Branch data Line data Source code
1 : : // SPDX-License-Identifier: GPL-2.0+
2 : : /*
3 : : * linux/fs/jbd2/transaction.c
4 : : *
5 : : * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
6 : : *
7 : : * Copyright 1998 Red Hat corp --- All Rights Reserved
8 : : *
9 : : * Generic filesystem transaction handling code; part of the ext2fs
10 : : * journaling system.
11 : : *
12 : : * This file manages transactions (compound commits managed by the
13 : : * journaling code) and handles (individual atomic operations by the
14 : : * filesystem).
15 : : */
16 : :
17 : : #include <linux/time.h>
18 : : #include <linux/fs.h>
19 : : #include <linux/jbd2.h>
20 : : #include <linux/errno.h>
21 : : #include <linux/slab.h>
22 : : #include <linux/timer.h>
23 : : #include <linux/mm.h>
24 : : #include <linux/highmem.h>
25 : : #include <linux/hrtimer.h>
26 : : #include <linux/backing-dev.h>
27 : : #include <linux/bug.h>
28 : : #include <linux/module.h>
29 : : #include <linux/sched/mm.h>
30 : :
31 : : #include <trace/events/jbd2.h>
32 : :
33 : : static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh);
34 : : static void __jbd2_journal_unfile_buffer(struct journal_head *jh);
35 : :
36 : : static struct kmem_cache *transaction_cache;
37 : 78 : int __init jbd2_journal_init_transaction_cache(void)
38 : : {
39 [ - + ]: 78 : J_ASSERT(!transaction_cache);
40 : 78 : transaction_cache = kmem_cache_create("jbd2_transaction_s",
41 : : sizeof(transaction_t),
42 : : 0,
43 : : SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY,
44 : : NULL);
45 [ - + ]: 78 : if (!transaction_cache) {
46 : 0 : pr_emerg("JBD2: failed to create transaction cache\n");
47 : 0 : return -ENOMEM;
48 : : }
49 : : return 0;
50 : : }
51 : :
52 : 0 : void jbd2_journal_destroy_transaction_cache(void)
53 : : {
54 : 0 : kmem_cache_destroy(transaction_cache);
55 : 0 : transaction_cache = NULL;
56 : 0 : }
57 : :
58 : 170575 : void jbd2_journal_free_transaction(transaction_t *transaction)
59 : : {
60 [ + - ]: 253 : if (unlikely(ZERO_OR_NULL_PTR(transaction)))
61 : : return;
62 : 253 : kmem_cache_free(transaction_cache, transaction);
63 : : }
64 : :
65 : : /*
66 : : * Base amount of descriptor blocks we reserve for each transaction.
67 : : */
68 : 637 : static int jbd2_descriptor_blocks_per_trans(journal_t *journal)
69 : : {
70 : 637 : int tag_space = journal->j_blocksize - sizeof(journal_header_t);
71 : 637 : int tags_per_block;
72 : :
73 : : /* Subtract UUID */
74 : 637 : tag_space -= 16;
75 [ + - ]: 637 : if (jbd2_journal_has_csum_v2or3(journal))
76 : 637 : tag_space -= sizeof(struct jbd2_journal_block_tail);
77 : : /* Commit code leaves a slack space of 16 bytes at the end of block */
78 : 637 : tags_per_block = (tag_space - 16) / journal_tag_bytes(journal);
79 : : /*
80 : : * Revoke descriptors are accounted separately so we need to reserve
81 : : * space for commit block and normal transaction descriptor blocks.
82 : : */
83 : 637 : return 1 + DIV_ROUND_UP(journal->j_max_transaction_buffers,
84 : : tags_per_block);
85 : : }
86 : :
87 : : /*
88 : : * jbd2_get_transaction: obtain a new transaction_t object.
89 : : *
90 : : * Simply initialise a new transaction. Initialize it in
91 : : * RUNNING state and add it to the current journal (which should not
92 : : * have an existing running transaction: we only make a new transaction
93 : : * once we have started to commit the old one).
94 : : *
95 : : * Preconditions:
96 : : * The journal MUST be locked. We don't perform atomic mallocs on the
97 : : * new transaction and we can't block without protecting against other
98 : : * processes trying to touch the journal while it is in transition.
99 : : *
100 : : */
101 : :
102 : 637 : static void jbd2_get_transaction(journal_t *journal,
103 : : transaction_t *transaction)
104 : : {
105 : 637 : transaction->t_journal = journal;
106 : 637 : transaction->t_state = T_RUNNING;
107 : 637 : transaction->t_start_time = ktime_get();
108 : 637 : transaction->t_tid = journal->j_transaction_sequence++;
109 : 637 : transaction->t_expires = jiffies + journal->j_commit_interval;
110 : 637 : spin_lock_init(&transaction->t_handle_lock);
111 : 637 : atomic_set(&transaction->t_updates, 0);
112 : 637 : atomic_set(&transaction->t_outstanding_credits,
113 : 637 : jbd2_descriptor_blocks_per_trans(journal) +
114 : 637 : atomic_read(&journal->j_reserved_credits));
115 : 637 : atomic_set(&transaction->t_outstanding_revokes, 0);
116 : 637 : atomic_set(&transaction->t_handle_count, 0);
117 : 637 : INIT_LIST_HEAD(&transaction->t_inode_list);
118 : 637 : INIT_LIST_HEAD(&transaction->t_private_list);
119 : :
120 : : /* Set up the commit timer for the new transaction. */
121 : 637 : journal->j_commit_timer.expires = round_jiffies_up(transaction->t_expires);
122 : 637 : add_timer(&journal->j_commit_timer);
123 : :
124 [ - + ]: 637 : J_ASSERT(journal->j_running_transaction == NULL);
125 : 637 : journal->j_running_transaction = transaction;
126 : 637 : transaction->t_max_wait = 0;
127 : 637 : transaction->t_start = jiffies;
128 : 637 : transaction->t_requested = 0;
129 : 637 : }
130 : :
131 : : /*
132 : : * Handle management.
133 : : *
134 : : * A handle_t is an object which represents a single atomic update to a
135 : : * filesystem, and which tracks all of the modifications which form part
136 : : * of that one update.
137 : : */
138 : :
139 : : /*
140 : : * Update transaction's maximum wait time, if debugging is enabled.
141 : : *
142 : : * In order for t_max_wait to be reliable, it must be protected by a
143 : : * lock. But doing so will mean that start_this_handle() can not be
144 : : * run in parallel on SMP systems, which limits our scalability. So
145 : : * unless debugging is enabled, we no longer update t_max_wait, which
146 : : * means that maximum wait time reported by the jbd2_run_stats
147 : : * tracepoint will always be zero.
148 : : */
149 : 170322 : static inline void update_t_max_wait(transaction_t *transaction,
150 : : unsigned long ts)
151 : : {
152 : : #ifdef CONFIG_JBD2_DEBUG
153 : : if (jbd2_journal_enable_debug &&
154 : : time_after(transaction->t_start, ts)) {
155 : : ts = jbd2_time_diff(ts, transaction->t_start);
156 : : spin_lock(&transaction->t_handle_lock);
157 : : if (ts > transaction->t_max_wait)
158 : : transaction->t_max_wait = ts;
159 : : spin_unlock(&transaction->t_handle_lock);
160 : : }
161 : : #endif
162 : 170322 : }
163 : :
164 : : /*
165 : : * Wait until running transaction passes to T_FLUSH state and new transaction
166 : : * can thus be started. Also starts the commit if needed. The function expects
167 : : * running transaction to exist and releases j_state_lock.
168 : : */
169 : 0 : static void wait_transaction_locked(journal_t *journal)
170 : : __releases(journal->j_state_lock)
171 : : {
172 : 0 : DEFINE_WAIT(wait);
173 : 0 : int need_to_start;
174 : 0 : tid_t tid = journal->j_running_transaction->t_tid;
175 : :
176 : 0 : prepare_to_wait(&journal->j_wait_transaction_locked, &wait,
177 : : TASK_UNINTERRUPTIBLE);
178 : 0 : need_to_start = !tid_geq(journal->j_commit_request, tid);
179 : 0 : read_unlock(&journal->j_state_lock);
180 [ # # ]: 0 : if (need_to_start)
181 : 0 : jbd2_log_start_commit(journal, tid);
182 : 0 : jbd2_might_wait_for_commit(journal);
183 : 0 : schedule();
184 : 0 : finish_wait(&journal->j_wait_transaction_locked, &wait);
185 : 0 : }
186 : :
187 : : /*
188 : : * Wait until running transaction transitions from T_SWITCH to T_FLUSH
189 : : * state and new transaction can thus be started. The function releases
190 : : * j_state_lock.
191 : : */
192 : 0 : static void wait_transaction_switching(journal_t *journal)
193 : : __releases(journal->j_state_lock)
194 : : {
195 [ # # ]: 0 : DEFINE_WAIT(wait);
196 : :
197 [ # # # # : 0 : if (WARN_ON(!journal->j_running_transaction ||
# # # # ]
198 : : journal->j_running_transaction->t_state != T_SWITCH))
199 : 0 : return;
200 : 0 : prepare_to_wait(&journal->j_wait_transaction_locked, &wait,
201 : : TASK_UNINTERRUPTIBLE);
202 : 0 : read_unlock(&journal->j_state_lock);
203 : : /*
204 : : * We don't call jbd2_might_wait_for_commit() here as there's no
205 : : * waiting for outstanding handles happening anymore in T_SWITCH state
206 : : * and handling of reserved handles actually relies on that for
207 : : * correctness.
208 : : */
209 : 0 : schedule();
210 : 0 : finish_wait(&journal->j_wait_transaction_locked, &wait);
211 : : }
212 : :
213 : 1248 : static void sub_reserved_credits(journal_t *journal, int blocks)
214 : : {
215 : 1248 : atomic_sub(blocks, &journal->j_reserved_credits);
216 : 1248 : wake_up(&journal->j_wait_reserved);
217 : 1248 : }
218 : :
219 : : /*
220 : : * Wait until we can add credits for handle to the running transaction. Called
221 : : * with j_state_lock held for reading. Returns 0 if handle joined the running
222 : : * transaction. Returns 1 if we had to wait, j_state_lock is dropped, and
223 : : * caller must retry.
224 : : */
225 : 169698 : static int add_transaction_credits(journal_t *journal, int blocks,
226 : : int rsv_blocks)
227 : : {
228 : 169698 : transaction_t *t = journal->j_running_transaction;
229 : 169698 : int needed;
230 : 169698 : int total = blocks + rsv_blocks;
231 : :
232 : : /*
233 : : * If the current transaction is locked down for commit, wait
234 : : * for the lock to be released.
235 : : */
236 [ - + ]: 169698 : if (t->t_state != T_RUNNING) {
237 [ # # ]: 0 : WARN_ON_ONCE(t->t_state >= T_FLUSH);
238 : 0 : wait_transaction_locked(journal);
239 : 0 : return 1;
240 : : }
241 : :
242 : : /*
243 : : * If there is not enough space left in the log to write all
244 : : * potential buffers requested by this operation, we need to
245 : : * stall pending a log checkpoint to free some more log space.
246 : : */
247 : 169698 : needed = atomic_add_return(total, &t->t_outstanding_credits);
248 [ - + ]: 169698 : if (needed > journal->j_max_transaction_buffers) {
249 : : /*
250 : : * If the current transaction is already too large,
251 : : * then start to commit it: we can then go back and
252 : : * attach this handle to a new transaction.
253 : : */
254 : 0 : atomic_sub(total, &t->t_outstanding_credits);
255 : :
256 : : /*
257 : : * Is the number of reserved credits in the current transaction too
258 : : * big to fit this handle? Wait until reserved credits are freed.
259 : : */
260 : 0 : if (atomic_read(&journal->j_reserved_credits) + total >
261 [ # # ]: 0 : journal->j_max_transaction_buffers) {
262 : 0 : read_unlock(&journal->j_state_lock);
263 : 0 : jbd2_might_wait_for_commit(journal);
264 [ # # # # ]: 0 : wait_event(journal->j_wait_reserved,
265 : : atomic_read(&journal->j_reserved_credits) + total <=
266 : : journal->j_max_transaction_buffers);
267 : 0 : return 1;
268 : : }
269 : :
270 : 0 : wait_transaction_locked(journal);
271 : 0 : return 1;
272 : : }
273 : :
274 : : /*
275 : : * The commit code assumes that it can get enough log space
276 : : * without forcing a checkpoint. This is *critical* for
277 : : * correctness: a checkpoint of a buffer which is also
278 : : * associated with a committing transaction creates a deadlock,
279 : : * so commit simply cannot force through checkpoints.
280 : : *
281 : : * We must therefore ensure the necessary space in the journal
282 : : * *before* starting to dirty potentially checkpointed buffers
283 : : * in the new transaction.
284 : : */
285 [ + + - + ]: 169792 : if (jbd2_log_space_left(journal) < journal->j_max_transaction_buffers) {
286 : 0 : atomic_sub(total, &t->t_outstanding_credits);
287 : 0 : read_unlock(&journal->j_state_lock);
288 : 0 : jbd2_might_wait_for_commit(journal);
289 : 0 : write_lock(&journal->j_state_lock);
290 [ # # ]: 0 : if (jbd2_log_space_left(journal) <
291 [ # # ]: 0 : journal->j_max_transaction_buffers)
292 : 0 : __jbd2_log_wait_for_space(journal);
293 : 0 : write_unlock(&journal->j_state_lock);
294 : 0 : return 1;
295 : : }
296 : :
297 : : /* No reservation? We are done... */
298 [ + + ]: 169698 : if (!rsv_blocks)
299 : : return 0;
300 : :
301 : 1248 : needed = atomic_add_return(rsv_blocks, &journal->j_reserved_credits);
302 : : /* We allow at most half of a transaction to be reserved */
303 [ - + ]: 1248 : if (needed > journal->j_max_transaction_buffers / 2) {
304 : 0 : sub_reserved_credits(journal, rsv_blocks);
305 : 0 : atomic_sub(total, &t->t_outstanding_credits);
306 : 0 : read_unlock(&journal->j_state_lock);
307 : 0 : jbd2_might_wait_for_commit(journal);
308 [ # # # # ]: 0 : wait_event(journal->j_wait_reserved,
309 : : atomic_read(&journal->j_reserved_credits) + rsv_blocks
310 : : <= journal->j_max_transaction_buffers / 2);
311 : 0 : return 1;
312 : : }
313 : : return 0;
314 : : }
315 : :
316 : : /*
317 : : * start_this_handle: Given a handle, deal with any locking or stalling
318 : : * needed to make sure that there is enough journal space for the handle
319 : : * to begin. Attach the handle to a transaction and set up the
320 : : * transaction's buffer credits.
321 : : */
322 : :
323 : 170322 : static int start_this_handle(journal_t *journal, handle_t *handle,
324 : : gfp_t gfp_mask)
325 : : {
326 : 170322 : transaction_t *transaction, *new_transaction = NULL;
327 : 170322 : int blocks = handle->h_total_credits;
328 : 170322 : int rsv_blocks = 0;
329 : 170322 : unsigned long ts = jiffies;
330 : :
331 [ + + ]: 170322 : if (handle->h_rsv_handle)
332 : 1248 : rsv_blocks = handle->h_rsv_handle->h_total_credits;
333 : :
334 : : /*
335 : : * Limit the number of reserved credits to 1/2 of maximum transaction
336 : : * size and limit the number of total credits to not exceed maximum
337 : : * transaction size per operation.
338 : : */
339 [ + - ]: 170322 : if ((rsv_blocks > journal->j_max_transaction_buffers / 2) ||
340 [ - + ]: 170322 : (rsv_blocks + blocks > journal->j_max_transaction_buffers)) {
341 : 0 : printk(KERN_ERR "JBD2: %s wants too many credits "
342 : : "credits:%d rsv_credits:%d max:%d\n",
343 : 0 : current->comm, blocks, rsv_blocks,
344 : : journal->j_max_transaction_buffers);
345 : 0 : WARN_ON(1);
346 : 0 : return -ENOSPC;
347 : : }
348 : :
349 : 170322 : alloc_transaction:
350 [ + + ]: 170322 : if (!journal->j_running_transaction) {
351 : : /*
352 : : * If __GFP_FS is not present, then we may be being called from
353 : : * inside the fs writeback layer, so we MUST NOT fail.
