]>
Commit | Line | Data |
---|---|---|
1 | // SPDX-License-Identifier: GPL-2.0 | |
2 | /* | |
3 | * Write ahead logging implementation copyright Chris Mason 2000 | |
4 | * | |
5 | * The background commits make this code very interrelated, and | |
6 | * overly complex. I need to rethink things a bit....The major players: | |
7 | * | |
8 | * journal_begin -- call with the number of blocks you expect to log. | |
9 | * If the current transaction is too | |
10 | * old, it will block until the current transaction is | |
11 | * finished, and then start a new one. | |
12 | * Usually, your transaction will get joined in with | |
13 | * previous ones for speed. | |
14 | * | |
15 | * journal_join -- same as journal_begin, but won't block on the current | |
16 | * transaction regardless of age. Don't ever call | |
17 | * this. Ever. There are only two places it should be | |
18 | * called from, and they are both inside this file. | |
19 | * | |
20 | * journal_mark_dirty -- adds blocks into this transaction. clears any flags | |
21 | * that might make them get sent to disk | |
22 | * and then marks them BH_JDirty. Puts the buffer head | |
23 | * into the current transaction hash. | |
24 | * | |
25 | * journal_end -- if the current transaction is batchable, it does nothing | |
26 | * otherwise, it could do an async/synchronous commit, or | |
27 | * a full flush of all log and real blocks in the | |
28 | * transaction. | |
29 | * | |
30 | * flush_old_commits -- if the current transaction is too old, it is ended and | |
31 | * commit blocks are sent to disk. Forces commit blocks | |
32 | * to disk for all backgrounded commits that have been | |
33 | * around too long. | |
34 | * -- Note, if you call this as an immediate flush from | |
35 | * within kupdate, it will ignore the immediate flag | |
36 | */ | |
37 | ||
38 | #include <linux/time.h> | |
39 | #include <linux/semaphore.h> | |
40 | #include <linux/vmalloc.h> | |
41 | #include "reiserfs.h" | |
42 | #include <linux/kernel.h> | |
43 | #include <linux/errno.h> | |
44 | #include <linux/fcntl.h> | |
45 | #include <linux/stat.h> | |
46 | #include <linux/string.h> | |
47 | #include <linux/buffer_head.h> | |
48 | #include <linux/workqueue.h> | |
49 | #include <linux/writeback.h> | |
50 | #include <linux/blkdev.h> | |
51 | #include <linux/backing-dev.h> | |
52 | #include <linux/uaccess.h> | |
53 | #include <linux/slab.h> | |
54 | ||
55 | ||
56 | /* gets a struct reiserfs_journal_list * from a list head */ | |
57 | #define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \ | |
58 | j_list)) | |
59 | ||
60 | /* must be correct to keep the desc and commit structs at 4k */ | |
61 | #define JOURNAL_TRANS_HALF 1018 | |
62 | #define BUFNR 64 /*read ahead */ | |
63 | ||
64 | /* cnode stat bits. Move these into reiserfs_fs.h */ | |
65 | ||
66 | /* this block was freed, and can't be written. */ | |
67 | #define BLOCK_FREED 2 | |
68 | /* this block was freed during this transaction, and can't be written */ | |
69 | #define BLOCK_FREED_HOLDER 3 | |
70 | ||
71 | /* used in flush_journal_list */ | |
72 | #define BLOCK_NEEDS_FLUSH 4 | |
73 | #define BLOCK_DIRTIED 5 | |
74 | ||
75 | /* journal list state bits */ | |
76 | #define LIST_TOUCHED 1 | |
77 | #define LIST_DIRTY 2 | |
78 | #define LIST_COMMIT_PENDING 4 /* someone will commit this list */ | |
79 | ||
80 | /* flags for do_journal_end */ | |
81 | #define FLUSH_ALL 1 /* flush commit and real blocks */ | |
82 | #define COMMIT_NOW 2 /* end and commit this transaction */ | |
83 | #define WAIT 4 /* wait for the log blocks to hit the disk */ | |
84 | ||
85 | static int do_journal_end(struct reiserfs_transaction_handle *, int flags); | |
86 | static int flush_journal_list(struct super_block *s, | |
87 | struct reiserfs_journal_list *jl, int flushall); | |
88 | static int flush_commit_list(struct super_block *s, | |
89 | struct reiserfs_journal_list *jl, int flushall); | |
90 | static int can_dirty(struct reiserfs_journal_cnode *cn); | |
91 | static int journal_join(struct reiserfs_transaction_handle *th, | |
92 | struct super_block *sb); | |
93 | static void release_journal_dev(struct super_block *super, | |
94 | struct reiserfs_journal *journal); | |
95 | static void dirty_one_transaction(struct super_block *s, | |
96 | struct reiserfs_journal_list *jl); | |
97 | static void flush_async_commits(struct work_struct *work); | |
98 | static void queue_log_writer(struct super_block *s); | |
99 | ||
100 | /* values for join in do_journal_begin_r */ | |
101 | enum { | |
102 | JBEGIN_REG = 0, /* regular journal begin */ | |
103 | /* join the running transaction if at all possible */ | |
104 | JBEGIN_JOIN = 1, | |
105 | /* called from cleanup code, ignores aborted flag */ | |
106 | JBEGIN_ABORT = 2, | |
107 | }; | |
108 | ||
109 | static int do_journal_begin_r(struct reiserfs_transaction_handle *th, | |
110 | struct super_block *sb, | |
111 | unsigned long nblocks, int join); | |
112 | ||
113 | static void init_journal_hash(struct super_block *sb) | |
114 | { | |
115 | struct reiserfs_journal *journal = SB_JOURNAL(sb); | |
116 | memset(journal->j_hash_table, 0, | |
117 | JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *)); | |
118 | } | |
119 | ||
120 | /* | |
121 | * clears BH_Dirty and sticks the buffer on the clean list. Called because | |
122 | * I can't allow refile_buffer to make schedule happen after I've freed a | |
123 | * block. Look at remove_from_transaction and journal_mark_freed for | |
124 | * more details. | |
125 | */ | |
126 | static int reiserfs_clean_and_file_buffer(struct buffer_head *bh) | |
127 | { | |
128 | if (bh) { | |
129 | clear_buffer_dirty(bh); | |
130 | clear_buffer_journal_test(bh); | |
131 | } | |
132 | return 0; | |
133 | } | |
134 | ||
135 | static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block | |
136 | *sb) | |
137 | { | |
138 | struct reiserfs_bitmap_node *bn; | |
139 | static int id; | |
140 | ||
141 | bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS); | |
142 | if (!bn) { | |
143 | return NULL; | |
144 | } | |
145 | bn->data = kzalloc(sb->s_blocksize, GFP_NOFS); | |
146 | if (!bn->data) { | |
147 | kfree(bn); | |
148 | return NULL; | |
149 | } | |
150 | bn->id = id++; | |
151 | INIT_LIST_HEAD(&bn->list); | |
152 | return bn; | |
153 | } | |
154 | ||
155 | static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *sb) | |
156 | { | |
157 | struct reiserfs_journal *journal = SB_JOURNAL(sb); | |
158 | struct reiserfs_bitmap_node *bn = NULL; | |
159 | struct list_head *entry = journal->j_bitmap_nodes.next; | |
160 | ||
161 | journal->j_used_bitmap_nodes++; | |
162 | repeat: | |
163 | ||
164 | if (entry != &journal->j_bitmap_nodes) { | |
165 | bn = list_entry(entry, struct reiserfs_bitmap_node, list); | |
166 | list_del(entry); | |
167 | memset(bn->data, 0, sb->s_blocksize); | |
168 | journal->j_free_bitmap_nodes--; | |
169 | return bn; | |
170 | } | |
171 | bn = allocate_bitmap_node(sb); | |
172 | if (!bn) { | |
173 | yield(); | |
174 | goto repeat; | |
175 | } | |
176 | return bn; | |
177 | } | |
178 | static inline void free_bitmap_node(struct super_block *sb, | |
179 | struct reiserfs_bitmap_node *bn) | |
180 | { | |
181 | struct reiserfs_journal *journal = SB_JOURNAL(sb); | |
182 | journal->j_used_bitmap_nodes--; | |
183 | if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) { | |
184 | kfree(bn->data); | |
185 | kfree(bn); | |
186 | } else { | |
187 | list_add(&bn->list, &journal->j_bitmap_nodes); | |
188 | journal->j_free_bitmap_nodes++; | |
189 | } | |
190 | } | |
191 | ||
192 | static void allocate_bitmap_nodes(struct super_block *sb) | |
193 | { | |
194 | int i; | |
195 | struct reiserfs_journal *journal = SB_JOURNAL(sb); | |
196 | struct reiserfs_bitmap_node *bn = NULL; | |
197 | for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) { | |
198 | bn = allocate_bitmap_node(sb); | |
199 | if (bn) { | |
200 | list_add(&bn->list, &journal->j_bitmap_nodes); | |
201 | journal->j_free_bitmap_nodes++; | |
202 | } else { | |
203 | /* this is ok, we'll try again when more are needed */ | |
204 | break; | |
205 | } | |
206 | } | |
207 | } | |
208 | ||
209 | static int set_bit_in_list_bitmap(struct super_block *sb, | |
210 | b_blocknr_t block, | |
211 | struct reiserfs_list_bitmap *jb) | |
212 | { | |
213 | unsigned int bmap_nr = block / (sb->s_blocksize << 3); | |
214 | unsigned int bit_nr = block % (sb->s_blocksize << 3); | |
215 | ||
216 | if (!jb->bitmaps[bmap_nr]) { | |
217 | jb->bitmaps[bmap_nr] = get_bitmap_node(sb); | |
218 | } | |
219 | set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data); | |
220 | return 0; | |
221 | } | |
222 | ||
223 | static void cleanup_bitmap_list(struct super_block *sb, | |
224 | struct reiserfs_list_bitmap *jb) | |
225 | { | |
226 | int i; | |
227 | if (jb->bitmaps == NULL) | |
228 | return; | |
229 | ||
230 | for (i = 0; i < reiserfs_bmap_count(sb); i++) { | |
231 | if (jb->bitmaps[i]) { | |
232 | free_bitmap_node(sb, jb->bitmaps[i]); | |
233 | jb->bitmaps[i] = NULL; | |
234 | } | |
235 | } | |
236 | } | |
237 | ||
238 | /* | |
239 | * only call this on FS unmount. | |
240 | */ | |
241 | static int free_list_bitmaps(struct super_block *sb, | |
242 | struct reiserfs_list_bitmap *jb_array) | |
243 | { | |
244 | int i; | |
245 | struct reiserfs_list_bitmap *jb; | |
246 | for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) { | |
247 | jb = jb_array + i; | |
248 | jb->journal_list = NULL; | |
249 | cleanup_bitmap_list(sb, jb); | |
250 | vfree(jb->bitmaps); | |
251 | jb->bitmaps = NULL; | |
252 | } | |
253 | return 0; | |
254 | } | |
255 | ||
256 | static int free_bitmap_nodes(struct super_block *sb) | |
257 | { | |
258 | struct reiserfs_journal *journal = SB_JOURNAL(sb); | |
259 | struct list_head *next = journal->j_bitmap_nodes.next; | |
260 | struct reiserfs_bitmap_node *bn; | |
261 | ||
262 | while (next != &journal->j_bitmap_nodes) { | |
263 | bn = list_entry(next, struct reiserfs_bitmap_node, list); | |
264 | list_del(next); | |
265 | kfree(bn->data); | |
266 | kfree(bn); | |
267 | next = journal->j_bitmap_nodes.next; | |
268 | journal->j_free_bitmap_nodes--; | |
269 | } | |
270 | ||
271 | return 0; | |
272 | } | |
273 | ||
274 | /* | |
275 | * get memory for JOURNAL_NUM_BITMAPS worth of bitmaps. | |
276 | * jb_array is the array to be filled in. | |
277 | */ | |
278 | int reiserfs_allocate_list_bitmaps(struct super_block *sb, | |
279 | struct reiserfs_list_bitmap *jb_array, | |
280 | unsigned int bmap_nr) | |
281 | { | |
282 | int i; | |
283 | int failed = 0; | |
284 | struct reiserfs_list_bitmap *jb; | |
285 | int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *); | |
286 | ||
287 | for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) { | |
288 | jb = jb_array + i; | |
289 | jb->journal_list = NULL; | |
290 | jb->bitmaps = vzalloc(mem); | |
291 | if (!jb->bitmaps) { | |
292 | reiserfs_warning(sb, "clm-2000", "unable to " | |
293 | "allocate bitmaps for journal lists"); | |
294 | failed = 1; | |
295 | break; | |
296 | } | |
297 | } | |
298 | if (failed) { | |
299 | free_list_bitmaps(sb, jb_array); | |
300 | return -1; | |
301 | } | |
302 | return 0; | |
303 | } | |
304 | ||
305 | /* | |
306 | * find an available list bitmap. If you can't find one, flush a commit list | |
307 | * and try again | |
308 | */ | |
309 | static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *sb, | |
310 | struct reiserfs_journal_list | |
311 | *jl) | |
312 | { | |
313 | int i, j; | |
314 | struct reiserfs_journal *journal = SB_JOURNAL(sb); | |
315 | struct reiserfs_list_bitmap *jb = NULL; | |
316 | ||
317 | for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) { | |
318 | i = journal->j_list_bitmap_index; | |
319 | journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS; | |
320 | jb = journal->j_list_bitmap + i; | |
321 | if (journal->j_list_bitmap[i].journal_list) { | |
322 | flush_commit_list(sb, | |
323 | journal->j_list_bitmap[i]. | |
324 | journal_list, 1); | |
325 | if (!journal->j_list_bitmap[i].journal_list) { | |
326 | break; | |
327 | } | |
328 | } else { | |
329 | break; | |
330 | } | |
331 | } | |
332 | /* double check to make sure if flushed correctly */ | |
333 | if (jb->journal_list) | |
334 | return NULL; | |
335 | jb->journal_list = jl; | |
336 | return jb; | |
337 | } | |
338 | ||
339 | /* | |
340 | * allocates a new chunk of X nodes, and links them all together as a list. | |
341 | * Uses the cnode->next and cnode->prev pointers | |
342 | * returns NULL on failure | |
343 | */ | |
344 | static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes) | |
345 | { | |
346 | struct reiserfs_journal_cnode *head; | |
347 | int i; | |
348 | if (num_cnodes <= 0) { | |
349 | return NULL; | |
350 | } | |
351 | head = vzalloc(array_size(num_cnodes, | |
352 | sizeof(struct reiserfs_journal_cnode))); | |
353 | if (!head) { | |
354 | return NULL; | |
355 | } | |
356 | head[0].prev = NULL; | |
357 | head[0].next = head + 1; | |
358 | for (i = 1; i < num_cnodes; i++) { | |
359 | head[i].prev = head + (i - 1); | |
360 | head[i].next = head + (i + 1); /* if last one, overwrite it after the if */ | |
361 | } | |
362 | head[num_cnodes - 1].next = NULL; | |
363 | return head; | |
364 | } | |
365 | ||
366 | /* pulls a cnode off the free list, or returns NULL on failure */ | |
367 | static struct reiserfs_journal_cnode *get_cnode(struct super_block *sb) | |
368 | { | |
369 | struct reiserfs_journal_cnode *cn; | |
370 | struct reiserfs_journal *journal = SB_JOURNAL(sb); | |
371 | ||
372 | reiserfs_check_lock_depth(sb, "get_cnode"); | |
373 | ||
374 | if (journal->j_cnode_free <= 0) { | |
375 | return NULL; | |
376 | } | |
377 | journal->j_cnode_used++; | |
378 | journal->j_cnode_free--; | |
379 | cn = journal->j_cnode_free_list; | |
380 | if (!cn) { | |
381 | return cn; | |
382 | } | |
383 | if (cn->next) { | |
384 | cn->next->prev = NULL; | |
385 | } | |
386 | journal->j_cnode_free_list = cn->next; | |
387 | memset(cn, 0, sizeof(struct reiserfs_journal_cnode)); | |
388 | return cn; | |
389 | } | |
390 | ||
391 | /* | |
392 | * returns a cnode to the free list | |
393 | */ | |
394 | static void free_cnode(struct super_block *sb, | |
395 | struct reiserfs_journal_cnode *cn) | |
396 | { | |
397 | struct reiserfs_journal *journal = SB_JOURNAL(sb); | |
398 | ||
399 | reiserfs_check_lock_depth(sb, "free_cnode"); | |
400 | ||
401 | journal->j_cnode_used--; | |
402 | journal->j_cnode_free++; | |
403 | /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */ | |
404 | cn->next = journal->j_cnode_free_list; | |
405 | if (journal->j_cnode_free_list) { | |
406 | journal->j_cnode_free_list->prev = cn; | |
407 | } | |
408 | cn->prev = NULL; /* not needed with the memset, but I might kill the memset, and forget to do this */ | |
409 | journal->j_cnode_free_list = cn; | |
410 | } | |
411 | ||
412 | static void clear_prepared_bits(struct buffer_head *bh) | |
413 | { | |
414 | clear_buffer_journal_prepared(bh); | |
415 | clear_buffer_journal_restore_dirty(bh); | |
416 | } | |
417 | ||
418 | /* | |
419 | * return a cnode with same dev, block number and size in table, | |
420 | * or null if not found | |
421 | */ | |
422 | static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct | |
423 | super_block | |
424 | *sb, | |
425 | struct | |
426 | reiserfs_journal_cnode | |
427 | **table, | |
428 | long bl) | |
429 | { | |
430 | struct reiserfs_journal_cnode *cn; | |
431 | cn = journal_hash(table, sb, bl); | |
432 | while (cn) { | |
433 | if (cn->blocknr == bl && cn->sb == sb) | |
434 | return cn; | |
435 | cn = cn->hnext; | |
436 | } | |
437 | return (struct reiserfs_journal_cnode *)0; | |
438 | } | |
439 | ||
440 | /* | |
441 | * this actually means 'can this block be reallocated yet?'. If you set | |
442 | * search_all, a block can only be allocated if it is not in the current | |
443 | * transaction, was not freed by the current transaction, and has no chance | |
444 | * of ever being overwritten by a replay after crashing. | |
445 | * | |
446 | * If you don't set search_all, a block can only be allocated if it is not | |
447 | * in the current transaction. Since deleting a block removes it from the | |
448 | * current transaction, this case should never happen. If you don't set | |
449 | * search_all, make sure you never write the block without logging it. | |
450 | * | |
451 | * next_zero_bit is a suggestion about the next block to try for find_forward. | |
452 | * when bl is rejected because it is set in a journal list bitmap, we search | |
453 | * for the next zero bit in the bitmap that rejected bl. Then, we return | |
454 | * that through next_zero_bit for find_forward to try. | |
455 | * | |
456 | * Just because we return something in next_zero_bit does not mean we won't | |
457 | * reject it on the next call to reiserfs_in_journal | |
458 | */ | |
459 | int reiserfs_in_journal(struct super_block *sb, | |
460 | unsigned int bmap_nr, int bit_nr, int search_all, | |
461 | b_blocknr_t * next_zero_bit) | |
462 | { | |
463 | struct reiserfs_journal *journal = SB_JOURNAL(sb); | |
464 | struct reiserfs_journal_cnode *cn; | |
465 | struct reiserfs_list_bitmap *jb; | |
466 | int i; | |
467 | unsigned long bl; | |
468 | ||
469 | *next_zero_bit = 0; /* always start this at zero. */ | |
470 | ||
471 | PROC_INFO_INC(sb, journal.in_journal); | |
472 | /* | |
473 | * If we aren't doing a search_all, this is a metablock, and it | |
474 | * will be logged before use. if we crash before the transaction | |
475 | * that freed it commits, this transaction won't have committed | |
476 | * either, and the block will never be written | |
477 | */ | |
478 | if (search_all) { | |
479 | for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) { | |
480 | PROC_INFO_INC(sb, journal.in_journal_bitmap); | |
481 | jb = journal->j_list_bitmap + i; | |
482 | if (jb->journal_list && jb->bitmaps[bmap_nr] && | |
483 | test_bit(bit_nr, | |
484 | (unsigned long *)jb->bitmaps[bmap_nr]-> | |
485 | data)) { | |
486 | *next_zero_bit = | |
487 | find_next_zero_bit((unsigned long *) | |
488 | (jb->bitmaps[bmap_nr]-> | |
489 | data), | |
490 | sb->s_blocksize << 3, | |
491 | bit_nr + 1); | |
492 | return 1; | |
493 | } | |
494 | } | |
495 | } | |
496 | ||
497 | bl = bmap_nr * (sb->s_blocksize << 3) + bit_nr; | |
498 | /* is it in any old transactions? */ | |
499 | if (search_all | |
500 | && (cn = | |
501 | get_journal_hash_dev(sb, journal->j_list_hash_table, bl))) { | |
502 | return 1; | |
503 | } | |
504 | ||
505 | /* is it in the current transaction. This should never happen */ | |
506 | if ((cn = get_journal_hash_dev(sb, journal->j_hash_table, bl))) { | |
507 | BUG(); | |
508 | return 1; | |
509 | } | |
510 | ||
511 | PROC_INFO_INC(sb, journal.in_journal_reusable); | |
512 | /* safe for reuse */ | |
513 | return 0; | |
514 | } | |
515 | ||
516 | /* insert cn into table */ | |
517 | static inline void insert_journal_hash(struct reiserfs_journal_cnode **table, | |
518 | struct reiserfs_journal_cnode *cn) | |
519 | { | |
520 | struct reiserfs_journal_cnode *cn_orig; | |
521 | ||
