1 /* 2 * linux/fs/ext4/inode.c 3 * 4 * Copyright (C) 1992, 1993, 1994, 1995 5 * Remy Card (card@masi.ibp.fr) 6 * Laboratoire MASI - Institut Blaise Pascal 7 * Universite Pierre et Marie Curie (Paris VI) 8 * 9 * from 10 * 11 * linux/fs/minix/inode.c 12 * 13 * Copyright (C) 1991, 1992 Linus Torvalds 14 * 15 * 64-bit file support on 64-bit platforms by Jakub Jelinek 16 * (jj@sunsite.ms.mff.cuni.cz) 17 * 18 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000 19 */ 20 21 #include <linux/fs.h> 22 #include <linux/time.h> 23 #include <linux/highuid.h> 24 #include <linux/pagemap.h> 25 #include <linux/quotaops.h> 26 #include <linux/string.h> 27 #include <linux/buffer_head.h> 28 #include <linux/writeback.h> 29 #include <linux/pagevec.h> 30 #include <linux/mpage.h> 31 #include <linux/namei.h> 32 #include <linux/uio.h> 33 #include <linux/bio.h> 34 #include <linux/workqueue.h> 35 #include <linux/kernel.h> 36 #include <linux/printk.h> 37 #include <linux/slab.h> 38 #include <linux/bitops.h> 39 40 #include "ext4_jbd2.h" 41 #include "xattr.h" 42 #include "acl.h" 43 #include "truncate.h" 44 45 #include <trace/events/ext4.h> 46 47 #define MPAGE_DA_EXTENT_TAIL 0x01 48 49 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw, 50 struct ext4_inode_info *ei) 51 { 52 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 53 __u16 csum_lo; 54 __u16 csum_hi = 0; 55 __u32 csum; 56 57 csum_lo = le16_to_cpu(raw->i_checksum_lo); 58 raw->i_checksum_lo = 0; 59 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && 60 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) { 61 csum_hi = le16_to_cpu(raw->i_checksum_hi); 62 raw->i_checksum_hi = 0; 63 } 64 65 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, 66 EXT4_INODE_SIZE(inode->i_sb)); 67 68 raw->i_checksum_lo = cpu_to_le16(csum_lo); 69 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && 70 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) 71 raw->i_checksum_hi = cpu_to_le16(csum_hi); 72 73 return csum; 74 } 75 76 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw, 77 struct ext4_inode_info *ei) 78 { 79 __u32 provided, calculated; 80 81 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != 82 cpu_to_le32(EXT4_OS_LINUX) || 83 !ext4_has_metadata_csum(inode->i_sb)) 84 return 1; 85 86 provided = le16_to_cpu(raw->i_checksum_lo); 87 calculated = ext4_inode_csum(inode, raw, ei); 88 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && 89 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) 90 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16; 91 else 92 calculated &= 0xFFFF; 93 94 return provided == calculated; 95 } 96 97 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw, 98 struct ext4_inode_info *ei) 99 { 100 __u32 csum; 101 102 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != 103 cpu_to_le32(EXT4_OS_LINUX) || 104 !ext4_has_metadata_csum(inode->i_sb)) 105 return; 106 107 csum = ext4_inode_csum(inode, raw, ei); 108 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF); 109 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && 110 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) 111 raw->i_checksum_hi = cpu_to_le16(csum >> 16); 112 } 113 114 static inline int ext4_begin_ordered_truncate(struct inode *inode, 115 loff_t new_size) 116 { 117 trace_ext4_begin_ordered_truncate(inode, new_size); 118 /* 119 * If jinode is zero, then we never opened the file for 120 * writing, so there's no need to call 121 * jbd2_journal_begin_ordered_truncate() since there's no 122 * outstanding writes we need to flush. 123 */ 124 if (!EXT4_I(inode)->jinode) 125 return 0; 126 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode), 127 EXT4_I(inode)->jinode, 128 new_size); 129 } 130 131 static void ext4_invalidatepage(struct page *page, unsigned int offset, 132 unsigned int length); 133 static int __ext4_journalled_writepage(struct page *page, unsigned int len); 134 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh); 135 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, 136 int pextents); 137 138 /* 139 * Test whether an inode is a fast symlink. 140 */ 141 int ext4_inode_is_fast_symlink(struct inode *inode) 142 { 143 int ea_blocks = EXT4_I(inode)->i_file_acl ? 144 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0; 145 146 if (ext4_has_inline_data(inode)) 147 return 0; 148 149 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); 150 } 151 152 /* 153 * Restart the transaction associated with *handle. This does a commit, 154 * so before we call here everything must be consistently dirtied against 155 * this transaction. 156 */ 157 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode, 158 int nblocks) 159 { 160 int ret; 161 162 /* 163 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this 164 * moment, get_block can be called only for blocks inside i_size since 165 * page cache has been already dropped and writes are blocked by 166 * i_mutex. So we can safely drop the i_data_sem here. 167 */ 168 BUG_ON(EXT4_JOURNAL(inode) == NULL); 169 jbd_debug(2, "restarting handle %p\n", handle); 170 up_write(&EXT4_I(inode)->i_data_sem); 171 ret = ext4_journal_restart(handle, nblocks); 172 down_write(&EXT4_I(inode)->i_data_sem); 173 ext4_discard_preallocations(inode); 174 175 return ret; 176 } 177 178 /* 179 * Called at the last iput() if i_nlink is zero. 180 */ 181 void ext4_evict_inode(struct inode *inode) 182 { 183 handle_t *handle; 184 int err; 185 186 trace_ext4_evict_inode(inode); 187 188 if (inode->i_nlink) { 189 /* 190 * When journalling data dirty buffers are tracked only in the 191 * journal. So although mm thinks everything is clean and 192 * ready for reaping the inode might still have some pages to 193 * write in the running transaction or waiting to be 194 * checkpointed. Thus calling jbd2_journal_invalidatepage() 195 * (via truncate_inode_pages()) to discard these buffers can 196 * cause data loss. Also even if we did not discard these 197 * buffers, we would have no way to find them after the inode 198 * is reaped and thus user could see stale data if he tries to 199 * read them before the transaction is checkpointed. So be 200 * careful and force everything to disk here... We use 201 * ei->i_datasync_tid to store the newest transaction 202 * containing inode's data. 203 * 204 * Note that directories do not have this problem because they 205 * don't use page cache. 206 */ 207 if (ext4_should_journal_data(inode) && 208 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) && 209 inode->i_ino != EXT4_JOURNAL_INO) { 210 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; 211 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid; 212 213 jbd2_complete_transaction(journal, commit_tid); 214 filemap_write_and_wait(&inode->i_data); 215 } 216 truncate_inode_pages_final(&inode->i_data); 217 218 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count)); 219 goto no_delete; 220 } 221 222 if (is_bad_inode(inode)) 223 goto no_delete; 224 dquot_initialize(inode); 225 226 if (ext4_should_order_data(inode)) 227 ext4_begin_ordered_truncate(inode, 0); 228 truncate_inode_pages_final(&inode->i_data); 229 230 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count)); 231 232 /* 233 * Protect us against freezing - iput() caller didn't have to have any 234 * protection against it 235 */ 236 sb_start_intwrite(inode->i_sb); 237 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, 238 ext4_blocks_for_truncate(inode)+3); 239 if (IS_ERR(handle)) { 240 ext4_std_error(inode->i_sb, PTR_ERR(handle)); 241 /* 242 * If we're going to skip the normal cleanup, we still need to 243 * make sure that the in-core orphan linked list is properly 244 * cleaned up. 245 */ 246 ext4_orphan_del(NULL, inode); 247 sb_end_intwrite(inode->i_sb); 248 goto no_delete; 249 } 250 251 if (IS_SYNC(inode)) 252 ext4_handle_sync(handle); 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 ext4_truncate(inode); 262 263 /* 264 * ext4_ext_truncate() doesn't reserve any slop when it 265 * restarts journal transactions; therefore there may not be 266 * enough credits left in the handle to remove the inode from 267 * the orphan list and set the dtime field. 268 */ 269 if (!ext4_handle_has_enough_credits(handle, 3)) { 270 err = ext4_journal_extend(handle, 3); 271 if (err > 0) 272 err = ext4_journal_restart(handle, 3); 273 if (err != 0) { 274 ext4_warning(inode->i_sb, 275 "couldn't extend journal (err %d)", err); 276 stop_handle: 277 ext4_journal_stop(handle); 278 ext4_orphan_del(NULL, inode); 279 sb_end_intwrite(inode->i_sb); 280 goto no_delete; 281 } 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 = get_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 sb_end_intwrite(inode->i_sb); 309 return; 310 no_delete: 311 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */ 312 } 313 314 #ifdef CONFIG_QUOTA 315 qsize_t *ext4_get_reserved_space(struct inode *inode) 316 { 317 return &EXT4_I(inode)->i_reserved_quota; 318 } 319 #endif 320 321 /* 322 * Called with i_data_sem down, which is important since we can call 323 * ext4_discard_preallocations() from here. 324 */ 325 void ext4_da_update_reserve_space(struct inode *inode, 326 int used, int quota_claim) 327 { 328 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 329 struct ext4_inode_info *ei = EXT4_I(inode); 330 331 spin_lock(&ei->i_block_reservation_lock); 332 trace_ext4_da_update_reserve_space(inode, used, quota_claim); 333 if (unlikely(used > ei->i_reserved_data_blocks)) { 334 ext4_warning(inode->i_sb, "%s: ino %lu, used %d " 335 "with only %d reserved data blocks", 336 __func__, inode->i_ino, used, 337 ei->i_reserved_data_blocks); 338 WARN_ON(1); 339 used = ei->i_reserved_data_blocks; 340 } 341 342 /* Update per-inode reservations */ 343 ei->i_reserved_data_blocks -= used; 344 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used); 345 346 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); 347 348 /* Update quota subsystem for data blocks */ 349 if (quota_claim) 350 dquot_claim_block(inode, EXT4_C2B(sbi, used)); 351 else { 352 /* 353 * We did fallocate with an offset that is already delayed 354 * allocated. So on delayed allocated writeback we should 355 * not re-claim the quota for fallocated blocks. 356 */ 357 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used)); 358 } 359 360 /* 361 * If we have done all the pending block allocations and if 362 * there aren't any writers on the inode, we can discard the 363 * inode's preallocations. 364 */ 365 if ((ei->i_reserved_data_blocks == 0) && 366 (atomic_read(&inode->i_writecount) == 0)) 367 ext4_discard_preallocations(inode); 368 } 369 370 static int __check_block_validity(struct inode *inode, const char *func, 371 unsigned int line, 372 struct ext4_map_blocks *map) 373 { 374 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk, 375 map->m_len)) { 376 ext4_error_inode(inode, func, line, map->m_pblk, 377 "lblock %lu mapped to illegal pblock " 378 "(length %d)", (unsigned long) map->m_lblk, 379 map->m_len); 380 return -EIO; 381 } 382 return 0; 383 } 384 385 #define check_block_validity(inode, map) \ 386 __check_block_validity((inode), __func__, __LINE__, (map)) 387 388 #ifdef ES_AGGRESSIVE_TEST 389 static void ext4_map_blocks_es_recheck(handle_t *handle, 390 struct inode *inode, 391 struct ext4_map_blocks *es_map, 392 struct ext4_map_blocks *map, 393 int flags) 394 { 395 int retval; 396 397 map->m_flags = 0; 398 /* 399 * There is a race window that the result is not the same. 400 * e.g. xfstests #223 when dioread_nolock enables. The reason 401 * is that we lookup a block mapping in extent status tree with 402 * out taking i_data_sem. So at the time the unwritten extent 403 * could be converted. 404 */ 405 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) 406 down_read(&EXT4_I(inode)->i_data_sem); 407 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { 408 retval = ext4_ext_map_blocks(handle, inode, map, flags & 409 EXT4_GET_BLOCKS_KEEP_SIZE); 410 } else { 411 retval = ext4_ind_map_blocks(handle, inode, map, flags & 412 EXT4_GET_BLOCKS_KEEP_SIZE); 413 } 414 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) 415 up_read((&EXT4_I(inode)->i_data_sem)); 416 417 /* 418 * We don't check m_len because extent will be collpased in status 419 * tree. So the m_len might not equal. 420 */ 421 if (es_map->m_lblk != map->m_lblk || 422 es_map->m_flags != map->m_flags || 423 es_map->m_pblk != map->m_pblk) { 424 printk("ES cache assertion failed for inode: %lu " 425 "es_cached ex [%d/%d/%llu/%x] != " 426 "found ex [%d/%d/%llu/%x] retval %d flags %x\n", 427 inode->i_ino, es_map->m_lblk, es_map->m_len, 428 es_map->m_pblk, es_map->m_flags, map->m_lblk, 429 map->m_len, map->m_pblk, map->m_flags, 430 retval, flags); 431 } 432 } 433 #endif /* ES_AGGRESSIVE_TEST */ 434 435 /* 436 * The ext4_map_blocks() function tries to look up the requested blocks, 437 * and returns if the blocks are already mapped. 438 * 439 * Otherwise it takes the write lock of the i_data_sem and allocate blocks 440 * and store the allocated blocks in the result buffer head and mark it 441 * mapped. 442 * 443 * If file type is extents based, it will call ext4_ext_map_blocks(), 444 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping 445 * based files 446 * 447 * On success, it returns the number of blocks being mapped or allocated. 448 * if create==0 and the blocks are pre-allocated and unwritten block, 449 * the result buffer head is unmapped. If the create ==1, it will make sure 450 * the buffer head is mapped. 451 * 452 * It returns 0 if plain look up failed (blocks have not been allocated), in 453 * that case, buffer head is unmapped 454 * 455 * It returns the error in case of allocation failure. 456 */ 457 int ext4_map_blocks(handle_t *handle, struct inode *inode, 458 struct ext4_map_blocks *map, int flags) 459 { 460 struct extent_status es; 461 int retval; 462 int ret = 0; 463 #ifdef ES_AGGRESSIVE_TEST 464 struct ext4_map_blocks orig_map; 465 466 memcpy(&orig_map, map, sizeof(*map)); 467 #endif 468 469 map->m_flags = 0; 470 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u," 471 "logical block %lu\n", inode->i_ino, flags, map->m_len, 472 (unsigned long) map->m_lblk); 473 474 /* 475 * ext4_map_blocks returns an int, and m_len is an unsigned int 476 */ 477 if (unlikely(map->m_len > INT_MAX)) 478 map->m_len = INT_MAX; 479 480 /* We can handle the block number less than EXT_MAX_BLOCKS */ 481 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS)) 482 return -EIO; 483 484 /* Lookup extent status tree firstly */ 485 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) { 486 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) { 487 map->m_pblk = ext4_es_pblock(&es) + 488 map->m_lblk - es.es_lblk; 489 map->m_flags |= ext4_es_is_written(&es) ? 490 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN; 491 retval = es.es_len - (map->m_lblk - es.es_lblk); 492 if (retval > map->m_len) 493 retval = map->m_len; 494 map->m_len = retval; 495 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) { 496 retval = 0; 497 } else { 498 BUG_ON(1); 499 } 500 #ifdef ES_AGGRESSIVE_TEST 501 ext4_map_blocks_es_recheck(handle, inode, map, 502 &orig_map, flags); 503 #endif 504 goto found; 505 } 506 507 /* 508 * Try to see if we can get the block without requesting a new 509 * file system block. 510 */ 511 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) 512 down_read(&EXT4_I(inode)->i_data_sem); 513 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { 514 retval = ext4_ext_map_blocks(handle, inode, map, flags & 515 EXT4_GET_BLOCKS_KEEP_SIZE); 516 } else { 517 retval = ext4_ind_map_blocks(handle, inode, map, flags & 518 EXT4_GET_BLOCKS_KEEP_SIZE); 519 } 520 if (retval > 0) { 521 unsigned int status; 522 523 if (unlikely(retval != map->m_len)) { 524 ext4_warning(inode->i_sb, 525 "ES len assertion failed for inode " 526 "%lu: retval %d != map->m_len %d", 527 inode->i_ino, retval, map->m_len); 528 WARN_ON(1); 529 } 530 531 status = map->m_flags & EXT4_MAP_UNWRITTEN ? 532 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; 533 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && 534 !(status & EXTENT_STATUS_WRITTEN) && 535 ext4_find_delalloc_range(inode, map->m_lblk, 536 map->m_lblk + map->m_len - 1)) 537 status |= EXTENT_STATUS_DELAYED; 538 ret = ext4_es_insert_extent(inode, map->m_lblk, 539 map->m_len, map->m_pblk, status); 540 if (ret < 0) 541 retval = ret; 542 } 543 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) 544 up_read((&EXT4_I(inode)->i_data_sem)); 545 546 found: 547 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { 548 ret = check_block_validity(inode, map); 549 if (ret != 0) 550 return ret; 551 } 552 553 /* If it is only a block(s) look up */ 554 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) 555 return retval; 556 557 /* 558 * Returns if the blocks have already allocated 559 * 560 * Note that if blocks have been preallocated 561 * ext4_ext_get_block() returns the create = 0 562 * with buffer head unmapped. 563 */ 564 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) 565 /* 566 * If we need to convert extent to unwritten 567 * we continue and do the actual work in 568 * ext4_ext_map_blocks() 569 */ 570 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) 571 return retval; 572 573 /* 574 * Here we clear m_flags because after allocating an new extent, 575 * it will be set again. 576 */ 577 map->m_flags &= ~EXT4_MAP_FLAGS; 578 579 /* 580 * New blocks allocate and/or writing to unwritten extent 581 * will possibly result in updating i_data, so we take 582 * the write lock of i_data_sem, and call get_block() 583 * with create == 1 flag. 584 */ 585 down_write(&EXT4_I(inode)->i_data_sem); 586 587 /* 588 * We need to check for EXT4 here because migrate 589 * could have changed the inode type in between 590 */ 591 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { 592 retval = ext4_ext_map_blocks(handle, inode, map, flags); 593 } else { 594 retval = ext4_ind_map_blocks(handle, inode, map, flags); 595 596 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) { 597 /* 598 * We allocated new blocks which will result in 599 * i_data's format changing. Force the migrate 600 * to fail by clearing migrate flags 601 */ 602 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE); 603 } 604 605 /* 606 * Update reserved blocks/metadata blocks after successful 607 * block allocation which had been deferred till now. We don't 608 * support fallocate for non extent files. So we can update 609 * reserve space here. 610 */ 611 if ((retval > 0) && 612 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)) 613 ext4_da_update_reserve_space(inode, retval, 1); 614 } 615 616 if (retval > 0) { 617 unsigned int status; 618 619 if (unlikely(retval != map->m_len)) { 620 ext4_warning(inode->i_sb, 621 "ES len assertion failed for inode " 622 "%lu: retval %d != map->m_len %d", 623 inode->i_ino, retval, map->m_len); 624 WARN_ON(1); 625 } 626 627 /* 628 * If the extent has been zeroed out, we don't need to update 629 * extent status tree. 630 */ 631 if ((flags & EXT4_GET_BLOCKS_PRE_IO) && 632 ext4_es_lookup_extent(inode, map->m_lblk, &es)) { 633 if (ext4_es_is_written(&es)) 634 goto has_zeroout; 635 } 636 status = map->m_flags & EXT4_MAP_UNWRITTEN ? 637 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; 638 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && 639 !(status & EXTENT_STATUS_WRITTEN) && 640 ext4_find_delalloc_range(inode, map->m_lblk, 641 map->m_lblk + map->m_len - 1)) 642 status |= EXTENT_STATUS_DELAYED; 643 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, 644 map->m_pblk, status); 645 if (ret < 0) 646 retval = ret; 647 } 648 649 has_zeroout: 650 up_write((&EXT4_I(inode)->i_data_sem)); 651 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { 652 ret = check_block_validity(inode, map); 653 if (ret != 0) 654 return ret; 655 } 656 return retval; 657 } 658 659 /* Maximum number of blocks we map for direct IO at once. */ 660 #define DIO_MAX_BLOCKS 4096 661 662 static int _ext4_get_block(struct inode *inode, sector_t iblock, 663 struct buffer_head *bh, int flags) 664 { 665 handle_t *handle = ext4_journal_current_handle(); 666 struct ext4_map_blocks map; 667 int ret = 0, started = 0; 668 int dio_credits; 669 670 if (ext4_has_inline_data(inode)) 671 return -ERANGE; 672 673 map.m_lblk = iblock; 674 map.m_len = bh->b_size >> inode->i_blkbits; 675 676 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) { 677 /* Direct IO write... */ 678 if (map.m_len > DIO_MAX_BLOCKS) 679 map.m_len = DIO_MAX_BLOCKS; 680 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len); 681 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, 682 dio_credits); 683 if (IS_ERR(handle)) { 684 ret = PTR_ERR(handle); 685 return ret; 686 } 687 started = 1; 688 } 689 690 ret = ext4_map_blocks(handle, inode, &map, flags); 691 if (ret > 0) { 692 ext4_io_end_t *io_end = ext4_inode_aio(inode); 693 694 map_bh(bh, inode->i_sb, map.m_pblk); 695 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; 696 if (IS_DAX(inode) && buffer_unwritten(bh)) { 697 /* 698 * dgc: I suspect unwritten conversion on ext4+DAX is 699 * fundamentally broken here when there are concurrent 700 * read/write in progress on this inode. 