| 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/jbd2.h> |
| 24 | #include <linux/highuid.h> |
| 25 | #include <linux/pagemap.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/ratelimit.h> |
| 40 | |
| 41 | #include "ext4_jbd2.h" |
| 42 | #include "xattr.h" |
| 43 | #include "acl.h" |
| 44 | #include "truncate.h" |
| 45 | |
| 46 | #include <trace/events/ext4.h> |
| 47 | |
| 48 | #define MPAGE_DA_EXTENT_TAIL 0x01 |
| 49 | |
| 50 | static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw, |
| 51 | struct ext4_inode_info *ei) |
| 52 | { |
| 53 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 54 | __u16 csum_lo; |
| 55 | __u16 csum_hi = 0; |
| 56 | __u32 csum; |
| 57 | |
| 58 | csum_lo = raw->i_checksum_lo; |
| 59 | raw->i_checksum_lo = 0; |
| 60 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && |
| 61 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) { |
| 62 | csum_hi = raw->i_checksum_hi; |
| 63 | raw->i_checksum_hi = 0; |
| 64 | } |
| 65 | |
| 66 | csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, |
| 67 | EXT4_INODE_SIZE(inode->i_sb)); |
| 68 | |
| 69 | raw->i_checksum_lo = csum_lo; |
| 70 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && |
| 71 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) |
| 72 | raw->i_checksum_hi = csum_hi; |
| 73 | |
| 74 | return csum; |
| 75 | } |
| 76 | |
| 77 | static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw, |
| 78 | struct ext4_inode_info *ei) |
| 79 | { |
| 80 | __u32 provided, calculated; |
| 81 | |
| 82 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != |
| 83 | cpu_to_le32(EXT4_OS_LINUX) || |
| 84 | !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, |
| 85 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) |
| 86 | return 1; |
| 87 | |
| 88 | provided = le16_to_cpu(raw->i_checksum_lo); |
| 89 | calculated = ext4_inode_csum(inode, raw, ei); |
| 90 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && |
| 91 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) |
| 92 | provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16; |
| 93 | else |
| 94 | calculated &= 0xFFFF; |
| 95 | |
| 96 | return provided == calculated; |
| 97 | } |
| 98 | |
| 99 | static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw, |
| 100 | struct ext4_inode_info *ei) |
| 101 | { |
| 102 | __u32 csum; |
| 103 | |
| 104 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != |
| 105 | cpu_to_le32(EXT4_OS_LINUX) || |
| 106 | !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, |
| 107 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) |
| 108 | return; |
| 109 | |
| 110 | csum = ext4_inode_csum(inode, raw, ei); |
| 111 | raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF); |
| 112 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && |
| 113 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) |
| 114 | raw->i_checksum_hi = cpu_to_le16(csum >> 16); |
| 115 | } |
| 116 | |
| 117 | static inline int ext4_begin_ordered_truncate(struct inode *inode, |
| 118 | loff_t new_size) |
| 119 | { |
| 120 | trace_ext4_begin_ordered_truncate(inode, new_size); |
| 121 | /* |
| 122 | * If jinode is zero, then we never opened the file for |
| 123 | * writing, so there's no need to call |
| 124 | * jbd2_journal_begin_ordered_truncate() since there's no |
| 125 | * outstanding writes we need to flush. |
| 126 | */ |
| 127 | if (!EXT4_I(inode)->jinode) |
| 128 | return 0; |
| 129 | return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode), |
| 130 | EXT4_I(inode)->jinode, |
| 131 | new_size); |
| 132 | } |
| 133 | |
| 134 | static void ext4_invalidatepage(struct page *page, unsigned long offset); |
| 135 | static int __ext4_journalled_writepage(struct page *page, unsigned int len); |
| 136 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh); |
| 137 | static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle, |
| 138 | struct inode *inode, struct page *page, loff_t from, |
| 139 | loff_t length, int flags); |
| 140 | |
| 141 | /* |
| 142 | * Test whether an inode is a fast symlink. |
| 143 | */ |
| 144 | static int ext4_inode_is_fast_symlink(struct inode *inode) |
| 145 | { |
| 146 | int ea_blocks = EXT4_I(inode)->i_file_acl ? |
| 147 | (inode->i_sb->s_blocksize >> 9) : 0; |
| 148 | |
| 149 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); |
| 150 | } |
| 151 | |
| 152 | /* |
| 153 | * Restart the transaction associated with *handle. This does a commit, |
| 154 | * so before we call here everything must be consistently dirtied against |
| 155 | * this transaction. |
| 156 | */ |
| 157 | int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode, |
| 158 | int nblocks) |
| 159 | { |
| 160 | int ret; |
| 161 | |
| 162 | /* |
| 163 | * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this |
| 164 | * moment, get_block can be called only for blocks inside i_size since |
| 165 | * page cache has been already dropped and writes are blocked by |
| 166 | * i_mutex. So we can safely drop the i_data_sem here. |
| 167 | */ |
| 168 | BUG_ON(EXT4_JOURNAL(inode) == NULL); |
| 169 | jbd_debug(2, "restarting handle %p\n", handle); |
| 170 | up_write(&EXT4_I(inode)->i_data_sem); |
| 171 | ret = ext4_journal_restart(handle, nblocks); |
| 172 | down_write(&EXT4_I(inode)->i_data_sem); |
| 173 | ext4_discard_preallocations(inode); |
| 174 | |
| 175 | return ret; |
| 176 | } |
| 177 | |
| 178 | /* |
| 179 | * Called at the last iput() if i_nlink is zero. |
| 180 | */ |
| 181 | void ext4_evict_inode(struct inode *inode) |
| 182 | { |
| 183 | handle_t *handle; |
| 184 | int err; |
| 185 | |
| 186 | trace_ext4_evict_inode(inode); |
| 187 | |
| 188 | ext4_ioend_wait(inode); |
| 189 | |
| 190 | if (inode->i_nlink) { |
| 191 | /* |
| 192 | * When journalling data dirty buffers are tracked only in the |
| 193 | * journal. So although mm thinks everything is clean and |
| 194 | * ready for reaping the inode might still have some pages to |
| 195 | * write in the running transaction or waiting to be |
| 196 | * checkpointed. Thus calling jbd2_journal_invalidatepage() |
| 197 | * (via truncate_inode_pages()) to discard these buffers can |
| 198 | * cause data loss. Also even if we did not discard these |
| 199 | * buffers, we would have no way to find them after the inode |
| 200 | * is reaped and thus user could see stale data if he tries to |
| 201 | * read them before the transaction is checkpointed. So be |
| 202 | * careful and force everything to disk here... We use |
| 203 | * ei->i_datasync_tid to store the newest transaction |
| 204 | * containing inode's data. |
| 205 | * |
| 206 | * Note that directories do not have this problem because they |
| 207 | * don't use page cache. |
| 208 | */ |
| 209 | if (ext4_should_journal_data(inode) && |
| 210 | (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) { |
| 211 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; |
| 212 | tid_t commit_tid = EXT4_I(inode)->i_datasync_tid; |
| 213 | |
| 214 | jbd2_log_start_commit(journal, commit_tid); |
| 215 | jbd2_log_wait_commit(journal, commit_tid); |
| 216 | filemap_write_and_wait(&inode->i_data); |
| 217 | } |
| 218 | truncate_inode_pages(&inode->i_data, 0); |
| 219 | goto no_delete; |
| 220 | } |
| 221 | |
| 222 | if (!is_bad_inode(inode)) |
| 223 | dquot_initialize(inode); |
| 224 | |
| 225 | if (ext4_should_order_data(inode)) |
| 226 | ext4_begin_ordered_truncate(inode, 0); |
| 227 | truncate_inode_pages(&inode->i_data, 0); |
| 228 | |
| 229 | if (is_bad_inode(inode)) |
| 230 | goto no_delete; |
| 231 | |
| 232 | /* |
| 233 | * Protect us against freezing - iput() caller didn't have to have any |
| 234 | * protection against it |
| 235 | */ |
| 236 | sb_start_intwrite(inode->i_sb); |
| 237 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, |
| 238 | ext4_blocks_for_truncate(inode)+3); |
| 239 | if (IS_ERR(handle)) { |
| 240 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); |
| 241 | /* |
| 242 | * If we're going to skip the normal cleanup, we still need to |
| 243 | * make sure that the in-core orphan linked list is properly |
| 244 | * cleaned up. |
| 245 | */ |
| 246 | ext4_orphan_del(NULL, inode); |
| 247 | sb_end_intwrite(inode->i_sb); |
| 248 | goto no_delete; |
| 249 | } |
| 250 | |
| 251 | if (IS_SYNC(inode)) |
| 252 | ext4_handle_sync(handle); |
| 253 | inode->i_size = 0; |
| 254 | err = ext4_mark_inode_dirty(handle, inode); |
| 255 | if (err) { |
| 256 | ext4_warning(inode->i_sb, |
| 257 | "couldn't mark inode dirty (err %d)", err); |
| 258 | goto stop_handle; |
| 259 | } |
| 260 | if (inode->i_blocks) |
| 261 | ext4_truncate(inode); |
| 262 | |
| 263 | /* |
| 264 | * ext4_ext_truncate() doesn't reserve any slop when it |
| 265 | * restarts journal transactions; therefore there may not be |
| 266 | * enough credits left in the handle to remove the inode from |
| 267 | * the orphan list and set the dtime field. |
| 268 | */ |
| 269 | if (!ext4_handle_has_enough_credits(handle, 3)) { |
| 270 | err = ext4_journal_extend(handle, 3); |
| 271 | if (err > 0) |
| 272 | err = ext4_journal_restart(handle, 3); |
| 273 | if (err != 0) { |
| 274 | ext4_warning(inode->i_sb, |
| 275 | "couldn't extend journal (err %d)", err); |
| 276 | stop_handle: |
| 277 | ext4_journal_stop(handle); |
| 278 | ext4_orphan_del(NULL, inode); |
| 279 | sb_end_intwrite(inode->i_sb); |
| 280 | goto no_delete; |
| 281 | } |
| 282 | } |
| 283 | |
| 284 | /* |
| 285 | * Kill off the orphan record which ext4_truncate created. |
| 286 | * AKPM: I think this can be inside the above `if'. |
| 287 | * Note that ext4_orphan_del() has to be able to cope with the |
| 288 | * deletion of a non-existent orphan - this is because we don't |
| 289 | * know if ext4_truncate() actually created an orphan record. |
| 290 | * (Well, we could do this if we need to, but heck - it works) |
| 291 | */ |
| 292 | ext4_orphan_del(handle, inode); |
| 293 | EXT4_I(inode)->i_dtime = get_seconds(); |
| 294 | |
| 295 | /* |
| 296 | * One subtle ordering requirement: if anything has gone wrong |
| 297 | * (transaction abort, IO errors, whatever), then we can still |
| 298 | * do these next steps (the fs will already have been marked as |
| 299 | * having errors), but we can't free the inode if the mark_dirty |
| 300 | * fails. |
| 301 | */ |
| 302 | if (ext4_mark_inode_dirty(handle, inode)) |
| 303 | /* If that failed, just do the required in-core inode clear. */ |
| 304 | ext4_clear_inode(inode); |
| 305 | else |
| 306 | ext4_free_inode(handle, inode); |
| 307 | ext4_journal_stop(handle); |
| 308 | sb_end_intwrite(inode->i_sb); |
| 309 | return; |
| 310 | no_delete: |
| 311 | ext4_clear_inode(inode); /* We must guarantee clearing of inode... */ |
| 312 | } |
| 313 | |
| 314 | #ifdef CONFIG_QUOTA |
| 315 | qsize_t *ext4_get_reserved_space(struct inode *inode) |
| 316 | { |
| 317 | return &EXT4_I(inode)->i_reserved_quota; |
| 318 | } |
| 319 | #endif |
| 320 | |
| 321 | /* |
| 322 | * Calculate the number of metadata blocks need to reserve |
| 323 | * to allocate a block located at @lblock |
| 324 | */ |
| 325 | static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock) |
| 326 | { |
| 327 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
| 328 | return ext4_ext_calc_metadata_amount(inode, lblock); |
| 329 | |
| 330 | return ext4_ind_calc_metadata_amount(inode, lblock); |
| 331 | } |
| 332 | |
| 333 | /* |
| 334 | * Called with i_data_sem down, which is important since we can call |
| 335 | * ext4_discard_preallocations() from here. |
| 336 | */ |
| 337 | void ext4_da_update_reserve_space(struct inode *inode, |
| 338 | int used, int quota_claim) |
| 339 | { |
| 340 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 341 | struct ext4_inode_info *ei = EXT4_I(inode); |
| 342 | |
| 343 | spin_lock(&ei->i_block_reservation_lock); |
| 344 | trace_ext4_da_update_reserve_space(inode, used, quota_claim); |
| 345 | if (unlikely(used > ei->i_reserved_data_blocks)) { |
| 346 | ext4_warning(inode->i_sb, "%s: ino %lu, used %d " |
| 347 | "with only %d reserved data blocks", |
| 348 | __func__, inode->i_ino, used, |
| 349 | ei->i_reserved_data_blocks); |
| 350 | WARN_ON(1); |
| 351 | used = ei->i_reserved_data_blocks; |
| 352 | } |
| 353 | |
| 354 | if (unlikely(ei->i_allocated_meta_blocks > ei->i_reserved_meta_blocks)) { |
| 355 | ext4_warning(inode->i_sb, "ino %lu, allocated %d " |
| 356 | "with only %d reserved metadata blocks " |
| 357 | "(releasing %d blocks with reserved %d data blocks)", |
| 358 | inode->i_ino, ei->i_allocated_meta_blocks, |
| 359 | ei->i_reserved_meta_blocks, used, |
| 360 | ei->i_reserved_data_blocks); |
| 361 | WARN_ON(1); |
| 362 | ei->i_allocated_meta_blocks = ei->i_reserved_meta_blocks; |
| 363 | } |
| 364 | |
| 365 | /* Update per-inode reservations */ |
| 366 | ei->i_reserved_data_blocks -= used; |
| 367 | ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks; |
| 368 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, |
| 369 | used + ei->i_allocated_meta_blocks); |
| 370 | ei->i_allocated_meta_blocks = 0; |
| 371 | |
| 372 | if (ei->i_reserved_data_blocks == 0) { |
| 373 | /* |
| 374 | * We can release all of the reserved metadata blocks |
| 375 | * only when we have written all of the delayed |
| 376 | * allocation blocks. |
| 377 | */ |
| 378 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, |
| 379 | ei->i_reserved_meta_blocks); |
| 380 | ei->i_reserved_meta_blocks = 0; |
| 381 | ei->i_da_metadata_calc_len = 0; |
| 382 | } |
| 383 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
| 384 | |
| 385 | /* Update quota subsystem for data blocks */ |
| 386 | if (quota_claim) |
| 387 | dquot_claim_block(inode, EXT4_C2B(sbi, used)); |
| 388 | else { |
| 389 | /* |
| 390 | * We did fallocate with an offset that is already delayed |
| 391 | * allocated. So on delayed allocated writeback we should |
| 392 | * not re-claim the quota for fallocated blocks. |
| 393 | */ |
| 394 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, used)); |
| 395 | } |
| 396 | |
| 397 | /* |
| 398 | * If we have done all the pending block allocations and if |
| 399 | * there aren't any writers on the inode, we can discard the |
| 400 | * inode's preallocations. |
| 401 | */ |
| 402 | if ((ei->i_reserved_data_blocks == 0) && |
| 403 | (atomic_read(&inode->i_writecount) == 0)) |
| 404 | ext4_discard_preallocations(inode); |
| 405 | } |
| 406 | |
| 407 | static int __check_block_validity(struct inode *inode, const char *func, |
| 408 | unsigned int line, |
| 409 | struct ext4_map_blocks *map) |
| 410 | { |
| 411 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk, |
| 412 | map->m_len)) { |
| 413 | ext4_error_inode(inode, func, line, map->m_pblk, |
| 414 | "lblock %lu mapped to illegal pblock " |
| 415 | "(length %d)", (unsigned long) map->m_lblk, |
| 416 | map->m_len); |
| 417 | return -EIO; |
| 418 | } |
| 419 | return 0; |
| 420 | } |
| 421 | |
| 422 | #define check_block_validity(inode, map) \ |
| 423 | __check_block_validity((inode), __func__, __LINE__, (map)) |
| 424 | |
| 425 | /* |
| 426 | * Return the number of contiguous dirty pages in a given inode |
| 427 | * starting at page frame idx. |
| 428 | */ |
| 429 | static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx, |
| 430 | unsigned int max_pages) |
| 431 | { |
| 432 | struct address_space *mapping = inode->i_mapping; |
| 433 | pgoff_t index; |
| 434 | struct pagevec pvec; |
| 435 | pgoff_t num = 0; |
| 436 | int i, nr_pages, done = 0; |
| 437 | |
| 438 | if (max_pages == 0) |
| 439 | return 0; |
| 440 | pagevec_init(&pvec, 0); |
| 441 | while (!done) { |
| 442 | index = idx; |
| 443 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, |
| 444 | PAGECACHE_TAG_DIRTY, |
| 445 | (pgoff_t)PAGEVEC_SIZE); |
| 446 | if (nr_pages == 0) |
| 447 | break; |
| 448 | for (i = 0; i < nr_pages; i++) { |
| 449 | struct page *page = pvec.pages[i]; |
| 450 | struct buffer_head *bh, *head; |
| 451 | |
| 452 | lock_page(page); |
| 453 | if (unlikely(page->mapping != mapping) || |
| 454 | !PageDirty(page) || |
| 455 | PageWriteback(page) || |
| 456 | page->index != idx) { |
| 457 | done = 1; |
| 458 | unlock_page(page); |
| 459 | break; |
| 460 | } |
| 461 | if (page_has_buffers(page)) { |
| 462 | bh = head = page_buffers(page); |
| 463 | do { |
| 464 | if (!buffer_delay(bh) && |
| 465 | !buffer_unwritten(bh)) |
| 466 | done = 1; |
| 467 | bh = bh->b_this_page; |
| 468 | } while (!done && (bh != head)); |
| 469 | } |
| 470 | unlock_page(page); |
| 471 | if (done) |
| 472 | break; |
| 473 | idx++; |
| 474 | num++; |
| 475 | if (num >= max_pages) { |
| 476 | done = 1; |
| 477 | break; |
| 478 | } |
| 479 | } |
| 480 | pagevec_release(&pvec); |
| 481 | } |
| 482 | return num; |
| 483 | } |
| 484 | |
| 485 | /* |
| 486 | * The ext4_map_blocks() function tries to look up the requested blocks, |
| 487 | * and returns if the blocks are already mapped. |
| 488 | * |
| 489 | * Otherwise it takes the write lock of the i_data_sem and allocate blocks |
| 490 | * and store the allocated blocks in the result buffer head and mark it |
| 491 | * mapped. |
| 492 | * |
| 493 | * If file type is extents based, it will call ext4_ext_map_blocks(), |
| 494 | * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping |
| 495 | * based files |
| 496 | * |
| 497 | * On success, it returns the number of blocks being mapped or allocate. |
| 498 | * if create==0 and the blocks are pre-allocated and uninitialized block, |
| 499 | * the result buffer head is unmapped. If the create ==1, it will make sure |
| 500 | * the buffer head is mapped. |
| 501 | * |
| 502 | * It returns 0 if plain look up failed (blocks have not been allocated), in |
| 503 | * that case, buffer head is unmapped |
| 504 | * |
| 505 | * It returns the error in case of allocation failure. |
| 506 | */ |
| 507 | int ext4_map_blocks(handle_t *handle, struct inode *inode, |
| 508 | struct ext4_map_blocks *map, int flags) |
| 509 | { |
| 510 | int retval; |
| 511 | |
| 512 | map->m_flags = 0; |
| 513 | ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u," |
| 514 | "logical block %lu\n", inode->i_ino, flags, map->m_len, |
| 515 | (unsigned long) map->m_lblk); |
| 516 | /* |
| 517 | * Try to see if we can get the block without requesting a new |
| 518 | * file system block. |
| 519 | */ |
| 520 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) |
| 521 | down_read((&EXT4_I(inode)->i_data_sem)); |
| 522 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
| 523 | retval = ext4_ext_map_blocks(handle, inode, map, flags & |
| 524 | EXT4_GET_BLOCKS_KEEP_SIZE); |
| 525 | } else { |
| 526 | retval = ext4_ind_map_blocks(handle, inode, map, flags & |
| 527 | EXT4_GET_BLOCKS_KEEP_SIZE); |
| 528 | } |
| 529 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) |
| 530 | up_read((&EXT4_I(inode)->i_data_sem)); |
| 531 | |
| 532 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
| 533 | int ret; |
| 534 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) { |
| 535 | /* delayed alloc may be allocated by fallocate and |
| 536 | * coverted to initialized by directIO. |
| 537 | * we need to handle delayed extent here. |
| 538 | */ |
| 539 | down_write((&EXT4_I(inode)->i_data_sem)); |
| 540 | goto delayed_mapped; |
| 541 | } |
| 542 | ret = check_block_validity(inode, map); |
| 543 | if (ret != 0) |
| 544 | return ret; |
| 545 | } |
| 546 | |
| 547 | /* If it is only a block(s) look up */ |
| 548 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) |
| 549 | return retval; |
| 550 | |
| 551 | /* |
| 552 | * Returns if the blocks have already allocated |
| 553 | * |
| 554 | * Note that if blocks have been preallocated |
| 555 | * ext4_ext_get_block() returns the create = 0 |
| 556 | * with buffer head unmapped. |
| 557 | */ |
| 558 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) |
| 559 | return retval; |
| 560 | |
| 561 | /* |
| 562 | * When we call get_blocks without the create flag, the |
| 563 | * BH_Unwritten flag could have gotten set if the blocks |
| 564 | * requested were part of a uninitialized extent. We need to |
| 565 | * clear this flag now that we are committed to convert all or |
| 566 | * part of the uninitialized extent to be an initialized |
| 567 | * extent. This is because we need to avoid the combination |
| 568 | * of BH_Unwritten and BH_Mapped flags being simultaneously |
| 569 | * set on the buffer_head. |
| 570 | */ |
| 571 | map->m_flags &= ~EXT4_MAP_UNWRITTEN; |
| 572 | |
| 573 | /* |
| 574 | * New blocks allocate and/or writing to uninitialized extent |
| 575 | * will possibly result in updating i_data, so we take |
| 576 | * the write lock of i_data_sem, and call get_blocks() |
| 577 | * with create == 1 flag. |
| 578 | */ |
| 579 | down_write((&EXT4_I(inode)->i_data_sem)); |
| 580 | |
| 581 | /* |
| 582 | * if the caller is from delayed allocation writeout path |
| 583 | * we have already reserved fs blocks for allocation |
| 584 | * let the underlying get_block() function know to |
| 585 | * avoid double accounting |
| 586 | */ |
| 587 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
| 588 | ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED); |
| 589 | /* |
| 590 | * We need to check for EXT4 here because migrate |
| 591 | * could have changed the inode type in between |
| 592 | */ |
| 593 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
| 594 | retval = ext4_ext_map_blocks(handle, inode, map, flags); |
| 595 | } else { |
| 596 | retval = ext4_ind_map_blocks(handle, inode, map, flags); |
| 597 | |
| 598 | if (retval > 0 && map->m_flags & EXT4_MAP_NEW) { |
| 599 | /* |
| 600 | * We allocated new blocks which will result in |
| 601 | * i_data's format changing. Force the migrate |
| 602 | * to fail by clearing migrate flags |
| 603 | */ |
| 604 | ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE); |
| 605 | } |
| 606 | |
| 607 | /* |
| 608 | * Update reserved blocks/metadata blocks after successful |
| 609 | * block allocation which had been deferred till now. We don't |
| 610 | * support fallocate for non extent files. So we can update |
| 611 | * reserve space here. |
| 612 | */ |
| 613 | if ((retval > 0) && |
| 614 | (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)) |
| 615 | ext4_da_update_reserve_space(inode, retval, 1); |
| 616 | } |
| 617 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) { |
| 618 | ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED); |
| 619 | |
| 620 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
| 621 | int ret; |
| 622 | delayed_mapped: |
| 623 | /* delayed allocation blocks has been allocated */ |
| 624 | ret = ext4_es_remove_extent(inode, map->m_lblk, |
| 625 | map->m_len); |
| 626 | if (ret < 0) |
| 627 | retval = ret; |
| 628 | } |
| 629 | } |
| 630 | |
| 631 | up_write((&EXT4_I(inode)->i_data_sem)); |
| 632 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
| 633 | int ret = check_block_validity(inode, map); |
| 634 | if (ret != 0) |
| 635 | return ret; |
| 636 | } |
| 637 | return retval; |
| 638 | } |
| 639 | |
