| 1 | // SPDX-License-Identifier: GPL-2.0+ |
| 2 | /* |
| 3 | * ext4_jbd2.h |
| 4 | * |
| 5 | * Written by Stephen C. Tweedie <sct@redhat.com>, 1999 |
| 6 | * |
| 7 | * Copyright 1998--1999 Red Hat corp --- All Rights Reserved |
| 8 | * |
| 9 | * Ext4-specific journaling extensions. |
| 10 | */ |
| 11 | |
| 12 | #ifndef _EXT4_JBD2_H |
| 13 | #define _EXT4_JBD2_H |
| 14 | |
| 15 | #include <linux/fs.h> |
| 16 | #include <linux/jbd2.h> |
| 17 | #include "ext4.h" |
| 18 | |
| 19 | #define EXT4_JOURNAL(inode) (EXT4_SB((inode)->i_sb)->s_journal) |
| 20 | |
| 21 | /* Define the number of blocks we need to account to a transaction to |
| 22 | * modify one block of data. |
| 23 | * |
| 24 | * We may have to touch one inode, one bitmap buffer, up to three |
| 25 | * indirection blocks, the group and superblock summaries, and the data |
| 26 | * block to complete the transaction. |
| 27 | * |
| 28 | * For extents-enabled fs we may have to allocate and modify up to |
| 29 | * 5 levels of tree, data block (for each of these we need bitmap + group |
| 30 | * summaries), root which is stored in the inode, sb |
| 31 | */ |
| 32 | |
| 33 | #define EXT4_SINGLEDATA_TRANS_BLOCKS(sb) \ |
| 34 | (ext4_has_feature_extents(sb) ? 20U : 8U) |
| 35 | |
| 36 | /* Extended attribute operations touch at most two data buffers, |
| 37 | * two bitmap buffers, and two group summaries, in addition to the inode |
| 38 | * and the superblock, which are already accounted for. */ |
| 39 | |
| 40 | #define EXT4_XATTR_TRANS_BLOCKS 6U |
| 41 | |
| 42 | /* Define the minimum size for a transaction which modifies data. This |
| 43 | * needs to take into account the fact that we may end up modifying two |
| 44 | * quota files too (one for the group, one for the user quota). The |
| 45 | * superblock only gets updated once, of course, so don't bother |
| 46 | * counting that again for the quota updates. */ |
| 47 | |
| 48 | #define EXT4_DATA_TRANS_BLOCKS(sb) (EXT4_SINGLEDATA_TRANS_BLOCKS(sb) + \ |
| 49 | EXT4_XATTR_TRANS_BLOCKS - 2 + \ |
| 50 | EXT4_MAXQUOTAS_TRANS_BLOCKS(sb)) |
| 51 | |
| 52 | /* |
| 53 | * Define the number of metadata blocks we need to account to modify data. |
| 54 | * |
| 55 | * This include super block, inode block, quota blocks and xattr blocks |
| 56 | */ |
| 57 | #define EXT4_META_TRANS_BLOCKS(sb) (EXT4_XATTR_TRANS_BLOCKS + \ |
| 58 | EXT4_MAXQUOTAS_TRANS_BLOCKS(sb)) |
| 59 | |
| 60 | /* Define an arbitrary limit for the amount of data we will anticipate |
| 61 | * writing to any given transaction. For unbounded transactions such as |
| 62 | * write(2) and truncate(2) we can write more than this, but we always |
| 63 | * start off at the maximum transaction size and grow the transaction |
| 64 | * optimistically as we go. */ |
| 65 | |
| 66 | #define EXT4_MAX_TRANS_DATA 64U |
| 67 | |
| 68 | /* We break up a large truncate or write transaction once the handle's |
| 69 | * buffer credits gets this low, we need either to extend the |
| 70 | * transaction or to start a new one. Reserve enough space here for |
| 71 | * inode, bitmap, superblock, group and indirection updates for at least |
| 72 | * one block, plus two quota updates. Quota allocations are not |
