shrink) { 5948 error = ext4_orphan_add(handle, inode); 5949 orphan = 1; 5950 } 5951 /* 5952 * Update c/mtime and tail zero the EOF folio on 5953 * truncate up. ext4_truncate() handles the shrink case 5954 * below. 5955 */ 5956 if (!shrink) { 5957 inode_set_mtime_to_ts(inode, 5958 inode_set_ctime_current(inode)); 5959 if (oldsize & (inode->i_sb->s_blocksize - 1)) 5960 ext4_block_truncate_page(handle, 5961 inode->i_mapping, oldsize); 5962 } 5963 5964 if (shrink) 5965 ext4_fc_track_range(handle, inode, 5966 (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >> 5967 inode->i_sb->s_blocksize_bits, 5968 EXT_MAX_BLOCKS - 1); 5969 else 5970 ext4_fc_track_range( 5971 handle, inode, 5972 (oldsize > 0 ? oldsize - 1 : oldsize) >> 5973 inode->i_sb->s_blocksize_bits, 5974 (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >> 5975 inode->i_sb->s_blocksize_bits); 5976 5977 down_write(&EXT4_I(inode)->i_data_sem); 5978 old_disksize = EXT4_I(inode)->i_disksize; 5979 EXT4_I(inode)->i_disksize = attr->ia_size; 5980 5981 /* 5982 * We have to update i_size under i_data_sem together 5983 * with i_disksize to avoid races with writeback code 5984 * running ext4_wb_update_i_disksize(). 5985 */ 5986 if (!error) 5987 i_size_write(inode, attr->ia_size); 5988 else 5989 EXT4_I(inode)->i_disksize = old_disksize; 5990 up_write(&EXT4_I(inode)->i_data_sem); 5991 rc = ext4_mark_inode_dirty(handle, inode); 5992 if (!error) 5993 error = rc; 5994 ext4_journal_stop(handle); 5995 if (error) 5996 goto out_mmap_sem; 5997 if (!shrink) { 5998 pagecache_isize_extended(inode, oldsize, 5999 inode->i_size); 6000 } else if (ext4_should_journal_data(inode)) { 6001 ext4_wait_for_tail_page_commit(inode); 6002 } 6003 } 6004 6005 /* 6006 * Truncate pagecache after we've waited for commit 6007 * in data=journal mode to make pages freeable. 6008 */ 6009 truncate_pagecache(inode, inode->i_size); 6010 /* 6011 * Call ext4_truncate() even if i_size didn't change to 6012 * truncate possible preallocated blocks. 6013 */ 6014 if (attr->ia_size <= oldsize) { 6015 rc = ext4_truncate(inode); 6016 if (rc) 6017 error = rc; 6018 } 6019 out_mmap_sem: 6020 filemap_invalidate_unlock(inode->i_mapping); 6021 } 6022 6023 if (!error) { 6024 if (inc_ivers) 6025 inode_inc_iversion(inode); 6026 setattr_copy(idmap, inode, attr); 6027 mark_inode_dirty(inode); 6028 } 6029 6030 /* 6031 * If the call to ext4_truncate failed to get a transaction handle at 6032 * all, we need to clean up the in-core orphan list manually. 6033 */ 6034 if (orphan && inode->i_nlink) 6035 ext4_orphan_del(NULL, inode); 6036 6037 if (!error && (ia_valid & ATTR_MODE)) 6038 rc = posix_acl_chmod(idmap, dentry, inode->i_mode); 6039 6040 err_out: 6041 if (error) 6042 ext4_std_error(inode->i_sb, error); 6043 if (!error) 6044 error = rc; 6045 return error; 6046 } 6047 6048 u32 ext4_dio_alignment(struct inode *inode) 6049 { 6050 if (fsverity_active(inode)) 6051 return 0; 6052 if (ext4_should_journal_data(inode)) 6053 return 0; 6054 if (ext4_has_inline_data(inode)) 6055 return 0; 6056 if (IS_ENCRYPTED(inode)) { 6057 if (!fscrypt_dio_supported(inode)) 6058 return 0; 6059 return i_blocksize(inode); 6060 } 6061 return 1; /* use the iomap defaults */ 6062 } 6063 6064 int ext4_getattr(struct mnt_idmap *idmap, const struct path *path, 6065 struct kstat *stat, u32 request_mask, unsigned int query_flags) 6066 { 6067 struct inode *inode = d_inode(path->dentry); 6068 struct ext4_inode *raw_inode; 6069 struct ext4_inode_info *ei = EXT4_I(inode); 6070 unsigned int flags; 6071 6072 if ((request_mask & STATX_BTIME) && 6073 EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) { 6074 stat->result_mask |= STATX_BTIME; 6075 stat->btime.tv_sec = ei->i_crtime.tv_sec; 6076 stat->btime.tv_nsec = ei->i_crtime.tv_nsec; 6077 } 6078 6079 /* 6080 * Return the DIO alignment restrictions if requested. We only return 6081 * this information when requested, since on encrypted files it might 6082 * take a fair bit of work to get if the file wasn't opened recently. 