diff options
Diffstat (limited to 'fs/jbd2')
-rw-r--r-- | fs/jbd2/checkpoint.c | 18 | ||||
-rw-r--r-- | fs/jbd2/commit.c | 50 | ||||
-rw-r--r-- | fs/jbd2/journal.c | 121 | ||||
-rw-r--r-- | fs/jbd2/recovery.c | 10 | ||||
-rw-r--r-- | fs/jbd2/transaction.c | 233 |
5 files changed, 236 insertions, 196 deletions
diff --git a/fs/jbd2/checkpoint.c b/fs/jbd2/checkpoint.c index 076d1cc44f95..1c23a0f4e8a3 100644 --- a/fs/jbd2/checkpoint.c +++ b/fs/jbd2/checkpoint.c | |||
@@ -118,13 +118,13 @@ static int __try_to_free_cp_buf(struct journal_head *jh) | |||
118 | void __jbd2_log_wait_for_space(journal_t *journal) | 118 | void __jbd2_log_wait_for_space(journal_t *journal) |
119 | { | 119 | { |
120 | int nblocks, space_left; | 120 | int nblocks, space_left; |
121 | assert_spin_locked(&journal->j_state_lock); | 121 | /* assert_spin_locked(&journal->j_state_lock); */ |
122 | 122 | ||
123 | nblocks = jbd_space_needed(journal); | 123 | nblocks = jbd_space_needed(journal); |
124 | while (__jbd2_log_space_left(journal) < nblocks) { | 124 | while (__jbd2_log_space_left(journal) < nblocks) { |
125 | if (journal->j_flags & JBD2_ABORT) | 125 | if (journal->j_flags & JBD2_ABORT) |
126 | return; | 126 | return; |
127 | spin_unlock(&journal->j_state_lock); | 127 | write_unlock(&journal->j_state_lock); |
128 | mutex_lock(&journal->j_checkpoint_mutex); | 128 | mutex_lock(&journal->j_checkpoint_mutex); |
129 | 129 | ||
130 | /* | 130 | /* |
@@ -138,7 +138,7 @@ void __jbd2_log_wait_for_space(journal_t *journal) | |||
138 | * filesystem, so abort the journal and leave a stack | 138 | * filesystem, so abort the journal and leave a stack |
139 | * trace for forensic evidence. | 139 | * trace for forensic evidence. |
140 | */ | 140 | */ |
141 | spin_lock(&journal->j_state_lock); | 141 | write_lock(&journal->j_state_lock); |
142 | spin_lock(&journal->j_list_lock); | 142 | spin_lock(&journal->j_list_lock); |
143 | nblocks = jbd_space_needed(journal); | 143 | nblocks = jbd_space_needed(journal); |
144 | space_left = __jbd2_log_space_left(journal); | 144 | space_left = __jbd2_log_space_left(journal); |
@@ -149,7 +149,7 @@ void __jbd2_log_wait_for_space(journal_t *journal) | |||
149 | if (journal->j_committing_transaction) | 149 | if (journal->j_committing_transaction) |
150 | tid = journal->j_committing_transaction->t_tid; | 150 | tid = journal->j_committing_transaction->t_tid; |
151 | spin_unlock(&journal->j_list_lock); | 151 | spin_unlock(&journal->j_list_lock); |
152 | spin_unlock(&journal->j_state_lock); | 152 | write_unlock(&journal->j_state_lock); |
153 | if (chkpt) { | 153 | if (chkpt) { |
154 | jbd2_log_do_checkpoint(journal); | 154 | jbd2_log_do_checkpoint(journal); |
155 | } else if (jbd2_cleanup_journal_tail(journal) == 0) { | 155 | } else if (jbd2_cleanup_journal_tail(journal) == 0) { |
@@ -167,7 +167,7 @@ void __jbd2_log_wait_for_space(journal_t *journal) | |||
167 | WARN_ON(1); | 167 | WARN_ON(1); |
168 | jbd2_journal_abort(journal, 0); | 168 | jbd2_journal_abort(journal, 0); |
169 | } | 169 | } |
170 | spin_lock(&journal->j_state_lock); | 170 | write_lock(&journal->j_state_lock); |
171 | } else { | 171 | } else { |
172 | spin_unlock(&journal->j_list_lock); | 172 | spin_unlock(&journal->j_list_lock); |
173 | } | 173 | } |
@@ -474,7 +474,7 @@ int jbd2_cleanup_journal_tail(journal_t *journal) | |||
474 | * next transaction ID we will write, and where it will | 474 | * next transaction ID we will write, and where it will |
475 | * start. */ | 475 | * start. */ |
476 | 476 | ||
477 | spin_lock(&journal->j_state_lock); | 477 | write_lock(&journal->j_state_lock); |
478 | spin_lock(&journal->j_list_lock); | 478 | spin_lock(&journal->j_list_lock); |
479 | transaction = journal->j_checkpoint_transactions; | 479 | transaction = journal->j_checkpoint_transactions; |
480 | if (transaction) { | 480 | if (transaction) { |
@@ -496,7 +496,7 @@ int jbd2_cleanup_journal_tail(journal_t *journal) | |||
496 | /* If the oldest pinned transaction is at the tail of the log | 496 | /* If the oldest pinned transaction is at the tail of the log |
497 | already then there's not much we can do right now. */ | 497 | already then there's not much we can do right now. */ |
498 | if (journal->j_tail_sequence == first_tid) { | 498 | if (journal->j_tail_sequence == first_tid) { |
499 | spin_unlock(&journal->j_state_lock); | 499 | write_unlock(&journal->j_state_lock); |
500 | return 1; | 500 | return 1; |
501 | } | 501 | } |
502 | 502 | ||
@@ -516,7 +516,7 @@ int jbd2_cleanup_journal_tail(journal_t *journal) | |||
516 | journal->j_free += freed; | 516 | journal->j_free += freed; |
517 | journal->j_tail_sequence = first_tid; | 517 | journal->j_tail_sequence = first_tid; |
518 | journal->j_tail = blocknr; | 518 | journal->j_tail = blocknr; |
519 | spin_unlock(&journal->j_state_lock); | 519 | write_unlock(&journal->j_state_lock); |
520 | 520 | ||
521 | /* | 521 | /* |
522 | * If there is an external journal, we need to make sure that | 522 | * If there is an external journal, we need to make sure that |
@@ -775,7 +775,7 @@ void __jbd2_journal_drop_transaction(journal_t *journal, transaction_t *transact | |||
775 | J_ASSERT(transaction->t_log_list == NULL); | 775 | J_ASSERT(transaction->t_log_list == NULL); |
776 | J_ASSERT(transaction->t_checkpoint_list == NULL); | 776 | J_ASSERT(transaction->t_checkpoint_list == NULL); |
777 | J_ASSERT(transaction->t_checkpoint_io_list == NULL); | 777 | J_ASSERT(transaction->t_checkpoint_io_list == NULL); |
778 | J_ASSERT(transaction->t_updates == 0); | 778 | J_ASSERT(atomic_read(&transaction->t_updates) == 0); |
779 | J_ASSERT(journal->j_committing_transaction != transaction); | 779 | J_ASSERT(journal->j_committing_transaction != transaction); |
780 | J_ASSERT(journal->j_running_transaction != transaction); | 780 | J_ASSERT(journal->j_running_transaction != transaction); |
781 | 781 | ||
diff --git a/fs/jbd2/commit.c b/fs/jbd2/commit.c index 75716d3d2be0..f52e5e8049f1 100644 --- a/fs/jbd2/commit.c +++ b/fs/jbd2/commit.c | |||
@@ -150,11 +150,11 @@ static int journal_submit_commit_record(journal_t *journal, | |||
150 | */ | 150 | */ |
151 | if (ret == -EOPNOTSUPP && barrier_done) { | 151 | if (ret == -EOPNOTSUPP && barrier_done) { |
152 | printk(KERN_WARNING | 152 | printk(KERN_WARNING |
153 | "JBD: barrier-based sync failed on %s - " | 153 | "JBD2: Disabling barriers on %s, " |
154 | "disabling barriers\n", journal->j_devname); | 154 | "not supported by device\n", journal->j_devname); |
155 | spin_lock(&journal->j_state_lock); | 155 | write_lock(&journal->j_state_lock); |
156 | journal->j_flags &= ~JBD2_BARRIER; | 156 | journal->j_flags &= ~JBD2_BARRIER; |
157 | spin_unlock(&journal->j_state_lock); | 157 | write_unlock(&journal->j_state_lock); |
158 | 158 | ||
159 | /* And try again, without the barrier */ | 159 | /* And try again, without the barrier */ |
160 | lock_buffer(bh); | 160 | lock_buffer(bh); |
@@ -180,11 +180,11 @@ retry: | |||
180 | wait_on_buffer(bh); | 180 | wait_on_buffer(bh); |
181 | if (buffer_eopnotsupp(bh) && (journal->j_flags & JBD2_BARRIER)) { | 181 | if (buffer_eopnotsupp(bh) && (journal->j_flags & JBD2_BARRIER)) { |
182 | printk(KERN_WARNING | 182 | printk(KERN_WARNING |
183 | "JBD2: wait_on_commit_record: sync failed on %s - " | 183 | "JBD2: %s: disabling barries on %s - not supported " |
184 | "disabling barriers\n", journal->j_devname); | 184 | "by device\n", __func__, journal->j_devname); |
185 | spin_lock(&journal->j_state_lock); | 185 | write_lock(&journal->j_state_lock); |
186 | journal->j_flags &= ~JBD2_BARRIER; | 186 | journal->j_flags &= ~JBD2_BARRIER; |
187 | spin_unlock(&journal->j_state_lock); | 187 | write_unlock(&journal->j_state_lock); |
188 | 188 | ||
189 | lock_buffer(bh); | 189 | lock_buffer(bh); |
190 | clear_buffer_dirty(bh); | 190 | clear_buffer_dirty(bh); |
@@ -400,7 +400,7 @@ void jbd2_journal_commit_transaction(journal_t *journal) | |||
400 | jbd_debug(1, "JBD: starting commit of transaction %d\n", | 400 | jbd_debug(1, "JBD: starting commit of transaction %d\n", |
401 | commit_transaction->t_tid); | 401 | commit_transaction->t_tid); |
402 | 402 | ||
403 | spin_lock(&journal->j_state_lock); | 403 | write_lock(&journal->j_state_lock); |
404 | commit_transaction->t_state = T_LOCKED; | 404 | commit_transaction->t_state = T_LOCKED; |
405 | 405 | ||
406 | /* | 406 | /* |
@@ -417,23 +417,23 @@ void jbd2_journal_commit_transaction(journal_t *journal) | |||
417 | stats.run.rs_locked); | 417 | stats.run.rs_locked); |
418 | 418 | ||
419 | spin_lock(&commit_transaction->t_handle_lock); | 419 | spin_lock(&commit_transaction->t_handle_lock); |
420 | while (commit_transaction->t_updates) { | 420 | while (atomic_read(&commit_transaction->t_updates)) { |
421 | DEFINE_WAIT(wait); | 421 | DEFINE_WAIT(wait); |
422 | 422 | ||
423 | prepare_to_wait(&journal->j_wait_updates, &wait, | 423 | prepare_to_wait(&journal->j_wait_updates, &wait, |
424 | TASK_UNINTERRUPTIBLE); | 424 | TASK_UNINTERRUPTIBLE); |