| 73 | * needed. */ |
| 74 | |
| 75 | #define EXT4_RESERVE_TRANS_BLOCKS 12U |
| 76 | |
| 77 | /* |
| 78 | * Number of credits needed if we need to insert an entry into a |
| 79 | * directory. For each new index block, we need 4 blocks (old index |
| 80 | * block, new index block, bitmap block, bg summary). For normal |
| 81 | * htree directories there are 2 levels; if the largedir feature |
| 82 | * enabled it's 3 levels. |
| 83 | */ |
| 84 | #define 12U |
| 85 | |
| 86 | #ifdef CONFIG_QUOTA |
| 87 | /* Amount of blocks needed for quota update - we know that the structure was |
| 88 | * allocated so we need to update only data block */ |
| 89 | #define EXT4_QUOTA_TRANS_BLOCKS(sb) ((ext4_quota_capable(sb)) ? 1 : 0) |
| 90 | /* Amount of blocks needed for quota insert/delete - we do some block writes |
| 91 | * but inode, sb and group updates are done only once */ |
| 92 | #define EXT4_QUOTA_INIT_BLOCKS(sb) ((ext4_quota_capable(sb)) ?\ |
| 93 | (DQUOT_INIT_ALLOC*(EXT4_SINGLEDATA_TRANS_BLOCKS(sb)-3)\ |
| 94 | +3+DQUOT_INIT_REWRITE) : 0) |
| 95 | |
| 96 | #define EXT4_QUOTA_DEL_BLOCKS(sb) ((ext4_quota_capable(sb)) ?\ |
| 97 | (DQUOT_DEL_ALLOC*(EXT4_SINGLEDATA_TRANS_BLOCKS(sb)-3)\ |
| 98 | +3+DQUOT_DEL_REWRITE) : 0) |
| 99 | #else |
| 100 | #define EXT4_QUOTA_TRANS_BLOCKS(sb) 0 |
| 101 | #define EXT4_QUOTA_INIT_BLOCKS(sb) 0 |
| 102 | #define EXT4_QUOTA_DEL_BLOCKS(sb) 0 |
| 103 | #endif |
| 104 | #define EXT4_MAXQUOTAS_TRANS_BLOCKS(sb) (EXT4_MAXQUOTAS*EXT4_QUOTA_TRANS_BLOCKS(sb)) |
| 105 | #define EXT4_MAXQUOTAS_INIT_BLOCKS(sb) (EXT4_MAXQUOTAS*EXT4_QUOTA_INIT_BLOCKS(sb)) |
| 106 | #define EXT4_MAXQUOTAS_DEL_BLOCKS(sb) (EXT4_MAXQUOTAS*EXT4_QUOTA_DEL_BLOCKS(sb)) |
| 107 | |
| 108 | /* |
| 109 | * Ext4 handle operation types -- for logging purposes |
| 110 | */ |
| 111 | #define EXT4_HT_MISC 0 |
| 112 | #define EXT4_HT_INODE 1 |
| 113 | #define EXT4_HT_WRITE_PAGE 2 |
| 114 | #define EXT4_HT_MAP_BLOCKS 3 |
| 115 | #define EXT4_HT_DIR 4 |
| 116 | #define EXT4_HT_TRUNCATE 5 |
| 117 | #define EXT4_HT_QUOTA 6 |
| 118 | #define EXT4_HT_RESIZE 7 |
| 119 | #define EXT4_HT_MIGRATE 8 |
| 120 | #define EXT4_HT_MOVE_EXTENTS 9 |
| 121 | #define EXT4_HT_XATTR 10 |
| 122 | #define EXT4_HT_EXT_CONVERT 11 |
| 123 | #define EXT4_HT_MAX 12 |
| 124 | |
| 125 | int |
| 126 | ext4_mark_iloc_dirty(handle_t *handle, |
| 127 | struct inode *inode, |
| 128 | struct ext4_iloc *iloc); |
| 129 | |
| 130 | /* |
| 131 | * On success, We end up with an outstanding reference count against |
| 132 | * iloc->bh. This _must_ be cleaned up later. |
| 133 | */ |
| 134 | |
| 135 | int ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
| 136 | struct ext4_iloc *iloc); |
| 137 | |
| 138 | #define ext4_mark_inode_dirty(__h, __i) \ |
| 139 | __ext4_mark_inode_dirty((__h), (__i), __func__, __LINE__) |
| 140 | int __ext4_mark_inode_dirty(handle_t *handle, struct inode *inode, |
| 141 | const char *func, unsigned int line); |
| 142 | |
| 143 | int ext4_expand_extra_isize(struct inode *inode, |
| 144 | unsigned int , |
| 145 | struct ext4_iloc *iloc); |
| 146 | /* |
| 147 | * Wrapper functions with which ext4 calls into JBD. |
| 148 | */ |