6083 */ 6084 if ((request_mask & STATX_DIOALIGN) && S_ISREG(inode->i_mode)) { 6085 u32 dio_align = ext4_dio_alignment(inode); 6086 6087 stat->result_mask |= STATX_DIOALIGN; 6088 if (dio_align == 1) { 6089 struct block_device *bdev = inode->i_sb->s_bdev; 6090 6091 /* iomap defaults */ 6092 stat->dio_mem_align = bdev_dma_alignment(bdev) + 1; 6093 stat->dio_offset_align = bdev_logical_block_size(bdev); 6094 } else { 6095 stat->dio_mem_align = dio_align; 6096 stat->dio_offset_align = dio_align; 6097 } 6098 } 6099 6100 if ((request_mask & STATX_WRITE_ATOMIC) && S_ISREG(inode->i_mode)) { 6101 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 6102 unsigned int awu_min = 0, awu_max = 0; 6103 6104 if (ext4_inode_can_atomic_write(inode)) { 6105 awu_min = sbi->s_awu_min; 6106 awu_max = sbi->s_awu_max; 6107 } 6108 6109 generic_fill_statx_atomic_writes(stat, awu_min, awu_max, 0); 6110 } 6111 6112 flags = ei->i_flags & EXT4_FL_USER_VISIBLE; 6113 if (flags & EXT4_APPEND_FL) 6114 stat->attributes |= STATX_ATTR_APPEND; 6115 if (flags & EXT4_COMPR_FL) 6116 stat->attributes |= STATX_ATTR_COMPRESSED; 6117 if (flags & EXT4_ENCRYPT_FL) 6118 stat->attributes |= STATX_ATTR_ENCRYPTED; 6119 if (flags & EXT4_IMMUTABLE_FL) 6120 stat->attributes |= STATX_ATTR_IMMUTABLE; 6121 if (flags & EXT4_NODUMP_FL) 6122 stat->attributes |= STATX_ATTR_NODUMP; 6123 if (flags & EXT4_VERITY_FL) 6124 stat->attributes |= STATX_ATTR_VERITY; 6125 6126 stat->attributes_mask |= (STATX_ATTR_APPEND | 6127 STATX_ATTR_COMPRESSED | 6128 STATX_ATTR_ENCRYPTED | 6129 STATX_ATTR_IMMUTABLE | 6130 STATX_ATTR_NODUMP | 6131 STATX_ATTR_VERITY); 6132 6133 generic_fillattr(idmap, request_mask, inode, stat); 6134 return 0; 6135 } 6136 6137 int ext4_file_getattr(struct mnt_idmap *idmap, 6138 const struct path *path, struct kstat *stat, 6139 u32 request_mask, unsigned int query_flags) 6140 { 6141 struct inode *inode = d_inode(path->dentry); 6142 u64 delalloc_blocks; 6143 6144 ext4_getattr(idmap, path, stat, request_mask, query_flags); 6145 6146 /* 6147 * If there is inline data in the inode, the inode will normally not 6148 * have data blocks allocated (it may have an external xattr block). 6149 * Report at least one sector for such files, so tools like tar, rsync, 6150 * others don't incorrectly think the file is completely sparse. 6151 */ 6152 if (unlikely(ext4_has_inline_data(inode))) 6153 stat->blocks += (stat->size + 511) >> 9; 6154 6155 /* 6156 * We can't update i_blocks if the block allocation is delayed 6157 * otherwise in the case of system crash before the real block 6158 * allocation is done, we will have i_blocks inconsistent with 6159 * on-disk file blocks. 6160 * We always keep i_blocks updated together with real 6161 * allocation. But to not confuse with user, stat 6162 * will return the blocks that include the delayed allocation 6163 * blocks for this file. 6164 */ 6165 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb), 6166 EXT4_I(inode)->i_reserved_data_blocks); 6167 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9); 6168 return 0; 6169 } 6170 6171 static int ext4_index_trans_blocks(struct inode *inode, int lblocks, 6172 int pextents) 6173 { 6174 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) 6175 return ext4_ind_trans_blocks(inode, lblocks); 6176 return ext4_ext_index_trans_blocks(inode, pextents); 6177 } 6178 6179 /* 6180 * Account for index blocks, block groups bitmaps and block group 6181 * descriptor blocks if modify datablocks and index blocks 6182 * worse case, the indexs blocks spread over different block groups 6183 * 6184 * If datablocks are discontiguous, they are possible to spread over 6185 * different block groups too. If they are contiguous, with flexbg, 6186 * they could still across block group boundary. 