| 640 | /* Maximum number of blocks we map for direct IO at once. */ |
| 641 | #define DIO_MAX_BLOCKS 4096 |
| 642 | |
| 643 | static int _ext4_get_block(struct inode *inode, sector_t iblock, |
| 644 | struct buffer_head *bh, int flags) |
| 645 | { |
| 646 | handle_t *handle = ext4_journal_current_handle(); |
| 647 | struct ext4_map_blocks map; |
| 648 | int ret = 0, started = 0; |
| 649 | int dio_credits; |
| 650 | |
| 651 | if (ext4_has_inline_data(inode)) |
| 652 | return -ERANGE; |
| 653 | |
| 654 | map.m_lblk = iblock; |
| 655 | map.m_len = bh->b_size >> inode->i_blkbits; |
| 656 | |
| 657 | if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) { |
| 658 | /* Direct IO write... */ |
| 659 | if (map.m_len > DIO_MAX_BLOCKS) |
| 660 | map.m_len = DIO_MAX_BLOCKS; |
| 661 | dio_credits = ext4_chunk_trans_blocks(inode, map.m_len); |
| 662 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, |
| 663 | dio_credits); |
| 664 | if (IS_ERR(handle)) { |
| 665 | ret = PTR_ERR(handle); |
| 666 | return ret; |
| 667 | } |
| 668 | started = 1; |
| 669 | } |
| 670 | |
| 671 | ret = ext4_map_blocks(handle, inode, &map, flags); |
| 672 | if (ret > 0) { |
| 673 | map_bh(bh, inode->i_sb, map.m_pblk); |
| 674 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; |
| 675 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; |
| 676 | ret = 0; |
| 677 | } |
| 678 | if (started) |
| 679 | ext4_journal_stop(handle); |
| 680 | return ret; |
| 681 | } |
| 682 | |
| 683 | int ext4_get_block(struct inode *inode, sector_t iblock, |
| 684 | struct buffer_head *bh, int create) |
| 685 | { |
| 686 | return _ext4_get_block(inode, iblock, bh, |
| 687 | create ? EXT4_GET_BLOCKS_CREATE : 0); |
| 688 | } |
| 689 | |
| 690 | /* |
| 691 | * `handle' can be NULL if create is zero |
| 692 | */ |
| 693 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
| 694 | ext4_lblk_t block, int create, int *errp) |
| 695 | { |
| 696 | struct ext4_map_blocks map; |
| 697 | struct buffer_head *bh; |
| 698 | int fatal = 0, err; |
| 699 | |
| 700 | J_ASSERT(handle != NULL || create == 0); |
| 701 | |
| 702 | map.m_lblk = block; |
| 703 | map.m_len = 1; |
| 704 | err = ext4_map_blocks(handle, inode, &map, |
| 705 | create ? EXT4_GET_BLOCKS_CREATE : 0); |
| 706 | |
| 707 | /* ensure we send some value back into *errp */ |
| 708 | *errp = 0; |
| 709 | |
| 710 | if (create && err == 0) |
| 711 | err = -ENOSPC; /* should never happen */ |
| 712 | if (err < 0) |
| 713 | *errp = err; |
| 714 | if (err <= 0) |
| 715 | return NULL; |
| 716 | |
| 717 | bh = sb_getblk(inode->i_sb, map.m_pblk); |
| 718 | if (unlikely(!bh)) { |
| 719 | *errp = -ENOMEM; |
| 720 | return NULL; |
| 721 | } |
| 722 | if (map.m_flags & EXT4_MAP_NEW) { |
| 723 | J_ASSERT(create != 0); |
| 724 | J_ASSERT(handle != NULL); |
| 725 | |
| 726 | /* |
| 727 | * Now that we do not always journal data, we should |
| 728 | * keep in mind whether this should always journal the |
| 729 | * new buffer as metadata. For now, regular file |
| 730 | * writes use ext4_get_block instead, so it's not a |
| 731 | * problem. |
| 732 | */ |
| 733 | lock_buffer(bh); |
| 734 | BUFFER_TRACE(bh, "call get_create_access"); |
| 735 | fatal = ext4_journal_get_create_access(handle, bh); |
| 736 | if (!fatal && !buffer_uptodate(bh)) { |
| 737 | memset(bh->b_data, 0, inode->i_sb->s_blocksize); |
| 738 | set_buffer_uptodate(bh); |
| 739 | } |
| 740 | unlock_buffer(bh); |
| 741 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
| 742 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
| 743 | if (!fatal) |
| 744 | fatal = err; |
| 745 | } else { |
| 746 | BUFFER_TRACE(bh, "not a new buffer"); |
| 747 | } |
| 748 | if (fatal) { |
| 749 | *errp = fatal; |
| 750 | brelse(bh); |
| 751 | bh = NULL; |
| 752 | } |
| 753 | return bh; |
| 754 | } |
| 755 | |
| 756 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
| 757 | ext4_lblk_t block, int create, int *err) |
| 758 | { |
| 759 | struct buffer_head *bh; |
| 760 | |
| 761 | bh = ext4_getblk(handle, inode, block, create, err); |
| 762 | if (!bh) |
| 763 | return bh; |
| 764 | if (buffer_uptodate(bh)) |
| 765 | return bh; |
| 766 | ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh); |
| 767 | wait_on_buffer(bh); |
| 768 | if (buffer_uptodate(bh)) |
| 769 | return bh; |
| 770 | put_bh(bh); |
| 771 | *err = -EIO; |
| 772 | return NULL; |
| 773 | } |
| 774 | |
| 775 | int ext4_walk_page_buffers(handle_t *handle, |
| 776 | struct buffer_head *head, |
| 777 | unsigned from, |
| 778 | unsigned to, |
| 779 | int *partial, |
| 780 | int (*fn)(handle_t *handle, |
| 781 | struct buffer_head *bh)) |
| 782 | { |
| 783 | struct buffer_head *bh; |
| 784 | unsigned block_start, block_end; |
| 785 | unsigned blocksize = head->b_size; |
| 786 | int err, ret = 0; |
| 787 | struct buffer_head *next; |
| 788 | |
| 789 | for (bh = head, block_start = 0; |
| 790 | ret == 0 && (bh != head || !block_start); |
| 791 | block_start = block_end, bh = next) { |
| 792 | next = bh->b_this_page; |
| 793 | block_end = block_start + blocksize; |
| 794 | if (block_end <= from || block_start >= to) { |
| 795 | if (partial && !buffer_uptodate(bh)) |
| 796 | *partial = 1; |
| 797 | continue; |
| 798 | } |
| 799 | err = (*fn)(handle, bh); |
| 800 | if (!ret) |
| 801 | ret = err; |
| 802 | } |
| 803 | return ret; |
| 804 | } |
| 805 | |
| 806 | /* |
| 807 | * To preserve ordering, it is essential that the hole instantiation and |
| 808 | * the data write be encapsulated in a single transaction. We cannot |
| 809 | * close off a transaction and start a new one between the ext4_get_block() |
| 810 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
| 811 | * prepare_write() is the right place. |
| 812 | * |
| 813 | * Also, this function can nest inside ext4_writepage(). In that case, we |
| 814 | * *know* that ext4_writepage() has generated enough buffer credits to do the |
| 815 | * whole page. So we won't block on the journal in that case, which is good, |
| 816 | * because the caller may be PF_MEMALLOC. |
| 817 | * |
| 818 | * By accident, ext4 can be reentered when a transaction is open via |
| 819 | * quota file writes. If we were to commit the transaction while thus |
| 820 | * reentered, there can be a deadlock - we would be holding a quota |
| 821 | * lock, and the commit would never complete if another thread had a |
| 822 | * transaction open and was blocking on the quota lock - a ranking |
| 823 | * violation. |
| 824 | * |
| 825 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
| 826 | * will _not_ run commit under these circumstances because handle->h_ref |
| 827 | * is elevated. We'll still have enough credits for the tiny quotafile |
| 828 | * write. |
| 829 | */ |
| 830 | int do_journal_get_write_access(handle_t *handle, |
| 831 | struct buffer_head *bh) |
| 832 | { |
| 833 | int dirty = buffer_dirty(bh); |
| 834 | int ret; |
| 835 | |
| 836 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
| 837 | return 0; |
| 838 | /* |
| 839 | * __block_write_begin() could have dirtied some buffers. Clean |
| 840 | * the dirty bit as jbd2_journal_get_write_access() could complain |
| 841 | * otherwise about fs integrity issues. Setting of the dirty bit |
| 842 | * by __block_write_begin() isn't a real problem here as we clear |
| 843 | * the bit before releasing a page lock and thus writeback cannot |
| 844 | * ever write the buffer. |
| 845 | */ |
| 846 | if (dirty) |
| 847 | clear_buffer_dirty(bh); |
| 848 | ret = ext4_journal_get_write_access(handle, bh); |
| 849 | if (!ret && dirty) |
| 850 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); |
| 851 | return ret; |
| 852 | } |
| 853 | |
| 854 | static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock, |
| 855 | struct buffer_head *bh_result, int create); |
| 856 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
| 857 | loff_t pos, unsigned len, unsigned flags, |
| 858 | struct page **pagep, void **fsdata) |
| 859 | { |
| 860 | struct inode *inode = mapping->host; |
| 861 | int ret, needed_blocks; |
| 862 | handle_t *handle; |
| 863 | int retries = 0; |
| 864 | struct page *page; |
| 865 | pgoff_t index; |
| 866 | unsigned from, to; |
| 867 | |
| 868 | trace_ext4_write_begin(inode, pos, len, flags); |
| 869 | /* |
| 870 | * Reserve one block more for addition to orphan list in case |
| 871 | * we allocate blocks but write fails for some reason |
| 872 | */ |
| 873 | needed_blocks = ext4_writepage_trans_blocks(inode) + 1; |
| 874 | index = pos >> PAGE_CACHE_SHIFT; |
| 875 | from = pos & (PAGE_CACHE_SIZE - 1); |
| 876 | to = from + len; |
| 877 | |
| 878 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { |
| 879 | ret = ext4_try_to_write_inline_data(mapping, inode, pos, len, |
| 880 | flags, pagep); |
| 881 | if (ret < 0) |
| 882 | return ret; |
| 883 | if (ret == 1) |
| 884 | return 0; |
| 885 | } |
| 886 | |
| 887 | /* |
| 888 | * grab_cache_page_write_begin() can take a long time if the |
| 889 | * system is thrashing due to memory pressure, or if the page |
| 890 | * is being written back. So grab it first before we start |
| 891 | * the transaction handle. This also allows us to allocate |
| 892 | * the page (if needed) without using GFP_NOFS. |
| 893 | */ |
| 894 | retry_grab: |
| 895 | page = grab_cache_page_write_begin(mapping, index, flags); |
| 896 | if (!page) |
| 897 | return -ENOMEM; |
| 898 | unlock_page(page); |
| 899 | |
| 900 | retry_journal: |
| 901 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); |
| 902 | if (IS_ERR(handle)) { |
| 903 | page_cache_release(page); |
| 904 | return PTR_ERR(handle); |
| 905 | } |
| 906 | |
| 907 | lock_page(page); |
| 908 | if (page->mapping != mapping) { |
| 909 | /* The page got truncated from under us */ |
| 910 | unlock_page(page); |
| 911 | page_cache_release(page); |
| 912 | ext4_journal_stop(handle); |
| 913 | goto retry_grab; |
| 914 | } |
| 915 | wait_on_page_writeback(page); |
| 916 | |
| 917 | if (ext4_should_dioread_nolock(inode)) |
| 918 | ret = __block_write_begin(page, pos, len, ext4_get_block_write); |
| 919 | else |
| 920 | ret = __block_write_begin(page, pos, len, ext4_get_block); |
| 921 | |
| 922 | if (!ret && ext4_should_journal_data(inode)) { |
| 923 | ret = ext4_walk_page_buffers(handle, page_buffers(page), |
| 924 | from, to, NULL, |
| 925 | do_journal_get_write_access); |
| 926 | } |
| 927 | |
| 928 | if (ret) { |
| 929 | unlock_page(page); |
| 930 | /* |
| 931 | * __block_write_begin may have instantiated a few blocks |
| 932 | * outside i_size. Trim these off again. Don't need |
| 933 | * i_size_read because we hold i_mutex. |
| 934 | * |
| 935 | * Add inode to orphan list in case we crash before |
| 936 | * truncate finishes |
| 937 | */ |
| 938 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
| 939 | ext4_orphan_add(handle, inode); |
| 940 | |
| 941 | ext4_journal_stop(handle); |
| 942 | if (pos + len > inode->i_size) { |
| 943 | ext4_truncate_failed_write(inode); |
| 944 | /* |
| 945 | * If truncate failed early the inode might |
| 946 | * still be on the orphan list; we need to |
| 947 | * make sure the inode is removed from the |
| 948 | * orphan list in that case. |
| 949 | */ |
| 950 | if (inode->i_nlink) |
| 951 | ext4_orphan_del(NULL, inode); |
| 952 | } |
| 953 | |
| 954 | if (ret == -ENOSPC && |
| 955 | ext4_should_retry_alloc(inode->i_sb, &retries)) |
| 956 | goto retry_journal; |
| 957 | page_cache_release(page); |
| 958 | return ret; |
| 959 | } |
| 960 | *pagep = page; |
| 961 | return ret; |
| 962 | } |
| 963 | |
| 964 | /* For write_end() in data=journal mode */ |
| 965 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) |
| 966 | { |
| 967 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
| 968 | return 0; |
| 969 | set_buffer_uptodate(bh); |
| 970 | return ext4_handle_dirty_metadata(handle, NULL, bh); |
| 971 | } |
| 972 | |
| 973 | static int ext4_generic_write_end(struct file *file, |
| 974 | struct address_space *mapping, |
| 975 | loff_t pos, unsigned len, unsigned copied, |
| 976 | struct page *page, void *fsdata) |
| 977 | { |
| 978 | int i_size_changed = 0; |
| 979 | struct inode *inode = mapping->host; |
| 980 | handle_t *handle = ext4_journal_current_handle(); |
| 981 | |
| 982 | if (ext4_has_inline_data(inode)) |
| 983 | copied = ext4_write_inline_data_end(inode, pos, len, |
| 984 | copied, page); |
| 985 | else |
| 986 | copied = block_write_end(file, mapping, pos, |
| 987 | len, copied, page, fsdata); |
| 988 | |
| 989 | /* |
| 990 | * No need to use i_size_read() here, the i_size |
| 991 | * cannot change under us because we hold i_mutex. |
| 992 | * |
| 993 | * But it's important to update i_size while still holding page lock: |
| 994 | * page writeout could otherwise come in and zero beyond i_size. |
| 995 | */ |
| 996 | if (pos + copied > inode->i_size) { |
| 997 | i_size_write(inode, pos + copied); |
| 998 | i_size_changed = 1; |
| 999 | } |
| 1000 | |
| 1001 | if (pos + copied > EXT4_I(inode)->i_disksize) { |
| 1002 | /* We need to mark inode dirty even if |
| 1003 | * new_i_size is less that inode->i_size |
| 1004 | * bu greater than i_disksize.(hint delalloc) |
| 1005 | */ |
| 1006 | ext4_update_i_disksize(inode, (pos + copied)); |
| 1007 | i_size_changed = 1; |
| 1008 | } |
| 1009 | unlock_page(page); |
| 1010 | page_cache_release(page); |
| 1011 | |
| 1012 | /* |
| 1013 | * Don't mark the inode dirty under page lock. First, it unnecessarily |
| 1014 | * makes the holding time of page lock longer. Second, it forces lock |
| 1015 | * ordering of page lock and transaction start for journaling |
| 1016 | * filesystems. |
| 1017 | */ |
| 1018 | if (i_size_changed) |
| 1019 | ext4_mark_inode_dirty(handle, inode); |
| 1020 | |
| 1021 | return copied; |
| 1022 | } |
| 1023 | |
| 1024 | /* |
| 1025 | * We need to pick up the new inode size which generic_commit_write gave us |
| 1026 | * `file' can be NULL - eg, when called from page_symlink(). |
| 1027 | * |
| 1028 | * ext4 never places buffers on inode->i_mapping->private_list. metadata |
| 1029 | * buffers are managed internally. |
| 1030 | */ |
| 1031 | static int ext4_ordered_write_end(struct file *file, |
| 1032 | struct address_space *mapping, |
| 1033 | loff_t pos, unsigned len, unsigned copied, |
| 1034 | struct page *page, void *fsdata) |
| 1035 | { |
| 1036 | handle_t *handle = ext4_journal_current_handle(); |
| 1037 | struct inode *inode = mapping->host; |
| 1038 | int ret = 0, ret2; |
| 1039 | |
| 1040 | trace_ext4_ordered_write_end(inode, pos, len, copied); |
| 1041 | ret = ext4_jbd2_file_inode(handle, inode); |
| 1042 | |
| 1043 | if (ret == 0) { |
| 1044 | ret2 = ext4_generic_write_end(file, mapping, pos, len, copied, |
| 1045 | page, fsdata); |
| 1046 | copied = ret2; |
| 1047 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
| 1048 | /* if we have allocated more blocks and copied |
| 1049 | * less. We will have blocks allocated outside |
| 1050 | * inode->i_size. So truncate them |
| 1051 | */ |
| 1052 | ext4_orphan_add(handle, inode); |
| 1053 | if (ret2 < 0) |
| 1054 | ret = ret2; |
| 1055 | } else { |
| 1056 | unlock_page(page); |
| 1057 | page_cache_release(page); |
| 1058 | } |
| 1059 | |
| 1060 | ret2 = ext4_journal_stop(handle); |
| 1061 | if (!ret) |
| 1062 | ret = ret2; |
| 1063 | |
| 1064 | if (pos + len > inode->i_size) { |
| 1065 | ext4_truncate_failed_write(inode); |
| 1066 | /* |
| 1067 | * If truncate failed early the inode might still be |
| 1068 | * on the orphan list; we need to make sure the inode |
| 1069 | * is removed from the orphan list in that case. |
| 1070 | */ |
| 1071 | if (inode->i_nlink) |
| 1072 | ext4_orphan_del(NULL, inode); |
| 1073 | } |
| 1074 | |
| 1075 | |
| 1076 | return ret ? ret : copied; |
| 1077 | } |
| 1078 | |
| 1079 | static int ext4_writeback_write_end(struct file *file, |
| 1080 | struct address_space *mapping, |
| 1081 | loff_t pos, unsigned len, unsigned copied, |
| 1082 | struct page *page, void *fsdata) |
| 1083 | { |
| 1084 | handle_t *handle = ext4_journal_current_handle(); |
| 1085 | struct inode *inode = mapping->host; |
| 1086 | int ret = 0, ret2; |
| 1087 | |
| 1088 | trace_ext4_writeback_write_end(inode, pos, len, copied); |
| 1089 | ret2 = ext4_generic_write_end(file, mapping, pos, len, copied, |
| 1090 | page, fsdata); |
| 1091 | copied = ret2; |
| 1092 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
| 1093 | /* if we have allocated more blocks and copied |
| 1094 | * less. We will have blocks allocated outside |
| 1095 | * inode->i_size. So truncate them |
| 1096 | */ |
| 1097 | ext4_orphan_add(handle, inode); |
| 1098 | |
| 1099 | if (ret2 < 0) |
| 1100 | ret = ret2; |
| 1101 | |
| 1102 | ret2 = ext4_journal_stop(handle); |
| 1103 | if (!ret) |
| 1104 | ret = ret2; |
| 1105 | |
| 1106 | if (pos + len > inode->i_size) { |
| 1107 | ext4_truncate_failed_write(inode); |
| 1108 | /* |
| 1109 | * If truncate failed early the inode might still be |
| 1110 | * on the orphan list; we need to make sure the inode |
| 1111 | * is removed from the orphan list in that case. |
| 1112 | */ |
| 1113 | if (inode->i_nlink) |
| 1114 | ext4_orphan_del(NULL, inode); |
| 1115 | } |
| 1116 | |
| 1117 | return ret ? ret : copied; |
| 1118 | } |
| 1119 | |
| 1120 | static int ext4_journalled_write_end(struct file *file, |
| 1121 | struct address_space *mapping, |
| 1122 | loff_t pos, unsigned len, unsigned copied, |
| 1123 | struct page *page, void *fsdata) |
| 1124 | { |
| 1125 | handle_t *handle = ext4_journal_current_handle(); |
| 1126 | struct inode *inode = mapping->host; |
| 1127 | int ret = 0, ret2; |
| 1128 | int partial = 0; |
| 1129 | unsigned from, to; |
| 1130 | loff_t new_i_size; |
| 1131 | |
| 1132 | trace_ext4_journalled_write_end(inode, pos, len, copied); |
| 1133 | from = pos & (PAGE_CACHE_SIZE - 1); |
| 1134 | to = from + len; |
| 1135 | |
| 1136 | BUG_ON(!ext4_handle_valid(handle)); |
| 1137 | |
| 1138 | if (ext4_has_inline_data(inode)) |
| 1139 | copied = ext4_write_inline_data_end(inode, pos, len, |
| 1140 | copied, page); |
| 1141 | else { |
| 1142 | if (copied < len) { |
| 1143 | if (!PageUptodate(page)) |
| 1144 | copied = 0; |
| 1145 | page_zero_new_buffers(page, from+copied, to); |
| 1146 | } |
| 1147 | |
| 1148 | ret = ext4_walk_page_buffers(handle, page_buffers(page), from, |
| 1149 | to, &partial, write_end_fn); |
| 1150 | if (!partial) |
| 1151 | SetPageUptodate(page); |
| 1152 | } |
| 1153 | new_i_size = pos + copied; |
| 1154 | if (new_i_size > inode->i_size) |
| 1155 | i_size_write(inode, pos+copied); |
| 1156 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
| 1157 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
| 1158 | if (new_i_size > EXT4_I(inode)->i_disksize) { |
| 1159 | ext4_update_i_disksize(inode, new_i_size); |
| 1160 | ret2 = ext4_mark_inode_dirty(handle, inode); |
| 1161 | if (!ret) |
| 1162 | ret = ret2; |
| 1163 | } |
| 1164 | |
| 1165 | unlock_page(page); |
| 1166 | page_cache_release(page); |
| 1167 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
| 1168 | /* if we have allocated more blocks and copied |
| 1169 | * less. We will have blocks allocated outside |
| 1170 | * inode->i_size. So truncate them |
| 1171 | */ |
| 1172 | ext4_orphan_add(handle, inode); |
| 1173 | |
| 1174 | ret2 = ext4_journal_stop(handle); |
| 1175 | if (!ret) |
| 1176 | ret = ret2; |
| 1177 | if (pos + len > inode->i_size) { |
| 1178 | ext4_truncate_failed_write(inode); |
| 1179 | /* |
| 1180 | * If truncate failed early the inode might still be |
| 1181 | * on the orphan list; we need to make sure the inode |
| 1182 | * is removed from the orphan list in that case. |
| 1183 | */ |
| 1184 | if (inode->i_nlink) |
| 1185 | ext4_orphan_del(NULL, inode); |
| 1186 | } |
| 1187 | |
| 1188 | return ret ? ret : copied; |
| 1189 | } |
| 1190 | |
| 1191 | /* |
| 1192 | * Reserve a single cluster located at lblock |
| 1193 | */ |
| 1194 | static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock) |
| 1195 | { |
| 1196 | int retries = 0; |
| 1197 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 1198 | struct ext4_inode_info *ei = EXT4_I(inode); |
| 1199 | unsigned int md_needed; |
| 1200 | int ret; |
| 1201 | ext4_lblk_t save_last_lblock; |
| 1202 | int save_len; |
| 1203 | |
| 1204 | /* |
| 1205 | * We will charge metadata quota at writeout time; this saves |
| 1206 | * us from metadata over-estimation, though we may go over by |
| 1207 | * a small amount in the end. Here we just reserve for data. |
| 1208 | */ |
| 1209 | ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1)); |
| 1210 | if (ret) |
| 1211 | return ret; |
| 1212 | |
| 1213 | /* |
| 1214 | * recalculate the amount of metadata blocks to reserve |
| 1215 | * in order to allocate nrblocks |
| 1216 | * worse case is one extent per block |
| 1217 | */ |
| 1218 | repeat: |
| 1219 | spin_lock(&ei->i_block_reservation_lock); |
| 1220 | /* |
| 1221 | * ext4_calc_metadata_amount() has side effects, which we have |
| 1222 | * to be prepared undo if we fail to claim space. |
| 1223 | */ |
| 1224 | save_len = ei->i_da_metadata_calc_len; |
| 1225 | save_last_lblock = ei->i_da_metadata_calc_last_lblock; |
| 1226 | md_needed = EXT4_NUM_B2C(sbi, |
| 1227 | ext4_calc_metadata_amount(inode, lblock)); |
| 1228 | trace_ext4_da_reserve_space(inode, md_needed); |
| 1229 | |
| 1230 | /* |
| 1231 | * We do still charge estimated metadata to the sb though; |
| 1232 | * we cannot afford to run out of free blocks. |
| 1233 | */ |
| 1234 | if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) { |
| 1235 | ei->i_da_metadata_calc_len = save_len; |
| 1236 | ei->i_da_metadata_calc_last_lblock = save_last_lblock; |
| 1237 | spin_unlock(&ei->i_block_reservation_lock); |
| 1238 | if (ext4_should_retry_alloc(inode->i_sb, &retries)) { |
| 1239 | yield(); |
| 1240 | goto repeat; |
| 1241 | } |
| 1242 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1)); |
| 1243 | return -ENOSPC; |
| 1244 | } |
| 1245 | ei->i_reserved_data_blocks++; |
| 1246 | ei->i_reserved_meta_blocks += md_needed; |
| 1247 | spin_unlock(&ei->i_block_reservation_lock); |
| 1248 | |
| 1249 | return 0; /* success */ |
| 1250 | } |
| 1251 | |
| 1252 | static void ext4_da_release_space(struct inode *inode, int to_free) |
| 1253 | { |
| 1254 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 1255 | struct ext4_inode_info *ei = EXT4_I(inode); |
| 1256 | |
| 1257 | if (!to_free) |
| 1258 | return; /* Nothing to release, exit */ |
| 1259 | |
| 1260 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
| 1261 | |
| 1262 | trace_ext4_da_release_space(inode, to_free); |
| 1263 | if (unlikely(to_free > ei->i_reserved_data_blocks)) { |
| 1264 | /* |
| 1265 | * if there aren't enough reserved blocks, then the |
| 1266 | * counter is messed up somewhere. Since this |
| 1267 | * function is called from invalidate page, it's |
| 1268 | * harmless to return without any action. |
| 1269 | */ |
| 1270 | ext4_warning(inode->i_sb, "ext4_da_release_space: " |
| 1271 | "ino %lu, to_free %d with only %d reserved " |
