| 1 | // SPDX-License-Identifier: GPL-2.0+ | 
|---|
| 2 | /* | 
|---|
| 3 | * linux/fs/jbd2/checkpoint.c | 
|---|
| 4 | * | 
|---|
| 5 | * Written by Stephen C. Tweedie <sct@redhat.com>, 1999 | 
|---|
| 6 | * | 
|---|
| 7 | * Copyright 1999 Red Hat Software --- All Rights Reserved | 
|---|
| 8 | * | 
|---|
| 9 | * Checkpoint routines for the generic filesystem journaling code. | 
|---|
| 10 | * Part of the ext2fs journaling system. | 
|---|
| 11 | * | 
|---|
| 12 | * Checkpointing is the process of ensuring that a section of the log is | 
|---|
| 13 | * committed fully to disk, so that that portion of the log can be | 
|---|
| 14 | * reused. | 
|---|
| 15 | */ | 
|---|
| 16 |  | 
|---|
| 17 | #include <linux/time.h> | 
|---|
| 18 | #include <linux/fs.h> | 
|---|
| 19 | #include <linux/jbd2.h> | 
|---|
| 20 | #include <linux/errno.h> | 
|---|
| 21 | #include <linux/slab.h> | 
|---|
| 22 | #include <linux/blkdev.h> | 
|---|
| 23 | #include <trace/events/jbd2.h> | 
|---|
| 24 |  | 
|---|
| 25 | /* | 
|---|
| 26 | * Unlink a buffer from a transaction checkpoint list. | 
|---|
| 27 | * | 
|---|
| 28 | * Called with j_list_lock held. | 
|---|
| 29 | */ | 
|---|
| 30 | static inline void __buffer_unlink(struct journal_head *jh) | 
|---|
| 31 | { | 
|---|
| 32 | transaction_t *transaction = jh->b_cp_transaction; | 
|---|
| 33 |  | 
|---|
| 34 | jh->b_cpnext->b_cpprev = jh->b_cpprev; | 
|---|
| 35 | jh->b_cpprev->b_cpnext = jh->b_cpnext; | 
|---|
| 36 | if (transaction->t_checkpoint_list == jh) { | 
|---|
| 37 | transaction->t_checkpoint_list = jh->b_cpnext; | 
|---|
| 38 | if (transaction->t_checkpoint_list == jh) | 
|---|
| 39 | transaction->t_checkpoint_list = NULL; | 
|---|
| 40 | } | 
|---|
| 41 | } | 
|---|
| 42 |  | 
|---|
| 43 | /* | 
|---|
| 44 | * __jbd2_log_wait_for_space: wait until there is space in the journal. | 
|---|
| 45 | * | 
|---|
| 46 | * Called under j-state_lock *only*.  It will be unlocked if we have to wait | 
|---|
| 47 | * for a checkpoint to free up some space in the log. | 
|---|
| 48 | */ | 
|---|
| 49 | void __jbd2_log_wait_for_space(journal_t *journal) | 
|---|
| 50 | __acquires(&journal->j_state_lock) | 
|---|
| 51 | __releases(&journal->j_state_lock) | 
|---|
| 52 | { | 
|---|
| 53 | int nblocks, space_left; | 
|---|
| 54 | /* assert_spin_locked(&journal->j_state_lock); */ | 
|---|
| 55 |  | 
|---|
| 56 | nblocks = journal->j_max_transaction_buffers; | 
|---|
| 57 | while (jbd2_log_space_left(journal) < nblocks) { | 
|---|
| 58 | write_unlock(&journal->j_state_lock); | 
|---|
| 59 | mutex_lock_io(lock: &journal->j_checkpoint_mutex); | 
|---|
| 60 |  | 
|---|
| 61 | /* | 
|---|
| 62 | * Test again, another process may have checkpointed while we | 
|---|
| 63 | * were waiting for the checkpoint lock. If there are no | 
|---|
| 64 | * transactions ready to be checkpointed, try to recover | 
|---|
| 65 | * journal space by calling cleanup_journal_tail(), and if | 
|---|
| 66 | * that doesn't work, by waiting for the currently committing | 
|---|
| 67 | * transaction to complete.  If there is absolutely no way | 
|---|
| 68 | * to make progress, this is either a BUG or corrupted | 
|---|
| 69 | * filesystem, so abort the journal and leave a stack | 
|---|
| 70 | * trace for forensic evidence. | 
|---|
| 71 | */ | 
|---|
| 72 | write_lock(&journal->j_state_lock); | 
|---|
| 73 | if (journal->j_flags & JBD2_ABORT) { | 
|---|
| 74 | mutex_unlock(lock: &journal->j_checkpoint_mutex); | 
|---|
| 75 | return; | 
|---|
| 76 | } | 
|---|
| 77 | spin_lock(lock: &journal->j_list_lock); | 
|---|
| 78 | space_left = jbd2_log_space_left(journal); | 
|---|
| 79 | if (space_left < nblocks) { | 
|---|
| 80 | int chkpt = journal->j_checkpoint_transactions != NULL; | 
|---|
| 81 | tid_t tid = 0; | 
|---|
| 82 | bool has_transaction = false; | 
|---|
| 83 |  | 
|---|
| 84 | if (journal->j_committing_transaction) { | 
|---|
| 85 | tid = journal->j_committing_transaction->t_tid; | 
|---|
| 86 | has_transaction = true; | 
|---|
| 87 | } | 
|---|
| 88 | spin_unlock(lock: &journal->j_list_lock); | 
