Branch data Line data Source code
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 : 0 : int __init jbd2_journal_init_transaction_cache(void)
40 : : {
41 [ # # ]: 0 : J_ASSERT(!transaction_cache);
42 : 0 : transaction_cache = kmem_cache_create("jbd2_transaction_s",
43 : : sizeof(transaction_t),
44 : : 0,
45 : : SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY,
46 : : NULL);
47 [ # # ]: 0 : if (transaction_cache)
48 : : return 0;
49 : 0 : return -ENOMEM;
50 : : }
51 : :
52 : 0 : void jbd2_journal_destroy_transaction_cache(void)
53 : : {
54 [ # # ]: 0 : if (transaction_cache) {
55 : 0 : kmem_cache_destroy(transaction_cache);
56 : 0 : transaction_cache = NULL;
57 : : }
58 : 0 : }
59 : :
60 : 0 : void jbd2_journal_free_transaction(transaction_t *transaction)
61 : : {
62 [ # # + + ]: 8749372 : if (unlikely(ZERO_OR_NULL_PTR(transaction)))
[ + - ]
63 : 0 : return;
64 : 20682 : 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 : 0 : jbd2_get_transaction(journal_t *journal, transaction_t *transaction)
84 : : {
85 : 19773 : transaction->t_journal = journal;
86 : 19773 : transaction->t_state = T_RUNNING;
87 : 19773 : transaction->t_start_time = ktime_get();
88 : 19773 : transaction->t_tid = journal->j_transaction_sequence++;
89 : 19773 : transaction->t_expires = jiffies + journal->j_commit_interval;
90 : 19773 : spin_lock_init(&transaction->t_handle_lock);
91 : 19773 : atomic_set(&transaction->t_updates, 0);
92 : 19773 : atomic_set(&transaction->t_outstanding_credits,
93 : : atomic_read(&journal->j_reserved_credits));
94 : 19773 : atomic_set(&transaction->t_handle_count, 0);
95 : 19773 : INIT_LIST_HEAD(&transaction->t_inode_list);
96 : 19773 : INIT_LIST_HEAD(&transaction->t_private_list);
97 : :
98 : : /* Set up the commit timer for the new transaction. */
99 : 19773 : journal->j_commit_timer.expires = round_jiffies_up(transaction->t_expires);
100 : 19773 : add_timer(&journal->j_commit_timer);
101 : :
102 [ - + ]: 19773 : J_ASSERT(journal->j_running_transaction == NULL);
103 : 19773 : journal->j_running_transaction = transaction;
104 : 19773 : transaction->t_max_wait = 0;
105 : 19773 : transaction->t_start = jiffies;
106 : 19773 : transaction->t_requested = 0;
107 : :
108 : 19773 : 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 : 0 : static void wait_transaction_locked(journal_t *journal)
150 : : __releases(journal->j_state_lock)
151 : : {
152 : 996 : DEFINE_WAIT(wait);
153 : : int need_to_start;
154 : 498 : tid_t tid = journal->j_running_transaction->t_tid;
155 : :
156 : 498 : prepare_to_wait(&journal->j_wait_transaction_locked, &wait,
157 : : TASK_UNINTERRUPTIBLE);
158 : 498 : need_to_start = !tid_geq(journal->j_commit_request, tid);
159 : : read_unlock(&journal->j_state_lock);
160 [ - + ]: 498 : if (need_to_start)
161 : 0 : jbd2_log_start_commit(journal, tid);
162 : 498 : schedule();
163 : 497 : finish_wait(&journal->j_wait_transaction_locked, &wait);
164 : 496 : }
165 : :
166 : 0 : static void sub_reserved_credits(journal_t *journal, int blocks)
167 : : {
168 : 0 : atomic_sub(blocks, &journal->j_reserved_credits);
169 : 0 : wake_up(&journal->j_wait_reserved);
170 : 0 : }
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 : 0 : static int add_transaction_credits(journal_t *journal, int blocks,
179 : : int rsv_blocks)
180 : : {
181 : 8725629 : transaction_t *t = journal->j_running_transaction;
182 : : int needed;
183 : 8725629 : 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 [ + + ]: 8725629 : if (t->t_state == T_LOCKED) {
190 : 498 : wait_transaction_locked(journal);
191 : 496 : 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 : 8725131 : needed = atomic_add_return(total, &t->t_outstanding_credits);
200 [ - + ]: 8723758 : 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 : 0 : wait_transaction_locked(journal);
208 : 0 : 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 [ - + ]: 8723758 : 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 : 0 : write_lock(&journal->j_state_lock);
226 [ # # ]: 0 : if (jbd2_log_space_left(journal) < jbd2_space_needed(journal))
227 : 0 : __jbd2_log_wait_for_space(journal);
228 : : write_unlock(&journal->j_state_lock);
229 : 0 : return 1;
230 : : }
231 : :
232 : : /* No reservation? We are done... */
233 [ - + ]: 8723758 : if (!rsv_blocks)
234 : : return 0;
235 : :
236 : 0 : needed = atomic_add_return(rsv_blocks, &journal->j_reserved_credits);
237 : : /* We allow at most half of a transaction to be reserved */
238 [ # # ]: 0 : if (needed > journal->j_max_transaction_buffers / 2) {
239 : 0 : sub_reserved_credits(journal, rsv_blocks);
240 : : atomic_sub(total, &t->t_outstanding_credits);
241 : : read_unlock(&journal->j_state_lock);
242 [ # # ][ # # ]: 0 : 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 : 0 : 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 : 8727743 : int blocks = handle->h_buffer_credits;
262 : : int rsv_blocks = 0;
263 : 8727743 : 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 [ - + ][ - + ]: 8727743 : if (WARN_ON(blocks > journal->j_max_transaction_buffers / 2)) {
270 : 0 : printk(KERN_ERR "JBD2: %s wants too many credits (%d > %d)\n",
271 : 0 : current->comm, blocks,
272 : 0 : journal->j_max_transaction_buffers / 2);
273 : 0 : return -ENOSPC;
274 : : }
275 : :
276 [ - + ]: 8727358 : if (handle->h_rsv_handle)
277 : 8727358 : rsv_blocks = handle->h_rsv_handle->h_buffer_credits;
278 : :
279 : : alloc_transaction:
280 [ + + ]: 8727837 : if (!journal->j_running_transaction) {
281 : 19981 : new_transaction = kmem_cache_zalloc(transaction_cache,
282 : : gfp_mask);
283 [ + - ]: 8728241 : 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 [ # # ]: 0 : if ((gfp_mask & __GFP_FS) == 0) {
292 : 20995 : congestion_wait(BLK_RW_ASYNC, HZ/50);
293 : 479 : 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 : 8749236 : read_lock(&journal->j_state_lock);
307 [ - + ]: 8748186 : BUG_ON(journal->j_flags & JBD2_UNMOUNT);
308 [ + ][ - + ]: 8748186 : if (is_journal_aborted(journal) ||
309 [ # # ]: 0 : (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 [ + ][ - + ]: 8746297 : if (!handle->h_reserved && journal->j_barrier_count) {
321 : : read_unlock(&journal->j_state_lock);
322 [ # # ][ # # ]: 0 : wait_event(journal->j_wait_transaction_locked,
323 : : journal->j_barrier_count == 0);
324 : : goto repeat;
325 : : }
326 : :
327 [ + + ]: 8746297 : if (!journal->j_running_transaction) {
328 : : read_unlock(&journal->j_state_lock);
329 [ + + ]: 20979 : if (!new_transaction)
330 : : goto alloc_transaction;
331 : 20500 : write_lock(&journal->j_state_lock);
332 [ + + ][ + - ]: 20500 : if (!journal->j_running_transaction &&
333 [ + - ]: 19773 : (handle->h_reserved || !journal->j_barrier_count)) {
334 : 19773 : 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 [ + - ]: 8725320 : if (!handle->h_reserved) {
344 : : /* We may have dropped j_state_lock - restart in that case */
345 [ + + ]: 8725320 : 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 : 0 : sub_reserved_credits(journal, blocks);
354 : 0 : 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 : 8723842 : handle->h_transaction = transaction;
362 : 8723842 : handle->h_requested_credits = blocks;
363 : 8723842 : handle->h_start_jiffies = jiffies;
364 : 8723842 : atomic_inc(&transaction->t_updates);
365 : 8727611 : 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 : 8729600 : 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 : 0 : static handle_t *new_handle(int nblocks)
382 : : {
383 : : handle_t *handle = jbd2_alloc_handle(GFP_NOFS);
384 [ + ]: 8728185 : if (!handle)
385 : : return NULL;
386 : 8728217 : handle->h_buffer_credits = nblocks;
387 : 8728217 : handle->h_ref = 1;
388 : :
389 : : lockdep_init_map(&handle->h_lockdep_map, "jbd2_handle",
390 : : &jbd2_handle_key, 0);
391 : :
392 : 8728217 : 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 : 0 : 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 [ + ]: 12842680 : if (!journal)
422 : : return ERR_PTR(-EROFS);
423 : :
424 [ + + ]: 12845977 : if (handle) {
425 [ - + ]: 4117151 : J_ASSERT(handle->h_transaction->t_journal == journal);
426 : 4117151 : handle->h_ref++;
427 : 4117151 : return handle;
428 : : }
429 : :
430 : 8728826 : handle = new_handle(nblocks);
431 [ + ]: 8726232 : if (!handle)
432 : : return ERR_PTR(-ENOMEM);
433 [ - + ]: 8728849 : if (rsv_blocks) {
434 : : handle_t *rsv_handle;
435 : :
436 : 0 : rsv_handle = new_handle(rsv_blocks);
437 [ # # ]: 0 : if (!rsv_handle) {
438 : : jbd2_free_handle(handle);
439 : 0 : return ERR_PTR(-ENOMEM);
440 : : }
441 : 0 : rsv_handle->h_reserved = 1;
442 : 0 : rsv_handle->h_journal = journal;
443 : 0 : handle->h_rsv_handle = rsv_handle;
444 : : }
445 : :
446 : 8728849 : err = start_this_handle(journal, handle, gfp_mask);
447 [ - + ]: 8729216 : if (err < 0) {
448 [ # # ]: 0 : if (handle->h_rsv_handle)
449 : : jbd2_free_handle(handle->h_rsv_handle);
450 : : jbd2_free_handle(handle);
451 : 0 : return ERR_PTR(err);
452 : : }
453 : 8729216 : handle->h_type = type;
454 : 8729216 : handle->h_line_no = line_no;
455 : 8729216 : trace_jbd2_handle_start(journal->j_fs_dev->bd_dev,
456 : 8729216 : handle->h_transaction->t_tid, type,
457 : : line_no, nblocks);
458 : 8729218 : return handle;
459 : : }
460 : : EXPORT_SYMBOL(jbd2__journal_start);
461 : :
462 : :
463 : 0 : handle_t *jbd2_journal_start(journal_t *journal, int nblocks)
464 : : {
465 : 0 : return jbd2__journal_start(journal, nblocks, 0, GFP_NOFS, 0, 0);
466 : : }
467 : : EXPORT_SYMBOL(jbd2_journal_start);
468 : :
469 : 0 : void jbd2_journal_free_reserved(handle_t *handle)
470 : : {
471 : 0 : journal_t *journal = handle->h_journal;
472 : :
473 [ # # ]: 0 : WARN_ON(!handle->h_reserved);
474 : 0 : sub_reserved_credits(journal, handle->h_buffer_credits);
475 : : jbd2_free_handle(handle);
476 : 0 : }
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 : 0 : int jbd2_journal_start_reserved(handle_t *handle, unsigned int type,
492 : : unsigned int line_no)
493 : : {
494 : 0 : journal_t *journal = handle->h_journal;
495 : : int ret = -EIO;
496 : :
497 [ # # ][ # # ]: 0 : if (WARN_ON(!handle->h_reserved)) {
498 : : /* Someone passed in normal handle? Just stop it. */
499 : 0 : jbd2_journal_stop(handle);
500 : 0 : 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 [ # # ][ # # ]: 0 : if (WARN_ON(current->journal_info)) {
507 : 0 : jbd2_journal_free_reserved(handle);
508 : 0 : return ret;
509 : : }
510 : :
511 : 0 : handle->h_journal = NULL;
512 : : /*
513 : : * GFP_NOFS is here because callers are likely from writeback or
514 : : * similarly constrained call sites
515 : : */
516 : 0 : ret = start_this_handle(journal, handle, GFP_NOFS);
517 [ # # ]: 0 : if (ret < 0)
518 : 0 : jbd2_journal_free_reserved(handle);
519 : 0 : handle->h_type = type;
520 : 0 : handle->h_line_no = line_no;
521 : 0 : return ret;
522 : : }
523 : : EXPORT_SYMBOL(jbd2_journal_start_reserved);
524 : :
525 : : /**
526 : : * int jbd2_journal_extend() - extend buffer credits.
527 : : * @handle: handle to 'extend'
528 : : * @nblocks: nr blocks to try to extend by.
529 : : *
530 : : * Some transactions, such as large extends and truncates, can be done
531 : : * atomically all at once or in several stages. The operation requests
532 : : * a credit for a number of buffer modications in advance, but can
533 : : * extend its credit if it needs more.
534 : : *
535 : : * jbd2_journal_extend tries to give the running handle more buffer credits.
536 : : * It does not guarantee that allocation - this is a best-effort only.
537 : : * The calling process MUST be able to deal cleanly with a failure to
538 : : * extend here.
539 : : *
540 : : * Return 0 on success, non-zero on failure.
541 : : *
542 : : * return code < 0 implies an error
543 : : * return code > 0 implies normal transaction-full status.
544 : : */
545 : 0 : int jbd2_journal_extend(handle_t *handle, int nblocks)
546 : : {
547 : 3387 : transaction_t *transaction = handle->h_transaction;
548 : 3383 : journal_t *journal;
549 : : int result;
550 : : int wanted;
551 : :
552 [ - + ]: 3387 : WARN_ON(!transaction);
553 [ + - ]: 3387 : if (is_handle_aborted(handle))
554 : : return -EROFS;
555 : 3387 : journal = transaction->t_journal;
556 : :
557 : : result = 1;
558 : :
559 : 3387 : read_lock(&journal->j_state_lock);
560 : :
561 : : /* Don't extend a locked-down transaction! */
562 [ + + ]: 3387 : if (transaction->t_state != T_RUNNING) {
563 : : jbd_debug(3, "denied handle %p %d blocks: "
564 : : "transaction not running\n", handle, nblocks);
565 : : goto error_out;
566 : : }
567 : :
568 : : spin_lock(&transaction->t_handle_lock);
569 : 3383 : wanted = atomic_add_return(nblocks,
570 : : &transaction->t_outstanding_credits);
571 : :
572 [ - + ]: 6770 : if (wanted > journal->j_max_transaction_buffers) {
573 : : jbd_debug(3, "denied handle %p %d blocks: "
574 : : "transaction too large\n", handle, nblocks);
575 : : atomic_sub(nblocks, &transaction->t_outstanding_credits);
576 : : goto unlock;
577 : : }
578 : :
579 [ - + ]: 3383 : if (wanted + (wanted >> JBD2_CONTROL_BLOCKS_SHIFT) >
580 : : jbd2_log_space_left(journal)) {
581 : : jbd_debug(3, "denied handle %p %d blocks: "
582 : : "insufficient log space\n", handle, nblocks);
583 : : atomic_sub(nblocks, &transaction->t_outstanding_credits);
584 : : goto unlock;
585 : : }
586 : :
587 : 10149 : trace_jbd2_handle_extend(journal->j_fs_dev->bd_dev,
588 : 3383 : transaction->t_tid,
589 : 6766 : handle->h_type, handle->h_line_no,
590 : : handle->h_buffer_credits,
591 : : nblocks);
592 : :
593 : 3383 : handle->h_buffer_credits += nblocks;
594 : 3383 : handle->h_requested_credits += nblocks;
595 : : result = 0;
596 : :
597 : : jbd_debug(3, "extended handle %p by %d\n", handle, nblocks);
598 : : unlock:
599 : : spin_unlock(&transaction->t_handle_lock);
600 : : error_out:
601 : : read_unlock(&journal->j_state_lock);
602 : 3387 : return result;
603 : : }
604 : :
605 : :
606 : : /**
607 : : * int jbd2_journal_restart() - restart a handle .
608 : : * @handle: handle to restart
609 : : * @nblocks: nr credits requested
610 : : *
611 : : * Restart a handle for a multi-transaction filesystem
612 : : * operation.
613 : : *
614 : : * If the jbd2_journal_extend() call above fails to grant new buffer credits
615 : : * to a running handle, a call to jbd2_journal_restart will commit the
616 : : * handle's transaction so far and reattach the handle to a new
617 : : * transaction capabable of guaranteeing the requested number of
618 : : * credits. We preserve reserved handle if there's any attached to the
619 : : * passed in handle.
620 : : */
621 : 0 : int jbd2__journal_restart(handle_t *handle, int nblocks, gfp_t gfp_mask)
622 : : {
623 : 3 : transaction_t *transaction = handle->h_transaction;
624 : : journal_t *journal;
625 : : tid_t tid;
626 : : int need_to_start, ret;
627 : :
628 [ - + ]: 3 : WARN_ON(!transaction);
629 : : /* If we've had an abort of any type, don't even think about
630 : : * actually doing the restart! */
631 [ + - ]: 3 : if (is_handle_aborted(handle))
632 : : return 0;
633 : 3 : journal = transaction->t_journal;
634 : :
635 : : /*
636 : : * First unlink the handle from its current transaction, and start the
637 : : * commit on that.
638 : : */
639 [ - + ]: 3 : J_ASSERT(atomic_read(&transaction->t_updates) > 0);
640 [ - + ]: 3 : J_ASSERT(journal_current_handle() == handle);
641 : :
642 : 3 : read_lock(&journal->j_state_lock);
643 : : spin_lock(&transaction->t_handle_lock);
644 : 3 : atomic_sub(handle->h_buffer_credits,
645 : : &transaction->t_outstanding_credits);
646 [ - + ]: 3 : if (handle->h_rsv_handle) {
647 : 0 : sub_reserved_credits(journal,
648 : : handle->h_rsv_handle->h_buffer_credits);
649 : : }
650 [ + - ]: 3 : if (atomic_dec_and_test(&transaction->t_updates))
651 : 3 : wake_up(&journal->j_wait_updates);
652 : 3 : tid = transaction->t_tid;
653 : : spin_unlock(&transaction->t_handle_lock);
654 : 3 : handle->h_transaction = NULL;
655 : 3 : current->journal_info = NULL;
656 : :
657 : : jbd_debug(2, "restarting handle %p\n", handle);
658 : 3 : need_to_start = !tid_geq(journal->j_commit_request, tid);
659 : : read_unlock(&journal->j_state_lock);
660 [ - + ]: 3 : if (need_to_start)
661 : 0 : jbd2_log_start_commit(journal, tid);
662 : :
663 : : lock_map_release(&handle->h_lockdep_map);
664 : 3 : handle->h_buffer_credits = nblocks;
665 : 3 : ret = start_this_handle(journal, handle, gfp_mask);
666 : 3 : return ret;
667 : : }
668 : : EXPORT_SYMBOL(jbd2__journal_restart);
669 : :
670 : :
671 : 0 : int jbd2_journal_restart(handle_t *handle, int nblocks)
672 : : {
673 : 3 : 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 : 0 : void jbd2_journal_lock_updates(journal_t *journal)
688 : : {
689 : 0 : DEFINE_WAIT(wait);
690 : :
691 : 0 : write_lock(&journal->j_state_lock);
692 : 0 : ++journal->j_barrier_count;
693 : :
694 : : /* Wait until there are no reserved handles */
695 [ # # ]: 0 : if (atomic_read(&journal->j_reserved_credits)) {
696 : : write_unlock(&journal->j_state_lock);
697 [ # # ][ # # ]: 0 : wait_event(journal->j_wait_reserved,
698 : : atomic_read(&journal->j_reserved_credits) == 0);
699 : 0 : write_lock(&journal->j_state_lock);
700 : : }
701 : :
702 : : /* Wait until there are no running updates */
703 : : while (1) {
704 : 0 : transaction_t *transaction = journal->j_running_transaction;
705 : :
706 [ # # ]: 0 : if (!transaction)
707 : : break;
708 : :
709 : : spin_lock(&transaction->t_handle_lock);
710 : 0 : prepare_to_wait(&journal->j_wait_updates, &wait,
711 : : TASK_UNINTERRUPTIBLE);
712 [ # # ]: 0 : if (!atomic_read(&transaction->t_updates)) {
713 : : spin_unlock(&transaction->t_handle_lock);
714 : 0 : finish_wait(&journal->j_wait_updates, &wait);
715 : 0 : break;
716 : : }
717 : : spin_unlock(&transaction->t_handle_lock);
718 : : write_unlock(&journal->j_state_lock);
719 : 0 : schedule();
720 : 0 : finish_wait(&journal->j_wait_updates, &wait);
721 : 0 : write_lock(&journal->j_state_lock);
722 : 0 : }
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 : 0 : mutex_lock(&journal->j_barrier);
732 : 0 : }
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 : 0 : void jbd2_journal_unlock_updates (journal_t *journal)
743 : : {
744 [ # # ]: 0 : J_ASSERT(journal->j_barrier_count != 0);
745 : :
746 : 0 : mutex_unlock(&journal->j_barrier);
747 : 0 : write_lock(&journal->j_state_lock);
748 : 0 : --journal->j_barrier_count;
749 : : write_unlock(&journal->j_state_lock);
750 : 0 : wake_up(&journal->j_wait_transaction_locked);
751 : 0 : }
752 : :
753 : 0 : static void warn_dirty_buffer(struct buffer_head *bh)
754 : : {
755 : : char b[BDEVNAME_SIZE];
756 : :
757 : 0 : 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 : 0 : }
763 : :
764 : 0 : static int sleep_on_shadow_bh(void *word)
765 : : {
766 : 1508 : io_schedule();
767 : 1504 : 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 : 0 : do_get_write_access(handle_t *handle, struct journal_head *jh,
782 : : int force_copy)
783 : : {
784 : 0 : struct buffer_head *bh;
785 : 13116093 : 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 [ - + ]: 13116093 : WARN_ON(!transaction);
793 [ + - ]: 13121201 : if (is_handle_aborted(handle))
794 : : return -EROFS;
795 : 13123530 : journal = transaction->t_journal;
796 : :
797 : : jbd_debug(5, "journal_head %p, force_copy %d\n", jh, force_copy);
798 : :
799 : : JBUFFER_TRACE(jh, "entry");
800 : : repeat:
801 : : bh = jh2bh(jh);
802 : :
803 : : /* @@@ Need to check for errors here at some point. */
804 : :
805 : 13123530 : start_lock = jiffies;
806 : : lock_buffer(bh);
807 : : jbd_lock_bh_state(bh);
808 : :
809 : : /* If it takes too long to lock the buffer, trace it */
810 : 13119299 : time_lock = jbd2_time_diff(start_lock, jiffies);
811 [ + + ]: 13119299 : if (time_lock > HZ/10)
812 : 113 : trace_jbd2_lock_buffer_stall(bh->b_bdev->bd_dev,
813 : 113 : jiffies_to_msecs(time_lock));
814 : :
815 : : /* We now hold the buffer lock so it is safe to query the buffer
816 : : * state. Is the buffer dirty?
817 : : *
818 : : * If so, there are two possibilities. The buffer may be
819 : : * non-journaled, and undergoing a quite legitimate writeback.
820 : : * Otherwise, it is journaled, and we don't expect dirty buffers
821 : : * in that state (the buffers should be marked JBD_Dirty
822 : : * instead.) So either the IO is being done under our own
823 : : * control and this is a bug, or it's a third party IO such as
824 : : * dump(8) (which may leave the buffer scheduled for read ---
825 : : * ie. locked but not dirty) or tune2fs (which may actually have
826 : : * the buffer dirtied, ugh.) */
827 : :
828 [ + + ]: 13119487 : if (buffer_dirty(bh)) {
829 : : /*
830 : : * First question: is this buffer already part of the current
831 : : * transaction or the existing committing transaction?
832 : : */
833 [ - + ]: 43039 : if (jh->b_transaction) {
834 [ # # ][ # # ]: 0 : J_ASSERT_JH(jh,
835 : : jh->b_transaction == transaction ||
836 : : jh->b_transaction ==
837 : : journal->j_committing_transaction);
838 [ # # ]: 0 : if (jh->b_next_transaction)
839 [ # # ]: 0 : J_ASSERT_JH(jh, jh->b_next_transaction ==
840 : : transaction);
841 : 0 : warn_dirty_buffer(bh);
842 : : }
843 : : /*
844 : : * In any case we need to clean the dirty flag and we must
845 : : * do it under the buffer lock to be sure we don't race
846 : : * with running write-out.
847 : : */
848 : : JBUFFER_TRACE(jh, "Journalling dirty buffer");
849 : : clear_buffer_dirty(bh);
850 : : set_buffer_jbddirty(bh);
851 : : }
852 : :
853 : 13119481 : unlock_buffer(bh);
854 : :
855 : : error = -EROFS;
856 [ - + ]: 13121987 : if (is_handle_aborted(handle)) {
857 : : jbd_unlock_bh_state(bh);
858 : : goto out;
859 : : }
860 : : error = 0;
861 : :
862 : : /*
863 : : * The buffer is already part of this transaction if b_transaction or
864 : : * b_next_transaction points to it
865 : : */
866 [ + + ][ + + ]: 13121987 : if (jh->b_transaction == transaction ||
867 : 305296 : jh->b_next_transaction == transaction)
868 : : goto done;
869 : :
870 : : /*
871 : : * this is the first time this transaction is touching this buffer,
872 : : * reset the modified flag
873 : : */
874 : 77256 : jh->b_modified = 0;
875 : :
876 : : /*
877 : : * If there is already a copy-out version of this buffer, then we don't
878 : : * need to make another one
879 : : */
880 [ - + ]: 77256 : if (jh->b_frozen_data) {
881 : : JBUFFER_TRACE(jh, "has frozen data");
882 [ # # ]: 0 : J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
883 : 0 : jh->b_next_transaction = transaction;
884 : 0 : goto done;
885 : : }
886 : :
887 : : /* Is there data here we need to preserve? */
888 : :
889 [ + + ][ + ]: 77256 : if (jh->b_transaction && jh->b_transaction != transaction) {
890 : : JBUFFER_TRACE(jh, "owned by older transaction");
891 [ - + ]: 13123689 : J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
892 [ - + ]: 13123689 : J_ASSERT_JH(jh, jh->b_transaction ==
893 : : journal->j_committing_transaction);
894 : :
895 : : /* There is one case we have to be very careful about.
896 : : * If the committing transaction is currently writing
897 : : * this buffer out to disk and has NOT made a copy-out,
898 : : * then we cannot modify the buffer contents at all
899 : : * right now. The essence of copy-out is that it is the
900 : : * extra copy, not the primary copy, which gets
901 : : * journaled. If the primary copy is already going to
902 : : * disk then we cannot do copy-out here. */
903 : :
904 [ + + ]: 7596 : if (buffer_shadow(bh)) {
905 : : JBUFFER_TRACE(jh, "on shadow: sleep");
906 : : jbd_unlock_bh_state(bh);
907 : : wait_on_bit(&bh->b_state, BH_Shadow,
908 : : sleep_on_shadow_bh, TASK_UNINTERRUPTIBLE);
909 : : goto repeat;
910 : : }
911 : :
912 : : /*
913 : : * Only do the copy if the currently-owning transaction still
914 : : * needs it. If buffer isn't on BJ_Metadata list, the
915 : : * committing transaction is past that stage (here we use the
916 : : * fact that BH_Shadow is set under bh_state lock together with
917 : : * refiling to BJ_Shadow list and at this point we know the
918 : : * buffer doesn't have BH_Shadow set).
919 : : *
920 : : * Subtle point, though: if this is a get_undo_access,
921 : : * then we will be relying on the frozen_data to contain
922 : : * the new value of the committed_data record after the
923 : : * transaction, so we HAVE to force the frozen_data copy
924 : : * in that case.
925 : : */
926 [ + + ][ + ]: 6087 : if (jh->b_jlist == BJ_Metadata || force_copy) {
927 : : JBUFFER_TRACE(jh, "generate frozen data");
928 [ + + ]: 1632 : if (!frozen_buffer) {
929 : : JBUFFER_TRACE(jh, "allocate memory for buffer");
930 : : jbd_unlock_bh_state(bh);
931 : 823 : frozen_buffer =
932 : 823 : jbd2_alloc(jh2bh(jh)->b_size,
933 : : GFP_NOFS);
934 [ + - ]: 823 : if (!frozen_buffer) {
935 : 0 : printk(KERN_ERR
936 : : "%s: OOM for frozen_buffer\n",
937 : : __func__);
938 : : JBUFFER_TRACE(jh, "oom!");
939 : : error = -ENOMEM;
940 : : jbd_lock_bh_state(bh);
941 : : goto done;
942 : : }
943 : : goto repeat;
944 : : }
945 : 809 : jh->b_frozen_data = frozen_buffer;
946 : : frozen_buffer = NULL;
947 : : need_copy = 1;
948 : : }
949 : 5264 : jh->b_next_transaction = transaction;
950 : : }
951 : :
952 : :
953 : : /*
954 : : * Finally, if the buffer is not journaled right now, we need to make
955 : : * sure it doesn't get written to disk before the caller actually
956 : : * commits the new data
957 : : */
958 [ + ]: 0 : if (!jh->b_transaction) {
959 : : JBUFFER_TRACE(jh, "no transaction");
960 [ - + ]: 69677 : J_ASSERT_JH(jh, !jh->b_next_transaction);
961 : : JBUFFER_TRACE(jh, "file as BJ_Reserved");
962 : : spin_lock(&journal->j_list_lock);
963 : 69680 : __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
964 : : spin_unlock(&journal->j_list_lock);
965 : : }
966 : :
967 : : done:
968 [ + + ]: 13119167 : if (need_copy) {
969 : : struct page *page;
970 : : int offset;
971 : : char *source;
972 : :
973 [ - + ]: 809 : J_EXPECT_JH(jh, buffer_uptodate(jh2bh(jh)),
974 : : "Possible IO failure.\n");
975 : 809 : page = jh2bh(jh)->b_page;
976 : 809 : offset = offset_in_page(jh2bh(jh)->b_data);
977 : 809 : source = kmap_atomic(page);
978 : : /* Fire data frozen trigger just before we copy the data */
979 : 809 : jbd2_buffer_frozen_trigger(jh, source + offset,
980 : : jh->b_triggers);
981 : 1618 : memcpy(jh->b_frozen_data, source+offset, jh2bh(jh)->b_size);
982 : 809 : kunmap_atomic(source);
983 : :
984 : : /*
985 : : * Now that the frozen data is saved off, we need to store
986 : : * any matching triggers.
987 : : */
988 : 808 : jh->b_frozen_triggers = jh->b_triggers;
989 : : }
990 : : jbd_unlock_bh_state(bh);
991 : :
992 : : /*
993 : : * If we are about to journal a buffer, then any revoke pending on it is
994 : : * no longer valid
995 : : */
996 : 13115979 : jbd2_journal_cancel_revoke(handle, jh);
997 : :
998 : : out:
999 [ + + ]: 13117189 : if (unlikely(frozen_buffer)) /* It's usually NULL */
1000 : 14 : jbd2_free(frozen_buffer, bh->b_size);
1001 : :
1002 : : JBUFFER_TRACE(jh, "exit");
1003 : 13117189 : return error;
1004 : : }
1005 : :
1006 : : /**
1007 : : * int jbd2_journal_get_write_access() - notify intent to modify a buffer for metadata (not data) update.
1008 : : * @handle: transaction to add buffer modifications to
1009 : : * @bh: bh to be used for metadata writes
1010 : : *
1011 : : * Returns an error code or 0 on success.
1012 : : *
1013 : : * In full data journalling mode the buffer may be of type BJ_AsyncData,
1014 : : * because we're write()ing a buffer which is also part of a shared mapping.
1015 : : */
1016 : :
1017 : 0 : int jbd2_journal_get_write_access(handle_t *handle, struct buffer_head *bh)
1018 : : {
1019 : 13118939 : struct journal_head *jh = jbd2_journal_add_journal_head(bh);
1020 : : int rc;
1021 : :
1022 : : /* We do not want to get caught playing with fields which the
1023 : : * log thread also manipulates. Make sure that the buffer
1024 : : * completes any outstanding IO before proceeding. */
1025 : 13116678 : rc = do_get_write_access(handle, jh, 0);
1026 : 13117537 : jbd2_journal_put_journal_head(jh);
1027 : 13115852 : return rc;
1028 : : }
1029 : :
1030 : :
1031 : : /*
1032 : : * When the user wants to journal a newly created buffer_head
1033 : : * (ie. getblk() returned a new buffer and we are going to populate it
1034 : : * manually rather than reading off disk), then we need to keep the
1035 : : * buffer_head locked until it has been completely filled with new
1036 : : * data. In this case, we should be able to make the assertion that
1037 : : * the bh is not already part of an existing transaction.
1038 : : *
1039 : : * The buffer should already be locked by the caller by this point.
1040 : : * There is no lock ranking violation: it was a newly created,
1041 : : * unlocked buffer beforehand. */
1042 : :
1043 : : /**
1044 : : * int jbd2_journal_get_create_access () - notify intent to use newly created bh
1045 : : * @handle: transaction to new buffer to
1046 : : * @bh: new buffer.
1047 : : *
1048 : : * Call this if you create a new bh.
1049 : : */
1050 : 0 : int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh)
1051 : : {
1052 : 35839 : transaction_t *transaction = handle->h_transaction;
1053 : : journal_t *journal;
1054 : 35839 : struct journal_head *jh = jbd2_journal_add_journal_head(bh);
1055 : : int err;
1056 : :
1057 : : jbd_debug(5, "journal_head %p\n", jh);
1058 [ - + ]: 35882 : WARN_ON(!transaction);
1059 : : err = -EROFS;
1060 [ + - ]: 35827 : if (is_handle_aborted(handle))
1061 : : goto out;
1062 : 35827 : journal = transaction->t_journal;
1063 : : err = 0;
1064 : :
1065 : : JBUFFER_TRACE(jh, "entry");
1066 : : /*
1067 : : * The buffer may already belong to this transaction due to pre-zeroing
1068 : : * in the filesystem's new_block code. It may also be on the previous,
1069 : : * committing transaction's lists, but it HAS to be in Forget state in
1070 : : * that case: the transaction must have deleted the buffer for it to be
1071 : : * reused here.
1072 : : */
1073 : : jbd_lock_bh_state(bh);
1074 : : spin_lock(&journal->j_list_lock);
1075 [ + + ][ + + ]: 71757 : J_ASSERT_JH(jh, (jh->b_transaction == transaction ||
[ - + ][ # # ]
[ # # ][ # # ]
1076 : : jh->b_transaction == NULL ||
1077 : : (jh->b_transaction == journal->j_committing_transaction &&
1078 : : jh->b_jlist == BJ_Forget)));
1079 : :
1080 [ - + ]: 35918 : J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1081 [ - + ]: 35918 : J_ASSERT_JH(jh, buffer_locked(jh2bh(jh)));
1082 : :
1083 [ + - ]: 35918 : if (jh->b_transaction == NULL) {
1084 : : /*
1085 : : * Previous jbd2_journal_forget() could have left the buffer
1086 : : * with jbddirty bit set because it was being committed. When
1087 : : * the commit finished, we've filed the buffer for
1088 : : * checkpointing and marked it dirty. Now we are reallocating
1089 : : * the buffer so the transaction freeing it must have
1090 : : * committed and so it's safe to clear the dirty bit.
1091 : : */
1092 : : clear_buffer_dirty(jh2bh(jh));
1093 : : /* first access by this transaction */
1094 : 35918 : jh->b_modified = 0;
1095 : :
1096 : : JBUFFER_TRACE(jh, "file as BJ_Reserved");
1097 : 35918 : __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
1098 [ # # ]: 0 : } else if (jh->b_transaction == journal->j_committing_transaction) {
1099 : : /* first access by this transaction */
1100 : 0 : jh->b_modified = 0;
1101 : :
1102 : : JBUFFER_TRACE(jh, "set next transaction");
1103 : 0 : jh->b_next_transaction = transaction;
1104 : : }
1105 : : spin_unlock(&journal->j_list_lock);
1106 : : jbd_unlock_bh_state(bh);
1107 : :
1108 : : /*
1109 : : * akpm: I added this. ext3_alloc_branch can pick up new indirect
1110 : : * blocks which contain freed but then revoked metadata. We need
1111 : : * to cancel the revoke in case we end up freeing it yet again
1112 : : * and the reallocating as data - this would cause a second revoke,
1113 : : * which hits an assertion error.
1114 : : */
1115 : : JBUFFER_TRACE(jh, "cancelling revoke");
1116 : 35918 : jbd2_journal_cancel_revoke(handle, jh);
1117 : : out:
1118 : 35918 : jbd2_journal_put_journal_head(jh);
1119 : 35918 : return err;
1120 : : }
1121 : :
1122 : : /**
1123 : : * int jbd2_journal_get_undo_access() - Notify intent to modify metadata with
1124 : : * non-rewindable consequences
1125 : : * @handle: transaction
1126 : : * @bh: buffer to undo
1127 : : *
1128 : : * Sometimes there is a need to distinguish between metadata which has
1129 : : * been committed to disk and that which has not. The ext3fs code uses
1130 : : * this for freeing and allocating space, we have to make sure that we
1131 : : * do not reuse freed space until the deallocation has been committed,
1132 : : * since if we overwrote that space we would make the delete
1133 : : * un-rewindable in case of a crash.
1134 : : *
1135 : : * To deal with that, jbd2_journal_get_undo_access requests write access to a
1136 : : * buffer for parts of non-rewindable operations such as delete
1137 : : * operations on the bitmaps. The journaling code must keep a copy of
1138 : : * the buffer's contents prior to the undo_access call until such time
1139 : : * as we know that the buffer has definitely been committed to disk.
1140 : : *
1141 : : * We never need to know which transaction the committed data is part
1142 : : * of, buffers touched here are guaranteed to be dirtied later and so
1143 : : * will be committed to a new transaction in due course, at which point
1144 : : * we can discard the old committed data pointer.
1145 : : *
1146 : : * Returns error number or 0 on success.
1147 : : */
1148 : 0 : int jbd2_journal_get_undo_access(handle_t *handle, struct buffer_head *bh)
1149 : : {
1150 : : int err;
1151 : 0 : struct journal_head *jh = jbd2_journal_add_journal_head(bh);
1152 : : char *committed_data = NULL;
1153 : :
1154 : : JBUFFER_TRACE(jh, "entry");
1155 : :
1156 : : /*
1157 : : * Do this first --- it can drop the journal lock, so we want to
1158 : : * make sure that obtaining the committed_data is done
1159 : : * atomically wrt. completion of any outstanding commits.
1160 : : */
1161 : 0 : err = do_get_write_access(handle, jh, 1);
1162 [ # # ]: 0 : if (err)
1163 : : goto out;
1164 : :
1165 : : repeat:
1166 [ # # ]: 0 : if (!jh->b_committed_data) {
1167 : 0 : committed_data = jbd2_alloc(jh2bh(jh)->b_size, GFP_NOFS);
1168 [ # # ]: 0 : if (!committed_data) {
1169 : 0 : printk(KERN_ERR "%s: No memory for committed data\n",
1170 : : __func__);
1171 : : err = -ENOMEM;
1172 : 0 : goto out;
1173 : : }
1174 : : }
1175 : :
1176 : : jbd_lock_bh_state(bh);
1177 [ # # ]: 0 : if (!jh->b_committed_data) {
1178 : : /* Copy out the current buffer contents into the
1179 : : * preserved, committed copy. */
1180 : : JBUFFER_TRACE(jh, "generate b_committed data");
1181 [ # # ]: 0 : if (!committed_data) {
1182 : : jbd_unlock_bh_state(bh);
1183 : : goto repeat;
1184 : : }
1185 : :
1186 : 0 : jh->b_committed_data = committed_data;
1187 : : committed_data = NULL;
1188 : 0 : memcpy(jh->b_committed_data, bh->b_data, bh->b_size);
1189 : : }
1190 : : jbd_unlock_bh_state(bh);
1191 : : out:
1192 : 0 : jbd2_journal_put_journal_head(jh);
1193 [ # # ]: 0 : if (unlikely(committed_data))
1194 : 0 : jbd2_free(committed_data, bh->b_size);
1195 : 0 : return err;
1196 : : }
1197 : :
1198 : : /**
1199 : : * void jbd2_journal_set_triggers() - Add triggers for commit writeout
1200 : : * @bh: buffer to trigger on
1201 : : * @type: struct jbd2_buffer_trigger_type containing the trigger(s).
1202 : : *
1203 : : * Set any triggers on this journal_head. This is always safe, because
1204 : : * triggers for a committing buffer will be saved off, and triggers for
1205 : : * a running transaction will match the buffer in that transaction.
1206 : : *
1207 : : * Call with NULL to clear the triggers.
1208 : : */
1209 : 0 : void jbd2_journal_set_triggers(struct buffer_head *bh,
1210 : : struct jbd2_buffer_trigger_type *type)
1211 : : {
1212 : 0 : struct journal_head *jh = jbd2_journal_grab_journal_head(bh);
1213 : :
1214 [ # # ][ # # ]: 0 : if (WARN_ON(!jh))
1215 : 0 : return;
1216 : 0 : jh->b_triggers = type;
1217 : 0 : jbd2_journal_put_journal_head(jh);
1218 : : }
1219 : :
1220 : 0 : void jbd2_buffer_frozen_trigger(struct journal_head *jh, void *mapped_data,
1221 : : struct jbd2_buffer_trigger_type *triggers)
1222 : : {
1223 : : struct buffer_head *bh = jh2bh(jh);
1224 : :
1225 [ - + ]: 75996 : if (!triggers || !triggers->t_frozen)
[ # # - + ]
[ # # ]
1226 : 0 : return;
1227 : :
1228 : 0 : triggers->t_frozen(triggers, bh, mapped_data, bh->b_size);
1229 : : }
1230 : :
1231 : 0 : void jbd2_buffer_abort_trigger(struct journal_head *jh,
1232 : : struct jbd2_buffer_trigger_type *triggers)
1233 : : {
1234 [ # # ][ # # ]: 0 : if (!triggers || !triggers->t_abort)
1235 : 0 : return;
1236 : :
1237 : 0 : triggers->t_abort(triggers, jh2bh(jh));
1238 : : }
1239 : :
1240 : :
1241 : :
1242 : : /**
1243 : : * int jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata
1244 : : * @handle: transaction to add buffer to.
1245 : : * @bh: buffer to mark
1246 : : *
1247 : : * mark dirty metadata which needs to be journaled as part of the current
1248 : : * transaction.
1249 : : *
1250 : : * The buffer must have previously had jbd2_journal_get_write_access()
1251 : : * called so that it has a valid journal_head attached to the buffer
1252 : : * head.
1253 : : *
1254 : : * The buffer is placed on the transaction's metadata list and is marked
1255 : : * as belonging to the transaction.
1256 : : *
1257 : : * Returns error number or 0 on success.
1258 : : *
1259 : : * Special care needs to be taken if the buffer already belongs to the
1260 : : * current committing transaction (in which case we should have frozen
1261 : : * data present for that commit). In that case, we don't relink the
1262 : : * buffer: that only gets done when the old transaction finally
1263 : : * completes its commit.
1264 : : */
1265 : 0 : int jbd2_journal_dirty_metadata(handle_t *handle, struct buffer_head *bh)
1266 : : {
1267 : 13107498 : transaction_t *transaction = handle->h_transaction;
1268 : : journal_t *journal;
1269 : : struct journal_head *jh;
1270 : : int ret = 0;
1271 : :
1272 [ - + ]: 13107498 : WARN_ON(!transaction);
1273 [ + - ]: 13113718 : if (is_handle_aborted(handle))
1274 : : return -EROFS;
1275 : 13113718 : journal = transaction->t_journal;
1276 : 13113718 : jh = jbd2_journal_grab_journal_head(bh);
1277 [ + ]: 13109653 : if (!jh) {
1278 : : ret = -EUCLEAN;
1279 : : goto out;
1280 : : }
1281 : : jbd_debug(5, "journal_head %p\n", jh);
1282 : : JBUFFER_TRACE(jh, "entry");
1283 : :
1284 : : jbd_lock_bh_state(bh);
1285 : :
1286 [ + + ]: 13107483 : if (jh->b_modified == 0) {
1287 : : /*
1288 : : * This buffer's got modified and becoming part
1289 : : * of the transaction. This needs to be done
1290 : : * once a transaction -bzzz
1291 : : */
1292 : 110444 : jh->b_modified = 1;
1293 [ + ]: 110444 : if (handle->h_buffer_credits <= 0) {
1294 : : ret = -ENOSPC;
1295 : : goto out_unlock_bh;
1296 : : }
1297 : 110792 : handle->h_buffer_credits--;
1298 : : }
1299 : :
1300 : : /*
1301 : : * fastpath, to avoid expensive locking. If this buffer is already
1302 : : * on the running transaction's metadata list there is nothing to do.
1303 : : * Nobody can take it off again because there is a handle open.
1304 : : * I _think_ we're OK here with SMP barriers - a mistaken decision will
1305 : : * result in this test being false, so we go in and take the locks.
1306 : : */
1307 [ + + ][ + + ]: 13107831 : if (jh->b_transaction == transaction && jh->b_jlist == BJ_Metadata) {
1308 : : JBUFFER_TRACE(jh, "fastpath");
1309 [ - + ]: 12771799 : if (unlikely(jh->b_transaction !=
1310 : : journal->j_running_transaction)) {
1311 [ # # ][ # # ]: 0 : printk(KERN_ERR "JBD2: %s: "
1312 : : "jh->b_transaction (%llu, %p, %u) != "
1313 : : "journal->j_running_transaction (%p, %u)",
1314 : 0 : journal->j_devname,
1315 : : (unsigned long long) bh->b_blocknr,
1316 : : jh->b_transaction,
1317 : : jh->b_transaction ? jh->b_transaction->t_tid : 0,
1318 : : journal->j_running_transaction,
1319 : : journal->j_running_transaction ?
1320 : : journal->j_running_transaction->t_tid : 0);
1321 : : ret = -EINVAL;
1322 : : }
1323 : : goto out_unlock_bh;
1324 : : }
1325 : :
1326 : : set_buffer_jbddirty(bh);
1327 : :
1328 : : /*
1329 : : * Metadata already on the current transaction list doesn't
1330 : : * need to be filed. Metadata on another transaction's list must
1331 : : * be committing, and will be refiled once the commit completes:
1332 : : * leave it alone for now.
1333 : : */
1334 [ + + ]: 338698 : if (jh->b_transaction != transaction) {
1335 : : JBUFFER_TRACE(jh, "already on other transaction");
1336 [ - + ]: 233168 : if (unlikely(jh->b_transaction !=
1337 : : journal->j_committing_transaction)) {
1338 [ # # ][ # # ]: 0 : printk(KERN_ERR "JBD2: %s: "
1339 : : "jh->b_transaction (%llu, %p, %u) != "
1340 : : "journal->j_committing_transaction (%p, %u)",
1341 : 0 : journal->j_devname,
1342 : : (unsigned long long) bh->b_blocknr,
1343 : : jh->b_transaction,
1344 : : jh->b_transaction ? jh->b_transaction->t_tid : 0,
1345 : : journal->j_committing_transaction,
1346 : : journal->j_committing_transaction ?
1347 : : journal->j_committing_transaction->t_tid : 0);
1348 : : ret = -EINVAL;
1349 : : }
1350 [ - + ]: 233210 : if (unlikely(jh->b_next_transaction != transaction)) {
1351 [ # # ]: 0 : printk(KERN_ERR "JBD2: %s: "
1352 : : "jh->b_next_transaction (%llu, %p, %u) != "
1353 : : "transaction (%p, %u)",
1354 : 0 : journal->j_devname,
1355 : : (unsigned long long) bh->b_blocknr,
1356 : : jh->b_next_transaction,
1357 : : jh->b_next_transaction ?
1358 : : jh->b_next_transaction->t_tid : 0,
1359 : : transaction, transaction->t_tid);
1360 : : ret = -EINVAL;
1361 : : }
1362 : : /* And this case is illegal: we can't reuse another
1363 : : * transaction's data buffer, ever. */
1364 : : goto out_unlock_bh;
1365 : : }
1366 : :
1367 : : /* That test should have eliminated the following case: */
1368 [ - + ]: 105530 : J_ASSERT_JH(jh, jh->b_frozen_data == NULL);
1369 : :
1370 : : JBUFFER_TRACE(jh, "file as BJ_Metadata");
1371 : : spin_lock(&journal->j_list_lock);
1372 : 105591 : __jbd2_journal_file_buffer(jh, transaction, BJ_Metadata);
1373 : : spin_unlock(&journal->j_list_lock);
1374 : : out_unlock_bh:
1375 : : jbd_unlock_bh_state(bh);
1376 : 13106880 : jbd2_journal_put_journal_head(jh);
1377 : : out:
1378 : : JBUFFER_TRACE(jh, "exit");
1379 : 13111080 : return ret;
1380 : : }
1381 : :
1382 : : /**
1383 : : * void jbd2_journal_forget() - bforget() for potentially-journaled buffers.
1384 : : * @handle: transaction handle
1385 : : * @bh: bh to 'forget'
1386 : : *
1387 : : * We can only do the bforget if there are no commits pending against the
1388 : : * buffer. If the buffer is dirty in the current running transaction we
1389 : : * can safely unlink it.
1390 : : *
1391 : : * bh may not be a journalled buffer at all - it may be a non-JBD
1392 : : * buffer which came off the hashtable. Check for this.
1393 : : *
1394 : : * Decrements bh->b_count by one.
1395 : : *
1396 : : * Allow this call even if the handle has aborted --- it may be part of
1397 : : * the caller's cleanup after an abort.
1398 : : */
1399 : 0 : int jbd2_journal_forget (handle_t *handle, struct buffer_head *bh)
1400 : : {
1401 : 36351 : transaction_t *transaction = handle->h_transaction;
1402 : : journal_t *journal;
1403 : : struct journal_head *jh;
1404 : : int drop_reserve = 0;
1405 : : int err = 0;
1406 : : int was_modified = 0;
1407 : :
1408 [ - + ]: 36351 : WARN_ON(!transaction);
1409 [ + - ]: 36350 : if (is_handle_aborted(handle))
1410 : : return -EROFS;
1411 : 36350 : journal = transaction->t_journal;
1412 : :
1413 : : BUFFER_TRACE(bh, "entry");
1414 : :
1415 : : jbd_lock_bh_state(bh);
1416 : : spin_lock(&journal->j_list_lock);
1417 : :
1418 [ + + ]: 36352 : if (!buffer_jbd(bh))
1419 : : goto not_jbd;
1420 : : jh = bh2jh(bh);
1421 : :
1422 : : /* Critical error: attempting to delete a bitmap buffer, maybe?
1423 : : * Don't do any jbd operations, and return an error. */
1424 [ - + ][ + - ]: 35459 : if (!J_EXPECT_JH(jh, !jh->b_committed_data,
1425 : : "inconsistent data on disk")) {
1426 : : err = -EIO;
1427 : : goto not_jbd;
1428 : : }
1429 : :
1430 : : /* keep track of whether or not this transaction modified us */
1431 : 35459 : was_modified = jh->b_modified;
1432 : :
1433 : : /*
1434 : : * The buffer's going from the transaction, we must drop
1435 : : * all references -bzzz
1436 : : */
1437 : 35459 : jh->b_modified = 0;
1438 : :
1439 [ + + ]: 35459 : if (jh->b_transaction == transaction) {
1440 [ - + ]: 32515 : J_ASSERT_JH(jh, !jh->b_frozen_data);
1441 : :
1442 : : /* If we are forgetting a buffer which is already part
1443 : : * of this transaction, then we can just drop it from
1444 : : * the transaction immediately. */
1445 : : clear_buffer_dirty(bh);
1446 : : clear_buffer_jbddirty(bh);
1447 : :
1448 : : JBUFFER_TRACE(jh, "belongs to current transaction: unfile");
1449 : :
1450 : : /*
1451 : : * we only want to drop a reference if this transaction
1452 : : * modified the buffer
1453 : : */
1454 [ + + ]: 32515 : if (was_modified)
1455 : : drop_reserve = 1;
1456 : :
1457 : : /*
1458 : : * We are no longer going to journal this buffer.
1459 : : * However, the commit of this transaction is still
1460 : : * important to the buffer: the delete that we are now
1461 : : * processing might obsolete an old log entry, so by
1462 : : * committing, we can satisfy the buffer's checkpoint.
1463 : : *
1464 : : * So, if we have a checkpoint on the buffer, we should
1465 : : * now refile the buffer on our BJ_Forget list so that
1466 : : * we know to remove the checkpoint after we commit.
1467 : : */
1468 : :
1469 [ + + ]: 32515 : if (jh->b_cp_transaction) {
1470 : 1251 : __jbd2_journal_temp_unlink_buffer(jh);
1471 : 1251 : __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
1472 : : } else {
1473 : : __jbd2_journal_unfile_buffer(jh);
1474 [ + - ]: 31264 : if (!buffer_jbd(bh)) {
1475 : : spin_unlock(&journal->j_list_lock);
1476 : : jbd_unlock_bh_state(bh);
1477 : 31264 : __bforget(bh);
1478 : 31264 : goto drop;
1479 : : }
1480 : : }
1481 [ + + ]: 2944 : } else if (jh->b_transaction) {
1482 [ - + ]: 1482 : J_ASSERT_JH(jh, (jh->b_transaction ==
1483 : : journal->j_committing_transaction));
1484 : : /* However, if the buffer is still owned by a prior
1485 : : * (committing) transaction, we can't drop it yet... */
1486 : : JBUFFER_TRACE(jh, "belongs to older transaction");
1487 : : /* ... but we CAN drop it from the new transaction if we
1488 : : * have also modified it since the original commit. */
1489 : :
1490 [ + + ]: 1482 : if (jh->b_next_transaction) {
1491 [ - + ]: 499 : J_ASSERT(jh->b_next_transaction == transaction);
1492 : 499 : jh->b_next_transaction = NULL;
1493 : :
1494 : : /*
1495 : : * only drop a reference if this transaction modified
1496 : : * the buffer
1497 : : */
1498 [ + + ]: 499 : if (was_modified)
1499 : : drop_reserve = 1;
1500 : : }
1501 : : }
1502 : :
1503 : : not_jbd:
1504 : : spin_unlock(&journal->j_list_lock);
1505 : : jbd_unlock_bh_state(bh);
1506 : 5088 : __brelse(bh);
1507 : : drop:
1508 [ + + ]: 36352 : if (drop_reserve) {
1509 : : /* no need to reserve log space for this block -bzzz */
1510 : 33003 : handle->h_buffer_credits++;
1511 : : }
1512 : 36352 : return err;
1513 : : }
1514 : :
1515 : : /**
1516 : : * int jbd2_journal_stop() - complete a transaction
1517 : : * @handle: tranaction to complete.
1518 : : *
1519 : : * All done for a particular handle.
1520 : : *
1521 : : * There is not much action needed here. We just return any remaining
1522 : : * buffer credits to the transaction and remove the handle. The only
1523 : : * complication is that we need to start a commit operation if the
1524 : : * filesystem is marked for synchronous update.
1525 : : *
1526 : : * jbd2_journal_stop itself will not usually return an error, but it may
1527 : : * do so in unusual circumstances. In particular, expect it to
1528 : : * return -EIO if a jbd2_journal_abort has been executed since the
1529 : : * transaction began.
1530 : : */
1531 : 0 : int jbd2_journal_stop(handle_t *handle)
1532 : : {
1533 : 12844485 : transaction_t *transaction = handle->h_transaction;
1534 : : journal_t *journal;
1535 : : int err = 0, wait_for_commit = 0;
1536 : : tid_t tid;
1537 : : pid_t pid;
1538 : :
1539 [ + ]: 12844485 : if (!transaction)
1540 : : goto free_and_exit;
1541 : 12844648 : journal = transaction->t_journal;
1542 : :
1543 [ - + ]: 12844648 : J_ASSERT(journal_current_handle() == handle);
1544 : :
1545 [ + ]: 12844648 : if (is_handle_aborted(handle))
1546 : : err = -EIO;
1547 : : else
1548 [ - + ]: 12847163 : J_ASSERT(atomic_read(&transaction->t_updates) > 0);
1549 : :
1550 [ + + ]: 12844648 : if (--handle->h_ref > 0) {
1551 : : jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1552 : : handle->h_ref);
1553 : : return err;
1554 : : }
1555 : :
1556 : : jbd_debug(4, "Handle %p going down\n", handle);
1557 : 61081804 : trace_jbd2_handle_stats(journal->j_fs_dev->bd_dev,
1558 : 8725972 : transaction->t_tid,
1559 : 17451944 : handle->h_type, handle->h_line_no,
1560 : 8725972 : jiffies - handle->h_start_jiffies,
1561 : 17451944 : handle->h_sync, handle->h_requested_credits,
1562 : 8725972 : (handle->h_requested_credits -
1563 : 8725972 : handle->h_buffer_credits));
1564 : :
1565 : : /*
1566 : : * Implement synchronous transaction batching. If the handle
1567 : : * was synchronous, don't force a commit immediately. Let's
1568 : : * yield and let another thread piggyback onto this
1569 : : * transaction. Keep doing that while new threads continue to
1570 : : * arrive. It doesn't cost much - we're about to run a commit
1571 : : * and sleep on IO anyway. Speeds up many-threaded, many-dir
1572 : : * operations by 30x or more...
1573 : : *
1574 : : * We try and optimize the sleep time against what the
1575 : : * underlying disk can do, instead of having a static sleep
1576 : : * time. This is useful for the case where our storage is so
1577 : : * fast that it is more optimal to go ahead and force a flush
1578 : : * and wait for the transaction to be committed than it is to
1579 : : * wait for an arbitrary amount of time for new writers to
1580 : : * join the transaction. We achieve this by measuring how
1581 : : * long it takes to commit a transaction, and compare it with
1582 : : * how long this transaction has been running, and if run time
1583 : : * < commit time then we sleep for the delta and commit. This
1584 : : * greatly helps super fast disks that would see slowdowns as
1585 : : * more threads started doing fsyncs.
1586 : : *
1587 : : * But don't do this if this process was the most recent one
1588 : : * to perform a synchronous write. We do this to detect the
1589 : : * case where a single process is doing a stream of sync
1590 : : * writes. No point in waiting for joiners in that case.
1591 : : */
1592 : 8728163 : pid = current->pid;
1593 [ - + ][ # # ]: 8728163 : if (handle->h_sync && journal->j_last_sync_writer != pid) {
1594 : : u64 commit_time, trans_time;
1595 : :
1596 : 0 : journal->j_last_sync_writer = pid;
1597 : :
1598 : 0 : read_lock(&journal->j_state_lock);
1599 : 0 : commit_time = journal->j_average_commit_time;
1600 : : read_unlock(&journal->j_state_lock);
1601 : :
1602 : 0 : trans_time = ktime_to_ns(ktime_sub(ktime_get(),
1603 : : transaction->t_start_time));
1604 : :
1605 : 0 : commit_time = max_t(u64, commit_time,
1606 : : 1000*journal->j_min_batch_time);
1607 : 0 : commit_time = min_t(u64, commit_time,
1608 : : 1000*journal->j_max_batch_time);
1609 : :
1610 [ # # ]: 0 : if (trans_time < commit_time) {
1611 : 0 : ktime_t expires = ktime_add_ns(ktime_get(),
1612 : : commit_time);
1613 : 0 : set_current_state(TASK_UNINTERRUPTIBLE);
1614 : 0 : schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1615 : : }
1616 : : }
1617 : :
1618 [ - + ]: 8728163 : if (handle->h_sync)
1619 : 0 : transaction->t_synchronous_commit = 1;
1620 : 8728163 : current->journal_info = NULL;
1621 : 8728163 : atomic_sub(handle->h_buffer_credits,
1622 : : &transaction->t_outstanding_credits);
1623 : :
1624 : : /*
1625 : : * If the handle is marked SYNC, we need to set another commit
1626 : : * going! We also want to force a commit if the current
1627 : : * transaction is occupying too much of the log, or if the
1628 : : * transaction is too old now.
1629 : : */
1630 [ + ][ + ]: 8725587 : if (handle->h_sync ||
1631 : 8727168 : (atomic_read(&transaction->t_outstanding_credits) >
1632 : 8727168 : journal->j_max_transaction_buffers) ||
1633 [ + + ]: 8727643 : time_after_eq(jiffies, transaction->t_expires)) {
1634 : : /* Do this even for aborted journals: an abort still
1635 : : * completes the commit thread, it just doesn't write
1636 : : * anything to disk. */
1637 : :
1638 : : jbd_debug(2, "transaction too old, requesting commit for "
1639 : : "handle %p\n", handle);
1640 : : /* This is non-blocking */
1641 : 3055 : jbd2_log_start_commit(journal, transaction->t_tid);
1642 : :
1643 : : /*
1644 : : * Special case: JBD2_SYNC synchronous updates require us
1645 : : * to wait for the commit to complete.
1646 : : */
1647 [ - + ][ # # ]: 4548 : if (handle->h_sync && !(current->flags & PF_MEMALLOC))
1648 : : wait_for_commit = 1;
1649 : : }
1650 : :
1651 : : /*
1652 : : * Once we drop t_updates, if it goes to zero the transaction
1653 : : * could start committing on us and eventually disappear. So
1654 : : * once we do this, we must not dereference transaction
1655 : : * pointer again.
1656 : : */
1657 : 8727080 : tid = transaction->t_tid;
1658 [ + + ]: 8725705 : if (atomic_dec_and_test(&transaction->t_updates)) {
1659 : 7979226 : wake_up(&journal->j_wait_updates);
1660 [ - + ]: 7979226 : if (journal->j_barrier_count)
1661 : 0 : wake_up(&journal->j_wait_transaction_locked);
1662 : : }
1663 : :
1664 [ - + ]: 8725967 : if (wait_for_commit)
1665 : 0 : err = jbd2_log_wait_commit(journal, tid);
1666 : :
1667 : : lock_map_release(&handle->h_lockdep_map);
1668 : :
1669 [ - + ]: 8725967 : if (handle->h_rsv_handle)
1670 : 0 : jbd2_journal_free_reserved(handle->h_rsv_handle);
1671 : : free_and_exit:
1672 : : jbd2_free_handle(handle);
1673 : 8729092 : return err;
1674 : : }
1675 : :
1676 : : /*
1677 : : *
1678 : : * List management code snippets: various functions for manipulating the
1679 : : * transaction buffer lists.
1680 : : *
1681 : : */
1682 : :
1683 : : /*
1684 : : * Append a buffer to a transaction list, given the transaction's list head
1685 : : * pointer.
1686 : : *
1687 : : * j_list_lock is held.
1688 : : *
1689 : : * jbd_lock_bh_state(jh2bh(jh)) is held.
1690 : : */
1691 : :
1692 : : static inline void
1693 : : __blist_add_buffer(struct journal_head **list, struct journal_head *jh)
1694 : : {
1695 [ + + ]: 369277 : if (!*list) {
1696 : 156926 : jh->b_tnext = jh->b_tprev = jh;
1697 : 156926 : *list = jh;
1698 : : } else {
1699 : : /* Insert at the tail of the list to preserve order */
1700 : 212351 : struct journal_head *first = *list, *last = first->b_tprev;
1701 : 212351 : jh->b_tprev = last;
1702 : 212351 : jh->b_tnext = first;
1703 : 212351 : last->b_tnext = first->b_tprev = jh;
1704 : : }
1705 : : }
1706 : :
1707 : : /*
1708 : : * Remove a buffer from a transaction list, given the transaction's list
1709 : : * head pointer.
1710 : : *
1711 : : * Called with j_list_lock held, and the journal may not be locked.
1712 : : *
1713 : : * jbd_lock_bh_state(jh2bh(jh)) is held.
1714 : : */
1715 : :
1716 : : static inline void
1717 : : __blist_del_buffer(struct journal_head **list, struct journal_head *jh)
1718 : : {
1719 [ + + ]: 368813 : if (*list == jh) {
1720 : 274787 : *list = jh->b_tnext;
1721 [ + + ]: 274787 : if (*list == jh)
1722 : 156925 : *list = NULL;
1723 : : }
1724 : 368813 : jh->b_tprev->b_tnext = jh->b_tnext;
1725 : 368813 : jh->b_tnext->b_tprev = jh->b_tprev;
1726 : : }
1727 : :
1728 : : /*
1729 : : * Remove a buffer from the appropriate transaction list.
1730 : : *
1731 : : * Note that this function can *change* the value of
1732 : : * bh->b_transaction->t_buffers, t_forget, t_shadow_list, t_log_list or
1733 : : * t_reserved_list. If the caller is holding onto a copy of one of these
1734 : : * pointers, it could go bad. Generally the caller needs to re-read the
1735 : : * pointer from the transaction_t.
1736 : : *
1737 : : * Called under j_list_lock.
1738 : : */
1739 : 0 : static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh)
1740 : : {
1741 : : struct journal_head **list = NULL;
1742 : : transaction_t *transaction;
1743 : : struct buffer_head *bh = jh2bh(jh);
1744 : :
1745 [ - + ]: 374909 : J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
1746 : 374909 : transaction = jh->b_transaction;
1747 [ + + ]: 374909 : if (transaction)
1748 [ - + ]: 374831 : assert_spin_locked(&transaction->t_journal->j_list_lock);
1749 : :
1750 [ - + ]: 374909 : J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
1751 [ + + ]: 374909 : if (jh->b_jlist != BJ_None)
1752 [ - + ]: 368813 : J_ASSERT_JH(jh, transaction != NULL);
1753 : :
1754 [ + + + + : 374909 : switch (jh->b_jlist) {
- + ]
1755 : : case BJ_None:
1756 : 0 : return;
1757 : : case BJ_Metadata:
1758 : 109890 : transaction->t_nr_buffers--;
1759 [ - + ]: 109890 : J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0);
1760 : 109890 : list = &transaction->t_buffers;
1761 : 109890 : break;
1762 : : case BJ_Forget:
1763 : 77327 : list = &transaction->t_forget;
1764 : 77327 : break;
1765 : : case BJ_Shadow:
1766 : 75996 : list = &transaction->t_shadow_list;
1767 : 75996 : break;
1768 : : case BJ_Reserved:
1769 : 105600 : list = &transaction->t_reserved_list;
1770 : 105600 : break;
1771 : : }
1772 : :
1773 : : __blist_del_buffer(list, jh);
1774 : 368813 : jh->b_jlist = BJ_None;
1775 [ + + ]: 368813 : if (test_clear_buffer_jbddirty(bh))
1776 : 72617 : mark_buffer_dirty(bh); /* Expose it to the VM */
1777 : : }
1778 : :
1779 : : /*
1780 : : * Remove buffer from all transactions.
1781 : : *
1782 : : * Called with bh_state lock and j_list_lock
1783 : : *
1784 : : * jh and bh may be already freed when this function returns.
1785 : : */
1786 : : static void __jbd2_journal_unfile_buffer(struct journal_head *jh)
1787 : : {
1788 : 105212 : __jbd2_journal_temp_unlink_buffer(jh);
1789 : 105212 : jh->b_transaction = NULL;
1790 : 105212 : jbd2_journal_put_journal_head(jh);
1791 : : }
1792 : :
1793 : 0 : void jbd2_journal_unfile_buffer(journal_t *journal, struct journal_head *jh)
1794 : : {
1795 : : struct buffer_head *bh = jh2bh(jh);
1796 : :
1797 : : /* Get reference so that buffer cannot be freed before we unlock it */
1798 : : get_bh(bh);
1799 : : jbd_lock_bh_state(bh);
1800 : : spin_lock(&journal->j_list_lock);
1801 : : __jbd2_journal_unfile_buffer(jh);
1802 : : spin_unlock(&journal->j_list_lock);
1803 : : jbd_unlock_bh_state(bh);
1804 : 0 : __brelse(bh);
1805 : 0 : }
1806 : :
1807 : : /*
1808 : : * Called from jbd2_journal_try_to_free_buffers().
1809 : : *
1810 : : * Called under jbd_lock_bh_state(bh)
1811 : : */
1812 : : static void
1813 : 0 : __journal_try_to_free_buffer(journal_t *journal, struct buffer_head *bh)
1814 : : {
1815 : : struct journal_head *jh;
1816 : :
1817 : : jh = bh2jh(bh);
1818 : :
1819 [ + - ][ + - ]: 1260 : if (buffer_locked(bh) || buffer_dirty(bh))
1820 : : goto out;
1821 : :
1822 [ + - ]: 1260 : if (jh->b_next_transaction != NULL)
1823 : : goto out;
1824 : :
1825 : : spin_lock(&journal->j_list_lock);
1826 [ + + ][ + + ]: 1260 : if (jh->b_cp_transaction != NULL && jh->b_transaction == NULL) {
1827 : : /* written-back checkpointed metadata buffer */
1828 : : JBUFFER_TRACE(jh, "remove from checkpoint list");
1829 : 23 : __jbd2_journal_remove_checkpoint(jh);
1830 : : }
1831 : : spin_unlock(&journal->j_list_lock);
1832 : : out:
1833 : 1260 : return;
1834 : : }
1835 : :
1836 : : /**
1837 : : * int jbd2_journal_try_to_free_buffers() - try to free page buffers.
1838 : : * @journal: journal for operation
1839 : : * @page: to try and free
1840 : : * @gfp_mask: we use the mask to detect how hard should we try to release
1841 : : * buffers. If __GFP_WAIT and __GFP_FS is set, we wait for commit code to
1842 : : * release the buffers.
1843 : : *
1844 : : *
1845 : : * For all the buffers on this page,
1846 : : * if they are fully written out ordered data, move them onto BUF_CLEAN
1847 : : * so try_to_free_buffers() can reap them.
1848 : : *
1849 : : * This function returns non-zero if we wish try_to_free_buffers()
1850 : : * to be called. We do this if the page is releasable by try_to_free_buffers().
1851 : : * We also do it if the page has locked or dirty buffers and the caller wants
1852 : : * us to perform sync or async writeout.
1853 : : *
1854 : : * This complicates JBD locking somewhat. We aren't protected by the
1855 : : * BKL here. We wish to remove the buffer from its committing or
1856 : : * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer.
1857 : : *
1858 : : * This may *change* the value of transaction_t->t_datalist, so anyone
1859 : : * who looks at t_datalist needs to lock against this function.
1860 : : *
1861 : : * Even worse, someone may be doing a jbd2_journal_dirty_data on this
1862 : : * buffer. So we need to lock against that. jbd2_journal_dirty_data()
1863 : : * will come out of the lock with the buffer dirty, which makes it
1864 : : * ineligible for release here.
1865 : : *
1866 : : * Who else is affected by this? hmm... Really the only contender
1867 : : * is do_get_write_access() - it could be looking at the buffer while
1868 : : * journal_try_to_free_buffer() is changing its state. But that
1869 : : * cannot happen because we never reallocate freed data as metadata
1870 : : * while the data is part of a transaction. Yes?
1871 : : *
1872 : : * Return 0 on failure, 1 on success
1873 : : */
1874 : 0 : int jbd2_journal_try_to_free_buffers(journal_t *journal,
1875 : : struct page *page, gfp_t gfp_mask)
1876 : : {
1877 : : struct buffer_head *head;
1878 : : struct buffer_head *bh;
1879 : : int ret = 0;
1880 : :
1881 [ - + ]: 1897893 : J_ASSERT(PageLocked(page));
1882 : :
1883 [ - + ]: 1897893 : head = page_buffers(page);
1884 : : bh = head;
1885 : : do {
1886 : : struct journal_head *jh;
1887 : :
1888 : : /*
1889 : : * We take our own ref against the journal_head here to avoid
1890 : : * having to add tons of locking around each instance of
1891 : : * jbd2_journal_put_journal_head().
1892 : : */
1893 : 1897893 : jh = jbd2_journal_grab_journal_head(bh);
1894 [ + + ]: 1897731 : if (!jh)
1895 : 1896471 : continue;
1896 : :
1897 : : jbd_lock_bh_state(bh);
1898 : 1260 : __journal_try_to_free_buffer(journal, bh);
1899 : 1260 : jbd2_journal_put_journal_head(jh);
1900 : : jbd_unlock_bh_state(bh);
1901 [ + + ]: 1260 : if (buffer_jbd(bh))
1902 : : goto busy;
1903 [ - + ]: 1896525 : } while ((bh = bh->b_this_page) != head);
1904 : :
1905 : 1896525 : ret = try_to_free_buffers(page);
1906 : :
1907 : : busy:
1908 : 1897862 : return ret;
1909 : : }
1910 : :
1911 : : /*
1912 : : * This buffer is no longer needed. If it is on an older transaction's
1913 : : * checkpoint list we need to record it on this transaction's forget list
1914 : : * to pin this buffer (and hence its checkpointing transaction) down until
1915 : : * this transaction commits. If the buffer isn't on a checkpoint list, we
1916 : : * release it.
1917 : : * Returns non-zero if JBD no longer has an interest in the buffer.
1918 : : *
1919 : : * Called under j_list_lock.
1920 : : *
1921 : : * Called under jbd_lock_bh_state(bh).
1922 : : */
1923 : 0 : static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction)
1924 : : {
1925 : : int may_free = 1;
1926 : : struct buffer_head *bh = jh2bh(jh);
1927 : :
1928 [ + + ]: 1466 : if (jh->b_cp_transaction) {
1929 : : JBUFFER_TRACE(jh, "on running+cp transaction");
1930 : 78 : __jbd2_journal_temp_unlink_buffer(jh);
1931 : : /*
1932 : : * We don't want to write the buffer anymore, clear the
1933 : : * bit so that we don't confuse checks in
1934 : : * __journal_file_buffer
1935 : : */
1936 : : clear_buffer_dirty(bh);
1937 : 78 : __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
1938 : : may_free = 0;
1939 : : } else {
1940 : : JBUFFER_TRACE(jh, "on running transaction");
1941 : : __jbd2_journal_unfile_buffer(jh);
1942 : : }
1943 : 1466 : return may_free;
1944 : : }
1945 : :
1946 : : /*
1947 : : * jbd2_journal_invalidatepage
1948 : : *
1949 : : * This code is tricky. It has a number of cases to deal with.
1950 : : *
1951 : : * There are two invariants which this code relies on:
1952 : : *
1953 : : * i_size must be updated on disk before we start calling invalidatepage on the
1954 : : * data.
1955 : : *
1956 : : * This is done in ext3 by defining an ext3_setattr method which
1957 : : * updates i_size before truncate gets going. By maintaining this
1958 : : * invariant, we can be sure that it is safe to throw away any buffers
1959 : : * attached to the current transaction: once the transaction commits,
1960 : : * we know that the data will not be needed.
1961 : : *
1962 : : * Note however that we can *not* throw away data belonging to the
1963 : : * previous, committing transaction!
1964 : : *
1965 : : * Any disk blocks which *are* part of the previous, committing
1966 : : * transaction (and which therefore cannot be discarded immediately) are
1967 : : * not going to be reused in the new running transaction
1968 : : *
1969 : : * The bitmap committed_data images guarantee this: any block which is
1970 : : * allocated in one transaction and removed in the next will be marked
1971 : : * as in-use in the committed_data bitmap, so cannot be reused until
1972 : : * the next transaction to delete the block commits. This means that
1973 : : * leaving committing buffers dirty is quite safe: the disk blocks
1974 : : * cannot be reallocated to a different file and so buffer aliasing is
1975 : : * not possible.
1976 : : *
1977 : : *
1978 : : * The above applies mainly to ordered data mode. In writeback mode we
1979 : : * don't make guarantees about the order in which data hits disk --- in
1980 : : * particular we don't guarantee that new dirty data is flushed before
1981 : : * transaction commit --- so it is always safe just to discard data
1982 : : * immediately in that mode. --sct
1983 : : */
1984 : :
1985 : : /*
1986 : : * The journal_unmap_buffer helper function returns zero if the buffer
1987 : : * concerned remains pinned as an anonymous buffer belonging to an older
1988 : : * transaction.
1989 : : *
1990 : : * We're outside-transaction here. Either or both of j_running_transaction
1991 : : * and j_committing_transaction may be NULL.
1992 : : */
1993 : 0 : static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh,
1994 : : int partial_page)
1995 : : {
1996 : : transaction_t *transaction;
1997 : : struct journal_head *jh;
1998 : : int may_free = 1;
1999 : :
2000 : : BUFFER_TRACE(bh, "entry");
2001 : :
2002 : : /*
2003 : : * It is safe to proceed here without the j_list_lock because the
2004 : : * buffers cannot be stolen by try_to_free_buffers as long as we are
2005 : : * holding the page lock. --sct
2006 : : */
2007 : :
2008 [ + + ]: 5291 : if (!buffer_jbd(bh))
2009 : : goto zap_buffer_unlocked;
2010 : :
2011 : : /* OK, we have data buffer in journaled mode */
2012 : 1468 : write_lock(&journal->j_state_lock);
2013 : : jbd_lock_bh_state(bh);
2014 : : spin_lock(&journal->j_list_lock);
2015 : :
2016 : 1468 : jh = jbd2_journal_grab_journal_head(bh);
2017 [ + - ]: 1468 : if (!jh)
2018 : : goto zap_buffer_no_jh;
2019 : :
2020 : : /*
2021 : : * We cannot remove the buffer from checkpoint lists until the
2022 : : * transaction adding inode to orphan list (let's call it T)
2023 : : * is committed. Otherwise if the transaction changing the
2024 : : * buffer would be cleaned from the journal before T is
2025 : : * committed, a crash will cause that the correct contents of
2026 : : * the buffer will be lost. On the other hand we have to
2027 : : * clear the buffer dirty bit at latest at the moment when the
2028 : : * transaction marking the buffer as freed in the filesystem
2029 : : * structures is committed because from that moment on the
2030 : : * block can be reallocated and used by a different page.
2031 : : * Since the block hasn't been freed yet but the inode has
2032 : : * already been added to orphan list, it is safe for us to add
2033 : : * the buffer to BJ_Forget list of the newest transaction.
2034 : : *
2035 : : * Also we have to clear buffer_mapped flag of a truncated buffer
2036 : : * because the buffer_head may be attached to the page straddling
2037 : : * i_size (can happen only when blocksize < pagesize) and thus the
2038 : : * buffer_head can be reused when the file is extended again. So we end
2039 : : * up keeping around invalidated buffers attached to transactions'
2040 : : * BJ_Forget list just to stop checkpointing code from cleaning up
2041 : : * the transaction this buffer was modified in.
2042 : : */
2043 : 1468 : transaction = jh->b_transaction;
2044 [ + + ]: 1468 : if (transaction == NULL) {
2045 : : /* First case: not on any transaction. If it
2046 : : * has no checkpoint link, then we can zap it:
2047 : : * it's a writeback-mode buffer so we don't care
2048 : : * if it hits disk safely. */
2049 [ + - ]: 78 : if (!jh->b_cp_transaction) {
2050 : : JBUFFER_TRACE(jh, "not on any transaction: zap");
2051 : : goto zap_buffer;
2052 : : }
2053 : :
2054 [ + - ]: 78 : if (!buffer_dirty(bh)) {
2055 : : /* bdflush has written it. We can drop it now */
2056 : : goto zap_buffer;
2057 : : }
2058 : :
2059 : : /* OK, it must be in the journal but still not
2060 : : * written fully to disk: it's metadata or
2061 : : * journaled data... */
2062 : :
2063 [ + - ]: 78 : if (journal->j_running_transaction) {
2064 : : /* ... and once the current transaction has
2065 : : * committed, the buffer won't be needed any
2066 : : * longer. */
2067 : : JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget");
2068 : 78 : may_free = __dispose_buffer(jh,
2069 : : journal->j_running_transaction);
2070 : 78 : goto zap_buffer;
2071 : : } else {
2072 : : /* There is no currently-running transaction. So the
2073 : : * orphan record which we wrote for this file must have
2074 : : * passed into commit. We must attach this buffer to
2075 : : * the committing transaction, if it exists. */
2076 [ # # ]: 0 : if (journal->j_committing_transaction) {
2077 : : JBUFFER_TRACE(jh, "give to committing trans");
2078 : 0 : may_free = __dispose_buffer(jh,
2079 : : journal->j_committing_transaction);
2080 : 0 : goto zap_buffer;
2081 : : } else {
2082 : : /* The orphan record's transaction has
2083 : : * committed. We can cleanse this buffer */
2084 : : clear_buffer_jbddirty(bh);
2085 : : goto zap_buffer;
2086 : : }
2087 : : }
2088 [ + + ]: 1390 : } else if (transaction == journal->j_committing_transaction) {
2089 : : JBUFFER_TRACE(jh, "on committing transaction");
2090 : : /*
2091 : : * The buffer is committing, we simply cannot touch
2092 : : * it. If the page is straddling i_size we have to wait
2093 : : * for commit and try again.
2094 : : */
2095 [ - + ]: 2 : if (partial_page) {
2096 : 0 : jbd2_journal_put_journal_head(jh);
2097 : : spin_unlock(&journal->j_list_lock);
2098 : : jbd_unlock_bh_state(bh);
2099 : : write_unlock(&journal->j_state_lock);
2100 : 0 : return -EBUSY;
2101 : : }
2102 : : /*
2103 : : * OK, buffer won't be reachable after truncate. We just set
2104 : : * j_next_transaction to the running transaction (if there is
2105 : : * one) and mark buffer as freed so that commit code knows it
2106 : : * should clear dirty bits when it is done with the buffer.
2107 : : */
2108 : : set_buffer_freed(bh);
2109 [ + - ][ + - ]: 2 : if (journal->j_running_transaction && buffer_jbddirty(bh))
2110 : 2 : jh->b_next_transaction = journal->j_running_transaction;
2111 : 2 : jbd2_journal_put_journal_head(jh);
2112 : : spin_unlock(&journal->j_list_lock);
2113 : : jbd_unlock_bh_state(bh);
2114 : : write_unlock(&journal->j_state_lock);
2115 : 2 : return 0;
2116 : : } else {
2117 : : /* Good, the buffer belongs to the running transaction.
2118 : : * We are writing our own transaction's data, not any
2119 : : * previous one's, so it is safe to throw it away
2120 : : * (remember that we expect the filesystem to have set
2121 : : * i_size already for this truncate so recovery will not
2122 : : * expose the disk blocks we are discarding here.) */
2123 [ - + ]: 1388 : J_ASSERT_JH(jh, transaction == journal->j_running_transaction);
2124 : : JBUFFER_TRACE(jh, "on running transaction");
2125 : 1388 : may_free = __dispose_buffer(jh, transaction);
2126 : : }
2127 : :
2128 : : zap_buffer:
2129 : : /*
2130 : : * This is tricky. Although the buffer is truncated, it may be reused
2131 : : * if blocksize < pagesize and it is attached to the page straddling
2132 : : * EOF. Since the buffer might have been added to BJ_Forget list of the
2133 : : * running transaction, journal_get_write_access() won't clear
2134 : : * b_modified and credit accounting gets confused. So clear b_modified
2135 : : * here.
2136 : : */
2137 : 1466 : jh->b_modified = 0;
2138 : 1466 : jbd2_journal_put_journal_head(jh);
2139 : : zap_buffer_no_jh:
2140 : : spin_unlock(&journal->j_list_lock);
2141 : : jbd_unlock_bh_state(bh);
2142 : : write_unlock(&journal->j_state_lock);
2143 : : zap_buffer_unlocked:
2144 : : clear_buffer_dirty(bh);
2145 [ - + ]: 5289 : J_ASSERT_BH(bh, !buffer_jbddirty(bh));
2146 : : clear_buffer_mapped(bh);
2147 : : clear_buffer_req(bh);
2148 : : clear_buffer_new(bh);
2149 : : clear_buffer_delay(bh);
2150 : : clear_buffer_unwritten(bh);
2151 : 5289 : bh->b_bdev = NULL;
2152 : 5289 : return may_free;
2153 : : }
2154 : :
2155 : : /**
2156 : : * void jbd2_journal_invalidatepage()
2157 : : * @journal: journal to use for flush...
2158 : : * @page: page to flush
2159 : : * @offset: start of the range to invalidate
2160 : : * @length: length of the range to invalidate
2161 : : *
2162 : : * Reap page buffers containing data after in the specified range in page.
2163 : : * Can return -EBUSY if buffers are part of the committing transaction and
2164 : : * the page is straddling i_size. Caller then has to wait for current commit
2165 : : * and try again.
2166 : : */
2167 : 0 : int jbd2_journal_invalidatepage(journal_t *journal,
2168 : : struct page *page,
2169 : : unsigned int offset,
2170 : : unsigned int length)
2171 : : {
2172 : : struct buffer_head *head, *bh, *next;
2173 : 5291 : unsigned int stop = offset + length;
2174 : : unsigned int curr_off = 0;
2175 : 5291 : int partial_page = (offset || length < PAGE_CACHE_SIZE);
2176 : : int may_free = 1;
2177 : : int ret = 0;
2178 : :
2179 [ - + ]: 5291 : if (!PageLocked(page))
2180 : 0 : BUG();
2181 [ + - ]: 5291 : if (!page_has_buffers(page))
2182 : : return 0;
2183 : :
2184 [ - + ]: 5291 : BUG_ON(stop > PAGE_CACHE_SIZE || stop < length);
2185 : :
2186 : : /* We will potentially be playing with lists other than just the
2187 : : * data lists (especially for journaled data mode), so be
2188 : : * cautious in our locking. */
2189 : :
2190 [ - + ]: 5291 : head = bh = page_buffers(page);
2191 : : do {
2192 : 5291 : unsigned int next_off = curr_off + bh->b_size;
2193 : 5291 : next = bh->b_this_page;
2194 : :
2195 [ + - ]: 5291 : if (next_off > stop)
2196 : : return 0;
2197 : :
2198 [ + - ]: 5291 : if (offset <= curr_off) {
2199 : : /* This block is wholly outside the truncation point */
2200 : : lock_buffer(bh);
2201 : 5291 : ret = journal_unmap_buffer(journal, bh, partial_page);
2202 : 5291 : unlock_buffer(bh);
2203 [ + - ]: 5291 : if (ret < 0)
2204 : : return ret;
2205 : 5291 : may_free &= ret;
2206 : : }
2207 : : curr_off = next_off;
2208 : : bh = next;
2209 : :
2210 [ - + ]: 5291 : } while (bh != head);
2211 : :
2212 [ + - ]: 5291 : if (!partial_page) {
2213 [ + + ][ + - ]: 5291 : if (may_free && try_to_free_buffers(page))
2214 [ - + ]: 5211 : J_ASSERT(!page_has_buffers(page));
2215 : : }
2216 : : return 0;
2217 : : }
2218 : :
2219 : : /*
2220 : : * File a buffer on the given transaction list.