| 1272 | "data blocks", inode->i_ino, to_free, |
| 1273 | ei->i_reserved_data_blocks); |
| 1274 | WARN_ON(1); |
| 1275 | to_free = ei->i_reserved_data_blocks; |
| 1276 | } |
| 1277 | ei->i_reserved_data_blocks -= to_free; |
| 1278 | |
| 1279 | if (ei->i_reserved_data_blocks == 0) { |
| 1280 | /* |
| 1281 | * We can release all of the reserved metadata blocks |
| 1282 | * only when we have written all of the delayed |
| 1283 | * allocation blocks. |
| 1284 | * Note that in case of bigalloc, i_reserved_meta_blocks, |
| 1285 | * i_reserved_data_blocks, etc. refer to number of clusters. |
| 1286 | */ |
| 1287 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, |
| 1288 | ei->i_reserved_meta_blocks); |
| 1289 | ei->i_reserved_meta_blocks = 0; |
| 1290 | ei->i_da_metadata_calc_len = 0; |
| 1291 | } |
| 1292 | |
| 1293 | /* update fs dirty data blocks counter */ |
| 1294 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free); |
| 1295 | |
| 1296 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
| 1297 | |
| 1298 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free)); |
| 1299 | } |
| 1300 | |
| 1301 | static void ext4_da_page_release_reservation(struct page *page, |
| 1302 | unsigned long offset) |
| 1303 | { |
| 1304 | int to_release = 0; |
| 1305 | struct buffer_head *head, *bh; |
| 1306 | unsigned int curr_off = 0; |
| 1307 | struct inode *inode = page->mapping->host; |
| 1308 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 1309 | int num_clusters; |
| 1310 | ext4_fsblk_t lblk; |
| 1311 | |
| 1312 | head = page_buffers(page); |
| 1313 | bh = head; |
| 1314 | do { |
| 1315 | unsigned int next_off = curr_off + bh->b_size; |
| 1316 | |
| 1317 | if ((offset <= curr_off) && (buffer_delay(bh))) { |
| 1318 | to_release++; |
| 1319 | clear_buffer_delay(bh); |
| 1320 | } |
| 1321 | curr_off = next_off; |
| 1322 | } while ((bh = bh->b_this_page) != head); |
| 1323 | |
| 1324 | if (to_release) { |
| 1325 | lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits); |
| 1326 | ext4_es_remove_extent(inode, lblk, to_release); |
| 1327 | } |
| 1328 | |
| 1329 | /* If we have released all the blocks belonging to a cluster, then we |
| 1330 | * need to release the reserved space for that cluster. */ |
| 1331 | num_clusters = EXT4_NUM_B2C(sbi, to_release); |
| 1332 | while (num_clusters > 0) { |
| 1333 | lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) + |
| 1334 | ((num_clusters - 1) << sbi->s_cluster_bits); |
| 1335 | if (sbi->s_cluster_ratio == 1 || |
| 1336 | !ext4_find_delalloc_cluster(inode, lblk)) |
| 1337 | ext4_da_release_space(inode, 1); |
| 1338 | |
| 1339 | num_clusters--; |
| 1340 | } |
| 1341 | } |
| 1342 | |
| 1343 | /* |
| 1344 | * Delayed allocation stuff |
| 1345 | */ |
| 1346 | |
| 1347 | /* |
| 1348 | * mpage_da_submit_io - walks through extent of pages and try to write |
| 1349 | * them with writepage() call back |
| 1350 | * |
| 1351 | * @mpd->inode: inode |
| 1352 | * @mpd->first_page: first page of the extent |
| 1353 | * @mpd->next_page: page after the last page of the extent |
| 1354 | * |
| 1355 | * By the time mpage_da_submit_io() is called we expect all blocks |
| 1356 | * to be allocated. this may be wrong if allocation failed. |
| 1357 | * |
| 1358 | * As pages are already locked by write_cache_pages(), we can't use it |
| 1359 | */ |
| 1360 | static int mpage_da_submit_io(struct mpage_da_data *mpd, |
| 1361 | struct ext4_map_blocks *map) |
| 1362 | { |
| 1363 | struct pagevec pvec; |
| 1364 | unsigned long index, end; |
| 1365 | int ret = 0, err, nr_pages, i; |
| 1366 | struct inode *inode = mpd->inode; |
| 1367 | struct address_space *mapping = inode->i_mapping; |
| 1368 | loff_t size = i_size_read(inode); |
| 1369 | unsigned int len, block_start; |
| 1370 | struct buffer_head *bh, *page_bufs = NULL; |
| 1371 | sector_t pblock = 0, cur_logical = 0; |
| 1372 | struct ext4_io_submit io_submit; |
| 1373 | |
| 1374 | BUG_ON(mpd->next_page <= mpd->first_page); |
| 1375 | memset(&io_submit, 0, sizeof(io_submit)); |
| 1376 | /* |
| 1377 | * We need to start from the first_page to the next_page - 1 |
| 1378 | * to make sure we also write the mapped dirty buffer_heads. |
| 1379 | * If we look at mpd->b_blocknr we would only be looking |
| 1380 | * at the currently mapped buffer_heads. |
| 1381 | */ |
| 1382 | index = mpd->first_page; |
| 1383 | end = mpd->next_page - 1; |
| 1384 | |
| 1385 | pagevec_init(&pvec, 0); |
| 1386 | while (index <= end) { |
| 1387 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); |
| 1388 | if (nr_pages == 0) |
| 1389 | break; |
| 1390 | for (i = 0; i < nr_pages; i++) { |
| 1391 | int skip_page = 0; |
| 1392 | struct page *page = pvec.pages[i]; |
| 1393 | |
| 1394 | index = page->index; |
| 1395 | if (index > end) |
| 1396 | break; |
| 1397 | |
| 1398 | if (index == size >> PAGE_CACHE_SHIFT) |
| 1399 | len = size & ~PAGE_CACHE_MASK; |
| 1400 | else |
| 1401 | len = PAGE_CACHE_SIZE; |
| 1402 | if (map) { |
| 1403 | cur_logical = index << (PAGE_CACHE_SHIFT - |
| 1404 | inode->i_blkbits); |
| 1405 | pblock = map->m_pblk + (cur_logical - |
| 1406 | map->m_lblk); |
| 1407 | } |
| 1408 | index++; |
| 1409 | |
| 1410 | BUG_ON(!PageLocked(page)); |
| 1411 | BUG_ON(PageWriteback(page)); |
| 1412 | |
| 1413 | bh = page_bufs = page_buffers(page); |
| 1414 | block_start = 0; |
| 1415 | do { |
| 1416 | if (map && (cur_logical >= map->m_lblk) && |
| 1417 | (cur_logical <= (map->m_lblk + |
| 1418 | (map->m_len - 1)))) { |
| 1419 | if (buffer_delay(bh)) { |
| 1420 | clear_buffer_delay(bh); |
| 1421 | bh->b_blocknr = pblock; |
| 1422 | } |
| 1423 | if (buffer_unwritten(bh) || |
| 1424 | buffer_mapped(bh)) |
| 1425 | BUG_ON(bh->b_blocknr != pblock); |
| 1426 | if (map->m_flags & EXT4_MAP_UNINIT) |
| 1427 | set_buffer_uninit(bh); |
| 1428 | clear_buffer_unwritten(bh); |
| 1429 | } |
| 1430 | |
| 1431 | /* |
| 1432 | * skip page if block allocation undone and |
| 1433 | * block is dirty |
| 1434 | */ |
| 1435 | if (ext4_bh_delay_or_unwritten(NULL, bh)) |
| 1436 | skip_page = 1; |
| 1437 | bh = bh->b_this_page; |
| 1438 | block_start += bh->b_size; |
| 1439 | cur_logical++; |
| 1440 | pblock++; |
| 1441 | } while (bh != page_bufs); |
| 1442 | |
| 1443 | if (skip_page) { |
| 1444 | unlock_page(page); |
| 1445 | continue; |
| 1446 | } |
| 1447 | |
| 1448 | clear_page_dirty_for_io(page); |
| 1449 | err = ext4_bio_write_page(&io_submit, page, len, |
| 1450 | mpd->wbc); |
| 1451 | if (!err) |
| 1452 | mpd->pages_written++; |
| 1453 | /* |
| 1454 | * In error case, we have to continue because |
| 1455 | * remaining pages are still locked |
| 1456 | */ |
| 1457 | if (ret == 0) |
| 1458 | ret = err; |
| 1459 | } |
| 1460 | pagevec_release(&pvec); |
| 1461 | } |
| 1462 | ext4_io_submit(&io_submit); |
| 1463 | return ret; |
| 1464 | } |
| 1465 | |
| 1466 | static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd) |
| 1467 | { |
| 1468 | int nr_pages, i; |
| 1469 | pgoff_t index, end; |
| 1470 | struct pagevec pvec; |
| 1471 | struct inode *inode = mpd->inode; |
| 1472 | struct address_space *mapping = inode->i_mapping; |
| 1473 | ext4_lblk_t start, last; |
| 1474 | |
| 1475 | index = mpd->first_page; |
| 1476 | end = mpd->next_page - 1; |
| 1477 | |
| 1478 | start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits); |
| 1479 | last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits); |
| 1480 | ext4_es_remove_extent(inode, start, last - start + 1); |
| 1481 | |
| 1482 | pagevec_init(&pvec, 0); |
| 1483 | while (index <= end) { |
| 1484 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); |
| 1485 | if (nr_pages == 0) |
| 1486 | break; |
| 1487 | for (i = 0; i < nr_pages; i++) { |
| 1488 | struct page *page = pvec.pages[i]; |
| 1489 | if (page->index > end) |
| 1490 | break; |
| 1491 | BUG_ON(!PageLocked(page)); |
| 1492 | BUG_ON(PageWriteback(page)); |
| 1493 | block_invalidatepage(page, 0); |
| 1494 | ClearPageUptodate(page); |
| 1495 | unlock_page(page); |
| 1496 | } |
| 1497 | index = pvec.pages[nr_pages - 1]->index + 1; |
| 1498 | pagevec_release(&pvec); |
| 1499 | } |
| 1500 | return; |
| 1501 | } |
| 1502 | |
| 1503 | static void ext4_print_free_blocks(struct inode *inode) |
| 1504 | { |
| 1505 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 1506 | struct super_block *sb = inode->i_sb; |
| 1507 | |
| 1508 | ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld", |
| 1509 | EXT4_C2B(EXT4_SB(inode->i_sb), |
| 1510 | ext4_count_free_clusters(inode->i_sb))); |
| 1511 | ext4_msg(sb, KERN_CRIT, "Free/Dirty block details"); |
| 1512 | ext4_msg(sb, KERN_CRIT, "free_blocks=%lld", |
| 1513 | (long long) EXT4_C2B(EXT4_SB(inode->i_sb), |
| 1514 | percpu_counter_sum(&sbi->s_freeclusters_counter))); |
| 1515 | ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld", |
| 1516 | (long long) EXT4_C2B(EXT4_SB(inode->i_sb), |
| 1517 | percpu_counter_sum(&sbi->s_dirtyclusters_counter))); |
| 1518 | ext4_msg(sb, KERN_CRIT, "Block reservation details"); |
| 1519 | ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u", |
| 1520 | EXT4_I(inode)->i_reserved_data_blocks); |
| 1521 | ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u", |
| 1522 | EXT4_I(inode)->i_reserved_meta_blocks); |
| 1523 | return; |
| 1524 | } |
| 1525 | |
| 1526 | /* |
| 1527 | * mpage_da_map_and_submit - go through given space, map them |
| 1528 | * if necessary, and then submit them for I/O |
| 1529 | * |
| 1530 | * @mpd - bh describing space |
| 1531 | * |
| 1532 | * The function skips space we know is already mapped to disk blocks. |
| 1533 | * |
| 1534 | */ |
| 1535 | static void mpage_da_map_and_submit(struct mpage_da_data *mpd) |
| 1536 | { |
| 1537 | int err, blks, get_blocks_flags; |
| 1538 | struct ext4_map_blocks map, *mapp = NULL; |
| 1539 | sector_t next = mpd->b_blocknr; |
| 1540 | unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits; |
| 1541 | loff_t disksize = EXT4_I(mpd->inode)->i_disksize; |
| 1542 | handle_t *handle = NULL; |
| 1543 | |
| 1544 | /* |
| 1545 | * If the blocks are mapped already, or we couldn't accumulate |
| 1546 | * any blocks, then proceed immediately to the submission stage. |
| 1547 | */ |
| 1548 | if ((mpd->b_size == 0) || |
| 1549 | ((mpd->b_state & (1 << BH_Mapped)) && |
| 1550 | !(mpd->b_state & (1 << BH_Delay)) && |
| 1551 | !(mpd->b_state & (1 << BH_Unwritten)))) |
| 1552 | goto submit_io; |
| 1553 | |
| 1554 | handle = ext4_journal_current_handle(); |
| 1555 | BUG_ON(!handle); |
| 1556 | |
| 1557 | /* |
| 1558 | * Call ext4_map_blocks() to allocate any delayed allocation |
| 1559 | * blocks, or to convert an uninitialized extent to be |
| 1560 | * initialized (in the case where we have written into |
| 1561 | * one or more preallocated blocks). |
| 1562 | * |
| 1563 | * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to |
| 1564 | * indicate that we are on the delayed allocation path. This |
| 1565 | * affects functions in many different parts of the allocation |
| 1566 | * call path. This flag exists primarily because we don't |
| 1567 | * want to change *many* call functions, so ext4_map_blocks() |
| 1568 | * will set the EXT4_STATE_DELALLOC_RESERVED flag once the |
| 1569 | * inode's allocation semaphore is taken. |
| 1570 | * |
| 1571 | * If the blocks in questions were delalloc blocks, set |
| 1572 | * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting |
| 1573 | * variables are updated after the blocks have been allocated. |
| 1574 | */ |
| 1575 | map.m_lblk = next; |
| 1576 | map.m_len = max_blocks; |
| 1577 | get_blocks_flags = EXT4_GET_BLOCKS_CREATE; |
| 1578 | if (ext4_should_dioread_nolock(mpd->inode)) |
| 1579 | get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; |
| 1580 | if (mpd->b_state & (1 << BH_Delay)) |
| 1581 | get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; |
| 1582 | |
| 1583 | blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags); |
| 1584 | if (blks < 0) { |
| 1585 | struct super_block *sb = mpd->inode->i_sb; |
| 1586 | |
| 1587 | err = blks; |
| 1588 | /* |
| 1589 | * If get block returns EAGAIN or ENOSPC and there |
| 1590 | * appears to be free blocks we will just let |
| 1591 | * mpage_da_submit_io() unlock all of the pages. |
| 1592 | */ |
| 1593 | if (err == -EAGAIN) |
| 1594 | goto submit_io; |
| 1595 | |
| 1596 | if (err == -ENOSPC && ext4_count_free_clusters(sb)) { |
| 1597 | mpd->retval = err; |
| 1598 | goto submit_io; |
| 1599 | } |
| 1600 | |
| 1601 | /* |
| 1602 | * get block failure will cause us to loop in |
| 1603 | * writepages, because a_ops->writepage won't be able |
| 1604 | * to make progress. The page will be redirtied by |
| 1605 | * writepage and writepages will again try to write |
| 1606 | * the same. |
| 1607 | */ |
| 1608 | if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) { |
| 1609 | ext4_msg(sb, KERN_CRIT, |
| 1610 | "delayed block allocation failed for inode %lu " |
| 1611 | "at logical offset %llu with max blocks %zd " |
| 1612 | "with error %d", mpd->inode->i_ino, |
| 1613 | (unsigned long long) next, |
| 1614 | mpd->b_size >> mpd->inode->i_blkbits, err); |
| 1615 | ext4_msg(sb, KERN_CRIT, |
| 1616 | "This should not happen!! Data will be lost"); |
| 1617 | if (err == -ENOSPC) |
| 1618 | ext4_print_free_blocks(mpd->inode); |
| 1619 | } |
| 1620 | /* invalidate all the pages */ |
| 1621 | ext4_da_block_invalidatepages(mpd); |
| 1622 | |
| 1623 | /* Mark this page range as having been completed */ |
| 1624 | mpd->io_done = 1; |
| 1625 | return; |
| 1626 | } |
| 1627 | BUG_ON(blks == 0); |
| 1628 | |
| 1629 | mapp = ↦ |
| 1630 | if (map.m_flags & EXT4_MAP_NEW) { |
| 1631 | struct block_device *bdev = mpd->inode->i_sb->s_bdev; |
| 1632 | int i; |
| 1633 | |
| 1634 | for (i = 0; i < map.m_len; i++) |
| 1635 | unmap_underlying_metadata(bdev, map.m_pblk + i); |
| 1636 | } |
| 1637 | |
| 1638 | /* |
| 1639 | * Update on-disk size along with block allocation. |
| 1640 | */ |
| 1641 | disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits; |
| 1642 | if (disksize > i_size_read(mpd->inode)) |
| 1643 | disksize = i_size_read(mpd->inode); |
| 1644 | if (disksize > EXT4_I(mpd->inode)->i_disksize) { |
| 1645 | ext4_update_i_disksize(mpd->inode, disksize); |
| 1646 | err = ext4_mark_inode_dirty(handle, mpd->inode); |
| 1647 | if (err) |
| 1648 | ext4_error(mpd->inode->i_sb, |
| 1649 | "Failed to mark inode %lu dirty", |
| 1650 | mpd->inode->i_ino); |
| 1651 | } |
| 1652 | |
| 1653 | submit_io: |
| 1654 | mpage_da_submit_io(mpd, mapp); |
| 1655 | mpd->io_done = 1; |
| 1656 | } |
| 1657 | |
| 1658 | #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \ |
| 1659 | (1 << BH_Delay) | (1 << BH_Unwritten)) |
| 1660 | |
| 1661 | /* |
| 1662 | * mpage_add_bh_to_extent - try to add one more block to extent of blocks |
| 1663 | * |
| 1664 | * @mpd->lbh - extent of blocks |
| 1665 | * @logical - logical number of the block in the file |
| 1666 | * @b_state - b_state of the buffer head added |
| 1667 | * |
| 1668 | * the function is used to collect contig. blocks in same state |
| 1669 | */ |
| 1670 | static void mpage_add_bh_to_extent(struct mpage_da_data *mpd, sector_t logical, |
| 1671 | unsigned long b_state) |
| 1672 | { |
| 1673 | sector_t next; |
| 1674 | int blkbits = mpd->inode->i_blkbits; |
| 1675 | int nrblocks = mpd->b_size >> blkbits; |
| 1676 | |
| 1677 | /* |
| 1678 | * XXX Don't go larger than mballoc is willing to allocate |
| 1679 | * This is a stopgap solution. We eventually need to fold |
| 1680 | * mpage_da_submit_io() into this function and then call |
| 1681 | * ext4_map_blocks() multiple times in a loop |
| 1682 | */ |
| 1683 | if (nrblocks >= (8*1024*1024 >> blkbits)) |
| 1684 | goto flush_it; |
| 1685 | |
| 1686 | /* check if the reserved journal credits might overflow */ |
| 1687 | if (!ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS)) { |
| 1688 | if (nrblocks >= EXT4_MAX_TRANS_DATA) { |
| 1689 | /* |
| 1690 | * With non-extent format we are limited by the journal |
| 1691 | * credit available. Total credit needed to insert |
| 1692 | * nrblocks contiguous blocks is dependent on the |
| 1693 | * nrblocks. So limit nrblocks. |
| 1694 | */ |
| 1695 | goto flush_it; |
| 1696 | } |
| 1697 | } |
| 1698 | /* |
| 1699 | * First block in the extent |
| 1700 | */ |
| 1701 | if (mpd->b_size == 0) { |
| 1702 | mpd->b_blocknr = logical; |
| 1703 | mpd->b_size = 1 << blkbits; |
| 1704 | mpd->b_state = b_state & BH_FLAGS; |
| 1705 | return; |
| 1706 | } |
| 1707 | |
| 1708 | next = mpd->b_blocknr + nrblocks; |
| 1709 | /* |
| 1710 | * Can we merge the block to our big extent? |
| 1711 | */ |
| 1712 | if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) { |
| 1713 | mpd->b_size += 1 << blkbits; |
| 1714 | return; |
| 1715 | } |
| 1716 | |
| 1717 | flush_it: |
| 1718 | /* |
| 1719 | * We couldn't merge the block to our extent, so we |
| 1720 | * need to flush current extent and start new one |
| 1721 | */ |
| 1722 | mpage_da_map_and_submit(mpd); |
| 1723 | return; |
| 1724 | } |
| 1725 | |
| 1726 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh) |
| 1727 | { |
| 1728 | return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh); |
| 1729 | } |
| 1730 | |
| 1731 | /* |
| 1732 | * This function is grabs code from the very beginning of |
| 1733 | * ext4_map_blocks, but assumes that the caller is from delayed write |
| 1734 | * time. This function looks up the requested blocks and sets the |
| 1735 | * buffer delay bit under the protection of i_data_sem. |
| 1736 | */ |
| 1737 | static int ext4_da_map_blocks(struct inode *inode, sector_t iblock, |
| 1738 | struct ext4_map_blocks *map, |
| 1739 | struct buffer_head *bh) |
| 1740 | { |
| 1741 | int retval; |
| 1742 | sector_t invalid_block = ~((sector_t) 0xffff); |
| 1743 | |
| 1744 | if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) |
| 1745 | invalid_block = ~0; |
| 1746 | |
| 1747 | map->m_flags = 0; |
| 1748 | ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u," |
| 1749 | "logical block %lu\n", inode->i_ino, map->m_len, |
| 1750 | (unsigned long) map->m_lblk); |
| 1751 | /* |
| 1752 | * Try to see if we can get the block without requesting a new |
| 1753 | * file system block. |
| 1754 | */ |
| 1755 | down_read((&EXT4_I(inode)->i_data_sem)); |
| 1756 | if (ext4_has_inline_data(inode)) { |
| 1757 | /* |
| 1758 | * We will soon create blocks for this page, and let |
| 1759 | * us pretend as if the blocks aren't allocated yet. |
| 1760 | * In case of clusters, we have to handle the work |
| 1761 | * of mapping from cluster so that the reserved space |
| 1762 | * is calculated properly. |
| 1763 | */ |
| 1764 | if ((EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) && |
| 1765 | ext4_find_delalloc_cluster(inode, map->m_lblk)) |
| 1766 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; |
| 1767 | retval = 0; |
| 1768 | } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
| 1769 | retval = ext4_ext_map_blocks(NULL, inode, map, 0); |
| 1770 | else |
| 1771 | retval = ext4_ind_map_blocks(NULL, inode, map, 0); |
| 1772 | |
| 1773 | if (retval == 0) { |
| 1774 | /* |
| 1775 | * XXX: __block_prepare_write() unmaps passed block, |
| 1776 | * is it OK? |
| 1777 | */ |
| 1778 | /* If the block was allocated from previously allocated cluster, |
| 1779 | * then we dont need to reserve it again. */ |
| 1780 | if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) { |
| 1781 | retval = ext4_da_reserve_space(inode, iblock); |
| 1782 | if (retval) |
| 1783 | /* not enough space to reserve */ |
| 1784 | goto out_unlock; |
| 1785 | } |
| 1786 | |
| 1787 | retval = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, |
| 1788 | ~0, EXTENT_STATUS_DELAYED); |
| 1789 | if (retval) |
| 1790 | goto out_unlock; |
| 1791 | |
| 1792 | /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served |
| 1793 | * and it should not appear on the bh->b_state. |
| 1794 | */ |
| 1795 | map->m_flags &= ~EXT4_MAP_FROM_CLUSTER; |
| 1796 | |
| 1797 | map_bh(bh, inode->i_sb, invalid_block); |
| 1798 | set_buffer_new(bh); |
| 1799 | set_buffer_delay(bh); |
| 1800 | } |
| 1801 | |
| 1802 | out_unlock: |
| 1803 | up_read((&EXT4_I(inode)->i_data_sem)); |
| 1804 | |
| 1805 | return retval; |
| 1806 | } |
| 1807 | |
| 1808 | /* |
| 1809 | * This is a special get_blocks_t callback which is used by |
| 1810 | * ext4_da_write_begin(). It will either return mapped block or |
| 1811 | * reserve space for a single block. |
| 1812 | * |
| 1813 | * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. |
| 1814 | * We also have b_blocknr = -1 and b_bdev initialized properly |
| 1815 | * |
| 1816 | * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. |
| 1817 | * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev |
| 1818 | * initialized properly. |
| 1819 | */ |
| 1820 | int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, |
| 1821 | struct buffer_head *bh, int create) |
| 1822 | { |
| 1823 | struct ext4_map_blocks map; |
| 1824 | int ret = 0; |
| 1825 | |
| 1826 | BUG_ON(create == 0); |
| 1827 | BUG_ON(bh->b_size != inode->i_sb->s_blocksize); |
| 1828 | |
| 1829 | map.m_lblk = iblock; |
| 1830 | map.m_len = 1; |
| 1831 | |
| 1832 | /* |
| 1833 | * first, we need to know whether the block is allocated already |
| 1834 | * preallocated blocks are unmapped but should treated |
| 1835 | * the same as allocated blocks. |
| 1836 | */ |
| 1837 | ret = ext4_da_map_blocks(inode, iblock, &map, bh); |
| 1838 | if (ret <= 0) |
| 1839 | return ret; |
| 1840 | |
| 1841 | map_bh(bh, inode->i_sb, map.m_pblk); |
| 1842 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; |
| 1843 | |
| 1844 | if (buffer_unwritten(bh)) { |
| 1845 | /* A delayed write to unwritten bh should be marked |
| 1846 | * new and mapped. Mapped ensures that we don't do |
| 1847 | * get_block multiple times when we write to the same |
| 1848 | * offset and new ensures that we do proper zero out |
| 1849 | * for partial write. |
| 1850 | */ |
| 1851 | set_buffer_new(bh); |
| 1852 | set_buffer_mapped(bh); |
| 1853 | } |
| 1854 | return 0; |
| 1855 | } |
| 1856 | |
| 1857 | static int bget_one(handle_t *handle, struct buffer_head *bh) |
| 1858 | { |
| 1859 | get_bh(bh); |
| 1860 | return 0; |
| 1861 | } |
| 1862 | |
| 1863 | static int bput_one(handle_t *handle, struct buffer_head *bh) |
| 1864 | { |
| 1865 | put_bh(bh); |
| 1866 | return 0; |
| 1867 | } |
| 1868 | |
| 1869 | static int __ext4_journalled_writepage(struct page *page, |
| 1870 | unsigned int len) |
| 1871 | { |
| 1872 | struct address_space *mapping = page->mapping; |
| 1873 | struct inode *inode = mapping->host; |
| 1874 | struct buffer_head *page_bufs = NULL; |
| 1875 | handle_t *handle = NULL; |
| 1876 | int ret = 0, err = 0; |
| 1877 | int inline_data = ext4_has_inline_data(inode); |
| 1878 | struct buffer_head *inode_bh = NULL; |
| 1879 | |
| 1880 | ClearPageChecked(page); |
| 1881 | |
| 1882 | if (inline_data) { |
| 1883 | BUG_ON(page->index != 0); |
| 1884 | BUG_ON(len > ext4_get_max_inline_size(inode)); |
| 1885 | inode_bh = ext4_journalled_write_inline_data(inode, len, page); |
