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

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

Version: ~ [ linux-5.3-rc4 ] ~ [ linux-5.2.8 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.66 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.138 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.189 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.189 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.71 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ 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 /*
  2  * linux/fs/jbd2/transaction.c
  3  *
  4  * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
  5  *
  6  * Copyright 1998 Red Hat corp --- All Rights Reserved
  7  *
  8  * This file is part of the Linux kernel and is made available under
  9  * the terms of the GNU General Public License, version 2, or at your
 10  * option, any later version, incorporated herein by reference.
 11  *
 12  * Generic filesystem transaction handling code; part of the ext2fs
 13  * journaling system.
 14  *
 15  * This file manages transactions (compound commits managed by the
 16  * journaling code) and handles (individual atomic operations by the
 17  * filesystem).
 18  */
 19 
 20 #include <linux/time.h>
 21 #include <linux/fs.h>
 22 #include <linux/jbd2.h>
 23 #include <linux/errno.h>
 24 #include <linux/slab.h>
 25 #include <linux/timer.h>
 26 #include <linux/mm.h>
 27 #include <linux/highmem.h>
 28 #include <linux/hrtimer.h>
 29 #include <linux/backing-dev.h>
 30 #include <linux/bug.h>
 31 #include <linux/module.h>
 32 
 33 #include <trace/events/jbd2.h>
 34 
 35 static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh);
 36 static void __jbd2_journal_unfile_buffer(struct journal_head *jh);
 37 
 38 static struct kmem_cache *transaction_cache;
 39 int __init jbd2_journal_init_transaction_cache(void)
 40 {
 41         J_ASSERT(!transaction_cache);
 42         transaction_cache = kmem_cache_create("jbd2_transaction_s",
 43                                         sizeof(transaction_t),
 44                                         0,
 45                                         SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY,
 46                                         NULL);
 47         if (transaction_cache)
 48                 return 0;
 49         return -ENOMEM;
 50 }
 51 
 52 void jbd2_journal_destroy_transaction_cache(void)
 53 {
 54         if (transaction_cache) {
 55                 kmem_cache_destroy(transaction_cache);
 56                 transaction_cache = NULL;
 57         }
 58 }
 59 
 60 void jbd2_journal_free_transaction(transaction_t *transaction)
 61 {
 62         if (unlikely(ZERO_OR_NULL_PTR(transaction)))
 63                 return;
 64         kmem_cache_free(transaction_cache, transaction);
 65 }
 66 
 67 /*
 68  * jbd2_get_transaction: obtain a new transaction_t object.
 69  *
 70  * Simply allocate and initialise a new transaction.  Create it in
 71  * RUNNING state and add it to the current journal (which should not
 72  * have an existing running transaction: we only make a new transaction
 73  * once we have started to commit the old one).
 74  *
 75  * Preconditions:
 76  *      The journal MUST be locked.  We don't perform atomic mallocs on the
 77  *      new transaction and we can't block without protecting against other
 78  *      processes trying to touch the journal while it is in transition.
 79  *
 80  */
 81 
 82 static transaction_t *
 83 jbd2_get_transaction(journal_t *journal, transaction_t *transaction)
 84 {
 85         transaction->t_journal = journal;
 86         transaction->t_state = T_RUNNING;
 87         transaction->t_start_time = ktime_get();
 88         transaction->t_tid = journal->j_transaction_sequence++;
 89         transaction->t_expires = jiffies + journal->j_commit_interval;
 90         spin_lock_init(&transaction->t_handle_lock);
 91         atomic_set(&transaction->t_updates, 0);
 92         atomic_set(&transaction->t_outstanding_credits,
 93                    atomic_read(&journal->j_reserved_credits));
 94         atomic_set(&transaction->t_handle_count, 0);
 95         INIT_LIST_HEAD(&transaction->t_inode_list);
 96         INIT_LIST_HEAD(&transaction->t_private_list);
 97 
 98         /* Set up the commit timer for the new transaction. */
 99         journal->j_commit_timer.expires = round_jiffies_up(transaction->t_expires);
100         add_timer(&journal->j_commit_timer);
101 
102         J_ASSERT(journal->j_running_transaction == NULL);
103         journal->j_running_transaction = transaction;
104         transaction->t_max_wait = 0;
105         transaction->t_start = jiffies;
106         transaction->t_requested = 0;
107 
108         return transaction;
109 }
110 
111 /*
112  * Handle management.
113  *
114  * A handle_t is an object which represents a single atomic update to a
115  * filesystem, and which tracks all of the modifications which form part
116  * of that one update.
117  */
118 
119 /*
120  * Update transaction's maximum wait time, if debugging is enabled.
121  *
122  * In order for t_max_wait to be reliable, it must be protected by a
123  * lock.  But doing so will mean that start_this_handle() can not be
124  * run in parallel on SMP systems, which limits our scalability.  So
125  * unless debugging is enabled, we no longer update t_max_wait, which
126  * means that maximum wait time reported by the jbd2_run_stats
127  * tracepoint will always be zero.
128  */
129 static inline void update_t_max_wait(transaction_t *transaction,
130                                      unsigned long ts)
131 {
132 #ifdef CONFIG_JBD2_DEBUG
133         if (jbd2_journal_enable_debug &&
134             time_after(transaction->t_start, ts)) {
135                 ts = jbd2_time_diff(ts, transaction->t_start);
136                 spin_lock(&transaction->t_handle_lock);
137                 if (ts > transaction->t_max_wait)
138                         transaction->t_max_wait = ts;
139                 spin_unlock(&transaction->t_handle_lock);
140         }
141 #endif
142 }
143 
144 /*
145  * Wait until running transaction passes T_LOCKED state. Also starts the commit
146  * if needed. The function expects running transaction to exist and releases
147  * j_state_lock.
148  */
149 static void wait_transaction_locked(journal_t *journal)
150         __releases(journal->j_state_lock)
151 {
152         DEFINE_WAIT(wait);
153         int need_to_start;
154         tid_t tid = journal->j_running_transaction->t_tid;
155 
156         prepare_to_wait(&journal->j_wait_transaction_locked, &wait,
157                         TASK_UNINTERRUPTIBLE);
158         need_to_start = !tid_geq(journal->j_commit_request, tid);
159         read_unlock(&journal->j_state_lock);
160         if (need_to_start)
161                 jbd2_log_start_commit(journal, tid);
162         schedule();
163         finish_wait(&journal->j_wait_transaction_locked, &wait);
164 }
165 
166 static void sub_reserved_credits(journal_t *journal, int blocks)
167 {
168         atomic_sub(blocks, &journal->j_reserved_credits);
169         wake_up(&journal->j_wait_reserved);
170 }
171 
172 /*
173  * Wait until we can add credits for handle to the running transaction.  Called
174  * with j_state_lock held for reading. Returns 0 if handle joined the running
175  * transaction. Returns 1 if we had to wait, j_state_lock is dropped, and
176  * caller must retry.
177  */
178 static int add_transaction_credits(journal_t *journal, int blocks,
179                                    int rsv_blocks)
180 {
181         transaction_t *t = journal->j_running_transaction;
182         int needed;
183         int total = blocks + rsv_blocks;
184 
185         /*
186          * If the current transaction is locked down for commit, wait
187          * for the lock to be released.
188          */
189         if (t->t_state == T_LOCKED) {
190                 wait_transaction_locked(journal);
191                 return 1;
192         }
193 
194         /*
195          * If there is not enough space left in the log to write all
196          * potential buffers requested by this operation, we need to
197          * stall pending a log checkpoint to free some more log space.
198          */
199         needed = atomic_add_return(total, &t->t_outstanding_credits);
200         if (needed > journal->j_max_transaction_buffers) {
201                 /*
202                  * If the current transaction is already too large,
203                  * then start to commit it: we can then go back and
204                  * attach this handle to a new transaction.
205                  */
206                 atomic_sub(total, &t->t_outstanding_credits);
207                 wait_transaction_locked(journal);
208                 return 1;
209         }
210 
211         /*
212          * The commit code assumes that it can get enough log space
213          * without forcing a checkpoint.  This is *critical* for
214          * correctness: a checkpoint of a buffer which is also
215          * associated with a committing transaction creates a deadlock,
216          * so commit simply cannot force through checkpoints.
217          *
218          * We must therefore ensure the necessary space in the journal
219          * *before* starting to dirty potentially checkpointed buffers
220          * in the new transaction.
221          */
222         if (jbd2_log_space_left(journal) < jbd2_space_needed(journal)) {
223                 atomic_sub(total, &t->t_outstanding_credits);
224                 read_unlock(&journal->j_state_lock);
225                 write_lock(&journal->j_state_lock);
226                 if (jbd2_log_space_left(journal) < jbd2_space_needed(journal))
227                         __jbd2_log_wait_for_space(journal);
228                 write_unlock(&journal->j_state_lock);
229                 return 1;
230         }
231 
232         /* No reservation? We are done... */
233         if (!rsv_blocks)
234                 return 0;
235 
236         needed = atomic_add_return(rsv_blocks, &journal->j_reserved_credits);
237         /* We allow at most half of a transaction to be reserved */
238         if (needed > journal->j_max_transaction_buffers / 2) {
239                 sub_reserved_credits(journal, rsv_blocks);
240                 atomic_sub(total, &t->t_outstanding_credits);
241                 read_unlock(&journal->j_state_lock);
242                 wait_event(journal->j_wait_reserved,
243                          atomic_read(&journal->j_reserved_credits) + rsv_blocks
244                          <= journal->j_max_transaction_buffers / 2);
245                 return 1;
246         }
247         return 0;
248 }
249 
250 /*
251  * start_this_handle: Given a handle, deal with any locking or stalling
252  * needed to make sure that there is enough journal space for the handle
253  * to begin.  Attach the handle to a transaction and set up the
254  * transaction's buffer credits.
255  */
256 
257 static int start_this_handle(journal_t *journal, handle_t *handle,
258                              gfp_t gfp_mask)
259 {
260         transaction_t   *transaction, *new_transaction = NULL;
261         int             blocks = handle->h_buffer_credits;
262         int             rsv_blocks = 0;
263         unsigned long ts = jiffies;
264 
265         /*
266          * 1/2 of transaction can be reserved so we can practically handle
267          * only 1/2 of maximum transaction size per operation
268          */
269         if (WARN_ON(blocks > journal->j_max_transaction_buffers / 2)) {
270                 printk(KERN_ERR "JBD2: %s wants too many credits (%d > %d)\n",
271                        current->comm, blocks,
272                        journal->j_max_transaction_buffers / 2);
273                 return -ENOSPC;
274         }
275 
276         if (handle->h_rsv_handle)
277                 rsv_blocks = handle->h_rsv_handle->h_buffer_credits;
278 
279 alloc_transaction:
280         if (!journal->j_running_transaction) {
281                 new_transaction = kmem_cache_zalloc(transaction_cache,
282                                                     gfp_mask);
283                 if (!new_transaction) {
284                         /*
285                          * If __GFP_FS is not present, then we may be
286                          * being called from inside the fs writeback
287                          * layer, so we MUST NOT fail.  Since
288                          * __GFP_NOFAIL is going away, we will arrange
289                          * to retry the allocation ourselves.
290                          */
291                         if ((gfp_mask & __GFP_FS) == 0) {
292                                 congestion_wait(BLK_RW_ASYNC, HZ/50);
293                                 goto alloc_transaction;
294                         }
295                         return -ENOMEM;
296                 }
297         }
298 
299         jbd_debug(3, "New handle %p going live.\n", handle);
300 
301         /*
302          * We need to hold j_state_lock until t_updates has been incremented,
303          * for proper journal barrier handling
304          */
305 repeat:
306         read_lock(&journal->j_state_lock);
307         BUG_ON(journal->j_flags & JBD2_UNMOUNT);
308         if (is_journal_aborted(journal) ||
309             (journal->j_errno != 0 && !(journal->j_flags & JBD2_ACK_ERR))) {
310                 read_unlock(&journal->j_state_lock);
311                 jbd2_journal_free_transaction(new_transaction);
312                 return -EROFS;
313         }
314 
315         /*
316          * Wait on the journal's transaction barrier if necessary. Specifically
317          * we allow reserved handles to proceed because otherwise commit could
318          * deadlock on page writeback not being able to complete.