| 149 | int __ext4_journal_get_write_access(const char *where, unsigned int line, |
| 150 | handle_t *handle, struct super_block *sb, |
| 151 | struct buffer_head *bh, |
| 152 | enum ext4_journal_trigger_type trigger_type); |
| 153 | |
| 154 | int __ext4_forget(const char *where, unsigned int line, handle_t *handle, |
| 155 | int is_metadata, struct inode *inode, |
| 156 | struct buffer_head *bh, ext4_fsblk_t blocknr); |
| 157 | |
| 158 | int __ext4_journal_get_create_access(const char *where, unsigned int line, |
| 159 | handle_t *handle, struct super_block *sb, |
| 160 | struct buffer_head *bh, |
| 161 | enum ext4_journal_trigger_type trigger_type); |
| 162 | |
| 163 | int __ext4_handle_dirty_metadata(const char *where, unsigned int line, |
| 164 | handle_t *handle, struct inode *inode, |
| 165 | struct buffer_head *bh); |
| 166 | |
| 167 | #define ext4_journal_get_write_access(handle, sb, bh, trigger_type) \ |
| 168 | __ext4_journal_get_write_access(__func__, __LINE__, (handle), (sb), \ |
| 169 | (bh), (trigger_type)) |
| 170 | #define ext4_forget(handle, is_metadata, inode, bh, block_nr) \ |
| 171 | __ext4_forget(__func__, __LINE__, (handle), (is_metadata), (inode), \ |
| 172 | (bh), (block_nr)) |
| 173 | #define ext4_journal_get_create_access(handle, sb, bh, trigger_type) \ |
| 174 | __ext4_journal_get_create_access(__func__, __LINE__, (handle), (sb), \ |
| 175 | (bh), (trigger_type)) |
| 176 | #define ext4_handle_dirty_metadata(handle, inode, bh) \ |
| 177 | __ext4_handle_dirty_metadata(__func__, __LINE__, (handle), (inode), \ |
| 178 | (bh)) |
| 179 | |
| 180 | handle_t *__ext4_journal_start_sb(struct inode *inode, struct super_block *sb, |
| 181 | unsigned int line, int type, int blocks, |
| 182 | int rsv_blocks, int revoke_creds); |
| 183 | int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle); |
| 184 | |
| 185 | #define EXT4_NOJOURNAL_MAX_REF_COUNT ((unsigned long) 4096) |
| 186 | |
| 187 | /* Note: Do not use this for NULL handles. This is only to determine if |
| 188 | * a properly allocated handle is using a journal or not. */ |
| 189 | static inline int ext4_handle_valid(handle_t *handle) |
| 190 | { |
| 191 | if ((unsigned long)handle < EXT4_NOJOURNAL_MAX_REF_COUNT) |
| 192 | return 0; |
| 193 | return 1; |
| 194 | } |
| 195 | |
| 196 | static inline void ext4_handle_sync(handle_t *handle) |
| 197 | { |
| 198 | if (ext4_handle_valid(handle)) |
| 199 | handle->h_sync = 1; |
| 200 | } |
| 201 | |
| 202 | static inline int ext4_handle_is_aborted(handle_t *handle) |
| 203 | { |
| 204 | if (ext4_handle_valid(handle)) |
| 205 | return is_handle_aborted(handle); |
| 206 | return 0; |
| 207 | } |
| 208 | |
| 209 | static inline int ext4_free_metadata_revoke_credits(struct super_block *sb, |
| 210 | int blocks) |
| 211 | { |
| 212 | /* Freeing each metadata block can result in freeing one cluster */ |
| 213 | return blocks * EXT4_SB(sb)->s_cluster_ratio; |
| 214 | } |
| 215 | |
| 216 | static inline int ext4_trans_default_revoke_credits(struct super_block *sb) |
| 217 | { |
| 218 | return ext4_free_metadata_revoke_credits(sb, blocks: 8); |
| 219 | } |
| 220 | |
| 221 | #define ext4_journal_start_sb(sb, type, nblocks) \ |