425 | if (commit_transaction->t_updates) { | 425 | if (atomic_read(&commit_transaction->t_updates)) { |
426 | spin_unlock(&commit_transaction->t_handle_lock); | 426 | spin_unlock(&commit_transaction->t_handle_lock); |
427 | spin_unlock(&journal->j_state_lock); | 427 | write_unlock(&journal->j_state_lock); |
428 | schedule(); | 428 | schedule(); |
429 | spin_lock(&journal->j_state_lock); | 429 | write_lock(&journal->j_state_lock); |
430 | spin_lock(&commit_transaction->t_handle_lock); | 430 | spin_lock(&commit_transaction->t_handle_lock); |
431 | } | 431 | } |
432 | finish_wait(&journal->j_wait_updates, &wait); | 432 | finish_wait(&journal->j_wait_updates, &wait); |
433 | } | 433 | } |
434 | spin_unlock(&commit_transaction->t_handle_lock); | 434 | spin_unlock(&commit_transaction->t_handle_lock); |
435 | 435 | ||
436 | J_ASSERT (commit_transaction->t_outstanding_credits <= | 436 | J_ASSERT (atomic_read(&commit_transaction->t_outstanding_credits) <= |
437 | journal->j_max_transaction_buffers); | 437 | journal->j_max_transaction_buffers); |
438 | 438 | ||
439 | /* | 439 | /* |
@@ -497,7 +497,7 @@ void jbd2_journal_commit_transaction(journal_t *journal) | |||
497 | start_time = ktime_get(); | 497 | start_time = ktime_get(); |
498 | commit_transaction->t_log_start = journal->j_head; | 498 | commit_transaction->t_log_start = journal->j_head; |
499 | wake_up(&journal->j_wait_transaction_locked); | 499 | wake_up(&journal->j_wait_transaction_locked); |
500 | spin_unlock(&journal->j_state_lock); | 500 | write_unlock(&journal->j_state_lock); |
501 | 501 | ||
502 | jbd_debug (3, "JBD: commit phase 2\n"); | 502 | jbd_debug (3, "JBD: commit phase 2\n"); |
503 | 503 | ||
@@ -519,19 +519,20 @@ void jbd2_journal_commit_transaction(journal_t *journal) | |||
519 | * transaction! Now comes the tricky part: we need to write out | 519 | * transaction! Now comes the tricky part: we need to write out |
520 | * metadata. Loop over the transaction's entire buffer list: | 520 | * metadata. Loop over the transaction's entire buffer list: |
521 | */ | 521 | */ |
522 | spin_lock(&journal->j_state_lock); | 522 | write_lock(&journal->j_state_lock); |
523 | commit_transaction->t_state = T_COMMIT; | 523 | commit_transaction->t_state = T_COMMIT; |
524 | spin_unlock(&journal->j_state_lock); | 524 | write_unlock(&journal->j_state_lock); |
525 | 525 | ||
526 | trace_jbd2_commit_logging(journal, commit_transaction); | 526 | trace_jbd2_commit_logging(journal, commit_transaction); |
527 | stats.run.rs_logging = jiffies; | 527 | stats.run.rs_logging = jiffies; |
528 | stats.run.rs_flushing = jbd2_time_diff(stats.run.rs_flushing, | 528 | stats.run.rs_flushing = jbd2_time_diff(stats.run.rs_flushing, |
529 | stats.run.rs_logging); | 529 | stats.run.rs_logging); |
530 | stats.run.rs_blocks = commit_transaction->t_outstanding_credits; | 530 | stats.run.rs_blocks = |
531 | atomic_read(&commit_transaction->t_outstanding_credits); | ||
531 | stats.run.rs_blocks_logged = 0; | 532 | stats.run.rs_blocks_logged = 0; |
532 | 533 | ||
533 | J_ASSERT(commit_transaction->t_nr_buffers <= | 534 | J_ASSERT(commit_transaction->t_nr_buffers <= |
534 | commit_transaction->t_outstanding_credits); | 535 | atomic_read(&commit_transaction->t_outstanding_credits)); |
535 | 536 | ||
536 | err = 0; | 537 | err = 0; |
537 | descriptor = NULL; | 538 | descriptor = NULL; |
@@ -616,7 +617,7 @@ void jbd2_journal_commit_transaction(journal_t *journal) | |||
616 | * the free space in the log, but this counter is changed | 617 | * the free space in the log, but this counter is changed |
617 | * by jbd2_journal_next_log_block() also. | 618 | * by jbd2_journal_next_log_block() also. |
618 | */ | 619 | */ |
619 | commit_transaction->t_outstanding_credits--; | 620 | atomic_dec(&commit_transaction->t_outstanding_credits); |
620 | 621 | ||
621 | /* Bump b_count to prevent truncate from stumbling over | 622 | /* Bump b_count to prevent truncate from stumbling over |
622 | the shadowed buffer! @@@ This can go if we ever get | 623 | the shadowed buffer! @@@ This can go if we ever get |
@@ -977,7 +978,7 @@ restart_loop: | |||
977 | * __jbd2_journal_drop_transaction(). Otherwise we could race with | 978 | * __jbd2_journal_drop_transaction(). Otherwise we could race with |
978 | * other checkpointing code processing the transaction... | 979 | * other checkpointing code processing the transaction... |
979 | */ | 980 | */ |
980 | spin_lock(&journal->j_state_lock); | 981 | write_lock(&journal->j_state_lock); |
981 | spin_lock(&journal->j_list_lock); | 982 | spin_lock(&journal->j_list_lock); |
982 | /* | 983 | /* |
983 | * Now recheck if some buffers did not get attached to the transaction | 984 | * Now recheck if some buffers did not get attached to the transaction |
@@ -985,7 +986,7 @@ restart_loop: | |||
985 | */ | 986 | */ |
986 | if (commit_transaction->t_forget) { | 987 | if (commit_transaction->t_forget) { |
987 | spin_unlock(&journal->j_list_lock); | 988 | spin_unlock(&journal->j_list_lock); |
988 | spin_unlock(&journal->j_state_lock); | 989 | write_unlock(&journal->j_state_lock); |
989 | goto restart_loop; | 990 | goto restart_loop; |
990 | } | 991 | } |
991 | 992 | ||
@@ -1003,7 +1004,8 @@ restart_loop: | |||
1003 | * File the transaction statistics | 1004 | * File the transaction statistics |
1004 | */ | 1005 | */ |
1005 | stats.ts_tid = commit_transaction->t_tid; | 1006 | stats.ts_tid = commit_transaction->t_tid; |
1006 | stats.run.rs_handle_count = commit_transaction->t_handle_count; | 1007 | stats.run.rs_handle_count = |
1008 | atomic_read(&commit_transaction->t_handle_count); | ||
1007 | trace_jbd2_run_stats(journal->j_fs_dev->bd_dev, | 1009 | trace_jbd2_run_stats(journal->j_fs_dev->bd_dev, |
1008 | commit_transaction->t_tid, &stats.run); | 1010 | commit_transaction->t_tid, &stats.run); |
1009 | 1011 | ||
@@ -1037,7 +1039,7 @@ restart_loop: | |||
1037 | journal->j_average_commit_time*3) / 4; | 1039 | journal->j_average_commit_time*3) / 4; |
1038 | else | 1040 | else |
1039 | journal->j_average_commit_time = commit_time; | 1041 | journal->j_average_commit_time = commit_time; |
1040 | spin_unlock(&journal->j_state_lock); | 1042 | write_unlock(&journal->j_state_lock); |
1041 | 1043 | ||
1042 | if (commit_transaction->t_checkpoint_list == NULL && | 1044 | if (commit_transaction->t_checkpoint_list == NULL && |
1043 | commit_transaction->t_checkpoint_io_list == NULL) { | 1045 | commit_transaction->t_checkpoint_io_list == NULL) { |
diff --git a/fs/jbd2/journal.c b/fs/jbd2/journal.c index 036880895bfc..ad5866aaf0f9 100644 --- a/fs/jbd2/journal.c +++ b/fs/jbd2/journal.c | |||
@@ -41,6 +41,7 @@ | |||
41 | #include <linux/hash.h> | 41 | #include <linux/hash.h> |
42 | #include <linux/log2.h> | 42 | #include <linux/log2.h> |
43 | #include <linux/vmalloc.h> | 43 | #include <linux/vmalloc.h> |
44 | #include <linux/backing-dev.h> | ||
44 | 45 | ||
45 | #define CREATE_TRACE_POINTS | 46 | #define CREATE_TRACE_POINTS |
46 | #include <trace/events/jbd2.h> | 47 | #include <trace/events/jbd2.h> |
@@ -48,8 +49,6 @@ | |||
48 | #include <asm/uaccess.h> | 49 | #include <asm/uaccess.h> |
49 | #include <asm/page.h> | 50 | #include <asm/page.h> |
50 | 51 | ||
51 | EXPORT_SYMBOL(jbd2_journal_start); | ||
52 | EXPORT_SYMBOL(jbd2_journal_restart); | ||
53 | EXPORT_SYMBOL(jbd2_journal_extend); | 52 | EXPORT_SYMBOL(jbd2_journal_extend); |
54 | EXPORT_SYMBOL(jbd2_journal_stop); | 53 | EXPORT_SYMBOL(jbd2_journal_stop); |
55 | EXPORT_SYMBOL(jbd2_journal_lock_updates); | 54 | EXPORT_SYMBOL(jbd2_journal_lock_updates); |
@@ -143,7 +142,7 @@ static int kjournald2(void *arg) | |||
143 | /* | 142 | /* |
144 | * And now, wait forever for commit wakeup events. | 143 | * And now, wait forever for commit wakeup events. |
145 | */ | 144 | */ |
146 | spin_lock(&journal->j_state_lock); | 145 | write_lock(&journal->j_state_lock); |
147 | 146 | ||
148 | loop: | 147 | loop: |
149 | if (journal->j_flags & JBD2_UNMOUNT) | 148 | if (journal->j_flags & JBD2_UNMOUNT) |
@@ -154,10 +153,10 @@ loop: | |||
154 | 153 | ||
155 | if (journal->j_commit_sequence != journal->j_commit_request) { | 154 | if (journal->j_commit_sequence != journal->j_commit_request) { |
156 | jbd_debug(1, "OK, requests differ\n"); | 155 | jbd_debug(1, "OK, requests differ\n"); |
157 | spin_unlock(&journal->j_state_lock); | 156 | write_unlock(&journal->j_state_lock); |
158 | del_timer_sync(&journal->j_commit_timer); | 157 | del_timer_sync(&journal->j_commit_timer); |
159 | jbd2_journal_commit_transaction(journal); | 158 | jbd2_journal_commit_transaction(journal); |
160 | spin_lock(&journal->j_state_lock); | 159 | write_lock(&journal->j_state_lock); |
161 | goto loop; | 160 | goto loop; |
162 | } | 161 | } |
163 | 162 | ||
@@ -169,9 +168,9 @@ loop: | |||
169 | * be already stopped. | 168 | * be already stopped. |
170 | */ | 169 | */ |
171 | jbd_debug(1, "Now suspending kjournald2\n"); | 170 | jbd_debug(1, "Now suspending kjournald2\n"); |