319          */
320         if (!handle->h_reserved && journal->j_barrier_count) {
321                 read_unlock(&journal->j_state_lock);
322                 wait_event(journal->j_wait_transaction_locked,
323                                 journal->j_barrier_count == 0);
324                 goto repeat;
325         }
326 
327         if (!journal->j_running_transaction) {
328                 read_unlock(&journal->j_state_lock);
329                 if (!new_transaction)
330                         goto alloc_transaction;
331                 write_lock(&journal->j_state_lock);
332                 if (!journal->j_running_transaction &&
333                     (handle->h_reserved || !journal->j_barrier_count)) {
334                         jbd2_get_transaction(journal, new_transaction);
335                         new_transaction = NULL;
336                 }
337                 write_unlock(&journal->j_state_lock);
338                 goto repeat;
339         }
340 
341         transaction = journal->j_running_transaction;
342 
343         if (!handle->h_reserved) {
344                 /* We may have dropped j_state_lock - restart in that case */
345                 if (add_transaction_credits(journal, blocks, rsv_blocks))
346                         goto repeat;
347         } else {
348                 /*
349                  * We have handle reserved so we are allowed to join T_LOCKED
350                  * transaction and we don't have to check for transaction size
351                  * and journal space.
352                  */
353                 sub_reserved_credits(journal, blocks);
354                 handle->h_reserved = 0;
355         }
356 
357         /* OK, account for the buffers that this operation expects to
358          * use and add the handle to the running transaction. 
359          */
360         update_t_max_wait(transaction, ts);
361         handle->h_transaction = transaction;
362         handle->h_requested_credits = blocks;
363         handle->h_start_jiffies = jiffies;
364         atomic_inc(&transaction->t_updates);
365         atomic_inc(&transaction->t_handle_count);
366         jbd_debug(4, "Handle %p given %d credits (total %d, free %lu)\n",
367                   handle, blocks,
368                   atomic_read(&transaction->t_outstanding_credits),
369                   jbd2_log_space_left(journal));
370         read_unlock(&journal->j_state_lock);
371         current->journal_info = handle;
372 
373         lock_map_acquire(&handle->h_lockdep_map);
374         jbd2_journal_free_transaction(new_transaction);
375         return 0;
376 }
377 
378 static struct lock_class_key jbd2_handle_key;
379 
380 /* Allocate a new handle.  This should probably be in a slab... */
381 static handle_t *new_handle(int nblocks)
382 {
383         handle_t *handle = jbd2_alloc_handle(GFP_NOFS);
384         if (!handle)
385                 return NULL;
386         handle->h_buffer_credits = nblocks;
387         handle->h_ref = 1;
388 
389         lockdep_init_map(&handle->h_lockdep_map, "jbd2_handle",
390                                                 &jbd2_handle_key, 0);
391 
392         return handle;
393 }
394 
395 /**
396  * handle_t *jbd2_journal_start() - Obtain a new handle.
397  * @journal: Journal to start transaction on.
398  * @nblocks: number of block buffer we might modify
399  *
400  * We make sure that the transaction can guarantee at least nblocks of
401  * modified buffers in the log.  We block until the log can guarantee
402  * that much space. Additionally, if rsv_blocks > 0, we also create another
403  * handle with rsv_blocks reserved blocks in the journal. This handle is
404  * is stored in h_rsv_handle. It is not attached to any particular transaction
405  * and thus doesn't block transaction commit. If the caller uses this reserved
406  * handle, it has to set h_rsv_handle to NULL as otherwise jbd2_journal_stop()
407  * on the parent handle will dispose the reserved one. Reserved handle has to
408  * be converted to a normal handle using jbd2_journal_start_reserved() before
409  * it can be used.
410  *
411  * Return a pointer to a newly allocated handle, or an ERR_PTR() value
412  * on failure.
413  */
414 handle_t *jbd2__journal_start(journal_t *journal, int nblocks, int rsv_blocks,
415                               gfp_t gfp_mask, unsigned int type,
416                               unsigned int line_no)
417 {
418         handle_t *handle = journal_current_handle();
419         int err;
420 
421         if (!journal)
422                 return ERR_PTR(-EROFS);
423 
424         if (handle) {
425                 J_ASSERT(handle->h_transaction->t_journal == journal);
426                 handle->h_ref++;
427                 return handle;
428         }
429 
430         handle = new_handle(nblocks);
431         if (!handle)
432                 return ERR_PTR(-ENOMEM);
433         if (rsv_blocks) {
434                 handle_t *rsv_handle;
435 
436                 rsv_handle = new_handle(rsv_blocks);
437                 if (!rsv_handle) {
438                         jbd2_free_handle(handle);
439                         return ERR_PTR(-ENOMEM);
440                 }
441                 rsv_handle->h_reserved = 1;
442                 rsv_handle->h_journal = journal;
443                 handle->h_rsv_handle = rsv_handle;
444         }
445 
446         err = start_this_handle(journal, handle, gfp_mask);
447         if (err < 0) {
448                 if (handle->h_rsv_handle)
449                         jbd2_free_handle(handle->h_rsv_handle);
450                 jbd2_free_handle(handle);
451                 return ERR_PTR(err);
452         }
453         handle->h_type = type;
454         handle->h_line_no = line_no;
455         trace_jbd2_handle_start(journal->j_fs_dev->bd_dev,
456                                 handle->h_transaction->t_tid, type,
457                                 line_no, nblocks);
458         return handle;
459 }
460 EXPORT_SYMBOL(jbd2__journal_start);
461 
462 
463 handle_t *jbd2_journal_start(journal_t *journal, int nblocks)
464 {
465         return jbd2__journal_start(journal, nblocks, 0, GFP_NOFS, 0, 0);
466 }
467 EXPORT_SYMBOL(jbd2_journal_start);
468 
469 void jbd2_journal_free_reserved(handle_t *handle)
470 {
471         journal_t *journal = handle->h_journal;
472 
473         WARN_ON(!handle->h_reserved);
474         sub_reserved_credits(journal, handle->h_buffer_credits);
475         jbd2_free_handle(handle);
476 }
477 EXPORT_SYMBOL(jbd2_journal_free_reserved);
478 
479 /**
480  * int jbd2_journal_start_reserved(handle_t *handle) - start reserved handle
481  * @handle: handle to start
482  *
483  * Start handle that has been previously reserved with jbd2_journal_reserve().
484  * This attaches @handle to the running transaction (or creates one if there's
485  * not transaction running). Unlike jbd2_journal_start() this function cannot
486  * block on journal commit, checkpointing, or similar stuff. It can block on
487  * memory allocation or frozen journal though.
488  *
489  * Return 0 on success, non-zero on error - handle is freed in that case.
490  */
491 int jbd2_journal_start_reserved(handle_t *handle, unsigned int type,
492                                 unsigned int line_no)
493 {
494         journal_t *journal = handle->h_journal;
495         int ret = -EIO;
496 
497         if (WARN_ON(!handle->h_reserved)) {
498                 /* Someone passed in normal handle? Just stop it. */
499                 jbd2_journal_stop(handle);
500                 return ret;
501         }
502         /*
503          * Usefulness of mixing of reserved and unreserved handles is
504          * questionable. So far nobody seems to need it so just error out.
505          */
506         if (WARN_ON(current->journal_info)) {
507                 jbd2_journal_free_reserved(handle);
508                 return ret;
509         }
510 
511         handle->h_journal = NULL;
512         /*
513          * GFP_NOFS is here because callers are likely from writeback or
514          * similarly constrained call sites
515          */
516         ret = start_this_handle(journal, handle, GFP_NOFS);
517         if (ret < 0)
518                 jbd2_journal_free_reserved(handle);
519         handle->h_type = type;
520         handle->h_line_no = line_no;
521         return ret;
522 }
523 EXPORT_SYMBOL(jbd2_journal_start_reserved);
524 
525 /**
526  * int jbd2_journal_extend() - extend buffer credits.
527  * @handle:  handle to 'extend'
528  * @nblocks: nr blocks to try to extend by.
529  *
530  * Some transactions, such as large extends and truncates, can be done
531  * atomically all at once or in several stages.  The operation requests
532  * a credit for a number of buffer modications in advance, but can
533  * extend its credit if it needs more.
534  *
535  * jbd2_journal_extend tries to give the running handle more buffer credits.
536  * It does not guarantee that allocation - this is a best-effort only.
537  * The calling process MUST be able to deal cleanly with a failure to
538  * extend here.
539  *
540  * Return 0 on success, non-zero on failure.
541  *
542  * return code < 0 implies an error
543  * return code > 0 implies normal transaction-full status.
544  */
545 int jbd2_journal_extend(handle_t *handle, int nblocks)
546 {
547         transaction_t *transaction = handle->h_transaction;
548         journal_t *journal;
549         int result;
550         int wanted;
551 
552         WARN_ON(!transaction);
553         if (is_handle_aborted(handle))
554                 return -EROFS;
555         journal = transaction->t_journal;
556 
557         result = 1;
558 
559         read_lock(&journal->j_state_lock);
560 
561         /* Don't extend a locked-down transaction! */
562         if (transaction->t_state != T_RUNNING) {
563                 jbd_debug(3, "denied handle %p %d blocks: "
564                           "transaction not running\n", handle, nblocks);
565                 goto error_out;
566         }
567 
568         spin_lock(&transaction->t_handle_lock);
569         wanted = atomic_add_return(nblocks,
570                                    &transaction->t_outstanding_credits);
571 
572         if (wanted > journal->j_max_transaction_buffers) {
573                 jbd_debug(3, "denied handle %p %d blocks: "
574                           "transaction too large\n", handle, nblocks);
575                 atomic_sub(nblocks, &transaction->t_outstanding_credits);
576                 goto unlock;
577         }
578 
579         if (wanted + (wanted >> JBD2_CONTROL_BLOCKS_SHIFT) >
580             jbd2_log_space_left(journal)) {
581                 jbd_debug(3, "denied handle %p %d blocks: "
582                           "insufficient log space\n", handle, nblocks);
583                 atomic_sub(nblocks, &transaction->t_outstanding_credits);
584                 goto unlock;
585         }
586 
587         trace_jbd2_handle_extend(journal->j_fs_dev->bd_dev,
588                                  transaction->t_tid,
589                                  handle->h_type, handle->h_line_no,
590                                  handle->h_buffer_credits,
591                                  nblocks);
592 
593         handle->h_buffer_credits += nblocks;
594         handle->h_requested_credits += nblocks;
595         result = 0;
596 
597         jbd_debug(3, "extended handle %p by %d\n", handle, nblocks);
598 unlock:
599         spin_unlock(&transaction->t_handle_lock);
600 error_out:
601         read_unlock(&journal->j_state_lock);
602         return result;
603 }
604 
605 
606 /**
607  * int jbd2_journal_restart() - restart a handle .
608  * @handle:  handle to restart
609  * @nblocks: nr credits requested
610  *
611  * Restart a handle for a multi-transaction filesystem
612  * operation.
613  *
614  * If the jbd2_journal_extend() call above fails to grant new buffer credits
615  * to a running handle, a call to jbd2_journal_restart will commit the
616  * handle's transaction so far and reattach the handle to a new
617  * transaction capabable of guaranteeing the requested number of
618  * credits. We preserve reserved handle if there's any attached to the
619  * passed in handle.
620  */
621 int jbd2__journal_restart(handle_t *handle, int nblocks, gfp_t gfp_mask)
622 {
623         transaction_t *transaction = handle->h_transaction;
624         journal_t *journal;
625         tid_t           tid;
626         int             need_to_start, ret;
627 
628         WARN_ON(!transaction);
629         /* If we've had an abort of any type, don't even think about
630          * actually doing the restart! */
631         if (is_handle_aborted(handle))
632                 return 0;
633         journal = transaction->t_journal;
634 
635         /*
636          * First unlink the handle from its current transaction, and start the
637          * commit on that.