| 222 | __ext4_journal_start_sb(NULL, (sb), __LINE__, (type), (nblocks), 0,\ |
| 223 | ext4_trans_default_revoke_credits(sb)) |
| 224 | |
| 225 | #define ext4_journal_start(inode, type, nblocks) \ |
| 226 | __ext4_journal_start((inode), __LINE__, (type), (nblocks), 0, \ |
| 227 | ext4_trans_default_revoke_credits((inode)->i_sb)) |
| 228 | |
| 229 | #define ext4_journal_start_with_reserve(inode, type, blocks, rsv_blocks)\ |
| 230 | __ext4_journal_start((inode), __LINE__, (type), (blocks), (rsv_blocks),\ |
| 231 | ext4_trans_default_revoke_credits((inode)->i_sb)) |
| 232 | |
| 233 | #define ext4_journal_start_with_revoke(inode, type, blocks, revoke_creds) \ |
| 234 | __ext4_journal_start((inode), __LINE__, (type), (blocks), 0, \ |
| 235 | (revoke_creds)) |
| 236 | |
| 237 | static inline handle_t *__ext4_journal_start(struct inode *inode, |
| 238 | unsigned int line, int type, |
| 239 | int blocks, int rsv_blocks, |
| 240 | int revoke_creds) |
| 241 | { |
| 242 | return __ext4_journal_start_sb(inode, sb: inode->i_sb, line, type, blocks, |
| 243 | rsv_blocks, revoke_creds); |
| 244 | } |
| 245 | |
| 246 | #define ext4_journal_stop(handle) \ |
| 247 | __ext4_journal_stop(__func__, __LINE__, (handle)) |
| 248 | |
| 249 | #define ext4_journal_start_reserved(handle, type) \ |
| 250 | __ext4_journal_start_reserved((handle), __LINE__, (type)) |
| 251 | |
| 252 | handle_t *__ext4_journal_start_reserved(handle_t *handle, unsigned int line, |
| 253 | int type); |
| 254 | |
| 255 | static inline handle_t *ext4_journal_current_handle(void) |
| 256 | { |
| 257 | return journal_current_handle(); |
| 258 | } |
| 259 | |
| 260 | static inline int ext4_journal_extend(handle_t *handle, int nblocks, int revoke) |
| 261 | { |
| 262 | if (ext4_handle_valid(handle)) |
| 263 | return jbd2_journal_extend(handle, nblocks, revoke_records: revoke); |
| 264 | return 0; |
| 265 | } |
| 266 | |
| 267 | static inline int ext4_journal_restart(handle_t *handle, int nblocks, |
| 268 | int revoke) |
| 269 | { |
| 270 | if (ext4_handle_valid(handle)) |
| 271 | return jbd2__journal_restart(handle, nblocks, revoke_records: revoke, GFP_NOFS); |
| 272 | return 0; |
| 273 | } |
| 274 | |
| 275 | int __ext4_journal_ensure_credits(handle_t *handle, int check_cred, |
| 276 | int extend_cred, int revoke_cred); |
| 277 | |
| 278 | |
| 279 | /* |
| 280 | * Ensure @handle has at least @check_creds credits available. If not, |
| 281 | * transaction will be extended or restarted to contain at least @extend_cred |
| 282 | * credits. Before restarting transaction @fn is executed to allow for cleanup |
| 283 | * before the transaction is restarted. |
| 284 | * |
| 285 | * The return value is < 0 in case of error, 0 in case the handle has enough |
| 286 | * credits or transaction extension succeeded, 1 in case transaction had to be |
| 287 | * restarted. |
| 288 | */ |
| 289 | #define ext4_journal_ensure_credits_fn(handle, check_cred, extend_cred, \ |
| 290 | revoke_cred, fn) \ |
| 291 | ({ \ |
| 292 | __label__ __ensure_end; \ |
| 293 | int err = __ext4_journal_ensure_credits((handle), (check_cred), \ |
| 294 | (extend_cred), (revoke_cred)); \ |
| 295 | \ |
| 296 | if (err <= 0) \ |