522 | cn_orig = journal_hash(table, cn->sb, cn->blocknr); | |
523 | cn->hnext = cn_orig; | |
524 | cn->hprev = NULL; | |
525 | if (cn_orig) { | |
526 | cn_orig->hprev = cn; | |
527 | } | |
528 | journal_hash(table, cn->sb, cn->blocknr) = cn; | |
529 | } | |
530 | ||
531 | /* lock the current transaction */ | |
532 | static inline void lock_journal(struct super_block *sb) | |
533 | { | |
534 | PROC_INFO_INC(sb, journal.lock_journal); | |
535 | ||
536 | reiserfs_mutex_lock_safe(&SB_JOURNAL(sb)->j_mutex, sb); | |
537 | } | |
538 | ||
539 | /* unlock the current transaction */ | |
540 | static inline void unlock_journal(struct super_block *sb) | |
541 | { | |
542 | mutex_unlock(&SB_JOURNAL(sb)->j_mutex); | |
543 | } | |
544 | ||
545 | static inline void get_journal_list(struct reiserfs_journal_list *jl) | |
546 | { | |
547 | jl->j_refcount++; | |
548 | } | |
549 | ||
550 | static inline void put_journal_list(struct super_block *s, | |
551 | struct reiserfs_journal_list *jl) | |
552 | { | |
553 | if (jl->j_refcount < 1) { | |
554 | reiserfs_panic(s, "journal-2", "trans id %u, refcount at %d", | |
555 | jl->j_trans_id, jl->j_refcount); | |
556 | } | |
557 | if (--jl->j_refcount == 0) | |
558 | kfree(jl); | |
559 | } | |
560 | ||
561 | /* | |
562 | * this used to be much more involved, and I'm keeping it just in case | |
563 | * things get ugly again. it gets called by flush_commit_list, and | |
564 | * cleans up any data stored about blocks freed during a transaction. | |
565 | */ | |
566 | static void cleanup_freed_for_journal_list(struct super_block *sb, | |
567 | struct reiserfs_journal_list *jl) | |
568 | { | |
569 | ||
570 | struct reiserfs_list_bitmap *jb = jl->j_list_bitmap; | |
571 | if (jb) { | |
572 | cleanup_bitmap_list(sb, jb); | |
573 | } | |
574 | jl->j_list_bitmap->journal_list = NULL; | |
575 | jl->j_list_bitmap = NULL; | |
576 | } | |
577 | ||
578 | static int journal_list_still_alive(struct super_block *s, | |
579 | unsigned int trans_id) | |
580 | { | |
581 | struct reiserfs_journal *journal = SB_JOURNAL(s); | |
582 | struct list_head *entry = &journal->j_journal_list; | |
583 | struct reiserfs_journal_list *jl; | |
584 | ||
585 | if (!list_empty(entry)) { | |
586 | jl = JOURNAL_LIST_ENTRY(entry->next); | |
587 | if (jl->j_trans_id <= trans_id) { | |
588 | return 1; | |
589 | } | |
590 | } | |
591 | return 0; | |
592 | } | |
593 | ||
594 | /* | |
595 | * If page->mapping was null, we failed to truncate this page for | |
596 | * some reason. Most likely because it was truncated after being | |
597 | * logged via data=journal. | |
598 | * | |
599 | * This does a check to see if the buffer belongs to one of these | |
600 | * lost pages before doing the final put_bh. If page->mapping was | |
601 | * null, it tries to free buffers on the page, which should make the | |
602 | * final put_page drop the page from the lru. | |
603 | */ | |
604 | static void release_buffer_page(struct buffer_head *bh) | |
605 | { | |
606 | struct page *page = bh->b_page; | |
607 | if (!page->mapping && trylock_page(page)) { | |
608 | get_page(page); | |
609 | put_bh(bh); | |
610 | if (!page->mapping) | |
611 | try_to_free_buffers(page); | |
612 | unlock_page(page); | |
613 | put_page(page); | |
614 | } else { | |
615 | put_bh(bh); | |
616 | } | |
617 | } | |
618 | ||
619 | static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate) | |
620 | { | |
621 | if (buffer_journaled(bh)) { | |
622 | reiserfs_warning(NULL, "clm-2084", | |
623 | "pinned buffer %lu:%pg sent to disk", | |
624 | bh->b_blocknr, bh->b_bdev); | |
625 | } | |
626 | if (uptodate) | |
627 | set_buffer_uptodate(bh); | |
628 | else | |
629 | clear_buffer_uptodate(bh); | |
630 | ||
631 | unlock_buffer(bh); | |
632 | release_buffer_page(bh); | |
633 | } | |
634 | ||
635 | static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate) | |
636 | { | |
637 | if (uptodate) | |
638 | set_buffer_uptodate(bh); | |
639 | else | |
640 | clear_buffer_uptodate(bh); | |
641 | unlock_buffer(bh); | |
642 | put_bh(bh); | |
643 | } | |
644 | ||
645 | static void submit_logged_buffer(struct buffer_head *bh) | |
646 | { | |
647 | get_bh(bh); | |
648 | bh->b_end_io = reiserfs_end_buffer_io_sync; | |
649 | clear_buffer_journal_new(bh); | |
650 | clear_buffer_dirty(bh); | |
651 | if (!test_clear_buffer_journal_test(bh)) | |
652 | BUG(); | |
653 | if (!buffer_uptodate(bh)) | |
654 | BUG(); | |
655 | submit_bh(REQ_OP_WRITE, 0, bh); | |
656 | } | |
657 | ||
658 | static void submit_ordered_buffer(struct buffer_head *bh) | |
659 | { | |
660 | get_bh(bh); | |
661 | bh->b_end_io = reiserfs_end_ordered_io; | |
662 | clear_buffer_dirty(bh); | |
663 | if (!buffer_uptodate(bh)) | |
664 | BUG(); | |
665 | submit_bh(REQ_OP_WRITE, 0, bh); | |
666 | } | |
667 | ||
668 | #define CHUNK_SIZE 32 | |
669 | struct buffer_chunk { | |
670 | struct buffer_head *bh[CHUNK_SIZE]; | |
671 | int nr; | |
672 | }; | |
673 | ||
674 | static void write_chunk(struct buffer_chunk *chunk) | |
675 | { | |
676 | int i; | |
677 | for (i = 0; i < chunk->nr; i++) { | |
678 | submit_logged_buffer(chunk->bh[i]); | |
679 | } | |
680 | chunk->nr = 0; | |
681 | } | |
682 | ||
683 | static void write_ordered_chunk(struct buffer_chunk *chunk) | |
684 | { | |
685 | int i; | |
686 | for (i = 0; i < chunk->nr; i++) { | |
687 | submit_ordered_buffer(chunk->bh[i]); | |
688 | } | |
689 | chunk->nr = 0; | |
690 | } | |
691 | ||
692 | static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh, | |
693 | spinlock_t * lock, void (fn) (struct buffer_chunk *)) | |
694 | { | |
695 | int ret = 0; | |
696 | BUG_ON(chunk->nr >= CHUNK_SIZE); | |
697 | chunk->bh[chunk->nr++] = bh; | |
698 | if (chunk->nr >= CHUNK_SIZE) { | |
699 | ret = 1; | |
700 | if (lock) { | |
701 | spin_unlock(lock); | |
702 | fn(chunk); | |
703 | spin_lock(lock); | |
704 | } else { | |
705 | fn(chunk); | |
706 | } | |
707 | } | |
708 | return ret; | |
709 | } | |
710 | ||
711 | static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0); | |
712 | static struct reiserfs_jh *alloc_jh(void) | |
713 | { | |
714 | struct reiserfs_jh *jh; | |
715 | while (1) { | |
716 | jh = kmalloc(sizeof(*jh), GFP_NOFS); | |
717 | if (jh) { | |
718 | atomic_inc(&nr_reiserfs_jh); | |
719 | return jh; | |
720 | } | |
721 | yield(); | |
722 | } | |
723 | } | |
724 | ||
725 | /* | |
726 | * we want to free the jh when the buffer has been written | |
727 | * and waited on | |
728 | */ | |
729 | void reiserfs_free_jh(struct buffer_head *bh) | |
730 | { | |
731 | struct reiserfs_jh *jh; | |
732 | ||
733 | jh = bh->b_private; | |
734 | if (jh) { | |
735 | bh->b_private = NULL; | |
736 | jh->bh = NULL; | |
737 | list_del_init(&jh->list); | |
738 | kfree(jh); | |
739 | if (atomic_read(&nr_reiserfs_jh) <= 0) | |
740 | BUG(); | |
741 | atomic_dec(&nr_reiserfs_jh); | |
742 | put_bh(bh); | |
743 | } | |
744 | } | |
745 | ||
746 | static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh, | |
747 | int tail) | |
748 | { | |
749 | struct reiserfs_jh *jh; | |
750 | ||
751 | if (bh->b_private) { | |
752 | spin_lock(&j->j_dirty_buffers_lock); | |
753 | if (!bh->b_private) { | |
754 | spin_unlock(&j->j_dirty_buffers_lock); | |
755 | goto no_jh; | |
756 | } | |
757 | jh = bh->b_private; | |
758 | list_del_init(&jh->list); | |
759 | } else { | |
760 | no_jh: | |
761 | get_bh(bh); | |
762 | jh = alloc_jh(); | |
763 | spin_lock(&j->j_dirty_buffers_lock); | |
764 | /* | |
765 | * buffer must be locked for __add_jh, should be able to have | |
766 | * two adds at the same time | |
767 | */ | |
768 | BUG_ON(bh->b_private); | |
769 | jh->bh = bh; | |
770 | bh->b_private = jh; | |
771 | } | |
772 | jh->jl = j->j_current_jl; | |
773 | if (tail) | |
774 | list_add_tail(&jh->list, &jh->jl->j_tail_bh_list); | |
775 | else { | |
776 | list_add_tail(&jh->list, &jh->jl->j_bh_list); | |
777 | } | |
778 | spin_unlock(&j->j_dirty_buffers_lock); | |
779 | return 0; | |
780 | } | |
781 | ||
782 | int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh) | |
783 | { | |
784 | return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1); | |
785 | } | |
786 | int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh) | |
787 | { | |
788 | return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0); | |
789 | } | |
790 | ||
791 | #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list) | |
792 | static int write_ordered_buffers(spinlock_t * lock, | |
793 | struct reiserfs_journal *j, | |
794 | struct reiserfs_journal_list *jl, | |
795 | struct list_head *list) | |
796 | { | |
797 | struct buffer_head *bh; | |
798 | struct reiserfs_jh *jh; | |
799 | int ret = j->j_errno; | |
800 | struct buffer_chunk chunk; | |
801 | struct list_head tmp; | |
802 | INIT_LIST_HEAD(&tmp); | |
803 | ||
804 | chunk.nr = 0; | |
805 | spin_lock(lock); | |
806 | while (!list_empty(list)) { | |
807 | jh = JH_ENTRY(list->next); | |
808 | bh = jh->bh; | |
809 | get_bh(bh); | |
810 | if (!trylock_buffer(bh)) { | |
811 | if (!buffer_dirty(bh)) { | |
812 | list_move(&jh->list, &tmp); | |
813 | goto loop_next; | |
814 | } | |
815 | spin_unlock(lock); | |
816 | if (chunk.nr) | |
817 | write_ordered_chunk(&chunk); | |
818 | wait_on_buffer(bh); | |
819 | cond_resched(); | |
820 | spin_lock(lock); | |
821 | goto loop_next; | |
822 | } | |
823 | /* | |
824 | * in theory, dirty non-uptodate buffers should never get here, | |
825 | * but the upper layer io error paths still have a few quirks. | |
826 | * Handle them here as gracefully as we can | |
827 | */ | |
828 | if (!buffer_uptodate(bh) && buffer_dirty(bh)) { | |
829 | clear_buffer_dirty(bh); | |
830 | ret = -EIO; | |
831 | } | |
832 | if (buffer_dirty(bh)) { | |
833 | list_move(&jh->list, &tmp); | |
834 | add_to_chunk(&chunk, bh, lock, write_ordered_chunk); | |
835 | } else { | |
836 | reiserfs_free_jh(bh); | |
837 | unlock_buffer(bh); | |
838 | } | |
839 | loop_next: | |
840 | put_bh(bh); | |
841 | cond_resched_lock(lock); | |
842 | } | |
843 | if (chunk.nr) { | |
844 | spin_unlock(lock); | |
845 | write_ordered_chunk(&chunk); | |
846 | spin_lock(lock); | |
847 | } | |
848 | while (!list_empty(&tmp)) { | |
849 | jh = JH_ENTRY(tmp.prev); | |
850 | bh = jh->bh; | |
851 | get_bh(bh); | |
852 | reiserfs_free_jh(bh); | |
853 | ||
854 | if (buffer_locked(bh)) { | |
855 | spin_unlock(lock); | |
856 | wait_on_buffer(bh); | |
857 | spin_lock(lock); | |
858 | } | |
859 | if (!buffer_uptodate(bh)) { | |
860 | ret = -EIO; | |
861 | } | |
862 | /* | |
863 | * ugly interaction with invalidatepage here. | |
864 | * reiserfs_invalidate_page will pin any buffer that has a | |
865 | * valid journal head from an older transaction. If someone | |
866 | * else sets our buffer dirty after we write it in the first | |
867 | * loop, and then someone truncates the page away, nobody | |
868 | * will ever write the buffer. We're safe if we write the | |
869 | * page one last time after freeing the journal header. | |
870 | */ | |
871 | if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) { | |
872 | spin_unlock(lock); | |
873 | ll_rw_block(REQ_OP_WRITE, 0, 1, &bh); | |
874 | spin_lock(lock); | |
875 | } | |
876 | put_bh(bh); | |
877 | cond_resched_lock(lock); | |
878 | } | |
879 | spin_unlock(lock); | |
880 | return ret; | |
881 | } | |
882 | ||
883 | static int flush_older_commits(struct super_block *s, | |
884 | struct reiserfs_journal_list *jl) | |
885 | { | |
886 | struct reiserfs_journal *journal = SB_JOURNAL(s); | |
887 | struct reiserfs_journal_list *other_jl; | |
888 | struct reiserfs_journal_list *first_jl; | |
889 | struct list_head *entry; | |
890 | unsigned int trans_id = jl->j_trans_id; | |
891 | unsigned int other_trans_id; | |
892 | ||
893 | find_first: | |
894 | /* | |
895 | * first we walk backwards to find the oldest uncommitted transation | |
896 | */ | |
897 | first_jl = jl; | |
898 | entry = jl->j_list.prev; | |
899 | while (1) { | |
900 | other_jl = JOURNAL_LIST_ENTRY(entry); | |
901 | if (entry == &journal->j_journal_list || | |
902 | atomic_read(&other_jl->j_older_commits_done)) | |
903 | break; | |
904 | ||
905 | first_jl = other_jl; | |
906 | entry = other_jl->j_list.prev; | |
907 | } | |
908 | ||
909 | /* if we didn't find any older uncommitted transactions, return now */ | |
910 | if (first_jl == jl) { | |
911 | return 0; | |
912 | } | |
913 | ||
914 | entry = &first_jl->j_list; | |
915 | while (1) { | |
916 | other_jl = JOURNAL_LIST_ENTRY(entry); | |
917 | other_trans_id = other_jl->j_trans_id; | |
918 | ||
919 | if (other_trans_id < trans_id) { | |
920 | if (atomic_read(&other_jl->j_commit_left) != 0) { | |
921 | flush_commit_list(s, other_jl, 0); | |
922 | ||
923 | /* list we were called with is gone, return */ | |
924 | if (!journal_list_still_alive(s, trans_id)) | |
925 | return 1; | |
926 | ||
927 | /* | |
928 | * the one we just flushed is gone, this means | |
929 | * all older lists are also gone, so first_jl | |
930 | * is no longer valid either. Go back to the | |
931 | * beginning. | |
932 | */ | |
933 | if (!journal_list_still_alive | |
934 | (s, other_trans_id)) { | |
935 | goto find_first; | |
936 | } | |
937 | } | |
938 | entry = entry->next; | |
939 | if (entry == &journal->j_journal_list) | |
940 | return 0; | |
941 | } else { | |
942 | return 0; | |
943 | } | |
944 | } | |
945 | return 0; | |
946 | } | |
947 | ||
948 | static int reiserfs_async_progress_wait(struct super_block *s) | |
949 | { | |
950 | struct reiserfs_journal *j = SB_JOURNAL(s); | |
951 | ||
952 | if (atomic_read(&j->j_async_throttle)) { | |
953 | int depth; | |
954 | ||
955 | depth = reiserfs_write_unlock_nested(s); | |
956 | congestion_wait(BLK_RW_ASYNC, HZ / 10); | |
957 | reiserfs_write_lock_nested(s, depth); | |
958 | } | |
959 | ||
960 | return 0; | |
961 | } | |
962 | ||
963 | /* | |
964 | * if this journal list still has commit blocks unflushed, send them to disk. | |
965 | * | |
966 | * log areas must be flushed in order (transaction 2 can't commit before | |
967 | * transaction 1) Before the commit block can by written, every other log | |
968 | * block must be safely on disk | |
969 | */ | |
970 | static int flush_commit_list(struct super_block *s, | |
971 | struct reiserfs_journal_list *jl, int flushall) | |
972 | { | |
973 | int i; | |
974 | b_blocknr_t bn; | |
975 | struct buffer_head *tbh = NULL; | |
976 | unsigned int trans_id = jl->j_trans_id; | |
977 | struct reiserfs_journal *journal = SB_JOURNAL(s); | |
978 | int retval = 0; | |
979 | int write_len; | |
980 | int depth; | |
981 | ||
982 | reiserfs_check_lock_depth(s, "flush_commit_list"); | |
983 | ||
984 | if (atomic_read(&jl->j_older_commits_done)) { | |
985 | return 0; | |
986 | } | |
987 | ||
988 | /* | |
989 | * before we can put our commit blocks on disk, we have to make | |
990 | * sure everyone older than us is on disk too | |
991 | */ | |
992 | BUG_ON(jl->j_len <= 0); | |
993 | BUG_ON(trans_id == journal->j_trans_id); | |
994 | ||
995 | get_journal_list(jl); | |
996 | if (flushall) { | |
997 | if (flush_older_commits(s, jl) == 1) { | |
998 | /* | |
999 | * list disappeared during flush_older_commits. | |
1000 | * return | |
1001 | */ | |
1002 | goto put_jl; | |
1003 | } | |
1004 | } | |
1005 | ||
1006 | /* make sure nobody is trying to flush this one at the same time */ | |
1007 | reiserfs_mutex_lock_safe(&jl->j_commit_mutex, s); | |
1008 | ||
1009 | if (!journal_list_still_alive(s, trans_id)) { | |
1010 | mutex_unlock(&jl->j_commit_mutex); | |
1011 | goto put_jl; | |
1012 | } | |
1013 | BUG_ON(jl->j_trans_id == 0); | |
1014 | ||
1015 | /* this commit is done, exit */ | |
1016 | if (atomic_read(&jl->j_commit_left) <= 0) { | |
1017 | if (flushall) { | |
1018 | atomic_set(&jl->j_older_commits_done, 1); | |
1019 | } | |
1020 | mutex_unlock(&jl->j_commit_mutex); | |
1021 | goto put_jl; | |
1022 | } | |
1023 | ||
1024 | if (!list_empty(&jl->j_bh_list)) { | |
1025 | int ret; | |
1026 | ||
1027 | /* | |
1028 | * We might sleep in numerous places inside | |
1029 | * write_ordered_buffers. Relax the write lock. | |
1030 | */ | |
1031 | depth = reiserfs_write_unlock_nested(s); | |
1032 | ret = write_ordered_buffers(&journal->j_dirty_buffers_lock, | |
1033 | journal, jl, &jl->j_bh_list); | |
1034 | if (ret < 0 && retval == 0) | |
1035 | retval = ret; | |
1036 | reiserfs_write_lock_nested(s, depth); | |
1037 | } | |
1038 | BUG_ON(!list_empty(&jl->j_bh_list)); | |
1039 | /* | |
1040 | * for the description block and all the log blocks, submit any buffers | |
1041 | * that haven't already reached the disk. Try to write at least 256 | |
1042 | * log blocks. later on, we will only wait on blocks that correspond | |
1043 | * to this transaction, but while we're unplugging we might as well | |
1044 | * get a chunk of data on there. | |
1045 | */ | |
1046 | atomic_inc(&journal->j_async_throttle); | |
1047 | write_len = jl->j_len + 1; | |
1048 | if (write_len < 256) | |
1049 | write_len = 256; | |
1050 | for (i = 0 ; i < write_len ; i++) { | |
1051 | bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) % | |
1052 | SB_ONDISK_JOURNAL_SIZE(s); | |
1053 | tbh = journal_find_get_block(s, bn); | |
1054 | if (tbh) { | |
1055 | if (buffer_dirty(tbh)) { | |
1056 | depth = reiserfs_write_unlock_nested(s); | |
1057 | ll_rw_block(REQ_OP_WRITE, 0, 1, &tbh); | |
1058 | reiserfs_write_lock_nested(s, depth); | |
1059 | } | |
1060 | put_bh(tbh) ; | |
1061 | } | |
1062 | } | |
1063 | atomic_dec(&journal->j_async_throttle); | |
1064 | ||
1065 | for (i = 0; i < (jl->j_len + 1); i++) { | |
1066 | bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + | |
1067 | (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s); | |
1068 | tbh = journal_find_get_block(s, bn); | |
1069 | ||
1070 | depth = reiserfs_write_unlock_nested(s); | |
1071 | __wait_on_buffer(tbh); | |
1072 | reiserfs_write_lock_nested(s, depth); | |
1073 | /* | |
1074 | * since we're using ll_rw_blk above, it might have skipped | |
1075 | * over a locked buffer. Double check here | |
1076 | */ | |
1077 | /* redundant, sync_dirty_buffer() checks */ | |
1078 | if (buffer_dirty(tbh)) { | |
1079 | depth = reiserfs_write_unlock_nested(s); | |
1080 | sync_dirty_buffer(tbh); | |
1081 | reiserfs_write_lock_nested(s, depth); | |
1082 | } | |
1083 | if (unlikely(!buffer_uptodate(tbh))) { | |
1084 | #ifdef CONFIG_REISERFS_CHECK | |
1085 | reiserfs_warning(s, "journal-601", | |
1086 | "buffer write failed"); | |
1087 | #endif | |
1088 | retval = -EIO; | |
1089 | } | |
1090 | /* once for journal_find_get_block */ | |
1091 | put_bh(tbh); | |
1092 | /* once due to original getblk in do_journal_end */ | |
1093 | put_bh(tbh); | |
1094 | atomic_dec(&jl->j_commit_left); | |
1095 | } | |
1096 | ||
1097 | BUG_ON(atomic_read(&jl->j_commit_left) != 1); | |
1098 | ||
1099 | /* | |
1100 | * If there was a write error in the journal - we can't commit | |
1101 | * this transaction - it will be invalid and, if successful, | |
1102 | * will just end up propagating the write error out to | |
1103 | * the file system. | |
1104 | */ | |
1105 | if (likely(!retval && !reiserfs_is_journal_aborted (journal))) { | |
1106 | if (buffer_dirty(jl->j_commit_bh)) | |