172 | spin_unlock(&journal->j_state_lock); | 171 | write_unlock(&journal->j_state_lock); |
173 | refrigerator(); | 172 | refrigerator(); |
174 | spin_lock(&journal->j_state_lock); | 173 | write_lock(&journal->j_state_lock); |
175 | } else { | 174 | } else { |
176 | /* | 175 | /* |
177 | * We assume on resume that commits are already there, | 176 | * We assume on resume that commits are already there, |
@@ -191,9 +190,9 @@ loop: | |||
191 | if (journal->j_flags & JBD2_UNMOUNT) | 190 | if (journal->j_flags & JBD2_UNMOUNT) |
192 | should_sleep = 0; | 191 | should_sleep = 0; |
193 | if (should_sleep) { | 192 | if (should_sleep) { |
194 | spin_unlock(&journal->j_state_lock); | 193 | write_unlock(&journal->j_state_lock); |
195 | schedule(); | 194 | schedule(); |
196 | spin_lock(&journal->j_state_lock); | 195 | write_lock(&journal->j_state_lock); |
197 | } | 196 | } |
198 | finish_wait(&journal->j_wait_commit, &wait); | 197 | finish_wait(&journal->j_wait_commit, &wait); |
199 | } | 198 | } |
@@ -211,7 +210,7 @@ loop: | |||
211 | goto loop; | 210 | goto loop; |
212 | 211 | ||
213 | end_loop: | 212 | end_loop: |
214 | spin_unlock(&journal->j_state_lock); | 213 | write_unlock(&journal->j_state_lock); |
215 | del_timer_sync(&journal->j_commit_timer); | 214 | del_timer_sync(&journal->j_commit_timer); |
216 | journal->j_task = NULL; | 215 | journal->j_task = NULL; |
217 | wake_up(&journal->j_wait_done_commit); | 216 | wake_up(&journal->j_wait_done_commit); |
@@ -234,16 +233,16 @@ static int jbd2_journal_start_thread(journal_t *journal) | |||
234 | 233 | ||
235 | static void journal_kill_thread(journal_t *journal) | 234 | static void journal_kill_thread(journal_t *journal) |
236 | { | 235 | { |
237 | spin_lock(&journal->j_state_lock); | 236 | write_lock(&journal->j_state_lock); |
238 | journal->j_flags |= JBD2_UNMOUNT; | 237 | journal->j_flags |= JBD2_UNMOUNT; |
239 | 238 | ||
240 | while (journal->j_task) { | 239 | while (journal->j_task) { |
241 | wake_up(&journal->j_wait_commit); | 240 | wake_up(&journal->j_wait_commit); |
242 | spin_unlock(&journal->j_state_lock); | 241 | write_unlock(&journal->j_state_lock); |
243 | wait_event(journal->j_wait_done_commit, journal->j_task == NULL); | 242 | wait_event(journal->j_wait_done_commit, journal->j_task == NULL); |
244 | spin_lock(&journal->j_state_lock); | 243 | write_lock(&journal->j_state_lock); |
245 | } | 244 | } |
246 | spin_unlock(&journal->j_state_lock); | 245 | write_unlock(&journal->j_state_lock); |
247 | } | 246 | } |
248 | 247 | ||
249 | /* | 248 | /* |
@@ -310,7 +309,17 @@ int jbd2_journal_write_metadata_buffer(transaction_t *transaction, | |||
310 | */ | 309 | */ |
311 | J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in)); | 310 | J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in)); |
312 | 311 | ||
313 | new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL); | 312 | retry_alloc: |
313 | new_bh = alloc_buffer_head(GFP_NOFS); | ||
314 | if (!new_bh) { | ||
315 | /* | ||
316 | * Failure is not an option, but __GFP_NOFAIL is going | ||
317 | * away; so we retry ourselves here. | ||
318 | */ | ||
319 | congestion_wait(BLK_RW_ASYNC, HZ/50); | ||
320 | goto retry_alloc; | ||
321 | } | ||
322 | |||
314 | /* keep subsequent assertions sane */ | 323 | /* keep subsequent assertions sane */ |
315 | new_bh->b_state = 0; | 324 | new_bh->b_state = 0; |
316 | init_buffer(new_bh, NULL, NULL); | 325 | init_buffer(new_bh, NULL, NULL); |
@@ -442,7 +451,7 @@ int __jbd2_log_space_left(journal_t *journal) | |||
442 | { | 451 | { |
443 | int left = journal->j_free; | 452 | int left = journal->j_free; |
444 | 453 | ||
445 | assert_spin_locked(&journal->j_state_lock); | 454 | /* assert_spin_locked(&journal->j_state_lock); */ |
446 | 455 | ||
447 | /* | 456 | /* |
448 | * Be pessimistic here about the number of those free blocks which | 457 | * Be pessimistic here about the number of those free blocks which |
@@ -487,9 +496,9 @@ int jbd2_log_start_commit(journal_t *journal, tid_t tid) | |||
487 | { | 496 | { |
488 | int ret; | 497 | int ret; |
489 | 498 | ||
490 | spin_lock(&journal->j_state_lock); | 499 | write_lock(&journal->j_state_lock); |
491 | ret = __jbd2_log_start_commit(journal, tid); | 500 | ret = __jbd2_log_start_commit(journal, tid); |
492 | spin_unlock(&journal->j_state_lock); | 501 | write_unlock(&journal->j_state_lock); |
493 | return ret; | 502 | return ret; |
494 | } | 503 | } |
495 | 504 | ||
@@ -508,7 +517,7 @@ int jbd2_journal_force_commit_nested(journal_t *journal) | |||
508 | transaction_t *transaction = NULL; | 517 | transaction_t *transaction = NULL; |
509 | tid_t tid; | 518 | tid_t tid; |
510 | 519 | ||
511 | spin_lock(&journal->j_state_lock); | 520 | read_lock(&journal->j_state_lock); |
512 | if (journal->j_running_transaction && !current->journal_info) { | 521 | if (journal->j_running_transaction && !current->journal_info) { |
513 | transaction = journal->j_running_transaction; | 522 | transaction = journal->j_running_transaction; |
514 | __jbd2_log_start_commit(journal, transaction->t_tid); | 523 | __jbd2_log_start_commit(journal, transaction->t_tid); |
@@ -516,12 +525,12 @@ int jbd2_journal_force_commit_nested(journal_t *journal) | |||
516 | transaction = journal->j_committing_transaction; | 525 | transaction = journal->j_committing_transaction; |
517 | 526 | ||
518 | if (!transaction) { | 527 | if (!transaction) { |
519 | spin_unlock(&journal->j_state_lock); | 528 | read_unlock(&journal->j_state_lock); |
520 | return 0; /* Nothing to retry */ | 529 | return 0; /* Nothing to retry */ |
521 | } | 530 | } |
522 | 531 | ||
523 | tid = transaction->t_tid; | 532 | tid = transaction->t_tid; |
524 | spin_unlock(&journal->j_state_lock); | 533 | read_unlock(&journal->j_state_lock); |
525 | jbd2_log_wait_commit(journal, tid); | 534 | jbd2_log_wait_commit(journal, tid); |
526 | return 1; | 535 | return 1; |
527 | } | 536 | } |
@@ -535,7 +544,7 @@ int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid) | |||
535 | { | 544 | { |
536 | int ret = 0; | 545 | int ret = 0; |
537 | 546 | ||
538 | spin_lock(&journal->j_state_lock); | 547 | write_lock(&journal->j_state_lock); |
539 | if (journal->j_running_transaction) { | 548 | if (journal->j_running_transaction) { |
540 | tid_t tid = journal->j_running_transaction->t_tid; | 549 | tid_t tid = journal->j_running_transaction->t_tid; |
541 | 550 | ||
@@ -554,7 +563,7 @@ int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid) | |||
554 | *ptid = journal->j_committing_transaction->t_tid; | 563 | *ptid = journal->j_committing_transaction->t_tid; |
555 | ret = 1; | 564 | ret = 1; |
556 | } | 565 | } |
557 | spin_unlock(&journal->j_state_lock); | 566 | write_unlock(&journal->j_state_lock); |
558 | return ret; | 567 | return ret; |
559 | } | 568 | } |
560 | 569 | ||
@@ -566,26 +575,24 @@ int jbd2_log_wait_commit(journal_t *journal, tid_t tid) | |||
566 | { | 575 | { |
567 | int err = 0; | 576 | int err = 0; |
568 | 577 | ||
578 | read_lock(&journal->j_state_lock); | ||
569 | #ifdef CONFIG_JBD2_DEBUG | 579 | #ifdef CONFIG_JBD2_DEBUG |
570 | spin_lock(&journal->j_state_lock); | ||
571 | if (!tid_geq(journal->j_commit_request, tid)) { | 580 | if (!tid_geq(journal->j_commit_request, tid)) { |
572 | printk(KERN_EMERG | 581 | printk(KERN_EMERG |
573 | "%s: error: j_commit_request=%d, tid=%d\n", | 582 | "%s: error: j_commit_request=%d, tid=%d\n", |
574 | __func__, journal->j_commit_request, tid); | 583 | __func__, journal->j_commit_request, tid); |
575 | } | 584 | } |
576 | spin_unlock(&journal->j_state_lock); | ||
577 | #endif | 585 | #endif |
578 | spin_lock(&journal->j_state_lock); | ||
579 | while (tid_gt(tid, journal->j_commit_sequence)) { | 586 | while (tid_gt(tid, journal->j_commit_sequence)) { |
580 | jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n", | 587 | jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n", |
581 | tid, journal->j_commit_sequence); | 588 | tid, journal->j_commit_sequence); |
582 | wake_up(&journal->j_wait_commit); | 589 | wake_up(&journal->j_wait_commit); |
583 | spin_unlock(&journal->j_state_lock); | 590 | read_unlock(&journal->j_state_lock); |
584 | wait_event(journal->j_wait_done_commit, | 591 | wait_event(journal->j_wait_done_commit, |
585 | !tid_gt(tid, journal->j_commit_sequence)); | 592 | !tid_gt(tid, journal->j_commit_sequence)); |
586 | spin_lock(&journal->j_state_lock); | 593 | read_lock(&journal->j_state_lock); |
587 | } | 594 | } |
588 | spin_unlock(&journal->j_state_lock); | 595 | read_unlock(&journal->j_state_lock); |
589 | 596 | ||
590 | if (unlikely(is_journal_aborted(journal))) { | 597 | if (unlikely(is_journal_aborted(journal))) { |
591 | printk(KERN_EMERG "journal commit I/O error\n"); | 598 | printk(KERN_EMERG "journal commit I/O error\n"); |
@@ -602,7 +609,7 @@ int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp) | |||
602 | { | 609 | { |
603 | unsigned long blocknr; | 610 | unsigned long blocknr; |
604 | 611 | ||
605 | spin_lock(&journal->j_state_lock); | 612 | write_lock(&journal->j_state_lock); |
606 | J_ASSERT(journal->j_free > 1); | 613 | J_ASSERT(journal->j_free > 1); |
607 | 614 | ||
608 | blocknr = journal->j_head; | 615 | blocknr = journal->j_head; |
@@ -610,7 +617,7 @@ int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp) | |||
610 | journal->j_free--; | 617 | journal->j_free--; |
611 | if (journal->j_head == journal->j_last) | 618 | if (journal->j_head == journal->j_last) |
612 | journal->j_head = journal->j_first; | 619 | journal->j_head = journal->j_first; |
613 | spin_unlock(&journal->j_state_lock); | 620 | write_unlock(&journal->j_state_lock); |
614 | return jbd2_journal_bmap(journal, blocknr, retp); | 621 | return jbd2_journal_bmap(journal, blocknr, retp); |
615 | } | 622 | } |
616 | 623 | ||
@@ -830,7 +837,7 @@ static journal_t * journal_init_common (void) | |||
830 | mutex_init(&journal->j_checkpoint_mutex); | 837 | mutex_init(&journal->j_checkpoint_mutex); |
831 | spin_lock_init(&journal->j_revoke_lock); | 838 | spin_lock_init(&journal->j_revoke_lock); |
832 | spin_lock_init(&journal->j_list_lock); | 839 | spin_lock_init(&journal->j_list_lock); |
833 | spin_lock_init(&journal->j_state_lock); | 840 | rwlock_init(&journal->j_state_lock); |
834 | 841 | ||
835 | journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE); | 842 | journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE); |
836 | journal->j_min_batch_time = 0; | 843 | journal->j_min_batch_time = 0; |
@@ -1096,14 +1103,14 @@ void jbd2_journal_update_superblock(journal_t *journal, int wait) | |||
1096 | set_buffer_uptodate(bh); | 1103 | set_buffer_uptodate(bh); |
1097 | } | 1104 | } |
1098 | 1105 | ||
1099 | spin_lock(&journal->j_state_lock); | 1106 | read_lock(&journal->j_state_lock); |
1100 | jbd_debug(1,"JBD: updating superblock (start %ld, seq %d, errno %d)\n", | 1107 | jbd_debug(1,"JBD: updating superblock (start %ld, seq %d, errno %d)\n", |
1101 | journal->j_tail, journal->j_tail_sequence, journal->j_errno); | 1108 | journal->j_tail, journal->j_tail_sequence, journal->j_errno); |
1102 | 1109 | ||
1103 | sb->s_sequence = cpu_to_be32(journal->j_tail_sequence); | 1110 | sb->s_sequence = cpu_to_be32(journal->j_tail_sequence); |
1104 | sb->s_start = cpu_to_be32(journal->j_tail); | 1111 | sb->s_start = cpu_to_be32(journal->j_tail); |
1105 | sb->s_errno = cpu_to_be32(journal->j_errno); | 1112 | sb->s_errno = cpu_to_be32(journal->j_errno); |
1106 | spin_unlock(&journal->j_state_lock); | 1113 | read_unlock(&journal->j_state_lock); |
1107 | 1114 | ||
1108 | BUFFER_TRACE(bh, "marking dirty"); | 1115 | BUFFER_TRACE(bh, "marking dirty"); |
1109 | mark_buffer_dirty(bh); | 1116 | mark_buffer_dirty(bh); |
@@ -1124,12 +1131,12 @@ out: | |||
1124 | * any future commit will have to be careful to update the | 1131 | * any future commit will have to be careful to update the |
1125 | * superblock again to re-record the true start of the log. */ | 1132 | * superblock again to re-record the true start of the log. */ |
1126 | 1133 | ||
1127 | spin_lock(&journal->j_state_lock); | 1134 | write_lock(&journal->j_state_lock); |
1128 | if (sb->s_start) | 1135 | if (sb->s_start) |
1129 | journal->j_flags &= ~JBD2_FLUSHED; | 1136 | journal->j_flags &= ~JBD2_FLUSHED; |
1130 | else | 1137 | else |
1131 | journal->j_flags |= JBD2_FLUSHED; | 1138 | journal->j_flags |= JBD2_FLUSHED; |
1132 | spin_unlock(&journal->j_state_lock); | 1139 | write_unlock(&journal->j_state_lock); |
1133 | } | 1140 | } |
1134 | 1141 | ||
1135 | /* | 1142 | /* |
@@ -1391,13 +1398,9 @@ int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat, | |||
1391 | int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat, | 1398 | int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat, |
1392 | unsigned long ro, unsigned long incompat) | 1399 | unsigned long ro, unsigned long incompat) |
1393 | { | 1400 | { |
1394 | journal_superblock_t *sb; | ||
1395 | |||
1396 | if (!compat && !ro && !incompat) | 1401 | if (!compat && !ro && !incompat) |
1397 | return 1; | 1402 | return 1; |
1398 | 1403 | ||
1399 | sb = journal->j_superblock; | ||
1400 | |||
1401 | /* We can support any known requested features iff the | 1404 | /* We can support any known requested features iff the |
1402 | * superblock is in version 2. Otherwise we fail to support any | 1405 | * superblock is in version 2. Otherwise we fail to support any |
1403 | * extended sb features. */ | 1406 | * extended sb features. */ |
@@ -1545,7 +1548,7 @@ int jbd2_journal_flush(journal_t *journal) | |||
1545 | transaction_t *transaction = NULL; | 1548 | transaction_t *transaction = NULL; |
1546 | unsigned long old_tail; | 1549 | unsigned long old_tail; |
1547 | 1550 | ||
1548 | spin_lock(&journal->j_state_lock); | 1551 | write_lock(&journal->j_state_lock); |
1549 | 1552 | ||
1550 | /* Force everything buffered to the log... */ | 1553 | /* Force everything buffered to the log... */ |
1551 | if (journal->j_running_transaction) { | 1554 | if (journal->j_running_transaction) { |
@@ -1558,10 +1561,10 @@ int jbd2_journal_flush(journal_t *journal) | |||
1558 | if (transaction) { | 1561 | if (transaction) { |
1559 | tid_t tid = transaction->t_tid; | 1562 | tid_t tid = transaction->t_tid; |
1560 | 1563 | ||
1561 | spin_unlock(&journal->j_state_lock); | 1564 | write_unlock(&journal->j_state_lock); |
1562 | jbd2_log_wait_commit(journal, tid); | 1565 | jbd2_log_wait_commit(journal, tid); |
1563 | } else { | 1566 | } else { |
1564 | spin_unlock(&journal->j_state_lock); | 1567 | write_unlock(&journal->j_state_lock); |
1565 | } | 1568 | } |
1566 | 1569 | ||
1567 | /* ...and flush everything in the log out to disk. */ | 1570 | /* ...and flush everything in the log out to disk. */ |
@@ -1585,12 +1588,12 @@ int jbd2_journal_flush(journal_t *journal) | |||
1585 | * the magic code for a fully-recovered superblock. Any future | 1588 | * the magic code for a fully-recovered superblock. Any future |
1586 | * commits of data to the journal will restore the current | 1589 | * commits of data to the journal will restore the current |
1587 | * s_start value. */ | 1590 | * s_start value. */ |
1588 | spin_lock(&journal->j_state_lock); | 1591 | write_lock(&journal->j_state_lock); |
1589 | old_tail = journal->j_tail; | 1592 | old_tail = journal->j_tail; |
1590 | journal->j_tail = 0; | 1593 | journal->j_tail = 0; |
1591 | spin_unlock(&journal->j_state_lock); | 1594 | write_unlock(&journal->j_state_lock); |
1592 | jbd2_journal_update_superblock(journal, 1); | 1595 | jbd2_journal_update_superblock(journal, 1); |
1593 | spin_lock(&journal->j_state_lock); | 1596 | write_lock(&journal->j_state_lock); |
1594 | journal->j_tail = old_tail; | 1597 | journal->j_tail = old_tail; |
1595 | 1598 | ||
1596 | J_ASSERT(!journal->j_running_transaction); | 1599 | J_ASSERT(!journal->j_running_transaction); |
@@ -1598,7 +1601,7 @@ int jbd2_journal_flush(journal_t *journal) | |||
1598 | J_ASSERT(!journal->j_checkpoint_transactions); | 1601 | J_ASSERT(!journal->j_checkpoint_transactions); |
1599 | J_ASSERT(journal->j_head == journal->j_tail); | 1602 | J_ASSERT(journal->j_head == journal->j_tail); |
1600 | J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence); | 1603 | J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence); |
1601 | spin_unlock(&journal->j_state_lock); | 1604 | write_unlock(&journal->j_state_lock); |
1602 | return 0; | 1605 | return 0; |
1603 | } | 1606 | } |
1604 | 1607 | ||
@@ -1617,7 +1620,6 @@ int jbd2_journal_flush(journal_t *journal) | |||
1617 | 1620 | ||
1618 | int jbd2_journal_wipe(journal_t *journal, int write) | 1621 | int jbd2_journal_wipe(journal_t *journal, int write) |
1619 | { | 1622 | { |
1620 | journal_superblock_t *sb; | ||
1621 | int err = 0; | 1623 | int err = 0; |
1622 | 1624 | ||
1623 | J_ASSERT (!(journal->j_flags & JBD2_LOADED)); | 1625 | J_ASSERT (!(journal->j_flags & JBD2_LOADED)); |
@@ -1626,8 +1628,6 @@ int jbd2_journal_wipe(journal_t *journal, int write) | |||
1626 | if (err) | 1628 | if (err) |
1627 | return err; | 1629 | return err; |
1628 | 1630 | ||
1629 | sb = journal->j_superblock; | ||
1630 | |||
1631 | if (!journal->j_tail) | 1631 | if (!journal->j_tail) |
1632 | goto no_recovery; | 1632 | goto no_recovery; |
1633 | 1633 | ||
@@ -1665,12 +1665,12 @@ void __jbd2_journal_abort_hard(journal_t *journal) | |||
1665 | printk(KERN_ERR "Aborting journal on device %s.\n", | 1665 | printk(KERN_ERR "Aborting journal on device %s.\n", |
1666 | journal->j_devname); | 1666 | journal->j_devname); |
1667 | 1667 | ||
1668 | spin_lock(&journal->j_state_lock); | 1668 | write_lock(&journal->j_state_lock); |
1669 | journal->j_flags |= JBD2_ABORT; | 1669 | journal->j_flags |= JBD2_ABORT; |
1670 | transaction = journal->j_running_transaction; | 1670 | transaction = journal->j_running_transaction; |
1671 | if (transaction) | 1671 | if (transaction) |
1672 | __jbd2_log_start_commit(journal, transaction->t_tid); | 1672 | __jbd2_log_start_commit(journal, transaction->t_tid); |
1673 | spin_unlock(&journal->j_state_lock); | 1673 | write_unlock(&journal->j_state_lock); |
1674 | } | 1674 | } |
1675 | 1675 | ||
1676 | /* Soft abort: record the abort error status in the journal superblock, | 1676 | /* Soft abort: record the abort error status in the journal superblock, |
@@ -1755,12 +1755,12 @@ int jbd2_journal_errno(journal_t *journal) | |||
1755 | { | 1755 | { |
1756 | int err; | 1756 | int err; |
1757 | 1757 | ||
1758 | spin_lock(&journal->j_state_lock); | 1758 | read_lock(&journal->j_state_lock); |
1759 | if (journal->j_flags & JBD2_ABORT) | 1759 | if (journal->j_flags & JBD2_ABORT) |
1760 | err = -EROFS; | 1760 | err = -EROFS; |
1761 | else | 1761 | else |
1762 | err = journal->j_errno; | 1762 | err = journal->j_errno; |
1763 | spin_unlock(&journal->j_state_lock); | 1763 | read_unlock(&journal->j_state_lock); |
1764 | return err; | 1764 | return err; |
1765 | } | 1765 | } |
1766 | 1766 | ||
@@ -1775,12 +1775,12 @@ int jbd2_journal_clear_err(journal_t *journal) | |||
1775 | { | 1775 | { |
1776 | int err = 0; | 1776 | int err = 0; |
1777 | 1777 | ||
1778 | spin_lock(&journal->j_state_lock); | 1778 | write_lock(&journal->j_state_lock); |
1779 | if (journal->j_flags & JBD2_ABORT) | 1779 | if (journal->j_flags & JBD2_ABORT) |
1780 | err = -EROFS; | 1780 | err = -EROFS; |
1781 | else | 1781 | else |
1782 | journal->j_errno = 0; | 1782 | journal->j_errno = 0; |
1783 | spin_unlock(&journal->j_state_lock); | 1783 | write_unlock(&journal->j_state_lock); |
1784 | return err; | 1784 | return err; |
1785 | } | 1785 | } |
1786 | 1786 | ||
@@ -1793,10 +1793,10 @@ int jbd2_journal_clear_err(journal_t *journal) | |||
1793 | */ | 1793 | */ |
1794 | void jbd2_journal_ack_err(journal_t *journal) | 1794 | void jbd2_journal_ack_err(journal_t *journal) |
1795 | { | 1795 | { |
1796 | spin_lock(&journal->j_state_lock); | 1796 | write_lock(&journal->j_state_lock); |
1797 | if (journal->j_errno) | 1797 | if (journal->j_errno) |
1798 | journal->j_flags |= JBD2_ACK_ERR; | 1798 | journal->j_flags |= JBD2_ACK_ERR; |
1799 | spin_unlock(&journal->j_state_lock); | 1799 | write_unlock(&journal->j_state_lock); |
1800 | } | 1800 | } |
1801 | 1801 | ||
1802 | int jbd2_journal_blocks_per_page(struct inode *inode) | 1802 | int jbd2_journal_blocks_per_page(struct inode *inode) |
@@ -2201,8 +2201,6 @@ void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode) | |||
2201 | void jbd2_journal_release_jbd_inode(journal_t *journal, | 2201 | void jbd2_journal_release_jbd_inode(journal_t *journal, |
2202 | struct jbd2_inode *jinode) | 2202 | struct jbd2_inode *jinode) |
2203 | { | 2203 | { |
2204 | int writeout = 0; | ||
2205 | |||
2206 | if (!journal) | 2204 | if (!journal) |
2207 | return; | 2205 | return; |
2208 | restart: | 2206 | restart: |
@@ -2219,9 +2217,6 @@ restart: | |||
2219 | goto restart; | 2217 | goto restart; |
2220 | } | 2218 | } |
2221 | 2219 | ||
2222 | /* Do we need to wait for data writeback? */ | ||
2223 | if (journal->j_committing_transaction == jinode->i_transaction) | ||
2224 | writeout = 1; | ||
2225 | if (jinode->i_transaction) { | 2220 | if (jinode->i_transaction) { |
2226 | list_del(&jinode->i_list); | 2221 | list_del(&jinode->i_list); |
2227 | jinode->i_transaction = NULL; | 2222 | jinode->i_transaction = NULL; |
diff --git a/fs/jbd2/recovery.c b/fs/jbd2/recovery.c index 049281b7cb89..2bc4d5f116f1 100644 --- a/fs/jbd2/recovery.c +++ b/fs/jbd2/recovery.c | |||
@@ -285,12 +285,10 @@ int jbd2_journal_recover(journal_t *journal) | |||
285 | int jbd2_journal_skip_recovery(journal_t *journal) | 285 | int jbd2_journal_skip_recovery(journal_t *journal) |
286 | { | 286 | { |
287 | int err; | 287 | int err; |
288 | journal_superblock_t * sb; | ||
289 | 288 | ||
290 | struct recovery_info info; | 289 | struct recovery_info info; |
291 | 290 | ||
292 | memset (&info, 0, sizeof(info)); | 291 | memset (&info, 0, sizeof(info)); |
293 | sb = journal->j_superblock; | ||
294 | 292 | ||
295 | err = do_one_pass(journal, &info, PASS_SCAN); | 293 | err = do_one_pass(journal, &info, PASS_SCAN); |
296 | 294 | ||
@@ -299,7 +297,8 @@ int jbd2_journal_skip_recovery(journal_t *journal) | |||
299 | ++journal->j_transaction_sequence; | 297 | ++journal->j_transaction_sequence; |
300 | } else { | 298 | } else { |
301 | #ifdef CONFIG_JBD2_DEBUG | 299 | #ifdef CONFIG_JBD2_DEBUG |
302 | int dropped = info.end_transaction - be32_to_cpu(sb->s_sequence); | 300 | int dropped = info.end_transaction - |
301 | be32_to_cpu(journal->j_superblock->s_sequence); | ||
303 | #endif | 302 | #endif |
304 | jbd_debug(1, | 303 | jbd_debug(1, |
305 | "JBD: ignoring %d transaction%s from the journal.\n", | 304 | "JBD: ignoring %d transaction%s from the journal.\n", |
@@ -365,11 +364,6 @@ static int do_one_pass(journal_t *journal, | |||
365 | int tag_bytes = journal_tag_bytes(journal); | 364 | int tag_bytes = journal_tag_bytes(journal); |
366 | __u32 crc32_sum = ~0; /* Transactional Checksums */ | 365 | __u32 crc32_sum = ~0; /* Transactional Checksums */ |
367 | 366 | ||
368 | /* Precompute the maximum metadata descriptors in a descriptor block */ | ||
369 | int MAX_BLOCKS_PER_DESC; | ||
370 | MAX_BLOCKS_PER_DESC = ((journal->j_blocksize-sizeof(journal_header_t)) | ||
371 | / tag_bytes); | ||
372 | |||
373 | /* | 367 | /* |
374 | * First thing is to establish what we expect to find in the log | 368 | * First thing is to establish what we expect to find in the log |
375 | * (in terms of transaction IDs), and where (in terms of log | 369 | * (in terms of transaction IDs), and where (in terms of log |
diff --git a/fs/jbd2/transaction.c b/fs/jbd2/transaction.c index b8e0806681bb..d95cc9d0401d 100644 --- a/fs/jbd2/transaction.c +++ b/fs/jbd2/transaction.c | |||
@@ -26,6 +26,8 @@ | |||
26 | #include <linux/mm.h> | 26 | #include <linux/mm.h> |
27 | #include <linux/highmem.h> | 27 | #include <linux/highmem.h> |
28 | #include <linux/hrtimer.h> | 28 | #include <linux/hrtimer.h> |
29 | #include <linux/backing-dev.h> | ||
30 | #include <linux/module.h> | ||
29 | 31 | ||
30 | static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh); | 32 | static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh); |
31 | 33 | ||
@@ -53,6 +55,9 @@ jbd2_get_transaction(journal_t *journal, transaction_t *transaction) | |||
53 | transaction->t_tid = journal->j_transaction_sequence++; | 55 | transaction->t_tid = journal->j_transaction_sequence++; |
54 | transaction->t_expires = jiffies + journal->j_commit_interval; | 56 | transaction->t_expires = jiffies + journal->j_commit_interval; |
55 | spin_lock_init(&transaction->t_handle_lock); | 57 | spin_lock_init(&transaction->t_handle_lock); |
58 | atomic_set(&transaction->t_updates, 0); | ||
59 | atomic_set(&transaction->t_outstanding_credits, 0); | ||
60 | atomic_set(&transaction->t_handle_count, 0); | ||
56 | INIT_LIST_HEAD(&transaction->t_inode_list); | 61 | INIT_LIST_HEAD(&transaction->t_inode_list); |
57 | INIT_LIST_HEAD(&transaction->t_private_list); | 62 | INIT_LIST_HEAD(&transaction->t_private_list); |
58 | 63 | ||
@@ -83,65 +88,75 @@ jbd2_get_transaction(journal_t *journal, transaction_t *transaction) | |||
83 | * transaction's buffer credits. | 88 | * transaction's buffer credits. |
84 | */ | 89 | */ |
85 | 90 | ||
86 | static int start_this_handle(journal_t *journal, handle_t *handle) | 91 | static int start_this_handle(journal_t *journal, handle_t *handle, |
92 | int gfp_mask) | ||
87 | { | 93 | { |
88 | transaction_t *transaction; | 94 | transaction_t *transaction; |
89 | int needed; | 95 | int needed; |
90 | int nblocks = handle->h_buffer_credits; | 96 | int nblocks = handle->h_buffer_credits; |
91 | transaction_t *new_transaction = NULL; | 97 | transaction_t *new_transaction = NULL; |
92 | int ret = 0; | ||
93 | unsigned long ts = jiffies; | 98 | unsigned long ts = jiffies; |
94 | 99 | ||
95 | if (nblocks > journal->j_max_transaction_buffers) { | 100 | if (nblocks > journal->j_max_transaction_buffers) { |
96 | printk(KERN_ERR "JBD: %s wants too many credits (%d > %d)\n", | 101 | printk(KERN_ERR "JBD: %s wants too many credits (%d > %d)\n", |
97 | current->comm, nblocks, | 102 | current->comm, nblocks, |
98 | journal->j_max_transaction_buffers); | 103 | journal->j_max_transaction_buffers); |
99 | ret = -ENOSPC; | 104 | return -ENOSPC; |
100 | goto out; | ||
101 | } | 105 | } |
102 | 106 | ||
103 | alloc_transaction: | 107 | alloc_transaction: |
104 | if (!journal->j_running_transaction) { | 108 | if (!journal->j_running_transaction) { |
105 | new_transaction = kzalloc(sizeof(*new_transaction), | 109 | new_transaction = kzalloc(sizeof(*new_transaction), gfp_mask); |
106 | GFP_NOFS|__GFP_NOFAIL); | ||
107 | if (!new_transaction) { | 110 | if (!new_transaction) { |
108 | ret = -ENOMEM; | 111 | /* |
109 | goto out; | 112 | * If __GFP_FS is not present, then we may be |
113 | * being called from inside the fs writeback | ||
114 | * layer, so we MUST NOT fail. Since | ||
115 | * __GFP_NOFAIL is going away, we will arrange | ||
