~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/jbd2/transaction.c

Version: ~ [ linux-5.13-rc7 ] ~ [ linux-5.12.12 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.45 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.127 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.195 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.237 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.273 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.273 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 // SPDX-License-Identifier: GPL-2.0+
  2 /*
  3  * linux/fs/jbd2/transaction.c
  4  *
  5  * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
  6  *
  7  * Copyright 1998 Red Hat corp --- All Rights Reserved
  8  *
  9  * Generic filesystem transaction handling code; part of the ext2fs
 10  * journaling system.
 11  *
 12  * This file manages transactions (compound commits managed by the
 13  * journaling code) and handles (individual atomic operations by the
 14  * filesystem).
 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/timer.h>
 23 #include <linux/mm.h>
 24 #include <linux/highmem.h>
 25 #include <linux/hrtimer.h>
 26 #include <linux/backing-dev.h>
 27 #include <linux/bug.h>
 28 #include <linux/module.h>
 29 #include <linux/sched/mm.h>
 30 
 31 #include <trace/events/jbd2.h>
 32 
 33 static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh);
 34 static void __jbd2_journal_unfile_buffer(struct journal_head *jh);
 35 
 36 static struct kmem_cache *transaction_cache;
 37 int __init jbd2_journal_init_transaction_cache(void)
 38 {
 39         J_ASSERT(!transaction_cache);
 40         transaction_cache = kmem_cache_create("jbd2_transaction_s",
 41                                         sizeof(transaction_t),
 42                                         0,
 43                                         SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY,
 44                                         NULL);
 45         if (transaction_cache)
 46                 return 0;
 47         return -ENOMEM;
 48 }
 49 
 50 void jbd2_journal_destroy_transaction_cache(void)
 51 {
 52         if (transaction_cache) {
 53                 kmem_cache_destroy(transaction_cache);
 54                 transaction_cache = NULL;
 55         }
 56 }
 57 
 58 void jbd2_journal_free_transaction(transaction_t *transaction)
 59 {
 60         if (unlikely(ZERO_OR_NULL_PTR(transaction)))
 61                 return;
 62         kmem_cache_free(transaction_cache, transaction);
 63 }
 64 
 65 /*
 66  * jbd2_get_transaction: obtain a new transaction_t object.
 67  *
 68  * Simply allocate and initialise a new transaction.  Create it in
 69  * RUNNING state and add it to the current journal (which should not
 70  * have an existing running transaction: we only make a new transaction
 71  * once we have started to commit the old one).
 72  *
 73  * Preconditions:
 74  *      The journal MUST be locked.  We don't perform atomic mallocs on the
 75  *      new transaction and we can't block without protecting against other
 76  *      processes trying to touch the journal while it is in transition.
 77  *
 78  */
 79 
 80 static transaction_t *
 81 jbd2_get_transaction(journal_t *journal, transaction_t *transaction)
 82 {
 83         transaction->t_journal = journal;
 84         transaction->t_state = T_RUNNING;
 85         transaction->t_start_time = ktime_get();
 86         transaction->t_tid = journal->j_transaction_sequence++;
 87         transaction->t_expires = jiffies + journal->j_commit_interval;
 88         spin_lock_init(&transaction->t_handle_lock);
 89         atomic_set(&transaction->t_updates, 0);
 90         atomic_set(&transaction->t_outstanding_credits,
 91                    atomic_read(&journal->j_reserved_credits));
 92         atomic_set(&transaction->t_handle_count, 0);
 93         INIT_LIST_HEAD(&transaction->t_inode_list);
 94         INIT_LIST_HEAD(&transaction->t_private_list);
 95 
 96         /* Set up the commit timer for the new transaction. */
 97         journal->j_commit_timer.expires = round_jiffies_up(transaction->t_expires);
 98         add_timer(&journal->j_commit_timer);
 99 
100         J_ASSERT(journal->j_running_transaction == NULL);
101         journal->j_running_transaction = transaction;
102         transaction->t_max_wait = 0;
103         transaction->t_start = jiffies;
104         transaction->t_requested = 0;
105 
106         return transaction;
107 }
108 
109 /*
110  * Handle management.
111  *
112  * A handle_t is an object which represents a single atomic update to a
113  * filesystem, and which tracks all of the modifications which form part
114  * of that one update.
115  */
116 
117 /*
118  * Update transaction's maximum wait time, if debugging is enabled.
119  *
120  * In order for t_max_wait to be reliable, it must be protected by a
121  * lock.  But doing so will mean that start_this_handle() can not be
122  * run in parallel on SMP systems, which limits our scalability.  So
123  * unless debugging is enabled, we no longer update t_max_wait, which
124  * means that maximum wait time reported by the jbd2_run_stats
125  * tracepoint will always be zero.
126  */
127 static inline void update_t_max_wait(transaction_t *transaction,
128                                      unsigned long ts)
129 {
130 #ifdef CONFIG_JBD2_DEBUG
131         if (jbd2_journal_enable_debug &&
132             time_after(transaction->t_start, ts)) {
133                 ts = jbd2_time_diff(ts, transaction->t_start);
134                 spin_lock(&transaction->t_handle_lock);
135                 if (ts > transaction->t_max_wait)
136                         transaction->t_max_wait = ts;
137                 spin_unlock(&transaction->t_handle_lock);
138         }
139 #endif
140 }
141 
142 /*
143  * Wait until running transaction passes T_LOCKED state. Also starts the commit
144  * if needed. The function expects running transaction to exist and releases
145  * j_state_lock.
146  */
147 static void wait_transaction_locked(journal_t *journal)
148         __releases(journal->j_state_lock)
149 {
150         DEFINE_WAIT(wait);
151         int need_to_start;
152         tid_t tid = journal->j_running_transaction->t_tid;
153 
154         prepare_to_wait(&journal->j_wait_transaction_locked, &wait,
155                         TASK_UNINTERRUPTIBLE);
156         need_to_start = !tid_geq(journal->j_commit_request, tid);
157         read_unlock(&journal->j_state_lock);
158         if (need_to_start)
159                 jbd2_log_start_commit(journal, tid);
160         jbd2_might_wait_for_commit(journal);
161         schedule();
162         finish_wait(&journal->j_wait_transaction_locked, &wait);
163 }
164 
165 static void sub_reserved_credits(journal_t *journal, int blocks)
166 {
167         atomic_sub(blocks, &journal->j_reserved_credits);
168         wake_up(&journal->j_wait_reserved);
169 }
170 
171 /*
172  * Wait until we can add credits for handle to the running transaction.  Called
173  * with j_state_lock held for reading. Returns 0 if handle joined the running
174  * transaction. Returns 1 if we had to wait, j_state_lock is dropped, and
175  * caller must retry.
176  */
177 static int add_transaction_credits(journal_t *journal, int blocks,
178                                    int rsv_blocks)
179 {
180         transaction_t *t = journal->j_running_transaction;
181         int needed;
182         int total = blocks + rsv_blocks;
183 
184         /*
185          * If the current transaction is locked down for commit, wait
186          * for the lock to be released.
187          */
188         if (t->t_state == T_LOCKED) {
189                 wait_transaction_locked(journal);
190                 return 1;
191         }
192 
193         /*
194          * If there is not enough space left in the log to write all
195          * potential buffers requested by this operation, we need to
196          * stall pending a log checkpoint to free some more log space.
197          */
198         needed = atomic_add_return(total, &t->t_outstanding_credits);
199         if (needed > journal->j_max_transaction_buffers) {
200                 /*
201                  * If the current transaction is already too large,
202                  * then start to commit it: we can then go back and
203                  * attach this handle to a new transaction.
204                  */
205                 atomic_sub(total, &t->t_outstanding_credits);
206 
207                 /*
208                  * Is the number of reserved credits in the current transaction too
209                  * big to fit this handle? Wait until reserved credits are freed.
210                  */
211                 if (atomic_read(&journal->j_reserved_credits) + total >
212                     journal->j_max_transaction_buffers) {
213                         read_unlock(&journal->j_state_lock);
214                         jbd2_might_wait_for_commit(journal);
215                         wait_event(journal->j_wait_reserved,
216                                    atomic_read(&journal->j_reserved_credits) + total <=
217                                    journal->j_max_transaction_buffers);
218                         return 1;
219                 }
220 
221                 wait_transaction_locked(journal);
222                 return 1;
223         }
224 
225         /*
226          * The commit code assumes that it can get enough log space
227          * without forcing a checkpoint.  This is *critical* for
228          * correctness: a checkpoint of a buffer which is also
229          * associated with a committing transaction creates a deadlock,
230          * so commit simply cannot force through checkpoints.
231          *
232          * We must therefore ensure the necessary space in the journal
233          * *before* starting to dirty potentially checkpointed buffers
234          * in the new transaction.
235          */
236         if (jbd2_log_space_left(journal) < jbd2_space_needed(journal)) {
237                 atomic_sub(total, &t->t_outstanding_credits);
238                 read_unlock(&journal->j_state_lock);
239                 jbd2_might_wait_for_commit(journal);
240                 write_lock(&journal->j_state_lock);
241                 if (jbd2_log_space_left(journal) < jbd2_space_needed(journal))
242                         __jbd2_log_wait_for_space(journal);
243                 write_unlock(&journal->j_state_lock);
244                 return 1;
245         }
246 
247         /* No reservation? We are done... */
248         if (!rsv_blocks)
249                 return 0;
250 
251         needed = atomic_add_return(rsv_blocks, &journal->j_reserved_credits);
252         /* We allow at most half of a transaction to be reserved */
253         if (needed > journal->j_max_transaction_buffers / 2) {
254                 sub_reserved_credits(journal, rsv_blocks);
255                 atomic_sub(total, &t->t_outstanding_credits);
256                 read_unlock(&journal->j_state_lock);
257                 jbd2_might_wait_for_commit(journal);
258                 wait_event(journal->j_wait_reserved,
259                          atomic_read(&journal->j_reserved_credits) + rsv_blocks
260                          <= journal->j_max_transaction_buffers / 2);
261                 return 1;
262         }
263         return 0;
264 }
265 
266 /*
267  * start_this_handle: Given a handle, deal with any locking or stalling
268  * needed to make sure that there is enough journal space for the handle
269  * to begin.  Attach the handle to a transaction and set up the
270  * transaction's buffer credits.
271  */
272 
273 static int start_this_handle(journal_t *journal, handle_t *handle,
274                              gfp_t gfp_mask)
275 {
276         transaction_t   *transaction, *new_transaction = NULL;
277         int             blocks = handle->h_buffer_credits;
278         int             rsv_blocks = 0;
279         unsigned long ts = jiffies;
280 
281         if (handle->h_rsv_handle)
282                 rsv_blocks = handle->h_rsv_handle->h_buffer_credits;
283 
284         /*
285          * Limit the number of reserved credits to 1/2 of maximum transaction
286          * size and limit the number of total credits to not exceed maximum
287          * transaction size per operation.
288          */
289         if ((rsv_blocks > journal->j_max_transaction_buffers / 2) ||
290             (rsv_blocks + blocks > journal->j_max_transaction_buffers)) {
291                 printk(KERN_ERR "JBD2: %s wants too many credits "
292                        "credits:%d rsv_credits:%d max:%d\n",
293                        current->comm, blocks, rsv_blocks,
294                        journal->j_max_transaction_buffers);
295                 WARN_ON(1);
296                 return -ENOSPC;
297         }
298 
299 alloc_transaction:
300         if (!journal->j_running_transaction) {
301                 /*
302                  * If __GFP_FS is not present, then we may be being called from
303                  * inside the fs writeback layer, so we MUST NOT fail.
304                  */
305                 if ((gfp_mask & __GFP_FS) == 0)
306                         gfp_mask |= __GFP_NOFAIL;
307                 new_transaction = kmem_cache_zalloc(transaction_cache,
308                                                     gfp_mask);
309                 if (!new_transaction)
310                         return -ENOMEM;
311         }
312 
313         jbd_debug(3, "New handle %p going live.\n", handle);
314 
315         /*
316          * We need to hold j_state_lock until t_updates has been incremented,
317          * for proper journal barrier handling
318          */
319 repeat:
320         read_lock(&journal->j_state_lock);
321         BUG_ON(journal->j_flags & JBD2_UNMOUNT);
322         if (is_journal_aborted(journal) ||
323             (journal->j_errno != 0 && !(journal->j_flags & JBD2_ACK_ERR))) {
324                 read_unlock(&journal->j_state_lock);
325                 jbd2_journal_free_transaction(new_transaction);
326                 return -EROFS;
327         }
328 
329         /*
330          * Wait on the journal's transaction barrier if necessary. Specifically
331          * we allow reserved handles to proceed because otherwise commit could
332          * deadlock on page writeback not being able to complete.
