| 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/dax.h> |
| 26 | #include <linux/quotaops.h> |
| 27 | #include <linux/string.h> |
| 28 | #include <linux/buffer_head.h> |
| 29 | #include <linux/writeback.h> |
| 30 | #include <linux/pagevec.h> |
| 31 | #include <linux/mpage.h> |
| 32 | #include <linux/namei.h> |
| 33 | #include <linux/uio.h> |
| 34 | #include <linux/bio.h> |
| 35 | #include <linux/workqueue.h> |
| 36 | #include <linux/kernel.h> |
| 37 | #include <linux/printk.h> |
| 38 | #include <linux/slab.h> |
| 39 | #include <linux/bitops.h> |
| 40 | #include <linux/iomap.h> |
| 41 | |
| 42 | #include "ext4_jbd2.h" |
| 43 | #include "xattr.h" |
| 44 | #include "acl.h" |
| 45 | #include "truncate.h" |
| 46 | |
| 47 | #include <trace/events/ext4.h> |
| 48 | |
| 49 | #define MPAGE_DA_EXTENT_TAIL 0x01 |
| 50 | |
| 51 | static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw, |
| 52 | struct ext4_inode_info *ei) |
| 53 | { |
| 54 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 55 | __u32 csum; |
| 56 | __u16 dummy_csum = 0; |
| 57 | int offset = offsetof(struct ext4_inode, i_checksum_lo); |
| 58 | unsigned int csum_size = sizeof(dummy_csum); |
| 59 | |
| 60 | csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset); |
| 61 | csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size); |
| 62 | offset += csum_size; |
| 63 | csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset, |
| 64 | EXT4_GOOD_OLD_INODE_SIZE - offset); |
| 65 | |
| 66 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
| 67 | offset = offsetof(struct ext4_inode, i_checksum_hi); |
| 68 | csum = ext4_chksum(sbi, csum, (__u8 *)raw + |
| 69 | EXT4_GOOD_OLD_INODE_SIZE, |
| 70 | offset - EXT4_GOOD_OLD_INODE_SIZE); |
| 71 | if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) { |
| 72 | csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, |
| 73 | csum_size); |
| 74 | offset += csum_size; |
| 75 | } |
| 76 | csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset, |
| 77 | EXT4_INODE_SIZE(inode->i_sb) - offset); |
| 78 | } |
| 79 | |
| 80 | return csum; |
| 81 | } |
| 82 | |
| 83 | static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw, |
| 84 | struct ext4_inode_info *ei) |
| 85 | { |
| 86 | __u32 provided, calculated; |
| 87 | |
| 88 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != |
| 89 | cpu_to_le32(EXT4_OS_LINUX) || |
| 90 | !ext4_has_metadata_csum(inode->i_sb)) |
| 91 | return 1; |
| 92 | |
| 93 | provided = le16_to_cpu(raw->i_checksum_lo); |
| 94 | calculated = ext4_inode_csum(inode, raw, ei); |
| 95 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && |
| 96 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) |
| 97 | provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16; |
| 98 | else |
| 99 | calculated &= 0xFFFF; |
| 100 | |
| 101 | return provided == calculated; |
| 102 | } |
| 103 | |
| 104 | static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw, |
| 105 | struct ext4_inode_info *ei) |
| 106 | { |
| 107 | __u32 csum; |
| 108 | |
| 109 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != |
| 110 | cpu_to_le32(EXT4_OS_LINUX) || |
| 111 | !ext4_has_metadata_csum(inode->i_sb)) |
| 112 | return; |
| 113 | |
| 114 | csum = ext4_inode_csum(inode, raw, ei); |
| 115 | raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF); |
| 116 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && |
| 117 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) |
| 118 | raw->i_checksum_hi = cpu_to_le16(csum >> 16); |
| 119 | } |
| 120 | |
| 121 | static inline int ext4_begin_ordered_truncate(struct inode *inode, |
| 122 | loff_t new_size) |
| 123 | { |
| 124 | trace_ext4_begin_ordered_truncate(inode, new_size); |
| 125 | /* |
| 126 | * If jinode is zero, then we never opened the file for |
| 127 | * writing, so there's no need to call |
| 128 | * jbd2_journal_begin_ordered_truncate() since there's no |
| 129 | * outstanding writes we need to flush. |
| 130 | */ |
| 131 | if (!EXT4_I(inode)->jinode) |
| 132 | return 0; |
| 133 | return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode), |
| 134 | EXT4_I(inode)->jinode, |
| 135 | new_size); |
| 136 | } |
| 137 | |
| 138 | static void ext4_invalidatepage(struct page *page, unsigned int offset, |
| 139 | unsigned int length); |
| 140 | static int __ext4_journalled_writepage(struct page *page, unsigned int len); |
| 141 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh); |
| 142 | |
| 143 | /* |
| 144 | * Test whether an inode is a fast symlink. |
| 145 | */ |
| 146 | int ext4_inode_is_fast_symlink(struct inode *inode) |
| 147 | { |
| 148 | int ea_blocks = EXT4_I(inode)->i_file_acl ? |
| 149 | EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0; |
| 150 | |
| 151 | if (ext4_has_inline_data(inode)) |
| 152 | return 0; |
| 153 | |
| 154 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); |
| 155 | } |
| 156 | |
| 157 | /* |
| 158 | * Restart the transaction associated with *handle. This does a commit, |
| 159 | * so before we call here everything must be consistently dirtied against |
| 160 | * this transaction. |
| 161 | */ |
| 162 | int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode, |
| 163 | int nblocks) |
| 164 | { |
| 165 | int ret; |
| 166 | |
| 167 | /* |
| 168 | * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this |
| 169 | * moment, get_block can be called only for blocks inside i_size since |
| 170 | * page cache has been already dropped and writes are blocked by |
| 171 | * i_mutex. So we can safely drop the i_data_sem here. |
| 172 | */ |
| 173 | BUG_ON(EXT4_JOURNAL(inode) == NULL); |
| 174 | jbd_debug(2, "restarting handle %p\n", handle); |
| 175 | up_write(&EXT4_I(inode)->i_data_sem); |
| 176 | ret = ext4_journal_restart(handle, nblocks); |
| 177 | down_write(&EXT4_I(inode)->i_data_sem); |
| 178 | ext4_discard_preallocations(inode); |
| 179 | |
| 180 | return ret; |
| 181 | } |
| 182 | |
| 183 | /* |
| 184 | * Called at the last iput() if i_nlink is zero. |
| 185 | */ |
| 186 | void ext4_evict_inode(struct inode *inode) |
| 187 | { |
| 188 | handle_t *handle; |
| 189 | int err; |
| 190 | int extra_credits = 3; |
| 191 | struct ext4_xattr_ino_array *lea_ino_array = NULL; |
| 192 | |
| 193 | trace_ext4_evict_inode(inode); |
| 194 | |
| 195 | if (inode->i_nlink) { |
| 196 | /* |
| 197 | * When journalling data dirty buffers are tracked only in the |
| 198 | * journal. So although mm thinks everything is clean and |
| 199 | * ready for reaping the inode might still have some pages to |
| 200 | * write in the running transaction or waiting to be |
| 201 | * checkpointed. Thus calling jbd2_journal_invalidatepage() |
| 202 | * (via truncate_inode_pages()) to discard these buffers can |
| 203 | * cause data loss. Also even if we did not discard these |
| 204 | * buffers, we would have no way to find them after the inode |
| 205 | * is reaped and thus user could see stale data if he tries to |
| 206 | * read them before the transaction is checkpointed. So be |
| 207 | * careful and force everything to disk here... We use |
| 208 | * ei->i_datasync_tid to store the newest transaction |
| 209 | * containing inode's data. |
| 210 | * |
| 211 | * Note that directories do not have this problem because they |
| 212 | * don't use page cache. |
| 213 | */ |
| 214 | if (inode->i_ino != EXT4_JOURNAL_INO && |
| 215 | ext4_should_journal_data(inode) && |
| 216 | (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) { |
| 217 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; |
| 218 | tid_t commit_tid = EXT4_I(inode)->i_datasync_tid; |
| 219 | |
| 220 | jbd2_complete_transaction(journal, commit_tid); |
| 221 | filemap_write_and_wait(&inode->i_data); |
| 222 | } |
| 223 | truncate_inode_pages_final(&inode->i_data); |
| 224 | |
| 225 | goto no_delete; |
| 226 | } |
| 227 | |
| 228 | if (is_bad_inode(inode)) |
| 229 | goto no_delete; |
| 230 | dquot_initialize(inode); |
| 231 | |
| 232 | if (ext4_should_order_data(inode)) |
| 233 | ext4_begin_ordered_truncate(inode, 0); |
| 234 | truncate_inode_pages_final(&inode->i_data); |
| 235 | |
| 236 | /* |
| 237 | * Protect us against freezing - iput() caller didn't have to have any |
| 238 | * protection against it |
| 239 | */ |
| 240 | sb_start_intwrite(inode->i_sb); |
| 241 | |
| 242 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, extra_credits); |
| 243 | if (IS_ERR(handle)) { |
| 244 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); |
| 245 | /* |
| 246 | * If we're going to skip the normal cleanup, we still need to |
| 247 | * make sure that the in-core orphan linked list is properly |
| 248 | * cleaned up. |
| 249 | */ |
| 250 | ext4_orphan_del(NULL, inode); |
| 251 | sb_end_intwrite(inode->i_sb); |
| 252 | goto no_delete; |
| 253 | } |
| 254 | if (IS_SYNC(inode)) |
| 255 | ext4_handle_sync(handle); |
| 256 | |
| 257 | /* |
| 258 | * Delete xattr inode before deleting the main inode. |
| 259 | */ |
| 260 | err = ext4_xattr_delete_inode(handle, inode, &lea_ino_array); |
| 261 | if (err) { |
| 262 | ext4_warning(inode->i_sb, |
| 263 | "couldn't delete inode's xattr (err %d)", err); |
| 264 | goto stop_handle; |
| 265 | } |
| 266 | |
| 267 | if (!IS_NOQUOTA(inode)) |
| 268 | extra_credits += 2 * EXT4_QUOTA_DEL_BLOCKS(inode->i_sb); |
| 269 | |
| 270 | if (!ext4_handle_has_enough_credits(handle, |
| 271 | ext4_blocks_for_truncate(inode) + extra_credits)) { |
| 272 | err = ext4_journal_extend(handle, |
| 273 | ext4_blocks_for_truncate(inode) + extra_credits); |
| 274 | if (err > 0) |
| 275 | err = ext4_journal_restart(handle, |
| 276 | ext4_blocks_for_truncate(inode) + extra_credits); |
| 277 | if (err != 0) { |
| 278 | ext4_warning(inode->i_sb, |
| 279 | "couldn't extend journal (err %d)", err); |
| 280 | goto stop_handle; |
| 281 | } |
| 282 | } |
| 283 | |
| 284 | inode->i_size = 0; |
| 285 | err = ext4_mark_inode_dirty(handle, inode); |
| 286 | if (err) { |
| 287 | ext4_warning(inode->i_sb, |
| 288 | "couldn't mark inode dirty (err %d)", err); |
| 289 | goto stop_handle; |
| 290 | } |
| 291 | if (inode->i_blocks) { |
| 292 | err = ext4_truncate(inode); |
| 293 | if (err) { |
| 294 | ext4_error(inode->i_sb, |
| 295 | "couldn't truncate inode %lu (err %d)", |
| 296 | inode->i_ino, err); |
| 297 | goto stop_handle; |
| 298 | } |
| 299 | } |
| 300 | |
| 301 | /* |
| 302 | * ext4_ext_truncate() doesn't reserve any slop when it |
| 303 | * restarts journal transactions; therefore there may not be |
| 304 | * enough credits left in the handle to remove the inode from |
| 305 | * the orphan list and set the dtime field. |
| 306 | */ |
| 307 | if (!ext4_handle_has_enough_credits(handle, extra_credits)) { |
| 308 | err = ext4_journal_extend(handle, extra_credits); |
| 309 | if (err > 0) |
| 310 | err = ext4_journal_restart(handle, extra_credits); |
| 311 | if (err != 0) { |
| 312 | ext4_warning(inode->i_sb, |
| 313 | "couldn't extend journal (err %d)", err); |
| 314 | stop_handle: |
| 315 | ext4_journal_stop(handle); |
| 316 | ext4_orphan_del(NULL, inode); |
| 317 | sb_end_intwrite(inode->i_sb); |
| 318 | goto no_delete; |
| 319 | } |
| 320 | } |
| 321 | |
| 322 | /* |
| 323 | * Kill off the orphan record which ext4_truncate created. |
| 324 | * AKPM: I think this can be inside the above `if'. |
| 325 | * Note that ext4_orphan_del() has to be able to cope with the |
| 326 | * deletion of a non-existent orphan - this is because we don't |
| 327 | * know if ext4_truncate() actually created an orphan record. |
| 328 | * (Well, we could do this if we need to, but heck - it works) |
| 329 | */ |
| 330 | ext4_orphan_del(handle, inode); |
| 331 | EXT4_I(inode)->i_dtime = get_seconds(); |
| 332 | |
| 333 | /* |
| 334 | * One subtle ordering requirement: if anything has gone wrong |
| 335 | * (transaction abort, IO errors, whatever), then we can still |
| 336 | * do these next steps (the fs will already have been marked as |
| 337 | * having errors), but we can't free the inode if the mark_dirty |
| 338 | * fails. |
| 339 | */ |
| 340 | if (ext4_mark_inode_dirty(handle, inode)) |
| 341 | /* If that failed, just do the required in-core inode clear. */ |
| 342 | ext4_clear_inode(inode); |
| 343 | else |
| 344 | ext4_free_inode(handle, inode); |
| 345 | |
| 346 | ext4_journal_stop(handle); |
| 347 | sb_end_intwrite(inode->i_sb); |
| 348 | |
| 349 | if (lea_ino_array != NULL) |
| 350 | ext4_xattr_inode_array_free(inode, lea_ino_array); |
| 351 | return; |
| 352 | no_delete: |
| 353 | ext4_clear_inode(inode); /* We must guarantee clearing of inode... */ |
| 354 | } |
| 355 | |
| 356 | #ifdef CONFIG_QUOTA |
| 357 | qsize_t *ext4_get_reserved_space(struct inode *inode) |
| 358 | { |
| 359 | return &EXT4_I(inode)->i_reserved_quota; |
| 360 | } |
| 361 | #endif |
| 362 | |
| 363 | /* |
| 364 | * Called with i_data_sem down, which is important since we can call |
| 365 | * ext4_discard_preallocations() from here. |
| 366 | */ |
| 367 | void ext4_da_update_reserve_space(struct inode *inode, |
| 368 | int used, int quota_claim) |
| 369 | { |
| 370 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 371 | struct ext4_inode_info *ei = EXT4_I(inode); |
| 372 | |
| 373 | spin_lock(&ei->i_block_reservation_lock); |
| 374 | trace_ext4_da_update_reserve_space(inode, used, quota_claim); |
| 375 | if (unlikely(used > ei->i_reserved_data_blocks)) { |
| 376 | ext4_warning(inode->i_sb, "%s: ino %lu, used %d " |
| 377 | "with only %d reserved data blocks", |
| 378 | __func__, inode->i_ino, used, |
| 379 | ei->i_reserved_data_blocks); |
| 380 | WARN_ON(1); |
| 381 | used = ei->i_reserved_data_blocks; |
| 382 | } |
| 383 | |
| 384 | /* Update per-inode reservations */ |
| 385 | ei->i_reserved_data_blocks -= used; |
| 386 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, used); |
| 387 | |
| 388 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
| 389 | |
| 390 | /* Update quota subsystem for data blocks */ |
| 391 | if (quota_claim) |
| 392 | dquot_claim_block(inode, EXT4_C2B(sbi, used)); |
| 393 | else { |
| 394 | /* |
| 395 | * We did fallocate with an offset that is already delayed |
| 396 | * allocated. So on delayed allocated writeback we should |
| 397 | * not re-claim the quota for fallocated blocks. |
| 398 | */ |
| 399 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, used)); |
| 400 | } |
| 401 | |
| 402 | /* |
| 403 | * If we have done all the pending block allocations and if |
| 404 | * there aren't any writers on the inode, we can discard the |
| 405 | * inode's preallocations. |
| 406 | */ |
| 407 | if ((ei->i_reserved_data_blocks == 0) && |
| 408 | (atomic_read(&inode->i_writecount) == 0)) |
| 409 | ext4_discard_preallocations(inode); |
| 410 | } |
| 411 | |
| 412 | static int __check_block_validity(struct inode *inode, const char *func, |
| 413 | unsigned int line, |
| 414 | struct ext4_map_blocks *map) |
| 415 | { |
| 416 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk, |
| 417 | map->m_len)) { |
| 418 | ext4_error_inode(inode, func, line, map->m_pblk, |
| 419 | "lblock %lu mapped to illegal pblock " |
| 420 | "(length %d)", (unsigned long) map->m_lblk, |
| 421 | map->m_len); |
| 422 | return -EFSCORRUPTED; |
| 423 | } |
| 424 | return 0; |
| 425 | } |
| 426 | |
| 427 | int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk, |
| 428 | ext4_lblk_t len) |
| 429 | { |
| 430 | int ret; |
| 431 | |
| 432 | if (ext4_encrypted_inode(inode)) |
| 433 | return fscrypt_zeroout_range(inode, lblk, pblk, len); |
| 434 | |
| 435 | ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS); |
| 436 | if (ret > 0) |
| 437 | ret = 0; |
| 438 | |
| 439 | return ret; |
| 440 | } |
| 441 | |
| 442 | #define check_block_validity(inode, map) \ |
| 443 | __check_block_validity((inode), __func__, __LINE__, (map)) |
| 444 | |
| 445 | #ifdef ES_AGGRESSIVE_TEST |
| 446 | static void ext4_map_blocks_es_recheck(handle_t *handle, |
| 447 | struct inode *inode, |
| 448 | struct ext4_map_blocks *es_map, |
| 449 | struct ext4_map_blocks *map, |
| 450 | int flags) |
| 451 | { |
| 452 | int retval; |
| 453 | |
| 454 | map->m_flags = 0; |
| 455 | /* |
| 456 | * There is a race window that the result is not the same. |
| 457 | * e.g. xfstests #223 when dioread_nolock enables. The reason |
| 458 | * is that we lookup a block mapping in extent status tree with |
| 459 | * out taking i_data_sem. So at the time the unwritten extent |
| 460 | * could be converted. |
| 461 | */ |
| 462 | down_read(&EXT4_I(inode)->i_data_sem); |
| 463 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
| 464 | retval = ext4_ext_map_blocks(handle, inode, map, flags & |
| 465 | EXT4_GET_BLOCKS_KEEP_SIZE); |
| 466 | } else { |
| 467 | retval = ext4_ind_map_blocks(handle, inode, map, flags & |
| 468 | EXT4_GET_BLOCKS_KEEP_SIZE); |
| 469 | } |
| 470 | up_read((&EXT4_I(inode)->i_data_sem)); |
| 471 | |
| 472 | /* |
| 473 | * We don't check m_len because extent will be collpased in status |
| 474 | * tree. So the m_len might not equal. |
| 475 | */ |
| 476 | if (es_map->m_lblk != map->m_lblk || |
| 477 | es_map->m_flags != map->m_flags || |
| 478 | es_map->m_pblk != map->m_pblk) { |
| 479 | printk("ES cache assertion failed for inode: %lu " |
| 480 | "es_cached ex [%d/%d/%llu/%x] != " |
| 481 | "found ex [%d/%d/%llu/%x] retval %d flags %x\n", |
| 482 | inode->i_ino, es_map->m_lblk, es_map->m_len, |
| 483 | es_map->m_pblk, es_map->m_flags, map->m_lblk, |
| 484 | map->m_len, map->m_pblk, map->m_flags, |
| 485 | retval, flags); |
| 486 | } |
| 487 | } |
| 488 | #endif /* ES_AGGRESSIVE_TEST */ |
| 489 | |
| 490 | /* |
| 491 | * The ext4_map_blocks() function tries to look up the requested blocks, |
| 492 | * and returns if the blocks are already mapped. |
| 493 | * |
| 494 | * Otherwise it takes the write lock of the i_data_sem and allocate blocks |
| 495 | * and store the allocated blocks in the result buffer head and mark it |
| 496 | * mapped. |
| 497 | * |
| 498 | * If file type is extents based, it will call ext4_ext_map_blocks(), |
| 499 | * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping |
| 500 | * based files |
| 501 | * |
| 502 | * On success, it returns the number of blocks being mapped or allocated. if |
| 503 | * create==0 and the blocks are pre-allocated and unwritten, the resulting @map |
| 504 | * is marked as unwritten. If the create == 1, it will mark @map as mapped. |
| 505 | * |
| 506 | * It returns 0 if plain look up failed (blocks have not been allocated), in |
| 507 | * that case, @map is returned as unmapped but we still do fill map->m_len to |
| 508 | * indicate the length of a hole starting at map->m_lblk. |
| 509 | * |
| 510 | * It returns the error in case of allocation failure. |
| 511 | */ |
| 512 | int ext4_map_blocks(handle_t *handle, struct inode *inode, |
| 513 | struct ext4_map_blocks *map, int flags) |
| 514 | { |
| 515 | struct extent_status es; |
| 516 | int retval; |
| 517 | int ret = 0; |
| 518 | #ifdef ES_AGGRESSIVE_TEST |
| 519 | struct ext4_map_blocks orig_map; |
| 520 | |
| 521 | memcpy(&orig_map, map, sizeof(*map)); |
| 522 | #endif |
| 523 | |
| 524 | map->m_flags = 0; |
| 525 | ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u," |
| 526 | "logical block %lu\n", inode->i_ino, flags, map->m_len, |
| 527 | (unsigned long) map->m_lblk); |
| 528 | |
| 529 | /* |
| 530 | * ext4_map_blocks returns an int, and m_len is an unsigned int |
| 531 | */ |
| 532 | if (unlikely(map->m_len > INT_MAX)) |
| 533 | map->m_len = INT_MAX; |
| 534 | |
| 535 | /* We can handle the block number less than EXT_MAX_BLOCKS */ |
| 536 | if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS)) |
| 537 | return -EFSCORRUPTED; |
| 538 | |
| 539 | /* Lookup extent status tree firstly */ |
| 540 | if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) { |
| 541 | if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) { |
| 542 | map->m_pblk = ext4_es_pblock(&es) + |
| 543 | map->m_lblk - es.es_lblk; |
| 544 | map->m_flags |= ext4_es_is_written(&es) ? |
| 545 | EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN; |
| 546 | retval = es.es_len - (map->m_lblk - es.es_lblk); |
| 547 | if (retval > map->m_len) |
| 548 | retval = map->m_len; |
| 549 | map->m_len = retval; |
| 550 | } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) { |
| 551 | map->m_pblk = 0; |
| 552 | retval = es.es_len - (map->m_lblk - es.es_lblk); |
| 553 | if (retval > map->m_len) |
| 554 | retval = map->m_len; |
| 555 | map->m_len = retval; |
| 556 | retval = 0; |
| 557 | } else { |
| 558 | BUG_ON(1); |
| 559 | } |
| 560 | #ifdef ES_AGGRESSIVE_TEST |
| 561 | ext4_map_blocks_es_recheck(handle, inode, map, |
| 562 | &orig_map, flags); |
| 563 | #endif |
| 564 | goto found; |
| 565 | } |
| 566 | |
| 567 | /* |
| 568 | * Try to see if we can get the block without requesting a new |
| 569 | * file system block. |
| 570 | */ |
| 571 | down_read(&EXT4_I(inode)->i_data_sem); |
| 572 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
| 573 | retval = ext4_ext_map_blocks(handle, inode, map, flags & |
| 574 | EXT4_GET_BLOCKS_KEEP_SIZE); |
| 575 | } else { |
| 576 | retval = ext4_ind_map_blocks(handle, inode, map, flags & |
| 577 | EXT4_GET_BLOCKS_KEEP_SIZE); |
| 578 | } |
| 579 | if (retval > 0) { |
| 580 | unsigned int status; |
| 581 | |
| 582 | if (unlikely(retval != map->m_len)) { |
| 583 | ext4_warning(inode->i_sb, |
| 584 | "ES len assertion failed for inode " |
| 585 | "%lu: retval %d != map->m_len %d", |
| 586 | inode->i_ino, retval, map->m_len); |
| 587 | WARN_ON(1); |
| 588 | } |
| 589 | |
| 590 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
| 591 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; |
| 592 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && |
| 593 | !(status & EXTENT_STATUS_WRITTEN) && |
| 594 | ext4_find_delalloc_range(inode, map->m_lblk, |
| 595 | map->m_lblk + map->m_len - 1)) |
| 596 | status |= EXTENT_STATUS_DELAYED; |
| 597 | ret = ext4_es_insert_extent(inode, map->m_lblk, |
| 598 | map->m_len, map->m_pblk, status); |
| 599 | if (ret < 0) |
| 600 | retval = ret; |
| 601 | } |
| 602 | up_read((&EXT4_I(inode)->i_data_sem)); |
| 603 | |
| 604 | found: |
| 605 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
| 606 | ret = check_block_validity(inode, map); |
| 607 | if (ret != 0) |
| 608 | return ret; |
| 609 | } |
| 610 | |
| 611 | /* If it is only a block(s) look up */ |
| 612 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) |
| 613 | return retval; |
| 614 | |
| 615 | /* |
| 616 | * Returns if the blocks have already allocated |
| 617 | * |
| 618 | * Note that if blocks have been preallocated |
| 619 | * ext4_ext_get_block() returns the create = 0 |
| 620 | * with buffer head unmapped. |
| 621 | */ |
| 622 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) |
| 623 | /* |
| 624 | * If we need to convert extent to unwritten |
| 625 | * we continue and do the actual work in |
| 626 | * ext4_ext_map_blocks() |
| 627 | */ |
| 628 | if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) |
| 629 | return retval; |
| 630 | |
| 631 | /* |
| 632 | * Here we clear m_flags because after allocating an new extent, |
| 633 | * it will be set again. |
| 634 | */ |
| 635 | map->m_flags &= ~EXT4_MAP_FLAGS; |
| 636 | |
| 637 | /* |
| 638 | * New blocks allocate and/or writing to unwritten extent |
| 639 | * will possibly result in updating i_data, so we take |
| 640 | * the write lock of i_data_sem, and call get_block() |
| 641 | * with create == 1 flag. |
| 642 | */ |
| 643 | down_write(&EXT4_I(inode)->i_data_sem); |
| 644 | |
| 645 | /* |
| 646 | * We need to check for EXT4 here because migrate |
| 647 | * could have changed the inode type in between |
| 648 | */ |
| 649 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
| 650 | retval = ext4_ext_map_blocks(handle, inode, map, flags); |
| 651 | } else { |
| 652 | retval = ext4_ind_map_blocks(handle, inode, map, flags); |
| 653 | |
| 654 | if (retval > 0 && map->m_flags & EXT4_MAP_NEW) { |
| 655 | /* |
| 656 | * We allocated new blocks which will result in |
| 657 | * i_data's format changing. Force the migrate |
| 658 | * to fail by clearing migrate flags |
| 659 | */ |
| 660 | ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE); |
| 661 | } |
| 662 | |
| 663 | /* |
| 664 | * Update reserved blocks/metadata blocks after successful |
| 665 | * block allocation which had been deferred till now. We don't |
| 666 | * support fallocate for non extent files. So we can update |
| 667 | * reserve space here. |
| 668 | */ |
| 669 | if ((retval > 0) && |
| 670 | (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)) |
| 671 | ext4_da_update_reserve_space(inode, retval, 1); |
| 672 | } |
| 673 | |
| 674 | if (retval > 0) { |
| 675 | unsigned int status; |
| 676 | |
| 677 | if (unlikely(retval != map->m_len)) { |
| 678 | ext4_warning(inode->i_sb, |
| 679 | "ES len assertion failed for inode " |
| 680 | "%lu: retval %d != map->m_len %d", |
| 681 | inode->i_ino, retval, map->m_len); |
| 682 | WARN_ON(1); |
| 683 | } |
| 684 | |
| 685 | /* |
| 686 | * We have to zeroout blocks before inserting them into extent |
| 687 | * status tree. Otherwise someone could look them up there and |
| 688 | * use them before they are really zeroed. We also have to |
| 689 | * unmap metadata before zeroing as otherwise writeback can |
| 690 | * overwrite zeros with stale data from block device. |
| 691 | */ |
| 692 | if (flags & EXT4_GET_BLOCKS_ZERO && |
| 693 | map->m_flags & EXT4_MAP_MAPPED && |
| 694 | map->m_flags & EXT4_MAP_NEW) { |
| 695 | clean_bdev_aliases(inode->i_sb->s_bdev, map->m_pblk, |
| 696 | map->m_len); |
| 697 | ret = ext4_issue_zeroout(inode, map->m_lblk, |
| 698 | map->m_pblk, map->m_len); |
| 699 | if (ret) { |
| 700 | retval = ret; |
| 701 | goto out_sem; |
| 702 | } |
| 703 | } |
| 704 | |
| 705 | /* |
| 706 | * If the extent has been zeroed out, we don't need to update |
| 707 | * extent status tree. |
| 708 | */ |
| 709 | if ((flags & EXT4_GET_BLOCKS_PRE_IO) && |
| 710 | ext4_es_lookup_extent(inode, map->m_lblk, &es)) { |
| 711 | if (ext4_es_is_written(&es)) |
| 712 | goto out_sem; |
| 713 | } |
| 714 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
| 715 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; |
| 716 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && |
| 717 | !(status & EXTENT_STATUS_WRITTEN) && |
| 718 | ext4_find_delalloc_range(inode, map->m_lblk, |
| 719 | map->m_lblk + map->m_len - 1)) |
| 720 | status |= EXTENT_STATUS_DELAYED; |
| 721 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, |
| 722 | map->m_pblk, status); |
| 723 | if (ret < 0) { |
| 724 | retval = ret; |
| 725 | goto out_sem; |
| 726 | } |
| 727 | } |
| 728 | |
| 729 | out_sem: |
| 730 | up_write((&EXT4_I(inode)->i_data_sem)); |
| 731 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
| 732 | ret = check_block_validity(inode, map); |
| 733 | if (ret != 0) |
| 734 | return ret; |
| 735 | |
| 736 | /* |
| 737 | * Inodes with freshly allocated blocks where contents will be |
| 738 | * visible after transaction commit must be on transaction's |
| 739 | * ordered data list. |
| 740 | */ |
| 741 | if (map->m_flags & EXT4_MAP_NEW && |
| 742 | !(map->m_flags & EXT4_MAP_UNWRITTEN) && |
| 743 | !(flags & EXT4_GET_BLOCKS_ZERO) && |
| 744 | !IS_NOQUOTA(inode) && |
| 745 | ext4_should_order_data(inode)) { |
| 746 | if (flags & EXT4_GET_BLOCKS_IO_SUBMIT) |
| 747 | ret = ext4_jbd2_inode_add_wait(handle, inode); |
| 748 | else |
| 749 | ret = ext4_jbd2_inode_add_write(handle, inode); |
| 750 | if (ret) |
| 751 | return ret; |
| 752 | } |
| 753 | } |
| 754 | return retval; |
| 755 | } |
| 756 | |
| 757 | /* |
| 758 | * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages |
| 759 | * we have to be careful as someone else may be manipulating b_state as well. |
| 760 | */ |
| 761 | static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags) |
| 762 | { |
| 763 | unsigned long old_state; |
| 764 | unsigned long new_state; |
| 765 | |
| 766 | flags &= EXT4_MAP_FLAGS; |
| 767 | |
| 768 | /* Dummy buffer_head? Set non-atomically. */ |
| 769 | if (!bh->b_page) { |
| 770 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags; |
| 771 | return; |
| 772 | } |
| 773 | /* |
| 774 | * Someone else may be modifying b_state. Be careful! This is ugly but |
| 775 | * once we get rid of using bh as a container for mapping information |
| 776 | * to pass to / from get_block functions, this can go away. |
| 777 | */ |
| 778 | do { |
| 779 | old_state = READ_ONCE(bh->b_state); |
| 780 | new_state = (old_state & ~EXT4_MAP_FLAGS) | flags; |
| 781 | } while (unlikely( |
| 782 | cmpxchg(&bh->b_state, old_state, new_state) != old_state)); |
| 783 | } |
| 784 | |
| 785 | static int _ext4_get_block(struct inode *inode, sector_t iblock, |
| 786 | struct buffer_head *bh, int flags) |
| 787 | { |
| 788 | struct ext4_map_blocks map; |
| 789 | int ret = 0; |
| 790 | |
| 791 | if (ext4_has_inline_data(inode)) |
| 792 | return -ERANGE; |
| 793 | |
| 794 | map.m_lblk = iblock; |
| 795 | map.m_len = bh->b_size >> inode->i_blkbits; |
| 796 | |