638          */
639         J_ASSERT(atomic_read(&transaction->t_updates) > 0);
640         J_ASSERT(journal_current_handle() == handle);
641 
642         read_lock(&journal->j_state_lock);
643         spin_lock(&transaction->t_handle_lock);
644         atomic_sub(handle->h_buffer_credits,
645                    &transaction->t_outstanding_credits);
646         if (handle->h_rsv_handle) {
647                 sub_reserved_credits(journal,
648                                      handle->h_rsv_handle->h_buffer_credits);
649         }
650         if (atomic_dec_and_test(&transaction->t_updates))
651                 wake_up(&journal->j_wait_updates);
652         tid = transaction->t_tid;
653         spin_unlock(&transaction->t_handle_lock);
654         handle->h_transaction = NULL;
655         current->journal_info = NULL;
656 
657         jbd_debug(2, "restarting handle %p\n", handle);
658         need_to_start = !tid_geq(journal->j_commit_request, tid);
659         read_unlock(&journal->j_state_lock);
660         if (need_to_start)
661                 jbd2_log_start_commit(journal, tid);
662 
663         lock_map_release(&handle->h_lockdep_map);
664         handle->h_buffer_credits = nblocks;
665         ret = start_this_handle(journal, handle, gfp_mask);
666         return ret;
667 }
668 EXPORT_SYMBOL(jbd2__journal_restart);
669 
670 
671 int jbd2_journal_restart(handle_t *handle, int nblocks)
672 {
673         return jbd2__journal_restart(handle, nblocks, GFP_NOFS);
674 }
675 EXPORT_SYMBOL(jbd2_journal_restart);
676 
677 /**
678  * void jbd2_journal_lock_updates () - establish a transaction barrier.
679  * @journal:  Journal to establish a barrier on.
680  *
681  * This locks out any further updates from being started, and blocks
682  * until all existing updates have completed, returning only once the
683  * journal is in a quiescent state with no updates running.
684  *
685  * The journal lock should not be held on entry.
686  */
687 void jbd2_journal_lock_updates(journal_t *journal)
688 {
689         DEFINE_WAIT(wait);
690 
691         write_lock(&journal->j_state_lock);
692         ++journal->j_barrier_count;
693 
694         /* Wait until there are no reserved handles */
695         if (atomic_read(&journal->j_reserved_credits)) {
696                 write_unlock(&journal->j_state_lock);
697                 wait_event(journal->j_wait_reserved,
698                            atomic_read(&journal->j_reserved_credits) == 0);
699                 write_lock(&journal->j_state_lock);
700         }
701 
702         /* Wait until there are no running updates */
703         while (1) {
704                 transaction_t *transaction = journal->j_running_transaction;
705 
706                 if (!transaction)
707                         break;
708 
709                 spin_lock(&transaction->t_handle_lock);
710                 prepare_to_wait(&journal->j_wait_updates, &wait,
711                                 TASK_UNINTERRUPTIBLE);
712                 if (!atomic_read(&transaction->t_updates)) {
713                         spin_unlock(&transaction->t_handle_lock);
714                         finish_wait(&journal->j_wait_updates, &wait);
715                         break;
716                 }
717                 spin_unlock(&transaction->t_handle_lock);
718                 write_unlock(&journal->j_state_lock);
719                 schedule();
720                 finish_wait(&journal->j_wait_updates, &wait);
721                 write_lock(&journal->j_state_lock);
722         }
723         write_unlock(&journal->j_state_lock);
724 
725         /*
726          * We have now established a barrier against other normal updates, but
727          * we also need to barrier against other jbd2_journal_lock_updates() calls
728          * to make sure that we serialise special journal-locked operations
729          * too.
730          */
731         mutex_lock(&journal->j_barrier);
732 }
733 
734 /**
735  * void jbd2_journal_unlock_updates (journal_t* journal) - release barrier
736  * @journal:  Journal to release the barrier on.
737  *
738  * Release a transaction barrier obtained with jbd2_journal_lock_updates().
739  *
740  * Should be called without the journal lock held.
741  */
742 void jbd2_journal_unlock_updates (journal_t *journal)
743 {
744         J_ASSERT(journal->j_barrier_count != 0);
745 
746         mutex_unlock(&journal->j_barrier);
747         write_lock(&journal->j_state_lock);
748         --journal->j_barrier_count;
749         write_unlock(&journal->j_state_lock);
750         wake_up(&journal->j_wait_transaction_locked);
751 }
752 
753 static void warn_dirty_buffer(struct buffer_head *bh)
754 {
755         char b[BDEVNAME_SIZE];
756 
757         printk(KERN_WARNING
758                "JBD2: Spotted dirty metadata buffer (dev = %s, blocknr = %llu). "
759                "There's a risk of filesystem corruption in case of system "
760                "crash.\n",
761                bdevname(bh->b_bdev, b), (unsigned long long)bh->b_blocknr);
762 }
763 
764 static int sleep_on_shadow_bh(void *word)
765 {
766         io_schedule();
767         return 0;
768 }
769 
770 /*
771  * If the buffer is already part of the current transaction, then there
772  * is nothing we need to do.  If it is already part of a prior
773  * transaction which we are still committing to disk, then we need to
774  * make sure that we do not overwrite the old copy: we do copy-out to
775  * preserve the copy going to disk.  We also account the buffer against
776  * the handle's metadata buffer credits (unless the buffer is already
777  * part of the transaction, that is).
778  *
779  */
780 static int
781 do_get_write_access(handle_t *handle, struct journal_head *jh,
782                         int force_copy)
783 {
784         struct buffer_head *bh;
785         transaction_t *transaction = handle->h_transaction;
786         journal_t *journal;
787         int error;
788         char *frozen_buffer = NULL;
789         int need_copy = 0;
790         unsigned long start_lock, time_lock;
791 
792         WARN_ON(!transaction);
793         if (is_handle_aborted(handle))
794                 return -EROFS;
795         journal = transaction->t_journal;
796 
797         jbd_debug(5, "journal_head %p, force_copy %d\n", jh, force_copy);
798 
799         JBUFFER_TRACE(jh, "entry");
800 repeat:
801         bh = jh2bh(jh);
802 
803         /* @@@ Need to check for errors here at some point. */
804 
805         start_lock = jiffies;
806         lock_buffer(bh);
807         jbd_lock_bh_state(bh);
808 
809         /* If it takes too long to lock the buffer, trace it */
810         time_lock = jbd2_time_diff(start_lock, jiffies);
811         if (time_lock > HZ/10)
812                 trace_jbd2_lock_buffer_stall(bh->b_bdev->bd_dev,
813                         jiffies_to_msecs(time_lock));
814 
815         /* We now hold the buffer lock so it is safe to query the buffer
816          * state.  Is the buffer dirty?
817          *
818          * If so, there are two possibilities.  The buffer may be
819          * non-journaled, and undergoing a quite legitimate writeback.
820          * Otherwise, it is journaled, and we don't expect dirty buffers
821          * in that state (the buffers should be marked JBD_Dirty
822          * instead.)  So either the IO is being done under our own
823          * control and this is a bug, or it's a third party IO such as
824          * dump(8) (which may leave the buffer scheduled for read ---
825          * ie. locked but not dirty) or tune2fs (which may actually have
826          * the buffer dirtied, ugh.)  */
827 
828         if (buffer_dirty(bh)) {
829                 /*
830                  * First question: is this buffer already part of the current
831                  * transaction or the existing committing transaction?
832                  */
833                 if (jh->b_transaction) {
834                         J_ASSERT_JH(jh,
835                                 jh->b_transaction == transaction ||
836                                 jh->b_transaction ==
837                                         journal->j_committing_transaction);
838                         if (jh->b_next_transaction)
839                                 J_ASSERT_JH(jh, jh->b_next_transaction ==
840                                                         transaction);
841                         warn_dirty_buffer(bh);
842                 }
843                 /*
844                  * In any case we need to clean the dirty flag and we must
845                  * do it under the buffer lock to be sure we don't race
846                  * with running write-out.
847                  */
848                 JBUFFER_TRACE(jh, "Journalling dirty buffer");
849                 clear_buffer_dirty(bh);
850                 set_buffer_jbddirty(bh);
851         }
852 
853         unlock_buffer(bh);
854 
855         error = -EROFS;
856         if (is_handle_aborted(handle)) {
857                 jbd_unlock_bh_state(bh);
858                 goto out;
859         }
860         error = 0;
861 
862         /*
863          * The buffer is already part of this transaction if b_transaction or
864          * b_next_transaction points to it
865          */
866         if (jh->b_transaction == transaction ||
867             jh->b_next_transaction == transaction)
868                 goto done;
869 
870         /*
871          * this is the first time this transaction is touching this buffer,
872          * reset the modified flag
873          */
874        jh->b_modified = 0;
875 
876         /*
877          * If there is already a copy-out version of this buffer, then we don't
878          * need to make another one
879          */
880         if (jh->b_frozen_data) {
881                 JBUFFER_TRACE(jh, "has frozen data");
882                 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
883                 jh->b_next_transaction = transaction;
884                 goto done;
885         }
886 
887         /* Is there data here we need to preserve? */
888 
889         if (jh->b_transaction && jh->b_transaction != transaction) {
890                 JBUFFER_TRACE(jh, "owned by older transaction");
891                 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
892                 J_ASSERT_JH(jh, jh->b_transaction ==
893                                         journal->j_committing_transaction);
894 
895                 /* There is one case we have to be very careful about.
896                  * If the committing transaction is currently writing
897                  * this buffer out to disk and has NOT made a copy-out,
898                  * then we cannot modify the buffer contents at all
899                  * right now.  The essence of copy-out is that it is the
900                  * extra copy, not the primary copy, which gets
901                  * journaled.  If the primary copy is already going to
902                  * disk then we cannot do copy-out here. */
903 
904                 if (buffer_shadow(bh)) {
905                         JBUFFER_TRACE(jh, "on shadow: sleep");
906                         jbd_unlock_bh_state(bh);
907                         wait_on_bit(&bh->b_state, BH_Shadow,
908                                     sleep_on_shadow_bh, TASK_UNINTERRUPTIBLE);
909                         goto repeat;
910                 }
911 
912                 /*
913                  * Only do the copy if the currently-owning transaction still
914                  * needs it. If buffer isn't on BJ_Metadata list, the
915                  * committing transaction is past that stage (here we use the
916                  * fact that BH_Shadow is set under bh_state lock together with
917                  * refiling to BJ_Shadow list and at this point we know the
918                  * buffer doesn't have BH_Shadow set).
919                  *
920                  * Subtle point, though: if this is a get_undo_access,
921                  * then we will be relying on the frozen_data to contain
922                  * the new value of the committed_data record after the
923                  * transaction, so we HAVE to force the frozen_data copy
924                  * in that case.