333          */
334         if (!handle->h_reserved && journal->j_barrier_count) {
335                 read_unlock(&journal->j_state_lock);
336                 wait_event(journal->j_wait_transaction_locked,
337                                 journal->j_barrier_count == 0);
338                 goto repeat;
339         }
340 
341         if (!journal->j_running_transaction) {
342                 read_unlock(&journal->j_state_lock);
343                 if (!new_transaction)
344                         goto alloc_transaction;
345                 write_lock(&journal->j_state_lock);
346                 if (!journal->j_running_transaction &&
347                     (handle->h_reserved || !journal->j_barrier_count)) {
348                         jbd2_get_transaction(journal, new_transaction);
349                         new_transaction = NULL;
350                 }
351                 write_unlock(&journal->j_state_lock);
352                 goto repeat;
353         }
354 
355         transaction = journal->j_running_transaction;
356 
357         if (!handle->h_reserved) {
358                 /* We may have dropped j_state_lock - restart in that case */
359                 if (add_transaction_credits(journal, blocks, rsv_blocks))
360                         goto repeat;
361         } else {
362                 /*
363                  * We have handle reserved so we are allowed to join T_LOCKED
364                  * transaction and we don't have to check for transaction size
365                  * and journal space.
366                  */
367                 sub_reserved_credits(journal, blocks);
368                 handle->h_reserved = 0;
369         }
370 
371         /* OK, account for the buffers that this operation expects to
372          * use and add the handle to the running transaction. 
373          */
374         update_t_max_wait(transaction, ts);
375         handle->h_transaction = transaction;
376         handle->h_requested_credits = blocks;
377         handle->h_start_jiffies = jiffies;
378         atomic_inc(&transaction->t_updates);
379         atomic_inc(&transaction->t_handle_count);
380         jbd_debug(4, "Handle %p given %d credits (total %d, free %lu)\n",
381                   handle, blocks,
382                   atomic_read(&transaction->t_outstanding_credits),
383                   jbd2_log_space_left(journal));
384         read_unlock(&journal->j_state_lock);
385         current->journal_info = handle;
386 
387         rwsem_acquire_read(&journal->j_trans_commit_map, 0, 0, _THIS_IP_);
388         jbd2_journal_free_transaction(new_transaction);
389         /*
390          * Ensure that no allocations done while the transaction is open are
391          * going to recurse back to the fs layer.
392          */
393         handle->saved_alloc_context = memalloc_nofs_save();
394         return 0;
395 }
396 
397 /* Allocate a new handle.  This should probably be in a slab... */
398 static handle_t *new_handle(int nblocks)
399 {
400         handle_t *handle = jbd2_alloc_handle(GFP_NOFS);
401         if (!handle)
402                 return NULL;
403         handle->h_buffer_credits = nblocks;
404         handle->h_ref = 1;
405 
406         return handle;
407 }
408 
409 handle_t *jbd2__journal_start(journal_t *journal, int nblocks, int rsv_blocks,
410                               gfp_t gfp_mask, unsigned int type,
411                               unsigned int line_no)
412 {
413         handle_t *handle = journal_current_handle();
414         int err;
415 
416         if (!journal)
417                 return ERR_PTR(-EROFS);
418 
419         if (handle) {
420                 J_ASSERT(handle->h_transaction->t_journal == journal);
421                 handle->h_ref++;
422                 return handle;
423         }
424 
425         handle = new_handle(nblocks);
426         if (!handle)
427                 return ERR_PTR(-ENOMEM);
428         if (rsv_blocks) {
429                 handle_t *rsv_handle;
430 
431                 rsv_handle = new_handle(rsv_blocks);
432                 if (!rsv_handle) {
433                         jbd2_free_handle(handle);
434                         return ERR_PTR(-ENOMEM);
435                 }
436                 rsv_handle->h_reserved = 1;
437                 rsv_handle->h_journal = journal;
438                 handle->h_rsv_handle = rsv_handle;
439         }
440 
441         err = start_this_handle(journal, handle, gfp_mask);
442         if (err < 0) {
443                 if (handle->h_rsv_handle)
444                         jbd2_free_handle(handle->h_rsv_handle);
445                 jbd2_free_handle(handle);
446                 return ERR_PTR(err);
447         }
448         handle->h_type = type;
449         handle->h_line_no = line_no;
450         trace_jbd2_handle_start(journal->j_fs_dev->bd_dev,
451                                 handle->h_transaction->t_tid, type,
452                                 line_no, nblocks);
453 
454         return handle;
455 }
456 EXPORT_SYMBOL(jbd2__journal_start);
457 
458 
459 /**
460  * handle_t *jbd2_journal_start() - Obtain a new handle.
461  * @journal: Journal to start transaction on.
462  * @nblocks: number of block buffer we might modify
463  *
464  * We make sure that the transaction can guarantee at least nblocks of
465  * modified buffers in the log.  We block until the log can guarantee
466  * that much space. Additionally, if rsv_blocks > 0, we also create another
467  * handle with rsv_blocks reserved blocks in the journal. This handle is
468  * is stored in h_rsv_handle. It is not attached to any particular transaction
469  * and thus doesn't block transaction commit. If the caller uses this reserved
470  * handle, it has to set h_rsv_handle to NULL as otherwise jbd2_journal_stop()
471  * on the parent handle will dispose the reserved one. Reserved handle has to
472  * be converted to a normal handle using jbd2_journal_start_reserved() before
473  * it can be used.
474  *
475  * Return a pointer to a newly allocated handle, or an ERR_PTR() value
476  * on failure.
477  */
478 handle_t *jbd2_journal_start(journal_t *journal, int nblocks)
479 {
480         return jbd2__journal_start(journal, nblocks, 0, GFP_NOFS, 0, 0);
481 }
482 EXPORT_SYMBOL(jbd2_journal_start);
483 
484 void jbd2_journal_free_reserved(handle_t *handle)
485 {
486         journal_t *journal = handle->h_journal;
487 
488         WARN_ON(!handle->h_reserved);
489         sub_reserved_credits(journal, handle->h_buffer_credits);
490         jbd2_free_handle(handle);
491 }
492 EXPORT_SYMBOL(jbd2_journal_free_reserved);
493 
494 /**
495  * int jbd2_journal_start_reserved() - start reserved handle
496  * @handle: handle to start
497  * @type: for handle statistics
498  * @line_no: for handle statistics
499  *
500  * Start handle that has been previously reserved with jbd2_journal_reserve().
501  * This attaches @handle to the running transaction (or creates one if there's
502  * not transaction running). Unlike jbd2_journal_start() this function cannot
503  * block on journal commit, checkpointing, or similar stuff. It can block on
504  * memory allocation or frozen journal though.
505  *
506  * Return 0 on success, non-zero on error - handle is freed in that case.
507  */
508 int jbd2_journal_start_reserved(handle_t *handle, unsigned int type,
509                                 unsigned int line_no)
510 {
511         journal_t *journal = handle->h_journal;
512         int ret = -EIO;
513 
514         if (WARN_ON(!handle->h_reserved)) {
515                 /* Someone passed in normal handle? Just stop it. */
516                 jbd2_journal_stop(handle);
517                 return ret;
518         }
519         /*
520          * Usefulness of mixing of reserved and unreserved handles is
521          * questionable. So far nobody seems to need it so just error out.
522          */
523         if (WARN_ON(current->journal_info)) {
524                 jbd2_journal_free_reserved(handle);
525                 return ret;
526         }
527 
528         handle->h_journal = NULL;
529         /*
530          * GFP_NOFS is here because callers are likely from writeback or
531          * similarly constrained call sites
532          */
533         ret = start_this_handle(journal, handle, GFP_NOFS);
534         if (ret < 0) {
535                 handle->h_journal = journal;
536                 jbd2_journal_free_reserved(handle);
537                 return ret;
538         }
539         handle->h_type = type;
540         handle->h_line_no = line_no;
541         return 0;
542 }
543 EXPORT_SYMBOL(jbd2_journal_start_reserved);
544 
545 /**
546  * int jbd2_journal_extend() - extend buffer credits.
547  * @handle:  handle to 'extend'
548  * @nblocks: nr blocks to try to extend by.
549  *
550  * Some transactions, such as large extends and truncates, can be done
551  * atomically all at once or in several stages.  The operation requests
552  * a credit for a number of buffer modifications in advance, but can
553  * extend its credit if it needs more.
554  *
555  * jbd2_journal_extend tries to give the running handle more buffer credits.
556  * It does not guarantee that allocation - this is a best-effort only.
557  * The calling process MUST be able to deal cleanly with a failure to
558  * extend here.
559  *
560  * Return 0 on success, non-zero on failure.
561  *
562  * return code < 0 implies an error
563  * return code > 0 implies normal transaction-full status.
564  */
565 int jbd2_journal_extend(handle_t *handle, int nblocks)
566 {
567         transaction_t *transaction = handle->h_transaction;
568         journal_t *journal;
569         int result;
570         int wanted;
571 
572         if (is_handle_aborted(handle))
573                 return -EROFS;
574         journal = transaction->t_journal;
575 
576         result = 1;
577 
578         read_lock(&journal->j_state_lock);
579 
580         /* Don't extend a locked-down transaction! */
581         if (transaction->t_state != T_RUNNING) {
582                 jbd_debug(3, "denied handle %p %d blocks: "
583                           "transaction not running\n", handle, nblocks);
584                 goto error_out;
585         }
586 
587         spin_lock(&transaction->t_handle_lock);
588         wanted = atomic_add_return(nblocks,
589                                    &transaction->t_outstanding_credits);
590 
591         if (wanted > journal->j_max_transaction_buffers) {
592                 jbd_debug(3, "denied handle %p %d blocks: "
593                           "transaction too large\n", handle, nblocks);
594                 atomic_sub(nblocks, &transaction->t_outstanding_credits);
595                 goto unlock;
596         }
597 
598         if (wanted + (wanted >> JBD2_CONTROL_BLOCKS_SHIFT) >
599             jbd2_log_space_left(journal)) {
600                 jbd_debug(3, "denied handle %p %d blocks: "
601                           "insufficient log space\n", handle, nblocks);
602                 atomic_sub(nblocks, &transaction->t_outstanding_credits);
603                 goto unlock;
604         }
605 
606         trace_jbd2_handle_extend(journal->j_fs_dev->bd_dev,
607                                  transaction->t_tid,
608                                  handle->h_type, handle->h_line_no,
609                                  handle->h_buffer_credits,
610                                  nblocks);
611 
612         handle->h_buffer_credits += nblocks;
613         handle->h_requested_credits += nblocks;
614         result = 0;
615 
616         jbd_debug(3, "extended handle %p by %d\n", handle, nblocks);
617 unlock:
618         spin_unlock(&transaction->t_handle_lock);
619 error_out:
620         read_unlock(&journal->j_state_lock);
621         return result;
622 }
623 
624 
625 /**
626  * int jbd2_journal_restart() - restart a handle .
627  * @handle:  handle to restart
628  * @nblocks: nr credits requested
629  * @gfp_mask: memory allocation flags (for start_this_handle)
630  *
631  * Restart a handle for a multi-transaction filesystem
632  * operation.
633  *
634  * If the jbd2_journal_extend() call above fails to grant new buffer credits
635  * to a running handle, a call to jbd2_journal_restart will commit the
636  * handle's transaction so far and reattach the handle to a new
637  * transaction capable of guaranteeing the requested number of
638  * credits. We preserve reserved handle if there's any attached to the
639  * passed in handle.
640  */
641 int jbd2__journal_restart(handle_t *handle, int nblocks, gfp_t gfp_mask)
642 {
643         transaction_t *transaction = handle->h_transaction;
644         journal_t *journal;
645         tid_t           tid;
646         int             need_to_start, ret;
647 
648         /* If we've had an abort of any type, don't even think about
649          * actually doing the restart! */
650         if (is_handle_aborted(handle))
651                 return 0;
652         journal = transaction->t_journal;
653 
654         /*
655          * First unlink the handle from its current transaction, and start the
656          * commit on that.
657          */
658         J_ASSERT(atomic_read(&transaction->t_updates) > 0);
659         J_ASSERT(journal_current_handle() == handle);
660 
661         read_lock(&journal->j_state_lock);
662         spin_lock(&transaction->t_handle_lock);
663         atomic_sub(handle->h_buffer_credits,
664                    &transaction->t_outstanding_credits);
665         if (handle->h_rsv_handle) {
666                 sub_reserved_credits(journal,
667                                      handle->h_rsv_handle->h_buffer_credits);
668         }
669         if (atomic_dec_and_test(&transaction->t_updates))
670                 wake_up(&journal->j_wait_updates);
671         tid = transaction->t_tid;
672         spin_unlock(&transaction->t_handle_lock);
673         handle->h_transaction = NULL;
674         current->journal_info = NULL;
675 
676         jbd_debug(2, "restarting handle %p\n", handle);
677         need_to_start = !tid_geq(journal->j_commit_request, tid);
678         read_unlock(&journal->j_state_lock);
679         if (need_to_start)
680                 jbd2_log_start_commit(journal, tid);
681 
682         rwsem_release(&journal->j_trans_commit_map, 1, _THIS_IP_);
683         handle->h_buffer_credits = nblocks;
684         /*
685          * Restore the original nofs context because the journal restart
686          * is basically the same thing as journal stop and start.