6187 * 6188 * Also account for superblock, inode, quota and xattr blocks 6189 */ 6190 int ext4_meta_trans_blocks(struct inode *inode, int lblocks, int pextents) 6191 { 6192 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); 6193 int gdpblocks; 6194 int idxblocks; 6195 int ret; 6196 6197 /* 6198 * How many index and leaf blocks need to touch to map @lblocks 6199 * logical blocks to @pextents physical extents? 6200 */ 6201 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents); 6202 6203 /* 6204 * Now let's see how many group bitmaps and group descriptors need 6205 * to account 6206 */ 6207 groups = idxblocks + pextents; 6208 gdpblocks = groups; 6209 if (groups > ngroups) 6210 groups = ngroups; 6211 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) 6212 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; 6213 6214 /* bitmaps and block group descriptor blocks */ 6215 ret = idxblocks + groups + gdpblocks; 6216 6217 /* Blocks for super block, inode, quota and xattr blocks */ 6218 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); 6219 6220 return ret; 6221 } 6222 6223 /* 6224 * Calculate the journal credits for modifying the number of blocks 6225 * in a single extent within one transaction. 'nrblocks' is used only 6226 * for non-extent inodes. For extent type inodes, 'nrblocks' can be 6227 * zero if the exact number of blocks is unknown. 6228 */ 6229 int ext4_chunk_trans_extent(struct inode *inode, int nrblocks) 6230 { 6231 int ret; 6232 6233 ret = ext4_meta_trans_blocks(inode, nrblocks, 1); 6234 /* Account for data blocks for journalled mode */ 6235 if (ext4_should_journal_data(inode)) 6236 ret += nrblocks; 6237 return ret; 6238 } 6239 6240 /* 6241 * Calculate the journal credits for a chunk of data modification. 6242 * 6243 * This is called from DIO, fallocate or whoever calling 6244 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. 6245 * 6246 * journal buffers for data blocks are not included here, as DIO 6247 * and fallocate do no need to journal data buffers. 6248 */ 6249 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) 6250 { 6251 return ext4_meta_trans_blocks(inode, nrblocks, 1); 6252 } 6253 6254 /* 6255 * The caller must have previously called ext4_reserve_inode_write(). 6256 * Give this, we know that the caller already has write access to iloc->bh. 6257 */ 6258 int ext4_mark_iloc_dirty(handle_t *handle, 6259 struct inode *inode, struct ext4_iloc *iloc) 6260 { 6261 int err = 0; 6262 6263 err = ext4_emergency_state(inode->i_sb); 6264 if (unlikely(err)) { 6265 put_bh(iloc->bh); 6266 return err; 6267 } 6268 ext4_fc_track_inode(handle, inode); 6269 6270 /* the do_update_inode consumes one bh->b_count */ 6271 get_bh(iloc->bh); 6272 6273 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ 6274 err = ext4_do_update_inode(handle, inode, iloc); 6275 put_bh(iloc->bh); 6276 return err; 6277 } 6278 6279 /* 6280 * On success, We end up with an outstanding reference count against 6281 * iloc->bh. This _must_ be cleaned up later. 