925                  */
926                 if (jh->b_jlist == BJ_Metadata || force_copy) {
927                         JBUFFER_TRACE(jh, "generate frozen data");
928                         if (!frozen_buffer) {
929                                 JBUFFER_TRACE(jh, "allocate memory for buffer");
930                                 jbd_unlock_bh_state(bh);
931                                 frozen_buffer =
932                                         jbd2_alloc(jh2bh(jh)->b_size,
933                                                          GFP_NOFS);
934                                 if (!frozen_buffer) {
935                                         printk(KERN_EMERG
936                                                "%s: OOM for frozen_buffer\n",
937                                                __func__);
938                                         JBUFFER_TRACE(jh, "oom!");
939                                         error = -ENOMEM;
940                                         jbd_lock_bh_state(bh);
941                                         goto done;
942                                 }
943                                 goto repeat;
944                         }
945                         jh->b_frozen_data = frozen_buffer;
946                         frozen_buffer = NULL;
947                         need_copy = 1;
948                 }
949                 jh->b_next_transaction = transaction;
950         }
951 
952 
953         /*
954          * Finally, if the buffer is not journaled right now, we need to make
955          * sure it doesn't get written to disk before the caller actually
956          * commits the new data
957          */
958         if (!jh->b_transaction) {
959                 JBUFFER_TRACE(jh, "no transaction");
960                 J_ASSERT_JH(jh, !jh->b_next_transaction);
961                 JBUFFER_TRACE(jh, "file as BJ_Reserved");
962                 spin_lock(&journal->j_list_lock);
963                 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
964                 spin_unlock(&journal->j_list_lock);
965         }
966 
967 done:
968         if (need_copy) {
969                 struct page *page;
970                 int offset;
971                 char *source;
972 
973                 J_EXPECT_JH(jh, buffer_uptodate(jh2bh(jh)),
974                             "Possible IO failure.\n");
975                 page = jh2bh(jh)->b_page;
976                 offset = offset_in_page(jh2bh(jh)->b_data);
977                 source = kmap_atomic(page);
978                 /* Fire data frozen trigger just before we copy the data */
979                 jbd2_buffer_frozen_trigger(jh, source + offset,
980                                            jh->b_triggers);
981                 memcpy(jh->b_frozen_data, source+offset, jh2bh(jh)->b_size);
982                 kunmap_atomic(source);
983 
984                 /*
985                  * Now that the frozen data is saved off, we need to store
986                  * any matching triggers.
987                  */
988                 jh->b_frozen_triggers = jh->b_triggers;
989         }
990         jbd_unlock_bh_state(bh);
991 
992         /*
993          * If we are about to journal a buffer, then any revoke pending on it is
994          * no longer valid
995          */
996         jbd2_journal_cancel_revoke(handle, jh);
997 
998 out:
999         if (unlikely(frozen_buffer))    /* It's usually NULL */
1000                 jbd2_free(frozen_buffer, bh->b_size);
1001 
1002         JBUFFER_TRACE(jh, "exit");
1003         return error;
1004 }
1005 
1006 /**
1007  * int jbd2_journal_get_write_access() - notify intent to modify a buffer for metadata (not data) update.
1008  * @handle: transaction to add buffer modifications to
1009  * @bh:     bh to be used for metadata writes
1010  *
1011  * Returns an error code or 0 on success.
1012  *
1013  * In full data journalling mode the buffer may be of type BJ_AsyncData,
1014  * because we're write()ing a buffer which is also part of a shared mapping.
1015  */
1016 
1017 int jbd2_journal_get_write_access(handle_t *handle, struct buffer_head *bh)
1018 {
1019         struct journal_head *jh = jbd2_journal_add_journal_head(bh);
1020         int rc;
1021 
1022         /* We do not want to get caught playing with fields which the
1023          * log thread also manipulates.  Make sure that the buffer
1024          * completes any outstanding IO before proceeding. */
1025         rc = do_get_write_access(handle, jh, 0);
1026         jbd2_journal_put_journal_head(jh);
1027         return rc;
1028 }
1029 
1030 
1031 /*
1032  * When the user wants to journal a newly created buffer_head
1033  * (ie. getblk() returned a new buffer and we are going to populate it
1034  * manually rather than reading off disk), then we need to keep the
1035  * buffer_head locked until it has been completely filled with new
1036  * data.  In this case, we should be able to make the assertion that
1037  * the bh is not already part of an existing transaction.
1038  *
1039  * The buffer should already be locked by the caller by this point.
1040  * There is no lock ranking violation: it was a newly created,
1041  * unlocked buffer beforehand. */
1042 
1043 /**
1044  * int jbd2_journal_get_create_access () - notify intent to use newly created bh
1045  * @handle: transaction to new buffer to
1046  * @bh: new buffer.
1047  *
1048  * Call this if you create a new bh.
1049  */
1050 int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh)
1051 {
1052         transaction_t *transaction = handle->h_transaction;
1053         journal_t *journal;
1054         struct journal_head *jh = jbd2_journal_add_journal_head(bh);
1055         int err;
1056 
1057         jbd_debug(5, "journal_head %p\n", jh);
1058         WARN_ON(!transaction);
1059         err = -EROFS;
1060         if (is_handle_aborted(handle))
1061                 goto out;
1062         journal = transaction->t_journal;
1063         err = 0;
1064 
1065         JBUFFER_TRACE(jh, "entry");
1066         /*
1067          * The buffer may already belong to this transaction due to pre-zeroing
1068          * in the filesystem's new_block code.  It may also be on the previous,
1069          * committing transaction's lists, but it HAS to be in Forget state in
1070          * that case: the transaction must have deleted the buffer for it to be
1071          * reused here.
1072          */
1073         jbd_lock_bh_state(bh);
1074         spin_lock(&journal->j_list_lock);
1075         J_ASSERT_JH(jh, (jh->b_transaction == transaction ||
1076                 jh->b_transaction == NULL ||
1077                 (jh->b_transaction == journal->j_committing_transaction &&
1078                           jh->b_jlist == BJ_Forget)));
1079 
1080         J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1081         J_ASSERT_JH(jh, buffer_locked(jh2bh(jh)));
1082 
1083         if (jh->b_transaction == NULL) {
1084                 /*
1085                  * Previous jbd2_journal_forget() could have left the buffer
1086                  * with jbddirty bit set because it was being committed. When
1087                  * the commit finished, we've filed the buffer for
1088                  * checkpointing and marked it dirty. Now we are reallocating
1089                  * the buffer so the transaction freeing it must have
1090                  * committed and so it's safe to clear the dirty bit.
1091                  */
1092                 clear_buffer_dirty(jh2bh(jh));
1093                 /* first access by this transaction */
1094                 jh->b_modified = 0;
1095 
1096                 JBUFFER_TRACE(jh, "file as BJ_Reserved");
1097                 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
1098         } else if (jh->b_transaction == journal->j_committing_transaction) {
1099                 /* first access by this transaction */
1100                 jh->b_modified = 0;
1101 
1102                 JBUFFER_TRACE(jh, "set next transaction");
1103                 jh->b_next_transaction = transaction;
1104         }
1105         spin_unlock(&journal->j_list_lock);
1106         jbd_unlock_bh_state(bh);
1107 
1108         /*
1109          * akpm: I added this.  ext3_alloc_branch can pick up new indirect
1110          * blocks which contain freed but then revoked metadata.  We need
1111          * to cancel the revoke in case we end up freeing it yet again
1112          * and the reallocating as data - this would cause a second revoke,
1113          * which hits an assertion error.
1114          */
1115         JBUFFER_TRACE(jh, "cancelling revoke");
1116         jbd2_journal_cancel_revoke(handle, jh);
1117 out:
1118         jbd2_journal_put_journal_head(jh);
1119         return err;
1120 }
1121 
1122 /**
1123  * int jbd2_journal_get_undo_access() -  Notify intent to modify metadata with
1124  *     non-rewindable consequences
1125  * @handle: transaction
1126  * @bh: buffer to undo
1127  *
1128  * Sometimes there is a need to distinguish between metadata which has
1129  * been committed to disk and that which has not.  The ext3fs code uses
1130  * this for freeing and allocating space, we have to make sure that we
1131  * do not reuse freed space until the deallocation has been committed,
1132  * since if we overwrote that space we would make the delete
1133  * un-rewindable in case of a crash.
1134  *
1135  * To deal with that, jbd2_journal_get_undo_access requests write access to a
1136  * buffer for parts of non-rewindable operations such as delete
1137  * operations on the bitmaps.  The journaling code must keep a copy of
1138  * the buffer's contents prior to the undo_access call until such time
1139  * as we know that the buffer has definitely been committed to disk.
1140  *
1141  * We never need to know which transaction the committed data is part
1142  * of, buffers touched here are guaranteed to be dirtied later and so
1143  * will be committed to a new transaction in due course, at which point
1144  * we can discard the old committed data pointer.
1145  *
1146  * Returns error number or 0 on success.
1147  */
1148 int jbd2_journal_get_undo_access(handle_t *handle, struct buffer_head *bh)
1149 {
1150         int err;
1151         struct journal_head *jh = jbd2_journal_add_journal_head(bh);
1152         char *committed_data = NULL;
1153 
1154         JBUFFER_TRACE(jh, "entry");
1155 
1156         /*
1157          * Do this first --- it can drop the journal lock, so we want to
1158          * make sure that obtaining the committed_data is done
1159          * atomically wrt. completion of any outstanding commits.
1160          */
1161         err = do_get_write_access(handle, jh, 1);
1162         if (err)
1163                 goto out;
1164 
1165 repeat:
1166         if (!jh->b_committed_data) {
1167                 committed_data = jbd2_alloc(jh2bh(jh)->b_size, GFP_NOFS);
1168                 if (!committed_data) {
1169                         printk(KERN_EMERG "%s: No memory for committed data\n",
1170                                 __func__);
1171                         err = -ENOMEM;
1172                         goto out;
1173                 }
1174         }
1175 
1176         jbd_lock_bh_state(bh);
1177         if (!jh->b_committed_data) {
1178                 /* Copy out the current buffer contents into the
1179                  * preserved, committed copy. */
1180                 JBUFFER_TRACE(jh, "generate b_committed data");
1181                 if (!committed_data) {
1182                         jbd_unlock_bh_state(bh);
1183                         goto repeat;
1184                 }
1185 
1186                 jh->b_committed_data = committed_data;
1187                 committed_data = NULL;
1188                 memcpy(jh->b_committed_data, bh->b_data, bh->b_size);
1189         }
1190         jbd_unlock_bh_state(bh);
1191 out:
1192         jbd2_journal_put_journal_head(jh);
1193         if (unlikely(committed_data))
1194                 jbd2_free(committed_data, bh->b_size);
1195         return err;
1196 }
1197 
1198 /**
1199  * void jbd2_journal_set_triggers() - Add triggers for commit writeout
1200  * @bh: buffer to trigger on
1201  * @type: struct jbd2_buffer_trigger_type containing the trigger(s).
1202  *
1203  * Set any triggers on this journal_head.  This is always safe, because
1204  * triggers for a committing buffer will be saved off, and triggers for
1205  * a running transaction will match the buffer in that transaction.
1206  *
1207  * Call with NULL to clear the triggers.
1208  */
1209 void jbd2_journal_set_triggers(struct buffer_head *bh,
1210                                struct jbd2_buffer_trigger_type *type)
1211 {
1212         struct journal_head *jh = jbd2_journal_grab_journal_head(bh);
1213 
1214         if (WARN_ON(!jh))
1215                 return;
1216         jh->b_triggers = type;
1217         jbd2_journal_put_journal_head(jh);
1218 }
1219 
1220 void jbd2_buffer_frozen_trigger(struct journal_head *jh, void *mapped_data,
1221                                 struct jbd2_buffer_trigger_type *triggers)
1222 {
1223         struct buffer_head *bh = jh2bh(jh);
1224 
1225         if (!triggers || !triggers->t_frozen)
1226                 return;
1227 
1228         triggers->t_frozen(triggers, bh, mapped_data, bh->b_size);
1229 }
1230 
1231 void jbd2_buffer_abort_trigger(struct journal_head *jh,
1232                                struct jbd2_buffer_trigger_type *triggers)
1233 {
1234         if (!triggers || !triggers->t_abort)
1235                 return;
1236 
1237         triggers->t_abort(triggers, jh2bh(jh));
1238 }
1239 
1240 
1241 
1242 /**
1243  * int jbd2_journal_dirty_metadata() -  mark a buffer as containing dirty metadata
1244  * @handle: transaction to add buffer to.
1245  * @bh: buffer to mark
1246  *
1247  * mark dirty metadata which needs to be journaled as part of the current
1248  * transaction.
1249  *
1250  * The buffer must have previously had jbd2_journal_get_write_access()
1251  * called so that it has a valid journal_head attached to the buffer
1252  * head.
1253  *
1254  * The buffer is placed on the transaction's metadata list and is marked
1255  * as belonging to the transaction.
1256  *
1257  * Returns error number or 0 on success.
1258  *
1259  * Special care needs to be taken if the buffer already belongs to the
1260  * current committing transaction (in which case we should have frozen
1261  * data present for that commit).  In that case, we don't relink the
1262  * buffer: that only gets done when the old transaction finally
1263  * completes its commit.
1264  */
1265 int jbd2_journal_dirty_metadata(handle_t *handle, struct buffer_head *bh)
1266 {
1267         transaction_t *transaction = handle->h_transaction;
1268         journal_t *journal;
1269         struct journal_head *jh;
1270         int ret = 0;
1271 
1272         WARN_ON(!transaction);
1273         if (is_handle_aborted(handle))
1274                 return -EROFS;
1275         journal = transaction->t_journal;
1276         jh = jbd2_journal_grab_journal_head(bh);
1277         if (!jh) {
1278                 ret = -EUCLEAN;
1279                 goto out;
1280         }
1281         jbd_debug(5, "journal_head %p\n", jh);
1282         JBUFFER_TRACE(jh, "entry");
1283 
1284         jbd_lock_bh_state(bh);
1285 
1286         if (jh->b_modified == 0) {
1287                 /*
1288                  * This buffer's got modified and becoming part
1289                  * of the transaction. This needs to be done
1290                  * once a transaction -bzzz
1291                  */
1292                 jh->b_modified = 1;
1293                 J_ASSERT_JH(jh, handle->h_buffer_credits > 0);
1294                 handle->h_buffer_credits--;
1295         }
1296 
1297         /*
1298          * fastpath, to avoid expensive locking.  If this buffer is already
1299          * on the running transaction's metadata list there is nothing to do.
1300          * Nobody can take it off again because there is a handle open.
1301          * I _think_ we're OK here with SMP barriers - a mistaken decision will
1302          * result in this test being false, so we go in and take the locks.
1303          */
1304         if (jh->b_transaction == transaction && jh->b_jlist == BJ_Metadata) {
1305                 JBUFFER_TRACE(jh, "fastpath");
1306                 if (unlikely(jh->b_transaction !=
1307                              journal->j_running_transaction)) {
1308                         printk(KERN_EMERG "JBD: %s: "
1309                                "jh->b_transaction (%llu, %p, %u) != "
1310                                "journal->j_running_transaction (%p, %u)",
1311                                journal->j_devname,
1312                                (unsigned long long) bh->b_blocknr,
1313                                jh->b_transaction,
1314                                jh->b_transaction ? jh->b_transaction->t_tid : 0,
1315                                journal->j_running_transaction,
1316                                journal->j_running_transaction ?
1317                                journal->j_running_transaction->t_tid : 0);
1318                         ret = -EINVAL;
1319                 }
1320                 goto out_unlock_bh;
1321         }
1322 
1323         set_buffer_jbddirty(bh);
1324 
1325         /*
1326          * Metadata already on the current transaction list doesn't
1327          * need to be filed.  Metadata on another transaction's list must
1328          * be committing, and will be refiled once the commit completes:
1329          * leave it alone for now.
1330          */
1331         if (jh->b_transaction != transaction) {
1332                 JBUFFER_TRACE(jh, "already on other transaction");
1333                 if (unlikely(jh->b_transaction !=
1334                              journal->j_committing_transaction)) {
1335                         printk(KERN_EMERG "JBD: %s: "
1336                                "jh->b_transaction (%llu, %p, %u) != "
1337                                "journal->j_committing_transaction (%p, %u)",
1338                                journal->j_devname,
1339                                (unsigned long long) bh->b_blocknr,
1340                                jh->b_transaction,
1341                                jh->b_transaction ? jh->b_transaction->t_tid : 0,
1342                                journal->j_committing_transaction,
1343                                journal->j_committing_transaction ?
1344                                journal->j_committing_transaction->t_tid : 0);
1345                         ret = -EINVAL;
1346                 }
1347                 if (unlikely(jh->b_next_transaction != transaction)) {
1348                         printk(KERN_EMERG "JBD: %s: "
1349                                "jh->b_next_transaction (%llu, %p, %u) != "
1350                                "transaction (%p, %u)",
1351                                journal->j_devname,
1352                                (unsigned long long) bh->b_blocknr,
1353                                jh->b_next_transaction,
1354                                jh->b_next_transaction ?
1355                                jh->b_next_transaction->t_tid : 0,
1356                                transaction, transaction->t_tid);
1357                         ret = -EINVAL;
1358                 }
1359                 /* And this case is illegal: we can't reuse another
1360                  * transaction's data buffer, ever. */
1361                 goto out_unlock_bh;
1362         }
1363 
1364         /* That test should have eliminated the following case: */
1365         J_ASSERT_JH(jh, jh->b_frozen_data == NULL);
1366 
1367         JBUFFER_TRACE(jh, "file as BJ_Metadata");
1368         spin_lock(&journal->j_list_lock);
1369         __jbd2_journal_file_buffer(jh, transaction, BJ_Metadata);
1370         spin_unlock(&journal->j_list_lock);
1371 out_unlock_bh:
1372         jbd_unlock_bh_state(bh);
1373         jbd2_journal_put_journal_head(jh);
1374 out:
1375         JBUFFER_TRACE(jh, "exit");
1376         WARN_ON(ret);   /* All errors are bugs, so dump the stack */
1377         return ret;
1378 }
1379 
1380 /**
1381  * void jbd2_journal_forget() - bforget() for potentially-journaled buffers.
1382  * @handle: transaction handle
1383  * @bh:     bh to 'forget'
1384  *
1385  * We can only do the bforget if there are no commits pending against the
1386  * buffer.  If the buffer is dirty in the current running transaction we
1387  * can safely unlink it.
1388  *
1389  * bh may not be a journalled buffer at all - it may be a non-JBD
1390  * buffer which came off the hashtable.  Check for this.
1391  *
1392  * Decrements bh->b_count by one.
1393  *
1394  * Allow this call even if the handle has aborted --- it may be part of
1395  * the caller's cleanup after an abort.
1396  */
1397 int jbd2_journal_forget (handle_t *handle, struct buffer_head *bh)
1398 {
1399         transaction_t *transaction = handle->h_transaction;
1400         journal_t *journal;
1401         struct journal_head *jh;
1402         int drop_reserve = 0;
1403         int err = 0;
1404         int was_modified = 0;
1405 
1406         WARN_ON(!transaction);
1407         if (is_handle_aborted(handle))
1408                 return -EROFS;
1409         journal = transaction->t_journal;
1410 
1411         BUFFER_TRACE(bh, "entry");
1412 
1413         jbd_lock_bh_state(bh);
1414         spin_lock(&journal->j_list_lock);
1415 
1416         if (!buffer_jbd(bh))
1417                 goto not_jbd;
1418         jh = bh2jh(bh);
1419 
1420         /* Critical error: attempting to delete a bitmap buffer, maybe?
1421          * Don't do any jbd operations, and return an error. */
1422         if (!J_EXPECT_JH(jh, !jh->b_committed_data,
1423                          "inconsistent data on disk")) {
1424                 err = -EIO;
1425                 goto not_jbd;
1426         }
1427 
1428         /* keep track of whether or not this transaction modified us */
1429         was_modified = jh->b_modified;
1430 
1431         /*
1432          * The buffer's going from the transaction, we must drop
1433          * all references -bzzz
1434          */
1435         jh->b_modified = 0;
1436 
1437         if (jh->b_transaction == transaction) {
1438                 J_ASSERT_JH(jh, !jh->b_frozen_data);
1439 
1440                 /* If we are forgetting a buffer which is already part
1441                  * of this transaction, then we can just drop it from
1442                  * the transaction immediately. */
1443                 clear_buffer_dirty(bh);
1444                 clear_buffer_jbddirty(bh);
1445 
1446                 JBUFFER_TRACE(jh, "belongs to current transaction: unfile");
1447 
1448                 /*
1449                  * we only want to drop a reference if this transaction
1450                  * modified the buffer
1451                  */
1452                 if (was_modified)
1453                         drop_reserve = 1;
1454 
1455                 /*
1456                  * We are no longer going to journal this buffer.
1457                  * However, the commit of this transaction is still
1458                  * important to the buffer: the delete that we are now
1459                  * processing might obsolete an old log entry, so by
1460                  * committing, we can satisfy the buffer's checkpoint.
1461                  *
1462                  * So, if we have a checkpoint on the buffer, we should
1463                  * now refile the buffer on our BJ_Forget list so that
1464                  * we know to remove the checkpoint after we commit.
1465                  */
1466 
1467                 if (jh->b_cp_transaction) {
1468                         __jbd2_journal_temp_unlink_buffer(jh);
1469                         __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
1470                 } else {
1471                         __jbd2_journal_unfile_buffer(jh);
1472                         if (!buffer_jbd(bh)) {
1473                                 spin_unlock(&journal->j_list_lock);
1474                                 jbd_unlock_bh_state(bh);
1475                                 __bforget(bh);
1476                                 goto drop;
1477                         }
1478                 }
1479         } else if (jh->b_transaction) {
1480                 J_ASSERT_JH(jh, (jh->b_transaction ==
1481                                  journal->j_committing_transaction));
1482                 /* However, if the buffer is still owned by a prior
1483                  * (committing) transaction, we can't drop it yet... */
1484                 JBUFFER_TRACE(jh, "belongs to older transaction");
1485                 /* ... but we CAN drop it from the new transaction if we
1486                  * have also modified it since the original commit. */
1487 
1488                 if (jh->b_next_transaction) {
1489                         J_ASSERT(jh->b_next_transaction == transaction);
1490                         jh->b_next_transaction = NULL;
1491 
1492                         /*
1493                          * only drop a reference if this transaction modified
1494                          * the buffer
1495                          */
1496                         if (was_modified)
1497                                 drop_reserve = 1;
1498                 }
1499         }
1500 
1501 not_jbd:
1502         spin_unlock(&journal->j_list_lock);
1503         jbd_unlock_bh_state(bh);
1504         __brelse(bh);
1505 drop:
1506         if (drop_reserve) {
1507                 /* no need to reserve log space for this block -bzzz */
1508                 handle->h_buffer_credits++;
1509         }
1510         return err;
1511 }
1512 
1513 /**
1514  * int jbd2_journal_stop() - complete a transaction
1515  * @handle: tranaction to complete.
1516  *
1517  * All done for a particular handle.
1518  *
1519  * There is not much action needed here.  We just return any remaining
1520  * buffer credits to the transaction and remove the handle.  The only
1521  * complication is that we need to start a commit operation if the
1522  * filesystem is marked for synchronous update.
1523  *
1524  * jbd2_journal_stop itself will not usually return an error, but it may
1525  * do so in unusual circumstances.  In particular, expect it to
1526  * return -EIO if a jbd2_journal_abort has been executed since the
1527  * transaction began.
1528  */
1529 int jbd2_journal_stop(handle_t *handle)
1530 {
1531         transaction_t *transaction = handle->h_transaction;
1532         journal_t *journal;
1533         int err = 0, wait_for_commit = 0;
1534         tid_t tid;
1535         pid_t pid;
1536 
1537         if (!transaction)
1538                 goto free_and_exit;
1539         journal = transaction->t_journal;
1540 
1541         J_ASSERT(journal_current_handle() == handle);
1542 
1543         if (is_handle_aborted(handle))
1544                 err = -EIO;
1545         else
1546                 J_ASSERT(atomic_read(&transaction->t_updates) > 0);
1547 
1548         if (--handle->h_ref > 0) {
1549                 jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1550                           handle->h_ref);
1551                 return err;
1552         }
1553 
1554         jbd_debug(4, "Handle %p going down\n", handle);
1555         trace_jbd2_handle_stats(journal->j_fs_dev->bd_dev,
1556                                 transaction->t_tid,
1557                                 handle->h_type, handle->h_line_no,
1558                                 jiffies - handle->h_start_jiffies,
1559                                 handle->h_sync, handle->h_requested_credits,
1560                                 (handle->h_requested_credits -
1561                                  handle->h_buffer_credits));
1562 
1563         /*
1564          * Implement synchronous transaction batching.  If the handle
1565          * was synchronous, don't force a commit immediately.  Let's
1566          * yield and let another thread piggyback onto this
1567          * transaction.  Keep doing that while new threads continue to
1568          * arrive.  It doesn't cost much - we're about to run a commit
1569          * and sleep on IO anyway.  Speeds up many-threaded, many-dir
1570          * operations by 30x or more...