687          * start_this_handle will start a new nofs context.
688          */
689         memalloc_nofs_restore(handle->saved_alloc_context);
690         ret = start_this_handle(journal, handle, gfp_mask);
691         return ret;
692 }
693 EXPORT_SYMBOL(jbd2__journal_restart);
694 
695 
696 int jbd2_journal_restart(handle_t *handle, int nblocks)
697 {
698         return jbd2__journal_restart(handle, nblocks, GFP_NOFS);
699 }
700 EXPORT_SYMBOL(jbd2_journal_restart);
701 
702 /**
703  * void jbd2_journal_lock_updates () - establish a transaction barrier.
704  * @journal:  Journal to establish a barrier on.
705  *
706  * This locks out any further updates from being started, and blocks
707  * until all existing updates have completed, returning only once the
708  * journal is in a quiescent state with no updates running.
709  *
710  * The journal lock should not be held on entry.
711  */
712 void jbd2_journal_lock_updates(journal_t *journal)
713 {
714         DEFINE_WAIT(wait);
715 
716         jbd2_might_wait_for_commit(journal);
717 
718         write_lock(&journal->j_state_lock);
719         ++journal->j_barrier_count;
720 
721         /* Wait until there are no reserved handles */
722         if (atomic_read(&journal->j_reserved_credits)) {
723                 write_unlock(&journal->j_state_lock);
724                 wait_event(journal->j_wait_reserved,
725                            atomic_read(&journal->j_reserved_credits) == 0);
726                 write_lock(&journal->j_state_lock);
727         }
728 
729         /* Wait until there are no running updates */
730         while (1) {
731                 transaction_t *transaction = journal->j_running_transaction;
732 
733                 if (!transaction)
734                         break;
735 
736                 spin_lock(&transaction->t_handle_lock);
737                 prepare_to_wait(&journal->j_wait_updates, &wait,
738                                 TASK_UNINTERRUPTIBLE);
739                 if (!atomic_read(&transaction->t_updates)) {
740                         spin_unlock(&transaction->t_handle_lock);
741                         finish_wait(&journal->j_wait_updates, &wait);
742                         break;
743                 }
744                 spin_unlock(&transaction->t_handle_lock);
745                 write_unlock(&journal->j_state_lock);
746                 schedule();
747                 finish_wait(&journal->j_wait_updates, &wait);
748                 write_lock(&journal->j_state_lock);
749         }
750         write_unlock(&journal->j_state_lock);
751 
752         /*
753          * We have now established a barrier against other normal updates, but
754          * we also need to barrier against other jbd2_journal_lock_updates() calls
755          * to make sure that we serialise special journal-locked operations
756          * too.
757          */
758         mutex_lock(&journal->j_barrier);
759 }
760 
761 /**
762  * void jbd2_journal_unlock_updates (journal_t* journal) - release barrier
763  * @journal:  Journal to release the barrier on.
764  *
765  * Release a transaction barrier obtained with jbd2_journal_lock_updates().
766  *
767  * Should be called without the journal lock held.
768  */
769 void jbd2_journal_unlock_updates (journal_t *journal)
770 {
771         J_ASSERT(journal->j_barrier_count != 0);
772 
773         mutex_unlock(&journal->j_barrier);
774         write_lock(&journal->j_state_lock);
775         --journal->j_barrier_count;
776         write_unlock(&journal->j_state_lock);
777         wake_up(&journal->j_wait_transaction_locked);
778 }
779 
780 static void warn_dirty_buffer(struct buffer_head *bh)
781 {
782         printk(KERN_WARNING
783                "JBD2: Spotted dirty metadata buffer (dev = %pg, blocknr = %llu). "
784                "There's a risk of filesystem corruption in case of system "
785                "crash.\n",
786                bh->b_bdev, (unsigned long long)bh->b_blocknr);
787 }
788 
789 /* Call t_frozen trigger and copy buffer data into jh->b_frozen_data. */
790 static void jbd2_freeze_jh_data(struct journal_head *jh)
791 {
792         struct page *page;
793         int offset;
794         char *source;
795         struct buffer_head *bh = jh2bh(jh);
796 
797         J_EXPECT_JH(jh, buffer_uptodate(bh), "Possible IO failure.\n");
798         page = bh->b_page;
799         offset = offset_in_page(bh->b_data);
800         source = kmap_atomic(page);
801         /* Fire data frozen trigger just before we copy the data */
802         jbd2_buffer_frozen_trigger(jh, source + offset, jh->b_triggers);
803         memcpy(jh->b_frozen_data, source + offset, bh->b_size);
804         kunmap_atomic(source);
805 
806         /*
807          * Now that the frozen data is saved off, we need to store any matching
808          * triggers.
809          */
810         jh->b_frozen_triggers = jh->b_triggers;
811 }
812 
813 /*
814  * If the buffer is already part of the current transaction, then there
815  * is nothing we need to do.  If it is already part of a prior
816  * transaction which we are still committing to disk, then we need to
817  * make sure that we do not overwrite the old copy: we do copy-out to
818  * preserve the copy going to disk.  We also account the buffer against
819  * the handle's metadata buffer credits (unless the buffer is already
820  * part of the transaction, that is).
821  *
822  */
823 static int
824 do_get_write_access(handle_t *handle, struct journal_head *jh,
825                         int force_copy)
826 {
827         struct buffer_head *bh;
828         transaction_t *transaction = handle->h_transaction;
829         journal_t *journal;
830         int error;
831         char *frozen_buffer = NULL;
832         unsigned long start_lock, time_lock;
833 
834         if (is_handle_aborted(handle))
835                 return -EROFS;
836         journal = transaction->t_journal;
837 
838         jbd_debug(5, "journal_head %p, force_copy %d\n", jh, force_copy);
839 
840         JBUFFER_TRACE(jh, "entry");
841 repeat:
842         bh = jh2bh(jh);
843 
844         /* @@@ Need to check for errors here at some point. */
845 
846         start_lock = jiffies;
847         lock_buffer(bh);
848         jbd_lock_bh_state(bh);
849 
850         /* If it takes too long to lock the buffer, trace it */
851         time_lock = jbd2_time_diff(start_lock, jiffies);
852         if (time_lock > HZ/10)
853                 trace_jbd2_lock_buffer_stall(bh->b_bdev->bd_dev,
854                         jiffies_to_msecs(time_lock));
855 
856         /* We now hold the buffer lock so it is safe to query the buffer
857          * state.  Is the buffer dirty?
858          *
859          * If so, there are two possibilities.  The buffer may be
860          * non-journaled, and undergoing a quite legitimate writeback.
861          * Otherwise, it is journaled, and we don't expect dirty buffers
862          * in that state (the buffers should be marked JBD_Dirty
863          * instead.)  So either the IO is being done under our own
864          * control and this is a bug, or it's a third party IO such as
865          * dump(8) (which may leave the buffer scheduled for read ---
866          * ie. locked but not dirty) or tune2fs (which may actually have
867          * the buffer dirtied, ugh.)  */
868 
869         if (buffer_dirty(bh)) {
870                 /*
871                  * First question: is this buffer already part of the current
872                  * transaction or the existing committing transaction?
873                  */
874                 if (jh->b_transaction) {
875                         J_ASSERT_JH(jh,
876                                 jh->b_transaction == transaction ||
877                                 jh->b_transaction ==
878                                         journal->j_committing_transaction);
879                         if (jh->b_next_transaction)
880                                 J_ASSERT_JH(jh, jh->b_next_transaction ==
881                                                         transaction);
882                         warn_dirty_buffer(bh);
883                 }
884                 /*
885                  * In any case we need to clean the dirty flag and we must
886                  * do it under the buffer lock to be sure we don't race
887                  * with running write-out.
888                  */
889                 JBUFFER_TRACE(jh, "Journalling dirty buffer");
890                 clear_buffer_dirty(bh);
891                 set_buffer_jbddirty(bh);
892         }
893 
894         unlock_buffer(bh);
895 
896         error = -EROFS;
897         if (is_handle_aborted(handle)) {
898                 jbd_unlock_bh_state(bh);
899                 goto out;
900         }
901         error = 0;
902 
903         /*
904          * The buffer is already part of this transaction if b_transaction or
905          * b_next_transaction points to it
906          */
907         if (jh->b_transaction == transaction ||
908             jh->b_next_transaction == transaction)
909                 goto done;
910 
911         /*
912          * this is the first time this transaction is touching this buffer,
913          * reset the modified flag
914          */
915        jh->b_modified = 0;
916 
917         /*
918          * If the buffer is not journaled right now, we need to make sure it
919          * doesn't get written to disk before the caller actually commits the
920          * new data
921          */
922         if (!jh->b_transaction) {
923                 JBUFFER_TRACE(jh, "no transaction");
924                 J_ASSERT_JH(jh, !jh->b_next_transaction);
925                 JBUFFER_TRACE(jh, "file as BJ_Reserved");
926                 /*
927                  * Make sure all stores to jh (b_modified, b_frozen_data) are
928                  * visible before attaching it to the running transaction.
929                  * Paired with barrier in jbd2_write_access_granted()
930                  */
931                 smp_wmb();
932                 spin_lock(&journal->j_list_lock);
933                 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
934                 spin_unlock(&journal->j_list_lock);
935                 goto done;
936         }
937         /*
938          * If there is already a copy-out version of this buffer, then we don't
939          * need to make another one
940          */
941         if (jh->b_frozen_data) {
942                 JBUFFER_TRACE(jh, "has frozen data");
943                 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
944                 goto attach_next;
945         }
946 
947         JBUFFER_TRACE(jh, "owned by older transaction");
948         J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
949         J_ASSERT_JH(jh, jh->b_transaction == journal->j_committing_transaction);
950 
951         /*
952          * There is one case we have to be very careful about.  If the
953          * committing transaction is currently writing this buffer out to disk
954          * and has NOT made a copy-out, then we cannot modify the buffer
955          * contents at all right now.  The essence of copy-out is that it is
956          * the extra copy, not the primary copy, which gets journaled.  If the
957          * primary copy is already going to disk then we cannot do copy-out
958          * here.
959          */
960         if (buffer_shadow(bh)) {
961                 JBUFFER_TRACE(jh, "on shadow: sleep");
962                 jbd_unlock_bh_state(bh);
963                 wait_on_bit_io(&bh->b_state, BH_Shadow, TASK_UNINTERRUPTIBLE);
964                 goto repeat;
965         }
966 
967         /*
968          * Only do the copy if the currently-owning transaction still needs it.
969          * If buffer isn't on BJ_Metadata list, the committing transaction is
970          * past that stage (here we use the fact that BH_Shadow is set under
971          * bh_state lock together with refiling to BJ_Shadow list and at this
972          * point we know the buffer doesn't have BH_Shadow set).
973          *
974          * Subtle point, though: if this is a get_undo_access, then we will be
975          * relying on the frozen_data to contain the new value of the
976          * committed_data record after the transaction, so we HAVE to force the
977          * frozen_data copy in that case.