| 1886 | if (inode_bh == NULL) |
| 1887 | goto out; |
| 1888 | } else { |
| 1889 | page_bufs = page_buffers(page); |
| 1890 | if (!page_bufs) { |
| 1891 | BUG(); |
| 1892 | goto out; |
| 1893 | } |
| 1894 | ext4_walk_page_buffers(handle, page_bufs, 0, len, |
| 1895 | NULL, bget_one); |
| 1896 | } |
| 1897 | /* As soon as we unlock the page, it can go away, but we have |
| 1898 | * references to buffers so we are safe */ |
| 1899 | unlock_page(page); |
| 1900 | |
| 1901 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
| 1902 | ext4_writepage_trans_blocks(inode)); |
| 1903 | if (IS_ERR(handle)) { |
| 1904 | ret = PTR_ERR(handle); |
| 1905 | goto out; |
| 1906 | } |
| 1907 | |
| 1908 | BUG_ON(!ext4_handle_valid(handle)); |
| 1909 | |
| 1910 | if (inline_data) { |
| 1911 | ret = ext4_journal_get_write_access(handle, inode_bh); |
| 1912 | |
| 1913 | err = ext4_handle_dirty_metadata(handle, inode, inode_bh); |
| 1914 | |
| 1915 | } else { |
| 1916 | ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, |
| 1917 | do_journal_get_write_access); |
| 1918 | |
| 1919 | err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, |
| 1920 | write_end_fn); |
| 1921 | } |
| 1922 | if (ret == 0) |
| 1923 | ret = err; |
| 1924 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
| 1925 | err = ext4_journal_stop(handle); |
| 1926 | if (!ret) |
| 1927 | ret = err; |
| 1928 | |
| 1929 | if (!ext4_has_inline_data(inode)) |
| 1930 | ext4_walk_page_buffers(handle, page_bufs, 0, len, |
| 1931 | NULL, bput_one); |
| 1932 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
| 1933 | out: |
| 1934 | brelse(inode_bh); |
| 1935 | return ret; |
| 1936 | } |
| 1937 | |
| 1938 | /* |
| 1939 | * Note that we don't need to start a transaction unless we're journaling data |
| 1940 | * because we should have holes filled from ext4_page_mkwrite(). We even don't |
| 1941 | * need to file the inode to the transaction's list in ordered mode because if |
| 1942 | * we are writing back data added by write(), the inode is already there and if |
| 1943 | * we are writing back data modified via mmap(), no one guarantees in which |
| 1944 | * transaction the data will hit the disk. In case we are journaling data, we |
| 1945 | * cannot start transaction directly because transaction start ranks above page |
| 1946 | * lock so we have to do some magic. |
| 1947 | * |
| 1948 | * This function can get called via... |
| 1949 | * - ext4_da_writepages after taking page lock (have journal handle) |
| 1950 | * - journal_submit_inode_data_buffers (no journal handle) |
| 1951 | * - shrink_page_list via the kswapd/direct reclaim (no journal handle) |
| 1952 | * - grab_page_cache when doing write_begin (have journal handle) |
| 1953 | * |
| 1954 | * We don't do any block allocation in this function. If we have page with |
| 1955 | * multiple blocks we need to write those buffer_heads that are mapped. This |
| 1956 | * is important for mmaped based write. So if we do with blocksize 1K |
| 1957 | * truncate(f, 1024); |
| 1958 | * a = mmap(f, 0, 4096); |
| 1959 | * a[0] = 'a'; |
| 1960 | * truncate(f, 4096); |
| 1961 | * we have in the page first buffer_head mapped via page_mkwrite call back |
| 1962 | * but other buffer_heads would be unmapped but dirty (dirty done via the |
| 1963 | * do_wp_page). So writepage should write the first block. If we modify |
| 1964 | * the mmap area beyond 1024 we will again get a page_fault and the |
| 1965 | * page_mkwrite callback will do the block allocation and mark the |
| 1966 | * buffer_heads mapped. |
| 1967 | * |
| 1968 | * We redirty the page if we have any buffer_heads that is either delay or |
| 1969 | * unwritten in the page. |
| 1970 | * |
| 1971 | * We can get recursively called as show below. |
| 1972 | * |
| 1973 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> |
| 1974 | * ext4_writepage() |
| 1975 | * |
| 1976 | * But since we don't do any block allocation we should not deadlock. |
| 1977 | * Page also have the dirty flag cleared so we don't get recurive page_lock. |
| 1978 | */ |
| 1979 | static int ext4_writepage(struct page *page, |
| 1980 | struct writeback_control *wbc) |
| 1981 | { |
| 1982 | int ret = 0; |
| 1983 | loff_t size; |
| 1984 | unsigned int len; |
| 1985 | struct buffer_head *page_bufs = NULL; |
| 1986 | struct inode *inode = page->mapping->host; |
| 1987 | struct ext4_io_submit io_submit; |
| 1988 | |
| 1989 | trace_ext4_writepage(page); |
| 1990 | size = i_size_read(inode); |
| 1991 | if (page->index == size >> PAGE_CACHE_SHIFT) |
| 1992 | len = size & ~PAGE_CACHE_MASK; |
| 1993 | else |
| 1994 | len = PAGE_CACHE_SIZE; |
| 1995 | |
| 1996 | page_bufs = page_buffers(page); |
| 1997 | /* |
| 1998 | * We cannot do block allocation or other extent handling in this |
| 1999 | * function. If there are buffers needing that, we have to redirty |
| 2000 | * the page. But we may reach here when we do a journal commit via |
| 2001 | * journal_submit_inode_data_buffers() and in that case we must write |
| 2002 | * allocated buffers to achieve data=ordered mode guarantees. |
| 2003 | */ |
| 2004 | if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL, |
| 2005 | ext4_bh_delay_or_unwritten)) { |
| 2006 | redirty_page_for_writepage(wbc, page); |
| 2007 | if (current->flags & PF_MEMALLOC) { |
| 2008 | /* |
| 2009 | * For memory cleaning there's no point in writing only |
| 2010 | * some buffers. So just bail out. Warn if we came here |
| 2011 | * from direct reclaim. |
| 2012 | */ |
| 2013 | WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) |
| 2014 | == PF_MEMALLOC); |
| 2015 | unlock_page(page); |
| 2016 | return 0; |
| 2017 | } |
| 2018 | } |
| 2019 | |
| 2020 | if (PageChecked(page) && ext4_should_journal_data(inode)) |
| 2021 | /* |
| 2022 | * It's mmapped pagecache. Add buffers and journal it. There |
| 2023 | * doesn't seem much point in redirtying the page here. |
| 2024 | */ |
| 2025 | return __ext4_journalled_writepage(page, len); |
| 2026 | |
| 2027 | memset(&io_submit, 0, sizeof(io_submit)); |
| 2028 | ret = ext4_bio_write_page(&io_submit, page, len, wbc); |
| 2029 | ext4_io_submit(&io_submit); |
| 2030 | return ret; |
| 2031 | } |
| 2032 | |
| 2033 | /* |
| 2034 | * This is called via ext4_da_writepages() to |
| 2035 | * calculate the total number of credits to reserve to fit |
| 2036 | * a single extent allocation into a single transaction, |
| 2037 | * ext4_da_writpeages() will loop calling this before |
| 2038 | * the block allocation. |
| 2039 | */ |
| 2040 | |
| 2041 | static int ext4_da_writepages_trans_blocks(struct inode *inode) |
| 2042 | { |
| 2043 | int max_blocks = EXT4_I(inode)->i_reserved_data_blocks; |
| 2044 | |
| 2045 | /* |
| 2046 | * With non-extent format the journal credit needed to |
| 2047 | * insert nrblocks contiguous block is dependent on |
| 2048 | * number of contiguous block. So we will limit |
| 2049 | * number of contiguous block to a sane value |
| 2050 | */ |
| 2051 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) && |
| 2052 | (max_blocks > EXT4_MAX_TRANS_DATA)) |
| 2053 | max_blocks = EXT4_MAX_TRANS_DATA; |
| 2054 | |
| 2055 | return ext4_chunk_trans_blocks(inode, max_blocks); |
| 2056 | } |
| 2057 | |
| 2058 | /* |
| 2059 | * write_cache_pages_da - walk the list of dirty pages of the given |
| 2060 | * address space and accumulate pages that need writing, and call |
| 2061 | * mpage_da_map_and_submit to map a single contiguous memory region |
| 2062 | * and then write them. |
| 2063 | */ |
| 2064 | static int write_cache_pages_da(handle_t *handle, |
| 2065 | struct address_space *mapping, |
| 2066 | struct writeback_control *wbc, |
| 2067 | struct mpage_da_data *mpd, |
| 2068 | pgoff_t *done_index) |
| 2069 | { |
| 2070 | struct buffer_head *bh, *head; |
| 2071 | struct inode *inode = mapping->host; |
| 2072 | struct pagevec pvec; |
| 2073 | unsigned int nr_pages; |
| 2074 | sector_t logical; |
| 2075 | pgoff_t index, end; |
| 2076 | long nr_to_write = wbc->nr_to_write; |
| 2077 | int i, tag, ret = 0; |
| 2078 | |
| 2079 | memset(mpd, 0, sizeof(struct mpage_da_data)); |
| 2080 | mpd->wbc = wbc; |
| 2081 | mpd->inode = inode; |
| 2082 | pagevec_init(&pvec, 0); |
| 2083 | index = wbc->range_start >> PAGE_CACHE_SHIFT; |
| 2084 | end = wbc->range_end >> PAGE_CACHE_SHIFT; |
| 2085 | |
| 2086 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
| 2087 | tag = PAGECACHE_TAG_TOWRITE; |
| 2088 | else |
| 2089 | tag = PAGECACHE_TAG_DIRTY; |
| 2090 | |
| 2091 | *done_index = index; |
| 2092 | while (index <= end) { |
| 2093 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
| 2094 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); |
| 2095 | if (nr_pages == 0) |
| 2096 | return 0; |
| 2097 | |
| 2098 | for (i = 0; i < nr_pages; i++) { |
| 2099 | struct page *page = pvec.pages[i]; |
| 2100 | |
| 2101 | /* |
| 2102 | * At this point, the page may be truncated or |
| 2103 | * invalidated (changing page->mapping to NULL), or |
| 2104 | * even swizzled back from swapper_space to tmpfs file |
| 2105 | * mapping. However, page->index will not change |
| 2106 | * because we have a reference on the page. |
| 2107 | */ |
| 2108 | if (page->index > end) |
| 2109 | goto out; |
| 2110 | |
| 2111 | *done_index = page->index + 1; |
| 2112 | |
| 2113 | /* |
| 2114 | * If we can't merge this page, and we have |
| 2115 | * accumulated an contiguous region, write it |
| 2116 | */ |
| 2117 | if ((mpd->next_page != page->index) && |
| 2118 | (mpd->next_page != mpd->first_page)) { |
| 2119 | mpage_da_map_and_submit(mpd); |
| 2120 | goto ret_extent_tail; |
| 2121 | } |
| 2122 | |
| 2123 | lock_page(page); |
| 2124 | |
| 2125 | /* |
| 2126 | * If the page is no longer dirty, or its |
| 2127 | * mapping no longer corresponds to inode we |
| 2128 | * are writing (which means it has been |
| 2129 | * truncated or invalidated), or the page is |
| 2130 | * already under writeback and we are not |
| 2131 | * doing a data integrity writeback, skip the page |
| 2132 | */ |
| 2133 | if (!PageDirty(page) || |
| 2134 | (PageWriteback(page) && |
| 2135 | (wbc->sync_mode == WB_SYNC_NONE)) || |
| 2136 | unlikely(page->mapping != mapping)) { |
| 2137 | unlock_page(page); |
| 2138 | continue; |
| 2139 | } |
| 2140 | |
| 2141 | wait_on_page_writeback(page); |
| 2142 | BUG_ON(PageWriteback(page)); |
| 2143 | |
| 2144 | /* |
| 2145 | * If we have inline data and arrive here, it means that |
| 2146 | * we will soon create the block for the 1st page, so |
| 2147 | * we'd better clear the inline data here. |
| 2148 | */ |
| 2149 | if (ext4_has_inline_data(inode)) { |
| 2150 | BUG_ON(ext4_test_inode_state(inode, |
| 2151 | EXT4_STATE_MAY_INLINE_DATA)); |
| 2152 | ext4_destroy_inline_data(handle, inode); |
| 2153 | } |
| 2154 | |
| 2155 | if (mpd->next_page != page->index) |
| 2156 | mpd->first_page = page->index; |
| 2157 | mpd->next_page = page->index + 1; |
| 2158 | logical = (sector_t) page->index << |
| 2159 | (PAGE_CACHE_SHIFT - inode->i_blkbits); |
| 2160 | |
| 2161 | /* Add all dirty buffers to mpd */ |
| 2162 | head = page_buffers(page); |
| 2163 | bh = head; |
| 2164 | do { |
| 2165 | BUG_ON(buffer_locked(bh)); |
| 2166 | /* |
| 2167 | * We need to try to allocate unmapped blocks |
| 2168 | * in the same page. Otherwise we won't make |
| 2169 | * progress with the page in ext4_writepage |
| 2170 | */ |
| 2171 | if (ext4_bh_delay_or_unwritten(NULL, bh)) { |
| 2172 | mpage_add_bh_to_extent(mpd, logical, |
| 2173 | bh->b_state); |
| 2174 | if (mpd->io_done) |
| 2175 | goto ret_extent_tail; |
| 2176 | } else if (buffer_dirty(bh) && |
| 2177 | buffer_mapped(bh)) { |
| 2178 | /* |
| 2179 | * mapped dirty buffer. We need to |
| 2180 | * update the b_state because we look |
| 2181 | * at b_state in mpage_da_map_blocks. |
| 2182 | * We don't update b_size because if we |
| 2183 | * find an unmapped buffer_head later |
| 2184 | * we need to use the b_state flag of |
| 2185 | * that buffer_head. |
| 2186 | */ |
| 2187 | if (mpd->b_size == 0) |
| 2188 | mpd->b_state = |
| 2189 | bh->b_state & BH_FLAGS; |
| 2190 | } |
| 2191 | logical++; |
| 2192 | } while ((bh = bh->b_this_page) != head); |
| 2193 | |
| 2194 | if (nr_to_write > 0) { |
| 2195 | nr_to_write--; |
| 2196 | if (nr_to_write == 0 && |
| 2197 | wbc->sync_mode == WB_SYNC_NONE) |
| 2198 | /* |
| 2199 | * We stop writing back only if we are |
| 2200 | * not doing integrity sync. In case of |
| 2201 | * integrity sync we have to keep going |
| 2202 | * because someone may be concurrently |
| 2203 | * dirtying pages, and we might have |
| 2204 | * synced a lot of newly appeared dirty |
| 2205 | * pages, but have not synced all of the |
| 2206 | * old dirty pages. |
| 2207 | */ |
| 2208 | goto out; |
| 2209 | } |
| 2210 | } |
| 2211 | pagevec_release(&pvec); |
| 2212 | cond_resched(); |
| 2213 | } |
| 2214 | return 0; |
| 2215 | ret_extent_tail: |
| 2216 | ret = MPAGE_DA_EXTENT_TAIL; |
| 2217 | out: |
| 2218 | pagevec_release(&pvec); |
| 2219 | cond_resched(); |
| 2220 | return ret; |
| 2221 | } |
| 2222 | |
| 2223 | |
| 2224 | static int ext4_da_writepages(struct address_space *mapping, |
| 2225 | struct writeback_control *wbc) |
| 2226 | { |
| 2227 | pgoff_t index; |
| 2228 | int range_whole = 0; |
| 2229 | handle_t *handle = NULL; |
| 2230 | struct mpage_da_data mpd; |
| 2231 | struct inode *inode = mapping->host; |
| 2232 | int pages_written = 0; |
| 2233 | unsigned int max_pages; |
| 2234 | int range_cyclic, cycled = 1, io_done = 0; |
| 2235 | int needed_blocks, ret = 0; |
| 2236 | long desired_nr_to_write, nr_to_writebump = 0; |
| 2237 | loff_t range_start = wbc->range_start; |
| 2238 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
| 2239 | pgoff_t done_index = 0; |
| 2240 | pgoff_t end; |
| 2241 | struct blk_plug plug; |
| 2242 | |
| 2243 | trace_ext4_da_writepages(inode, wbc); |
| 2244 | |
| 2245 | /* |
| 2246 | * No pages to write? This is mainly a kludge to avoid starting |
| 2247 | * a transaction for special inodes like journal inode on last iput() |
| 2248 | * because that could violate lock ordering on umount |
| 2249 | */ |
| 2250 | if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
| 2251 | return 0; |
| 2252 | |
| 2253 | /* |
| 2254 | * If the filesystem has aborted, it is read-only, so return |
| 2255 | * right away instead of dumping stack traces later on that |
| 2256 | * will obscure the real source of the problem. We test |
| 2257 | * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because |
| 2258 | * the latter could be true if the filesystem is mounted |
| 2259 | * read-only, and in that case, ext4_da_writepages should |
| 2260 | * *never* be called, so if that ever happens, we would want |
| 2261 | * the stack trace. |
| 2262 | */ |
| 2263 | if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) |
| 2264 | return -EROFS; |
| 2265 | |
| 2266 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
| 2267 | range_whole = 1; |
| 2268 | |
| 2269 | range_cyclic = wbc->range_cyclic; |
| 2270 | if (wbc->range_cyclic) { |
| 2271 | index = mapping->writeback_index; |
| 2272 | if (index) |
| 2273 | cycled = 0; |
| 2274 | wbc->range_start = index << PAGE_CACHE_SHIFT; |
| 2275 | wbc->range_end = LLONG_MAX; |
| 2276 | wbc->range_cyclic = 0; |
| 2277 | end = -1; |
| 2278 | } else { |
| 2279 | index = wbc->range_start >> PAGE_CACHE_SHIFT; |
| 2280 | end = wbc->range_end >> PAGE_CACHE_SHIFT; |
| 2281 | } |
| 2282 | |
| 2283 | /* |
| 2284 | * This works around two forms of stupidity. The first is in |
| 2285 | * the writeback code, which caps the maximum number of pages |
| 2286 | * written to be 1024 pages. This is wrong on multiple |
| 2287 | * levels; different architectues have a different page size, |
| 2288 | * which changes the maximum amount of data which gets |
| 2289 | * written. Secondly, 4 megabytes is way too small. XFS |
| 2290 | * forces this value to be 16 megabytes by multiplying |
| 2291 | * nr_to_write parameter by four, and then relies on its |
| 2292 | * allocator to allocate larger extents to make them |
| 2293 | * contiguous. Unfortunately this brings us to the second |
| 2294 | * stupidity, which is that ext4's mballoc code only allocates |
| 2295 | * at most 2048 blocks. So we force contiguous writes up to |
| 2296 | * the number of dirty blocks in the inode, or |
| 2297 | * sbi->max_writeback_mb_bump whichever is smaller. |
| 2298 | */ |
| 2299 | max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT); |
| 2300 | if (!range_cyclic && range_whole) { |
| 2301 | if (wbc->nr_to_write == LONG_MAX) |
| 2302 | desired_nr_to_write = wbc->nr_to_write; |
| 2303 | else |
| 2304 | desired_nr_to_write = wbc->nr_to_write * 8; |
| 2305 | } else |
| 2306 | desired_nr_to_write = ext4_num_dirty_pages(inode, index, |
| 2307 | max_pages); |
| 2308 | if (desired_nr_to_write > max_pages) |
| 2309 | desired_nr_to_write = max_pages; |
| 2310 | |
| 2311 | if (wbc->nr_to_write < desired_nr_to_write) { |
| 2312 | nr_to_writebump = desired_nr_to_write - wbc->nr_to_write; |
| 2313 | wbc->nr_to_write = desired_nr_to_write; |
| 2314 | } |
| 2315 | |
| 2316 | retry: |
| 2317 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
| 2318 | tag_pages_for_writeback(mapping, index, end); |
| 2319 | |
| 2320 | blk_start_plug(&plug); |
| 2321 | while (!ret && wbc->nr_to_write > 0) { |
| 2322 | |
| 2323 | /* |
| 2324 | * we insert one extent at a time. So we need |
| 2325 | * credit needed for single extent allocation. |
| 2326 | * journalled mode is currently not supported |
| 2327 | * by delalloc |
| 2328 | */ |
| 2329 | BUG_ON(ext4_should_journal_data(inode)); |
| 2330 | needed_blocks = ext4_da_writepages_trans_blocks(inode); |
| 2331 | |
| 2332 | /* start a new transaction*/ |
| 2333 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
| 2334 | needed_blocks); |
| 2335 | if (IS_ERR(handle)) { |
| 2336 | ret = PTR_ERR(handle); |
| 2337 | ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " |
| 2338 | "%ld pages, ino %lu; err %d", __func__, |
| 2339 | wbc->nr_to_write, inode->i_ino, ret); |
| 2340 | blk_finish_plug(&plug); |
| 2341 | goto out_writepages; |
| 2342 | } |
| 2343 | |
| 2344 | /* |
| 2345 | * Now call write_cache_pages_da() to find the next |
| 2346 | * contiguous region of logical blocks that need |
| 2347 | * blocks to be allocated by ext4 and submit them. |
| 2348 | */ |
| 2349 | ret = write_cache_pages_da(handle, mapping, |
| 2350 | wbc, &mpd, &done_index); |
| 2351 | /* |
| 2352 | * If we have a contiguous extent of pages and we |
| 2353 | * haven't done the I/O yet, map the blocks and submit |
| 2354 | * them for I/O. |
| 2355 | */ |
| 2356 | if (!mpd.io_done && mpd.next_page != mpd.first_page) { |
| 2357 | mpage_da_map_and_submit(&mpd); |
| 2358 | ret = MPAGE_DA_EXTENT_TAIL; |
| 2359 | } |
| 2360 | trace_ext4_da_write_pages(inode, &mpd); |
| 2361 | wbc->nr_to_write -= mpd.pages_written; |
| 2362 | |
| 2363 | ext4_journal_stop(handle); |
| 2364 | |
| 2365 | if ((mpd.retval == -ENOSPC) && sbi->s_journal) { |
| 2366 | /* commit the transaction which would |
| 2367 | * free blocks released in the transaction |
| 2368 | * and try again |
| 2369 | */ |
| 2370 | jbd2_journal_force_commit_nested(sbi->s_journal); |
| 2371 | ret = 0; |
| 2372 | } else if (ret == MPAGE_DA_EXTENT_TAIL) { |
| 2373 | /* |
| 2374 | * Got one extent now try with rest of the pages. |
| 2375 | * If mpd.retval is set -EIO, journal is aborted. |
| 2376 | * So we don't need to write any more. |
| 2377 | */ |
| 2378 | pages_written += mpd.pages_written; |
| 2379 | ret = mpd.retval; |
| 2380 | io_done = 1; |
| 2381 | } else if (wbc->nr_to_write) |
| 2382 | /* |
| 2383 | * There is no more writeout needed |
| 2384 | * or we requested for a noblocking writeout |
| 2385 | * and we found the device congested |
| 2386 | */ |
| 2387 | break; |
| 2388 | } |
| 2389 | blk_finish_plug(&plug); |
| 2390 | if (!io_done && !cycled) { |
| 2391 | cycled = 1; |
| 2392 | index = 0; |
| 2393 | wbc->range_start = index << PAGE_CACHE_SHIFT; |
| 2394 | wbc->range_end = mapping->writeback_index - 1; |
| 2395 | goto retry; |
| 2396 | } |
| 2397 | |
| 2398 | /* Update index */ |
| 2399 | wbc->range_cyclic = range_cyclic; |
| 2400 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
| 2401 | /* |
| 2402 | * set the writeback_index so that range_cyclic |
| 2403 | * mode will write it back later |
| 2404 | */ |
| 2405 | mapping->writeback_index = done_index; |
| 2406 | |
| 2407 | out_writepages: |
| 2408 | wbc->nr_to_write -= nr_to_writebump; |
| 2409 | wbc->range_start = range_start; |
| 2410 | trace_ext4_da_writepages_result(inode, wbc, ret, pages_written); |
| 2411 | return ret; |
| 2412 | } |
| 2413 | |
| 2414 | static int ext4_nonda_switch(struct super_block *sb) |
| 2415 | { |
| 2416 | s64 free_blocks, dirty_blocks; |
| 2417 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
| 2418 | |
| 2419 | /* |
| 2420 | * switch to non delalloc mode if we are running low |
| 2421 | * on free block. The free block accounting via percpu |
| 2422 | * counters can get slightly wrong with percpu_counter_batch getting |
| 2423 | * accumulated on each CPU without updating global counters |
| 2424 | * Delalloc need an accurate free block accounting. So switch |
| 2425 | * to non delalloc when we are near to error range. |
| 2426 | */ |
| 2427 | free_blocks = EXT4_C2B(sbi, |
| 2428 | percpu_counter_read_positive(&sbi->s_freeclusters_counter)); |
| 2429 | dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyclusters_counter); |
| 2430 | /* |
| 2431 | * Start pushing delalloc when 1/2 of free blocks are dirty. |
| 2432 | */ |
| 2433 | if (dirty_blocks && (free_blocks < 2 * dirty_blocks) && |
| 2434 | !writeback_in_progress(sb->s_bdi) && |
| 2435 | down_read_trylock(&sb->s_umount)) { |
| 2436 | writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE); |
| 2437 | up_read(&sb->s_umount); |
| 2438 | } |
| 2439 | |
| 2440 | if (2 * free_blocks < 3 * dirty_blocks || |
| 2441 | free_blocks < (dirty_blocks + EXT4_FREECLUSTERS_WATERMARK)) { |
| 2442 | /* |
| 2443 | * free block count is less than 150% of dirty blocks |
| 2444 | * or free blocks is less than watermark |
| 2445 | */ |
| 2446 | return 1; |
| 2447 | } |
| 2448 | return 0; |
| 2449 | } |
| 2450 | |
| 2451 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, |
| 2452 | loff_t pos, unsigned len, unsigned flags, |
| 2453 | struct page **pagep, void **fsdata) |
| 2454 | { |
| 2455 | int ret, retries = 0; |
| 2456 | struct page *page; |
| 2457 | pgoff_t index; |
| 2458 | struct inode *inode = mapping->host; |
| 2459 | handle_t *handle; |
| 2460 | |
| 2461 | index = pos >> PAGE_CACHE_SHIFT; |
| 2462 | |
| 2463 | if (ext4_nonda_switch(inode->i_sb)) { |
| 2464 | *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; |
| 2465 | return ext4_write_begin(file, mapping, pos, |
| 2466 | len, flags, pagep, fsdata); |
| 2467 | } |
| 2468 | *fsdata = (void *)0; |
| 2469 | trace_ext4_da_write_begin(inode, pos, len, flags); |
| 2470 | |