| 297 | goto __ensure_end; \ |
| 298 | err = (fn); \ |
| 299 | if (err < 0) \ |
| 300 | goto __ensure_end; \ |
| 301 | err = ext4_journal_restart((handle), (extend_cred), (revoke_cred)); \ |
| 302 | if (err == 0) \ |
| 303 | err = 1; \ |
| 304 | __ensure_end: \ |
| 305 | err; \ |
| 306 | }) |
| 307 | |
| 308 | /* |
| 309 | * Ensure given handle has at least requested amount of credits available, |
| 310 | * possibly restarting transaction if needed. We also make sure the transaction |
| 311 | * has space for at least ext4_trans_default_revoke_credits(sb) revoke records |
| 312 | * as freeing one or two blocks is very common pattern and requesting this is |
| 313 | * very cheap. |
| 314 | */ |
| 315 | static inline int ext4_journal_ensure_credits(handle_t *handle, int credits, |
| 316 | int revoke_creds) |
| 317 | { |
| 318 | return ext4_journal_ensure_credits_fn(handle, credits, credits, |
| 319 | revoke_creds, 0); |
| 320 | } |
| 321 | |
| 322 | static inline int ext4_journal_blocks_per_folio(struct inode *inode) |
| 323 | { |
| 324 | if (EXT4_JOURNAL(inode) != NULL) |
| 325 | return jbd2_journal_blocks_per_folio(inode); |
| 326 | return 0; |
| 327 | } |
| 328 | |
| 329 | static inline int ext4_journal_force_commit(journal_t *journal) |
| 330 | { |
| 331 | if (journal) |
| 332 | return jbd2_journal_force_commit(journal); |
| 333 | return 0; |
| 334 | } |
| 335 | |
| 336 | static inline int ext4_jbd2_inode_add_write(handle_t *handle, |
| 337 | struct inode *inode, loff_t start_byte, loff_t length) |
| 338 | { |
| 339 | if (ext4_handle_valid(handle)) |
| 340 | return jbd2_journal_inode_ranged_write(handle, |
| 341 | inode: EXT4_I(inode)->jinode, start_byte, length); |
| 342 | return 0; |
| 343 | } |
| 344 | |
| 345 | static inline int ext4_jbd2_inode_add_wait(handle_t *handle, |
| 346 | struct inode *inode, loff_t start_byte, loff_t length) |
| 347 | { |
| 348 | if (ext4_handle_valid(handle)) |
| 349 | return jbd2_journal_inode_ranged_wait(handle, |
| 350 | inode: EXT4_I(inode)->jinode, start_byte, length); |
| 351 | return 0; |
| 352 | } |
| 353 | |
| 354 | static inline void ext4_update_inode_fsync_trans(handle_t *handle, |
| 355 | struct inode *inode, |
| 356 | int datasync) |
| 357 | { |
| 358 | struct ext4_inode_info *ei = EXT4_I(inode); |
| 359 | |
| 360 | if (ext4_handle_valid(handle) && !is_handle_aborted(handle)) { |
| 361 | ei->i_sync_tid = handle->h_transaction->t_tid; |
| 362 | if (datasync) |
| 363 | ei->i_datasync_tid = handle->h_transaction->t_tid; |
| 364 | } |
| 365 | } |
| 366 | |
| 367 | /* super.c */ |
| 368 | int ext4_force_commit(struct super_block *sb); |
| 369 | |
| 370 | /* |
| 371 | * Ext4 inode journal modes |
| 372 | */ |
| 373 | #define EXT4_INODE_JOURNAL_DATA_MODE 0x01 /* journal data mode */ |
| 374 | #define EXT4_INODE_ORDERED_DATA_MODE 0x02 /* ordered data mode */ |
| 375 | #define EXT4_INODE_WRITEBACK_DATA_MODE 0x04 /* writeback data mode */ |
| 376 | |
| 377 | int ext4_inode_journal_mode(struct inode *inode); |
| 378 | |
| 379 | static inline int ext4_should_journal_data(struct inode *inode) |
| 380 | { |
| 381 | return ext4_inode_journal_mode(inode) & EXT4_INODE_JOURNAL_DATA_MODE; |
| 382 | } |
| 383 | |
| 384 | static inline int ext4_should_order_data(struct inode *inode) |
| 385 | { |
| 386 | return ext4_inode_journal_mode(inode) & EXT4_INODE_ORDERED_DATA_MODE; |
| 387 | } |
| 388 | |
| 389 | static inline int ext4_should_writeback_data(struct inode *inode) |
| 390 | { |
| 391 | return ext4_inode_journal_mode(inode) & EXT4_INODE_WRITEBACK_DATA_MODE; |
| 392 | } |
| 393 | |
| 394 | static inline int ext4_free_data_revoke_credits(struct inode *inode, int blocks) |
| 395 | { |
| 396 | if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) |
| 397 | return 0; |
| 398 | if (!ext4_should_journal_data(inode)) |
| 399 | return 0; |
| 400 | /* |
| 401 | * Data blocks in one extent are contiguous, just account for partial |
| 402 | * clusters at extent boundaries |
| 403 | */ |
| 404 | return blocks + 2*(EXT4_SB(sb: inode->i_sb)->s_cluster_ratio - 1); |
| 405 | } |
| 406 | |
| 407 | /* |
| 408 | * This function controls whether or not we should try to go down the |
| 409 | * dioread_nolock code paths, which makes it safe to avoid taking |
| 410 | * i_rwsem for direct I/O reads. This only works for extent-based |
| 411 | * files, and it doesn't work if data journaling is enabled, since the |
| 412 | * dioread_nolock code uses b_private to pass information back to the |
| 413 | * I/O completion handler, and this conflicts with the jbd's use of |
| 414 | * b_private. |
| 415 | */ |
| 416 | static inline int ext4_should_dioread_nolock(struct inode *inode) |
| 417 | { |
| 418 | if (!test_opt(inode->i_sb, DIOREAD_NOLOCK)) |
| 419 | return 0; |
| 420 | if (!S_ISREG(inode->i_mode)) |
| 421 | return 0; |
| 422 | if (!(ext4_test_inode_flag(inode, bit: EXT4_INODE_EXTENTS))) |
| 423 | return 0; |
| 424 | if (ext4_should_journal_data(inode)) |
| 425 | return 0; |
| 426 | /* temporary fix to prevent generic/422 test failures */ |
| 427 | if (!test_opt(inode->i_sb, DELALLOC)) |
| 428 | return 0; |
| 429 | return 1; |
| 430 | } |
| 431 | |
| 432 | /* |
| 433 | * Pass journal explicitly as it may not be cached in the sbi->s_journal in some |
| 434 | * cases |
| 435 | */ |
| 436 | static inline int ext4_journal_destroy(struct ext4_sb_info *sbi, journal_t *journal) |
| 437 | { |
| 438 | int err = 0; |
| 439 | |
| 440 | /* |
| 441 | * At this point only two things can be operating on the journal. |
| 442 | * JBD2 thread performing transaction commit and s_sb_upd_work |
| 443 | * issuing sb update through the journal. Once we set |
| 444 | * EXT4_JOURNAL_DESTROY, new ext4_handle_error() calls will not |
| 445 | * queue s_sb_upd_work and ext4_force_commit() makes sure any |
| 446 | * ext4_handle_error() calls from the running transaction commit are |
| 447 | * finished. Hence no new s_sb_upd_work can be queued after we |
| 448 | * flush it here. |
| 449 | */ |
| 450 | ext4_set_mount_flag(sb: sbi->s_sb, bit: EXT4_MF_JOURNAL_DESTROY); |
| 451 | |
| 452 | ext4_force_commit(sb: sbi->s_sb); |
| 453 | flush_work(work: &sbi->s_sb_upd_work); |
| 454 | |
| 455 | err = jbd2_journal_destroy(journal); |
| 456 | sbi->s_journal = NULL; |
| 457 | |
| 458 | return err; |
| 459 | } |
| 460 | |
| 461 | #endif /* _EXT4_JBD2_H */ |
| 462 | |