978          */
979         if (jh->b_jlist == BJ_Metadata || force_copy) {
980                 JBUFFER_TRACE(jh, "generate frozen data");
981                 if (!frozen_buffer) {
982                         JBUFFER_TRACE(jh, "allocate memory for buffer");
983                         jbd_unlock_bh_state(bh);
984                         frozen_buffer = jbd2_alloc(jh2bh(jh)->b_size,
985                                                    GFP_NOFS | __GFP_NOFAIL);
986                         goto repeat;
987                 }
988                 jh->b_frozen_data = frozen_buffer;
989                 frozen_buffer = NULL;
990                 jbd2_freeze_jh_data(jh);
991         }
992 attach_next:
993         /*
994          * Make sure all stores to jh (b_modified, b_frozen_data) are visible
995          * before attaching it to the running transaction. Paired with barrier
996          * in jbd2_write_access_granted()
997          */
998         smp_wmb();
999         jh->b_next_transaction = transaction;
1000 
1001 done:
1002         jbd_unlock_bh_state(bh);
1003 
1004         /*
1005          * If we are about to journal a buffer, then any revoke pending on it is
1006          * no longer valid
1007          */
1008         jbd2_journal_cancel_revoke(handle, jh);
1009 
1010 out:
1011         if (unlikely(frozen_buffer))    /* It's usually NULL */
1012                 jbd2_free(frozen_buffer, bh->b_size);
1013 
1014         JBUFFER_TRACE(jh, "exit");
1015         return error;
1016 }
1017 
1018 /* Fast check whether buffer is already attached to the required transaction */
1019 static bool jbd2_write_access_granted(handle_t *handle, struct buffer_head *bh,
1020                                                         bool undo)
1021 {
1022         struct journal_head *jh;
1023         bool ret = false;
1024 
1025         /* Dirty buffers require special handling... */
1026         if (buffer_dirty(bh))
1027                 return false;
1028 
1029         /*
1030          * RCU protects us from dereferencing freed pages. So the checks we do
1031          * are guaranteed not to oops. However the jh slab object can get freed
1032          * & reallocated while we work with it. So we have to be careful. When
1033          * we see jh attached to the running transaction, we know it must stay
1034          * so until the transaction is committed. Thus jh won't be freed and
1035          * will be attached to the same bh while we run.  However it can
1036          * happen jh gets freed, reallocated, and attached to the transaction
1037          * just after we get pointer to it from bh. So we have to be careful
1038          * and recheck jh still belongs to our bh before we return success.
1039          */
1040         rcu_read_lock();
1041         if (!buffer_jbd(bh))
1042                 goto out;
1043         /* This should be bh2jh() but that doesn't work with inline functions */
1044         jh = READ_ONCE(bh->b_private);
1045         if (!jh)
1046                 goto out;
1047         /* For undo access buffer must have data copied */
1048         if (undo && !jh->b_committed_data)
1049                 goto out;
1050         if (jh->b_transaction != handle->h_transaction &&
1051             jh->b_next_transaction != handle->h_transaction)
1052                 goto out;
1053         /*
1054          * There are two reasons for the barrier here:
1055          * 1) Make sure to fetch b_bh after we did previous checks so that we
1056          * detect when jh went through free, realloc, attach to transaction
1057          * while we were checking. Paired with implicit barrier in that path.
1058          * 2) So that access to bh done after jbd2_write_access_granted()
1059          * doesn't get reordered and see inconsistent state of concurrent
1060          * do_get_write_access().
1061          */
1062         smp_mb();
1063         if (unlikely(jh->b_bh != bh))
1064                 goto out;
1065         ret = true;
1066 out:
1067         rcu_read_unlock();
1068         return ret;
1069 }
1070 
1071 /**
1072  * int jbd2_journal_get_write_access() - notify intent to modify a buffer for metadata (not data) update.
1073  * @handle: transaction to add buffer modifications to
1074  * @bh:     bh to be used for metadata writes
1075  *
1076  * Returns: error code or 0 on success.
1077  *
1078  * In full data journalling mode the buffer may be of type BJ_AsyncData,
1079  * because we're ``write()ing`` a buffer which is also part of a shared mapping.
1080  */
1081 
1082 int jbd2_journal_get_write_access(handle_t *handle, struct buffer_head *bh)
1083 {
1084         struct journal_head *jh;
1085         int rc;
1086 
1087         if (jbd2_write_access_granted(handle, bh, false))
1088                 return 0;
1089 
1090         jh = jbd2_journal_add_journal_head(bh);
1091         /* We do not want to get caught playing with fields which the
1092          * log thread also manipulates.  Make sure that the buffer
1093          * completes any outstanding IO before proceeding. */
1094         rc = do_get_write_access(handle, jh, 0);
1095         jbd2_journal_put_journal_head(jh);
1096         return rc;
1097 }
1098 
1099 
1100 /*
1101  * When the user wants to journal a newly created buffer_head
1102  * (ie. getblk() returned a new buffer and we are going to populate it
1103  * manually rather than reading off disk), then we need to keep the
1104  * buffer_head locked until it has been completely filled with new
1105  * data.  In this case, we should be able to make the assertion that
1106  * the bh is not already part of an existing transaction.
1107  *
1108  * The buffer should already be locked by the caller by this point.
1109  * There is no lock ranking violation: it was a newly created,
1110  * unlocked buffer beforehand. */
1111 
1112 /**
1113  * int jbd2_journal_get_create_access () - notify intent to use newly created bh
1114  * @handle: transaction to new buffer to
1115  * @bh: new buffer.
1116  *
1117  * Call this if you create a new bh.
1118  */
1119 int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh)
1120 {
1121         transaction_t *transaction = handle->h_transaction;
1122         journal_t *journal;
1123         struct journal_head *jh = jbd2_journal_add_journal_head(bh);
1124         int err;
1125 
1126         jbd_debug(5, "journal_head %p\n", jh);
1127         err = -EROFS;
1128         if (is_handle_aborted(handle))
1129                 goto out;
1130         journal = transaction->t_journal;
1131         err = 0;
1132 
1133         JBUFFER_TRACE(jh, "entry");
1134         /*
1135          * The buffer may already belong to this transaction due to pre-zeroing
1136          * in the filesystem's new_block code.  It may also be on the previous,
1137          * committing transaction's lists, but it HAS to be in Forget state in
1138          * that case: the transaction must have deleted the buffer for it to be
1139          * reused here.
1140          */
1141         jbd_lock_bh_state(bh);
1142         J_ASSERT_JH(jh, (jh->b_transaction == transaction ||
1143                 jh->b_transaction == NULL ||
1144                 (jh->b_transaction == journal->j_committing_transaction &&
1145                           jh->b_jlist == BJ_Forget)));
1146 
1147         J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1148         J_ASSERT_JH(jh, buffer_locked(jh2bh(jh)));
1149 
1150         if (jh->b_transaction == NULL) {
1151                 /*
1152                  * Previous jbd2_journal_forget() could have left the buffer
1153                  * with jbddirty bit set because it was being committed. When
1154                  * the commit finished, we've filed the buffer for
1155                  * checkpointing and marked it dirty. Now we are reallocating
1156                  * the buffer so the transaction freeing it must have
1157                  * committed and so it's safe to clear the dirty bit.
1158                  */
1159                 clear_buffer_dirty(jh2bh(jh));
1160                 /* first access by this transaction */
1161                 jh->b_modified = 0;
1162 
1163                 JBUFFER_TRACE(jh, "file as BJ_Reserved");
1164                 spin_lock(&journal->j_list_lock);
1165                 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
1166                 spin_unlock(&journal->j_list_lock);
1167         } else if (jh->b_transaction == journal->j_committing_transaction) {
1168                 /* first access by this transaction */
1169                 jh->b_modified = 0;
1170 
1171                 JBUFFER_TRACE(jh, "set next transaction");
1172                 spin_lock(&journal->j_list_lock);
1173                 jh->b_next_transaction = transaction;
1174                 spin_unlock(&journal->j_list_lock);
1175         }
1176         jbd_unlock_bh_state(bh);
1177 
1178         /*
1179          * akpm: I added this.  ext3_alloc_branch can pick up new indirect
1180          * blocks which contain freed but then revoked metadata.  We need
1181          * to cancel the revoke in case we end up freeing it yet again
1182          * and the reallocating as data - this would cause a second revoke,
1183          * which hits an assertion error.
1184          */
1185         JBUFFER_TRACE(jh, "cancelling revoke");
1186         jbd2_journal_cancel_revoke(handle, jh);
1187 out:
1188         jbd2_journal_put_journal_head(jh);
1189         return err;
1190 }
1191 
1192 /**
1193  * int jbd2_journal_get_undo_access() -  Notify intent to modify metadata with
1194  *     non-rewindable consequences
1195  * @handle: transaction
1196  * @bh: buffer to undo
1197  *
1198  * Sometimes there is a need to distinguish between metadata which has
1199  * been committed to disk and that which has not.  The ext3fs code uses
1200  * this for freeing and allocating space, we have to make sure that we
1201  * do not reuse freed space until the deallocation has been committed,
1202  * since if we overwrote that space we would make the delete
1203  * un-rewindable in case of a crash.
1204  *
1205  * To deal with that, jbd2_journal_get_undo_access requests write access to a
1206  * buffer for parts of non-rewindable operations such as delete
1207  * operations on the bitmaps.  The journaling code must keep a copy of
1208  * the buffer's contents prior to the undo_access call until such time
1209  * as we know that the buffer has definitely been committed to disk.
1210  *
1211  * We never need to know which transaction the committed data is part
1212  * of, buffers touched here are guaranteed to be dirtied later and so
1213  * will be committed to a new transaction in due course, at which point
1214  * we can discard the old committed data pointer.
1215  *
1216  * Returns error number or 0 on success.
1217  */
1218 int jbd2_journal_get_undo_access(handle_t *handle, struct buffer_head *bh)
1219 {
1220         int err;
1221         struct journal_head *jh;
1222         char *committed_data = NULL;
1223 
1224         JBUFFER_TRACE(jh, "entry");
1225         if (jbd2_write_access_granted(handle, bh, true))
1226                 return 0;
1227 
1228         jh = jbd2_journal_add_journal_head(bh);
1229         /*
1230          * Do this first --- it can drop the journal lock, so we want to
1231          * make sure that obtaining the committed_data is done
1232          * atomically wrt. completion of any outstanding commits.
1233          */
1234         err = do_get_write_access(handle, jh, 1);
1235         if (err)
1236                 goto out;
1237 
1238 repeat:
1239         if (!jh->b_committed_data)
1240                 committed_data = jbd2_alloc(jh2bh(jh)->b_size,
1241                                             GFP_NOFS|__GFP_NOFAIL);
1242 
1243         jbd_lock_bh_state(bh);
1244         if (!jh->b_committed_data) {
1245                 /* Copy out the current buffer contents into the
1246                  * preserved, committed copy. */
1247                 JBUFFER_TRACE(jh, "generate b_committed data");
1248                 if (!committed_data) {
1249                         jbd_unlock_bh_state(bh);
1250                         goto repeat;
1251                 }
1252 
1253                 jh->b_committed_data = committed_data;
1254                 committed_data = NULL;
1255                 memcpy(jh->b_committed_data, bh->b_data, bh->b_size);
1256         }
1257         jbd_unlock_bh_state(bh);
1258 out:
1259         jbd2_journal_put_journal_head(jh);
1260         if (unlikely(committed_data))
1261                 jbd2_free(committed_data, bh->b_size);
1262         return err;
1263 }
1264 
1265 /**
1266  * void jbd2_journal_set_triggers() - Add triggers for commit writeout
1267  * @bh: buffer to trigger on
1268  * @type: struct jbd2_buffer_trigger_type containing the trigger(s).
1269  *
1270  * Set any triggers on this journal_head.  This is always safe, because
1271  * triggers for a committing buffer will be saved off, and triggers for
1272  * a running transaction will match the buffer in that transaction.
1273  *
1274  * Call with NULL to clear the triggers.
1275  */
1276 void jbd2_journal_set_triggers(struct buffer_head *bh,
1277                                struct jbd2_buffer_trigger_type *type)
1278 {
1279         struct journal_head *jh = jbd2_journal_grab_journal_head(bh);
1280 
1281         if (WARN_ON(!jh))
1282                 return;
1283         jh->b_triggers = type;
1284         jbd2_journal_put_journal_head(jh);
1285 }
1286 
1287 void jbd2_buffer_frozen_trigger(struct journal_head *jh, void *mapped_data,
1288                                 struct jbd2_buffer_trigger_type *triggers)
1289 {
1290         struct buffer_head *bh = jh2bh(jh);
1291 
1292         if (!triggers || !triggers->t_frozen)
1293                 return;
1294 
1295         triggers->t_frozen(triggers, bh, mapped_data, bh->b_size);
1296 }
1297 
1298 void jbd2_buffer_abort_trigger(struct journal_head *jh,
1299                                struct jbd2_buffer_trigger_type *triggers)
1300 {
1301         if (!triggers || !triggers->t_abort)
1302                 return;
1303 
1304         triggers->t_abort(triggers, jh2bh(jh));
1305 }
1306 
1307 /**
1308  * int jbd2_journal_dirty_metadata() -  mark a buffer as containing dirty metadata
1309  * @handle: transaction to add buffer to.
1310  * @bh: buffer to mark
1311  *
1312  * mark dirty metadata which needs to be journaled as part of the current
1313  * transaction.
1314  *
1315  * The buffer must have previously had jbd2_journal_get_write_access()
1316  * called so that it has a valid journal_head attached to the buffer
1317  * head.
1318  *
1319  * The buffer is placed on the transaction's metadata list and is marked
1320  * as belonging to the transaction.
1321  *
1322  * Returns error number or 0 on success.