116 | * to retry the allocation ourselves. | ||
117 | */ | ||
118 | if ((gfp_mask & __GFP_FS) == 0) { | ||
119 | congestion_wait(BLK_RW_ASYNC, HZ/50); | ||
120 | goto alloc_transaction; | ||
121 | } | ||
122 | return -ENOMEM; | ||
110 | } | 123 | } |
111 | } | 124 | } |
112 | 125 | ||
113 | jbd_debug(3, "New handle %p going live.\n", handle); | 126 | jbd_debug(3, "New handle %p going live.\n", handle); |
114 | 127 | ||
115 | repeat: | ||
116 | |||
117 | /* | 128 | /* |
118 | * We need to hold j_state_lock until t_updates has been incremented, | 129 | * We need to hold j_state_lock until t_updates has been incremented, |
119 | * for proper journal barrier handling | 130 | * for proper journal barrier handling |
120 | */ | 131 | */ |
121 | spin_lock(&journal->j_state_lock); | 132 | repeat: |
122 | repeat_locked: | 133 | read_lock(&journal->j_state_lock); |
123 | if (is_journal_aborted(journal) || | 134 | if (is_journal_aborted(journal) || |
124 | (journal->j_errno != 0 && !(journal->j_flags & JBD2_ACK_ERR))) { | 135 | (journal->j_errno != 0 && !(journal->j_flags & JBD2_ACK_ERR))) { |
125 | spin_unlock(&journal->j_state_lock); | 136 | read_unlock(&journal->j_state_lock); |
126 | ret = -EROFS; | 137 | kfree(new_transaction); |
127 | goto out; | 138 | return -EROFS; |
128 | } | 139 | } |
129 | 140 | ||
130 | /* Wait on the journal's transaction barrier if necessary */ | 141 | /* Wait on the journal's transaction barrier if necessary */ |
131 | if (journal->j_barrier_count) { | 142 | if (journal->j_barrier_count) { |
132 | spin_unlock(&journal->j_state_lock); | 143 | read_unlock(&journal->j_state_lock); |
133 | wait_event(journal->j_wait_transaction_locked, | 144 | wait_event(journal->j_wait_transaction_locked, |
134 | journal->j_barrier_count == 0); | 145 | journal->j_barrier_count == 0); |
135 | goto repeat; | 146 | goto repeat; |
136 | } | 147 | } |
137 | 148 | ||
138 | if (!journal->j_running_transaction) { | 149 | if (!journal->j_running_transaction) { |
139 | if (!new_transaction) { | 150 | read_unlock(&journal->j_state_lock); |
140 | spin_unlock(&journal->j_state_lock); | 151 | if (!new_transaction) |
141 | goto alloc_transaction; | 152 | goto alloc_transaction; |
153 | write_lock(&journal->j_state_lock); | ||
154 | if (!journal->j_running_transaction) { | ||
155 | jbd2_get_transaction(journal, new_transaction); | ||
156 | new_transaction = NULL; | ||
142 | } | 157 | } |
143 | jbd2_get_transaction(journal, new_transaction); | 158 | write_unlock(&journal->j_state_lock); |
144 | new_transaction = NULL; | 159 | goto repeat; |
145 | } | 160 | } |
146 | 161 | ||
147 | transaction = journal->j_running_transaction; | 162 | transaction = journal->j_running_transaction; |
@@ -155,7 +170,7 @@ repeat_locked: | |||
155 | 170 | ||
156 | prepare_to_wait(&journal->j_wait_transaction_locked, | 171 | prepare_to_wait(&journal->j_wait_transaction_locked, |
157 | &wait, TASK_UNINTERRUPTIBLE); | 172 | &wait, TASK_UNINTERRUPTIBLE); |
158 | spin_unlock(&journal->j_state_lock); | 173 | read_unlock(&journal->j_state_lock); |
159 | schedule(); | 174 | schedule(); |
160 | finish_wait(&journal->j_wait_transaction_locked, &wait); | 175 | finish_wait(&journal->j_wait_transaction_locked, &wait); |
161 | goto repeat; | 176 | goto repeat; |
@@ -166,8 +181,8 @@ repeat_locked: | |||
166 | * buffers requested by this operation, we need to stall pending a log | 181 | * buffers requested by this operation, we need to stall pending a log |
167 | * checkpoint to free some more log space. | 182 | * checkpoint to free some more log space. |
168 | */ | 183 | */ |
169 | spin_lock(&transaction->t_handle_lock); | 184 | needed = atomic_add_return(nblocks, |
170 | needed = transaction->t_outstanding_credits + nblocks; | 185 | &transaction->t_outstanding_credits); |
171 | 186 | ||
172 | if (needed > journal->j_max_transaction_buffers) { | 187 | if (needed > journal->j_max_transaction_buffers) { |
173 | /* | 188 | /* |
@@ -178,11 +193,11 @@ repeat_locked: | |||
178 | DEFINE_WAIT(wait); | 193 | DEFINE_WAIT(wait); |
179 | 194 | ||
180 | jbd_debug(2, "Handle %p starting new commit...\n", handle); | 195 | jbd_debug(2, "Handle %p starting new commit...\n", handle); |
181 | spin_unlock(&transaction->t_handle_lock); | 196 | atomic_sub(nblocks, &transaction->t_outstanding_credits); |
182 | prepare_to_wait(&journal->j_wait_transaction_locked, &wait, | 197 | prepare_to_wait(&journal->j_wait_transaction_locked, &wait, |
183 | TASK_UNINTERRUPTIBLE); | 198 | TASK_UNINTERRUPTIBLE); |
184 | __jbd2_log_start_commit(journal, transaction->t_tid); | 199 | __jbd2_log_start_commit(journal, transaction->t_tid); |
185 | spin_unlock(&journal->j_state_lock); | 200 | read_unlock(&journal->j_state_lock); |
186 | schedule(); | 201 | schedule(); |
187 | finish_wait(&journal->j_wait_transaction_locked, &wait); | 202 | finish_wait(&journal->j_wait_transaction_locked, &wait); |
188 | goto repeat; | 203 | goto repeat; |
@@ -215,35 +230,48 @@ repeat_locked: | |||
215 | */ | 230 | */ |
216 | if (__jbd2_log_space_left(journal) < jbd_space_needed(journal)) { | 231 | if (__jbd2_log_space_left(journal) < jbd_space_needed(journal)) { |
217 | jbd_debug(2, "Handle %p waiting for checkpoint...\n", handle); | 232 | jbd_debug(2, "Handle %p waiting for checkpoint...\n", handle); |
218 | spin_unlock(&transaction->t_handle_lock); | 233 | atomic_sub(nblocks, &transaction->t_outstanding_credits); |
219 | __jbd2_log_wait_for_space(journal); | 234 | read_unlock(&journal->j_state_lock); |
220 | goto repeat_locked; | 235 | write_lock(&journal->j_state_lock); |
236 | if (__jbd2_log_space_left(journal) < jbd_space_needed(journal)) | ||
237 | __jbd2_log_wait_for_space(journal); | ||
238 | write_unlock(&journal->j_state_lock); | ||
239 | goto repeat; | ||
221 | } | 240 | } |
222 | 241 | ||
223 | /* OK, account for the buffers that this operation expects to | 242 | /* OK, account for the buffers that this operation expects to |
224 | * use and add the handle to the running transaction. */ | 243 | * use and add the handle to the running transaction. |
225 | 244 | * | |
226 | if (time_after(transaction->t_start, ts)) { | 245 | * In order for t_max_wait to be reliable, it must be |
246 | * protected by a lock. But doing so will mean that | ||
247 | * start_this_handle() can not be run in parallel on SMP | ||
248 | * systems, which limits our scalability. So we only enable | ||
249 | * it when debugging is enabled. We may want to use a | ||
250 | * separate flag, eventually, so we can enable this | ||
251 | * independently of debugging. | ||
252 | */ | ||
253 | #ifdef CONFIG_JBD2_DEBUG | ||
254 | if (jbd2_journal_enable_debug && | ||
255 | time_after(transaction->t_start, ts)) { | ||
227 | ts = jbd2_time_diff(ts, transaction->t_start); | 256 | ts = jbd2_time_diff(ts, transaction->t_start); |
257 | spin_lock(&transaction->t_handle_lock); | ||
228 | if (ts > transaction->t_max_wait) | 258 | if (ts > transaction->t_max_wait) |
229 | transaction->t_max_wait = ts; | 259 | transaction->t_max_wait = ts; |
260 | spin_unlock(&transaction->t_handle_lock); | ||
230 | } | 261 | } |
231 | 262 | #endif | |
232 | handle->h_transaction = transaction; | 263 | handle->h_transaction = transaction; |
233 | transaction->t_outstanding_credits += nblocks; | 264 | atomic_inc(&transaction->t_updates); |
234 | transaction->t_updates++; | 265 | atomic_inc(&transaction->t_handle_count); |
235 | transaction->t_handle_count++; | ||
236 | jbd_debug(4, "Handle %p given %d credits (total %d, free %d)\n", | 266 | jbd_debug(4, "Handle %p given %d credits (total %d, free %d)\n", |
237 | handle, nblocks, transaction->t_outstanding_credits, | 267 | handle, nblocks, |
268 | atomic_read(&transaction->t_outstanding_credits), | ||
238 | __jbd2_log_space_left(journal)); | 269 | __jbd2_log_space_left(journal)); |
239 | spin_unlock(&transaction->t_handle_lock); | 270 | read_unlock(&journal->j_state_lock); |
240 | spin_unlock(&journal->j_state_lock); | ||
241 | 271 | ||
242 | lock_map_acquire(&handle->h_lockdep_map); | 272 | lock_map_acquire(&handle->h_lockdep_map); |
243 | out: | 273 | kfree(new_transaction); |
244 | if (unlikely(new_transaction)) /* It's usually NULL */ | 274 | return 0; |
245 | kfree(new_transaction); | ||
246 | return ret; | ||
247 | } | 275 | } |
248 | 276 | ||
249 | static struct lock_class_key jbd2_handle_key; | 277 | static struct lock_class_key jbd2_handle_key; |
@@ -278,7 +306,7 @@ static handle_t *new_handle(int nblocks) | |||
278 | * | 306 | * |
279 | * Return a pointer to a newly allocated handle, or NULL on failure | 307 | * Return a pointer to a newly allocated handle, or NULL on failure |
280 | */ | 308 | */ |
281 | handle_t *jbd2_journal_start(journal_t *journal, int nblocks) | 309 | handle_t *jbd2__journal_start(journal_t *journal, int nblocks, int gfp_mask) |
282 | { | 310 | { |
283 | handle_t *handle = journal_current_handle(); | 311 | handle_t *handle = journal_current_handle(); |
284 | int err; | 312 | int err; |
@@ -298,7 +326,7 @@ handle_t *jbd2_journal_start(journal_t *journal, int nblocks) | |||
298 | 326 | ||
299 | current->journal_info = handle; | 327 | current->journal_info = handle; |
300 | 328 | ||
301 | err = start_this_handle(journal, handle); | 329 | err = start_this_handle(journal, handle, gfp_mask); |
302 | if (err < 0) { | 330 | if (err < 0) { |
303 | jbd2_free_handle(handle); | 331 | jbd2_free_handle(handle); |
304 | current->journal_info = NULL; | 332 | current->journal_info = NULL; |
@@ -308,6 +336,15 @@ handle_t *jbd2_journal_start(journal_t *journal, int nblocks) | |||
308 | out: | 336 | out: |
309 | return handle; | 337 | return handle; |
310 | } | 338 | } |
339 | EXPORT_SYMBOL(jbd2__journal_start); | ||
340 | |||
341 | |||
342 | handle_t *jbd2_journal_start(journal_t *journal, int nblocks) | ||
343 | { | ||
344 | return jbd2__journal_start(journal, nblocks, GFP_NOFS); | ||
345 | } | ||
346 | EXPORT_SYMBOL(jbd2_journal_start); | ||
347 | |||
311 | 348 | ||
312 | /** | 349 | /** |
313 | * int jbd2_journal_extend() - extend buffer credits. | 350 | * int jbd2_journal_extend() - extend buffer credits. |
@@ -342,7 +379,7 @@ int jbd2_journal_extend(handle_t *handle, int nblocks) | |||
342 | 379 | ||
343 | result = 1; | 380 | result = 1; |
344 | 381 | ||
345 | spin_lock(&journal->j_state_lock); | 382 | read_lock(&journal->j_state_lock); |
346 | 383 | ||
347 | /* Don't extend a locked-down transaction! */ | 384 | /* Don't extend a locked-down transaction! */ |
348 | if (handle->h_transaction->t_state != T_RUNNING) { | 385 | if (handle->h_transaction->t_state != T_RUNNING) { |
@@ -352,7 +389,7 @@ int jbd2_journal_extend(handle_t *handle, int nblocks) | |||
352 | } | 389 | } |
353 | 390 | ||
354 | spin_lock(&transaction->t_handle_lock); | 391 | spin_lock(&transaction->t_handle_lock); |
355 | wanted = transaction->t_outstanding_credits + nblocks; | 392 | wanted = atomic_read(&transaction->t_outstanding_credits) + nblocks; |
356 | 393 | ||
357 | if (wanted > journal->j_max_transaction_buffers) { | 394 | if (wanted > journal->j_max_transaction_buffers) { |
358 | jbd_debug(3, "denied handle %p %d blocks: " | 395 | jbd_debug(3, "denied handle %p %d blocks: " |
@@ -367,14 +404,14 @@ int jbd2_journal_extend(handle_t *handle, int nblocks) | |||
367 | } | 404 | } |
368 | 405 | ||
369 | handle->h_buffer_credits += nblocks; | 406 | handle->h_buffer_credits += nblocks; |
370 | transaction->t_outstanding_credits += nblocks; | 407 | atomic_add(nblocks, &transaction->t_outstanding_credits); |
371 | result = 0; | 408 | result = 0; |
372 | 409 | ||
373 | jbd_debug(3, "extended handle %p by %d\n", handle, nblocks); | 410 | jbd_debug(3, "extended handle %p by %d\n", handle, nblocks); |
374 | unlock: | 411 | unlock: |
375 | spin_unlock(&transaction->t_handle_lock); | 412 | spin_unlock(&transaction->t_handle_lock); |
376 | error_out: | 413 | error_out: |
377 | spin_unlock(&journal->j_state_lock); | 414 | read_unlock(&journal->j_state_lock); |
378 | out: | 415 | out: |
379 | return result; | 416 | return result; |
380 | } | 417 | } |
@@ -394,8 +431,7 @@ out: | |||
394 | * transaction capabable of guaranteeing the requested number of | 431 | * transaction capabable of guaranteeing the requested number of |
395 | * credits. | 432 | * credits. |
396 | */ | 433 | */ |
397 | 434 | int jbd2__journal_restart(handle_t *handle, int nblocks, int gfp_mask) | |
398 | int jbd2_journal_restart(handle_t *handle, int nblocks) | ||
399 | { | 435 | { |
400 | transaction_t *transaction = handle->h_transaction; | 436 | transaction_t *transaction = handle->h_transaction; |
401 | journal_t *journal = transaction->t_journal; | 437 | journal_t *journal = transaction->t_journal; |
@@ -410,29 +446,35 @@ int jbd2_journal_restart(handle_t *handle, int nblocks) | |||
410 | * First unlink the handle from its current transaction, and start the | 446 | * First unlink the handle from its current transaction, and start the |
411 | * commit on that. | 447 | * commit on that. |
412 | */ | 448 | */ |
413 | J_ASSERT(transaction->t_updates > 0); | 449 | J_ASSERT(atomic_read(&transaction->t_updates) > 0); |
414 | J_ASSERT(journal_current_handle() == handle); | 450 | J_ASSERT(journal_current_handle() == handle); |
415 | 451 | ||
416 | spin_lock(&journal->j_state_lock); | 452 | read_lock(&journal->j_state_lock); |
417 | spin_lock(&transaction->t_handle_lock); | 453 | spin_lock(&transaction->t_handle_lock); |
418 | transaction->t_outstanding_credits -= handle->h_buffer_credits; | 454 | atomic_sub(handle->h_buffer_credits, |
419 | transaction->t_updates--; | 455 | &transaction->t_outstanding_credits); |
420 | 456 | if (atomic_dec_and_test(&transaction->t_updates)) | |
421 | if (!transaction->t_updates) | ||
422 | wake_up(&journal->j_wait_updates); | 457 | wake_up(&journal->j_wait_updates); |
423 | spin_unlock(&transaction->t_handle_lock); | 458 | spin_unlock(&transaction->t_handle_lock); |
424 | 459 | ||
425 | jbd_debug(2, "restarting handle %p\n", handle); | 460 | jbd_debug(2, "restarting handle %p\n", handle); |
426 | __jbd2_log_start_commit(journal, transaction->t_tid); | 461 | __jbd2_log_start_commit(journal, transaction->t_tid); |
427 | spin_unlock(&journal->j_state_lock); | 462 | read_unlock(&journal->j_state_lock); |
428 | 463 | ||
429 | lock_map_release(&handle->h_lockdep_map); | 464 | lock_map_release(&handle->h_lockdep_map); |
430 | handle->h_buffer_credits = nblocks; | 465 | handle->h_buffer_credits = nblocks; |
431 | ret = start_this_handle(journal, handle); | 466 | ret = start_this_handle(journal, handle, gfp_mask); |
432 | return ret; | 467 | return ret; |
433 | } | 468 | } |
469 | EXPORT_SYMBOL(jbd2__journal_restart); | ||
434 | 470 | ||
435 | 471 | ||
472 | int jbd2_journal_restart(handle_t *handle, int nblocks) | ||
473 | { | ||
474 | return jbd2__journal_restart(handle, nblocks, GFP_NOFS); | ||
475 | } | ||
476 | EXPORT_SYMBOL(jbd2_journal_restart); | ||
477 | |||
436 | /** | 478 | /** |
437 | * void jbd2_journal_lock_updates () - establish a transaction barrier. | 479 | * void jbd2_journal_lock_updates () - establish a transaction barrier. |
438 | * @journal: Journal to establish a barrier on. | 480 | * @journal: Journal to establish a barrier on. |
@@ -447,7 +489,7 @@ void jbd2_journal_lock_updates(journal_t *journal) | |||
447 | { | 489 | { |
448 | DEFINE_WAIT(wait); | 490 | DEFINE_WAIT(wait); |
449 | 491 | ||
450 | spin_lock(&journal->j_state_lock); | 492 | write_lock(&journal->j_state_lock); |
451 | ++journal->j_barrier_count; | 493 | ++journal->j_barrier_count; |
452 | 494 | ||
453 | /* Wait until there are no running updates */ | 495 | /* Wait until there are no running updates */ |
@@ -458,19 +500,19 @@ void jbd2_journal_lock_updates(journal_t *journal) | |||
458 | break; | 500 | break; |
459 | 501 | ||
460 | spin_lock(&transaction->t_handle_lock); | 502 | spin_lock(&transaction->t_handle_lock); |
461 | if (!transaction->t_updates) { | 503 | if (!atomic_read(&transaction->t_updates)) { |
462 | spin_unlock(&transaction->t_handle_lock); | 504 | spin_unlock(&transaction->t_handle_lock); |
463 | break; | 505 | break; |
464 | } | 506 | } |
465 | prepare_to_wait(&journal->j_wait_updates, &wait, | 507 | prepare_to_wait(&journal->j_wait_updates, &wait, |
466 | TASK_UNINTERRUPTIBLE); | 508 | TASK_UNINTERRUPTIBLE); |
467 | spin_unlock(&transaction->t_handle_lock); | 509 | spin_unlock(&transaction->t_handle_lock); |
468 | spin_unlock(&journal->j_state_lock); | 510 | write_unlock(&journal->j_state_lock); |
469 | schedule(); | 511 | schedule(); |
470 | finish_wait(&journal->j_wait_updates, &wait); | 512 | finish_wait(&journal->j_wait_updates, &wait); |
471 | spin_lock(&journal->j_state_lock); | 513 | write_lock(&journal->j_state_lock); |
472 | } | 514 | } |
473 | spin_unlock(&journal->j_state_lock); | 515 | write_unlock(&journal->j_state_lock); |
474 | 516 | ||
475 | /* | 517 | /* |
476 | * We have now established a barrier against other normal updates, but | 518 | * We have now established a barrier against other normal updates, but |
@@ -494,9 +536,9 @@ void jbd2_journal_unlock_updates (journal_t *journal) | |||
494 | J_ASSERT(journal->j_barrier_count != 0); | 536 | J_ASSERT(journal->j_barrier_count != 0); |
495 | 537 | ||
496 | mutex_unlock(&journal->j_barrier); | 538 | mutex_unlock(&journal->j_barrier); |
497 | spin_lock(&journal->j_state_lock); | 539 | write_lock(&journal->j_state_lock); |
498 | --journal->j_barrier_count; | 540 | --journal->j_barrier_count; |
499 | spin_unlock(&journal->j_state_lock); | 541 | write_unlock(&journal->j_state_lock); |
500 | wake_up(&journal->j_wait_transaction_locked); | 542 | wake_up(&journal->j_wait_transaction_locked); |
501 | } | 543 | } |
502 | 544 | ||
@@ -1238,7 +1280,8 @@ int jbd2_journal_stop(handle_t *handle) | |||
1238 | { | 1280 | { |
1239 | transaction_t *transaction = handle->h_transaction; | 1281 | transaction_t *transaction = handle->h_transaction; |
1240 | journal_t *journal = transaction->t_journal; | 1282 | journal_t *journal = transaction->t_journal; |
1241 | int err; | 1283 | int err, wait_for_commit = 0; |
1284 | tid_t tid; | ||
1242 | pid_t pid; | 1285 | pid_t pid; |
1243 | 1286 | ||
1244 | J_ASSERT(journal_current_handle() == handle); | 1287 | J_ASSERT(journal_current_handle() == handle); |
@@ -1246,7 +1289,7 @@ int jbd2_journal_stop(handle_t *handle) | |||
1246 | if (is_handle_aborted(handle)) | 1289 | if (is_handle_aborted(handle)) |
1247 | err = -EIO; | 1290 | err = -EIO; |
1248 | else { | 1291 | else { |
1249 | J_ASSERT(transaction->t_updates > 0); | 1292 | J_ASSERT(atomic_read(&transaction->t_updates) > 0); |
1250 | err = 0; | 1293 | err = 0; |
1251 | } | 1294 | } |
1252 | 1295 | ||
@@ -1291,9 +1334,9 @@ int jbd2_journal_stop(handle_t *handle) | |||
1291 | 1334 | ||
1292 | journal->j_last_sync_writer = pid; | 1335 | journal->j_last_sync_writer = pid; |
1293 | 1336 | ||
1294 | spin_lock(&journal->j_state_lock); | 1337 | read_lock(&journal->j_state_lock); |
1295 | commit_time = journal->j_average_commit_time; | 1338 | commit_time = journal->j_average_commit_time; |
1296 | spin_unlock(&journal->j_state_lock); | 1339 | read_unlock(&journal->j_state_lock); |
1297 | 1340 | ||
1298 | trans_time = ktime_to_ns(ktime_sub(ktime_get(), | 1341 | trans_time = ktime_to_ns(ktime_sub(ktime_get(), |
1299 | transaction->t_start_time)); | 1342 | transaction->t_start_time)); |
@@ -1314,14 +1357,8 @@ int jbd2_journal_stop(handle_t *handle) | |||
1314 | if (handle->h_sync) | 1357 | if (handle->h_sync) |
1315 | transaction->t_synchronous_commit = 1; | 1358 | transaction->t_synchronous_commit = 1; |
1316 | current->journal_info = NULL; | 1359 | current->journal_info = NULL; |
1317 | spin_lock(&transaction->t_handle_lock); | 1360 | atomic_sub(handle->h_buffer_credits, |
1318 | transaction->t_outstanding_credits -= handle->h_buffer_credits; | 1361 | &transaction->t_outstanding_credits); |
1319 | transaction->t_updates--; | ||
1320 | if (!transaction->t_updates) { | ||
1321 | wake_up(&journal->j_wait_updates); | ||
1322 | if (journal->j_barrier_count) | ||
1323 | wake_up(&journal->j_wait_transaction_locked); | ||
1324 | } | ||
1325 | 1362 | ||
1326 | /* | 1363 | /* |
1327 | * If the handle is marked SYNC, we need to set another commit | 1364 | * If the handle is marked SYNC, we need to set another commit |
@@ -1330,15 +1367,13 @@ int jbd2_journal_stop(handle_t *handle) | |||
1330 | * transaction is too old now. | 1367 | * transaction is too old now. |
1331 | */ | 1368 | */ |
1332 | if (handle->h_sync || | 1369 | if (handle->h_sync || |
1333 | transaction->t_outstanding_credits > | 1370 | (atomic_read(&transaction->t_outstanding_credits) > |
1334 | journal->j_max_transaction_buffers || | 1371 | journal->j_max_transaction_buffers) || |
1335 | time_after_eq(jiffies, transaction->t_expires)) { | 1372 | time_after_eq(jiffies, transaction->t_expires)) { |
1336 | /* Do this even for aborted journals: an abort still | 1373 | /* Do this even for aborted journals: an abort still |
1337 | * completes the commit thread, it just doesn't write | 1374 | * completes the commit thread, it just doesn't write |
1338 | * anything to disk. */ | 1375 | * anything to disk. */ |
1339 | tid_t tid = transaction->t_tid; | ||
1340 | 1376 | ||
1341 | spin_unlock(&transaction->t_handle_lock); | ||
1342 | jbd_debug(2, "transaction too old, requesting commit for " | 1377 | jbd_debug(2, "transaction too old, requesting commit for " |
1343 | "handle %p\n", handle); | 1378 | "handle %p\n", handle); |
1344 | /* This is non-blocking */ | 1379 | /* This is non-blocking */ |
@@ -1349,11 +1384,25 @@ int jbd2_journal_stop(handle_t *handle) | |||
1349 | * to wait for the commit to complete. | 1384 | * to wait for the commit to complete. |
1350 | */ | 1385 | */ |
1351 | if (handle->h_sync && !(current->flags & PF_MEMALLOC)) | 1386 | if (handle->h_sync && !(current->flags & PF_MEMALLOC)) |
1352 | err = jbd2_log_wait_commit(journal, tid); | 1387 | wait_for_commit = 1; |
1353 | } else { | ||
1354 | spin_unlock(&transaction->t_handle_lock); | ||
1355 | } | 1388 | } |
1356 | 1389 | ||
1390 | /* | ||
1391 | * Once we drop t_updates, if it goes to zero the transaction | ||
1392 | * could start commiting on us and eventually disappear. So | ||
1393 | * once we do this, we must not dereference transaction | ||
1394 | * pointer again. | ||
1395 | */ | ||
1396 | tid = transaction->t_tid; | ||
1397 | if (atomic_dec_and_test(&transaction->t_updates)) { | ||
1398 | wake_up(&journal->j_wait_updates); | ||
1399 | if (journal->j_barrier_count) | ||
1400 | wake_up(&journal->j_wait_transaction_locked); | ||
1401 | } | ||
1402 | |||
1403 | if (wait_for_commit) | ||
1404 | err = jbd2_log_wait_commit(journal, tid); | ||
1405 | |||
1357 | lock_map_release(&handle->h_lockdep_map); | 1406 | lock_map_release(&handle->h_lockdep_map); |
1358 | 1407 | ||
1359 | jbd2_free_handle(handle); | 1408 | jbd2_free_handle(handle); |
@@ -1719,7 +1768,7 @@ static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh) | |||
1719 | goto zap_buffer_unlocked; | 1768 | goto zap_buffer_unlocked; |
1720 | 1769 | ||
1721 | /* OK, we have data buffer in journaled mode */ | 1770 | /* OK, we have data buffer in journaled mode */ |
1722 | spin_lock(&journal->j_state_lock); | 1771 | write_lock(&journal->j_state_lock); |
1723 | jbd_lock_bh_state(bh); | 1772 | jbd_lock_bh_state(bh); |
1724 | spin_lock(&journal->j_list_lock); | 1773 | spin_lock(&journal->j_list_lock); |
1725 | 1774 | ||
@@ -1772,7 +1821,7 @@ static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh) | |||
1772 | jbd2_journal_put_journal_head(jh); | 1821 | jbd2_journal_put_journal_head(jh); |
1773 | spin_unlock(&journal->j_list_lock); | 1822 | spin_unlock(&journal->j_list_lock); |
1774 | jbd_unlock_bh_state(bh); | 1823 | jbd_unlock_bh_state(bh); |
1775 | spin_unlock(&journal->j_state_lock); | 1824 | write_unlock(&journal->j_state_lock); |
1776 | return ret; | 1825 | return ret; |
1777 | } else { | 1826 | } else { |
1778 | /* There is no currently-running transaction. So the | 1827 | /* There is no currently-running transaction. So the |
@@ -1786,7 +1835,7 @@ static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh) | |||
1786 | jbd2_journal_put_journal_head(jh); | 1835 | jbd2_journal_put_journal_head(jh); |
1787 | spin_unlock(&journal->j_list_lock); | 1836 | spin_unlock(&journal->j_list_lock); |
1788 | jbd_unlock_bh_state(bh); | 1837 | jbd_unlock_bh_state(bh); |
1789 | spin_unlock(&journal->j_state_lock); | 1838 | write_unlock(&journal->j_state_lock); |
1790 | return ret; | 1839 | return ret; |
1791 | } else { | 1840 | } else { |
1792 | /* The orphan record's transaction has | 1841 | /* The orphan record's transaction has |
@@ -1810,7 +1859,7 @@ static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh) | |||
1810 | jbd2_journal_put_journal_head(jh); | 1859 | jbd2_journal_put_journal_head(jh); |
1811 | spin_unlock(&journal->j_list_lock); | 1860 | spin_unlock(&journal->j_list_lock); |
1812 | jbd_unlock_bh_state(bh); | 1861 | jbd_unlock_bh_state(bh); |
1813 | spin_unlock(&journal->j_state_lock); | 1862 | write_unlock(&journal->j_state_lock); |
1814 | return 0; | 1863 | return 0; |
1815 | } else { | 1864 | } else { |
1816 | /* Good, the buffer belongs to the running transaction. | 1865 | /* Good, the buffer belongs to the running transaction. |
@@ -1829,7 +1878,7 @@ zap_buffer: | |||
1829 | zap_buffer_no_jh: | 1878 | zap_buffer_no_jh: |
1830 | spin_unlock(&journal->j_list_lock); | 1879 | spin_unlock(&journal->j_list_lock); |
1831 | jbd_unlock_bh_state(bh); | 1880 | jbd_unlock_bh_state(bh); |
1832 | spin_unlock(&journal->j_state_lock); | 1881 | write_unlock(&journal->j_state_lock); |
1833 | zap_buffer_unlocked: | 1882 | zap_buffer_unlocked: |
1834 | clear_buffer_dirty(bh); | 1883 | clear_buffer_dirty(bh); |
1835 | J_ASSERT_BH(bh, !buffer_jbddirty(bh)); | 1884 | J_ASSERT_BH(bh, !buffer_jbddirty(bh)); |
@@ -2136,9 +2185,9 @@ int jbd2_journal_begin_ordered_truncate(journal_t *journal, | |||
2136 | /* Locks are here just to force reading of recent values, it is | 2185 | /* Locks are here just to force reading of recent values, it is |
2137 | * enough that the transaction was not committing before we started | 2186 | * enough that the transaction was not committing before we started |
2138 | * a transaction adding the inode to orphan list */ | 2187 | * a transaction adding the inode to orphan list */ |
2139 | spin_lock(&journal->j_state_lock); | 2188 | read_lock(&journal->j_state_lock); |
2140 | commit_trans = journal->j_committing_transaction; | 2189 | commit_trans = journal->j_committing_transaction; |
2141 | spin_unlock(&journal->j_state_lock); | 2190 | read_unlock(&journal->j_state_lock); |
2142 | spin_lock(&journal->j_list_lock); | 2191 | spin_lock(&journal->j_list_lock); |
2143 | inode_trans = jinode->i_transaction; | 2192 | inode_trans = jinode->i_transaction; |
2144 | spin_unlock(&journal->j_list_lock); | 2193 | spin_unlock(&journal->j_list_lock); |