| 797 | ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map, |
| 798 | flags); |
| 799 | if (ret > 0) { |
| 800 | map_bh(bh, inode->i_sb, map.m_pblk); |
| 801 | ext4_update_bh_state(bh, map.m_flags); |
| 802 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; |
| 803 | ret = 0; |
| 804 | } else if (ret == 0) { |
| 805 | /* hole case, need to fill in bh->b_size */ |
| 806 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; |
| 807 | } |
| 808 | return ret; |
| 809 | } |
| 810 | |
| 811 | int ext4_get_block(struct inode *inode, sector_t iblock, |
| 812 | struct buffer_head *bh, int create) |
| 813 | { |
| 814 | return _ext4_get_block(inode, iblock, bh, |
| 815 | create ? EXT4_GET_BLOCKS_CREATE : 0); |
| 816 | } |
| 817 | |
| 818 | /* |
| 819 | * Get block function used when preparing for buffered write if we require |
| 820 | * creating an unwritten extent if blocks haven't been allocated. The extent |
| 821 | * will be converted to written after the IO is complete. |
| 822 | */ |
| 823 | int ext4_get_block_unwritten(struct inode *inode, sector_t iblock, |
| 824 | struct buffer_head *bh_result, int create) |
| 825 | { |
| 826 | ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n", |
| 827 | inode->i_ino, create); |
| 828 | return _ext4_get_block(inode, iblock, bh_result, |
| 829 | EXT4_GET_BLOCKS_IO_CREATE_EXT); |
| 830 | } |
| 831 | |
| 832 | /* Maximum number of blocks we map for direct IO at once. */ |
| 833 | #define DIO_MAX_BLOCKS 4096 |
| 834 | |
| 835 | /* |
| 836 | * Get blocks function for the cases that need to start a transaction - |
| 837 | * generally difference cases of direct IO and DAX IO. It also handles retries |
| 838 | * in case of ENOSPC. |
| 839 | */ |
| 840 | static int ext4_get_block_trans(struct inode *inode, sector_t iblock, |
| 841 | struct buffer_head *bh_result, int flags) |
| 842 | { |
| 843 | int dio_credits; |
| 844 | handle_t *handle; |
| 845 | int retries = 0; |
| 846 | int ret; |
| 847 | |
| 848 | /* Trim mapping request to maximum we can map at once for DIO */ |
| 849 | if (bh_result->b_size >> inode->i_blkbits > DIO_MAX_BLOCKS) |
| 850 | bh_result->b_size = DIO_MAX_BLOCKS << inode->i_blkbits; |
| 851 | dio_credits = ext4_chunk_trans_blocks(inode, |
| 852 | bh_result->b_size >> inode->i_blkbits); |
| 853 | retry: |
| 854 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits); |
| 855 | if (IS_ERR(handle)) |
| 856 | return PTR_ERR(handle); |
| 857 | |
| 858 | ret = _ext4_get_block(inode, iblock, bh_result, flags); |
| 859 | ext4_journal_stop(handle); |
| 860 | |
| 861 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
| 862 | goto retry; |
| 863 | return ret; |
| 864 | } |
| 865 | |
| 866 | /* Get block function for DIO reads and writes to inodes without extents */ |
| 867 | int ext4_dio_get_block(struct inode *inode, sector_t iblock, |
| 868 | struct buffer_head *bh, int create) |
| 869 | { |
| 870 | /* We don't expect handle for direct IO */ |
| 871 | WARN_ON_ONCE(ext4_journal_current_handle()); |
| 872 | |
| 873 | if (!create) |
| 874 | return _ext4_get_block(inode, iblock, bh, 0); |
| 875 | return ext4_get_block_trans(inode, iblock, bh, EXT4_GET_BLOCKS_CREATE); |
| 876 | } |
| 877 | |
| 878 | /* |
| 879 | * Get block function for AIO DIO writes when we create unwritten extent if |
| 880 | * blocks are not allocated yet. The extent will be converted to written |
| 881 | * after IO is complete. |
| 882 | */ |
| 883 | static int ext4_dio_get_block_unwritten_async(struct inode *inode, |
| 884 | sector_t iblock, struct buffer_head *bh_result, int create) |
| 885 | { |
| 886 | int ret; |
| 887 | |
| 888 | /* We don't expect handle for direct IO */ |
| 889 | WARN_ON_ONCE(ext4_journal_current_handle()); |
| 890 | |
| 891 | ret = ext4_get_block_trans(inode, iblock, bh_result, |
| 892 | EXT4_GET_BLOCKS_IO_CREATE_EXT); |
| 893 | |
| 894 | /* |
| 895 | * When doing DIO using unwritten extents, we need io_end to convert |
| 896 | * unwritten extents to written on IO completion. We allocate io_end |
| 897 | * once we spot unwritten extent and store it in b_private. Generic |
| 898 | * DIO code keeps b_private set and furthermore passes the value to |
| 899 | * our completion callback in 'private' argument. |
| 900 | */ |
| 901 | if (!ret && buffer_unwritten(bh_result)) { |
| 902 | if (!bh_result->b_private) { |
| 903 | ext4_io_end_t *io_end; |
| 904 | |
| 905 | io_end = ext4_init_io_end(inode, GFP_KERNEL); |
| 906 | if (!io_end) |
| 907 | return -ENOMEM; |
| 908 | bh_result->b_private = io_end; |
| 909 | ext4_set_io_unwritten_flag(inode, io_end); |
| 910 | } |
| 911 | set_buffer_defer_completion(bh_result); |
| 912 | } |
| 913 | |
| 914 | return ret; |
| 915 | } |
| 916 | |
| 917 | /* |
| 918 | * Get block function for non-AIO DIO writes when we create unwritten extent if |
| 919 | * blocks are not allocated yet. The extent will be converted to written |
| 920 | * after IO is complete from ext4_ext_direct_IO() function. |
| 921 | */ |
| 922 | static int ext4_dio_get_block_unwritten_sync(struct inode *inode, |
| 923 | sector_t iblock, struct buffer_head *bh_result, int create) |
| 924 | { |
| 925 | int ret; |
| 926 | |
| 927 | /* We don't expect handle for direct IO */ |
| 928 | WARN_ON_ONCE(ext4_journal_current_handle()); |
| 929 | |
| 930 | ret = ext4_get_block_trans(inode, iblock, bh_result, |
| 931 | EXT4_GET_BLOCKS_IO_CREATE_EXT); |
| 932 | |
| 933 | /* |
| 934 | * Mark inode as having pending DIO writes to unwritten extents. |
| 935 | * ext4_ext_direct_IO() checks this flag and converts extents to |
| 936 | * written. |
| 937 | */ |
| 938 | if (!ret && buffer_unwritten(bh_result)) |
| 939 | ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); |
| 940 | |
| 941 | return ret; |
| 942 | } |
| 943 | |
| 944 | static int ext4_dio_get_block_overwrite(struct inode *inode, sector_t iblock, |
| 945 | struct buffer_head *bh_result, int create) |
| 946 | { |
| 947 | int ret; |
| 948 | |
| 949 | ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n", |
| 950 | inode->i_ino, create); |
| 951 | /* We don't expect handle for direct IO */ |
| 952 | WARN_ON_ONCE(ext4_journal_current_handle()); |
| 953 | |
| 954 | ret = _ext4_get_block(inode, iblock, bh_result, 0); |
| 955 | /* |
| 956 | * Blocks should have been preallocated! ext4_file_write_iter() checks |
| 957 | * that. |
| 958 | */ |
| 959 | WARN_ON_ONCE(!buffer_mapped(bh_result) || buffer_unwritten(bh_result)); |
| 960 | |
| 961 | return ret; |
| 962 | } |
| 963 | |
| 964 | |
| 965 | /* |
| 966 | * `handle' can be NULL if create is zero |
| 967 | */ |
| 968 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
| 969 | ext4_lblk_t block, int map_flags) |
| 970 | { |
| 971 | struct ext4_map_blocks map; |
| 972 | struct buffer_head *bh; |
| 973 | int create = map_flags & EXT4_GET_BLOCKS_CREATE; |
| 974 | int err; |
| 975 | |
| 976 | J_ASSERT(handle != NULL || create == 0); |
| 977 | |
| 978 | map.m_lblk = block; |
| 979 | map.m_len = 1; |
| 980 | err = ext4_map_blocks(handle, inode, &map, map_flags); |
| 981 | |
| 982 | if (err == 0) |
| 983 | return create ? ERR_PTR(-ENOSPC) : NULL; |
| 984 | if (err < 0) |
| 985 | return ERR_PTR(err); |
| 986 | |
| 987 | bh = sb_getblk(inode->i_sb, map.m_pblk); |
| 988 | if (unlikely(!bh)) |
| 989 | return ERR_PTR(-ENOMEM); |
| 990 | if (map.m_flags & EXT4_MAP_NEW) { |
| 991 | J_ASSERT(create != 0); |
| 992 | J_ASSERT(handle != NULL); |
| 993 | |
| 994 | /* |
| 995 | * Now that we do not always journal data, we should |
| 996 | * keep in mind whether this should always journal the |
| 997 | * new buffer as metadata. For now, regular file |
| 998 | * writes use ext4_get_block instead, so it's not a |
| 999 | * problem. |
| 1000 | */ |
| 1001 | lock_buffer(bh); |
| 1002 | BUFFER_TRACE(bh, "call get_create_access"); |
| 1003 | err = ext4_journal_get_create_access(handle, bh); |
| 1004 | if (unlikely(err)) { |
| 1005 | unlock_buffer(bh); |
| 1006 | goto errout; |
| 1007 | } |
| 1008 | if (!buffer_uptodate(bh)) { |
| 1009 | memset(bh->b_data, 0, inode->i_sb->s_blocksize); |
| 1010 | set_buffer_uptodate(bh); |
| 1011 | } |
| 1012 | unlock_buffer(bh); |
| 1013 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
| 1014 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
| 1015 | if (unlikely(err)) |
| 1016 | goto errout; |
| 1017 | } else |
| 1018 | BUFFER_TRACE(bh, "not a new buffer"); |
| 1019 | return bh; |
| 1020 | errout: |
| 1021 | brelse(bh); |
| 1022 | return ERR_PTR(err); |
| 1023 | } |
| 1024 | |
| 1025 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
| 1026 | ext4_lblk_t block, int map_flags) |
| 1027 | { |
| 1028 | struct buffer_head *bh; |
| 1029 | |
| 1030 | bh = ext4_getblk(handle, inode, block, map_flags); |
| 1031 | if (IS_ERR(bh)) |
| 1032 | return bh; |
| 1033 | if (!bh || buffer_uptodate(bh)) |
| 1034 | return bh; |
| 1035 | ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh); |
| 1036 | wait_on_buffer(bh); |
| 1037 | if (buffer_uptodate(bh)) |
| 1038 | return bh; |
| 1039 | put_bh(bh); |
| 1040 | return ERR_PTR(-EIO); |
| 1041 | } |
| 1042 | |
| 1043 | int ext4_walk_page_buffers(handle_t *handle, |
| 1044 | struct buffer_head *head, |
| 1045 | unsigned from, |
| 1046 | unsigned to, |
| 1047 | int *partial, |
| 1048 | int (*fn)(handle_t *handle, |
| 1049 | struct buffer_head *bh)) |
| 1050 | { |
| 1051 | struct buffer_head *bh; |
| 1052 | unsigned block_start, block_end; |
| 1053 | unsigned blocksize = head->b_size; |
| 1054 | int err, ret = 0; |
| 1055 | struct buffer_head *next; |
| 1056 | |
| 1057 | for (bh = head, block_start = 0; |
| 1058 | ret == 0 && (bh != head || !block_start); |
| 1059 | block_start = block_end, bh = next) { |
| 1060 | next = bh->b_this_page; |
| 1061 | block_end = block_start + blocksize; |
| 1062 | if (block_end <= from || block_start >= to) { |
| 1063 | if (partial && !buffer_uptodate(bh)) |
| 1064 | *partial = 1; |
| 1065 | continue; |
| 1066 | } |
| 1067 | err = (*fn)(handle, bh); |
| 1068 | if (!ret) |
| 1069 | ret = err; |
| 1070 | } |
| 1071 | return ret; |
| 1072 | } |
| 1073 | |
| 1074 | /* |
| 1075 | * To preserve ordering, it is essential that the hole instantiation and |
| 1076 | * the data write be encapsulated in a single transaction. We cannot |
| 1077 | * close off a transaction and start a new one between the ext4_get_block() |
| 1078 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
| 1079 | * prepare_write() is the right place. |
| 1080 | * |
| 1081 | * Also, this function can nest inside ext4_writepage(). In that case, we |
| 1082 | * *know* that ext4_writepage() has generated enough buffer credits to do the |
| 1083 | * whole page. So we won't block on the journal in that case, which is good, |
| 1084 | * because the caller may be PF_MEMALLOC. |
| 1085 | * |
| 1086 | * By accident, ext4 can be reentered when a transaction is open via |
| 1087 | * quota file writes. If we were to commit the transaction while thus |
| 1088 | * reentered, there can be a deadlock - we would be holding a quota |
| 1089 | * lock, and the commit would never complete if another thread had a |
| 1090 | * transaction open and was blocking on the quota lock - a ranking |
| 1091 | * violation. |
| 1092 | * |
| 1093 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
| 1094 | * will _not_ run commit under these circumstances because handle->h_ref |
| 1095 | * is elevated. We'll still have enough credits for the tiny quotafile |
| 1096 | * write. |
| 1097 | */ |
| 1098 | int do_journal_get_write_access(handle_t *handle, |
| 1099 | struct buffer_head *bh) |
| 1100 | { |
| 1101 | int dirty = buffer_dirty(bh); |
| 1102 | int ret; |
| 1103 | |
| 1104 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
| 1105 | return 0; |
| 1106 | /* |
| 1107 | * __block_write_begin() could have dirtied some buffers. Clean |
| 1108 | * the dirty bit as jbd2_journal_get_write_access() could complain |
| 1109 | * otherwise about fs integrity issues. Setting of the dirty bit |
| 1110 | * by __block_write_begin() isn't a real problem here as we clear |
| 1111 | * the bit before releasing a page lock and thus writeback cannot |
| 1112 | * ever write the buffer. |
| 1113 | */ |
| 1114 | if (dirty) |
| 1115 | clear_buffer_dirty(bh); |
| 1116 | BUFFER_TRACE(bh, "get write access"); |
| 1117 | ret = ext4_journal_get_write_access(handle, bh); |
| 1118 | if (!ret && dirty) |
| 1119 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); |
| 1120 | return ret; |
| 1121 | } |
| 1122 | |
| 1123 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
| 1124 | static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len, |
| 1125 | get_block_t *get_block) |
| 1126 | { |
| 1127 | unsigned from = pos & (PAGE_SIZE - 1); |
| 1128 | unsigned to = from + len; |
| 1129 | struct inode *inode = page->mapping->host; |
| 1130 | unsigned block_start, block_end; |
| 1131 | sector_t block; |
| 1132 | int err = 0; |
| 1133 | unsigned blocksize = inode->i_sb->s_blocksize; |
| 1134 | unsigned bbits; |
| 1135 | struct buffer_head *bh, *head, *wait[2], **wait_bh = wait; |
| 1136 | bool decrypt = false; |
| 1137 | |
| 1138 | BUG_ON(!PageLocked(page)); |
| 1139 | BUG_ON(from > PAGE_SIZE); |
| 1140 | BUG_ON(to > PAGE_SIZE); |
| 1141 | BUG_ON(from > to); |
| 1142 | |
| 1143 | if (!page_has_buffers(page)) |
| 1144 | create_empty_buffers(page, blocksize, 0); |
| 1145 | head = page_buffers(page); |
| 1146 | bbits = ilog2(blocksize); |
| 1147 | block = (sector_t)page->index << (PAGE_SHIFT - bbits); |
| 1148 | |
| 1149 | for (bh = head, block_start = 0; bh != head || !block_start; |
| 1150 | block++, block_start = block_end, bh = bh->b_this_page) { |
| 1151 | block_end = block_start + blocksize; |
| 1152 | if (block_end <= from || block_start >= to) { |
| 1153 | if (PageUptodate(page)) { |
| 1154 | if (!buffer_uptodate(bh)) |
| 1155 | set_buffer_uptodate(bh); |
| 1156 | } |
| 1157 | continue; |
| 1158 | } |
| 1159 | if (buffer_new(bh)) |
| 1160 | clear_buffer_new(bh); |
| 1161 | if (!buffer_mapped(bh)) { |
| 1162 | WARN_ON(bh->b_size != blocksize); |
| 1163 | err = get_block(inode, block, bh, 1); |
| 1164 | if (err) |
| 1165 | break; |
| 1166 | if (buffer_new(bh)) { |
| 1167 | clean_bdev_bh_alias(bh); |
| 1168 | if (PageUptodate(page)) { |
| 1169 | clear_buffer_new(bh); |
| 1170 | set_buffer_uptodate(bh); |
| 1171 | mark_buffer_dirty(bh); |
| 1172 | continue; |
| 1173 | } |
| 1174 | if (block_end > to || block_start < from) |
| 1175 | zero_user_segments(page, to, block_end, |
| 1176 | block_start, from); |
| 1177 | continue; |
| 1178 | } |
| 1179 | } |
| 1180 | if (PageUptodate(page)) { |
| 1181 | if (!buffer_uptodate(bh)) |
| 1182 | set_buffer_uptodate(bh); |
| 1183 | continue; |
| 1184 | } |
| 1185 | if (!buffer_uptodate(bh) && !buffer_delay(bh) && |
| 1186 | !buffer_unwritten(bh) && |
| 1187 | (block_start < from || block_end > to)) { |
| 1188 | ll_rw_block(REQ_OP_READ, 0, 1, &bh); |
| 1189 | *wait_bh++ = bh; |
| 1190 | decrypt = ext4_encrypted_inode(inode) && |
| 1191 | S_ISREG(inode->i_mode); |
| 1192 | } |
| 1193 | } |
| 1194 | /* |
| 1195 | * If we issued read requests, let them complete. |
| 1196 | */ |
| 1197 | while (wait_bh > wait) { |
| 1198 | wait_on_buffer(*--wait_bh); |
| 1199 | if (!buffer_uptodate(*wait_bh)) |
| 1200 | err = -EIO; |
| 1201 | } |
| 1202 | if (unlikely(err)) |
| 1203 | page_zero_new_buffers(page, from, to); |
| 1204 | else if (decrypt) |
| 1205 | err = fscrypt_decrypt_page(page->mapping->host, page, |
| 1206 | PAGE_SIZE, 0, page->index); |
| 1207 | return err; |
| 1208 | } |
| 1209 | #endif |
| 1210 | |
| 1211 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
| 1212 | loff_t pos, unsigned len, unsigned flags, |
| 1213 | struct page **pagep, void **fsdata) |
| 1214 | { |
| 1215 | struct inode *inode = mapping->host; |
| 1216 | int ret, needed_blocks; |
| 1217 | handle_t *handle; |
| 1218 | int retries = 0; |
| 1219 | struct page *page; |
| 1220 | pgoff_t index; |
| 1221 | unsigned from, to; |
| 1222 | |
| 1223 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
| 1224 | return -EIO; |
| 1225 | |
| 1226 | trace_ext4_write_begin(inode, pos, len, flags); |
| 1227 | /* |
| 1228 | * Reserve one block more for addition to orphan list in case |
| 1229 | * we allocate blocks but write fails for some reason |
| 1230 | */ |
| 1231 | needed_blocks = ext4_writepage_trans_blocks(inode) + 1; |
| 1232 | index = pos >> PAGE_SHIFT; |
| 1233 | from = pos & (PAGE_SIZE - 1); |
| 1234 | to = from + len; |
| 1235 | |
| 1236 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { |
| 1237 | ret = ext4_try_to_write_inline_data(mapping, inode, pos, len, |
| 1238 | flags, pagep); |
| 1239 | if (ret < 0) |
| 1240 | return ret; |
| 1241 | if (ret == 1) |
| 1242 | return 0; |
| 1243 | } |
| 1244 | |
| 1245 | /* |
| 1246 | * grab_cache_page_write_begin() can take a long time if the |
| 1247 | * system is thrashing due to memory pressure, or if the page |
| 1248 | * is being written back. So grab it first before we start |
| 1249 | * the transaction handle. This also allows us to allocate |
| 1250 | * the page (if needed) without using GFP_NOFS. |
| 1251 | */ |
| 1252 | retry_grab: |
| 1253 | page = grab_cache_page_write_begin(mapping, index, flags); |
| 1254 | if (!page) |
| 1255 | return -ENOMEM; |
| 1256 | unlock_page(page); |
| 1257 | |
| 1258 | retry_journal: |
| 1259 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); |
| 1260 | if (IS_ERR(handle)) { |
| 1261 | put_page(page); |
| 1262 | return PTR_ERR(handle); |
| 1263 | } |
| 1264 | |
| 1265 | lock_page(page); |
| 1266 | if (page->mapping != mapping) { |
| 1267 | /* The page got truncated from under us */ |
| 1268 | unlock_page(page); |
| 1269 | put_page(page); |
| 1270 | ext4_journal_stop(handle); |
| 1271 | goto retry_grab; |
| 1272 | } |
| 1273 | /* In case writeback began while the page was unlocked */ |
| 1274 | wait_for_stable_page(page); |
| 1275 | |
| 1276 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
| 1277 | if (ext4_should_dioread_nolock(inode)) |
| 1278 | ret = ext4_block_write_begin(page, pos, len, |
| 1279 | ext4_get_block_unwritten); |
| 1280 | else |
| 1281 | ret = ext4_block_write_begin(page, pos, len, |
| 1282 | ext4_get_block); |
| 1283 | #else |
| 1284 | if (ext4_should_dioread_nolock(inode)) |
| 1285 | ret = __block_write_begin(page, pos, len, |
| 1286 | ext4_get_block_unwritten); |
| 1287 | else |
| 1288 | ret = __block_write_begin(page, pos, len, ext4_get_block); |
| 1289 | #endif |
| 1290 | if (!ret && ext4_should_journal_data(inode)) { |
| 1291 | ret = ext4_walk_page_buffers(handle, page_buffers(page), |
| 1292 | from, to, NULL, |
| 1293 | do_journal_get_write_access); |
| 1294 | } |
| 1295 | |
| 1296 | if (ret) { |
| 1297 | unlock_page(page); |
| 1298 | /* |
| 1299 | * __block_write_begin may have instantiated a few blocks |
| 1300 | * outside i_size. Trim these off again. Don't need |
| 1301 | * i_size_read because we hold i_mutex. |
| 1302 | * |
| 1303 | * Add inode to orphan list in case we crash before |
| 1304 | * truncate finishes |
| 1305 | */ |
| 1306 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
| 1307 | ext4_orphan_add(handle, inode); |
| 1308 | |
| 1309 | ext4_journal_stop(handle); |
| 1310 | if (pos + len > inode->i_size) { |
| 1311 | ext4_truncate_failed_write(inode); |
| 1312 | /* |
| 1313 | * If truncate failed early the inode might |
| 1314 | * still be on the orphan list; we need to |
| 1315 | * make sure the inode is removed from the |
| 1316 | * orphan list in that case. |
| 1317 | */ |
| 1318 | if (inode->i_nlink) |
| 1319 | ext4_orphan_del(NULL, inode); |
| 1320 | } |
| 1321 | |
| 1322 | if (ret == -ENOSPC && |
| 1323 | ext4_should_retry_alloc(inode->i_sb, &retries)) |
| 1324 | goto retry_journal; |
| 1325 | put_page(page); |
| 1326 | return ret; |
| 1327 | } |
| 1328 | *pagep = page; |
| 1329 | return ret; |
| 1330 | } |
| 1331 | |
| 1332 | /* For write_end() in data=journal mode */ |
| 1333 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) |
| 1334 | { |
| 1335 | int ret; |
| 1336 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
| 1337 | return 0; |
| 1338 | set_buffer_uptodate(bh); |
| 1339 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); |
| 1340 | clear_buffer_meta(bh); |
| 1341 | clear_buffer_prio(bh); |
| 1342 | return ret; |
| 1343 | } |
| 1344 | |
| 1345 | /* |
| 1346 | * We need to pick up the new inode size which generic_commit_write gave us |
| 1347 | * `file' can be NULL - eg, when called from page_symlink(). |
| 1348 | * |
| 1349 | * ext4 never places buffers on inode->i_mapping->private_list. metadata |
| 1350 | * buffers are managed internally. |
| 1351 | */ |
| 1352 | static int ext4_write_end(struct file *file, |
| 1353 | struct address_space *mapping, |
| 1354 | loff_t pos, unsigned len, unsigned copied, |
| 1355 | struct page *page, void *fsdata) |
| 1356 | { |
| 1357 | handle_t *handle = ext4_journal_current_handle(); |
| 1358 | struct inode *inode = mapping->host; |
| 1359 | loff_t old_size = inode->i_size; |
| 1360 | int ret = 0, ret2; |
| 1361 | int i_size_changed = 0; |
| 1362 | |
| 1363 | trace_ext4_write_end(inode, pos, len, copied); |
| 1364 | if (ext4_has_inline_data(inode)) { |
| 1365 | ret = ext4_write_inline_data_end(inode, pos, len, |
| 1366 | copied, page); |
| 1367 | if (ret < 0) { |
| 1368 | unlock_page(page); |
| 1369 | put_page(page); |
| 1370 | goto errout; |
| 1371 | } |
| 1372 | copied = ret; |
| 1373 | } else |
| 1374 | copied = block_write_end(file, mapping, pos, |
| 1375 | len, copied, page, fsdata); |
| 1376 | /* |
| 1377 | * it's important to update i_size while still holding page lock: |
| 1378 | * page writeout could otherwise come in and zero beyond i_size. |
| 1379 | */ |
| 1380 | i_size_changed = ext4_update_inode_size(inode, pos + copied); |
| 1381 | unlock_page(page); |
| 1382 | put_page(page); |
| 1383 | |
| 1384 | if (old_size < pos) |
| 1385 | pagecache_isize_extended(inode, old_size, pos); |
| 1386 | /* |
| 1387 | * Don't mark the inode dirty under page lock. First, it unnecessarily |
| 1388 | * makes the holding time of page lock longer. Second, it forces lock |
| 1389 | * ordering of page lock and transaction start for journaling |
| 1390 | * filesystems. |
| 1391 | */ |
| 1392 | if (i_size_changed) |
| 1393 | ext4_mark_inode_dirty(handle, inode); |
| 1394 | |
| 1395 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
| 1396 | /* if we have allocated more blocks and copied |
| 1397 | * less. We will have blocks allocated outside |
| 1398 | * inode->i_size. So truncate them |
| 1399 | */ |
| 1400 | ext4_orphan_add(handle, inode); |
| 1401 | errout: |
| 1402 | ret2 = ext4_journal_stop(handle); |
| 1403 | if (!ret) |
| 1404 | ret = ret2; |
| 1405 | |
| 1406 | if (pos + len > inode->i_size) { |
| 1407 | ext4_truncate_failed_write(inode); |
| 1408 | /* |
| 1409 | * If truncate failed early the inode might still be |
| 1410 | * on the orphan list; we need to make sure the inode |
| 1411 | * is removed from the orphan list in that case. |
| 1412 | */ |
| 1413 | if (inode->i_nlink) |
| 1414 | ext4_orphan_del(NULL, inode); |
| 1415 | } |
| 1416 | |
| 1417 | return ret ? ret : copied; |
| 1418 | } |
| 1419 | |
| 1420 | /* |
| 1421 | * This is a private version of page_zero_new_buffers() which doesn't |
| 1422 | * set the buffer to be dirty, since in data=journalled mode we need |
| 1423 | * to call ext4_handle_dirty_metadata() instead. |
| 1424 | */ |
| 1425 | static void ext4_journalled_zero_new_buffers(handle_t *handle, |
| 1426 | struct page *page, |
| 1427 | unsigned from, unsigned to) |
| 1428 | { |
| 1429 | unsigned int block_start = 0, block_end; |
| 1430 | struct buffer_head *head, *bh; |
| 1431 | |
| 1432 | bh = head = page_buffers(page); |
| 1433 | do { |
| 1434 | block_end = block_start + bh->b_size; |
| 1435 | if (buffer_new(bh)) { |
| 1436 | if (block_end > from && block_start < to) { |
| 1437 | if (!PageUptodate(page)) { |
| 1438 | unsigned start, size; |
| 1439 | |
| 1440 | start = max(from, block_start); |
| 1441 | size = min(to, block_end) - start; |
| 1442 | |
| 1443 | zero_user(page, start, size); |
| 1444 | write_end_fn(handle, bh); |
| 1445 | } |
| 1446 | clear_buffer_new(bh); |
| 1447 | } |
| 1448 | } |
| 1449 | block_start = block_end; |
| 1450 | bh = bh->b_this_page; |
| 1451 | } while (bh != head); |
| 1452 | } |
| 1453 | |
| 1454 | static int ext4_journalled_write_end(struct file *file, |
| 1455 | struct address_space *mapping, |
| 1456 | loff_t pos, unsigned len, unsigned copied, |
| 1457 | struct page *page, void *fsdata) |
| 1458 | { |
| 1459 | handle_t *handle = ext4_journal_current_handle(); |
| 1460 | struct inode *inode = mapping->host; |
| 1461 | loff_t old_size = inode->i_size; |
| 1462 | int ret = 0, ret2; |
| 1463 | int partial = 0; |
| 1464 | unsigned from, to; |
| 1465 | int size_changed = 0; |
| 1466 | |
| 1467 | trace_ext4_journalled_write_end(inode, pos, len, copied); |
| 1468 | from = pos & (PAGE_SIZE - 1); |
| 1469 | to = from + len; |
| 1470 | |
| 1471 | BUG_ON(!ext4_handle_valid(handle)); |
| 1472 | |
| 1473 | if (ext4_has_inline_data(inode)) { |
| 1474 | ret = ext4_write_inline_data_end(inode, pos, len, |
| 1475 | copied, page); |
| 1476 | if (ret < 0) { |
| 1477 | unlock_page(page); |
| 1478 | put_page(page); |
| 1479 | goto errout; |
| 1480 | } |
| 1481 | copied = ret; |
| 1482 | } else if (unlikely(copied < len) && !PageUptodate(page)) { |
| 1483 | copied = 0; |
| 1484 | ext4_journalled_zero_new_buffers(handle, page, from, to); |
| 1485 | } else { |
| 1486 | if (unlikely(copied < len)) |
| 1487 | ext4_journalled_zero_new_buffers(handle, page, |
| 1488 | from + copied, to); |
| 1489 | ret = ext4_walk_page_buffers(handle, page_buffers(page), from, |
| 1490 | from + copied, &partial, |
| 1491 | write_end_fn); |
| 1492 | if (!partial) |
| 1493 | SetPageUptodate(page); |
| 1494 | } |
| 1495 | size_changed = ext4_update_inode_size(inode, pos + copied); |
| 1496 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
| 1497 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
| 1498 | unlock_page(page); |
| 1499 | put_page(page); |
| 1500 | |
| 1501 | if (old_size < pos) |
| 1502 | pagecache_isize_extended(inode, old_size, pos); |
| 1503 | |
| 1504 | if (size_changed) { |
| 1505 | ret2 = ext4_mark_inode_dirty(handle, inode); |
| 1506 | if (!ret) |
| 1507 | ret = ret2; |
| 1508 | } |
| 1509 | |
| 1510 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
| 1511 | /* if we have allocated more blocks and copied |
| 1512 | * less. We will have blocks allocated outside |
| 1513 | * inode->i_size. So truncate them |
| 1514 | */ |
| 1515 | ext4_orphan_add(handle, inode); |
| 1516 | |
| 1517 | errout: |
| 1518 | ret2 = ext4_journal_stop(handle); |
| 1519 | if (!ret) |
| 1520 | ret = ret2; |
| 1521 | if (pos + len > inode->i_size) { |
| 1522 | ext4_truncate_failed_write(inode); |
| 1523 | /* |
| 1524 | * If truncate failed early the inode might still be |
| 1525 | * on the orphan list; we need to make sure the inode |
| 1526 | * is removed from the orphan list in that case. |
| 1527 | */ |
| 1528 | if (inode->i_nlink) |
| 1529 | ext4_orphan_del(NULL, inode); |
| 1530 | } |
| 1531 | |
| 1532 | return ret ? ret : copied; |
| 1533 | } |
| 1534 | |
| 1535 | /* |
| 1536 | * Reserve space for a single cluster |
| 1537 | */ |
| 1538 | static int ext4_da_reserve_space(struct inode *inode) |
| 1539 | { |
| 1540 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 1541 | struct ext4_inode_info *ei = EXT4_I(inode); |
| 1542 | int ret; |
| 1543 | |
| 1544 | /* |
| 1545 | * We will charge metadata quota at writeout time; this saves |
| 1546 | * us from metadata over-estimation, though we may go over by |
| 1547 | * a small amount in the end. Here we just reserve for data. |
| 1548 | */ |
| 1549 | ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1)); |
| 1550 | if (ret) |
| 1551 | return ret; |
| 1552 | |
| 1553 | spin_lock(&ei->i_block_reservation_lock); |
| 1554 | if (ext4_claim_free_clusters(sbi, 1, 0)) { |
| 1555 | spin_unlock(&ei->i_block_reservation_lock); |
| 1556 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1)); |
| 1557 | return -ENOSPC; |
| 1558 | } |
| 1559 | ei->i_reserved_data_blocks++; |
| 1560 | trace_ext4_da_reserve_space(inode); |
| 1561 | spin_unlock(&ei->i_block_reservation_lock); |
| 1562 | |
| 1563 | return 0; /* success */ |
| 1564 | } |
| 1565 | |
| 1566 | static void ext4_da_release_space(struct inode *inode, int to_free) |
| 1567 | { |
| 1568 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 1569 | struct ext4_inode_info *ei = EXT4_I(inode); |
| 1570 | |
| 1571 | if (!to_free) |
| 1572 | return; /* Nothing to release, exit */ |
| 1573 | |
| 1574 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
| 1575 | |
| 1576 | trace_ext4_da_release_space(inode, to_free); |
| 1577 | if (unlikely(to_free > ei->i_reserved_data_blocks)) { |
| 1578 | /* |
| 1579 | * if there aren't enough reserved blocks, then the |
| 1580 | * counter is messed up somewhere. Since this |
| 1581 | * function is called from invalidate page, it's |
| 1582 | * harmless to return without any action. |