1323  *
1324  * Special care needs to be taken if the buffer already belongs to the
1325  * current committing transaction (in which case we should have frozen
1326  * data present for that commit).  In that case, we don't relink the
1327  * buffer: that only gets done when the old transaction finally
1328  * completes its commit.
1329  */
1330 int jbd2_journal_dirty_metadata(handle_t *handle, struct buffer_head *bh)
1331 {
1332         transaction_t *transaction = handle->h_transaction;
1333         journal_t *journal;
1334         struct journal_head *jh;
1335         int ret = 0;
1336 
1337         if (is_handle_aborted(handle))
1338                 return -EROFS;
1339         if (!buffer_jbd(bh)) {
1340                 ret = -EUCLEAN;
1341                 goto out;
1342         }
1343         /*
1344          * We don't grab jh reference here since the buffer must be part
1345          * of the running transaction.
1346          */
1347         jh = bh2jh(bh);
1348         /*
1349          * This and the following assertions are unreliable since we may see jh
1350          * in inconsistent state unless we grab bh_state lock. But this is
1351          * crucial to catch bugs so let's do a reliable check until the
1352          * lockless handling is fully proven.
1353          */
1354         if (jh->b_transaction != transaction &&
1355             jh->b_next_transaction != transaction) {
1356                 jbd_lock_bh_state(bh);
1357                 J_ASSERT_JH(jh, jh->b_transaction == transaction ||
1358                                 jh->b_next_transaction == transaction);
1359                 jbd_unlock_bh_state(bh);
1360         }
1361         if (jh->b_modified == 1) {
1362                 /* If it's in our transaction it must be in BJ_Metadata list. */
1363                 if (jh->b_transaction == transaction &&
1364                     jh->b_jlist != BJ_Metadata) {
1365                         jbd_lock_bh_state(bh);
1366                         if (jh->b_transaction == transaction &&
1367                             jh->b_jlist != BJ_Metadata)
1368                                 pr_err("JBD2: assertion failure: h_type=%u "
1369                                        "h_line_no=%u block_no=%llu jlist=%u\n",
1370                                        handle->h_type, handle->h_line_no,
1371                                        (unsigned long long) bh->b_blocknr,
1372                                        jh->b_jlist);
1373                         J_ASSERT_JH(jh, jh->b_transaction != transaction ||
1374                                         jh->b_jlist == BJ_Metadata);
1375                         jbd_unlock_bh_state(bh);
1376                 }
1377                 goto out;
1378         }
1379 
1380         journal = transaction->t_journal;
1381         jbd_debug(5, "journal_head %p\n", jh);
1382         JBUFFER_TRACE(jh, "entry");
1383 
1384         jbd_lock_bh_state(bh);
1385 
1386         if (jh->b_modified == 0) {
1387                 /*
1388                  * This buffer's got modified and becoming part
1389                  * of the transaction. This needs to be done
1390                  * once a transaction -bzzz
1391                  */
1392                 if (handle->h_buffer_credits <= 0) {
1393                         ret = -ENOSPC;
1394                         goto out_unlock_bh;
1395                 }
1396                 jh->b_modified = 1;
1397                 handle->h_buffer_credits--;
1398         }
1399 
1400         /*
1401          * fastpath, to avoid expensive locking.  If this buffer is already
1402          * on the running transaction's metadata list there is nothing to do.
1403          * Nobody can take it off again because there is a handle open.
1404          * I _think_ we're OK here with SMP barriers - a mistaken decision will
1405          * result in this test being false, so we go in and take the locks.
1406          */
1407         if (jh->b_transaction == transaction && jh->b_jlist == BJ_Metadata) {
1408                 JBUFFER_TRACE(jh, "fastpath");
1409                 if (unlikely(jh->b_transaction !=
1410                              journal->j_running_transaction)) {
1411                         printk(KERN_ERR "JBD2: %s: "
1412                                "jh->b_transaction (%llu, %p, %u) != "
1413                                "journal->j_running_transaction (%p, %u)\n",
1414                                journal->j_devname,
1415                                (unsigned long long) bh->b_blocknr,
1416                                jh->b_transaction,
1417                                jh->b_transaction ? jh->b_transaction->t_tid : 0,
1418                                journal->j_running_transaction,
1419                                journal->j_running_transaction ?
1420                                journal->j_running_transaction->t_tid : 0);
1421                         ret = -EINVAL;
1422                 }
1423                 goto out_unlock_bh;
1424         }
1425 
1426         set_buffer_jbddirty(bh);
1427 
1428         /*
1429          * Metadata already on the current transaction list doesn't
1430          * need to be filed.  Metadata on another transaction's list must
1431          * be committing, and will be refiled once the commit completes:
1432          * leave it alone for now.
1433          */
1434         if (jh->b_transaction != transaction) {
1435                 JBUFFER_TRACE(jh, "already on other transaction");
1436                 if (unlikely(((jh->b_transaction !=
1437                                journal->j_committing_transaction)) ||
1438                              (jh->b_next_transaction != transaction))) {
1439                         printk(KERN_ERR "jbd2_journal_dirty_metadata: %s: "
1440                                "bad jh for block %llu: "
1441                                "transaction (%p, %u), "
1442                                "jh->b_transaction (%p, %u), "
1443                                "jh->b_next_transaction (%p, %u), jlist %u\n",
1444                                journal->j_devname,
1445                                (unsigned long long) bh->b_blocknr,
1446                                transaction, transaction->t_tid,
1447                                jh->b_transaction,
1448                                jh->b_transaction ?
1449                                jh->b_transaction->t_tid : 0,
1450                                jh->b_next_transaction,
1451                                jh->b_next_transaction ?
1452                                jh->b_next_transaction->t_tid : 0,
1453                                jh->b_jlist);
1454                         WARN_ON(1);
1455                         ret = -EINVAL;
1456                 }
1457                 /* And this case is illegal: we can't reuse another
1458                  * transaction's data buffer, ever. */
1459                 goto out_unlock_bh;
1460         }
1461 
1462         /* That test should have eliminated the following case: */
1463         J_ASSERT_JH(jh, jh->b_frozen_data == NULL);
1464 
1465         JBUFFER_TRACE(jh, "file as BJ_Metadata");
1466         spin_lock(&journal->j_list_lock);
1467         __jbd2_journal_file_buffer(jh, transaction, BJ_Metadata);
1468         spin_unlock(&journal->j_list_lock);
1469 out_unlock_bh:
1470         jbd_unlock_bh_state(bh);
1471 out:
1472         JBUFFER_TRACE(jh, "exit");
1473         return ret;
1474 }
1475 
1476 /**
1477  * void jbd2_journal_forget() - bforget() for potentially-journaled buffers.
1478  * @handle: transaction handle
1479  * @bh:     bh to 'forget'
1480  *
1481  * We can only do the bforget if there are no commits pending against the
1482  * buffer.  If the buffer is dirty in the current running transaction we
1483  * can safely unlink it.
1484  *
1485  * bh may not be a journalled buffer at all - it may be a non-JBD
1486  * buffer which came off the hashtable.  Check for this.
1487  *
1488  * Decrements bh->b_count by one.
1489  *
1490  * Allow this call even if the handle has aborted --- it may be part of
1491  * the caller's cleanup after an abort.
1492  */
1493 int jbd2_journal_forget (handle_t *handle, struct buffer_head *bh)
1494 {
1495         transaction_t *transaction = handle->h_transaction;
1496         journal_t *journal;
1497         struct journal_head *jh;
1498         int drop_reserve = 0;
1499         int err = 0;
1500         int was_modified = 0;
1501 
1502         if (is_handle_aborted(handle))
1503                 return -EROFS;
1504         journal = transaction->t_journal;
1505 
1506         BUFFER_TRACE(bh, "entry");
1507 
1508         jbd_lock_bh_state(bh);
1509 
1510         if (!buffer_jbd(bh))
1511                 goto not_jbd;
1512         jh = bh2jh(bh);
1513 
1514         /* Critical error: attempting to delete a bitmap buffer, maybe?
1515          * Don't do any jbd operations, and return an error. */
1516         if (!J_EXPECT_JH(jh, !jh->b_committed_data,
1517                          "inconsistent data on disk")) {
1518                 err = -EIO;
1519                 goto not_jbd;
1520         }
1521 
1522         /* keep track of whether or not this transaction modified us */
1523         was_modified = jh->b_modified;
1524 
1525         /*
1526          * The buffer's going from the transaction, we must drop
1527          * all references -bzzz
1528          */
1529         jh->b_modified = 0;
1530 
1531         if (jh->b_transaction == transaction) {
1532                 J_ASSERT_JH(jh, !jh->b_frozen_data);
1533 
1534                 /* If we are forgetting a buffer which is already part
1535                  * of this transaction, then we can just drop it from
1536                  * the transaction immediately. */
1537                 clear_buffer_dirty(bh);
1538                 clear_buffer_jbddirty(bh);
1539 
1540                 JBUFFER_TRACE(jh, "belongs to current transaction: unfile");
1541 
1542                 /*
1543                  * we only want to drop a reference if this transaction
1544                  * modified the buffer
1545                  */
1546                 if (was_modified)
1547                         drop_reserve = 1;
1548 
1549                 /*
1550                  * We are no longer going to journal this buffer.
1551                  * However, the commit of this transaction is still
1552                  * important to the buffer: the delete that we are now
1553                  * processing might obsolete an old log entry, so by
1554                  * committing, we can satisfy the buffer's checkpoint.
1555                  *
1556                  * So, if we have a checkpoint on the buffer, we should
1557                  * now refile the buffer on our BJ_Forget list so that
1558                  * we know to remove the checkpoint after we commit.
1559                  */
1560 
1561                 spin_lock(&journal->j_list_lock);
1562                 if (jh->b_cp_transaction) {
1563                         __jbd2_journal_temp_unlink_buffer(jh);
1564                         __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
1565                 } else {
1566                         __jbd2_journal_unfile_buffer(jh);
1567                         if (!buffer_jbd(bh)) {
1568                                 spin_unlock(&journal->j_list_lock);
1569                                 jbd_unlock_bh_state(bh);
1570                                 __bforget(bh);
1571                                 goto drop;
1572                         }
1573                 }
1574                 spin_unlock(&journal->j_list_lock);
1575         } else if (jh->b_transaction) {
1576                 J_ASSERT_JH(jh, (jh->b_transaction ==
1577                                  journal->j_committing_transaction));
1578                 /* However, if the buffer is still owned by a prior
1579                  * (committing) transaction, we can't drop it yet... */
1580                 JBUFFER_TRACE(jh, "belongs to older transaction");
1581                 /* ... but we CAN drop it from the new transaction if we
1582                  * have also modified it since the original commit. */
1583 
1584                 if (jh->b_next_transaction) {
1585                         J_ASSERT(jh->b_next_transaction == transaction);
1586                         spin_lock(&journal->j_list_lock);
1587                         jh->b_next_transaction = NULL;
1588                         spin_unlock(&journal->j_list_lock);
1589 
1590                         /*
1591                          * only drop a reference if this transaction modified
1592                          * the buffer
1593                          */
1594                         if (was_modified)
1595                                 drop_reserve = 1;
1596                 }
1597         }
1598 
1599 not_jbd:
1600         jbd_unlock_bh_state(bh);
1601         __brelse(bh);
1602 drop:
1603         if (drop_reserve) {
1604                 /* no need to reserve log space for this block -bzzz */
1605                 handle->h_buffer_credits++;
1606         }
1607         return err;
1608 }
1609 
1610 /**
1611  * int jbd2_journal_stop() - complete a transaction
1612  * @handle: transaction to complete.
1613  *
1614  * All done for a particular handle.
1615  *
1616  * There is not much action needed here.  We just return any remaining
1617  * buffer credits to the transaction and remove the handle.  The only
1618  * complication is that we need to start a commit operation if the
1619  * filesystem is marked for synchronous update.
1620  *
1621  * jbd2_journal_stop itself will not usually return an error, but it may
1622  * do so in unusual circumstances.  In particular, expect it to
1623  * return -EIO if a jbd2_journal_abort has been executed since the
1624  * transaction began.