1107 | BUG(); | |
1108 | mark_buffer_dirty(jl->j_commit_bh) ; | |
1109 | depth = reiserfs_write_unlock_nested(s); | |
1110 | if (reiserfs_barrier_flush(s)) | |
1111 | __sync_dirty_buffer(jl->j_commit_bh, | |
1112 | REQ_SYNC | REQ_PREFLUSH | REQ_FUA); | |
1113 | else | |
1114 | sync_dirty_buffer(jl->j_commit_bh); | |
1115 | reiserfs_write_lock_nested(s, depth); | |
1116 | } | |
1117 | ||
1118 | /* | |
1119 | * If there was a write error in the journal - we can't commit this | |
1120 | * transaction - it will be invalid and, if successful, will just end | |
1121 | * up propagating the write error out to the filesystem. | |
1122 | */ | |
1123 | if (unlikely(!buffer_uptodate(jl->j_commit_bh))) { | |
1124 | #ifdef CONFIG_REISERFS_CHECK | |
1125 | reiserfs_warning(s, "journal-615", "buffer write failed"); | |
1126 | #endif | |
1127 | retval = -EIO; | |
1128 | } | |
1129 | bforget(jl->j_commit_bh); | |
1130 | if (journal->j_last_commit_id != 0 && | |
1131 | (jl->j_trans_id - journal->j_last_commit_id) != 1) { | |
1132 | reiserfs_warning(s, "clm-2200", "last commit %lu, current %lu", | |
1133 | journal->j_last_commit_id, jl->j_trans_id); | |
1134 | } | |
1135 | journal->j_last_commit_id = jl->j_trans_id; | |
1136 | ||
1137 | /* | |
1138 | * now, every commit block is on the disk. It is safe to allow | |
1139 | * blocks freed during this transaction to be reallocated | |
1140 | */ | |
1141 | cleanup_freed_for_journal_list(s, jl); | |
1142 | ||
1143 | retval = retval ? retval : journal->j_errno; | |
1144 | ||
1145 | /* mark the metadata dirty */ | |
1146 | if (!retval) | |
1147 | dirty_one_transaction(s, jl); | |
1148 | atomic_dec(&jl->j_commit_left); | |
1149 | ||
1150 | if (flushall) { | |
1151 | atomic_set(&jl->j_older_commits_done, 1); | |
1152 | } | |
1153 | mutex_unlock(&jl->j_commit_mutex); | |
1154 | put_jl: | |
1155 | put_journal_list(s, jl); | |
1156 | ||
1157 | if (retval) | |
1158 | reiserfs_abort(s, retval, "Journal write error in %s", | |
1159 | __func__); | |
1160 | return retval; | |
1161 | } | |
1162 | ||
1163 | /* | |
1164 | * flush_journal_list frequently needs to find a newer transaction for a | |
1165 | * given block. This does that, or returns NULL if it can't find anything | |
1166 | */ | |
1167 | static struct reiserfs_journal_list *find_newer_jl_for_cn(struct | |
1168 | reiserfs_journal_cnode | |
1169 | *cn) | |
1170 | { | |
1171 | struct super_block *sb = cn->sb; | |
1172 | b_blocknr_t blocknr = cn->blocknr; | |
1173 | ||
1174 | cn = cn->hprev; | |
1175 | while (cn) { | |
1176 | if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) { | |
1177 | return cn->jlist; | |
1178 | } | |
1179 | cn = cn->hprev; | |
1180 | } | |
1181 | return NULL; | |
1182 | } | |
1183 | ||
1184 | static void remove_journal_hash(struct super_block *, | |
1185 | struct reiserfs_journal_cnode **, | |
1186 | struct reiserfs_journal_list *, unsigned long, | |
1187 | int); | |
1188 | ||
1189 | /* | |
1190 | * once all the real blocks have been flushed, it is safe to remove them | |
1191 | * from the journal list for this transaction. Aside from freeing the | |
1192 | * cnode, this also allows the block to be reallocated for data blocks | |
1193 | * if it had been deleted. | |
1194 | */ | |
1195 | static void remove_all_from_journal_list(struct super_block *sb, | |
1196 | struct reiserfs_journal_list *jl, | |
1197 | int debug) | |
1198 | { | |
1199 | struct reiserfs_journal *journal = SB_JOURNAL(sb); | |
1200 | struct reiserfs_journal_cnode *cn, *last; | |
1201 | cn = jl->j_realblock; | |
1202 | ||
1203 | /* | |
1204 | * which is better, to lock once around the whole loop, or | |
1205 | * to lock for each call to remove_journal_hash? | |
1206 | */ | |
1207 | while (cn) { | |
1208 | if (cn->blocknr != 0) { | |
1209 | if (debug) { | |
1210 | reiserfs_warning(sb, "reiserfs-2201", | |
1211 | "block %u, bh is %d, state %ld", | |
1212 | cn->blocknr, cn->bh ? 1 : 0, | |
1213 | cn->state); | |
1214 | } | |
1215 | cn->state = 0; | |
1216 | remove_journal_hash(sb, journal->j_list_hash_table, | |
1217 | jl, cn->blocknr, 1); | |
1218 | } | |
1219 | last = cn; | |
1220 | cn = cn->next; | |
1221 | free_cnode(sb, last); | |
1222 | } | |
1223 | jl->j_realblock = NULL; | |
1224 | } | |
1225 | ||
1226 | /* | |
1227 | * if this timestamp is greater than the timestamp we wrote last to the | |
1228 | * header block, write it to the header block. once this is done, I can | |
1229 | * safely say the log area for this transaction won't ever be replayed, | |
1230 | * and I can start releasing blocks in this transaction for reuse as data | |
1231 | * blocks. called by flush_journal_list, before it calls | |
1232 | * remove_all_from_journal_list | |
1233 | */ | |
1234 | static int _update_journal_header_block(struct super_block *sb, | |
1235 | unsigned long offset, | |
1236 | unsigned int trans_id) | |
1237 | { | |
1238 | struct reiserfs_journal_header *jh; | |
1239 | struct reiserfs_journal *journal = SB_JOURNAL(sb); | |
1240 | int depth; | |
1241 | ||
1242 | if (reiserfs_is_journal_aborted(journal)) | |
1243 | return -EIO; | |
1244 | ||
1245 | if (trans_id >= journal->j_last_flush_trans_id) { | |
1246 | if (buffer_locked((journal->j_header_bh))) { | |
1247 | depth = reiserfs_write_unlock_nested(sb); | |
1248 | __wait_on_buffer(journal->j_header_bh); | |
1249 | reiserfs_write_lock_nested(sb, depth); | |
1250 | if (unlikely(!buffer_uptodate(journal->j_header_bh))) { | |
1251 | #ifdef CONFIG_REISERFS_CHECK | |
1252 | reiserfs_warning(sb, "journal-699", | |
1253 | "buffer write failed"); | |
1254 | #endif | |
1255 | return -EIO; | |
1256 | } | |
1257 | } | |
1258 | journal->j_last_flush_trans_id = trans_id; | |
1259 | journal->j_first_unflushed_offset = offset; | |
1260 | jh = (struct reiserfs_journal_header *)(journal->j_header_bh-> | |
1261 | b_data); | |
1262 | jh->j_last_flush_trans_id = cpu_to_le32(trans_id); | |
1263 | jh->j_first_unflushed_offset = cpu_to_le32(offset); | |
1264 | jh->j_mount_id = cpu_to_le32(journal->j_mount_id); | |
1265 | ||
1266 | set_buffer_dirty(journal->j_header_bh); | |
1267 | depth = reiserfs_write_unlock_nested(sb); | |
1268 | ||
1269 | if (reiserfs_barrier_flush(sb)) | |
1270 | __sync_dirty_buffer(journal->j_header_bh, | |
1271 | REQ_SYNC | REQ_PREFLUSH | REQ_FUA); | |
1272 | else | |
1273 | sync_dirty_buffer(journal->j_header_bh); | |
1274 | ||
1275 | reiserfs_write_lock_nested(sb, depth); | |
1276 | if (!buffer_uptodate(journal->j_header_bh)) { | |
1277 | reiserfs_warning(sb, "journal-837", | |
1278 | "IO error during journal replay"); | |
1279 | return -EIO; | |
1280 | } | |
1281 | } | |
1282 | return 0; | |
1283 | } | |
1284 | ||
1285 | static int update_journal_header_block(struct super_block *sb, | |
1286 | unsigned long offset, | |
1287 | unsigned int trans_id) | |
1288 | { | |
1289 | return _update_journal_header_block(sb, offset, trans_id); | |
1290 | } | |
1291 | ||
1292 | /* | |
1293 | ** flush any and all journal lists older than you are | |
1294 | ** can only be called from flush_journal_list | |
1295 | */ | |
1296 | static int flush_older_journal_lists(struct super_block *sb, | |
1297 | struct reiserfs_journal_list *jl) | |
1298 | { | |
1299 | struct list_head *entry; | |
1300 | struct reiserfs_journal_list *other_jl; | |
1301 | struct reiserfs_journal *journal = SB_JOURNAL(sb); | |
1302 | unsigned int trans_id = jl->j_trans_id; | |
1303 | ||
1304 | /* | |
1305 | * we know we are the only ones flushing things, no extra race | |
1306 | * protection is required. | |
1307 | */ | |
1308 | restart: | |
1309 | entry = journal->j_journal_list.next; | |
1310 | /* Did we wrap? */ | |
1311 | if (entry == &journal->j_journal_list) | |
1312 | return 0; | |
1313 | other_jl = JOURNAL_LIST_ENTRY(entry); | |
1314 | if (other_jl->j_trans_id < trans_id) { | |
1315 | BUG_ON(other_jl->j_refcount <= 0); | |
1316 | /* do not flush all */ | |
1317 | flush_journal_list(sb, other_jl, 0); | |
1318 | ||
1319 | /* other_jl is now deleted from the list */ | |
1320 | goto restart; | |
1321 | } | |
1322 | return 0; | |
1323 | } | |
1324 | ||
1325 | static void del_from_work_list(struct super_block *s, | |
1326 | struct reiserfs_journal_list *jl) | |
1327 | { | |
1328 | struct reiserfs_journal *journal = SB_JOURNAL(s); | |
1329 | if (!list_empty(&jl->j_working_list)) { | |
1330 | list_del_init(&jl->j_working_list); | |
1331 | journal->j_num_work_lists--; | |
1332 | } | |
1333 | } | |
1334 | ||
1335 | /* | |
1336 | * flush a journal list, both commit and real blocks | |
1337 | * | |
1338 | * always set flushall to 1, unless you are calling from inside | |
1339 | * flush_journal_list | |
1340 | * | |
1341 | * IMPORTANT. This can only be called while there are no journal writers, | |
1342 | * and the journal is locked. That means it can only be called from | |
1343 | * do_journal_end, or by journal_release | |
1344 | */ | |
1345 | static int flush_journal_list(struct super_block *s, | |
1346 | struct reiserfs_journal_list *jl, int flushall) | |
1347 | { | |
1348 | struct reiserfs_journal_list *pjl; | |
1349 | struct reiserfs_journal_cnode *cn; | |
1350 | int count; | |
1351 | int was_jwait = 0; | |
1352 | int was_dirty = 0; | |
1353 | struct buffer_head *saved_bh; | |
1354 | unsigned long j_len_saved = jl->j_len; | |
1355 | struct reiserfs_journal *journal = SB_JOURNAL(s); | |
1356 | int err = 0; | |
1357 | int depth; | |
1358 | ||
1359 | BUG_ON(j_len_saved <= 0); | |
1360 | ||
1361 | if (atomic_read(&journal->j_wcount) != 0) { | |
1362 | reiserfs_warning(s, "clm-2048", "called with wcount %d", | |
1363 | atomic_read(&journal->j_wcount)); | |
1364 | } | |
1365 | ||
1366 | /* if flushall == 0, the lock is already held */ | |
1367 | if (flushall) { | |
1368 | reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s); | |
1369 | } else if (mutex_trylock(&journal->j_flush_mutex)) { | |
1370 | BUG(); | |
1371 | } | |
1372 | ||
1373 | count = 0; | |
1374 | if (j_len_saved > journal->j_trans_max) { | |
1375 | reiserfs_panic(s, "journal-715", "length is %lu, trans id %lu", | |
1376 | j_len_saved, jl->j_trans_id); | |
1377 | return 0; | |
1378 | } | |
1379 | ||
1380 | /* if all the work is already done, get out of here */ | |
1381 | if (atomic_read(&jl->j_nonzerolen) <= 0 && | |
1382 | atomic_read(&jl->j_commit_left) <= 0) { | |
1383 | goto flush_older_and_return; | |
1384 | } | |
1385 | ||
1386 | /* | |
1387 | * start by putting the commit list on disk. This will also flush | |
1388 | * the commit lists of any olders transactions | |
1389 | */ | |
1390 | flush_commit_list(s, jl, 1); | |
1391 | ||
1392 | if (!(jl->j_state & LIST_DIRTY) | |
1393 | && !reiserfs_is_journal_aborted(journal)) | |
1394 | BUG(); | |
1395 | ||
1396 | /* are we done now? */ | |
1397 | if (atomic_read(&jl->j_nonzerolen) <= 0 && | |
1398 | atomic_read(&jl->j_commit_left) <= 0) { | |
1399 | goto flush_older_and_return; | |
1400 | } | |
1401 | ||
1402 | /* | |
1403 | * loop through each cnode, see if we need to write it, | |
1404 | * or wait on a more recent transaction, or just ignore it | |
1405 | */ | |
1406 | if (atomic_read(&journal->j_wcount) != 0) { | |
1407 | reiserfs_panic(s, "journal-844", "journal list is flushing, " | |
1408 | "wcount is not 0"); | |
1409 | } | |
1410 | cn = jl->j_realblock; | |
1411 | while (cn) { | |
1412 | was_jwait = 0; | |
1413 | was_dirty = 0; | |
1414 | saved_bh = NULL; | |
1415 | /* blocknr of 0 is no longer in the hash, ignore it */ | |
1416 | if (cn->blocknr == 0) { | |
1417 | goto free_cnode; | |
1418 | } | |
1419 | ||
1420 | /* | |
1421 | * This transaction failed commit. | |
1422 | * Don't write out to the disk | |
1423 | */ | |
1424 | if (!(jl->j_state & LIST_DIRTY)) | |
1425 | goto free_cnode; | |
1426 | ||
1427 | pjl = find_newer_jl_for_cn(cn); | |
1428 | /* | |
1429 | * the order is important here. We check pjl to make sure we | |
1430 | * don't clear BH_JDirty_wait if we aren't the one writing this | |
1431 | * block to disk | |
1432 | */ | |
1433 | if (!pjl && cn->bh) { | |
1434 | saved_bh = cn->bh; | |
1435 | ||
1436 | /* | |
1437 | * we do this to make sure nobody releases the | |
1438 | * buffer while we are working with it | |
1439 | */ | |
1440 | get_bh(saved_bh); | |
1441 | ||
1442 | if (buffer_journal_dirty(saved_bh)) { | |
1443 | BUG_ON(!can_dirty(cn)); | |
1444 | was_jwait = 1; | |
1445 | was_dirty = 1; | |
1446 | } else if (can_dirty(cn)) { | |
1447 | /* | |
1448 | * everything with !pjl && jwait | |
1449 | * should be writable | |
1450 | */ | |
1451 | BUG(); | |
1452 | } | |
1453 | } | |
1454 | ||
1455 | /* | |
1456 | * if someone has this block in a newer transaction, just make | |
1457 | * sure they are committed, and don't try writing it to disk | |
1458 | */ | |
1459 | if (pjl) { | |
1460 | if (atomic_read(&pjl->j_commit_left)) | |
1461 | flush_commit_list(s, pjl, 1); | |
1462 | goto free_cnode; | |
1463 | } | |
1464 | ||
1465 | /* | |
1466 | * bh == NULL when the block got to disk on its own, OR, | |
1467 | * the block got freed in a future transaction | |
1468 | */ | |
1469 | if (saved_bh == NULL) { | |
1470 | goto free_cnode; | |
1471 | } | |
1472 | ||
1473 | /* | |
1474 | * this should never happen. kupdate_one_transaction has | |
1475 | * this list locked while it works, so we should never see a | |
1476 | * buffer here that is not marked JDirty_wait | |
1477 | */ | |
1478 | if ((!was_jwait) && !buffer_locked(saved_bh)) { | |
1479 | reiserfs_warning(s, "journal-813", | |
1480 | "BAD! buffer %llu %cdirty %cjwait, " | |
1481 | "not in a newer transaction", | |
1482 | (unsigned long long)saved_bh-> | |
1483 | b_blocknr, was_dirty ? ' ' : '!', | |
1484 | was_jwait ? ' ' : '!'); | |
1485 | } | |
1486 | if (was_dirty) { | |
1487 | /* | |
1488 | * we inc again because saved_bh gets decremented | |
1489 | * at free_cnode | |
1490 | */ | |
1491 | get_bh(saved_bh); | |
1492 | set_bit(BLOCK_NEEDS_FLUSH, &cn->state); | |
1493 | lock_buffer(saved_bh); | |
1494 | BUG_ON(cn->blocknr != saved_bh->b_blocknr); | |
1495 | if (buffer_dirty(saved_bh)) | |
1496 | submit_logged_buffer(saved_bh); | |
1497 | else | |
1498 | unlock_buffer(saved_bh); | |
1499 | count++; | |
1500 | } else { | |
1501 | reiserfs_warning(s, "clm-2082", | |
1502 | "Unable to flush buffer %llu in %s", | |
1503 | (unsigned long long)saved_bh-> | |
1504 | b_blocknr, __func__); | |
1505 | } | |
1506 | free_cnode: | |
1507 | cn = cn->next; | |
1508 | if (saved_bh) { | |
1509 | /* | |
1510 | * we incremented this to keep others from | |
1511 | * taking the buffer head away | |
1512 | */ | |
1513 | put_bh(saved_bh); | |
1514 | if (atomic_read(&saved_bh->b_count) < 0) { | |
1515 | reiserfs_warning(s, "journal-945", | |
1516 | "saved_bh->b_count < 0"); | |
1517 | } | |
1518 | } | |
1519 | } | |
1520 | if (count > 0) { | |
1521 | cn = jl->j_realblock; | |
1522 | while (cn) { | |
1523 | if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) { | |
1524 | if (!cn->bh) { | |
1525 | reiserfs_panic(s, "journal-1011", | |
1526 | "cn->bh is NULL"); | |
1527 | } | |
1528 | ||
1529 | depth = reiserfs_write_unlock_nested(s); | |
1530 | __wait_on_buffer(cn->bh); | |
1531 | reiserfs_write_lock_nested(s, depth); | |
1532 | ||
1533 | if (!cn->bh) { | |
1534 | reiserfs_panic(s, "journal-1012", | |
1535 | "cn->bh is NULL"); | |
1536 | } | |
1537 | if (unlikely(!buffer_uptodate(cn->bh))) { | |
1538 | #ifdef CONFIG_REISERFS_CHECK | |
1539 | reiserfs_warning(s, "journal-949", | |
1540 | "buffer write failed"); | |
1541 | #endif | |
1542 | err = -EIO; | |
1543 | } | |
1544 | /* | |
1545 | * note, we must clear the JDirty_wait bit | |
1546 | * after the up to date check, otherwise we | |
1547 | * race against our flushpage routine | |
1548 | */ | |
1549 | BUG_ON(!test_clear_buffer_journal_dirty | |
1550 | (cn->bh)); | |
1551 | ||
1552 | /* drop one ref for us */ | |
1553 | put_bh(cn->bh); | |
1554 | /* drop one ref for journal_mark_dirty */ | |
1555 | release_buffer_page(cn->bh); | |
1556 | } | |
1557 | cn = cn->next; | |
1558 | } | |
1559 | } | |
1560 | ||
1561 | if (err) | |
1562 | reiserfs_abort(s, -EIO, | |
1563 | "Write error while pushing transaction to disk in %s", | |
1564 | __func__); | |
1565 | flush_older_and_return: | |
1566 | ||
1567 | /* | |
1568 | * before we can update the journal header block, we _must_ flush all | |
1569 | * real blocks from all older transactions to disk. This is because | |
1570 | * once the header block is updated, this transaction will not be | |
1571 | * replayed after a crash | |
1572 | */ | |
1573 | if (flushall) { | |
1574 | flush_older_journal_lists(s, jl); | |
1575 | } | |
1576 | ||
1577 | err = journal->j_errno; | |
1578 | /* | |
1579 | * before we can remove everything from the hash tables for this | |
1580 | * transaction, we must make sure it can never be replayed | |
1581 | * | |
1582 | * since we are only called from do_journal_end, we know for sure there | |
1583 | * are no allocations going on while we are flushing journal lists. So, | |
1584 | * we only need to update the journal header block for the last list | |
1585 | * being flushed | |
1586 | */ | |
1587 | if (!err && flushall) { | |
1588 | err = | |
1589 | update_journal_header_block(s, | |
1590 | (jl->j_start + jl->j_len + | |
1591 | 2) % SB_ONDISK_JOURNAL_SIZE(s), | |
1592 | jl->j_trans_id); | |
1593 | if (err) | |
1594 | reiserfs_abort(s, -EIO, | |
1595 | "Write error while updating journal header in %s", | |
1596 | __func__); | |
1597 | } | |
1598 | remove_all_from_journal_list(s, jl, 0); | |
1599 | list_del_init(&jl->j_list); | |
1600 | journal->j_num_lists--; | |
1601 | del_from_work_list(s, jl); | |
1602 | ||
1603 | if (journal->j_last_flush_id != 0 && | |
1604 | (jl->j_trans_id - journal->j_last_flush_id) != 1) { | |
1605 | reiserfs_warning(s, "clm-2201", "last flush %lu, current %lu", | |
1606 | journal->j_last_flush_id, jl->j_trans_id); | |
1607 | } | |
1608 | journal->j_last_flush_id = jl->j_trans_id; | |
1609 | ||
1610 | /* | |
1611 | * not strictly required since we are freeing the list, but it should | |
1612 | * help find code using dead lists later on | |
1613 | */ | |
1614 | jl->j_len = 0; | |
1615 | atomic_set(&jl->j_nonzerolen, 0); | |
1616 | jl->j_start = 0; | |
1617 | jl->j_realblock = NULL; | |
1618 | jl->j_commit_bh = NULL; | |
1619 | jl->j_trans_id = 0; | |
1620 | jl->j_state = 0; | |
1621 | put_journal_list(s, jl); | |
1622 | if (flushall) | |
1623 | mutex_unlock(&journal->j_flush_mutex); | |
1624 | return err; | |
1625 | } | |
1626 | ||
1627 | static int write_one_transaction(struct super_block *s, | |
1628 | struct reiserfs_journal_list *jl, | |
1629 | struct buffer_chunk *chunk) | |
1630 | { | |
1631 | struct reiserfs_journal_cnode *cn; | |
1632 | int ret = 0; | |
1633 | ||
1634 | jl->j_state |= LIST_TOUCHED; | |
1635 | del_from_work_list(s, jl); | |
1636 | if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) { | |
1637 | return 0; | |
1638 | } | |
1639 | ||
1640 | cn = jl->j_realblock; | |
1641 | while (cn) { | |
1642 | /* | |
1643 | * if the blocknr == 0, this has been cleared from the hash, | |
1644 | * skip it | |
1645 | */ | |
1646 | if (cn->blocknr == 0) { | |