| 2471 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { |
| 2472 | ret = ext4_da_write_inline_data_begin(mapping, inode, |
| 2473 | pos, len, flags, |
| 2474 | pagep, fsdata); |
| 2475 | if (ret < 0) |
| 2476 | return ret; |
| 2477 | if (ret == 1) |
| 2478 | return 0; |
| 2479 | } |
| 2480 | |
| 2481 | /* |
| 2482 | * grab_cache_page_write_begin() can take a long time if the |
| 2483 | * system is thrashing due to memory pressure, or if the page |
| 2484 | * is being written back. So grab it first before we start |
| 2485 | * the transaction handle. This also allows us to allocate |
| 2486 | * the page (if needed) without using GFP_NOFS. |
| 2487 | */ |
| 2488 | retry_grab: |
| 2489 | page = grab_cache_page_write_begin(mapping, index, flags); |
| 2490 | if (!page) |
| 2491 | return -ENOMEM; |
| 2492 | unlock_page(page); |
| 2493 | |
| 2494 | /* |
| 2495 | * With delayed allocation, we don't log the i_disksize update |
| 2496 | * if there is delayed block allocation. But we still need |
| 2497 | * to journalling the i_disksize update if writes to the end |
| 2498 | * of file which has an already mapped buffer. |
| 2499 | */ |
| 2500 | retry_journal: |
| 2501 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, 1); |
| 2502 | if (IS_ERR(handle)) { |
| 2503 | page_cache_release(page); |
| 2504 | return PTR_ERR(handle); |
| 2505 | } |
| 2506 | |
| 2507 | lock_page(page); |
| 2508 | if (page->mapping != mapping) { |
| 2509 | /* The page got truncated from under us */ |
| 2510 | unlock_page(page); |
| 2511 | page_cache_release(page); |
| 2512 | ext4_journal_stop(handle); |
| 2513 | goto retry_grab; |
| 2514 | } |
| 2515 | /* In case writeback began while the page was unlocked */ |
| 2516 | wait_on_page_writeback(page); |
| 2517 | |
| 2518 | ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep); |
| 2519 | if (ret < 0) { |
| 2520 | unlock_page(page); |
| 2521 | ext4_journal_stop(handle); |
| 2522 | /* |
| 2523 | * block_write_begin may have instantiated a few blocks |
| 2524 | * outside i_size. Trim these off again. Don't need |
| 2525 | * i_size_read because we hold i_mutex. |
| 2526 | */ |
| 2527 | if (pos + len > inode->i_size) |
| 2528 | ext4_truncate_failed_write(inode); |
| 2529 | |
| 2530 | if (ret == -ENOSPC && |
| 2531 | ext4_should_retry_alloc(inode->i_sb, &retries)) |
| 2532 | goto retry_journal; |
| 2533 | |
| 2534 | page_cache_release(page); |
| 2535 | return ret; |
| 2536 | } |
| 2537 | |
| 2538 | *pagep = page; |
| 2539 | return ret; |
| 2540 | } |
| 2541 | |
| 2542 | /* |
| 2543 | * Check if we should update i_disksize |
| 2544 | * when write to the end of file but not require block allocation |
| 2545 | */ |
| 2546 | static int ext4_da_should_update_i_disksize(struct page *page, |
| 2547 | unsigned long offset) |
| 2548 | { |
| 2549 | struct buffer_head *bh; |
| 2550 | struct inode *inode = page->mapping->host; |
| 2551 | unsigned int idx; |
| 2552 | int i; |
| 2553 | |
| 2554 | bh = page_buffers(page); |
| 2555 | idx = offset >> inode->i_blkbits; |
| 2556 | |
| 2557 | for (i = 0; i < idx; i++) |
| 2558 | bh = bh->b_this_page; |
| 2559 | |
| 2560 | if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
| 2561 | return 0; |
| 2562 | return 1; |
| 2563 | } |
| 2564 | |
| 2565 | static int ext4_da_write_end(struct file *file, |
| 2566 | struct address_space *mapping, |
| 2567 | loff_t pos, unsigned len, unsigned copied, |
| 2568 | struct page *page, void *fsdata) |
| 2569 | { |
| 2570 | struct inode *inode = mapping->host; |
| 2571 | int ret = 0, ret2; |
| 2572 | handle_t *handle = ext4_journal_current_handle(); |
| 2573 | loff_t new_i_size; |
| 2574 | unsigned long start, end; |
| 2575 | int write_mode = (int)(unsigned long)fsdata; |
| 2576 | |
| 2577 | if (write_mode == FALL_BACK_TO_NONDELALLOC) { |
| 2578 | switch (ext4_inode_journal_mode(inode)) { |
| 2579 | case EXT4_INODE_ORDERED_DATA_MODE: |
| 2580 | return ext4_ordered_write_end(file, mapping, pos, |
| 2581 | len, copied, page, fsdata); |
| 2582 | case EXT4_INODE_WRITEBACK_DATA_MODE: |
| 2583 | return ext4_writeback_write_end(file, mapping, pos, |
| 2584 | len, copied, page, fsdata); |
| 2585 | default: |
| 2586 | BUG(); |
| 2587 | } |
| 2588 | } |
| 2589 | |
| 2590 | trace_ext4_da_write_end(inode, pos, len, copied); |
| 2591 | start = pos & (PAGE_CACHE_SIZE - 1); |
| 2592 | end = start + copied - 1; |
| 2593 | |
| 2594 | /* |
| 2595 | * generic_write_end() will run mark_inode_dirty() if i_size |
| 2596 | * changes. So let's piggyback the i_disksize mark_inode_dirty |
| 2597 | * into that. |
| 2598 | */ |
| 2599 | new_i_size = pos + copied; |
| 2600 | if (copied && new_i_size > EXT4_I(inode)->i_disksize) { |
| 2601 | if (ext4_has_inline_data(inode) || |
| 2602 | ext4_da_should_update_i_disksize(page, end)) { |
| 2603 | down_write(&EXT4_I(inode)->i_data_sem); |
| 2604 | if (new_i_size > EXT4_I(inode)->i_disksize) |
| 2605 | EXT4_I(inode)->i_disksize = new_i_size; |
| 2606 | up_write(&EXT4_I(inode)->i_data_sem); |
| 2607 | /* We need to mark inode dirty even if |
| 2608 | * new_i_size is less that inode->i_size |
| 2609 | * bu greater than i_disksize.(hint delalloc) |
| 2610 | */ |
| 2611 | ext4_mark_inode_dirty(handle, inode); |
| 2612 | } |
| 2613 | } |
| 2614 | |
| 2615 | if (write_mode != CONVERT_INLINE_DATA && |
| 2616 | ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) && |
| 2617 | ext4_has_inline_data(inode)) |
| 2618 | ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied, |
| 2619 | page); |
| 2620 | else |
| 2621 | ret2 = generic_write_end(file, mapping, pos, len, copied, |
| 2622 | page, fsdata); |
| 2623 | |
| 2624 | copied = ret2; |
| 2625 | if (ret2 < 0) |
| 2626 | ret = ret2; |
| 2627 | ret2 = ext4_journal_stop(handle); |
| 2628 | if (!ret) |
| 2629 | ret = ret2; |
| 2630 | |
| 2631 | return ret ? ret : copied; |
| 2632 | } |
| 2633 | |
| 2634 | static void ext4_da_invalidatepage(struct page *page, unsigned long offset) |
| 2635 | { |
| 2636 | /* |
| 2637 | * Drop reserved blocks |
| 2638 | */ |
| 2639 | BUG_ON(!PageLocked(page)); |
| 2640 | if (!page_has_buffers(page)) |
| 2641 | goto out; |
| 2642 | |
| 2643 | ext4_da_page_release_reservation(page, offset); |
| 2644 | |
| 2645 | out: |
| 2646 | ext4_invalidatepage(page, offset); |
| 2647 | |
| 2648 | return; |
| 2649 | } |
| 2650 | |
| 2651 | /* |
| 2652 | * Force all delayed allocation blocks to be allocated for a given inode. |
| 2653 | */ |
| 2654 | int ext4_alloc_da_blocks(struct inode *inode) |
| 2655 | { |
| 2656 | trace_ext4_alloc_da_blocks(inode); |
| 2657 | |
| 2658 | if (!EXT4_I(inode)->i_reserved_data_blocks && |
| 2659 | !EXT4_I(inode)->i_reserved_meta_blocks) |
| 2660 | return 0; |
| 2661 | |
| 2662 | /* |
| 2663 | * We do something simple for now. The filemap_flush() will |
| 2664 | * also start triggering a write of the data blocks, which is |
| 2665 | * not strictly speaking necessary (and for users of |
| 2666 | * laptop_mode, not even desirable). However, to do otherwise |
| 2667 | * would require replicating code paths in: |
| 2668 | * |
| 2669 | * ext4_da_writepages() -> |
| 2670 | * write_cache_pages() ---> (via passed in callback function) |
| 2671 | * __mpage_da_writepage() --> |
| 2672 | * mpage_add_bh_to_extent() |
| 2673 | * mpage_da_map_blocks() |
| 2674 | * |
| 2675 | * The problem is that write_cache_pages(), located in |
| 2676 | * mm/page-writeback.c, marks pages clean in preparation for |
| 2677 | * doing I/O, which is not desirable if we're not planning on |
| 2678 | * doing I/O at all. |
| 2679 | * |
| 2680 | * We could call write_cache_pages(), and then redirty all of |
| 2681 | * the pages by calling redirty_page_for_writepage() but that |
| 2682 | * would be ugly in the extreme. So instead we would need to |
| 2683 | * replicate parts of the code in the above functions, |
| 2684 | * simplifying them because we wouldn't actually intend to |
| 2685 | * write out the pages, but rather only collect contiguous |
| 2686 | * logical block extents, call the multi-block allocator, and |
| 2687 | * then update the buffer heads with the block allocations. |
| 2688 | * |
| 2689 | * For now, though, we'll cheat by calling filemap_flush(), |
| 2690 | * which will map the blocks, and start the I/O, but not |
| 2691 | * actually wait for the I/O to complete. |
| 2692 | */ |
| 2693 | return filemap_flush(inode->i_mapping); |
| 2694 | } |
| 2695 | |
| 2696 | /* |
| 2697 | * bmap() is special. It gets used by applications such as lilo and by |
| 2698 | * the swapper to find the on-disk block of a specific piece of data. |
| 2699 | * |
| 2700 | * Naturally, this is dangerous if the block concerned is still in the |
| 2701 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
| 2702 | * filesystem and enables swap, then they may get a nasty shock when the |
| 2703 | * data getting swapped to that swapfile suddenly gets overwritten by |
| 2704 | * the original zero's written out previously to the journal and |
| 2705 | * awaiting writeback in the kernel's buffer cache. |
| 2706 | * |
| 2707 | * So, if we see any bmap calls here on a modified, data-journaled file, |
| 2708 | * take extra steps to flush any blocks which might be in the cache. |
| 2709 | */ |
| 2710 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
| 2711 | { |
| 2712 | struct inode *inode = mapping->host; |
| 2713 | journal_t *journal; |
| 2714 | int err; |
| 2715 | |
| 2716 | /* |
| 2717 | * We can get here for an inline file via the FIBMAP ioctl |
| 2718 | */ |
| 2719 | if (ext4_has_inline_data(inode)) |
| 2720 | return 0; |
| 2721 | |
| 2722 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
| 2723 | test_opt(inode->i_sb, DELALLOC)) { |
| 2724 | /* |
| 2725 | * With delalloc we want to sync the file |
| 2726 | * so that we can make sure we allocate |
| 2727 | * blocks for file |
| 2728 | */ |
| 2729 | filemap_write_and_wait(mapping); |
| 2730 | } |
| 2731 | |
| 2732 | if (EXT4_JOURNAL(inode) && |
| 2733 | ext4_test_inode_state(inode, EXT4_STATE_JDATA)) { |
| 2734 | /* |
| 2735 | * This is a REALLY heavyweight approach, but the use of |
| 2736 | * bmap on dirty files is expected to be extremely rare: |
| 2737 | * only if we run lilo or swapon on a freshly made file |
| 2738 | * do we expect this to happen. |
| 2739 | * |
| 2740 | * (bmap requires CAP_SYS_RAWIO so this does not |
| 2741 | * represent an unprivileged user DOS attack --- we'd be |
| 2742 | * in trouble if mortal users could trigger this path at |
| 2743 | * will.) |
| 2744 | * |
| 2745 | * NB. EXT4_STATE_JDATA is not set on files other than |
| 2746 | * regular files. If somebody wants to bmap a directory |
| 2747 | * or symlink and gets confused because the buffer |
| 2748 | * hasn't yet been flushed to disk, they deserve |
| 2749 | * everything they get. |
| 2750 | */ |
| 2751 | |
| 2752 | ext4_clear_inode_state(inode, EXT4_STATE_JDATA); |
| 2753 | journal = EXT4_JOURNAL(inode); |
| 2754 | jbd2_journal_lock_updates(journal); |
| 2755 | err = jbd2_journal_flush(journal); |
| 2756 | jbd2_journal_unlock_updates(journal); |
| 2757 | |
| 2758 | if (err) |
| 2759 | return 0; |
| 2760 | } |
| 2761 | |
| 2762 | return generic_block_bmap(mapping, block, ext4_get_block); |
| 2763 | } |
| 2764 | |
| 2765 | static int ext4_readpage(struct file *file, struct page *page) |
| 2766 | { |
| 2767 | int ret = -EAGAIN; |
| 2768 | struct inode *inode = page->mapping->host; |
| 2769 | |
| 2770 | trace_ext4_readpage(page); |
| 2771 | |
| 2772 | if (ext4_has_inline_data(inode)) |
| 2773 | ret = ext4_readpage_inline(inode, page); |
| 2774 | |
| 2775 | if (ret == -EAGAIN) |
| 2776 | return mpage_readpage(page, ext4_get_block); |
| 2777 | |
| 2778 | return ret; |
| 2779 | } |
| 2780 | |
| 2781 | static int |
| 2782 | ext4_readpages(struct file *file, struct address_space *mapping, |
| 2783 | struct list_head *pages, unsigned nr_pages) |
| 2784 | { |
| 2785 | struct inode *inode = mapping->host; |
| 2786 | |
| 2787 | /* If the file has inline data, no need to do readpages. */ |
| 2788 | if (ext4_has_inline_data(inode)) |
| 2789 | return 0; |
| 2790 | |
| 2791 | return mpage_readpages(mapping, pages, nr_pages, ext4_get_block); |
| 2792 | } |
| 2793 | |
| 2794 | static void ext4_invalidatepage(struct page *page, unsigned long offset) |
| 2795 | { |
| 2796 | trace_ext4_invalidatepage(page, offset); |
| 2797 | |
| 2798 | /* No journalling happens on data buffers when this function is used */ |
| 2799 | WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page))); |
| 2800 | |
| 2801 | block_invalidatepage(page, offset); |
| 2802 | } |
| 2803 | |
| 2804 | static int __ext4_journalled_invalidatepage(struct page *page, |
| 2805 | unsigned long offset) |
| 2806 | { |
| 2807 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
| 2808 | |
| 2809 | trace_ext4_journalled_invalidatepage(page, offset); |
| 2810 | |
| 2811 | /* |
| 2812 | * If it's a full truncate we just forget about the pending dirtying |
| 2813 | */ |
| 2814 | if (offset == 0) |
| 2815 | ClearPageChecked(page); |
| 2816 | |
| 2817 | return jbd2_journal_invalidatepage(journal, page, offset); |
| 2818 | } |
| 2819 | |
| 2820 | /* Wrapper for aops... */ |
| 2821 | static void ext4_journalled_invalidatepage(struct page *page, |
| 2822 | unsigned long offset) |
| 2823 | { |
| 2824 | WARN_ON(__ext4_journalled_invalidatepage(page, offset) < 0); |
| 2825 | } |
| 2826 | |
| 2827 | static int ext4_releasepage(struct page *page, gfp_t wait) |
| 2828 | { |
| 2829 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
| 2830 | |
| 2831 | trace_ext4_releasepage(page); |
| 2832 | |
| 2833 | WARN_ON(PageChecked(page)); |
| 2834 | if (!page_has_buffers(page)) |
| 2835 | return 0; |
| 2836 | if (journal) |
| 2837 | return jbd2_journal_try_to_free_buffers(journal, page, wait); |
| 2838 | else |
| 2839 | return try_to_free_buffers(page); |
| 2840 | } |
| 2841 | |
| 2842 | /* |
| 2843 | * ext4_get_block used when preparing for a DIO write or buffer write. |
| 2844 | * We allocate an uinitialized extent if blocks haven't been allocated. |
| 2845 | * The extent will be converted to initialized after the IO is complete. |
| 2846 | */ |
| 2847 | int ext4_get_block_write(struct inode *inode, sector_t iblock, |
| 2848 | struct buffer_head *bh_result, int create) |
| 2849 | { |
| 2850 | ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n", |
| 2851 | inode->i_ino, create); |
| 2852 | return _ext4_get_block(inode, iblock, bh_result, |
| 2853 | EXT4_GET_BLOCKS_IO_CREATE_EXT); |
| 2854 | } |
| 2855 | |
| 2856 | static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock, |
| 2857 | struct buffer_head *bh_result, int create) |
| 2858 | { |
| 2859 | ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n", |
| 2860 | inode->i_ino, create); |
| 2861 | return _ext4_get_block(inode, iblock, bh_result, |
| 2862 | EXT4_GET_BLOCKS_NO_LOCK); |
| 2863 | } |
| 2864 | |
| 2865 | static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset, |
| 2866 | ssize_t size, void *private, int ret, |
| 2867 | bool is_async) |
| 2868 | { |
| 2869 | struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode; |
| 2870 | ext4_io_end_t *io_end = iocb->private; |
| 2871 | |
| 2872 | /* if not async direct IO or dio with 0 bytes write, just return */ |
| 2873 | if (!io_end || !size) |
| 2874 | goto out; |
| 2875 | |
| 2876 | ext_debug("ext4_end_io_dio(): io_end 0x%p " |
| 2877 | "for inode %lu, iocb 0x%p, offset %llu, size %zd\n", |
| 2878 | iocb->private, io_end->inode->i_ino, iocb, offset, |
| 2879 | size); |
| 2880 | |
| 2881 | iocb->private = NULL; |
| 2882 | |
| 2883 | /* if not aio dio with unwritten extents, just free io and return */ |
| 2884 | if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) { |
| 2885 | ext4_free_io_end(io_end); |
| 2886 | out: |
| 2887 | inode_dio_done(inode); |
| 2888 | if (is_async) |
| 2889 | aio_complete(iocb, ret, 0); |
| 2890 | return; |
| 2891 | } |
| 2892 | |
| 2893 | io_end->offset = offset; |
| 2894 | io_end->size = size; |
| 2895 | if (is_async) { |
| 2896 | io_end->iocb = iocb; |
| 2897 | io_end->result = ret; |
| 2898 | } |
| 2899 | |
| 2900 | ext4_add_complete_io(io_end); |
| 2901 | } |
| 2902 | |
| 2903 | /* |
| 2904 | * For ext4 extent files, ext4 will do direct-io write to holes, |
| 2905 | * preallocated extents, and those write extend the file, no need to |
| 2906 | * fall back to buffered IO. |
| 2907 | * |
| 2908 | * For holes, we fallocate those blocks, mark them as uninitialized |
| 2909 | * If those blocks were preallocated, we mark sure they are split, but |
| 2910 | * still keep the range to write as uninitialized. |
| 2911 | * |
| 2912 | * The unwritten extents will be converted to written when DIO is completed. |
| 2913 | * For async direct IO, since the IO may still pending when return, we |
| 2914 | * set up an end_io call back function, which will do the conversion |
| 2915 | * when async direct IO completed. |
| 2916 | * |
| 2917 | * If the O_DIRECT write will extend the file then add this inode to the |
| 2918 | * orphan list. So recovery will truncate it back to the original size |
| 2919 | * if the machine crashes during the write. |
| 2920 | * |
| 2921 | */ |
| 2922 | static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb, |
| 2923 | const struct iovec *iov, loff_t offset, |
| 2924 | unsigned long nr_segs) |
| 2925 | { |
| 2926 | struct file *file = iocb->ki_filp; |
| 2927 | struct inode *inode = file->f_mapping->host; |
| 2928 | ssize_t ret; |
| 2929 | size_t count = iov_length(iov, nr_segs); |
| 2930 | int overwrite = 0; |
| 2931 | get_block_t *get_block_func = NULL; |
| 2932 | int dio_flags = 0; |
| 2933 | loff_t final_size = offset + count; |
| 2934 | |
| 2935 | /* Use the old path for reads and writes beyond i_size. */ |
| 2936 | if (rw != WRITE || final_size > inode->i_size) |
| 2937 | return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs); |
| 2938 | |
| 2939 | BUG_ON(iocb->private == NULL); |
| 2940 | |
| 2941 | /* If we do a overwrite dio, i_mutex locking can be released */ |
| 2942 | overwrite = *((int *)iocb->private); |
| 2943 | |
| 2944 | if (overwrite) { |
| 2945 | atomic_inc(&inode->i_dio_count); |
| 2946 | down_read(&EXT4_I(inode)->i_data_sem); |
| 2947 | mutex_unlock(&inode->i_mutex); |
| 2948 | } |
| 2949 | |
| 2950 | /* |
| 2951 | * We could direct write to holes and fallocate. |
| 2952 | * |
| 2953 | * Allocated blocks to fill the hole are marked as |
| 2954 | * uninitialized to prevent parallel buffered read to expose |
| 2955 | * the stale data before DIO complete the data IO. |
| 2956 | * |
| 2957 | * As to previously fallocated extents, ext4 get_block will |
| 2958 | * just simply mark the buffer mapped but still keep the |
| 2959 | * extents uninitialized. |
| 2960 | * |
| 2961 | * For non AIO case, we will convert those unwritten extents |
| 2962 | * to written after return back from blockdev_direct_IO. |
| 2963 | * |
| 2964 | * For async DIO, the conversion needs to be deferred when the |
| 2965 | * IO is completed. The ext4 end_io callback function will be |
| 2966 | * called to take care of the conversion work. Here for async |
| 2967 | * case, we allocate an io_end structure to hook to the iocb. |
| 2968 | */ |
| 2969 | iocb->private = NULL; |
| 2970 | ext4_inode_aio_set(inode, NULL); |
| 2971 | if (!is_sync_kiocb(iocb)) { |
| 2972 | ext4_io_end_t *io_end = ext4_init_io_end(inode, GFP_NOFS); |
| 2973 | if (!io_end) { |
| 2974 | ret = -ENOMEM; |
| 2975 | goto retake_lock; |
| 2976 | } |
| 2977 | io_end->flag |= EXT4_IO_END_DIRECT; |
| 2978 | iocb->private = io_end; |
| 2979 | /* |
| 2980 | * we save the io structure for current async direct |
| 2981 | * IO, so that later ext4_map_blocks() could flag the |
| 2982 | * io structure whether there is a unwritten extents |
| 2983 | * needs to be converted when IO is completed. |
| 2984 | */ |
| 2985 | ext4_inode_aio_set(inode, io_end); |
| 2986 | } |
| 2987 | |
| 2988 | if (overwrite) { |
| 2989 | get_block_func = ext4_get_block_write_nolock; |
| 2990 | } else { |
| 2991 | get_block_func = ext4_get_block_write; |
| 2992 | dio_flags = DIO_LOCKING; |
| 2993 | } |
| 2994 | ret = __blockdev_direct_IO(rw, iocb, inode, |
| 2995 | inode->i_sb->s_bdev, iov, |
| 2996 | offset, nr_segs, |
| 2997 | get_block_func, |
| 2998 | ext4_end_io_dio, |
| 2999 | NULL, |
| 3000 | dio_flags); |
| 3001 | |
| 3002 | if (iocb->private) |
| 3003 | ext4_inode_aio_set(inode, NULL); |
| 3004 | /* |
| 3005 | * The io_end structure takes a reference to the inode, that |
| 3006 | * structure needs to be destroyed and the reference to the |
| 3007 | * inode need to be dropped, when IO is complete, even with 0 |
| 3008 | * byte write, or failed. |
| 3009 | * |
| 3010 | * In the successful AIO DIO case, the io_end structure will |
| 3011 | * be destroyed and the reference to the inode will be dropped |
| 3012 | * after the end_io call back function is called. |
| 3013 | * |
| 3014 | * In the case there is 0 byte write, or error case, since VFS |
| 3015 | * direct IO won't invoke the end_io call back function, we |
| 3016 | * need to free the end_io structure here. |
| 3017 | */ |
| 3018 | if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) { |
| 3019 | ext4_free_io_end(iocb->private); |
| 3020 | iocb->private = NULL; |
| 3021 | } else if (ret > 0 && !overwrite && ext4_test_inode_state(inode, |
| 3022 | EXT4_STATE_DIO_UNWRITTEN)) { |
| 3023 | int err; |
| 3024 | /* |
| 3025 | * for non AIO case, since the IO is already |
| 3026 | * completed, we could do the conversion right here |
| 3027 | */ |
| 3028 | err = ext4_convert_unwritten_extents(inode, |
| 3029 | offset, ret); |
| 3030 | if (err < 0) |
| 3031 | ret = err; |
| 3032 | ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); |
| 3033 | } |
| 3034 | |
| 3035 | retake_lock: |
| 3036 | /* take i_mutex locking again if we do a ovewrite dio */ |
| 3037 | if (overwrite) { |
| 3038 | inode_dio_done(inode); |
| 3039 | up_read(&EXT4_I(inode)->i_data_sem); |
| 3040 | mutex_lock(&inode->i_mutex); |
| 3041 | } |
| 3042 | |
| 3043 | return ret; |
| 3044 | } |
| 3045 | |
| 3046 | static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb, |
| 3047 | const struct iovec *iov, loff_t offset, |
| 3048 | unsigned long nr_segs) |
| 3049 | { |
| 3050 | struct file *file = iocb->ki_filp; |
| 3051 | struct inode *inode = file->f_mapping->host; |
| 3052 | ssize_t ret; |
| 3053 | |
| 3054 | /* |
| 3055 | * If we are doing data journalling we don't support O_DIRECT |
| 3056 | */ |
| 3057 | if (ext4_should_journal_data(inode)) |
| 3058 | return 0; |
| 3059 | |
| 3060 | /* Let buffer I/O handle the inline data case. */ |
| 3061 | if (ext4_has_inline_data(inode)) |
| 3062 | return 0; |
| 3063 | |
| 3064 | trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw); |
| 3065 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
| 3066 | ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs); |
| 3067 | else |
| 3068 | ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs); |
| 3069 | trace_ext4_direct_IO_exit(inode, offset, |
| 3070 | iov_length(iov, nr_segs), rw, ret); |
| 3071 | return ret; |
| 3072 | } |
| 3073 | |
| 3074 | /* |
| 3075 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
| 3076 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
| 3077 | * much here because ->set_page_dirty is called under VFS locks. The page is |
| 3078 | * not necessarily locked. |
| 3079 | * |
| 3080 | * We cannot just dirty the page and leave attached buffers clean, because the |
| 3081 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty |
| 3082 | * or jbddirty because all the journalling code will explode. |
| 3083 | * |
| 3084 | * So what we do is to mark the page "pending dirty" and next time writepage |
| 3085 | * is called, propagate that into the buffers appropriately. |
| 3086 | */ |
| 3087 | static int ext4_journalled_set_page_dirty(struct page *page) |
| 3088 | { |
| 3089 | SetPageChecked(page); |
| 3090 | return __set_page_dirty_nobuffers(page); |
| 3091 | } |
| 3092 | |
| 3093 | static const struct address_space_operations ext4_ordered_aops = { |
| 3094 | .readpage = ext4_readpage, |
| 3095 | .readpages = ext4_readpages, |
| 3096 | .writepage = ext4_writepage, |
| 3097 | .write_begin = ext4_write_begin, |
| 3098 | .write_end = ext4_ordered_write_end, |
| 3099 | .bmap = ext4_bmap, |
| 3100 | .invalidatepage = ext4_invalidatepage, |
| 3101 | .releasepage = ext4_releasepage, |
| 3102 | .direct_IO = ext4_direct_IO, |
| 3103 | .migratepage = buffer_migrate_page, |
| 3104 | .is_partially_uptodate = block_is_partially_uptodate, |
| 3105 | .error_remove_page = generic_error_remove_page, |
| 3106 | }; |
| 3107 | |
| 3108 | static const struct address_space_operations ext4_writeback_aops = { |
| 3109 | .readpage = ext4_readpage, |
| 3110 | .readpages = ext4_readpages, |
| 3111 | .writepage = ext4_writepage, |
| 3112 | .write_begin = ext4_write_begin, |
| 3113 | .write_end = ext4_writeback_write_end, |
| 3114 | .bmap = ext4_bmap, |
| 3115 | .invalidatepage = ext4_invalidatepage, |
| 3116 | .releasepage = ext4_releasepage, |
| 3117 | .direct_IO = ext4_direct_IO, |
| 3118 | .migratepage = buffer_migrate_page, |
| 3119 | .is_partially_uptodate = block_is_partially_uptodate, |
| 3120 | .error_remove_page = generic_error_remove_page, |
| 3121 | }; |
| 3122 | |
| 3123 | static const struct address_space_operations ext4_journalled_aops = { |
| 3124 | .readpage = ext4_readpage, |
| 3125 | .readpages = ext4_readpages, |
| 3126 | .writepage = ext4_writepage, |
| 3127 | .write_begin = ext4_write_begin, |
| 3128 | .write_end = ext4_journalled_write_end, |
| 3129 | .set_page_dirty = ext4_journalled_set_page_dirty, |
| 3130 | .bmap = ext4_bmap, |
| 3131 | .invalidatepage = ext4_journalled_invalidatepage, |
| 3132 | .releasepage = ext4_releasepage, |
| 3133 | .direct_IO = ext4_direct_IO, |
| 3134 | .is_partially_uptodate = block_is_partially_uptodate, |
| 3135 | .error_remove_page = generic_error_remove_page, |
| 3136 | }; |
| 3137 | |
| 3138 | static const struct address_space_operations ext4_da_aops = { |
| 3139 | .readpage = ext4_readpage, |
| 3140 | .readpages = ext4_readpages, |
| 3141 | .writepage = ext4_writepage, |
| 3142 | .writepages = ext4_da_writepages, |
| 3143 | .write_begin = ext4_da_write_begin, |
| 3144 | .write_end = ext4_da_write_end, |
| 3145 | .bmap = ext4_bmap, |
| 3146 | .invalidatepage = ext4_da_invalidatepage, |
| 3147 | .releasepage = ext4_releasepage, |
| 3148 | .direct_IO = ext4_direct_IO, |
| 3149 | .migratepage = buffer_migrate_page, |
| 3150 | .is_partially_uptodate = block_is_partially_uptodate, |
| 3151 | .error_remove_page = generic_error_remove_page, |
| 3152 | }; |
| 3153 | |
| 3154 | void ext4_set_aops(struct inode *inode) |
| 3155 | { |
| 3156 | switch (ext4_inode_journal_mode(inode)) { |
| 3157 | case EXT4_INODE_ORDERED_DATA_MODE: |
| 3158 | if (test_opt(inode->i_sb, DELALLOC)) |
| 3159 | inode->i_mapping->a_ops = &ext4_da_aops; |
| 3160 | else |
| 3161 | inode->i_mapping->a_ops = &ext4_ordered_aops; |
| 3162 | break; |
| 3163 | case EXT4_INODE_WRITEBACK_DATA_MODE: |
| 3164 | if (test_opt(inode->i_sb, DELALLOC)) |
| 3165 | inode->i_mapping->a_ops = &ext4_da_aops; |
| 3166 | else |
| 3167 | inode->i_mapping->a_ops = &ext4_writeback_aops; |
| 3168 | break; |
| 3169 | case EXT4_INODE_JOURNAL_DATA_MODE: |
| 3170 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
| 3171 | break; |
| 3172 | default: |
| 3173 | BUG(); |
| 3174 | } |
| 3175 | } |
| 3176 | |
| 3177 | |
| 3178 | /* |
| 3179 | * ext4_discard_partial_page_buffers() |
| 3180 | * Wrapper function for ext4_discard_partial_page_buffers_no_lock. |
| 3181 | * This function finds and locks the page containing the offset |
| 3182 | * "from" and passes it to ext4_discard_partial_page_buffers_no_lock. |
| 3183 | * Calling functions that already have the page locked should call |
| 3184 | * ext4_discard_partial_page_buffers_no_lock directly. |
| 3185 | */ |
| 3186 | int ext4_discard_partial_page_buffers(handle_t *handle, |
| 3187 | struct address_space *mapping, loff_t from, |
| 3188 | loff_t length, int flags) |
| 3189 | { |
| 3190 | struct inode *inode = mapping->host; |
| 3191 | struct page *page; |
| 3192 | int err = 0; |
| 3193 | |
| 3194 | page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT, |
| 3195 | mapping_gfp_mask(mapping) & ~__GFP_FS); |
| 3196 | if (!page) |
| 3197 | return -ENOMEM; |
| 3198 | |
| 3199 | err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page, |
| 3200 | from, length, flags); |
| 3201 | |
| 3202 | unlock_page(page); |
| 3203 | page_cache_release(page); |
| 3204 | return err; |
| 3205 | } |
| 3206 | |
| 3207 | /* |
| 3208 | * ext4_discard_partial_page_buffers_no_lock() |
| 3209 | * Zeros a page range of length 'length' starting from offset 'from'. |
| 3210 | * Buffer heads that correspond to the block aligned regions of the |
| 3211 | * zeroed range will be unmapped. Unblock aligned regions |
| 3212 | * will have the corresponding buffer head mapped if needed so that |
| 3213 | * that region of the page can be updated with the partial zero out. |
| 3214 | * |
| 3215 | * This function assumes that the page has already been locked. The |
| 3216 | * The range to be discarded must be contained with in the given page. |
| 3217 | * If the specified range exceeds the end of the page it will be shortened |
| 3218 | * to the end of the page that corresponds to 'from'. This function is |
| 3219 | * appropriate for updating a page and it buffer heads to be unmapped and |
| 3220 | * zeroed for blocks that have been either released, or are going to be |
| 3221 | * released. |
| 3222 | * |
| 3223 | * handle: The journal handle |
| 3224 | * inode: The files inode |
| 3225 | * page: A locked page that contains the offset "from" |
| 3226 | * from: The starting byte offset (from the beginning of the file) |
| 3227 | * to begin discarding |
| 3228 | * len: The length of bytes to discard |
| 3229 | * flags: Optional flags that may be used: |
| 3230 | * |
| 3231 | * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED |
| 3232 | * Only zero the regions of the page whose buffer heads |
| 3233 | * have already been unmapped. This flag is appropriate |
| 3234 | * for updating the contents of a page whose blocks may |
| 3235 | * have already been released, and we only want to zero |
| 3236 | * out the regions that correspond to those released blocks. |
| 3237 | * |
| 3238 | * Returns zero on success or negative on failure. |
| 3239 | */ |
| 3240 | static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle, |
| 3241 | struct inode *inode, struct page *page, loff_t from, |
| 3242 | loff_t length, int flags) |
| 3243 | { |
| 3244 | ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT; |
| 3245 | unsigned int offset = from & (PAGE_CACHE_SIZE-1); |
| 3246 | unsigned int blocksize, max, pos; |
| 3247 | ext4_lblk_t iblock; |
| 3248 | struct buffer_head *bh; |
| 3249 | int err = 0; |
| 3250 | |
| 3251 | blocksize = inode->i_sb->s_blocksize; |
| 3252 | max = PAGE_CACHE_SIZE - offset; |
| 3253 | |
| 3254 | if (index != page->index) |
| 3255 | return -EINVAL; |
| 3256 | |
| 3257 | /* |
| 3258 | * correct length if it does not fall between |
| 3259 | * 'from' and the end of the page |
| 3260 | */ |
| 3261 | if (length > max || length < 0) |
| 3262 | length = max; |
| 3263 | |
| 3264 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); |
| 3265 | |
| 3266 | if (!page_has_buffers(page)) |
| 3267 | create_empty_buffers(page, blocksize, 0); |
| 3268 | |
| 3269 | /* Find the buffer that contains "offset" */ |
| 3270 | bh = page_buffers(page); |
| 3271 | pos = blocksize; |
| 3272 | while (offset >= pos) { |
| 3273 | bh = bh->b_this_page; |
| 3274 | iblock++; |
| 3275 | pos += blocksize; |
| 3276 | } |
| 3277 | |
| 3278 | pos = offset; |
| 3279 | while (pos < offset + length) { |
| 3280 | unsigned int end_of_block, range_to_discard; |
| 3281 | |
| 3282 | err = 0; |
| 3283 | |
| 3284 | /* The length of space left to zero and unmap */ |
| 3285 | range_to_discard = offset + length - pos; |
| 3286 | |
| 3287 | /* The length of space until the end of the block */ |
| 3288 | end_of_block = blocksize - (pos & (blocksize-1)); |
| 3289 | |
| 3290 | /* |
| 3291 | * Do not unmap or zero past end of block |
| 3292 | * for this buffer head |
| 3293 | */ |
| 3294 | if (range_to_discard > end_of_block) |
| 3295 | range_to_discard = end_of_block; |
| 3296 | |
| 3297 | |
| 3298 | /* |
| 3299 | * Skip this buffer head if we are only zeroing unampped |
| 3300 | * regions of the page |
| 3301 | */ |
| 3302 | if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED && |
| 3303 | buffer_mapped(bh)) |
| 3304 | goto next; |
| 3305 | |
| 3306 | /* If the range is block aligned, unmap */ |
| 3307 | if (range_to_discard == blocksize) { |
| 3308 | clear_buffer_dirty(bh); |
| 3309 | bh->b_bdev = NULL; |
| 3310 | clear_buffer_mapped(bh); |
| 3311 | clear_buffer_req(bh); |
| 3312 | clear_buffer_new(bh); |
| 3313 | clear_buffer_delay(bh); |
| 3314 | clear_buffer_unwritten(bh); |
| 3315 | clear_buffer_uptodate(bh); |
| 3316 | zero_user(page, pos, range_to_discard); |
| 3317 | BUFFER_TRACE(bh, "Buffer discarded"); |
| 3318 | goto next; |
| 3319 | } |
| 3320 | |
| 3321 | /* |
| 3322 | * If this block is not completely contained in the range |
| 3323 | * to be discarded, then it is not going to be released. Because |
| 3324 | * we need to keep this block, we need to make sure this part |
| 3325 | * of the page is uptodate before we modify it by writeing |
| 3326 | * partial zeros on it. |
| 3327 | */ |
| 3328 | if (!buffer_mapped(bh)) { |
| 3329 | /* |
| 3330 | * Buffer head must be mapped before we can read |
| 3331 | * from the block |
| 3332 | */ |
| 3333 | BUFFER_TRACE(bh, "unmapped"); |
| 3334 | ext4_get_block(inode, iblock, bh, 0); |
| 3335 | /* unmapped? It's a hole - nothing to do */ |
| 3336 | if (!buffer_mapped(bh)) { |
| 3337 | BUFFER_TRACE(bh, "still unmapped"); |
| 3338 | goto next; |
| 3339 | } |
| 3340 | } |
| 3341 | |
| 3342 | /* Ok, it's mapped. Make sure it's up-to-date */ |
| 3343 | if (PageUptodate(page)) |
| 3344 | set_buffer_uptodate(bh); |
| 3345 | |
| 3346 | if (!buffer_uptodate(bh)) { |
| 3347 | err = -EIO; |
| 3348 | ll_rw_block(READ, 1, &bh); |
| 3349 | wait_on_buffer(bh); |
| 3350 | /* Uhhuh. Read error. Complain and punt.*/ |
| 3351 | if (!buffer_uptodate(bh)) |
| 3352 | goto next; |
| 3353 | } |
| 3354 | |
| 3355 | if (ext4_should_journal_data(inode)) { |
| 3356 | BUFFER_TRACE(bh, "get write access"); |
| 3357 | err = ext4_journal_get_write_access(handle, bh); |
| 3358 | if (err) |
| 3359 | goto next; |
| 3360 | } |
| 3361 | |
| 3362 | zero_user(page, pos, range_to_discard); |
| 3363 | |
| 3364 | err = 0; |
| 3365 | if (ext4_should_journal_data(inode)) { |
| 3366 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
| 3367 | } else |
| 3368 | mark_buffer_dirty(bh); |
| 3369 | |
| 3370 | BUFFER_TRACE(bh, "Partial buffer zeroed"); |
| 3371 | next: |
| 3372 | bh = bh->b_this_page; |
| 3373 | iblock++; |
| 3374 | pos += range_to_discard; |
| 3375 | } |
| 3376 | |
| 3377 | return err; |
| 3378 | } |
| 3379 | |
| 3380 | int ext4_can_truncate(struct inode *inode) |
| 3381 | { |
| 3382 | if (S_ISREG(inode->i_mode)) |
| 3383 | return 1; |
| 3384 | if (S_ISDIR(inode->i_mode)) |
| 3385 | return 1; |
| 3386 | if (S_ISLNK(inode->i_mode)) |
| 3387 | return !ext4_inode_is_fast_symlink(inode); |
| 3388 | return 0; |
| 3389 | } |
| 3390 | |
| 3391 | /* |
| 3392 | * ext4_punch_hole: punches a hole in a file by releaseing the blocks |
| 3393 | * associated with the given offset and length |
| 3394 | * |
| 3395 | * @inode: File inode |
| 3396 | * @offset: The offset where the hole will begin |
| 3397 | * @len: The length of the hole |
| 3398 | * |
| 3399 | * Returns: 0 on success or negative on failure |
| 3400 | */ |
| 3401 | |
| 3402 | int ext4_punch_hole(struct file *file, loff_t offset, loff_t length) |
| 3403 | { |
| 3404 | struct inode *inode = file->f_path.dentry->d_inode; |
| 3405 | if (!S_ISREG(inode->i_mode)) |
| 3406 | return -EOPNOTSUPP; |
| 3407 | |
| 3408 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
| 3409 | return ext4_ind_punch_hole(file, offset, length); |
| 3410 | |
| 3411 | if (EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) { |
| 3412 | /* TODO: Add support for bigalloc file systems */ |
| 3413 | return -EOPNOTSUPP; |
| 3414 | } |
| 3415 | |
| 3416 | trace_ext4_punch_hole(inode, offset, length); |
| 3417 | |
| 3418 | return ext4_ext_punch_hole(file, offset, length); |
| 3419 | } |
| 3420 | |
| 3421 | /* |
| 3422 | * ext4_truncate() |
| 3423 | * |
| 3424 | * We block out ext4_get_block() block instantiations across the entire |
| 3425 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run |
| 3426 | * simultaneously on behalf of the same inode. |
| 3427 | * |
| 3428 | * As we work through the truncate and commit bits of it to the journal there |
| 3429 | * is one core, guiding principle: the file's tree must always be consistent on |
| 3430 | * disk. We must be able to restart the truncate after a crash. |
| 3431 | * |
| 3432 | * The file's tree may be transiently inconsistent in memory (although it |
| 3433 | * probably isn't), but whenever we close off and commit a journal transaction, |
| 3434 | * the contents of (the filesystem + the journal) must be consistent and |
| 3435 | * restartable. It's pretty simple, really: bottom up, right to left (although |
| 3436 | * left-to-right works OK too). |
| 3437 | * |
| 3438 | * Note that at recovery time, journal replay occurs *before* the restart of |
| 3439 | * truncate against the orphan inode list. |
| 3440 | * |
| 3441 | * The committed inode has the new, desired i_size (which is the same as |
| 3442 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
| 3443 | * that this inode's truncate did not complete and it will again call |
| 3444 | * ext4_truncate() to have another go. So there will be instantiated blocks |
| 3445 | * to the right of the truncation point in a crashed ext4 filesystem. But |
| 3446 | * that's fine - as long as they are linked from the inode, the post-crash |
| 3447 | * ext4_truncate() run will find them and release them. |
| 3448 | */ |
| 3449 | void ext4_truncate(struct inode *inode) |
| 3450 | { |
| 3451 | trace_ext4_truncate_enter(inode); |
| 3452 | |
| 3453 | if (!ext4_can_truncate(inode)) |
| 3454 | return; |
| 3455 | |
| 3456 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
| 3457 | |
| 3458 | if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) |
| 3459 | ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
| 3460 | |
| 3461 | if (ext4_has_inline_data(inode)) { |
| 3462 | int has_inline = 1; |
| 3463 | |
| 3464 | ext4_inline_data_truncate(inode, &has_inline); |
| 3465 | if (has_inline) |
| 3466 | return; |
| 3467 | } |
| 3468 | |
| 3469 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
| 3470 | ext4_ext_truncate(inode); |
| 3471 | else |
| 3472 | ext4_ind_truncate(inode); |
| 3473 | |
| 3474 | trace_ext4_truncate_exit(inode); |
| 3475 | } |
| 3476 | |
| 3477 | /* |
| 3478 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
| 3479 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
| 3480 | * data in memory that is needed to recreate the on-disk version of this |
| 3481 | * inode. |
| 3482 | */ |
| 3483 | static int __ext4_get_inode_loc(struct inode *inode, |
| 3484 | struct ext4_iloc *iloc, int in_mem) |
| 3485 | { |
| 3486 | struct ext4_group_desc *gdp; |
| 3487 | struct buffer_head *bh; |
| 3488 | struct super_block *sb = inode->i_sb; |
| 3489 | ext4_fsblk_t block; |
| 3490 | int inodes_per_block, inode_offset; |
| 3491 | |
| 3492 | iloc->bh = NULL; |
| 3493 | if (!ext4_valid_inum(sb, inode->i_ino)) |
| 3494 | return -EIO; |
| 3495 | |
| 3496 | iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb); |
| 3497 | gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); |
| 3498 | if (!gdp) |
| 3499 | return -EIO; |
| 3500 | |
| 3501 | /* |
| 3502 | * Figure out the offset within the block group inode table |
| 3503 | */ |
| 3504 | inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; |
| 3505 | inode_offset = ((inode->i_ino - 1) % |
| 3506 | EXT4_INODES_PER_GROUP(sb)); |
| 3507 | block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block); |
| 3508 | iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); |
| 3509 | |
| 3510 | bh = sb_getblk(sb, block); |
| 3511 | if (unlikely(!bh)) |
| 3512 | return -ENOMEM; |
| 3513 | if (!buffer_uptodate(bh)) { |
| 3514 | lock_buffer(bh); |
| 3515 | |
| 3516 | /* |
| 3517 | * If the buffer has the write error flag, we have failed |
| 3518 | * to write out another inode in the same block. In this |
| 3519 | * case, we don't have to read the block because we may |
| 3520 | * read the old inode data successfully. |
| 3521 | */ |
| 3522 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) |
| 3523 | set_buffer_uptodate(bh); |
| 3524 | |
| 3525 | if (buffer_uptodate(bh)) { |
| 3526 | /* someone brought it uptodate while we waited */ |
| 3527 | unlock_buffer(bh); |
| 3528 | goto has_buffer; |
| 3529 | } |
| 3530 | |
| 3531 | /* |
| 3532 | * If we have all information of the inode in memory and this |
| 3533 | * is the only valid inode in the block, we need not read the |
| 3534 | * block. |
| 3535 | */ |
| 3536 | if (in_mem) { |
| 3537 | struct buffer_head *bitmap_bh; |
| 3538 | int i, start; |
| 3539 | |
| 3540 | start = inode_offset & ~(inodes_per_block - 1); |
| 3541 | |
| 3542 | /* Is the inode bitmap in cache? */ |
| 3543 | bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); |
| 3544 | if (unlikely(!bitmap_bh)) |
| 3545 | goto make_io; |
| 3546 | |
| 3547 | /* |
| 3548 | * If the inode bitmap isn't in cache then the |
| 3549 | * optimisation may end up performing two reads instead |
| 3550 | * of one, so skip it. |
| 3551 | */ |
| 3552 | if (!buffer_uptodate(bitmap_bh)) { |
| 3553 | brelse(bitmap_bh); |
| 3554 | goto make_io; |
| 3555 | } |
| 3556 | for (i = start; i < start + inodes_per_block; i++) { |
| 3557 | if (i == inode_offset) |
| 3558 | continue; |
| 3559 | if (ext4_test_bit(i, bitmap_bh->b_data)) |
| 3560 | break; |
| 3561 | } |
| 3562 | brelse(bitmap_bh); |
| 3563 | if (i == start + inodes_per_block) { |
| 3564 | /* all other inodes are free, so skip I/O */ |
| 3565 | memset(bh->b_data, 0, bh->b_size); |
| 3566 | set_buffer_uptodate(bh); |
| 3567 | unlock_buffer(bh); |
| 3568 | goto has_buffer; |
| 3569 | } |
| 3570 | } |
| 3571 | |
| 3572 | make_io: |
| 3573 | /* |
| 3574 | * If we need to do any I/O, try to pre-readahead extra |
| 3575 | * blocks from the inode table. |
| 3576 | */ |
| 3577 | if (EXT4_SB(sb)->s_inode_readahead_blks) { |
| 3578 | ext4_fsblk_t b, end, table; |
| 3579 | unsigned num; |
| 3580 | |
| 3581 | table = ext4_inode_table(sb, gdp); |
| 3582 | /* s_inode_readahead_blks is always a power of 2 */ |
| 3583 | b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1); |
| 3584 | if (table > b) |
| 3585 | b = table; |
| 3586 | end = b + EXT4_SB(sb)->s_inode_readahead_blks; |
| 3587 | num = EXT4_INODES_PER_GROUP(sb); |
| 3588 | if (ext4_has_group_desc_csum(sb)) |
| 3589 | num -= ext4_itable_unused_count(sb, gdp); |
| 3590 | table += num / inodes_per_block; |
| 3591 | if (end > table) |
| 3592 | end = table; |
| 3593 | while (b <= end) |
| 3594 | sb_breadahead(sb, b++); |
| 3595 | } |
| 3596 | |
| 3597 | /* |
| 3598 | * There are other valid inodes in the buffer, this inode |
| 3599 | * has in-inode xattrs, or we don't have this inode in memory. |
| 3600 | * Read the block from disk. |
| 3601 | */ |
| 3602 | trace_ext4_load_inode(inode); |
| 3603 | get_bh(bh); |
| 3604 | bh->b_end_io = end_buffer_read_sync; |
| 3605 | submit_bh(READ | REQ_META | REQ_PRIO, bh); |
| 3606 | wait_on_buffer(bh); |
| 3607 | if (!buffer_uptodate(bh)) { |
| 3608 | EXT4_ERROR_INODE_BLOCK(inode, block, |
| 3609 | "unable to read itable block"); |
| 3610 | brelse(bh); |
| 3611 | return -EIO; |
| 3612 | } |
| 3613 | } |
| 3614 | has_buffer: |
| 3615 | iloc->bh = bh; |
| 3616 | return 0; |
| 3617 | } |
| 3618 | |
| 3619 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
| 3620 | { |
| 3621 | /* We have all inode data except xattrs in memory here. */ |
| 3622 | return __ext4_get_inode_loc(inode, iloc, |
| 3623 | !ext4_test_inode_state(inode, EXT4_STATE_XATTR)); |
| 3624 | } |
| 3625 | |
| 3626 | void ext4_set_inode_flags(struct inode *inode) |
| 3627 | { |
| 3628 | unsigned int flags = EXT4_I(inode)->i_flags; |
| 3629 | |
| 3630 | inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC); |
| 3631 | if (flags & EXT4_SYNC_FL) |
| 3632 | inode->i_flags |= S_SYNC; |
| 3633 | if (flags & EXT4_APPEND_FL) |
| 3634 | inode->i_flags |= S_APPEND; |
| 3635 | if (flags & EXT4_IMMUTABLE_FL) |
| 3636 | inode->i_flags |= S_IMMUTABLE; |
| 3637 | if (flags & EXT4_NOATIME_FL) |
| 3638 | inode->i_flags |= S_NOATIME; |
| 3639 | if (flags & EXT4_DIRSYNC_FL) |
| 3640 | inode->i_flags |= S_DIRSYNC; |
| 3641 | } |
| 3642 | |
| 3643 | /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */ |
| 3644 | void ext4_get_inode_flags(struct ext4_inode_info *ei) |
| 3645 | { |
| 3646 | unsigned int vfs_fl; |
| 3647 | unsigned long old_fl, new_fl; |
| 3648 | |
| 3649 | do { |
| 3650 | vfs_fl = ei->vfs_inode.i_flags; |
| 3651 | old_fl = ei->i_flags; |
| 3652 | new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL| |
| 3653 | EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL| |
| 3654 | EXT4_DIRSYNC_FL); |
| 3655 | if (vfs_fl & S_SYNC) |
| 3656 | new_fl |= EXT4_SYNC_FL; |
| 3657 | if (vfs_fl & S_APPEND) |
| 3658 | new_fl |= EXT4_APPEND_FL; |
| 3659 | if (vfs_fl & S_IMMUTABLE) |
| 3660 | new_fl |= EXT4_IMMUTABLE_FL; |
| 3661 | if (vfs_fl & S_NOATIME) |
| 3662 | new_fl |= EXT4_NOATIME_FL; |
| 3663 | if (vfs_fl & S_DIRSYNC) |
| 3664 | new_fl |= EXT4_DIRSYNC_FL; |
| 3665 | } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl); |
| 3666 | } |
| 3667 | |
| 3668 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
| 3669 | struct ext4_inode_info *ei) |
| 3670 | { |
| 3671 | blkcnt_t i_blocks ; |