1625  */
1626 int jbd2_journal_stop(handle_t *handle)
1627 {
1628         transaction_t *transaction = handle->h_transaction;
1629         journal_t *journal;
1630         int err = 0, wait_for_commit = 0;
1631         tid_t tid;
1632         pid_t pid;
1633 
1634         if (!transaction) {
1635                 /*
1636                  * Handle is already detached from the transaction so
1637                  * there is nothing to do other than decrease a refcount,
1638                  * or free the handle if refcount drops to zero
1639                  */
1640                 if (--handle->h_ref > 0) {
1641                         jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1642                                                          handle->h_ref);
1643                         return err;
1644                 } else {
1645                         if (handle->h_rsv_handle)
1646                                 jbd2_free_handle(handle->h_rsv_handle);
1647                         goto free_and_exit;
1648                 }
1649         }
1650         journal = transaction->t_journal;
1651 
1652         J_ASSERT(journal_current_handle() == handle);
1653 
1654         if (is_handle_aborted(handle))
1655                 err = -EIO;
1656         else
1657                 J_ASSERT(atomic_read(&transaction->t_updates) > 0);
1658 
1659         if (--handle->h_ref > 0) {
1660                 jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1661                           handle->h_ref);
1662                 return err;
1663         }
1664 
1665         jbd_debug(4, "Handle %p going down\n", handle);
1666         trace_jbd2_handle_stats(journal->j_fs_dev->bd_dev,
1667                                 transaction->t_tid,
1668                                 handle->h_type, handle->h_line_no,
1669                                 jiffies - handle->h_start_jiffies,
1670                                 handle->h_sync, handle->h_requested_credits,
1671                                 (handle->h_requested_credits -
1672                                  handle->h_buffer_credits));
1673 
1674         /*
1675          * Implement synchronous transaction batching.  If the handle
1676          * was synchronous, don't force a commit immediately.  Let's
1677          * yield and let another thread piggyback onto this
1678          * transaction.  Keep doing that while new threads continue to
1679          * arrive.  It doesn't cost much - we're about to run a commit
1680          * and sleep on IO anyway.  Speeds up many-threaded, many-dir
1681          * operations by 30x or more...
1682          *
1683          * We try and optimize the sleep time against what the
1684          * underlying disk can do, instead of having a static sleep
1685          * time.  This is useful for the case where our storage is so
1686          * fast that it is more optimal to go ahead and force a flush
1687          * and wait for the transaction to be committed than it is to
1688          * wait for an arbitrary amount of time for new writers to
1689          * join the transaction.  We achieve this by measuring how
1690          * long it takes to commit a transaction, and compare it with
1691          * how long this transaction has been running, and if run time
1692          * < commit time then we sleep for the delta and commit.  This
1693          * greatly helps super fast disks that would see slowdowns as
1694          * more threads started doing fsyncs.
1695          *
1696          * But don't do this if this process was the most recent one
1697          * to perform a synchronous write.  We do this to detect the
1698          * case where a single process is doing a stream of sync
1699          * writes.  No point in waiting for joiners in that case.
1700          *
1701          * Setting max_batch_time to 0 disables this completely.
1702          */
1703         pid = current->pid;
1704         if (handle->h_sync && journal->j_last_sync_writer != pid &&
1705             journal->j_max_batch_time) {
1706                 u64 commit_time, trans_time;
1707 
1708                 journal->j_last_sync_writer = pid;
1709 
1710                 read_lock(&journal->j_state_lock);
1711                 commit_time = journal->j_average_commit_time;
1712                 read_unlock(&journal->j_state_lock);
1713 
1714                 trans_time = ktime_to_ns(ktime_sub(ktime_get(),
1715                                                    transaction->t_start_time));
1716 
1717                 commit_time = max_t(u64, commit_time,
1718                                     1000*journal->j_min_batch_time);
1719                 commit_time = min_t(u64, commit_time,
1720                                     1000*journal->j_max_batch_time);
1721 
1722                 if (trans_time < commit_time) {
1723                         ktime_t expires = ktime_add_ns(ktime_get(),
1724                                                        commit_time);
1725                         set_current_state(TASK_UNINTERRUPTIBLE);
1726                         schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1727                 }
1728         }
1729 
1730         if (handle->h_sync)
1731                 transaction->t_synchronous_commit = 1;
1732         current->journal_info = NULL;
1733         atomic_sub(handle->h_buffer_credits,
1734                    &transaction->t_outstanding_credits);
1735 
1736         /*
1737          * If the handle is marked SYNC, we need to set another commit
1738          * going!  We also want to force a commit if the current
1739          * transaction is occupying too much of the log, or if the
1740          * transaction is too old now.
1741          */
1742         if (handle->h_sync ||
1743             (atomic_read(&transaction->t_outstanding_credits) >
1744              journal->j_max_transaction_buffers) ||
1745             time_after_eq(jiffies, transaction->t_expires)) {
1746                 /* Do this even for aborted journals: an abort still
1747                  * completes the commit thread, it just doesn't write
1748                  * anything to disk. */
1749 
1750                 jbd_debug(2, "transaction too old, requesting commit for "
1751                                         "handle %p\n", handle);
1752                 /* This is non-blocking */
1753                 jbd2_log_start_commit(journal, transaction->t_tid);
1754 
1755                 /*
1756                  * Special case: JBD2_SYNC synchronous updates require us
1757                  * to wait for the commit to complete.
1758                  */
1759                 if (handle->h_sync && !(current->flags & PF_MEMALLOC))
1760                         wait_for_commit = 1;
1761         }
1762 
1763         /*
1764          * Once we drop t_updates, if it goes to zero the transaction
1765          * could start committing on us and eventually disappear.  So
1766          * once we do this, we must not dereference transaction
1767          * pointer again.
1768          */
1769         tid = transaction->t_tid;
1770         if (atomic_dec_and_test(&transaction->t_updates)) {
1771                 wake_up(&journal->j_wait_updates);
1772                 if (journal->j_barrier_count)
1773                         wake_up(&journal->j_wait_transaction_locked);
1774         }
1775 
1776         rwsem_release(&journal->j_trans_commit_map, 1, _THIS_IP_);
1777 
1778         if (wait_for_commit)
1779                 err = jbd2_log_wait_commit(journal, tid);
1780 
1781         if (handle->h_rsv_handle)
1782                 jbd2_journal_free_reserved(handle->h_rsv_handle);
1783 free_and_exit:
1784         /*
1785          * Scope of the GFP_NOFS context is over here and so we can restore the
1786          * original alloc context.
1787          */
1788         memalloc_nofs_restore(handle->saved_alloc_context);
1789         jbd2_free_handle(handle);
1790         return err;
1791 }
1792 
1793 /*
1794  *
1795  * List management code snippets: various functions for manipulating the
1796  * transaction buffer lists.
1797  *
1798  */
1799 
1800 /*
1801  * Append a buffer to a transaction list, given the transaction's list head
1802  * pointer.
1803  *
1804  * j_list_lock is held.
1805  *
1806  * jbd_lock_bh_state(jh2bh(jh)) is held.
1807  */
1808 
1809 static inline void
1810 __blist_add_buffer(struct journal_head **list, struct journal_head *jh)
1811 {
1812         if (!*list) {
1813                 jh->b_tnext = jh->b_tprev = jh;
1814                 *list = jh;
1815         } else {
1816                 /* Insert at the tail of the list to preserve order */
1817                 struct journal_head *first = *list, *last = first->b_tprev;
1818                 jh->b_tprev = last;
1819                 jh->b_tnext = first;
1820                 last->b_tnext = first->b_tprev = jh;
1821         }
1822 }
1823 
1824 /*
1825  * Remove a buffer from a transaction list, given the transaction's list
1826  * head pointer.
1827  *
1828  * Called with j_list_lock held, and the journal may not be locked.
1829  *
1830  * jbd_lock_bh_state(jh2bh(jh)) is held.
1831  */
1832 
1833 static inline void
1834 __blist_del_buffer(struct journal_head **list, struct journal_head *jh)
1835 {
1836         if (*list == jh) {
1837                 *list = jh->b_tnext;
1838                 if (*list == jh)
1839                         *list = NULL;
1840         }
1841         jh->b_tprev->b_tnext = jh->b_tnext;
1842         jh->b_tnext->b_tprev = jh->b_tprev;
1843 }
1844 
1845 /*
1846  * Remove a buffer from the appropriate transaction list.
1847  *
1848  * Note that this function can *change* the value of
1849  * bh->b_transaction->t_buffers, t_forget, t_shadow_list, t_log_list or
1850  * t_reserved_list.  If the caller is holding onto a copy of one of these
1851  * pointers, it could go bad.  Generally the caller needs to re-read the
1852  * pointer from the transaction_t.
1853  *
1854  * Called under j_list_lock.
1855  */
1856 static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh)
1857 {
1858         struct journal_head **list = NULL;
1859         transaction_t *transaction;
1860         struct buffer_head *bh = jh2bh(jh);
1861 
1862         J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
1863         transaction = jh->b_transaction;
1864         if (transaction)
1865                 assert_spin_locked(&transaction->t_journal->j_list_lock);
1866 
1867         J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
1868         if (jh->b_jlist != BJ_None)
1869                 J_ASSERT_JH(jh, transaction != NULL);
1870 
1871         switch (jh->b_jlist) {
1872         case BJ_None:
1873                 return;
1874         case BJ_Metadata:
1875                 transaction->t_nr_buffers--;
1876                 J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0);
1877                 list = &transaction->t_buffers;
1878                 break;
1879         case BJ_Forget:
1880                 list = &transaction->t_forget;
1881                 break;
1882         case BJ_Shadow:
1883                 list = &transaction->t_shadow_list;
1884                 break;
1885         case BJ_Reserved:
1886                 list = &transaction->t_reserved_list;
1887                 break;
1888         }
1889 
1890         __blist_del_buffer(list, jh);
1891         jh->b_jlist = BJ_None;
1892         if (transaction && is_journal_aborted(transaction->t_journal))
1893                 clear_buffer_jbddirty(bh);
1894         else if (test_clear_buffer_jbddirty(bh))
1895                 mark_buffer_dirty(bh);  /* Expose it to the VM */
1896 }
1897 
1898 /*
1899  * Remove buffer from all transactions.
1900  *
1901  * Called with bh_state lock and j_list_lock
1902  *
1903  * jh and bh may be already freed when this function returns.
1904  */
1905 static void __jbd2_journal_unfile_buffer(struct journal_head *jh)
1906 {
1907         __jbd2_journal_temp_unlink_buffer(jh);
1908         jh->b_transaction = NULL;
1909         jbd2_journal_put_journal_head(jh);
1910 }
1911 
1912 void jbd2_journal_unfile_buffer(journal_t *journal, struct journal_head *jh)
1913 {
1914         struct buffer_head *bh = jh2bh(jh);
1915 
1916         /* Get reference so that buffer cannot be freed before we unlock it */
1917         get_bh(bh);
1918         jbd_lock_bh_state(bh);
1919         spin_lock(&journal->j_list_lock);
1920         __jbd2_journal_unfile_buffer(jh);
1921         spin_unlock(&journal->j_list_lock);
1922         jbd_unlock_bh_state(bh);
1923         __brelse(bh);
1924 }
1925 
1926 /*
1927  * Called from jbd2_journal_try_to_free_buffers().
1928  *
1929  * Called under jbd_lock_bh_state(bh)
1930  */
1931 static void
1932 __journal_try_to_free_buffer(journal_t *journal, struct buffer_head *bh)
1933 {
1934         struct journal_head *jh;
1935 
1936         jh = bh2jh(bh);
1937 
1938         if (buffer_locked(bh) || buffer_dirty(bh))
1939                 goto out;
1940 
1941         if (jh->b_next_transaction != NULL || jh->b_transaction != NULL)
1942                 goto out;
1943 
1944         spin_lock(&journal->j_list_lock);
1945         if (jh->b_cp_transaction != NULL) {
1946                 /* written-back checkpointed metadata buffer */
1947                 JBUFFER_TRACE(jh, "remove from checkpoint list");
1948                 __jbd2_journal_remove_checkpoint(jh);
1949         }
1950         spin_unlock(&journal->j_list_lock);
1951 out:
1952         return;
1953 }
1954 
1955 /**
1956  * int jbd2_journal_try_to_free_buffers() - try to free page buffers.
1957  * @journal: journal for operation
1958  * @page: to try and free
1959  * @gfp_mask: we use the mask to detect how hard should we try to release
1960  * buffers. If __GFP_DIRECT_RECLAIM and __GFP_FS is set, we wait for commit
1961  * code to release the buffers.
1962  *
1963  *
1964  * For all the buffers on this page,
1965  * if they are fully written out ordered data, move them onto BUF_CLEAN
1966  * so try_to_free_buffers() can reap them.
1967  *
1968  * This function returns non-zero if we wish try_to_free_buffers()
1969  * to be called. We do this if the page is releasable by try_to_free_buffers().
1970  * We also do it if the page has locked or dirty buffers and the caller wants
1971  * us to perform sync or async writeout.
1972  *
1973  * This complicates JBD locking somewhat.  We aren't protected by the
1974  * BKL here.  We wish to remove the buffer from its committing or
1975  * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer.
1976  *
1977  * This may *change* the value of transaction_t->t_datalist, so anyone
1978  * who looks at t_datalist needs to lock against this function.