1647 | goto next; | |
1648 | } | |
1649 | if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) { | |
1650 | struct buffer_head *tmp_bh; | |
1651 | /* | |
1652 | * we can race against journal_mark_freed when we try | |
1653 | * to lock_buffer(cn->bh), so we have to inc the buffer | |
1654 | * count, and recheck things after locking | |
1655 | */ | |
1656 | tmp_bh = cn->bh; | |
1657 | get_bh(tmp_bh); | |
1658 | lock_buffer(tmp_bh); | |
1659 | if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) { | |
1660 | if (!buffer_journal_dirty(tmp_bh) || | |
1661 | buffer_journal_prepared(tmp_bh)) | |
1662 | BUG(); | |
1663 | add_to_chunk(chunk, tmp_bh, NULL, write_chunk); | |
1664 | ret++; | |
1665 | } else { | |
1666 | /* note, cn->bh might be null now */ | |
1667 | unlock_buffer(tmp_bh); | |
1668 | } | |
1669 | put_bh(tmp_bh); | |
1670 | } | |
1671 | next: | |
1672 | cn = cn->next; | |
1673 | cond_resched(); | |
1674 | } | |
1675 | return ret; | |
1676 | } | |
1677 | ||
1678 | /* used by flush_commit_list */ | |
1679 | static void dirty_one_transaction(struct super_block *s, | |
1680 | struct reiserfs_journal_list *jl) | |
1681 | { | |
1682 | struct reiserfs_journal_cnode *cn; | |
1683 | struct reiserfs_journal_list *pjl; | |
1684 | ||
1685 | jl->j_state |= LIST_DIRTY; | |
1686 | cn = jl->j_realblock; | |
1687 | while (cn) { | |
1688 | /* | |
1689 | * look for a more recent transaction that logged this | |
1690 | * buffer. Only the most recent transaction with a buffer in | |
1691 | * it is allowed to send that buffer to disk | |
1692 | */ | |
1693 | pjl = find_newer_jl_for_cn(cn); | |
1694 | if (!pjl && cn->blocknr && cn->bh | |
1695 | && buffer_journal_dirty(cn->bh)) { | |
1696 | BUG_ON(!can_dirty(cn)); | |
1697 | /* | |
1698 | * if the buffer is prepared, it will either be logged | |
1699 | * or restored. If restored, we need to make sure | |
1700 | * it actually gets marked dirty | |
1701 | */ | |
1702 | clear_buffer_journal_new(cn->bh); | |
1703 | if (buffer_journal_prepared(cn->bh)) { | |
1704 | set_buffer_journal_restore_dirty(cn->bh); | |
1705 | } else { | |
1706 | set_buffer_journal_test(cn->bh); | |
1707 | mark_buffer_dirty(cn->bh); | |
1708 | } | |
1709 | } | |
1710 | cn = cn->next; | |
1711 | } | |
1712 | } | |
1713 | ||
1714 | static int kupdate_transactions(struct super_block *s, | |
1715 | struct reiserfs_journal_list *jl, | |
1716 | struct reiserfs_journal_list **next_jl, | |
1717 | unsigned int *next_trans_id, | |
1718 | int num_blocks, int num_trans) | |
1719 | { | |
1720 | int ret = 0; | |
1721 | int written = 0; | |
1722 | int transactions_flushed = 0; | |
1723 | unsigned int orig_trans_id = jl->j_trans_id; | |
1724 | struct buffer_chunk chunk; | |
1725 | struct list_head *entry; | |
1726 | struct reiserfs_journal *journal = SB_JOURNAL(s); | |
1727 | chunk.nr = 0; | |
1728 | ||
1729 | reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s); | |
1730 | if (!journal_list_still_alive(s, orig_trans_id)) { | |
1731 | goto done; | |
1732 | } | |
1733 | ||
1734 | /* | |
1735 | * we've got j_flush_mutex held, nobody is going to delete any | |
1736 | * of these lists out from underneath us | |
1737 | */ | |
1738 | while ((num_trans && transactions_flushed < num_trans) || | |
1739 | (!num_trans && written < num_blocks)) { | |
1740 | ||
1741 | if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) || | |
1742 | atomic_read(&jl->j_commit_left) | |
1743 | || !(jl->j_state & LIST_DIRTY)) { | |
1744 | del_from_work_list(s, jl); | |
1745 | break; | |
1746 | } | |
1747 | ret = write_one_transaction(s, jl, &chunk); | |
1748 | ||
1749 | if (ret < 0) | |
1750 | goto done; | |
1751 | transactions_flushed++; | |
1752 | written += ret; | |
1753 | entry = jl->j_list.next; | |
1754 | ||
1755 | /* did we wrap? */ | |
1756 | if (entry == &journal->j_journal_list) { | |
1757 | break; | |
1758 | } | |
1759 | jl = JOURNAL_LIST_ENTRY(entry); | |
1760 | ||
1761 | /* don't bother with older transactions */ | |
1762 | if (jl->j_trans_id <= orig_trans_id) | |
1763 | break; | |
1764 | } | |
1765 | if (chunk.nr) { | |
1766 | write_chunk(&chunk); | |
1767 | } | |
1768 | ||
1769 | done: | |
1770 | mutex_unlock(&journal->j_flush_mutex); | |
1771 | return ret; | |
1772 | } | |
1773 | ||
1774 | /* | |
1775 | * for o_sync and fsync heavy applications, they tend to use | |
1776 | * all the journa list slots with tiny transactions. These | |
1777 | * trigger lots and lots of calls to update the header block, which | |
1778 | * adds seeks and slows things down. | |
1779 | * | |
1780 | * This function tries to clear out a large chunk of the journal lists | |
1781 | * at once, which makes everything faster since only the newest journal | |
1782 | * list updates the header block | |
1783 | */ | |
1784 | static int flush_used_journal_lists(struct super_block *s, | |
1785 | struct reiserfs_journal_list *jl) | |
1786 | { | |
1787 | unsigned long len = 0; | |
1788 | unsigned long cur_len; | |
1789 | int i; | |
1790 | int limit = 256; | |
1791 | struct reiserfs_journal_list *tjl; | |
1792 | struct reiserfs_journal_list *flush_jl; | |
1793 | unsigned int trans_id; | |
1794 | struct reiserfs_journal *journal = SB_JOURNAL(s); | |
1795 | ||
1796 | flush_jl = tjl = jl; | |
1797 | ||
1798 | /* in data logging mode, try harder to flush a lot of blocks */ | |
1799 | if (reiserfs_data_log(s)) | |
1800 | limit = 1024; | |
1801 | /* flush for 256 transactions or limit blocks, whichever comes first */ | |
1802 | for (i = 0; i < 256 && len < limit; i++) { | |
1803 | if (atomic_read(&tjl->j_commit_left) || | |
1804 | tjl->j_trans_id < jl->j_trans_id) { | |
1805 | break; | |
1806 | } | |
1807 | cur_len = atomic_read(&tjl->j_nonzerolen); | |
1808 | if (cur_len > 0) { | |
1809 | tjl->j_state &= ~LIST_TOUCHED; | |
1810 | } | |
1811 | len += cur_len; | |
1812 | flush_jl = tjl; | |
1813 | if (tjl->j_list.next == &journal->j_journal_list) | |
1814 | break; | |
1815 | tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next); | |
1816 | } | |
1817 | get_journal_list(jl); | |
1818 | get_journal_list(flush_jl); | |
1819 | ||
1820 | /* | |
1821 | * try to find a group of blocks we can flush across all the | |
1822 | * transactions, but only bother if we've actually spanned | |
1823 | * across multiple lists | |
1824 | */ | |
1825 | if (flush_jl != jl) | |
1826 | kupdate_transactions(s, jl, &tjl, &trans_id, len, i); | |
1827 | ||
1828 | flush_journal_list(s, flush_jl, 1); | |
1829 | put_journal_list(s, flush_jl); | |
1830 | put_journal_list(s, jl); | |
1831 | return 0; | |
1832 | } | |
1833 | ||
1834 | /* | |
1835 | * removes any nodes in table with name block and dev as bh. | |
1836 | * only touchs the hnext and hprev pointers. | |
1837 | */ | |
1838 | static void remove_journal_hash(struct super_block *sb, | |
1839 | struct reiserfs_journal_cnode **table, | |
1840 | struct reiserfs_journal_list *jl, | |
1841 | unsigned long block, int remove_freed) | |
1842 | { | |
1843 | struct reiserfs_journal_cnode *cur; | |
1844 | struct reiserfs_journal_cnode **head; | |
1845 | ||
1846 | head = &(journal_hash(table, sb, block)); | |
1847 | if (!head) { | |
1848 | return; | |
1849 | } | |
1850 | cur = *head; | |
1851 | while (cur) { | |
1852 | if (cur->blocknr == block && cur->sb == sb | |
1853 | && (jl == NULL || jl == cur->jlist) | |
1854 | && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) { | |
1855 | if (cur->hnext) { | |
1856 | cur->hnext->hprev = cur->hprev; | |
1857 | } | |
1858 | if (cur->hprev) { | |
1859 | cur->hprev->hnext = cur->hnext; | |
1860 | } else { | |
1861 | *head = cur->hnext; | |
1862 | } | |
1863 | cur->blocknr = 0; | |
1864 | cur->sb = NULL; | |
1865 | cur->state = 0; | |
1866 | /* | |
1867 | * anybody who clears the cur->bh will also | |
1868 | * dec the nonzerolen | |
1869 | */ | |
1870 | if (cur->bh && cur->jlist) | |
1871 | atomic_dec(&cur->jlist->j_nonzerolen); | |
1872 | cur->bh = NULL; | |
1873 | cur->jlist = NULL; | |
1874 | } | |
1875 | cur = cur->hnext; | |
1876 | } | |
1877 | } | |
1878 | ||
1879 | static void free_journal_ram(struct super_block *sb) | |
1880 | { | |
1881 | struct reiserfs_journal *journal = SB_JOURNAL(sb); | |
1882 | kfree(journal->j_current_jl); | |
1883 | journal->j_num_lists--; | |
1884 | ||
1885 | vfree(journal->j_cnode_free_orig); | |
1886 | free_list_bitmaps(sb, journal->j_list_bitmap); | |
1887 | free_bitmap_nodes(sb); /* must be after free_list_bitmaps */ | |
1888 | if (journal->j_header_bh) { | |
1889 | brelse(journal->j_header_bh); | |
1890 | } | |
1891 | /* | |
1892 | * j_header_bh is on the journal dev, make sure | |
1893 | * not to release the journal dev until we brelse j_header_bh | |
1894 | */ | |
1895 | release_journal_dev(sb, journal); | |
1896 | vfree(journal); | |
1897 | } | |
1898 | ||
1899 | /* | |
1900 | * call on unmount. Only set error to 1 if you haven't made your way out | |
1901 | * of read_super() yet. Any other caller must keep error at 0. | |
1902 | */ | |
1903 | static int do_journal_release(struct reiserfs_transaction_handle *th, | |
1904 | struct super_block *sb, int error) | |
1905 | { | |
1906 | struct reiserfs_transaction_handle myth; | |
1907 | struct reiserfs_journal *journal = SB_JOURNAL(sb); | |
1908 | ||
1909 | /* | |
1910 | * we only want to flush out transactions if we were | |
1911 | * called with error == 0 | |
1912 | */ | |
1913 | if (!error && !sb_rdonly(sb)) { | |
1914 | /* end the current trans */ | |
1915 | BUG_ON(!th->t_trans_id); | |
1916 | do_journal_end(th, FLUSH_ALL); | |
1917 | ||
1918 | /* | |
1919 | * make sure something gets logged to force | |
1920 | * our way into the flush code | |
1921 | */ | |
1922 | if (!journal_join(&myth, sb)) { | |
1923 | reiserfs_prepare_for_journal(sb, | |
1924 | SB_BUFFER_WITH_SB(sb), | |
1925 | 1); | |
1926 | journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb)); | |
1927 | do_journal_end(&myth, FLUSH_ALL); | |
1928 | } | |
1929 | } | |
1930 | ||
1931 | /* this also catches errors during the do_journal_end above */ | |
1932 | if (!error && reiserfs_is_journal_aborted(journal)) { | |
1933 | memset(&myth, 0, sizeof(myth)); | |
1934 | if (!journal_join_abort(&myth, sb)) { | |
1935 | reiserfs_prepare_for_journal(sb, | |
1936 | SB_BUFFER_WITH_SB(sb), | |
1937 | 1); | |
1938 | journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb)); | |
1939 | do_journal_end(&myth, FLUSH_ALL); | |
1940 | } | |
1941 | } | |
1942 | ||
1943 | ||
1944 | /* | |
1945 | * We must release the write lock here because | |
1946 | * the workqueue job (flush_async_commit) needs this lock | |
1947 | */ | |
1948 | reiserfs_write_unlock(sb); | |
1949 | ||
1950 | /* | |
1951 | * Cancel flushing of old commits. Note that neither of these works | |
1952 | * will be requeued because superblock is being shutdown and doesn't | |
1953 | * have SB_ACTIVE set. | |
1954 | */ | |
1955 | reiserfs_cancel_old_flush(sb); | |
1956 | /* wait for all commits to finish */ | |
1957 | cancel_delayed_work_sync(&SB_JOURNAL(sb)->j_work); | |
1958 | ||
1959 | free_journal_ram(sb); | |
1960 | ||
1961 | reiserfs_write_lock(sb); | |
1962 | ||
1963 | return 0; | |
1964 | } | |
1965 | ||
1966 | /* * call on unmount. flush all journal trans, release all alloc'd ram */ | |
1967 | int journal_release(struct reiserfs_transaction_handle *th, | |
1968 | struct super_block *sb) | |
1969 | { | |
1970 | return do_journal_release(th, sb, 0); | |
1971 | } | |
1972 | ||
1973 | /* only call from an error condition inside reiserfs_read_super! */ | |
1974 | int journal_release_error(struct reiserfs_transaction_handle *th, | |
1975 | struct super_block *sb) | |
1976 | { | |
1977 | return do_journal_release(th, sb, 1); | |
1978 | } | |
1979 | ||
1980 | /* | |
1981 | * compares description block with commit block. | |
1982 | * returns 1 if they differ, 0 if they are the same | |
1983 | */ | |
1984 | static int journal_compare_desc_commit(struct super_block *sb, | |
1985 | struct reiserfs_journal_desc *desc, | |
1986 | struct reiserfs_journal_commit *commit) | |
1987 | { | |
1988 | if (get_commit_trans_id(commit) != get_desc_trans_id(desc) || | |
1989 | get_commit_trans_len(commit) != get_desc_trans_len(desc) || | |
1990 | get_commit_trans_len(commit) > SB_JOURNAL(sb)->j_trans_max || | |
1991 | get_commit_trans_len(commit) <= 0) { | |
1992 | return 1; | |
1993 | } | |
1994 | return 0; | |
1995 | } | |
1996 | ||
1997 | /* | |
1998 | * returns 0 if it did not find a description block | |
1999 | * returns -1 if it found a corrupt commit block | |
2000 | * returns 1 if both desc and commit were valid | |
2001 | * NOTE: only called during fs mount | |
2002 | */ | |
2003 | static int journal_transaction_is_valid(struct super_block *sb, | |
2004 | struct buffer_head *d_bh, | |
2005 | unsigned int *oldest_invalid_trans_id, | |
2006 | unsigned long *newest_mount_id) | |
2007 | { | |
2008 | struct reiserfs_journal_desc *desc; | |
2009 | struct reiserfs_journal_commit *commit; | |
2010 | struct buffer_head *c_bh; | |
2011 | unsigned long offset; | |
2012 | ||
2013 | if (!d_bh) | |
2014 | return 0; | |
2015 | ||
2016 | desc = (struct reiserfs_journal_desc *)d_bh->b_data; | |
2017 | if (get_desc_trans_len(desc) > 0 | |
2018 | && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) { | |
2019 | if (oldest_invalid_trans_id && *oldest_invalid_trans_id | |
2020 | && get_desc_trans_id(desc) > *oldest_invalid_trans_id) { | |
2021 | reiserfs_debug(sb, REISERFS_DEBUG_CODE, | |
2022 | "journal-986: transaction " | |
2023 | "is valid returning because trans_id %d is greater than " | |
2024 | "oldest_invalid %lu", | |
2025 | get_desc_trans_id(desc), | |
2026 | *oldest_invalid_trans_id); | |
2027 | return 0; | |
2028 | } | |
2029 | if (newest_mount_id | |
2030 | && *newest_mount_id > get_desc_mount_id(desc)) { | |
2031 | reiserfs_debug(sb, REISERFS_DEBUG_CODE, | |
2032 | "journal-1087: transaction " | |
2033 | "is valid returning because mount_id %d is less than " | |
2034 | "newest_mount_id %lu", | |
2035 | get_desc_mount_id(desc), | |
2036 | *newest_mount_id); | |
2037 | return -1; | |
2038 | } | |
2039 | if (get_desc_trans_len(desc) > SB_JOURNAL(sb)->j_trans_max) { | |
2040 | reiserfs_warning(sb, "journal-2018", | |
2041 | "Bad transaction length %d " | |
2042 | "encountered, ignoring transaction", | |
2043 | get_desc_trans_len(desc)); | |
2044 | return -1; | |
2045 | } | |
2046 | offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb); | |
2047 | ||
2048 | /* | |
2049 | * ok, we have a journal description block, | |
2050 | * let's see if the transaction was valid | |
2051 | */ | |
2052 | c_bh = | |
2053 | journal_bread(sb, | |
2054 | SB_ONDISK_JOURNAL_1st_BLOCK(sb) + | |
2055 | ((offset + get_desc_trans_len(desc) + | |
2056 | 1) % SB_ONDISK_JOURNAL_SIZE(sb))); | |
2057 | if (!c_bh) | |
2058 | return 0; | |
2059 | commit = (struct reiserfs_journal_commit *)c_bh->b_data; | |
2060 | if (journal_compare_desc_commit(sb, desc, commit)) { | |
2061 | reiserfs_debug(sb, REISERFS_DEBUG_CODE, | |
2062 | "journal_transaction_is_valid, commit offset %ld had bad " | |
2063 | "time %d or length %d", | |
2064 | c_bh->b_blocknr - | |
2065 | SB_ONDISK_JOURNAL_1st_BLOCK(sb), | |
2066 | get_commit_trans_id(commit), | |
2067 | get_commit_trans_len(commit)); | |
2068 | brelse(c_bh); | |
2069 | if (oldest_invalid_trans_id) { | |
2070 | *oldest_invalid_trans_id = | |
2071 | get_desc_trans_id(desc); | |
2072 | reiserfs_debug(sb, REISERFS_DEBUG_CODE, | |
2073 | "journal-1004: " | |
2074 | "transaction_is_valid setting oldest invalid trans_id " | |
2075 | "to %d", | |
2076 | get_desc_trans_id(desc)); | |
2077 | } | |
2078 | return -1; | |
2079 | } | |
2080 | brelse(c_bh); | |
2081 | reiserfs_debug(sb, REISERFS_DEBUG_CODE, | |
2082 | "journal-1006: found valid " | |
2083 | "transaction start offset %llu, len %d id %d", | |
2084 | d_bh->b_blocknr - | |
2085 | SB_ONDISK_JOURNAL_1st_BLOCK(sb), | |
2086 | get_desc_trans_len(desc), | |
2087 | get_desc_trans_id(desc)); | |
2088 | return 1; | |
2089 | } else { | |
2090 | return 0; | |
2091 | } | |
2092 | } | |
2093 | ||
2094 | static void brelse_array(struct buffer_head **heads, int num) | |
2095 | { | |
2096 | int i; | |
2097 | for (i = 0; i < num; i++) { | |
2098 | brelse(heads[i]); | |
2099 | } | |
2100 | } | |
2101 | ||
2102 | /* | |
2103 | * given the start, and values for the oldest acceptable transactions, | |
2104 | * this either reads in a replays a transaction, or returns because the | |
2105 | * transaction is invalid, or too old. | |
2106 | * NOTE: only called during fs mount | |
2107 | */ | |
2108 | static int journal_read_transaction(struct super_block *sb, | |
2109 | unsigned long cur_dblock, | |
2110 | unsigned long oldest_start, | |
2111 | unsigned int oldest_trans_id, | |
2112 | unsigned long newest_mount_id) | |
2113 | { | |
2114 | struct reiserfs_journal *journal = SB_JOURNAL(sb); | |
2115 | struct reiserfs_journal_desc *desc; | |
2116 | struct reiserfs_journal_commit *commit; | |
2117 | unsigned int trans_id = 0; | |
2118 | struct buffer_head *c_bh; | |
2119 | struct buffer_head *d_bh; | |
2120 | struct buffer_head **log_blocks = NULL; | |
2121 | struct buffer_head **real_blocks = NULL; | |
2122 | unsigned int trans_offset; | |
2123 | int i; | |
2124 | int trans_half; | |
2125 | ||
2126 | d_bh = journal_bread(sb, cur_dblock); | |
2127 | if (!d_bh) | |
2128 | return 1; | |
2129 | desc = (struct reiserfs_journal_desc *)d_bh->b_data; | |
2130 | trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb); | |
2131 | reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1037: " | |
2132 | "journal_read_transaction, offset %llu, len %d mount_id %d", | |
2133 | d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb), | |
2134 | get_desc_trans_len(desc), get_desc_mount_id(desc)); | |
2135 | if (get_desc_trans_id(desc) < oldest_trans_id) { | |
2136 | reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1039: " | |
2137 | "journal_read_trans skipping because %lu is too old", | |
2138 | cur_dblock - | |
2139 | SB_ONDISK_JOURNAL_1st_BLOCK(sb)); | |
2140 | brelse(d_bh); | |
2141 | return 1; | |
2142 | } | |
2143 | if (get_desc_mount_id(desc) != newest_mount_id) { | |
2144 | reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1146: " | |
2145 | "journal_read_trans skipping because %d is != " | |
2146 | "newest_mount_id %lu", get_desc_mount_id(desc), | |
2147 | newest_mount_id); | |
2148 | brelse(d_bh); | |
2149 | return 1; | |
2150 | } | |
2151 | c_bh = journal_bread(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) + | |
2152 | ((trans_offset + get_desc_trans_len(desc) + 1) % | |
2153 | SB_ONDISK_JOURNAL_SIZE(sb))); | |
2154 | if (!c_bh) { | |
2155 | brelse(d_bh); | |
2156 | return 1; | |
2157 | } | |
2158 | commit = (struct reiserfs_journal_commit *)c_bh->b_data; | |
2159 | if (journal_compare_desc_commit(sb, desc, commit)) { | |
2160 | reiserfs_debug(sb, REISERFS_DEBUG_CODE, | |
2161 | "journal_read_transaction, " | |
2162 | "commit offset %llu had bad time %d or length %d", | |
2163 | c_bh->b_blocknr - | |
2164 | SB_ONDISK_JOURNAL_1st_BLOCK(sb), | |
2165 | get_commit_trans_id(commit), | |