701 */ 702 WARN_ON_ONCE(io_end); 703 bh->b_assoc_map = inode->i_mapping; 704 bh->b_private = (void *)(unsigned long)iblock; 705 } 706 if (io_end && io_end->flag & EXT4_IO_END_UNWRITTEN) 707 set_buffer_defer_completion(bh); 708 bh->b_size = inode->i_sb->s_blocksize * map.m_len; 709 ret = 0; 710 } 711 if (started) 712 ext4_journal_stop(handle); 713 return ret; 714 } 715 716 int ext4_get_block(struct inode *inode, sector_t iblock, 717 struct buffer_head *bh, int create) 718 { 719 return _ext4_get_block(inode, iblock, bh, 720 create ? EXT4_GET_BLOCKS_CREATE : 0); 721 } 722 723 /* 724 * `handle' can be NULL if create is zero 725 */ 726 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, 727 ext4_lblk_t block, int map_flags) 728 { 729 struct ext4_map_blocks map; 730 struct buffer_head *bh; 731 int create = map_flags & EXT4_GET_BLOCKS_CREATE; 732 int err; 733 734 J_ASSERT(handle != NULL || create == 0); 735 736 map.m_lblk = block; 737 map.m_len = 1; 738 err = ext4_map_blocks(handle, inode, &map, map_flags); 739 740 if (err == 0) 741 return create ? ERR_PTR(-ENOSPC) : NULL; 742 if (err < 0) 743 return ERR_PTR(err); 744 745 bh = sb_getblk(inode->i_sb, map.m_pblk); 746 if (unlikely(!bh)) 747 return ERR_PTR(-ENOMEM); 748 if (map.m_flags & EXT4_MAP_NEW) { 749 J_ASSERT(create != 0); 750 J_ASSERT(handle != NULL); 751 752 /* 753 * Now that we do not always journal data, we should 754 * keep in mind whether this should always journal the 755 * new buffer as metadata. For now, regular file 756 * writes use ext4_get_block instead, so it's not a 757 * problem. 758 */ 759 lock_buffer(bh); 760 BUFFER_TRACE(bh, "call get_create_access"); 761 err = ext4_journal_get_create_access(handle, bh); 762 if (unlikely(err)) { 763 unlock_buffer(bh); 764 goto errout; 765 } 766 if (!buffer_uptodate(bh)) { 767 memset(bh->b_data, 0, inode->i_sb->s_blocksize); 768 set_buffer_uptodate(bh); 769 } 770 unlock_buffer(bh); 771 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); 772 err = ext4_handle_dirty_metadata(handle, inode, bh); 773 if (unlikely(err)) 774 goto errout; 775 } else 776 BUFFER_TRACE(bh, "not a new buffer"); 777 return bh; 778 errout: 779 brelse(bh); 780 return ERR_PTR(err); 781 } 782 783 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, 784 ext4_lblk_t block, int map_flags) 785 { 786 struct buffer_head *bh; 787 788 bh = ext4_getblk(handle, inode, block, map_flags); 789 if (IS_ERR(bh)) 790 return bh; 791 if (!bh || buffer_uptodate(bh)) 792 return bh; 793 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh); 794 wait_on_buffer(bh); 795 if (buffer_uptodate(bh)) 796 return bh; 797 put_bh(bh); 798 return ERR_PTR(-EIO); 799 } 800 801 int ext4_walk_page_buffers(handle_t *handle, 802 struct buffer_head *head, 803 unsigned from, 804 unsigned to, 805 int *partial, 806 int (*fn)(handle_t *handle, 807 struct buffer_head *bh)) 808 { 809 struct buffer_head *bh; 810 unsigned block_start, block_end; 811 unsigned blocksize = head->b_size; 812 int err, ret = 0; 813 struct buffer_head *next; 814 815 for (bh = head, block_start = 0; 816 ret == 0 && (bh != head || !block_start); 817 block_start = block_end, bh = next) { 818 next = bh->b_this_page; 819 block_end = block_start + blocksize; 820 if (block_end <= from || block_start >= to) { 821 if (partial && !buffer_uptodate(bh)) 822 *partial = 1; 823 continue; 824 } 825 err = (*fn)(handle, bh); 826 if (!ret) 827 ret = err; 828 } 829 return ret; 830 } 831 832 /* 833 * To preserve ordering, it is essential that the hole instantiation and 834 * the data write be encapsulated in a single transaction. We cannot 835 * close off a transaction and start a new one between the ext4_get_block() 836 * and the commit_write(). So doing the jbd2_journal_start at the start of 837 * prepare_write() is the right place. 838 * 839 * Also, this function can nest inside ext4_writepage(). In that case, we 840 * *know* that ext4_writepage() has generated enough buffer credits to do the 841 * whole page. So we won't block on the journal in that case, which is good, 842 * because the caller may be PF_MEMALLOC. 843 * 844 * By accident, ext4 can be reentered when a transaction is open via 845 * quota file writes. If we were to commit the transaction while thus 846 * reentered, there can be a deadlock - we would be holding a quota 847 * lock, and the commit would never complete if another thread had a 848 * transaction open and was blocking on the quota lock - a ranking 849 * violation. 850 * 851 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start 852 * will _not_ run commit under these circumstances because handle->h_ref 853 * is elevated. We'll still have enough credits for the tiny quotafile 854 * write. 855 */ 856 int do_journal_get_write_access(handle_t *handle, 857 struct buffer_head *bh) 858 { 859 int dirty = buffer_dirty(bh); 860 int ret; 861 862 if (!buffer_mapped(bh) || buffer_freed(bh)) 863 return 0; 864 /* 865 * __block_write_begin() could have dirtied some buffers. Clean 866 * the dirty bit as jbd2_journal_get_write_access() could complain 867 * otherwise about fs integrity issues. Setting of the dirty bit 868 * by __block_write_begin() isn't a real problem here as we clear 869 * the bit before releasing a page lock and thus writeback cannot 870 * ever write the buffer. 871 */ 872 if (dirty) 873 clear_buffer_dirty(bh); 874 BUFFER_TRACE(bh, "get write access"); 875 ret = ext4_journal_get_write_access(handle, bh); 876 if (!ret && dirty) 877 ret = ext4_handle_dirty_metadata(handle, NULL, bh); 878 return ret; 879 } 880 881 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock, 882 struct buffer_head *bh_result, int create); 883 884 #ifdef CONFIG_EXT4_FS_ENCRYPTION 885 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len, 886 get_block_t *get_block) 887 { 888 unsigned from = pos & (PAGE_CACHE_SIZE - 1); 889 unsigned to = from + len; 890 struct inode *inode = page->mapping->host; 891 unsigned block_start, block_end; 892 sector_t block; 893 int err = 0; 894 unsigned blocksize = inode->i_sb->s_blocksize; 895 unsigned bbits; 896 struct buffer_head *bh, *head, *wait[2], **wait_bh = wait; 897 bool decrypt = false; 898 899 BUG_ON(!PageLocked(page)); 900 BUG_ON(from > PAGE_CACHE_SIZE); 901 BUG_ON(to > PAGE_CACHE_SIZE); 902 BUG_ON(from > to); 903 904 if (!page_has_buffers(page)) 905 create_empty_buffers(page, blocksize, 0); 906 head = page_buffers(page); 907 bbits = ilog2(blocksize); 908 block = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits); 909 910 for (bh = head, block_start = 0; bh != head || !block_start; 911 block++, block_start = block_end, bh = bh->b_this_page) { 912 block_end = block_start + blocksize; 913 if (block_end <= from || block_start >= to) { 914 if (PageUptodate(page)) { 915 if (!buffer_uptodate(bh)) 916 set_buffer_uptodate(bh); 917 } 918 continue; 919 } 920 if (buffer_new(bh)) 921 clear_buffer_new(bh); 922 if (!buffer_mapped(bh)) { 923 WARN_ON(bh->b_size != blocksize); 924 err = get_block(inode, block, bh, 1); 925 if (err) 926 break; 927 if (buffer_new(bh)) { 928 unmap_underlying_metadata(bh->b_bdev, 929 bh->b_blocknr); 930 if (PageUptodate(page)) { 931 clear_buffer_new(bh); 932 set_buffer_uptodate(bh); 933 mark_buffer_dirty(bh); 934 continue; 935 } 936 if (block_end > to || block_start < from) 937 zero_user_segments(page, to, block_end, 938 block_start, from); 939 continue; 940 } 941 } 942 if (PageUptodate(page)) { 943 if (!buffer_uptodate(bh)) 944 set_buffer_uptodate(bh); 945 continue; 946 } 947 if (!buffer_uptodate(bh) && !buffer_delay(bh) && 948 !buffer_unwritten(bh) && 949 (block_start < from || block_end > to)) { 950 ll_rw_block(READ, 1, &bh); 951 *wait_bh++ = bh; 952 decrypt = ext4_encrypted_inode(inode) && 953 S_ISREG(inode->i_mode); 954 } 955 } 956 /* 957 * If we issued read requests, let them complete. 958 */ 959 while (wait_bh > wait) { 960 wait_on_buffer(*--wait_bh); 961 if (!buffer_uptodate(*wait_bh)) 962 err = -EIO; 963 } 964 if (unlikely(err)) 965 page_zero_new_buffers(page, from, to); 966 else if (decrypt) 967 err = ext4_decrypt_one(inode, page); 968 return err; 969 } 970 #endif 971 972 static int ext4_write_begin(struct file *file, struct address_space *mapping, 973 loff_t pos, unsigned len, unsigned flags, 974 struct page **pagep, void **fsdata) 975 { 976 struct inode *inode = mapping->host; 977 int ret, needed_blocks; 978 handle_t *handle; 979 int retries = 0; 980 struct page *page; 981 pgoff_t index; 982 unsigned from, to; 983 984 trace_ext4_write_begin(inode, pos, len, flags); 985 /* 986 * Reserve one block more for addition to orphan list in case 987 * we allocate blocks but write fails for some reason 988 */ 989 needed_blocks = ext4_writepage_trans_blocks(inode) + 1; 990 index = pos >> PAGE_CACHE_SHIFT; 991 from = pos & (PAGE_CACHE_SIZE - 1); 992 to = from + len; 993 994 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { 995 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len, 996 flags, pagep); 997 if (ret < 0) 998 return ret; 999 if (ret == 1) 1000 return 0; 1001 } 1002 1003 /* 1004 * grab_cache_page_write_begin() can take a long time if the 1005 * system is thrashing due to memory pressure, or if the page 1006 * is being written back. So grab it first before we start 1007 * the transaction handle. This also allows us to allocate 1008 * the page (if needed) without using GFP_NOFS. 1009 */ 1010 retry_grab: 1011 page = grab_cache_page_write_begin(mapping, index, flags); 1012 if (!page) 1013 return -ENOMEM; 1014 unlock_page(page); 1015 1016 retry_journal: 1017 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); 1018 if (IS_ERR(handle)) { 1019 page_cache_release(page); 1020 return PTR_ERR(handle); 1021 } 1022 1023 lock_page(page); 1024 if (page->mapping != mapping) { 1025 /* The page got truncated from under us */ 1026 unlock_page(page); 1027 page_cache_release(page); 1028 ext4_journal_stop(handle); 1029 goto retry_grab; 1030 } 1031 /* In case writeback began while the page was unlocked */ 1032 wait_for_stable_page(page); 1033 1034 #ifdef CONFIG_EXT4_FS_ENCRYPTION 1035 if (ext4_should_dioread_nolock(inode)) 1036 ret = ext4_block_write_begin(page, pos, len, 1037 ext4_get_block_write); 1038 else 1039 ret = ext4_block_write_begin(page, pos, len, 1040 ext4_get_block); 1041 #else 1042 if (ext4_should_dioread_nolock(inode)) 1043 ret = __block_write_begin(page, pos, len, ext4_get_block_write); 1044 else 1045 ret = __block_write_begin(page, pos, len, ext4_get_block); 1046 #endif 1047 if (!ret && ext4_should_journal_data(inode)) { 1048 ret = ext4_walk_page_buffers(handle, page_buffers(page), 1049 from, to, NULL, 1050 do_journal_get_write_access); 1051 } 1052 1053 if (ret) { 1054 unlock_page(page); 1055 /* 1056 * __block_write_begin may have instantiated a few blocks 1057 * outside i_size. Trim these off again. Don't need 1058 * i_size_read because we hold i_mutex. 1059 * 1060 * Add inode to orphan list in case we crash before 1061 * truncate finishes 1062 */ 1063 if (pos + len > inode->i_size && ext4_can_truncate(inode)) 1064 ext4_orphan_add(handle, inode); 1065 1066 ext4_journal_stop(handle); 1067 if (pos + len > inode->i_size) { 1068 ext4_truncate_failed_write(inode); 1069 /* 1070 * If truncate failed early the inode might 1071 * still be on the orphan list; we need to 1072 * make sure the inode is removed from the 1073 * orphan list in that case. 1074 */ 1075 if (inode->i_nlink) 1076 ext4_orphan_del(NULL, inode); 1077 } 1078 1079 if (ret == -ENOSPC && 1080 ext4_should_retry_alloc(inode->i_sb, &retries)) 1081 goto retry_journal; 1082 page_cache_release(page); 1083 return ret; 1084 } 1085 *pagep = page; 1086 return ret; 1087 } 1088 1089 /* For write_end() in data=journal mode */ 1090 static int write_end_fn(handle_t *handle, struct buffer_head *bh) 1091 { 1092 int ret; 1093 if (!buffer_mapped(bh) || buffer_freed(bh)) 1094 return 0; 1095 set_buffer_uptodate(bh); 1096 ret = ext4_handle_dirty_metadata(handle, NULL, bh); 1097 clear_buffer_meta(bh); 1098 clear_buffer_prio(bh); 1099 return ret; 1100 } 1101 1102 /* 1103 * We need to pick up the new inode size which generic_commit_write gave us 1104 * `file' can be NULL - eg, when called from page_symlink(). 1105 * 1106 * ext4 never places buffers on inode->i_mapping->private_list. metadata 1107 * buffers are managed internally. 1108 */ 1109 static int ext4_write_end(struct file *file, 1110 struct address_space *mapping, 1111 loff_t pos, unsigned len, unsigned copied, 1112 struct page *page, void *fsdata) 1113 { 1114 handle_t *handle = ext4_journal_current_handle(); 1115 struct inode *inode = mapping->host; 1116 loff_t old_size = inode->i_size; 1117 int ret = 0, ret2; 1118 int i_size_changed = 0; 1119 1120 trace_ext4_write_end(inode, pos, len, copied); 1121 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) { 1122 ret = ext4_jbd2_file_inode(handle, inode); 1123 if (ret) { 1124 unlock_page(page); 1125 page_cache_release(page); 1126 goto errout; 1127 } 1128 } 1129 1130 if (ext4_has_inline_data(inode)) { 1131 ret = ext4_write_inline_data_end(inode, pos, len, 1132 copied, page); 1133 if (ret < 0) 1134 goto errout; 1135 copied = ret; 1136 } else 1137 copied = block_write_end(file, mapping, pos, 1138 len, copied, page, fsdata); 1139 /* 1140 * it's important to update i_size while still holding page lock: 1141 * page writeout could otherwise come in and zero beyond i_size. 1142 */ 1143 i_size_changed = ext4_update_inode_size(inode, pos + copied); 1144 unlock_page(page); 1145 page_cache_release(page); 1146 1147 if (old_size < pos) 1148 pagecache_isize_extended(inode, old_size, pos); 1149 /* 1150 * Don't mark the inode dirty under page lock. First, it unnecessarily 1151 * makes the holding time of page lock longer. Second, it forces lock 1152 * ordering of page lock and transaction start for journaling 1153 * filesystems. 1154 */ 1155 if (i_size_changed) 1156 ext4_mark_inode_dirty(handle, inode); 1157 1158 if (pos + len > inode->i_size && ext4_can_truncate(inode)) 1159 /* if we have allocated more blocks and copied 1160 * less. We will have blocks allocated outside 1161 * inode->i_size. So truncate them 1162 */ 1163 ext4_orphan_add(handle, inode); 1164 errout: 1165 ret2 = ext4_journal_stop(handle); 1166 if (!ret) 1167 ret = ret2; 1168 1169 if (pos + len > inode->i_size) { 1170 ext4_truncate_failed_write(inode); 1171 /* 1172 * If truncate failed early the inode might still be 1173 * on the orphan list; we need to make sure the inode 1174 * is removed from the orphan list in that case. 1175 */ 1176 if (inode->i_nlink) 1177 ext4_orphan_del(NULL, inode); 1178 } 1179 1180 return ret ? ret : copied; 1181 } 1182 1183 static int ext4_journalled_write_end(struct file *file, 1184 struct address_space *mapping, 1185 loff_t pos, unsigned len, unsigned copied, 1186 struct page *page, void *fsdata) 1187 { 1188 handle_t *handle = ext4_journal_current_handle(); 1189 struct inode *inode = mapping->host; 1190 loff_t old_size = inode->i_size; 1191 int ret = 0, ret2; 1192 int partial = 0; 1193 unsigned from, to; 1194 int size_changed = 0; 1195 1196 trace_ext4_journalled_write_end(inode, pos, len, copied); 1197 from = pos & (PAGE_CACHE_SIZE - 1); 1198 to = from + len; 1199 1200 BUG_ON(!ext4_handle_valid(handle)); 1201 1202 if (ext4_has_inline_data(inode)) 1203 copied = ext4_write_inline_data_end(inode, pos, len, 1204 copied, page); 1205 else { 1206 if (copied < len) { 1207 if (!PageUptodate(page)) 1208 copied = 0; 1209 page_zero_new_buffers(page, from+copied, to); 1210 } 1211 1212 ret = ext4_walk_page_buffers(handle, page_buffers(page), from, 1213 to, &partial, write_end_fn); 1214 if (!partial) 1215 SetPageUptodate(page); 1216 } 1217 size_changed = ext4_update_inode_size(inode, pos + copied); 1218 ext4_set_inode_state(inode, EXT4_STATE_JDATA); 1219 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; 1220 unlock_page(page); 1221 page_cache_release(page); 1222 1223 if (old_size < pos) 1224 pagecache_isize_extended(inode, old_size, pos); 1225 1226 if (size_changed) { 1227 ret2 = ext4_mark_inode_dirty(handle, inode); 1228 if (!ret) 1229 ret = ret2; 1230 } 1231 1232 if (pos + len > inode->i_size && ext4_can_truncate(inode)) 1233 /* if we have allocated more blocks and copied 1234 * less. We will have blocks allocated outside 1235 * inode->i_size. So truncate them 1236 */ 1237 ext4_orphan_add(handle, inode); 1238 1239 ret2 = ext4_journal_stop(handle); 1240 if (!ret) 1241 ret = ret2; 1242 if (pos + len > inode->i_size) { 1243 ext4_truncate_failed_write(inode); 1244 /* 1245 * If truncate failed early the inode might still be 1246 * on the orphan list; we need to make sure the inode 1247 * is removed from the orphan list in that case. 1248 */ 1249 if (inode->i_nlink) 1250 ext4_orphan_del(NULL, inode); 1251 } 1252 1253 return ret ? ret : copied; 1254 } 1255 1256 /* 1257 * Reserve space for a single cluster 1258 */ 1259 static int ext4_da_reserve_space(struct inode *inode) 1260 { 1261 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 1262 struct ext4_inode_info *ei = EXT4_I(inode); 1263 int ret; 1264 1265 /* 1266 * We will charge metadata quota at writeout time; this saves 1267 * us from metadata over-estimation, though we may go over by 1268 * a small amount in the end. Here we just reserve for data. 1269 */ 1270 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1)); 1271 if (ret) 1272 return ret; 1273 1274 spin_lock(&ei->i_block_reservation_lock); 1275 if (ext4_claim_free_clusters(sbi, 1, 0)) { 1276 spin_unlock(&ei->i_block_reservation_lock); 1277 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1)); 1278 return -ENOSPC; 1279 } 1280 ei->i_reserved_data_blocks++; 1281 trace_ext4_da_reserve_space(inode); 1282 spin_unlock(&ei->i_block_reservation_lock); 1283 1284 return 0; /* success */ 1285 } 1286 1287 static void ext4_da_release_space(struct inode *inode, int to_free) 1288 { 1289 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 1290 struct ext4_inode_info *ei = EXT4_I(inode); 1291 1292 if (!to_free) 1293 return; /* Nothing to release, exit */ 1294 1295 spin_lock(&EXT4_I(inode)->i_block_reservation_lock); 1296 1297 trace_ext4_da_release_space(inode, to_free); 1298 if (unlikely(to_free > ei->i_reserved_data_blocks)) { 1299 /* 1300 * if there aren't enough reserved blocks, then the 1301 * counter is messed up somewhere. Since this 1302 * function is called from invalidate page, it's 1303 * harmless to return without any action. 1304 */ 1305 ext4_warning(inode->i_sb, "ext4_da_release_space: " 1306 "ino %lu, to_free %d with only %d reserved " 1307 "data blocks", inode->i_ino, to_free, 1308 ei->i_reserved_data_blocks); 1309 WARN_ON(1); 1310 to_free = ei->i_reserved_data_blocks; 1311 } 1312 ei->i_reserved_data_blocks -= to_free; 1313 1314 /* update fs dirty data blocks counter */ 1315 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free); 1316 1317 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); 1318 1319 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free)); 1320 } 1321 1322 static void ext4_da_page_release_reservation(struct page *page, 1323 unsigned int offset, 1324 unsigned int length) 1325 { 1326 int to_release = 0; 1327 struct buffer_head *head, *bh; 1328 unsigned int curr_off = 0; 1329 struct inode *inode = page->mapping->host; 1330 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 1331 unsigned int stop = offset + length; 1332 int num_clusters; 1333 ext4_fsblk_t lblk; 1334 1335 BUG_ON(stop > PAGE_CACHE_SIZE || stop < length); 1336 1337 head = page_buffers(page); 1338 bh = head; 1339 do { 1340 unsigned int next_off = curr_off + bh->b_size; 1341 1342 if (next_off > stop) 1343 break; 1344 1345 if ((offset <= curr_off) && (buffer_delay(bh))) { 1346 to_release++; 1347 clear_buffer_delay(bh); 1348 } 1349 curr_off = next_off; 1350 } while ((bh = bh->b_this_page) != head); 1351 1352 if (to_release) { 1353 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits); 1354 ext4_es_remove_extent(inode, lblk, to_release); 1355 } 1356 1357 /* If we have released all the blocks belonging to a cluster, then we 1358 * need to release the reserved space for that cluster. */ 1359 num_clusters = EXT4_NUM_B2C(sbi, to_release); 1360 while (num_clusters > 0) { 1361 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) + 1362 ((num_clusters - 1) << sbi->s_cluster_bits); 1363 if (sbi->s_cluster_ratio == 1 || 1364 !ext4_find_delalloc_cluster(inode, lblk)) 1365 ext4_da_release_space(inode, 1); 1366 1367 num_clusters--; 1368 } 1369 } 1370 1371 /* 1372 * Delayed allocation stuff 1373 */ 1374 1375 struct mpage_da_data { 1376 struct inode *inode; 1377 struct writeback_control *wbc; 1378 1379 pgoff_t first_page; /* The first page to write */ 1380 pgoff_t next_page; /* Current page to examine */ 1381 pgoff_t last_page; /* Last page to examine */ 1382 /* 1383 * Extent to map - this can be after first_page because that can be 1384 * fully mapped. We somewhat abuse m_flags to store whether the extent 1385 * is delalloc or unwritten. 