|---|
| 89 | write_unlock(&journal->j_state_lock); | 
|---|
| 90 | if (chkpt) { | 
|---|
| 91 | jbd2_log_do_checkpoint(journal); | 
|---|
| 92 | } else if (jbd2_cleanup_journal_tail(journal) <= 0) { | 
|---|
| 93 | /* | 
|---|
| 94 | * We were able to recover space or the | 
|---|
| 95 | * journal was aborted due to an error. | 
|---|
| 96 | */ | 
|---|
| 97 | ; | 
|---|
| 98 | } else if (has_transaction) { | 
|---|
| 99 | /* | 
|---|
| 100 | * jbd2_journal_commit_transaction() may want | 
|---|
| 101 | * to take the checkpoint_mutex if JBD2_FLUSHED | 
|---|
| 102 | * is set.  So we need to temporarily drop it. | 
|---|
| 103 | */ | 
|---|
| 104 | mutex_unlock(lock: &journal->j_checkpoint_mutex); | 
|---|
| 105 | jbd2_log_wait_commit(journal, tid); | 
|---|
| 106 | write_lock(&journal->j_state_lock); | 
|---|
| 107 | continue; | 
|---|
| 108 | } else { | 
|---|
| 109 | printk(KERN_ERR "%s: needed %d blocks and " | 
|---|
| 110 | "only had %d space available\n", | 
|---|
| 111 | __func__, nblocks, space_left); | 
|---|
| 112 | printk(KERN_ERR "%s: no way to get more " | 
|---|
| 113 | "journal space in %s\n", __func__, | 
|---|
| 114 | journal->j_devname); | 
|---|
| 115 | WARN_ON(1); | 
|---|
| 116 | jbd2_journal_abort(journal, -EIO); | 
|---|
| 117 | } | 
|---|
| 118 | write_lock(&journal->j_state_lock); | 
|---|
| 119 | } else { | 
|---|
| 120 | spin_unlock(lock: &journal->j_list_lock); | 
|---|
| 121 | } | 
|---|
| 122 | mutex_unlock(lock: &journal->j_checkpoint_mutex); | 
|---|
| 123 | } | 
|---|
| 124 | } | 
|---|
| 125 |  | 
|---|
| 126 | static void | 
|---|
| 127 | __flush_batch(journal_t *journal, int *batch_count) | 
|---|
| 128 | { | 
|---|
| 129 | int i; | 
|---|
| 130 | struct blk_plug plug; | 
|---|
| 131 |  | 
|---|
| 132 | blk_start_plug(&plug); | 
|---|
| 133 | for (i = 0; i < *batch_count; i++) | 
|---|
| 134 | write_dirty_buffer(bh: journal->j_chkpt_bhs[i], JBD2_JOURNAL_REQ_FLAGS); | 
|---|
| 135 | blk_finish_plug(&plug); | 
|---|
| 136 |  | 
|---|
| 137 | for (i = 0; i < *batch_count; i++) { | 
|---|
| 138 | struct buffer_head *bh = journal->j_chkpt_bhs[i]; | 
|---|
| 139 | BUFFER_TRACE(bh, "brelse"); | 
|---|
| 140 | __brelse(bh); | 
|---|
| 141 | journal->j_chkpt_bhs[i] = NULL; | 
|---|
| 142 | } | 
|---|
| 143 | *batch_count = 0; | 
|---|
| 144 | } | 
|---|
| 145 |  | 
|---|
| 146 | /* | 
|---|
| 147 | * Perform an actual checkpoint. We take the first transaction on the | 
|---|
| 148 | * list of transactions to be checkpointed and send all its buffers | 
|---|
| 149 | * to disk. We submit larger chunks of data at once. | 
|---|
| 150 | * | 
|---|
| 151 | * The journal should be locked before calling this function. | 
|---|
| 152 | * Called with j_checkpoint_mutex held. | 
|---|
| 153 | */ | 
|---|
| 154 | int jbd2_log_do_checkpoint(journal_t *journal) | 
|---|
| 155 | { | 
|---|
| 156 | struct journal_head	*jh; | 
|---|
| 157 | struct buffer_head	*bh; | 
|---|
| 158 | transaction_t		*transaction; | 
|---|
| 159 | tid_t			this_tid; | 
|---|
| 160 | int			result, batch_count = 0; | 
|---|
| 161 |  | 
|---|
| 162 | jbd2_debug(1, "Start checkpoint\n"); | 
|---|
| 163 |  | 
|---|
| 164 | /* | 
|---|
| 165 | * First thing: if there are any transactions in the log which | 
|---|
| 166 | * don't need checkpointing, just eliminate them from the | 
|---|
| 167 | * journal straight away. | 
|---|
| 168 | */ | 
|---|
| 169 | result = jbd2_cleanup_journal_tail(journal); | 
|---|
| 170 | trace_jbd2_checkpoint(journal, result); | 
|---|
| 171 | jbd2_debug(1, "cleanup_journal_tail returned %d\n", result); | 
|---|
| 172 | if (result <= 0) | 
|---|
| 173 | return result; | 
|---|
| 174 |  | 
|---|
| 175 | /* | 
|---|
| 176 | * OK, we need to start writing disk blocks.  Take one transaction | 
|---|
| 177 | * and write it. | 
|---|
| 178 | */ | 
|---|
| 179 | spin_lock(lock: &journal->j_list_lock); | 
|---|
| 180 | if (!journal->j_checkpoint_transactions) | 
|---|
| 181 | goto out; | 
|---|
| 182 | transaction = journal->j_checkpoint_transactions; | 
|---|
| 183 | if (transaction->t_chp_stats.cs_chp_time == 0) | 
|---|
| 184 | transaction->t_chp_stats.cs_chp_time = jiffies; | 
|---|
| 185 | this_tid = transaction->t_tid; | 
|---|
| 186 | restart: | 
|---|
| 187 | /* | 
|---|
| 188 | * If someone cleaned up this transaction while we slept, we're | 
|---|
| 189 | * done (maybe it's a new transaction, but it fell at the same | 
|---|
| 190 | * address). | 
|---|
| 191 | */ | 
|---|
| 192 | if (journal->j_checkpoint_transactions != transaction || | 
|---|
| 193 | transaction->t_tid != this_tid) | 
|---|
| 194 | goto out; | 
|---|
| 195 |  | 
|---|
| 196 | /* checkpoint all of the transaction's buffers */ | 
|---|
| 197 | while (transaction->t_checkpoint_list) { | 
|---|
| 198 | jh = transaction->t_checkpoint_list; | 
|---|
| 199 | bh = jh2bh(jh); | 
|---|
| 200 |  | 
|---|
| 201 | if (jh->b_transaction != NULL) { | 
|---|
| 202 | transaction_t *t = jh->b_transaction; | 
|---|
| 203 | tid_t tid = t->t_tid; | 
|---|
| 204 |  | 
|---|
| 205 | transaction->t_chp_stats.cs_forced_to_close++; | 
|---|
| 206 | spin_unlock(lock: &journal->j_list_lock); | 
|---|
| 207 | if (unlikely(journal->j_flags & JBD2_UNMOUNT)) | 
|---|
| 208 | /* | 
|---|
| 209 | * The journal thread is dead; so | 
|---|
| 210 | * starting and waiting for a commit | 
|---|
| 211 | * to finish will cause us to wait for | 
|---|
| 212 | * a _very_ long time. | 
|---|
| 213 | */ | 
|---|
| 214 | printk(KERN_ERR | 
|---|
| 215 | "JBD2: %s: Waiting for Godot: block %llu\n", | 
|---|
| 216 | journal->j_devname, (unsigned long long) bh->b_blocknr); | 
|---|
| 217 |  | 
|---|
| 218 | if (batch_count) | 
|---|
| 219 | __flush_batch(journal, batch_count: &batch_count); | 
|---|
| 220 | jbd2_log_start_commit(journal, tid); | 
|---|
| 221 | /* | 
|---|
| 222 | * jbd2_journal_commit_transaction() may want | 
|---|
| 223 | * to take the checkpoint_mutex if JBD2_FLUSHED | 
|---|
| 224 | * is set, jbd2_update_log_tail() called by | 
|---|
| 225 | * jbd2_journal_commit_transaction() may also take | 
|---|
| 226 | * checkpoint_mutex.  So we need to temporarily | 
|---|
| 227 | * drop it. | 
|---|
| 228 | */ | 
|---|
| 229 | mutex_unlock(lock: &journal->j_checkpoint_mutex); | 
|---|
| 230 | jbd2_log_wait_commit(journal, tid); | 
|---|
| 231 | mutex_lock_io(lock: &journal->j_checkpoint_mutex); | 
|---|
| 232 | spin_lock(lock: &journal->j_list_lock); | 
|---|
| 233 | goto restart; | 
|---|
| 234 | } | 
|---|
| 235 | if (!trylock_buffer(bh)) { | 
|---|
| 236 | /* | 
|---|
| 237 | * The buffer is locked, it may be writing back, or | 
|---|
| 238 | * flushing out in the last couple of cycles, or | 
|---|
| 239 | * re-adding into a new transaction, need to check | 
|---|
| 240 | * it again until it's unlocked. | 
|---|
| 241 | */ | 
|---|
| 242 | get_bh(bh); | 
|---|
| 243 | spin_unlock(lock: &journal->j_list_lock); | 
|---|
| 244 | wait_on_buffer(bh); | 
|---|
| 245 | /* the journal_head may have gone by now */ | 
|---|
| 246 | BUFFER_TRACE(bh, "brelse"); | 
|---|
| 247 | __brelse(bh); | 
|---|
| 248 | goto retry; | 
|---|
| 249 | } else if (!buffer_dirty(bh)) { | 
|---|
| 250 | unlock_buffer(bh); | 
|---|
| 251 | BUFFER_TRACE(bh, "remove from checkpoint"); | 
|---|
| 252 | /* | 
|---|
| 253 | * If the transaction was released or the checkpoint | 
|---|
| 254 | * list was empty, we're done. | 
|---|
| 255 | */ | 
|---|
| 256 | if (__jbd2_journal_remove_checkpoint(jh) || | 
|---|
| 257 | !transaction->t_checkpoint_list) | 
|---|
| 258 | goto out; | 
|---|
| 259 | } else { | 
|---|
| 260 | unlock_buffer(bh); | 
|---|
| 261 | /* | 
|---|
| 262 | * We are about to write the buffer, it could be | 
|---|
| 263 | * raced by some other transaction shrink or buffer | 
|---|
| 264 | * re-log logic once we release the j_list_lock, | 
|---|
| 265 | * leave it on the checkpoint list and check status | 
|---|
| 266 | * again to make sure it's clean. | 
|---|
| 267 | */ | 
|---|
| 268 | BUFFER_TRACE(bh, "queue"); | 
|---|
| 269 | get_bh(bh); | 
|---|
| 270 | J_ASSERT_BH(bh, !buffer_jwrite(bh)); | 