6282 */ 6283 6284 int 6285 ext4_reserve_inode_write(handle_t *handle, struct inode *inode, 6286 struct ext4_iloc *iloc) 6287 { 6288 int err; 6289 6290 err = ext4_emergency_state(inode->i_sb); 6291 if (unlikely(err)) 6292 return err; 6293 6294 err = ext4_get_inode_loc(inode, iloc); 6295 if (!err) { 6296 BUFFER_TRACE(iloc->bh, "get_write_access"); 6297 err = ext4_journal_get_write_access(handle, inode->i_sb, 6298 iloc->bh, EXT4_JTR_NONE); 6299 if (err) { 6300 brelse(iloc->bh); 6301 iloc->bh = NULL; 6302 } 6303 ext4_fc_track_inode(handle, inode); 6304 } 6305 ext4_std_error(inode->i_sb, err); 6306 return err; 6307 } 6308 6309 static int __ext4_expand_extra_isize(struct inode *inode, 6310 unsigned int new_extra_isize, 6311 struct ext4_iloc *iloc, 6312 handle_t *handle, int *no_expand) 6313 { 6314 struct ext4_inode *raw_inode; 6315 struct ext4_xattr_ibody_header *header; 6316 unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb); 6317 struct ext4_inode_info *ei = EXT4_I(inode); 6318 int error; 6319 6320 /* this was checked at iget time, but double check for good measure */ 6321 if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) || 6322 (ei->i_extra_isize & 3)) { 6323 EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)", 6324 ei->i_extra_isize, 6325 EXT4_INODE_SIZE(inode->i_sb)); 6326 return -EFSCORRUPTED; 6327 } 6328 if ((new_extra_isize < ei->i_extra_isize) || 6329 (new_extra_isize < 4) || 6330 (new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE)) 6331 return -EINVAL; /* Should never happen */ 6332 6333 raw_inode = ext4_raw_inode(iloc); 6334 6335 header = IHDR(inode, raw_inode); 6336 6337 /* No extended attributes present */ 6338 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || 6339 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { 6340 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE + 6341 EXT4_I(inode)->i_extra_isize, 0, 6342 new_extra_isize - EXT4_I(inode)->i_extra_isize); 6343 EXT4_I(inode)->i_extra_isize = new_extra_isize; 6344 return 0; 6345 } 6346 6347 /* 6348 * We may need to allocate external xattr block so we need quotas 6349 * initialized. Here we can be called with various locks held so we 6350 * cannot affort to initialize quotas ourselves. So just bail. 6351 */ 6352 if (dquot_initialize_needed(inode)) 6353 return -EAGAIN; 6354 6355 /* try to expand with EAs present */ 6356 error = ext4_expand_extra_isize_ea(inode, new_extra_isize, 6357 raw_inode, handle); 6358 if (error) { 6359 /* 6360 * Inode size expansion failed; don't try again 6361 */ 6362 *no_expand = 1; 6363 } 6364 6365 return error; 6366 } 6367 6368 /* 6369 * Expand an inode by new_extra_isize bytes. 6370 * Returns 0 on success or negative error number on failure. 6371 */ 6372 static int ext4_try_to_expand_extra_isize(struct inode *inode, 6373 unsigned int new_extra_isize, 6374 struct ext4_iloc iloc, 6375 handle_t *handle) 6376 { 6377 int no_expand; 6378 int error; 6379 6380 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) 6381 return -EOVERFLOW; 6382 6383 /* 6384 * In nojournal mode, we can immediately attempt to expand 6385 * the inode. When journaled, we first need to obtain extra 6386 * buffer credits since we may write into the EA block 6387 * with this same handle. If journal_extend fails, then it will 6388 * only result in a minor loss of functionality for that inode. 6389 * If this is felt to be critical, then e2fsck should be run to 6390 * force a large enough s_min_extra_isize. 