1386 */ 1387 struct ext4_map_blocks map; 1388 struct ext4_io_submit io_submit; /* IO submission data */ 1389 }; 1390 1391 static void mpage_release_unused_pages(struct mpage_da_data *mpd, 1392 bool invalidate) 1393 { 1394 int nr_pages, i; 1395 pgoff_t index, end; 1396 struct pagevec pvec; 1397 struct inode *inode = mpd->inode; 1398 struct address_space *mapping = inode->i_mapping; 1399 1400 /* This is necessary when next_page == 0. */ 1401 if (mpd->first_page >= mpd->next_page) 1402 return; 1403 1404 index = mpd->first_page; 1405 end = mpd->next_page - 1; 1406 if (invalidate) { 1407 ext4_lblk_t start, last; 1408 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits); 1409 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits); 1410 ext4_es_remove_extent(inode, start, last - start + 1); 1411 } 1412 1413 pagevec_init(&pvec, 0); 1414 while (index <= end) { 1415 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); 1416 if (nr_pages == 0) 1417 break; 1418 for (i = 0; i < nr_pages; i++) { 1419 struct page *page = pvec.pages[i]; 1420 if (page->index > end) 1421 break; 1422 BUG_ON(!PageLocked(page)); 1423 BUG_ON(PageWriteback(page)); 1424 if (invalidate) { 1425 block_invalidatepage(page, 0, PAGE_CACHE_SIZE); 1426 ClearPageUptodate(page); 1427 } 1428 unlock_page(page); 1429 } 1430 index = pvec.pages[nr_pages - 1]->index + 1; 1431 pagevec_release(&pvec); 1432 } 1433 } 1434 1435 static void ext4_print_free_blocks(struct inode *inode) 1436 { 1437 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 1438 struct super_block *sb = inode->i_sb; 1439 struct ext4_inode_info *ei = EXT4_I(inode); 1440 1441 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld", 1442 EXT4_C2B(EXT4_SB(inode->i_sb), 1443 ext4_count_free_clusters(sb))); 1444 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details"); 1445 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld", 1446 (long long) EXT4_C2B(EXT4_SB(sb), 1447 percpu_counter_sum(&sbi->s_freeclusters_counter))); 1448 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld", 1449 (long long) EXT4_C2B(EXT4_SB(sb), 1450 percpu_counter_sum(&sbi->s_dirtyclusters_counter))); 1451 ext4_msg(sb, KERN_CRIT, "Block reservation details"); 1452 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u", 1453 ei->i_reserved_data_blocks); 1454 return; 1455 } 1456 1457 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh) 1458 { 1459 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh); 1460 } 1461 1462 /* 1463 * This function is grabs code from the very beginning of 1464 * ext4_map_blocks, but assumes that the caller is from delayed write 1465 * time. This function looks up the requested blocks and sets the 1466 * buffer delay bit under the protection of i_data_sem. 1467 */ 1468 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock, 1469 struct ext4_map_blocks *map, 1470 struct buffer_head *bh) 1471 { 1472 struct extent_status es; 1473 int retval; 1474 sector_t invalid_block = ~((sector_t) 0xffff); 1475 #ifdef ES_AGGRESSIVE_TEST 1476 struct ext4_map_blocks orig_map; 1477 1478 memcpy(&orig_map, map, sizeof(*map)); 1479 #endif 1480 1481 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) 1482 invalid_block = ~0; 1483 1484 map->m_flags = 0; 1485 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u," 1486 "logical block %lu\n", inode->i_ino, map->m_len, 1487 (unsigned long) map->m_lblk); 1488 1489 /* Lookup extent status tree firstly */ 1490 if (ext4_es_lookup_extent(inode, iblock, &es)) { 1491 if (ext4_es_is_hole(&es)) { 1492 retval = 0; 1493 down_read(&EXT4_I(inode)->i_data_sem); 1494 goto add_delayed; 1495 } 1496 1497 /* 1498 * Delayed extent could be allocated by fallocate. 1499 * So we need to check it. 1500 */ 1501 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) { 1502 map_bh(bh, inode->i_sb, invalid_block); 1503 set_buffer_new(bh); 1504 set_buffer_delay(bh); 1505 return 0; 1506 } 1507 1508 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk; 1509 retval = es.es_len - (iblock - es.es_lblk); 1510 if (retval > map->m_len) 1511 retval = map->m_len; 1512 map->m_len = retval; 1513 if (ext4_es_is_written(&es)) 1514 map->m_flags |= EXT4_MAP_MAPPED; 1515 else if (ext4_es_is_unwritten(&es)) 1516 map->m_flags |= EXT4_MAP_UNWRITTEN; 1517 else 1518 BUG_ON(1); 1519 1520 #ifdef ES_AGGRESSIVE_TEST 1521 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0); 1522 #endif 1523 return retval; 1524 } 1525 1526 /* 1527 * Try to see if we can get the block without requesting a new 1528 * file system block. 1529 */ 1530 down_read(&EXT4_I(inode)->i_data_sem); 1531 if (ext4_has_inline_data(inode)) 1532 retval = 0; 1533 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) 1534 retval = ext4_ext_map_blocks(NULL, inode, map, 0); 1535 else 1536 retval = ext4_ind_map_blocks(NULL, inode, map, 0); 1537 1538 add_delayed: 1539 if (retval == 0) { 1540 int ret; 1541 /* 1542 * XXX: __block_prepare_write() unmaps passed block, 1543 * is it OK? 1544 */ 1545 /* 1546 * If the block was allocated from previously allocated cluster, 1547 * then we don't need to reserve it again. However we still need 1548 * to reserve metadata for every block we're going to write. 1549 */ 1550 if (EXT4_SB(inode->i_sb)->s_cluster_ratio == 1 || 1551 !ext4_find_delalloc_cluster(inode, map->m_lblk)) { 1552 ret = ext4_da_reserve_space(inode); 1553 if (ret) { 1554 /* not enough space to reserve */ 1555 retval = ret; 1556 goto out_unlock; 1557 } 1558 } 1559 1560 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, 1561 ~0, EXTENT_STATUS_DELAYED); 1562 if (ret) { 1563 retval = ret; 1564 goto out_unlock; 1565 } 1566 1567 map_bh(bh, inode->i_sb, invalid_block); 1568 set_buffer_new(bh); 1569 set_buffer_delay(bh); 1570 } else if (retval > 0) { 1571 int ret; 1572 unsigned int status; 1573 1574 if (unlikely(retval != map->m_len)) { 1575 ext4_warning(inode->i_sb, 1576 "ES len assertion failed for inode " 1577 "%lu: retval %d != map->m_len %d", 1578 inode->i_ino, retval, map->m_len); 1579 WARN_ON(1); 1580 } 1581 1582 status = map->m_flags & EXT4_MAP_UNWRITTEN ? 1583 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; 1584 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, 1585 map->m_pblk, status); 1586 if (ret != 0) 1587 retval = ret; 1588 } 1589 1590 out_unlock: 1591 up_read((&EXT4_I(inode)->i_data_sem)); 1592 1593 return retval; 1594 } 1595 1596 /* 1597 * This is a special get_block_t callback which is used by 1598 * ext4_da_write_begin(). It will either return mapped block or 1599 * reserve space for a single block. 1600 * 1601 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. 1602 * We also have b_blocknr = -1 and b_bdev initialized properly 1603 * 1604 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. 1605 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev 1606 * initialized properly. 1607 */ 1608 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, 1609 struct buffer_head *bh, int create) 1610 { 1611 struct ext4_map_blocks map; 1612 int ret = 0; 1613 1614 BUG_ON(create == 0); 1615 BUG_ON(bh->b_size != inode->i_sb->s_blocksize); 1616 1617 map.m_lblk = iblock; 1618 map.m_len = 1; 1619 1620 /* 1621 * first, we need to know whether the block is allocated already 1622 * preallocated blocks are unmapped but should treated 1623 * the same as allocated blocks. 1624 */ 1625 ret = ext4_da_map_blocks(inode, iblock, &map, bh); 1626 if (ret <= 0) 1627 return ret; 1628 1629 map_bh(bh, inode->i_sb, map.m_pblk); 1630 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; 1631 1632 if (buffer_unwritten(bh)) { 1633 /* A delayed write to unwritten bh should be marked 1634 * new and mapped. Mapped ensures that we don't do 1635 * get_block multiple times when we write to the same 1636 * offset and new ensures that we do proper zero out 1637 * for partial write. 1638 */ 1639 set_buffer_new(bh); 1640 set_buffer_mapped(bh); 1641 } 1642 return 0; 1643 } 1644 1645 static int bget_one(handle_t *handle, struct buffer_head *bh) 1646 { 1647 get_bh(bh); 1648 return 0; 1649 } 1650 1651 static int bput_one(handle_t *handle, struct buffer_head *bh) 1652 { 1653 put_bh(bh); 1654 return 0; 1655 } 1656 1657 static int __ext4_journalled_writepage(struct page *page, 1658 unsigned int len) 1659 { 1660 struct address_space *mapping = page->mapping; 1661 struct inode *inode = mapping->host; 1662 struct buffer_head *page_bufs = NULL; 1663 handle_t *handle = NULL; 1664 int ret = 0, err = 0; 1665 int inline_data = ext4_has_inline_data(inode); 1666 struct buffer_head *inode_bh = NULL; 1667 1668 ClearPageChecked(page); 1669 1670 if (inline_data) { 1671 BUG_ON(page->index != 0); 1672 BUG_ON(len > ext4_get_max_inline_size(inode)); 1673 inode_bh = ext4_journalled_write_inline_data(inode, len, page); 1674 if (inode_bh == NULL) 1675 goto out; 1676 } else { 1677 page_bufs = page_buffers(page); 1678 if (!page_bufs) { 1679 BUG(); 1680 goto out; 1681 } 1682 ext4_walk_page_buffers(handle, page_bufs, 0, len, 1683 NULL, bget_one); 1684 } 1685 /* 1686 * We need to release the page lock before we start the 1687 * journal, so grab a reference so the page won't disappear 1688 * out from under us. 1689 */ 1690 get_page(page); 1691 unlock_page(page); 1692 1693 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, 1694 ext4_writepage_trans_blocks(inode)); 1695 if (IS_ERR(handle)) { 1696 ret = PTR_ERR(handle); 1697 put_page(page); 1698 goto out_no_pagelock; 1699 } 1700 BUG_ON(!ext4_handle_valid(handle)); 1701 1702 lock_page(page); 1703 put_page(page); 1704 if (page->mapping != mapping) { 1705 /* The page got truncated from under us */ 1706 ext4_journal_stop(handle); 1707 ret = 0; 1708 goto out; 1709 } 1710 1711 if (inline_data) { 1712 BUFFER_TRACE(inode_bh, "get write access"); 1713 ret = ext4_journal_get_write_access(handle, inode_bh); 1714 1715 err = ext4_handle_dirty_metadata(handle, inode, inode_bh); 1716 1717 } else { 1718 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, 1719 do_journal_get_write_access); 1720 1721 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, 1722 write_end_fn); 1723 } 1724 if (ret == 0) 1725 ret = err; 1726 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; 1727 err = ext4_journal_stop(handle); 1728 if (!ret) 1729 ret = err; 1730 1731 if (!ext4_has_inline_data(inode)) 1732 ext4_walk_page_buffers(NULL, page_bufs, 0, len, 1733 NULL, bput_one); 1734 ext4_set_inode_state(inode, EXT4_STATE_JDATA); 1735 out: 1736 unlock_page(page); 1737 out_no_pagelock: 1738 brelse(inode_bh); 1739 return ret; 1740 } 1741 1742 /* 1743 * Note that we don't need to start a transaction unless we're journaling data 1744 * because we should have holes filled from ext4_page_mkwrite(). We even don't 1745 * need to file the inode to the transaction's list in ordered mode because if 1746 * we are writing back data added by write(), the inode is already there and if 1747 * we are writing back data modified via mmap(), no one guarantees in which 1748 * transaction the data will hit the disk. In case we are journaling data, we 1749 * cannot start transaction directly because transaction start ranks above page 1750 * lock so we have to do some magic. 1751 * 1752 * This function can get called via... 1753 * - ext4_writepages after taking page lock (have journal handle) 1754 * - journal_submit_inode_data_buffers (no journal handle) 1755 * - shrink_page_list via the kswapd/direct reclaim (no journal handle) 1756 * - grab_page_cache when doing write_begin (have journal handle) 1757 * 1758 * We don't do any block allocation in this function. If we have page with 1759 * multiple blocks we need to write those buffer_heads that are mapped. This 1760 * is important for mmaped based write. So if we do with blocksize 1K 1761 * truncate(f, 1024); 1762 * a = mmap(f, 0, 4096); 1763 * a[0] = 'a'; 1764 * truncate(f, 4096); 1765 * we have in the page first buffer_head mapped via page_mkwrite call back 1766 * but other buffer_heads would be unmapped but dirty (dirty done via the 1767 * do_wp_page). So writepage should write the first block. If we modify 1768 * the mmap area beyond 1024 we will again get a page_fault and the 1769 * page_mkwrite callback will do the block allocation and mark the 1770 * buffer_heads mapped. 1771 * 1772 * We redirty the page if we have any buffer_heads that is either delay or 1773 * unwritten in the page. 1774 * 1775 * We can get recursively called as show below. 1776 * 1777 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> 1778 * ext4_writepage() 1779 * 1780 * But since we don't do any block allocation we should not deadlock. 1781 * Page also have the dirty flag cleared so we don't get recurive page_lock. 1782 */ 1783 static int ext4_writepage(struct page *page, 1784 struct writeback_control *wbc) 1785 { 1786 int ret = 0; 1787 loff_t size; 1788 unsigned int len; 1789 struct buffer_head *page_bufs = NULL; 1790 struct inode *inode = page->mapping->host; 1791 struct ext4_io_submit io_submit; 1792 bool keep_towrite = false; 1793 1794 trace_ext4_writepage(page); 1795 size = i_size_read(inode); 1796 if (page->index == size >> PAGE_CACHE_SHIFT) 1797 len = size & ~PAGE_CACHE_MASK; 1798 else 1799 len = PAGE_CACHE_SIZE; 1800 1801 page_bufs = page_buffers(page); 1802 /* 1803 * We cannot do block allocation or other extent handling in this 1804 * function. If there are buffers needing that, we have to redirty 1805 * the page. But we may reach here when we do a journal commit via 1806 * journal_submit_inode_data_buffers() and in that case we must write 1807 * allocated buffers to achieve data=ordered mode guarantees. 1808 */ 1809 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL, 1810 ext4_bh_delay_or_unwritten)) { 1811 redirty_page_for_writepage(wbc, page); 1812 if (current->flags & PF_MEMALLOC) { 1813 /* 1814 * For memory cleaning there's no point in writing only 1815 * some buffers. So just bail out. Warn if we came here 1816 * from direct reclaim. 1817 */ 1818 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) 1819 == PF_MEMALLOC); 1820 unlock_page(page); 1821 return 0; 1822 } 1823 keep_towrite = true; 1824 } 1825 1826 if (PageChecked(page) && ext4_should_journal_data(inode)) 1827 /* 1828 * It's mmapped pagecache. Add buffers and journal it. There 1829 * doesn't seem much point in redirtying the page here. 1830 */ 1831 return __ext4_journalled_writepage(page, len); 1832 1833 ext4_io_submit_init(&io_submit, wbc); 1834 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS); 1835 if (!io_submit.io_end) { 1836 redirty_page_for_writepage(wbc, page); 1837 unlock_page(page); 1838 return -ENOMEM; 1839 } 1840 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite); 1841 ext4_io_submit(&io_submit); 1842 /* Drop io_end reference we got from init */ 1843 ext4_put_io_end_defer(io_submit.io_end); 1844 return ret; 1845 } 1846 1847 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page) 1848 { 1849 int len; 1850 loff_t size = i_size_read(mpd->inode); 1851 int err; 1852 1853 BUG_ON(page->index != mpd->first_page); 1854 if (page->index == size >> PAGE_CACHE_SHIFT) 1855 len = size & ~PAGE_CACHE_MASK; 1856 else 1857 len = PAGE_CACHE_SIZE; 1858 clear_page_dirty_for_io(page); 1859 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false); 1860 if (!err) 1861 mpd->wbc->nr_to_write--; 1862 mpd->first_page++; 1863 1864 return err; 1865 } 1866 1867 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay)) 1868 1869 /* 1870 * mballoc gives us at most this number of blocks... 1871 * XXX: That seems to be only a limitation of ext4_mb_normalize_request(). 1872 * The rest of mballoc seems to handle chunks up to full group size. 1873 */ 1874 #define MAX_WRITEPAGES_EXTENT_LEN 2048 1875 1876 /* 1877 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map 1878 * 1879 * @mpd - extent of blocks 1880 * @lblk - logical number of the block in the file 1881 * @bh - buffer head we want to add to the extent 1882 * 1883 * The function is used to collect contig. blocks in the same state. If the 1884 * buffer doesn't require mapping for writeback and we haven't started the 1885 * extent of buffers to map yet, the function returns 'true' immediately - the 1886 * caller can write the buffer right away. Otherwise the function returns true 1887 * if the block has been added to the extent, false if the block couldn't be 1888 * added. 1889 */ 1890 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk, 1891 struct buffer_head *bh) 1892 { 1893 struct ext4_map_blocks *map = &mpd->map; 1894 1895 /* Buffer that doesn't need mapping for writeback? */ 1896 if (!buffer_dirty(bh) || !buffer_mapped(bh) || 1897 (!buffer_delay(bh) && !buffer_unwritten(bh))) { 1898 /* So far no extent to map => we write the buffer right away */ 1899 if (map->m_len == 0) 1900 return true; 1901 return false; 1902 } 1903 1904 /* First block in the extent? */ 1905 if (map->m_len == 0) { 1906 map->m_lblk = lblk; 1907 map->m_len = 1; 1908 map->m_flags = bh->b_state & BH_FLAGS; 1909 return true; 1910 } 1911 1912 /* Don't go larger than mballoc is willing to allocate */ 1913 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN) 1914 return false; 1915 1916 /* Can we merge the block to our big extent? */ 1917 if (lblk == map->m_lblk + map->m_len && 1918 (bh->b_state & BH_FLAGS) == map->m_flags) { 1919 map->m_len++; 1920 return true; 1921 } 1922 return false; 1923 } 1924 1925 /* 1926 * mpage_process_page_bufs - submit page buffers for IO or add them to extent 1927 * 1928 * @mpd - extent of blocks for mapping 1929 * @head - the first buffer in the page 1930 * @bh - buffer we should start processing from 1931 * @lblk - logical number of the block in the file corresponding to @bh 1932 * 1933 * Walk through page buffers from @bh upto @head (exclusive) and either submit 1934 * the page for IO if all buffers in this page were mapped and there's no 1935 * accumulated extent of buffers to map or add buffers in the page to the 1936 * extent of buffers to map. The function returns 1 if the caller can continue 1937 * by processing the next page, 0 if it should stop adding buffers to the 1938 * extent to map because we cannot extend it anymore. It can also return value 1939 * < 0 in case of error during IO submission. 1940 */ 1941 static int mpage_process_page_bufs(struct mpage_da_data *mpd, 1942 struct buffer_head *head, 1943 struct buffer_head *bh, 1944 ext4_lblk_t lblk) 1945 { 1946 struct inode *inode = mpd->inode; 1947 int err; 1948 ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1) 1949 >> inode->i_blkbits; 1950 1951 do { 1952 BUG_ON(buffer_locked(bh)); 1953 1954 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) { 1955 /* Found extent to map? */ 1956 if (mpd->map.m_len) 1957 return 0; 1958 /* Everything mapped so far and we hit EOF */ 1959 break; 1960 } 1961 } while (lblk++, (bh = bh->b_this_page) != head); 1962 /* So far everything mapped? Submit the page for IO. */ 1963 if (mpd->map.m_len == 0) { 1964 err = mpage_submit_page(mpd, head->b_page); 1965 if (err < 0) 1966 return err; 1967 } 1968 return lblk < blocks; 1969 } 1970 1971 /* 1972 * mpage_map_buffers - update buffers corresponding to changed extent and 1973 * submit fully mapped pages for IO 1974 * 1975 * @mpd - description of extent to map, on return next extent to map 1976 * 1977 * Scan buffers corresponding to changed extent (we expect corresponding pages 1978 * to be already locked) and update buffer state according to new extent state. 1979 * We map delalloc buffers to their physical location, clear unwritten bits, 1980 * and mark buffers as uninit when we perform writes to unwritten extents 1981 * and do extent conversion after IO is finished. If the last page is not fully 1982 * mapped, we update @map to the next extent in the last page that needs 1983 * mapping. Otherwise we submit the page for IO. 1984 */ 1985 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd) 1986 { 1987 struct pagevec pvec; 1988 int nr_pages, i; 1989 struct inode *inode = mpd->inode; 1990 struct buffer_head *head, *bh; 1991 int bpp_bits = PAGE_CACHE_SHIFT - inode->i_blkbits; 1992 pgoff_t start, end; 1993 ext4_lblk_t lblk; 1994 sector_t pblock; 1995 int err; 1996 1997 start = mpd->map.m_lblk >> bpp_bits; 1998 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits; 1999 lblk = start << bpp_bits; 2000 pblock = mpd->map.m_pblk; 2001 2002 pagevec_init(&pvec, 0); 2003 while (start <= end) { 2004 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start, 2005 PAGEVEC_SIZE); 2006 if (nr_pages == 0) 2007 break; 2008 for (i = 0; i < nr_pages; i++) { 2009 struct page *page = pvec.pages[i]; 2010 2011 if (page->index > end) 2012 break; 2013 /* Up to 'end' pages must be contiguous */ 2014 BUG_ON(page->index != start); 2015 bh = head = page_buffers(page); 2016 do { 2017 if (lblk < mpd->map.m_lblk) 2018 continue; 2019 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) { 2020 /* 2021 * Buffer after end of mapped extent. 2022 * Find next buffer in the page to map. 2023 */ 2024 mpd->map.m_len = 0; 2025 mpd->map.m_flags = 0; 2026 /* 2027 * FIXME: If dioread_nolock supports 2028 * blocksize < pagesize, we need to make 2029 * sure we add size mapped so far to 2030 * io_end->size as the following call 2031 * can submit the page for IO. 2032 */ 2033 err = mpage_process_page_bufs(mpd, head, 2034 bh, lblk); 2035 pagevec_release(&pvec); 2036 if (err > 0) 2037 err = 0; 2038 return err; 2039 } 2040 if (buffer_delay(bh)) { 2041 clear_buffer_delay(bh); 2042 bh->b_blocknr = pblock++; 2043 } 2044 clear_buffer_unwritten(bh); 2045 } while (lblk++, (bh = bh->b_this_page) != head); 2046 2047 /* 2048 * FIXME: This is going to break if dioread_nolock 2049 * supports blocksize < pagesize as we will try to 2050 * convert potentially unmapped parts of inode. 