|---|
| 271 | journal->j_chkpt_bhs[batch_count++] = bh; | 
|---|
| 272 | transaction->t_chp_stats.cs_written++; | 
|---|
| 273 | transaction->t_checkpoint_list = jh->b_cpnext; | 
|---|
| 274 | } | 
|---|
| 275 |  | 
|---|
| 276 | if ((batch_count == JBD2_NR_BATCH) || | 
|---|
| 277 | need_resched() || spin_needbreak(lock: &journal->j_list_lock) || | 
|---|
| 278 | jh2bh(jh: transaction->t_checkpoint_list) == journal->j_chkpt_bhs[0]) | 
|---|
| 279 | goto unlock_and_flush; | 
|---|
| 280 | } | 
|---|
| 281 |  | 
|---|
| 282 | if (batch_count) { | 
|---|
| 283 | unlock_and_flush: | 
|---|
| 284 | spin_unlock(lock: &journal->j_list_lock); | 
|---|
| 285 | retry: | 
|---|
| 286 | if (batch_count) | 
|---|
| 287 | __flush_batch(journal, batch_count: &batch_count); | 
|---|
| 288 | cond_resched(); | 
|---|
| 289 | spin_lock(lock: &journal->j_list_lock); | 
|---|
| 290 | goto restart; | 
|---|
| 291 | } | 
|---|
| 292 |  | 
|---|
| 293 | out: | 
|---|
| 294 | spin_unlock(lock: &journal->j_list_lock); | 
|---|
| 295 | result = jbd2_cleanup_journal_tail(journal); | 
|---|
| 296 |  | 
|---|
| 297 | return (result < 0) ? result : 0; | 
|---|
| 298 | } | 
|---|
| 299 |  | 
|---|
| 300 | /* | 
|---|
| 301 | * Check the list of checkpoint transactions for the journal to see if | 
|---|
| 302 | * we have already got rid of any since the last update of the log tail | 
|---|
| 303 | * in the journal superblock.  If so, we can instantly roll the | 
|---|
| 304 | * superblock forward to remove those transactions from the log. | 
|---|
| 305 | * | 
|---|
| 306 | * Return <0 on error, 0 on success, 1 if there was nothing to clean up. | 
|---|
| 307 | * | 
|---|
| 308 | * Called with the journal lock held. | 
|---|
| 309 | * | 
|---|
| 310 | * This is the only part of the journaling code which really needs to be | 
|---|
| 311 | * aware of transaction aborts.  Checkpointing involves writing to the | 
|---|
| 312 | * main filesystem area rather than to the journal, so it can proceed | 
|---|
| 313 | * even in abort state, but we must not update the super block if | 
|---|
| 314 | * checkpointing may have failed.  Otherwise, we would lose some metadata | 
|---|
| 315 | * buffers which should be written-back to the filesystem. | 
|---|
| 316 | */ | 
|---|
| 317 |  | 
|---|
| 318 | int jbd2_cleanup_journal_tail(journal_t *journal) | 
|---|
| 319 | { | 
|---|
| 320 | tid_t		first_tid; | 
|---|
| 321 | unsigned long	blocknr; | 
|---|
| 322 |  | 
|---|
| 323 | if (is_journal_aborted(journal)) | 
|---|
| 324 | return -EIO; | 
|---|
| 325 |  | 
|---|
| 326 | if (!jbd2_journal_get_log_tail(journal, tid: &first_tid, block: &blocknr)) | 
|---|
| 327 | return 1; | 
|---|
| 328 | J_ASSERT(blocknr != 0); | 
|---|
| 329 |  | 
|---|
| 330 | /* | 
|---|
| 331 | * We need to make sure that any blocks that were recently written out | 
|---|
| 332 | * --- perhaps by jbd2_log_do_checkpoint() --- are flushed out before | 
|---|
| 333 | * we drop the transactions from the journal. It's unlikely this will | 
|---|
| 334 | * be necessary, especially with an appropriately sized journal, but we | 
|---|
| 335 | * need this to guarantee correctness.  Fortunately | 
|---|
| 336 | * jbd2_cleanup_journal_tail() doesn't get called all that often. | 
|---|
| 337 | */ | 
|---|
| 338 | if (journal->j_flags & JBD2_BARRIER) | 
|---|
| 339 | blkdev_issue_flush(bdev: journal->j_fs_dev); | 
|---|
| 340 |  | 
|---|
| 341 | return __jbd2_update_log_tail(journal, tid: first_tid, block: blocknr); | 
|---|
| 342 | } | 
|---|
| 343 |  | 
|---|
| 344 |  | 
|---|
| 345 | /* Checkpoint list management */ | 
|---|
| 346 |  | 
|---|
| 347 | /* | 
|---|
| 348 | * journal_shrink_one_cp_list | 
|---|
| 349 | * | 
|---|
| 350 | * Find all the written-back checkpoint buffers in the given list | 
|---|
| 351 | * and try to release them. If the whole transaction is released, set | 
|---|
| 352 | * the 'released' parameter. Return the number of released checkpointed | 
|---|
| 353 | * buffers. | 
|---|
| 354 | * | 
|---|