| 1583 | */ |
| 1584 | ext4_warning(inode->i_sb, "ext4_da_release_space: " |
| 1585 | "ino %lu, to_free %d with only %d reserved " |
| 1586 | "data blocks", inode->i_ino, to_free, |
| 1587 | ei->i_reserved_data_blocks); |
| 1588 | WARN_ON(1); |
| 1589 | to_free = ei->i_reserved_data_blocks; |
| 1590 | } |
| 1591 | ei->i_reserved_data_blocks -= to_free; |
| 1592 | |
| 1593 | /* update fs dirty data blocks counter */ |
| 1594 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free); |
| 1595 | |
| 1596 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
| 1597 | |
| 1598 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free)); |
| 1599 | } |
| 1600 | |
| 1601 | static void ext4_da_page_release_reservation(struct page *page, |
| 1602 | unsigned int offset, |
| 1603 | unsigned int length) |
| 1604 | { |
| 1605 | int to_release = 0, contiguous_blks = 0; |
| 1606 | struct buffer_head *head, *bh; |
| 1607 | unsigned int curr_off = 0; |
| 1608 | struct inode *inode = page->mapping->host; |
| 1609 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 1610 | unsigned int stop = offset + length; |
| 1611 | int num_clusters; |
| 1612 | ext4_fsblk_t lblk; |
| 1613 | |
| 1614 | BUG_ON(stop > PAGE_SIZE || stop < length); |
| 1615 | |
| 1616 | head = page_buffers(page); |
| 1617 | bh = head; |
| 1618 | do { |
| 1619 | unsigned int next_off = curr_off + bh->b_size; |
| 1620 | |
| 1621 | if (next_off > stop) |
| 1622 | break; |
| 1623 | |
| 1624 | if ((offset <= curr_off) && (buffer_delay(bh))) { |
| 1625 | to_release++; |
| 1626 | contiguous_blks++; |
| 1627 | clear_buffer_delay(bh); |
| 1628 | } else if (contiguous_blks) { |
| 1629 | lblk = page->index << |
| 1630 | (PAGE_SHIFT - inode->i_blkbits); |
| 1631 | lblk += (curr_off >> inode->i_blkbits) - |
| 1632 | contiguous_blks; |
| 1633 | ext4_es_remove_extent(inode, lblk, contiguous_blks); |
| 1634 | contiguous_blks = 0; |
| 1635 | } |
| 1636 | curr_off = next_off; |
| 1637 | } while ((bh = bh->b_this_page) != head); |
| 1638 | |
| 1639 | if (contiguous_blks) { |
| 1640 | lblk = page->index << (PAGE_SHIFT - inode->i_blkbits); |
| 1641 | lblk += (curr_off >> inode->i_blkbits) - contiguous_blks; |
| 1642 | ext4_es_remove_extent(inode, lblk, contiguous_blks); |
| 1643 | } |
| 1644 | |
| 1645 | /* If we have released all the blocks belonging to a cluster, then we |
| 1646 | * need to release the reserved space for that cluster. */ |
| 1647 | num_clusters = EXT4_NUM_B2C(sbi, to_release); |
| 1648 | while (num_clusters > 0) { |
| 1649 | lblk = (page->index << (PAGE_SHIFT - inode->i_blkbits)) + |
| 1650 | ((num_clusters - 1) << sbi->s_cluster_bits); |
| 1651 | if (sbi->s_cluster_ratio == 1 || |
| 1652 | !ext4_find_delalloc_cluster(inode, lblk)) |
| 1653 | ext4_da_release_space(inode, 1); |
| 1654 | |
| 1655 | num_clusters--; |
| 1656 | } |
| 1657 | } |
| 1658 | |
| 1659 | /* |
| 1660 | * Delayed allocation stuff |
| 1661 | */ |
| 1662 | |
| 1663 | struct mpage_da_data { |
| 1664 | struct inode *inode; |
| 1665 | struct writeback_control *wbc; |
| 1666 | |
| 1667 | pgoff_t first_page; /* The first page to write */ |
| 1668 | pgoff_t next_page; /* Current page to examine */ |
| 1669 | pgoff_t last_page; /* Last page to examine */ |
| 1670 | /* |
| 1671 | * Extent to map - this can be after first_page because that can be |
| 1672 | * fully mapped. We somewhat abuse m_flags to store whether the extent |
| 1673 | * is delalloc or unwritten. |
| 1674 | */ |
| 1675 | struct ext4_map_blocks map; |
| 1676 | struct ext4_io_submit io_submit; /* IO submission data */ |
| 1677 | unsigned int do_map:1; |
| 1678 | }; |
| 1679 | |
| 1680 | static void mpage_release_unused_pages(struct mpage_da_data *mpd, |
| 1681 | bool invalidate) |
| 1682 | { |
| 1683 | int nr_pages, i; |
| 1684 | pgoff_t index, end; |
| 1685 | struct pagevec pvec; |
| 1686 | struct inode *inode = mpd->inode; |
| 1687 | struct address_space *mapping = inode->i_mapping; |
| 1688 | |
| 1689 | /* This is necessary when next_page == 0. */ |
| 1690 | if (mpd->first_page >= mpd->next_page) |
| 1691 | return; |
| 1692 | |
| 1693 | index = mpd->first_page; |
| 1694 | end = mpd->next_page - 1; |
| 1695 | if (invalidate) { |
| 1696 | ext4_lblk_t start, last; |
| 1697 | start = index << (PAGE_SHIFT - inode->i_blkbits); |
| 1698 | last = end << (PAGE_SHIFT - inode->i_blkbits); |
| 1699 | ext4_es_remove_extent(inode, start, last - start + 1); |
| 1700 | } |
| 1701 | |
| 1702 | pagevec_init(&pvec, 0); |
| 1703 | while (index <= end) { |
| 1704 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); |
| 1705 | if (nr_pages == 0) |
| 1706 | break; |
| 1707 | for (i = 0; i < nr_pages; i++) { |
| 1708 | struct page *page = pvec.pages[i]; |
| 1709 | if (page->index > end) |
| 1710 | break; |
| 1711 | BUG_ON(!PageLocked(page)); |
| 1712 | BUG_ON(PageWriteback(page)); |
| 1713 | if (invalidate) { |
| 1714 | if (page_mapped(page)) |
| 1715 | clear_page_dirty_for_io(page); |
| 1716 | block_invalidatepage(page, 0, PAGE_SIZE); |
| 1717 | ClearPageUptodate(page); |
| 1718 | } |
| 1719 | unlock_page(page); |
| 1720 | } |
| 1721 | index = pvec.pages[nr_pages - 1]->index + 1; |
| 1722 | pagevec_release(&pvec); |
| 1723 | } |
| 1724 | } |
| 1725 | |
| 1726 | static void ext4_print_free_blocks(struct inode *inode) |
| 1727 | { |
| 1728 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 1729 | struct super_block *sb = inode->i_sb; |
| 1730 | struct ext4_inode_info *ei = EXT4_I(inode); |
| 1731 | |
| 1732 | ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld", |
| 1733 | EXT4_C2B(EXT4_SB(inode->i_sb), |
| 1734 | ext4_count_free_clusters(sb))); |
| 1735 | ext4_msg(sb, KERN_CRIT, "Free/Dirty block details"); |
| 1736 | ext4_msg(sb, KERN_CRIT, "free_blocks=%lld", |
| 1737 | (long long) EXT4_C2B(EXT4_SB(sb), |
| 1738 | percpu_counter_sum(&sbi->s_freeclusters_counter))); |
| 1739 | ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld", |
| 1740 | (long long) EXT4_C2B(EXT4_SB(sb), |
| 1741 | percpu_counter_sum(&sbi->s_dirtyclusters_counter))); |
| 1742 | ext4_msg(sb, KERN_CRIT, "Block reservation details"); |
| 1743 | ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u", |
| 1744 | ei->i_reserved_data_blocks); |
| 1745 | return; |
| 1746 | } |
| 1747 | |
| 1748 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh) |
| 1749 | { |
| 1750 | return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh); |
| 1751 | } |
| 1752 | |
| 1753 | /* |
| 1754 | * This function is grabs code from the very beginning of |
| 1755 | * ext4_map_blocks, but assumes that the caller is from delayed write |
| 1756 | * time. This function looks up the requested blocks and sets the |
| 1757 | * buffer delay bit under the protection of i_data_sem. |
| 1758 | */ |
| 1759 | static int ext4_da_map_blocks(struct inode *inode, sector_t iblock, |
| 1760 | struct ext4_map_blocks *map, |
| 1761 | struct buffer_head *bh) |
| 1762 | { |
| 1763 | struct extent_status es; |
| 1764 | int retval; |
| 1765 | sector_t invalid_block = ~((sector_t) 0xffff); |
| 1766 | #ifdef ES_AGGRESSIVE_TEST |
| 1767 | struct ext4_map_blocks orig_map; |
| 1768 | |
| 1769 | memcpy(&orig_map, map, sizeof(*map)); |
| 1770 | #endif |
| 1771 | |
| 1772 | if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) |
| 1773 | invalid_block = ~0; |
| 1774 | |
| 1775 | map->m_flags = 0; |
| 1776 | ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u," |
| 1777 | "logical block %lu\n", inode->i_ino, map->m_len, |
| 1778 | (unsigned long) map->m_lblk); |
| 1779 | |
| 1780 | /* Lookup extent status tree firstly */ |
| 1781 | if (ext4_es_lookup_extent(inode, iblock, &es)) { |
| 1782 | if (ext4_es_is_hole(&es)) { |
| 1783 | retval = 0; |
| 1784 | down_read(&EXT4_I(inode)->i_data_sem); |
| 1785 | goto add_delayed; |
| 1786 | } |
| 1787 | |
| 1788 | /* |
| 1789 | * Delayed extent could be allocated by fallocate. |
| 1790 | * So we need to check it. |
| 1791 | */ |
| 1792 | if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) { |
| 1793 | map_bh(bh, inode->i_sb, invalid_block); |
| 1794 | set_buffer_new(bh); |
| 1795 | set_buffer_delay(bh); |
| 1796 | return 0; |
| 1797 | } |
| 1798 | |
| 1799 | map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk; |
| 1800 | retval = es.es_len - (iblock - es.es_lblk); |
| 1801 | if (retval > map->m_len) |
| 1802 | retval = map->m_len; |
| 1803 | map->m_len = retval; |
| 1804 | if (ext4_es_is_written(&es)) |
| 1805 | map->m_flags |= EXT4_MAP_MAPPED; |
| 1806 | else if (ext4_es_is_unwritten(&es)) |
| 1807 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
| 1808 | else |
| 1809 | BUG_ON(1); |
| 1810 | |
| 1811 | #ifdef ES_AGGRESSIVE_TEST |
| 1812 | ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0); |
| 1813 | #endif |
| 1814 | return retval; |
| 1815 | } |
| 1816 | |
| 1817 | /* |
| 1818 | * Try to see if we can get the block without requesting a new |
| 1819 | * file system block. |
| 1820 | */ |
| 1821 | down_read(&EXT4_I(inode)->i_data_sem); |
| 1822 | if (ext4_has_inline_data(inode)) |
| 1823 | retval = 0; |
| 1824 | else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
| 1825 | retval = ext4_ext_map_blocks(NULL, inode, map, 0); |
| 1826 | else |
| 1827 | retval = ext4_ind_map_blocks(NULL, inode, map, 0); |
| 1828 | |
| 1829 | add_delayed: |
| 1830 | if (retval == 0) { |
| 1831 | int ret; |
| 1832 | /* |
| 1833 | * XXX: __block_prepare_write() unmaps passed block, |
| 1834 | * is it OK? |
| 1835 | */ |
| 1836 | /* |
| 1837 | * If the block was allocated from previously allocated cluster, |
| 1838 | * then we don't need to reserve it again. However we still need |
| 1839 | * to reserve metadata for every block we're going to write. |
| 1840 | */ |
| 1841 | if (EXT4_SB(inode->i_sb)->s_cluster_ratio == 1 || |
| 1842 | !ext4_find_delalloc_cluster(inode, map->m_lblk)) { |
| 1843 | ret = ext4_da_reserve_space(inode); |
| 1844 | if (ret) { |
| 1845 | /* not enough space to reserve */ |
| 1846 | retval = ret; |
| 1847 | goto out_unlock; |
| 1848 | } |
| 1849 | } |
| 1850 | |
| 1851 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, |
| 1852 | ~0, EXTENT_STATUS_DELAYED); |
| 1853 | if (ret) { |
| 1854 | retval = ret; |
| 1855 | goto out_unlock; |
| 1856 | } |
| 1857 | |
| 1858 | map_bh(bh, inode->i_sb, invalid_block); |
| 1859 | set_buffer_new(bh); |
| 1860 | set_buffer_delay(bh); |
| 1861 | } else if (retval > 0) { |
| 1862 | int ret; |
| 1863 | unsigned int status; |
| 1864 | |
| 1865 | if (unlikely(retval != map->m_len)) { |
| 1866 | ext4_warning(inode->i_sb, |
| 1867 | "ES len assertion failed for inode " |
| 1868 | "%lu: retval %d != map->m_len %d", |
| 1869 | inode->i_ino, retval, map->m_len); |
| 1870 | WARN_ON(1); |
| 1871 | } |
| 1872 | |
| 1873 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
| 1874 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; |
| 1875 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, |
| 1876 | map->m_pblk, status); |
| 1877 | if (ret != 0) |
| 1878 | retval = ret; |
| 1879 | } |
| 1880 | |
| 1881 | out_unlock: |
| 1882 | up_read((&EXT4_I(inode)->i_data_sem)); |
| 1883 | |
| 1884 | return retval; |
| 1885 | } |
| 1886 | |
| 1887 | /* |
| 1888 | * This is a special get_block_t callback which is used by |
| 1889 | * ext4_da_write_begin(). It will either return mapped block or |
| 1890 | * reserve space for a single block. |
| 1891 | * |
| 1892 | * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. |
| 1893 | * We also have b_blocknr = -1 and b_bdev initialized properly |
| 1894 | * |
| 1895 | * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. |
| 1896 | * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev |
| 1897 | * initialized properly. |
| 1898 | */ |
| 1899 | int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, |
| 1900 | struct buffer_head *bh, int create) |
| 1901 | { |
| 1902 | struct ext4_map_blocks map; |
| 1903 | int ret = 0; |
| 1904 | |
| 1905 | BUG_ON(create == 0); |
| 1906 | BUG_ON(bh->b_size != inode->i_sb->s_blocksize); |
| 1907 | |
| 1908 | map.m_lblk = iblock; |
| 1909 | map.m_len = 1; |
| 1910 | |
| 1911 | /* |
| 1912 | * first, we need to know whether the block is allocated already |
| 1913 | * preallocated blocks are unmapped but should treated |
| 1914 | * the same as allocated blocks. |
| 1915 | */ |
| 1916 | ret = ext4_da_map_blocks(inode, iblock, &map, bh); |
| 1917 | if (ret <= 0) |
| 1918 | return ret; |
| 1919 | |
| 1920 | map_bh(bh, inode->i_sb, map.m_pblk); |
| 1921 | ext4_update_bh_state(bh, map.m_flags); |
| 1922 | |
| 1923 | if (buffer_unwritten(bh)) { |
| 1924 | /* A delayed write to unwritten bh should be marked |
| 1925 | * new and mapped. Mapped ensures that we don't do |
| 1926 | * get_block multiple times when we write to the same |
| 1927 | * offset and new ensures that we do proper zero out |
| 1928 | * for partial write. |
| 1929 | */ |
| 1930 | set_buffer_new(bh); |
| 1931 | set_buffer_mapped(bh); |
| 1932 | } |
| 1933 | return 0; |
| 1934 | } |
| 1935 | |
| 1936 | static int bget_one(handle_t *handle, struct buffer_head *bh) |
| 1937 | { |
| 1938 | get_bh(bh); |
| 1939 | return 0; |
| 1940 | } |
| 1941 | |
| 1942 | static int bput_one(handle_t *handle, struct buffer_head *bh) |
| 1943 | { |
| 1944 | put_bh(bh); |
| 1945 | return 0; |
| 1946 | } |
| 1947 | |
| 1948 | static int __ext4_journalled_writepage(struct page *page, |
| 1949 | unsigned int len) |
| 1950 | { |
| 1951 | struct address_space *mapping = page->mapping; |
| 1952 | struct inode *inode = mapping->host; |
| 1953 | struct buffer_head *page_bufs = NULL; |
| 1954 | handle_t *handle = NULL; |
| 1955 | int ret = 0, err = 0; |
| 1956 | int inline_data = ext4_has_inline_data(inode); |
| 1957 | struct buffer_head *inode_bh = NULL; |
| 1958 | |
| 1959 | ClearPageChecked(page); |
| 1960 | |
| 1961 | if (inline_data) { |
| 1962 | BUG_ON(page->index != 0); |
| 1963 | BUG_ON(len > ext4_get_max_inline_size(inode)); |
| 1964 | inode_bh = ext4_journalled_write_inline_data(inode, len, page); |
| 1965 | if (inode_bh == NULL) |
| 1966 | goto out; |
| 1967 | } else { |
| 1968 | page_bufs = page_buffers(page); |
| 1969 | if (!page_bufs) { |
| 1970 | BUG(); |
| 1971 | goto out; |
| 1972 | } |
| 1973 | ext4_walk_page_buffers(handle, page_bufs, 0, len, |
| 1974 | NULL, bget_one); |
| 1975 | } |
| 1976 | /* |
| 1977 | * We need to release the page lock before we start the |
| 1978 | * journal, so grab a reference so the page won't disappear |
| 1979 | * out from under us. |
| 1980 | */ |
| 1981 | get_page(page); |
| 1982 | unlock_page(page); |
| 1983 | |
| 1984 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
| 1985 | ext4_writepage_trans_blocks(inode)); |
| 1986 | if (IS_ERR(handle)) { |
| 1987 | ret = PTR_ERR(handle); |
| 1988 | put_page(page); |
| 1989 | goto out_no_pagelock; |
| 1990 | } |
| 1991 | BUG_ON(!ext4_handle_valid(handle)); |
| 1992 | |
| 1993 | lock_page(page); |
| 1994 | put_page(page); |
| 1995 | if (page->mapping != mapping) { |
| 1996 | /* The page got truncated from under us */ |
| 1997 | ext4_journal_stop(handle); |
| 1998 | ret = 0; |
| 1999 | goto out; |
| 2000 | } |
| 2001 | |
| 2002 | if (inline_data) { |
| 2003 | BUFFER_TRACE(inode_bh, "get write access"); |
| 2004 | ret = ext4_journal_get_write_access(handle, inode_bh); |
| 2005 | |
| 2006 | err = ext4_handle_dirty_metadata(handle, inode, inode_bh); |
| 2007 | |
| 2008 | } else { |
| 2009 | ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, |
| 2010 | do_journal_get_write_access); |
| 2011 | |
| 2012 | err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, |
| 2013 | write_end_fn); |
| 2014 | } |
| 2015 | if (ret == 0) |
| 2016 | ret = err; |
| 2017 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
| 2018 | err = ext4_journal_stop(handle); |
| 2019 | if (!ret) |
| 2020 | ret = err; |
| 2021 | |
| 2022 | if (!ext4_has_inline_data(inode)) |
| 2023 | ext4_walk_page_buffers(NULL, page_bufs, 0, len, |
| 2024 | NULL, bput_one); |
| 2025 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
| 2026 | out: |
| 2027 | unlock_page(page); |
| 2028 | out_no_pagelock: |
| 2029 | brelse(inode_bh); |
| 2030 | return ret; |
| 2031 | } |
| 2032 | |
| 2033 | /* |
| 2034 | * Note that we don't need to start a transaction unless we're journaling data |
| 2035 | * because we should have holes filled from ext4_page_mkwrite(). We even don't |
| 2036 | * need to file the inode to the transaction's list in ordered mode because if |
| 2037 | * we are writing back data added by write(), the inode is already there and if |
| 2038 | * we are writing back data modified via mmap(), no one guarantees in which |
| 2039 | * transaction the data will hit the disk. In case we are journaling data, we |
| 2040 | * cannot start transaction directly because transaction start ranks above page |
| 2041 | * lock so we have to do some magic. |
| 2042 | * |
| 2043 | * This function can get called via... |
| 2044 | * - ext4_writepages after taking page lock (have journal handle) |
| 2045 | * - journal_submit_inode_data_buffers (no journal handle) |
| 2046 | * - shrink_page_list via the kswapd/direct reclaim (no journal handle) |
| 2047 | * - grab_page_cache when doing write_begin (have journal handle) |
| 2048 | * |
| 2049 | * We don't do any block allocation in this function. If we have page with |
| 2050 | * multiple blocks we need to write those buffer_heads that are mapped. This |
| 2051 | * is important for mmaped based write. So if we do with blocksize 1K |
| 2052 | * truncate(f, 1024); |
| 2053 | * a = mmap(f, 0, 4096); |
| 2054 | * a[0] = 'a'; |
| 2055 | * truncate(f, 4096); |
| 2056 | * we have in the page first buffer_head mapped via page_mkwrite call back |
| 2057 | * but other buffer_heads would be unmapped but dirty (dirty done via the |
| 2058 | * do_wp_page). So writepage should write the first block. If we modify |
| 2059 | * the mmap area beyond 1024 we will again get a page_fault and the |
| 2060 | * page_mkwrite callback will do the block allocation and mark the |
| 2061 | * buffer_heads mapped. |
| 2062 | * |
| 2063 | * We redirty the page if we have any buffer_heads that is either delay or |
| 2064 | * unwritten in the page. |
| 2065 | * |
| 2066 | * We can get recursively called as show below. |
| 2067 | * |
| 2068 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> |
| 2069 | * ext4_writepage() |
| 2070 | * |
| 2071 | * But since we don't do any block allocation we should not deadlock. |
| 2072 | * Page also have the dirty flag cleared so we don't get recurive page_lock. |
| 2073 | */ |
| 2074 | static int ext4_writepage(struct page *page, |
| 2075 | struct writeback_control *wbc) |
| 2076 | { |
| 2077 | int ret = 0; |
| 2078 | loff_t size; |
| 2079 | unsigned int len; |
| 2080 | struct buffer_head *page_bufs = NULL; |
| 2081 | struct inode *inode = page->mapping->host; |
| 2082 | struct ext4_io_submit io_submit; |
| 2083 | bool keep_towrite = false; |
| 2084 | |
| 2085 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) { |
| 2086 | ext4_invalidatepage(page, 0, PAGE_SIZE); |
| 2087 | unlock_page(page); |
| 2088 | return -EIO; |
| 2089 | } |
| 2090 | |
| 2091 | trace_ext4_writepage(page); |
| 2092 | size = i_size_read(inode); |
| 2093 | if (page->index == size >> PAGE_SHIFT) |
| 2094 | len = size & ~PAGE_MASK; |
| 2095 | else |
| 2096 | len = PAGE_SIZE; |
| 2097 | |
| 2098 | page_bufs = page_buffers(page); |
| 2099 | /* |
| 2100 | * We cannot do block allocation or other extent handling in this |
| 2101 | * function. If there are buffers needing that, we have to redirty |
| 2102 | * the page. But we may reach here when we do a journal commit via |
| 2103 | * journal_submit_inode_data_buffers() and in that case we must write |
| 2104 | * allocated buffers to achieve data=ordered mode guarantees. |
| 2105 | * |
| 2106 | * Also, if there is only one buffer per page (the fs block |
| 2107 | * size == the page size), if one buffer needs block |
| 2108 | * allocation or needs to modify the extent tree to clear the |
| 2109 | * unwritten flag, we know that the page can't be written at |
| 2110 | * all, so we might as well refuse the write immediately. |
| 2111 | * Unfortunately if the block size != page size, we can't as |
| 2112 | * easily detect this case using ext4_walk_page_buffers(), but |
| 2113 | * for the extremely common case, this is an optimization that |
| 2114 | * skips a useless round trip through ext4_bio_write_page(). |
| 2115 | */ |
| 2116 | if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL, |
| 2117 | ext4_bh_delay_or_unwritten)) { |
| 2118 | redirty_page_for_writepage(wbc, page); |
| 2119 | if ((current->flags & PF_MEMALLOC) || |
| 2120 | (inode->i_sb->s_blocksize == PAGE_SIZE)) { |
| 2121 | /* |
| 2122 | * For memory cleaning there's no point in writing only |
| 2123 | * some buffers. So just bail out. Warn if we came here |
| 2124 | * from direct reclaim. |
| 2125 | */ |
| 2126 | WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) |
| 2127 | == PF_MEMALLOC); |
| 2128 | unlock_page(page); |
| 2129 | return 0; |
| 2130 | } |
| 2131 | keep_towrite = true; |
| 2132 | } |
| 2133 | |
| 2134 | if (PageChecked(page) && ext4_should_journal_data(inode)) |
| 2135 | /* |
| 2136 | * It's mmapped pagecache. Add buffers and journal it. There |
| 2137 | * doesn't seem much point in redirtying the page here. |
| 2138 | */ |
| 2139 | return __ext4_journalled_writepage(page, len); |
| 2140 | |
| 2141 | ext4_io_submit_init(&io_submit, wbc); |
| 2142 | io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS); |
| 2143 | if (!io_submit.io_end) { |
| 2144 | redirty_page_for_writepage(wbc, page); |
| 2145 | unlock_page(page); |
| 2146 | return -ENOMEM; |
| 2147 | } |
| 2148 | ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite); |
| 2149 | ext4_io_submit(&io_submit); |
| 2150 | /* Drop io_end reference we got from init */ |
| 2151 | ext4_put_io_end_defer(io_submit.io_end); |
| 2152 | return ret; |
| 2153 | } |
| 2154 | |
| 2155 | static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page) |
| 2156 | { |
| 2157 | int len; |
| 2158 | loff_t size; |
| 2159 | int err; |
| 2160 | |
| 2161 | BUG_ON(page->index != mpd->first_page); |
| 2162 | clear_page_dirty_for_io(page); |
| 2163 | /* |
| 2164 | * We have to be very careful here! Nothing protects writeback path |
| 2165 | * against i_size changes and the page can be writeably mapped into |
| 2166 | * page tables. So an application can be growing i_size and writing |
| 2167 | * data through mmap while writeback runs. clear_page_dirty_for_io() |
| 2168 | * write-protects our page in page tables and the page cannot get |
| 2169 | * written to again until we release page lock. So only after |
| 2170 | * clear_page_dirty_for_io() we are safe to sample i_size for |
| 2171 | * ext4_bio_write_page() to zero-out tail of the written page. We rely |
| 2172 | * on the barrier provided by TestClearPageDirty in |
| 2173 | * clear_page_dirty_for_io() to make sure i_size is really sampled only |
| 2174 | * after page tables are updated. |
| 2175 | */ |
| 2176 | size = i_size_read(mpd->inode); |
| 2177 | if (page->index == size >> PAGE_SHIFT) |
| 2178 | len = size & ~PAGE_MASK; |
| 2179 | else |
| 2180 | len = PAGE_SIZE; |
| 2181 | err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false); |
| 2182 | if (!err) |
| 2183 | mpd->wbc->nr_to_write--; |
| 2184 | mpd->first_page++; |
| 2185 | |
| 2186 | return err; |
| 2187 | } |
| 2188 | |
| 2189 | #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay)) |
| 2190 | |
| 2191 | /* |
| 2192 | * mballoc gives us at most this number of blocks... |
| 2193 | * XXX: That seems to be only a limitation of ext4_mb_normalize_request(). |
| 2194 | * The rest of mballoc seems to handle chunks up to full group size. |
| 2195 | */ |
| 2196 | #define MAX_WRITEPAGES_EXTENT_LEN 2048 |
| 2197 | |
| 2198 | /* |
| 2199 | * mpage_add_bh_to_extent - try to add bh to extent of blocks to map |
| 2200 | * |
| 2201 | * @mpd - extent of blocks |
| 2202 | * @lblk - logical number of the block in the file |
| 2203 | * @bh - buffer head we want to add to the extent |
| 2204 | * |
| 2205 | * The function is used to collect contig. blocks in the same state. If the |
| 2206 | * buffer doesn't require mapping for writeback and we haven't started the |
| 2207 | * extent of buffers to map yet, the function returns 'true' immediately - the |
| 2208 | * caller can write the buffer right away. Otherwise the function returns true |
| 2209 | * if the block has been added to the extent, false if the block couldn't be |
| 2210 | * added. |
| 2211 | */ |
| 2212 | static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk, |
| 2213 | struct buffer_head *bh) |
| 2214 | { |
| 2215 | struct ext4_map_blocks *map = &mpd->map; |
| 2216 | |
| 2217 | /* Buffer that doesn't need mapping for writeback? */ |
| 2218 | if (!buffer_dirty(bh) || !buffer_mapped(bh) || |
| 2219 | (!buffer_delay(bh) && !buffer_unwritten(bh))) { |
| 2220 | /* So far no extent to map => we write the buffer right away */ |
| 2221 | if (map->m_len == 0) |
| 2222 | return true; |
| 2223 | return false; |
| 2224 | } |
| 2225 | |
| 2226 | /* First block in the extent? */ |
| 2227 | if (map->m_len == 0) { |
| 2228 | /* We cannot map unless handle is started... */ |
| 2229 | if (!mpd->do_map) |
| 2230 | return false; |
| 2231 | map->m_lblk = lblk; |
| 2232 | map->m_len = 1; |
| 2233 | map->m_flags = bh->b_state & BH_FLAGS; |
| 2234 | return true; |
| 2235 | } |
| 2236 | |
| 2237 | /* Don't go larger than mballoc is willing to allocate */ |
| 2238 | if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN) |
| 2239 | return false; |
| 2240 | |
| 2241 | /* Can we merge the block to our big extent? */ |
| 2242 | if (lblk == map->m_lblk + map->m_len && |
| 2243 | (bh->b_state & BH_FLAGS) == map->m_flags) { |
| 2244 | map->m_len++; |
| 2245 | return true; |
| 2246 | } |
| 2247 | return false; |
| 2248 | } |
| 2249 | |
| 2250 | /* |
| 2251 | * mpage_process_page_bufs - submit page buffers for IO or add them to extent |
| 2252 | * |
| 2253 | * @mpd - extent of blocks for mapping |
| 2254 | * @head - the first buffer in the page |
| 2255 | * @bh - buffer we should start processing from |
| 2256 | * @lblk - logical number of the block in the file corresponding to @bh |
| 2257 | * |
| 2258 | * Walk through page buffers from @bh upto @head (exclusive) and either submit |
| 2259 | * the page for IO if all buffers in this page were mapped and there's no |
| 2260 | * accumulated extent of buffers to map or add buffers in the page to the |
| 2261 | * extent of buffers to map. The function returns 1 if the caller can continue |
| 2262 | * by processing the next page, 0 if it should stop adding buffers to the |
| 2263 | * extent to map because we cannot extend it anymore. It can also return value |
| 2264 | * < 0 in case of error during IO submission. |
| 2265 | */ |
| 2266 | static int mpage_process_page_bufs(struct mpage_da_data *mpd, |
| 2267 | struct buffer_head *head, |
| 2268 | struct buffer_head *bh, |
| 2269 | ext4_lblk_t lblk) |
| 2270 | { |
| 2271 | struct inode *inode = mpd->inode; |
| 2272 | int err; |
| 2273 | ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1) |
| 2274 | >> inode->i_blkbits; |
| 2275 | |
| 2276 | do { |
| 2277 | BUG_ON(buffer_locked(bh)); |
| 2278 | |
| 2279 | if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) { |
| 2280 | /* Found extent to map? */ |
| 2281 | if (mpd->map.m_len) |
| 2282 | return 0; |
| 2283 | /* Buffer needs mapping and handle is not started? */ |
| 2284 | if (!mpd->do_map) |
| 2285 | return 0; |
| 2286 | /* Everything mapped so far and we hit EOF */ |
| 2287 | break; |
| 2288 | } |
| 2289 | } while (lblk++, (bh = bh->b_this_page) != head); |
| 2290 | /* So far everything mapped? Submit the page for IO. */ |
| 2291 | if (mpd->map.m_len == 0) { |
| 2292 | err = mpage_submit_page(mpd, head->b_page); |
| 2293 | if (err < 0) |
| 2294 | return err; |
| 2295 | } |
| 2296 | return lblk < blocks; |
| 2297 | } |
| 2298 | |
| 2299 | /* |
| 2300 | * mpage_map_buffers - update buffers corresponding to changed extent and |
| 2301 | * submit fully mapped pages for IO |
| 2302 | * |
| 2303 | * @mpd - description of extent to map, on return next extent to map |
| 2304 | * |
| 2305 | * Scan buffers corresponding to changed extent (we expect corresponding pages |
| 2306 | * to be already locked) and update buffer state according to new extent state. |
| 2307 | * We map delalloc buffers to their physical location, clear unwritten bits, |
| 2308 | * and mark buffers as uninit when we perform writes to unwritten extents |
| 2309 | * and do extent conversion after IO is finished. If the last page is not fully |
| 2310 | * mapped, we update @map to the next extent in the last page that needs |
| 2311 | * mapping. Otherwise we submit the page for IO. |
| 2312 | */ |
| 2313 | static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd) |
| 2314 | { |
| 2315 | struct pagevec pvec; |
| 2316 | int nr_pages, i; |
| 2317 | struct inode *inode = mpd->inode; |
| 2318 | struct buffer_head *head, *bh; |