2166 | get_commit_trans_len(commit)); | |
2167 | brelse(c_bh); | |
2168 | brelse(d_bh); | |
2169 | return 1; | |
2170 | } | |
2171 | ||
2172 | if (bdev_read_only(sb->s_bdev)) { | |
2173 | reiserfs_warning(sb, "clm-2076", | |
2174 | "device is readonly, unable to replay log"); | |
2175 | brelse(c_bh); | |
2176 | brelse(d_bh); | |
2177 | return -EROFS; | |
2178 | } | |
2179 | ||
2180 | trans_id = get_desc_trans_id(desc); | |
2181 | /* | |
2182 | * now we know we've got a good transaction, and it was | |
2183 | * inside the valid time ranges | |
2184 | */ | |
2185 | log_blocks = kmalloc_array(get_desc_trans_len(desc), | |
2186 | sizeof(struct buffer_head *), | |
2187 | GFP_NOFS); | |
2188 | real_blocks = kmalloc_array(get_desc_trans_len(desc), | |
2189 | sizeof(struct buffer_head *), | |
2190 | GFP_NOFS); | |
2191 | if (!log_blocks || !real_blocks) { | |
2192 | brelse(c_bh); | |
2193 | brelse(d_bh); | |
2194 | kfree(log_blocks); | |
2195 | kfree(real_blocks); | |
2196 | reiserfs_warning(sb, "journal-1169", | |
2197 | "kmalloc failed, unable to mount FS"); | |
2198 | return -1; | |
2199 | } | |
2200 | /* get all the buffer heads */ | |
2201 | trans_half = journal_trans_half(sb->s_blocksize); | |
2202 | for (i = 0; i < get_desc_trans_len(desc); i++) { | |
2203 | log_blocks[i] = | |
2204 | journal_getblk(sb, | |
2205 | SB_ONDISK_JOURNAL_1st_BLOCK(sb) + | |
2206 | (trans_offset + 1 + | |
2207 | i) % SB_ONDISK_JOURNAL_SIZE(sb)); | |
2208 | if (i < trans_half) { | |
2209 | real_blocks[i] = | |
2210 | sb_getblk(sb, | |
2211 | le32_to_cpu(desc->j_realblock[i])); | |
2212 | } else { | |
2213 | real_blocks[i] = | |
2214 | sb_getblk(sb, | |
2215 | le32_to_cpu(commit-> | |
2216 | j_realblock[i - trans_half])); | |
2217 | } | |
2218 | if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(sb)) { | |
2219 | reiserfs_warning(sb, "journal-1207", | |
2220 | "REPLAY FAILURE fsck required! " | |
2221 | "Block to replay is outside of " | |
2222 | "filesystem"); | |
2223 | goto abort_replay; | |
2224 | } | |
2225 | /* make sure we don't try to replay onto log or reserved area */ | |
2226 | if (is_block_in_log_or_reserved_area | |
2227 | (sb, real_blocks[i]->b_blocknr)) { | |
2228 | reiserfs_warning(sb, "journal-1204", | |
2229 | "REPLAY FAILURE fsck required! " | |
2230 | "Trying to replay onto a log block"); | |
2231 | abort_replay: | |
2232 | brelse_array(log_blocks, i); | |
2233 | brelse_array(real_blocks, i); | |
2234 | brelse(c_bh); | |
2235 | brelse(d_bh); | |
2236 | kfree(log_blocks); | |
2237 | kfree(real_blocks); | |
2238 | return -1; | |
2239 | } | |
2240 | } | |
2241 | /* read in the log blocks, memcpy to the corresponding real block */ | |
2242 | ll_rw_block(REQ_OP_READ, 0, get_desc_trans_len(desc), log_blocks); | |
2243 | for (i = 0; i < get_desc_trans_len(desc); i++) { | |
2244 | ||
2245 | wait_on_buffer(log_blocks[i]); | |
2246 | if (!buffer_uptodate(log_blocks[i])) { | |
2247 | reiserfs_warning(sb, "journal-1212", | |
2248 | "REPLAY FAILURE fsck required! " | |
2249 | "buffer write failed"); | |
2250 | brelse_array(log_blocks + i, | |
2251 | get_desc_trans_len(desc) - i); | |
2252 | brelse_array(real_blocks, get_desc_trans_len(desc)); | |
2253 | brelse(c_bh); | |
2254 | brelse(d_bh); | |
2255 | kfree(log_blocks); | |
2256 | kfree(real_blocks); | |
2257 | return -1; | |
2258 | } | |
2259 | memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data, | |
2260 | real_blocks[i]->b_size); | |
2261 | set_buffer_uptodate(real_blocks[i]); | |
2262 | brelse(log_blocks[i]); | |
2263 | } | |
2264 | /* flush out the real blocks */ | |
2265 | for (i = 0; i < get_desc_trans_len(desc); i++) { | |
2266 | set_buffer_dirty(real_blocks[i]); | |
2267 | write_dirty_buffer(real_blocks[i], 0); | |
2268 | } | |
2269 | for (i = 0; i < get_desc_trans_len(desc); i++) { | |
2270 | wait_on_buffer(real_blocks[i]); | |
2271 | if (!buffer_uptodate(real_blocks[i])) { | |
2272 | reiserfs_warning(sb, "journal-1226", | |
2273 | "REPLAY FAILURE, fsck required! " | |
2274 | "buffer write failed"); | |
2275 | brelse_array(real_blocks + i, | |
2276 | get_desc_trans_len(desc) - i); | |
2277 | brelse(c_bh); | |
2278 | brelse(d_bh); | |
2279 | kfree(log_blocks); | |
2280 | kfree(real_blocks); | |
2281 | return -1; | |
2282 | } | |
2283 | brelse(real_blocks[i]); | |
2284 | } | |
2285 | cur_dblock = | |
2286 | SB_ONDISK_JOURNAL_1st_BLOCK(sb) + | |
2287 | ((trans_offset + get_desc_trans_len(desc) + | |
2288 | 2) % SB_ONDISK_JOURNAL_SIZE(sb)); | |
2289 | reiserfs_debug(sb, REISERFS_DEBUG_CODE, | |
2290 | "journal-1095: setting journal " "start to offset %ld", | |
2291 | cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb)); | |
2292 | ||
2293 | /* | |
2294 | * init starting values for the first transaction, in case | |
2295 | * this is the last transaction to be replayed. | |
2296 | */ | |
2297 | journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb); | |
2298 | journal->j_last_flush_trans_id = trans_id; | |
2299 | journal->j_trans_id = trans_id + 1; | |
2300 | /* check for trans_id overflow */ | |
2301 | if (journal->j_trans_id == 0) | |
2302 | journal->j_trans_id = 10; | |
2303 | brelse(c_bh); | |
2304 | brelse(d_bh); | |
2305 | kfree(log_blocks); | |
2306 | kfree(real_blocks); | |
2307 | return 0; | |
2308 | } | |
2309 | ||
2310 | /* | |
2311 | * This function reads blocks starting from block and to max_block of bufsize | |
2312 | * size (but no more than BUFNR blocks at a time). This proved to improve | |
2313 | * mounting speed on self-rebuilding raid5 arrays at least. | |
2314 | * Right now it is only used from journal code. But later we might use it | |
2315 | * from other places. | |
2316 | * Note: Do not use journal_getblk/sb_getblk functions here! | |
2317 | */ | |
2318 | static struct buffer_head *reiserfs_breada(struct block_device *dev, | |
2319 | b_blocknr_t block, int bufsize, | |
2320 | b_blocknr_t max_block) | |
2321 | { | |
2322 | struct buffer_head *bhlist[BUFNR]; | |
2323 | unsigned int blocks = BUFNR; | |
2324 | struct buffer_head *bh; | |
2325 | int i, j; | |
2326 | ||
2327 | bh = __getblk(dev, block, bufsize); | |
2328 | if (buffer_uptodate(bh)) | |
2329 | return (bh); | |
2330 | ||
2331 | if (block + BUFNR > max_block) { | |
2332 | blocks = max_block - block; | |
2333 | } | |
2334 | bhlist[0] = bh; | |
2335 | j = 1; | |
2336 | for (i = 1; i < blocks; i++) { | |
2337 | bh = __getblk(dev, block + i, bufsize); | |
2338 | if (buffer_uptodate(bh)) { | |
2339 | brelse(bh); | |
2340 | break; | |
2341 | } else | |
2342 | bhlist[j++] = bh; | |
2343 | } | |
2344 | ll_rw_block(REQ_OP_READ, 0, j, bhlist); | |
2345 | for (i = 1; i < j; i++) | |
2346 | brelse(bhlist[i]); | |
2347 | bh = bhlist[0]; | |
2348 | wait_on_buffer(bh); | |
2349 | if (buffer_uptodate(bh)) | |
2350 | return bh; | |
2351 | brelse(bh); | |
2352 | return NULL; | |
2353 | } | |
2354 | ||
2355 | /* | |
2356 | * read and replay the log | |
2357 | * on a clean unmount, the journal header's next unflushed pointer will be | |
2358 | * to an invalid transaction. This tests that before finding all the | |
2359 | * transactions in the log, which makes normal mount times fast. | |
2360 | * | |
2361 | * After a crash, this starts with the next unflushed transaction, and | |
2362 | * replays until it finds one too old, or invalid. | |
2363 | * | |
2364 | * On exit, it sets things up so the first transaction will work correctly. | |
2365 | * NOTE: only called during fs mount | |
2366 | */ | |
2367 | static int journal_read(struct super_block *sb) | |
2368 | { | |
2369 | struct reiserfs_journal *journal = SB_JOURNAL(sb); | |
2370 | struct reiserfs_journal_desc *desc; | |
2371 | unsigned int oldest_trans_id = 0; | |
2372 | unsigned int oldest_invalid_trans_id = 0; | |
2373 | time64_t start; | |
2374 | unsigned long oldest_start = 0; | |
2375 | unsigned long cur_dblock = 0; | |
2376 | unsigned long newest_mount_id = 9; | |
2377 | struct buffer_head *d_bh; | |
2378 | struct reiserfs_journal_header *jh; | |
2379 | int valid_journal_header = 0; | |
2380 | int replay_count = 0; | |
2381 | int continue_replay = 1; | |
2382 | int ret; | |
2383 | ||
2384 | cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(sb); | |
2385 | reiserfs_info(sb, "checking transaction log (%pg)\n", | |
2386 | journal->j_dev_bd); | |
2387 | start = ktime_get_seconds(); | |
2388 | ||
2389 | /* | |
2390 | * step 1, read in the journal header block. Check the transaction | |
2391 | * it says is the first unflushed, and if that transaction is not | |
2392 | * valid, replay is done | |
2393 | */ | |
2394 | journal->j_header_bh = journal_bread(sb, | |
2395 | SB_ONDISK_JOURNAL_1st_BLOCK(sb) | |
2396 | + SB_ONDISK_JOURNAL_SIZE(sb)); | |
2397 | if (!journal->j_header_bh) { | |
2398 | return 1; | |
2399 | } | |
2400 | jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data); | |
2401 | if (le32_to_cpu(jh->j_first_unflushed_offset) < | |
2402 | SB_ONDISK_JOURNAL_SIZE(sb) | |
2403 | && le32_to_cpu(jh->j_last_flush_trans_id) > 0) { | |
2404 | oldest_start = | |
2405 | SB_ONDISK_JOURNAL_1st_BLOCK(sb) + | |
2406 | le32_to_cpu(jh->j_first_unflushed_offset); | |
2407 | oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1; | |
2408 | newest_mount_id = le32_to_cpu(jh->j_mount_id); | |
2409 | reiserfs_debug(sb, REISERFS_DEBUG_CODE, | |
2410 | "journal-1153: found in " | |
2411 | "header: first_unflushed_offset %d, last_flushed_trans_id " | |
2412 | "%lu", le32_to_cpu(jh->j_first_unflushed_offset), | |
2413 | le32_to_cpu(jh->j_last_flush_trans_id)); | |
2414 | valid_journal_header = 1; | |
2415 | ||
2416 | /* | |
2417 | * now, we try to read the first unflushed offset. If it | |
2418 | * is not valid, there is nothing more we can do, and it | |
2419 | * makes no sense to read through the whole log. | |
2420 | */ | |
2421 | d_bh = | |
2422 | journal_bread(sb, | |
2423 | SB_ONDISK_JOURNAL_1st_BLOCK(sb) + | |
2424 | le32_to_cpu(jh->j_first_unflushed_offset)); | |
2425 | ret = journal_transaction_is_valid(sb, d_bh, NULL, NULL); | |
2426 | if (!ret) { | |
2427 | continue_replay = 0; | |
2428 | } | |
2429 | brelse(d_bh); | |
2430 | goto start_log_replay; | |
2431 | } | |
2432 | ||
2433 | /* | |
2434 | * ok, there are transactions that need to be replayed. start | |
2435 | * with the first log block, find all the valid transactions, and | |
2436 | * pick out the oldest. | |
2437 | */ | |
2438 | while (continue_replay | |
2439 | && cur_dblock < | |
2440 | (SB_ONDISK_JOURNAL_1st_BLOCK(sb) + | |
2441 | SB_ONDISK_JOURNAL_SIZE(sb))) { | |
2442 | /* | |
2443 | * Note that it is required for blocksize of primary fs | |
2444 | * device and journal device to be the same | |
2445 | */ | |
2446 | d_bh = | |
2447 | reiserfs_breada(journal->j_dev_bd, cur_dblock, | |
2448 | sb->s_blocksize, | |
2449 | SB_ONDISK_JOURNAL_1st_BLOCK(sb) + | |
2450 | SB_ONDISK_JOURNAL_SIZE(sb)); | |
2451 | ret = | |
2452 | journal_transaction_is_valid(sb, d_bh, | |
2453 | &oldest_invalid_trans_id, | |
2454 | &newest_mount_id); | |
2455 | if (ret == 1) { | |
2456 | desc = (struct reiserfs_journal_desc *)d_bh->b_data; | |
2457 | if (oldest_start == 0) { /* init all oldest_ values */ | |
2458 | oldest_trans_id = get_desc_trans_id(desc); | |
2459 | oldest_start = d_bh->b_blocknr; | |
2460 | newest_mount_id = get_desc_mount_id(desc); | |
2461 | reiserfs_debug(sb, REISERFS_DEBUG_CODE, | |
2462 | "journal-1179: Setting " | |
2463 | "oldest_start to offset %llu, trans_id %lu", | |
2464 | oldest_start - | |
2465 | SB_ONDISK_JOURNAL_1st_BLOCK | |
2466 | (sb), oldest_trans_id); | |
2467 | } else if (oldest_trans_id > get_desc_trans_id(desc)) { | |
2468 | /* one we just read was older */ | |
2469 | oldest_trans_id = get_desc_trans_id(desc); | |
2470 | oldest_start = d_bh->b_blocknr; | |
2471 | reiserfs_debug(sb, REISERFS_DEBUG_CODE, | |
2472 | "journal-1180: Resetting " | |
2473 | "oldest_start to offset %lu, trans_id %lu", | |
2474 | oldest_start - | |
2475 | SB_ONDISK_JOURNAL_1st_BLOCK | |
2476 | (sb), oldest_trans_id); | |
2477 | } | |
2478 | if (newest_mount_id < get_desc_mount_id(desc)) { | |
2479 | newest_mount_id = get_desc_mount_id(desc); | |
2480 | reiserfs_debug(sb, REISERFS_DEBUG_CODE, | |
2481 | "journal-1299: Setting " | |
2482 | "newest_mount_id to %d", | |
2483 | get_desc_mount_id(desc)); | |
2484 | } | |
2485 | cur_dblock += get_desc_trans_len(desc) + 2; | |
2486 | } else { | |
2487 | cur_dblock++; | |
2488 | } | |
2489 | brelse(d_bh); | |
2490 | } | |
2491 | ||
2492 | start_log_replay: | |
2493 | cur_dblock = oldest_start; | |
2494 | if (oldest_trans_id) { | |
2495 | reiserfs_debug(sb, REISERFS_DEBUG_CODE, | |
2496 | "journal-1206: Starting replay " | |
2497 | "from offset %llu, trans_id %lu", | |
2498 | cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb), | |
2499 | oldest_trans_id); | |
2500 | ||
2501 | } | |
2502 | replay_count = 0; | |
2503 | while (continue_replay && oldest_trans_id > 0) { | |
2504 | ret = | |
2505 | journal_read_transaction(sb, cur_dblock, oldest_start, | |
2506 | oldest_trans_id, newest_mount_id); | |
2507 | if (ret < 0) { | |
2508 | return ret; | |
2509 | } else if (ret != 0) { | |
2510 | break; | |
2511 | } | |
2512 | cur_dblock = | |
2513 | SB_ONDISK_JOURNAL_1st_BLOCK(sb) + journal->j_start; | |
2514 | replay_count++; | |
2515 | if (cur_dblock == oldest_start) | |
2516 | break; | |
2517 | } | |
2518 | ||
2519 | if (oldest_trans_id == 0) { | |
2520 | reiserfs_debug(sb, REISERFS_DEBUG_CODE, | |
2521 | "journal-1225: No valid " "transactions found"); | |
2522 | } | |
2523 | /* | |
2524 | * j_start does not get set correctly if we don't replay any | |
2525 | * transactions. if we had a valid journal_header, set j_start | |
2526 | * to the first unflushed transaction value, copy the trans_id | |
2527 | * from the header | |
2528 | */ | |
2529 | if (valid_journal_header && replay_count == 0) { | |
2530 | journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset); | |
2531 | journal->j_trans_id = | |
2532 | le32_to_cpu(jh->j_last_flush_trans_id) + 1; | |
2533 | /* check for trans_id overflow */ | |
2534 | if (journal->j_trans_id == 0) | |
2535 | journal->j_trans_id = 10; | |
2536 | journal->j_last_flush_trans_id = | |
2537 | le32_to_cpu(jh->j_last_flush_trans_id); | |
2538 | journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1; | |
2539 | } else { | |
2540 | journal->j_mount_id = newest_mount_id + 1; | |
2541 | } | |
2542 | reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1299: Setting " | |
2543 | "newest_mount_id to %lu", journal->j_mount_id); | |
2544 | journal->j_first_unflushed_offset = journal->j_start; | |
2545 | if (replay_count > 0) { | |
2546 | reiserfs_info(sb, | |
2547 | "replayed %d transactions in %lu seconds\n", | |
2548 | replay_count, ktime_get_seconds() - start); | |
2549 | } | |
2550 | /* needed to satisfy the locking in _update_journal_header_block */ | |
2551 | reiserfs_write_lock(sb); | |
2552 | if (!bdev_read_only(sb->s_bdev) && | |
2553 | _update_journal_header_block(sb, journal->j_start, | |
2554 | journal->j_last_flush_trans_id)) { | |
2555 | reiserfs_write_unlock(sb); | |
2556 | /* | |
2557 | * replay failed, caller must call free_journal_ram and abort | |
2558 | * the mount | |
2559 | */ | |
2560 | return -1; | |
2561 | } | |
2562 | reiserfs_write_unlock(sb); | |
2563 | return 0; | |
2564 | } | |
2565 | ||
2566 | static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s) | |
2567 | { | |
2568 | struct reiserfs_journal_list *jl; | |
2569 | jl = kzalloc(sizeof(struct reiserfs_journal_list), | |
2570 | GFP_NOFS | __GFP_NOFAIL); | |
2571 | INIT_LIST_HEAD(&jl->j_list); | |
2572 | INIT_LIST_HEAD(&jl->j_working_list); | |
2573 | INIT_LIST_HEAD(&jl->j_tail_bh_list); | |
2574 | INIT_LIST_HEAD(&jl->j_bh_list); | |
2575 | mutex_init(&jl->j_commit_mutex); | |
2576 | SB_JOURNAL(s)->j_num_lists++; | |
2577 | get_journal_list(jl); | |
2578 | return jl; | |
2579 | } | |
2580 | ||
2581 | static void journal_list_init(struct super_block *sb) | |
2582 | { | |
2583 | SB_JOURNAL(sb)->j_current_jl = alloc_journal_list(sb); | |
2584 | } | |
2585 | ||
2586 | static void release_journal_dev(struct super_block *super, | |
2587 | struct reiserfs_journal *journal) | |
2588 | { | |
2589 | if (journal->j_dev_bd != NULL) { | |
2590 | blkdev_put(journal->j_dev_bd, journal->j_dev_mode); | |
2591 | journal->j_dev_bd = NULL; | |
2592 | } | |
2593 | } | |
2594 | ||
2595 | static int journal_init_dev(struct super_block *super, | |
2596 | struct reiserfs_journal *journal, | |
2597 | const char *jdev_name) | |
2598 | { | |
2599 | int result; | |
2600 | dev_t jdev; | |
2601 | fmode_t blkdev_mode = FMODE_READ | FMODE_WRITE | FMODE_EXCL; | |
2602 | ||
2603 | result = 0; | |
2604 | ||
2605 | journal->j_dev_bd = NULL; | |
2606 | jdev = SB_ONDISK_JOURNAL_DEVICE(super) ? | |
2607 | new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev; | |
2608 | ||
2609 | if (bdev_read_only(super->s_bdev)) | |
2610 | blkdev_mode = FMODE_READ; | |
2611 | ||
2612 | /* there is no "jdev" option and journal is on separate device */ | |
2613 | if ((!jdev_name || !jdev_name[0])) { | |
2614 | if (jdev == super->s_dev) | |
2615 | blkdev_mode &= ~FMODE_EXCL; | |
2616 | journal->j_dev_bd = blkdev_get_by_dev(jdev, blkdev_mode, | |
2617 | journal); | |
2618 | journal->j_dev_mode = blkdev_mode; | |
2619 | if (IS_ERR(journal->j_dev_bd)) { | |
2620 | result = PTR_ERR(journal->j_dev_bd); | |
2621 | journal->j_dev_bd = NULL; | |
2622 | reiserfs_warning(super, "sh-458", | |
2623 | "cannot init journal device unknown-block(%u,%u): %i", | |
2624 | MAJOR(jdev), MINOR(jdev), result); | |
2625 | return result; | |
2626 | } else if (jdev != super->s_dev) | |
2627 | set_blocksize(journal->j_dev_bd, super->s_blocksize); | |
2628 | ||
2629 | return 0; | |
2630 | } | |
2631 | ||
2632 | journal->j_dev_mode = blkdev_mode; | |
2633 | journal->j_dev_bd = blkdev_get_by_path(jdev_name, blkdev_mode, journal); | |
2634 | if (IS_ERR(journal->j_dev_bd)) { | |
2635 | result = PTR_ERR(journal->j_dev_bd); | |
2636 | journal->j_dev_bd = NULL; | |
2637 | reiserfs_warning(super, "sh-457", | |
2638 | "journal_init_dev: Cannot open '%s': %i", | |
2639 | jdev_name, result); | |
2640 | return result; | |
2641 | } | |
2642 | ||
2643 | set_blocksize(journal->j_dev_bd, super->s_blocksize); | |
2644 | reiserfs_info(super, | |
2645 | "journal_init_dev: journal device: %pg\n", | |
2646 | journal->j_dev_bd); | |
2647 | return 0; | |
2648 | } | |
2649 | ||
2650 | /* | |
2651 | * When creating/tuning a file system user can assign some | |
2652 | * journal params within boundaries which depend on the ratio | |
2653 | * blocksize/standard_blocksize. | |
2654 | * | |
2655 | * For blocks >= standard_blocksize transaction size should | |
2656 | * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more | |
2657 | * then JOURNAL_TRANS_MAX_DEFAULT. | |
2658 | * | |
2659 | * For blocks < standard_blocksize these boundaries should be | |
2660 | * decreased proportionally. | |
2661 | */ | |
2662 | #define REISERFS_STANDARD_BLKSIZE (4096) | |
2663 | ||
2664 | static int check_advise_trans_params(struct super_block *sb, | |
2665 | struct reiserfs_journal *journal) | |
2666 | { | |
2667 | if (journal->j_trans_max) { | |
2668 | /* Non-default journal params. Do sanity check for them. */ | |
2669 | int ratio = 1; | |
2670 | if (sb->s_blocksize < REISERFS_STANDARD_BLKSIZE) | |
2671 | ratio = REISERFS_STANDARD_BLKSIZE / sb->s_blocksize; | |
2672 | ||
2673 | if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio || | |
2674 | journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio || | |
2675 | SB_ONDISK_JOURNAL_SIZE(sb) / journal->j_trans_max < | |
2676 | JOURNAL_MIN_RATIO) { | |
2677 | reiserfs_warning(sb, "sh-462", | |
2678 | "bad transaction max size (%u). " | |
2679 | "FSCK?", journal->j_trans_max); | |
2680 | return 1; | |
2681 | } | |
2682 | if (journal->j_max_batch != (journal->j_trans_max) * | |
2683 | JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) { | |
2684 | reiserfs_warning(sb, "sh-463", | |
2685 | "bad transaction max batch (%u). " | |
2686 | "FSCK?", journal->j_max_batch); | |
2687 | return 1; | |
2688 | } | |
2689 | } else { | |
2690 | /* | |
2691 | * Default journal params. | |
2692 | * The file system was created by old version | |
2693 | * of mkreiserfs, so some fields contain zeros, | |
2694 | * and we need to advise proper values for them | |
2695 | */ | |
2696 | if (sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) { | |
2697 | reiserfs_warning(sb, "sh-464", "bad blocksize (%u)", | |
2698 | sb->s_blocksize); | |
2699 | return 1; | |
2700 | } | |
2701 | journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT; | |
2702 | journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT; | |
2703 | journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE; | |
2704 | } | |
2705 | return 0; | |
2706 | } | |
2707 | ||
2708 | /* must be called once on fs mount. calls journal_read for you */ | |
2709 | int journal_init(struct super_block *sb, const char *j_dev_name, | |
2710 | int old_format, unsigned int commit_max_age) | |
2711 | { | |
2712 | int num_cnodes = SB_ONDISK_JOURNAL_SIZE(sb) * 2; | |
2713 | struct buffer_head *bhjh; | |
2714 | struct reiserfs_super_block *rs; | |
2715 | struct reiserfs_journal_header *jh; | |
2716 | struct reiserfs_journal *journal; | |
2717 | struct reiserfs_journal_list *jl; | |
2718 | int ret; | |
2719 | ||
2720 | journal = SB_JOURNAL(sb) = vzalloc(sizeof(struct reiserfs_journal)); | |
2721 | if (!journal) { | |
2722 | reiserfs_warning(sb, "journal-1256", | |
2723 | "unable to get memory for journal structure"); | |
2724 | return 1; | |
2725 | } | |
2726 | INIT_LIST_HEAD(&journal->j_bitmap_nodes); | |
2727 | INIT_LIST_HEAD(&journal->j_prealloc_list); | |
2728 | INIT_LIST_HEAD(&journal->j_working_list); | |
2729 | INIT_LIST_HEAD(&journal->j_journal_list); | |
2730 | journal->j_persistent_trans = 0; | |
2731 | if (reiserfs_allocate_list_bitmaps(sb, journal->j_list_bitmap, | |
2732 | reiserfs_bmap_count(sb))) | |
2733 | goto free_and_return; | |
2734 | ||
2735 | allocate_bitmap_nodes(sb); | |
2736 | ||
2737 | /* reserved for journal area support */ | |
2738 | SB_JOURNAL_1st_RESERVED_BLOCK(sb) = (old_format ? | |
2739 | REISERFS_OLD_DISK_OFFSET_IN_BYTES | |
2740 | / sb->s_blocksize + | |
2741 | reiserfs_bmap_count(sb) + | |
2742 | 1 : | |
2743 | REISERFS_DISK_OFFSET_IN_BYTES / | |
2744 | sb->s_blocksize + 2); | |
2745 | ||
2746 | /* | |
2747 | * Sanity check to see is the standard journal fitting | |
2748 | * within first bitmap (actual for small blocksizes) | |
2749 | */ | |
2750 | if (!SB_ONDISK_JOURNAL_DEVICE(sb) && | |
2751 | (SB_JOURNAL_1st_RESERVED_BLOCK(sb) + | |
2752 | SB_ONDISK_JOURNAL_SIZE(sb) > sb->s_blocksize * 8)) { | |
2753 | reiserfs_warning(sb, "journal-1393", | |
2754 | "journal does not fit for area addressed " | |
2755 | "by first of bitmap blocks. It starts at " | |
2756 | "%u and its size is %u. Block size %ld", | |
2757 | SB_JOURNAL_1st_RESERVED_BLOCK(sb), | |
2758 | SB_ONDISK_JOURNAL_SIZE(sb), | |
2759 | sb->s_blocksize); | |
2760 | goto free_and_return; | |
2761 | } | |
2762 | ||
2763 | if (journal_init_dev(sb, journal, j_dev_name) != 0) { | |
2764 | reiserfs_warning(sb, "sh-462", | |
2765 | "unable to initialize journal device"); | |
2766 | goto free_and_return; | |
2767 | } | |
2768 | ||
2769 | rs = SB_DISK_SUPER_BLOCK(sb); | |
2770 | ||
2771 | /* read journal header */ | |
2772 | bhjh = journal_bread(sb, | |
2773 | SB_ONDISK_JOURNAL_1st_BLOCK(sb) + | |
2774 | SB_ONDISK_JOURNAL_SIZE(sb)); | |
2775 | if (!bhjh) { | |
2776 | reiserfs_warning(sb, "sh-459", | |
2777 | "unable to read journal header"); | |
2778 | goto free_and_return; | |
2779 | } | |
2780 | jh = (struct reiserfs_journal_header *)(bhjh->b_data); | |
2781 | ||
2782 | /* make sure that journal matches to the super block */ | |
2783 | if (is_reiserfs_jr(rs) | |
2784 | && (le32_to_cpu(jh->jh_journal.jp_journal_magic) != | |
2785 | sb_jp_journal_magic(rs))) { | |
2786 | reiserfs_warning(sb, "sh-460", | |
2787 | "journal header magic %x (device %pg) does " | |
2788 | "not match to magic found in super block %x", | |
2789 | jh->jh_journal.jp_journal_magic, | |
2790 | journal->j_dev_bd, | |
2791 | sb_jp_journal_magic(rs)); | |
2792 | brelse(bhjh); | |
2793 | goto free_and_return; | |
2794 | } | |
2795 | ||
2796 | journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max); | |
2797 | journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch); | |
2798 | journal->j_max_commit_age = | |
2799 | le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age); | |
2800 | journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE; | |
2801 | ||
2802 | if (check_advise_trans_params(sb, journal) != 0) | |
2803 | goto free_and_return; | |
2804 | journal->j_default_max_commit_age = journal->j_max_commit_age; | |
2805 | ||
2806 | if (commit_max_age != 0) { | |
2807 | journal->j_max_commit_age = commit_max_age; | |
2808 | journal->j_max_trans_age = commit_max_age; | |
2809 | } | |
2810 | ||
2811 | reiserfs_info(sb, "journal params: device %pg, size %u, " | |
2812 | "journal first block %u, max trans len %u, max batch %u, " | |
2813 | "max commit age %u, max trans age %u\n", | |
2814 | journal->j_dev_bd, | |
2815 | SB_ONDISK_JOURNAL_SIZE(sb), | |
2816 | SB_ONDISK_JOURNAL_1st_BLOCK(sb), | |
2817 | journal->j_trans_max, | |
2818 | journal->j_max_batch, | |
2819 | journal->j_max_commit_age, journal->j_max_trans_age); | |
2820 | ||
2821 | brelse(bhjh); | |
2822 | ||
2823 | journal->j_list_bitmap_index = 0; | |
2824 | journal_list_init(sb); | |
2825 | ||
2826 | memset(journal->j_list_hash_table, 0, | |
2827 | JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *)); | |
2828 | ||
2829 | INIT_LIST_HEAD(&journal->j_dirty_buffers); | |
2830 | spin_lock_init(&journal->j_dirty_buffers_lock); | |
2831 | ||
2832 | journal->j_start = 0; | |
2833 | journal->j_len = 0; | |
2834 | journal->j_len_alloc = 0; | |
2835 | atomic_set(&journal->j_wcount, 0); | |
2836 | atomic_set(&journal->j_async_throttle, 0); | |
2837 | journal->j_bcount = 0; | |
2838 | journal->j_trans_start_time = 0; | |
2839 | journal->j_last = NULL; | |
2840 | journal->j_first = NULL; | |
2841 | init_waitqueue_head(&journal->j_join_wait); | |
2842 | mutex_init(&journal->j_mutex); | |
2843 | mutex_init(&journal->j_flush_mutex); | |
2844 | ||
2845 | journal->j_trans_id = 10; | |
2846 | journal->j_mount_id = 10; | |
2847 | journal->j_state = 0; | |
2848 | atomic_set(&journal->j_jlock, 0); | |
2849 | journal->j_cnode_free_list = allocate_cnodes(num_cnodes); | |
2850 | journal->j_cnode_free_orig = journal->j_cnode_free_list; | |
2851 | journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0; | |
2852 | journal->j_cnode_used = 0; | |
2853 | journal->j_must_wait = 0; | |
2854 | ||
2855 | if (journal->j_cnode_free == 0) { | |
2856 | reiserfs_warning(sb, "journal-2004", "Journal cnode memory " | |
2857 | "allocation failed (%ld bytes). Journal is " | |
2858 | "too large for available memory. Usually " | |
2859 | "this is due to a journal that is too large.", | |
2860 | sizeof (struct reiserfs_journal_cnode) * num_cnodes); | |
2861 | goto free_and_return; | |
2862 | } | |
2863 | ||
2864 | init_journal_hash(sb); | |
2865 | jl = journal->j_current_jl; | |
2866 | ||
2867 | /* | |
2868 | * get_list_bitmap() may call flush_commit_list() which | |
2869 | * requires the lock. Calling flush_commit_list() shouldn't happen | |
2870 | * this early but I like to be paranoid. | |
2871 | */ | |
2872 | reiserfs_write_lock(sb); | |
2873 | jl->j_list_bitmap = get_list_bitmap(sb, jl); | |
2874 | reiserfs_write_unlock(sb); | |
2875 | if (!jl->j_list_bitmap) { | |
2876 | reiserfs_warning(sb, "journal-2005", | |
2877 | "get_list_bitmap failed for journal list 0"); | |
2878 | goto free_and_return; | |
2879 | } | |
2880 | ||
2881 | ret = journal_read(sb); | |
2882 | if (ret < 0) { | |
2883 | reiserfs_warning(sb, "reiserfs-2006", | |
2884 | "Replay Failure, unable to mount"); | |
2885 | goto free_and_return; | |
2886 | } | |
2887 | ||
2888 | INIT_DELAYED_WORK(&journal->j_work, flush_async_commits); | |
2889 | journal->j_work_sb = sb; | |
2890 | return 0; | |
2891 | free_and_return: | |
2892 | free_journal_ram(sb); | |
2893 | return 1; | |
2894 | } | |
2895 | ||
2896 | /* | |
2897 | * test for a polite end of the current transaction. Used by file_write, | |
2898 | * and should be used by delete to make sure they don't write more than | |
2899 | * can fit inside a single transaction | |
2900 | */ | |
2901 | int journal_transaction_should_end(struct reiserfs_transaction_handle *th, | |
2902 | int new_alloc) | |
2903 | { | |
2904 | struct reiserfs_journal *journal = SB_JOURNAL(th->t_super); | |
2905 | time64_t now = ktime_get_seconds(); | |
2906 | /* cannot restart while nested */ | |
2907 | BUG_ON(!th->t_trans_id); | |
2908 | if (th->t_refcount > 1) | |
2909 | return 0; | |
2910 | if (journal->j_must_wait > 0 || | |
2911 | (journal->j_len_alloc + new_alloc) >= journal->j_max_batch || | |
2912 | atomic_read(&journal->j_jlock) || | |
2913 | (now - journal->j_trans_start_time) > journal->j_max_trans_age || | |
2914 | journal->j_cnode_free < (journal->j_trans_max * 3)) { | |
2915 | return 1; | |
2916 | } | |
2917 | ||
2918 | journal->j_len_alloc += new_alloc; | |
2919 | th->t_blocks_allocated += new_alloc ; | |
2920 | return 0; | |
2921 | } | |
2922 | ||
2923 | /* this must be called inside a transaction */ | |
2924 | void reiserfs_block_writes(struct reiserfs_transaction_handle *th) | |
2925 | { | |
2926 | struct reiserfs_journal *journal = SB_JOURNAL(th->t_super); | |
2927 | BUG_ON(!th->t_trans_id); | |
2928 | journal->j_must_wait = 1; | |
2929 | set_bit(J_WRITERS_BLOCKED, &journal->j_state); | |
2930 | return; | |
2931 | } | |
2932 | ||
2933 | /* this must be called without a transaction started */ | |
2934 | void reiserfs_allow_writes(struct super_block *s) | |
2935 | { | |
2936 | struct reiserfs_journal *journal = SB_JOURNAL(s); | |
2937 | clear_bit(J_WRITERS_BLOCKED, &journal->j_state); | |
2938 | wake_up(&journal->j_join_wait); | |
2939 | } | |
2940 | ||
2941 | /* this must be called without a transaction started */ | |
2942 | void reiserfs_wait_on_write_block(struct super_block *s) | |
2943 | { | |
2944 | struct reiserfs_journal *journal = SB_JOURNAL(s); | |
2945 | wait_event(journal->j_join_wait, | |
2946 | !test_bit(J_WRITERS_BLOCKED, &journal->j_state)); | |
2947 | } | |
2948 | ||
2949 | static void queue_log_writer(struct super_block *s) | |
2950 | { | |
2951 | wait_queue_entry_t wait; | |
2952 | struct reiserfs_journal *journal = SB_JOURNAL(s); | |
2953 | set_bit(J_WRITERS_QUEUED, &journal->j_state); | |
2954 | ||
2955 | /* | |
2956 | * we don't want to use wait_event here because | |
2957 | * we only want to wait once. | |
2958 | */ | |
2959 | init_waitqueue_entry(&wait, current); | |
2960 | add_wait_queue(&journal->j_join_wait, &wait); | |
2961 | set_current_state(TASK_UNINTERRUPTIBLE); | |
2962 | if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) { | |
2963 | int depth = reiserfs_write_unlock_nested(s); | |
2964 | schedule(); | |
2965 | reiserfs_write_lock_nested(s, depth); | |
2966 | } | |
2967 | __set_current_state(TASK_RUNNING); | |
2968 | remove_wait_queue(&journal->j_join_wait, &wait); | |
2969 | } | |
2970 | ||
2971 | static void wake_queued_writers(struct super_block *s) | |
2972 | { | |
2973 | struct reiserfs_journal *journal = SB_JOURNAL(s); | |
2974 | if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state)) | |
2975 | wake_up(&journal->j_join_wait); | |
2976 | } | |
2977 | ||
2978 | static void let_transaction_grow(struct super_block *sb, unsigned int trans_id) | |
2979 | { | |
2980 | struct reiserfs_journal *journal = SB_JOURNAL(sb); | |
2981 | unsigned long bcount = journal->j_bcount; | |
2982 | while (1) { | |
2983 | int depth; | |
2984 | ||
2985 | depth = reiserfs_write_unlock_nested(sb); | |
2986 | schedule_timeout_uninterruptible(1); | |
2987 | reiserfs_write_lock_nested(sb, depth); | |
2988 | ||
2989 | journal->j_current_jl->j_state |= LIST_COMMIT_PENDING; | |
2990 | while ((atomic_read(&journal->j_wcount) > 0 || | |
2991 | atomic_read(&journal->j_jlock)) && | |
2992 | journal->j_trans_id == trans_id) { | |
2993 | queue_log_writer(sb); | |
2994 | } | |
2995 | if (journal->j_trans_id != trans_id) | |
2996 | break; | |
2997 | if (bcount == journal->j_bcount) | |
2998 | break; | |
2999 | bcount = journal->j_bcount; | |
3000 | } | |
3001 | } | |
3002 | ||
3003 | /* | |
3004 | * join == true if you must join an existing transaction. | |
3005 | * join == false if you can deal with waiting for others to finish | |
3006 | * | |
3007 | * this will block until the transaction is joinable. send the number of | |
3008 | * blocks you expect to use in nblocks. | |
3009 | */ | |
3010 | static int do_journal_begin_r(struct reiserfs_transaction_handle *th, | |
3011 | struct super_block *sb, unsigned long nblocks, | |
3012 | int join) | |
3013 | { | |
3014 | time64_t now = ktime_get_seconds(); | |
3015 | unsigned int old_trans_id; | |
3016 | struct reiserfs_journal *journal = SB_JOURNAL(sb); | |
3017 | struct reiserfs_transaction_handle myth; | |
3018 | int sched_count = 0; | |
3019 | int retval; | |
3020 | int depth; | |
3021 | ||
3022 | reiserfs_check_lock_depth(sb, "journal_begin"); | |
3023 | BUG_ON(nblocks > journal->j_trans_max); | |
3024 | ||
3025 | PROC_INFO_INC(sb, journal.journal_being); | |
3026 | /* set here for journal_join */ | |
3027 | th->t_refcount = 1; | |
3028 | th->t_super = sb; | |
3029 | ||
3030 | relock: | |
3031 | lock_journal(sb); | |
3032 | if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) { | |
3033 | unlock_journal(sb); | |
3034 | retval = journal->j_errno; | |
3035 | goto out_fail; | |
3036 | } | |
3037 | journal->j_bcount++; | |
3038 | ||
3039 | if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) { | |
3040 | unlock_journal(sb); | |
3041 | depth = reiserfs_write_unlock_nested(sb); | |
3042 | reiserfs_wait_on_write_block(sb); | |
3043 | reiserfs_write_lock_nested(sb, depth); | |
3044 | PROC_INFO_INC(sb, journal.journal_relock_writers); | |
3045 | goto relock; | |
3046 | } | |
3047 | now = ktime_get_seconds(); | |
3048 | ||
3049 | /* | |
3050 | * if there is no room in the journal OR | |
3051 | * if this transaction is too old, and we weren't called joinable, | |
3052 | * wait for it to finish before beginning we don't sleep if there | |
3053 | * aren't other writers | |
3054 | */ | |
3055 | ||
3056 | if ((!join && journal->j_must_wait > 0) || | |
3057 | (!join | |
3058 | && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch) | |
3059 | || (!join && atomic_read(&journal->j_wcount) > 0 | |
3060 | && journal->j_trans_start_time > 0 | |
3061 | && (now - journal->j_trans_start_time) > | |
3062 | journal->j_max_trans_age) || (!join | |
3063 | && atomic_read(&journal->j_jlock)) | |
3064 | || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) { | |
3065 | ||
3066 | old_trans_id = journal->j_trans_id; | |
3067 | /* allow others to finish this transaction */ | |
3068 | unlock_journal(sb); | |
3069 | ||
3070 | if (!join && (journal->j_len_alloc + nblocks + 2) >= | |
3071 | journal->j_max_batch && | |
3072 | ((journal->j_len + nblocks + 2) * 100) < | |
3073 | (journal->j_len_alloc * 75)) { | |
3074 | if (atomic_read(&journal->j_wcount) > 10) { | |
3075 | sched_count++; | |
3076 | queue_log_writer(sb); | |
3077 | goto relock; | |
3078 | } | |
3079 | } | |
3080 | /* | |
3081 | * don't mess with joining the transaction if all we | |
3082 | * have to do is wait for someone else to do a commit | |
3083 | */ | |
3084 | if (atomic_read(&journal->j_jlock)) { | |
3085 | while (journal->j_trans_id == old_trans_id && | |
3086 | atomic_read(&journal->j_jlock)) { | |
3087 | queue_log_writer(sb); | |
3088 | } | |
3089 | goto relock; | |
3090 | } | |
3091 | retval = journal_join(&myth, sb); | |
3092 | if (retval) | |
3093 | goto out_fail; | |
3094 | ||
3095 | /* someone might have ended the transaction while we joined */ | |
3096 | if (old_trans_id != journal->j_trans_id) { | |
3097 | retval = do_journal_end(&myth, 0); | |
3098 | } else { | |
3099 | retval = do_journal_end(&myth, COMMIT_NOW); | |
3100 | } | |
3101 | ||
3102 | if (retval) | |
3103 | goto out_fail; | |
3104 | ||
3105 | PROC_INFO_INC(sb, journal.journal_relock_wcount); | |
3106 | goto relock; | |
3107 | } | |
3108 | /* we are the first writer, set trans_id */ | |
3109 | if (journal->j_trans_start_time == 0) { | |
3110 | journal->j_trans_start_time = ktime_get_seconds(); | |
3111 | } | |
3112 | atomic_inc(&journal->j_wcount); | |
3113 | journal->j_len_alloc += nblocks; | |
3114 | th->t_blocks_logged = 0; | |
3115 | th->t_blocks_allocated = nblocks; | |
3116 | th->t_trans_id = journal->j_trans_id; | |
3117 | unlock_journal(sb); | |
3118 | INIT_LIST_HEAD(&th->t_list); | |
3119 | return 0; | |
3120 | ||
3121 | out_fail: | |
3122 | memset(th, 0, sizeof(*th)); | |
3123 | /* | |
3124 | * Re-set th->t_super, so we can properly keep track of how many | |
3125 | * persistent transactions there are. We need to do this so if this | |
3126 | * call is part of a failed restart_transaction, we can free it later | |
3127 | */ | |
3128 | th->t_super = sb; | |
3129 | return retval; | |
3130 | } | |
3131 | ||
3132 | struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct | |
3133 | super_block | |
3134 | *s, | |
3135 | int nblocks) | |
3136 | { | |
3137 | int ret; | |
3138 | struct reiserfs_transaction_handle *th; | |
3139 | ||
3140 | /* | |
3141 | * if we're nesting into an existing transaction. It will be | |
3142 | * persistent on its own | |
3143 | */ | |
3144 | if (reiserfs_transaction_running(s)) { | |
3145 | th = current->journal_info; | |
3146 | th->t_refcount++; | |
3147 | BUG_ON(th->t_refcount < 2); | |
3148 | ||
3149 | return th; | |
3150 | } | |
3151 | th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS); | |
3152 | if (!th) | |
3153 | return NULL; | |
3154 | ret = journal_begin(th, s, nblocks); | |
3155 | if (ret) { | |
3156 | kfree(th); | |
3157 | return NULL; | |
3158 | } | |
3159 | ||
3160 | SB_JOURNAL(s)->j_persistent_trans++; | |
3161 | return th; | |
3162 | } | |
3163 | ||
3164 | int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th) | |
3165 | { | |
3166 | struct super_block *s = th->t_super; | |
3167 | int ret = 0; | |
3168 | if (th->t_trans_id) | |
3169 | ret = journal_end(th); | |
3170 | else | |
3171 | ret = -EIO; | |
3172 | if (th->t_refcount == 0) { | |
3173 | SB_JOURNAL(s)->j_persistent_trans--; | |
3174 | kfree(th); | |
3175 | } | |
3176 | return ret; | |
3177 | } | |
3178 | ||
3179 | static int journal_join(struct reiserfs_transaction_handle *th, | |
3180 | struct super_block *sb) | |
3181 | { | |
3182 | struct reiserfs_transaction_handle *cur_th = current->journal_info; | |
3183 | ||
3184 | /* | |
3185 | * this keeps do_journal_end from NULLing out the | |
3186 | * current->journal_info pointer | |
3187 | */ | |
3188 | th->t_handle_save = cur_th; | |
3189 | BUG_ON(cur_th && cur_th->t_refcount > 1); | |
3190 | return do_journal_begin_r(th, sb, 1, JBEGIN_JOIN); | |
3191 | } | |
3192 | ||
3193 | int journal_join_abort(struct reiserfs_transaction_handle *th, | |
3194 | struct super_block *sb) | |
3195 | { | |
3196 | struct reiserfs_transaction_handle *cur_th = current->journal_info; | |
3197 | ||
3198 | /* | |
3199 | * this keeps do_journal_end from NULLing out the | |
3200 | * current->journal_info pointer | |
3201 | */ | |
3202 | th->t_handle_save = cur_th; | |
3203 | BUG_ON(cur_th && cur_th->t_refcount > 1); | |
3204 | return do_journal_begin_r(th, sb, 1, JBEGIN_ABORT); | |
3205 | } | |
3206 | ||
3207 | int journal_begin(struct reiserfs_transaction_handle *th, | |
3208 | struct super_block *sb, unsigned long nblocks) | |
3209 | { | |
3210 | struct reiserfs_transaction_handle *cur_th = current->journal_info; | |
3211 | int ret; | |
3212 | ||
3213 | th->t_handle_save = NULL; | |
3214 | if (cur_th) { | |
3215 | /* we are nesting into the current transaction */ | |
3216 | if (cur_th->t_super == sb) { | |
3217 | BUG_ON(!cur_th->t_refcount); | |
3218 | cur_th->t_refcount++; | |
3219 | memcpy(th, cur_th, sizeof(*th)); | |
3220 | if (th->t_refcount <= 1) | |
3221 | reiserfs_warning(sb, "reiserfs-2005", | |
3222 | "BAD: refcount <= 1, but " | |
3223 | "journal_info != 0"); | |
3224 | return 0; | |
3225 | } else { | |
3226 | /* | |
3227 | * we've ended up with a handle from a different | |
3228 | * filesystem. save it and restore on journal_end. | |
3229 | * This should never really happen... | |
3230 | */ | |
3231 | reiserfs_warning(sb, "clm-2100", | |
3232 | "nesting info a different FS"); | |
3233 | th->t_handle_save = current->journal_info; | |
3234 | current->journal_info = th; | |
3235 | } | |
3236 | } else { | |
3237 | current->journal_info = th; | |
3238 | } | |
3239 | ret = do_journal_begin_r(th, sb, nblocks, JBEGIN_REG); | |
3240 | BUG_ON(current->journal_info != th); | |
3241 | ||
3242 | /* | |
3243 | * I guess this boils down to being the reciprocal of clm-2100 above. | |
3244 | * If do_journal_begin_r fails, we need to put it back, since | |
3245 | * journal_end won't be called to do it. */ | |
3246 | if (ret) | |
3247 | current->journal_info = th->t_handle_save; | |
3248 | else | |
3249 | BUG_ON(!th->t_refcount); | |
3250 | ||
3251 | return ret; | |
3252 | } | |
3253 | ||
3254 | /* | |
3255 | * puts bh into the current transaction. If it was already there, reorders | |
3256 | * removes the old pointers from the hash, and puts new ones in (to make | |
3257 | * sure replay happen in the right order). | |
3258 | * | |
3259 | * if it was dirty, cleans and files onto the clean list. I can't let it | |
3260 | * be dirty again until the transaction is committed. | |
3261 | * | |
3262 | * if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len. | |
3263 | */ | |
3264 | int journal_mark_dirty(struct reiserfs_transaction_handle *th, | |
3265 | struct buffer_head *bh) | |
3266 | { | |
3267 | struct super_block *sb = th->t_super; | |
3268 | struct reiserfs_journal *journal = SB_JOURNAL(sb); | |
3269 | struct reiserfs_journal_cnode *cn = NULL; | |
3270 | int count_already_incd = 0; | |
3271 | int prepared = 0; | |
3272 | BUG_ON(!th->t_trans_id); | |
3273 | ||
3274 | PROC_INFO_INC(sb, journal.mark_dirty); | |
3275 | if (th->t_trans_id != journal->j_trans_id) { | |
3276 | reiserfs_panic(th->t_super, "journal-1577", | |
3277 | "handle trans id %ld != current trans id %ld", | |
3278 | th->t_trans_id, journal->j_trans_id); | |
3279 | } | |
3280 | ||
3281 | prepared = test_clear_buffer_journal_prepared(bh); | |
3282 | clear_buffer_journal_restore_dirty(bh); | |
3283 | /* already in this transaction, we are done */ | |
3284 | if (buffer_journaled(bh)) { | |
3285 | PROC_INFO_INC(sb, journal.mark_dirty_already); | |
3286 | return 0; | |
3287 | } | |
3288 | ||
3289 | /* | |
3290 | * this must be turned into a panic instead of a warning. We can't | |
3291 | * allow a dirty or journal_dirty or locked buffer to be logged, as | |
3292 | * some changes could get to disk too early. NOT GOOD. | |
3293 | */ | |
3294 | if (!prepared || buffer_dirty(bh)) { | |
3295 | reiserfs_warning(sb, "journal-1777", | |
3296 | "buffer %llu bad state " | |
3297 | "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT", | |
3298 | (unsigned long long)bh->b_blocknr, | |
3299 | prepared ? ' ' : '!', | |
3300 | buffer_locked(bh) ? ' ' : '!', | |
3301 | buffer_dirty(bh) ? ' ' : '!', | |
3302 | buffer_journal_dirty(bh) ? ' ' : '!'); | |
3303 | } | |
3304 | ||
3305 | if (atomic_read(&journal->j_wcount) <= 0) { | |
3306 | reiserfs_warning(sb, "journal-1409", | |
3307 | "returning because j_wcount was %d", | |
3308 | atomic_read(&journal->j_wcount)); | |
3309 | return 1; | |
3310 | } | |
3311 | /* | |
3312 | * this error means I've screwed up, and we've overflowed | |
3313 | * the transaction. Nothing can be done here, except make the | |
3314 | * FS readonly or panic. | |
3315 | */ | |
3316 | if (journal->j_len >= journal->j_trans_max) { | |
3317 | reiserfs_panic(th->t_super, "journal-1413", | |
3318 | "j_len (%lu) is too big", | |
3319 | journal->j_len); | |
3320 | } | |
3321 | ||
3322 | if (buffer_journal_dirty(bh)) { | |
3323 | count_already_incd = 1; | |
3324 | PROC_INFO_INC(sb, journal.mark_dirty_notjournal); | |
3325 | clear_buffer_journal_dirty(bh); | |
3326 | } | |
3327 | ||
3328 | if (journal->j_len > journal->j_len_alloc) { | |
3329 | journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT; | |
3330 | } | |
3331 | ||
3332 | set_buffer_journaled(bh); | |
3333 | ||
3334 | /* now put this guy on the end */ | |
3335 | if (!cn) { | |
3336 | cn = get_cnode(sb); | |
3337 | if (!cn) { | |
3338 | reiserfs_panic(sb, "journal-4", "get_cnode failed!"); | |
3339 | } | |
3340 | ||
3341 | if (th->t_blocks_logged == th->t_blocks_allocated) { | |
3342 | th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT; | |
3343 | journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT; | |
3344 | } | |
3345 | th->t_blocks_logged++; | |
3346 | journal->j_len++; | |
3347 | ||
3348 | cn->bh = bh; | |
3349 | cn->blocknr = bh->b_blocknr; | |
3350 | cn->sb = sb; | |
3351 | cn->jlist = NULL; | |
3352 | insert_journal_hash(journal->j_hash_table, cn); | |
3353 | if (!count_already_incd) { | |
3354 | get_bh(bh); | |
3355 | } | |
3356 | } | |
3357 | cn->next = NULL; | |
3358 | cn->prev = journal->j_last; | |
3359 | cn->bh = bh; | |
3360 | if (journal->j_last) { | |
3361 | journal->j_last->next = cn; | |
3362 | journal->j_last = cn; | |
3363 | } else { | |
3364 | journal->j_first = cn; | |
3365 | journal->j_last = cn; | |
3366 | } | |
3367 | reiserfs_schedule_old_flush(sb); | |
3368 | return 0; | |
3369 | } | |
3370 | ||
3371 | int journal_end(struct reiserfs_transaction_handle *th) | |
3372 | { | |
3373 | struct super_block *sb = th->t_super; | |
3374 | if (!current->journal_info && th->t_refcount > 1) | |
3375 | reiserfs_warning(sb, "REISER-NESTING", | |
3376 | "th NULL, refcount %d", th->t_refcount); | |
3377 | ||
3378 | if (!th->t_trans_id) { | |
3379 | WARN_ON(1); | |
3380 | return -EIO; | |
3381 | } | |
3382 | ||
3383 | th->t_refcount--; | |
3384 | if (th->t_refcount > 0) { | |
3385 | struct reiserfs_transaction_handle *cur_th = | |
3386 | current->journal_info; | |
3387 | ||
3388 | /* | |
3389 | * we aren't allowed to close a nested transaction on a | |
3390 | * different filesystem from the one in the task struct | |
3391 | */ | |
3392 | BUG_ON(cur_th->t_super != th->t_super); | |
3393 | ||
3394 | if (th != cur_th) { | |
3395 | memcpy(current->journal_info, th, sizeof(*th)); | |
3396 | th->t_trans_id = 0; | |
3397 | } | |
3398 | return 0; | |
3399 | } else { | |
3400 | return do_journal_end(th, 0); | |
3401 | } | |
3402 | } | |
3403 | ||
3404 | /* | |
3405 | * removes from the current transaction, relsing and descrementing any counters. | |
3406 | * also files the removed buffer directly onto the clean list | |
3407 | * | |
3408 | * called by journal_mark_freed when a block has been deleted | |
3409 | * | |
3410 | * returns 1 if it cleaned and relsed the buffer. 0 otherwise | |
3411 | */ | |
3412 | static int remove_from_transaction(struct super_block *sb, | |
3413 | b_blocknr_t blocknr, int already_cleaned) | |
3414 | { | |
3415 | struct buffer_head *bh; | |
3416 | struct reiserfs_journal_cnode *cn; | |
3417 | struct reiserfs_journal *journal = SB_JOURNAL(sb); | |
3418 | int ret = 0; | |
3419 | ||
3420 | cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr); | |
3421 | if (!cn || !cn->bh) { | |
3422 | return ret; | |
3423 | } | |
3424 | bh = cn->bh; | |
3425 | if (cn->prev) { | |
3426 | cn->prev->next = cn->next; | |
3427 | } | |
3428 | if (cn->next) { | |
3429 | cn->next->prev = cn->prev; | |
3430 | } | |
3431 | if (cn == journal->j_first) { | |
3432 | journal->j_first = cn->next; | |
3433 | } | |
3434 | if (cn == journal->j_last) { | |
3435 | journal->j_last = cn->prev; | |
3436 | } | |
3437 | remove_journal_hash(sb, journal->j_hash_table, NULL, | |
3438 | bh->b_blocknr, 0); | |
3439 | clear_buffer_journaled(bh); /* don't log this one */ | |
3440 | ||
3441 | if (!already_cleaned) { | |
3442 | clear_buffer_journal_dirty(bh); | |
3443 | clear_buffer_dirty(bh); | |
3444 | clear_buffer_journal_test(bh); | |
3445 | put_bh(bh); | |
3446 | if (atomic_read(&bh->b_count) < 0) { | |
3447 | reiserfs_warning(sb, "journal-1752", | |
3448 | "b_count < 0"); | |
3449 | } | |
3450 | ret = 1; | |
3451 | } | |
3452 | journal->j_len--; | |
3453 | journal->j_len_alloc--; | |
3454 | free_cnode(sb, cn); | |
3455 | return ret; | |
3456 | } | |
3457 | ||
3458 | /* | |
3459 | * for any cnode in a journal list, it can only be dirtied of all the | |
3460 | * transactions that include it are committed to disk. | |
3461 | * this checks through each transaction, and returns 1 if you are allowed | |
3462 | * to dirty, and 0 if you aren't | |
3463 | * | |
3464 | * it is called by dirty_journal_list, which is called after | |
3465 | * flush_commit_list has gotten all the log blocks for a given | |
3466 | * transaction on disk | |
3467 | * | |
3468 | */ | |
3469 | static int can_dirty(struct reiserfs_journal_cnode *cn) | |
3470 | { | |
3471 | struct super_block *sb = cn->sb; | |
3472 | b_blocknr_t blocknr = cn->blocknr; | |
3473 | struct reiserfs_journal_cnode *cur = cn->hprev; | |
3474 | int can_dirty = 1; | |
3475 | ||
3476 | /* | |
3477 | * first test hprev. These are all newer than cn, so any node here | |
3478 | * with the same block number and dev means this node can't be sent | |
3479 | * to disk right now. | |
3480 | */ | |
3481 | while (cur && can_dirty) { | |
3482 | if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb && | |
3483 | cur->blocknr == blocknr) { | |
3484 | can_dirty = 0; | |
3485 | } | |
3486 | cur = cur->hprev; | |
3487 | } | |
3488 | /* | |
3489 | * then test hnext. These are all older than cn. As long as they | |
3490 | * are committed to the log, it is safe to write cn to disk | |
3491 | */ | |
3492 | cur = cn->hnext; | |
3493 | while (cur && can_dirty) { | |
3494 | if (cur->jlist && cur->jlist->j_len > 0 && | |
3495 | atomic_read(&cur->jlist->j_commit_left) > 0 && cur->bh && | |
3496 | cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) { | |
3497 | can_dirty = 0; | |
3498 | } | |
3499 | cur = cur->hnext; | |
3500 | } | |
3501 | return can_dirty; | |
3502 | } | |
3503 | ||
3504 | /* | |
3505 | * syncs the commit blocks, but does not force the real buffers to disk | |
3506 | * will wait until the current transaction is done/committed before returning | |
3507 | */ | |
3508 | int journal_end_sync(struct reiserfs_transaction_handle *th) | |
3509 | { | |
3510 | struct super_block *sb = th->t_super; | |
3511 | struct reiserfs_journal *journal = SB_JOURNAL(sb); | |
3512 | ||
3513 | BUG_ON(!th->t_trans_id); | |
3514 | /* you can sync while nested, very, very bad */ | |
3515 | BUG_ON(th->t_refcount > 1); | |
3516 | if (journal->j_len == 0) { | |
3517 | reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb), | |
3518 | 1); | |
3519 | journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb)); | |
3520 | } | |
3521 | return do_journal_end(th, COMMIT_NOW | WAIT); | |
3522 | } | |
3523 | ||
3524 | /* writeback the pending async commits to disk */ | |
3525 | static void flush_async_commits(struct work_struct *work) | |
3526 | { | |
3527 | struct reiserfs_journal *journal = | |
3528 | container_of(work, struct reiserfs_journal, j_work.work); | |
3529 | struct super_block *sb = journal->j_work_sb; | |
3530 | struct reiserfs_journal_list *jl; | |
3531 | struct list_head *entry; | |
3532 | ||
3533 | reiserfs_write_lock(sb); | |
3534 | if (!list_empty(&journal->j_journal_list)) { | |
3535 | /* last entry is the youngest, commit it and you get everything */ | |
3536 | entry = journal->j_journal_list.prev; | |
3537 | jl = JOURNAL_LIST_ENTRY(entry); | |
3538 | flush_commit_list(sb, jl, 1); | |
3539 | } | |
3540 | reiserfs_write_unlock(sb); | |
3541 | } | |
3542 | ||
3543 | /* | |
3544 | * flushes any old transactions to disk | |
3545 | * ends the current transaction if it is too old | |
3546 | */ | |
3547 | void reiserfs_flush_old_commits(struct super_block *sb) | |
3548 | { | |
3549 | time64_t now; | |
3550 | struct reiserfs_transaction_handle th; | |
3551 | struct reiserfs_journal *journal = SB_JOURNAL(sb); | |
3552 | ||
3553 | now = ktime_get_seconds(); | |
3554 | /* | |
3555 | * safety check so we don't flush while we are replaying the log during | |
3556 | * mount | |
3557 | */ | |
3558 | if (list_empty(&journal->j_journal_list)) | |
3559 | return; | |
3560 | ||
3561 | /* | |
3562 | * check the current transaction. If there are no writers, and it is | |
3563 | * too old, finish it, and force the commit blocks to disk | |
3564 | */ | |
3565 | if (atomic_read(&journal->j_wcount) <= 0 && | |
3566 | journal->j_trans_start_time > 0 && | |
3567 | journal->j_len > 0 && | |
3568 | (now - journal->j_trans_start_time) > journal->j_max_trans_age) { | |
3569 | if (!journal_join(&th, sb)) { | |
3570 | reiserfs_prepare_for_journal(sb, | |
3571 | SB_BUFFER_WITH_SB(sb), | |
3572 | 1); | |
3573 | journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb)); | |
3574 | ||
3575 | /* | |
3576 | * we're only being called from kreiserfsd, it makes | |
3577 | * no sense to do an async commit so that kreiserfsd | |
3578 | * can do it later | |
3579 | */ | |
3580 | do_journal_end(&th, COMMIT_NOW | WAIT); | |
3581 | } | |
3582 | } | |
3583 | } | |
3584 | ||
3585 | /* | |
3586 | * returns 0 if do_journal_end should return right away, returns 1 if | |
3587 | * do_journal_end should finish the commit | |
3588 | * | |
3589 | * if the current transaction is too old, but still has writers, this will | |
3590 | * wait on j_join_wait until all the writers are done. By the time it | |
3591 | * wakes up, the transaction it was called has already ended, so it just | |
3592 | * flushes the commit list and returns 0. | |
3593 | * | |
3594 | * Won't batch when flush or commit_now is set. Also won't batch when | |
3595 | * others are waiting on j_join_wait. | |
3596 | * | |
3597 | * Note, we can't allow the journal_end to proceed while there are still | |
3598 | * writers in the log. | |
3599 | */ | |
3600 | static int check_journal_end(struct reiserfs_transaction_handle *th, int flags) | |
3601 | { | |
3602 | ||
3603 | time64_t now; | |
3604 | int flush = flags & FLUSH_ALL; | |
3605 | int commit_now = flags & COMMIT_NOW; | |
3606 | int wait_on_commit = flags & WAIT; | |
3607 | struct reiserfs_journal_list *jl; | |
3608 | struct super_block *sb = th->t_super; | |
3609 | struct reiserfs_journal *journal = SB_JOURNAL(sb); | |
3610 | ||
3611 | BUG_ON(!th->t_trans_id); | |
3612 | ||
3613 | if (th->t_trans_id != journal->j_trans_id) { | |
3614 | reiserfs_panic(th->t_super, "journal-1577", | |
3615 | "handle trans id %ld != current trans id %ld", | |
3616 | th->t_trans_id, journal->j_trans_id); | |
3617 | } | |
3618 | ||
3619 | journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged); | |
3620 | /* <= 0 is allowed. unmounting might not call begin */ | |
3621 | if (atomic_read(&journal->j_wcount) > 0) | |
3622 | atomic_dec(&journal->j_wcount); | |
3623 | ||
3624 | /* | |
3625 | * BUG, deal with case where j_len is 0, but people previously | |
3626 | * freed blocks need to be released will be dealt with by next | |
3627 | * transaction that actually writes something, but should be taken | |
3628 | * care of in this trans | |
3629 | */ | |
3630 | BUG_ON(journal->j_len == 0); | |
3631 | ||
3632 | /* | |
3633 | * if wcount > 0, and we are called to with flush or commit_now, | |
3634 | * we wait on j_join_wait. We will wake up when the last writer has | |
3635 | * finished the transaction, and started it on its way to the disk. | |
3636 | * Then, we flush the commit or journal list, and just return 0 | |
3637 | * because the rest of journal end was already done for this | |
3638 | * transaction. | |
3639 | */ | |
3640 | if (atomic_read(&journal->j_wcount) > 0) { | |
3641 | if (flush || commit_now) { | |
3642 | unsigned trans_id; | |
3643 | ||
3644 | jl = journal->j_current_jl; | |
3645 | trans_id = jl->j_trans_id; | |
3646 | if (wait_on_commit) | |
3647 | jl->j_state |= LIST_COMMIT_PENDING; | |
3648 | atomic_set(&journal->j_jlock, 1); | |
3649 | if (flush) { | |
3650 | journal->j_next_full_flush = 1; | |
3651 | } | |
3652 | unlock_journal(sb); | |
3653 | ||
3654 | /* | |
3655 | * sleep while the current transaction is | |
3656 | * still j_jlocked | |
3657 | */ | |
3658 | while (journal->j_trans_id == trans_id) { | |
3659 | if (atomic_read(&journal->j_jlock)) { | |
3660 | queue_log_writer(sb); | |
3661 | } else { | |
3662 | lock_journal(sb); | |
3663 | if (journal->j_trans_id == trans_id) { | |
3664 | atomic_set(&journal->j_jlock, | |
3665 | 1); | |
3666 | } | |
3667 | unlock_journal(sb); | |
3668 | } | |
3669 | } | |
3670 | BUG_ON(journal->j_trans_id == trans_id); | |
3671 | ||
3672 | if (commit_now | |
3673 | && journal_list_still_alive(sb, trans_id) | |
3674 | && wait_on_commit) { | |
3675 | flush_commit_list(sb, jl, 1); | |
3676 | } | |
3677 | return 0; | |
3678 | } | |
3679 | unlock_journal(sb); | |
3680 | return 0; | |
3681 | } | |
3682 | ||