| 355 | * Called with j_list_lock held. | 
|---|
| 356 | */ | 
|---|
| 357 | static unsigned long journal_shrink_one_cp_list(struct journal_head *jh, | 
|---|
| 358 | enum jbd2_shrink_type type, | 
|---|
| 359 | bool *released) | 
|---|
| 360 | { | 
|---|
| 361 | struct journal_head *last_jh; | 
|---|
| 362 | struct journal_head *next_jh = jh; | 
|---|
| 363 | unsigned long nr_freed = 0; | 
|---|
| 364 | int ret; | 
|---|
| 365 |  | 
|---|
| 366 | *released = false; | 
|---|
| 367 | if (!jh) | 
|---|
| 368 | return 0; | 
|---|
| 369 |  | 
|---|
| 370 | last_jh = jh->b_cpprev; | 
|---|
| 371 | do { | 
|---|
| 372 | jh = next_jh; | 
|---|
| 373 | next_jh = jh->b_cpnext; | 
|---|
| 374 |  | 
|---|
| 375 | if (type == JBD2_SHRINK_DESTROY) { | 
|---|
| 376 | ret = __jbd2_journal_remove_checkpoint(jh); | 
|---|
| 377 | } else { | 
|---|
| 378 | ret = jbd2_journal_try_remove_checkpoint(jh); | 
|---|
| 379 | if (ret < 0) { | 
|---|
| 380 | if (type == JBD2_SHRINK_BUSY_SKIP) | 
|---|
| 381 | continue; | 
|---|
| 382 | break; | 
|---|
| 383 | } | 
|---|
| 384 | } | 
|---|
| 385 |  | 
|---|
| 386 | nr_freed++; | 
|---|
| 387 | if (ret) { | 
|---|
| 388 | *released = true; | 
|---|
| 389 | break; | 
|---|
| 390 | } | 
|---|
| 391 |  | 
|---|
| 392 | if (need_resched()) | 
|---|
| 393 | break; | 
|---|
| 394 | } while (jh != last_jh); | 
|---|
| 395 |  | 
|---|
| 396 | return nr_freed; | 
|---|
| 397 | } | 
|---|
| 398 |  | 
|---|
| 399 | /* | 
|---|
| 400 | * jbd2_journal_shrink_checkpoint_list | 
|---|
| 401 | * | 
|---|
| 402 | * Find 'nr_to_scan' written-back checkpoint buffers in the journal | 
|---|
| 403 | * and try to release them. Return the number of released checkpointed | 
|---|
| 404 | * buffers. | 
|---|
| 405 | * | 
|---|
| 406 | * Called with j_list_lock held. | 
|---|
| 407 | */ | 
|---|
| 408 | unsigned long jbd2_journal_shrink_checkpoint_list(journal_t *journal, | 
|---|
| 409 | unsigned long *nr_to_scan) | 
|---|
| 410 | { | 
|---|
| 411 | transaction_t *transaction, *last_transaction, *next_transaction; | 
|---|
| 412 | bool __maybe_unused released; | 
|---|
| 413 | tid_t first_tid = 0, last_tid = 0, next_tid = 0; | 
|---|
| 414 | tid_t tid = 0; | 
|---|
| 415 | unsigned long nr_freed = 0; | 
|---|
| 416 | unsigned long freed; | 
|---|
| 417 | bool first_set = false; | 
|---|
| 418 |  | 
|---|
| 419 | again: | 
|---|
| 420 | spin_lock(lock: &journal->j_list_lock); | 
|---|
| 421 | if (!journal->j_checkpoint_transactions) { | 
|---|
| 422 | spin_unlock(lock: &journal->j_list_lock); | 
|---|
| 423 | goto out; | 
|---|
| 424 | } | 
|---|
| 425 |  | 
|---|
| 426 | /* | 
|---|
| 427 | * Get next shrink transaction, resume previous scan or start | 
|---|
| 428 | * over again. If some others do checkpoint and drop transaction | 
|---|
| 429 | * from the checkpoint list, we ignore saved j_shrink_transaction | 
|---|
| 430 | * and start over unconditionally. | 
|---|
| 431 | */ | 
|---|
| 432 | if (journal->j_shrink_transaction) | 
|---|
| 433 | transaction = journal->j_shrink_transaction; | 
|---|
| 434 | else | 
|---|
| 435 | transaction = journal->j_checkpoint_transactions; | 
|---|
| 436 |  | 
|---|
| 437 | if (!first_set) { | 
|---|
| 438 | first_tid = transaction->t_tid; | 
|---|
| 439 | first_set = true; | 
|---|
| 440 | } | 
|---|
| 441 | last_transaction = journal->j_checkpoint_transactions->t_cpprev; | 
|---|
| 442 | next_transaction = transaction; | 
|---|
| 443 | last_tid = last_transaction->t_tid; | 
|---|
| 444 | do { | 
|---|
| 445 | transaction = next_transaction; | 
|---|
| 446 | next_transaction = transaction->t_cpnext; | 
|---|
| 447 | tid = transaction->t_tid; | 
|---|
| 448 |  | 
|---|
| 449 | freed = journal_shrink_one_cp_list(jh: transaction->t_checkpoint_list, | 
|---|
| 450 | type: JBD2_SHRINK_BUSY_SKIP, released: &released); | 
|---|
| 451 | nr_freed += freed; | 
|---|
| 452 | (*nr_to_scan) -= min(*nr_to_scan, freed); | 
|---|
| 453 | if (*nr_to_scan == 0) | 
|---|
| 454 | break; | 
|---|
| 455 | if (need_resched() || spin_needbreak(lock: &journal->j_list_lock)) | 