2221 : : */
2222 : 0 : void __jbd2_journal_file_buffer(struct journal_head *jh,
2223 : : transaction_t *transaction, int jlist)
2224 : : {
2225 : : struct journal_head **list = NULL;
2226 : : int was_dirty = 0;
2227 : 0 : struct buffer_head *bh = jh2bh(jh);
2228 : :
2229 [ - + ]: 369277 : J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
2230 [ - + ]: 369277 : assert_spin_locked(&transaction->t_journal->j_list_lock);
2231 : :
2232 [ - + ]: 369277 : J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
2233 [ + + ][ - + ]: 369277 : J_ASSERT_JH(jh, jh->b_transaction == transaction ||
2234 : : jh->b_transaction == NULL);
2235 : :
2236 [ + + ][ + - ]: 369277 : if (jh->b_transaction && jh->b_jlist == jlist)
2237 : : return;
2238 : :
2239 [ + + ]: 369277 : if (jlist == BJ_Metadata || jlist == BJ_Reserved ||
2240 [ + - ]: 153323 : jlist == BJ_Shadow || jlist == BJ_Forget) {
2241 : : /*
2242 : : * For metadata buffers, we track dirty bit in buffer_jbddirty
2243 : : * instead of buffer_dirty. We should not see a dirty bit set
2244 : : * here because we clear it in do_get_write_access but e.g.
2245 : : * tune2fs can modify the sb and set the dirty bit at any time
2246 : : * so we try to gracefully handle that.
2247 : : */
2248 [ - + ]: 369277 : if (buffer_dirty(bh))
2249 : 0 : warn_dirty_buffer(bh);
2250 [ + - + + ]: 738554 : if (test_clear_buffer_dirty(bh) ||
2251 : : test_clear_buffer_jbddirty(bh))
2252 : : was_dirty = 1;
2253 : : }
2254 : :
2255 [ + + ]: 369277 : if (jh->b_transaction)
2256 : 263601 : __jbd2_journal_temp_unlink_buffer(jh);
2257 : : else
2258 : 105676 : jbd2_journal_grab_journal_head(bh);
2259 : 369277 : jh->b_transaction = transaction;
2260 : :
2261 [ - + + + : 738554 : switch (jlist) {
+ - ]
2262 : : case BJ_None:
2263 [ # # ]: 0 : J_ASSERT_JH(jh, !jh->b_committed_data);
2264 [ # # ]: 0 : J_ASSERT_JH(jh, !jh->b_frozen_data);
2265 : : return;
2266 : : case BJ_Metadata:
2267 : 110354 : transaction->t_nr_buffers++;
2268 : 110354 : list = &transaction->t_buffers;
2269 : 110354 : break;
2270 : : case BJ_Forget:
2271 : 77327 : list = &transaction->t_forget;
2272 : 77327 : break;
2273 : : case BJ_Shadow:
2274 : 75996 : list = &transaction->t_shadow_list;
2275 : 75996 : break;
2276 : : case BJ_Reserved:
2277 : 105600 : list = &transaction->t_reserved_list;
2278 : 105600 : break;
2279 : : }
2280 : :
2281 : : __blist_add_buffer(list, jh);
2282 : 369277 : jh->b_jlist = jlist;
2283 : :
2284 [ + + ]: 369277 : if (was_dirty)
2285 : : set_buffer_jbddirty(bh);
2286 : : }
2287 : :
2288 : 0 : void jbd2_journal_file_buffer(struct journal_head *jh,
2289 : : transaction_t *transaction, int jlist)
2290 : : {
2291 : : jbd_lock_bh_state(jh2bh(jh));
2292 : 75996 : spin_lock(&transaction->t_journal->j_list_lock);
2293 : 75996 : __jbd2_journal_file_buffer(jh, transaction, jlist);
2294 : 75996 : spin_unlock(&transaction->t_journal->j_list_lock);
2295 : : jbd_unlock_bh_state(jh2bh(jh));
2296 : 75996 : }
2297 : :
2298 : : /*
2299 : : * Remove a buffer from its current buffer list in preparation for
2300 : : * dropping it from its current transaction entirely. If the buffer has
2301 : : * already started to be used by a subsequent transaction, refile the
2302 : : * buffer on that transaction's metadata list.
2303 : : *
2304 : : * Called under j_list_lock
2305 : : * Called under jbd_lock_bh_state(jh2bh(jh))
2306 : : *
2307 : : * jh and bh may be already free when this function returns
2308 : : */
2309 : 0 : void __jbd2_journal_refile_buffer(struct journal_head *jh)
2310 : : {
2311 : : int was_dirty, jlist;
2312 : : struct buffer_head *bh = jh2bh(jh);
2313 : :
2314 [ - + ]: 77327 : J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
2315 [ + - ]: 77327 : if (jh->b_transaction)
2316 [ - + ]: 77327 : assert_spin_locked(&jh->b_transaction->t_journal->j_list_lock);
2317 : :
2318 : : /* If the buffer is now unused, just drop it. */
2319 [ + + ]: 77327 : if (jh->b_next_transaction == NULL) {
2320 : : __jbd2_journal_unfile_buffer(jh);
2321 : 77327 : return;
2322 : : }
2323 : :
2324 : : /*
2325 : : * It has been modified by a later transaction: add it to the new
2326 : : * transaction's metadata list.
2327 : : */
2328 : :
2329 : : was_dirty = test_clear_buffer_jbddirty(bh);
2330 : 4767 : __jbd2_journal_temp_unlink_buffer(jh);
2331 : : /*
2332 : : * We set b_transaction here because b_next_transaction will inherit
2333 : : * our jh reference and thus __jbd2_journal_file_buffer() must not
2334 : : * take a new one.
2335 : : */
2336 : 4767 : jh->b_transaction = jh->b_next_transaction;
2337 : 4767 : jh->b_next_transaction = NULL;
2338 [ + + ]: 4767 : if (buffer_freed(bh))
2339 : : jlist = BJ_Forget;
2340 [ + + ]: 4765 : else if (jh->b_modified)
2341 : : jlist = BJ_Metadata;
2342 : : else
2343 : : jlist = BJ_Reserved;
2344 : 4767 : __jbd2_journal_file_buffer(jh, jh->b_transaction, jlist);
2345 [ - + ]: 4767 : J_ASSERT_JH(jh, jh->b_transaction->t_state == T_RUNNING);
2346 : :
2347 [ + - ]: 4767 : if (was_dirty)
2348 : : set_buffer_jbddirty(bh);
2349 : : }
2350 : :
2351 : : /*
2352 : : * __jbd2_journal_refile_buffer() with necessary locking added. We take our
2353 : : * bh reference so that we can safely unlock bh.
2354 : : *
2355 : : * The jh and bh may be freed by this call.
2356 : : */
2357 : 0 : void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh)
2358 : : {
2359 : : struct buffer_head *bh = jh2bh(jh);
2360 : :
2361 : : /* Get reference so that buffer cannot be freed before we unlock it */
2362 : : get_bh(bh);
2363 : : jbd_lock_bh_state(bh);
2364 : : spin_lock(&journal->j_list_lock);
2365 : 0 : __jbd2_journal_refile_buffer(jh);
2366 : : jbd_unlock_bh_state(bh);
2367 : : spin_unlock(&journal->j_list_lock);
2368 : 0 : __brelse(bh);
2369 : 0 : }
2370 : :
2371 : : /*
2372 : : * File inode in the inode list of the handle's transaction
2373 : : */
2374 : 0 : int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode)
2375 : : {
2376 : 14849 : transaction_t *transaction = handle->h_transaction;
2377 : : journal_t *journal;
2378 : :
2379 [ - + ]: 14849 : WARN_ON(!transaction);
2380 [ + - ]: 14849 : if (is_handle_aborted(handle))
2381 : : return -EROFS;
2382 : 14849 : journal = transaction->t_journal;
2383 : :
2384 : : jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode->i_vfs_inode->i_ino,
2385 : : transaction->t_tid);
2386 : :
2387 : : /*
2388 : : * First check whether inode isn't already on the transaction's
2389 : : * lists without taking the lock. Note that this check is safe
2390 : : * without the lock as we cannot race with somebody removing inode
2391 : : * from the transaction. The reason is that we remove inode from the
2392 : : * transaction only in journal_release_jbd_inode() and when we commit
2393 : : * the transaction. We are guarded from the first case by holding
2394 : : * a reference to the inode. We are safe against the second case
2395 : : * because if jinode->i_transaction == transaction, commit code
2396 : : * cannot touch the transaction because we hold reference to it,
2397 : : * and if jinode->i_next_transaction == transaction, commit code
2398 : : * will only file the inode where we want it.
2399 : : */
2400 [ + + ][ + - ]: 14849 : if (jinode->i_transaction == transaction ||
2401 : 1054 : jinode->i_next_transaction == transaction)
2402 : : return 0;
2403 : :
2404 : : spin_lock(&journal->j_list_lock);
2405 : :
2406 [ + - ][ + ]: 1054 : if (jinode->i_transaction == transaction ||
2407 : 1054 : jinode->i_next_transaction == transaction)
2408 : : goto done;
2409 : :
2410 : : /*
2411 : : * We only ever set this variable to 1 so the test is safe. Since
2412 : : * t_need_data_flush is likely to be set, we do the test to save some
2413 : : * cacheline bouncing
2414 : : */
2415 [ + + ]: 15903 : if (!transaction->t_need_data_flush)
2416 : 530 : transaction->t_need_data_flush = 1;
2417 : : /* On some different transaction's list - should be
2418 : : * the committing one */
2419 [ + + ]: 15903 : if (jinode->i_transaction) {
2420 [ - + ]: 19 : J_ASSERT(jinode->i_next_transaction == NULL);
2421 [ - + ]: 19 : J_ASSERT(jinode->i_transaction ==
2422 : : journal->j_committing_transaction);
2423 : 19 : jinode->i_next_transaction = transaction;
2424 : 19 : goto done;
2425 : : }
2426 : : /* Not on any transaction list... */
2427 [ - + ]: 1035 : J_ASSERT(!jinode->i_next_transaction);
2428 : 1035 : jinode->i_transaction = transaction;
2429 : 1035 : list_add(&jinode->i_list, &transaction->t_inode_list);
2430 : : done:
2431 : : spin_unlock(&journal->j_list_lock);
2432 : :
2433 : 1054 : return 0;
2434 : : }
2435 : :
2436 : : /*
2437 : : * File truncate and transaction commit interact with each other in a
2438 : : * non-trivial way. If a transaction writing data block A is
2439 : : * committing, we cannot discard the data by truncate until we have
2440 : : * written them. Otherwise if we crashed after the transaction with
2441 : : * write has committed but before the transaction with truncate has
2442 : : * committed, we could see stale data in block A. This function is a
2443 : : * helper to solve this problem. It starts writeout of the truncated
2444 : : * part in case it is in the committing transaction.
2445 : : *
2446 : : * Filesystem code must call this function when inode is journaled in
2447 : : * ordered mode before truncation happens and after the inode has been
2448 : : * placed on orphan list with the new inode size. The second condition
2449 : : * avoids the race that someone writes new data and we start
2450 : : * committing the transaction after this function has been called but
2451 : : * before a transaction for truncate is started (and furthermore it
2452 : : * allows us to optimize the case where the addition to orphan list
2453 : : * happens in the same transaction as write --- we don't have to write
2454 : : * any data in such case).
2455 : : */
2456 : 0 : int jbd2_journal_begin_ordered_truncate(journal_t *journal,
2457 : : struct jbd2_inode *jinode,
2458 : : loff_t new_size)
2459 : : {
2460 : : transaction_t *inode_trans, *commit_trans;
2461 : : int ret = 0;
2462 : :
2463 : : /* This is a quick check to avoid locking if not necessary */
2464 [ + + ]: 266837 : if (!jinode->i_transaction)
2465 : : goto out;
2466 : : /* Locks are here just to force reading of recent values, it is
2467 : : * enough that the transaction was not committing before we started
2468 : : * a transaction adding the inode to orphan list */
2469 : 16473 : read_lock(&journal->j_state_lock);
2470 : 16473 : commit_trans = journal->j_committing_transaction;
2471 : : read_unlock(&journal->j_state_lock);
2472 : : spin_lock(&journal->j_list_lock);
2473 : 16473 : inode_trans = jinode->i_transaction;
2474 : : spin_unlock(&journal->j_list_lock);
2475 [ + + ]: 16473 : if (inode_trans == commit_trans) {
2476 : 53 : ret = filemap_fdatawrite_range(jinode->i_vfs_inode->i_mapping,
2477 : : new_size, LLONG_MAX);
2478 [ - + ]: 53 : if (ret)
2479 : 0 : jbd2_journal_abort(journal, ret);
2480 : : }
2481 : : out:
2482 : 266837 : return ret;
2483 : : }
|