| 3672 | struct inode *inode = &(ei->vfs_inode); |
| 3673 | struct super_block *sb = inode->i_sb; |
| 3674 | |
| 3675 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, |
| 3676 | EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) { |
| 3677 | /* we are using combined 48 bit field */ |
| 3678 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | |
| 3679 | le32_to_cpu(raw_inode->i_blocks_lo); |
| 3680 | if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { |
| 3681 | /* i_blocks represent file system block size */ |
| 3682 | return i_blocks << (inode->i_blkbits - 9); |
| 3683 | } else { |
| 3684 | return i_blocks; |
| 3685 | } |
| 3686 | } else { |
| 3687 | return le32_to_cpu(raw_inode->i_blocks_lo); |
| 3688 | } |
| 3689 | } |
| 3690 | |
| 3691 | static inline void ext4_iget_extra_inode(struct inode *inode, |
| 3692 | struct ext4_inode *raw_inode, |
| 3693 | struct ext4_inode_info *ei) |
| 3694 | { |
| 3695 | __le32 *magic = (void *)raw_inode + |
| 3696 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize; |
| 3697 | if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) { |
| 3698 | ext4_set_inode_state(inode, EXT4_STATE_XATTR); |
| 3699 | ext4_find_inline_data_nolock(inode); |
| 3700 | } else |
| 3701 | EXT4_I(inode)->i_inline_off = 0; |
| 3702 | } |
| 3703 | |
| 3704 | struct inode *ext4_iget(struct super_block *sb, unsigned long ino) |
| 3705 | { |
| 3706 | struct ext4_iloc iloc; |
| 3707 | struct ext4_inode *raw_inode; |
| 3708 | struct ext4_inode_info *ei; |
| 3709 | struct inode *inode; |
| 3710 | journal_t *journal = EXT4_SB(sb)->s_journal; |
| 3711 | long ret; |
| 3712 | int block; |
| 3713 | uid_t i_uid; |
| 3714 | gid_t i_gid; |
| 3715 | |
| 3716 | inode = iget_locked(sb, ino); |
| 3717 | if (!inode) |
| 3718 | return ERR_PTR(-ENOMEM); |
| 3719 | if (!(inode->i_state & I_NEW)) |
| 3720 | return inode; |
| 3721 | |
| 3722 | ei = EXT4_I(inode); |
| 3723 | iloc.bh = NULL; |
| 3724 | |
| 3725 | ret = __ext4_get_inode_loc(inode, &iloc, 0); |
| 3726 | if (ret < 0) |
| 3727 | goto bad_inode; |
| 3728 | raw_inode = ext4_raw_inode(&iloc); |
| 3729 | |
| 3730 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
| 3731 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); |
| 3732 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > |
| 3733 | EXT4_INODE_SIZE(inode->i_sb)) { |
| 3734 | EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)", |
| 3735 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize, |
| 3736 | EXT4_INODE_SIZE(inode->i_sb)); |
| 3737 | ret = -EIO; |
| 3738 | goto bad_inode; |
| 3739 | } |
| 3740 | } else |
| 3741 | ei->i_extra_isize = 0; |
| 3742 | |
| 3743 | /* Precompute checksum seed for inode metadata */ |
| 3744 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, |
| 3745 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) { |
| 3746 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 3747 | __u32 csum; |
| 3748 | __le32 inum = cpu_to_le32(inode->i_ino); |
| 3749 | __le32 gen = raw_inode->i_generation; |
| 3750 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, |
| 3751 | sizeof(inum)); |
| 3752 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, |
| 3753 | sizeof(gen)); |
| 3754 | } |
| 3755 | |
| 3756 | if (!ext4_inode_csum_verify(inode, raw_inode, ei)) { |
| 3757 | EXT4_ERROR_INODE(inode, "checksum invalid"); |
| 3758 | ret = -EIO; |
| 3759 | goto bad_inode; |
| 3760 | } |
| 3761 | |
| 3762 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
| 3763 | i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); |
| 3764 | i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); |
| 3765 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
| 3766 | i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
| 3767 | i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; |
| 3768 | } |
| 3769 | i_uid_write(inode, i_uid); |
| 3770 | i_gid_write(inode, i_gid); |
| 3771 | set_nlink(inode, le16_to_cpu(raw_inode->i_links_count)); |
| 3772 | |
| 3773 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
| 3774 | ei->i_inline_off = 0; |
| 3775 | ei->i_dir_start_lookup = 0; |
| 3776 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); |
| 3777 | /* We now have enough fields to check if the inode was active or not. |
| 3778 | * This is needed because nfsd might try to access dead inodes |
| 3779 | * the test is that same one that e2fsck uses |
| 3780 | * NeilBrown 1999oct15 |
| 3781 | */ |
| 3782 | if (inode->i_nlink == 0) { |
| 3783 | if (inode->i_mode == 0 || |
| 3784 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) { |
| 3785 | /* this inode is deleted */ |
| 3786 | ret = -ESTALE; |
| 3787 | goto bad_inode; |
| 3788 | } |
| 3789 | /* The only unlinked inodes we let through here have |
| 3790 | * valid i_mode and are being read by the orphan |
| 3791 | * recovery code: that's fine, we're about to complete |
| 3792 | * the process of deleting those. */ |
| 3793 | } |
| 3794 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
| 3795 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
| 3796 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
| 3797 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) |
| 3798 | ei->i_file_acl |= |
| 3799 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; |
| 3800 | inode->i_size = ext4_isize(raw_inode); |
| 3801 | ei->i_disksize = inode->i_size; |
| 3802 | #ifdef CONFIG_QUOTA |
| 3803 | ei->i_reserved_quota = 0; |
| 3804 | #endif |
| 3805 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
| 3806 | ei->i_block_group = iloc.block_group; |
| 3807 | ei->i_last_alloc_group = ~0; |
| 3808 | /* |
| 3809 | * NOTE! The in-memory inode i_data array is in little-endian order |
| 3810 | * even on big-endian machines: we do NOT byteswap the block numbers! |
| 3811 | */ |
| 3812 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
| 3813 | ei->i_data[block] = raw_inode->i_block[block]; |
| 3814 | INIT_LIST_HEAD(&ei->i_orphan); |
| 3815 | |
| 3816 | /* |
| 3817 | * Set transaction id's of transactions that have to be committed |
| 3818 | * to finish f[data]sync. We set them to currently running transaction |
| 3819 | * as we cannot be sure that the inode or some of its metadata isn't |
| 3820 | * part of the transaction - the inode could have been reclaimed and |
| 3821 | * now it is reread from disk. |
| 3822 | */ |
| 3823 | if (journal) { |
| 3824 | transaction_t *transaction; |
| 3825 | tid_t tid; |
| 3826 | |
| 3827 | read_lock(&journal->j_state_lock); |
| 3828 | if (journal->j_running_transaction) |
| 3829 | transaction = journal->j_running_transaction; |
| 3830 | else |
| 3831 | transaction = journal->j_committing_transaction; |
| 3832 | if (transaction) |
| 3833 | tid = transaction->t_tid; |
| 3834 | else |
| 3835 | tid = journal->j_commit_sequence; |
| 3836 | read_unlock(&journal->j_state_lock); |
| 3837 | ei->i_sync_tid = tid; |
| 3838 | ei->i_datasync_tid = tid; |
| 3839 | } |
| 3840 | |
| 3841 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
| 3842 | if (ei->i_extra_isize == 0) { |
| 3843 | /* The extra space is currently unused. Use it. */ |
| 3844 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
| 3845 | EXT4_GOOD_OLD_INODE_SIZE; |
| 3846 | } else { |
| 3847 | ext4_iget_extra_inode(inode, raw_inode, ei); |
| 3848 | } |
| 3849 | } |
| 3850 | |
| 3851 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
| 3852 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); |
| 3853 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); |
| 3854 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); |
| 3855 | |
| 3856 | inode->i_version = le32_to_cpu(raw_inode->i_disk_version); |
| 3857 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
| 3858 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) |
| 3859 | inode->i_version |= |
| 3860 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; |
| 3861 | } |
| 3862 | |
| 3863 | ret = 0; |
| 3864 | if (ei->i_file_acl && |
| 3865 | !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) { |
| 3866 | EXT4_ERROR_INODE(inode, "bad extended attribute block %llu", |
| 3867 | ei->i_file_acl); |
| 3868 | ret = -EIO; |
| 3869 | goto bad_inode; |
| 3870 | } else if (!ext4_has_inline_data(inode)) { |
| 3871 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
| 3872 | if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || |
| 3873 | (S_ISLNK(inode->i_mode) && |
| 3874 | !ext4_inode_is_fast_symlink(inode)))) |
| 3875 | /* Validate extent which is part of inode */ |
| 3876 | ret = ext4_ext_check_inode(inode); |
| 3877 | } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || |
| 3878 | (S_ISLNK(inode->i_mode) && |
| 3879 | !ext4_inode_is_fast_symlink(inode))) { |
| 3880 | /* Validate block references which are part of inode */ |
| 3881 | ret = ext4_ind_check_inode(inode); |
| 3882 | } |
| 3883 | } |
| 3884 | if (ret) |
| 3885 | goto bad_inode; |
| 3886 | |
| 3887 | if (S_ISREG(inode->i_mode)) { |
| 3888 | inode->i_op = &ext4_file_inode_operations; |
| 3889 | inode->i_fop = &ext4_file_operations; |
| 3890 | ext4_set_aops(inode); |
| 3891 | } else if (S_ISDIR(inode->i_mode)) { |
| 3892 | inode->i_op = &ext4_dir_inode_operations; |
| 3893 | inode->i_fop = &ext4_dir_operations; |
| 3894 | } else if (S_ISLNK(inode->i_mode)) { |
| 3895 | if (ext4_inode_is_fast_symlink(inode)) { |
| 3896 | inode->i_op = &ext4_fast_symlink_inode_operations; |
| 3897 | nd_terminate_link(ei->i_data, inode->i_size, |
| 3898 | sizeof(ei->i_data) - 1); |
| 3899 | } else { |
| 3900 | inode->i_op = &ext4_symlink_inode_operations; |
| 3901 | ext4_set_aops(inode); |
| 3902 | } |
| 3903 | } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
| 3904 | S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { |
| 3905 | inode->i_op = &ext4_special_inode_operations; |
| 3906 | if (raw_inode->i_block[0]) |
| 3907 | init_special_inode(inode, inode->i_mode, |
| 3908 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); |
| 3909 | else |
| 3910 | init_special_inode(inode, inode->i_mode, |
| 3911 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); |
| 3912 | } else { |
| 3913 | ret = -EIO; |
| 3914 | EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode); |
| 3915 | goto bad_inode; |
| 3916 | } |
| 3917 | brelse(iloc.bh); |
| 3918 | ext4_set_inode_flags(inode); |
| 3919 | unlock_new_inode(inode); |
| 3920 | return inode; |
| 3921 | |
| 3922 | bad_inode: |
| 3923 | brelse(iloc.bh); |
| 3924 | iget_failed(inode); |
| 3925 | return ERR_PTR(ret); |
| 3926 | } |
| 3927 | |
| 3928 | static int ext4_inode_blocks_set(handle_t *handle, |
| 3929 | struct ext4_inode *raw_inode, |
| 3930 | struct ext4_inode_info *ei) |
| 3931 | { |
| 3932 | struct inode *inode = &(ei->vfs_inode); |
| 3933 | u64 i_blocks = inode->i_blocks; |
| 3934 | struct super_block *sb = inode->i_sb; |
| 3935 | |
| 3936 | if (i_blocks <= ~0U) { |
| 3937 | /* |
| 3938 | * i_blocks can be represented in a 32 bit variable |
| 3939 | * as multiple of 512 bytes |
| 3940 | */ |
| 3941 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
| 3942 | raw_inode->i_blocks_high = 0; |
| 3943 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
| 3944 | return 0; |
| 3945 | } |
| 3946 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) |
| 3947 | return -EFBIG; |
| 3948 | |
| 3949 | if (i_blocks <= 0xffffffffffffULL) { |
| 3950 | /* |
| 3951 | * i_blocks can be represented in a 48 bit variable |
| 3952 | * as multiple of 512 bytes |
| 3953 | */ |
| 3954 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
| 3955 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
| 3956 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
| 3957 | } else { |
| 3958 | ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
| 3959 | /* i_block is stored in file system block size */ |
| 3960 | i_blocks = i_blocks >> (inode->i_blkbits - 9); |
| 3961 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
| 3962 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
| 3963 | } |
| 3964 | return 0; |
| 3965 | } |
| 3966 | |
| 3967 | /* |
| 3968 | * Post the struct inode info into an on-disk inode location in the |
| 3969 | * buffer-cache. This gobbles the caller's reference to the |
| 3970 | * buffer_head in the inode location struct. |
| 3971 | * |
| 3972 | * The caller must have write access to iloc->bh. |
| 3973 | */ |
| 3974 | static int ext4_do_update_inode(handle_t *handle, |
| 3975 | struct inode *inode, |
| 3976 | struct ext4_iloc *iloc) |
| 3977 | { |
| 3978 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
| 3979 | struct ext4_inode_info *ei = EXT4_I(inode); |
| 3980 | struct buffer_head *bh = iloc->bh; |
| 3981 | int err = 0, rc, block; |
| 3982 | int need_datasync = 0; |
| 3983 | uid_t i_uid; |
| 3984 | gid_t i_gid; |
| 3985 | |
| 3986 | /* For fields not not tracking in the in-memory inode, |
| 3987 | * initialise them to zero for new inodes. */ |
| 3988 | if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) |
| 3989 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); |
| 3990 | |
| 3991 | ext4_get_inode_flags(ei); |
| 3992 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
| 3993 | i_uid = i_uid_read(inode); |
| 3994 | i_gid = i_gid_read(inode); |
| 3995 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
| 3996 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid)); |
| 3997 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid)); |
| 3998 | /* |
| 3999 | * Fix up interoperability with old kernels. Otherwise, old inodes get |
| 4000 | * re-used with the upper 16 bits of the uid/gid intact |
| 4001 | */ |
| 4002 | if (!ei->i_dtime) { |
| 4003 | raw_inode->i_uid_high = |
| 4004 | cpu_to_le16(high_16_bits(i_uid)); |
| 4005 | raw_inode->i_gid_high = |
| 4006 | cpu_to_le16(high_16_bits(i_gid)); |
| 4007 | } else { |
| 4008 | raw_inode->i_uid_high = 0; |
| 4009 | raw_inode->i_gid_high = 0; |
| 4010 | } |
| 4011 | } else { |
| 4012 | raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid)); |
| 4013 | raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid)); |
| 4014 | raw_inode->i_uid_high = 0; |
| 4015 | raw_inode->i_gid_high = 0; |
| 4016 | } |
| 4017 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); |
| 4018 | |
| 4019 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); |
| 4020 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); |
| 4021 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); |
| 4022 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); |
| 4023 | |
| 4024 | if (ext4_inode_blocks_set(handle, raw_inode, ei)) |
| 4025 | goto out_brelse; |
| 4026 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
| 4027 | raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF); |
| 4028 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != |
| 4029 | cpu_to_le32(EXT4_OS_HURD)) |
| 4030 | raw_inode->i_file_acl_high = |
| 4031 | cpu_to_le16(ei->i_file_acl >> 32); |
| 4032 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
| 4033 | if (ei->i_disksize != ext4_isize(raw_inode)) { |
| 4034 | ext4_isize_set(raw_inode, ei->i_disksize); |
| 4035 | need_datasync = 1; |
| 4036 | } |
| 4037 | if (ei->i_disksize > 0x7fffffffULL) { |
| 4038 | struct super_block *sb = inode->i_sb; |
| 4039 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, |
| 4040 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE) || |
| 4041 | EXT4_SB(sb)->s_es->s_rev_level == |
| 4042 | cpu_to_le32(EXT4_GOOD_OLD_REV)) { |
| 4043 | /* If this is the first large file |
| 4044 | * created, add a flag to the superblock. |
| 4045 | */ |
| 4046 | err = ext4_journal_get_write_access(handle, |
| 4047 | EXT4_SB(sb)->s_sbh); |
| 4048 | if (err) |
| 4049 | goto out_brelse; |
| 4050 | ext4_update_dynamic_rev(sb); |
| 4051 | EXT4_SET_RO_COMPAT_FEATURE(sb, |
| 4052 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE); |
| 4053 | ext4_handle_sync(handle); |
| 4054 | err = ext4_handle_dirty_super(handle, sb); |
| 4055 | } |
| 4056 | } |
| 4057 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); |
| 4058 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { |
| 4059 | if (old_valid_dev(inode->i_rdev)) { |
| 4060 | raw_inode->i_block[0] = |
| 4061 | cpu_to_le32(old_encode_dev(inode->i_rdev)); |
| 4062 | raw_inode->i_block[1] = 0; |
| 4063 | } else { |
| 4064 | raw_inode->i_block[0] = 0; |
| 4065 | raw_inode->i_block[1] = |
| 4066 | cpu_to_le32(new_encode_dev(inode->i_rdev)); |
| 4067 | raw_inode->i_block[2] = 0; |
| 4068 | } |
| 4069 | } else if (!ext4_has_inline_data(inode)) { |
| 4070 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
| 4071 | raw_inode->i_block[block] = ei->i_data[block]; |
| 4072 | } |
| 4073 | |
| 4074 | raw_inode->i_disk_version = cpu_to_le32(inode->i_version); |
| 4075 | if (ei->i_extra_isize) { |
| 4076 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) |
| 4077 | raw_inode->i_version_hi = |
| 4078 | cpu_to_le32(inode->i_version >> 32); |
| 4079 | raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize); |
| 4080 | } |
| 4081 | |
| 4082 | ext4_inode_csum_set(inode, raw_inode, ei); |
| 4083 | |
| 4084 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
| 4085 | rc = ext4_handle_dirty_metadata(handle, NULL, bh); |
| 4086 | if (!err) |
| 4087 | err = rc; |
| 4088 | ext4_clear_inode_state(inode, EXT4_STATE_NEW); |
| 4089 | |
| 4090 | ext4_update_inode_fsync_trans(handle, inode, need_datasync); |
| 4091 | out_brelse: |
| 4092 | brelse(bh); |
| 4093 | ext4_std_error(inode->i_sb, err); |
| 4094 | return err; |
| 4095 | } |
| 4096 | |
| 4097 | /* |
| 4098 | * ext4_write_inode() |
| 4099 | * |
| 4100 | * We are called from a few places: |
| 4101 | * |
| 4102 | * - Within generic_file_write() for O_SYNC files. |
| 4103 | * Here, there will be no transaction running. We wait for any running |
| 4104 | * transaction to commit. |
| 4105 | * |
| 4106 | * - Within sys_sync(), kupdate and such. |
| 4107 | * We wait on commit, if tol to. |
| 4108 | * |
| 4109 | * - Within prune_icache() (PF_MEMALLOC == true) |
| 4110 | * Here we simply return. We can't afford to block kswapd on the |
| 4111 | * journal commit. |
| 4112 | * |
| 4113 | * In all cases it is actually safe for us to return without doing anything, |
| 4114 | * because the inode has been copied into a raw inode buffer in |
| 4115 | * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for |
| 4116 | * knfsd. |
| 4117 | * |
| 4118 | * Note that we are absolutely dependent upon all inode dirtiers doing the |
| 4119 | * right thing: they *must* call mark_inode_dirty() after dirtying info in |
| 4120 | * which we are interested. |
| 4121 | * |
| 4122 | * It would be a bug for them to not do this. The code: |
| 4123 | * |
| 4124 | * mark_inode_dirty(inode) |
| 4125 | * stuff(); |
| 4126 | * inode->i_size = expr; |
| 4127 | * |
| 4128 | * is in error because a kswapd-driven write_inode() could occur while |
| 4129 | * `stuff()' is running, and the new i_size will be lost. Plus the inode |
| 4130 | * will no longer be on the superblock's dirty inode list. |
| 4131 | */ |
| 4132 | int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) |
| 4133 | { |
| 4134 | int err; |
| 4135 | |
| 4136 | if (current->flags & PF_MEMALLOC) |
| 4137 | return 0; |
| 4138 | |
| 4139 | if (EXT4_SB(inode->i_sb)->s_journal) { |
| 4140 | if (ext4_journal_current_handle()) { |
| 4141 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); |
| 4142 | dump_stack(); |
| 4143 | return -EIO; |
| 4144 | } |
| 4145 | |
| 4146 | if (wbc->sync_mode != WB_SYNC_ALL) |
| 4147 | return 0; |
| 4148 | |
| 4149 | err = ext4_force_commit(inode->i_sb); |
| 4150 | } else { |
| 4151 | struct ext4_iloc iloc; |
| 4152 | |
| 4153 | err = __ext4_get_inode_loc(inode, &iloc, 0); |
| 4154 | if (err) |
| 4155 | return err; |
| 4156 | if (wbc->sync_mode == WB_SYNC_ALL) |
| 4157 | sync_dirty_buffer(iloc.bh); |
| 4158 | if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { |
| 4159 | EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr, |
| 4160 | "IO error syncing inode"); |
| 4161 | err = -EIO; |
| 4162 | } |
| 4163 | brelse(iloc.bh); |
| 4164 | } |
| 4165 | return err; |
| 4166 | } |
| 4167 | |
| 4168 | /* |
| 4169 | * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate |
| 4170 | * buffers that are attached to a page stradding i_size and are undergoing |
| 4171 | * commit. In that case we have to wait for commit to finish and try again. |
| 4172 | */ |
| 4173 | static void ext4_wait_for_tail_page_commit(struct inode *inode) |
| 4174 | { |
| 4175 | struct page *page; |
| 4176 | unsigned offset; |
| 4177 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; |
| 4178 | tid_t commit_tid = 0; |
| 4179 | int ret; |
| 4180 | |
| 4181 | offset = inode->i_size & (PAGE_CACHE_SIZE - 1); |
| 4182 | /* |
| 4183 | * All buffers in the last page remain valid? Then there's nothing to |
| 4184 | * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE == |
| 4185 | * blocksize case |
| 4186 | */ |
| 4187 | if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits)) |
| 4188 | return; |
| 4189 | while (1) { |
| 4190 | page = find_lock_page(inode->i_mapping, |
| 4191 | inode->i_size >> PAGE_CACHE_SHIFT); |
| 4192 | if (!page) |
| 4193 | return; |
| 4194 | ret = __ext4_journalled_invalidatepage(page, offset); |
| 4195 | unlock_page(page); |
| 4196 | page_cache_release(page); |
| 4197 | if (ret != -EBUSY) |
| 4198 | return; |
| 4199 | commit_tid = 0; |
| 4200 | read_lock(&journal->j_state_lock); |
| 4201 | if (journal->j_committing_transaction) |
| 4202 | commit_tid = journal->j_committing_transaction->t_tid; |
| 4203 | read_unlock(&journal->j_state_lock); |
| 4204 | if (commit_tid) |
| 4205 | jbd2_log_wait_commit(journal, commit_tid); |
| 4206 | } |
| 4207 | } |
| 4208 | |
| 4209 | /* |
| 4210 | * ext4_setattr() |
| 4211 | * |
| 4212 | * Called from notify_change. |
| 4213 | * |
| 4214 | * We want to trap VFS attempts to truncate the file as soon as |
| 4215 | * possible. In particular, we want to make sure that when the VFS |
| 4216 | * shrinks i_size, we put the inode on the orphan list and modify |
| 4217 | * i_disksize immediately, so that during the subsequent flushing of |
| 4218 | * dirty pages and freeing of disk blocks, we can guarantee that any |
| 4219 | * commit will leave the blocks being flushed in an unused state on |
| 4220 | * disk. (On recovery, the inode will get truncated and the blocks will |
| 4221 | * be freed, so we have a strong guarantee that no future commit will |
| 4222 | * leave these blocks visible to the user.) |
| 4223 | * |
| 4224 | * Another thing we have to assure is that if we are in ordered mode |
| 4225 | * and inode is still attached to the committing transaction, we must |
| 4226 | * we start writeout of all the dirty pages which are being truncated. |
| 4227 | * This way we are sure that all the data written in the previous |
| 4228 | * transaction are already on disk (truncate waits for pages under |
| 4229 | * writeback). |
| 4230 | * |
| 4231 | * Called with inode->i_mutex down. |
| 4232 | */ |
| 4233 | int ext4_setattr(struct dentry *dentry, struct iattr *attr) |