| 2319 | int bpp_bits = PAGE_SHIFT - inode->i_blkbits; |
| 2320 | pgoff_t start, end; |
| 2321 | ext4_lblk_t lblk; |
| 2322 | sector_t pblock; |
| 2323 | int err; |
| 2324 | |
| 2325 | start = mpd->map.m_lblk >> bpp_bits; |
| 2326 | end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits; |
| 2327 | lblk = start << bpp_bits; |
| 2328 | pblock = mpd->map.m_pblk; |
| 2329 | |
| 2330 | pagevec_init(&pvec, 0); |
| 2331 | while (start <= end) { |
| 2332 | nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start, |
| 2333 | PAGEVEC_SIZE); |
| 2334 | if (nr_pages == 0) |
| 2335 | break; |
| 2336 | for (i = 0; i < nr_pages; i++) { |
| 2337 | struct page *page = pvec.pages[i]; |
| 2338 | |
| 2339 | if (page->index > end) |
| 2340 | break; |
| 2341 | /* Up to 'end' pages must be contiguous */ |
| 2342 | BUG_ON(page->index != start); |
| 2343 | bh = head = page_buffers(page); |
| 2344 | do { |
| 2345 | if (lblk < mpd->map.m_lblk) |
| 2346 | continue; |
| 2347 | if (lblk >= mpd->map.m_lblk + mpd->map.m_len) { |
| 2348 | /* |
| 2349 | * Buffer after end of mapped extent. |
| 2350 | * Find next buffer in the page to map. |
| 2351 | */ |
| 2352 | mpd->map.m_len = 0; |
| 2353 | mpd->map.m_flags = 0; |
| 2354 | /* |
| 2355 | * FIXME: If dioread_nolock supports |
| 2356 | * blocksize < pagesize, we need to make |
| 2357 | * sure we add size mapped so far to |
| 2358 | * io_end->size as the following call |
| 2359 | * can submit the page for IO. |
| 2360 | */ |
| 2361 | err = mpage_process_page_bufs(mpd, head, |
| 2362 | bh, lblk); |
| 2363 | pagevec_release(&pvec); |
| 2364 | if (err > 0) |
| 2365 | err = 0; |
| 2366 | return err; |
| 2367 | } |
| 2368 | if (buffer_delay(bh)) { |
| 2369 | clear_buffer_delay(bh); |
| 2370 | bh->b_blocknr = pblock++; |
| 2371 | } |
| 2372 | clear_buffer_unwritten(bh); |
| 2373 | } while (lblk++, (bh = bh->b_this_page) != head); |
| 2374 | |
| 2375 | /* |
| 2376 | * FIXME: This is going to break if dioread_nolock |
| 2377 | * supports blocksize < pagesize as we will try to |
| 2378 | * convert potentially unmapped parts of inode. |
| 2379 | */ |
| 2380 | mpd->io_submit.io_end->size += PAGE_SIZE; |
| 2381 | /* Page fully mapped - let IO run! */ |
| 2382 | err = mpage_submit_page(mpd, page); |
| 2383 | if (err < 0) { |
| 2384 | pagevec_release(&pvec); |
| 2385 | return err; |
| 2386 | } |
| 2387 | start++; |
| 2388 | } |
| 2389 | pagevec_release(&pvec); |
| 2390 | } |
| 2391 | /* Extent fully mapped and matches with page boundary. We are done. */ |
| 2392 | mpd->map.m_len = 0; |
| 2393 | mpd->map.m_flags = 0; |
| 2394 | return 0; |
| 2395 | } |
| 2396 | |
| 2397 | static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd) |
| 2398 | { |
| 2399 | struct inode *inode = mpd->inode; |
| 2400 | struct ext4_map_blocks *map = &mpd->map; |
| 2401 | int get_blocks_flags; |
| 2402 | int err, dioread_nolock; |
| 2403 | |
| 2404 | trace_ext4_da_write_pages_extent(inode, map); |
| 2405 | /* |
| 2406 | * Call ext4_map_blocks() to allocate any delayed allocation blocks, or |
| 2407 | * to convert an unwritten extent to be initialized (in the case |
| 2408 | * where we have written into one or more preallocated blocks). It is |
| 2409 | * possible that we're going to need more metadata blocks than |
| 2410 | * previously reserved. However we must not fail because we're in |
| 2411 | * writeback and there is nothing we can do about it so it might result |
| 2412 | * in data loss. So use reserved blocks to allocate metadata if |
| 2413 | * possible. |
| 2414 | * |
| 2415 | * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if |
| 2416 | * the blocks in question are delalloc blocks. This indicates |
| 2417 | * that the blocks and quotas has already been checked when |
| 2418 | * the data was copied into the page cache. |
| 2419 | */ |
| 2420 | get_blocks_flags = EXT4_GET_BLOCKS_CREATE | |
| 2421 | EXT4_GET_BLOCKS_METADATA_NOFAIL | |
| 2422 | EXT4_GET_BLOCKS_IO_SUBMIT; |
| 2423 | dioread_nolock = ext4_should_dioread_nolock(inode); |
| 2424 | if (dioread_nolock) |
| 2425 | get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; |
| 2426 | if (map->m_flags & (1 << BH_Delay)) |
| 2427 | get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; |
| 2428 | |
| 2429 | err = ext4_map_blocks(handle, inode, map, get_blocks_flags); |
| 2430 | if (err < 0) |
| 2431 | return err; |
| 2432 | if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) { |
| 2433 | if (!mpd->io_submit.io_end->handle && |
| 2434 | ext4_handle_valid(handle)) { |
| 2435 | mpd->io_submit.io_end->handle = handle->h_rsv_handle; |
| 2436 | handle->h_rsv_handle = NULL; |
| 2437 | } |
| 2438 | ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end); |
| 2439 | } |
| 2440 | |
| 2441 | BUG_ON(map->m_len == 0); |
| 2442 | if (map->m_flags & EXT4_MAP_NEW) { |
| 2443 | clean_bdev_aliases(inode->i_sb->s_bdev, map->m_pblk, |
| 2444 | map->m_len); |
| 2445 | } |
| 2446 | return 0; |
| 2447 | } |
| 2448 | |
| 2449 | /* |
| 2450 | * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length |
| 2451 | * mpd->len and submit pages underlying it for IO |
| 2452 | * |
| 2453 | * @handle - handle for journal operations |
| 2454 | * @mpd - extent to map |
| 2455 | * @give_up_on_write - we set this to true iff there is a fatal error and there |
| 2456 | * is no hope of writing the data. The caller should discard |
| 2457 | * dirty pages to avoid infinite loops. |
| 2458 | * |
| 2459 | * The function maps extent starting at mpd->lblk of length mpd->len. If it is |
| 2460 | * delayed, blocks are allocated, if it is unwritten, we may need to convert |
| 2461 | * them to initialized or split the described range from larger unwritten |
| 2462 | * extent. Note that we need not map all the described range since allocation |
| 2463 | * can return less blocks or the range is covered by more unwritten extents. We |
| 2464 | * cannot map more because we are limited by reserved transaction credits. On |
| 2465 | * the other hand we always make sure that the last touched page is fully |
| 2466 | * mapped so that it can be written out (and thus forward progress is |
| 2467 | * guaranteed). After mapping we submit all mapped pages for IO. |
| 2468 | */ |
| 2469 | static int mpage_map_and_submit_extent(handle_t *handle, |
| 2470 | struct mpage_da_data *mpd, |
| 2471 | bool *give_up_on_write) |
| 2472 | { |
| 2473 | struct inode *inode = mpd->inode; |
| 2474 | struct ext4_map_blocks *map = &mpd->map; |
| 2475 | int err; |
| 2476 | loff_t disksize; |
| 2477 | int progress = 0; |
| 2478 | |
| 2479 | mpd->io_submit.io_end->offset = |
| 2480 | ((loff_t)map->m_lblk) << inode->i_blkbits; |
| 2481 | do { |
| 2482 | err = mpage_map_one_extent(handle, mpd); |
| 2483 | if (err < 0) { |
| 2484 | struct super_block *sb = inode->i_sb; |
| 2485 | |
| 2486 | if (ext4_forced_shutdown(EXT4_SB(sb)) || |
| 2487 | EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED) |
| 2488 | goto invalidate_dirty_pages; |
| 2489 | /* |
| 2490 | * Let the uper layers retry transient errors. |
| 2491 | * In the case of ENOSPC, if ext4_count_free_blocks() |
| 2492 | * is non-zero, a commit should free up blocks. |
| 2493 | */ |
| 2494 | if ((err == -ENOMEM) || |
| 2495 | (err == -ENOSPC && ext4_count_free_clusters(sb))) { |
| 2496 | if (progress) |
| 2497 | goto update_disksize; |
| 2498 | return err; |
| 2499 | } |
| 2500 | ext4_msg(sb, KERN_CRIT, |
| 2501 | "Delayed block allocation failed for " |
| 2502 | "inode %lu at logical offset %llu with" |
| 2503 | " max blocks %u with error %d", |
| 2504 | inode->i_ino, |
| 2505 | (unsigned long long)map->m_lblk, |
| 2506 | (unsigned)map->m_len, -err); |
| 2507 | ext4_msg(sb, KERN_CRIT, |
| 2508 | "This should not happen!! Data will " |
| 2509 | "be lost\n"); |
| 2510 | if (err == -ENOSPC) |
| 2511 | ext4_print_free_blocks(inode); |
| 2512 | invalidate_dirty_pages: |
| 2513 | *give_up_on_write = true; |
| 2514 | return err; |
| 2515 | } |
| 2516 | progress = 1; |
| 2517 | /* |
| 2518 | * Update buffer state, submit mapped pages, and get us new |
| 2519 | * extent to map |
| 2520 | */ |
| 2521 | err = mpage_map_and_submit_buffers(mpd); |
| 2522 | if (err < 0) |
| 2523 | goto update_disksize; |
| 2524 | } while (map->m_len); |
| 2525 | |
| 2526 | update_disksize: |
| 2527 | /* |
| 2528 | * Update on-disk size after IO is submitted. Races with |
| 2529 | * truncate are avoided by checking i_size under i_data_sem. |
| 2530 | */ |
| 2531 | disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT; |
| 2532 | if (disksize > EXT4_I(inode)->i_disksize) { |
| 2533 | int err2; |
| 2534 | loff_t i_size; |
| 2535 | |
| 2536 | down_write(&EXT4_I(inode)->i_data_sem); |
| 2537 | i_size = i_size_read(inode); |
| 2538 | if (disksize > i_size) |
| 2539 | disksize = i_size; |
| 2540 | if (disksize > EXT4_I(inode)->i_disksize) |
| 2541 | EXT4_I(inode)->i_disksize = disksize; |
| 2542 | up_write(&EXT4_I(inode)->i_data_sem); |
| 2543 | err2 = ext4_mark_inode_dirty(handle, inode); |
| 2544 | if (err2) |
| 2545 | ext4_error(inode->i_sb, |
| 2546 | "Failed to mark inode %lu dirty", |
| 2547 | inode->i_ino); |
| 2548 | if (!err) |
| 2549 | err = err2; |
| 2550 | } |
| 2551 | return err; |
| 2552 | } |
| 2553 | |
| 2554 | /* |
| 2555 | * Calculate the total number of credits to reserve for one writepages |
| 2556 | * iteration. This is called from ext4_writepages(). We map an extent of |
| 2557 | * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping |
| 2558 | * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN + |
| 2559 | * bpp - 1 blocks in bpp different extents. |
| 2560 | */ |
| 2561 | static int ext4_da_writepages_trans_blocks(struct inode *inode) |
| 2562 | { |
| 2563 | int bpp = ext4_journal_blocks_per_page(inode); |
| 2564 | |
| 2565 | return ext4_meta_trans_blocks(inode, |
| 2566 | MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp); |
| 2567 | } |
| 2568 | |
| 2569 | /* |
| 2570 | * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages |
| 2571 | * and underlying extent to map |
| 2572 | * |
| 2573 | * @mpd - where to look for pages |
| 2574 | * |
| 2575 | * Walk dirty pages in the mapping. If they are fully mapped, submit them for |
| 2576 | * IO immediately. When we find a page which isn't mapped we start accumulating |
| 2577 | * extent of buffers underlying these pages that needs mapping (formed by |
| 2578 | * either delayed or unwritten buffers). We also lock the pages containing |
| 2579 | * these buffers. The extent found is returned in @mpd structure (starting at |
| 2580 | * mpd->lblk with length mpd->len blocks). |
| 2581 | * |
| 2582 | * Note that this function can attach bios to one io_end structure which are |
| 2583 | * neither logically nor physically contiguous. Although it may seem as an |
| 2584 | * unnecessary complication, it is actually inevitable in blocksize < pagesize |
| 2585 | * case as we need to track IO to all buffers underlying a page in one io_end. |
| 2586 | */ |
| 2587 | static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd) |
| 2588 | { |
| 2589 | struct address_space *mapping = mpd->inode->i_mapping; |
| 2590 | struct pagevec pvec; |
| 2591 | unsigned int nr_pages; |
| 2592 | long left = mpd->wbc->nr_to_write; |
| 2593 | pgoff_t index = mpd->first_page; |
| 2594 | pgoff_t end = mpd->last_page; |
| 2595 | int tag; |
| 2596 | int i, err = 0; |
| 2597 | int blkbits = mpd->inode->i_blkbits; |
| 2598 | ext4_lblk_t lblk; |
| 2599 | struct buffer_head *head; |
| 2600 | |
| 2601 | if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages) |
| 2602 | tag = PAGECACHE_TAG_TOWRITE; |
| 2603 | else |
| 2604 | tag = PAGECACHE_TAG_DIRTY; |
| 2605 | |
| 2606 | pagevec_init(&pvec, 0); |
| 2607 | mpd->map.m_len = 0; |
| 2608 | mpd->next_page = index; |
| 2609 | while (index <= end) { |
| 2610 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
| 2611 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); |
| 2612 | if (nr_pages == 0) |
| 2613 | goto out; |
| 2614 | |
| 2615 | for (i = 0; i < nr_pages; i++) { |
| 2616 | struct page *page = pvec.pages[i]; |
| 2617 | |
| 2618 | /* |
| 2619 | * At this point, the page may be truncated or |
| 2620 | * invalidated (changing page->mapping to NULL), or |
| 2621 | * even swizzled back from swapper_space to tmpfs file |
| 2622 | * mapping. However, page->index will not change |
| 2623 | * because we have a reference on the page. |
| 2624 | */ |
| 2625 | if (page->index > end) |
| 2626 | goto out; |
| 2627 | |
| 2628 | /* |
| 2629 | * Accumulated enough dirty pages? This doesn't apply |
| 2630 | * to WB_SYNC_ALL mode. For integrity sync we have to |
| 2631 | * keep going because someone may be concurrently |
| 2632 | * dirtying pages, and we might have synced a lot of |
| 2633 | * newly appeared dirty pages, but have not synced all |
| 2634 | * of the old dirty pages. |
| 2635 | */ |
| 2636 | if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0) |
| 2637 | goto out; |
| 2638 | |
| 2639 | /* If we can't merge this page, we are done. */ |
| 2640 | if (mpd->map.m_len > 0 && mpd->next_page != page->index) |
| 2641 | goto out; |
| 2642 | |
| 2643 | lock_page(page); |
| 2644 | /* |
| 2645 | * If the page is no longer dirty, or its mapping no |
| 2646 | * longer corresponds to inode we are writing (which |
| 2647 | * means it has been truncated or invalidated), or the |
| 2648 | * page is already under writeback and we are not doing |
| 2649 | * a data integrity writeback, skip the page |
| 2650 | */ |
| 2651 | if (!PageDirty(page) || |
| 2652 | (PageWriteback(page) && |
| 2653 | (mpd->wbc->sync_mode == WB_SYNC_NONE)) || |
| 2654 | unlikely(page->mapping != mapping)) { |
| 2655 | unlock_page(page); |
| 2656 | continue; |
| 2657 | } |
| 2658 | |
| 2659 | wait_on_page_writeback(page); |
| 2660 | BUG_ON(PageWriteback(page)); |
| 2661 | |
| 2662 | if (mpd->map.m_len == 0) |
| 2663 | mpd->first_page = page->index; |
| 2664 | mpd->next_page = page->index + 1; |
| 2665 | /* Add all dirty buffers to mpd */ |
| 2666 | lblk = ((ext4_lblk_t)page->index) << |
| 2667 | (PAGE_SHIFT - blkbits); |
| 2668 | head = page_buffers(page); |
| 2669 | err = mpage_process_page_bufs(mpd, head, head, lblk); |
| 2670 | if (err <= 0) |
| 2671 | goto out; |
| 2672 | err = 0; |
| 2673 | left--; |
| 2674 | } |
| 2675 | pagevec_release(&pvec); |
| 2676 | cond_resched(); |
| 2677 | } |
| 2678 | return 0; |
| 2679 | out: |
| 2680 | pagevec_release(&pvec); |
| 2681 | return err; |
| 2682 | } |
| 2683 | |
| 2684 | static int __writepage(struct page *page, struct writeback_control *wbc, |
| 2685 | void *data) |
| 2686 | { |
| 2687 | struct address_space *mapping = data; |
| 2688 | int ret = ext4_writepage(page, wbc); |
| 2689 | mapping_set_error(mapping, ret); |
| 2690 | return ret; |
| 2691 | } |
| 2692 | |
| 2693 | static int ext4_writepages(struct address_space *mapping, |
| 2694 | struct writeback_control *wbc) |
| 2695 | { |
| 2696 | pgoff_t writeback_index = 0; |
| 2697 | long nr_to_write = wbc->nr_to_write; |
| 2698 | int range_whole = 0; |
| 2699 | int cycled = 1; |
| 2700 | handle_t *handle = NULL; |
| 2701 | struct mpage_da_data mpd; |
| 2702 | struct inode *inode = mapping->host; |
| 2703 | int needed_blocks, rsv_blocks = 0, ret = 0; |
| 2704 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
| 2705 | bool done; |
| 2706 | struct blk_plug plug; |
| 2707 | bool give_up_on_write = false; |
| 2708 | |
| 2709 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
| 2710 | return -EIO; |
| 2711 | |
| 2712 | percpu_down_read(&sbi->s_journal_flag_rwsem); |
| 2713 | trace_ext4_writepages(inode, wbc); |
| 2714 | |
| 2715 | if (dax_mapping(mapping)) { |
| 2716 | ret = dax_writeback_mapping_range(mapping, inode->i_sb->s_bdev, |
| 2717 | wbc); |
| 2718 | goto out_writepages; |
| 2719 | } |
| 2720 | |
| 2721 | /* |
| 2722 | * No pages to write? This is mainly a kludge to avoid starting |
| 2723 | * a transaction for special inodes like journal inode on last iput() |
| 2724 | * because that could violate lock ordering on umount |
| 2725 | */ |
| 2726 | if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
| 2727 | goto out_writepages; |
| 2728 | |
| 2729 | if (ext4_should_journal_data(inode)) { |
| 2730 | struct blk_plug plug; |
| 2731 | |
| 2732 | blk_start_plug(&plug); |
| 2733 | ret = write_cache_pages(mapping, wbc, __writepage, mapping); |
| 2734 | blk_finish_plug(&plug); |
| 2735 | goto out_writepages; |
| 2736 | } |
| 2737 | |
| 2738 | /* |
| 2739 | * If the filesystem has aborted, it is read-only, so return |
| 2740 | * right away instead of dumping stack traces later on that |
| 2741 | * will obscure the real source of the problem. We test |
| 2742 | * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because |
| 2743 | * the latter could be true if the filesystem is mounted |
| 2744 | * read-only, and in that case, ext4_writepages should |
| 2745 | * *never* be called, so if that ever happens, we would want |
| 2746 | * the stack trace. |
| 2747 | */ |
| 2748 | if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) || |
| 2749 | sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) { |
| 2750 | ret = -EROFS; |
| 2751 | goto out_writepages; |
| 2752 | } |
| 2753 | |
| 2754 | if (ext4_should_dioread_nolock(inode)) { |
| 2755 | /* |
| 2756 | * We may need to convert up to one extent per block in |
| 2757 | * the page and we may dirty the inode. |
| 2758 | */ |
| 2759 | rsv_blocks = 1 + (PAGE_SIZE >> inode->i_blkbits); |
| 2760 | } |
| 2761 | |
| 2762 | /* |
| 2763 | * If we have inline data and arrive here, it means that |
| 2764 | * we will soon create the block for the 1st page, so |
| 2765 | * we'd better clear the inline data here. |
| 2766 | */ |
| 2767 | if (ext4_has_inline_data(inode)) { |
| 2768 | /* Just inode will be modified... */ |
| 2769 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); |
| 2770 | if (IS_ERR(handle)) { |
| 2771 | ret = PTR_ERR(handle); |
| 2772 | goto out_writepages; |
| 2773 | } |
| 2774 | BUG_ON(ext4_test_inode_state(inode, |
| 2775 | EXT4_STATE_MAY_INLINE_DATA)); |
| 2776 | ext4_destroy_inline_data(handle, inode); |
| 2777 | ext4_journal_stop(handle); |
| 2778 | } |
| 2779 | |
| 2780 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
| 2781 | range_whole = 1; |
| 2782 | |
| 2783 | if (wbc->range_cyclic) { |
| 2784 | writeback_index = mapping->writeback_index; |
| 2785 | if (writeback_index) |
| 2786 | cycled = 0; |
| 2787 | mpd.first_page = writeback_index; |
| 2788 | mpd.last_page = -1; |
| 2789 | } else { |
| 2790 | mpd.first_page = wbc->range_start >> PAGE_SHIFT; |
| 2791 | mpd.last_page = wbc->range_end >> PAGE_SHIFT; |
| 2792 | } |
| 2793 | |
| 2794 | mpd.inode = inode; |
| 2795 | mpd.wbc = wbc; |
| 2796 | ext4_io_submit_init(&mpd.io_submit, wbc); |
| 2797 | retry: |
| 2798 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
| 2799 | tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page); |
| 2800 | done = false; |
| 2801 | blk_start_plug(&plug); |
| 2802 | |
| 2803 | /* |
| 2804 | * First writeback pages that don't need mapping - we can avoid |
| 2805 | * starting a transaction unnecessarily and also avoid being blocked |
| 2806 | * in the block layer on device congestion while having transaction |
| 2807 | * started. |
| 2808 | */ |
| 2809 | mpd.do_map = 0; |
| 2810 | mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); |
| 2811 | if (!mpd.io_submit.io_end) { |
| 2812 | ret = -ENOMEM; |
| 2813 | goto unplug; |
| 2814 | } |
| 2815 | ret = mpage_prepare_extent_to_map(&mpd); |
| 2816 | /* Submit prepared bio */ |
| 2817 | ext4_io_submit(&mpd.io_submit); |
| 2818 | ext4_put_io_end_defer(mpd.io_submit.io_end); |
| 2819 | mpd.io_submit.io_end = NULL; |
| 2820 | /* Unlock pages we didn't use */ |
| 2821 | mpage_release_unused_pages(&mpd, false); |
| 2822 | if (ret < 0) |
| 2823 | goto unplug; |
| 2824 | |
| 2825 | while (!done && mpd.first_page <= mpd.last_page) { |
| 2826 | /* For each extent of pages we use new io_end */ |
| 2827 | mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); |
| 2828 | if (!mpd.io_submit.io_end) { |
| 2829 | ret = -ENOMEM; |
| 2830 | break; |
| 2831 | } |
| 2832 | |
| 2833 | /* |
| 2834 | * We have two constraints: We find one extent to map and we |
| 2835 | * must always write out whole page (makes a difference when |
| 2836 | * blocksize < pagesize) so that we don't block on IO when we |
| 2837 | * try to write out the rest of the page. Journalled mode is |
| 2838 | * not supported by delalloc. |
| 2839 | */ |
| 2840 | BUG_ON(ext4_should_journal_data(inode)); |
| 2841 | needed_blocks = ext4_da_writepages_trans_blocks(inode); |
| 2842 | |
| 2843 | /* start a new transaction */ |
| 2844 | handle = ext4_journal_start_with_reserve(inode, |
| 2845 | EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks); |
| 2846 | if (IS_ERR(handle)) { |
| 2847 | ret = PTR_ERR(handle); |
| 2848 | ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " |
| 2849 | "%ld pages, ino %lu; err %d", __func__, |
| 2850 | wbc->nr_to_write, inode->i_ino, ret); |
| 2851 | /* Release allocated io_end */ |
| 2852 | ext4_put_io_end(mpd.io_submit.io_end); |
| 2853 | mpd.io_submit.io_end = NULL; |
| 2854 | break; |
| 2855 | } |
| 2856 | mpd.do_map = 1; |
| 2857 | |
| 2858 | trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc); |
| 2859 | ret = mpage_prepare_extent_to_map(&mpd); |
| 2860 | if (!ret) { |
| 2861 | if (mpd.map.m_len) |
| 2862 | ret = mpage_map_and_submit_extent(handle, &mpd, |
| 2863 | &give_up_on_write); |
| 2864 | else { |
| 2865 | /* |
| 2866 | * We scanned the whole range (or exhausted |
| 2867 | * nr_to_write), submitted what was mapped and |
| 2868 | * didn't find anything needing mapping. We are |
| 2869 | * done. |
| 2870 | */ |
| 2871 | done = true; |
| 2872 | } |
| 2873 | } |
| 2874 | /* |
| 2875 | * Caution: If the handle is synchronous, |
| 2876 | * ext4_journal_stop() can wait for transaction commit |
| 2877 | * to finish which may depend on writeback of pages to |
| 2878 | * complete or on page lock to be released. In that |
| 2879 | * case, we have to wait until after after we have |
| 2880 | * submitted all the IO, released page locks we hold, |
| 2881 | * and dropped io_end reference (for extent conversion |
| 2882 | * to be able to complete) before stopping the handle. |
| 2883 | */ |
| 2884 | if (!ext4_handle_valid(handle) || handle->h_sync == 0) { |
| 2885 | ext4_journal_stop(handle); |
| 2886 | handle = NULL; |
| 2887 | mpd.do_map = 0; |
| 2888 | } |
| 2889 | /* Submit prepared bio */ |
| 2890 | ext4_io_submit(&mpd.io_submit); |
| 2891 | /* Unlock pages we didn't use */ |
| 2892 | mpage_release_unused_pages(&mpd, give_up_on_write); |
| 2893 | /* |
| 2894 | * Drop our io_end reference we got from init. We have |
| 2895 | * to be careful and use deferred io_end finishing if |
| 2896 | * we are still holding the transaction as we can |
| 2897 | * release the last reference to io_end which may end |
| 2898 | * up doing unwritten extent conversion. |
| 2899 | */ |
| 2900 | if (handle) { |
| 2901 | ext4_put_io_end_defer(mpd.io_submit.io_end); |
| 2902 | ext4_journal_stop(handle); |
| 2903 | } else |
| 2904 | ext4_put_io_end(mpd.io_submit.io_end); |
| 2905 | mpd.io_submit.io_end = NULL; |
| 2906 | |
| 2907 | if (ret == -ENOSPC && sbi->s_journal) { |
| 2908 | /* |
| 2909 | * Commit the transaction which would |
| 2910 | * free blocks released in the transaction |
| 2911 | * and try again |
| 2912 | */ |
| 2913 | jbd2_journal_force_commit_nested(sbi->s_journal); |
| 2914 | ret = 0; |
| 2915 | continue; |
| 2916 | } |
| 2917 | /* Fatal error - ENOMEM, EIO... */ |
| 2918 | if (ret) |
| 2919 | break; |
| 2920 | } |
| 2921 | unplug: |
| 2922 | blk_finish_plug(&plug); |
| 2923 | if (!ret && !cycled && wbc->nr_to_write > 0) { |
| 2924 | cycled = 1; |
| 2925 | mpd.last_page = writeback_index - 1; |
| 2926 | mpd.first_page = 0; |
| 2927 | goto retry; |
| 2928 | } |
| 2929 | |
| 2930 | /* Update index */ |
| 2931 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
| 2932 | /* |
| 2933 | * Set the writeback_index so that range_cyclic |
| 2934 | * mode will write it back later |
| 2935 | */ |
| 2936 | mapping->writeback_index = mpd.first_page; |
| 2937 | |
| 2938 | out_writepages: |
| 2939 | trace_ext4_writepages_result(inode, wbc, ret, |
| 2940 | nr_to_write - wbc->nr_to_write); |
| 2941 | percpu_up_read(&sbi->s_journal_flag_rwsem); |
| 2942 | return ret; |
| 2943 | } |
| 2944 | |
| 2945 | static int ext4_nonda_switch(struct super_block *sb) |
| 2946 | { |
| 2947 | s64 free_clusters, dirty_clusters; |
| 2948 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
| 2949 | |
| 2950 | /* |
| 2951 | * switch to non delalloc mode if we are running low |
| 2952 | * on free block. The free block accounting via percpu |
| 2953 | * counters can get slightly wrong with percpu_counter_batch getting |
| 2954 | * accumulated on each CPU without updating global counters |
| 2955 | * Delalloc need an accurate free block accounting. So switch |
| 2956 | * to non delalloc when we are near to error range. |
| 2957 | */ |
| 2958 | free_clusters = |
| 2959 | percpu_counter_read_positive(&sbi->s_freeclusters_counter); |
| 2960 | dirty_clusters = |
| 2961 | percpu_counter_read_positive(&sbi->s_dirtyclusters_counter); |
| 2962 | /* |
| 2963 | * Start pushing delalloc when 1/2 of free blocks are dirty. |
| 2964 | */ |
| 2965 | if (dirty_clusters && (free_clusters < 2 * dirty_clusters)) |
| 2966 | try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE); |
| 2967 | |
| 2968 | if (2 * free_clusters < 3 * dirty_clusters || |
| 2969 | free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) { |
| 2970 | /* |
| 2971 | * free block count is less than 150% of dirty blocks |
| 2972 | * or free blocks is less than watermark |
| 2973 | */ |
| 2974 | return 1; |
| 2975 | } |
| 2976 | return 0; |
| 2977 | } |
| 2978 | |
| 2979 | /* We always reserve for an inode update; the superblock could be there too */ |
| 2980 | static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len) |
| 2981 | { |
| 2982 | if (likely(ext4_has_feature_large_file(inode->i_sb))) |
| 2983 | return 1; |
| 2984 | |
| 2985 | if (pos + len <= 0x7fffffffULL) |
| 2986 | return 1; |
| 2987 | |
| 2988 | /* We might need to update the superblock to set LARGE_FILE */ |
| 2989 | return 2; |
| 2990 | } |
| 2991 | |
| 2992 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, |
| 2993 | loff_t pos, unsigned len, unsigned flags, |
| 2994 | struct page **pagep, void **fsdata) |
| 2995 | { |
| 2996 | int ret, retries = 0; |
| 2997 | struct page *page; |
| 2998 | pgoff_t index; |
| 2999 | struct inode *inode = mapping->host; |
| 3000 | handle_t *handle; |
| 3001 | |
| 3002 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
| 3003 | return -EIO; |
| 3004 | |
| 3005 | index = pos >> PAGE_SHIFT; |
| 3006 | |
| 3007 | if (ext4_nonda_switch(inode->i_sb) || |
| 3008 | S_ISLNK(inode->i_mode)) { |
| 3009 | *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; |
| 3010 | return ext4_write_begin(file, mapping, pos, |
| 3011 | len, flags, pagep, fsdata); |
| 3012 | } |
| 3013 | *fsdata = (void *)0; |
| 3014 | trace_ext4_da_write_begin(inode, pos, len, flags); |
| 3015 | |
| 3016 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { |
| 3017 | ret = ext4_da_write_inline_data_begin(mapping, inode, |
| 3018 | pos, len, flags, |
| 3019 | pagep, fsdata); |
| 3020 | if (ret < 0) |
| 3021 | return ret; |
| 3022 | if (ret == 1) |
| 3023 | return 0; |
| 3024 | } |
| 3025 | |
| 3026 | /* |
| 3027 | * grab_cache_page_write_begin() can take a long time if the |
| 3028 | * system is thrashing due to memory pressure, or if the page |
| 3029 | * is being written back. So grab it first before we start |
| 3030 | * the transaction handle. This also allows us to allocate |
| 3031 | * the page (if needed) without using GFP_NOFS. |
| 3032 | */ |
| 3033 | retry_grab: |
| 3034 | page = grab_cache_page_write_begin(mapping, index, flags); |
| 3035 | if (!page) |
| 3036 | return -ENOMEM; |
| 3037 | unlock_page(page); |
| 3038 | |
| 3039 | /* |
| 3040 | * With delayed allocation, we don't log the i_disksize update |
| 3041 | * if there is delayed block allocation. But we still need |
| 3042 | * to journalling the i_disksize update if writes to the end |
| 3043 | * of file which has an already mapped buffer. |
| 3044 | */ |
| 3045 | retry_journal: |
| 3046 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
| 3047 | ext4_da_write_credits(inode, pos, len)); |
| 3048 | if (IS_ERR(handle)) { |
| 3049 | put_page(page); |
| 3050 | return PTR_ERR(handle); |
| 3051 | } |
| 3052 | |
| 3053 | lock_page(page); |
| 3054 | if (page->mapping != mapping) { |
| 3055 | /* The page got truncated from under us */ |
| 3056 | unlock_page(page); |
| 3057 | put_page(page); |
| 3058 | ext4_journal_stop(handle); |
| 3059 | goto retry_grab; |
| 3060 | } |
| 3061 | /* In case writeback began while the page was unlocked */ |
| 3062 | wait_for_stable_page(page); |
| 3063 | |
| 3064 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