1979  *
1980  * Even worse, someone may be doing a jbd2_journal_dirty_data on this
1981  * buffer.  So we need to lock against that.  jbd2_journal_dirty_data()
1982  * will come out of the lock with the buffer dirty, which makes it
1983  * ineligible for release here.
1984  *
1985  * Who else is affected by this?  hmm...  Really the only contender
1986  * is do_get_write_access() - it could be looking at the buffer while
1987  * journal_try_to_free_buffer() is changing its state.  But that
1988  * cannot happen because we never reallocate freed data as metadata
1989  * while the data is part of a transaction.  Yes?
1990  *
1991  * Return 0 on failure, 1 on success
1992  */
1993 int jbd2_journal_try_to_free_buffers(journal_t *journal,
1994                                 struct page *page, gfp_t gfp_mask)
1995 {
1996         struct buffer_head *head;
1997         struct buffer_head *bh;
1998         int ret = 0;
1999 
2000         J_ASSERT(PageLocked(page));
2001 
2002         head = page_buffers(page);
2003         bh = head;
2004         do {
2005                 struct journal_head *jh;
2006 
2007                 /*
2008                  * We take our own ref against the journal_head here to avoid
2009                  * having to add tons of locking around each instance of
2010                  * jbd2_journal_put_journal_head().
2011                  */
2012                 jh = jbd2_journal_grab_journal_head(bh);
2013                 if (!jh)
2014                         continue;
2015 
2016                 jbd_lock_bh_state(bh);
2017                 __journal_try_to_free_buffer(journal, bh);
2018                 jbd2_journal_put_journal_head(jh);
2019                 jbd_unlock_bh_state(bh);
2020                 if (buffer_jbd(bh))
2021                         goto busy;
2022         } while ((bh = bh->b_this_page) != head);
2023 
2024         ret = try_to_free_buffers(page);
2025 
2026 busy:
2027         return ret;
2028 }
2029 
2030 /*
2031  * This buffer is no longer needed.  If it is on an older transaction's
2032  * checkpoint list we need to record it on this transaction's forget list
2033  * to pin this buffer (and hence its checkpointing transaction) down until
2034  * this transaction commits.  If the buffer isn't on a checkpoint list, we
2035  * release it.
2036  * Returns non-zero if JBD no longer has an interest in the buffer.
2037  *
2038  * Called under j_list_lock.
2039  *
2040  * Called under jbd_lock_bh_state(bh).
2041  */
2042 static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction)
2043 {
2044         int may_free = 1;
2045         struct buffer_head *bh = jh2bh(jh);
2046 
2047         if (jh->b_cp_transaction) {
2048                 JBUFFER_TRACE(jh, "on running+cp transaction");
2049                 __jbd2_journal_temp_unlink_buffer(jh);
2050                 /*
2051                  * We don't want to write the buffer anymore, clear the
2052                  * bit so that we don't confuse checks in
2053                  * __journal_file_buffer
2054                  */
2055                 clear_buffer_dirty(bh);
2056                 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
2057                 may_free = 0;
2058         } else {
2059                 JBUFFER_TRACE(jh, "on running transaction");
2060                 __jbd2_journal_unfile_buffer(jh);
2061         }
2062         return may_free;
2063 }
2064 
2065 /*
2066  * jbd2_journal_invalidatepage
2067  *
2068  * This code is tricky.  It has a number of cases to deal with.
2069  *
2070  * There are two invariants which this code relies on:
2071  *
2072  * i_size must be updated on disk before we start calling invalidatepage on the
2073  * data.
2074  *
2075  *  This is done in ext3 by defining an ext3_setattr method which
2076  *  updates i_size before truncate gets going.  By maintaining this
2077  *  invariant, we can be sure that it is safe to throw away any buffers
2078  *  attached to the current transaction: once the transaction commits,
2079  *  we know that the data will not be needed.
2080  *
2081  *  Note however that we can *not* throw away data belonging to the
2082  *  previous, committing transaction!
2083  *
2084  * Any disk blocks which *are* part of the previous, committing
2085  * transaction (and which therefore cannot be discarded immediately) are
2086  * not going to be reused in the new running transaction
2087  *
2088  *  The bitmap committed_data images guarantee this: any block which is
2089  *  allocated in one transaction and removed in the next will be marked
2090  *  as in-use in the committed_data bitmap, so cannot be reused until
2091  *  the next transaction to delete the block commits.  This means that
2092  *  leaving committing buffers dirty is quite safe: the disk blocks
2093  *  cannot be reallocated to a different file and so buffer aliasing is
2094  *  not possible.
2095  *
2096  *
2097  * The above applies mainly to ordered data mode.  In writeback mode we
2098  * don't make guarantees about the order in which data hits disk --- in
2099  * particular we don't guarantee that new dirty data is flushed before
2100  * transaction commit --- so it is always safe just to discard data
2101  * immediately in that mode.  --sct
2102  */
2103 
2104 /*
2105  * The journal_unmap_buffer helper function returns zero if the buffer
2106  * concerned remains pinned as an anonymous buffer belonging to an older
2107  * transaction.
2108  *
2109  * We're outside-transaction here.  Either or both of j_running_transaction
2110  * and j_committing_transaction may be NULL.
2111  */
2112 static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh,
2113                                 int partial_page)
2114 {
2115         transaction_t *transaction;
2116         struct journal_head *jh;
2117         int may_free = 1;
2118 
2119         BUFFER_TRACE(bh, "entry");
2120 
2121         /*
2122          * It is safe to proceed here without the j_list_lock because the
2123          * buffers cannot be stolen by try_to_free_buffers as long as we are
2124          * holding the page lock. --sct
2125          */
2126 
2127         if (!buffer_jbd(bh))
2128                 goto zap_buffer_unlocked;
2129 
2130         /* OK, we have data buffer in journaled mode */
2131         write_lock(&journal->j_state_lock);
2132         jbd_lock_bh_state(bh);
2133         spin_lock(&journal->j_list_lock);
2134 
2135         jh = jbd2_journal_grab_journal_head(bh);
2136         if (!jh)
2137                 goto zap_buffer_no_jh;
2138 
2139         /*
2140          * We cannot remove the buffer from checkpoint lists until the
2141          * transaction adding inode to orphan list (let's call it T)
2142          * is committed.  Otherwise if the transaction changing the
2143          * buffer would be cleaned from the journal before T is
2144          * committed, a crash will cause that the correct contents of
2145          * the buffer will be lost.  On the other hand we have to
2146          * clear the buffer dirty bit at latest at the moment when the
2147          * transaction marking the buffer as freed in the filesystem
2148          * structures is committed because from that moment on the
2149          * block can be reallocated and used by a different page.
2150          * Since the block hasn't been freed yet but the inode has
2151          * already been added to orphan list, it is safe for us to add
2152          * the buffer to BJ_Forget list of the newest transaction.
2153          *
2154          * Also we have to clear buffer_mapped flag of a truncated buffer
2155          * because the buffer_head may be attached to the page straddling
2156          * i_size (can happen only when blocksize < pagesize) and thus the
2157          * buffer_head can be reused when the file is extended again. So we end
2158          * up keeping around invalidated buffers attached to transactions'
2159          * BJ_Forget list just to stop checkpointing code from cleaning up
2160          * the transaction this buffer was modified in.
2161          */
2162         transaction = jh->b_transaction;
2163         if (transaction == NULL) {
2164                 /* First case: not on any transaction.  If it
2165                  * has no checkpoint link, then we can zap it:
2166                  * it's a writeback-mode buffer so we don't care
2167                  * if it hits disk safely. */
2168                 if (!jh->b_cp_transaction) {
2169                         JBUFFER_TRACE(jh, "not on any transaction: zap");
2170                         goto zap_buffer;
2171                 }
2172 
2173                 if (!buffer_dirty(bh)) {
2174                         /* bdflush has written it.  We can drop it now */
2175                         __jbd2_journal_remove_checkpoint(jh);
2176                         goto zap_buffer;
2177                 }
2178 
2179                 /* OK, it must be in the journal but still not
2180                  * written fully to disk: it's metadata or
2181                  * journaled data... */
2182 
2183                 if (journal->j_running_transaction) {
2184                         /* ... and once the current transaction has
2185                          * committed, the buffer won't be needed any
2186                          * longer. */
2187                         JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget");
2188                         may_free = __dispose_buffer(jh,
2189                                         journal->j_running_transaction);
2190                         goto zap_buffer;
2191                 } else {
2192                         /* There is no currently-running transaction. So the
2193                          * orphan record which we wrote for this file must have
2194                          * passed into commit.  We must attach this buffer to
2195                          * the committing transaction, if it exists. */
2196                         if (journal->j_committing_transaction) {
2197                                 JBUFFER_TRACE(jh, "give to committing trans");
2198                                 may_free = __dispose_buffer(jh,
2199                                         journal->j_committing_transaction);
2200                                 goto zap_buffer;
2201                         } else {
2202                                 /* The orphan record's transaction has
2203                                  * committed.  We can cleanse this buffer */
2204                                 clear_buffer_jbddirty(bh);
2205                                 __jbd2_journal_remove_checkpoint(jh);
2206                                 goto zap_buffer;
2207                         }
2208                 }
2209         } else if (transaction == journal->j_committing_transaction) {
2210                 JBUFFER_TRACE(jh, "on committing transaction");
2211                 /*
2212                  * The buffer is committing, we simply cannot touch
2213                  * it. If the page is straddling i_size we have to wait
2214                  * for commit and try again.
2215                  */
2216                 if (partial_page) {
2217                         jbd2_journal_put_journal_head(jh);
2218                         spin_unlock(&journal->j_list_lock);
2219                         jbd_unlock_bh_state(bh);
2220                         write_unlock(&journal->j_state_lock);
2221                         return -EBUSY;
2222                 }
2223                 /*
2224                  * OK, buffer won't be reachable after truncate. We just set
2225                  * j_next_transaction to the running transaction (if there is
2226                  * one) and mark buffer as freed so that commit code knows it
2227                  * should clear dirty bits when it is done with the buffer.
2228                  */
2229                 set_buffer_freed(bh);
2230                 if (journal->j_running_transaction && buffer_jbddirty(bh))
2231                         jh->b_next_transaction = journal->j_running_transaction;
2232                 jbd2_journal_put_journal_head(jh);
2233                 spin_unlock(&journal->j_list_lock);
2234                 jbd_unlock_bh_state(bh);
2235                 write_unlock(&journal->j_state_lock);
2236                 return 0;
2237         } else {
2238                 /* Good, the buffer belongs to the running transaction.
2239                  * We are writing our own transaction's data, not any
2240                  * previous one's, so it is safe to throw it away
2241                  * (remember that we expect the filesystem to have set
2242                  * i_size already for this truncate so recovery will not
2243                  * expose the disk blocks we are discarding here.) */
2244                 J_ASSERT_JH(jh, transaction == journal->j_running_transaction);
2245                 JBUFFER_TRACE(jh, "on running transaction");
2246                 may_free = __dispose_buffer(jh, transaction);
2247         }
2248 
2249 zap_buffer:
2250         /*
2251          * This is tricky. Although the buffer is truncated, it may be reused
2252          * if blocksize < pagesize and it is attached to the page straddling
2253          * EOF. Since the buffer might have been added to BJ_Forget list of the
2254          * running transaction, journal_get_write_access() won't clear
2255          * b_modified and credit accounting gets confused. So clear b_modified
2256          * here.
2257          */
2258         jh->b_modified = 0;
2259         jbd2_journal_put_journal_head(jh);
2260 zap_buffer_no_jh:
2261         spin_unlock(&journal->j_list_lock);
2262         jbd_unlock_bh_state(bh);
2263         write_unlock(&journal->j_state_lock);
2264 zap_buffer_unlocked:
2265         clear_buffer_dirty(bh);
2266         J_ASSERT_BH(bh, !buffer_jbddirty(bh));
2267         clear_buffer_mapped(bh);
2268         clear_buffer_req(bh);
2269         clear_buffer_new(bh);
2270         clear_buffer_delay(bh);
2271         clear_buffer_unwritten(bh);
2272         bh->b_bdev = NULL;
2273         return may_free;
2274 }
2275 
2276 /**
2277  * void jbd2_journal_invalidatepage()
2278  * @journal: journal to use for flush...
2279  * @page:    page to flush
2280  * @offset:  start of the range to invalidate
2281  * @length:  length of the range to invalidate
2282  *
2283  * Reap page buffers containing data after in the specified range in page.
2284  * Can return -EBUSY if buffers are part of the committing transaction and
2285  * the page is straddling i_size. Caller then has to wait for current commit
2286  * and try again.