| 4234 | { |
| 4235 | struct inode *inode = dentry->d_inode; |
| 4236 | int error, rc = 0; |
| 4237 | int orphan = 0; |
| 4238 | const unsigned int ia_valid = attr->ia_valid; |
| 4239 | |
| 4240 | error = inode_change_ok(inode, attr); |
| 4241 | if (error) |
| 4242 | return error; |
| 4243 | |
| 4244 | if (is_quota_modification(inode, attr)) |
| 4245 | dquot_initialize(inode); |
| 4246 | if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) || |
| 4247 | (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) { |
| 4248 | handle_t *handle; |
| 4249 | |
| 4250 | /* (user+group)*(old+new) structure, inode write (sb, |
| 4251 | * inode block, ? - but truncate inode update has it) */ |
| 4252 | handle = ext4_journal_start(inode, EXT4_HT_QUOTA, |
| 4253 | (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) + |
| 4254 | EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3); |
| 4255 | if (IS_ERR(handle)) { |
| 4256 | error = PTR_ERR(handle); |
| 4257 | goto err_out; |
| 4258 | } |
| 4259 | error = dquot_transfer(inode, attr); |
| 4260 | if (error) { |
| 4261 | ext4_journal_stop(handle); |
| 4262 | return error; |
| 4263 | } |
| 4264 | /* Update corresponding info in inode so that everything is in |
| 4265 | * one transaction */ |
| 4266 | if (attr->ia_valid & ATTR_UID) |
| 4267 | inode->i_uid = attr->ia_uid; |
| 4268 | if (attr->ia_valid & ATTR_GID) |
| 4269 | inode->i_gid = attr->ia_gid; |
| 4270 | error = ext4_mark_inode_dirty(handle, inode); |
| 4271 | ext4_journal_stop(handle); |
| 4272 | } |
| 4273 | |
| 4274 | if (attr->ia_valid & ATTR_SIZE) { |
| 4275 | |
| 4276 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
| 4277 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 4278 | |
| 4279 | if (attr->ia_size > sbi->s_bitmap_maxbytes) |
| 4280 | return -EFBIG; |
| 4281 | } |
| 4282 | } |
| 4283 | |
| 4284 | if (S_ISREG(inode->i_mode) && |
| 4285 | attr->ia_valid & ATTR_SIZE && |
| 4286 | (attr->ia_size < inode->i_size)) { |
| 4287 | handle_t *handle; |
| 4288 | |
| 4289 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 3); |
| 4290 | if (IS_ERR(handle)) { |
| 4291 | error = PTR_ERR(handle); |
| 4292 | goto err_out; |
| 4293 | } |
| 4294 | if (ext4_handle_valid(handle)) { |
| 4295 | error = ext4_orphan_add(handle, inode); |
| 4296 | orphan = 1; |
| 4297 | } |
| 4298 | EXT4_I(inode)->i_disksize = attr->ia_size; |
| 4299 | rc = ext4_mark_inode_dirty(handle, inode); |
| 4300 | if (!error) |
| 4301 | error = rc; |
| 4302 | ext4_journal_stop(handle); |
| 4303 | |
| 4304 | if (ext4_should_order_data(inode)) { |
| 4305 | error = ext4_begin_ordered_truncate(inode, |
| 4306 | attr->ia_size); |
| 4307 | if (error) { |
| 4308 | /* Do as much error cleanup as possible */ |
| 4309 | handle = ext4_journal_start(inode, |
| 4310 | EXT4_HT_INODE, 3); |
| 4311 | if (IS_ERR(handle)) { |
| 4312 | ext4_orphan_del(NULL, inode); |
| 4313 | goto err_out; |
| 4314 | } |
| 4315 | ext4_orphan_del(handle, inode); |
| 4316 | orphan = 0; |
| 4317 | ext4_journal_stop(handle); |
| 4318 | goto err_out; |
| 4319 | } |
| 4320 | } |
| 4321 | } |
| 4322 | |
| 4323 | if (attr->ia_valid & ATTR_SIZE) { |
| 4324 | if (attr->ia_size != inode->i_size) { |
| 4325 | loff_t oldsize = inode->i_size; |
| 4326 | |
| 4327 | i_size_write(inode, attr->ia_size); |
| 4328 | /* |
| 4329 | * Blocks are going to be removed from the inode. Wait |
| 4330 | * for dio in flight. Temporarily disable |
| 4331 | * dioread_nolock to prevent livelock. |
| 4332 | */ |
| 4333 | if (orphan) { |
| 4334 | if (!ext4_should_journal_data(inode)) { |
| 4335 | ext4_inode_block_unlocked_dio(inode); |
| 4336 | inode_dio_wait(inode); |
| 4337 | ext4_inode_resume_unlocked_dio(inode); |
| 4338 | } else |
| 4339 | ext4_wait_for_tail_page_commit(inode); |
| 4340 | } |
| 4341 | /* |
| 4342 | * Truncate pagecache after we've waited for commit |
| 4343 | * in data=journal mode to make pages freeable. |
| 4344 | */ |
| 4345 | truncate_pagecache(inode, oldsize, inode->i_size); |
| 4346 | } |
| 4347 | ext4_truncate(inode); |
| 4348 | } |
| 4349 | |
| 4350 | if (!rc) { |
| 4351 | setattr_copy(inode, attr); |
| 4352 | mark_inode_dirty(inode); |
| 4353 | } |
| 4354 | |
| 4355 | /* |
| 4356 | * If the call to ext4_truncate failed to get a transaction handle at |
| 4357 | * all, we need to clean up the in-core orphan list manually. |
| 4358 | */ |
| 4359 | if (orphan && inode->i_nlink) |
| 4360 | ext4_orphan_del(NULL, inode); |
| 4361 | |
| 4362 | if (!rc && (ia_valid & ATTR_MODE)) |
| 4363 | rc = ext4_acl_chmod(inode); |
| 4364 | |
| 4365 | err_out: |
| 4366 | ext4_std_error(inode->i_sb, error); |
| 4367 | if (!error) |
| 4368 | error = rc; |
| 4369 | return error; |
| 4370 | } |
| 4371 | |
| 4372 | int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry, |
| 4373 | struct kstat *stat) |
| 4374 | { |
| 4375 | struct inode *inode; |
| 4376 | unsigned long delalloc_blocks; |
| 4377 | |
| 4378 | inode = dentry->d_inode; |
| 4379 | generic_fillattr(inode, stat); |
| 4380 | |
| 4381 | /* |
| 4382 | * We can't update i_blocks if the block allocation is delayed |
| 4383 | * otherwise in the case of system crash before the real block |
| 4384 | * allocation is done, we will have i_blocks inconsistent with |
| 4385 | * on-disk file blocks. |
| 4386 | * We always keep i_blocks updated together with real |
| 4387 | * allocation. But to not confuse with user, stat |
| 4388 | * will return the blocks that include the delayed allocation |
| 4389 | * blocks for this file. |
| 4390 | */ |
| 4391 | delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb), |
| 4392 | EXT4_I(inode)->i_reserved_data_blocks); |
| 4393 | |
| 4394 | stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9; |
| 4395 | return 0; |
| 4396 | } |
| 4397 | |
| 4398 | static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk) |
| 4399 | { |
| 4400 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
| 4401 | return ext4_ind_trans_blocks(inode, nrblocks, chunk); |
| 4402 | return ext4_ext_index_trans_blocks(inode, nrblocks, chunk); |
| 4403 | } |
| 4404 | |
| 4405 | /* |
| 4406 | * Account for index blocks, block groups bitmaps and block group |
| 4407 | * descriptor blocks if modify datablocks and index blocks |
| 4408 | * worse case, the indexs blocks spread over different block groups |
| 4409 | * |
| 4410 | * If datablocks are discontiguous, they are possible to spread over |
| 4411 | * different block groups too. If they are contiguous, with flexbg, |
| 4412 | * they could still across block group boundary. |
| 4413 | * |
| 4414 | * Also account for superblock, inode, quota and xattr blocks |
| 4415 | */ |
| 4416 | static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk) |
| 4417 | { |
| 4418 | ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
| 4419 | int gdpblocks; |
| 4420 | int idxblocks; |
| 4421 | int ret = 0; |
| 4422 | |
| 4423 | /* |
| 4424 | * How many index blocks need to touch to modify nrblocks? |
| 4425 | * The "Chunk" flag indicating whether the nrblocks is |
| 4426 | * physically contiguous on disk |
| 4427 | * |
| 4428 | * For Direct IO and fallocate, they calls get_block to allocate |
| 4429 | * one single extent at a time, so they could set the "Chunk" flag |
| 4430 | */ |
| 4431 | idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk); |
| 4432 | |
| 4433 | ret = idxblocks; |
| 4434 | |
| 4435 | /* |
| 4436 | * Now let's see how many group bitmaps and group descriptors need |
| 4437 | * to account |
| 4438 | */ |
| 4439 | groups = idxblocks; |
| 4440 | if (chunk) |
| 4441 | groups += 1; |
| 4442 | else |
| 4443 | groups += nrblocks; |
| 4444 | |
| 4445 | gdpblocks = groups; |
| 4446 | if (groups > ngroups) |
| 4447 | groups = ngroups; |
| 4448 | if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
| 4449 | gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; |
| 4450 | |
| 4451 | /* bitmaps and block group descriptor blocks */ |
| 4452 | ret += groups + gdpblocks; |
| 4453 | |
| 4454 | /* Blocks for super block, inode, quota and xattr blocks */ |
| 4455 | ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); |
| 4456 | |
| 4457 | return ret; |
| 4458 | } |
| 4459 | |
| 4460 | /* |
| 4461 | * Calculate the total number of credits to reserve to fit |
| 4462 | * the modification of a single pages into a single transaction, |
| 4463 | * which may include multiple chunks of block allocations. |
| 4464 | * |
| 4465 | * This could be called via ext4_write_begin() |
| 4466 | * |
| 4467 | * We need to consider the worse case, when |
| 4468 | * one new block per extent. |
| 4469 | */ |
| 4470 | int ext4_writepage_trans_blocks(struct inode *inode) |
| 4471 | { |
| 4472 | int bpp = ext4_journal_blocks_per_page(inode); |
| 4473 | int ret; |
| 4474 | |
| 4475 | ret = ext4_meta_trans_blocks(inode, bpp, 0); |
| 4476 | |
| 4477 | /* Account for data blocks for journalled mode */ |
| 4478 | if (ext4_should_journal_data(inode)) |
| 4479 | ret += bpp; |
| 4480 | return ret; |
| 4481 | } |
| 4482 | |
| 4483 | /* |
| 4484 | * Calculate the journal credits for a chunk of data modification. |
| 4485 | * |
| 4486 | * This is called from DIO, fallocate or whoever calling |
| 4487 | * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. |
| 4488 | * |
| 4489 | * journal buffers for data blocks are not included here, as DIO |
| 4490 | * and fallocate do no need to journal data buffers. |
| 4491 | */ |
| 4492 | int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) |
| 4493 | { |
| 4494 | return ext4_meta_trans_blocks(inode, nrblocks, 1); |
| 4495 | } |
| 4496 | |
| 4497 | /* |
| 4498 | * The caller must have previously called ext4_reserve_inode_write(). |
| 4499 | * Give this, we know that the caller already has write access to iloc->bh. |
| 4500 | */ |
| 4501 | int ext4_mark_iloc_dirty(handle_t *handle, |
| 4502 | struct inode *inode, struct ext4_iloc *iloc) |
| 4503 | { |
| 4504 | int err = 0; |
| 4505 | |
| 4506 | if (IS_I_VERSION(inode)) |
| 4507 | inode_inc_iversion(inode); |
| 4508 | |
| 4509 | /* the do_update_inode consumes one bh->b_count */ |
| 4510 | get_bh(iloc->bh); |
| 4511 | |
| 4512 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
| 4513 | err = ext4_do_update_inode(handle, inode, iloc); |
| 4514 | put_bh(iloc->bh); |
| 4515 | return err; |
| 4516 | } |
| 4517 | |
| 4518 | /* |
| 4519 | * On success, We end up with an outstanding reference count against |
| 4520 | * iloc->bh. This _must_ be cleaned up later. |
| 4521 | */ |
| 4522 | |
| 4523 | int |
| 4524 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
| 4525 | struct ext4_iloc *iloc) |
| 4526 | { |
| 4527 | int err; |
| 4528 | |
| 4529 | err = ext4_get_inode_loc(inode, iloc); |
| 4530 | if (!err) { |
| 4531 | BUFFER_TRACE(iloc->bh, "get_write_access"); |
| 4532 | err = ext4_journal_get_write_access(handle, iloc->bh); |
| 4533 | if (err) { |
| 4534 | brelse(iloc->bh); |
| 4535 | iloc->bh = NULL; |
| 4536 | } |
| 4537 | } |
| 4538 | ext4_std_error(inode->i_sb, err); |
| 4539 | return err; |
| 4540 | } |
| 4541 | |
| 4542 | /* |
| 4543 | * Expand an inode by new_extra_isize bytes. |
| 4544 | * Returns 0 on success or negative error number on failure. |
| 4545 | */ |
| 4546 | static int ext4_expand_extra_isize(struct inode *inode, |
| 4547 | unsigned int new_extra_isize, |
| 4548 | struct ext4_iloc iloc, |
| 4549 | handle_t *handle) |
| 4550 | { |
| 4551 | struct ext4_inode *raw_inode; |
| 4552 | struct ext4_xattr_ibody_header *header; |
| 4553 | |
| 4554 | if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) |
| 4555 | return 0; |
| 4556 | |
| 4557 | raw_inode = ext4_raw_inode(&iloc); |
| 4558 | |
| 4559 | header = IHDR(inode, raw_inode); |
| 4560 | |
| 4561 | /* No extended attributes present */ |
| 4562 | if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || |
| 4563 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { |
| 4564 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0, |
| 4565 | new_extra_isize); |
| 4566 | EXT4_I(inode)->i_extra_isize = new_extra_isize; |
| 4567 | return 0; |
| 4568 | } |
| 4569 | |
| 4570 | /* try to expand with EAs present */ |
| 4571 | return ext4_expand_extra_isize_ea(inode, new_extra_isize, |
| 4572 | raw_inode, handle); |
| 4573 | } |
| 4574 | |
| 4575 | /* |
| 4576 | * What we do here is to mark the in-core inode as clean with respect to inode |
| 4577 | * dirtiness (it may still be data-dirty). |
| 4578 | * This means that the in-core inode may be reaped by prune_icache |
| 4579 | * without having to perform any I/O. This is a very good thing, |
| 4580 | * because *any* task may call prune_icache - even ones which |
| 4581 | * have a transaction open against a different journal. |
| 4582 | * |
| 4583 | * Is this cheating? Not really. Sure, we haven't written the |
| 4584 | * inode out, but prune_icache isn't a user-visible syncing function. |
| 4585 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) |
| 4586 | * we start and wait on commits. |
| 4587 | */ |
| 4588 | int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) |
| 4589 | { |
| 4590 | struct ext4_iloc iloc; |
| 4591 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 4592 | static unsigned int mnt_count; |
| 4593 | int err, ret; |
| 4594 | |
| 4595 | might_sleep(); |
| 4596 | trace_ext4_mark_inode_dirty(inode, _RET_IP_); |
| 4597 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
| 4598 | if (ext4_handle_valid(handle) && |
| 4599 | EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize && |
| 4600 | !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) { |
| 4601 | /* |
| 4602 | * We need extra buffer credits since we may write into EA block |
| 4603 | * with this same handle. If journal_extend fails, then it will |
| 4604 | * only result in a minor loss of functionality for that inode. |
| 4605 | * If this is felt to be critical, then e2fsck should be run to |
| 4606 | * force a large enough s_min_extra_isize. |
| 4607 | */ |
| 4608 | if ((jbd2_journal_extend(handle, |
| 4609 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) { |
| 4610 | ret = ext4_expand_extra_isize(inode, |
| 4611 | sbi->s_want_extra_isize, |
| 4612 | iloc, handle); |
| 4613 | if (ret) { |
| 4614 | ext4_set_inode_state(inode, |
| 4615 | EXT4_STATE_NO_EXPAND); |
| 4616 | if (mnt_count != |
| 4617 | le16_to_cpu(sbi->s_es->s_mnt_count)) { |
| 4618 | ext4_warning(inode->i_sb, |
| 4619 | "Unable to expand inode %lu. Delete" |
| 4620 | " some EAs or run e2fsck.", |
| 4621 | inode->i_ino); |
| 4622 | mnt_count = |
| 4623 | le16_to_cpu(sbi->s_es->s_mnt_count); |
| 4624 | } |
| 4625 | } |
| 4626 | } |
| 4627 | } |
| 4628 | if (!err) |
| 4629 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); |
| 4630 | return err; |
| 4631 | } |
| 4632 | |
| 4633 | /* |
| 4634 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
| 4635 | * |
| 4636 | * We're really interested in the case where a file is being extended. |
| 4637 | * i_size has been changed by generic_commit_write() and we thus need |
| 4638 | * to include the updated inode in the current transaction. |
| 4639 | * |
| 4640 | * Also, dquot_alloc_block() will always dirty the inode when blocks |
| 4641 | * are allocated to the file. |
| 4642 | * |
| 4643 | * If the inode is marked synchronous, we don't honour that here - doing |
| 4644 | * so would cause a commit on atime updates, which we don't bother doing. |
| 4645 | * We handle synchronous inodes at the highest possible level. |
| 4646 | */ |
| 4647 | void ext4_dirty_inode(struct inode *inode, int flags) |
| 4648 | { |
| 4649 | handle_t *handle; |
| 4650 | |
| 4651 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); |
| 4652 | if (IS_ERR(handle)) |
| 4653 | goto out; |
| 4654 | |
| 4655 | ext4_mark_inode_dirty(handle, inode); |
| 4656 | |
| 4657 | ext4_journal_stop(handle); |
| 4658 | out: |
| 4659 | return; |
| 4660 | } |
| 4661 | |
| 4662 | #if 0 |
| 4663 | /* |
| 4664 | * Bind an inode's backing buffer_head into this transaction, to prevent |
| 4665 | * it from being flushed to disk early. Unlike |
| 4666 | * ext4_reserve_inode_write, this leaves behind no bh reference and |
| 4667 | * returns no iloc structure, so the caller needs to repeat the iloc |
| 4668 | * lookup to mark the inode dirty later. |
| 4669 | */ |
| 4670 | static int ext4_pin_inode(handle_t *handle, struct inode *inode) |
| 4671 | { |
| 4672 | struct ext4_iloc iloc; |
| 4673 | |
| 4674 | int err = 0; |
| 4675 | if (handle) { |
| 4676 | err = ext4_get_inode_loc(inode, &iloc); |
| 4677 | if (!err) { |
| 4678 | BUFFER_TRACE(iloc.bh, "get_write_access"); |
| 4679 | err = jbd2_journal_get_write_access(handle, iloc.bh); |
| 4680 | if (!err) |
| 4681 | err = ext4_handle_dirty_metadata(handle, |
| 4682 | NULL, |
| 4683 | iloc.bh); |
| 4684 | brelse(iloc.bh); |
| 4685 | } |
| 4686 | } |
| 4687 | ext4_std_error(inode->i_sb, err); |
| 4688 | return err; |
| 4689 | } |
| 4690 | #endif |
| 4691 | |
| 4692 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
| 4693 | { |
| 4694 | journal_t *journal; |
| 4695 | handle_t *handle; |
| 4696 | int err; |
| 4697 | |
| 4698 | /* |
| 4699 | * We have to be very careful here: changing a data block's |
| 4700 | * journaling status dynamically is dangerous. If we write a |
| 4701 | * data block to the journal, change the status and then delete |
| 4702 | * that block, we risk forgetting to revoke the old log record |
| 4703 | * from the journal and so a subsequent replay can corrupt data. |
| 4704 | * So, first we make sure that the journal is empty and that |
| 4705 | * nobody is changing anything. |
| 4706 | */ |
| 4707 | |
| 4708 | journal = EXT4_JOURNAL(inode); |
| 4709 | if (!journal) |
| 4710 | return 0; |
| 4711 | if (is_journal_aborted(journal)) |
| 4712 | return -EROFS; |
| 4713 | /* We have to allocate physical blocks for delalloc blocks |
| 4714 | * before flushing journal. otherwise delalloc blocks can not |
| 4715 | * be allocated any more. even more truncate on delalloc blocks |
| 4716 | * could trigger BUG by flushing delalloc blocks in journal. |
| 4717 | * There is no delalloc block in non-journal data mode. |
| 4718 | */ |
| 4719 | if (val && test_opt(inode->i_sb, DELALLOC)) { |
| 4720 | err = ext4_alloc_da_blocks(inode); |
| 4721 | if (err < 0) |
| 4722 | return err; |
| 4723 | } |
| 4724 | |
| 4725 | /* Wait for all existing dio workers */ |
| 4726 | ext4_inode_block_unlocked_dio(inode); |
| 4727 | inode_dio_wait(inode); |
| 4728 | |
| 4729 | jbd2_journal_lock_updates(journal); |
| 4730 | |
| 4731 | /* |
| 4732 | * OK, there are no updates running now, and all cached data is |
| 4733 | * synced to disk. We are now in a completely consistent state |
| 4734 | * which doesn't have anything in the journal, and we know that |
| 4735 | * no filesystem updates are running, so it is safe to modify |
| 4736 | * the inode's in-core data-journaling state flag now. |
| 4737 | */ |
| 4738 | |
| 4739 | if (val) |
| 4740 | ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
| 4741 | else { |
| 4742 | jbd2_journal_flush(journal); |
| 4743 | ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
| 4744 | } |
| 4745 | ext4_set_aops(inode); |
| 4746 | |
| 4747 | jbd2_journal_unlock_updates(journal); |
| 4748 | ext4_inode_resume_unlocked_dio(inode); |
| 4749 | |
| 4750 | /* Finally we can mark the inode as dirty. */ |
| 4751 | |
| 4752 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); |
| 4753 | if (IS_ERR(handle)) |
| 4754 | return PTR_ERR(handle); |
| 4755 | |
| 4756 | err = ext4_mark_inode_dirty(handle, inode); |
| 4757 | ext4_handle_sync(handle); |
| 4758 | ext4_journal_stop(handle); |
| 4759 | ext4_std_error(inode->i_sb, err); |
| 4760 | |
| 4761 | return err; |
| 4762 | } |
| 4763 | |
| 4764 | static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh) |
| 4765 | { |
| 4766 | return !buffer_mapped(bh); |
| 4767 | } |
| 4768 | |
| 4769 | int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
| 4770 | { |
| 4771 | struct page *page = vmf->page; |
| 4772 | loff_t size; |
| 4773 | unsigned long len; |
| 4774 | int ret; |
| 4775 | struct file *file = vma->vm_file; |
| 4776 | struct inode *inode = file->f_path.dentry->d_inode; |
| 4777 | struct address_space *mapping = inode->i_mapping; |
| 4778 | handle_t *handle; |
| 4779 | get_block_t *get_block; |
| 4780 | int retries = 0; |
| 4781 | |
| 4782 | sb_start_pagefault(inode->i_sb); |
| 4783 | file_update_time(vma->vm_file); |
| 4784 | /* Delalloc case is easy... */ |
| 4785 | if (test_opt(inode->i_sb, DELALLOC) && |
| 4786 | !ext4_should_journal_data(inode) && |
| 4787 | !ext4_nonda_switch(inode->i_sb)) { |
| 4788 | do { |
| 4789 | ret = __block_page_mkwrite(vma, vmf, |
| 4790 | ext4_da_get_block_prep); |
| 4791 | } while (ret == -ENOSPC && |
| 4792 | ext4_should_retry_alloc(inode->i_sb, &retries)); |
| 4793 | goto out_ret; |
| 4794 | } |
| 4795 | |
| 4796 | lock_page(page); |
| 4797 | size = i_size_read(inode); |
| 4798 | /* Page got truncated from under us? */ |
| 4799 | if (page->mapping != mapping || page_offset(page) > size) { |
| 4800 | unlock_page(page); |
| 4801 | ret = VM_FAULT_NOPAGE; |
| 4802 | goto out; |
| 4803 | } |
| 4804 | |
| 4805 | if (page->index == size >> PAGE_CACHE_SHIFT) |
| 4806 | len = size & ~PAGE_CACHE_MASK; |
| 4807 | else |
| 4808 | len = PAGE_CACHE_SIZE; |
| 4809 | /* |
| 4810 | * Return if we have all the buffers mapped. This avoids the need to do |
| 4811 | * journal_start/journal_stop which can block and take a long time |
| 4812 | */ |
| 4813 | if (page_has_buffers(page)) { |
| 4814 | if (!ext4_walk_page_buffers(NULL, page_buffers(page), |
| 4815 | 0, len, NULL, |
| 4816 | ext4_bh_unmapped)) { |
| 4817 | /* Wait so that we don't change page under IO */ |
| 4818 | wait_on_page_writeback(page); |
| 4819 | ret = VM_FAULT_LOCKED; |
| 4820 | goto out; |
| 4821 | } |
| 4822 | } |
| 4823 | unlock_page(page); |
| 4824 | /* OK, we need to fill the hole... */ |
| 4825 | if (ext4_should_dioread_nolock(inode)) |
| 4826 | get_block = ext4_get_block_write; |
| 4827 | else |
| 4828 | get_block = ext4_get_block; |
| 4829 | retry_alloc: |
| 4830 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
| 4831 | ext4_writepage_trans_blocks(inode)); |
| 4832 | if (IS_ERR(handle)) { |
| 4833 | ret = VM_FAULT_SIGBUS; |
| 4834 | goto out; |
| 4835 | } |
| 4836 | ret = __block_page_mkwrite(vma, vmf, get_block); |
| 4837 | if (!ret && ext4_should_journal_data(inode)) { |
| 4838 | if (ext4_walk_page_buffers(handle, page_buffers(page), 0, |
| 4839 | PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) { |
| 4840 | unlock_page(page); |
| 4841 | ret = VM_FAULT_SIGBUS; |
| 4842 | ext4_journal_stop(handle); |
| 4843 | goto out; |
| 4844 | } |
| 4845 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
| 4846 | } |
| 4847 | ext4_journal_stop(handle); |
| 4848 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
| 4849 | goto retry_alloc; |
| 4850 | out_ret: |
| 4851 | ret = block_page_mkwrite_return(ret); |
| 4852 | out: |
| 4853 | sb_end_pagefault(inode->i_sb); |
| 4854 | return ret; |
| 4855 | } |