| 3065 | ret = ext4_block_write_begin(page, pos, len, |
| 3066 | ext4_da_get_block_prep); |
| 3067 | #else |
| 3068 | ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep); |
| 3069 | #endif |
| 3070 | if (ret < 0) { |
| 3071 | unlock_page(page); |
| 3072 | ext4_journal_stop(handle); |
| 3073 | /* |
| 3074 | * block_write_begin may have instantiated a few blocks |
| 3075 | * outside i_size. Trim these off again. Don't need |
| 3076 | * i_size_read because we hold i_mutex. |
| 3077 | */ |
| 3078 | if (pos + len > inode->i_size) |
| 3079 | ext4_truncate_failed_write(inode); |
| 3080 | |
| 3081 | if (ret == -ENOSPC && |
| 3082 | ext4_should_retry_alloc(inode->i_sb, &retries)) |
| 3083 | goto retry_journal; |
| 3084 | |
| 3085 | put_page(page); |
| 3086 | return ret; |
| 3087 | } |
| 3088 | |
| 3089 | *pagep = page; |
| 3090 | return ret; |
| 3091 | } |
| 3092 | |
| 3093 | /* |
| 3094 | * Check if we should update i_disksize |
| 3095 | * when write to the end of file but not require block allocation |
| 3096 | */ |
| 3097 | static int ext4_da_should_update_i_disksize(struct page *page, |
| 3098 | unsigned long offset) |
| 3099 | { |
| 3100 | struct buffer_head *bh; |
| 3101 | struct inode *inode = page->mapping->host; |
| 3102 | unsigned int idx; |
| 3103 | int i; |
| 3104 | |
| 3105 | bh = page_buffers(page); |
| 3106 | idx = offset >> inode->i_blkbits; |
| 3107 | |
| 3108 | for (i = 0; i < idx; i++) |
| 3109 | bh = bh->b_this_page; |
| 3110 | |
| 3111 | if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
| 3112 | return 0; |
| 3113 | return 1; |
| 3114 | } |
| 3115 | |
| 3116 | static int ext4_da_write_end(struct file *file, |
| 3117 | struct address_space *mapping, |
| 3118 | loff_t pos, unsigned len, unsigned copied, |
| 3119 | struct page *page, void *fsdata) |
| 3120 | { |
| 3121 | struct inode *inode = mapping->host; |
| 3122 | int ret = 0, ret2; |
| 3123 | handle_t *handle = ext4_journal_current_handle(); |
| 3124 | loff_t new_i_size; |
| 3125 | unsigned long start, end; |
| 3126 | int write_mode = (int)(unsigned long)fsdata; |
| 3127 | |
| 3128 | if (write_mode == FALL_BACK_TO_NONDELALLOC) |
| 3129 | return ext4_write_end(file, mapping, pos, |
| 3130 | len, copied, page, fsdata); |
| 3131 | |
| 3132 | trace_ext4_da_write_end(inode, pos, len, copied); |
| 3133 | start = pos & (PAGE_SIZE - 1); |
| 3134 | end = start + copied - 1; |
| 3135 | |
| 3136 | /* |
| 3137 | * generic_write_end() will run mark_inode_dirty() if i_size |
| 3138 | * changes. So let's piggyback the i_disksize mark_inode_dirty |
| 3139 | * into that. |
| 3140 | */ |
| 3141 | new_i_size = pos + copied; |
| 3142 | if (copied && new_i_size > EXT4_I(inode)->i_disksize) { |
| 3143 | if (ext4_has_inline_data(inode) || |
| 3144 | ext4_da_should_update_i_disksize(page, end)) { |
| 3145 | ext4_update_i_disksize(inode, new_i_size); |
| 3146 | /* We need to mark inode dirty even if |
| 3147 | * new_i_size is less that inode->i_size |
| 3148 | * bu greater than i_disksize.(hint delalloc) |
| 3149 | */ |
| 3150 | ext4_mark_inode_dirty(handle, inode); |
| 3151 | } |
| 3152 | } |
| 3153 | |
| 3154 | if (write_mode != CONVERT_INLINE_DATA && |
| 3155 | ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) && |
| 3156 | ext4_has_inline_data(inode)) |
| 3157 | ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied, |
| 3158 | page); |
| 3159 | else |
| 3160 | ret2 = generic_write_end(file, mapping, pos, len, copied, |
| 3161 | page, fsdata); |
| 3162 | |
| 3163 | copied = ret2; |
| 3164 | if (ret2 < 0) |
| 3165 | ret = ret2; |
| 3166 | ret2 = ext4_journal_stop(handle); |
| 3167 | if (!ret) |
| 3168 | ret = ret2; |
| 3169 | |
| 3170 | return ret ? ret : copied; |
| 3171 | } |
| 3172 | |
| 3173 | static void ext4_da_invalidatepage(struct page *page, unsigned int offset, |
| 3174 | unsigned int length) |
| 3175 | { |
| 3176 | /* |
| 3177 | * Drop reserved blocks |
| 3178 | */ |
| 3179 | BUG_ON(!PageLocked(page)); |
| 3180 | if (!page_has_buffers(page)) |
| 3181 | goto out; |
| 3182 | |
| 3183 | ext4_da_page_release_reservation(page, offset, length); |
| 3184 | |
| 3185 | out: |
| 3186 | ext4_invalidatepage(page, offset, length); |
| 3187 | |
| 3188 | return; |
| 3189 | } |
| 3190 | |
| 3191 | /* |
| 3192 | * Force all delayed allocation blocks to be allocated for a given inode. |
| 3193 | */ |
| 3194 | int ext4_alloc_da_blocks(struct inode *inode) |
| 3195 | { |
| 3196 | trace_ext4_alloc_da_blocks(inode); |
| 3197 | |
| 3198 | if (!EXT4_I(inode)->i_reserved_data_blocks) |
| 3199 | return 0; |
| 3200 | |
| 3201 | /* |
| 3202 | * We do something simple for now. The filemap_flush() will |
| 3203 | * also start triggering a write of the data blocks, which is |
| 3204 | * not strictly speaking necessary (and for users of |
| 3205 | * laptop_mode, not even desirable). However, to do otherwise |
| 3206 | * would require replicating code paths in: |
| 3207 | * |
| 3208 | * ext4_writepages() -> |
| 3209 | * write_cache_pages() ---> (via passed in callback function) |
| 3210 | * __mpage_da_writepage() --> |
| 3211 | * mpage_add_bh_to_extent() |
| 3212 | * mpage_da_map_blocks() |
| 3213 | * |
| 3214 | * The problem is that write_cache_pages(), located in |
| 3215 | * mm/page-writeback.c, marks pages clean in preparation for |
| 3216 | * doing I/O, which is not desirable if we're not planning on |
| 3217 | * doing I/O at all. |
| 3218 | * |
| 3219 | * We could call write_cache_pages(), and then redirty all of |
| 3220 | * the pages by calling redirty_page_for_writepage() but that |
| 3221 | * would be ugly in the extreme. So instead we would need to |
| 3222 | * replicate parts of the code in the above functions, |
| 3223 | * simplifying them because we wouldn't actually intend to |
| 3224 | * write out the pages, but rather only collect contiguous |
| 3225 | * logical block extents, call the multi-block allocator, and |
| 3226 | * then update the buffer heads with the block allocations. |
| 3227 | * |
| 3228 | * For now, though, we'll cheat by calling filemap_flush(), |
| 3229 | * which will map the blocks, and start the I/O, but not |
| 3230 | * actually wait for the I/O to complete. |
| 3231 | */ |
| 3232 | return filemap_flush(inode->i_mapping); |
| 3233 | } |
| 3234 | |
| 3235 | /* |
| 3236 | * bmap() is special. It gets used by applications such as lilo and by |
| 3237 | * the swapper to find the on-disk block of a specific piece of data. |
| 3238 | * |
| 3239 | * Naturally, this is dangerous if the block concerned is still in the |
| 3240 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
| 3241 | * filesystem and enables swap, then they may get a nasty shock when the |
| 3242 | * data getting swapped to that swapfile suddenly gets overwritten by |
| 3243 | * the original zero's written out previously to the journal and |
| 3244 | * awaiting writeback in the kernel's buffer cache. |
| 3245 | * |
| 3246 | * So, if we see any bmap calls here on a modified, data-journaled file, |
| 3247 | * take extra steps to flush any blocks which might be in the cache. |
| 3248 | */ |
| 3249 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
| 3250 | { |
| 3251 | struct inode *inode = mapping->host; |
| 3252 | journal_t *journal; |
| 3253 | int err; |
| 3254 | |
| 3255 | /* |
| 3256 | * We can get here for an inline file via the FIBMAP ioctl |
| 3257 | */ |
| 3258 | if (ext4_has_inline_data(inode)) |
| 3259 | return 0; |
| 3260 | |
| 3261 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
| 3262 | test_opt(inode->i_sb, DELALLOC)) { |
| 3263 | /* |
| 3264 | * With delalloc we want to sync the file |
| 3265 | * so that we can make sure we allocate |
| 3266 | * blocks for file |
| 3267 | */ |
| 3268 | filemap_write_and_wait(mapping); |
| 3269 | } |
| 3270 | |
| 3271 | if (EXT4_JOURNAL(inode) && |
| 3272 | ext4_test_inode_state(inode, EXT4_STATE_JDATA)) { |
| 3273 | /* |
| 3274 | * This is a REALLY heavyweight approach, but the use of |
| 3275 | * bmap on dirty files is expected to be extremely rare: |
| 3276 | * only if we run lilo or swapon on a freshly made file |
| 3277 | * do we expect this to happen. |
| 3278 | * |
| 3279 | * (bmap requires CAP_SYS_RAWIO so this does not |
| 3280 | * represent an unprivileged user DOS attack --- we'd be |
| 3281 | * in trouble if mortal users could trigger this path at |
| 3282 | * will.) |
| 3283 | * |
| 3284 | * NB. EXT4_STATE_JDATA is not set on files other than |
| 3285 | * regular files. If somebody wants to bmap a directory |
| 3286 | * or symlink and gets confused because the buffer |
| 3287 | * hasn't yet been flushed to disk, they deserve |
| 3288 | * everything they get. |
| 3289 | */ |
| 3290 | |
| 3291 | ext4_clear_inode_state(inode, EXT4_STATE_JDATA); |
| 3292 | journal = EXT4_JOURNAL(inode); |
| 3293 | jbd2_journal_lock_updates(journal); |
| 3294 | err = jbd2_journal_flush(journal); |
| 3295 | jbd2_journal_unlock_updates(journal); |
| 3296 | |
| 3297 | if (err) |
| 3298 | return 0; |
| 3299 | } |
| 3300 | |
| 3301 | return generic_block_bmap(mapping, block, ext4_get_block); |
| 3302 | } |
| 3303 | |
| 3304 | static int ext4_readpage(struct file *file, struct page *page) |
| 3305 | { |
| 3306 | int ret = -EAGAIN; |
| 3307 | struct inode *inode = page->mapping->host; |
| 3308 | |
| 3309 | trace_ext4_readpage(page); |
| 3310 | |
| 3311 | if (ext4_has_inline_data(inode)) |
| 3312 | ret = ext4_readpage_inline(inode, page); |
| 3313 | |
| 3314 | if (ret == -EAGAIN) |
| 3315 | return ext4_mpage_readpages(page->mapping, NULL, page, 1); |
| 3316 | |
| 3317 | return ret; |
| 3318 | } |
| 3319 | |
| 3320 | static int |
| 3321 | ext4_readpages(struct file *file, struct address_space *mapping, |
| 3322 | struct list_head *pages, unsigned nr_pages) |
| 3323 | { |
| 3324 | struct inode *inode = mapping->host; |
| 3325 | |
| 3326 | /* If the file has inline data, no need to do readpages. */ |
| 3327 | if (ext4_has_inline_data(inode)) |
| 3328 | return 0; |
| 3329 | |
| 3330 | return ext4_mpage_readpages(mapping, pages, NULL, nr_pages); |
| 3331 | } |
| 3332 | |
| 3333 | static void ext4_invalidatepage(struct page *page, unsigned int offset, |
| 3334 | unsigned int length) |
| 3335 | { |
| 3336 | trace_ext4_invalidatepage(page, offset, length); |
| 3337 | |
| 3338 | /* No journalling happens on data buffers when this function is used */ |
| 3339 | WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page))); |
| 3340 | |
| 3341 | block_invalidatepage(page, offset, length); |
| 3342 | } |
| 3343 | |
| 3344 | static int __ext4_journalled_invalidatepage(struct page *page, |
| 3345 | unsigned int offset, |
| 3346 | unsigned int length) |
| 3347 | { |
| 3348 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
| 3349 | |
| 3350 | trace_ext4_journalled_invalidatepage(page, offset, length); |
| 3351 | |
| 3352 | /* |
| 3353 | * If it's a full truncate we just forget about the pending dirtying |
| 3354 | */ |
| 3355 | if (offset == 0 && length == PAGE_SIZE) |
| 3356 | ClearPageChecked(page); |
| 3357 | |
| 3358 | return jbd2_journal_invalidatepage(journal, page, offset, length); |
| 3359 | } |
| 3360 | |
| 3361 | /* Wrapper for aops... */ |
| 3362 | static void ext4_journalled_invalidatepage(struct page *page, |
| 3363 | unsigned int offset, |
| 3364 | unsigned int length) |
| 3365 | { |
| 3366 | WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0); |
| 3367 | } |
| 3368 | |
| 3369 | static int ext4_releasepage(struct page *page, gfp_t wait) |
| 3370 | { |
| 3371 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
| 3372 | |
| 3373 | trace_ext4_releasepage(page); |
| 3374 | |
| 3375 | /* Page has dirty journalled data -> cannot release */ |
| 3376 | if (PageChecked(page)) |
| 3377 | return 0; |
| 3378 | if (journal) |
| 3379 | return jbd2_journal_try_to_free_buffers(journal, page, wait); |
| 3380 | else |
| 3381 | return try_to_free_buffers(page); |
| 3382 | } |
| 3383 | |
| 3384 | #ifdef CONFIG_FS_DAX |
| 3385 | static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length, |
| 3386 | unsigned flags, struct iomap *iomap) |
| 3387 | { |
| 3388 | struct block_device *bdev; |
| 3389 | unsigned int blkbits = inode->i_blkbits; |
| 3390 | unsigned long first_block = offset >> blkbits; |
| 3391 | unsigned long last_block = (offset + length - 1) >> blkbits; |
| 3392 | struct ext4_map_blocks map; |
| 3393 | int ret; |
| 3394 | |
| 3395 | if (WARN_ON_ONCE(ext4_has_inline_data(inode))) |
| 3396 | return -ERANGE; |
| 3397 | |
| 3398 | map.m_lblk = first_block; |
| 3399 | map.m_len = last_block - first_block + 1; |
| 3400 | |
| 3401 | if (!(flags & IOMAP_WRITE)) { |
| 3402 | ret = ext4_map_blocks(NULL, inode, &map, 0); |
| 3403 | } else { |
| 3404 | int dio_credits; |
| 3405 | handle_t *handle; |
| 3406 | int retries = 0; |
| 3407 | |
| 3408 | /* Trim mapping request to maximum we can map at once for DIO */ |
| 3409 | if (map.m_len > DIO_MAX_BLOCKS) |
| 3410 | map.m_len = DIO_MAX_BLOCKS; |
| 3411 | dio_credits = ext4_chunk_trans_blocks(inode, map.m_len); |
| 3412 | retry: |
| 3413 | /* |
| 3414 | * Either we allocate blocks and then we don't get unwritten |
| 3415 | * extent so we have reserved enough credits, or the blocks |
| 3416 | * are already allocated and unwritten and in that case |
| 3417 | * extent conversion fits in the credits as well. |
| 3418 | */ |
| 3419 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, |
| 3420 | dio_credits); |
| 3421 | if (IS_ERR(handle)) |
| 3422 | return PTR_ERR(handle); |
| 3423 | |
| 3424 | ret = ext4_map_blocks(handle, inode, &map, |
| 3425 | EXT4_GET_BLOCKS_CREATE_ZERO); |
| 3426 | if (ret < 0) { |
| 3427 | ext4_journal_stop(handle); |
| 3428 | if (ret == -ENOSPC && |
| 3429 | ext4_should_retry_alloc(inode->i_sb, &retries)) |
| 3430 | goto retry; |
| 3431 | return ret; |
| 3432 | } |
| 3433 | |
| 3434 | /* |
| 3435 | * If we added blocks beyond i_size, we need to make sure they |
| 3436 | * will get truncated if we crash before updating i_size in |
| 3437 | * ext4_iomap_end(). For faults we don't need to do that (and |
| 3438 | * even cannot because for orphan list operations inode_lock is |
| 3439 | * required) - if we happen to instantiate block beyond i_size, |
| 3440 | * it is because we race with truncate which has already added |
| 3441 | * the inode to the orphan list. |
| 3442 | */ |
| 3443 | if (!(flags & IOMAP_FAULT) && first_block + map.m_len > |
| 3444 | (i_size_read(inode) + (1 << blkbits) - 1) >> blkbits) { |
| 3445 | int err; |
| 3446 | |
| 3447 | err = ext4_orphan_add(handle, inode); |
| 3448 | if (err < 0) { |
| 3449 | ext4_journal_stop(handle); |
| 3450 | return err; |
| 3451 | } |
| 3452 | } |
| 3453 | ext4_journal_stop(handle); |
| 3454 | } |
| 3455 | |
| 3456 | iomap->flags = 0; |
| 3457 | bdev = inode->i_sb->s_bdev; |
| 3458 | iomap->bdev = bdev; |
| 3459 | if (blk_queue_dax(bdev->bd_queue)) |
| 3460 | iomap->dax_dev = fs_dax_get_by_host(bdev->bd_disk->disk_name); |
| 3461 | else |
| 3462 | iomap->dax_dev = NULL; |
| 3463 | iomap->offset = first_block << blkbits; |
| 3464 | |
| 3465 | if (ret == 0) { |
| 3466 | iomap->type = IOMAP_HOLE; |
| 3467 | iomap->blkno = IOMAP_NULL_BLOCK; |
| 3468 | iomap->length = (u64)map.m_len << blkbits; |
| 3469 | } else { |
| 3470 | if (map.m_flags & EXT4_MAP_MAPPED) { |
| 3471 | iomap->type = IOMAP_MAPPED; |
| 3472 | } else if (map.m_flags & EXT4_MAP_UNWRITTEN) { |
| 3473 | iomap->type = IOMAP_UNWRITTEN; |
| 3474 | } else { |
| 3475 | WARN_ON_ONCE(1); |
| 3476 | return -EIO; |
| 3477 | } |
| 3478 | iomap->blkno = (sector_t)map.m_pblk << (blkbits - 9); |
| 3479 | iomap->length = (u64)map.m_len << blkbits; |
| 3480 | } |
| 3481 | |
| 3482 | if (map.m_flags & EXT4_MAP_NEW) |
| 3483 | iomap->flags |= IOMAP_F_NEW; |
| 3484 | return 0; |
| 3485 | } |
| 3486 | |
| 3487 | static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length, |
| 3488 | ssize_t written, unsigned flags, struct iomap *iomap) |
| 3489 | { |
| 3490 | int ret = 0; |
| 3491 | handle_t *handle; |
| 3492 | int blkbits = inode->i_blkbits; |
| 3493 | bool truncate = false; |
| 3494 | |
| 3495 | fs_put_dax(iomap->dax_dev); |
| 3496 | if (!(flags & IOMAP_WRITE) || (flags & IOMAP_FAULT)) |
| 3497 | return 0; |
| 3498 | |
| 3499 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); |
| 3500 | if (IS_ERR(handle)) { |
| 3501 | ret = PTR_ERR(handle); |
| 3502 | goto orphan_del; |
| 3503 | } |
| 3504 | if (ext4_update_inode_size(inode, offset + written)) |
| 3505 | ext4_mark_inode_dirty(handle, inode); |
| 3506 | /* |
| 3507 | * We may need to truncate allocated but not written blocks beyond EOF. |
| 3508 | */ |
| 3509 | if (iomap->offset + iomap->length > |
| 3510 | ALIGN(inode->i_size, 1 << blkbits)) { |
| 3511 | ext4_lblk_t written_blk, end_blk; |
| 3512 | |
| 3513 | written_blk = (offset + written) >> blkbits; |
| 3514 | end_blk = (offset + length) >> blkbits; |
| 3515 | if (written_blk < end_blk && ext4_can_truncate(inode)) |
| 3516 | truncate = true; |
| 3517 | } |
| 3518 | /* |
| 3519 | * Remove inode from orphan list if we were extending a inode and |
| 3520 | * everything went fine. |
| 3521 | */ |
| 3522 | if (!truncate && inode->i_nlink && |
| 3523 | !list_empty(&EXT4_I(inode)->i_orphan)) |
| 3524 | ext4_orphan_del(handle, inode); |
| 3525 | ext4_journal_stop(handle); |
| 3526 | if (truncate) { |
| 3527 | ext4_truncate_failed_write(inode); |
| 3528 | orphan_del: |
| 3529 | /* |
| 3530 | * If truncate failed early the inode might still be on the |
| 3531 | * orphan list; we need to make sure the inode is removed from |
| 3532 | * the orphan list in that case. |
| 3533 | */ |
| 3534 | if (inode->i_nlink) |
| 3535 | ext4_orphan_del(NULL, inode); |
| 3536 | } |
| 3537 | return ret; |
| 3538 | } |
| 3539 | |
| 3540 | const struct iomap_ops ext4_iomap_ops = { |
| 3541 | .iomap_begin = ext4_iomap_begin, |
| 3542 | .iomap_end = ext4_iomap_end, |
| 3543 | }; |
| 3544 | |
| 3545 | #endif |
| 3546 | |
| 3547 | static int ext4_end_io_dio(struct kiocb *iocb, loff_t offset, |
| 3548 | ssize_t size, void *private) |
| 3549 | { |
| 3550 | ext4_io_end_t *io_end = private; |
| 3551 | |
| 3552 | /* if not async direct IO just return */ |
| 3553 | if (!io_end) |
| 3554 | return 0; |
| 3555 | |
| 3556 | ext_debug("ext4_end_io_dio(): io_end 0x%p " |
| 3557 | "for inode %lu, iocb 0x%p, offset %llu, size %zd\n", |
| 3558 | io_end, io_end->inode->i_ino, iocb, offset, size); |
| 3559 | |
| 3560 | /* |
| 3561 | * Error during AIO DIO. We cannot convert unwritten extents as the |
| 3562 | * data was not written. Just clear the unwritten flag and drop io_end. |
| 3563 | */ |
| 3564 | if (size <= 0) { |
| 3565 | ext4_clear_io_unwritten_flag(io_end); |
| 3566 | size = 0; |
| 3567 | } |
| 3568 | io_end->offset = offset; |
| 3569 | io_end->size = size; |
| 3570 | ext4_put_io_end(io_end); |
| 3571 | |
| 3572 | return 0; |
| 3573 | } |
| 3574 | |
| 3575 | /* |
| 3576 | * Handling of direct IO writes. |
| 3577 | * |
| 3578 | * For ext4 extent files, ext4 will do direct-io write even to holes, |
| 3579 | * preallocated extents, and those write extend the file, no need to |
| 3580 | * fall back to buffered IO. |
| 3581 | * |
| 3582 | * For holes, we fallocate those blocks, mark them as unwritten |
| 3583 | * If those blocks were preallocated, we mark sure they are split, but |
| 3584 | * still keep the range to write as unwritten. |
| 3585 | * |
| 3586 | * The unwritten extents will be converted to written when DIO is completed. |
| 3587 | * For async direct IO, since the IO may still pending when return, we |
| 3588 | * set up an end_io call back function, which will do the conversion |
| 3589 | * when async direct IO completed. |
| 3590 | * |
| 3591 | * If the O_DIRECT write will extend the file then add this inode to the |
| 3592 | * orphan list. So recovery will truncate it back to the original size |
| 3593 | * if the machine crashes during the write. |
| 3594 | * |
| 3595 | */ |
| 3596 | static ssize_t ext4_direct_IO_write(struct kiocb *iocb, struct iov_iter *iter) |
| 3597 | { |
| 3598 | struct file *file = iocb->ki_filp; |
| 3599 | struct inode *inode = file->f_mapping->host; |
| 3600 | struct ext4_inode_info *ei = EXT4_I(inode); |
| 3601 | ssize_t ret; |
| 3602 | loff_t offset = iocb->ki_pos; |
| 3603 | size_t count = iov_iter_count(iter); |
| 3604 | int overwrite = 0; |
| 3605 | get_block_t *get_block_func = NULL; |
| 3606 | int dio_flags = 0; |
| 3607 | loff_t final_size = offset + count; |
| 3608 | int orphan = 0; |
| 3609 | handle_t *handle; |
| 3610 | |
| 3611 | if (final_size > inode->i_size) { |
| 3612 | /* Credits for sb + inode write */ |
| 3613 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); |
| 3614 | if (IS_ERR(handle)) { |
| 3615 | ret = PTR_ERR(handle); |
| 3616 | goto out; |
| 3617 | } |
| 3618 | ret = ext4_orphan_add(handle, inode); |
| 3619 | if (ret) { |
| 3620 | ext4_journal_stop(handle); |
| 3621 | goto out; |
| 3622 | } |
| 3623 | orphan = 1; |
| 3624 | ei->i_disksize = inode->i_size; |
| 3625 | ext4_journal_stop(handle); |
| 3626 | } |
| 3627 | |
| 3628 | BUG_ON(iocb->private == NULL); |
| 3629 | |
| 3630 | /* |
| 3631 | * Make all waiters for direct IO properly wait also for extent |
| 3632 | * conversion. This also disallows race between truncate() and |
| 3633 | * overwrite DIO as i_dio_count needs to be incremented under i_mutex. |
| 3634 | */ |
| 3635 | inode_dio_begin(inode); |
| 3636 | |
| 3637 | /* If we do a overwrite dio, i_mutex locking can be released */ |
| 3638 | overwrite = *((int *)iocb->private); |
| 3639 | |
| 3640 | if (overwrite) |
| 3641 | inode_unlock(inode); |
| 3642 | |
| 3643 | /* |
| 3644 | * For extent mapped files we could direct write to holes and fallocate. |
| 3645 | * |
| 3646 | * Allocated blocks to fill the hole are marked as unwritten to prevent |
| 3647 | * parallel buffered read to expose the stale data before DIO complete |
| 3648 | * the data IO. |
| 3649 | * |
| 3650 | * As to previously fallocated extents, ext4 get_block will just simply |
| 3651 | * mark the buffer mapped but still keep the extents unwritten. |
| 3652 | * |
| 3653 | * For non AIO case, we will convert those unwritten extents to written |
| 3654 | * after return back from blockdev_direct_IO. That way we save us from |
| 3655 | * allocating io_end structure and also the overhead of offloading |
| 3656 | * the extent convertion to a workqueue. |
| 3657 | * |
| 3658 | * For async DIO, the conversion needs to be deferred when the |
| 3659 | * IO is completed. The ext4 end_io callback function will be |
| 3660 | * called to take care of the conversion work. Here for async |
| 3661 | * case, we allocate an io_end structure to hook to the iocb. |
| 3662 | */ |
| 3663 | iocb->private = NULL; |
| 3664 | if (overwrite) |
| 3665 | get_block_func = ext4_dio_get_block_overwrite; |
| 3666 | else if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) || |
| 3667 | round_down(offset, i_blocksize(inode)) >= inode->i_size) { |
| 3668 | get_block_func = ext4_dio_get_block; |
| 3669 | dio_flags = DIO_LOCKING | DIO_SKIP_HOLES; |
| 3670 | } else if (is_sync_kiocb(iocb)) { |
| 3671 | get_block_func = ext4_dio_get_block_unwritten_sync; |
| 3672 | dio_flags = DIO_LOCKING; |
| 3673 | } else { |
| 3674 | get_block_func = ext4_dio_get_block_unwritten_async; |
| 3675 | dio_flags = DIO_LOCKING; |
| 3676 | } |
| 3677 | ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, iter, |
| 3678 | get_block_func, ext4_end_io_dio, NULL, |
| 3679 | dio_flags); |
| 3680 | |
| 3681 | if (ret > 0 && !overwrite && ext4_test_inode_state(inode, |
| 3682 | EXT4_STATE_DIO_UNWRITTEN)) { |
| 3683 | int err; |
| 3684 | /* |
| 3685 | * for non AIO case, since the IO is already |
| 3686 | * completed, we could do the conversion right here |
| 3687 | */ |
| 3688 | err = ext4_convert_unwritten_extents(NULL, inode, |
| 3689 | offset, ret); |
| 3690 | if (err < 0) |
| 3691 | ret = err; |
| 3692 | ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); |
| 3693 | } |
| 3694 | |
| 3695 | inode_dio_end(inode); |
| 3696 | /* take i_mutex locking again if we do a ovewrite dio */ |
| 3697 | if (overwrite) |
| 3698 | inode_lock(inode); |
| 3699 | |
| 3700 | if (ret < 0 && final_size > inode->i_size) |
| 3701 | ext4_truncate_failed_write(inode); |
| 3702 | |
| 3703 | /* Handle extending of i_size after direct IO write */ |
| 3704 | if (orphan) { |
| 3705 | int err; |
| 3706 | |
| 3707 | /* Credits for sb + inode write */ |
| 3708 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); |
| 3709 | if (IS_ERR(handle)) { |
| 3710 | /* This is really bad luck. We've written the data |
| 3711 | * but cannot extend i_size. Bail out and pretend |
| 3712 | * the write failed... */ |
| 3713 | ret = PTR_ERR(handle); |
| 3714 | if (inode->i_nlink) |
| 3715 | ext4_orphan_del(NULL, inode); |
| 3716 | |
| 3717 | goto out; |
| 3718 | } |
| 3719 | if (inode->i_nlink) |
| 3720 | ext4_orphan_del(handle, inode); |
| 3721 | if (ret > 0) { |
| 3722 | loff_t end = offset + ret; |
| 3723 | if (end > inode->i_size) { |
| 3724 | ei->i_disksize = end; |
| 3725 | i_size_write(inode, end); |
| 3726 | /* |
| 3727 | * We're going to return a positive `ret' |
| 3728 | * here due to non-zero-length I/O, so there's |
| 3729 | * no way of reporting error returns from |
| 3730 | * ext4_mark_inode_dirty() to userspace. So |
| 3731 | * ignore it. |
| 3732 | */ |
| 3733 | ext4_mark_inode_dirty(handle, inode); |
| 3734 | } |
| 3735 | } |
| 3736 | err = ext4_journal_stop(handle); |
| 3737 | if (ret == 0) |
| 3738 | ret = err; |
| 3739 | } |
| 3740 | out: |
| 3741 | return ret; |
| 3742 | } |
| 3743 | |
| 3744 | static ssize_t ext4_direct_IO_read(struct kiocb *iocb, struct iov_iter *iter) |
| 3745 | { |
| 3746 | struct address_space *mapping = iocb->ki_filp->f_mapping; |
| 3747 | struct inode *inode = mapping->host; |
| 3748 | size_t count = iov_iter_count(iter); |
| 3749 | ssize_t ret; |
| 3750 | |
| 3751 | /* |
| 3752 | * Shared inode_lock is enough for us - it protects against concurrent |
| 3753 | * writes & truncates and since we take care of writing back page cache, |
| 3754 | * we are protected against page writeback as well. |
| 3755 | */ |
| 3756 | inode_lock_shared(inode); |
| 3757 | ret = filemap_write_and_wait_range(mapping, iocb->ki_pos, |
| 3758 | iocb->ki_pos + count - 1); |
| 3759 | if (ret) |
| 3760 | goto out_unlock; |
| 3761 | ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, |
| 3762 | iter, ext4_dio_get_block, NULL, NULL, 0); |
| 3763 | out_unlock: |
| 3764 | inode_unlock_shared(inode); |
| 3765 | return ret; |
| 3766 | } |
| 3767 | |
| 3768 | static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
| 3769 | { |
| 3770 | struct file *file = iocb->ki_filp; |
| 3771 | struct inode *inode = file->f_mapping->host; |
| 3772 | size_t count = iov_iter_count(iter); |
| 3773 | loff_t offset = iocb->ki_pos; |
| 3774 | ssize_t ret; |
| 3775 | |
| 3776 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
| 3777 | if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode)) |
| 3778 | return 0; |
| 3779 | #endif |
| 3780 | |
| 3781 | /* |
| 3782 | * If we are doing data journalling we don't support O_DIRECT |
| 3783 | */ |
| 3784 | if (ext4_should_journal_data(inode)) |
| 3785 | return 0; |
| 3786 | |
| 3787 | /* Let buffer I/O handle the inline data case. */ |
| 3788 | if (ext4_has_inline_data(inode)) |
| 3789 | return 0; |
| 3790 | |
| 3791 | /* DAX uses iomap path now */ |
| 3792 | if (WARN_ON_ONCE(IS_DAX(inode))) |
| 3793 | return 0; |
| 3794 | |
| 3795 | trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter)); |
| 3796 | if (iov_iter_rw(iter) == READ) |
| 3797 | ret = ext4_direct_IO_read(iocb, iter); |
| 3798 | else |
| 3799 | ret = ext4_direct_IO_write(iocb, iter); |
| 3800 | trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret); |
| 3801 | return ret; |
| 3802 | } |
| 3803 | |
| 3804 | /* |
| 3805 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
| 3806 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
| 3807 | * much here because ->set_page_dirty is called under VFS locks. The page is |
| 3808 | * not necessarily locked. |
| 3809 | * |
| 3810 | * We cannot just dirty the page and leave attached buffers clean, because the |
| 3811 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty |
| 3812 | * or jbddirty because all the journalling code will explode. |
| 3813 | * |
| 3814 | * So what we do is to mark the page "pending dirty" and next time writepage |
| 3815 | * is called, propagate that into the buffers appropriately. |
| 3816 | */ |
| 3817 | static int ext4_journalled_set_page_dirty(struct page *page) |
| 3818 | { |
| 3819 | SetPageChecked(page); |
| 3820 | return __set_page_dirty_nobuffers(page); |
| 3821 | } |
| 3822 | |
| 3823 | static int ext4_set_page_dirty(struct page *page) |
| 3824 | { |
| 3825 | WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page)); |
| 3826 | WARN_ON_ONCE(!page_has_buffers(page)); |
| 3827 | return __set_page_dirty_buffers(page); |
| 3828 | } |
| 3829 | |
| 3830 | static const struct address_space_operations ext4_aops = { |
| 3831 | .readpage = ext4_readpage, |
| 3832 | .readpages = ext4_readpages, |
| 3833 | .writepage = ext4_writepage, |
| 3834 | .writepages = ext4_writepages, |
| 3835 | .write_begin = ext4_write_begin, |
| 3836 | .write_end = ext4_write_end, |
| 3837 | .set_page_dirty = ext4_set_page_dirty, |
| 3838 | .bmap = ext4_bmap, |
| 3839 | .invalidatepage = ext4_invalidatepage, |
| 3840 | .releasepage = ext4_releasepage, |
| 3841 | .direct_IO = ext4_direct_IO, |
| 3842 | .migratepage = buffer_migrate_page, |
| 3843 | .is_partially_uptodate = block_is_partially_uptodate, |
| 3844 | .error_remove_page = generic_error_remove_page, |
| 3845 | }; |
| 3846 | |
| 3847 | static const struct address_space_operations ext4_journalled_aops = { |
| 3848 | .readpage = ext4_readpage, |
| 3849 | .readpages = ext4_readpages, |
| 3850 | .writepage = ext4_writepage, |
| 3851 | .writepages = ext4_writepages, |
| 3852 | .write_begin = ext4_write_begin, |
| 3853 | .write_end = ext4_journalled_write_end, |
| 3854 | .set_page_dirty = ext4_journalled_set_page_dirty, |
| 3855 | .bmap = ext4_bmap, |
| 3856 | .invalidatepage = ext4_journalled_invalidatepage, |
| 3857 | .releasepage = ext4_releasepage, |
| 3858 | .direct_IO = ext4_direct_IO, |
| 3859 | .is_partially_uptodate = block_is_partially_uptodate, |
| 3860 | .error_remove_page = generic_error_remove_page, |
| 3861 | }; |
| 3862 | |
| 3863 | static const struct address_space_operations ext4_da_aops = { |
| 3864 | .readpage = ext4_readpage, |
| 3865 | .readpages = ext4_readpages, |
| 3866 | .writepage = ext4_writepage, |
| 3867 | .writepages = ext4_writepages, |
| 3868 | .write_begin = ext4_da_write_begin, |
| 3869 | .write_end = ext4_da_write_end, |
| 3870 | .set_page_dirty = ext4_set_page_dirty, |
| 3871 | .bmap = ext4_bmap, |
| 3872 | .invalidatepage = ext4_da_invalidatepage, |
| 3873 | .releasepage = ext4_releasepage, |
| 3874 | .direct_IO = ext4_direct_IO, |
| 3875 | .migratepage = buffer_migrate_page, |
| 3876 | .is_partially_uptodate = block_is_partially_uptodate, |
| 3877 | .error_remove_page = generic_error_remove_page, |
| 3878 | }; |
| 3879 | |
| 3880 | void ext4_set_aops(struct inode *inode) |
| 3881 | { |
| 3882 | switch (ext4_inode_journal_mode(inode)) { |
| 3883 | case EXT4_INODE_ORDERED_DATA_MODE: |
| 3884 | case EXT4_INODE_WRITEBACK_DATA_MODE: |
| 3885 | break; |
| 3886 | case EXT4_INODE_JOURNAL_DATA_MODE: |
| 3887 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
| 3888 | return; |
| 3889 | default: |
| 3890 | BUG(); |
| 3891 | } |
| 3892 | if (test_opt(inode->i_sb, DELALLOC)) |
| 3893 | inode->i_mapping->a_ops = &ext4_da_aops; |
| 3894 | else |
| 3895 | inode->i_mapping->a_ops = &ext4_aops; |
| 3896 | } |
| 3897 | |
| 3898 | static int __ext4_block_zero_page_range(handle_t *handle, |
| 3899 | struct address_space *mapping, loff_t from, loff_t length) |
| 3900 | { |
| 3901 | ext4_fsblk_t index = from >> PAGE_SHIFT; |
| 3902 | unsigned offset = from & (PAGE_SIZE-1); |
| 3903 | unsigned blocksize, pos; |
| 3904 | ext4_lblk_t iblock; |
| 3905 | struct inode *inode = mapping->host; |
| 3906 | struct buffer_head *bh; |
| 3907 | struct page *page; |
| 3908 | int err = 0; |
| 3909 | |
| 3910 | page = find_or_create_page(mapping, from >> PAGE_SHIFT, |
| 3911 | mapping_gfp_constraint(mapping, ~__GFP_FS)); |
| 3912 | if (!page) |
| 3913 | return -ENOMEM; |
| 3914 | |
| 3915 | blocksize = inode->i_sb->s_blocksize; |
| 3916 | |
| 3917 | iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits); |
| 3918 | |
| 3919 | if (!page_has_buffers(page)) |
| 3920 | create_empty_buffers(page, blocksize, 0); |
| 3921 | |
| 3922 | /* Find the buffer that contains "offset" */ |
| 3923 | bh = page_buffers(page); |
| 3924 | pos = blocksize; |
| 3925 | while (offset >= pos) { |
| 3926 | bh = bh->b_this_page; |
| 3927 | iblock++; |
| 3928 | pos += blocksize; |
| 3929 | } |
| 3930 | if (buffer_freed(bh)) { |
| 3931 | BUFFER_TRACE(bh, "freed: skip"); |
| 3932 | goto unlock; |
| 3933 | } |
| 3934 | if (!buffer_mapped(bh)) { |
| 3935 | BUFFER_TRACE(bh, "unmapped"); |
| 3936 | ext4_get_block(inode, iblock, bh, 0); |
| 3937 | /* unmapped? It's a hole - nothing to do */ |
| 3938 | if (!buffer_mapped(bh)) { |
| 3939 | BUFFER_TRACE(bh, "still unmapped"); |
| 3940 | goto unlock; |
| 3941 | } |
| 3942 | } |
| 3943 | |
| 3944 | /* Ok, it's mapped. Make sure it's up-to-date */ |
| 3945 | if (PageUptodate(page)) |
| 3946 | set_buffer_uptodate(bh); |
| 3947 | |
| 3948 | if (!buffer_uptodate(bh)) { |
| 3949 | err = -EIO; |
| 3950 | ll_rw_block(REQ_OP_READ, 0, 1, &bh); |
| 3951 | wait_on_buffer(bh); |
| 3952 | /* Uhhuh. Read error. Complain and punt. */ |
| 3953 | if (!buffer_uptodate(bh)) |
| 3954 | goto unlock; |
| 3955 | if (S_ISREG(inode->i_mode) && |
| 3956 | ext4_encrypted_inode(inode)) { |
| 3957 | /* We expect the key to be set. */ |
| 3958 | BUG_ON(!fscrypt_has_encryption_key(inode)); |
| 3959 | BUG_ON(blocksize != PAGE_SIZE); |
| 3960 | WARN_ON_ONCE(fscrypt_decrypt_page(page->mapping->host, |
| 3961 | page, PAGE_SIZE, 0, page->index)); |
| 3962 | } |
| 3963 | } |
| 3964 | if (ext4_should_journal_data(inode)) { |
| 3965 | BUFFER_TRACE(bh, "get write access"); |
| 3966 | err = ext4_journal_get_write_access(handle, bh); |
| 3967 | if (err) |
| 3968 | goto unlock; |
| 3969 | } |
| 3970 | zero_user(page, offset, length); |
| 3971 | BUFFER_TRACE(bh, "zeroed end of block"); |
| 3972 | |
| 3973 | if (ext4_should_journal_data(inode)) { |
| 3974 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
| 3975 | } else { |
| 3976 | err = 0; |
| 3977 | mark_buffer_dirty(bh); |
| 3978 | if (ext4_should_order_data(inode)) |
| 3979 | err = ext4_jbd2_inode_add_write(handle, inode); |
| 3980 | } |
| 3981 | |
| 3982 | unlock: |
| 3983 | unlock_page(page); |
| 3984 | put_page(page); |
| 3985 | return err; |
| 3986 | } |
| 3987 | |
| 3988 | /* |
| 3989 | * ext4_block_zero_page_range() zeros out a mapping of length 'length' |
| 3990 | * starting from file offset 'from'. The range to be zero'd must |
| 3991 | * be contained with in one block. If the specified range exceeds |
| 3992 | * the end of the block it will be shortened to end of the block |
| 3993 | * that cooresponds to 'from' |
| 3994 | */ |
| 3995 | static int ext4_block_zero_page_range(handle_t *handle, |
| 3996 | struct address_space *mapping, loff_t from, loff_t length) |
| 3997 | { |
| 3998 | struct inode *inode = mapping->host; |
| 3999 | unsigned offset = from & (PAGE_SIZE-1); |
| 4000 | unsigned blocksize = inode->i_sb->s_blocksize; |
| 4001 | unsigned max = blocksize - (offset & (blocksize - 1)); |
| 4002 | |
| 4003 | /* |
| 4004 | * correct length if it does not fall between |
| 4005 | * 'from' and the end of the block |
| 4006 | */ |
| 4007 | if (length > max || length < 0) |
| 4008 | length = max; |
| 4009 | |
| 4010 | if (IS_DAX(inode)) { |
| 4011 | return iomap_zero_range(inode, from, length, NULL, |
| 4012 | &ext4_iomap_ops); |
| 4013 | } |
| 4014 | return __ext4_block_zero_page_range(handle, mapping, from, length); |
| 4015 | } |
| 4016 | |
| 4017 | /* |
| 4018 | * ext4_block_truncate_page() zeroes out a mapping from file offset `from' |
| 4019 | * up to the end of the block which corresponds to `from'. |
| 4020 | * This required during truncate. We need to physically zero the tail end |
| 4021 | * of that block so it doesn't yield old data if the file is later grown. |
| 4022 | */ |
| 4023 | static int ext4_block_truncate_page(handle_t *handle, |
| 4024 | struct address_space *mapping, loff_t from) |
| 4025 | { |
| 4026 | unsigned offset = from & (PAGE_SIZE-1); |
| 4027 | unsigned length; |
| 4028 | unsigned blocksize; |
| 4029 | struct inode *inode = mapping->host; |
| 4030 | |
| 4031 | /* If we are processing an encrypted inode during orphan list handling */ |
| 4032 | if (ext4_encrypted_inode(inode) && !fscrypt_has_encryption_key(inode)) |
| 4033 | return 0; |
| 4034 | |
| 4035 | blocksize = inode->i_sb->s_blocksize; |
| 4036 | length = blocksize - (offset & (blocksize - 1)); |
| 4037 | |
| 4038 | return ext4_block_zero_page_range(handle, mapping, from, length); |
| 4039 | } |
| 4040 | |
| 4041 | int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode, |
| 4042 | loff_t lstart, loff_t length) |
| 4043 | { |
| 4044 | struct super_block *sb = inode->i_sb; |
| 4045 | struct address_space *mapping = inode->i_mapping; |
| 4046 | unsigned partial_start, partial_end; |
| 4047 | ext4_fsblk_t start, end; |
| 4048 | loff_t byte_end = (lstart + length - 1); |
| 4049 | int err = 0; |
| 4050 | |
| 4051 | partial_start = lstart & (sb->s_blocksize - 1); |
| 4052 | partial_end = byte_end & (sb->s_blocksize - 1); |
| 4053 | |
| 4054 | start = lstart >> sb->s_blocksize_bits; |
| 4055 | end = byte_end >> sb->s_blocksize_bits; |
| 4056 | |
| 4057 | /* Handle partial zero within the single block */ |
| 4058 | if (start == end && |
| 4059 | (partial_start || (partial_end != sb->s_blocksize - 1))) { |
| 4060 | err = ext4_block_zero_page_range(handle, mapping, |
| 4061 | lstart, length); |
| 4062 | return err; |
| 4063 | } |
| 4064 | /* Handle partial zero out on the start of the range */ |
| 4065 | if (partial_start) { |
| 4066 | err = ext4_block_zero_page_range(handle, mapping, |
| 4067 | lstart, sb->s_blocksize); |
| 4068 | if (err) |
| 4069 | return err; |
| 4070 | } |
| 4071 | /* Handle partial zero out on the end of the range */ |
| 4072 | if (partial_end != sb->s_blocksize - 1) |
| 4073 | err = ext4_block_zero_page_range(handle, mapping, |
| 4074 | byte_end - partial_end, |
| 4075 | partial_end + 1); |
| 4076 | return err; |
| 4077 | } |
| 4078 | |
| 4079 | int ext4_can_truncate(struct inode *inode) |
| 4080 | { |
| 4081 | if (S_ISREG(inode->i_mode)) |
| 4082 | return 1; |
| 4083 | if (S_ISDIR(inode->i_mode)) |
| 4084 | return 1; |
| 4085 | if (S_ISLNK(inode->i_mode)) |
| 4086 | return !ext4_inode_is_fast_symlink(inode); |
| 4087 | return 0; |
| 4088 | } |
| 4089 | |
| 4090 | /* |
| 4091 | * We have to make sure i_disksize gets properly updated before we truncate |
| 4092 | * page cache due to hole punching or zero range. Otherwise i_disksize update |
| 4093 | * can get lost as it may have been postponed to submission of writeback but |
| 4094 | * that will never happen after we truncate page cache. |
| 4095 | */ |
| 4096 | int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset, |
| 4097 | loff_t len) |
| 4098 | { |
| 4099 | handle_t *handle; |
| 4100 | loff_t size = i_size_read(inode); |
| 4101 | |
| 4102 | WARN_ON(!inode_is_locked(inode)); |
| 4103 | if (offset > size || offset + len < size) |
| 4104 | return 0; |
| 4105 | |
| 4106 | if (EXT4_I(inode)->i_disksize >= size) |
| 4107 | return 0; |
| 4108 | |
| 4109 | handle = ext4_journal_start(inode, EXT4_HT_MISC, 1); |
| 4110 | if (IS_ERR(handle)) |
| 4111 | return PTR_ERR(handle); |
| 4112 | ext4_update_i_disksize(inode, size); |
| 4113 | ext4_mark_inode_dirty(handle, inode); |
| 4114 | ext4_journal_stop(handle); |
| 4115 | |
| 4116 | return 0; |
| 4117 | } |
| 4118 | |
| 4119 | /* |
| 4120 | * ext4_punch_hole: punches a hole in a file by releasing the blocks |
| 4121 | * associated with the given offset and length |
| 4122 | * |
| 4123 | * @inode: File inode |
| 4124 | * @offset: The offset where the hole will begin |
| 4125 | * @len: The length of the hole |
| 4126 | * |
| 4127 | * Returns: 0 on success or negative on failure |
| 4128 | */ |
| 4129 | |
| 4130 | int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length) |
| 4131 | { |
| 4132 | struct super_block *sb = inode->i_sb; |
| 4133 | ext4_lblk_t first_block, stop_block; |
| 4134 | struct address_space *mapping = inode->i_mapping; |
| 4135 | loff_t first_block_offset, last_block_offset; |
| 4136 | handle_t *handle; |
| 4137 | unsigned int credits; |
| 4138 | int ret = 0; |
| 4139 | |
| 4140 | if (!S_ISREG(inode->i_mode)) |
| 4141 | return -EOPNOTSUPP; |
| 4142 | |
| 4143 | trace_ext4_punch_hole(inode, offset, length, 0); |
| 4144 | |
| 4145 | /* |
| 4146 | * Write out all dirty pages to avoid race conditions |
| 4147 | * Then release them. |
| 4148 | */ |
| 4149 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { |
| 4150 | ret = filemap_write_and_wait_range(mapping, offset, |
| 4151 | offset + length - 1); |
| 4152 | if (ret) |
| 4153 | return ret; |
| 4154 | } |
| 4155 | |
| 4156 | inode_lock(inode); |
| 4157 | |
| 4158 | /* No need to punch hole beyond i_size */ |
| 4159 | if (offset >= inode->i_size) |
| 4160 | goto out_mutex; |
| 4161 | |
| 4162 | /* |
| 4163 | * If the hole extends beyond i_size, set the hole |
| 4164 | * to end after the page that contains i_size |
| 4165 | */ |
| 4166 | if (offset + length > inode->i_size) { |
| 4167 | length = inode->i_size + |
| 4168 | PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) - |
| 4169 | offset; |
| 4170 | } |
| 4171 | |
| 4172 | if (offset & (sb->s_blocksize - 1) || |
| 4173 | (offset + length) & (sb->s_blocksize - 1)) { |
| 4174 | /* |
| 4175 | * Attach jinode to inode for jbd2 if we do any zeroing of |
| 4176 | * partial block |
| 4177 | */ |
| 4178 | ret = ext4_inode_attach_jinode(inode); |
| 4179 | if (ret < 0) |
| 4180 | goto out_mutex; |
| 4181 | |
| 4182 | } |
| 4183 | |
| 4184 | /* Wait all existing dio workers, newcomers will block on i_mutex */ |
| 4185 | ext4_inode_block_unlocked_dio(inode); |
| 4186 | inode_dio_wait(inode); |
| 4187 | |
| 4188 | /* |
| 4189 | * Prevent page faults from reinstantiating pages we have released from |
| 4190 | * page cache. |
| 4191 | */ |
| 4192 | down_write(&EXT4_I(inode)->i_mmap_sem); |
| 4193 | first_block_offset = round_up(offset, sb->s_blocksize); |
| 4194 | last_block_offset = round_down((offset + length), sb->s_blocksize) - 1; |
| 4195 | |
| 4196 | /* Now release the pages and zero block aligned part of pages*/ |
| 4197 | if (last_block_offset > first_block_offset) { |
| 4198 | ret = ext4_update_disksize_before_punch(inode, offset, length); |
| 4199 | if (ret) |
| 4200 | goto out_dio; |
| 4201 | truncate_pagecache_range(inode, first_block_offset, |
| 4202 | last_block_offset); |
| 4203 | } |
| 4204 | |
| 4205 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
| 4206 | credits = ext4_writepage_trans_blocks(inode); |
| 4207 | else |
| 4208 | credits = ext4_blocks_for_truncate(inode); |
| 4209 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); |
| 4210 | if (IS_ERR(handle)) { |
| 4211 | ret = PTR_ERR(handle); |
| 4212 | ext4_std_error(sb, ret); |
| 4213 | goto out_dio; |
| 4214 | } |
| 4215 | |
| 4216 | ret = ext4_zero_partial_blocks(handle, inode, offset, |
| 4217 | length); |
| 4218 | if (ret) |
| 4219 | goto out_stop; |
| 4220 | |
| 4221 | first_block = (offset + sb->s_blocksize - 1) >> |
| 4222 | EXT4_BLOCK_SIZE_BITS(sb); |
| 4223 | stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb); |
| 4224 | |
| 4225 | /* If there are no blocks to remove, return now */ |
| 4226 | if (first_block >= stop_block) |
| 4227 | goto out_stop; |
| 4228 | |
| 4229 | down_write(&EXT4_I(inode)->i_data_sem); |
| 4230 | ext4_discard_preallocations(inode); |
| 4231 | |
| 4232 | ret = ext4_es_remove_extent(inode, first_block, |
| 4233 | stop_block - first_block); |
| 4234 | if (ret) { |
| 4235 | up_write(&EXT4_I(inode)->i_data_sem); |
| 4236 | goto out_stop; |
| 4237 | } |
| 4238 | |
| 4239 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
| 4240 | ret = ext4_ext_remove_space(inode, first_block, |
| 4241 | stop_block - 1); |
| 4242 | else |
| 4243 | ret = ext4_ind_remove_space(handle, inode, first_block, |
| 4244 | stop_block); |
| 4245 | |
| 4246 | up_write(&EXT4_I(inode)->i_data_sem); |
| 4247 | if (IS_SYNC(inode)) |
| 4248 | ext4_handle_sync(handle); |
| 4249 | |
| 4250 | inode->i_mtime = inode->i_ctime = current_time(inode); |
| 4251 | ext4_mark_inode_dirty(handle, inode); |
| 4252 | if (ret >= 0) |
| 4253 | ext4_update_inode_fsync_trans(handle, inode, 1); |
| 4254 | out_stop: |
| 4255 | ext4_journal_stop(handle); |
| 4256 | out_dio: |
| 4257 | up_write(&EXT4_I(inode)->i_mmap_sem); |
| 4258 | ext4_inode_resume_unlocked_dio(inode); |
| 4259 | out_mutex: |
| 4260 | inode_unlock(inode); |
| 4261 | return ret; |
| 4262 | } |
| 4263 | |
| 4264 | int ext4_inode_attach_jinode(struct inode *inode) |
| 4265 | { |
| 4266 | struct ext4_inode_info *ei = EXT4_I(inode); |
| 4267 | struct jbd2_inode *jinode; |
| 4268 | |
| 4269 | if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal) |
| 4270 | return 0; |
| 4271 | |
| 4272 | jinode = jbd2_alloc_inode(GFP_KERNEL); |
| 4273 | spin_lock(&inode->i_lock); |
| 4274 | if (!ei->jinode) { |
| 4275 | if (!jinode) { |
| 4276 | spin_unlock(&inode->i_lock); |
| 4277 | return -ENOMEM; |
| 4278 | } |
| 4279 | ei->jinode = jinode; |
| 4280 | jbd2_journal_init_jbd_inode(ei->jinode, inode); |
| 4281 | jinode = NULL; |
| 4282 | } |
| 4283 | spin_unlock(&inode->i_lock); |
| 4284 | if (unlikely(jinode != NULL)) |
| 4285 | jbd2_free_inode(jinode); |
| 4286 | return 0; |
| 4287 | } |
| 4288 | |
| 4289 | /* |
| 4290 | * ext4_truncate() |
| 4291 | * |
| 4292 | * We block out ext4_get_block() block instantiations across the entire |
| 4293 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run |
| 4294 | * simultaneously on behalf of the same inode. |
| 4295 | * |
| 4296 | * As we work through the truncate and commit bits of it to the journal there |
| 4297 | * is one core, guiding principle: the file's tree must always be consistent on |
| 4298 | * disk. We must be able to restart the truncate after a crash. |
| 4299 | * |
| 4300 | * The file's tree may be transiently inconsistent in memory (although it |
| 4301 | * probably isn't), but whenever we close off and commit a journal transaction, |
| 4302 | * the contents of (the filesystem + the journal) must be consistent and |
| 4303 | * restartable. It's pretty simple, really: bottom up, right to left (although |
| 4304 | * left-to-right works OK too). |
| 4305 | * |
| 4306 | * Note that at recovery time, journal replay occurs *before* the restart of |
| 4307 | * truncate against the orphan inode list. |
| 4308 | * |
| 4309 | * The committed inode has the new, desired i_size (which is the same as |
| 4310 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
| 4311 | * that this inode's truncate did not complete and it will again call |
| 4312 | * ext4_truncate() to have another go. So there will be instantiated blocks |
| 4313 | * to the right of the truncation point in a crashed ext4 filesystem. But |
| 4314 | * that's fine - as long as they are linked from the inode, the post-crash |
| 4315 | * ext4_truncate() run will find them and release them. |
| 4316 | */ |
| 4317 | int ext4_truncate(struct inode *inode) |
| 4318 | { |
| 4319 | struct ext4_inode_info *ei = EXT4_I(inode); |
| 4320 | unsigned int credits; |
| 4321 | int err = 0; |
| 4322 | handle_t *handle; |
| 4323 | struct address_space *mapping = inode->i_mapping; |
| 4324 | |
| 4325 | /* |
| 4326 | * There is a possibility that we're either freeing the inode |
| 4327 | * or it's a completely new inode. In those cases we might not |
| 4328 | * have i_mutex locked because it's not necessary. |
| 4329 | */ |
| 4330 | if (!(inode->i_state & (I_NEW|I_FREEING))) |
| 4331 | WARN_ON(!inode_is_locked(inode)); |
| 4332 | trace_ext4_truncate_enter(inode); |
| 4333 | |
| 4334 | if (!ext4_can_truncate(inode)) |
| 4335 | return 0; |
| 4336 | |
| 4337 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
| 4338 | |
| 4339 | if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) |
| 4340 | ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
| 4341 | |
| 4342 | if (ext4_has_inline_data(inode)) { |
| 4343 | int has_inline = 1; |
| 4344 | |
| 4345 | err = ext4_inline_data_truncate(inode, &has_inline); |
| 4346 | if (err) |
| 4347 | return err; |
| 4348 | if (has_inline) |
| 4349 | return 0; |
| 4350 | } |
| 4351 | |
| 4352 | /* If we zero-out tail of the page, we have to create jinode for jbd2 */ |
| 4353 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) { |
| 4354 | if (ext4_inode_attach_jinode(inode) < 0) |
| 4355 | return 0; |
| 4356 | } |
| 4357 | |
| 4358 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
| 4359 | credits = ext4_writepage_trans_blocks(inode); |
| 4360 | else |
| 4361 | credits = ext4_blocks_for_truncate(inode); |
| 4362 | |
| 4363 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); |
| 4364 | if (IS_ERR(handle)) |
| 4365 | return PTR_ERR(handle); |
| 4366 | |
| 4367 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) |
| 4368 | ext4_block_truncate_page(handle, mapping, inode->i_size); |
| 4369 | |
| 4370 | /* |
| 4371 | * We add the inode to the orphan list, so that if this |
| 4372 | * truncate spans multiple transactions, and we crash, we will |
| 4373 | * resume the truncate when the filesystem recovers. It also |
| 4374 | * marks the inode dirty, to catch the new size. |
| 4375 | * |
| 4376 | * Implication: the file must always be in a sane, consistent |
| 4377 | * truncatable state while each transaction commits. |
| 4378 | */ |
| 4379 | err = ext4_orphan_add(handle, inode); |
| 4380 | if (err) |
| 4381 | goto out_stop; |
| 4382 | |
| 4383 | down_write(&EXT4_I(inode)->i_data_sem); |
| 4384 | |
| 4385 | ext4_discard_preallocations(inode); |
| 4386 | |
| 4387 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
| 4388 | err = ext4_ext_truncate(handle, inode); |
| 4389 | else |
| 4390 | ext4_ind_truncate(handle, inode); |
| 4391 | |
| 4392 | up_write(&ei->i_data_sem); |
| 4393 | if (err) |
| 4394 | goto out_stop; |
| 4395 | |
| 4396 | if (IS_SYNC(inode)) |
| 4397 | ext4_handle_sync(handle); |
| 4398 | |
| 4399 | out_stop: |
| 4400 | /* |
| 4401 | * If this was a simple ftruncate() and the file will remain alive, |
| 4402 | * then we need to clear up the orphan record which we created above. |
| 4403 | * However, if this was a real unlink then we were called by |
| 4404 | * ext4_evict_inode(), and we allow that function to clean up the |
| 4405 | * orphan info for us. |
| 4406 | */ |
| 4407 | if (inode->i_nlink) |
| 4408 | ext4_orphan_del(handle, inode); |
| 4409 | |
| 4410 | inode->i_mtime = inode->i_ctime = current_time(inode); |
| 4411 | ext4_mark_inode_dirty(handle, inode); |
| 4412 | ext4_journal_stop(handle); |
| 4413 | |
| 4414 | trace_ext4_truncate_exit(inode); |
| 4415 | return err; |
| 4416 | } |
| 4417 | |
| 4418 | /* |
| 4419 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
| 4420 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
| 4421 | * data in memory that is needed to recreate the on-disk version of this |
| 4422 | * inode. |
| 4423 | */ |
| 4424 | static int __ext4_get_inode_loc(struct inode *inode, |
| 4425 | struct ext4_iloc *iloc, int in_mem) |
| 4426 | { |
| 4427 | struct ext4_group_desc *gdp; |
| 4428 | struct buffer_head *bh; |
| 4429 | struct super_block *sb = inode->i_sb; |
| 4430 | ext4_fsblk_t block; |
| 4431 | int inodes_per_block, inode_offset; |
| 4432 | |
| 4433 | iloc->bh = NULL; |
| 4434 | if (!ext4_valid_inum(sb, inode->i_ino)) |
| 4435 | return -EFSCORRUPTED; |
| 4436 | |
| 4437 | iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb); |
| 4438 | gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); |
| 4439 | if (!gdp) |
| 4440 | return -EIO; |
| 4441 | |
| 4442 | /* |
| 4443 | * Figure out the offset within the block group inode table |
| 4444 | */ |
| 4445 | inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; |
| 4446 | inode_offset = ((inode->i_ino - 1) % |
| 4447 | EXT4_INODES_PER_GROUP(sb)); |
| 4448 | block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block); |
| 4449 | iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); |
| 4450 | |
| 4451 | bh = sb_getblk(sb, block); |
| 4452 | if (unlikely(!bh)) |
| 4453 | return -ENOMEM; |
| 4454 | if (!buffer_uptodate(bh)) { |
| 4455 | lock_buffer(bh); |
| 4456 | |
| 4457 | /* |
| 4458 | * If the buffer has the write error flag, we have failed |
| 4459 | * to write out another inode in the same block. In this |
| 4460 | * case, we don't have to read the block because we may |
| 4461 | * read the old inode data successfully. |
| 4462 | */ |
| 4463 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) |
| 4464 | set_buffer_uptodate(bh); |
| 4465 | |
| 4466 | if (buffer_uptodate(bh)) { |
| 4467 | /* someone brought it uptodate while we waited */ |
| 4468 | unlock_buffer(bh); |
| 4469 | goto has_buffer; |
| 4470 | } |
| 4471 | |
| 4472 | /* |
| 4473 | * If we have all information of the inode in memory and this |
| 4474 | * is the only valid inode in the block, we need not read the |
| 4475 | * block. |
| 4476 | */ |
| 4477 | if (in_mem) { |
| 4478 | struct buffer_head *bitmap_bh; |
| 4479 | int i, start; |
| 4480 | |
| 4481 | start = inode_offset & ~(inodes_per_block - 1); |
| 4482 | |
| 4483 | /* Is the inode bitmap in cache? */ |
| 4484 | bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); |
| 4485 | if (unlikely(!bitmap_bh)) |
| 4486 | goto make_io; |
| 4487 | |
| 4488 | /* |
| 4489 | * If the inode bitmap isn't in cache then the |
| 4490 | * optimisation may end up performing two reads instead |
| 4491 | * of one, so skip it. |
| 4492 | */ |
| 4493 | if (!buffer_uptodate(bitmap_bh)) { |
| 4494 | brelse(bitmap_bh); |
| 4495 | goto make_io; |
| 4496 | } |
| 4497 | for (i = start; i < start + inodes_per_block; i++) { |
| 4498 | if (i == inode_offset) |
| 4499 | continue; |
| 4500 | if (ext4_test_bit(i, bitmap_bh->b_data)) |
| 4501 | break; |
| 4502 | } |
| 4503 | brelse(bitmap_bh); |
| 4504 | if (i == start + inodes_per_block) { |
| 4505 | /* all other inodes are free, so skip I/O */ |
| 4506 | memset(bh->b_data, 0, bh->b_size); |
| 4507 | set_buffer_uptodate(bh); |
| 4508 | unlock_buffer(bh); |
| 4509 | goto has_buffer; |
| 4510 | } |
| 4511 | } |
| 4512 | |
| 4513 | make_io: |
| 4514 | /* |
| 4515 | * If we need to do any I/O, try to pre-readahead extra |
| 4516 | * blocks from the inode table. |
| 4517 | */ |
| 4518 | if (EXT4_SB(sb)->s_inode_readahead_blks) { |
| 4519 | ext4_fsblk_t b, end, table; |
| 4520 | unsigned num; |
| 4521 | __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks; |
| 4522 | |
| 4523 | table = ext4_inode_table(sb, gdp); |
| 4524 | /* s_inode_readahead_blks is always a power of 2 */ |
| 4525 | b = block & ~((ext4_fsblk_t) ra_blks - 1); |
| 4526 | if (table > b) |
| 4527 | b = table; |
| 4528 | end = b + ra_blks; |
| 4529 | num = EXT4_INODES_PER_GROUP(sb); |
| 4530 | if (ext4_has_group_desc_csum(sb)) |
| 4531 | num -= ext4_itable_unused_count(sb, gdp); |
| 4532 | table += num / inodes_per_block; |
| 4533 | if (end > table) |
| 4534 | end = table; |
| 4535 | while (b <= end) |
| 4536 | sb_breadahead(sb, b++); |
| 4537 | } |
| 4538 | |
| 4539 | /* |
| 4540 | * There are other valid inodes in the buffer, this inode |
| 4541 | * has in-inode xattrs, or we don't have this inode in memory. |
| 4542 | * Read the block from disk. |
| 4543 | */ |
| 4544 | trace_ext4_load_inode(inode); |
| 4545 | get_bh(bh); |
| 4546 | bh->b_end_io = end_buffer_read_sync; |
| 4547 | submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh); |
| 4548 | wait_on_buffer(bh); |
| 4549 | if (!buffer_uptodate(bh)) { |
| 4550 | EXT4_ERROR_INODE_BLOCK(inode, block, |
| 4551 | "unable to read itable block"); |
| 4552 | brelse(bh); |
| 4553 | return -EIO; |
| 4554 | } |
| 4555 | } |
| 4556 | has_buffer: |
| 4557 | iloc->bh = bh; |
| 4558 | return 0; |
| 4559 | } |
| 4560 | |
| 4561 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
| 4562 | { |
| 4563 | /* We have all inode data except xattrs in memory here. */ |
| 4564 | return __ext4_get_inode_loc(inode, iloc, |
| 4565 | !ext4_test_inode_state(inode, EXT4_STATE_XATTR)); |
| 4566 | } |
| 4567 | |
| 4568 | void ext4_set_inode_flags(struct inode *inode) |
| 4569 | { |
| 4570 | unsigned int flags = EXT4_I(inode)->i_flags; |
| 4571 | unsigned int new_fl = 0; |
| 4572 | |
| 4573 | if (flags & EXT4_SYNC_FL) |
| 4574 | new_fl |= S_SYNC; |
| 4575 | if (flags & EXT4_APPEND_FL) |
| 4576 | new_fl |= S_APPEND; |
| 4577 | if (flags & EXT4_IMMUTABLE_FL) |
| 4578 | new_fl |= S_IMMUTABLE; |
| 4579 | if (flags & EXT4_NOATIME_FL) |
| 4580 | new_fl |= S_NOATIME; |
| 4581 | if (flags & EXT4_DIRSYNC_FL) |
| 4582 | new_fl |= S_DIRSYNC; |
| 4583 | if (test_opt(inode->i_sb, DAX) && S_ISREG(inode->i_mode) && |
| 4584 | !ext4_should_journal_data(inode) && !ext4_has_inline_data(inode) && |
| 4585 | !ext4_encrypted_inode(inode)) |
| 4586 | new_fl |= S_DAX; |
| 4587 | inode_set_flags(inode, new_fl, |
| 4588 | S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX); |
| 4589 | } |
| 4590 | |
| 4591 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
| 4592 | struct ext4_inode_info *ei) |
| 4593 | { |
| 4594 | blkcnt_t i_blocks ; |
| 4595 | struct inode *inode = &(ei->vfs_inode); |
| 4596 | struct super_block *sb = inode->i_sb; |
| 4597 | |
| 4598 | if (ext4_has_feature_huge_file(sb)) { |
| 4599 | /* we are using combined 48 bit field */ |
| 4600 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | |
| 4601 | le32_to_cpu(raw_inode->i_blocks_lo); |
| 4602 | if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { |
| 4603 | /* i_blocks represent file system block size */ |