|---|
| 456 | break; | 
|---|
| 457 | } while (transaction != last_transaction); | 
|---|
| 458 |  | 
|---|
| 459 | if (transaction != last_transaction) { | 
|---|
| 460 | journal->j_shrink_transaction = next_transaction; | 
|---|
| 461 | next_tid = next_transaction->t_tid; | 
|---|
| 462 | } else { | 
|---|
| 463 | journal->j_shrink_transaction = NULL; | 
|---|
| 464 | next_tid = 0; | 
|---|
| 465 | } | 
|---|
| 466 |  | 
|---|
| 467 | spin_unlock(lock: &journal->j_list_lock); | 
|---|
| 468 | cond_resched(); | 
|---|
| 469 |  | 
|---|
| 470 | if (*nr_to_scan && journal->j_shrink_transaction) | 
|---|
| 471 | goto again; | 
|---|
| 472 | out: | 
|---|
| 473 | trace_jbd2_shrink_checkpoint_list(journal, first_tid, tid, last_tid, | 
|---|
| 474 | nr_freed, next_tid); | 
|---|
| 475 |  | 
|---|
| 476 | return nr_freed; | 
|---|
| 477 | } | 
|---|
| 478 |  | 
|---|
| 479 | /* | 
|---|
| 480 | * journal_clean_checkpoint_list | 
|---|
| 481 | * | 
|---|
| 482 | * Find all the written-back checkpoint buffers in the journal and release them. | 
|---|
| 483 | * If 'type' is JBD2_SHRINK_DESTROY, release all buffers unconditionally. If | 
|---|
| 484 | * 'type' is JBD2_SHRINK_BUSY_STOP, will stop release buffers if encounters a | 
|---|
| 485 | * busy buffer. To avoid wasting CPU cycles scanning the buffer list in some | 
|---|
| 486 | * cases, don't pass JBD2_SHRINK_BUSY_SKIP 'type' for this function. | 
|---|
| 487 | * | 
|---|
| 488 | * Called with j_list_lock held. | 
|---|
| 489 | */ | 
|---|
| 490 | void __jbd2_journal_clean_checkpoint_list(journal_t *journal, | 
|---|
| 491 | enum jbd2_shrink_type type) | 
|---|
| 492 | { | 
|---|
| 493 | transaction_t *transaction, *last_transaction, *next_transaction; | 
|---|
| 494 | bool released; | 
|---|
| 495 |  | 
|---|
| 496 | WARN_ON_ONCE(type == JBD2_SHRINK_BUSY_SKIP); | 
|---|
| 497 |  | 
|---|
| 498 | transaction = journal->j_checkpoint_transactions; | 
|---|
| 499 | if (!transaction) | 
|---|
| 500 | return; | 
|---|
| 501 |  | 
|---|
| 502 | last_transaction = transaction->t_cpprev; | 
|---|
| 503 | next_transaction = transaction; | 
|---|
| 504 | do { | 
|---|
| 505 | transaction = next_transaction; | 
|---|
| 506 | next_transaction = transaction->t_cpnext; | 
|---|
| 507 | journal_shrink_one_cp_list(jh: transaction->t_checkpoint_list, | 
|---|
| 508 | type, released: &released); | 
|---|
| 509 | /* | 
|---|
| 510 | * This function only frees up some memory if possible so we | 
|---|
| 511 | * dont have an obligation to finish processing. Bail out if | 
|---|
| 512 | * preemption requested: | 
|---|
| 513 | */ | 
|---|
| 514 | if (need_resched()) | 
|---|
| 515 | return; | 
|---|
| 516 | /* | 
|---|
| 517 | * Stop scanning if we couldn't free the transaction. This | 
|---|
| 518 | * avoids pointless scanning of transactions which still | 
|---|
| 519 | * weren't checkpointed. | 
|---|
| 520 | */ | 
|---|
| 521 | if (!released) | 
|---|
| 522 | return; | 
|---|
| 523 | } while (transaction != last_transaction); | 
|---|
| 524 | } | 
|---|
| 525 |  | 
|---|
| 526 | /* | 
|---|
| 527 | * Remove buffers from all checkpoint lists as journal is aborted and we just | 
|---|
| 528 | * need to free memory | 
|---|
| 529 | */ | 
|---|
| 530 | void jbd2_journal_destroy_checkpoint(journal_t *journal) | 
|---|
| 531 | { | 
|---|
| 532 | /* | 
|---|
| 533 | * We loop because __jbd2_journal_clean_checkpoint_list() may abort | 
|---|
| 534 | * early due to a need of rescheduling. | 
|---|
| 535 | */ | 
|---|
| 536 | while (1) { | 
|---|
| 537 | spin_lock(lock: &journal->j_list_lock); | 
|---|
| 538 | if (!journal->j_checkpoint_transactions) { | 
|---|
| 539 | spin_unlock(lock: &journal->j_list_lock); | 
|---|
| 540 | break; | 
|---|
| 541 | } | 
|---|
| 542 | __jbd2_journal_clean_checkpoint_list(journal, type: JBD2_SHRINK_DESTROY); | 
|---|
| 543 | spin_unlock(lock: &journal->j_list_lock); | 
|---|
| 544 | cond_resched(); | 
|---|
| 545 | } | 
|---|
| 546 | } | 
|---|
| 547 |  | 
|---|
| 548 | /* | 
|---|
| 549 | * journal_remove_checkpoint: called after a buffer has been committed | 
|---|
| 550 | * to disk (either by being write-back flushed to disk, or being | 
|---|
| 551 | * committed to the log). | 
|---|
| 552 | * | 
|---|
| 553 | * We cannot safely clean a transaction out of the log until all of the | 
|---|
| 554 | * buffer updates committed in that transaction have safely been stored | 
|---|
| 555 | * elsewhere on disk.  To achieve this, all of the buffers in a | 
|---|
| 556 | * transaction need to be maintained on the transaction's checkpoint | 
|---|
| 557 | * lists until they have been rewritten, at which point this function is | 
|---|
| 558 | * called to remove the buffer from the existing transaction's | 
|---|
| 559 | * checkpoint lists. | 
|---|
| 560 | * | 
|---|
| 561 | * The function returns 1 if it frees the transaction, 0 otherwise. | 
|---|
| 562 | * The function can free jh and bh. | 
|---|
| 563 | * | 
|---|
| 564 | * This function is called with j_list_lock held. | 
|---|
| 565 | */ | 
|---|
| 566 | int __jbd2_journal_remove_checkpoint(struct journal_head *jh) | 
|---|
| 567 | { | 
|---|
| 568 | struct transaction_chp_stats_s *stats; | 
|---|
| 569 | transaction_t *transaction; | 
|---|
| 570 | journal_t *journal; | 
|---|
| 571 |  | 
|---|
| 572 | JBUFFER_TRACE(jh, "entry"); | 
|---|
| 573 |  | 
|---|
| 574 | transaction = jh->b_cp_transaction; | 
|---|
| 575 | if (!transaction) { | 
|---|
| 576 | JBUFFER_TRACE(jh, "not on transaction"); | 
|---|
| 577 | return 0; | 
|---|
| 578 | } | 
|---|
| 579 | journal = transaction->t_journal; | 
|---|
| 580 |  | 
|---|
| 581 | JBUFFER_TRACE(jh, "removing from transaction"); | 
|---|
| 582 |  | 
|---|
| 583 | __buffer_unlink(jh); | 
|---|
| 584 | jh->b_cp_transaction = NULL; | 
|---|
| 585 | percpu_counter_dec(fbc: &journal->j_checkpoint_jh_count); | 
|---|
| 586 | jbd2_journal_put_journal_head(jh); | 
|---|
| 587 |  | 
|---|
| 588 | /* Is this transaction empty? */ | 
|---|
| 589 | if (transaction->t_checkpoint_list) | 
|---|
| 590 | return 0; | 
|---|
| 591 |  | 
|---|
| 592 | /* | 
|---|
| 593 | * There is one special case to worry about: if we have just pulled the | 
|---|
| 594 | * buffer off a running or committing transaction's checkpoing list, | 
|---|
| 595 | * then even if the checkpoint list is empty, the transaction obviously | 
|---|
| 596 | * cannot be dropped! | 
|---|
| 597 | * | 
|---|
| 598 | * The locking here around t_state is a bit sleazy. | 
|---|
| 599 | * See the comment at the end of jbd2_journal_commit_transaction(). | 
|---|
| 600 | */ | 
|---|
| 601 | if (transaction->t_state != T_FINISHED) | 
|---|
| 602 | return 0; | 
|---|
| 603 |  | 
|---|
| 604 | /* | 
|---|
| 605 | * OK, that was the last buffer for the transaction, we can now | 
|---|
| 606 | * safely remove this transaction from the log. | 
|---|
| 607 | */ | 
|---|
| 608 | stats = &transaction->t_chp_stats; | 
|---|
| 609 | if (stats->cs_chp_time) | 
|---|
| 610 | stats->cs_chp_time = jbd2_time_diff(start: stats->cs_chp_time, | 
|---|
| 611 | end: jiffies); | 
|---|
| 612 | trace_jbd2_checkpoint_stats(dev: journal->j_fs_dev->bd_dev, | 
|---|
| 613 | tid: transaction->t_tid, stats); | 
|---|
| 614 |  | 
|---|
| 615 | __jbd2_journal_drop_transaction(journal, transaction); | 
|---|
| 616 | jbd2_journal_free_transaction(transaction); | 
|---|
| 617 | return 1; | 
|---|
| 618 | } | 
|---|
| 619 |  | 
|---|
| 620 | /* | 
|---|
| 621 | * Check the checkpoint buffer and try to remove it from the checkpoint | 
|---|
| 622 | * list if it's clean. Returns -EBUSY if it is not clean, returns 1 if | 
|---|
| 623 | * it frees the transaction, 0 otherwise. | 
|---|
| 624 | * | 
|---|
| 625 | * This function is called with j_list_lock held. | 
|---|
| 626 | */ | 
|---|
| 627 | int jbd2_journal_try_remove_checkpoint(struct journal_head *jh) | 
|---|
| 628 | { | 
|---|
| 629 | struct buffer_head *bh = jh2bh(jh); | 
|---|
| 630 |  | 
|---|
| 631 | if (jh->b_transaction) | 
|---|
| 632 | return -EBUSY; | 
|---|
| 633 | if (!trylock_buffer(bh)) | 
|---|
| 634 | return -EBUSY; | 
|---|
| 635 | if (buffer_dirty(bh)) { | 
|---|
| 636 | unlock_buffer(bh); | 
|---|
| 637 | return -EBUSY; | 
|---|
| 638 | } | 
|---|
| 639 | unlock_buffer(bh); | 
|---|
| 640 |  | 
|---|
| 641 | /* | 
|---|
| 642 | * Buffer is clean and the IO has finished (we held the buffer | 
|---|
| 643 | * lock) so the checkpoint is done. We can safely remove the | 
|---|
| 644 | * buffer from this transaction. | 
|---|
| 645 | */ | 
|---|
| 646 | JBUFFER_TRACE(jh, "remove from checkpoint list"); | 
|---|
| 647 | return __jbd2_journal_remove_checkpoint(jh); | 
|---|
| 648 | } | 
|---|
| 649 |  | 
|---|
| 650 | /* | 
|---|
| 651 | * journal_insert_checkpoint: put a committed buffer onto a checkpoint | 
|---|
| 652 | * list so that we know when it is safe to clean the transaction out of | 
|---|
| 653 | * the log. | 
|---|
| 654 | * | 
|---|
| 655 | * Called with the journal locked. | 
|---|
| 656 | * Called with j_list_lock held. | 
|---|
| 657 | */ | 
|---|
| 658 | void __jbd2_journal_insert_checkpoint(struct journal_head *jh, | 
|---|
| 659 | transaction_t *transaction) | 
|---|
| 660 | { | 
|---|
| 661 | JBUFFER_TRACE(jh, "entry"); | 
|---|
| 662 | J_ASSERT_JH(jh, buffer_dirty(jh2bh(jh)) || buffer_jbddirty(jh2bh(jh))); | 
|---|
| 663 | J_ASSERT_JH(jh, jh->b_cp_transaction == NULL); | 
|---|
| 664 |  | 
|---|
| 665 | /* Get reference for checkpointing transaction */ | 
|---|
| 666 | jbd2_journal_grab_journal_head(bh: jh2bh(jh)); | 
|---|
| 667 | jh->b_cp_transaction = transaction; | 
|---|
| 668 |  | 
|---|
| 669 | if (!transaction->t_checkpoint_list) { | 
|---|
| 670 | jh->b_cpnext = jh->b_cpprev = jh; | 
|---|
| 671 | } else { | 
|---|
| 672 | jh->b_cpnext = transaction->t_checkpoint_list; | 
|---|
| 673 | jh->b_cpprev = transaction->t_checkpoint_list->b_cpprev; | 
|---|
| 674 | jh->b_cpprev->b_cpnext = jh; | 
|---|
| 675 | jh->b_cpnext->b_cpprev = jh; | 
|---|
| 676 | } | 
|---|
| 677 | transaction->t_checkpoint_list = jh; | 
|---|
| 678 | percpu_counter_inc(fbc: &transaction->t_journal->j_checkpoint_jh_count); | 
|---|
| 679 | } | 
|---|
| 680 |  | 
|---|
| 681 | /* | 
|---|
| 682 | * We've finished with this transaction structure: adios... | 
|---|
| 683 | * | 
|---|
| 684 | * The transaction must have no links except for the checkpoint by this | 
|---|
| 685 | * point. | 
|---|
| 686 | * | 
|---|
| 687 | * Called with the journal locked. | 
|---|
| 688 | * Called with j_list_lock held. | 
|---|
| 689 | */ | 
|---|
| 690 |  | 
|---|
| 691 | void __jbd2_journal_drop_transaction(journal_t *journal, transaction_t *transaction) | 
|---|
| 692 | { | 
|---|
| 693 | assert_spin_locked(&journal->j_list_lock); | 
|---|
| 694 |  | 
|---|
| 695 | journal->j_shrink_transaction = NULL; | 
|---|
| 696 | if (transaction->t_cpnext) { | 
|---|
| 697 | transaction->t_cpnext->t_cpprev = transaction->t_cpprev; | 
|---|
| 698 | transaction->t_cpprev->t_cpnext = transaction->t_cpnext; | 
|---|
| 699 | if (journal->j_checkpoint_transactions == transaction) | 
|---|
| 700 | journal->j_checkpoint_transactions = | 
|---|
| 701 | transaction->t_cpnext; | 
|---|
| 702 | if (journal->j_checkpoint_transactions == transaction) | 
|---|
| 703 | journal->j_checkpoint_transactions = NULL; | 
|---|
| 704 | } | 
|---|
| 705 |  | 
|---|
| 706 | J_ASSERT(transaction->t_state == T_FINISHED); | 
|---|
| 707 | J_ASSERT(transaction->t_buffers == NULL); | 
|---|
| 708 | J_ASSERT(transaction->t_forget == NULL); | 
|---|
| 709 | J_ASSERT(transaction->t_shadow_list == NULL); | 
|---|
| 710 | J_ASSERT(transaction->t_checkpoint_list == NULL); | 
|---|
| 711 | J_ASSERT(atomic_read(&transaction->t_updates) == 0); | 
|---|
| 712 | J_ASSERT(journal->j_committing_transaction != transaction); | 
|---|
| 713 | J_ASSERT(journal->j_running_transaction != transaction); | 
|---|
| 714 |  | 
|---|
| 715 | trace_jbd2_drop_transaction(journal, commit_transaction: transaction); | 
|---|
| 716 |  | 
|---|
| 717 | jbd2_debug(1, "Dropping transaction %d, all done\n", transaction->t_tid); | 
|---|
| 718 | } | 
|---|
| 719 |  | 
|---|