2287  */
2288 int jbd2_journal_invalidatepage(journal_t *journal,
2289                                 struct page *page,
2290                                 unsigned int offset,
2291                                 unsigned int length)
2292 {
2293         struct buffer_head *head, *bh, *next;
2294         unsigned int stop = offset + length;
2295         unsigned int curr_off = 0;
2296         int partial_page = (offset || length < PAGE_SIZE);
2297         int may_free = 1;
2298         int ret = 0;
2299 
2300         if (!PageLocked(page))
2301                 BUG();
2302         if (!page_has_buffers(page))
2303                 return 0;
2304 
2305         BUG_ON(stop > PAGE_SIZE || stop < length);
2306 
2307         /* We will potentially be playing with lists other than just the
2308          * data lists (especially for journaled data mode), so be
2309          * cautious in our locking. */
2310 
2311         head = bh = page_buffers(page);
2312         do {
2313                 unsigned int next_off = curr_off + bh->b_size;
2314                 next = bh->b_this_page;
2315 
2316                 if (next_off > stop)
2317                         return 0;
2318 
2319                 if (offset <= curr_off) {
2320                         /* This block is wholly outside the truncation point */
2321                         lock_buffer(bh);
2322                         ret = journal_unmap_buffer(journal, bh, partial_page);
2323                         unlock_buffer(bh);
2324                         if (ret < 0)
2325                                 return ret;
2326                         may_free &= ret;
2327                 }
2328                 curr_off = next_off;
2329                 bh = next;
2330 
2331         } while (bh != head);
2332 
2333         if (!partial_page) {
2334                 if (may_free && try_to_free_buffers(page))
2335                         J_ASSERT(!page_has_buffers(page));
2336         }
2337         return 0;
2338 }
2339 
2340 /*
2341  * File a buffer on the given transaction list.
2342  */
2343 void __jbd2_journal_file_buffer(struct journal_head *jh,
2344                         transaction_t *transaction, int jlist)
2345 {
2346         struct journal_head **list = NULL;
2347         int was_dirty = 0;
2348         struct buffer_head *bh = jh2bh(jh);
2349 
2350         J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
2351         assert_spin_locked(&transaction->t_journal->j_list_lock);
2352 
2353         J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
2354         J_ASSERT_JH(jh, jh->b_transaction == transaction ||
2355                                 jh->b_transaction == NULL);
2356 
2357         if (jh->b_transaction && jh->b_jlist == jlist)
2358                 return;
2359 
2360         if (jlist == BJ_Metadata || jlist == BJ_Reserved ||
2361             jlist == BJ_Shadow || jlist == BJ_Forget) {
2362                 /*
2363                  * For metadata buffers, we track dirty bit in buffer_jbddirty
2364                  * instead of buffer_dirty. We should not see a dirty bit set
2365                  * here because we clear it in do_get_write_access but e.g.
2366                  * tune2fs can modify the sb and set the dirty bit at any time
2367                  * so we try to gracefully handle that.
2368                  */
2369                 if (buffer_dirty(bh))
2370                         warn_dirty_buffer(bh);
2371                 if (test_clear_buffer_dirty(bh) ||
2372                     test_clear_buffer_jbddirty(bh))
2373                         was_dirty = 1;
2374         }
2375 
2376         if (jh->b_transaction)
2377                 __jbd2_journal_temp_unlink_buffer(jh);
2378         else
2379                 jbd2_journal_grab_journal_head(bh);
2380         jh->b_transaction = transaction;
2381 
2382         switch (jlist) {
2383         case BJ_None:
2384                 J_ASSERT_JH(jh, !jh->b_committed_data);
2385                 J_ASSERT_JH(jh, !jh->b_frozen_data);
2386                 return;
2387         case BJ_Metadata:
2388                 transaction->t_nr_buffers++;
2389                 list = &transaction->t_buffers;
2390                 break;
2391         case BJ_Forget:
2392                 list = &transaction->t_forget;
2393                 break;
2394         case BJ_Shadow:
2395                 list = &transaction->t_shadow_list;
2396                 break;
2397         case BJ_Reserved:
2398                 list = &transaction->t_reserved_list;
2399                 break;
2400         }
2401 
2402         __blist_add_buffer(list, jh);
2403         jh->b_jlist = jlist;
2404 
2405         if (was_dirty)
2406                 set_buffer_jbddirty(bh);
2407 }
2408 
2409 void jbd2_journal_file_buffer(struct journal_head *jh,
2410                                 transaction_t *transaction, int jlist)
2411 {
2412         jbd_lock_bh_state(jh2bh(jh));
2413         spin_lock(&transaction->t_journal->j_list_lock);
2414         __jbd2_journal_file_buffer(jh, transaction, jlist);
2415         spin_unlock(&transaction->t_journal->j_list_lock);
2416         jbd_unlock_bh_state(jh2bh(jh));
2417 }
2418 
2419 /*
2420  * Remove a buffer from its current buffer list in preparation for
2421  * dropping it from its current transaction entirely.  If the buffer has
2422  * already started to be used by a subsequent transaction, refile the
2423  * buffer on that transaction's metadata list.
2424  *
2425  * Called under j_list_lock
2426  * Called under jbd_lock_bh_state(jh2bh(jh))
2427  *
2428  * jh and bh may be already free when this function returns
2429  */
2430 void __jbd2_journal_refile_buffer(struct journal_head *jh)
2431 {
2432         int was_dirty, jlist;
2433         struct buffer_head *bh = jh2bh(jh);
2434 
2435         J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
2436         if (jh->b_transaction)
2437                 assert_spin_locked(&jh->b_transaction->t_journal->j_list_lock);
2438 
2439         /* If the buffer is now unused, just drop it. */
2440         if (jh->b_next_transaction == NULL) {
2441                 __jbd2_journal_unfile_buffer(jh);
2442                 return;
2443         }
2444 
2445         /*
2446          * It has been modified by a later transaction: add it to the new
2447          * transaction's metadata list.
2448          */
2449 
2450         was_dirty = test_clear_buffer_jbddirty(bh);
2451         __jbd2_journal_temp_unlink_buffer(jh);
2452         /*
2453          * We set b_transaction here because b_next_transaction will inherit
2454          * our jh reference and thus __jbd2_journal_file_buffer() must not
2455          * take a new one.
2456          */
2457         jh->b_transaction = jh->b_next_transaction;
2458         jh->b_next_transaction = NULL;
2459         if (buffer_freed(bh))
2460                 jlist = BJ_Forget;
2461         else if (jh->b_modified)
2462                 jlist = BJ_Metadata;
2463         else
2464                 jlist = BJ_Reserved;
2465         __jbd2_journal_file_buffer(jh, jh->b_transaction, jlist);
2466         J_ASSERT_JH(jh, jh->b_transaction->t_state == T_RUNNING);
2467 
2468         if (was_dirty)
2469                 set_buffer_jbddirty(bh);
2470 }
2471 
2472 /*
2473  * __jbd2_journal_refile_buffer() with necessary locking added. We take our
2474  * bh reference so that we can safely unlock bh.
2475  *
2476  * The jh and bh may be freed by this call.
2477  */
2478 void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh)
2479 {
2480         struct buffer_head *bh = jh2bh(jh);
2481 
2482         /* Get reference so that buffer cannot be freed before we unlock it */
2483         get_bh(bh);
2484         jbd_lock_bh_state(bh);
2485         spin_lock(&journal->j_list_lock);
2486         __jbd2_journal_refile_buffer(jh);
2487         jbd_unlock_bh_state(bh);
2488         spin_unlock(&journal->j_list_lock);
2489         __brelse(bh);
2490 }
2491 
2492 /*
2493  * File inode in the inode list of the handle's transaction
2494  */
2495 static int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode,
2496                                    unsigned long flags)
2497 {
2498         transaction_t *transaction = handle->h_transaction;
2499         journal_t *journal;
2500 
2501         if (is_handle_aborted(handle))
2502                 return -EROFS;
2503         journal = transaction->t_journal;
2504 
2505         jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode->i_vfs_inode->i_ino,
2506                         transaction->t_tid);
2507 
2508         /*
2509          * First check whether inode isn't already on the transaction's
2510          * lists without taking the lock. Note that this check is safe
2511          * without the lock as we cannot race with somebody removing inode
2512          * from the transaction. The reason is that we remove inode from the
2513          * transaction only in journal_release_jbd_inode() and when we commit
2514          * the transaction. We are guarded from the first case by holding
2515          * a reference to the inode. We are safe against the second case
2516          * because if jinode->i_transaction == transaction, commit code
2517          * cannot touch the transaction because we hold reference to it,
2518          * and if jinode->i_next_transaction == transaction, commit code
2519          * will only file the inode where we want it.
2520          */
2521         if ((jinode->i_transaction == transaction ||
2522             jinode->i_next_transaction == transaction) &&
2523             (jinode->i_flags & flags) == flags)
2524                 return 0;
2525 
2526         spin_lock(&journal->j_list_lock);
2527         jinode->i_flags |= flags;
2528         /* Is inode already attached where we need it? */
2529         if (jinode->i_transaction == transaction ||
2530             jinode->i_next_transaction == transaction)
2531                 goto done;
2532 
2533         /*
2534          * We only ever set this variable to 1 so the test is safe. Since
2535          * t_need_data_flush is likely to be set, we do the test to save some
2536          * cacheline bouncing
2537          */
2538         if (!transaction->t_need_data_flush)
2539                 transaction->t_need_data_flush = 1;
2540         /* On some different transaction's list - should be
2541          * the committing one */
2542         if (jinode->i_transaction) {
2543                 J_ASSERT(jinode->i_next_transaction == NULL);
2544                 J_ASSERT(jinode->i_transaction ==
2545                                         journal->j_committing_transaction);
2546                 jinode->i_next_transaction = transaction;
2547                 goto done;
2548         }
2549         /* Not on any transaction list... */
2550         J_ASSERT(!jinode->i_next_transaction);
2551         jinode->i_transaction = transaction;
2552         list_add(&jinode->i_list, &transaction->t_inode_list);
2553 done:
2554         spin_unlock(&journal->j_list_lock);
2555 
2556         return 0;
2557 }
2558 
2559 int jbd2_journal_inode_add_write(handle_t *handle, struct jbd2_inode *jinode)
2560 {
2561         return jbd2_journal_file_inode(handle, jinode,
2562                                        JI_WRITE_DATA | JI_WAIT_DATA);
2563 }
2564 
2565 int jbd2_journal_inode_add_wait(handle_t *handle, struct jbd2_inode *jinode)
2566 {
2567         return jbd2_journal_file_inode(handle, jinode, JI_WAIT_DATA);
2568 }
2569 
2570 /*
2571  * File truncate and transaction commit interact with each other in a
2572  * non-trivial way.  If a transaction writing data block A is
2573  * committing, we cannot discard the data by truncate until we have
2574  * written them.  Otherwise if we crashed after the transaction with
2575  * write has committed but before the transaction with truncate has
2576  * committed, we could see stale data in block A.  This function is a
2577  * helper to solve this problem.  It starts writeout of the truncated
2578  * part in case it is in the committing transaction.
2579  *
2580  * Filesystem code must call this function when inode is journaled in
2581  * ordered mode before truncation happens and after the inode has been
2582  * placed on orphan list with the new inode size. The second condition
2583  * avoids the race that someone writes new data and we start
2584  * committing the transaction after this function has been called but
2585  * before a transaction for truncate is started (and furthermore it
2586  * allows us to optimize the case where the addition to orphan list
2587  * happens in the same transaction as write --- we don't have to write
2588  * any data in such case).
2589  */
2590 int jbd2_journal_begin_ordered_truncate(journal_t *journal,
2591                                         struct jbd2_inode *jinode,
2592                                         loff_t new_size)
2593 {
2594         transaction_t *inode_trans, *commit_trans;
2595         int ret = 0;
2596 
2597         /* This is a quick check to avoid locking if not necessary */
2598         if (!jinode->i_transaction)
2599                 goto out;
2600         /* Locks are here just to force reading of recent values, it is
2601          * enough that the transaction was not committing before we started
2602          * a transaction adding the inode to orphan list */
2603         read_lock(&journal->j_state_lock);
2604         commit_trans = journal->j_committing_transaction;
2605         read_unlock(&journal->j_state_lock);
2606         spin_lock(&journal->j_list_lock);
2607         inode_trans = jinode->i_transaction;
2608         spin_unlock(&journal->j_list_lock);
2609         if (inode_trans == commit_trans) {
2610                 ret = filemap_fdatawrite_range(jinode->i_vfs_inode->i_mapping,
2611                         new_size, LLONG_MAX);
2612                 if (ret)
2613                         jbd2_journal_abort(journal, ret);
2614         }
2615 out:
2616         return ret;
2617 }
2618 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp