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

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

Version: ~ [ linux-5.4-rc3 ] ~ [ linux-5.3.6 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.79 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.149 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.196 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.196 ] ~ [ 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-3.16.75 ] ~ [ 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                 return ret;
520         }
521         handle->h_type = type;
522         handle->h_line_no = line_no;
523         return 0;
524 }
525 EXPORT_SYMBOL(jbd2_journal_start_reserved);
526 
527 /**
528  * int jbd2_journal_extend() - extend buffer credits.
529  * @handle:  handle to 'extend'
530  * @nblocks: nr blocks to try to extend by.
531  *
532  * Some transactions, such as large extends and truncates, can be done
533  * atomically all at once or in several stages.  The operation requests
534  * a credit for a number of buffer modications in advance, but can
535  * extend its credit if it needs more.
536  *
537  * jbd2_journal_extend tries to give the running handle more buffer credits.
538  * It does not guarantee that allocation - this is a best-effort only.
539  * The calling process MUST be able to deal cleanly with a failure to
540  * extend here.
541  *
542  * Return 0 on success, non-zero on failure.
543  *
544  * return code < 0 implies an error
545  * return code > 0 implies normal transaction-full status.
546  */
547 int jbd2_journal_extend(handle_t *handle, int nblocks)
548 {
549         transaction_t *transaction = handle->h_transaction;
550         journal_t *journal;
551         int result;
552         int wanted;
553 
554         if (is_handle_aborted(handle))
555                 return -EROFS;
556         journal = transaction->t_journal;
557 
558         result = 1;
559 
560         read_lock(&journal->j_state_lock);
561 
562         /* Don't extend a locked-down transaction! */
563         if (transaction->t_state != T_RUNNING) {
564                 jbd_debug(3, "denied handle %p %d blocks: "
565                           "transaction not running\n", handle, nblocks);
566                 goto error_out;
567         }
568 
569         spin_lock(&transaction->t_handle_lock);
570         wanted = atomic_add_return(nblocks,
571                                    &transaction->t_outstanding_credits);
572 
573         if (wanted > journal->j_max_transaction_buffers) {
574                 jbd_debug(3, "denied handle %p %d blocks: "
575                           "transaction too large\n", handle, nblocks);
576                 atomic_sub(nblocks, &transaction->t_outstanding_credits);
577                 goto unlock;
578         }
579 
580         if (wanted + (wanted >> JBD2_CONTROL_BLOCKS_SHIFT) >
581             jbd2_log_space_left(journal)) {
582                 jbd_debug(3, "denied handle %p %d blocks: "
583                           "insufficient log space\n", handle, nblocks);
584                 atomic_sub(nblocks, &transaction->t_outstanding_credits);
585                 goto unlock;
586         }
587 
588         trace_jbd2_handle_extend(journal->j_fs_dev->bd_dev,
589                                  transaction->t_tid,
590                                  handle->h_type, handle->h_line_no,
591                                  handle->h_buffer_credits,
592                                  nblocks);
593 
594         handle->h_buffer_credits += nblocks;
595         handle->h_requested_credits += nblocks;
596         result = 0;
597 
598         jbd_debug(3, "extended handle %p by %d\n", handle, nblocks);
599 unlock:
600         spin_unlock(&transaction->t_handle_lock);
601 error_out:
602         read_unlock(&journal->j_state_lock);
603         return result;
604 }
605 
606 
607 /**
608  * int jbd2_journal_restart() - restart a handle .
609  * @handle:  handle to restart
610  * @nblocks: nr credits requested
611  *
612  * Restart a handle for a multi-transaction filesystem
613  * operation.
614  *
615  * If the jbd2_journal_extend() call above fails to grant new buffer credits
616  * to a running handle, a call to jbd2_journal_restart will commit the
617  * handle's transaction so far and reattach the handle to a new
618  * transaction capabable of guaranteeing the requested number of
619  * credits. We preserve reserved handle if there's any attached to the
620  * passed in handle.
621  */
622 int jbd2__journal_restart(handle_t *handle, int nblocks, gfp_t gfp_mask)
623 {
624         transaction_t *transaction = handle->h_transaction;
625         journal_t *journal;
626         tid_t           tid;
627         int             need_to_start, ret;
628 
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         if (is_handle_aborted(handle))
793                 return -EROFS;
794         journal = transaction->t_journal;
795 
796         jbd_debug(5, "journal_head %p, force_copy %d\n", jh, force_copy);
797 
798         JBUFFER_TRACE(jh, "entry");
799 repeat:
800         bh = jh2bh(jh);
801 
802         /* @@@ Need to check for errors here at some point. */
803 
804         start_lock = jiffies;
805         lock_buffer(bh);
806         jbd_lock_bh_state(bh);
807 
808         /* If it takes too long to lock the buffer, trace it */
809         time_lock = jbd2_time_diff(start_lock, jiffies);
810         if (time_lock > HZ/10)
811                 trace_jbd2_lock_buffer_stall(bh->b_bdev->bd_dev,
812                         jiffies_to_msecs(time_lock));
813 
814         /* We now hold the buffer lock so it is safe to query the buffer
815          * state.  Is the buffer dirty?
816          *
817          * If so, there are two possibilities.  The buffer may be
818          * non-journaled, and undergoing a quite legitimate writeback.
819          * Otherwise, it is journaled, and we don't expect dirty buffers
820          * in that state (the buffers should be marked JBD_Dirty
821          * instead.)  So either the IO is being done under our own
822          * control and this is a bug, or it's a third party IO such as
823          * dump(8) (which may leave the buffer scheduled for read ---
824          * ie. locked but not dirty) or tune2fs (which may actually have
825          * the buffer dirtied, ugh.)  */
826 
827         if (buffer_dirty(bh)) {
828                 /*
829                  * First question: is this buffer already part of the current
830                  * transaction or the existing committing transaction?
831                  */
832                 if (jh->b_transaction) {
833                         J_ASSERT_JH(jh,
834                                 jh->b_transaction == transaction ||
835                                 jh->b_transaction ==
836                                         journal->j_committing_transaction);
837                         if (jh->b_next_transaction)
838                                 J_ASSERT_JH(jh, jh->b_next_transaction ==
839                                                         transaction);
840                         warn_dirty_buffer(bh);
841                 }
842                 /*
843                  * In any case we need to clean the dirty flag and we must
844                  * do it under the buffer lock to be sure we don't race
845                  * with running write-out.
846                  */
847                 JBUFFER_TRACE(jh, "Journalling dirty buffer");
848                 clear_buffer_dirty(bh);
849                 set_buffer_jbddirty(bh);
850         }
851 
852         unlock_buffer(bh);
853 
854         error = -EROFS;
855         if (is_handle_aborted(handle)) {
856                 jbd_unlock_bh_state(bh);
857                 goto out;
858         }
859         error = 0;
860 
861         /*
862          * The buffer is already part of this transaction if b_transaction or
863          * b_next_transaction points to it
864          */
865         if (jh->b_transaction == transaction ||
866             jh->b_next_transaction == transaction)
867                 goto done;
868 
869         /*
870          * this is the first time this transaction is touching this buffer,
871          * reset the modified flag
872          */
873        jh->b_modified = 0;
874 
875         /*
876          * If there is already a copy-out version of this buffer, then we don't
877          * need to make another one
878          */
879         if (jh->b_frozen_data) {
880                 JBUFFER_TRACE(jh, "has frozen data");
881                 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
882                 jh->b_next_transaction = transaction;
883                 goto done;
884         }
885 
886         /* Is there data here we need to preserve? */
887 
888         if (jh->b_transaction && jh->b_transaction != transaction) {
889                 JBUFFER_TRACE(jh, "owned by older transaction");
890                 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
891                 J_ASSERT_JH(jh, jh->b_transaction ==
892                                         journal->j_committing_transaction);
893 
894                 /* There is one case we have to be very careful about.
895                  * If the committing transaction is currently writing
896                  * this buffer out to disk and has NOT made a copy-out,
897                  * then we cannot modify the buffer contents at all
898                  * right now.  The essence of copy-out is that it is the
899                  * extra copy, not the primary copy, which gets
900                  * journaled.  If the primary copy is already going to
901                  * disk then we cannot do copy-out here. */
902 
903                 if (buffer_shadow(bh)) {
904                         JBUFFER_TRACE(jh, "on shadow: sleep");
905                         jbd_unlock_bh_state(bh);
906                         wait_on_bit(&bh->b_state, BH_Shadow,
907                                     sleep_on_shadow_bh, TASK_UNINTERRUPTIBLE);
908                         goto repeat;
909                 }
910 
911                 /*
912                  * Only do the copy if the currently-owning transaction still
913                  * needs it. If buffer isn't on BJ_Metadata list, the
914                  * committing transaction is past that stage (here we use the
915                  * fact that BH_Shadow is set under bh_state lock together with
916                  * refiling to BJ_Shadow list and at this point we know the
917                  * buffer doesn't have BH_Shadow set).
918                  *
919                  * Subtle point, though: if this is a get_undo_access,
920                  * then we will be relying on the frozen_data to contain
921                  * the new value of the committed_data record after the
922                  * transaction, so we HAVE to force the frozen_data copy
923                  * in that case.
924                  */
925                 if (jh->b_jlist == BJ_Metadata || force_copy) {
926                         JBUFFER_TRACE(jh, "generate frozen data");
927                         if (!frozen_buffer) {
928                                 JBUFFER_TRACE(jh, "allocate memory for buffer");
929                                 jbd_unlock_bh_state(bh);
930                                 frozen_buffer =
931                                         jbd2_alloc(jh2bh(jh)->b_size,
932                                                          GFP_NOFS);
933                                 if (!frozen_buffer) {
934                                         printk(KERN_EMERG
935                                                "%s: OOM for frozen_buffer\n",
936                                                __func__);
937                                         JBUFFER_TRACE(jh, "oom!");
938                                         error = -ENOMEM;
939                                         jbd_lock_bh_state(bh);
940                                         goto done;
941                                 }
942                                 goto repeat;
943                         }
944                         jh->b_frozen_data = frozen_buffer;
945                         frozen_buffer = NULL;
946                         need_copy = 1;
947                 }
948                 jh->b_next_transaction = transaction;
949         }
950 
951 
952         /*
953          * Finally, if the buffer is not journaled right now, we need to make
954          * sure it doesn't get written to disk before the caller actually
955          * commits the new data
956          */
957         if (!jh->b_transaction) {
958                 JBUFFER_TRACE(jh, "no transaction");
959                 J_ASSERT_JH(jh, !jh->b_next_transaction);
960                 JBUFFER_TRACE(jh, "file as BJ_Reserved");
961                 spin_lock(&journal->j_list_lock);
962                 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
963                 spin_unlock(&journal->j_list_lock);
964         }
965 
966 done:
967         if (need_copy) {
968                 struct page *page;
969                 int offset;
970                 char *source;
971 
972                 J_EXPECT_JH(jh, buffer_uptodate(jh2bh(jh)),
973                             "Possible IO failure.\n");
974                 page = jh2bh(jh)->b_page;
975                 offset = offset_in_page(jh2bh(jh)->b_data);
976                 source = kmap_atomic(page);
977                 /* Fire data frozen trigger just before we copy the data */
978                 jbd2_buffer_frozen_trigger(jh, source + offset,
979                                            jh->b_triggers);
980                 memcpy(jh->b_frozen_data, source+offset, jh2bh(jh)->b_size);
981                 kunmap_atomic(source);
982 
983                 /*
984                  * Now that the frozen data is saved off, we need to store
985                  * any matching triggers.
986                  */
987                 jh->b_frozen_triggers = jh->b_triggers;
988         }
989         jbd_unlock_bh_state(bh);
990 
991         /*
992          * If we are about to journal a buffer, then any revoke pending on it is
993          * no longer valid
994          */
995         jbd2_journal_cancel_revoke(handle, jh);
996 
997 out:
998         if (unlikely(frozen_buffer))    /* It's usually NULL */
999                 jbd2_free(frozen_buffer, bh->b_size);
1000 
1001         JBUFFER_TRACE(jh, "exit");
1002         return error;
1003 }
1004 
1005 /**
1006  * int jbd2_journal_get_write_access() - notify intent to modify a buffer for metadata (not data) update.
1007  * @handle: transaction to add buffer modifications to
1008  * @bh:     bh to be used for metadata writes
1009  *
1010  * Returns an error code or 0 on success.
1011  *
1012  * In full data journalling mode the buffer may be of type BJ_AsyncData,
1013  * because we're write()ing a buffer which is also part of a shared mapping.
1014  */
1015 
1016 int jbd2_journal_get_write_access(handle_t *handle, struct buffer_head *bh)
1017 {
1018         struct journal_head *jh = jbd2_journal_add_journal_head(bh);
1019         int rc;
1020 
1021         /* We do not want to get caught playing with fields which the
1022          * log thread also manipulates.  Make sure that the buffer
1023          * completes any outstanding IO before proceeding. */
1024         rc = do_get_write_access(handle, jh, 0);
1025         jbd2_journal_put_journal_head(jh);
1026         return rc;
1027 }
1028 
1029 
1030 /*
1031  * When the user wants to journal a newly created buffer_head
1032  * (ie. getblk() returned a new buffer and we are going to populate it
1033  * manually rather than reading off disk), then we need to keep the
1034  * buffer_head locked until it has been completely filled with new
1035  * data.  In this case, we should be able to make the assertion that
1036  * the bh is not already part of an existing transaction.
1037  *
1038  * The buffer should already be locked by the caller by this point.
1039  * There is no lock ranking violation: it was a newly created,
1040  * unlocked buffer beforehand. */
1041 
1042 /**
1043  * int jbd2_journal_get_create_access () - notify intent to use newly created bh
1044  * @handle: transaction to new buffer to
1045  * @bh: new buffer.
1046  *
1047  * Call this if you create a new bh.
1048  */
1049 int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh)
1050 {
1051         transaction_t *transaction = handle->h_transaction;
1052         journal_t *journal;
1053         struct journal_head *jh = jbd2_journal_add_journal_head(bh);
1054         int err;
1055 
1056         jbd_debug(5, "journal_head %p\n", jh);
1057         err = -EROFS;
1058         if (is_handle_aborted(handle))
1059                 goto out;
1060         journal = transaction->t_journal;
1061         err = 0;
1062 
1063         JBUFFER_TRACE(jh, "entry");
1064         /*
1065          * The buffer may already belong to this transaction due to pre-zeroing
1066          * in the filesystem's new_block code.  It may also be on the previous,
1067          * committing transaction's lists, but it HAS to be in Forget state in
1068          * that case: the transaction must have deleted the buffer for it to be
1069          * reused here.
1070          */
1071         jbd_lock_bh_state(bh);
1072         spin_lock(&journal->j_list_lock);
1073         J_ASSERT_JH(jh, (jh->b_transaction == transaction ||
1074                 jh->b_transaction == NULL ||
1075                 (jh->b_transaction == journal->j_committing_transaction &&
1076                           jh->b_jlist == BJ_Forget)));
1077 
1078         J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1079         J_ASSERT_JH(jh, buffer_locked(jh2bh(jh)));
1080 
1081         if (jh->b_transaction == NULL) {
1082                 /*
1083                  * Previous jbd2_journal_forget() could have left the buffer
1084                  * with jbddirty bit set because it was being committed. When
1085                  * the commit finished, we've filed the buffer for
1086                  * checkpointing and marked it dirty. Now we are reallocating
1087                  * the buffer so the transaction freeing it must have
1088                  * committed and so it's safe to clear the dirty bit.
1089                  */
1090                 clear_buffer_dirty(jh2bh(jh));
1091                 /* first access by this transaction */
1092                 jh->b_modified = 0;
1093 
1094                 JBUFFER_TRACE(jh, "file as BJ_Reserved");
1095                 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
1096         } else if (jh->b_transaction == journal->j_committing_transaction) {
1097                 /* first access by this transaction */
1098                 jh->b_modified = 0;
1099 
1100                 JBUFFER_TRACE(jh, "set next transaction");
1101                 jh->b_next_transaction = transaction;
1102         }
1103         spin_unlock(&journal->j_list_lock);
1104         jbd_unlock_bh_state(bh);
1105 
1106         /*
1107          * akpm: I added this.  ext3_alloc_branch can pick up new indirect
1108          * blocks which contain freed but then revoked metadata.  We need
1109          * to cancel the revoke in case we end up freeing it yet again
1110          * and the reallocating as data - this would cause a second revoke,
1111          * which hits an assertion error.
1112          */
1113         JBUFFER_TRACE(jh, "cancelling revoke");
1114         jbd2_journal_cancel_revoke(handle, jh);
1115 out:
1116         jbd2_journal_put_journal_head(jh);
1117         return err;
1118 }
1119 
1120 /**
1121  * int jbd2_journal_get_undo_access() -  Notify intent to modify metadata with
1122  *     non-rewindable consequences
1123  * @handle: transaction
1124  * @bh: buffer to undo
1125  *
1126  * Sometimes there is a need to distinguish between metadata which has
1127  * been committed to disk and that which has not.  The ext3fs code uses
1128  * this for freeing and allocating space, we have to make sure that we
1129  * do not reuse freed space until the deallocation has been committed,
1130  * since if we overwrote that space we would make the delete
1131  * un-rewindable in case of a crash.
1132  *
1133  * To deal with that, jbd2_journal_get_undo_access requests write access to a
1134  * buffer for parts of non-rewindable operations such as delete
1135  * operations on the bitmaps.  The journaling code must keep a copy of
1136  * the buffer's contents prior to the undo_access call until such time
1137  * as we know that the buffer has definitely been committed to disk.
1138  *
1139  * We never need to know which transaction the committed data is part
1140  * of, buffers touched here are guaranteed to be dirtied later and so
1141  * will be committed to a new transaction in due course, at which point
1142  * we can discard the old committed data pointer.
1143  *
1144  * Returns error number or 0 on success.
1145  */
1146 int jbd2_journal_get_undo_access(handle_t *handle, struct buffer_head *bh)
1147 {
1148         int err;
1149         struct journal_head *jh = jbd2_journal_add_journal_head(bh);
1150         char *committed_data = NULL;
1151 
1152         JBUFFER_TRACE(jh, "entry");
1153 
1154         /*
1155          * Do this first --- it can drop the journal lock, so we want to
1156          * make sure that obtaining the committed_data is done
1157          * atomically wrt. completion of any outstanding commits.
1158          */
1159         err = do_get_write_access(handle, jh, 1);
1160         if (err)
1161                 goto out;
1162 
1163 repeat:
1164         if (!jh->b_committed_data) {
1165                 committed_data = jbd2_alloc(jh2bh(jh)->b_size, GFP_NOFS);
1166                 if (!committed_data) {
1167                         printk(KERN_EMERG "%s: No memory for committed data\n",
1168                                 __func__);
1169                         err = -ENOMEM;
1170                         goto out;
1171                 }
1172         }
1173 
1174         jbd_lock_bh_state(bh);
1175         if (!jh->b_committed_data) {
1176                 /* Copy out the current buffer contents into the
1177                  * preserved, committed copy. */
1178                 JBUFFER_TRACE(jh, "generate b_committed data");
1179                 if (!committed_data) {
1180                         jbd_unlock_bh_state(bh);
1181                         goto repeat;
1182                 }
1183 
1184                 jh->b_committed_data = committed_data;
1185                 committed_data = NULL;
1186                 memcpy(jh->b_committed_data, bh->b_data, bh->b_size);
1187         }
1188         jbd_unlock_bh_state(bh);
1189 out:
1190         jbd2_journal_put_journal_head(jh);
1191         if (unlikely(committed_data))
1192                 jbd2_free(committed_data, bh->b_size);
1193         return err;
1194 }
1195 
1196 /**
1197  * void jbd2_journal_set_triggers() - Add triggers for commit writeout
1198  * @bh: buffer to trigger on
1199  * @type: struct jbd2_buffer_trigger_type containing the trigger(s).
1200  *
1201  * Set any triggers on this journal_head.  This is always safe, because
1202  * triggers for a committing buffer will be saved off, and triggers for
1203  * a running transaction will match the buffer in that transaction.
1204  *
1205  * Call with NULL to clear the triggers.
1206  */
1207 void jbd2_journal_set_triggers(struct buffer_head *bh,
1208                                struct jbd2_buffer_trigger_type *type)
1209 {
1210         struct journal_head *jh = jbd2_journal_grab_journal_head(bh);
1211 
1212         if (WARN_ON(!jh))
1213                 return;
1214         jh->b_triggers = type;
1215         jbd2_journal_put_journal_head(jh);
1216 }
1217 
1218 void jbd2_buffer_frozen_trigger(struct journal_head *jh, void *mapped_data,
1219                                 struct jbd2_buffer_trigger_type *triggers)
1220 {
1221         struct buffer_head *bh = jh2bh(jh);
1222 
1223         if (!triggers || !triggers->t_frozen)
1224                 return;
1225 
1226         triggers->t_frozen(triggers, bh, mapped_data, bh->b_size);
1227 }
1228 
1229 void jbd2_buffer_abort_trigger(struct journal_head *jh,
1230                                struct jbd2_buffer_trigger_type *triggers)
1231 {
1232         if (!triggers || !triggers->t_abort)
1233                 return;
1234 
1235         triggers->t_abort(triggers, jh2bh(jh));
1236 }
1237 
1238 
1239 
1240 /**
1241  * int jbd2_journal_dirty_metadata() -  mark a buffer as containing dirty metadata
1242  * @handle: transaction to add buffer to.
1243  * @bh: buffer to mark
1244  *
1245  * mark dirty metadata which needs to be journaled as part of the current
1246  * transaction.
1247  *
1248  * The buffer must have previously had jbd2_journal_get_write_access()
1249  * called so that it has a valid journal_head attached to the buffer
1250  * head.
1251  *
1252  * The buffer is placed on the transaction's metadata list and is marked
1253  * as belonging to the transaction.
1254  *
1255  * Returns error number or 0 on success.
1256  *
1257  * Special care needs to be taken if the buffer already belongs to the
1258  * current committing transaction (in which case we should have frozen
1259  * data present for that commit).  In that case, we don't relink the
1260  * buffer: that only gets done when the old transaction finally
1261  * completes its commit.
1262  */
1263 int jbd2_journal_dirty_metadata(handle_t *handle, struct buffer_head *bh)
1264 {
1265         transaction_t *transaction = handle->h_transaction;
1266         journal_t *journal;
1267         struct journal_head *jh;
1268         int ret = 0;
1269 
1270         if (is_handle_aborted(handle))
1271                 return -EROFS;
1272         journal = transaction->t_journal;
1273         jh = jbd2_journal_grab_journal_head(bh);
1274         if (!jh) {
1275                 ret = -EUCLEAN;
1276                 goto out;
1277         }
1278         jbd_debug(5, "journal_head %p\n", jh);
1279         JBUFFER_TRACE(jh, "entry");
1280 
1281         jbd_lock_bh_state(bh);
1282 
1283         if (jh->b_modified == 0) {
1284                 /*
1285                  * This buffer's got modified and becoming part
1286                  * of the transaction. This needs to be done
1287                  * once a transaction -bzzz
1288                  */
1289                 jh->b_modified = 1;
1290                 if (handle->h_buffer_credits <= 0) {
1291                         ret = -ENOSPC;
1292                         goto out_unlock_bh;
1293                 }
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         return ret;
1377 }
1378 
1379 /**
1380  * void jbd2_journal_forget() - bforget() for potentially-journaled buffers.
1381  * @handle: transaction handle
1382  * @bh:     bh to 'forget'
1383  *
1384  * We can only do the bforget if there are no commits pending against the
1385  * buffer.  If the buffer is dirty in the current running transaction we
1386  * can safely unlink it.
1387  *
1388  * bh may not be a journalled buffer at all - it may be a non-JBD
1389  * buffer which came off the hashtable.  Check for this.
1390  *
1391  * Decrements bh->b_count by one.
1392  *
1393  * Allow this call even if the handle has aborted --- it may be part of
1394  * the caller's cleanup after an abort.
1395  */
1396 int jbd2_journal_forget (handle_t *handle, struct buffer_head *bh)
1397 {
1398         transaction_t *transaction = handle->h_transaction;
1399         journal_t *journal;
1400         struct journal_head *jh;
1401         int drop_reserve = 0;
1402         int err = 0;
1403         int was_modified = 0;
1404 
1405         if (is_handle_aborted(handle))
1406                 return -EROFS;
1407         journal = transaction->t_journal;
1408 
1409         BUFFER_TRACE(bh, "entry");
1410 
1411         jbd_lock_bh_state(bh);
1412         spin_lock(&journal->j_list_lock);
1413 
1414         if (!buffer_jbd(bh))
1415                 goto not_jbd;
1416         jh = bh2jh(bh);
1417 
1418         /* Critical error: attempting to delete a bitmap buffer, maybe?
1419          * Don't do any jbd operations, and return an error. */
1420         if (!J_EXPECT_JH(jh, !jh->b_committed_data,
1421                          "inconsistent data on disk")) {
1422                 err = -EIO;
1423                 goto not_jbd;
1424         }
1425 
1426         /* keep track of whether or not this transaction modified us */
1427         was_modified = jh->b_modified;
1428 
1429         /*
1430          * The buffer's going from the transaction, we must drop
1431          * all references -bzzz
1432          */
1433         jh->b_modified = 0;
1434 
1435         if (jh->b_transaction == transaction) {
1436                 J_ASSERT_JH(jh, !jh->b_frozen_data);
1437 
1438                 /* If we are forgetting a buffer which is already part
1439                  * of this transaction, then we can just drop it from
1440                  * the transaction immediately. */
1441                 clear_buffer_dirty(bh);
1442                 clear_buffer_jbddirty(bh);
1443 
1444                 JBUFFER_TRACE(jh, "belongs to current transaction: unfile");
1445 
1446                 /*
1447                  * we only want to drop a reference if this transaction
1448                  * modified the buffer
1449                  */
1450                 if (was_modified)
1451                         drop_reserve = 1;
1452 
1453                 /*
1454                  * We are no longer going to journal this buffer.
1455                  * However, the commit of this transaction is still
1456                  * important to the buffer: the delete that we are now
1457                  * processing might obsolete an old log entry, so by
1458                  * committing, we can satisfy the buffer's checkpoint.
1459                  *
1460                  * So, if we have a checkpoint on the buffer, we should
1461                  * now refile the buffer on our BJ_Forget list so that
1462                  * we know to remove the checkpoint after we commit.
1463                  */
1464 
1465                 if (jh->b_cp_transaction) {
1466                         __jbd2_journal_temp_unlink_buffer(jh);
1467                         __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
1468                 } else {
1469                         __jbd2_journal_unfile_buffer(jh);
1470                         if (!buffer_jbd(bh)) {
1471                                 spin_unlock(&journal->j_list_lock);
1472                                 jbd_unlock_bh_state(bh);
1473                                 __bforget(bh);
1474                                 goto drop;
1475                         }
1476                 }
1477         } else if (jh->b_transaction) {
1478                 J_ASSERT_JH(jh, (jh->b_transaction ==
1479                                  journal->j_committing_transaction));
1480                 /* However, if the buffer is still owned by a prior
1481                  * (committing) transaction, we can't drop it yet... */
1482                 JBUFFER_TRACE(jh, "belongs to older transaction");
1483                 /* ... but we CAN drop it from the new transaction if we
1484                  * have also modified it since the original commit. */
1485 
1486                 if (jh->b_next_transaction) {
1487                         J_ASSERT(jh->b_next_transaction == transaction);
1488                         jh->b_next_transaction = NULL;
1489 
1490                         /*
1491                          * only drop a reference if this transaction modified
1492                          * the buffer
1493                          */
1494                         if (was_modified)
1495                                 drop_reserve = 1;
1496                 }
1497         }
1498 
1499 not_jbd:
1500         spin_unlock(&journal->j_list_lock);
1501         jbd_unlock_bh_state(bh);
1502         __brelse(bh);
1503 drop:
1504         if (drop_reserve) {
1505                 /* no need to reserve log space for this block -bzzz */
1506                 handle->h_buffer_credits++;
1507         }
1508         return err;
1509 }
1510 
1511 /**
1512  * int jbd2_journal_stop() - complete a transaction
1513  * @handle: tranaction to complete.
1514  *
1515  * All done for a particular handle.
1516  *
1517  * There is not much action needed here.  We just return any remaining
1518  * buffer credits to the transaction and remove the handle.  The only
1519  * complication is that we need to start a commit operation if the
1520  * filesystem is marked for synchronous update.
1521  *
1522  * jbd2_journal_stop itself will not usually return an error, but it may
1523  * do so in unusual circumstances.  In particular, expect it to
1524  * return -EIO if a jbd2_journal_abort has been executed since the
1525  * transaction began.
1526  */
1527 int jbd2_journal_stop(handle_t *handle)
1528 {
1529         transaction_t *transaction = handle->h_transaction;
1530         journal_t *journal;
1531         int err = 0, wait_for_commit = 0;
1532         tid_t tid;
1533         pid_t pid;
1534 
1535         if (!transaction) {
1536                 /*
1537                  * Handle is already detached from the transaction so
1538                  * there is nothing to do other than decrease a refcount,
1539                  * or free the handle if refcount drops to zero
1540                  */
1541                 if (--handle->h_ref > 0) {
1542                         jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1543                                                          handle->h_ref);
1544                         return err;
1545                 } else {
1546                         if (handle->h_rsv_handle)
1547                                 jbd2_free_handle(handle->h_rsv_handle);
1548                         goto free_and_exit;
1549                 }
1550         }
1551         journal = transaction->t_journal;
1552 
1553         J_ASSERT(journal_current_handle() == handle);
1554 
1555         if (is_handle_aborted(handle))
1556                 err = -EIO;
1557         else
1558                 J_ASSERT(atomic_read(&transaction->t_updates) > 0);
1559 
1560         if (--handle->h_ref > 0) {
1561                 jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1562                           handle->h_ref);
1563                 return err;
1564         }
1565 
1566         jbd_debug(4, "Handle %p going down\n", handle);
1567         trace_jbd2_handle_stats(journal->j_fs_dev->bd_dev,
1568                                 transaction->t_tid,
1569                                 handle->h_type, handle->h_line_no,
1570                                 jiffies - handle->h_start_jiffies,
1571                                 handle->h_sync, handle->h_requested_credits,
1572                                 (handle->h_requested_credits -
1573                                  handle->h_buffer_credits));
1574 
1575         /*
1576          * Implement synchronous transaction batching.  If the handle
1577          * was synchronous, don't force a commit immediately.  Let's
1578          * yield and let another thread piggyback onto this
1579          * transaction.  Keep doing that while new threads continue to
1580          * arrive.  It doesn't cost much - we're about to run a commit
1581          * and sleep on IO anyway.  Speeds up many-threaded, many-dir
1582          * operations by 30x or more...
1583          *
1584          * We try and optimize the sleep time against what the
1585          * underlying disk can do, instead of having a static sleep
1586          * time.  This is useful for the case where our storage is so
1587          * fast that it is more optimal to go ahead and force a flush
1588          * and wait for the transaction to be committed than it is to
1589          * wait for an arbitrary amount of time for new writers to
1590          * join the transaction.  We achieve this by measuring how
1591          * long it takes to commit a transaction, and compare it with
1592          * how long this transaction has been running, and if run time
1593          * < commit time then we sleep for the delta and commit.  This
1594          * greatly helps super fast disks that would see slowdowns as
1595          * more threads started doing fsyncs.
1596          *
1597          * But don't do this if this process was the most recent one
1598          * to perform a synchronous write.  We do this to detect the
1599          * case where a single process is doing a stream of sync
1600          * writes.  No point in waiting for joiners in that case.
1601          *
1602          * Setting max_batch_time to 0 disables this completely.
1603          */
1604         pid = current->pid;
1605         if (handle->h_sync && journal->j_last_sync_writer != pid &&
1606             journal->j_max_batch_time) {
1607                 u64 commit_time, trans_time;
1608 
1609                 journal->j_last_sync_writer = pid;
1610 
1611                 read_lock(&journal->j_state_lock);
1612                 commit_time = journal->j_average_commit_time;
1613                 read_unlock(&journal->j_state_lock);
1614 
1615                 trans_time = ktime_to_ns(ktime_sub(ktime_get(),
1616                                                    transaction->t_start_time));
1617 
1618                 commit_time = max_t(u64, commit_time,
1619                                     1000*journal->j_min_batch_time);
1620                 commit_time = min_t(u64, commit_time,
1621                                     1000*journal->j_max_batch_time);
1622 
1623                 if (trans_time < commit_time) {
1624                         ktime_t expires = ktime_add_ns(ktime_get(),
1625                                                        commit_time);
1626                         set_current_state(TASK_UNINTERRUPTIBLE);
1627                         schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1628                 }
1629         }
1630 
1631         if (handle->h_sync)
1632                 transaction->t_synchronous_commit = 1;
1633         current->journal_info = NULL;
1634         atomic_sub(handle->h_buffer_credits,
1635                    &transaction->t_outstanding_credits);
1636 
1637         /*
1638          * If the handle is marked SYNC, we need to set another commit
1639          * going!  We also want to force a commit if the current
1640          * transaction is occupying too much of the log, or if the
1641          * transaction is too old now.
1642          */
1643         if (handle->h_sync ||
1644             (atomic_read(&transaction->t_outstanding_credits) >
1645              journal->j_max_transaction_buffers) ||
1646             time_after_eq(jiffies, transaction->t_expires)) {
1647                 /* Do this even for aborted journals: an abort still
1648                  * completes the commit thread, it just doesn't write
1649                  * anything to disk. */
1650 
1651                 jbd_debug(2, "transaction too old, requesting commit for "
1652                                         "handle %p\n", handle);
1653                 /* This is non-blocking */
1654                 jbd2_log_start_commit(journal, transaction->t_tid);
1655 
1656                 /*
1657                  * Special case: JBD2_SYNC synchronous updates require us
1658                  * to wait for the commit to complete.
1659                  */
1660                 if (handle->h_sync && !(current->flags & PF_MEMALLOC))
1661                         wait_for_commit = 1;
1662         }
1663 
1664         /*
1665          * Once we drop t_updates, if it goes to zero the transaction
1666          * could start committing on us and eventually disappear.  So
1667          * once we do this, we must not dereference transaction
1668          * pointer again.
1669          */
1670         tid = transaction->t_tid;
1671         if (atomic_dec_and_test(&transaction->t_updates)) {
1672                 wake_up(&journal->j_wait_updates);
1673                 if (journal->j_barrier_count)
1674                         wake_up(&journal->j_wait_transaction_locked);
1675         }
1676 
1677         if (wait_for_commit)
1678                 err = jbd2_log_wait_commit(journal, tid);
1679 
1680         lock_map_release(&handle->h_lockdep_map);
1681 
1682         if (handle->h_rsv_handle)
1683                 jbd2_journal_free_reserved(handle->h_rsv_handle);
1684 free_and_exit:
1685         jbd2_free_handle(handle);
1686         return err;
1687 }
1688 
1689 /*
1690  *
1691  * List management code snippets: various functions for manipulating the
1692  * transaction buffer lists.
1693  *
1694  */
1695 
1696 /*
1697  * Append a buffer to a transaction list, given the transaction's list head
1698  * pointer.
1699  *
1700  * j_list_lock is held.
1701  *
1702  * jbd_lock_bh_state(jh2bh(jh)) is held.
1703  */
1704 
1705 static inline void
1706 __blist_add_buffer(struct journal_head **list, struct journal_head *jh)
1707 {
1708         if (!*list) {
1709                 jh->b_tnext = jh->b_tprev = jh;
1710                 *list = jh;
1711         } else {
1712                 /* Insert at the tail of the list to preserve order */
1713                 struct journal_head *first = *list, *last = first->b_tprev;
1714                 jh->b_tprev = last;
1715                 jh->b_tnext = first;
1716                 last->b_tnext = first->b_tprev = jh;
1717         }
1718 }
1719 
1720 /*
1721  * Remove a buffer from a transaction list, given the transaction's list
1722  * head pointer.
1723  *
1724  * Called with j_list_lock held, and the journal may not be locked.
1725  *
1726  * jbd_lock_bh_state(jh2bh(jh)) is held.
1727  */
1728 
1729 static inline void
1730 __blist_del_buffer(struct journal_head **list, struct journal_head *jh)
1731 {
1732         if (*list == jh) {
1733                 *list = jh->b_tnext;
1734                 if (*list == jh)
1735                         *list = NULL;
1736         }
1737         jh->b_tprev->b_tnext = jh->b_tnext;
1738         jh->b_tnext->b_tprev = jh->b_tprev;
1739 }
1740 
1741 /*
1742  * Remove a buffer from the appropriate transaction list.
1743  *
1744  * Note that this function can *change* the value of
1745  * bh->b_transaction->t_buffers, t_forget, t_shadow_list, t_log_list or
1746  * t_reserved_list.  If the caller is holding onto a copy of one of these
1747  * pointers, it could go bad.  Generally the caller needs to re-read the
1748  * pointer from the transaction_t.
1749  *
1750  * Called under j_list_lock.
1751  */
1752 static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh)
1753 {
1754         struct journal_head **list = NULL;
1755         transaction_t *transaction;
1756         struct buffer_head *bh = jh2bh(jh);
1757 
1758         J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
1759         transaction = jh->b_transaction;
1760         if (transaction)
1761                 assert_spin_locked(&transaction->t_journal->j_list_lock);
1762 
1763         J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
1764         if (jh->b_jlist != BJ_None)
1765                 J_ASSERT_JH(jh, transaction != NULL);
1766 
1767         switch (jh->b_jlist) {
1768         case BJ_None:
1769                 return;
1770         case BJ_Metadata:
1771                 transaction->t_nr_buffers--;
1772                 J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0);
1773                 list = &transaction->t_buffers;
1774                 break;
1775         case BJ_Forget:
1776                 list = &transaction->t_forget;
1777                 break;
1778         case BJ_Shadow:
1779                 list = &transaction->t_shadow_list;
1780                 break;
1781         case BJ_Reserved:
1782                 list = &transaction->t_reserved_list;
1783                 break;
1784         }
1785 
1786         __blist_del_buffer(list, jh);
1787         jh->b_jlist = BJ_None;
1788         if (transaction && is_journal_aborted(transaction->t_journal))
1789                 clear_buffer_jbddirty(bh);
1790         else if (test_clear_buffer_jbddirty(bh))
1791                 mark_buffer_dirty(bh);  /* Expose it to the VM */
1792 }
1793 
1794 /*
1795  * Remove buffer from all transactions.
1796  *
1797  * Called with bh_state lock and j_list_lock
1798  *
1799  * jh and bh may be already freed when this function returns.
1800  */
1801 static void __jbd2_journal_unfile_buffer(struct journal_head *jh)
1802 {
1803         __jbd2_journal_temp_unlink_buffer(jh);
1804         jh->b_transaction = NULL;
1805         jbd2_journal_put_journal_head(jh);
1806 }
1807 
1808 void jbd2_journal_unfile_buffer(journal_t *journal, struct journal_head *jh)
1809 {
1810         struct buffer_head *bh = jh2bh(jh);
1811 
1812         /* Get reference so that buffer cannot be freed before we unlock it */
1813         get_bh(bh);
1814         jbd_lock_bh_state(bh);
1815         spin_lock(&journal->j_list_lock);
1816         __jbd2_journal_unfile_buffer(jh);
1817         spin_unlock(&journal->j_list_lock);
1818         jbd_unlock_bh_state(bh);
1819         __brelse(bh);
1820 }
1821 
1822 /*
1823  * Called from jbd2_journal_try_to_free_buffers().
1824  *
1825  * Called under jbd_lock_bh_state(bh)
1826  */
1827 static void
1828 __journal_try_to_free_buffer(journal_t *journal, struct buffer_head *bh)
1829 {
1830         struct journal_head *jh;
1831 
1832         jh = bh2jh(bh);
1833 
1834         if (buffer_locked(bh) || buffer_dirty(bh))
1835                 goto out;
1836 
1837         if (jh->b_next_transaction != NULL)
1838                 goto out;
1839 
1840         spin_lock(&journal->j_list_lock);
1841         if (jh->b_cp_transaction != NULL && jh->b_transaction == NULL) {
1842                 /* written-back checkpointed metadata buffer */
1843                 JBUFFER_TRACE(jh, "remove from checkpoint list");
1844                 __jbd2_journal_remove_checkpoint(jh);
1845         }
1846         spin_unlock(&journal->j_list_lock);
1847 out:
1848         return;
1849 }
1850 
1851 /**
1852  * int jbd2_journal_try_to_free_buffers() - try to free page buffers.
1853  * @journal: journal for operation
1854  * @page: to try and free
1855  * @gfp_mask: we use the mask to detect how hard should we try to release
1856  * buffers. If __GFP_WAIT and __GFP_FS is set, we wait for commit code to
1857  * release the buffers.
1858  *
1859  *
1860  * For all the buffers on this page,
1861  * if they are fully written out ordered data, move them onto BUF_CLEAN
1862  * so try_to_free_buffers() can reap them.
1863  *
1864  * This function returns non-zero if we wish try_to_free_buffers()
1865  * to be called. We do this if the page is releasable by try_to_free_buffers().
1866  * We also do it if the page has locked or dirty buffers and the caller wants
1867  * us to perform sync or async writeout.
1868  *
1869  * This complicates JBD locking somewhat.  We aren't protected by the
1870  * BKL here.  We wish to remove the buffer from its committing or
1871  * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer.
1872  *
1873  * This may *change* the value of transaction_t->t_datalist, so anyone
1874  * who looks at t_datalist needs to lock against this function.
1875  *
1876  * Even worse, someone may be doing a jbd2_journal_dirty_data on this
1877  * buffer.  So we need to lock against that.  jbd2_journal_dirty_data()
1878  * will come out of the lock with the buffer dirty, which makes it
1879  * ineligible for release here.
1880  *
1881  * Who else is affected by this?  hmm...  Really the only contender
1882  * is do_get_write_access() - it could be looking at the buffer while
1883  * journal_try_to_free_buffer() is changing its state.  But that
1884  * cannot happen because we never reallocate freed data as metadata
1885  * while the data is part of a transaction.  Yes?
1886  *
1887  * Return 0 on failure, 1 on success
1888  */
1889 int jbd2_journal_try_to_free_buffers(journal_t *journal,
1890                                 struct page *page, gfp_t gfp_mask)
1891 {
1892         struct buffer_head *head;
1893         struct buffer_head *bh;
1894         int ret = 0;
1895 
1896         J_ASSERT(PageLocked(page));
1897 
1898         head = page_buffers(page);
1899         bh = head;
1900         do {
1901                 struct journal_head *jh;
1902 
1903                 /*
1904                  * We take our own ref against the journal_head here to avoid
1905                  * having to add tons of locking around each instance of
1906                  * jbd2_journal_put_journal_head().
1907                  */
1908                 jh = jbd2_journal_grab_journal_head(bh);
1909                 if (!jh)
1910                         continue;
1911 
1912                 jbd_lock_bh_state(bh);
1913                 __journal_try_to_free_buffer(journal, bh);
1914                 jbd2_journal_put_journal_head(jh);
1915                 jbd_unlock_bh_state(bh);
1916                 if (buffer_jbd(bh))
1917                         goto busy;
1918         } while ((bh = bh->b_this_page) != head);
1919 
1920         ret = try_to_free_buffers(page);
1921 
1922 busy:
1923         return ret;
1924 }
1925 
1926 /*
1927  * This buffer is no longer needed.  If it is on an older transaction's
1928  * checkpoint list we need to record it on this transaction's forget list
1929  * to pin this buffer (and hence its checkpointing transaction) down until
1930  * this transaction commits.  If the buffer isn't on a checkpoint list, we
1931  * release it.
1932  * Returns non-zero if JBD no longer has an interest in the buffer.
1933  *
1934  * Called under j_list_lock.
1935  *
1936  * Called under jbd_lock_bh_state(bh).
1937  */
1938 static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction)
1939 {
1940         int may_free = 1;
1941         struct buffer_head *bh = jh2bh(jh);
1942 
1943         if (jh->b_cp_transaction) {
1944                 JBUFFER_TRACE(jh, "on running+cp transaction");
1945                 __jbd2_journal_temp_unlink_buffer(jh);
1946                 /*
1947                  * We don't want to write the buffer anymore, clear the
1948                  * bit so that we don't confuse checks in
1949                  * __journal_file_buffer
1950                  */
1951                 clear_buffer_dirty(bh);
1952                 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
1953                 may_free = 0;
1954         } else {
1955                 JBUFFER_TRACE(jh, "on running transaction");
1956                 __jbd2_journal_unfile_buffer(jh);
1957         }
1958         return may_free;
1959 }
1960 
1961 /*
1962  * jbd2_journal_invalidatepage
1963  *
1964  * This code is tricky.  It has a number of cases to deal with.
1965  *
1966  * There are two invariants which this code relies on:
1967  *
1968  * i_size must be updated on disk before we start calling invalidatepage on the
1969  * data.
1970  *
1971  *  This is done in ext3 by defining an ext3_setattr method which
1972  *  updates i_size before truncate gets going.  By maintaining this
1973  *  invariant, we can be sure that it is safe to throw away any buffers
1974  *  attached to the current transaction: once the transaction commits,
1975  *  we know that the data will not be needed.
1976  *
1977  *  Note however that we can *not* throw away data belonging to the
1978  *  previous, committing transaction!
1979  *
1980  * Any disk blocks which *are* part of the previous, committing
1981  * transaction (and which therefore cannot be discarded immediately) are
1982  * not going to be reused in the new running transaction
1983  *
1984  *  The bitmap committed_data images guarantee this: any block which is
1985  *  allocated in one transaction and removed in the next will be marked
1986  *  as in-use in the committed_data bitmap, so cannot be reused until
1987  *  the next transaction to delete the block commits.  This means that
1988  *  leaving committing buffers dirty is quite safe: the disk blocks
1989  *  cannot be reallocated to a different file and so buffer aliasing is
1990  *  not possible.
1991  *
1992  *
1993  * The above applies mainly to ordered data mode.  In writeback mode we
1994  * don't make guarantees about the order in which data hits disk --- in
1995  * particular we don't guarantee that new dirty data is flushed before
1996  * transaction commit --- so it is always safe just to discard data
1997  * immediately in that mode.  --sct
1998  */
1999 
2000 /*
2001  * The journal_unmap_buffer helper function returns zero if the buffer
2002  * concerned remains pinned as an anonymous buffer belonging to an older
2003  * transaction.
2004  *
2005  * We're outside-transaction here.  Either or both of j_running_transaction
2006  * and j_committing_transaction may be NULL.
2007  */
2008 static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh,
2009                                 int partial_page)
2010 {
2011         transaction_t *transaction;
2012         struct journal_head *jh;
2013         int may_free = 1;
2014 
2015         BUFFER_TRACE(bh, "entry");
2016 
2017         /*
2018          * It is safe to proceed here without the j_list_lock because the
2019          * buffers cannot be stolen by try_to_free_buffers as long as we are
2020          * holding the page lock. --sct
2021          */
2022 
2023         if (!buffer_jbd(bh))
2024                 goto zap_buffer_unlocked;
2025 
2026         /* OK, we have data buffer in journaled mode */
2027         write_lock(&journal->j_state_lock);
2028         jbd_lock_bh_state(bh);
2029         spin_lock(&journal->j_list_lock);
2030 
2031         jh = jbd2_journal_grab_journal_head(bh);
2032         if (!jh)
2033                 goto zap_buffer_no_jh;
2034 
2035         /*
2036          * We cannot remove the buffer from checkpoint lists until the
2037          * transaction adding inode to orphan list (let's call it T)
2038          * is committed.  Otherwise if the transaction changing the
2039          * buffer would be cleaned from the journal before T is
2040          * committed, a crash will cause that the correct contents of
2041          * the buffer will be lost.  On the other hand we have to
2042          * clear the buffer dirty bit at latest at the moment when the
2043          * transaction marking the buffer as freed in the filesystem
2044          * structures is committed because from that moment on the
2045          * block can be reallocated and used by a different page.
2046          * Since the block hasn't been freed yet but the inode has
2047          * already been added to orphan list, it is safe for us to add
2048          * the buffer to BJ_Forget list of the newest transaction.
2049          *
2050          * Also we have to clear buffer_mapped flag of a truncated buffer
2051          * because the buffer_head may be attached to the page straddling
2052          * i_size (can happen only when blocksize < pagesize) and thus the
2053          * buffer_head can be reused when the file is extended again. So we end
2054          * up keeping around invalidated buffers attached to transactions'
2055          * BJ_Forget list just to stop checkpointing code from cleaning up
2056          * the transaction this buffer was modified in.
2057          */
2058         transaction = jh->b_transaction;
2059         if (transaction == NULL) {
2060                 /* First case: not on any transaction.  If it
2061                  * has no checkpoint link, then we can zap it:
2062                  * it's a writeback-mode buffer so we don't care
2063                  * if it hits disk safely. */
2064                 if (!jh->b_cp_transaction) {
2065                         JBUFFER_TRACE(jh, "not on any transaction: zap");
2066                         goto zap_buffer;
2067                 }
2068 
2069                 if (!buffer_dirty(bh)) {
2070                         /* bdflush has written it.  We can drop it now */
2071                         __jbd2_journal_remove_checkpoint(jh);
2072                         goto zap_buffer;
2073                 }
2074 
2075                 /* OK, it must be in the journal but still not
2076                  * written fully to disk: it's metadata or
2077                  * journaled data... */
2078 
2079                 if (journal->j_running_transaction) {
2080                         /* ... and once the current transaction has
2081                          * committed, the buffer won't be needed any
2082                          * longer. */
2083                         JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget");
2084                         may_free = __dispose_buffer(jh,
2085                                         journal->j_running_transaction);
2086                         goto zap_buffer;
2087                 } else {
2088                         /* There is no currently-running transaction. So the
2089                          * orphan record which we wrote for this file must have
2090                          * passed into commit.  We must attach this buffer to
2091                          * the committing transaction, if it exists. */
2092                         if (journal->j_committing_transaction) {
2093                                 JBUFFER_TRACE(jh, "give to committing trans");
2094                                 may_free = __dispose_buffer(jh,
2095                                         journal->j_committing_transaction);
2096                                 goto zap_buffer;
2097                         } else {
2098                                 /* The orphan record's transaction has
2099                                  * committed.  We can cleanse this buffer */
2100                                 clear_buffer_jbddirty(bh);
2101                                 __jbd2_journal_remove_checkpoint(jh);
2102                                 goto zap_buffer;
2103                         }
2104                 }
2105         } else if (transaction == journal->j_committing_transaction) {
2106                 JBUFFER_TRACE(jh, "on committing transaction");
2107                 /*
2108                  * The buffer is committing, we simply cannot touch
2109                  * it. If the page is straddling i_size we have to wait
2110                  * for commit and try again.
2111                  */
2112                 if (partial_page) {
2113                         jbd2_journal_put_journal_head(jh);
2114                         spin_unlock(&journal->j_list_lock);
2115                         jbd_unlock_bh_state(bh);
2116                         write_unlock(&journal->j_state_lock);
2117                         return -EBUSY;
2118                 }
2119                 /*
2120                  * OK, buffer won't be reachable after truncate. We just set
2121                  * j_next_transaction to the running transaction (if there is
2122                  * one) and mark buffer as freed so that commit code knows it
2123                  * should clear dirty bits when it is done with the buffer.
2124                  */
2125                 set_buffer_freed(bh);
2126                 if (journal->j_running_transaction && buffer_jbddirty(bh))
2127                         jh->b_next_transaction = journal->j_running_transaction;
2128                 jbd2_journal_put_journal_head(jh);
2129                 spin_unlock(&journal->j_list_lock);
2130                 jbd_unlock_bh_state(bh);
2131                 write_unlock(&journal->j_state_lock);
2132                 return 0;
2133         } else {
2134                 /* Good, the buffer belongs to the running transaction.
2135                  * We are writing our own transaction's data, not any
2136                  * previous one's, so it is safe to throw it away
2137                  * (remember that we expect the filesystem to have set
2138                  * i_size already for this truncate so recovery will not
2139                  * expose the disk blocks we are discarding here.) */
2140                 J_ASSERT_JH(jh, transaction == journal->j_running_transaction);
2141                 JBUFFER_TRACE(jh, "on running transaction");
2142                 may_free = __dispose_buffer(jh, transaction);
2143         }
2144 
2145 zap_buffer:
2146         /*
2147          * This is tricky. Although the buffer is truncated, it may be reused
2148          * if blocksize < pagesize and it is attached to the page straddling
2149          * EOF. Since the buffer might have been added to BJ_Forget list of the
2150          * running transaction, journal_get_write_access() won't clear
2151          * b_modified and credit accounting gets confused. So clear b_modified
2152          * here.
2153          */
2154         jh->b_modified = 0;
2155         jbd2_journal_put_journal_head(jh);
2156 zap_buffer_no_jh:
2157         spin_unlock(&journal->j_list_lock);
2158         jbd_unlock_bh_state(bh);
2159         write_unlock(&journal->j_state_lock);
2160 zap_buffer_unlocked:
2161         clear_buffer_dirty(bh);
2162         J_ASSERT_BH(bh, !buffer_jbddirty(bh));
2163         clear_buffer_mapped(bh);
2164         clear_buffer_req(bh);
2165         clear_buffer_new(bh);
2166         clear_buffer_delay(bh);
2167         clear_buffer_unwritten(bh);
2168         bh->b_bdev = NULL;
2169         return may_free;
2170 }
2171 
2172 /**
2173  * void jbd2_journal_invalidatepage()
2174  * @journal: journal to use for flush...
2175  * @page:    page to flush
2176  * @offset:  start of the range to invalidate
2177  * @length:  length of the range to invalidate
2178  *
2179  * Reap page buffers containing data after in the specified range in page.
2180  * Can return -EBUSY if buffers are part of the committing transaction and
2181  * the page is straddling i_size. Caller then has to wait for current commit
2182  * and try again.
2183  */
2184 int jbd2_journal_invalidatepage(journal_t *journal,
2185                                 struct page *page,
2186                                 unsigned int offset,
2187                                 unsigned int length)
2188 {
2189         struct buffer_head *head, *bh, *next;
2190         unsigned int stop = offset + length;
2191         unsigned int curr_off = 0;
2192         int partial_page = (offset || length < PAGE_CACHE_SIZE);
2193         int may_free = 1;
2194         int ret = 0;
2195 
2196         if (!PageLocked(page))
2197                 BUG();
2198         if (!page_has_buffers(page))
2199                 return 0;
2200 
2201         BUG_ON(stop > PAGE_CACHE_SIZE || stop < length);
2202 
2203         /* We will potentially be playing with lists other than just the
2204          * data lists (especially for journaled data mode), so be
2205          * cautious in our locking. */
2206 
2207         head = bh = page_buffers(page);
2208         do {
2209                 unsigned int next_off = curr_off + bh->b_size;
2210                 next = bh->b_this_page;
2211 
2212                 if (next_off > stop)
2213                         return 0;
2214 
2215                 if (offset <= curr_off) {
2216                         /* This block is wholly outside the truncation point */
2217                         lock_buffer(bh);
2218                         ret = journal_unmap_buffer(journal, bh, partial_page);
2219                         unlock_buffer(bh);
2220                         if (ret < 0)
2221                                 return ret;
2222                         may_free &= ret;
2223                 }
2224                 curr_off = next_off;
2225                 bh = next;
2226 
2227         } while (bh != head);
2228 
2229         if (!partial_page) {
2230                 if (may_free && try_to_free_buffers(page))
2231                         J_ASSERT(!page_has_buffers(page));
2232         }
2233         return 0;
2234 }
2235 
2236 /*
2237  * File a buffer on the given transaction list.
2238  */
2239 void __jbd2_journal_file_buffer(struct journal_head *jh,
2240                         transaction_t *transaction, int jlist)
2241 {
2242         struct journal_head **list = NULL;
2243         int was_dirty = 0;
2244         struct buffer_head *bh = jh2bh(jh);
2245 
2246         J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
2247         assert_spin_locked(&transaction->t_journal->j_list_lock);
2248 
2249         J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
2250         J_ASSERT_JH(jh, jh->b_transaction == transaction ||
2251                                 jh->b_transaction == NULL);
2252 
2253         if (jh->b_transaction && jh->b_jlist == jlist)
2254                 return;
2255 
2256         if (jlist == BJ_Metadata || jlist == BJ_Reserved ||
2257             jlist == BJ_Shadow || jlist == BJ_Forget) {
2258                 /*
2259                  * For metadata buffers, we track dirty bit in buffer_jbddirty
2260                  * instead of buffer_dirty. We should not see a dirty bit set
2261                  * here because we clear it in do_get_write_access but e.g.
2262                  * tune2fs can modify the sb and set the dirty bit at any time
2263                  * so we try to gracefully handle that.
2264                  */
2265                 if (buffer_dirty(bh))
2266                         warn_dirty_buffer(bh);
2267                 if (test_clear_buffer_dirty(bh) ||
2268                     test_clear_buffer_jbddirty(bh))
2269                         was_dirty = 1;
2270         }
2271 
2272         if (jh->b_transaction)
2273                 __jbd2_journal_temp_unlink_buffer(jh);
2274         else
2275                 jbd2_journal_grab_journal_head(bh);
2276         jh->b_transaction = transaction;
2277 
2278         switch (jlist) {
2279         case BJ_None:
2280                 J_ASSERT_JH(jh, !jh->b_committed_data);
2281                 J_ASSERT_JH(jh, !jh->b_frozen_data);
2282                 return;
2283         case BJ_Metadata:
2284                 transaction->t_nr_buffers++;
2285                 list = &transaction->t_buffers;
2286                 break;
2287         case BJ_Forget:
2288                 list = &transaction->t_forget;
2289                 break;
2290         case BJ_Shadow:
2291                 list = &transaction->t_shadow_list;
2292                 break;
2293         case BJ_Reserved:
2294                 list = &transaction->t_reserved_list;
2295                 break;
2296         }
2297 
2298         __blist_add_buffer(list, jh);
2299         jh->b_jlist = jlist;
2300 
2301         if (was_dirty)
2302                 set_buffer_jbddirty(bh);
2303 }
2304 
2305 void jbd2_journal_file_buffer(struct journal_head *jh,
2306                                 transaction_t *transaction, int jlist)
2307 {
2308         jbd_lock_bh_state(jh2bh(jh));
2309         spin_lock(&transaction->t_journal->j_list_lock);
2310         __jbd2_journal_file_buffer(jh, transaction, jlist);
2311         spin_unlock(&transaction->t_journal->j_list_lock);
2312         jbd_unlock_bh_state(jh2bh(jh));
2313 }
2314 
2315 /*
2316  * Remove a buffer from its current buffer list in preparation for
2317  * dropping it from its current transaction entirely.  If the buffer has
2318  * already started to be used by a subsequent transaction, refile the
2319  * buffer on that transaction's metadata list.
2320  *
2321  * Called under j_list_lock
2322  * Called under jbd_lock_bh_state(jh2bh(jh))
2323  *
2324  * jh and bh may be already free when this function returns
2325  */
2326 void __jbd2_journal_refile_buffer(struct journal_head *jh)
2327 {
2328         int was_dirty, jlist;
2329         struct buffer_head *bh = jh2bh(jh);
2330 
2331         J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
2332         if (jh->b_transaction)
2333                 assert_spin_locked(&jh->b_transaction->t_journal->j_list_lock);
2334 
2335         /* If the buffer is now unused, just drop it. */
2336         if (jh->b_next_transaction == NULL) {
2337                 __jbd2_journal_unfile_buffer(jh);
2338                 return;
2339         }
2340 
2341         /*
2342          * It has been modified by a later transaction: add it to the new
2343          * transaction's metadata list.
2344          */
2345 
2346         was_dirty = test_clear_buffer_jbddirty(bh);
2347         __jbd2_journal_temp_unlink_buffer(jh);
2348         /*
2349          * We set b_transaction here because b_next_transaction will inherit
2350          * our jh reference and thus __jbd2_journal_file_buffer() must not
2351          * take a new one.
2352          */
2353         jh->b_transaction = jh->b_next_transaction;
2354         jh->b_next_transaction = NULL;
2355         if (buffer_freed(bh))
2356                 jlist = BJ_Forget;
2357         else if (jh->b_modified)
2358                 jlist = BJ_Metadata;
2359         else
2360                 jlist = BJ_Reserved;
2361         __jbd2_journal_file_buffer(jh, jh->b_transaction, jlist);
2362         J_ASSERT_JH(jh, jh->b_transaction->t_state == T_RUNNING);
2363 
2364         if (was_dirty)
2365                 set_buffer_jbddirty(bh);
2366 }
2367 
2368 /*
2369  * __jbd2_journal_refile_buffer() with necessary locking added. We take our
2370  * bh reference so that we can safely unlock bh.
2371  *
2372  * The jh and bh may be freed by this call.
2373  */
2374 void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh)
2375 {
2376         struct buffer_head *bh = jh2bh(jh);
2377 
2378         /* Get reference so that buffer cannot be freed before we unlock it */
2379         get_bh(bh);
2380         jbd_lock_bh_state(bh);
2381         spin_lock(&journal->j_list_lock);
2382         __jbd2_journal_refile_buffer(jh);
2383         jbd_unlock_bh_state(bh);
2384         spin_unlock(&journal->j_list_lock);
2385         __brelse(bh);
2386 }
2387 
2388 /*
2389  * File inode in the inode list of the handle's transaction
2390  */
2391 int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode)
2392 {
2393         transaction_t *transaction = handle->h_transaction;
2394         journal_t *journal;
2395 
2396         if (is_handle_aborted(handle))
2397                 return -EROFS;
2398         journal = transaction->t_journal;
2399 
2400         jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode->i_vfs_inode->i_ino,
2401                         transaction->t_tid);
2402 
2403         /*
2404          * First check whether inode isn't already on the transaction's
2405          * lists without taking the lock. Note that this check is safe
2406          * without the lock as we cannot race with somebody removing inode
2407          * from the transaction. The reason is that we remove inode from the
2408          * transaction only in journal_release_jbd_inode() and when we commit
2409          * the transaction. We are guarded from the first case by holding
2410          * a reference to the inode. We are safe against the second case
2411          * because if jinode->i_transaction == transaction, commit code
2412          * cannot touch the transaction because we hold reference to it,
2413          * and if jinode->i_next_transaction == transaction, commit code
2414          * will only file the inode where we want it.
2415          */
2416         if (jinode->i_transaction == transaction ||
2417             jinode->i_next_transaction == transaction)
2418                 return 0;
2419 
2420         spin_lock(&journal->j_list_lock);
2421 
2422         if (jinode->i_transaction == transaction ||
2423             jinode->i_next_transaction == transaction)
2424                 goto done;
2425 
2426         /*
2427          * We only ever set this variable to 1 so the test is safe. Since
2428          * t_need_data_flush is likely to be set, we do the test to save some
2429          * cacheline bouncing
2430          */
2431         if (!transaction->t_need_data_flush)
2432                 transaction->t_need_data_flush = 1;
2433         /* On some different transaction's list - should be
2434          * the committing one */
2435         if (jinode->i_transaction) {
2436                 J_ASSERT(jinode->i_next_transaction == NULL);
2437                 J_ASSERT(jinode->i_transaction ==
2438                                         journal->j_committing_transaction);
2439                 jinode->i_next_transaction = transaction;
2440                 goto done;
2441         }
2442         /* Not on any transaction list... */
2443         J_ASSERT(!jinode->i_next_transaction);
2444         jinode->i_transaction = transaction;
2445         list_add(&jinode->i_list, &transaction->t_inode_list);
2446 done:
2447         spin_unlock(&journal->j_list_lock);
2448 
2449         return 0;
2450 }
2451 
2452 /*
2453  * File truncate and transaction commit interact with each other in a
2454  * non-trivial way.  If a transaction writing data block A is
2455  * committing, we cannot discard the data by truncate until we have
2456  * written them.  Otherwise if we crashed after the transaction with
2457  * write has committed but before the transaction with truncate has
2458  * committed, we could see stale data in block A.  This function is a
2459  * helper to solve this problem.  It starts writeout of the truncated
2460  * part in case it is in the committing transaction.
2461  *
2462  * Filesystem code must call this function when inode is journaled in
2463  * ordered mode before truncation happens and after the inode has been
2464  * placed on orphan list with the new inode size. The second condition
2465  * avoids the race that someone writes new data and we start
2466  * committing the transaction after this function has been called but
2467  * before a transaction for truncate is started (and furthermore it
2468  * allows us to optimize the case where the addition to orphan list
2469  * happens in the same transaction as write --- we don't have to write
2470  * any data in such case).
2471  */
2472 int jbd2_journal_begin_ordered_truncate(journal_t *journal,
2473                                         struct jbd2_inode *jinode,
2474                                         loff_t new_size)
2475 {
2476         transaction_t *inode_trans, *commit_trans;
2477         int ret = 0;
2478 
2479         /* This is a quick check to avoid locking if not necessary */
2480         if (!jinode->i_transaction)
2481                 goto out;
2482         /* Locks are here just to force reading of recent values, it is
2483          * enough that the transaction was not committing before we started
2484          * a transaction adding the inode to orphan list */
2485         read_lock(&journal->j_state_lock);
2486         commit_trans = journal->j_committing_transaction;
2487         read_unlock(&journal->j_state_lock);
2488         spin_lock(&journal->j_list_lock);
2489         inode_trans = jinode->i_transaction;
2490         spin_unlock(&journal->j_list_lock);
2491         if (inode_trans == commit_trans) {
2492                 ret = filemap_fdatawrite_range(jinode->i_vfs_inode->i_mapping,
2493                         new_size, LLONG_MAX);
2494                 if (ret)
2495                         jbd2_journal_abort(journal, ret);
2496         }
2497 out:
2498         return ret;
2499 }
2500 

~ [ 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