6391 */ 6392 if (ext4_journal_extend(handle, 6393 EXT4_DATA_TRANS_BLOCKS(inode->i_sb), 0) != 0) 6394 return -ENOSPC; 6395 6396 if (ext4_write_trylock_xattr(inode, &no_expand) == 0) 6397 return -EBUSY; 6398 6399 error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc, 6400 handle, &no_expand); 6401 ext4_write_unlock_xattr(inode, &no_expand); 6402 6403 return error; 6404 } 6405 6406 int ext4_expand_extra_isize(struct inode *inode, 6407 unsigned int new_extra_isize, 6408 struct ext4_iloc *iloc) 6409 { 6410 handle_t *handle; 6411 int no_expand; 6412 int error, rc; 6413 6414 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) { 6415 brelse(iloc->bh); 6416 return -EOVERFLOW; 6417 } 6418 6419 handle = ext4_journal_start(inode, EXT4_HT_INODE, 6420 EXT4_DATA_TRANS_BLOCKS(inode->i_sb)); 6421 if (IS_ERR(handle)) { 6422 error = PTR_ERR(handle); 6423 brelse(iloc->bh); 6424 return error; 6425 } 6426 6427 ext4_write_lock_xattr(inode, &no_expand); 6428 6429 BUFFER_TRACE(iloc->bh, "get_write_access"); 6430 error = ext4_journal_get_write_access(handle, inode->i_sb, iloc->bh, 6431 EXT4_JTR_NONE); 6432 if (error) { 6433 brelse(iloc->bh); 6434 goto out_unlock; 6435 } 6436 6437 error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc, 6438 handle, &no_expand); 6439 6440 rc = ext4_mark_iloc_dirty(handle, inode, iloc); 6441 if (!error) 6442 error = rc; 6443 6444 out_unlock: 6445 ext4_write_unlock_xattr(inode, &no_expand); 6446 ext4_journal_stop(handle); 6447 return error; 6448 } 6449 6450 /* 6451 * What we do here is to mark the in-core inode as clean with respect to inode 6452 * dirtiness (it may still be data-dirty). 6453 * This means that the in-core inode may be reaped by prune_icache 6454 * without having to perform any I/O. This is a very good thing, 6455 * because *any* task may call prune_icache - even ones which 6456 * have a transaction open against a different journal. 6457 * 6458 * Is this cheating? Not really. Sure, we haven't written the 6459 * inode out, but prune_icache isn't a user-visible syncing function. 6460 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) 6461 * we start and wait on commits. 6462 */ 6463 int __ext4_mark_inode_dirty(handle_t *handle, struct inode *inode, 6464 const char *func, unsigned int line) 6465 { 6466 struct ext4_iloc iloc; 6467 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 6468 int err; 6469 6470 might_sleep(); 6471 trace_ext4_mark_inode_dirty(inode, _RET_IP_); 6472 err = ext4_reserve_inode_write(handle, inode, &iloc); 6473 if (err) 6474 goto out; 6475 6476 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize) 6477 ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize, 6478 iloc, handle); 6479 6480 err = ext4_mark_iloc_dirty(handle, inode, &iloc); 6481 out: 6482 if (unlikely(err)) 6483 ext4_error_inode_err(inode, func, line, 0, err, 6484 "mark_inode_dirty error"); 6485 return err; 6486 } 6487 6488 /* 6489 * ext4_dirty_inode() is called from __mark_inode_dirty() 6490 * 6491 * We're really interested in the case where a file is being extended. 6492 * i_size has been changed by generic_commit_write() and we thus need 6493 * to include the updated inode in the current transaction. 6494 * 6495 * Also, dquot_alloc_block() will always dirty the inode when blocks 6496 * are allocated to the file. 6497 * 6498 * If the inode is marked synchronous, we don't honour that here - doing 6499 * so would cause a commit on atime updates, which we don't bother doing. 6500 * We handle synchronous inodes at the highest possible level. 6501 */ 6502 void ext4_dirty_inode(struct inode *inode, int flags) 6503 { 6504 handle_t *handle; 6505 6506 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); 6507 if (IS_ERR(handle)) 6508 return; 6509 ext4_mark_inode_dirty(handle, inode); 6510 ext4_journal_stop(handle); 6511 } 6512 6513 int ext4_change_inode_journal_flag(struct inode *inode, int val) 6514 { 6515 journal_t *journal; 6516 handle_t *handle; 6517 int err; 6518 int alloc_ctx; 6519 6520 /* 6521 * We have to be very careful here: changing a data block's 6522 * journaling status dynamically is dangerous. If we write a 6523 * data block to the journal, change the status and then delete 6524 * that block, we risk forgetting to revoke the old log record 6525 * from the journal and so a subsequent replay can corrupt data. 6526 * So, first we make sure that the journal is empty and that 6527 * nobody is changing anything. 6528 */ 6529 6530 journal = EXT4_JOURNAL(inode); 6531 if (!journal) 6532 return 0; 6533 if (is_journal_aborted(journal)) 6534 return -EROFS; 6535 6536 /* Wait for all existing dio workers */ 6537 inode_dio_wait(inode); 6538 6539 /* 6540 * Before flushing the journal and switching inode's aops, we have 6541 * to flush all dirty data the inode has. There can be outstanding 6542 * delayed allocations, there can be unwritten extents created by 6543 * fallocate or buffered writes in dioread_nolock mode covered by 6544 * dirty data which can be converted only after flushing the dirty 6545 * data (and journalled aops don't know how to handle these cases). 6546 */ 6547 if (val) { 6548 filemap_invalidate_lock(inode->i_mapping); 6549 err = filemap_write_and_wait(inode->i_mapping); 6550 if (err < 0) { 6551 filemap_invalidate_unlock(inode->i_mapping); 6552 return err; 6553 } 6554 } 6555 6556 alloc_ctx = ext4_writepages_down_write(inode->i_sb); 6557 jbd2_journal_lock_updates(journal); 6558 6559 /* 6560 * OK, there are no updates running now, and all cached data is 6561 * synced to disk. We are now in a completely consistent state 6562 * which doesn't have anything in the journal, and we know that 6563 * no filesystem updates are running, so it is safe to modify 6564 * the inode's in-core data-journaling state flag now. 6565 */ 6566 6567 if (val) 6568 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); 6569 else { 6570 err = jbd2_journal_flush(journal, 0); 6571 if (err < 0) { 6572 jbd2_journal_unlock_updates(journal); 6573 ext4_writepages_up_write(inode->i_sb, alloc_ctx); 6574 return err; 6575 } 6576 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); 6577 } 6578 ext4_set_aops(inode); 6579 6580 jbd2_journal_unlock_updates(journal); 6581 ext4_writepages_up_write(inode->i_sb, alloc_ctx); 6582 6583 if (val) 6584 filemap_invalidate_unlock(inode->i_mapping); 6585 6586 /* Finally we can mark the inode as dirty. */ 6587 6588 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); 6589 if (IS_ERR(handle)) 6590 return PTR_ERR(handle); 6591 6592 ext4_fc_mark_ineligible(inode->i_sb, 6593 EXT4_FC_REASON_JOURNAL_FLAG_CHANGE, handle); 6594 err = ext4_mark_inode_dirty(handle, inode); 6595 ext4_handle_sync(handle); 6596 ext4_journal_stop(handle); 6597 ext4_std_error(inode->i_sb, err); 6598 6599 return err; 6600 } 6601 6602 static int ext4_bh_unmapped(handle_t *handle, struct inode *inode, 6603 struct buffer_head *bh) 6604 { 6605 return !buffer_mapped(bh); 6606 } 6607 6608 static int ext4_block_page_mkwrite(struct inode *inode, struct folio *folio, 6609 get_block_t get_block) 6610 { 6611 handle_t *handle; 6612 loff_t size; 6613 unsigned long len; 6614 int credits; 6615 int ret; 6616 6617 credits = ext4_chunk_trans_extent(inode, 6618 ext4_journal_blocks_per_folio(inode)); 6619 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, credits); 6620 if (IS_ERR(handle)) 6621 return PTR_ERR(handle); 6622 6623 folio_lock(folio); 6624 size = i_size_read(inode); 6625 /* Page got truncated from under us? */ 6626 if (folio->mapping != inode->i_mapping || folio_pos(folio) > size) { 