1571          *
1572          * We try and optimize the sleep time against what the
1573          * underlying disk can do, instead of having a static sleep
1574          * time.  This is useful for the case where our storage is so
1575          * fast that it is more optimal to go ahead and force a flush
1576          * and wait for the transaction to be committed than it is to
1577          * wait for an arbitrary amount of time for new writers to
1578          * join the transaction.  We achieve this by measuring how
1579          * long it takes to commit a transaction, and compare it with
1580          * how long this transaction has been running, and if run time
1581          * < commit time then we sleep for the delta and commit.  This
1582          * greatly helps super fast disks that would see slowdowns as
1583          * more threads started doing fsyncs.
1584          *
1585          * But don't do this if this process was the most recent one
1586          * to perform a synchronous write.  We do this to detect the
1587          * case where a single process is doing a stream of sync
1588          * writes.  No point in waiting for joiners in that case.
1589          */
1590         pid = current->pid;
1591         if (handle->h_sync && journal->j_last_sync_writer != pid) {
1592                 u64 commit_time, trans_time;
1593 
1594                 journal->j_last_sync_writer = pid;
1595 
1596                 read_lock(&journal->j_state_lock);
1597                 commit_time = journal->j_average_commit_time;
1598                 read_unlock(&journal->j_state_lock);
1599 
1600                 trans_time = ktime_to_ns(ktime_sub(ktime_get(),
1601                                                    transaction->t_start_time));
1602 
1603                 commit_time = max_t(u64, commit_time,
1604                                     1000*journal->j_min_batch_time);
1605                 commit_time = min_t(u64, commit_time,
1606                                     1000*journal->j_max_batch_time);
1607 
1608                 if (trans_time < commit_time) {
1609                         ktime_t expires = ktime_add_ns(ktime_get(),
1610                                                        commit_time);
1611                         set_current_state(TASK_UNINTERRUPTIBLE);
1612                         schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1613                 }
1614         }
1615 
1616         if (handle->h_sync)
1617                 transaction->t_synchronous_commit = 1;
1618         current->journal_info = NULL;
1619         atomic_sub(handle->h_buffer_credits,
1620                    &transaction->t_outstanding_credits);
1621 
1622         /*
1623          * If the handle is marked SYNC, we need to set another commit
1624          * going!  We also want to force a commit if the current
1625          * transaction is occupying too much of the log, or if the
1626          * transaction is too old now.
1627          */
1628         if (handle->h_sync ||
1629             (atomic_read(&transaction->t_outstanding_credits) >
1630              journal->j_max_transaction_buffers) ||
1631             time_after_eq(jiffies, transaction->t_expires)) {
1632                 /* Do this even for aborted journals: an abort still
1633                  * completes the commit thread, it just doesn't write
1634                  * anything to disk. */
1635 
1636                 jbd_debug(2, "transaction too old, requesting commit for "
1637                                         "handle %p\n", handle);
1638                 /* This is non-blocking */
1639                 jbd2_log_start_commit(journal, transaction->t_tid);
1640 
1641                 /*
1642                  * Special case: JBD2_SYNC synchronous updates require us
1643                  * to wait for the commit to complete.
1644                  */
1645                 if (handle->h_sync && !(current->flags & PF_MEMALLOC))
1646                         wait_for_commit = 1;
1647         }
1648 
1649         /*
1650          * Once we drop t_updates, if it goes to zero the transaction
1651          * could start committing on us and eventually disappear.  So
1652          * once we do this, we must not dereference transaction
1653          * pointer again.
1654          */
1655         tid = transaction->t_tid;
1656         if (atomic_dec_and_test(&transaction->t_updates)) {
1657                 wake_up(&journal->j_wait_updates);
1658                 if (journal->j_barrier_count)
1659                         wake_up(&journal->j_wait_transaction_locked);
1660         }
1661 
1662         if (wait_for_commit)
1663                 err = jbd2_log_wait_commit(journal, tid);
1664 
1665         lock_map_release(&handle->h_lockdep_map);
1666 
1667         if (handle->h_rsv_handle)
1668                 jbd2_journal_free_reserved(handle->h_rsv_handle);
1669 free_and_exit:
1670         jbd2_free_handle(handle);
1671         return err;
1672 }
1673 
1674 /*
1675  *
1676  * List management code snippets: various functions for manipulating the
1677  * transaction buffer lists.
1678  *
1679  */
1680 
1681 /*
1682  * Append a buffer to a transaction list, given the transaction's list head
1683  * pointer.
1684  *
1685  * j_list_lock is held.
1686  *
1687  * jbd_lock_bh_state(jh2bh(jh)) is held.
1688  */
1689 
1690 static inline void
1691 __blist_add_buffer(struct journal_head **list, struct journal_head *jh)
1692 {
1693         if (!*list) {
1694                 jh->b_tnext = jh->b_tprev = jh;
1695                 *list = jh;
1696         } else {
1697                 /* Insert at the tail of the list to preserve order */
1698                 struct journal_head *first = *list, *last = first->b_tprev;
1699                 jh->b_tprev = last;
1700                 jh->b_tnext = first;
1701                 last->b_tnext = first->b_tprev = jh;
1702         }
1703 }
1704 
1705 /*
1706  * Remove a buffer from a transaction list, given the transaction's list
1707  * head pointer.
1708  *
1709  * Called with j_list_lock held, and the journal may not be locked.
1710  *
1711  * jbd_lock_bh_state(jh2bh(jh)) is held.
1712  */
1713 
1714 static inline void
1715 __blist_del_buffer(struct journal_head **list, struct journal_head *jh)
1716 {
1717         if (*list == jh) {
1718                 *list = jh->b_tnext;
1719                 if (*list == jh)
1720                         *list = NULL;
1721         }
1722         jh->b_tprev->b_tnext = jh->b_tnext;
1723         jh->b_tnext->b_tprev = jh->b_tprev;
1724 }
1725 
1726 /*
1727  * Remove a buffer from the appropriate transaction list.
1728  *
1729  * Note that this function can *change* the value of
1730  * bh->b_transaction->t_buffers, t_forget, t_shadow_list, t_log_list or
1731  * t_reserved_list.  If the caller is holding onto a copy of one of these
1732  * pointers, it could go bad.  Generally the caller needs to re-read the
1733  * pointer from the transaction_t.
1734  *
1735  * Called under j_list_lock.
1736  */
1737 static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh)
1738 {
1739         struct journal_head **list = NULL;
1740         transaction_t *transaction;
1741         struct buffer_head *bh = jh2bh(jh);
1742 
1743         J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
1744         transaction = jh->b_transaction;
1745         if (transaction)
1746                 assert_spin_locked(&transaction->t_journal->j_list_lock);
1747 
1748         J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
1749         if (jh->b_jlist != BJ_None)
1750                 J_ASSERT_JH(jh, transaction != NULL);
1751 
1752         switch (jh->b_jlist) {
1753         case BJ_None:
1754                 return;
1755         case BJ_Metadata:
1756                 transaction->t_nr_buffers--;
1757                 J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0);
1758                 list = &transaction->t_buffers;
1759                 break;
1760         case BJ_Forget:
1761                 list = &transaction->t_forget;
1762                 break;
1763         case BJ_Shadow:
1764                 list = &transaction->t_shadow_list;
1765                 break;
1766         case BJ_Reserved:
1767                 list = &transaction->t_reserved_list;
1768                 break;
1769         }
1770 
1771         __blist_del_buffer(list, jh);
1772         jh->b_jlist = BJ_None;
1773         if (test_clear_buffer_jbddirty(bh))
1774                 mark_buffer_dirty(bh);  /* Expose it to the VM */
1775 }
1776 
1777 /*
1778  * Remove buffer from all transactions.
1779  *
1780  * Called with bh_state lock and j_list_lock
1781  *
1782  * jh and bh may be already freed when this function returns.
1783  */
1784 static void __jbd2_journal_unfile_buffer(struct journal_head *jh)
1785 {
1786         __jbd2_journal_temp_unlink_buffer(jh);
1787         jh->b_transaction = NULL;
1788         jbd2_journal_put_journal_head(jh);
1789 }
1790 
1791 void jbd2_journal_unfile_buffer(journal_t *journal, struct journal_head *jh)
1792 {
1793         struct buffer_head *bh = jh2bh(jh);
1794 
1795         /* Get reference so that buffer cannot be freed before we unlock it */
1796         get_bh(bh);
1797         jbd_lock_bh_state(bh);
1798         spin_lock(&journal->j_list_lock);
1799         __jbd2_journal_unfile_buffer(jh);
1800         spin_unlock(&journal->j_list_lock);
1801         jbd_unlock_bh_state(bh);
1802         __brelse(bh);
1803 }
1804 
1805 /*
1806  * Called from jbd2_journal_try_to_free_buffers().
1807  *
1808  * Called under jbd_lock_bh_state(bh)
1809  */
1810 static void
1811 __journal_try_to_free_buffer(journal_t *journal, struct buffer_head *bh)
1812 {
1813         struct journal_head *jh;
1814 
1815         jh = bh2jh(bh);
1816 
1817         if (buffer_locked(bh) || buffer_dirty(bh))
1818                 goto out;
1819 
1820         if (jh->b_next_transaction != NULL)
1821                 goto out;
1822 
1823         spin_lock(&journal->j_list_lock);
1824         if (jh->b_cp_transaction != NULL && jh->b_transaction == NULL) {
1825                 /* written-back checkpointed metadata buffer */
1826                 JBUFFER_TRACE(jh, "remove from checkpoint list");
1827                 __jbd2_journal_remove_checkpoint(jh);
1828         }
1829         spin_unlock(&journal->j_list_lock);
1830 out:
1831         return;
1832 }
1833 
1834 /**
1835  * int jbd2_journal_try_to_free_buffers() - try to free page buffers.
1836  * @journal: journal for operation
1837  * @page: to try and free
1838  * @gfp_mask: we use the mask to detect how hard should we try to release
1839  * buffers. If __GFP_WAIT and __GFP_FS is set, we wait for commit code to
1840  * release the buffers.
1841  *
1842  *
1843  * For all the buffers on this page,
1844  * if they are fully written out ordered data, move them onto BUF_CLEAN
1845  * so try_to_free_buffers() can reap them.
1846  *
1847  * This function returns non-zero if we wish try_to_free_buffers()
1848  * to be called. We do this if the page is releasable by try_to_free_buffers().
1849  * We also do it if the page has locked or dirty buffers and the caller wants
1850  * us to perform sync or async writeout.
1851  *
1852  * This complicates JBD locking somewhat.  We aren't protected by the
1853  * BKL here.  We wish to remove the buffer from its committing or
1854  * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer.
1855  *
1856  * This may *change* the value of transaction_t->t_datalist, so anyone
1857  * who looks at t_datalist needs to lock against this function.
1858  *
1859  * Even worse, someone may be doing a jbd2_journal_dirty_data on this
1860  * buffer.  So we need to lock against that.  jbd2_journal_dirty_data()
1861  * will come out of the lock with the buffer dirty, which makes it
1862  * ineligible for release here.
1863  *
1864  * Who else is affected by this?  hmm...  Really the only contender
1865  * is do_get_write_access() - it could be looking at the buffer while
1866  * journal_try_to_free_buffer() is changing its state.  But that
1867  * cannot happen because we never reallocate freed data as metadata
1868  * while the data is part of a transaction.  Yes?
1869  *
1870  * Return 0 on failure, 1 on success
1871  */
1872 int jbd2_journal_try_to_free_buffers(journal_t *journal,
1873                                 struct page *page, gfp_t gfp_mask)
1874 {
1875         struct buffer_head *head;
1876         struct buffer_head *bh;
1877         int ret = 0;
1878 
1879         J_ASSERT(PageLocked(page));
1880 
1881         head = page_buffers(page);
1882         bh = head;
1883         do {
1884                 struct journal_head *jh;
1885 
1886                 /*
1887                  * We take our own ref against the journal_head here to avoid
1888                  * having to add tons of locking around each instance of
1889                  * jbd2_journal_put_journal_head().
1890                  */
1891                 jh = jbd2_journal_grab_journal_head(bh);
1892                 if (!jh)
1893                         continue;
1894 
1895                 jbd_lock_bh_state(bh);
1896                 __journal_try_to_free_buffer(journal, bh);
1897                 jbd2_journal_put_journal_head(jh);
1898                 jbd_unlock_bh_state(bh);
1899                 if (buffer_jbd(bh))
1900                         goto busy;
1901         } while ((bh = bh->b_this_page) != head);
1902 
1903         ret = try_to_free_buffers(page);
1904 
1905 busy:
1906         return ret;
1907 }
1908 
1909 /*
1910  * This buffer is no longer needed.  If it is on an older transaction's
1911  * checkpoint list we need to record it on this transaction's forget list
1912  * to pin this buffer (and hence its checkpointing transaction) down until
1913  * this transaction commits.  If the buffer isn't on a checkpoint list, we
1914  * release it.
1915  * Returns non-zero if JBD no longer has an interest in the buffer.
1916  *
1917  * Called under j_list_lock.
1918  *
1919  * Called under jbd_lock_bh_state(bh).
1920  */
1921 static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction)
1922 {
1923         int may_free = 1;
1924         struct buffer_head *bh = jh2bh(jh);
1925 
1926         if (jh->b_cp_transaction) {
1927                 JBUFFER_TRACE(jh, "on running+cp transaction");
1928                 __jbd2_journal_temp_unlink_buffer(jh);
1929                 /*
1930                  * We don't want to write the buffer anymore, clear the
1931                  * bit so that we don't confuse checks in
1932                  * __journal_file_buffer
1933                  */
1934                 clear_buffer_dirty(bh);
1935                 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
1936                 may_free = 0;
1937         } else {
1938                 JBUFFER_TRACE(jh, "on running transaction");
1939                 __jbd2_journal_unfile_buffer(jh);
1940         }
1941         return may_free;
1942 }
1943 
1944 /*
1945  * jbd2_journal_invalidatepage
1946  *
1947  * This code is tricky.  It has a number of cases to deal with.
1948  *
1949  * There are two invariants which this code relies on:
1950  *
1951  * i_size must be updated on disk before we start calling invalidatepage on the
1952  * data.
1953  *
1954  *  This is done in ext3 by defining an ext3_setattr method which
1955  *  updates i_size before truncate gets going.  By maintaining this
1956  *  invariant, we can be sure that it is safe to throw away any buffers
1957  *  attached to the current transaction: once the transaction commits,
1958  *  we know that the data will not be needed.
1959  *
1960  *  Note however that we can *not* throw away data belonging to the
1961  *  previous, committing transaction!
1962  *
1963  * Any disk blocks which *are* part of the previous, committing
1964  * transaction (and which therefore cannot be discarded immediately) are
1965  * not going to be reused in the new running transaction
1966  *
1967  *  The bitmap committed_data images guarantee this: any block which is
1968  *  allocated in one transaction and removed in the next will be marked
1969  *  as in-use in the committed_data bitmap, so cannot be reused until
1970  *  the next transaction to delete the block commits.  This means that
1971  *  leaving committing buffers dirty is quite safe: the disk blocks
1972  *  cannot be reallocated to a different file and so buffer aliasing is
1973  *  not possible.
1974  *
1975  *
1976  * The above applies mainly to ordered data mode.  In writeback mode we
1977  * don't make guarantees about the order in which data hits disk --- in
1978  * particular we don't guarantee that new dirty data is flushed before
1979  * transaction commit --- so it is always safe just to discard data
1980  * immediately in that mode.  --sct
1981  */
1982 
1983 /*
1984  * The journal_unmap_buffer helper function returns zero if the buffer
1985  * concerned remains pinned as an anonymous buffer belonging to an older
1986  * transaction.
1987  *
1988  * We're outside-transaction here.  Either or both of j_running_transaction
1989  * and j_committing_transaction may be NULL.
1990  */
1991 static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh,
1992                                 int partial_page)
1993 {
1994         transaction_t *transaction;
1995         struct journal_head *jh;
1996         int may_free = 1;
1997 
1998         BUFFER_TRACE(bh, "entry");
1999 
2000         /*
2001          * It is safe to proceed here without the j_list_lock because the
2002          * buffers cannot be stolen by try_to_free_buffers as long as we are
2003          * holding the page lock. --sct
2004          */
2005 
2006         if (!buffer_jbd(bh))
2007                 goto zap_buffer_unlocked;
2008 
2009         /* OK, we have data buffer in journaled mode */
2010         write_lock(&journal->j_state_lock);
2011         jbd_lock_bh_state(bh);
2012         spin_lock(&journal->j_list_lock);
2013 
2014         jh = jbd2_journal_grab_journal_head(bh);
2015         if (!jh)
2016                 goto zap_buffer_no_jh;
2017 
2018         /*
2019          * We cannot remove the buffer from checkpoint lists until the
2020          * transaction adding inode to orphan list (let's call it T)
2021          * is committed.  Otherwise if the transaction changing the
2022          * buffer would be cleaned from the journal before T is
2023          * committed, a crash will cause that the correct contents of
2024          * the buffer will be lost.  On the other hand we have to
2025          * clear the buffer dirty bit at latest at the moment when the
2026          * transaction marking the buffer as freed in the filesystem
2027          * structures is committed because from that moment on the
2028          * block can be reallocated and used by a different page.
2029          * Since the block hasn't been freed yet but the inode has
2030          * already been added to orphan list, it is safe for us to add
2031          * the buffer to BJ_Forget list of the newest transaction.
2032          *
2033          * Also we have to clear buffer_mapped flag of a truncated buffer
2034          * because the buffer_head may be attached to the page straddling
2035          * i_size (can happen only when blocksize < pagesize) and thus the
2036          * buffer_head can be reused when the file is extended again. So we end
2037          * up keeping around invalidated buffers attached to transactions'
2038          * BJ_Forget list just to stop checkpointing code from cleaning up
2039          * the transaction this buffer was modified in.
2040          */
2041         transaction = jh->b_transaction;
2042         if (transaction == NULL) {
2043                 /* First case: not on any transaction.  If it
2044                  * has no checkpoint link, then we can zap it:
2045                  * it's a writeback-mode buffer so we don't care
2046                  * if it hits disk safely. */
2047                 if (!jh->b_cp_transaction) {
2048                         JBUFFER_TRACE(jh, "not on any transaction: zap");
2049                         goto zap_buffer;
2050                 }
2051 
2052                 if (!buffer_dirty(bh)) {
2053                         /* bdflush has written it.  We can drop it now */
2054                         goto zap_buffer;
2055                 }
2056 
2057                 /* OK, it must be in the journal but still not
2058                  * written fully to disk: it's metadata or
2059                  * journaled data... */
2060 
2061                 if (journal->j_running_transaction) {
2062                         /* ... and once the current transaction has
2063                          * committed, the buffer won't be needed any
2064                          * longer. */
2065                         JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget");
2066                         may_free = __dispose_buffer(jh,
2067                                         journal->j_running_transaction);
2068                         goto zap_buffer;
2069                 } else {
2070                         /* There is no currently-running transaction. So the
2071                          * orphan record which we wrote for this file must have
2072                          * passed into commit.  We must attach this buffer to
2073                          * the committing transaction, if it exists. */
2074                         if (journal->j_committing_transaction) {
2075                                 JBUFFER_TRACE(jh, "give to committing trans");
2076                                 may_free = __dispose_buffer(jh,
2077                                         journal->j_committing_transaction);
2078                                 goto zap_buffer;
2079                         } else {
2080                                 /* The orphan record's transaction has
2081                                  * committed.  We can cleanse this buffer */
2082                                 clear_buffer_jbddirty(bh);
2083                                 goto zap_buffer;
2084                         }
2085                 }
2086         } else if (transaction == journal->j_committing_transaction) {
2087                 JBUFFER_TRACE(jh, "on committing transaction");
2088                 /*
2089                  * The buffer is committing, we simply cannot touch
2090                  * it. If the page is straddling i_size we have to wait
2091                  * for commit and try again.
2092                  */
2093                 if (partial_page) {
2094                         jbd2_journal_put_journal_head(jh);
2095                         spin_unlock(&journal->j_list_lock);
2096                         jbd_unlock_bh_state(bh);
2097                         write_unlock(&journal->j_state_lock);
2098                         return -EBUSY;
2099                 }
2100                 /*
2101                  * OK, buffer won't be reachable after truncate. We just set
2102                  * j_next_transaction to the running transaction (if there is
2103                  * one) and mark buffer as freed so that commit code knows it
2104                  * should clear dirty bits when it is done with the buffer.
2105                  */
2106                 set_buffer_freed(bh);
2107                 if (journal->j_running_transaction && buffer_jbddirty(bh))
2108                         jh->b_next_transaction = journal->j_running_transaction;
2109                 jbd2_journal_put_journal_head(jh);
2110                 spin_unlock(&journal->j_list_lock);
2111                 jbd_unlock_bh_state(bh);
2112                 write_unlock(&journal->j_state_lock);
2113                 return 0;
2114         } else {
2115                 /* Good, the buffer belongs to the running transaction.
2116                  * We are writing our own transaction's data, not any
2117                  * previous one's, so it is safe to throw it away
2118                  * (remember that we expect the filesystem to have set
2119                  * i_size already for this truncate so recovery will not
2120                  * expose the disk blocks we are discarding here.) */
2121                 J_ASSERT_JH(jh, transaction == journal->j_running_transaction);
2122                 JBUFFER_TRACE(jh, "on running transaction");
2123                 may_free = __dispose_buffer(jh, transaction);
2124         }
2125 
2126 zap_buffer:
2127         /*
2128          * This is tricky. Although the buffer is truncated, it may be reused
2129          * if blocksize < pagesize and it is attached to the page straddling
2130          * EOF. Since the buffer might have been added to BJ_Forget list of the
2131          * running transaction, journal_get_write_access() won't clear
2132          * b_modified and credit accounting gets confused. So clear b_modified
2133          * here.
2134          */
2135         jh->b_modified = 0;
2136         jbd2_journal_put_journal_head(jh);
2137 zap_buffer_no_jh:
2138         spin_unlock(&journal->j_list_lock);
2139         jbd_unlock_bh_state(bh);
2140         write_unlock(&journal->j_state_lock);
2141 zap_buffer_unlocked:
2142         clear_buffer_dirty(bh);
2143         J_ASSERT_BH(bh, !buffer_jbddirty(bh));
2144         clear_buffer_mapped(bh);
2145         clear_buffer_req(bh);
2146         clear_buffer_new(bh);
2147         clear_buffer_delay(bh);
2148         clear_buffer_unwritten(bh);
2149         bh->b_bdev = NULL;
2150         return may_free;
2151 }
2152 
2153 /**
2154  * void jbd2_journal_invalidatepage()
2155  * @journal: journal to use for flush...
2156  * @page:    page to flush
2157  * @offset:  start of the range to invalidate
2158  * @length:  length of the range to invalidate
2159  *
2160  * Reap page buffers containing data after in the specified range in page.
2161  * Can return -EBUSY if buffers are part of the committing transaction and
2162  * the page is straddling i_size. Caller then has to wait for current commit
2163  * and try again.