354 : : */
355 [ + - ]: 637 : if ((gfp_mask & __GFP_FS) == 0)
356 : 637 : gfp_mask |= __GFP_NOFAIL;
357 : 637 : new_transaction = kmem_cache_zalloc(transaction_cache,
358 : : gfp_mask);
359 [ + - ]: 637 : if (!new_transaction)
360 : : return -ENOMEM;
361 : : }
362 : :
363 : 170959 : jbd_debug(3, "New handle %p going live.\n", handle);
364 : :
365 : : /*
366 : : * We need to hold j_state_lock until t_updates has been incremented,
367 : : * for proper journal barrier handling
368 : : */
369 : 170322 : repeat:
370 : 170959 : read_lock(&journal->j_state_lock);
371 [ - + ]: 170959 : BUG_ON(journal->j_flags & JBD2_UNMOUNT);
372 [ + - ]: 170959 : if (is_journal_aborted(journal) ||
373 [ - + - - ]: 170959 : (journal->j_errno != 0 && !(journal->j_flags & JBD2_ACK_ERR))) {
374 : 0 : read_unlock(&journal->j_state_lock);
375 [ # # ]: 0 : jbd2_journal_free_transaction(new_transaction);
376 : 0 : return -EROFS;
377 : : }
378 : :
379 : : /*
380 : : * Wait on the journal's transaction barrier if necessary. Specifically
381 : : * we allow reserved handles to proceed because otherwise commit could
382 : : * deadlock on page writeback not being able to complete.
383 : : */
384 [ + + - + ]: 170959 : if (!handle->h_reserved && journal->j_barrier_count) {
385 : 0 : read_unlock(&journal->j_state_lock);
386 [ # # # # ]: 0 : wait_event(journal->j_wait_transaction_locked,
387 : : journal->j_barrier_count == 0);
388 : 0 : goto repeat;
389 : : }
390 : :
391 [ + + ]: 170959 : if (!journal->j_running_transaction) {
392 : 637 : read_unlock(&journal->j_state_lock);
393 [ - + ]: 637 : if (!new_transaction)
394 : 0 : goto alloc_transaction;
395 : 637 : write_lock(&journal->j_state_lock);
396 [ + - ]: 637 : if (!journal->j_running_transaction &&
397 [ + - + - ]: 637 : (handle->h_reserved || !journal->j_barrier_count)) {
398 : 637 : jbd2_get_transaction(journal, new_transaction);
399 : 637 : new_transaction = NULL;
400 : : }
401 : 637 : write_unlock(&journal->j_state_lock);
402 : 637 : goto repeat;
403 : : }
404 : :
405 : 170322 : transaction = journal->j_running_transaction;
406 : :
407 [ + + ]: 170322 : if (!handle->h_reserved) {
408 : : /* We may have dropped j_state_lock - restart in that case */
409 [ - + ]: 169698 : if (add_transaction_credits(journal, blocks, rsv_blocks))
410 : 0 : goto repeat;
411 : : } else {
412 : : /*
413 : : * We have handle reserved so we are allowed to join T_LOCKED
414 : : * transaction and we don't have to check for transaction size
415 : : * and journal space. But we still have to wait while running
416 : : * transaction is being switched to a committing one as it
417 : : * won't wait for any handles anymore.
418 : : */
419 [ - + ]: 624 : if (transaction->t_state == T_SWITCH) {
420 : 0 : wait_transaction_switching(journal);
421 : 0 : goto repeat;
422 : : }
423 : 624 : sub_reserved_credits(journal, blocks);
424 : 624 : handle->h_reserved = 0;
425 : : }
426 : :
427 : : /* OK, account for the buffers that this operation expects to
428 : : * use and add the handle to the running transaction.
429 : : */
430 : 170322 : update_t_max_wait(transaction, ts);
431 : 170322 : handle->h_transaction = transaction;
432 : 170322 : handle->h_requested_credits = blocks;
433 : 170322 : handle->h_revoke_credits_requested = handle->h_revoke_credits;
434 : 170322 : handle->h_start_jiffies = jiffies;
435 : 170322 : atomic_inc(&transaction->t_updates);
436 : 170322 : atomic_inc(&transaction->t_handle_count);
437 : : jbd_debug(4, "Handle %p given %d credits (total %d, free %lu)\n",
438 : : handle, blocks,
439 : : atomic_read(&transaction->t_outstanding_credits),
440 : 170322 : jbd2_log_space_left(journal));
441 : 170322 : read_unlock(&journal->j_state_lock);
442 [ - + ]: 170322 : current->journal_info = handle;
443 : :
444 : 170322 : rwsem_acquire_read(&journal->j_trans_commit_map, 0, 0, _THIS_IP_);
445 [ - + ]: 170322 : jbd2_journal_free_transaction(new_transaction);
446 : : /*
447 : : * Ensure that no allocations done while the transaction is open are
448 : : * going to recurse back to the fs layer.
449 : : */
450 : 170322 : handle->saved_alloc_context = memalloc_nofs_save();
451 : 170322 : return 0;
452 : : }
453 : :
454 : : /* Allocate a new handle. This should probably be in a slab... */
455 : 170946 : static handle_t *new_handle(int nblocks)
456 : : {
457 : 170946 : handle_t *handle = jbd2_alloc_handle(GFP_NOFS);
458 [ + - - + ]: 170946 : if (!handle)
459 : : return NULL;
460 : 170946 : handle->h_total_credits = nblocks;
461 : 170946 : handle->h_ref = 1;
462 : :
463 : 170946 : return handle;
464 : : }
465 : :
466 : 257167 : handle_t *jbd2__journal_start(journal_t *journal, int nblocks, int rsv_blocks,
467 : : int revoke_records, gfp_t gfp_mask,
468 : : unsigned int type, unsigned int line_no)
469 : : {
470 [ + - ]: 257167 : handle_t *handle = journal_current_handle();
471 : 257167 : int err;
472 : :
473 [ + - ]: 257167 : if (!journal)
474 : : return ERR_PTR(-EROFS);
475 : :
476 [ + + ]: 257167 : if (handle) {
477 [ - + ]: 87469 : J_ASSERT(handle->h_transaction->t_journal == journal);
478 : 87469 : handle->h_ref++;
479 : 87469 : return handle;
480 : : }
481 : :
482 : 169698 : nblocks += DIV_ROUND_UP(revoke_records,
483 : : journal->j_revoke_records_per_block);
484 : 169698 : handle = new_handle(nblocks);
485 : 169698 : if (!handle)
486 : : return ERR_PTR(-ENOMEM);
487 [ + + ]: 169698 : if (rsv_blocks) {
488 : 1248 : handle_t *rsv_handle;
489 : :
490 : 1248 : rsv_handle = new_handle(rsv_blocks);
491 : 1248 : if (!rsv_handle) {
492 : 0 : jbd2_free_handle(handle);
493 : 0 : return ERR_PTR(-ENOMEM);
494 : : }
495 : 1248 : rsv_handle->h_reserved = 1;
496 : 1248 : rsv_handle->h_journal = journal;
497 : 1248 : handle->h_rsv_handle = rsv_handle;
498 : : }
499 : 169698 : handle->h_revoke_credits = revoke_records;
500 : :
501 : 169698 : err = start_this_handle(journal, handle, gfp_mask);
502 [ - + ]: 169698 : if (err < 0) {
503 [ # # ]: 0 : if (handle->h_rsv_handle)
504 : 0 : jbd2_free_handle(handle->h_rsv_handle);
505 : 0 : jbd2_free_handle(handle);
506 : 0 : return ERR_PTR(err);
507 : : }
508 : 169698 : handle->h_type = type;
509 : 169698 : handle->h_line_no = line_no;
510 : 169698 : trace_jbd2_handle_start(journal->j_fs_dev->bd_dev,
511 : 169698 : handle->h_transaction->t_tid, type,
512 : : line_no, nblocks);
513 : :
514 : 169698 : return handle;
515 : : }
516 : : EXPORT_SYMBOL(jbd2__journal_start);
517 : :
518 : :
519 : : /**
520 : : * handle_t *jbd2_journal_start() - Obtain a new handle.
521 : : * @journal: Journal to start transaction on.
522 : : * @nblocks: number of block buffer we might modify
523 : : *
524 : : * We make sure that the transaction can guarantee at least nblocks of
525 : : * modified buffers in the log. We block until the log can guarantee
526 : : * that much space. Additionally, if rsv_blocks > 0, we also create another
527 : : * handle with rsv_blocks reserved blocks in the journal. This handle is
528 : : * stored in h_rsv_handle. It is not attached to any particular transaction
529 : : * and thus doesn't block transaction commit. If the caller uses this reserved
530 : : * handle, it has to set h_rsv_handle to NULL as otherwise jbd2_journal_stop()
531 : : * on the parent handle will dispose the reserved one. Reserved handle has to
532 : : * be converted to a normal handle using jbd2_journal_start_reserved() before
533 : : * it can be used.
534 : : *
535 : : * Return a pointer to a newly allocated handle, or an ERR_PTR() value
536 : : * on failure.
537 : : */
538 : 0 : handle_t *jbd2_journal_start(journal_t *journal, int nblocks)
539 : : {
540 : 0 : return jbd2__journal_start(journal, nblocks, 0, 0, GFP_NOFS, 0, 0);
541 : : }
542 : : EXPORT_SYMBOL(jbd2_journal_start);
543 : :
544 : 624 : static void __jbd2_journal_unreserve_handle(handle_t *handle)
545 : : {
546 : 624 : journal_t *journal = handle->h_journal;
547 : :
548 [ - + ]: 624 : WARN_ON(!handle->h_reserved);
549 : 624 : sub_reserved_credits(journal, handle->h_total_credits);
550 : 624 : }
551 : :
552 : 0 : void jbd2_journal_free_reserved(handle_t *handle)
553 : : {
554 : 0 : __jbd2_journal_unreserve_handle(handle);
555 : 0 : jbd2_free_handle(handle);
556 : 0 : }
557 : : EXPORT_SYMBOL(jbd2_journal_free_reserved);
558 : :
559 : : /**
560 : : * int jbd2_journal_start_reserved() - start reserved handle
561 : : * @handle: handle to start
562 : : * @type: for handle statistics
563 : : * @line_no: for handle statistics
564 : : *
565 : : * Start handle that has been previously reserved with jbd2_journal_reserve().
566 : : * This attaches @handle to the running transaction (or creates one if there's
567 : : * not transaction running). Unlike jbd2_journal_start() this function cannot
568 : : * block on journal commit, checkpointing, or similar stuff. It can block on
569 : : * memory allocation or frozen journal though.
570 : : *
571 : : * Return 0 on success, non-zero on error - handle is freed in that case.
572 : : */
573 : 624 : int jbd2_journal_start_reserved(handle_t *handle, unsigned int type,
574 : : unsigned int line_no)
575 : : {
576 : 624 : journal_t *journal = handle->h_journal;
577 : 624 : int ret = -EIO;
578 : :
579 [ - + - + ]: 624 : if (WARN_ON(!handle->h_reserved)) {
580 : : /* Someone passed in normal handle? Just stop it. */
581 : 0 : jbd2_journal_stop(handle);
582 : 0 : return ret;
583 : : }
584 : : /*
585 : : * Usefulness of mixing of reserved and unreserved handles is
586 : : * questionable. So far nobody seems to need it so just error out.
587 : : */
588 [ - + - + ]: 624 : if (WARN_ON(current->journal_info)) {
589 : 0 : jbd2_journal_free_reserved(handle);
590 : 0 : return ret;
591 : : }
592 : :
593 : 624 : handle->h_journal = NULL;
594 : : /*
595 : : * GFP_NOFS is here because callers are likely from writeback or
596 : : * similarly constrained call sites
597 : : */
598 : 624 : ret = start_this_handle(journal, handle, GFP_NOFS);
599 [ - + ]: 624 : if (ret < 0) {
600 : 0 : handle->h_journal = journal;
601 : 0 : jbd2_journal_free_reserved(handle);
602 : 0 : return ret;
603 : : }
604 : 624 : handle->h_type = type;
605 : 624 : handle->h_line_no = line_no;
606 : 624 : trace_jbd2_handle_start(journal->j_fs_dev->bd_dev,
607 : 624 : handle->h_transaction->t_tid, type,
608 : : line_no, handle->h_total_credits);
609 : 624 : return 0;
610 : : }
611 : : EXPORT_SYMBOL(jbd2_journal_start_reserved);
612 : :
613 : : /**
614 : : * int jbd2_journal_extend() - extend buffer credits.
615 : : * @handle: handle to 'extend'
616 : : * @nblocks: nr blocks to try to extend by.
617 : : * @revoke_records: number of revoke records to try to extend by.
618 : : *
619 : : * Some transactions, such as large extends and truncates, can be done
620 : : * atomically all at once or in several stages. The operation requests
621 : : * a credit for a number of buffer modifications in advance, but can
622 : : * extend its credit if it needs more.
623 : : *
624 : : * jbd2_journal_extend tries to give the running handle more buffer credits.
625 : : * It does not guarantee that allocation - this is a best-effort only.
626 : : * The calling process MUST be able to deal cleanly with a failure to
627 : : * extend here.
628 : : *
629 : : * Return 0 on success, non-zero on failure.
630 : : *
631 : : * return code < 0 implies an error
632 : : * return code > 0 implies normal transaction-full status.
633 : : */
634 : 0 : int jbd2_journal_extend(handle_t *handle, int nblocks, int revoke_records)
635 : : {
636 : 0 : transaction_t *transaction = handle->h_transaction;
637 : 0 : journal_t *journal;
638 : 0 : int result;
639 : 0 : int wanted;
640 : :
641 [ # # # # ]: 0 : if (is_handle_aborted(handle))
642 : : return -EROFS;
643 : 0 : journal = transaction->t_journal;
644 : :
645 : 0 : result = 1;
646 : :
647 : 0 : read_lock(&journal->j_state_lock);
648 : :
649 : : /* Don't extend a locked-down transaction! */
650 [ # # ]: 0 : if (transaction->t_state != T_RUNNING) {
651 : : jbd_debug(3, "denied handle %p %d blocks: "
652 : 0 : "transaction not running\n", handle, nblocks);
653 : 0 : goto error_out;
654 : : }
655 : :
656 : 0 : nblocks += DIV_ROUND_UP(
657 : : handle->h_revoke_credits_requested + revoke_records,
658 : 0 : journal->j_revoke_records_per_block) -
659 : 0 : DIV_ROUND_UP(
660 : : handle->h_revoke_credits_requested,
661 : : journal->j_revoke_records_per_block);
662 : 0 : spin_lock(&transaction->t_handle_lock);
663 : 0 : wanted = atomic_add_return(nblocks,
664 : : &transaction->t_outstanding_credits);
665 : :
666 [ # # ]: 0 : if (wanted > journal->j_max_transaction_buffers) {
667 : : jbd_debug(3, "denied handle %p %d blocks: "
668 : 0 : "transaction too large\n", handle, nblocks);
669 : 0 : atomic_sub(nblocks, &transaction->t_outstanding_credits);
670 : 0 : goto unlock;
671 : : }
672 : :
673 : 0 : trace_jbd2_handle_extend(journal->j_fs_dev->bd_dev,
674 : 0 : transaction->t_tid,
675 : 0 : handle->h_type, handle->h_line_no,
676 : : handle->h_total_credits,
677 : : nblocks);
678 : :
679 : 0 : handle->h_total_credits += nblocks;
680 : 0 : handle->h_requested_credits += nblocks;
681 : 0 : handle->h_revoke_credits += revoke_records;
682 : 0 : handle->h_revoke_credits_requested += revoke_records;
683 : 0 : result = 0;
684 : :
685 : 0 : jbd_debug(3, "extended handle %p by %d\n", handle, nblocks);
686 : 0 : unlock:
687 : 0 : spin_unlock(&transaction->t_handle_lock);
688 : 0 : error_out:
689 : 0 : read_unlock(&journal->j_state_lock);
690 : 0 : return result;
691 : : }
692 : :
693 : 170322 : static void stop_this_handle(handle_t *handle)
694 : : {
695 : 170322 : transaction_t *transaction = handle->h_transaction;
696 : 170322 : journal_t *journal = transaction->t_journal;
697 : 170322 : int revokes;
698 : :
699 [ - + ]: 170322 : J_ASSERT(journal_current_handle() == handle);
700 [ - + ]: 170322 : J_ASSERT(atomic_read(&transaction->t_updates) > 0);
701 [ + + ]: 170322 : current->journal_info = NULL;
702 : : /*
703 : : * Subtract necessary revoke descriptor blocks from handle credits. We
704 : : * take care to account only for revoke descriptor blocks the
705 : : * transaction will really need as large sequences of transactions with
706 : : * small numbers of revokes are relatively common.
707 : : */
708 : 170322 : revokes = handle->h_revoke_credits_requested - handle->h_revoke_credits;
709 [ + + ]: 170322 : if (revokes) {
710 : 156 : int t_revokes, revoke_descriptors;
711 : 156 : int rr_per_blk = journal->j_revoke_records_per_block;
712 : :
713 [ - + ]: 156 : WARN_ON_ONCE(DIV_ROUND_UP(revokes, rr_per_blk)
714 : : > handle->h_total_credits);
715 : 156 : t_revokes = atomic_add_return(revokes,
716 : : &transaction->t_outstanding_revokes);
717 : 156 : revoke_descriptors =
718 : 156 : DIV_ROUND_UP(t_revokes, rr_per_blk) -
719 : 156 : DIV_ROUND_UP(t_revokes - revokes, rr_per_blk);
720 : 156 : handle->h_total_credits -= revoke_descriptors;
721 : : }
722 : 170322 : atomic_sub(handle->h_total_credits,
723 : : &transaction->t_outstanding_credits);
724 [ + + ]: 170322 : if (handle->h_rsv_handle)
725 : 624 : __jbd2_journal_unreserve_handle(handle->h_rsv_handle);
726 [ + + ]: 170322 : if (atomic_dec_and_test(&transaction->t_updates))
727 : 170106 : wake_up(&journal->j_wait_updates);
728 : :
729 : 170322 : rwsem_release(&journal->j_trans_commit_map, _THIS_IP_);
730 : : /*
731 : : * Scope of the GFP_NOFS context is over here and so we can restore the
732 : : * original alloc context.
733 : : */
734 : 170322 : memalloc_nofs_restore(handle->saved_alloc_context);
735 : 170322 : }
736 : :
737 : : /**
738 : : * int jbd2_journal_restart() - restart a handle .
739 : : * @handle: handle to restart
740 : : * @nblocks: nr credits requested
741 : : * @revoke_records: number of revoke record credits requested
742 : : * @gfp_mask: memory allocation flags (for start_this_handle)
743 : : *
744 : : * Restart a handle for a multi-transaction filesystem
745 : : * operation.
746 : : *
747 : : * If the jbd2_journal_extend() call above fails to grant new buffer credits
748 : : * to a running handle, a call to jbd2_journal_restart will commit the
749 : : * handle's transaction so far and reattach the handle to a new
750 : : * transaction capable of guaranteeing the requested number of
751 : : * credits. We preserve reserved handle if there's any attached to the
752 : : * passed in handle.
753 : : */
754 : 0 : int jbd2__journal_restart(handle_t *handle, int nblocks, int revoke_records,
755 : : gfp_t gfp_mask)
756 : : {
757 : 0 : transaction_t *transaction = handle->h_transaction;
758 : 0 : journal_t *journal;
759 : 0 : tid_t tid;
760 : 0 : int need_to_start;
761 : 0 : int ret;
762 : :
763 : : /* If we've had an abort of any type, don't even think about
764 : : * actually doing the restart! */
765 [ # # # # ]: 0 : if (is_handle_aborted(handle))
766 : : return 0;
767 : 0 : journal = transaction->t_journal;
768 : 0 : tid = transaction->t_tid;
769 : :
770 : : /*
771 : : * First unlink the handle from its current transaction, and start the
772 : : * commit on that.
773 : : */
774 : 0 : jbd_debug(2, "restarting handle %p\n", handle);
775 : 0 : stop_this_handle(handle);
776 : 0 : handle->h_transaction = NULL;
777 : :
778 : : /*
779 : : * TODO: If we use READ_ONCE / WRITE_ONCE for j_commit_request we can
780 : : * get rid of pointless j_state_lock traffic like this.
781 : : */
782 : 0 : read_lock(&journal->j_state_lock);
783 : 0 : need_to_start = !tid_geq(journal->j_commit_request, tid);
784 : 0 : read_unlock(&journal->j_state_lock);
785 [ # # ]: 0 : if (need_to_start)
786 : 0 : jbd2_log_start_commit(journal, tid);
787 : 0 : handle->h_total_credits = nblocks +
788 : 0 : DIV_ROUND_UP(revoke_records,
789 : : journal->j_revoke_records_per_block);
790 : 0 : handle->h_revoke_credits = revoke_records;
791 : 0 : ret = start_this_handle(journal, handle, gfp_mask);
792 : 0 : trace_jbd2_handle_restart(journal->j_fs_dev->bd_dev,
793 : 0 : ret ? 0 : handle->h_transaction->t_tid,
794 [ # # ]: 0 : handle->h_type, handle->h_line_no,
795 : : handle->h_total_credits);
796 : 0 : return ret;
797 : : }
798 : : EXPORT_SYMBOL(jbd2__journal_restart);
799 : :
800 : :
801 : 0 : int jbd2_journal_restart(handle_t *handle, int nblocks)
802 : : {
803 : 0 : return jbd2__journal_restart(handle, nblocks, 0, GFP_NOFS);
804 : : }
805 : : EXPORT_SYMBOL(jbd2_journal_restart);
806 : :
807 : : /**
808 : : * void jbd2_journal_lock_updates () - establish a transaction barrier.
809 : : * @journal: Journal to establish a barrier on.
810 : : *
811 : : * This locks out any further updates from being started, and blocks
812 : : * until all existing updates have completed, returning only once the
813 : : * journal is in a quiescent state with no updates running.
814 : : *
815 : : * The journal lock should not be held on entry.
816 : : */
817 : 0 : void jbd2_journal_lock_updates(journal_t *journal)
818 : : {
819 : 0 : DEFINE_WAIT(wait);
820 : :
821 : 0 : jbd2_might_wait_for_commit(journal);
822 : :
823 : 0 : write_lock(&journal->j_state_lock);
824 : 0 : ++journal->j_barrier_count;
825 : :
826 : : /* Wait until there are no reserved handles */
827 [ # # ]: 0 : if (atomic_read(&journal->j_reserved_credits)) {
828 : 0 : write_unlock(&journal->j_state_lock);
829 [ # # # # ]: 0 : wait_event(journal->j_wait_reserved,
830 : : atomic_read(&journal->j_reserved_credits) == 0);
831 : 0 : write_lock(&journal->j_state_lock);
832 : : }
833 : :
834 : : /* Wait until there are no running updates */
835 : 0 : while (1) {
836 : 0 : transaction_t *transaction = journal->j_running_transaction;
837 : :
838 [ # # ]: 0 : if (!transaction)
839 : : break;
840 : :
841 : 0 : spin_lock(&transaction->t_handle_lock);
842 : 0 : prepare_to_wait(&journal->j_wait_updates, &wait,
843 : : TASK_UNINTERRUPTIBLE);
844 [ # # ]: 0 : if (!atomic_read(&transaction->t_updates)) {
845 : 0 : spin_unlock(&transaction->t_handle_lock);
846 : 0 : finish_wait(&journal->j_wait_updates, &wait);
847 : 0 : break;
848 : : }
849 : 0 : spin_unlock(&transaction->t_handle_lock);
850 : 0 : write_unlock(&journal->j_state_lock);
851 : 0 : schedule();
852 : 0 : finish_wait(&journal->j_wait_updates, &wait);
853 : 0 : write_lock(&journal->j_state_lock);
854 : : }
855 : 0 : write_unlock(&journal->j_state_lock);
856 : :
857 : : /*
858 : : * We have now established a barrier against other normal updates, but
859 : : * we also need to barrier against other jbd2_journal_lock_updates() calls
860 : : * to make sure that we serialise special journal-locked operations
861 : : * too.
862 : : */
863 : 0 : mutex_lock(&journal->j_barrier);
864 : 0 : }
865 : :
866 : : /**
867 : : * void jbd2_journal_unlock_updates (journal_t* journal) - release barrier
868 : : * @journal: Journal to release the barrier on.
869 : : *
870 : : * Release a transaction barrier obtained with jbd2_journal_lock_updates().
871 : : *
872 : : * Should be called without the journal lock held.
873 : : */
874 : 0 : void jbd2_journal_unlock_updates (journal_t *journal)
875 : : {
876 [ # # ]: 0 : J_ASSERT(journal->j_barrier_count != 0);
877 : :
878 : 0 : mutex_unlock(&journal->j_barrier);
879 : 0 : write_lock(&journal->j_state_lock);
880 : 0 : --journal->j_barrier_count;
881 : 0 : write_unlock(&journal->j_state_lock);
882 : 0 : wake_up(&journal->j_wait_transaction_locked);
883 : 0 : }
884 : :
885 : : static void warn_dirty_buffer(struct buffer_head *bh)
886 : : {
887 : : printk(KERN_WARNING
888 : : "JBD2: Spotted dirty metadata buffer (dev = %pg, blocknr = %llu). "
889 : : "There's a risk of filesystem corruption in case of system "
890 : : "crash.\n",
891 : : bh->b_bdev, (unsigned long long)bh->b_blocknr);
892 : : }
893 : :
894 : : /* Call t_frozen trigger and copy buffer data into jh->b_frozen_data. */
895 : 0 : static void jbd2_freeze_jh_data(struct journal_head *jh)
896 : : {
897 : 0 : struct page *page;
898 : 0 : int offset;
899 : 0 : char *source;
900 : 0 : struct buffer_head *bh = jh2bh(jh);
901 : :
902 [ # # ]: 0 : J_EXPECT_JH(jh, buffer_uptodate(bh), "Possible IO failure.\n");
903 : 0 : page = bh->b_page;
904 : 0 : offset = offset_in_page(bh->b_data);
905 : 0 : source = kmap_atomic(page);
906 : : /* Fire data frozen trigger just before we copy the data */
907 : 0 : jbd2_buffer_frozen_trigger(jh, source + offset, jh->b_triggers);
908 : 0 : memcpy(jh->b_frozen_data, source + offset, bh->b_size);
909 : 0 : kunmap_atomic(source);
910 : :
911 : : /*
912 : : * Now that the frozen data is saved off, we need to store any matching
913 : : * triggers.
914 : : */
915 : 0 : jh->b_frozen_triggers = jh->b_triggers;
916 : 0 : }
917 : :
918 : : /*
919 : : * If the buffer is already part of the current transaction, then there
920 : : * is nothing we need to do. If it is already part of a prior
921 : : * transaction which we are still committing to disk, then we need to
922 : : * make sure that we do not overwrite the old copy: we do copy-out to
923 : : * preserve the copy going to disk. We also account the buffer against
924 : : * the handle's metadata buffer credits (unless the buffer is already
925 : : * part of the transaction, that is).
926 : : *
927 : : */
928 : : static int
929 : 16659 : do_get_write_access(handle_t *handle, struct journal_head *jh,
930 : : int force_copy)
931 : : {
932 : 16659 : struct buffer_head *bh;
933 : 16659 : transaction_t *transaction = handle->h_transaction;
934 : 16659 : journal_t *journal;
935 : 16659 : int error;
936 : 16659 : char *frozen_buffer = NULL;
937 : 16659 : unsigned long start_lock, time_lock;
938 : :
939 : 16659 : journal = transaction->t_journal;
940 : :
941 : 16661 : jbd_debug(5, "journal_head %p, force_copy %d\n", jh, force_copy);
942 : :
943 : 16661 : JBUFFER_TRACE(jh, "entry");
944 : : repeat:
945 : 16661 : bh = jh2bh(jh);
946 : :
947 : : /* @@@ Need to check for errors here at some point. */
948 : :
949 : 16661 : start_lock = jiffies;
950 : 16661 : lock_buffer(bh);
951 : 16661 : spin_lock(&jh->b_state_lock);
952 : :
953 : : /* If it takes too long to lock the buffer, trace it */
954 [ + - ]: 16661 : time_lock = jbd2_time_diff(start_lock, jiffies);
955 [ - + ]: 16661 : if (time_lock > HZ/10)
956 : 0 : trace_jbd2_lock_buffer_stall(bh->b_bdev->bd_dev,
957 : 0 : jiffies_to_msecs(time_lock));
958 : :
959 : : /* We now hold the buffer lock so it is safe to query the buffer
960 : : * state. Is the buffer dirty?
961 : : *
962 : : * If so, there are two possibilities. The buffer may be
963 : : * non-journaled, and undergoing a quite legitimate writeback.
964 : : * Otherwise, it is journaled, and we don't expect dirty buffers
965 : : * in that state (the buffers should be marked JBD_Dirty
966 : : * instead.) So either the IO is being done under our own
967 : : * control and this is a bug, or it's a third party IO such as
968 : : * dump(8) (which may leave the buffer scheduled for read ---
969 : : * ie. locked but not dirty) or tune2fs (which may actually have
970 : : * the buffer dirtied, ugh.) */
971 : :
972 [ + + ]: 16661 : if (buffer_dirty(bh)) {
973 : : /*
974 : : * First question: is this buffer already part of the current
975 : : * transaction or the existing committing transaction?
976 : : */
977 [ - + ]: 6185 : if (jh->b_transaction) {
978 [ # # # # ]: 0 : J_ASSERT_JH(jh,
979 : : jh->b_transaction == transaction ||
980 : : jh->b_transaction ==
981 : : journal->j_committing_transaction);
982 [ # # ]: 0 : if (jh->b_next_transaction)
983 [ # # ]: 0 : J_ASSERT_JH(jh, jh->b_next_transaction ==
984 : : transaction);
985 : 0 : warn_dirty_buffer(bh);
986 : : }
987 : : /*
988 : : * In any case we need to clean the dirty flag and we must
989 : : * do it under the buffer lock to be sure we don't race
990 : : * with running write-out.
991 : : */
992 : 6185 : JBUFFER_TRACE(jh, "Journalling dirty buffer");
993 : 6185 : clear_buffer_dirty(bh);
994 : 6185 : set_buffer_jbddirty(bh);
995 : : }
996 : :
997 : 16661 : unlock_buffer(bh);
998 : :
999 : 16661 : error = -EROFS;
1000 [ + - - + ]: 33322 : if (is_handle_aborted(handle)) {
1001 : 0 : spin_unlock(&jh->b_state_lock);
1002 : 0 : goto out;
1003 : : }
1004 : 16661 : error = 0;
1005 : :
1006 : : /*
1007 : : * The buffer is already part of this transaction if b_transaction or
1008 : : * b_next_transaction points to it
1009 : : */
1010 [ + - ]: 16661 : if (jh->b_transaction == transaction ||
1011 [ - + ]: 16661 : jh->b_next_transaction == transaction)
1012 : 0 : goto done;
1013 : :
1014 : : /*
1015 : : * this is the first time this transaction is touching this buffer,
1016 : : * reset the modified flag
1017 : : */
1018 : 16661 : jh->b_modified = 0;
1019 : :
1020 : : /*
1021 : : * If the buffer is not journaled right now, we need to make sure it
1022 : : * doesn't get written to disk before the caller actually commits the
1023 : : * new data
1024 : : */
1025 [ + + ]: 16661 : if (!jh->b_transaction) {
1026 : 16652 : JBUFFER_TRACE(jh, "no transaction");
1027 [ - + ]: 16652 : J_ASSERT_JH(jh, !jh->b_next_transaction);
1028 : 16652 : JBUFFER_TRACE(jh, "file as BJ_Reserved");
1029 : : /*
1030 : : * Make sure all stores to jh (b_modified, b_frozen_data) are
1031 : : * visible before attaching it to the running transaction.
1032 : : * Paired with barrier in jbd2_write_access_granted()
1033 : : */
1034 : 16652 : smp_wmb();
1035 : 16652 : spin_lock(&journal->j_list_lock);
1036 : 16652 : __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
1037 : 16652 : spin_unlock(&journal->j_list_lock);
1038 : 16652 : goto done;
1039 : : }
1040 : : /*
1041 : : * If there is already a copy-out version of this buffer, then we don't
1042 : : * need to make another one
1043 : : */
1044 [ - + ]: 9 : if (jh->b_frozen_data) {
1045 : 0 : JBUFFER_TRACE(jh, "has frozen data");
1046 [ # # ]: 0 : J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1047 : 0 : goto attach_next;
1048 : : }
1049 : :
1050 : 9 : JBUFFER_TRACE(jh, "owned by older transaction");
1051 [ - + ]: 9 : J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1052 [ - + ]: 9 : J_ASSERT_JH(jh, jh->b_transaction == journal->j_committing_transaction);
1053 : :
1054 : : /*
1055 : : * There is one case we have to be very careful about. If the
1056 : : * committing transaction is currently writing this buffer out to disk
1057 : : * and has NOT made a copy-out, then we cannot modify the buffer
1058 : : * contents at all right now. The essence of copy-out is that it is
1059 : : * the extra copy, not the primary copy, which gets journaled. If the
1060 : : * primary copy is already going to disk then we cannot do copy-out
1061 : : * here.
1062 : : */
1063 [ + + ]: 9 : if (buffer_shadow(bh)) {
1064 : 2 : JBUFFER_TRACE(jh, "on shadow: sleep");
1065 : 2 : spin_unlock(&jh->b_state_lock);
1066 : 2 : wait_on_bit_io(&bh->b_state, BH_Shadow, TASK_UNINTERRUPTIBLE);
1067 : 2 : goto repeat;
1068 : : }
1069 : :
1070 : : /*
1071 : : * Only do the copy if the currently-owning transaction still needs it.
1072 : : * If buffer isn't on BJ_Metadata list, the committing transaction is
1073 : : * past that stage (here we use the fact that BH_Shadow is set under
1074 : : * bh_state lock together with refiling to BJ_Shadow list and at this
1075 : : * point we know the buffer doesn't have BH_Shadow set).
1076 : : *
1077 : : * Subtle point, though: if this is a get_undo_access, then we will be
1078 : : * relying on the frozen_data to contain the new value of the
1079 : : * committed_data record after the transaction, so we HAVE to force the
1080 : : * frozen_data copy in that case.
1081 : : */
1082 [ + - + - ]: 7 : if (jh->b_jlist == BJ_Metadata || force_copy) {
1083 : 0 : JBUFFER_TRACE(jh, "generate frozen data");
1084 [ # # ]: 0 : if (!frozen_buffer) {
1085 : 0 : JBUFFER_TRACE(jh, "allocate memory for buffer");
1086 : 0 : spin_unlock(&jh->b_state_lock);
1087 : 0 : frozen_buffer = jbd2_alloc(jh2bh(jh)->b_size,
1088 : : GFP_NOFS | __GFP_NOFAIL);
1089 : 0 : goto repeat;
1090 : : }
1091 : 0 : jh->b_frozen_data = frozen_buffer;
1092 : 0 : frozen_buffer = NULL;
1093 : 0 : jbd2_freeze_jh_data(jh);
1094 : : }
1095 : 7 : attach_next:
1096 : : /*
1097 : : * Make sure all stores to jh (b_modified, b_frozen_data) are visible
1098 : : * before attaching it to the running transaction. Paired with barrier
1099 : : * in jbd2_write_access_granted()
1100 : : */
1101 : 7 : smp_wmb();
1102 : 7 : jh->b_next_transaction = transaction;
1103 : :
1104 : 16659 : done:
1105 : 16659 : spin_unlock(&jh->b_state_lock);
1106 : :
1107 : : /*
1108 : : * If we are about to journal a buffer, then any revoke pending on it is
1109 : : * no longer valid
1110 : : */
1111 : 16659 : jbd2_journal_cancel_revoke(handle, jh);
1112 : :
1113 : 16659 : out:
1114 [ - + ]: 16659 : if (unlikely(frozen_buffer)) /* It's usually NULL */
1115 : 0 : jbd2_free(frozen_buffer, bh->b_size);
1116 : :
1117 : 16659 : JBUFFER_TRACE(jh, "exit");
1118 : 16659 : return error;
1119 : : }
1120 : :
1121 : : /* Fast check whether buffer is already attached to the required transaction */
1122 : 637386 : static bool jbd2_write_access_granted(handle_t *handle, struct buffer_head *bh,
1123 : : bool undo)
1124 : : {
1125 : 637386 : struct journal_head *jh;
1126 : 637386 : bool ret = false;
1127 : :
1128 : : /* Dirty buffers require special handling... */
1129 [ + + ]: 637386 : if (buffer_dirty(bh))
1130 : : return false;
1131 : :
1132 : : /*
1133 : : * RCU protects us from dereferencing freed pages. So the checks we do
1134 : : * are guaranteed not to oops. However the jh slab object can get freed
1135 : : * & reallocated while we work with it. So we have to be careful. When
1136 : : * we see jh attached to the running transaction, we know it must stay
1137 : : * so until the transaction is committed. Thus jh won't be freed and
1138 : : * will be attached to the same bh while we run. However it can
1139 : : * happen jh gets freed, reallocated, and attached to the transaction
1140 : : * just after we get pointer to it from bh. So we have to be careful
1141 : : * and recheck jh still belongs to our bh before we return success.
1142 : : */
1143 : 631202 : rcu_read_lock();
1144 [ + + ]: 631202 : if (!buffer_jbd(bh))
1145 : 10467 : goto out;
1146 : : /* This should be bh2jh() but that doesn't work with inline functions */
1147 [ - + ]: 620735 : jh = READ_ONCE(bh->b_private);
1148 [ - + ]: 620735 : if (!jh)
1149 : 0 : goto out;
1150 : : /* For undo access buffer must have data copied */
1151 [ - + - - ]: 620735 : if (undo && !jh->b_committed_data)
1152 : 0 : goto out;
1153 [ + + + + ]: 620735 : if (READ_ONCE(jh->b_transaction) != handle->h_transaction &&
1154 [ + + ]: 85 : READ_ONCE(jh->b_next_transaction) != handle->h_transaction)
1155 : 8 : goto out;
1156 : : /*
1157 : : * There are two reasons for the barrier here:
1158 : : * 1) Make sure to fetch b_bh after we did previous checks so that we
1159 : : * detect when jh went through free, realloc, attach to transaction
1160 : : * while we were checking. Paired with implicit barrier in that path.
1161 : : * 2) So that access to bh done after jbd2_write_access_granted()
1162 : : * doesn't get reordered and see inconsistent state of concurrent
1163 : : * do_get_write_access().
1164 : : */
1165 : 620727 : smp_mb();
1166 [ - + ]: 620727 : if (unlikely(jh->b_bh != bh))
1167 : 0 : goto out;
1168 : : ret = true;
1169 : 631202 : out:
1170 : 631202 : rcu_read_unlock();
1171 : 631202 : return ret;
1172 : : }
1173 : :
1174 : : /**
1175 : : * int jbd2_journal_get_write_access() - notify intent to modify a buffer for metadata (not data) update.
1176 : : * @handle: transaction to add buffer modifications to
1177 : : * @bh: bh to be used for metadata writes
1178 : : *
1179 : : * Returns: error code or 0 on success.
1180 : : *
1181 : : * In full data journalling mode the buffer may be of type BJ_AsyncData,
1182 : : * because we're ``write()ing`` a buffer which is also part of a shared mapping.
1183 : : */
1184 : :
1185 : 637386 : int jbd2_journal_get_write_access(handle_t *handle, struct buffer_head *bh)
1186 : : {
1187 : 637386 : struct journal_head *jh;
1188 : 637386 : int rc;
1189 : :
1190 [ + - + - ]: 1274772 : if (is_handle_aborted(handle))
1191 : : return -EROFS;
1192 : :
1193 [ + + ]: 637386 : if (jbd2_write_access_granted(handle, bh, false))
1194 : : return 0;
1195 : :
1196 : 16659 : jh = jbd2_journal_add_journal_head(bh);
1197 : : /* We do not want to get caught playing with fields which the
1198 : : * log thread also manipulates. Make sure that the buffer
1199 : : * completes any outstanding IO before proceeding. */
1200 : 16659 : rc = do_get_write_access(handle, jh, 0);
1201 : 16659 : jbd2_journal_put_journal_head(jh);
1202 : 16659 : return rc;
1203 : : }
1204 : :
1205 : :
1206 : : /*
1207 : : * When the user wants to journal a newly created buffer_head
1208 : : * (ie. getblk() returned a new buffer and we are going to populate it
1209 : : * manually rather than reading off disk), then we need to keep the
1210 : : * buffer_head locked until it has been completely filled with new
1211 : : * data. In this case, we should be able to make the assertion that
1212 : : * the bh is not already part of an existing transaction.
1213 : : *
1214 : : * The buffer should already be locked by the caller by this point.
1215 : : * There is no lock ranking violation: it was a newly created,
1216 : : * unlocked buffer beforehand. */
1217 : :
1218 : : /**
1219 : : * int jbd2_journal_get_create_access () - notify intent to use newly created bh
1220 : : * @handle: transaction to new buffer to
1221 : : * @bh: new buffer.
1222 : : *
1223 : : * Call this if you create a new bh.
1224 : : */
1225 : 19578 : int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh)
1226 : : {
1227 : 19578 : transaction_t *transaction = handle->h_transaction;
1228 : 19578 : journal_t *journal;
1229 : 19578 : struct journal_head *jh = jbd2_journal_add_journal_head(bh);
1230 : 19578 : int err;
1231 : :
1232 : 19578 : jbd_debug(5, "journal_head %p\n", jh);
1233 : 19578 : err = -EROFS;
1234 [ + - - + ]: 39156 : if (is_handle_aborted(handle))
1235 : 0 : goto out;
1236 : 19578 : journal = transaction->t_journal;
1237 : 19578 : err = 0;
1238 : :
1239 : 19578 : JBUFFER_TRACE(jh, "entry");
1240 : : /*
1241 : : * The buffer may already belong to this transaction due to pre-zeroing
1242 : : * in the filesystem's new_block code. It may also be on the previous,
1243 : : * committing transaction's lists, but it HAS to be in Forget state in
1244 : : * that case: the transaction must have deleted the buffer for it to be
1245 : : * reused here.
1246 : : */
1247 : 19578 : spin_lock(&jh->b_state_lock);
1248 [ + - - + : 19578 : J_ASSERT_JH(jh, (jh->b_transaction == transaction ||
- - - - ]
1249 : : jh->b_transaction == NULL ||
1250 : : (jh->b_transaction == journal->j_committing_transaction &&
1251 : : jh->b_jlist == BJ_Forget)));
1252 : :
1253 [ - + ]: 19578 : J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1254 [ - + ]: 19578 : J_ASSERT_JH(jh, buffer_locked(jh2bh(jh)));
1255 : :
1256 [ + - ]: 19578 : if (jh->b_transaction == NULL) {
1257 : : /*
1258 : : * Previous jbd2_journal_forget() could have left the buffer
1259 : : * with jbddirty bit set because it was being committed. When
1260 : : * the commit finished, we've filed the buffer for
1261 : : * checkpointing and marked it dirty. Now we are reallocating
1262 : : * the buffer so the transaction freeing it must have
1263 : : * committed and so it's safe to clear the dirty bit.
1264 : : */
1265 : 19578 : clear_buffer_dirty(jh2bh(jh));
1266 : : /* first access by this transaction */
1267 : 19578 : jh->b_modified = 0;
1268 : :
1269 : 19578 : JBUFFER_TRACE(jh, "file as BJ_Reserved");
1270 : 19578 : spin_lock(&journal->j_list_lock);
1271 : 19578 : __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
1272 : 19578 : spin_unlock(&journal->j_list_lock);
1273 [ # # ]: 0 : } else if (jh->b_transaction == journal->j_committing_transaction) {
1274 : : /* first access by this transaction */
1275 : 0 : jh->b_modified = 0;
1276 : :
1277 : 0 : JBUFFER_TRACE(jh, "set next transaction");
1278 : 0 : spin_lock(&journal->j_list_lock);
1279 : 0 : jh->b_next_transaction = transaction;
1280 : 0 : spin_unlock(&journal->j_list_lock);
1281 : : }
1282 : 19578 : spin_unlock(&jh->b_state_lock);
1283 : :
1284 : : /*
1285 : : * akpm: I added this. ext3_alloc_branch can pick up new indirect
1286 : : * blocks which contain freed but then revoked metadata. We need
1287 : : * to cancel the revoke in case we end up freeing it yet again
1288 : : * and the reallocating as data - this would cause a second revoke,
1289 : : * which hits an assertion error.
1290 : : */
1291 : 19578 : JBUFFER_TRACE(jh, "cancelling revoke");
1292 : 19578 : jbd2_journal_cancel_revoke(handle, jh);
1293 : 19578 : out:
1294 : 19578 : jbd2_journal_put_journal_head(jh);
1295 : 19578 : return err;
1296 : : }
1297 : :
1298 : : /**
1299 : : * int jbd2_journal_get_undo_access() - Notify intent to modify metadata with
1300 : : * non-rewindable consequences
1301 : : * @handle: transaction
1302 : : * @bh: buffer to undo
1303 : : *
1304 : : * Sometimes there is a need to distinguish between metadata which has
1305 : : * been committed to disk and that which has not. The ext3fs code uses
1306 : : * this for freeing and allocating space, we have to make sure that we
1307 : : * do not reuse freed space until the deallocation has been committed,
1308 : : * since if we overwrote that space we would make the delete
1309 : : * un-rewindable in case of a crash.
1310 : : *
1311 : : * To deal with that, jbd2_journal_get_undo_access requests write access to a
1312 : : * buffer for parts of non-rewindable operations such as delete
1313 : : * operations on the bitmaps. The journaling code must keep a copy of
1314 : : * the buffer's contents prior to the undo_access call until such time
1315 : : * as we know that the buffer has definitely been committed to disk.
1316 : : *
1317 : : * We never need to know which transaction the committed data is part
1318 : : * of, buffers touched here are guaranteed to be dirtied later and so
1319 : : * will be committed to a new transaction in due course, at which point
1320 : : * we can discard the old committed data pointer.
1321 : : *
1322 : : * Returns error number or 0 on success.
1323 : : */
1324 : 0 : int jbd2_journal_get_undo_access(handle_t *handle, struct buffer_head *bh)
1325 : : {
1326 : 0 : int err;
1327 : 0 : struct journal_head *jh;
1328 : 0 : char *committed_data = NULL;
1329 : :
1330 [ # # # # ]: 0 : if (is_handle_aborted(handle))
1331 : : return -EROFS;
1332 : :
1333 [ # # ]: 0 : if (jbd2_write_access_granted(handle, bh, true))
1334 : : return 0;
1335 : :
1336 : 0 : jh = jbd2_journal_add_journal_head(bh);
1337 : 0 : JBUFFER_TRACE(jh, "entry");
1338 : :
1339 : : /*
1340 : : * Do this first --- it can drop the journal lock, so we want to
1341 : : * make sure that obtaining the committed_data is done
1342 : : * atomically wrt. completion of any outstanding commits.
1343 : : */
1344 : 0 : err = do_get_write_access(handle, jh, 1);
1345 [ # # ]: 0 : if (err)
1346 : 0 : goto out;
1347 : :
1348 : 0 : repeat:
1349 [ # # ]: 0 : if (!jh->b_committed_data)
1350 : 0 : committed_data = jbd2_alloc(jh2bh(jh)->b_size,
1351 : : GFP_NOFS|__GFP_NOFAIL);
1352 : :
1353 : 0 : spin_lock(&jh->b_state_lock);
1354 [ # # ]: 0 : if (!jh->b_committed_data) {
1355 : : /* Copy out the current buffer contents into the
1356 : : * preserved, committed copy. */
1357 : 0 : JBUFFER_TRACE(jh, "generate b_committed data");
1358 [ # # ]: 0 : if (!committed_data) {
1359 : 0 : spin_unlock(&jh->b_state_lock);
1360 : 0 : goto repeat;
1361 : : }
1362 : :
1363 : 0 : jh->b_committed_data = committed_data;
1364 : 0 : committed_data = NULL;
1365 : 0 : memcpy(jh->b_committed_data, bh->b_data, bh->b_size);
1366 : : }
1367 : 0 : spin_unlock(&jh->b_state_lock);
1368 : 0 : out:
1369 : 0 : jbd2_journal_put_journal_head(jh);
1370 [ # # ]: 0 : if (unlikely(committed_data))
1371 : 0 : jbd2_free(committed_data, bh->b_size);
1372 : : return err;
1373 : : }
1374 : :
1375 : : /**
1376 : : * void jbd2_journal_set_triggers() - Add triggers for commit writeout
1377 : : * @bh: buffer to trigger on
1378 : : * @type: struct jbd2_buffer_trigger_type containing the trigger(s).
1379 : : *
1380 : : * Set any triggers on this journal_head. This is always safe, because
1381 : : * triggers for a committing buffer will be saved off, and triggers for
1382 : : * a running transaction will match the buffer in that transaction.
1383 : : *
1384 : : * Call with NULL to clear the triggers.
1385 : : */
1386 : 0 : void jbd2_journal_set_triggers(struct buffer_head *bh,
1387 : : struct jbd2_buffer_trigger_type *type)
1388 : : {
1389 : 0 : struct journal_head *jh = jbd2_journal_grab_journal_head(bh);
1390 : :
1391 [ # # # # ]: 0 : if (WARN_ON(!jh))
1392 : : return;
1393 : 0 : jh->b_triggers = type;
1394 : 0 : jbd2_journal_put_journal_head(jh);
1395 : : }
1396 : :
1397 : 7224 : void jbd2_buffer_frozen_trigger(struct journal_head *jh, void *mapped_data,
1398 : : struct jbd2_buffer_trigger_type *triggers)
1399 : : {
1400 [ - + - - ]: 7224 : struct buffer_head *bh = jh2bh(jh);
1401 : :
1402 [ - + - - : 7224 : if (!triggers || !triggers->t_frozen)
- - - - ]
1403 : : return;
1404 : :
1405 : 0 : triggers->t_frozen(triggers, bh, mapped_data, bh->b_size);
1406 : : }
1407 : :
1408 : 0 : void jbd2_buffer_abort_trigger(struct journal_head *jh,
1409 : : struct jbd2_buffer_trigger_type *triggers)
1410 : : {
1411 [ # # # # ]: 0 : if (!triggers || !triggers->t_abort)
1412 : : return;
1413 : :
1414 : 0 : triggers->t_abort(triggers, jh2bh(jh));
1415 : : }
1416 : :
1417 : : /**
1418 : : * int jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata
1419 : : * @handle: transaction to add buffer to.
1420 : : * @bh: buffer to mark
1421 : : *
1422 : : * mark dirty metadata which needs to be journaled as part of the current
1423 : : * transaction.
1424 : : *
1425 : : * The buffer must have previously had jbd2_journal_get_write_access()
1426 : : * called so that it has a valid journal_head attached to the buffer
1427 : : * head.
1428 : : *
1429 : : * The buffer is placed on the transaction's metadata list and is marked
1430 : : * as belonging to the transaction.
1431 : : *
1432 : : * Returns error number or 0 on success.
1433 : : *
1434 : : * Special care needs to be taken if the buffer already belongs to the
1435 : : * current committing transaction (in which case we should have frozen
1436 : : * data present for that commit). In that case, we don't relink the
1437 : : * buffer: that only gets done when the old transaction finally
1438 : : * completes its commit.
1439 : : */
1440 : 656808 : int jbd2_journal_dirty_metadata(handle_t *handle, struct buffer_head *bh)
1441 : : {
1442 : 656808 : transaction_t *transaction = handle->h_transaction;
1443 : 656808 : journal_t *journal;
1444 : 656808 : struct journal_head *jh;
1445 : 656808 : int ret = 0;
1446 : :
1447 [ + - + - ]: 1313616 : if (is_handle_aborted(handle))
1448 : : return -EROFS;
1449 [ + - ]: 656808 : if (!buffer_jbd(bh))
1450 : : return -EUCLEAN;
1451 : :
1452 : : /*
1453 : : * We don't grab jh reference here since the buffer must be part
1454 : : * of the running transaction.
1455 : : */
1456 [ + + ]: 656808 : jh = bh2jh(bh);
1457 : 656808 : jbd_debug(5, "journal_head %p\n", jh);
1458 : 656808 : JBUFFER_TRACE(jh, "entry");
1459 : :
1460 : : /*
1461 : : * This and the following assertions are unreliable since we may see jh
1462 : : * in inconsistent state unless we grab bh_state lock. But this is
1463 : : * crucial to catch bugs so let's do a reliable check until the
1464 : : * lockless handling is fully proven.
1465 : : */
1466 [ + + ]: 656808 : if (jh->b_transaction != transaction &&
1467 [ - + ]: 84 : jh->b_next_transaction != transaction) {
1468 : 0 : spin_lock(&jh->b_state_lock);
1469 [ # # # # ]: 0 : J_ASSERT_JH(jh, jh->b_transaction == transaction ||
1470 : : jh->b_next_transaction == transaction);
1471 : 0 : spin_unlock(&jh->b_state_lock);
1472 : : }
1473 [ + + ]: 656808 : if (jh->b_modified == 1) {
1474 : : /* If it's in our transaction it must be in BJ_Metadata list. */
1475 [ + + ]: 620571 : if (jh->b_transaction == transaction &&
1476 [ - + ]: 620494 : jh->b_jlist != BJ_Metadata) {
1477 : 0 : spin_lock(&jh->b_state_lock);
1478 [ # # ]: 0 : if (jh->b_transaction == transaction &&
1479 [ # # ]: 0 : jh->b_jlist != BJ_Metadata)
1480 : 0 : pr_err("JBD2: assertion failure: h_type=%u "
1481 : : "h_line_no=%u block_no=%llu jlist=%u\n",
1482 : : handle->h_type, handle->h_line_no,
1483 : : (unsigned long long) bh->b_blocknr,
1484 : : jh->b_jlist);
1485 [ # # # # ]: 0 : J_ASSERT_JH(jh, jh->b_transaction != transaction ||
1486 : : jh->b_jlist == BJ_Metadata);
1487 : 0 : spin_unlock(&jh->b_state_lock);
1488 : : }
1489 : 620571 : goto out;
1490 : : }
1491 : :
1492 : 36237 : journal = transaction->t_journal;
1493 : 36237 : spin_lock(&jh->b_state_lock);
1494 : :
1495 [ + - ]: 36237 : if (jh->b_modified == 0) {
1496 : : /*
1497 : : * This buffer's got modified and becoming part
1498 : : * of the transaction. This needs to be done
1499 : : * once a transaction -bzzz
1500 : : */
1501 [ + - - + : 72474 : if (WARN_ON_ONCE(jbd2_handle_buffer_credits(handle) <= 0)) {
- + ]
1502 : 0 : ret = -ENOSPC;
1503 : 0 : goto out_unlock_bh;
1504 : : }
1505 : 36237 : jh->b_modified = 1;
1506 : 36237 : handle->h_total_credits--;
1507 : : }
1508 : :
1509 : : /*
1510 : : * fastpath, to avoid expensive locking. If this buffer is already
1511 : : * on the running transaction's metadata list there is nothing to do.
1512 : : * Nobody can take it off again because there is a handle open.
1513 : : * I _think_ we're OK here with SMP barriers - a mistaken decision will
1514 : : * result in this test being false, so we go in and take the locks.
1515 : : */
1516 [ + + - + ]: 36237 : if (jh->b_transaction == transaction && jh->b_jlist == BJ_Metadata) {
1517 : 0 : JBUFFER_TRACE(jh, "fastpath");
1518 [ # # ]: 0 : if (unlikely(jh->b_transaction !=
1519 : : journal->j_running_transaction)) {
1520 [ # # ]: 0 : printk(KERN_ERR "JBD2: %s: "
1521 : : "jh->b_transaction (%llu, %p, %u) != "
1522 : : "journal->j_running_transaction (%p, %u)\n",
1523 : 0 : journal->j_devname,
1524 : 0 : (unsigned long long) bh->b_blocknr,
1525 : : jh->b_transaction,
1526 : : jh->b_transaction ? jh->b_transaction->t_tid : 0,
1527 : : journal->j_running_transaction,
1528 : : journal->j_running_transaction ?
1529 : : journal->j_running_transaction->t_tid : 0);
1530 : 0 : ret = -EINVAL;
1531 : : }
1532 : 0 : goto out_unlock_bh;
1533 : : }
1534 : :
1535 : 36237 : set_buffer_jbddirty(bh);
1536 : :
1537 : : /*
1538 : : * Metadata already on the current transaction list doesn't
1539 : : * need to be filed. Metadata on another transaction's list must
1540 : : * be committing, and will be refiled once the commit completes:
1541 : : * leave it alone for now.
1542 : : */
1543 [ + + ]: 36237 : if (jh->b_transaction != transaction) {
1544 : 7 : JBUFFER_TRACE(jh, "already on other transaction");
1545 [ + - - + ]: 7 : if (unlikely(((jh->b_transaction !=
1546 : : journal->j_committing_transaction)) ||
1547 : : (jh->b_next_transaction != transaction))) {
1548 [ # # ]: 0 : printk(KERN_ERR "jbd2_journal_dirty_metadata: %s: "
1549 : : "bad jh for block %llu: "
1550 : : "transaction (%p, %u), "
1551 : : "jh->b_transaction (%p, %u), "
1552 : : "jh->b_next_transaction (%p, %u), jlist %u\n",
1553 : 0 : journal->j_devname,
1554 : 0 : (unsigned long long) bh->b_blocknr,
1555 : : transaction, transaction->t_tid,
1556 : : jh->b_transaction,
1557 : : jh->b_transaction ?
1558 : : jh->b_transaction->t_tid : 0,
1559 : : jh->b_next_transaction,
1560 [ # # ]: 0 : jh->b_next_transaction ?
1561 : : jh->b_next_transaction->t_tid : 0,
1562 : : jh->b_jlist);
1563 : 0 : WARN_ON(1);
1564 : : ret = -EINVAL;
1565 : : }
1566 : : /* And this case is illegal: we can't reuse another
1567 : : * transaction's data buffer, ever. */
1568 : 7 : goto out_unlock_bh;
1569 : : }
1570 : :
1571 : : /* That test should have eliminated the following case: */
1572 [ - + ]: 36230 : J_ASSERT_JH(jh, jh->b_frozen_data == NULL);
1573 : :
1574 : 36230 : JBUFFER_TRACE(jh, "file as BJ_Metadata");
1575 : 36230 : spin_lock(&journal->j_list_lock);
1576 : 36230 : __jbd2_journal_file_buffer(jh, transaction, BJ_Metadata);
1577 : 36230 : spin_unlock(&journal->j_list_lock);
1578 : 36237 : out_unlock_bh:
1579 : 36237 : spin_unlock(&jh->b_state_lock);
1580 : : out:
1581 : : JBUFFER_TRACE(jh, "exit");
1582 : : return ret;
1583 : : }
1584 : :
1585 : : /**
1586 : : * void jbd2_journal_forget() - bforget() for potentially-journaled buffers.
1587 : : * @handle: transaction handle
1588 : : * @bh: bh to 'forget'
1589 : : *
1590 : : * We can only do the bforget if there are no commits pending against the
1591 : : * buffer. If the buffer is dirty in the current running transaction we
1592 : : * can safely unlink it.
1593 : : *
1594 : : * bh may not be a journalled buffer at all - it may be a non-JBD
1595 : : * buffer which came off the hashtable. Check for this.
1596 : : *
1597 : : * Decrements bh->b_count by one.
1598 : : *
1599 : : * Allow this call even if the handle has aborted --- it may be part of
1600 : : * the caller's cleanup after an abort.
1601 : : */
1602 : 156 : int jbd2_journal_forget(handle_t *handle, struct buffer_head *bh)
1603 : : {
1604 : 156 : transaction_t *transaction = handle->h_transaction;
1605 : 156 : journal_t *journal;
1606 : 156 : struct journal_head *jh;
1607 : 156 : int drop_reserve = 0;
1608 : 156 : int err = 0;
1609 : 156 : int was_modified = 0;
1610 : :
1611 [ + - + - ]: 312 : if (is_handle_aborted(handle))
1612 : : return -EROFS;
1613 : 156 : journal = transaction->t_journal;
1614 : :
1615 : 156 : BUFFER_TRACE(bh, "entry");
1616 : :
1617 : 156 : jh = jbd2_journal_grab_journal_head(bh);
1618 [ - + ]: 156 : if (!jh) {
1619 : 0 : __bforget(bh);
1620 : 0 : return 0;
1621 : : }
1622 : :
1623 : 156 : spin_lock(&jh->b_state_lock);
1624 : :
1625 : : /* Critical error: attempting to delete a bitmap buffer, maybe?
1626 : : * Don't do any jbd operations, and return an error. */
1627 [ - + - + ]: 156 : if (!J_EXPECT_JH(jh, !jh->b_committed_data,
1628 : : "inconsistent data on disk")) {
1629 : 0 : err = -EIO;
1630 : 0 : goto drop;
1631 : : }
1632 : :
1633 : : /* keep track of whether or not this transaction modified us */
1634 : 156 : was_modified = jh->b_modified;
1635 : :
1636 : : /*
1637 : : * The buffer's going from the transaction, we must drop
1638 : : * all references -bzzz
1639 : : */
1640 : 156 : jh->b_modified = 0;
1641 : :
1642 [ + - ]: 156 : if (jh->b_transaction == transaction) {
1643 [ - + ]: 156 : J_ASSERT_JH(jh, !jh->b_frozen_data);
1644 : :
1645 : : /* If we are forgetting a buffer which is already part
1646 : : * of this transaction, then we can just drop it from
1647 : : * the transaction immediately. */
1648 : 156 : clear_buffer_dirty(bh);
1649 : 156 : clear_buffer_jbddirty(bh);
1650 : :
1651 : 156 : JBUFFER_TRACE(jh, "belongs to current transaction: unfile");
1652 : :
1653 : : /*
1654 : : * we only want to drop a reference if this transaction
1655 : : * modified the buffer
1656 : : */
1657 [ + - ]: 156 : if (was_modified)
1658 : 156 : drop_reserve = 1;
1659 : :
1660 : : /*
1661 : : * We are no longer going to journal this buffer.
1662 : : * However, the commit of this transaction is still
1663 : : * important to the buffer: the delete that we are now
1664 : : * processing might obsolete an old log entry, so by
1665 : : * committing, we can satisfy the buffer's checkpoint.
1666 : : *
1667 : : * So, if we have a checkpoint on the buffer, we should
1668 : : * now refile the buffer on our BJ_Forget list so that
1669 : : * we know to remove the checkpoint after we commit.
1670 : : */
1671 : :
1672 : 156 : spin_lock(&journal->j_list_lock);
1673 [ - + ]: 156 : if (jh->b_cp_transaction) {
1674 : 0 : __jbd2_journal_temp_unlink_buffer(jh);
1675 : 0 : __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
1676 : : } else {
1677 : 156 : __jbd2_journal_unfile_buffer(jh);
1678 : 156 : jbd2_journal_put_journal_head(jh);
1679 : : }
1680 : 156 : spin_unlock(&journal->j_list_lock);
1681 [ # # ]: 0 : } else if (jh->b_transaction) {
1682 [ # # ]: 0 : J_ASSERT_JH(jh, (jh->b_transaction ==
1683 : : journal->j_committing_transaction));
1684 : : /* However, if the buffer is still owned by a prior
1685 : : * (committing) transaction, we can't drop it yet... */
1686 : 0 : JBUFFER_TRACE(jh, "belongs to older transaction");
1687 : : /* ... but we CAN drop it from the new transaction through
1688 : : * marking the buffer as freed and set j_next_transaction to
1689 : : * the new transaction, so that not only the commit code
1690 : : * knows it should clear dirty bits when it is done with the
1691 : : * buffer, but also the buffer can be checkpointed only
1692 : : * after the new transaction commits. */
1693 : :
1694 : 0 : set_buffer_freed(bh);
1695 : :
1696 [ # # ]: 0 : if (!jh->b_next_transaction) {
1697 : 0 : spin_lock(&journal->j_list_lock);
1698 : 0 : jh->b_next_transaction = transaction;
1699 : 0 : spin_unlock(&journal->j_list_lock);
1700 : : } else {
1701 [ # # ]: 0 : J_ASSERT(jh->b_next_transaction == transaction);
1702 : :
1703 : : /*
1704 : : * only drop a reference if this transaction modified
1705 : : * the buffer
1706 : : */
1707 [ # # ]: 0 : if (was_modified)
1708 : 0 : drop_reserve = 1;
1709 : : }
1710 : : } else {
1711 : : /*
1712 : : * Finally, if the buffer is not belongs to any
1713 : : * transaction, we can just drop it now if it has no
1714 : : * checkpoint.
1715 : : */
1716 : 0 : spin_lock(&journal->j_list_lock);
1717 [ # # ]: 0 : if (!jh->b_cp_transaction) {
1718 : 0 : JBUFFER_TRACE(jh, "belongs to none transaction");
1719 : 0 : spin_unlock(&journal->j_list_lock);
1720 : 0 : goto drop;
1721 : : }
1722 : :
1723 : : /*
1724 : : * Otherwise, if the buffer has been written to disk,
1725 : : * it is safe to remove the checkpoint and drop it.
1726 : : */
1727 [ # # ]: 0 : if (!buffer_dirty(bh)) {
1728 : 0 : __jbd2_journal_remove_checkpoint(jh);
1729 : 0 : spin_unlock(&journal->j_list_lock);
1730 : 0 : goto drop;
1731 : : }
1732 : :
1733 : : /*
1734 : : * The buffer is still not written to disk, we should
1735 : : * attach this buffer to current transaction so that the
1736 : : * buffer can be checkpointed only after the current
1737 : : * transaction commits.
1738 : : */
1739 : 0 : clear_buffer_dirty(bh);
1740 : 0 : __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
1741 : 0 : spin_unlock(&journal->j_list_lock);
1742 : : }
1743 : 156 : drop:
1744 : 156 : __brelse(bh);
1745 : 156 : spin_unlock(&jh->b_state_lock);
1746 : 156 : jbd2_journal_put_journal_head(jh);
1747 [ + - ]: 156 : if (drop_reserve) {
1748 : : /* no need to reserve log space for this block -bzzz */
1749 : 156 : handle->h_total_credits++;
1750 : : }
1751 : : return err;
1752 : : }
1753 : :
1754 : : /**
1755 : : * int jbd2_journal_stop() - complete a transaction
1756 : : * @handle: transaction to complete.
1757 : : *
1758 : : * All done for a particular handle.
1759 : : *
1760 : : * There is not much action needed here. We just return any remaining
1761 : : * buffer credits to the transaction and remove the handle. The only
1762 : : * complication is that we need to start a commit operation if the
1763 : : * filesystem is marked for synchronous update.
1764 : : *
1765 : : * jbd2_journal_stop itself will not usually return an error, but it may
1766 : : * do so in unusual circumstances. In particular, expect it to
1767 : : * return -EIO if a jbd2_journal_abort has been executed since the
1768 : : * transaction began.
1769 : : */
1770 : 257791 : int jbd2_journal_stop(handle_t *handle)
1771 : : {
1772 : 257791 : transaction_t *transaction = handle->h_transaction;
1773 : 257791 : journal_t *journal;
1774 : 257791 : int err = 0, wait_for_commit = 0;
1775 : 257791 : tid_t tid;
1776 : 257791 : pid_t pid;
1777 : :
1778 [ + + ]: 257791 : if (--handle->h_ref > 0) {
1779 : : jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1780 : 87469 : handle->h_ref);
1781 [ + - + - ]: 174938 : if (is_handle_aborted(handle))
1782 : : return -EIO;
1783 : 87469 : return 0;
1784 : : }
1785 [ - + ]: 170322 : if (!transaction) {
1786 : : /*
1787 : : * Handle is already detached from the transaction so there is
1788 : : * nothing to do other than free the handle.
1789 : : */
1790 : 0 : memalloc_nofs_restore(handle->saved_alloc_context);
1791 : 0 : goto free_and_exit;
1792 : : }
1793 : 170322 : journal = transaction->t_journal;
1794 : 170322 : tid = transaction->t_tid;
1795 : :
1796 [ + - - + ]: 170322 : if (is_handle_aborted(handle))
1797 : 0 : err = -EIO;
1798 : :
1799 : 170322 : jbd_debug(4, "Handle %p going down\n", handle);
1800 : 170322 : trace_jbd2_handle_stats(journal->j_fs_dev->bd_dev,
1801 : 170322 : tid, handle->h_type, handle->h_line_no,
1802 : 170322 : jiffies - handle->h_start_jiffies,
1803 : 170322 : handle->h_sync, handle->h_requested_credits,
1804 : 170322 : (handle->h_requested_credits -
1805 : 170322 : handle->h_total_credits));
1806 : :
1807 : : /*
1808 : : * Implement synchronous transaction batching. If the handle
1809 : : * was synchronous, don't force a commit immediately. Let's
1810 : : * yield and let another thread piggyback onto this
1811 : : * transaction. Keep doing that while new threads continue to
1812 : : * arrive. It doesn't cost much - we're about to run a commit
1813 : : * and sleep on IO anyway. Speeds up many-threaded, many-dir
1814 : : * operations by 30x or more...
1815 : : *
1816 : : * We try and optimize the sleep time against what the
1817 : : * underlying disk can do, instead of having a static sleep
1818 : : * time. This is useful for the case where our storage is so
1819 : : * fast that it is more optimal to go ahead and force a flush
1820 : : * and wait for the transaction to be committed than it is to
1821 : : * wait for an arbitrary amount of time for new writers to
1822 : : * join the transaction. We achieve this by measuring how
1823 : : * long it takes to commit a transaction, and compare it with
1824 : : * how long this transaction has been running, and if run time
1825 : : * < commit time then we sleep for the delta and commit. This
1826 : : * greatly helps super fast disks that would see slowdowns as
1827 : : * more threads started doing fsyncs.
1828 : : *
1829 : : * But don't do this if this process was the most recent one
1830 : : * to perform a synchronous write. We do this to detect the
1831 : : * case where a single process is doing a stream of sync
1832 : : * writes. No point in waiting for joiners in that case.
1833 : : *
1834 : : * Setting max_batch_time to 0 disables this completely.
1835 : : */
1836 [ - + ]: 170322 : pid = current->pid;
1837 [ - + - - ]: 170322 : if (handle->h_sync && journal->j_last_sync_writer != pid &&
1838 [ # # ]: 0 : journal->j_max_batch_time) {
1839 : 0 : u64 commit_time, trans_time;
1840 : :
1841 : 0 : journal->j_last_sync_writer = pid;
1842 : :
1843 : 0 : read_lock(&journal->j_state_lock);
1844 : 0 : commit_time = journal->j_average_commit_time;
1845 : 0 : read_unlock(&journal->j_state_lock);
1846 : :
1847 [ # # ]: 0 : trans_time = ktime_to_ns(ktime_sub(ktime_get(),
1848 : : transaction->t_start_time));
1849 : :
1850 : 0 : commit_time = max_t(u64, commit_time,
1851 : : 1000*journal->j_min_batch_time);
1852 : 0 : commit_time = min_t(u64, commit_time,
1853 : : 1000*journal->j_max_batch_time);
1854 : :
1855 [ # # ]: 0 : if (trans_time < commit_time) {
1856 : 0 : ktime_t expires = ktime_add_ns(ktime_get(),
1857 : : commit_time);
1858 : 0 : set_current_state(TASK_UNINTERRUPTIBLE);
1859 : 0 : schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1860 : : }
1861 : : }
1862 : :
1863 [ - + ]: 170322 : if (handle->h_sync)
1864 : 0 : transaction->t_synchronous_commit = 1;
1865 : :
1866 : : /*
1867 : : * If the handle is marked SYNC, we need to set another commit
1868 : : * going! We also want to force a commit if the transaction is too
1869 : : * old now.
1870 : : */
1871 [ + - ]: 170322 : if (handle->h_sync ||
1872 [ + + ]: 170322 : time_after_eq(jiffies, transaction->t_expires)) {
1873 : : /* Do this even for aborted journals: an abort still
1874 : : * completes the commit thread, it just doesn't write
1875 : : * anything to disk. */
1876 : :
1877 : : jbd_debug(2, "transaction too old, requesting commit for "
1878 : 30 : "handle %p\n", handle);
1879 : : /* This is non-blocking */
1880 : 30 : jbd2_log_start_commit(journal, tid);
1881 : :
1882 : : /*
1883 : : * Special case: JBD2_SYNC synchronous updates require us
1884 : : * to wait for the commit to complete.
1885 : : */
1886 [ - + - - ]: 30 : if (handle->h_sync && !(current->flags & PF_MEMALLOC))
1887 : 0 : wait_for_commit = 1;
1888 : : }
1889 : :
1890 : : /*
1891 : : * Once stop_this_handle() drops t_updates, the transaction could start
1892 : : * committing on us and eventually disappear. So we must not
1893 : : * dereference transaction pointer again after calling
1894 : : * stop_this_handle().
1895 : : */
1896 : 170322 : stop_this_handle(handle);
1897 : :
1898 [ + - ]: 170322 : if (wait_for_commit)
1899 : 0 : err = jbd2_log_wait_commit(journal, tid);
1900 : :
1901 : 170322 : free_and_exit:
1902 [ + + ]: 170322 : if (handle->h_rsv_handle)
1903 : 624 : jbd2_free_handle(handle->h_rsv_handle);
1904 : 170322 : jbd2_free_handle(handle);
1905 : 170322 : return err;
1906 : : }
1907 : :
1908 : : /*
1909 : : *
1910 : : * List management code snippets: various functions for manipulating the
1911 : : * transaction buffer lists.
1912 : : *
1913 : : */
1914 : :
1915 : : /*
1916 : : * Append a buffer to a transaction list, given the transaction's list head
1917 : : * pointer.
1918 : : *
1919 : : * j_list_lock is held.
1920 : : *
1921 : : * jh->b_state_lock is held.
1922 : : */
1923 : :
1924 : : static inline void
1925 : 86915 : __blist_add_buffer(struct journal_head **list, struct journal_head *jh)
1926 : : {
1927 : 86915 : if (!*list) {
1928 : 36835 : jh->b_tnext = jh->b_tprev = jh;
1929 : 36835 : *list = jh;
1930 : : } else {
1931 : : /* Insert at the tail of the list to preserve order */
1932 : 50080 : struct journal_head *first = *list, *last = first->b_tprev;
1933 : 50080 : jh->b_tprev = last;
1934 : 50080 : jh->b_tnext = first;
1935 : 50080 : last->b_tnext = first->b_tprev = jh;
1936 : : }
1937 : : }
1938 : :
1939 : : /*
1940 : : * Remove a buffer from a transaction list, given the transaction's list
1941 : : * head pointer.
1942 : : *
1943 : : * Called with j_list_lock held, and the journal may not be locked.
1944 : : *
1945 : : * jh->b_state_lock is held.
1946 : : */
1947 : :
1948 : : static inline void
1949 : 58058 : __blist_del_buffer(struct journal_head **list, struct journal_head *jh)
1950 : : {
1951 : 58058 : if (*list == jh) {
1952 : 50399 : *list = jh->b_tnext;
1953 [ + + ]: 50399 : if (*list == jh)
1954 : 36757 : *list = NULL;
1955 : : }
1956 : 58058 : jh->b_tprev->b_tnext = jh->b_tnext;
1957 : 58058 : jh->b_tnext->b_tprev = jh->b_tprev;
1958 : : }
1959 : :
1960 : : /*
1961 : : * Remove a buffer from the appropriate transaction list.
1962 : : *
1963 : : * Note that this function can *change* the value of
1964 : : * bh->b_transaction->t_buffers, t_forget, t_shadow_list, t_log_list or
1965 : : * t_reserved_list. If the caller is holding onto a copy of one of these
1966 : : * pointers, it could go bad. Generally the caller needs to re-read the
1967 : : * pointer from the transaction_t.
1968 : : *
1969 : : * Called under j_list_lock.
1970 : : */
1971 : 58065 : static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh)
1972 : : {
1973 : 58065 : struct journal_head **list = NULL;
1974 : 58065 : transaction_t *transaction;
1975 [ + - ]: 58065 : struct buffer_head *bh = jh2bh(jh);
1976 : :
1977 : 58065 : lockdep_assert_held(&jh->b_state_lock);
1978 : 58065 : transaction = jh->b_transaction;
1979 [ + - ]: 58065 : if (transaction)
1980 [ - + ]: 58065 : assert_spin_locked(&transaction->t_journal->j_list_lock);
1981 : :
1982 [ - + ]: 58065 : J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
1983 [ + + ]: 58065 : if (jh->b_jlist != BJ_None)
1984 [ - + ]: 58058 : J_ASSERT_JH(jh, transaction != NULL);
1985 : :
1986 [ + + + + : 58065 : switch (jh->b_jlist) {
- + ]
1987 : : case BJ_None:
1988 : : return;
1989 : 7380 : case BJ_Metadata:
1990 : 7380 : transaction->t_nr_buffers--;
1991 [ - + ]: 7380 : J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0);
1992 : 7380 : list = &transaction->t_buffers;
1993 : 7380 : break;
1994 : 7224 : case BJ_Forget:
1995 : 7224 : list = &transaction->t_forget;
1996 : 7224 : break;
1997 : 7224 : case BJ_Shadow:
1998 : 7224 : list = &transaction->t_shadow_list;
1999 : 7224 : break;
2000 : 36230 : case BJ_Reserved:
2001 : 36230 : list = &transaction->t_reserved_list;
2002 : 36230 : break;
2003 : : }
2004 : :
2005 [ + + ]: 58058 : __blist_del_buffer(list, jh);
2006 : 58058 : jh->b_jlist = BJ_None;
2007 [ + - - + ]: 58058 : if (transaction && is_journal_aborted(transaction->t_journal))
2008 : 0 : clear_buffer_jbddirty(bh);
2009 [ + + ]: 58058 : else if (test_clear_buffer_jbddirty(bh))
2010 : 7217 : mark_buffer_dirty(bh); /* Expose it to the VM */
2011 : : }
2012 : :
2013 : : /*
2014 : : * Remove buffer from all transactions. The caller is responsible for dropping
2015 : : * the jh reference that belonged to the transaction.
2016 : : *
2017 : : * Called with bh_state lock and j_list_lock
2018 : : */
2019 : 7373 : static void __jbd2_journal_unfile_buffer(struct journal_head *jh)
2020 : : {
2021 : 156 : __jbd2_journal_temp_unlink_buffer(jh);
2022 : 156 : jh->b_transaction = NULL;
2023 : : }
2024 : :
2025 : 0 : void jbd2_journal_unfile_buffer(journal_t *journal, struct journal_head *jh)
2026 : : {
2027 : 0 : struct buffer_head *bh = jh2bh(jh);
2028 : :
2029 : : /* Get reference so that buffer cannot be freed before we unlock it */
2030 : 0 : get_bh(bh);
2031 : 0 : spin_lock(&jh->b_state_lock);
2032 : 0 : spin_lock(&journal->j_list_lock);
2033 : 0 : __jbd2_journal_unfile_buffer(jh);
2034 : 0 : spin_unlock(&journal->j_list_lock);
2035 : 0 : spin_unlock(&jh->b_state_lock);
2036 : 0 : jbd2_journal_put_journal_head(jh);
2037 : 0 : __brelse(bh);
2038 : 0 : }
2039 : :
2040 : : /*
2041 : : * Called from jbd2_journal_try_to_free_buffers().
2042 : : *
2043 : : * Called under jh->b_state_lock
2044 : : */
2045 : : static void
2046 : 0 : __journal_try_to_free_buffer(journal_t *journal, struct buffer_head *bh)
2047 : : {
2048 : 0 : struct journal_head *jh;
2049 : :
2050 : 0 : jh = bh2jh(bh);
2051 : :
2052 [ # # # # ]: 0 : if (buffer_locked(bh) || buffer_dirty(bh))
2053 : 0 : goto out;
2054 : :
2055 [ # # # # ]: 0 : if (jh->b_next_transaction != NULL || jh->b_transaction != NULL)
2056 : 0 : goto out;
2057 : :
2058 : 0 : spin_lock(&journal->j_list_lock);
2059 [ # # ]: 0 : if (jh->b_cp_transaction != NULL) {
2060 : : /* written-back checkpointed metadata buffer */
2061 : 0 : JBUFFER_TRACE(jh, "remove from checkpoint list");
2062 : 0 : __jbd2_journal_remove_checkpoint(jh);
2063 : : }
2064 : 0 : spin_unlock(&journal->j_list_lock);
2065 : 0 : out:
2066 : 0 : return;
2067 : : }
2068 : :
2069 : : /**
2070 : : * int jbd2_journal_try_to_free_buffers() - try to free page buffers.
2071 : : * @journal: journal for operation
2072 : : * @page: to try and free
2073 : : * @gfp_mask: we use the mask to detect how hard should we try to release
2074 : : * buffers. If __GFP_DIRECT_RECLAIM and __GFP_FS is set, we wait for commit
2075 : : * code to release the buffers.
2076 : : *
2077 : : *
2078 : : * For all the buffers on this page,
2079 : : * if they are fully written out ordered data, move them onto BUF_CLEAN
2080 : : * so try_to_free_buffers() can reap them.
2081 : : *
2082 : : * This function returns non-zero if we wish try_to_free_buffers()
2083 : : * to be called. We do this if the page is releasable by try_to_free_buffers().
2084 : : * We also do it if the page has locked or dirty buffers and the caller wants
2085 : : * us to perform sync or async writeout.
2086 : : *
2087 : : * This complicates JBD locking somewhat. We aren't protected by the
2088 : : * BKL here. We wish to remove the buffer from its committing or
2089 : : * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer.
2090 : : *
2091 : : * This may *change* the value of transaction_t->t_datalist, so anyone
2092 : : * who looks at t_datalist needs to lock against this function.
2093 : : *
2094 : : * Even worse, someone may be doing a jbd2_journal_dirty_data on this
2095 : : * buffer. So we need to lock against that. jbd2_journal_dirty_data()
2096 : : * will come out of the lock with the buffer dirty, which makes it
2097 : : * ineligible for release here.
2098 : : *
2099 : : * Who else is affected by this? hmm... Really the only contender
2100 : : * is do_get_write_access() - it could be looking at the buffer while
2101 : : * journal_try_to_free_buffer() is changing its state. But that
2102 : : * cannot happen because we never reallocate freed data as metadata
2103 : : * while the data is part of a transaction. Yes?
2104 : : *
2105 : : * Return 0 on failure, 1 on success
2106 : : */
2107 : 0 : int jbd2_journal_try_to_free_buffers(journal_t *journal,
2108 : : struct page *page, gfp_t gfp_mask)
2109 : : {
2110 : 0 : struct buffer_head *head;
2111 : 0 : struct buffer_head *bh;
2112 : 0 : int ret = 0;
2113 : :
2114 [ # # # # ]: 0 : J_ASSERT(PageLocked(page));
2115 : :
2116 [ # # ]: 0 : head = page_buffers(page);
2117 : 0 : bh = head;
2118 : 0 : do {
2119 : 0 : struct journal_head *jh;
2120 : :
2121 : : /*
2122 : : * We take our own ref against the journal_head here to avoid
2123 : : * having to add tons of locking around each instance of
2124 : : * jbd2_journal_put_journal_head().
2125 : : */
2126 : 0 : jh = jbd2_journal_grab_journal_head(bh);
2127 [ # # ]: 0 : if (!jh)
2128 : 0 : continue;
2129 : :
2130 : 0 : spin_lock(&jh->b_state_lock);
2131 : 0 : __journal_try_to_free_buffer(journal, bh);
2132 : 0 : spin_unlock(&jh->b_state_lock);
2133 : 0 : jbd2_journal_put_journal_head(jh);
2134 [ # # ]: 0 : if (buffer_jbd(bh))
2135 : 0 : goto busy;
2136 [ # # ]: 0 : } while ((bh = bh->b_this_page) != head);
2137 : :
2138 : 0 : ret = try_to_free_buffers(page);
2139 : :
2140 : 0 : busy:
2141 : 0 : return ret;
2142 : : }
2143 : :
2144 : : /*
2145 : : * This buffer is no longer needed. If it is on an older transaction's
2146 : : * checkpoint list we need to record it on this transaction's forget list
2147 : : * to pin this buffer (and hence its checkpointing transaction) down until
2148 : : * this transaction commits. If the buffer isn't on a checkpoint list, we
2149 : : * release it.
2150 : : * Returns non-zero if JBD no longer has an interest in the buffer.
2151 : : *
2152 : : * Called under j_list_lock.
2153 : : *
2154 : : * Called under jh->b_state_lock.
2155 : : */
2156 : 0 : static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction)
2157 : : {
2158 : 0 : int may_free = 1;
2159 [ # # ]: 0 : struct buffer_head *bh = jh2bh(jh);
2160 : :
2161 [ # # ]: 0 : if (jh->b_cp_transaction) {
2162 : 0 : JBUFFER_TRACE(jh, "on running+cp transaction");
2163 : 0 : __jbd2_journal_temp_unlink_buffer(jh);
2164 : : /*
2165 : : * We don't want to write the buffer anymore, clear the
2166 : : * bit so that we don't confuse checks in
2167 : : * __journal_file_buffer
2168 : : */
2169 : 0 : clear_buffer_dirty(bh);
2170 : 0 : __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
2171 : 0 : may_free = 0;
2172 : : } else {
2173 : 0 : JBUFFER_TRACE(jh, "on running transaction");
2174 : 0 : __jbd2_journal_unfile_buffer(jh);
2175 : 0 : jbd2_journal_put_journal_head(jh);
2176 : : }
2177 : 0 : return may_free;
2178 : : }
2179 : :
2180 : : /*
2181 : : * jbd2_journal_invalidatepage
2182 : : *
2183 : : * This code is tricky. It has a number of cases to deal with.
2184 : : *
2185 : : * There are two invariants which this code relies on:
2186 : : *
2187 : : * i_size must be updated on disk before we start calling invalidatepage on the
2188 : : * data.
2189 : : *
2190 : : * This is done in ext3 by defining an ext3_setattr method which
2191 : : * updates i_size before truncate gets going. By maintaining this
2192 : : * invariant, we can be sure that it is safe to throw away any buffers
2193 : : * attached to the current transaction: once the transaction commits,
2194 : : * we know that the data will not be needed.
2195 : : *
2196 : : * Note however that we can *not* throw away data belonging to the
2197 : : * previous, committing transaction!
2198 : : *
2199 : : * Any disk blocks which *are* part of the previous, committing
2200 : : * transaction (and which therefore cannot be discarded immediately) are
2201 : : * not going to be reused in the new running transaction
2202 : : *
2203 : : * The bitmap committed_data images guarantee this: any block which is
2204 : : * allocated in one transaction and removed in the next will be marked
2205 : : * as in-use in the committed_data bitmap, so cannot be reused until
2206 : : * the next transaction to delete the block commits. This means that
2207 : : * leaving committing buffers dirty is quite safe: the disk blocks
2208 : : * cannot be reallocated to a different file and so buffer aliasing is
2209 : : * not possible.
2210 : : *
2211 : : *
2212 : : * The above applies mainly to ordered data mode. In writeback mode we
2213 : : * don't make guarantees about the order in which data hits disk --- in
2214 : : * particular we don't guarantee that new dirty data is flushed before
2215 : : * transaction commit --- so it is always safe just to discard data
2216 : : * immediately in that mode. --sct
2217 : : */
2218 : :
2219 : : /*
2220 : : * The journal_unmap_buffer helper function returns zero if the buffer
2221 : : * concerned remains pinned as an anonymous buffer belonging to an older
2222 : : * transaction.
2223 : : *
2224 : : * We're outside-transaction here. Either or both of j_running_transaction
2225 : : * and j_committing_transaction may be NULL.
2226 : : */
2227 : 0 : static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh,
2228 : : int partial_page)
2229 : : {
2230 : 0 : transaction_t *transaction;
2231 : 0 : struct journal_head *jh;
2232 : 0 : int may_free = 1;
2233 : :
2234 : 0 : BUFFER_TRACE(bh, "entry");
2235 : :
2236 : : /*
2237 : : * It is safe to proceed here without the j_list_lock because the
2238 : : * buffers cannot be stolen by try_to_free_buffers as long as we are
2239 : : * holding the page lock. --sct
2240 : : */
2241 : :
2242 : 0 : jh = jbd2_journal_grab_journal_head(bh);
2243 [ # # ]: 0 : if (!jh)
2244 : 0 : goto zap_buffer_unlocked;
2245 : :
2246 : : /* OK, we have data buffer in journaled mode */
2247 : 0 : write_lock(&journal->j_state_lock);
2248 : 0 : spin_lock(&jh->b_state_lock);
2249 : 0 : spin_lock(&journal->j_list_lock);
2250 : :
2251 : : /*
2252 : : * We cannot remove the buffer from checkpoint lists until the
2253 : : * transaction adding inode to orphan list (let's call it T)
2254 : : * is committed. Otherwise if the transaction changing the
2255 : : * buffer would be cleaned from the journal before T is
2256 : : * committed, a crash will cause that the correct contents of
2257 : : * the buffer will be lost. On the other hand we have to
2258 : : * clear the buffer dirty bit at latest at the moment when the
2259 : : * transaction marking the buffer as freed in the filesystem
2260 : : * structures is committed because from that moment on the
2261 : : * block can be reallocated and used by a different page.
2262 : : * Since the block hasn't been freed yet but the inode has
2263 : : * already been added to orphan list, it is safe for us to add
2264 : : * the buffer to BJ_Forget list of the newest transaction.
2265 : : *
2266 : : * Also we have to clear buffer_mapped flag of a truncated buffer
2267 : : * because the buffer_head may be attached to the page straddling
2268 : : * i_size (can happen only when blocksize < pagesize) and thus the
2269 : : * buffer_head can be reused when the file is extended again. So we end
2270 : : * up keeping around invalidated buffers attached to transactions'
2271 : : * BJ_Forget list just to stop checkpointing code from cleaning up
2272 : : * the transaction this buffer was modified in.
2273 : : */
2274 : 0 : transaction = jh->b_transaction;
2275 [ # # ]: 0 : if (transaction == NULL) {
2276 : : /* First case: not on any transaction. If it
2277 : : * has no checkpoint link, then we can zap it:
2278 : : * it's a writeback-mode buffer so we don't care
2279 : : * if it hits disk safely. */
2280 [ # # ]: 0 : if (!jh->b_cp_transaction) {
2281 : 0 : JBUFFER_TRACE(jh, "not on any transaction: zap");
2282 : 0 : goto zap_buffer;
2283 : : }
2284 : :
2285 [ # # ]: 0 : if (!buffer_dirty(bh)) {
2286 : : /* bdflush has written it. We can drop it now */
2287 : 0 : __jbd2_journal_remove_checkpoint(jh);
2288 : 0 : goto zap_buffer;
2289 : : }
2290 : :
2291 : : /* OK, it must be in the journal but still not
2292 : : * written fully to disk: it's metadata or
2293 : : * journaled data... */
2294 : :
2295 [ # # ]: 0 : if (journal->j_running_transaction) {
2296 : : /* ... and once the current transaction has
2297 : : * committed, the buffer won't be needed any
2298 : : * longer. */
2299 : 0 : JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget");
2300 : 0 : may_free = __dispose_buffer(jh,
2301 : : journal->j_running_transaction);
2302 : 0 : goto zap_buffer;
2303 : : } else {
2304 : : /* There is no currently-running transaction. So the
2305 : : * orphan record which we wrote for this file must have
2306 : : * passed into commit. We must attach this buffer to
2307 : : * the committing transaction, if it exists. */
2308 [ # # ]: 0 : if (journal->j_committing_transaction) {
2309 : 0 : JBUFFER_TRACE(jh, "give to committing trans");
2310 : 0 : may_free = __dispose_buffer(jh,
2311 : : journal->j_committing_transaction);
2312 : 0 : goto zap_buffer;
2313 : : } else {
2314 : : /* The orphan record's transaction has
2315 : : * committed. We can cleanse this buffer */
2316 : 0 : clear_buffer_jbddirty(bh);
2317 : 0 : __jbd2_journal_remove_checkpoint(jh);
2318 : 0 : goto zap_buffer;
2319 : : }
2320 : : }
2321 [ # # ]: 0 : } else if (transaction == journal->j_committing_transaction) {
2322 : 0 : JBUFFER_TRACE(jh, "on committing transaction");
2323 : : /*
2324 : : * The buffer is committing, we simply cannot touch
2325 : : * it. If the page is straddling i_size we have to wait
2326 : : * for commit and try again.
2327 : : */
2328 [ # # ]: 0 : if (partial_page) {
2329 : 0 : spin_unlock(&journal->j_list_lock);
2330 : 0 : spin_unlock(&jh->b_state_lock);
2331 : 0 : write_unlock(&journal->j_state_lock);
2332 : 0 : jbd2_journal_put_journal_head(jh);
2333 : 0 : return -EBUSY;
2334 : : }
2335 : : /*
2336 : : * OK, buffer won't be reachable after truncate. We just clear
2337 : : * b_modified to not confuse transaction credit accounting, and
2338 : : * set j_next_transaction to the running transaction (if there
2339 : : * is one) and mark buffer as freed so that commit code knows
2340 : : * it should clear dirty bits when it is done with the buffer.
2341 : : */
2342 : 0 : set_buffer_freed(bh);
2343 [ # # # # ]: 0 : if (journal->j_running_transaction && buffer_jbddirty(bh))
2344 : 0 : jh->b_next_transaction = journal->j_running_transaction;
2345 : 0 : jh->b_modified = 0;
2346 : 0 : spin_unlock(&journal->j_list_lock);
2347 : 0 : spin_unlock(&jh->b_state_lock);
2348 : 0 : write_unlock(&journal->j_state_lock);
2349 : 0 : jbd2_journal_put_journal_head(jh);
2350 : 0 : return 0;
2351 : : } else {
2352 : : /* Good, the buffer belongs to the running transaction.
2353 : : * We are writing our own transaction's data, not any
2354 : : * previous one's, so it is safe to throw it away
2355 : : * (remember that we expect the filesystem to have set
2356 : : * i_size already for this truncate so recovery will not
2357 : : * expose the disk blocks we are discarding here.) */
2358 [ # # ]: 0 : J_ASSERT_JH(jh, transaction == journal->j_running_transaction);
2359 : 0 : JBUFFER_TRACE(jh, "on running transaction");
2360 : 0 : may_free = __dispose_buffer(jh, transaction);
2361 : : }
2362 : :
2363 : 0 : zap_buffer:
2364 : : /*
2365 : : * This is tricky. Although the buffer is truncated, it may be reused
2366 : : * if blocksize < pagesize and it is attached to the page straddling
2367 : : * EOF. Since the buffer might have been added to BJ_Forget list of the
2368 : : * running transaction, journal_get_write_access() won't clear
2369 : : * b_modified and credit accounting gets confused. So clear b_modified
2370 : : * here.
2371 : : */
2372 : 0 : jh->b_modified = 0;
2373 : 0 : spin_unlock(&journal->j_list_lock);
2374 : 0 : spin_unlock(&jh->b_state_lock);
2375 : 0 : write_unlock(&journal->j_state_lock);
2376 : 0 : jbd2_journal_put_journal_head(jh);
2377 : 0 : zap_buffer_unlocked:
2378 : 0 : clear_buffer_dirty(bh);
2379 [ # # ]: 0 : J_ASSERT_BH(bh, !buffer_jbddirty(bh));
2380 : 0 : clear_buffer_mapped(bh);
2381 : 0 : clear_buffer_req(bh);
2382 : 0 : clear_buffer_new(bh);
2383 : 0 : clear_buffer_delay(bh);
2384 : 0 : clear_buffer_unwritten(bh);
2385 : 0 : bh->b_bdev = NULL;
2386 : 0 : return may_free;
2387 : : }
2388 : :
2389 : : /**
2390 : : * void jbd2_journal_invalidatepage()
2391 : : * @journal: journal to use for flush...
2392 : : * @page: page to flush
2393 : : * @offset: start of the range to invalidate
2394 : : * @length: length of the range to invalidate
2395 : : *
2396 : : * Reap page buffers containing data after in the specified range in page.
2397 : : * Can return -EBUSY if buffers are part of the committing transaction and
2398 : : * the page is straddling i_size. Caller then has to wait for current commit
2399 : : * and try again.
2400 : : */
2401 : 0 : int jbd2_journal_invalidatepage(journal_t *journal,
2402 : : struct page *page,
2403 : : unsigned int offset,
2404 : : unsigned int length)
2405 : : {
2406 : 0 : struct buffer_head *head, *bh, *next;
2407 : 0 : unsigned int stop = offset + length;
2408 : 0 : unsigned int curr_off = 0;
2409 : 0 : int partial_page = (offset || length < PAGE_SIZE);
2410 : 0 : int may_free = 1;
2411 : 0 : int ret = 0;
2412 : :
2413 [ # # # # ]: 0 : if (!PageLocked(page))
2414 : 0 : BUG();
2415 [ # # ]: 0 : if (!page_has_buffers(page))
2416 : : return 0;
2417 : :
2418 [ # # ]: 0 : BUG_ON(stop > PAGE_SIZE || stop < length);
2419 : :
2420 : : /* We will potentially be playing with lists other than just the
2421 : : * data lists (especially for journaled data mode), so be
2422 : : * cautious in our locking. */
2423 : :
2424 [ # # ]: 0 : head = bh = page_buffers(page);
2425 : 0 : do {
2426 : 0 : unsigned int next_off = curr_off + bh->b_size;
2427 : 0 : next = bh->b_this_page;
2428 : :
2429 [ # # ]: 0 : if (next_off > stop)
2430 : : return 0;
2431 : :
2432 [ # # ]: 0 : if (offset <= curr_off) {
2433 : : /* This block is wholly outside the truncation point */
2434 : 0 : lock_buffer(bh);
2435 : 0 : ret = journal_unmap_buffer(journal, bh, partial_page);
2436 : 0 : unlock_buffer(bh);
2437 [ # # ]: 0 : if (ret < 0)
2438 : 0 : return ret;
2439 : 0 : may_free &= ret;
2440 : : }
2441 : 0 : curr_off = next_off;
2442 : 0 : bh = next;
2443 : :
2444 [ # # ]: 0 : } while (bh != head);
2445 : :
2446 [ # # ]: 0 : if (!partial_page) {
2447 [ # # # # ]: 0 : if (may_free && try_to_free_buffers(page))
2448 [ # # ]: 0 : J_ASSERT(!page_has_buffers(page));
2449 : : }
2450 : : return 0;
2451 : : }
2452 : :
2453 : : /*
2454 : : * File a buffer on the given transaction list.
2455 : : */
2456 : 86915 : void __jbd2_journal_file_buffer(struct journal_head *jh,
2457 : : transaction_t *transaction, int jlist)
2458 : : {
2459 : 86915 : struct journal_head **list = NULL;
2460 : 86915 : int was_dirty = 0;
2461 : 86915 : struct buffer_head *bh = jh2bh(jh);
2462 : :
2463 : 86915 : lockdep_assert_held(&jh->b_state_lock);
2464 [ - + ]: 86915 : assert_spin_locked(&transaction->t_journal->j_list_lock);
2465 : :
2466 [ - + ]: 86915 : J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
2467 [ + + - + ]: 86915 : J_ASSERT_JH(jh, jh->b_transaction == transaction ||
2468 : : jh->b_transaction == NULL);
2469 : :
2470 [ + + + - ]: 86915 : if (jh->b_transaction && jh->b_jlist == jlist)
2471 : : return;
2472 : :
2473 [ + + ]: 86915 : if (jlist == BJ_Metadata || jlist == BJ_Reserved ||
2474 [ + - ]: 14448 : jlist == BJ_Shadow || jlist == BJ_Forget) {
2475 : : /*
2476 : : * For metadata buffers, we track dirty bit in buffer_jbddirty
2477 : : * instead of buffer_dirty. We should not see a dirty bit set
2478 : : * here because we clear it in do_get_write_access but e.g.
2479 : : * tune2fs can modify the sb and set the dirty bit at any time
2480 : : * so we try to gracefully handle that.
2481 : : */
2482 [ - + ]: 86915 : if (buffer_dirty(bh))
2483 : 0 : warn_dirty_buffer(bh);
2484 [ + - + + ]: 173830 : if (test_clear_buffer_dirty(bh) ||
2485 : : test_clear_buffer_jbddirty(bh))
2486 : : was_dirty = 1;
2487 : : }
2488 : :
2489 [ + + ]: 86915 : if (jh->b_transaction)
2490 : 50685 : __jbd2_journal_temp_unlink_buffer(jh);
2491 : : else
2492 : 36230 : jbd2_journal_grab_journal_head(bh);
2493 : 86915 : jh->b_transaction = transaction;
2494 : :
2495 [ - + + + : 86915 : switch (jlist) {
+ - ]
2496 : 0 : case BJ_None:
2497 [ # # ]: 0 : J_ASSERT_JH(jh, !jh->b_committed_data);
2498 [ # # ]: 0 : J_ASSERT_JH(jh, !jh->b_frozen_data);
2499 : : return;
2500 : 36237 : case BJ_Metadata:
2501 : 36237 : transaction->t_nr_buffers++;
2502 : 36237 : list = &transaction->t_buffers;
2503 : 36237 : break;
2504 : 7224 : case BJ_Forget:
2505 : 7224 : list = &transaction->t_forget;
2506 : 7224 : break;
2507 : 7224 : case BJ_Shadow:
2508 : 7224 : list = &transaction->t_shadow_list;
2509 : 7224 : break;
2510 : 36230 : case BJ_Reserved:
2511 : 36230 : list = &transaction->t_reserved_list;
2512 : 36230 : break;
2513 : : }
2514 : :
2515 [ + + ]: 86915 : __blist_add_buffer(list, jh);
2516 : 86915 : jh->b_jlist = jlist;
2517 : :
2518 [ + + ]: 86915 : if (was_dirty)
2519 : 56863 : set_buffer_jbddirty(bh);
2520 : : }
2521 : :
2522 : 7224 : void jbd2_journal_file_buffer(struct journal_head *jh,
2523 : : transaction_t *transaction, int jlist)
2524 : : {
2525 : 7224 : spin_lock(&jh->b_state_lock);
2526 : 7224 : spin_lock(&transaction->t_journal->j_list_lock);
2527 : 7224 : __jbd2_journal_file_buffer(jh, transaction, jlist);
2528 : 7224 : spin_unlock(&transaction->t_journal->j_list_lock);
2529 : 7224 : spin_unlock(&jh->b_state_lock);
2530 : 7224 : }
2531 : :
2532 : : /*
2533 : : * Remove a buffer from its current buffer list in preparation for
2534 : : * dropping it from its current transaction entirely. If the buffer has
2535 : : * already started to be used by a subsequent transaction, refile the
2536 : : * buffer on that transaction's metadata list.
2537 : : *
2538 : : * Called under j_list_lock
2539 : : * Called under jh->b_state_lock
2540 : : *
2541 : : * When this function returns true, there's no next transaction to refile to
2542 : : * and the caller has to drop jh reference through
2543 : : * jbd2_journal_put_journal_head().
2544 : : */
2545 : 7224 : bool __jbd2_journal_refile_buffer(struct journal_head *jh)
2546 : : {
2547 : 7224 : int was_dirty, jlist;
2548 [ + - ]: 7224 : struct buffer_head *bh = jh2bh(jh);
2549 : :
2550 : 7224 : lockdep_assert_held(&jh->b_state_lock);
2551 [ + - ]: 7224 : if (jh->b_transaction)
2552 [ - + ]: 7224 : assert_spin_locked(&jh->b_transaction->t_journal->j_list_lock);
2553 : :
2554 : : /* If the buffer is now unused, just drop it. */
2555 [ + + ]: 7224 : if (jh->b_next_transaction == NULL) {
2556 : 7217 : __jbd2_journal_unfile_buffer(jh);
2557 : 7217 : return true;
2558 : : }
2559 : :
2560 : : /*
2561 : : * It has been modified by a later transaction: add it to the new
2562 : : * transaction's metadata list.
2563 : : */
2564 : :
2565 : 7 : was_dirty = test_clear_buffer_jbddirty(bh);
2566 : 7 : __jbd2_journal_temp_unlink_buffer(jh);
2567 : : /*
2568 : : * We set b_transaction here because b_next_transaction will inherit
2569 : : * our jh reference and thus __jbd2_journal_file_buffer() must not
2570 : : * take a new one.
2571 : : */
2572 : 7 : WRITE_ONCE(jh->b_transaction, jh->b_next_transaction);
2573 : 7 : WRITE_ONCE(jh->b_next_transaction, NULL);
2574 [ + - ]: 7 : if (buffer_freed(bh))
2575 : : jlist = BJ_Forget;
2576 [ - + ]: 7 : else if (jh->b_modified)
2577 : : jlist = BJ_Metadata;
2578 : : else
2579 : 0 : jlist = BJ_Reserved;
2580 : 7 : __jbd2_journal_file_buffer(jh, jh->b_transaction, jlist);
2581 [ - + ]: 7 : J_ASSERT_JH(jh, jh->b_transaction->t_state == T_RUNNING);
2582 : :
2583 [ + - ]: 7 : if (was_dirty)
2584 : 7 : set_buffer_jbddirty(bh);
2585 : : return false;
2586 : : }
2587 : :
2588 : : /*
2589 : : * __jbd2_journal_refile_buffer() with necessary locking added. We take our
2590 : : * bh reference so that we can safely unlock bh.
2591 : : *
2592 : : * The jh and bh may be freed by this call.
2593 : : */
2594 : 0 : void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh)
2595 : : {
2596 : 0 : bool drop;
2597 : :
2598 : 0 : spin_lock(&jh->b_state_lock);
2599 : 0 : spin_lock(&journal->j_list_lock);
2600 : 0 : drop = __jbd2_journal_refile_buffer(jh);
2601 : 0 : spin_unlock(&jh->b_state_lock);
2602 : 0 : spin_unlock(&journal->j_list_lock);
2603 [ # # ]: 0 : if (drop)
2604 : 0 : jbd2_journal_put_journal_head(jh);
2605 : 0 : }
2606 : :
2607 : : /*
2608 : : * File inode in the inode list of the handle's transaction
2609 : : */
2610 : 0 : static int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode,
2611 : : unsigned long flags, loff_t start_byte, loff_t end_byte)
2612 : : {
2613 : 0 : transaction_t *transaction = handle->h_transaction;
2614 : 0 : journal_t *journal;
2615 : :
2616 [ # # # # ]: 0 : if (is_handle_aborted(handle))
2617 : : return -EROFS;
2618 : 0 : journal = transaction->t_journal;
2619 : :
2620 : : jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode->i_vfs_inode->i_ino,
2621 : 0 : transaction->t_tid);
2622 : :
2623 : 0 : spin_lock(&journal->j_list_lock);
2624 : 0 : jinode->i_flags |= flags;
2625 : :
2626 [ # # ]: 0 : if (jinode->i_dirty_end) {
2627 : 0 : jinode->i_dirty_start = min(jinode->i_dirty_start, start_byte);
2628 : 0 : jinode->i_dirty_end = max(jinode->i_dirty_end, end_byte);
2629 : : } else {
2630 : 0 : jinode->i_dirty_start = start_byte;
2631 : 0 : jinode->i_dirty_end = end_byte;
2632 : : }
2633 : :
2634 : : /* Is inode already attached where we need it? */
2635 [ # # ]: 0 : if (jinode->i_transaction == transaction ||
2636 [ # # ]: 0 : jinode->i_next_transaction == transaction)
2637 : 0 : goto done;
2638 : :
2639 : : /*
2640 : : * We only ever set this variable to 1 so the test is safe. Since
2641 : : * t_need_data_flush is likely to be set, we do the test to save some
2642 : : * cacheline bouncing
2643 : : */
2644 [ # # ]: 0 : if (!transaction->t_need_data_flush)
2645 : 0 : transaction->t_need_data_flush = 1;
2646 : : /* On some different transaction's list - should be
2647 : : * the committing one */
2648 [ # # ]: 0 : if (jinode->i_transaction) {
2649 [ # # ]: 0 : J_ASSERT(jinode->i_next_transaction == NULL);
2650 [ # # ]: 0 : J_ASSERT(jinode->i_transaction ==
2651 : : journal->j_committing_transaction);
2652 : 0 : jinode->i_next_transaction = transaction;
2653 : 0 : goto done;
2654 : : }
2655 : : /* Not on any transaction list... */
2656 [ # # ]: 0 : J_ASSERT(!jinode->i_next_transaction);
2657 : 0 : jinode->i_transaction = transaction;
2658 : 0 : list_add(&jinode->i_list, &transaction->t_inode_list);
2659 : 0 : done:
2660 : 0 : spin_unlock(&journal->j_list_lock);
2661 : :
2662 : 0 : return 0;
2663 : : }
2664 : :
2665 : 0 : int jbd2_journal_inode_ranged_write(handle_t *handle,
2666 : : struct jbd2_inode *jinode, loff_t start_byte, loff_t length)
2667 : : {
2668 : 0 : return jbd2_journal_file_inode(handle, jinode,
2669 : : JI_WRITE_DATA | JI_WAIT_DATA, start_byte,
2670 : 0 : start_byte + length - 1);
2671 : : }
2672 : :
2673 : 0 : int jbd2_journal_inode_ranged_wait(handle_t *handle, struct jbd2_inode *jinode,
2674 : : loff_t start_byte, loff_t length)
2675 : : {
2676 : 0 : return jbd2_journal_file_inode(handle, jinode, JI_WAIT_DATA,
2677 : 0 : start_byte, start_byte + length - 1);
2678 : : }
2679 : :
2680 : : /*
2681 : : * File truncate and transaction commit interact with each other in a
2682 : : * non-trivial way. If a transaction writing data block A is
2683 : : * committing, we cannot discard the data by truncate until we have
2684 : : * written them. Otherwise if we crashed after the transaction with
2685 : : * write has committed but before the transaction with truncate has
2686 : : * committed, we could see stale data in block A. This function is a
2687 : : * helper to solve this problem. It starts writeout of the truncated
2688 : : * part in case it is in the committing transaction.
2689 : : *
2690 : : * Filesystem code must call this function when inode is journaled in
2691 : : * ordered mode before truncation happens and after the inode has been
2692 : : * placed on orphan list with the new inode size. The second condition
2693 : : * avoids the race that someone writes new data and we start
2694 : : * committing the transaction after this function has been called but
2695 : : * before a transaction for truncate is started (and furthermore it
2696 : : * allows us to optimize the case where the addition to orphan list
2697 : : * happens in the same transaction as write --- we don't have to write
2698 : : * any data in such case).
2699 : : */
2700 : 0 : int jbd2_journal_begin_ordered_truncate(journal_t *journal,
2701 : : struct jbd2_inode *jinode,
2702 : : loff_t new_size)
2703 : : {
2704 : 0 : transaction_t *inode_trans, *commit_trans;
2705 : 0 : int ret = 0;
2706 : :
2707 : : /* This is a quick check to avoid locking if not necessary */
2708 [ # # ]: 0 : if (!jinode->i_transaction)
2709 : 0 : goto out;
2710 : : /* Locks are here just to force reading of recent values, it is
2711 : : * enough that the transaction was not committing before we started
2712 : : * a transaction adding the inode to orphan list */
2713 : 0 : read_lock(&journal->j_state_lock);
2714 : 0 : commit_trans = journal->j_committing_transaction;
2715 : 0 : read_unlock(&journal->j_state_lock);
2716 : 0 : spin_lock(&journal->j_list_lock);
2717 : 0 : inode_trans = jinode->i_transaction;
2718 : 0 : spin_unlock(&journal->j_list_lock);
2719 [ # # ]: 0 : if (inode_trans == commit_trans) {
2720 : 0 : ret = filemap_fdatawrite_range(jinode->i_vfs_inode->i_mapping,
2721 : : new_size, LLONG_MAX);
2722 [ # # ]: 0 : if (ret)
2723 : 0 : jbd2_journal_abort(journal, ret);
2724 : : }
2725 : 0 : out:
2726 : 0 : return ret;
2727 : : }
|