6627 ret = -EFAULT; 6628 goto out_error; 6629 } 6630 6631 len = folio_size(folio); 6632 if (folio_pos(folio) + len > size) 6633 len = size - folio_pos(folio); 6634 6635 ret = ext4_block_write_begin(handle, folio, 0, len, get_block); 6636 if (ret) 6637 goto out_error; 6638 6639 if (!ext4_should_journal_data(inode)) { 6640 block_commit_write(folio, 0, len); 6641 folio_mark_dirty(folio); 6642 } else { 6643 ret = ext4_journal_folio_buffers(handle, folio, len); 6644 if (ret) 6645 goto out_error; 6646 } 6647 ext4_journal_stop(handle); 6648 folio_wait_stable(folio); 6649 return ret; 6650 6651 out_error: 6652 folio_unlock(folio); 6653 ext4_journal_stop(handle); 6654 return ret; 6655 } 6656 6657 vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf) 6658 { 6659 struct vm_area_struct *vma = vmf->vma; 6660 struct folio *folio = page_folio(vmf->page); 6661 loff_t size; 6662 unsigned long len; 6663 int err; 6664 vm_fault_t ret; 6665 struct file *file = vma->vm_file; 6666 struct inode *inode = file_inode(file); 6667 struct address_space *mapping = inode->i_mapping; 6668 get_block_t *get_block = ext4_get_block; 6669 int retries = 0; 6670 6671 if (unlikely(IS_IMMUTABLE(inode))) 6672 return VM_FAULT_SIGBUS; 6673 6674 sb_start_pagefault(inode->i_sb); 6675 file_update_time(vma->vm_file); 6676 6677 filemap_invalidate_lock_shared(mapping); 6678 6679 err = ext4_convert_inline_data(inode); 6680 if (err) 6681 goto out_ret; 6682 6683 /* 6684 * On data journalling we skip straight to the transaction handle: 6685 * there's no delalloc; page truncated will be checked later; the 6686 * early return w/ all buffers mapped (calculates size/len) can't 6687 * be used; and there's no dioread_nolock, so only ext4_get_block. 6688 */ 6689 if (ext4_should_journal_data(inode)) 6690 goto retry_alloc; 6691 6692 /* Delalloc case is easy... */ 6693 if (test_opt(inode->i_sb, DELALLOC) && 6694 !ext4_nonda_switch(inode->i_sb)) { 6695 do { 6696 err = block_page_mkwrite(vma, vmf, 6697 ext4_da_get_block_prep); 6698 } while (err == -ENOSPC && 6699 ext4_should_retry_alloc(inode->i_sb, &retries)); 6700 goto out_ret; 6701 } 6702 6703 folio_lock(folio); 6704 size = i_size_read(inode); 6705 /* Page got truncated from under us? */ 6706 if (folio->mapping != mapping || folio_pos(folio) > size) { 6707 folio_unlock(folio); 6708 ret = VM_FAULT_NOPAGE; 6709 goto out; 6710 } 6711 6712 len = folio_size(folio); 6713 if (folio_pos(folio) + len > size) 6714 len = size - folio_pos(folio); 6715 /* 6716 * Return if we have all the buffers mapped. This avoids the need to do 6717 * journal_start/journal_stop which can block and take a long time 6718 * 6719 * This cannot be done for data journalling, as we have to add the 6720 * inode to the transaction's list to writeprotect pages on commit. 6721 */ 6722 if (folio_buffers(folio)) { 6723 if (!ext4_walk_page_buffers(NULL, inode, folio_buffers(folio), 6724 0, len, NULL, 6725 ext4_bh_unmapped)) { 6726 /* Wait so that we don't change page under IO */ 6727 folio_wait_stable(folio); 6728 ret = VM_FAULT_LOCKED; 6729 goto out; 6730 } 6731 } 6732 folio_unlock(folio); 6733 /* OK, we need to fill the hole... */ 6734 if (ext4_should_dioread_nolock(inode)) 6735 get_block = ext4_get_block_unwritten; 6736 retry_alloc: 6737 /* Start journal and allocate blocks */ 6738 err = ext4_block_page_mkwrite(inode, folio, get_block); 6739 if (err == -EAGAIN || 6740 (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))) 6741 goto retry_alloc; 6742 out_ret: 6743 ret = vmf_fs_error(err); 6744 out: 6745 filemap_invalidate_unlock_shared(mapping); 6746 sb_end_pagefault(inode->i_sb); 6747 return ret; 6748 } 6749