2164  */
2165 int jbd2_journal_invalidatepage(journal_t *journal,
2166                                 struct page *page,
2167                                 unsigned int offset,
2168                                 unsigned int length)
2169 {
2170         struct buffer_head *head, *bh, *next;
2171         unsigned int stop = offset + length;
2172         unsigned int curr_off = 0;
2173         int partial_page = (offset || length < PAGE_CACHE_SIZE);
2174         int may_free = 1;
2175         int ret = 0;
2176 
2177         if (!PageLocked(page))
2178                 BUG();
2179         if (!page_has_buffers(page))
2180                 return 0;
2181 
2182         BUG_ON(stop > PAGE_CACHE_SIZE || stop < length);
2183 
2184         /* We will potentially be playing with lists other than just the
2185          * data lists (especially for journaled data mode), so be
2186          * cautious in our locking. */
2187 
2188         head = bh = page_buffers(page);
2189         do {
2190                 unsigned int next_off = curr_off + bh->b_size;
2191                 next = bh->b_this_page;
2192 
2193                 if (next_off > stop)
2194                         return 0;
2195 
2196                 if (offset <= curr_off) {
2197                         /* This block is wholly outside the truncation point */
2198                         lock_buffer(bh);
2199                         ret = journal_unmap_buffer(journal, bh, partial_page);
2200                         unlock_buffer(bh);
2201                         if (ret < 0)
2202                                 return ret;
2203                         may_free &= ret;
2204                 }
2205                 curr_off = next_off;
2206                 bh = next;
2207 
2208         } while (bh != head);
2209 
2210         if (!partial_page) {
2211                 if (may_free && try_to_free_buffers(page))
2212                         J_ASSERT(!page_has_buffers(page));
2213         }
2214         return 0;
2215 }
2216 
2217 /*
2218  * File a buffer on the given transaction list.
2219  */
2220 void __jbd2_journal_file_buffer(struct journal_head *jh,
2221                         transaction_t *transaction, int jlist)
2222 {
2223         struct journal_head **list = NULL;
2224         int was_dirty = 0;
2225         struct buffer_head *bh = jh2bh(jh);
2226 
2227         J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
2228         assert_spin_locked(&transaction->t_journal->j_list_lock);
2229 
2230         J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
2231         J_ASSERT_JH(jh, jh->b_transaction == transaction ||
2232                                 jh->b_transaction == NULL);
2233 
2234         if (jh->b_transaction && jh->b_jlist == jlist)
2235                 return;
2236 
2237         if (jlist == BJ_Metadata || jlist == BJ_Reserved ||
2238             jlist == BJ_Shadow || jlist == BJ_Forget) {
2239                 /*
2240                  * For metadata buffers, we track dirty bit in buffer_jbddirty
2241                  * instead of buffer_dirty. We should not see a dirty bit set
2242                  * here because we clear it in do_get_write_access but e.g.
2243                  * tune2fs can modify the sb and set the dirty bit at any time
2244                  * so we try to gracefully handle that.
2245                  */
2246                 if (buffer_dirty(bh))
2247                         warn_dirty_buffer(bh);
2248                 if (test_clear_buffer_dirty(bh) ||
2249                     test_clear_buffer_jbddirty(bh))
2250                         was_dirty = 1;
2251         }
2252 
2253         if (jh->b_transaction)
2254                 __jbd2_journal_temp_unlink_buffer(jh);
2255         else
2256                 jbd2_journal_grab_journal_head(bh);
2257         jh->b_transaction = transaction;
2258 
2259         switch (jlist) {
2260         case BJ_None:
2261                 J_ASSERT_JH(jh, !jh->b_committed_data);
2262                 J_ASSERT_JH(jh, !jh->b_frozen_data);
2263                 return;
2264         case BJ_Metadata:
2265                 transaction->t_nr_buffers++;
2266                 list = &transaction->t_buffers;
2267                 break;
2268         case BJ_Forget:
2269                 list = &transaction->t_forget;
2270                 break;
2271         case BJ_Shadow:
2272                 list = &transaction->t_shadow_list;
2273                 break;
2274         case BJ_Reserved:
2275                 list = &transaction->t_reserved_list;
2276                 break;
2277         }
2278 
2279         __blist_add_buffer(list, jh);
2280         jh->b_jlist = jlist;
2281 
2282         if (was_dirty)
2283                 set_buffer_jbddirty(bh);
2284 }
2285 
2286 void jbd2_journal_file_buffer(struct journal_head *jh,
2287                                 transaction_t *transaction, int jlist)
2288 {
2289         jbd_lock_bh_state(jh2bh(jh));
2290         spin_lock(&transaction->t_journal->j_list_lock);
2291         __jbd2_journal_file_buffer(jh, transaction, jlist);
2292         spin_unlock(&transaction->t_journal->j_list_lock);
2293         jbd_unlock_bh_state(jh2bh(jh));
2294 }
2295 
2296 /*
2297  * Remove a buffer from its current buffer list in preparation for
2298  * dropping it from its current transaction entirely.  If the buffer has
2299  * already started to be used by a subsequent transaction, refile the
2300  * buffer on that transaction's metadata list.
2301  *
2302  * Called under j_list_lock
2303  * Called under jbd_lock_bh_state(jh2bh(jh))
2304  *
2305  * jh and bh may be already free when this function returns
2306  */
2307 void __jbd2_journal_refile_buffer(struct journal_head *jh)
2308 {
2309         int was_dirty, jlist;
2310         struct buffer_head *bh = jh2bh(jh);
2311 
2312         J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
2313         if (jh->b_transaction)
2314                 assert_spin_locked(&jh->b_transaction->t_journal->j_list_lock);
2315 
2316         /* If the buffer is now unused, just drop it. */
2317         if (jh->b_next_transaction == NULL) {
2318                 __jbd2_journal_unfile_buffer(jh);
2319                 return;
2320         }
2321 
2322         /*
2323          * It has been modified by a later transaction: add it to the new
2324          * transaction's metadata list.
2325          */
2326 
2327         was_dirty = test_clear_buffer_jbddirty(bh);
2328         __jbd2_journal_temp_unlink_buffer(jh);
2329         /*
2330          * We set b_transaction here because b_next_transaction will inherit
2331          * our jh reference and thus __jbd2_journal_file_buffer() must not
2332          * take a new one.
2333          */
2334         jh->b_transaction = jh->b_next_transaction;
2335         jh->b_next_transaction = NULL;
2336         if (buffer_freed(bh))
2337                 jlist = BJ_Forget;
2338         else if (jh->b_modified)
2339                 jlist = BJ_Metadata;
2340         else
2341                 jlist = BJ_Reserved;
2342         __jbd2_journal_file_buffer(jh, jh->b_transaction, jlist);
2343         J_ASSERT_JH(jh, jh->b_transaction->t_state == T_RUNNING);
2344 
2345         if (was_dirty)
2346                 set_buffer_jbddirty(bh);
2347 }
2348 
2349 /*
2350  * __jbd2_journal_refile_buffer() with necessary locking added. We take our
2351  * bh reference so that we can safely unlock bh.
2352  *
2353  * The jh and bh may be freed by this call.
2354  */
2355 void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh)
2356 {
2357         struct buffer_head *bh = jh2bh(jh);
2358 
2359         /* Get reference so that buffer cannot be freed before we unlock it */
2360         get_bh(bh);
2361         jbd_lock_bh_state(bh);
2362         spin_lock(&journal->j_list_lock);
2363         __jbd2_journal_refile_buffer(jh);
2364         jbd_unlock_bh_state(bh);
2365         spin_unlock(&journal->j_list_lock);
2366         __brelse(bh);
2367 }
2368 
2369 /*
2370  * File inode in the inode list of the handle's transaction
2371  */
2372 int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode)
2373 {
2374         transaction_t *transaction = handle->h_transaction;
2375         journal_t *journal;
2376 
2377         WARN_ON(!transaction);
2378         if (is_handle_aborted(handle))
2379                 return -EROFS;
2380         journal = transaction->t_journal;
2381 
2382         jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode->i_vfs_inode->i_ino,
2383                         transaction->t_tid);
2384 
2385         /*
2386          * First check whether inode isn't already on the transaction's
2387          * lists without taking the lock. Note that this check is safe
2388          * without the lock as we cannot race with somebody removing inode
2389          * from the transaction. The reason is that we remove inode from the
2390          * transaction only in journal_release_jbd_inode() and when we commit
2391          * the transaction. We are guarded from the first case by holding
2392          * a reference to the inode. We are safe against the second case
2393          * because if jinode->i_transaction == transaction, commit code
2394          * cannot touch the transaction because we hold reference to it,
2395          * and if jinode->i_next_transaction == transaction, commit code
2396          * will only file the inode where we want it.
2397          */
2398         if (jinode->i_transaction == transaction ||
2399             jinode->i_next_transaction == transaction)
2400                 return 0;
2401 
2402         spin_lock(&journal->j_list_lock);
2403 
2404         if (jinode->i_transaction == transaction ||
2405             jinode->i_next_transaction == transaction)
2406                 goto done;
2407 
2408         /*
2409          * We only ever set this variable to 1 so the test is safe. Since
2410          * t_need_data_flush is likely to be set, we do the test to save some
2411          * cacheline bouncing
2412          */
2413         if (!transaction->t_need_data_flush)
2414                 transaction->t_need_data_flush = 1;
2415         /* On some different transaction's list - should be
2416          * the committing one */
2417         if (jinode->i_transaction) {
2418                 J_ASSERT(jinode->i_next_transaction == NULL);
2419                 J_ASSERT(jinode->i_transaction ==
2420                                         journal->j_committing_transaction);
2421                 jinode->i_next_transaction = transaction;
2422                 goto done;
2423         }
2424         /* Not on any transaction list... */
2425         J_ASSERT(!jinode->i_next_transaction);
2426         jinode->i_transaction = transaction;
2427         list_add(&jinode->i_list, &transaction->t_inode_list);
2428 done:
2429         spin_unlock(&journal->j_list_lock);
2430 
2431         return 0;
2432 }
2433 
2434 /*
2435  * File truncate and transaction commit interact with each other in a
2436  * non-trivial way.  If a transaction writing data block A is
2437  * committing, we cannot discard the data by truncate until we have
2438  * written them.  Otherwise if we crashed after the transaction with
2439  * write has committed but before the transaction with truncate has
2440  * committed, we could see stale data in block A.  This function is a
2441  * helper to solve this problem.  It starts writeout of the truncated
2442  * part in case it is in the committing transaction.
2443  *
2444  * Filesystem code must call this function when inode is journaled in
2445  * ordered mode before truncation happens and after the inode has been
2446  * placed on orphan list with the new inode size. The second condition
2447  * avoids the race that someone writes new data and we start
2448  * committing the transaction after this function has been called but
2449  * before a transaction for truncate is started (and furthermore it
2450  * allows us to optimize the case where the addition to orphan list
2451  * happens in the same transaction as write --- we don't have to write
2452  * any data in such case).
2453  */
2454 int jbd2_journal_begin_ordered_truncate(journal_t *journal,
2455                                         struct jbd2_inode *jinode,
2456                                         loff_t new_size)
2457 {
2458         transaction_t *inode_trans, *commit_trans;
2459         int ret = 0;
2460 
2461         /* This is a quick check to avoid locking if not necessary */
2462         if (!jinode->i_transaction)
2463                 goto out;
2464         /* Locks are here just to force reading of recent values, it is
2465          * enough that the transaction was not committing before we started
2466          * a transaction adding the inode to orphan list */
2467         read_lock(&journal->j_state_lock);
2468         commit_trans = journal->j_committing_transaction;
2469         read_unlock(&journal->j_state_lock);
2470         spin_lock(&journal->j_list_lock);
2471         inode_trans = jinode->i_transaction;
2472         spin_unlock(&journal->j_list_lock);
2473         if (inode_trans == commit_trans) {
2474                 ret = filemap_fdatawrite_range(jinode->i_vfs_inode->i_mapping,
2475                         new_size, LLONG_MAX);
2476                 if (ret)
2477                         jbd2_journal_abort(journal, ret);
2478         }
2479 out:
2480         return ret;
2481 }
2482 

~ [ 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