2051 */ 2052 mpd->io_submit.io_end->size += PAGE_CACHE_SIZE; 2053 /* Page fully mapped - let IO run! */ 2054 err = mpage_submit_page(mpd, page); 2055 if (err < 0) { 2056 pagevec_release(&pvec); 2057 return err; 2058 } 2059 start++; 2060 } 2061 pagevec_release(&pvec); 2062 } 2063 /* Extent fully mapped and matches with page boundary. We are done. */ 2064 mpd->map.m_len = 0; 2065 mpd->map.m_flags = 0; 2066 return 0; 2067 } 2068 2069 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd) 2070 { 2071 struct inode *inode = mpd->inode; 2072 struct ext4_map_blocks *map = &mpd->map; 2073 int get_blocks_flags; 2074 int err, dioread_nolock; 2075 2076 trace_ext4_da_write_pages_extent(inode, map); 2077 /* 2078 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or 2079 * to convert an unwritten extent to be initialized (in the case 2080 * where we have written into one or more preallocated blocks). It is 2081 * possible that we're going to need more metadata blocks than 2082 * previously reserved. However we must not fail because we're in 2083 * writeback and there is nothing we can do about it so it might result 2084 * in data loss. So use reserved blocks to allocate metadata if 2085 * possible. 2086 * 2087 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if 2088 * the blocks in question are delalloc blocks. This indicates 2089 * that the blocks and quotas has already been checked when 2090 * the data was copied into the page cache. 2091 */ 2092 get_blocks_flags = EXT4_GET_BLOCKS_CREATE | 2093 EXT4_GET_BLOCKS_METADATA_NOFAIL; 2094 dioread_nolock = ext4_should_dioread_nolock(inode); 2095 if (dioread_nolock) 2096 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; 2097 if (map->m_flags & (1 << BH_Delay)) 2098 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; 2099 2100 err = ext4_map_blocks(handle, inode, map, get_blocks_flags); 2101 if (err < 0) 2102 return err; 2103 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) { 2104 if (!mpd->io_submit.io_end->handle && 2105 ext4_handle_valid(handle)) { 2106 mpd->io_submit.io_end->handle = handle->h_rsv_handle; 2107 handle->h_rsv_handle = NULL; 2108 } 2109 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end); 2110 } 2111 2112 BUG_ON(map->m_len == 0); 2113 if (map->m_flags & EXT4_MAP_NEW) { 2114 struct block_device *bdev = inode->i_sb->s_bdev; 2115 int i; 2116 2117 for (i = 0; i < map->m_len; i++) 2118 unmap_underlying_metadata(bdev, map->m_pblk + i); 2119 } 2120 return 0; 2121 } 2122 2123 /* 2124 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length 2125 * mpd->len and submit pages underlying it for IO 2126 * 2127 * @handle - handle for journal operations 2128 * @mpd - extent to map 2129 * @give_up_on_write - we set this to true iff there is a fatal error and there 2130 * is no hope of writing the data. The caller should discard 2131 * dirty pages to avoid infinite loops. 2132 * 2133 * The function maps extent starting at mpd->lblk of length mpd->len. If it is 2134 * delayed, blocks are allocated, if it is unwritten, we may need to convert 2135 * them to initialized or split the described range from larger unwritten 2136 * extent. Note that we need not map all the described range since allocation 2137 * can return less blocks or the range is covered by more unwritten extents. We 2138 * cannot map more because we are limited by reserved transaction credits. On 2139 * the other hand we always make sure that the last touched page is fully 2140 * mapped so that it can be written out (and thus forward progress is 2141 * guaranteed). After mapping we submit all mapped pages for IO. 2142 */ 2143 static int mpage_map_and_submit_extent(handle_t *handle, 2144 struct mpage_da_data *mpd, 2145 bool *give_up_on_write) 2146 { 2147 struct inode *inode = mpd->inode; 2148 struct ext4_map_blocks *map = &mpd->map; 2149 int err; 2150 loff_t disksize; 2151 int progress = 0; 2152 2153 mpd->io_submit.io_end->offset = 2154 ((loff_t)map->m_lblk) << inode->i_blkbits; 2155 do { 2156 err = mpage_map_one_extent(handle, mpd); 2157 if (err < 0) { 2158 struct super_block *sb = inode->i_sb; 2159 2160 if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED) 2161 goto invalidate_dirty_pages; 2162 /* 2163 * Let the uper layers retry transient errors. 2164 * In the case of ENOSPC, if ext4_count_free_blocks() 2165 * is non-zero, a commit should free up blocks. 2166 */ 2167 if ((err == -ENOMEM) || 2168 (err == -ENOSPC && ext4_count_free_clusters(sb))) { 2169 if (progress) 2170 goto update_disksize; 2171 return err; 2172 } 2173 ext4_msg(sb, KERN_CRIT, 2174 "Delayed block allocation failed for " 2175 "inode %lu at logical offset %llu with" 2176 " max blocks %u with error %d", 2177 inode->i_ino, 2178 (unsigned long long)map->m_lblk, 2179 (unsigned)map->m_len, -err); 2180 ext4_msg(sb, KERN_CRIT, 2181 "This should not happen!! Data will " 2182 "be lost\n"); 2183 if (err == -ENOSPC) 2184 ext4_print_free_blocks(inode); 2185 invalidate_dirty_pages: 2186 *give_up_on_write = true; 2187 return err; 2188 } 2189 progress = 1; 2190 /* 2191 * Update buffer state, submit mapped pages, and get us new 2192 * extent to map 2193 */ 2194 err = mpage_map_and_submit_buffers(mpd); 2195 if (err < 0) 2196 goto update_disksize; 2197 } while (map->m_len); 2198 2199 update_disksize: 2200 /* 2201 * Update on-disk size after IO is submitted. Races with 2202 * truncate are avoided by checking i_size under i_data_sem. 2203 */ 2204 disksize = ((loff_t)mpd->first_page) << PAGE_CACHE_SHIFT; 2205 if (disksize > EXT4_I(inode)->i_disksize) { 2206 int err2; 2207 loff_t i_size; 2208 2209 down_write(&EXT4_I(inode)->i_data_sem); 2210 i_size = i_size_read(inode); 2211 if (disksize > i_size) 2212 disksize = i_size; 2213 if (disksize > EXT4_I(inode)->i_disksize) 2214 EXT4_I(inode)->i_disksize = disksize; 2215 err2 = ext4_mark_inode_dirty(handle, inode); 2216 up_write(&EXT4_I(inode)->i_data_sem); 2217 if (err2) 2218 ext4_error(inode->i_sb, 2219 "Failed to mark inode %lu dirty", 2220 inode->i_ino); 2221 if (!err) 2222 err = err2; 2223 } 2224 return err; 2225 } 2226 2227 /* 2228 * Calculate the total number of credits to reserve for one writepages 2229 * iteration. This is called from ext4_writepages(). We map an extent of 2230 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping 2231 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN + 2232 * bpp - 1 blocks in bpp different extents. 2233 */ 2234 static int ext4_da_writepages_trans_blocks(struct inode *inode) 2235 { 2236 int bpp = ext4_journal_blocks_per_page(inode); 2237 2238 return ext4_meta_trans_blocks(inode, 2239 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp); 2240 } 2241 2242 /* 2243 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages 2244 * and underlying extent to map 2245 * 2246 * @mpd - where to look for pages 2247 * 2248 * Walk dirty pages in the mapping. If they are fully mapped, submit them for 2249 * IO immediately. When we find a page which isn't mapped we start accumulating 2250 * extent of buffers underlying these pages that needs mapping (formed by 2251 * either delayed or unwritten buffers). We also lock the pages containing 2252 * these buffers. The extent found is returned in @mpd structure (starting at 2253 * mpd->lblk with length mpd->len blocks). 2254 * 2255 * Note that this function can attach bios to one io_end structure which are 2256 * neither logically nor physically contiguous. Although it may seem as an 2257 * unnecessary complication, it is actually inevitable in blocksize < pagesize 2258 * case as we need to track IO to all buffers underlying a page in one io_end. 2259 */ 2260 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd) 2261 { 2262 struct address_space *mapping = mpd->inode->i_mapping; 2263 struct pagevec pvec; 2264 unsigned int nr_pages; 2265 long left = mpd->wbc->nr_to_write; 2266 pgoff_t index = mpd->first_page; 2267 pgoff_t end = mpd->last_page; 2268 int tag; 2269 int i, err = 0; 2270 int blkbits = mpd->inode->i_blkbits; 2271 ext4_lblk_t lblk; 2272 struct buffer_head *head; 2273 2274 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages) 2275 tag = PAGECACHE_TAG_TOWRITE; 2276 else 2277 tag = PAGECACHE_TAG_DIRTY; 2278 2279 pagevec_init(&pvec, 0); 2280 mpd->map.m_len = 0; 2281 mpd->next_page = index; 2282 while (index <= end) { 2283 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, 2284 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); 2285 if (nr_pages == 0) 2286 goto out; 2287 2288 for (i = 0; i < nr_pages; i++) { 2289 struct page *page = pvec.pages[i]; 2290 2291 /* 2292 * At this point, the page may be truncated or 2293 * invalidated (changing page->mapping to NULL), or 2294 * even swizzled back from swapper_space to tmpfs file 2295 * mapping. However, page->index will not change 2296 * because we have a reference on the page. 2297 */ 2298 if (page->index > end) 2299 goto out; 2300 2301 /* 2302 * Accumulated enough dirty pages? This doesn't apply 2303 * to WB_SYNC_ALL mode. For integrity sync we have to 2304 * keep going because someone may be concurrently 2305 * dirtying pages, and we might have synced a lot of 2306 * newly appeared dirty pages, but have not synced all 2307 * of the old dirty pages. 2308 */ 2309 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0) 2310 goto out; 2311 2312 /* If we can't merge this page, we are done. */ 2313 if (mpd->map.m_len > 0 && mpd->next_page != page->index) 2314 goto out; 2315 2316 lock_page(page); 2317 /* 2318 * If the page is no longer dirty, or its mapping no 2319 * longer corresponds to inode we are writing (which 2320 * means it has been truncated or invalidated), or the 2321 * page is already under writeback and we are not doing 2322 * a data integrity writeback, skip the page 2323 */ 2324 if (!PageDirty(page) || 2325 (PageWriteback(page) && 2326 (mpd->wbc->sync_mode == WB_SYNC_NONE)) || 2327 unlikely(page->mapping != mapping)) { 2328 unlock_page(page); 2329 continue; 2330 } 2331 2332 wait_on_page_writeback(page); 2333 BUG_ON(PageWriteback(page)); 2334 2335 if (mpd->map.m_len == 0) 2336 mpd->first_page = page->index; 2337 mpd->next_page = page->index + 1; 2338 /* Add all dirty buffers to mpd */ 2339 lblk = ((ext4_lblk_t)page->index) << 2340 (PAGE_CACHE_SHIFT - blkbits); 2341 head = page_buffers(page); 2342 err = mpage_process_page_bufs(mpd, head, head, lblk); 2343 if (err <= 0) 2344 goto out; 2345 err = 0; 2346 left--; 2347 } 2348 pagevec_release(&pvec); 2349 cond_resched(); 2350 } 2351 return 0; 2352 out: 2353 pagevec_release(&pvec); 2354 return err; 2355 } 2356 2357 static int __writepage(struct page *page, struct writeback_control *wbc, 2358 void *data) 2359 { 2360 struct address_space *mapping = data; 2361 int ret = ext4_writepage(page, wbc); 2362 mapping_set_error(mapping, ret); 2363 return ret; 2364 } 2365 2366 static int ext4_writepages(struct address_space *mapping, 2367 struct writeback_control *wbc) 2368 { 2369 pgoff_t writeback_index = 0; 2370 long nr_to_write = wbc->nr_to_write; 2371 int range_whole = 0; 2372 int cycled = 1; 2373 handle_t *handle = NULL; 2374 struct mpage_da_data mpd; 2375 struct inode *inode = mapping->host; 2376 int needed_blocks, rsv_blocks = 0, ret = 0; 2377 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); 2378 bool done; 2379 struct blk_plug plug; 2380 bool give_up_on_write = false; 2381 2382 trace_ext4_writepages(inode, wbc); 2383 2384 /* 2385 * No pages to write? This is mainly a kludge to avoid starting 2386 * a transaction for special inodes like journal inode on last iput() 2387 * because that could violate lock ordering on umount 2388 */ 2389 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 2390 goto out_writepages; 2391 2392 if (ext4_should_journal_data(inode)) { 2393 struct blk_plug plug; 2394 2395 blk_start_plug(&plug); 2396 ret = write_cache_pages(mapping, wbc, __writepage, mapping); 2397 blk_finish_plug(&plug); 2398 goto out_writepages; 2399 } 2400 2401 /* 2402 * If the filesystem has aborted, it is read-only, so return 2403 * right away instead of dumping stack traces later on that 2404 * will obscure the real source of the problem. We test 2405 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because 2406 * the latter could be true if the filesystem is mounted 2407 * read-only, and in that case, ext4_writepages should 2408 * *never* be called, so if that ever happens, we would want 2409 * the stack trace. 2410 */ 2411 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) { 2412 ret = -EROFS; 2413 goto out_writepages; 2414 } 2415 2416 if (ext4_should_dioread_nolock(inode)) { 2417 /* 2418 * We may need to convert up to one extent per block in 2419 * the page and we may dirty the inode. 2420 */ 2421 rsv_blocks = 1 + (PAGE_CACHE_SIZE >> inode->i_blkbits); 2422 } 2423 2424 /* 2425 * If we have inline data and arrive here, it means that 2426 * we will soon create the block for the 1st page, so 2427 * we'd better clear the inline data here. 2428 */ 2429 if (ext4_has_inline_data(inode)) { 2430 /* Just inode will be modified... */ 2431 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); 2432 if (IS_ERR(handle)) { 2433 ret = PTR_ERR(handle); 2434 goto out_writepages; 2435 } 2436 BUG_ON(ext4_test_inode_state(inode, 2437 EXT4_STATE_MAY_INLINE_DATA)); 2438 ext4_destroy_inline_data(handle, inode); 2439 ext4_journal_stop(handle); 2440 } 2441 2442 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) 2443 range_whole = 1; 2444 2445 if (wbc->range_cyclic) { 2446 writeback_index = mapping->writeback_index; 2447 if (writeback_index) 2448 cycled = 0; 2449 mpd.first_page = writeback_index; 2450 mpd.last_page = -1; 2451 } else { 2452 mpd.first_page = wbc->range_start >> PAGE_CACHE_SHIFT; 2453 mpd.last_page = wbc->range_end >> PAGE_CACHE_SHIFT; 2454 } 2455 2456 mpd.inode = inode; 2457 mpd.wbc = wbc; 2458 ext4_io_submit_init(&mpd.io_submit, wbc); 2459 retry: 2460 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) 2461 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page); 2462 done = false; 2463 blk_start_plug(&plug); 2464 while (!done && mpd.first_page <= mpd.last_page) { 2465 /* For each extent of pages we use new io_end */ 2466 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); 2467 if (!mpd.io_submit.io_end) { 2468 ret = -ENOMEM; 2469 break; 2470 } 2471 2472 /* 2473 * We have two constraints: We find one extent to map and we 2474 * must always write out whole page (makes a difference when 2475 * blocksize < pagesize) so that we don't block on IO when we 2476 * try to write out the rest of the page. Journalled mode is 2477 * not supported by delalloc. 2478 */ 2479 BUG_ON(ext4_should_journal_data(inode)); 2480 needed_blocks = ext4_da_writepages_trans_blocks(inode); 2481 2482 /* start a new transaction */ 2483 handle = ext4_journal_start_with_reserve(inode, 2484 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks); 2485 if (IS_ERR(handle)) { 2486 ret = PTR_ERR(handle); 2487 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " 2488 "%ld pages, ino %lu; err %d", __func__, 2489 wbc->nr_to_write, inode->i_ino, ret); 2490 /* Release allocated io_end */ 2491 ext4_put_io_end(mpd.io_submit.io_end); 2492 break; 2493 } 2494 2495 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc); 2496 ret = mpage_prepare_extent_to_map(&mpd); 2497 if (!ret) { 2498 if (mpd.map.m_len) 2499 ret = mpage_map_and_submit_extent(handle, &mpd, 2500 &give_up_on_write); 2501 else { 2502 /* 2503 * We scanned the whole range (or exhausted 2504 * nr_to_write), submitted what was mapped and 2505 * didn't find anything needing mapping. We are 2506 * done. 2507 */ 2508 done = true; 2509 } 2510 } 2511 ext4_journal_stop(handle); 2512 /* Submit prepared bio */ 2513 ext4_io_submit(&mpd.io_submit); 2514 /* Unlock pages we didn't use */ 2515 mpage_release_unused_pages(&mpd, give_up_on_write); 2516 /* Drop our io_end reference we got from init */ 2517 ext4_put_io_end(mpd.io_submit.io_end); 2518 2519 if (ret == -ENOSPC && sbi->s_journal) { 2520 /* 2521 * Commit the transaction which would 2522 * free blocks released in the transaction 2523 * and try again 2524 */ 2525 jbd2_journal_force_commit_nested(sbi->s_journal); 2526 ret = 0; 2527 continue; 2528 } 2529 /* Fatal error - ENOMEM, EIO... */ 2530 if (ret) 2531 break; 2532 } 2533 blk_finish_plug(&plug); 2534 if (!ret && !cycled && wbc->nr_to_write > 0) { 2535 cycled = 1; 2536 mpd.last_page = writeback_index - 1; 2537 mpd.first_page = 0; 2538 goto retry; 2539 } 2540 2541 /* Update index */ 2542 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) 2543 /* 2544 * Set the writeback_index so that range_cyclic 2545 * mode will write it back later 2546 */ 2547 mapping->writeback_index = mpd.first_page; 2548 2549 out_writepages: 2550 trace_ext4_writepages_result(inode, wbc, ret, 2551 nr_to_write - wbc->nr_to_write); 2552 return ret; 2553 } 2554 2555 static int ext4_nonda_switch(struct super_block *sb) 2556 { 2557 s64 free_clusters, dirty_clusters; 2558 struct ext4_sb_info *sbi = EXT4_SB(sb); 2559 2560 /* 2561 * switch to non delalloc mode if we are running low 2562 * on free block. The free block accounting via percpu 2563 * counters can get slightly wrong with percpu_counter_batch getting 2564 * accumulated on each CPU without updating global counters 2565 * Delalloc need an accurate free block accounting. So switch 2566 * to non delalloc when we are near to error range. 2567 */ 2568 free_clusters = 2569 percpu_counter_read_positive(&sbi->s_freeclusters_counter); 2570 dirty_clusters = 2571 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter); 2572 /* 2573 * Start pushing delalloc when 1/2 of free blocks are dirty. 2574 */ 2575 if (dirty_clusters && (free_clusters < 2 * dirty_clusters)) 2576 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE); 2577 2578 if (2 * free_clusters < 3 * dirty_clusters || 2579 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) { 2580 /* 2581 * free block count is less than 150% of dirty blocks 2582 * or free blocks is less than watermark 2583 */ 2584 return 1; 2585 } 2586 return 0; 2587 } 2588 2589 /* We always reserve for an inode update; the superblock could be there too */ 2590 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len) 2591 { 2592 if (likely(EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, 2593 EXT4_FEATURE_RO_COMPAT_LARGE_FILE))) 2594 return 1; 2595 2596 if (pos + len <= 0x7fffffffULL) 2597 return 1; 2598 2599 /* We might need to update the superblock to set LARGE_FILE */ 2600 return 2; 2601 } 2602 2603 static int ext4_da_write_begin(struct file *file, struct address_space *mapping, 2604 loff_t pos, unsigned len, unsigned flags, 2605 struct page **pagep, void **fsdata) 2606 { 2607 int ret, retries = 0; 2608 struct page *page; 2609 pgoff_t index; 2610 struct inode *inode = mapping->host; 2611 handle_t *handle; 2612 2613 index = pos >> PAGE_CACHE_SHIFT; 2614 2615 if (ext4_nonda_switch(inode->i_sb)) { 2616 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; 2617 return ext4_write_begin(file, mapping, pos, 2618 len, flags, pagep, fsdata); 2619 } 2620 *fsdata = (void *)0; 2621 trace_ext4_da_write_begin(inode, pos, len, flags); 2622 2623 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { 2624 ret = ext4_da_write_inline_data_begin(mapping, inode, 2625 pos, len, flags, 2626 pagep, fsdata); 2627 if (ret < 0) 2628 return ret; 2629 if (ret == 1) 2630 return 0; 2631 } 2632 2633 /* 2634 * grab_cache_page_write_begin() can take a long time if the 2635 * system is thrashing due to memory pressure, or if the page 2636 * is being written back. So grab it first before we start 2637 * the transaction handle. This also allows us to allocate 2638 * the page (if needed) without using GFP_NOFS. 2639 */ 2640 retry_grab: 2641 page = grab_cache_page_write_begin(mapping, index, flags); 2642 if (!page) 2643 return -ENOMEM; 2644 unlock_page(page); 2645 2646 /* 2647 * With delayed allocation, we don't log the i_disksize update 2648 * if there is delayed block allocation. But we still need 2649 * to journalling the i_disksize update if writes to the end 2650 * of file which has an already mapped buffer. 2651 */ 2652 retry_journal: 2653 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, 2654 ext4_da_write_credits(inode, pos, len)); 2655 if (IS_ERR(handle)) { 2656 page_cache_release(page); 2657 return PTR_ERR(handle); 2658 } 2659 2660 lock_page(page); 2661 if (page->mapping != mapping) { 2662 /* The page got truncated from under us */ 2663 unlock_page(page); 2664 page_cache_release(page); 2665 ext4_journal_stop(handle); 2666 goto retry_grab; 2667 } 2668 /* In case writeback began while the page was unlocked */ 2669 wait_for_stable_page(page); 2670 2671 #ifdef CONFIG_EXT4_FS_ENCRYPTION 2672 ret = ext4_block_write_begin(page, pos, len, 2673 ext4_da_get_block_prep); 2674 #else 2675 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep); 2676 #endif 2677 if (ret < 0) { 2678 unlock_page(page); 2679 ext4_journal_stop(handle); 2680 /* 2681 * block_write_begin may have instantiated a few blocks 2682 * outside i_size. Trim these off again. Don't need 2683 * i_size_read because we hold i_mutex. 2684 */ 2685 if (pos + len > inode->i_size) 2686 ext4_truncate_failed_write(inode); 2687 2688 if (ret == -ENOSPC && 2689 ext4_should_retry_alloc(inode->i_sb, &retries)) 2690 goto retry_journal; 2691 2692 page_cache_release(page); 2693 return ret; 2694 } 2695 2696 *pagep = page; 2697 return ret; 2698 } 2699 2700 /* 2701 * Check if we should update i_disksize 2702 * when write to the end of file but not require block allocation 2703 */ 2704 static int ext4_da_should_update_i_disksize(struct page *page, 2705 unsigned long offset) 2706 { 2707 struct buffer_head *bh; 2708 struct inode *inode = page->mapping->host; 2709 unsigned int idx; 2710 int i; 2711 2712 bh = page_buffers(page); 2713 idx = offset >> inode->i_blkbits; 2714 2715 for (i = 0; i < idx; i++) 2716 bh = bh->b_this_page; 2717 2718 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) 2719 return 0; 2720 return 1; 2721 } 2722 2723 static int ext4_da_write_end(struct file *file, 2724 struct address_space *mapping, 2725 loff_t pos, unsigned len, unsigned copied, 2726 struct page *page, void *fsdata) 2727 { 2728 struct inode *inode = mapping->host; 2729 int ret = 0, ret2; 2730 handle_t *handle = ext4_journal_current_handle(); 2731 loff_t new_i_size; 2732 unsigned long start, end; 2733 int write_mode = (int)(unsigned long)fsdata; 2734 2735 if (write_mode == FALL_BACK_TO_NONDELALLOC) 2736 return ext4_write_end(file, mapping, pos, 2737 len, copied, page, fsdata); 2738 2739 trace_ext4_da_write_end(inode, pos, len, copied); 2740 start = pos & (PAGE_CACHE_SIZE - 1); 2741 end = start + copied - 1; 2742 2743 /* 2744 * generic_write_end() will run mark_inode_dirty() if i_size 2745 * changes. So let's piggyback the i_disksize mark_inode_dirty 2746 * into that. 2747 */ 2748 new_i_size = pos + copied; 2749 if (copied && new_i_size > EXT4_I(inode)->i_disksize) { 2750 if (ext4_has_inline_data(inode) || 2751 ext4_da_should_update_i_disksize(page, end)) { 2752 ext4_update_i_disksize(inode, new_i_size); 2753 /* We need to mark inode dirty even if 2754 * new_i_size is less that inode->i_size 2755 * bu greater than i_disksize.(hint delalloc) 2756 */ 2757 ext4_mark_inode_dirty(handle, inode); 2758 } 2759 } 2760 2761 if (write_mode != CONVERT_INLINE_DATA && 2762 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) && 2763 ext4_has_inline_data(inode)) 2764 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied, 2765 page); 2766 else 2767 ret2 = generic_write_end(file, mapping, pos, len, copied, 2768 page, fsdata); 2769 2770 copied = ret2; 2771 if (ret2 < 0) 2772 ret = ret2; 2773 ret2 = ext4_journal_stop(handle); 2774 if (!ret) 2775 ret = ret2; 2776 2777 return ret ? ret : copied; 2778 } 2779 2780 static void ext4_da_invalidatepage(struct page *page, unsigned int offset, 2781 unsigned int length) 2782 { 2783 /* 2784 * Drop reserved blocks 2785 */ 2786 BUG_ON(!PageLocked(page)); 2787 if (!page_has_buffers(page)) 2788 goto out; 2789 2790 ext4_da_page_release_reservation(page, offset, length); 2791 2792 out: 2793 ext4_invalidatepage(page, offset, length); 2794 2795 return; 2796 } 2797 2798 /* 2799 * Force all delayed allocation blocks to be allocated for a given inode. 2800 */ 2801 int ext4_alloc_da_blocks(struct inode *inode) 2802 { 2803 trace_ext4_alloc_da_blocks(inode); 2804 2805 if (!EXT4_I(inode)->i_reserved_data_blocks) 2806 return 0; 2807 2808 /* 2809 * We do something simple for now. The filemap_flush() will 2810 * also start triggering a write of the data blocks, which is 2811 * not strictly speaking necessary (and for users of 2812 * laptop_mode, not even desirable). However, to do otherwise 2813 * would require replicating code paths in: 2814 * 2815 * ext4_writepages() -> 2816 * write_cache_pages() ---> (via passed in callback function) 2817 * __mpage_da_writepage() --> 2818 * mpage_add_bh_to_extent() 2819 * mpage_da_map_blocks() 2820 * 2821 * The problem is that write_cache_pages(), located in 2822 * mm/page-writeback.c, marks pages clean in preparation for 2823 * doing I/O, which is not desirable if we're not planning on 2824 * doing I/O at all. 2825 * 2826 * We could call write_cache_pages(), and then redirty all of 2827 * the pages by calling redirty_page_for_writepage() but that 2828 * would be ugly in the extreme. So instead we would need to 2829 * replicate parts of the code in the above functions, 2830 * simplifying them because we wouldn't actually intend to 2831 * write out the pages, but rather only collect contiguous 2832 * logical block extents, call the multi-block allocator, and 2833 * then update the buffer heads with the block allocations. 2834 * 2835 * For now, though, we'll cheat by calling filemap_flush(), 2836 * which will map the blocks, and start the I/O, but not 2837 * actually wait for the I/O to complete. 2838 */ 2839 return filemap_flush(inode->i_mapping); 2840 } 2841 2842 /* 2843 * bmap() is special. It gets used by applications such as lilo and by 2844 * the swapper to find the on-disk block of a specific piece of data. 2845 * 2846 * Naturally, this is dangerous if the block concerned is still in the 2847 * journal. If somebody makes a swapfile on an ext4 data-journaling 2848 * filesystem and enables swap, then they may get a nasty shock when the 2849 * data getting swapped to that swapfile suddenly gets overwritten by 2850 * the original zero's written out previously to the journal and 2851 * awaiting writeback in the kernel's buffer cache. 2852 * 2853 * So, if we see any bmap calls here on a modified, data-journaled file, 2854 * take extra steps to flush any blocks which might be in the cache. 2855 */ 2856 static sector_t ext4_bmap(struct address_space *mapping, sector_t block) 2857 { 2858 struct inode *inode = mapping->host; 2859 journal_t *journal; 2860 int err; 2861 2862 /* 2863 * We can get here for an inline file via the FIBMAP ioctl 2864 */ 2865 if (ext4_has_inline_data(inode)) 2866 return 0; 2867 2868 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && 2869 test_opt(inode->i_sb, DELALLOC)) { 2870 /* 2871 * With delalloc we want to sync the file 2872 * so that we can make sure we allocate 2873 * blocks for file 2874 */ 2875 filemap_write_and_wait(mapping); 2876 } 2877 2878 if (EXT4_JOURNAL(inode) && 2879 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) { 2880 /* 2881 * This is a REALLY heavyweight approach, but the use of 2882 * bmap on dirty files is expected to be extremely rare: 2883 * only if we run lilo or swapon on a freshly made file 2884 * do we expect this to happen. 2885 * 2886 * (bmap requires CAP_SYS_RAWIO so this does not 2887 * represent an unprivileged user DOS attack --- we'd be 2888 * in trouble if mortal users could trigger this path at 2889 * will.) 2890 * 2891 * NB. EXT4_STATE_JDATA is not set on files other than 2892 * regular files. If somebody wants to bmap a directory 2893 * or symlink and gets confused because the buffer 2894 * hasn't yet been flushed to disk, they deserve 2895 * everything they get. 2896 */ 2897 2898 ext4_clear_inode_state(inode, EXT4_STATE_JDATA); 2899 journal = EXT4_JOURNAL(inode); 2900 jbd2_journal_lock_updates(journal); 2901 err = jbd2_journal_flush(journal); 2902 jbd2_journal_unlock_updates(journal); 2903 2904 if (err) 2905 return 0; 2906 } 2907 2908 return generic_block_bmap(mapping, block, ext4_get_block); 2909 } 2910 2911 static int ext4_readpage(struct file *file, struct page *page) 2912 { 2913 int ret = -EAGAIN; 2914 struct inode *inode = page->mapping->host; 2915 2916 trace_ext4_readpage(page); 2917 2918 if (ext4_has_inline_data(inode)) 2919 ret = ext4_readpage_inline(inode, page); 2920 2921 if (ret == -EAGAIN) 2922 return ext4_mpage_readpages(page->mapping, NULL, page, 1); 2923 2924 return ret; 2925 } 2926 2927 static int 2928 ext4_readpages(struct file *file, struct address_space *mapping, 2929 struct list_head *pages, unsigned nr_pages) 2930 { 2931 struct inode *inode = mapping->host; 2932 2933 /* If the file has inline data, no need to do readpages. */ 2934 if (ext4_has_inline_data(inode)) 2935 return 0; 2936 2937 return ext4_mpage_readpages(mapping, pages, NULL, nr_pages); 2938 } 2939 2940 static void ext4_invalidatepage(struct page *page, unsigned int offset, 2941 unsigned int length) 2942 { 2943 trace_ext4_invalidatepage(page, offset, length); 2944 2945 /* No journalling happens on data buffers when this function is used */ 2946 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page))); 2947 2948 block_invalidatepage(page, offset, length); 2949 } 2950 2951 static int __ext4_journalled_invalidatepage(struct page *page, 2952 unsigned int offset, 2953 unsigned int length) 2954 { 2955 journal_t *journal = EXT4_JOURNAL(page->mapping->host); 2956 2957 trace_ext4_journalled_invalidatepage(page, offset, length); 2958 2959 /* 2960 * If it's a full truncate we just forget about the pending dirtying 2961 */ 2962 if (offset == 0 && length == PAGE_CACHE_SIZE) 2963 ClearPageChecked(page); 2964 2965 return jbd2_journal_invalidatepage(journal, page, offset, length); 2966 } 2967 2968 /* Wrapper for aops... */ 2969 static void ext4_journalled_invalidatepage(struct page *page, 2970 unsigned int offset, 2971 unsigned int length) 2972 { 2973 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0); 2974 } 2975 2976 static int ext4_releasepage(struct page *page, gfp_t wait) 2977 { 2978 journal_t *journal = EXT4_JOURNAL(page->mapping->host); 2979 2980 trace_ext4_releasepage(page); 2981 2982 /* Page has dirty journalled data -> cannot release */ 2983 if (PageChecked(page)) 2984 return 0; 2985 if (journal) 2986 return jbd2_journal_try_to_free_buffers(journal, page, wait); 2987 else 2988 return try_to_free_buffers(page); 2989 } 2990 2991 /* 2992 * ext4_get_block used when preparing for a DIO write or buffer write. 2993 * We allocate an uinitialized extent if blocks haven't been allocated. 2994 * The extent will be converted to initialized after the IO is complete. 2995 */ 2996 int ext4_get_block_write(struct inode *inode, sector_t iblock, 2997 struct buffer_head *bh_result, int create) 2998 { 2999 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n", 3000 inode->i_ino, create); 3001 return _ext4_get_block(inode, iblock, bh_result, 3002 EXT4_GET_BLOCKS_IO_CREATE_EXT); 3003 } 3004 3005 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock, 3006 struct buffer_head *bh_result, int create) 3007 { 3008 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n", 3009 inode->i_ino, create); 3010 return _ext4_get_block(inode, iblock, bh_result, 3011 EXT4_GET_BLOCKS_NO_LOCK); 3012 } 3013 3014 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset, 3015 ssize_t size, void *private) 3016 { 3017 ext4_io_end_t *io_end = iocb->private; 3018 3019 /* if not async direct IO just return */ 3020 if (!io_end) 3021 return; 3022 3023 ext_debug("ext4_end_io_dio(): io_end 0x%p " 3024 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n", 3025 iocb->private, io_end->inode->i_ino, iocb, offset, 3026 size); 3027 3028 iocb->private = NULL; 3029 io_end->offset = offset; 3030 io_end->size = size; 3031 ext4_put_io_end(io_end); 3032 } 3033 3034 /* 3035 * For ext4 extent files, ext4 will do direct-io write to holes, 3036 * preallocated extents, and those write extend the file, no need to 3037 * fall back to buffered IO. 3038 * 3039 * For holes, we fallocate those blocks, mark them as unwritten 3040 * If those blocks were preallocated, we mark sure they are split, but 3041 * still keep the range to write as unwritten. 3042 * 3043 * The unwritten extents will be converted to written when DIO is completed. 3044 * For async direct IO, since the IO may still pending when return, we 3045 * set up an end_io call back function, which will do the conversion 3046 * when async direct IO completed. 3047 * 3048 * If the O_DIRECT write will extend the file then add this inode to the 3049 * orphan list. So recovery will truncate it back to the original size 3050 * if the machine crashes during the write. 3051 * 3052 */ 3053 static ssize_t ext4_ext_direct_IO(struct kiocb *iocb, struct iov_iter *iter, 3054 loff_t offset) 3055 { 3056 struct file *file = iocb->ki_filp; 3057 struct inode *inode = file->f_mapping->host; 3058 ssize_t ret; 3059 size_t count = iov_iter_count(iter); 3060 int overwrite = 0; 3061 get_block_t *get_block_func = NULL; 3062 int dio_flags = 0; 3063 loff_t final_size = offset + count; 3064 ext4_io_end_t *io_end = NULL; 3065 3066 /* Use the old path for reads and writes beyond i_size. */ 3067 if (iov_iter_rw(iter) != WRITE || final_size > inode->i_size) 3068 return ext4_ind_direct_IO(iocb, iter, offset); 3069 3070 BUG_ON(iocb->private == NULL); 3071 3072 /* 3073 * Make all waiters for direct IO properly wait also for extent 3074 * conversion. This also disallows race between truncate() and 3075 * overwrite DIO as i_dio_count needs to be incremented under i_mutex. 3076 */ 3077 if (iov_iter_rw(iter) == WRITE) 3078 inode_dio_begin(inode); 3079 3080 /* If we do a overwrite dio, i_mutex locking can be released */ 3081 overwrite = *((int *)iocb->private); 3082 3083 if (overwrite) { 3084 down_read(&EXT4_I(inode)->i_data_sem); 3085 mutex_unlock(&inode->i_mutex); 3086 } 3087 3088 /* 3089 * We could direct write to holes and fallocate. 3090 * 3091 * Allocated blocks to fill the hole are marked as 3092 * unwritten to prevent parallel buffered read to expose 3093 * the stale data before DIO complete the data IO. 3094 * 3095 * As to previously fallocated extents, ext4 get_block will 3096 * just simply mark the buffer mapped but still keep the 3097 * extents unwritten. 3098 * 3099 * For non AIO case, we will convert those unwritten extents 3100 * to written after return back from blockdev_direct_IO. 3101 * 3102 * For async DIO, the conversion needs to be deferred when the 3103 * IO is completed. The ext4 end_io callback function will be 3104 * called to take care of the conversion work. Here for async 3105 * case, we allocate an io_end structure to hook to the iocb. 3106 */ 3107 iocb->private = NULL; 3108 ext4_inode_aio_set(inode, NULL); 3109 if (!is_sync_kiocb(iocb)) { 3110 io_end = ext4_init_io_end(inode, GFP_NOFS); 3111 if (!io_end) { 3112 ret = -ENOMEM; 3113 goto retake_lock; 3114 } 3115 /* 3116 * Grab reference for DIO. Will be dropped in ext4_end_io_dio() 3117 */ 3118 iocb->private = ext4_get_io_end(io_end); 3119 /* 3120 * we save the io structure for current async direct 3121 * IO, so that later ext4_map_blocks() could flag the 3122 * io structure whether there is a unwritten extents 3123 * needs to be converted when IO is completed. 3124 */ 3125 ext4_inode_aio_set(inode, io_end); 3126 } 3127 3128 if (overwrite) { 3129 get_block_func = ext4_get_block_write_nolock; 3130 } else { 3131 get_block_func = ext4_get_block_write; 3132 dio_flags = DIO_LOCKING; 3133 } 3134 #ifdef CONFIG_EXT4_FS_ENCRYPTION 3135 BUG_ON(ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode)); 3136 #endif 3137 if (IS_DAX(inode)) 3138 ret = dax_do_io(iocb, inode, iter, offset, get_block_func, 3139 ext4_end_io_dio, dio_flags); 3140 else 3141 ret = __blockdev_direct_IO(iocb, inode, 3142 inode->i_sb->s_bdev, iter, offset, 3143 get_block_func, 3144 ext4_end_io_dio, NULL, dio_flags); 3145 3146 /* 3147 * Put our reference to io_end. This can free the io_end structure e.g. 3148 * in sync IO case or in case of error. It can even perform extent 3149 * conversion if all bios we submitted finished before we got here. 3150 * Note that in that case iocb->private can be already set to NULL 3151 * here. 3152 */ 3153 if (io_end) { 3154 ext4_inode_aio_set(inode, NULL); 3155 ext4_put_io_end(io_end); 3156 /* 3157 * When no IO was submitted ext4_end_io_dio() was not 3158 * called so we have to put iocb's reference. 3159 */ 3160 if (ret <= 0 && ret != -EIOCBQUEUED && iocb->private) { 3161 WARN_ON(iocb->private != io_end); 3162 WARN_ON(io_end->flag & EXT4_IO_END_UNWRITTEN); 3163 ext4_put_io_end(io_end); 3164 iocb->private = NULL; 3165 } 3166 } 3167 if (ret > 0 && !overwrite && ext4_test_inode_state(inode, 3168 EXT4_STATE_DIO_UNWRITTEN)) { 3169 int err; 3170 /* 3171 * for non AIO case, since the IO is already 3172 * completed, we could do the conversion right here 3173 */ 3174 err = ext4_convert_unwritten_extents(NULL, inode, 3175 offset, ret); 3176 if (err < 0) 3177 ret = err; 3178 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); 3179 } 3180 3181 retake_lock: 3182 if (iov_iter_rw(iter) == WRITE) 3183 inode_dio_end(inode); 3184 /* take i_mutex locking again if we do a ovewrite dio */ 3185 if (overwrite) { 3186 up_read(&EXT4_I(inode)->i_data_sem); 3187 mutex_lock(&inode->i_mutex); 3188 } 3189 3190 return ret; 3191 } 3192 3193 static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter, 3194 loff_t offset) 3195 { 3196 struct file *file = iocb->ki_filp; 3197 struct inode *inode = file->f_mapping->host; 3198 size_t count = iov_iter_count(iter); 3199 ssize_t ret; 3200 3201 #ifdef CONFIG_EXT4_FS_ENCRYPTION 3202 if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode)) 3203 return 0; 3204 #endif 3205 3206 /* 3207 * If we are doing data journalling we don't support O_DIRECT 3208 */ 3209 if (ext4_should_journal_data(inode)) 3210 return 0; 3211 3212 /* Let buffer I/O handle the inline data case. */ 3213 if (ext4_has_inline_data(inode)) 3214 return 0; 3215 3216 trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter)); 3217 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) 3218 ret = ext4_ext_direct_IO(iocb, iter, offset); 3219 else 3220 ret = ext4_ind_direct_IO(iocb, iter, offset); 3221 trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret); 3222 return ret; 3223 } 3224 3225 /* 3226 * Pages can be marked dirty completely asynchronously from ext4's journalling 3227 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do 3228 * much here because ->set_page_dirty is called under VFS locks. The page is 3229 * not necessarily locked. 3230 * 3231 * We cannot just dirty the page and leave attached buffers clean, because the 3232 * buffers' dirty state is "definitive". We cannot just set the buffers dirty 3233 * or jbddirty because all the journalling code will explode. 3234 * 3235 * So what we do is to mark the page "pending dirty" and next time writepage 3236 * is called, propagate that into the buffers appropriately. 3237 */ 3238 static int ext4_journalled_set_page_dirty(struct page *page) 3239 { 3240 SetPageChecked(page); 3241 return __set_page_dirty_nobuffers(page); 3242 } 3243 3244 static const struct address_space_operations ext4_aops = { 3245 .readpage = ext4_readpage, 3246 .readpages = ext4_readpages, 3247 .writepage = ext4_writepage, 3248 .writepages = ext4_writepages, 3249 .write_begin = ext4_write_begin, 3250 .write_end = ext4_write_end, 3251 .bmap = ext4_bmap, 3252 .invalidatepage = ext4_invalidatepage, 3253 .releasepage = ext4_releasepage, 3254 .direct_IO = ext4_direct_IO, 3255 .migratepage = buffer_migrate_page, 3256 .is_partially_uptodate = block_is_partially_uptodate, 3257 .error_remove_page = generic_error_remove_page, 3258 }; 3259 3260 static const struct address_space_operations ext4_journalled_aops = { 3261 .readpage = ext4_readpage, 3262 .readpages = ext4_readpages, 3263 .writepage = ext4_writepage, 3264 .writepages = ext4_writepages, 3265 .write_begin = ext4_write_begin, 3266 .write_end = ext4_journalled_write_end, 3267 .set_page_dirty = ext4_journalled_set_page_dirty, 3268 .bmap = ext4_bmap, 3269 .invalidatepage = ext4_journalled_invalidatepage, 3270 .releasepage = ext4_releasepage, 3271 .direct_IO = ext4_direct_IO, 3272 .is_partially_uptodate = block_is_partially_uptodate, 3273 .error_remove_page = generic_error_remove_page, 3274 }; 3275 3276 static const struct address_space_operations ext4_da_aops = { 3277 .readpage = ext4_readpage, 3278 .readpages = ext4_readpages, 3279 .writepage = ext4_writepage, 3280 .writepages = ext4_writepages, 3281 .write_begin = ext4_da_write_begin, 3282 .write_end = ext4_da_write_end, 3283 .bmap = ext4_bmap, 3284 .invalidatepage = ext4_da_invalidatepage, 3285 .releasepage = ext4_releasepage, 3286 .direct_IO = ext4_direct_IO, 3287 .migratepage = buffer_migrate_page, 3288 .is_partially_uptodate = block_is_partially_uptodate, 3289 .error_remove_page = generic_error_remove_page, 3290 }; 3291 3292 void ext4_set_aops(struct inode *inode) 3293 { 3294 switch (ext4_inode_journal_mode(inode)) { 3295 case EXT4_INODE_ORDERED_DATA_MODE: 3296 ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE); 3297 break; 3298 case EXT4_INODE_WRITEBACK_DATA_MODE: 3299 ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE); 3300 break; 3301 case EXT4_INODE_JOURNAL_DATA_MODE: 3302 inode->i_mapping->a_ops = &ext4_journalled_aops; 3303 return; 3304 default: 3305 BUG(); 3306 } 3307 if (test_opt(inode->i_sb, DELALLOC)) 3308 inode->i_mapping->a_ops = &ext4_da_aops; 3309 else 3310 inode->i_mapping->a_ops = &ext4_aops; 3311 } 3312 3313 static int __ext4_block_zero_page_range(handle_t *handle, 3314 struct address_space *mapping, loff_t from, loff_t length) 3315 { 3316 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT; 3317 unsigned offset = from & (PAGE_CACHE_SIZE-1); 3318 unsigned blocksize, pos; 3319 ext4_lblk_t iblock; 3320 struct inode *inode = mapping->host; 3321 struct buffer_head *bh; 3322 struct page *page; 3323 int err = 0; 3324 3325 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT, 3326 mapping_gfp_mask(mapping) & ~__GFP_FS); 3327 if (!page) 3328 return -ENOMEM; 3329 3330 blocksize = inode->i_sb->s_blocksize; 3331 3332 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); 3333 3334 if (!page_has_buffers(page)) 3335 create_empty_buffers(page, blocksize, 0); 3336 3337 /* Find the buffer that contains "offset" */ 3338 bh = page_buffers(page); 3339 pos = blocksize; 3340 while (offset >= pos) { 3341 bh = bh->b_this_page; 3342 iblock++; 3343 pos += blocksize; 3344 } 3345 if (buffer_freed(bh)) { 3346 BUFFER_TRACE(bh, "freed: skip"); 3347 goto unlock; 3348 } 3349 if (!buffer_mapped(bh)) { 3350 BUFFER_TRACE(bh, "unmapped"); 3351 ext4_get_block(inode, iblock, bh, 0); 3352 /* unmapped? It's a hole - nothing to do */ 3353 if (!buffer_mapped(bh)) { 3354 BUFFER_TRACE(bh, "still unmapped"); 3355 goto unlock; 3356 } 3357 } 3358 3359 /* Ok, it's mapped. Make sure it's up-to-date */ 3360 if (PageUptodate(page)) 3361 set_buffer_uptodate(bh); 3362 3363 if (!buffer_uptodate(bh)) { 3364 err = -EIO; 3365 ll_rw_block(READ, 1, &bh); 3366 wait_on_buffer(bh); 3367 /* Uhhuh. Read error. Complain and punt. */ 3368 if (!buffer_uptodate(bh)) 3369 goto unlock; 3370 if (S_ISREG(inode->i_mode) && 3371 ext4_encrypted_inode(inode)) { 3372 /* We expect the key to be set. */ 3373 BUG_ON(!ext4_has_encryption_key(inode)); 3374 BUG_ON(blocksize != PAGE_CACHE_SIZE); 3375 WARN_ON_ONCE(ext4_decrypt_one(inode, page)); 3376 } 3377 } 3378 if (ext4_should_journal_data(inode)) { 3379 BUFFER_TRACE(bh, "get write access"); 3380 err = ext4_journal_get_write_access(handle, bh); 3381 if (err) 3382 goto unlock; 3383 } 3384 zero_user(page, offset, length); 3385 BUFFER_TRACE(bh, "zeroed end of block"); 3386 3387 if (ext4_should_journal_data(inode)) { 3388 err = ext4_handle_dirty_metadata(handle, inode, bh); 3389 } else { 3390 err = 0; 3391 mark_buffer_dirty(bh); 3392 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) 3393 err = ext4_jbd2_file_inode(handle, inode); 3394 } 3395 3396 unlock: 3397 unlock_page(page); 3398 page_cache_release(page); 3399 return err; 3400 } 3401 3402 /* 3403 * ext4_block_zero_page_range() zeros out a mapping of length 'length' 3404 * starting from file offset 'from'. The range to be zero'd must 3405 * be contained with in one block. If the specified range exceeds 3406 * the end of the block it will be shortened to end of the block 3407 * that cooresponds to 'from' 3408 */ 3409 static int ext4_block_zero_page_range(handle_t *handle, 3410 struct address_space *mapping, loff_t from, loff_t length) 3411 { 3412 struct inode *inode = mapping->host; 3413 unsigned offset = from & (PAGE_CACHE_SIZE-1); 3414 unsigned blocksize = inode->i_sb->s_blocksize; 3415 unsigned max = blocksize - (offset & (blocksize - 1)); 3416 3417 /* 3418 * correct length if it does not fall between 3419 * 'from' and the end of the block 3420 */ 3421 if (length > max || length < 0) 3422 length = max; 3423 3424 if (IS_DAX(inode)) 3425 return dax_zero_page_range(inode, from, length, ext4_get_block); 3426 return __ext4_block_zero_page_range(handle, mapping, from, length); 3427 } 3428 3429 /* 3430 * ext4_block_truncate_page() zeroes out a mapping from file offset `from' 3431 * up to the end of the block which corresponds to `from'. 3432 * This required during truncate. We need to physically zero the tail end 3433 * of that block so it doesn't yield old data if the file is later grown. 3434 */ 3435 static int ext4_block_truncate_page(handle_t *handle, 3436 struct address_space *mapping, loff_t from) 3437 { 3438 unsigned offset = from & (PAGE_CACHE_SIZE-1); 3439 unsigned length; 3440 unsigned blocksize; 3441 struct inode *inode = mapping->host; 3442 3443 blocksize = inode->i_sb->s_blocksize; 3444 length = blocksize - (offset & (blocksize - 1)); 3445 3446 return ext4_block_zero_page_range(handle, mapping, from, length); 3447 } 3448 3449 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode, 3450 loff_t lstart, loff_t length) 3451 { 3452 struct super_block *sb = inode->i_sb; 3453 struct address_space *mapping = inode->i_mapping; 3454 unsigned partial_start, partial_end; 3455 ext4_fsblk_t start, end; 3456 loff_t byte_end = (lstart + length - 1); 3457 int err = 0; 3458 3459 partial_start = lstart & (sb->s_blocksize - 1); 3460 partial_end = byte_end & (sb->s_blocksize - 1); 3461 3462 start = lstart >> sb->s_blocksize_bits; 3463 end = byte_end >> sb->s_blocksize_bits; 3464 3465 /* Handle partial zero within the single block */ 3466 if (start == end && 3467 (partial_start || (partial_end != sb->s_blocksize - 1))) { 3468 err = ext4_block_zero_page_range(handle, mapping, 3469 lstart, length); 3470 return err; 3471 } 3472 /* Handle partial zero out on the start of the range */ 3473 if (partial_start) { 3474 err = ext4_block_zero_page_range(handle, mapping, 3475 lstart, sb->s_blocksize); 3476 if (err) 3477 return err; 3478 } 3479 /* Handle partial zero out on the end of the range */ 3480 if (partial_end != sb->s_blocksize - 1) 3481 err = ext4_block_zero_page_range(handle, mapping, 3482 byte_end - partial_end, 3483 partial_end + 1); 3484 return err; 3485 } 3486 3487 int ext4_can_truncate(struct inode *inode) 3488 { 3489 if (S_ISREG(inode->i_mode)) 3490 return 1; 3491 if (S_ISDIR(inode->i_mode)) 3492 return 1; 3493 if (S_ISLNK(inode->i_mode)) 3494 return !ext4_inode_is_fast_symlink(inode); 3495 return 0; 3496 } 3497 3498 /* 3499 * ext4_punch_hole: punches a hole in a file by releaseing the blocks 3500 * associated with the given offset and length 3501 * 3502 * @inode: File inode 3503 * @offset: The offset where the hole will begin 3504 * @len: The length of the hole 3505 * 3506 * Returns: 0 on success or negative on failure 3507 */ 3508 3509 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length) 3510 { 3511 struct super_block *sb = inode->i_sb; 3512 ext4_lblk_t first_block, stop_block; 3513 struct address_space *mapping = inode->i_mapping; 3514 loff_t first_block_offset, last_block_offset; 3515 handle_t *handle; 3516 unsigned int credits; 3517 int ret = 0; 3518 3519 if (!S_ISREG(inode->i_mode)) 3520 return -EOPNOTSUPP; 3521 3522 trace_ext4_punch_hole(inode, offset, length, 0); 3523 3524 /* 3525 * Write out all dirty pages to avoid race conditions 3526 * Then release them. 3527 */ 3528 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { 3529 ret = filemap_write_and_wait_range(mapping, offset, 3530 offset + length - 1); 3531 if (ret) 3532 return ret; 3533 } 3534 3535 mutex_lock(&inode->i_mutex); 3536 3537 /* No need to punch hole beyond i_size */ 3538 if (offset >= inode->i_size) 3539 goto out_mutex; 3540 3541 /* 3542 * If the hole extends beyond i_size, set the hole 3543 * to end after the page that contains i_size 3544 */ 3545 if (offset + length > inode->i_size) { 3546 length = inode->i_size + 3547 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) - 3548 offset; 3549 } 3550 3551 if (offset & (sb->s_blocksize - 1) || 3552 (offset + length) & (sb->s_blocksize - 1)) { 3553 /* 3554 * Attach jinode to inode for jbd2 if we do any zeroing of 3555 * partial block 3556 */ 3557 ret = ext4_inode_attach_jinode(inode); 3558 if (ret < 0) 3559 goto out_mutex; 3560 3561 } 3562 3563 first_block_offset = round_up(offset, sb->s_blocksize); 3564 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1; 3565 3566 /* Now release the pages and zero block aligned part of pages*/ 3567 if (last_block_offset > first_block_offset) 3568 truncate_pagecache_range(inode, first_block_offset, 3569 last_block_offset); 3570 3571 /* Wait all existing dio workers, newcomers will block on i_mutex */ 3572 ext4_inode_block_unlocked_dio(inode); 3573 inode_dio_wait(inode); 3574 3575 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) 3576 credits = ext4_writepage_trans_blocks(inode); 3577 else 3578 credits = ext4_blocks_for_truncate(inode); 3579 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); 3580 if (IS_ERR(handle)) { 3581 ret = PTR_ERR(handle); 3582 ext4_std_error(sb, ret); 3583 goto out_dio; 3584 } 3585 3586 ret = ext4_zero_partial_blocks(handle, inode, offset, 3587 length); 3588 if (ret) 3589 goto out_stop; 3590 3591 first_block = (offset + sb->s_blocksize - 1) >> 3592 EXT4_BLOCK_SIZE_BITS(sb); 3593 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb); 3594 3595 /* If there are no blocks to remove, return now */ 3596 if (first_block >= stop_block) 3597 goto out_stop; 3598 3599 down_write(&EXT4_I(inode)->i_data_sem); 3600 ext4_discard_preallocations(inode); 3601 3602 ret = ext4_es_remove_extent(inode, first_block, 3603 stop_block - first_block); 3604 if (ret) { 3605 up_write(&EXT4_I(inode)->i_data_sem); 3606 goto out_stop; 3607 } 3608 3609 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) 3610 ret = ext4_ext_remove_space(inode, first_block, 3611 stop_block - 1); 3612 else 3613 ret = ext4_ind_remove_space(handle, inode, first_block, 3614 stop_block); 3615 3616 up_write(&EXT4_I(inode)->i_data_sem); 3617 if (IS_SYNC(inode)) 3618 ext4_handle_sync(handle); 3619 3620 /* Now release the pages again to reduce race window */ 3621 if (last_block_offset > first_block_offset) 3622 truncate_pagecache_range(inode, first_block_offset, 3623 last_block_offset); 3624 3625 inode->i_mtime = inode->i_ctime = ext4_current_time(inode); 3626 ext4_mark_inode_dirty(handle, inode); 3627 out_stop: 3628 ext4_journal_stop(handle); 3629 out_dio: 3630 ext4_inode_resume_unlocked_dio(inode); 3631 out_mutex: 3632 mutex_unlock(&inode->i_mutex); 3633 return ret; 3634 } 3635 3636 int ext4_inode_attach_jinode(struct inode *inode) 3637 { 3638 struct ext4_inode_info *ei = EXT4_I(inode); 3639 struct jbd2_inode *jinode; 3640 3641 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal) 3642 return 0; 3643 3644 jinode = jbd2_alloc_inode(GFP_KERNEL); 3645 spin_lock(&inode->i_lock); 3646 if (!ei->jinode) { 3647 if (!jinode) { 3648 spin_unlock(&inode->i_lock); 3649 return -ENOMEM; 3650 } 3651 ei->jinode = jinode; 3652 jbd2_journal_init_jbd_inode(ei->jinode, inode); 3653 jinode = NULL; 3654 } 3655 spin_unlock(&inode->i_lock); 3656 if (unlikely(jinode != NULL)) 3657 jbd2_free_inode(jinode); 3658 return 0; 3659 } 3660 3661 /* 3662 * ext4_truncate() 3663 * 3664 * We block out ext4_get_block() block instantiations across the entire 3665 * transaction, and VFS/VM ensures that ext4_truncate() cannot run 3666 * simultaneously on behalf of the same inode. 3667 * 3668 * As we work through the truncate and commit bits of it to the journal there 3669 * is one core, guiding principle: the file's tree must always be consistent on 3670 * disk. We must be able to restart the truncate after a crash. 3671 * 3672 * The file's tree may be transiently inconsistent in memory (although it 3673 * probably isn't), but whenever we close off and commit a journal transaction, 3674 * the contents of (the filesystem + the journal) must be consistent and 3675 * restartable. It's pretty simple, really: bottom up, right to left (although 3676 * left-to-right works OK too). 3677 * 3678 * Note that at recovery time, journal replay occurs *before* the restart of 3679 * truncate against the orphan inode list. 3680 * 3681 * The committed inode has the new, desired i_size (which is the same as 3682 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see 3683 * that this inode's truncate did not complete and it will again call 3684 * ext4_truncate() to have another go. So there will be instantiated blocks 3685 * to the right of the truncation point in a crashed ext4 filesystem. But 3686 * that's fine - as long as they are linked from the inode, the post-crash 3687 * ext4_truncate() run will find them and release them. 3688 */ 3689 void ext4_truncate(struct inode *inode) 3690 { 3691 struct ext4_inode_info *ei = EXT4_I(inode); 3692 unsigned int credits; 3693 handle_t *handle; 3694 struct address_space *mapping = inode->i_mapping; 3695 3696 /* 3697 * There is a possibility that we're either freeing the inode 3698 * or it's a completely new inode. In those cases we might not 3699 * have i_mutex locked because it's not necessary. 3700 */ 3701 if (!(inode->i_state & (I_NEW|I_FREEING))) 3702 WARN_ON(!mutex_is_locked(&inode->i_mutex)); 3703 trace_ext4_truncate_enter(inode); 3704 3705 if (!ext4_can_truncate(inode)) 3706 return; 3707 3708 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); 3709 3710 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) 3711 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); 3712 3713 if (ext4_has_inline_data(inode)) { 3714 int has_inline = 1; 3715 3716 ext4_inline_data_truncate(inode, &has_inline); 3717 if (has_inline) 3718 return; 3719 } 3720 3721 /* If we zero-out tail of the page, we have to create jinode for jbd2 */ 3722 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) { 3723 if (ext4_inode_attach_jinode(inode) < 0) 3724 return; 3725 } 3726 3727 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) 3728 credits = ext4_writepage_trans_blocks(inode); 3729 else 3730 credits = ext4_blocks_for_truncate(inode); 3731 3732 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); 3733 if (IS_ERR(handle)) { 3734 ext4_std_error(inode->i_sb, PTR_ERR(handle)); 3735 return; 3736 } 3737 3738 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) 3739 ext4_block_truncate_page(handle, mapping, inode->i_size); 3740 3741 /* 3742 * We add the inode to the orphan list, so that if this 3743 * truncate spans multiple transactions, and we crash, we will 3744 * resume the truncate when the filesystem recovers. It also 3745 * marks the inode dirty, to catch the new size. 3746 * 3747 * Implication: the file must always be in a sane, consistent 3748 * truncatable state while each transaction commits. 3749 */ 3750 if (ext4_orphan_add(handle, inode)) 3751 goto out_stop; 3752 3753 down_write(&EXT4_I(inode)->i_data_sem); 3754 3755 ext4_discard_preallocations(inode); 3756 3757 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) 3758 ext4_ext_truncate(handle, inode); 3759 else 3760 ext4_ind_truncate(handle, inode); 3761 3762 up_write(&ei->i_data_sem); 3763 3764 if (IS_SYNC(inode)) 3765 ext4_handle_sync(handle); 3766 3767 out_stop: 3768 /* 3769 * If this was a simple ftruncate() and the file will remain alive, 3770 * then we need to clear up the orphan record which we created above. 3771 * However, if this was a real unlink then we were called by 3772 * ext4_evict_inode(), and we allow that function to clean up the 3773 * orphan info for us. 3774 */ 3775 if (inode->i_nlink) 3776 ext4_orphan_del(handle, inode); 3777 3778 inode->i_mtime = inode->i_ctime = ext4_current_time(inode); 3779 ext4_mark_inode_dirty(handle, inode); 3780 ext4_journal_stop(handle); 3781 3782 trace_ext4_truncate_exit(inode); 3783 } 3784 3785 /* 3786 * ext4_get_inode_loc returns with an extra refcount against the inode's 3787 * underlying buffer_head on success. If 'in_mem' is true, we have all 3788 * data in memory that is needed to recreate the on-disk version of this 3789 * inode. 3790 */ 3791 static int __ext4_get_inode_loc(struct inode *inode, 3792 struct ext4_iloc *iloc, int in_mem) 3793 { 3794 struct ext4_group_desc *gdp; 3795 struct buffer_head *bh; 3796 struct super_block *sb = inode->i_sb; 3797 ext4_fsblk_t block; 3798 int inodes_per_block, inode_offset; 3799 3800 iloc->bh = NULL; 3801 if (!ext4_valid_inum(sb, inode->i_ino)) 3802 return -EIO; 3803 3804 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb); 3805 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); 3806 if (!gdp) 3807 return -EIO; 3808 3809 /* 3810 * Figure out the offset within the block group inode table 3811 */ 3812 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; 3813 inode_offset = ((inode->i_ino - 1) % 3814 EXT4_INODES_PER_GROUP(sb)); 3815 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block); 3816 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); 3817 3818 bh = sb_getblk(sb, block); 3819 if (unlikely(!bh)) 3820 return -ENOMEM; 3821 if (!buffer_uptodate(bh)) { 3822 lock_buffer(bh); 3823 3824 /* 3825 * If the buffer has the write error flag, we have failed 3826 * to write out another inode in the same block. In this 3827 * case, we don't have to read the block because we may 3828 * read the old inode data successfully. 3829 */ 3830 if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) 3831 set_buffer_uptodate(bh); 3832 3833 if (buffer_uptodate(bh)) { 3834 /* someone brought it uptodate while we waited */ 3835 unlock_buffer(bh); 3836 goto has_buffer; 3837 } 3838 3839 /* 3840 * If we have all information of the inode in memory and this 3841 * is the only valid inode in the block, we need not read the 3842 * block. 3843 */ 3844 if (in_mem) { 3845 struct buffer_head *bitmap_bh; 3846 int i, start; 3847 3848 start = inode_offset & ~(inodes_per_block - 1); 3849 3850 /* Is the inode bitmap in cache? */ 3851 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); 3852 if (unlikely(!bitmap_bh)) 3853 goto make_io; 3854 3855 /* 3856 * If the inode bitmap isn't in cache then the 3857 * optimisation may end up performing two reads instead 3858 * of one, so skip it. 3859 */ 3860 if (!buffer_uptodate(bitmap_bh)) { 3861 brelse(bitmap_bh); 3862 goto make_io; 3863 } 3864 for (i = start; i < start + inodes_per_block; i++) { 3865 if (i == inode_offset) 3866 continue; 3867 if (ext4_test_bit(i, bitmap_bh->b_data)) 3868 break; 3869 } 3870 brelse(bitmap_bh); 3871 if (i == start + inodes_per_block) { 3872 /* all other inodes are free, so skip I/O */ 3873 memset(bh->b_data, 0, bh->b_size); 3874 set_buffer_uptodate(bh); 3875 unlock_buffer(bh); 3876 goto has_buffer; 3877 } 3878 } 3879 3880 make_io: 3881 /* 3882 * If we need to do any I/O, try to pre-readahead extra 3883 * blocks from the inode table. 3884 */ 3885 if (EXT4_SB(sb)->s_inode_readahead_blks) { 3886 ext4_fsblk_t b, end, table; 3887 unsigned num; 3888 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks; 3889 3890 table = ext4_inode_table(sb, gdp); 3891 /* s_inode_readahead_blks is always a power of 2 */ 3892 b = block & ~((ext4_fsblk_t) ra_blks - 1); 3893 if (table > b) 3894 b = table; 3895 end = b + ra_blks; 3896 num = EXT4_INODES_PER_GROUP(sb); 3897 if (ext4_has_group_desc_csum(sb)) 3898 num -= ext4_itable_unused_count(sb, gdp); 3899 table += num / inodes_per_block; 3900 if (end > table) 3901 end = table; 3902 while (b <= end) 3903 sb_breadahead(sb, b++); 3904 } 3905 3906 /* 3907 * There are other valid inodes in the buffer, this inode 3908 * has in-inode xattrs, or we don't have this inode in memory. 3909 * Read the block from disk. 3910 */ 3911 trace_ext4_load_inode(inode); 3912 get_bh(bh); 3913 bh->b_end_io = end_buffer_read_sync; 3914 submit_bh(READ | REQ_META | REQ_PRIO, bh); 3915 wait_on_buffer(bh); 3916 if (!buffer_uptodate(bh)) { 3917 EXT4_ERROR_INODE_BLOCK(inode, block, 3918 "unable to read itable block"); 3919 brelse(bh); 3920 return -EIO; 3921 } 3922 } 3923 has_buffer: 3924 iloc->bh = bh; 3925 return 0; 3926 } 3927 3928 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) 3929 { 3930 /* We have all inode data except xattrs in memory here. */ 3931 return __ext4_get_inode_loc(inode, iloc, 3932 !ext4_test_inode_state(inode, EXT4_STATE_XATTR)); 3933 } 3934 3935 void ext4_set_inode_flags(struct inode *inode) 3936 { 3937 unsigned int flags = EXT4_I(inode)->i_flags; 3938 unsigned int new_fl = 0; 3939 3940 if (flags & EXT4_SYNC_FL) 3941 new_fl |= S_SYNC; 3942 if (flags & EXT4_APPEND_FL) 3943 new_fl |= S_APPEND; 3944 if (flags & EXT4_IMMUTABLE_FL) 3945 new_fl |= S_IMMUTABLE; 3946 if (flags & EXT4_NOATIME_FL) 3947 new_fl |= S_NOATIME; 3948 if (flags & EXT4_DIRSYNC_FL) 3949 new_fl |= S_DIRSYNC; 3950 if (test_opt(inode->i_sb, DAX)) 3951 new_fl |= S_DAX; 3952 inode_set_flags(inode, new_fl, 3953 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX); 3954 } 3955 3956 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */ 3957 void ext4_get_inode_flags(struct ext4_inode_info *ei) 3958 { 3959 unsigned int vfs_fl; 3960 unsigned long old_fl, new_fl; 3961 3962 do { 3963 vfs_fl = ei->vfs_inode.i_flags; 3964 old_fl = ei->i_flags; 3965 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL| 3966 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL| 3967 EXT4_DIRSYNC_FL); 3968 if (vfs_fl & S_SYNC) 3969 new_fl |= EXT4_SYNC_FL; 3970 if (vfs_fl & S_APPEND) 3971 new_fl |= EXT4_APPEND_FL; 3972 if (vfs_fl & S_IMMUTABLE) 3973 new_fl |= EXT4_IMMUTABLE_FL; 3974 if (vfs_fl & S_NOATIME) 3975 new_fl |= EXT4_NOATIME_FL; 3976 if (vfs_fl & S_DIRSYNC) 3977 new_fl |= EXT4_DIRSYNC_FL; 3978 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl); 3979 } 3980 3981 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, 3982 struct ext4_inode_info *ei) 3983 { 3984 blkcnt_t i_blocks ; 3985 struct inode *inode = &(ei->vfs_inode); 3986 struct super_block *sb = inode->i_sb; 3987 3988 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, 3989 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) { 3990 /* we are using combined 48 bit field */ 3991 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | 3992 le32_to_cpu(raw_inode->i_blocks_lo); 3993 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { 3994 /* i_blocks represent file system block size */ 3995 return i_blocks << (inode->i_blkbits - 9); 3996 } else { 3997 return i_blocks; 3998 } 3999 } else { 4000 return le32_to_cpu(raw_inode->i_blocks_lo); 4001 } 4002 } 4003 4004 static inline void ext4_iget_extra_inode(struct inode *inode, 4005 struct ext4_inode *raw_inode, 4006 struct ext4_inode_info *ei) 4007 { 4008 __le32 *magic = (void *)raw_inode + 4009 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize; 4010 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) { 4011 ext4_set_inode_state(inode, EXT4_STATE_XATTR); 4012 ext4_find_inline_data_nolock(inode); 4013 } else 4014 EXT4_I(inode)->i_inline_off = 0; 4015 } 4016 4017 struct inode *ext4_iget(struct super_block *sb, unsigned long ino) 4018 { 4019 struct ext4_iloc iloc; 4020 struct ext4_inode *raw_inode; 4021 struct ext4_inode_info *ei; 4022 struct inode *inode; 4023 journal_t *journal = EXT4_SB(sb)->s_journal; 4024 long ret; 4025 int block; 4026 uid_t i_uid; 4027 gid_t i_gid; 4028 4029 inode = iget_locked(sb, ino); 4030 if (!inode) 4031 return ERR_PTR(-ENOMEM); 4032 if (!(inode->i_state & I_NEW)) 4033 return inode; 4034 4035 ei = EXT4_I(inode); 4036 iloc.bh = NULL; 4037 4038 ret = __ext4_get_inode_loc(inode, &iloc, 0); 4039 if (ret < 0) 4040 goto bad_inode; 4041 raw_inode = ext4_raw_inode(&iloc); 4042 4043 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { 4044 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); 4045 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > 4046 EXT4_INODE_SIZE(inode->i_sb)) { 4047 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)", 4048 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize, 4049 EXT4_INODE_SIZE(inode->i_sb)); 4050 ret = -EIO; 4051 goto bad_inode; 4052 } 4053 } else 4054 ei->i_extra_isize = 0; 4055 4056 /* Precompute checksum seed for inode metadata */ 4057 if (ext4_has_metadata_csum(sb)) { 4058 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 4059 __u32 csum; 4060 __le32 inum = cpu_to_le32(inode->i_ino); 4061 __le32 gen = raw_inode->i_generation; 4062 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, 4063 sizeof(inum)); 4064 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, 4065 sizeof(gen)); 4066 } 4067 4068 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) { 4069 EXT4_ERROR_INODE(inode, "checksum invalid"); 4070 ret = -EIO; 4071 goto bad_inode; 4072 } 4073 4074 inode->i_mode = le16_to_cpu(raw_inode->i_mode); 4075 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); 4076 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); 4077 if (!(test_opt(inode->i_sb, NO_UID32))) { 4078 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; 4079 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; 4080 } 4081 i_uid_write(inode, i_uid); 4082 i_gid_write(inode, i_gid); 4083 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count)); 4084 4085 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ 4086 ei->i_inline_off = 0; 4087 ei->i_dir_start_lookup = 0; 4088 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); 4089 /* We now have enough fields to check if the inode was active or not. 4090 * This is needed because nfsd might try to access dead inodes 4091 * the test is that same one that e2fsck uses 4092 * NeilBrown 1999oct15 4093 */ 4094 if (inode->i_nlink == 0) { 4095 if ((inode->i_mode == 0 || 4096 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) && 4097 ino != EXT4_BOOT_LOADER_INO) { 4098 /* this inode is deleted */ 4099 ret = -ESTALE; 4100 goto bad_inode; 4101 } 4102 /* The only unlinked inodes we let through here have 4103 * valid i_mode and are being read by the orphan 4104 * recovery code: that's fine, we're about to complete 4105 * the process of deleting those. 4106 * OR it is the EXT4_BOOT_LOADER_INO which is 4107 * not initialized on a new filesystem. */ 4108 } 4109 ei->i_flags = le32_to_cpu(raw_inode->i_flags); 4110 inode->i_blocks = ext4_inode_blocks(raw_inode, ei); 4111 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); 4112 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) 4113 ei->i_file_acl |= 4114 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; 4115 inode->i_size = ext4_isize(raw_inode); 4116 ei->i_disksize = inode->i_size; 4117 #ifdef CONFIG_QUOTA 4118 ei->i_reserved_quota = 0; 4119 #endif 4120 inode->i_generation = le32_to_cpu(raw_inode->i_generation); 4121 ei->i_block_group = iloc.block_group; 4122 ei->i_last_alloc_group = ~0; 4123 /* 4124 * NOTE! The in-memory inode i_data array is in little-endian order 4125 * even on big-endian machines: we do NOT byteswap the block numbers! 4126 */ 4127 for (block = 0; block < EXT4_N_BLOCKS; block++) 4128 ei->i_data[block] = raw_inode->i_block[block]; 4129 INIT_LIST_HEAD(&ei->i_orphan); 4130 4131 /* 4132 * Set transaction id's of transactions that have to be committed 4133 * to finish f[data]sync. We set them to currently running transaction 4134 * as we cannot be sure that the inode or some of its metadata isn't 4135 * part of the transaction - the inode could have been reclaimed and 4136 * now it is reread from disk. 4137 */ 4138 if (journal) { 4139 transaction_t *transaction; 4140 tid_t tid; 4141 4142 read_lock(&journal->j_state_lock); 4143 if (journal->j_running_transaction) 4144 transaction = journal->j_running_transaction; 4145 else 4146 transaction = journal->j_committing_transaction; 4147 if (transaction) 4148 tid = transaction->t_tid; 4149 else 4150 tid = journal->j_commit_sequence; 4151 read_unlock(&journal->j_state_lock); 4152 ei->i_sync_tid = tid; 4153 ei->i_datasync_tid = tid; 4154 } 4155 4156 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { 4157 if (ei->i_extra_isize == 0) { 4158 /* The extra space is currently unused. Use it. */ 4159 ei->i_extra_isize = sizeof(struct ext4_inode) - 4160 EXT4_GOOD_OLD_INODE_SIZE; 4161 } else { 4162 ext4_iget_extra_inode(inode, raw_inode, ei); 4163 } 4164 } 4165 4166 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); 4167 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); 4168 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); 4169 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); 4170 4171 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { 4172 inode->i_version = le32_to_cpu(raw_inode->i_disk_version); 4173 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { 4174 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) 4175 inode->i_version |= 4176 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; 4177 } 4178 } 4179 4180 ret = 0; 4181 if (ei->i_file_acl && 4182 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) { 4183 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu", 4184 ei->i_file_acl); 4185 ret = -EIO; 4186 goto bad_inode; 4187 } else if (!ext4_has_inline_data(inode)) { 4188 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { 4189 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 4190 (S_ISLNK(inode->i_mode) && 4191 !ext4_inode_is_fast_symlink(inode)))) 4192 /* Validate extent which is part of inode */ 4193 ret = ext4_ext_check_inode(inode); 4194 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 4195 (S_ISLNK(inode->i_mode) && 4196 !ext4_inode_is_fast_symlink(inode))) { 4197 /* Validate block references which are part of inode */ 4198 ret = ext4_ind_check_inode(inode); 4199 } 4200 } 4201 if (ret) 4202 goto bad_inode; 4203 4204 if (S_ISREG(inode->i_mode)) { 4205 inode->i_op = &ext4_file_inode_operations; 4206 inode->i_fop = &ext4_file_operations; 4207 ext4_set_aops(inode); 4208 } else if (S_ISDIR(inode->i_mode)) { 4209 inode->i_op = &ext4_dir_inode_operations; 4210 inode->i_fop = &ext4_dir_operations; 4211 } else if (S_ISLNK(inode->i_mode)) { 4212 if (ext4_encrypted_inode(inode)) { 4213 inode->i_op = &ext4_encrypted_symlink_inode_operations; 4214 ext4_set_aops(inode); 4215 } else if (ext4_inode_is_fast_symlink(inode)) { 4216 inode->i_link = (char *)ei->i_data; 4217 inode->i_op = &ext4_fast_symlink_inode_operations; 4218 nd_terminate_link(ei->i_data, inode->i_size, 4219 sizeof(ei->i_data) - 1); 4220 } else { 4221 inode->i_op = &ext4_symlink_inode_operations; 4222 ext4_set_aops(inode); 4223 } 4224 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || 4225 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { 4226 inode->i_op = &ext4_special_inode_operations; 4227 if (raw_inode->i_block[0]) 4228 init_special_inode(inode, inode->i_mode, 4229 old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); 4230 else 4231 init_special_inode(inode, inode->i_mode, 4232 new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); 4233 } else if (ino == EXT4_BOOT_LOADER_INO) { 4234 make_bad_inode(inode); 4235 } else { 4236 ret = -EIO; 4237 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode); 4238 goto bad_inode; 4239 } 4240 brelse(iloc.bh); 4241 ext4_set_inode_flags(inode); 4242 unlock_new_inode(inode); 4243 return inode; 4244 4245 bad_inode: 4246 brelse(iloc.bh); 4247 iget_failed(inode); 4248 return ERR_PTR(ret); 4249 } 4250 4251 struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino) 4252 { 4253 if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO) 4254 return ERR_PTR(-EIO); 4255 return ext4_iget(sb, ino); 4256 } 4257 4258 static int ext4_inode_blocks_set(handle_t *handle, 4259 struct ext4_inode *raw_inode, 4260 struct ext4_inode_info *ei) 4261 { 4262 struct inode *inode = &(ei->vfs_inode); 4263 u64 i_blocks = inode->i_blocks; 4264 struct super_block *sb = inode->i_sb; 4265 4266 if (i_blocks <= ~0U) { 4267 /* 4268 * i_blocks can be represented in a 32 bit variable 4269 * as multiple of 512 bytes 4270 */ 4271 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); 4272 raw_inode->i_blocks_high = 0; 4273 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); 4274 return 0; 4275 } 4276 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) 4277 return -EFBIG; 4278 4279 if (i_blocks <= 0xffffffffffffULL) { 4280 /* 4281 * i_blocks can be represented in a 48 bit variable 4282 * as multiple of 512 bytes 4283 */ 4284 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); 4285 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); 4286 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); 4287 } else { 4288 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); 4289 /* i_block is stored in file system block size */ 4290 i_blocks = i_blocks >> (inode->i_blkbits - 9); 4291 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); 4292 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); 4293 } 4294 return 0; 4295 } 4296 4297 struct other_inode { 4298 unsigned long orig_ino; 4299 struct ext4_inode *raw_inode; 4300 }; 4301 4302 static int other_inode_match(struct inode * inode, unsigned long ino, 4303 void *data) 4304 { 4305 struct other_inode *oi = (struct other_inode *) data; 4306 4307 if ((inode->i_ino != ino) || 4308 (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW | 4309 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) || 4310 ((inode->i_state & I_DIRTY_TIME) == 0)) 4311 return 0; 4312 spin_lock(&inode->i_lock); 4313 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW | 4314 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) && 4315 (inode->i_state & I_DIRTY_TIME)) { 4316 struct ext4_inode_info *ei = EXT4_I(inode); 4317 4318 inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED); 4319 spin_unlock(&inode->i_lock); 4320 4321 spin_lock(&ei->i_raw_lock); 4322 EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode); 4323 EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode); 4324 EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode); 4325 ext4_inode_csum_set(inode, oi->raw_inode, ei); 4326 spin_unlock(&ei->i_raw_lock); 4327 trace_ext4_other_inode_update_time(inode, oi->orig_ino); 4328 return -1; 4329 } 4330 spin_unlock(&inode->i_lock); 4331 return -1; 4332 } 4333 4334 /* 4335 * Opportunistically update the other time fields for other inodes in 4336 * the same inode table block. 4337 */ 4338 static void ext4_update_other_inodes_time(struct super_block *sb, 4339 unsigned long orig_ino, char *buf) 4340 { 4341 struct other_inode oi; 4342 unsigned long ino; 4343 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; 4344 int inode_size = EXT4_INODE_SIZE(sb); 4345 4346 oi.orig_ino = orig_ino; 4347 ino = (orig_ino & ~(inodes_per_block - 1)) + 1; 4348 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) { 4349 if (ino == orig_ino) 4350 continue; 4351 oi.raw_inode = (struct ext4_inode *) buf; 4352 (void) find_inode_nowait(sb, ino, other_inode_match, &oi); 4353 } 4354 } 4355 4356 /* 4357 * Post the struct inode info into an on-disk inode location in the 4358 * buffer-cache. This gobbles the caller's reference to the 4359 * buffer_head in the inode location struct. 4360 * 4361 * The caller must have write access to iloc->bh. 4362 */ 4363 static int ext4_do_update_inode(handle_t *handle, 4364 struct inode *inode, 4365 struct ext4_iloc *iloc) 4366 { 4367 struct ext4_inode *raw_inode = ext4_raw_inode(iloc); 4368 struct ext4_inode_info *ei = EXT4_I(inode); 4369 struct buffer_head *bh = iloc->bh; 4370 struct super_block *sb = inode->i_sb; 4371 int err = 0, rc, block; 4372 int need_datasync = 0, set_large_file = 0; 4373 uid_t i_uid; 4374 gid_t i_gid; 4375 4376 spin_lock(&ei->i_raw_lock); 4377 4378 /* For fields not tracked in the in-memory inode, 4379 * initialise them to zero for new inodes. */ 4380 if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) 4381 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); 4382 4383 ext4_get_inode_flags(ei); 4384 raw_inode->i_mode = cpu_to_le16(inode->i_mode); 4385 i_uid = i_uid_read(inode); 4386 i_gid = i_gid_read(inode); 4387 if (!(test_opt(inode->i_sb, NO_UID32))) { 4388 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid)); 4389 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid)); 4390 /* 4391 * Fix up interoperability with old kernels. Otherwise, old inodes get 4392 * re-used with the upper 16 bits of the uid/gid intact 4393 */ 4394 if (!ei->i_dtime) { 4395 raw_inode->i_uid_high = 4396 cpu_to_le16(high_16_bits(i_uid)); 4397 raw_inode->i_gid_high = 4398 cpu_to_le16(high_16_bits(i_gid)); 4399 } else { 4400 raw_inode->i_uid_high = 0; 4401 raw_inode->i_gid_high = 0; 4402 } 4403 } else { 4404 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid)); 4405 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid)); 4406 raw_inode->i_uid_high = 0; 4407 raw_inode->i_gid_high = 0; 4408 } 4409 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); 4410 4411 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); 4412 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); 4413 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); 4414 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); 4415 4416 err = ext4_inode_blocks_set(handle, raw_inode, ei); 4417 if (err) { 4418 spin_unlock(&ei->i_raw_lock); 4419 goto out_brelse; 4420 } 4421 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); 4422 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF); 4423 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) 4424 raw_inode->i_file_acl_high = 4425 cpu_to_le16(ei->i_file_acl >> 32); 4426 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); 4427 if (ei->i_disksize != ext4_isize(raw_inode)) { 4428 ext4_isize_set(raw_inode, ei->i_disksize); 4429 need_datasync = 1; 4430 } 4431 if (ei->i_disksize > 0x7fffffffULL) { 4432 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, 4433 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) || 4434 EXT4_SB(sb)->s_es->s_rev_level == 4435 cpu_to_le32(EXT4_GOOD_OLD_REV)) 4436 set_large_file = 1; 4437 } 4438 raw_inode->i_generation = cpu_to_le32(inode->i_generation); 4439 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { 4440 if (old_valid_dev(inode->i_rdev)) { 4441 raw_inode->i_block[0] = 4442 cpu_to_le32(old_encode_dev(inode->i_rdev)); 4443 raw_inode->i_block[1] = 0; 4444 } else { 4445 raw_inode->i_block[0] = 0; 4446 raw_inode->i_block[1] = 4447 cpu_to_le32(new_encode_dev(inode->i_rdev)); 4448 raw_inode->i_block[2] = 0; 4449 } 4450 } else if (!ext4_has_inline_data(inode)) { 4451 for (block = 0; block < EXT4_N_BLOCKS; block++) 4452 raw_inode->i_block[block] = ei->i_data[block]; 4453 } 4454 4455 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { 4456 raw_inode->i_disk_version = cpu_to_le32(inode->i_version); 4457 if (ei->i_extra_isize) { 4458 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) 4459 raw_inode->i_version_hi = 4460 cpu_to_le32(inode->i_version >> 32); 4461 raw_inode->i_extra_isize = 4462 cpu_to_le16(ei->i_extra_isize); 4463 } 4464 } 4465 ext4_inode_csum_set(inode, raw_inode, ei); 4466 spin_unlock(&ei->i_raw_lock); 4467 if (inode->i_sb->s_flags & MS_LAZYTIME) 4468 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino, 4469 bh->b_data); 4470 4471 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); 4472 rc = ext4_handle_dirty_metadata(handle, NULL, bh); 4473 if (!err) 4474 err = rc; 4475 ext4_clear_inode_state(inode, EXT4_STATE_NEW); 4476 if (set_large_file) { 4477 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access"); 4478 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh); 4479 if (err) 4480 goto out_brelse; 4481 ext4_update_dynamic_rev(sb); 4482 EXT4_SET_RO_COMPAT_FEATURE(sb, 4483 EXT4_FEATURE_RO_COMPAT_LARGE_FILE); 4484 ext4_handle_sync(handle); 4485 err = ext4_handle_dirty_super(handle, sb); 4486 } 4487 ext4_update_inode_fsync_trans(handle, inode, need_datasync); 4488 out_brelse: 4489 brelse(bh); 4490 ext4_std_error(inode->i_sb, err); 4491 return err; 4492 } 4493 4494 /* 4495 * ext4_write_inode() 4496 * 4497 * We are called from a few places: 4498 * 4499 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files. 4500 * Here, there will be no transaction running. We wait for any running 4501 * transaction to commit. 4502 * 4503 * - Within flush work (sys_sync(), kupdate and such). 4504 * We wait on commit, if told to. 4505 * 4506 * - Within iput_final() -> write_inode_now() 4507 * We wait on commit, if told to. 4508 * 4509 * In all cases it is actually safe for us to return without doing anything, 4510 * because the inode has been copied into a raw inode buffer in 4511 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL 4512 * writeback. 4513 * 4514 * Note that we are absolutely dependent upon all inode dirtiers doing the 4515 * right thing: they *must* call mark_inode_dirty() after dirtying info in 4516 * which we are interested. 4517 * 4518 * It would be a bug for them to not do this. The code: 4519 * 4520 * mark_inode_dirty(inode) 4521 * stuff(); 4522 * inode->i_size = expr; 4523 * 4524 * is in error because write_inode() could occur while `stuff()' is running, 4525 * and the new i_size will be lost. Plus the inode will no longer be on the 4526 * superblock's dirty inode list. 4527 */ 4528 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) 4529 { 4530 int err; 4531 4532 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC)) 4533 return 0; 4534 4535 if (EXT4_SB(inode->i_sb)->s_journal) { 4536 if (ext4_journal_current_handle()) { 4537 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); 4538 dump_stack(); 4539 return -EIO; 4540 } 4541 4542 /* 4543 * No need to force transaction in WB_SYNC_NONE mode. Also 4544 * ext4_sync_fs() will force the commit after everything is 4545 * written. 4546 */ 4547 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync) 4548 return 0; 4549 4550 err = ext4_force_commit(inode->i_sb); 4551 } else { 4552 struct ext4_iloc iloc; 4553 4554 err = __ext4_get_inode_loc(inode, &iloc, 0); 4555 if (err) 4556 return err; 4557 /* 4558 * sync(2) will flush the whole buffer cache. No need to do 4559 * it here separately for each inode. 4560 */ 4561 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) 4562 sync_dirty_buffer(iloc.bh); 4563 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { 4564 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr, 4565 "IO error syncing inode"); 4566 err = -EIO; 4567 } 4568 brelse(iloc.bh); 4569 } 4570 return err; 4571 } 4572 4573 /* 4574 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate 4575 * buffers that are attached to a page stradding i_size and are undergoing 4576 * commit. In that case we have to wait for commit to finish and try again. 4577 */ 4578 static void ext4_wait_for_tail_page_commit(struct inode *inode) 4579 { 4580 struct page *page; 4581 unsigned offset; 4582 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; 4583 tid_t commit_tid = 0; 4584 int ret; 4585 4586 offset = inode->i_size & (PAGE_CACHE_SIZE - 1); 4587 /* 4588 * All buffers in the last page remain valid? Then there's nothing to 4589 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE == 4590 * blocksize case 4591 */ 4592 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits)) 4593 return; 4594 while (1) { 4595 page = find_lock_page(inode->i_mapping, 4596 inode->i_size >> PAGE_CACHE_SHIFT); 4597 if (!page) 4598 return; 4599 ret = __ext4_journalled_invalidatepage(page, offset, 4600 PAGE_CACHE_SIZE - offset); 4601 unlock_page(page); 4602 page_cache_release(page); 4603 if (ret != -EBUSY) 4604 return; 4605 commit_tid = 0; 4606 read_lock(&journal->j_state_lock); 4607 if (journal->j_committing_transaction) 4608 commit_tid = journal->j_committing_transaction->t_tid; 4609 read_unlock(&journal->j_state_lock); 4610 if (commit_tid) 4611 jbd2_log_wait_commit(journal, commit_tid); 4612 } 4613 } 4614 4615 /* 4616 * ext4_setattr() 4617 * 4618 * Called from notify_change. 4619 * 4620 * We want to trap VFS attempts to truncate the file as soon as 4621 * possible. In particular, we want to make sure that when the VFS 4622 * shrinks i_size, we put the inode on the orphan list and modify 4623 * i_disksize immediately, so that during the subsequent flushing of 4624 * dirty pages and freeing of disk blocks, we can guarantee that any 4625 * commit will leave the blocks being flushed in an unused state on 4626 * disk. (On recovery, the inode will get truncated and the blocks will 4627 * be freed, so we have a strong guarantee that no future commit will 4628 * leave these blocks visible to the user.) 4629 * 4630 * Another thing we have to assure is that if we are in ordered mode 4631 * and inode is still attached to the committing transaction, we must 4632 * we start writeout of all the dirty pages which are being truncated. 4633 * This way we are sure that all the data written in the previous 4634 * transaction are already on disk (truncate waits for pages under 4635 * writeback). 4636 * 4637 * Called with inode->i_mutex down. 4638 */ 4639 int ext4_setattr(struct dentry *dentry, struct iattr *attr) 4640 { 4641 struct inode *inode = d_inode(dentry); 4642 int error, rc = 0; 4643 int orphan = 0; 4644 const unsigned int ia_valid = attr->ia_valid; 4645 4646 error = inode_change_ok(inode, attr); 4647 if (error) 4648 return error; 4649 4650 if (is_quota_modification(inode, attr)) 4651 dquot_initialize(inode); 4652 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) || 4653 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) { 4654 handle_t *handle; 4655 4656 /* (user+group)*(old+new) structure, inode write (sb, 4657 * inode block, ? - but truncate inode update has it) */ 4658 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 4659 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) + 4660 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3); 4661 if (IS_ERR(handle)) { 4662 error = PTR_ERR(handle); 4663 goto err_out; 4664 } 4665 error = dquot_transfer(inode, attr); 4666 if (error) { 4667 ext4_journal_stop(handle); 4668 return error; 4669 } 4670 /* Update corresponding info in inode so that everything is in 4671 * one transaction */ 4672 if (attr->ia_valid & ATTR_UID) 4673 inode->i_uid = attr->ia_uid; 4674 if (attr->ia_valid & ATTR_GID) 4675 inode->i_gid = attr->ia_gid; 4676 error = ext4_mark_inode_dirty(handle, inode); 4677 ext4_journal_stop(handle); 4678 } 4679 4680 if (attr->ia_valid & ATTR_SIZE) { 4681 handle_t *handle; 4682 loff_t oldsize = inode->i_size; 4683 int shrink = (attr->ia_size <= inode->i_size); 4684 4685 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { 4686 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 4687 4688 if (attr->ia_size > sbi->s_bitmap_maxbytes) 4689 return -EFBIG; 4690 } 4691 if (!S_ISREG(inode->i_mode)) 4692 return -EINVAL; 4693 4694 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size) 4695 inode_inc_iversion(inode); 4696 4697 if (ext4_should_order_data(inode) && 4698 (attr->ia_size < inode->i_size)) { 4699 error = ext4_begin_ordered_truncate(inode, 4700 attr->ia_size); 4701 if (error) 4702 goto err_out; 4703 } 4704 if (attr->ia_size != inode->i_size) { 4705 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3); 4706 if (IS_ERR(handle)) { 4707 error = PTR_ERR(handle); 4708 goto err_out; 4709 } 4710 if (ext4_handle_valid(handle) && shrink) { 4711 error = ext4_orphan_add(handle, inode); 4712 orphan = 1; 4713 } 4714 down_write(&EXT4_I(inode)->i_data_sem); 4715 EXT4_I(inode)->i_disksize = attr->ia_size; 4716 rc = ext4_mark_inode_dirty(handle, inode); 4717 if (!error) 4718 error = rc; 4719 /* 4720 * We have to update i_size under i_data_sem together 4721 * with i_disksize to avoid races with writeback code 4722 * running ext4_wb_update_i_disksize(). 4723 */ 4724 if (!error) 4725 i_size_write(inode, attr->ia_size); 4726 up_write(&EXT4_I(inode)->i_data_sem); 4727 ext4_journal_stop(handle); 4728 if (error) { 4729 if (orphan) 4730 ext4_orphan_del(NULL, inode); 4731 goto err_out; 4732 } 4733 } 4734 if (!shrink) 4735 pagecache_isize_extended(inode, oldsize, inode->i_size); 4736 4737 /* 4738 * Blocks are going to be removed from the inode. Wait 4739 * for dio in flight. Temporarily disable 4740 * dioread_nolock to prevent livelock. 4741 */ 4742 if (orphan) { 4743 if (!ext4_should_journal_data(inode)) { 4744 ext4_inode_block_unlocked_dio(inode); 4745 inode_dio_wait(inode); 4746 ext4_inode_resume_unlocked_dio(inode); 4747 } else 4748 ext4_wait_for_tail_page_commit(inode); 4749 } 4750 /* 4751 * Truncate pagecache after we've waited for commit 4752 * in data=journal mode to make pages freeable. 4753 */ 4754 truncate_pagecache(inode, inode->i_size); 4755 if (shrink) 4756 ext4_truncate(inode); 4757 } 4758 4759 if (!rc) { 4760 setattr_copy(inode, attr); 4761 mark_inode_dirty(inode); 4762 } 4763 4764 /* 4765 * If the call to ext4_truncate failed to get a transaction handle at 4766 * all, we need to clean up the in-core orphan list manually. 4767 */ 4768 if (orphan && inode->i_nlink) 4769 ext4_orphan_del(NULL, inode); 4770 4771 if (!rc && (ia_valid & ATTR_MODE)) 4772 rc = posix_acl_chmod(inode, inode->i_mode); 4773 4774 err_out: 4775 ext4_std_error(inode->i_sb, error); 4776 if (!error) 4777 error = rc; 4778 return error; 4779 } 4780 4781 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry, 4782 struct kstat *stat) 4783 { 4784 struct inode *inode; 4785 unsigned long long delalloc_blocks; 4786 4787 inode = d_inode(dentry); 4788 generic_fillattr(inode, stat); 4789 4790 /* 4791 * If there is inline data in the inode, the inode will normally not 4792 * have data blocks allocated (it may have an external xattr block). 4793 * Report at least one sector for such files, so tools like tar, rsync, 4794 * others doen't incorrectly think the file is completely sparse. 4795 */ 4796 if (unlikely(ext4_has_inline_data(inode))) 4797 stat->blocks += (stat->size + 511) >> 9; 4798 4799 /* 4800 * We can't update i_blocks if the block allocation is delayed 4801 * otherwise in the case of system crash before the real block 4802 * allocation is done, we will have i_blocks inconsistent with 4803 * on-disk file blocks. 4804 * We always keep i_blocks updated together with real 4805 * allocation. But to not confuse with user, stat 4806 * will return the blocks that include the delayed allocation 4807 * blocks for this file. 4808 */ 4809 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb), 4810 EXT4_I(inode)->i_reserved_data_blocks); 4811 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9); 4812 return 0; 4813 } 4814 4815 static int ext4_index_trans_blocks(struct inode *inode, int lblocks, 4816 int pextents) 4817 { 4818 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) 4819 return ext4_ind_trans_blocks(inode, lblocks); 4820 return ext4_ext_index_trans_blocks(inode, pextents); 4821 } 4822 4823 /* 4824 * Account for index blocks, block groups bitmaps and block group 4825 * descriptor blocks if modify datablocks and index blocks 4826 * worse case, the indexs blocks spread over different block groups 4827 * 4828 * If datablocks are discontiguous, they are possible to spread over 4829 * different block groups too. If they are contiguous, with flexbg, 4830 * they could still across block group boundary. 4831 * 4832 * Also account for superblock, inode, quota and xattr blocks 4833 */ 4834 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, 4835 int pextents) 4836 { 4837 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); 4838 int gdpblocks; 4839 int idxblocks; 4840 int ret = 0; 4841 4842 /* 4843 * How many index blocks need to touch to map @lblocks logical blocks 4844 * to @pextents physical extents? 4845 */ 4846 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents); 4847 4848 ret = idxblocks; 4849 4850 /* 4851 * Now let's see how many group bitmaps and group descriptors need 4852 * to account 4853 */ 4854 groups = idxblocks + pextents; 4855 gdpblocks = groups; 4856 if (groups > ngroups) 4857 groups = ngroups; 4858 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) 4859 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; 4860 4861 /* bitmaps and block group descriptor blocks */ 4862 ret += groups + gdpblocks; 4863 4864 /* Blocks for super block, inode, quota and xattr blocks */ 4865 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); 4866 4867 return ret; 4868 } 4869 4870 /* 4871 * Calculate the total number of credits to reserve to fit 4872 * the modification of a single pages into a single transaction, 4873 * which may include multiple chunks of block allocations. 4874 * 4875 * This could be called via ext4_write_begin() 4876 * 4877 * We need to consider the worse case, when 4878 * one new block per extent. 4879 */ 4880 int ext4_writepage_trans_blocks(struct inode *inode) 4881 { 4882 int bpp = ext4_journal_blocks_per_page(inode); 4883 int ret; 4884 4885 ret = ext4_meta_trans_blocks(inode, bpp, bpp); 4886 4887 /* Account for data blocks for journalled mode */ 4888 if (ext4_should_journal_data(inode)) 4889 ret += bpp; 4890 return ret; 4891 } 4892 4893 /* 4894 * Calculate the journal credits for a chunk of data modification. 4895 * 4896 * This is called from DIO, fallocate or whoever calling 4897 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. 4898 * 4899 * journal buffers for data blocks are not included here, as DIO 4900 * and fallocate do no need to journal data buffers. 4901 */ 4902 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) 4903 { 4904 return ext4_meta_trans_blocks(inode, nrblocks, 1); 4905 } 4906 4907 /* 4908 * The caller must have previously called ext4_reserve_inode_write(). 4909 * Give this, we know that the caller already has write access to iloc->bh. 4910 */ 4911 int ext4_mark_iloc_dirty(handle_t *handle, 4912 struct inode *inode, struct ext4_iloc *iloc) 4913 { 4914 int err = 0; 4915 4916 if (IS_I_VERSION(inode)) 4917 inode_inc_iversion(inode); 4918 4919 /* the do_update_inode consumes one bh->b_count */ 4920 get_bh(iloc->bh); 4921 4922 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ 4923 err = ext4_do_update_inode(handle, inode, iloc); 4924 put_bh(iloc->bh); 4925 return err; 4926 } 4927 4928 /* 4929 * On success, We end up with an outstanding reference count against 4930 * iloc->bh. This _must_ be cleaned up later. 4931 */ 4932 4933 int 4934 ext4_reserve_inode_write(handle_t *handle, struct inode *inode, 4935 struct ext4_iloc *iloc) 4936 { 4937 int err; 4938 4939 err = ext4_get_inode_loc(inode, iloc); 4940 if (!err) { 4941 BUFFER_TRACE(iloc->bh, "get_write_access"); 4942 err = ext4_journal_get_write_access(handle, iloc->bh); 4943 if (err) { 4944 brelse(iloc->bh); 4945 iloc->bh = NULL; 4946 } 4947 } 4948 ext4_std_error(inode->i_sb, err); 4949 return err; 4950 } 4951 4952 /* 4953 * Expand an inode by new_extra_isize bytes. 4954 * Returns 0 on success or negative error number on failure. 4955 */ 4956 static int ext4_expand_extra_isize(struct inode *inode, 4957 unsigned int new_extra_isize, 4958 struct ext4_iloc iloc, 4959 handle_t *handle) 4960 { 4961 struct ext4_inode *raw_inode; 4962 struct ext4_xattr_ibody_header *header; 4963 4964 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) 4965 return 0; 4966 4967 raw_inode = ext4_raw_inode(&iloc); 4968 4969 header = IHDR(inode, raw_inode); 4970 4971 /* No extended attributes present */ 4972 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || 4973 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { 4974 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0, 4975 new_extra_isize); 4976 EXT4_I(inode)->i_extra_isize = new_extra_isize; 4977 return 0; 4978 } 4979 4980 /* try to expand with EAs present */ 4981 return ext4_expand_extra_isize_ea(inode, new_extra_isize, 4982 raw_inode, handle); 4983 } 4984 4985 /* 4986 * What we do here is to mark the in-core inode as clean with respect to inode 4987 * dirtiness (it may still be data-dirty). 4988 * This means that the in-core inode may be reaped by prune_icache 4989 * without having to perform any I/O. This is a very good thing, 4990 * because *any* task may call prune_icache - even ones which 4991 * have a transaction open against a different journal. 4992 * 4993 * Is this cheating? Not really. Sure, we haven't written the 4994 * inode out, but prune_icache isn't a user-visible syncing function. 4995 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) 4996 * we start and wait on commits. 4997 */ 4998 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) 4999 { 5000 struct ext4_iloc iloc; 5001 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 5002 static unsigned int mnt_count; 5003 int err, ret; 5004 5005 might_sleep(); 5006 trace_ext4_mark_inode_dirty(inode, _RET_IP_); 5007 err = ext4_reserve_inode_write(handle, inode, &iloc); 5008 if (ext4_handle_valid(handle) && 5009 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize && 5010 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) { 5011 /* 5012 * We need extra buffer credits since we may write into EA block 5013 * with this same handle. If journal_extend fails, then it will 5014 * only result in a minor loss of functionality for that inode. 5015 * If this is felt to be critical, then e2fsck should be run to 5016 * force a large enough s_min_extra_isize. 5017 */ 5018 if ((jbd2_journal_extend(handle, 5019 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) { 5020 ret = ext4_expand_extra_isize(inode, 5021 sbi->s_want_extra_isize, 5022 iloc, handle); 5023 if (ret) { 5024 ext4_set_inode_state(inode, 5025 EXT4_STATE_NO_EXPAND); 5026 if (mnt_count != 5027 le16_to_cpu(sbi->s_es->s_mnt_count)) { 5028 ext4_warning(inode->i_sb, 5029 "Unable to expand inode %lu. Delete" 5030 " some EAs or run e2fsck.", 5031 inode->i_ino); 5032 mnt_count = 5033 le16_to_cpu(sbi->s_es->s_mnt_count); 5034 } 5035 } 5036 } 5037 } 5038 if (!err) 5039 err = ext4_mark_iloc_dirty(handle, inode, &iloc); 5040 return err; 5041 } 5042 5043 /* 5044 * ext4_dirty_inode() is called from __mark_inode_dirty() 5045 * 5046 * We're really interested in the case where a file is being extended. 5047 * i_size has been changed by generic_commit_write() and we thus need 5048 * to include the updated inode in the current transaction. 5049 * 5050 * Also, dquot_alloc_block() will always dirty the inode when blocks 5051 * are allocated to the file. 5052 * 5053 * If the inode is marked synchronous, we don't honour that here - doing 5054 * so would cause a commit on atime updates, which we don't bother doing. 5055 * We handle synchronous inodes at the highest possible level. 5056 * 5057 * If only the I_DIRTY_TIME flag is set, we can skip everything. If 5058 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need 5059 * to copy into the on-disk inode structure are the timestamp files. 5060 */ 5061 void ext4_dirty_inode(struct inode *inode, int flags) 5062 { 5063 handle_t *handle; 5064 5065 if (flags == I_DIRTY_TIME) 5066 return; 5067 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); 5068 if (IS_ERR(handle)) 5069 goto out; 5070 5071 ext4_mark_inode_dirty(handle, inode); 5072 5073 ext4_journal_stop(handle); 5074 out: 5075 return; 5076 } 5077 5078 #if 0 5079 /* 5080 * Bind an inode's backing buffer_head into this transaction, to prevent 5081 * it from being flushed to disk early. Unlike 5082 * ext4_reserve_inode_write, this leaves behind no bh reference and 5083 * returns no iloc structure, so the caller needs to repeat the iloc 5084 * lookup to mark the inode dirty later. 5085 */ 5086 static int ext4_pin_inode(handle_t *handle, struct inode *inode) 5087 { 5088 struct ext4_iloc iloc; 5089 5090 int err = 0; 5091 if (handle) { 5092 err = ext4_get_inode_loc(inode, &iloc); 5093 if (!err) { 5094 BUFFER_TRACE(iloc.bh, "get_write_access"); 5095 err = jbd2_journal_get_write_access(handle, iloc.bh); 5096 if (!err) 5097 err = ext4_handle_dirty_metadata(handle, 5098 NULL, 5099 iloc.bh); 5100 brelse(iloc.bh); 5101 } 5102 } 5103 ext4_std_error(inode->i_sb, err); 5104 return err; 5105 } 5106 #endif 5107 5108 int ext4_change_inode_journal_flag(struct inode *inode, int val) 5109 { 5110 journal_t *journal; 5111 handle_t *handle; 5112 int err; 5113 5114 /* 5115 * We have to be very careful here: changing a data block's 5116 * journaling status dynamically is dangerous. If we write a 5117 * data block to the journal, change the status and then delete 5118 * that block, we risk forgetting to revoke the old log record 5119 * from the journal and so a subsequent replay can corrupt data. 5120 * So, first we make sure that the journal is empty and that 5121 * nobody is changing anything. 5122 */ 5123 5124 journal = EXT4_JOURNAL(inode); 5125 if (!journal) 5126 return 0; 5127 if (is_journal_aborted(journal)) 5128 return -EROFS; 5129 /* We have to allocate physical blocks for delalloc blocks 5130 * before flushing journal. otherwise delalloc blocks can not 5131 * be allocated any more. even more truncate on delalloc blocks 5132 * could trigger BUG by flushing delalloc blocks in journal. 5133 * There is no delalloc block in non-journal data mode. 5134 */ 5135 if (val && test_opt(inode->i_sb, DELALLOC)) { 5136 err = ext4_alloc_da_blocks(inode); 5137 if (err < 0) 5138 return err; 5139 } 5140 5141 /* Wait for all existing dio workers */ 5142 ext4_inode_block_unlocked_dio(inode); 5143 inode_dio_wait(inode); 5144 5145 jbd2_journal_lock_updates(journal); 5146 5147 /* 5148 * OK, there are no updates running now, and all cached data is 5149 * synced to disk. We are now in a completely consistent state 5150 * which doesn't have anything in the journal, and we know that 5151 * no filesystem updates are running, so it is safe to modify 5152 * the inode's in-core data-journaling state flag now. 5153 */ 5154 5155 if (val) 5156 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); 5157 else { 5158 err = jbd2_journal_flush(journal); 5159 if (err < 0) { 5160 jbd2_journal_unlock_updates(journal); 5161 ext4_inode_resume_unlocked_dio(inode); 5162 return err; 5163 } 5164 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); 5165 } 5166 ext4_set_aops(inode); 5167 5168 jbd2_journal_unlock_updates(journal); 5169 ext4_inode_resume_unlocked_dio(inode); 5170 5171 /* Finally we can mark the inode as dirty. */ 5172 5173 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); 5174 if (IS_ERR(handle)) 5175 return PTR_ERR(handle); 5176 5177 err = ext4_mark_inode_dirty(handle, inode); 5178 ext4_handle_sync(handle); 5179 ext4_journal_stop(handle); 5180 ext4_std_error(inode->i_sb, err); 5181 5182 return err; 5183 } 5184 5185 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh) 5186 { 5187 return !buffer_mapped(bh); 5188 } 5189 5190 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) 5191 { 5192 struct page *page = vmf->page; 5193 loff_t size; 5194 unsigned long len; 5195 int ret; 5196 struct file *file = vma->vm_file; 5197 struct inode *inode = file_inode(file); 5198 struct address_space *mapping = inode->i_mapping; 5199 handle_t *handle; 5200 get_block_t *get_block; 5201 int retries = 0; 5202 5203 sb_start_pagefault(inode->i_sb); 5204 file_update_time(vma->vm_file); 5205 /* Delalloc case is easy... */ 5206 if (test_opt(inode->i_sb, DELALLOC) && 5207 !ext4_should_journal_data(inode) && 5208 !ext4_nonda_switch(inode->i_sb)) { 5209 do { 5210 ret = __block_page_mkwrite(vma, vmf, 5211 ext4_da_get_block_prep); 5212 } while (ret == -ENOSPC && 5213 ext4_should_retry_alloc(inode->i_sb, &retries)); 5214 goto out_ret; 5215 } 5216 5217 lock_page(page); 5218 size = i_size_read(inode); 5219 /* Page got truncated from under us? */ 5220 if (page->mapping != mapping || page_offset(page) > size) { 5221 unlock_page(page); 5222 ret = VM_FAULT_NOPAGE; 5223 goto out; 5224 } 5225 5226 if (page->index == size >> PAGE_CACHE_SHIFT) 5227 len = size & ~PAGE_CACHE_MASK; 5228 else 5229 len = PAGE_CACHE_SIZE; 5230 /* 5231 * Return if we have all the buffers mapped. This avoids the need to do 5232 * journal_start/journal_stop which can block and take a long time 5233 */ 5234 if (page_has_buffers(page)) { 5235 if (!ext4_walk_page_buffers(NULL, page_buffers(page), 5236 0, len, NULL, 5237 ext4_bh_unmapped)) { 5238 /* Wait so that we don't change page under IO */ 5239 wait_for_stable_page(page); 5240 ret = VM_FAULT_LOCKED; 5241 goto out; 5242 } 5243 } 5244 unlock_page(page); 5245 /* OK, we need to fill the hole... */ 5246 if (ext4_should_dioread_nolock(inode)) 5247 get_block = ext4_get_block_write; 5248 else 5249 get_block = ext4_get_block; 5250 retry_alloc: 5251 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, 5252 ext4_writepage_trans_blocks(inode)); 5253 if (IS_ERR(handle)) { 5254 ret = VM_FAULT_SIGBUS; 5255 goto out; 5256 } 5257 ret = __block_page_mkwrite(vma, vmf, get_block); 5258 if (!ret && ext4_should_journal_data(inode)) { 5259 if (ext4_walk_page_buffers(handle, page_buffers(page), 0, 5260 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) { 5261 unlock_page(page); 5262 ret = VM_FAULT_SIGBUS; 5263 ext4_journal_stop(handle); 5264 goto out; 5265 } 5266 ext4_set_inode_state(inode, EXT4_STATE_JDATA); 5267 } 5268 ext4_journal_stop(handle); 5269 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) 5270 goto retry_alloc; 5271 out_ret: 5272 ret = block_page_mkwrite_return(ret); 5273 out: 5274 sb_end_pagefault(inode->i_sb); 5275 return ret; 5276 } 5277