| 4604 | return i_blocks << (inode->i_blkbits - 9); |
| 4605 | } else { |
| 4606 | return i_blocks; |
| 4607 | } |
| 4608 | } else { |
| 4609 | return le32_to_cpu(raw_inode->i_blocks_lo); |
| 4610 | } |
| 4611 | } |
| 4612 | |
| 4613 | static inline void ext4_iget_extra_inode(struct inode *inode, |
| 4614 | struct ext4_inode *raw_inode, |
| 4615 | struct ext4_inode_info *ei) |
| 4616 | { |
| 4617 | __le32 *magic = (void *)raw_inode + |
| 4618 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize; |
| 4619 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize + sizeof(__le32) <= |
| 4620 | EXT4_INODE_SIZE(inode->i_sb) && |
| 4621 | *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) { |
| 4622 | ext4_set_inode_state(inode, EXT4_STATE_XATTR); |
| 4623 | ext4_find_inline_data_nolock(inode); |
| 4624 | } else |
| 4625 | EXT4_I(inode)->i_inline_off = 0; |
| 4626 | } |
| 4627 | |
| 4628 | int ext4_get_projid(struct inode *inode, kprojid_t *projid) |
| 4629 | { |
| 4630 | if (!ext4_has_feature_project(inode->i_sb)) |
| 4631 | return -EOPNOTSUPP; |
| 4632 | *projid = EXT4_I(inode)->i_projid; |
| 4633 | return 0; |
| 4634 | } |
| 4635 | |
| 4636 | struct inode *ext4_iget(struct super_block *sb, unsigned long ino) |
| 4637 | { |
| 4638 | struct ext4_iloc iloc; |
| 4639 | struct ext4_inode *raw_inode; |
| 4640 | struct ext4_inode_info *ei; |
| 4641 | struct inode *inode; |
| 4642 | journal_t *journal = EXT4_SB(sb)->s_journal; |
| 4643 | long ret; |
| 4644 | loff_t size; |
| 4645 | int block; |
| 4646 | uid_t i_uid; |
| 4647 | gid_t i_gid; |
| 4648 | projid_t i_projid; |
| 4649 | |
| 4650 | inode = iget_locked(sb, ino); |
| 4651 | if (!inode) |
| 4652 | return ERR_PTR(-ENOMEM); |
| 4653 | if (!(inode->i_state & I_NEW)) |
| 4654 | return inode; |
| 4655 | |
| 4656 | ei = EXT4_I(inode); |
| 4657 | iloc.bh = NULL; |
| 4658 | |
| 4659 | ret = __ext4_get_inode_loc(inode, &iloc, 0); |
| 4660 | if (ret < 0) |
| 4661 | goto bad_inode; |
| 4662 | raw_inode = ext4_raw_inode(&iloc); |
| 4663 | |
| 4664 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
| 4665 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); |
| 4666 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > |
| 4667 | EXT4_INODE_SIZE(inode->i_sb) || |
| 4668 | (ei->i_extra_isize & 3)) { |
| 4669 | EXT4_ERROR_INODE(inode, |
| 4670 | "bad extra_isize %u (inode size %u)", |
| 4671 | ei->i_extra_isize, |
| 4672 | EXT4_INODE_SIZE(inode->i_sb)); |
| 4673 | ret = -EFSCORRUPTED; |
| 4674 | goto bad_inode; |
| 4675 | } |
| 4676 | } else |
| 4677 | ei->i_extra_isize = 0; |
| 4678 | |
| 4679 | /* Precompute checksum seed for inode metadata */ |
| 4680 | if (ext4_has_metadata_csum(sb)) { |
| 4681 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 4682 | __u32 csum; |
| 4683 | __le32 inum = cpu_to_le32(inode->i_ino); |
| 4684 | __le32 gen = raw_inode->i_generation; |
| 4685 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, |
| 4686 | sizeof(inum)); |
| 4687 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, |
| 4688 | sizeof(gen)); |
| 4689 | } |
| 4690 | |
| 4691 | if (!ext4_inode_csum_verify(inode, raw_inode, ei)) { |
| 4692 | EXT4_ERROR_INODE(inode, "checksum invalid"); |
| 4693 | ret = -EFSBADCRC; |
| 4694 | goto bad_inode; |
| 4695 | } |
| 4696 | |
| 4697 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
| 4698 | i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); |
| 4699 | i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); |
| 4700 | if (ext4_has_feature_project(sb) && |
| 4701 | EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE && |
| 4702 | EXT4_FITS_IN_INODE(raw_inode, ei, i_projid)) |
| 4703 | i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid); |
| 4704 | else |
| 4705 | i_projid = EXT4_DEF_PROJID; |
| 4706 | |
| 4707 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
| 4708 | i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
| 4709 | i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; |
| 4710 | } |
| 4711 | i_uid_write(inode, i_uid); |
| 4712 | i_gid_write(inode, i_gid); |
| 4713 | ei->i_projid = make_kprojid(&init_user_ns, i_projid); |
| 4714 | set_nlink(inode, le16_to_cpu(raw_inode->i_links_count)); |
| 4715 | |
| 4716 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
| 4717 | ei->i_inline_off = 0; |
| 4718 | ei->i_dir_start_lookup = 0; |
| 4719 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); |
| 4720 | /* We now have enough fields to check if the inode was active or not. |
| 4721 | * This is needed because nfsd might try to access dead inodes |
| 4722 | * the test is that same one that e2fsck uses |
| 4723 | * NeilBrown 1999oct15 |
| 4724 | */ |
| 4725 | if (inode->i_nlink == 0) { |
| 4726 | if ((inode->i_mode == 0 || |
| 4727 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) && |
| 4728 | ino != EXT4_BOOT_LOADER_INO) { |
| 4729 | /* this inode is deleted */ |
| 4730 | ret = -ESTALE; |
| 4731 | goto bad_inode; |
| 4732 | } |
| 4733 | /* The only unlinked inodes we let through here have |
| 4734 | * valid i_mode and are being read by the orphan |
| 4735 | * recovery code: that's fine, we're about to complete |
| 4736 | * the process of deleting those. |
| 4737 | * OR it is the EXT4_BOOT_LOADER_INO which is |
| 4738 | * not initialized on a new filesystem. */ |
| 4739 | } |
| 4740 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
| 4741 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
| 4742 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
| 4743 | if (ext4_has_feature_64bit(sb)) |
| 4744 | ei->i_file_acl |= |
| 4745 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; |
| 4746 | inode->i_size = ext4_isize(sb, raw_inode); |
| 4747 | if ((size = i_size_read(inode)) < 0) { |
| 4748 | EXT4_ERROR_INODE(inode, "bad i_size value: %lld", size); |
| 4749 | ret = -EFSCORRUPTED; |
| 4750 | goto bad_inode; |
| 4751 | } |
| 4752 | ei->i_disksize = inode->i_size; |
| 4753 | #ifdef CONFIG_QUOTA |
| 4754 | ei->i_reserved_quota = 0; |
| 4755 | #endif |
| 4756 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
| 4757 | ei->i_block_group = iloc.block_group; |
| 4758 | ei->i_last_alloc_group = ~0; |
| 4759 | /* |
| 4760 | * NOTE! The in-memory inode i_data array is in little-endian order |
| 4761 | * even on big-endian machines: we do NOT byteswap the block numbers! |
| 4762 | */ |
| 4763 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
| 4764 | ei->i_data[block] = raw_inode->i_block[block]; |
| 4765 | INIT_LIST_HEAD(&ei->i_orphan); |
| 4766 | |
| 4767 | /* |
| 4768 | * Set transaction id's of transactions that have to be committed |
| 4769 | * to finish f[data]sync. We set them to currently running transaction |
| 4770 | * as we cannot be sure that the inode or some of its metadata isn't |
| 4771 | * part of the transaction - the inode could have been reclaimed and |
| 4772 | * now it is reread from disk. |
| 4773 | */ |
| 4774 | if (journal) { |
| 4775 | transaction_t *transaction; |
| 4776 | tid_t tid; |
| 4777 | |
| 4778 | read_lock(&journal->j_state_lock); |
| 4779 | if (journal->j_running_transaction) |
| 4780 | transaction = journal->j_running_transaction; |
| 4781 | else |
| 4782 | transaction = journal->j_committing_transaction; |
| 4783 | if (transaction) |
| 4784 | tid = transaction->t_tid; |
| 4785 | else |
| 4786 | tid = journal->j_commit_sequence; |
| 4787 | read_unlock(&journal->j_state_lock); |
| 4788 | ei->i_sync_tid = tid; |
| 4789 | ei->i_datasync_tid = tid; |
| 4790 | } |
| 4791 | |
| 4792 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
| 4793 | if (ei->i_extra_isize == 0) { |
| 4794 | /* The extra space is currently unused. Use it. */ |
| 4795 | BUILD_BUG_ON(sizeof(struct ext4_inode) & 3); |
| 4796 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
| 4797 | EXT4_GOOD_OLD_INODE_SIZE; |
| 4798 | } else { |
| 4799 | ext4_iget_extra_inode(inode, raw_inode, ei); |
| 4800 | } |
| 4801 | } |
| 4802 | |
| 4803 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
| 4804 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); |
| 4805 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); |
| 4806 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); |
| 4807 | |
| 4808 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
| 4809 | inode->i_version = le32_to_cpu(raw_inode->i_disk_version); |
| 4810 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
| 4811 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) |
| 4812 | inode->i_version |= |
| 4813 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; |
| 4814 | } |
| 4815 | } |
| 4816 | |
| 4817 | ret = 0; |
| 4818 | if (ei->i_file_acl && |
| 4819 | !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) { |
| 4820 | EXT4_ERROR_INODE(inode, "bad extended attribute block %llu", |
| 4821 | ei->i_file_acl); |
| 4822 | ret = -EFSCORRUPTED; |
| 4823 | goto bad_inode; |
| 4824 | } else if (!ext4_has_inline_data(inode)) { |
| 4825 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
| 4826 | if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || |
| 4827 | (S_ISLNK(inode->i_mode) && |
| 4828 | !ext4_inode_is_fast_symlink(inode)))) |
| 4829 | /* Validate extent which is part of inode */ |
| 4830 | ret = ext4_ext_check_inode(inode); |
| 4831 | } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || |
| 4832 | (S_ISLNK(inode->i_mode) && |
| 4833 | !ext4_inode_is_fast_symlink(inode))) { |
| 4834 | /* Validate block references which are part of inode */ |
| 4835 | ret = ext4_ind_check_inode(inode); |
| 4836 | } |
| 4837 | } |
| 4838 | if (ret) |
| 4839 | goto bad_inode; |
| 4840 | |
| 4841 | if (S_ISREG(inode->i_mode)) { |
| 4842 | inode->i_op = &ext4_file_inode_operations; |
| 4843 | inode->i_fop = &ext4_file_operations; |
| 4844 | ext4_set_aops(inode); |
| 4845 | } else if (S_ISDIR(inode->i_mode)) { |
| 4846 | inode->i_op = &ext4_dir_inode_operations; |
| 4847 | inode->i_fop = &ext4_dir_operations; |
| 4848 | } else if (S_ISLNK(inode->i_mode)) { |
| 4849 | if (ext4_encrypted_inode(inode)) { |
| 4850 | inode->i_op = &ext4_encrypted_symlink_inode_operations; |
| 4851 | ext4_set_aops(inode); |
| 4852 | } else if (ext4_inode_is_fast_symlink(inode)) { |
| 4853 | inode->i_link = (char *)ei->i_data; |
| 4854 | inode->i_op = &ext4_fast_symlink_inode_operations; |
| 4855 | nd_terminate_link(ei->i_data, inode->i_size, |
| 4856 | sizeof(ei->i_data) - 1); |
| 4857 | } else { |
| 4858 | inode->i_op = &ext4_symlink_inode_operations; |
| 4859 | ext4_set_aops(inode); |
| 4860 | } |
| 4861 | inode_nohighmem(inode); |
| 4862 | } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
| 4863 | S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { |
| 4864 | inode->i_op = &ext4_special_inode_operations; |
| 4865 | if (raw_inode->i_block[0]) |
| 4866 | init_special_inode(inode, inode->i_mode, |
| 4867 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); |
| 4868 | else |
| 4869 | init_special_inode(inode, inode->i_mode, |
| 4870 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); |
| 4871 | } else if (ino == EXT4_BOOT_LOADER_INO) { |
| 4872 | make_bad_inode(inode); |
| 4873 | } else { |
| 4874 | ret = -EFSCORRUPTED; |
| 4875 | EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode); |
| 4876 | goto bad_inode; |
| 4877 | } |
| 4878 | brelse(iloc.bh); |
| 4879 | ext4_set_inode_flags(inode); |
| 4880 | if (ei->i_flags & EXT4_EA_INODE_FL) |
| 4881 | ext4_xattr_inode_set_class(inode); |
| 4882 | unlock_new_inode(inode); |
| 4883 | return inode; |
| 4884 | |
| 4885 | bad_inode: |
| 4886 | brelse(iloc.bh); |
| 4887 | iget_failed(inode); |
| 4888 | return ERR_PTR(ret); |
| 4889 | } |
| 4890 | |
| 4891 | struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino) |
| 4892 | { |
| 4893 | if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO) |
| 4894 | return ERR_PTR(-EFSCORRUPTED); |
| 4895 | return ext4_iget(sb, ino); |
| 4896 | } |
| 4897 | |
| 4898 | static int ext4_inode_blocks_set(handle_t *handle, |
| 4899 | struct ext4_inode *raw_inode, |
| 4900 | struct ext4_inode_info *ei) |
| 4901 | { |
| 4902 | struct inode *inode = &(ei->vfs_inode); |
| 4903 | u64 i_blocks = inode->i_blocks; |
| 4904 | struct super_block *sb = inode->i_sb; |
| 4905 | |
| 4906 | if (i_blocks <= ~0U) { |
| 4907 | /* |
| 4908 | * i_blocks can be represented in a 32 bit variable |
| 4909 | * as multiple of 512 bytes |
| 4910 | */ |
| 4911 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
| 4912 | raw_inode->i_blocks_high = 0; |
| 4913 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
| 4914 | return 0; |
| 4915 | } |
| 4916 | if (!ext4_has_feature_huge_file(sb)) |
| 4917 | return -EFBIG; |
| 4918 | |
| 4919 | if (i_blocks <= 0xffffffffffffULL) { |
| 4920 | /* |
| 4921 | * i_blocks can be represented in a 48 bit variable |
| 4922 | * as multiple of 512 bytes |
| 4923 | */ |
| 4924 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
| 4925 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
| 4926 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
| 4927 | } else { |
| 4928 | ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
| 4929 | /* i_block is stored in file system block size */ |
| 4930 | i_blocks = i_blocks >> (inode->i_blkbits - 9); |
| 4931 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
| 4932 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
| 4933 | } |
| 4934 | return 0; |
| 4935 | } |
| 4936 | |
| 4937 | struct other_inode { |
| 4938 | unsigned long orig_ino; |
| 4939 | struct ext4_inode *raw_inode; |
| 4940 | }; |
| 4941 | |
| 4942 | static int other_inode_match(struct inode * inode, unsigned long ino, |
| 4943 | void *data) |
| 4944 | { |
| 4945 | struct other_inode *oi = (struct other_inode *) data; |
| 4946 | |
| 4947 | if ((inode->i_ino != ino) || |
| 4948 | (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW | |
| 4949 | I_DIRTY_SYNC | I_DIRTY_DATASYNC)) || |
| 4950 | ((inode->i_state & I_DIRTY_TIME) == 0)) |
| 4951 | return 0; |
| 4952 | spin_lock(&inode->i_lock); |
| 4953 | if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW | |
| 4954 | I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) && |
| 4955 | (inode->i_state & I_DIRTY_TIME)) { |
| 4956 | struct ext4_inode_info *ei = EXT4_I(inode); |
| 4957 | |
| 4958 | inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED); |
| 4959 | spin_unlock(&inode->i_lock); |
| 4960 | |
| 4961 | spin_lock(&ei->i_raw_lock); |
| 4962 | EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode); |
| 4963 | EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode); |
| 4964 | EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode); |
| 4965 | ext4_inode_csum_set(inode, oi->raw_inode, ei); |
| 4966 | spin_unlock(&ei->i_raw_lock); |
| 4967 | trace_ext4_other_inode_update_time(inode, oi->orig_ino); |
| 4968 | return -1; |
| 4969 | } |
| 4970 | spin_unlock(&inode->i_lock); |
| 4971 | return -1; |
| 4972 | } |
| 4973 | |
| 4974 | /* |
| 4975 | * Opportunistically update the other time fields for other inodes in |
| 4976 | * the same inode table block. |
| 4977 | */ |
| 4978 | static void ext4_update_other_inodes_time(struct super_block *sb, |
| 4979 | unsigned long orig_ino, char *buf) |
| 4980 | { |
| 4981 | struct other_inode oi; |
| 4982 | unsigned long ino; |
| 4983 | int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; |
| 4984 | int inode_size = EXT4_INODE_SIZE(sb); |
| 4985 | |
| 4986 | oi.orig_ino = orig_ino; |
| 4987 | /* |
| 4988 | * Calculate the first inode in the inode table block. Inode |
| 4989 | * numbers are one-based. That is, the first inode in a block |
| 4990 | * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1). |
| 4991 | */ |
| 4992 | ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1; |
| 4993 | for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) { |
| 4994 | if (ino == orig_ino) |
| 4995 | continue; |
| 4996 | oi.raw_inode = (struct ext4_inode *) buf; |
| 4997 | (void) find_inode_nowait(sb, ino, other_inode_match, &oi); |
| 4998 | } |
| 4999 | } |
| 5000 | |
| 5001 | /* |
| 5002 | * Post the struct inode info into an on-disk inode location in the |
| 5003 | * buffer-cache. This gobbles the caller's reference to the |
| 5004 | * buffer_head in the inode location struct. |
| 5005 | * |
| 5006 | * The caller must have write access to iloc->bh. |
| 5007 | */ |
| 5008 | static int ext4_do_update_inode(handle_t *handle, |
| 5009 | struct inode *inode, |
| 5010 | struct ext4_iloc *iloc) |
| 5011 | { |
| 5012 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
| 5013 | struct ext4_inode_info *ei = EXT4_I(inode); |
| 5014 | struct buffer_head *bh = iloc->bh; |
| 5015 | struct super_block *sb = inode->i_sb; |
| 5016 | int err = 0, rc, block; |
| 5017 | int need_datasync = 0, set_large_file = 0; |
| 5018 | uid_t i_uid; |
| 5019 | gid_t i_gid; |
| 5020 | projid_t i_projid; |
| 5021 | |
| 5022 | spin_lock(&ei->i_raw_lock); |
| 5023 | |
| 5024 | /* For fields not tracked in the in-memory inode, |
| 5025 | * initialise them to zero for new inodes. */ |
| 5026 | if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) |
| 5027 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); |
| 5028 | |
| 5029 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
| 5030 | i_uid = i_uid_read(inode); |
| 5031 | i_gid = i_gid_read(inode); |
| 5032 | i_projid = from_kprojid(&init_user_ns, ei->i_projid); |
| 5033 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
| 5034 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid)); |
| 5035 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid)); |
| 5036 | /* |
| 5037 | * Fix up interoperability with old kernels. Otherwise, old inodes get |
| 5038 | * re-used with the upper 16 bits of the uid/gid intact |
| 5039 | */ |
| 5040 | if (ei->i_dtime && list_empty(&ei->i_orphan)) { |
| 5041 | raw_inode->i_uid_high = 0; |
| 5042 | raw_inode->i_gid_high = 0; |
| 5043 | } else { |
| 5044 | raw_inode->i_uid_high = |
| 5045 | cpu_to_le16(high_16_bits(i_uid)); |
| 5046 | raw_inode->i_gid_high = |
| 5047 | cpu_to_le16(high_16_bits(i_gid)); |
| 5048 | } |
| 5049 | } else { |
| 5050 | raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid)); |
| 5051 | raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid)); |
| 5052 | raw_inode->i_uid_high = 0; |
| 5053 | raw_inode->i_gid_high = 0; |
| 5054 | } |
| 5055 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); |
| 5056 | |
| 5057 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); |
| 5058 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); |
| 5059 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); |
| 5060 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); |
| 5061 | |
| 5062 | err = ext4_inode_blocks_set(handle, raw_inode, ei); |
| 5063 | if (err) { |
| 5064 | spin_unlock(&ei->i_raw_lock); |
| 5065 | goto out_brelse; |
| 5066 | } |
| 5067 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
| 5068 | raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF); |
| 5069 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) |
| 5070 | raw_inode->i_file_acl_high = |
| 5071 | cpu_to_le16(ei->i_file_acl >> 32); |
| 5072 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
| 5073 | if (ei->i_disksize != ext4_isize(inode->i_sb, raw_inode)) { |
| 5074 | ext4_isize_set(raw_inode, ei->i_disksize); |
| 5075 | need_datasync = 1; |
| 5076 | } |
| 5077 | if (ei->i_disksize > 0x7fffffffULL) { |
| 5078 | if (!ext4_has_feature_large_file(sb) || |
| 5079 | EXT4_SB(sb)->s_es->s_rev_level == |
| 5080 | cpu_to_le32(EXT4_GOOD_OLD_REV)) |
| 5081 | set_large_file = 1; |
| 5082 | } |
| 5083 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); |
| 5084 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { |
| 5085 | if (old_valid_dev(inode->i_rdev)) { |
| 5086 | raw_inode->i_block[0] = |
| 5087 | cpu_to_le32(old_encode_dev(inode->i_rdev)); |
| 5088 | raw_inode->i_block[1] = 0; |
| 5089 | } else { |
| 5090 | raw_inode->i_block[0] = 0; |
| 5091 | raw_inode->i_block[1] = |
| 5092 | cpu_to_le32(new_encode_dev(inode->i_rdev)); |
| 5093 | raw_inode->i_block[2] = 0; |
| 5094 | } |
| 5095 | } else if (!ext4_has_inline_data(inode)) { |
| 5096 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
| 5097 | raw_inode->i_block[block] = ei->i_data[block]; |
| 5098 | } |
| 5099 | |
| 5100 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
| 5101 | raw_inode->i_disk_version = cpu_to_le32(inode->i_version); |
| 5102 | if (ei->i_extra_isize) { |
| 5103 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) |
| 5104 | raw_inode->i_version_hi = |
| 5105 | cpu_to_le32(inode->i_version >> 32); |
| 5106 | raw_inode->i_extra_isize = |
| 5107 | cpu_to_le16(ei->i_extra_isize); |
| 5108 | } |
| 5109 | } |
| 5110 | |
| 5111 | BUG_ON(!ext4_has_feature_project(inode->i_sb) && |
| 5112 | i_projid != EXT4_DEF_PROJID); |
| 5113 | |
| 5114 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && |
| 5115 | EXT4_FITS_IN_INODE(raw_inode, ei, i_projid)) |
| 5116 | raw_inode->i_projid = cpu_to_le32(i_projid); |
| 5117 | |
| 5118 | ext4_inode_csum_set(inode, raw_inode, ei); |
| 5119 | spin_unlock(&ei->i_raw_lock); |
| 5120 | if (inode->i_sb->s_flags & MS_LAZYTIME) |
| 5121 | ext4_update_other_inodes_time(inode->i_sb, inode->i_ino, |
| 5122 | bh->b_data); |
| 5123 | |
| 5124 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
| 5125 | rc = ext4_handle_dirty_metadata(handle, NULL, bh); |
| 5126 | if (!err) |
| 5127 | err = rc; |
| 5128 | ext4_clear_inode_state(inode, EXT4_STATE_NEW); |
| 5129 | if (set_large_file) { |
| 5130 | BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access"); |
| 5131 | err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh); |
| 5132 | if (err) |
| 5133 | goto out_brelse; |
| 5134 | ext4_update_dynamic_rev(sb); |
| 5135 | ext4_set_feature_large_file(sb); |
| 5136 | ext4_handle_sync(handle); |
| 5137 | err = ext4_handle_dirty_super(handle, sb); |
| 5138 | } |
| 5139 | ext4_update_inode_fsync_trans(handle, inode, need_datasync); |
| 5140 | out_brelse: |
| 5141 | brelse(bh); |
| 5142 | ext4_std_error(inode->i_sb, err); |
| 5143 | return err; |
| 5144 | } |
| 5145 | |
| 5146 | /* |
| 5147 | * ext4_write_inode() |
| 5148 | * |
| 5149 | * We are called from a few places: |
| 5150 | * |
| 5151 | * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files. |
| 5152 | * Here, there will be no transaction running. We wait for any running |
| 5153 | * transaction to commit. |
| 5154 | * |
| 5155 | * - Within flush work (sys_sync(), kupdate and such). |
| 5156 | * We wait on commit, if told to. |
| 5157 | * |
| 5158 | * - Within iput_final() -> write_inode_now() |
| 5159 | * We wait on commit, if told to. |
| 5160 | * |
| 5161 | * In all cases it is actually safe for us to return without doing anything, |
| 5162 | * because the inode has been copied into a raw inode buffer in |
| 5163 | * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL |
| 5164 | * writeback. |
| 5165 | * |
| 5166 | * Note that we are absolutely dependent upon all inode dirtiers doing the |
| 5167 | * right thing: they *must* call mark_inode_dirty() after dirtying info in |
| 5168 | * which we are interested. |
| 5169 | * |
| 5170 | * It would be a bug for them to not do this. The code: |
| 5171 | * |
| 5172 | * mark_inode_dirty(inode) |
| 5173 | * stuff(); |
| 5174 | * inode->i_size = expr; |
| 5175 | * |
| 5176 | * is in error because write_inode() could occur while `stuff()' is running, |
| 5177 | * and the new i_size will be lost. Plus the inode will no longer be on the |
| 5178 | * superblock's dirty inode list. |
| 5179 | */ |
| 5180 | int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) |
| 5181 | { |
| 5182 | int err; |
| 5183 | |
| 5184 | if (WARN_ON_ONCE(current->flags & PF_MEMALLOC)) |
| 5185 | return 0; |
| 5186 | |
| 5187 | if (EXT4_SB(inode->i_sb)->s_journal) { |
| 5188 | if (ext4_journal_current_handle()) { |
| 5189 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); |
| 5190 | dump_stack(); |
| 5191 | return -EIO; |
| 5192 | } |
| 5193 | |
| 5194 | /* |
| 5195 | * No need to force transaction in WB_SYNC_NONE mode. Also |
| 5196 | * ext4_sync_fs() will force the commit after everything is |
| 5197 | * written. |
| 5198 | */ |
| 5199 | if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync) |
| 5200 | return 0; |
| 5201 | |
| 5202 | err = ext4_force_commit(inode->i_sb); |
| 5203 | } else { |
| 5204 | struct ext4_iloc iloc; |
| 5205 | |
| 5206 | err = __ext4_get_inode_loc(inode, &iloc, 0); |
| 5207 | if (err) |
| 5208 | return err; |
| 5209 | /* |
| 5210 | * sync(2) will flush the whole buffer cache. No need to do |
| 5211 | * it here separately for each inode. |
| 5212 | */ |
| 5213 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) |
| 5214 | sync_dirty_buffer(iloc.bh); |
| 5215 | if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { |
| 5216 | EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr, |
| 5217 | "IO error syncing inode"); |
| 5218 | err = -EIO; |
| 5219 | } |
| 5220 | brelse(iloc.bh); |
| 5221 | } |
| 5222 | return err; |
| 5223 | } |
| 5224 | |
| 5225 | /* |
| 5226 | * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate |
| 5227 | * buffers that are attached to a page stradding i_size and are undergoing |
| 5228 | * commit. In that case we have to wait for commit to finish and try again. |
| 5229 | */ |
| 5230 | static void ext4_wait_for_tail_page_commit(struct inode *inode) |
| 5231 | { |
| 5232 | struct page *page; |
| 5233 | unsigned offset; |
| 5234 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; |
| 5235 | tid_t commit_tid = 0; |
| 5236 | int ret; |
| 5237 | |
| 5238 | offset = inode->i_size & (PAGE_SIZE - 1); |
| 5239 | /* |
| 5240 | * All buffers in the last page remain valid? Then there's nothing to |
| 5241 | * do. We do the check mainly to optimize the common PAGE_SIZE == |
| 5242 | * blocksize case |
| 5243 | */ |
| 5244 | if (offset > PAGE_SIZE - i_blocksize(inode)) |
| 5245 | return; |
| 5246 | while (1) { |
| 5247 | page = find_lock_page(inode->i_mapping, |
| 5248 | inode->i_size >> PAGE_SHIFT); |
| 5249 | if (!page) |
| 5250 | return; |
| 5251 | ret = __ext4_journalled_invalidatepage(page, offset, |
| 5252 | PAGE_SIZE - offset); |
| 5253 | unlock_page(page); |
| 5254 | put_page(page); |
| 5255 | if (ret != -EBUSY) |
| 5256 | return; |
| 5257 | commit_tid = 0; |
| 5258 | read_lock(&journal->j_state_lock); |
| 5259 | if (journal->j_committing_transaction) |
| 5260 | commit_tid = journal->j_committing_transaction->t_tid; |
| 5261 | read_unlock(&journal->j_state_lock); |
| 5262 | if (commit_tid) |
| 5263 | jbd2_log_wait_commit(journal, commit_tid); |
| 5264 | } |
| 5265 | } |
| 5266 | |
| 5267 | /* |
| 5268 | * ext4_setattr() |
| 5269 | * |
| 5270 | * Called from notify_change. |
| 5271 | * |
| 5272 | * We want to trap VFS attempts to truncate the file as soon as |
| 5273 | * possible. In particular, we want to make sure that when the VFS |
| 5274 | * shrinks i_size, we put the inode on the orphan list and modify |
| 5275 | * i_disksize immediately, so that during the subsequent flushing of |
| 5276 | * dirty pages and freeing of disk blocks, we can guarantee that any |
| 5277 | * commit will leave the blocks being flushed in an unused state on |
| 5278 | * disk. (On recovery, the inode will get truncated and the blocks will |
| 5279 | * be freed, so we have a strong guarantee that no future commit will |
| 5280 | * leave these blocks visible to the user.) |
| 5281 | * |
| 5282 | * Another thing we have to assure is that if we are in ordered mode |
| 5283 | * and inode is still attached to the committing transaction, we must |
| 5284 | * we start writeout of all the dirty pages which are being truncated. |
| 5285 | * This way we are sure that all the data written in the previous |
| 5286 | * transaction are already on disk (truncate waits for pages under |
| 5287 | * writeback). |
| 5288 | * |
| 5289 | * Called with inode->i_mutex down. |
| 5290 | */ |
| 5291 | int ext4_setattr(struct dentry *dentry, struct iattr *attr) |
| 5292 | { |
| 5293 | struct inode *inode = d_inode(dentry); |
| 5294 | int error, rc = 0; |
| 5295 | int orphan = 0; |
| 5296 | const unsigned int ia_valid = attr->ia_valid; |
| 5297 | |
| 5298 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
| 5299 | return -EIO; |
| 5300 | |
| 5301 | error = setattr_prepare(dentry, attr); |
| 5302 | if (error) |
| 5303 | return error; |
| 5304 | |
| 5305 | if (is_quota_modification(inode, attr)) { |
| 5306 | error = dquot_initialize(inode); |
| 5307 | if (error) |
| 5308 | return error; |
| 5309 | } |
| 5310 | if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) || |
| 5311 | (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) { |
| 5312 | handle_t *handle; |
| 5313 | |
| 5314 | /* (user+group)*(old+new) structure, inode write (sb, |
| 5315 | * inode block, ? - but truncate inode update has it) */ |
| 5316 | handle = ext4_journal_start(inode, EXT4_HT_QUOTA, |
| 5317 | (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) + |
| 5318 | EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3); |
| 5319 | if (IS_ERR(handle)) { |
| 5320 | error = PTR_ERR(handle); |
| 5321 | goto err_out; |
| 5322 | } |
| 5323 | error = dquot_transfer(inode, attr); |
| 5324 | if (error) { |
| 5325 | ext4_journal_stop(handle); |
| 5326 | return error; |
| 5327 | } |
| 5328 | /* Update corresponding info in inode so that everything is in |
| 5329 | * one transaction */ |
| 5330 | if (attr->ia_valid & ATTR_UID) |
| 5331 | inode->i_uid = attr->ia_uid; |
| 5332 | if (attr->ia_valid & ATTR_GID) |
| 5333 | inode->i_gid = attr->ia_gid; |
| 5334 | error = ext4_mark_inode_dirty(handle, inode); |
| 5335 | ext4_journal_stop(handle); |
| 5336 | } |
| 5337 | |
| 5338 | if (attr->ia_valid & ATTR_SIZE) { |
| 5339 | handle_t *handle; |
| 5340 | loff_t oldsize = inode->i_size; |
| 5341 | int shrink = (attr->ia_size <= inode->i_size); |
| 5342 | |
| 5343 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
| 5344 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 5345 | |
| 5346 | if (attr->ia_size > sbi->s_bitmap_maxbytes) |
| 5347 | return -EFBIG; |
| 5348 | } |
| 5349 | if (!S_ISREG(inode->i_mode)) |
| 5350 | return -EINVAL; |
| 5351 | |
| 5352 | if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size) |
| 5353 | inode_inc_iversion(inode); |
| 5354 | |
| 5355 | if (ext4_should_order_data(inode) && |
| 5356 | (attr->ia_size < inode->i_size)) { |
| 5357 | error = ext4_begin_ordered_truncate(inode, |
| 5358 | attr->ia_size); |
| 5359 | if (error) |
| 5360 | goto err_out; |
| 5361 | } |
| 5362 | if (attr->ia_size != inode->i_size) { |
| 5363 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 3); |
| 5364 | if (IS_ERR(handle)) { |
| 5365 | error = PTR_ERR(handle); |
| 5366 | goto err_out; |
| 5367 | } |
| 5368 | if (ext4_handle_valid(handle) && shrink) { |
| 5369 | error = ext4_orphan_add(handle, inode); |
| 5370 | orphan = 1; |
| 5371 | } |
| 5372 | /* |
| 5373 | * Update c/mtime on truncate up, ext4_truncate() will |
| 5374 | * update c/mtime in shrink case below |
| 5375 | */ |
| 5376 | if (!shrink) { |
| 5377 | inode->i_mtime = current_time(inode); |
| 5378 | inode->i_ctime = inode->i_mtime; |
| 5379 | } |
| 5380 | down_write(&EXT4_I(inode)->i_data_sem); |
| 5381 | EXT4_I(inode)->i_disksize = attr->ia_size; |
| 5382 | rc = ext4_mark_inode_dirty(handle, inode); |
| 5383 | if (!error) |
| 5384 | error = rc; |
| 5385 | /* |
| 5386 | * We have to update i_size under i_data_sem together |
| 5387 | * with i_disksize to avoid races with writeback code |
| 5388 | * running ext4_wb_update_i_disksize(). |
| 5389 | */ |
| 5390 | if (!error) |
| 5391 | i_size_write(inode, attr->ia_size); |
| 5392 | up_write(&EXT4_I(inode)->i_data_sem); |
| 5393 | ext4_journal_stop(handle); |
| 5394 | if (error) { |
| 5395 | if (orphan) |
| 5396 | ext4_orphan_del(NULL, inode); |
| 5397 | goto err_out; |
| 5398 | } |
| 5399 | } |
| 5400 | if (!shrink) |
| 5401 | pagecache_isize_extended(inode, oldsize, inode->i_size); |
| 5402 | |
| 5403 | /* |
| 5404 | * Blocks are going to be removed from the inode. Wait |
| 5405 | * for dio in flight. Temporarily disable |
| 5406 | * dioread_nolock to prevent livelock. |
| 5407 | */ |
| 5408 | if (orphan) { |
| 5409 | if (!ext4_should_journal_data(inode)) { |
| 5410 | ext4_inode_block_unlocked_dio(inode); |
| 5411 | inode_dio_wait(inode); |
| 5412 | ext4_inode_resume_unlocked_dio(inode); |
| 5413 | } else |
| 5414 | ext4_wait_for_tail_page_commit(inode); |
| 5415 | } |
| 5416 | down_write(&EXT4_I(inode)->i_mmap_sem); |
| 5417 | /* |
| 5418 | * Truncate pagecache after we've waited for commit |
| 5419 | * in data=journal mode to make pages freeable. |
| 5420 | */ |
| 5421 | truncate_pagecache(inode, inode->i_size); |
| 5422 | if (shrink) { |
| 5423 | rc = ext4_truncate(inode); |
| 5424 | if (rc) |
| 5425 | error = rc; |
| 5426 | } |
| 5427 | up_write(&EXT4_I(inode)->i_mmap_sem); |
| 5428 | } |
| 5429 | |
| 5430 | if (!error) { |
| 5431 | setattr_copy(inode, attr); |
| 5432 | mark_inode_dirty(inode); |
| 5433 | } |
| 5434 | |
| 5435 | /* |
| 5436 | * If the call to ext4_truncate failed to get a transaction handle at |
| 5437 | * all, we need to clean up the in-core orphan list manually. |
| 5438 | */ |
| 5439 | if (orphan && inode->i_nlink) |
| 5440 | ext4_orphan_del(NULL, inode); |
| 5441 | |
| 5442 | if (!error && (ia_valid & ATTR_MODE)) |
| 5443 | rc = posix_acl_chmod(inode, inode->i_mode); |
| 5444 | |
| 5445 | err_out: |
| 5446 | ext4_std_error(inode->i_sb, error); |
| 5447 | if (!error) |
| 5448 | error = rc; |
| 5449 | return error; |
| 5450 | } |
| 5451 | |
| 5452 | int ext4_getattr(const struct path *path, struct kstat *stat, |
| 5453 | u32 request_mask, unsigned int query_flags) |
| 5454 | { |
| 5455 | struct inode *inode = d_inode(path->dentry); |
| 5456 | struct ext4_inode *raw_inode; |
| 5457 | struct ext4_inode_info *ei = EXT4_I(inode); |
| 5458 | unsigned int flags; |
| 5459 | |
| 5460 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) { |
| 5461 | stat->result_mask |= STATX_BTIME; |
| 5462 | stat->btime.tv_sec = ei->i_crtime.tv_sec; |
| 5463 | stat->btime.tv_nsec = ei->i_crtime.tv_nsec; |
| 5464 | } |
| 5465 | |
| 5466 | flags = ei->i_flags & EXT4_FL_USER_VISIBLE; |
| 5467 | if (flags & EXT4_APPEND_FL) |
| 5468 | stat->attributes |= STATX_ATTR_APPEND; |
| 5469 | if (flags & EXT4_COMPR_FL) |
| 5470 | stat->attributes |= STATX_ATTR_COMPRESSED; |
| 5471 | if (flags & EXT4_ENCRYPT_FL) |
| 5472 | stat->attributes |= STATX_ATTR_ENCRYPTED; |
| 5473 | if (flags & EXT4_IMMUTABLE_FL) |
| 5474 | stat->attributes |= STATX_ATTR_IMMUTABLE; |
| 5475 | if (flags & EXT4_NODUMP_FL) |
| 5476 | stat->attributes |= STATX_ATTR_NODUMP; |
| 5477 | |
| 5478 | stat->attributes_mask |= (STATX_ATTR_APPEND | |
| 5479 | STATX_ATTR_COMPRESSED | |
| 5480 | STATX_ATTR_ENCRYPTED | |
| 5481 | STATX_ATTR_IMMUTABLE | |
| 5482 | STATX_ATTR_NODUMP); |
| 5483 | |
| 5484 | generic_fillattr(inode, stat); |
| 5485 | return 0; |
| 5486 | } |
| 5487 | |
| 5488 | int ext4_file_getattr(const struct path *path, struct kstat *stat, |
| 5489 | u32 request_mask, unsigned int query_flags) |
| 5490 | { |
| 5491 | struct inode *inode = d_inode(path->dentry); |
| 5492 | u64 delalloc_blocks; |
| 5493 | |
| 5494 | ext4_getattr(path, stat, request_mask, query_flags); |
| 5495 | |
| 5496 | /* |
| 5497 | * If there is inline data in the inode, the inode will normally not |
| 5498 | * have data blocks allocated (it may have an external xattr block). |
| 5499 | * Report at least one sector for such files, so tools like tar, rsync, |
| 5500 | * others don't incorrectly think the file is completely sparse. |
| 5501 | */ |
| 5502 | if (unlikely(ext4_has_inline_data(inode))) |
| 5503 | stat->blocks += (stat->size + 511) >> 9; |
| 5504 | |
| 5505 | /* |
| 5506 | * We can't update i_blocks if the block allocation is delayed |
| 5507 | * otherwise in the case of system crash before the real block |
| 5508 | * allocation is done, we will have i_blocks inconsistent with |
| 5509 | * on-disk file blocks. |
| 5510 | * We always keep i_blocks updated together with real |
| 5511 | * allocation. But to not confuse with user, stat |
| 5512 | * will return the blocks that include the delayed allocation |
| 5513 | * blocks for this file. |
| 5514 | */ |
| 5515 | delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb), |
| 5516 | EXT4_I(inode)->i_reserved_data_blocks); |
| 5517 | stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9); |
| 5518 | return 0; |
| 5519 | } |
| 5520 | |
| 5521 | static int ext4_index_trans_blocks(struct inode *inode, int lblocks, |
| 5522 | int pextents) |
| 5523 | { |
| 5524 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
| 5525 | return ext4_ind_trans_blocks(inode, lblocks); |
| 5526 | return ext4_ext_index_trans_blocks(inode, pextents); |
| 5527 | } |
| 5528 | |
| 5529 | /* |
| 5530 | * Account for index blocks, block groups bitmaps and block group |
| 5531 | * descriptor blocks if modify datablocks and index blocks |
| 5532 | * worse case, the indexs blocks spread over different block groups |
| 5533 | * |
| 5534 | * If datablocks are discontiguous, they are possible to spread over |
| 5535 | * different block groups too. If they are contiguous, with flexbg, |
| 5536 | * they could still across block group boundary. |
| 5537 | * |
| 5538 | * Also account for superblock, inode, quota and xattr blocks |
| 5539 | */ |
| 5540 | int ext4_meta_trans_blocks(struct inode *inode, int lblocks, |
| 5541 | int pextents) |
| 5542 | { |
| 5543 | ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
| 5544 | int gdpblocks; |
| 5545 | int idxblocks; |
| 5546 | int ret = 0; |
| 5547 | |
| 5548 | /* |
| 5549 | * How many index blocks need to touch to map @lblocks logical blocks |
| 5550 | * to @pextents physical extents? |
| 5551 | */ |
| 5552 | idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents); |
| 5553 | |
| 5554 | ret = idxblocks; |
| 5555 | |
| 5556 | /* |
| 5557 | * Now let's see how many group bitmaps and group descriptors need |
| 5558 | * to account |
| 5559 | */ |
| 5560 | groups = idxblocks + pextents; |
| 5561 | gdpblocks = groups; |
| 5562 | if (groups > ngroups) |
| 5563 | groups = ngroups; |
| 5564 | if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
| 5565 | gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; |
| 5566 | |
| 5567 | /* bitmaps and block group descriptor blocks */ |
| 5568 | ret += groups + gdpblocks; |
| 5569 | |
| 5570 | /* Blocks for super block, inode, quota and xattr blocks */ |
| 5571 | ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); |
| 5572 | |
| 5573 | return ret; |
| 5574 | } |
| 5575 | |
| 5576 | /* |
| 5577 | * Calculate the total number of credits to reserve to fit |
| 5578 | * the modification of a single pages into a single transaction, |
| 5579 | * which may include multiple chunks of block allocations. |
| 5580 | * |
| 5581 | * This could be called via ext4_write_begin() |
| 5582 | * |
| 5583 | * We need to consider the worse case, when |
| 5584 | * one new block per extent. |
| 5585 | */ |
| 5586 | int ext4_writepage_trans_blocks(struct inode *inode) |
| 5587 | { |
| 5588 | int bpp = ext4_journal_blocks_per_page(inode); |
| 5589 | int ret; |
| 5590 | |
| 5591 | ret = ext4_meta_trans_blocks(inode, bpp, bpp); |
| 5592 | |
| 5593 | /* Account for data blocks for journalled mode */ |
| 5594 | if (ext4_should_journal_data(inode)) |
| 5595 | ret += bpp; |
| 5596 | return ret; |
| 5597 | } |
| 5598 | |
| 5599 | /* |
| 5600 | * Calculate the journal credits for a chunk of data modification. |
| 5601 | * |
| 5602 | * This is called from DIO, fallocate or whoever calling |
| 5603 | * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. |
| 5604 | * |
| 5605 | * journal buffers for data blocks are not included here, as DIO |
| 5606 | * and fallocate do no need to journal data buffers. |
| 5607 | */ |
| 5608 | int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) |
| 5609 | { |
| 5610 | return ext4_meta_trans_blocks(inode, nrblocks, 1); |
| 5611 | } |
| 5612 | |
| 5613 | /* |
| 5614 | * The caller must have previously called ext4_reserve_inode_write(). |
| 5615 | * Give this, we know that the caller already has write access to iloc->bh. |
| 5616 | */ |
| 5617 | int ext4_mark_iloc_dirty(handle_t *handle, |
| 5618 | struct inode *inode, struct ext4_iloc *iloc) |
| 5619 | { |
| 5620 | int err = 0; |
| 5621 | |
| 5622 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
| 5623 | return -EIO; |
| 5624 | |
| 5625 | if (IS_I_VERSION(inode)) |
| 5626 | inode_inc_iversion(inode); |
| 5627 | |
| 5628 | /* the do_update_inode consumes one bh->b_count */ |
| 5629 | get_bh(iloc->bh); |
| 5630 | |
| 5631 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
| 5632 | err = ext4_do_update_inode(handle, inode, iloc); |
| 5633 | put_bh(iloc->bh); |
| 5634 | return err; |
| 5635 | } |
| 5636 | |
| 5637 | /* |
| 5638 | * On success, We end up with an outstanding reference count against |
| 5639 | * iloc->bh. This _must_ be cleaned up later. |
| 5640 | */ |
| 5641 | |
| 5642 | int |
| 5643 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
| 5644 | struct ext4_iloc *iloc) |
| 5645 | { |
| 5646 | int err; |
| 5647 | |
| 5648 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
| 5649 | return -EIO; |
| 5650 | |
| 5651 | err = ext4_get_inode_loc(inode, iloc); |
| 5652 | if (!err) { |
| 5653 | BUFFER_TRACE(iloc->bh, "get_write_access"); |
| 5654 | err = ext4_journal_get_write_access(handle, iloc->bh); |
| 5655 | if (err) { |
| 5656 | brelse(iloc->bh); |
| 5657 | iloc->bh = NULL; |
| 5658 | } |
| 5659 | } |
| 5660 | ext4_std_error(inode->i_sb, err); |
| 5661 | return err; |
| 5662 | } |
| 5663 | |
| 5664 | /* |
| 5665 | * Expand an inode by new_extra_isize bytes. |
| 5666 | * Returns 0 on success or negative error number on failure. |
| 5667 | */ |
| 5668 | static int ext4_expand_extra_isize(struct inode *inode, |
| 5669 | unsigned int new_extra_isize, |
| 5670 | struct ext4_iloc iloc, |
| 5671 | handle_t *handle) |
| 5672 | { |
| 5673 | struct ext4_inode *raw_inode; |
| 5674 | struct ext4_xattr_ibody_header *header; |
| 5675 | |
| 5676 | if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) |
| 5677 | return 0; |
| 5678 | |
| 5679 | raw_inode = ext4_raw_inode(&iloc); |
| 5680 | |
| 5681 | header = IHDR(inode, raw_inode); |
| 5682 | |
| 5683 | /* No extended attributes present */ |
| 5684 | if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || |
| 5685 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { |
| 5686 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE + |
| 5687 | EXT4_I(inode)->i_extra_isize, 0, |
| 5688 | new_extra_isize - EXT4_I(inode)->i_extra_isize); |
| 5689 | EXT4_I(inode)->i_extra_isize = new_extra_isize; |
| 5690 | return 0; |
| 5691 | } |
| 5692 | |
| 5693 | /* try to expand with EAs present */ |
| 5694 | return ext4_expand_extra_isize_ea(inode, new_extra_isize, |
| 5695 | raw_inode, handle); |
| 5696 | } |
| 5697 | |
| 5698 | /* |
| 5699 | * What we do here is to mark the in-core inode as clean with respect to inode |
| 5700 | * dirtiness (it may still be data-dirty). |
| 5701 | * This means that the in-core inode may be reaped by prune_icache |
| 5702 | * without having to perform any I/O. This is a very good thing, |
| 5703 | * because *any* task may call prune_icache - even ones which |
| 5704 | * have a transaction open against a different journal. |
| 5705 | * |
| 5706 | * Is this cheating? Not really. Sure, we haven't written the |
| 5707 | * inode out, but prune_icache isn't a user-visible syncing function. |
| 5708 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) |
| 5709 | * we start and wait on commits. |
| 5710 | */ |
| 5711 | int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) |
| 5712 | { |
| 5713 | struct ext4_iloc iloc; |
| 5714 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 5715 | static unsigned int mnt_count; |
| 5716 | int err, ret; |
| 5717 | |
| 5718 | might_sleep(); |
| 5719 | trace_ext4_mark_inode_dirty(inode, _RET_IP_); |
| 5720 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
| 5721 | if (err) |
| 5722 | return err; |
| 5723 | if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize && |
| 5724 | !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) { |
| 5725 | /* |
| 5726 | * In nojournal mode, we can immediately attempt to expand |
| 5727 | * the inode. When journaled, we first need to obtain extra |
| 5728 | * buffer credits since we may write into the EA block |
| 5729 | * with this same handle. If journal_extend fails, then it will |
| 5730 | * only result in a minor loss of functionality for that inode. |
| 5731 | * If this is felt to be critical, then e2fsck should be run to |
| 5732 | * force a large enough s_min_extra_isize. |
| 5733 | */ |
| 5734 | if (!ext4_handle_valid(handle) || |
| 5735 | jbd2_journal_extend(handle, |
| 5736 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb)) == 0) { |
| 5737 | ret = ext4_expand_extra_isize(inode, |
| 5738 | sbi->s_want_extra_isize, |
| 5739 | iloc, handle); |
| 5740 | if (ret) { |
| 5741 | if (mnt_count != |
| 5742 | le16_to_cpu(sbi->s_es->s_mnt_count)) { |
| 5743 | ext4_warning(inode->i_sb, |
| 5744 | "Unable to expand inode %lu. Delete" |
| 5745 | " some EAs or run e2fsck.", |
| 5746 | inode->i_ino); |
| 5747 | mnt_count = |
| 5748 | le16_to_cpu(sbi->s_es->s_mnt_count); |
| 5749 | } |
| 5750 | } |
| 5751 | } |
| 5752 | } |
| 5753 | return ext4_mark_iloc_dirty(handle, inode, &iloc); |
| 5754 | } |
| 5755 | |
| 5756 | /* |
| 5757 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
| 5758 | * |
| 5759 | * We're really interested in the case where a file is being extended. |
| 5760 | * i_size has been changed by generic_commit_write() and we thus need |
| 5761 | * to include the updated inode in the current transaction. |
| 5762 | * |
| 5763 | * Also, dquot_alloc_block() will always dirty the inode when blocks |
| 5764 | * are allocated to the file. |
| 5765 | * |
| 5766 | * If the inode is marked synchronous, we don't honour that here - doing |
| 5767 | * so would cause a commit on atime updates, which we don't bother doing. |
| 5768 | * We handle synchronous inodes at the highest possible level. |
| 5769 | * |
| 5770 | * If only the I_DIRTY_TIME flag is set, we can skip everything. If |
| 5771 | * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need |
| 5772 | * to copy into the on-disk inode structure are the timestamp files. |
| 5773 | */ |
| 5774 | void ext4_dirty_inode(struct inode *inode, int flags) |
| 5775 | { |
| 5776 | handle_t *handle; |
| 5777 | |
| 5778 | if (flags == I_DIRTY_TIME) |
| 5779 | return; |
| 5780 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); |
| 5781 | if (IS_ERR(handle)) |
| 5782 | goto out; |
| 5783 | |
| 5784 | ext4_mark_inode_dirty(handle, inode); |
| 5785 | |
| 5786 | ext4_journal_stop(handle); |
| 5787 | out: |
| 5788 | return; |
| 5789 | } |
| 5790 | |
| 5791 | #if 0 |
| 5792 | /* |
| 5793 | * Bind an inode's backing buffer_head into this transaction, to prevent |
| 5794 | * it from being flushed to disk early. Unlike |
| 5795 | * ext4_reserve_inode_write, this leaves behind no bh reference and |
| 5796 | * returns no iloc structure, so the caller needs to repeat the iloc |
| 5797 | * lookup to mark the inode dirty later. |
| 5798 | */ |
| 5799 | static int ext4_pin_inode(handle_t *handle, struct inode *inode) |
| 5800 | { |
| 5801 | struct ext4_iloc iloc; |
| 5802 | |
| 5803 | int err = 0; |
| 5804 | if (handle) { |
| 5805 | err = ext4_get_inode_loc(inode, &iloc); |
| 5806 | if (!err) { |
| 5807 | BUFFER_TRACE(iloc.bh, "get_write_access"); |
| 5808 | err = jbd2_journal_get_write_access(handle, iloc.bh); |
| 5809 | if (!err) |
| 5810 | err = ext4_handle_dirty_metadata(handle, |
| 5811 | NULL, |
| 5812 | iloc.bh); |
| 5813 | brelse(iloc.bh); |
| 5814 | } |
| 5815 | } |
| 5816 | ext4_std_error(inode->i_sb, err); |
| 5817 | return err; |
| 5818 | } |
| 5819 | #endif |
| 5820 | |
| 5821 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
| 5822 | { |
| 5823 | journal_t *journal; |
| 5824 | handle_t *handle; |
| 5825 | int err; |
| 5826 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 5827 | |
| 5828 | /* |
| 5829 | * We have to be very careful here: changing a data block's |
| 5830 | * journaling status dynamically is dangerous. If we write a |
| 5831 | * data block to the journal, change the status and then delete |
| 5832 | * that block, we risk forgetting to revoke the old log record |
| 5833 | * from the journal and so a subsequent replay can corrupt data. |
| 5834 | * So, first we make sure that the journal is empty and that |
| 5835 | * nobody is changing anything. |
| 5836 | */ |
| 5837 | |
| 5838 | journal = EXT4_JOURNAL(inode); |
| 5839 | if (!journal) |
| 5840 | return 0; |
| 5841 | if (is_journal_aborted(journal)) |
| 5842 | return -EROFS; |
| 5843 | |
| 5844 | /* Wait for all existing dio workers */ |
| 5845 | ext4_inode_block_unlocked_dio(inode); |
| 5846 | inode_dio_wait(inode); |
| 5847 | |
| 5848 | /* |
| 5849 | * Before flushing the journal and switching inode's aops, we have |
| 5850 | * to flush all dirty data the inode has. There can be outstanding |
| 5851 | * delayed allocations, there can be unwritten extents created by |
| 5852 | * fallocate or buffered writes in dioread_nolock mode covered by |
| 5853 | * dirty data which can be converted only after flushing the dirty |
| 5854 | * data (and journalled aops don't know how to handle these cases). |
| 5855 | */ |
| 5856 | if (val) { |
| 5857 | down_write(&EXT4_I(inode)->i_mmap_sem); |
| 5858 | err = filemap_write_and_wait(inode->i_mapping); |
| 5859 | if (err < 0) { |
| 5860 | up_write(&EXT4_I(inode)->i_mmap_sem); |
| 5861 | ext4_inode_resume_unlocked_dio(inode); |
| 5862 | return err; |
| 5863 | } |
| 5864 | } |
| 5865 | |
| 5866 | percpu_down_write(&sbi->s_journal_flag_rwsem); |
| 5867 | jbd2_journal_lock_updates(journal); |
| 5868 | |
| 5869 | /* |
| 5870 | * OK, there are no updates running now, and all cached data is |
| 5871 | * synced to disk. We are now in a completely consistent state |
| 5872 | * which doesn't have anything in the journal, and we know that |
| 5873 | * no filesystem updates are running, so it is safe to modify |
| 5874 | * the inode's in-core data-journaling state flag now. |
| 5875 | */ |
| 5876 | |
| 5877 | if (val) |
| 5878 | ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
| 5879 | else { |
| 5880 | err = jbd2_journal_flush(journal); |
| 5881 | if (err < 0) { |
| 5882 | jbd2_journal_unlock_updates(journal); |
| 5883 | percpu_up_write(&sbi->s_journal_flag_rwsem); |
| 5884 | ext4_inode_resume_unlocked_dio(inode); |
| 5885 | return err; |
| 5886 | } |
| 5887 | ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
| 5888 | } |
| 5889 | ext4_set_aops(inode); |
| 5890 | /* |
| 5891 | * Update inode->i_flags after EXT4_INODE_JOURNAL_DATA was updated. |
| 5892 | * E.g. S_DAX may get cleared / set. |
| 5893 | */ |
| 5894 | ext4_set_inode_flags(inode); |
| 5895 | |
| 5896 | jbd2_journal_unlock_updates(journal); |
| 5897 | percpu_up_write(&sbi->s_journal_flag_rwsem); |
| 5898 | |
| 5899 | if (val) |
| 5900 | up_write(&EXT4_I(inode)->i_mmap_sem); |
| 5901 | ext4_inode_resume_unlocked_dio(inode); |
| 5902 | |
| 5903 | /* Finally we can mark the inode as dirty. */ |
| 5904 | |
| 5905 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); |
| 5906 | if (IS_ERR(handle)) |
| 5907 | return PTR_ERR(handle); |
| 5908 | |
| 5909 | err = ext4_mark_inode_dirty(handle, inode); |
| 5910 | ext4_handle_sync(handle); |
| 5911 | ext4_journal_stop(handle); |
| 5912 | ext4_std_error(inode->i_sb, err); |
| 5913 | |
| 5914 | return err; |
| 5915 | } |
| 5916 | |
| 5917 | static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh) |
| 5918 | { |
| 5919 | return !buffer_mapped(bh); |
| 5920 | } |
| 5921 | |
| 5922 | int ext4_page_mkwrite(struct vm_fault *vmf) |
| 5923 | { |
| 5924 | struct vm_area_struct *vma = vmf->vma; |
| 5925 | struct page *page = vmf->page; |
| 5926 | loff_t size; |
| 5927 | unsigned long len; |
| 5928 | int ret; |
| 5929 | struct file *file = vma->vm_file; |
| 5930 | struct inode *inode = file_inode(file); |
| 5931 | struct address_space *mapping = inode->i_mapping; |
| 5932 | handle_t *handle; |
| 5933 | get_block_t *get_block; |
| 5934 | int retries = 0; |
| 5935 | |
| 5936 | sb_start_pagefault(inode->i_sb); |
| 5937 | file_update_time(vma->vm_file); |
| 5938 | |
| 5939 | down_read(&EXT4_I(inode)->i_mmap_sem); |
| 5940 | |
| 5941 | ret = ext4_convert_inline_data(inode); |
| 5942 | if (ret) |
| 5943 | goto out_ret; |
| 5944 | |
| 5945 | /* Delalloc case is easy... */ |
| 5946 | if (test_opt(inode->i_sb, DELALLOC) && |
| 5947 | !ext4_should_journal_data(inode) && |
| 5948 | !ext4_nonda_switch(inode->i_sb)) { |
| 5949 | do { |
| 5950 | ret = block_page_mkwrite(vma, vmf, |
| 5951 | ext4_da_get_block_prep); |
| 5952 | } while (ret == -ENOSPC && |
| 5953 | ext4_should_retry_alloc(inode->i_sb, &retries)); |
| 5954 | goto out_ret; |
| 5955 | } |
| 5956 | |
| 5957 | lock_page(page); |
| 5958 | size = i_size_read(inode); |
| 5959 | /* Page got truncated from under us? */ |
| 5960 | if (page->mapping != mapping || page_offset(page) > size) { |
| 5961 | unlock_page(page); |
| 5962 | ret = VM_FAULT_NOPAGE; |
| 5963 | goto out; |
| 5964 | } |
| 5965 | |
| 5966 | if (page->index == size >> PAGE_SHIFT) |
| 5967 | len = size & ~PAGE_MASK; |
| 5968 | else |
| 5969 | len = PAGE_SIZE; |
| 5970 | /* |
| 5971 | * Return if we have all the buffers mapped. This avoids the need to do |
| 5972 | * journal_start/journal_stop which can block and take a long time |
| 5973 | */ |
| 5974 | if (page_has_buffers(page)) { |
| 5975 | if (!ext4_walk_page_buffers(NULL, page_buffers(page), |
| 5976 | 0, len, NULL, |
| 5977 | ext4_bh_unmapped)) { |
| 5978 | /* Wait so that we don't change page under IO */ |
| 5979 | wait_for_stable_page(page); |
| 5980 | ret = VM_FAULT_LOCKED; |
| 5981 | goto out; |
| 5982 | } |
| 5983 | } |
| 5984 | unlock_page(page); |
| 5985 | /* OK, we need to fill the hole... */ |
| 5986 | if (ext4_should_dioread_nolock(inode)) |
| 5987 | get_block = ext4_get_block_unwritten; |
| 5988 | else |
| 5989 | get_block = ext4_get_block; |
| 5990 | retry_alloc: |
| 5991 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
| 5992 | ext4_writepage_trans_blocks(inode)); |
| 5993 | if (IS_ERR(handle)) { |
| 5994 | ret = VM_FAULT_SIGBUS; |
| 5995 | goto out; |
| 5996 | } |
| 5997 | ret = block_page_mkwrite(vma, vmf, get_block); |
| 5998 | if (!ret && ext4_should_journal_data(inode)) { |
| 5999 | if (ext4_walk_page_buffers(handle, page_buffers(page), 0, |
| 6000 | PAGE_SIZE, NULL, do_journal_get_write_access)) { |
| 6001 | unlock_page(page); |
| 6002 | ret = VM_FAULT_SIGBUS; |
| 6003 | ext4_journal_stop(handle); |
| 6004 | goto out; |
| 6005 | } |
| 6006 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
| 6007 | } |
| 6008 | ext4_journal_stop(handle); |
| 6009 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
| 6010 | goto retry_alloc; |
| 6011 | out_ret: |
| 6012 | ret = block_page_mkwrite_return(ret); |
| 6013 | out: |
| 6014 | up_read(&EXT4_I(inode)->i_mmap_sem); |
| 6015 | sb_end_pagefault(inode->i_sb); |
| 6016 | return ret; |
| 6017 | } |
| 6018 | |
| 6019 | int ext4_filemap_fault(struct vm_fault *vmf) |
| 6020 | { |
| 6021 | struct inode *inode = file_inode(vmf->vma->vm_file); |
| 6022 | int err; |
| 6023 | |
| 6024 | down_read(&EXT4_I(inode)->i_mmap_sem); |
| 6025 | err = filemap_fault(vmf); |
| 6026 | up_read(&EXT4_I(inode)->i_mmap_sem); |
| 6027 | |
| 6028 | return err; |
| 6029 | } |
| 6030 | |
| 6031 | /* |
| 6032 | * Find the first extent at or after @lblk in an inode that is not a hole. |
| 6033 | * Search for @map_len blocks at most. The extent is returned in @result. |
| 6034 | * |
| 6035 | * The function returns 1 if we found an extent. The function returns 0 in |
| 6036 | * case there is no extent at or after @lblk and in that case also sets |
| 6037 | * @result->es_len to 0. In case of error, the error code is returned. |
| 6038 | */ |
| 6039 | int ext4_get_next_extent(struct inode *inode, ext4_lblk_t lblk, |
| 6040 | unsigned int map_len, struct extent_status *result) |
| 6041 | { |
| 6042 | struct ext4_map_blocks map; |
| 6043 | struct extent_status es = {}; |
| 6044 | int ret; |
| 6045 | |
| 6046 | map.m_lblk = lblk; |
| 6047 | map.m_len = map_len; |
| 6048 | |
| 6049 | /* |
| 6050 | * For non-extent based files this loop may iterate several times since |
| 6051 | * we do not determine full hole size. |
| 6052 | */ |
| 6053 | while (map.m_len > 0) { |
| 6054 | ret = ext4_map_blocks(NULL, inode, &map, 0); |
| 6055 | if (ret < 0) |
| 6056 | return ret; |
| 6057 | /* There's extent covering m_lblk? Just return it. */ |
| 6058 | if (ret > 0) { |
| 6059 | int status; |
| 6060 | |
| 6061 | ext4_es_store_pblock(result, map.m_pblk); |
| 6062 | result->es_lblk = map.m_lblk; |
| 6063 | result->es_len = map.m_len; |
| 6064 | if (map.m_flags & EXT4_MAP_UNWRITTEN) |
| 6065 | status = EXTENT_STATUS_UNWRITTEN; |
| 6066 | else |
| 6067 | status = EXTENT_STATUS_WRITTEN; |
| 6068 | ext4_es_store_status(result, status); |
| 6069 | return 1; |
| 6070 | } |
| 6071 | ext4_es_find_delayed_extent_range(inode, map.m_lblk, |
| 6072 | map.m_lblk + map.m_len - 1, |
| 6073 | &es); |
| 6074 | /* Is delalloc data before next block in extent tree? */ |
| 6075 | if (es.es_len && es.es_lblk < map.m_lblk + map.m_len) { |
| 6076 | ext4_lblk_t offset = 0; |
| 6077 | |
| 6078 | if (es.es_lblk < lblk) |
| 6079 | offset = lblk - es.es_lblk; |
| 6080 | result->es_lblk = es.es_lblk + offset; |
| 6081 | ext4_es_store_pblock(result, |
| 6082 | ext4_es_pblock(&es) + offset); |
| 6083 | result->es_len = es.es_len - offset; |
| 6084 | ext4_es_store_status(result, ext4_es_status(&es)); |
| 6085 | |
| 6086 | return 1; |
| 6087 | } |
| 6088 | /* There's a hole at m_lblk, advance us after it */ |
| 6089 | map.m_lblk += map.m_len; |
| 6090 | map_len -= map.m_len; |
| 6091 | map.m_len = map_len; |
| 6092 | cond_resched(); |
| 6093 | } |
| 6094 | result->es_len = 0; |
| 6095 | return 0; |
| 6096 | } |