3683 | /* deal with old transactions where we are the last writers */ | |
3684 | now = ktime_get_seconds(); | |
3685 | if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) { | |
3686 | commit_now = 1; | |
3687 | journal->j_next_async_flush = 1; | |
3688 | } | |
3689 | /* don't batch when someone is waiting on j_join_wait */ | |
3690 | /* don't batch when syncing the commit or flushing the whole trans */ | |
3691 | if (!(journal->j_must_wait > 0) && !(atomic_read(&journal->j_jlock)) | |
3692 | && !flush && !commit_now && (journal->j_len < journal->j_max_batch) | |
3693 | && journal->j_len_alloc < journal->j_max_batch | |
3694 | && journal->j_cnode_free > (journal->j_trans_max * 3)) { | |
3695 | journal->j_bcount++; | |
3696 | unlock_journal(sb); | |
3697 | return 0; | |
3698 | } | |
3699 | ||
3700 | if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(sb)) { | |
3701 | reiserfs_panic(sb, "journal-003", | |
3702 | "j_start (%ld) is too high", | |
3703 | journal->j_start); | |
3704 | } | |
3705 | return 1; | |
3706 | } | |
3707 | ||
3708 | /* | |
3709 | * Does all the work that makes deleting blocks safe. | |
3710 | * when deleting a block mark BH_JNew, just remove it from the current | |
3711 | * transaction, clean it's buffer_head and move on. | |
3712 | * | |
3713 | * otherwise: | |
3714 | * set a bit for the block in the journal bitmap. That will prevent it from | |
3715 | * being allocated for unformatted nodes before this transaction has finished. | |
3716 | * | |
3717 | * mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers. | |
3718 | * That will prevent any old transactions with this block from trying to flush | |
3719 | * to the real location. Since we aren't removing the cnode from the | |
3720 | * journal_list_hash, *the block can't be reallocated yet. | |
3721 | * | |
3722 | * Then remove it from the current transaction, decrementing any counters and | |
3723 | * filing it on the clean list. | |
3724 | */ | |
3725 | int journal_mark_freed(struct reiserfs_transaction_handle *th, | |
3726 | struct super_block *sb, b_blocknr_t blocknr) | |
3727 | { | |
3728 | struct reiserfs_journal *journal = SB_JOURNAL(sb); | |
3729 | struct reiserfs_journal_cnode *cn = NULL; | |
3730 | struct buffer_head *bh = NULL; | |
3731 | struct reiserfs_list_bitmap *jb = NULL; | |
3732 | int cleaned = 0; | |
3733 | BUG_ON(!th->t_trans_id); | |
3734 | ||
3735 | cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr); | |
3736 | if (cn && cn->bh) { | |
3737 | bh = cn->bh; | |
3738 | get_bh(bh); | |
3739 | } | |
3740 | /* if it is journal new, we just remove it from this transaction */ | |
3741 | if (bh && buffer_journal_new(bh)) { | |
3742 | clear_buffer_journal_new(bh); | |
3743 | clear_prepared_bits(bh); | |
3744 | reiserfs_clean_and_file_buffer(bh); | |
3745 | cleaned = remove_from_transaction(sb, blocknr, cleaned); | |
3746 | } else { | |
3747 | /* | |
3748 | * set the bit for this block in the journal bitmap | |
3749 | * for this transaction | |
3750 | */ | |
3751 | jb = journal->j_current_jl->j_list_bitmap; | |
3752 | if (!jb) { | |
3753 | reiserfs_panic(sb, "journal-1702", | |
3754 | "journal_list_bitmap is NULL"); | |
3755 | } | |
3756 | set_bit_in_list_bitmap(sb, blocknr, jb); | |
3757 | ||
3758 | /* Note, the entire while loop is not allowed to schedule. */ | |
3759 | ||
3760 | if (bh) { | |
3761 | clear_prepared_bits(bh); | |
3762 | reiserfs_clean_and_file_buffer(bh); | |
3763 | } | |
3764 | cleaned = remove_from_transaction(sb, blocknr, cleaned); | |
3765 | ||
3766 | /* | |
3767 | * find all older transactions with this block, | |
3768 | * make sure they don't try to write it out | |
3769 | */ | |
3770 | cn = get_journal_hash_dev(sb, journal->j_list_hash_table, | |
3771 | blocknr); | |
3772 | while (cn) { | |
3773 | if (sb == cn->sb && blocknr == cn->blocknr) { | |
3774 | set_bit(BLOCK_FREED, &cn->state); | |
3775 | if (cn->bh) { | |
3776 | /* | |
3777 | * remove_from_transaction will brelse | |
3778 | * the buffer if it was in the current | |
3779 | * trans | |
3780 | */ | |
3781 | if (!cleaned) { | |
3782 | clear_buffer_journal_dirty(cn-> | |
3783 | bh); | |
3784 | clear_buffer_dirty(cn->bh); | |
3785 | clear_buffer_journal_test(cn-> | |
3786 | bh); | |
3787 | cleaned = 1; | |
3788 | put_bh(cn->bh); | |
3789 | if (atomic_read | |
3790 | (&cn->bh->b_count) < 0) { | |
3791 | reiserfs_warning(sb, | |
3792 | "journal-2138", | |
3793 | "cn->bh->b_count < 0"); | |
3794 | } | |
3795 | } | |
3796 | /* | |
3797 | * since we are clearing the bh, | |
3798 | * we MUST dec nonzerolen | |
3799 | */ | |
3800 | if (cn->jlist) { | |
3801 | atomic_dec(&cn->jlist-> | |
3802 | j_nonzerolen); | |
3803 | } | |
3804 | cn->bh = NULL; | |
3805 | } | |
3806 | } | |
3807 | cn = cn->hnext; | |
3808 | } | |
3809 | } | |
3810 | ||
3811 | if (bh) | |
3812 | release_buffer_page(bh); /* get_hash grabs the buffer */ | |
3813 | return 0; | |
3814 | } | |
3815 | ||
3816 | void reiserfs_update_inode_transaction(struct inode *inode) | |
3817 | { | |
3818 | struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb); | |
3819 | REISERFS_I(inode)->i_jl = journal->j_current_jl; | |
3820 | REISERFS_I(inode)->i_trans_id = journal->j_trans_id; | |
3821 | } | |
3822 | ||
3823 | /* | |
3824 | * returns -1 on error, 0 if no commits/barriers were done and 1 | |
3825 | * if a transaction was actually committed and the barrier was done | |
3826 | */ | |
3827 | static int __commit_trans_jl(struct inode *inode, unsigned long id, | |
3828 | struct reiserfs_journal_list *jl) | |
3829 | { | |
3830 | struct reiserfs_transaction_handle th; | |
3831 | struct super_block *sb = inode->i_sb; | |
3832 | struct reiserfs_journal *journal = SB_JOURNAL(sb); | |
3833 | int ret = 0; | |
3834 | ||
3835 | /* | |
3836 | * is it from the current transaction, | |
3837 | * or from an unknown transaction? | |
3838 | */ | |
3839 | if (id == journal->j_trans_id) { | |
3840 | jl = journal->j_current_jl; | |
3841 | /* | |
3842 | * try to let other writers come in and | |
3843 | * grow this transaction | |
3844 | */ | |
3845 | let_transaction_grow(sb, id); | |
3846 | if (journal->j_trans_id != id) { | |
3847 | goto flush_commit_only; | |
3848 | } | |
3849 | ||
3850 | ret = journal_begin(&th, sb, 1); | |
3851 | if (ret) | |
3852 | return ret; | |
3853 | ||
3854 | /* someone might have ended this transaction while we joined */ | |
3855 | if (journal->j_trans_id != id) { | |
3856 | reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb), | |
3857 | 1); | |
3858 | journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb)); | |
3859 | ret = journal_end(&th); | |
3860 | goto flush_commit_only; | |
3861 | } | |
3862 | ||
3863 | ret = journal_end_sync(&th); | |
3864 | if (!ret) | |
3865 | ret = 1; | |
3866 | ||
3867 | } else { | |
3868 | /* | |
3869 | * this gets tricky, we have to make sure the journal list in | |
3870 | * the inode still exists. We know the list is still around | |
3871 | * if we've got a larger transaction id than the oldest list | |
3872 | */ | |
3873 | flush_commit_only: | |
3874 | if (journal_list_still_alive(inode->i_sb, id)) { | |
3875 | /* | |
3876 | * we only set ret to 1 when we know for sure | |
3877 | * the barrier hasn't been started yet on the commit | |
3878 | * block. | |
3879 | */ | |
3880 | if (atomic_read(&jl->j_commit_left) > 1) | |
3881 | ret = 1; | |
3882 | flush_commit_list(sb, jl, 1); | |
3883 | if (journal->j_errno) | |
3884 | ret = journal->j_errno; | |
3885 | } | |
3886 | } | |
3887 | /* otherwise the list is gone, and long since committed */ | |
3888 | return ret; | |
3889 | } | |
3890 | ||
3891 | int reiserfs_commit_for_inode(struct inode *inode) | |
3892 | { | |
3893 | unsigned int id = REISERFS_I(inode)->i_trans_id; | |
3894 | struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl; | |
3895 | ||
3896 | /* | |
3897 | * for the whole inode, assume unset id means it was | |
3898 | * changed in the current transaction. More conservative | |
3899 | */ | |
3900 | if (!id || !jl) { | |
3901 | reiserfs_update_inode_transaction(inode); | |
3902 | id = REISERFS_I(inode)->i_trans_id; | |
3903 | /* jl will be updated in __commit_trans_jl */ | |
3904 | } | |
3905 | ||
3906 | return __commit_trans_jl(inode, id, jl); | |
3907 | } | |
3908 | ||
3909 | void reiserfs_restore_prepared_buffer(struct super_block *sb, | |
3910 | struct buffer_head *bh) | |
3911 | { | |
3912 | struct reiserfs_journal *journal = SB_JOURNAL(sb); | |
3913 | PROC_INFO_INC(sb, journal.restore_prepared); | |
3914 | if (!bh) { | |
3915 | return; | |
3916 | } | |
3917 | if (test_clear_buffer_journal_restore_dirty(bh) && | |
3918 | buffer_journal_dirty(bh)) { | |
3919 | struct reiserfs_journal_cnode *cn; | |
3920 | reiserfs_write_lock(sb); | |
3921 | cn = get_journal_hash_dev(sb, | |
3922 | journal->j_list_hash_table, | |
3923 | bh->b_blocknr); | |
3924 | if (cn && can_dirty(cn)) { | |
3925 | set_buffer_journal_test(bh); | |
3926 | mark_buffer_dirty(bh); | |
3927 | } | |
3928 | reiserfs_write_unlock(sb); | |
3929 | } | |
3930 | clear_buffer_journal_prepared(bh); | |
3931 | } | |
3932 | ||
3933 | extern struct tree_balance *cur_tb; | |
3934 | /* | |
3935 | * before we can change a metadata block, we have to make sure it won't | |
3936 | * be written to disk while we are altering it. So, we must: | |
3937 | * clean it | |
3938 | * wait on it. | |
3939 | */ | |
3940 | int reiserfs_prepare_for_journal(struct super_block *sb, | |
3941 | struct buffer_head *bh, int wait) | |
3942 | { | |
3943 | PROC_INFO_INC(sb, journal.prepare); | |
3944 | ||
3945 | if (!trylock_buffer(bh)) { | |
3946 | if (!wait) | |
3947 | return 0; | |
3948 | lock_buffer(bh); | |
3949 | } | |
3950 | set_buffer_journal_prepared(bh); | |
3951 | if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) { | |
3952 | clear_buffer_journal_test(bh); | |
3953 | set_buffer_journal_restore_dirty(bh); | |
3954 | } | |
3955 | unlock_buffer(bh); | |
3956 | return 1; | |
3957 | } | |
3958 | ||
3959 | /* | |
3960 | * long and ugly. If flush, will not return until all commit | |
3961 | * blocks and all real buffers in the trans are on disk. | |
3962 | * If no_async, won't return until all commit blocks are on disk. | |
3963 | * | |
3964 | * keep reading, there are comments as you go along | |
3965 | * | |
3966 | * If the journal is aborted, we just clean up. Things like flushing | |
3967 | * journal lists, etc just won't happen. | |
3968 | */ | |
3969 | static int do_journal_end(struct reiserfs_transaction_handle *th, int flags) | |
3970 | { | |
3971 | struct super_block *sb = th->t_super; | |
3972 | struct reiserfs_journal *journal = SB_JOURNAL(sb); | |
3973 | struct reiserfs_journal_cnode *cn, *next, *jl_cn; | |
3974 | struct reiserfs_journal_cnode *last_cn = NULL; | |
3975 | struct reiserfs_journal_desc *desc; | |
3976 | struct reiserfs_journal_commit *commit; | |
3977 | struct buffer_head *c_bh; /* commit bh */ | |
3978 | struct buffer_head *d_bh; /* desc bh */ | |
3979 | int cur_write_start = 0; /* start index of current log write */ | |
3980 | int i; | |
3981 | int flush; | |
3982 | int wait_on_commit; | |
3983 | struct reiserfs_journal_list *jl, *temp_jl; | |
3984 | struct list_head *entry, *safe; | |
3985 | unsigned long jindex; | |
3986 | unsigned int commit_trans_id; | |
3987 | int trans_half; | |
3988 | int depth; | |
3989 | ||
3990 | BUG_ON(th->t_refcount > 1); | |
3991 | BUG_ON(!th->t_trans_id); | |
3992 | BUG_ON(!th->t_super); | |
3993 | ||
3994 | /* | |
3995 | * protect flush_older_commits from doing mistakes if the | |
3996 | * transaction ID counter gets overflowed. | |
3997 | */ | |
3998 | if (th->t_trans_id == ~0U) | |
3999 | flags |= FLUSH_ALL | COMMIT_NOW | WAIT; | |
4000 | flush = flags & FLUSH_ALL; | |
4001 | wait_on_commit = flags & WAIT; | |
4002 | ||
4003 | current->journal_info = th->t_handle_save; | |
4004 | reiserfs_check_lock_depth(sb, "journal end"); | |
4005 | if (journal->j_len == 0) { | |
4006 | reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb), | |
4007 | 1); | |
4008 | journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb)); | |
4009 | } | |
4010 | ||
4011 | lock_journal(sb); | |
4012 | if (journal->j_next_full_flush) { | |
4013 | flags |= FLUSH_ALL; | |
4014 | flush = 1; | |
4015 | } | |
4016 | if (journal->j_next_async_flush) { | |
4017 | flags |= COMMIT_NOW | WAIT; | |
4018 | wait_on_commit = 1; | |
4019 | } | |
4020 | ||
4021 | /* | |
4022 | * check_journal_end locks the journal, and unlocks if it does | |
4023 | * not return 1 it tells us if we should continue with the | |
4024 | * journal_end, or just return | |
4025 | */ | |
4026 | if (!check_journal_end(th, flags)) { | |
4027 | reiserfs_schedule_old_flush(sb); | |
4028 | wake_queued_writers(sb); | |
4029 | reiserfs_async_progress_wait(sb); | |
4030 | goto out; | |
4031 | } | |
4032 | ||
4033 | /* check_journal_end might set these, check again */ | |
4034 | if (journal->j_next_full_flush) { | |
4035 | flush = 1; | |
4036 | } | |
4037 | ||
4038 | /* | |
4039 | * j must wait means we have to flush the log blocks, and the | |
4040 | * real blocks for this transaction | |
4041 | */ | |
4042 | if (journal->j_must_wait > 0) { | |
4043 | flush = 1; | |
4044 | } | |
4045 | #ifdef REISERFS_PREALLOCATE | |
4046 | /* | |
4047 | * quota ops might need to nest, setup the journal_info pointer | |
4048 | * for them and raise the refcount so that it is > 0. | |
4049 | */ | |
4050 | current->journal_info = th; | |
4051 | th->t_refcount++; | |
4052 | ||
4053 | /* it should not involve new blocks into the transaction */ | |
4054 | reiserfs_discard_all_prealloc(th); | |
4055 | ||
4056 | th->t_refcount--; | |
4057 | current->journal_info = th->t_handle_save; | |
4058 | #endif | |
4059 | ||
4060 | /* setup description block */ | |
4061 | d_bh = | |
4062 | journal_getblk(sb, | |
4063 | SB_ONDISK_JOURNAL_1st_BLOCK(sb) + | |
4064 | journal->j_start); | |
4065 | set_buffer_uptodate(d_bh); | |
4066 | desc = (struct reiserfs_journal_desc *)(d_bh)->b_data; | |
4067 | memset(d_bh->b_data, 0, d_bh->b_size); | |
4068 | memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8); | |
4069 | set_desc_trans_id(desc, journal->j_trans_id); | |
4070 | ||
4071 | /* | |
4072 | * setup commit block. Don't write (keep it clean too) this one | |
4073 | * until after everyone else is written | |
4074 | */ | |
4075 | c_bh = journal_getblk(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) + | |
4076 | ((journal->j_start + journal->j_len + | |
4077 | 1) % SB_ONDISK_JOURNAL_SIZE(sb))); | |
4078 | commit = (struct reiserfs_journal_commit *)c_bh->b_data; | |
4079 | memset(c_bh->b_data, 0, c_bh->b_size); | |
4080 | set_commit_trans_id(commit, journal->j_trans_id); | |
4081 | set_buffer_uptodate(c_bh); | |
4082 | ||
4083 | /* init this journal list */ | |
4084 | jl = journal->j_current_jl; | |
4085 | ||
4086 | /* | |
4087 | * we lock the commit before doing anything because | |
4088 | * we want to make sure nobody tries to run flush_commit_list until | |
4089 | * the new transaction is fully setup, and we've already flushed the | |
4090 | * ordered bh list | |
4091 | */ | |
4092 | reiserfs_mutex_lock_safe(&jl->j_commit_mutex, sb); | |
4093 | ||
4094 | /* save the transaction id in case we need to commit it later */ | |
4095 | commit_trans_id = jl->j_trans_id; | |
4096 | ||
4097 | atomic_set(&jl->j_older_commits_done, 0); | |
4098 | jl->j_trans_id = journal->j_trans_id; | |
4099 | jl->j_timestamp = journal->j_trans_start_time; | |
4100 | jl->j_commit_bh = c_bh; | |
4101 | jl->j_start = journal->j_start; | |
4102 | jl->j_len = journal->j_len; | |
4103 | atomic_set(&jl->j_nonzerolen, journal->j_len); | |
4104 | atomic_set(&jl->j_commit_left, journal->j_len + 2); | |
4105 | jl->j_realblock = NULL; | |
4106 | ||
4107 | /* | |
4108 | * The ENTIRE FOR LOOP MUST not cause schedule to occur. | |
4109 | * for each real block, add it to the journal list hash, | |
4110 | * copy into real block index array in the commit or desc block | |
4111 | */ | |
4112 | trans_half = journal_trans_half(sb->s_blocksize); | |
4113 | for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) { | |
4114 | if (buffer_journaled(cn->bh)) { | |
4115 | jl_cn = get_cnode(sb); | |
4116 | if (!jl_cn) { | |
4117 | reiserfs_panic(sb, "journal-1676", | |
4118 | "get_cnode returned NULL"); | |
4119 | } | |
4120 | if (i == 0) { | |
4121 | jl->j_realblock = jl_cn; | |
4122 | } | |
4123 | jl_cn->prev = last_cn; | |
4124 | jl_cn->next = NULL; | |
4125 | if (last_cn) { | |
4126 | last_cn->next = jl_cn; | |
4127 | } | |
4128 | last_cn = jl_cn; | |
4129 | /* | |
4130 | * make sure the block we are trying to log | |
4131 | * is not a block of journal or reserved area | |
4132 | */ | |
4133 | if (is_block_in_log_or_reserved_area | |
4134 | (sb, cn->bh->b_blocknr)) { | |
4135 | reiserfs_panic(sb, "journal-2332", | |
4136 | "Trying to log block %lu, " | |
4137 | "which is a log block", | |
4138 | cn->bh->b_blocknr); | |
4139 | } | |
4140 | jl_cn->blocknr = cn->bh->b_blocknr; | |
4141 | jl_cn->state = 0; | |
4142 | jl_cn->sb = sb; | |
4143 | jl_cn->bh = cn->bh; | |
4144 | jl_cn->jlist = jl; | |
4145 | insert_journal_hash(journal->j_list_hash_table, jl_cn); | |
4146 | if (i < trans_half) { | |
4147 | desc->j_realblock[i] = | |
4148 | cpu_to_le32(cn->bh->b_blocknr); | |
4149 | } else { | |
4150 | commit->j_realblock[i - trans_half] = | |
4151 | cpu_to_le32(cn->bh->b_blocknr); | |
4152 | } | |
4153 | } else { | |
4154 | i--; | |
4155 | } | |
4156 | } | |
4157 | set_desc_trans_len(desc, journal->j_len); | |
4158 | set_desc_mount_id(desc, journal->j_mount_id); | |
4159 | set_desc_trans_id(desc, journal->j_trans_id); | |
4160 | set_commit_trans_len(commit, journal->j_len); | |
4161 | ||
4162 | /* | |
4163 | * special check in case all buffers in the journal | |
4164 | * were marked for not logging | |
4165 | */ | |
4166 | BUG_ON(journal->j_len == 0); | |
4167 | ||
4168 | /* | |
4169 | * we're about to dirty all the log blocks, mark the description block | |
4170 | * dirty now too. Don't mark the commit block dirty until all the | |
4171 | * others are on disk | |
4172 | */ | |
4173 | mark_buffer_dirty(d_bh); | |
4174 | ||
4175 | /* | |
4176 | * first data block is j_start + 1, so add one to | |
4177 | * cur_write_start wherever you use it | |
4178 | */ | |
4179 | cur_write_start = journal->j_start; | |
4180 | cn = journal->j_first; | |
4181 | jindex = 1; /* start at one so we don't get the desc again */ | |
4182 | while (cn) { | |
4183 | clear_buffer_journal_new(cn->bh); | |
4184 | /* copy all the real blocks into log area. dirty log blocks */ | |
4185 | if (buffer_journaled(cn->bh)) { | |
4186 | struct buffer_head *tmp_bh; | |
4187 | char *addr; | |
4188 | struct page *page; | |
4189 | tmp_bh = | |
4190 | journal_getblk(sb, | |
4191 | SB_ONDISK_JOURNAL_1st_BLOCK(sb) + | |
4192 | ((cur_write_start + | |
4193 | jindex) % | |
4194 | SB_ONDISK_JOURNAL_SIZE(sb))); | |
4195 | set_buffer_uptodate(tmp_bh); | |
4196 | page = cn->bh->b_page; | |
4197 | addr = kmap(page); | |
4198 | memcpy(tmp_bh->b_data, | |
4199 | addr + offset_in_page(cn->bh->b_data), | |
4200 | cn->bh->b_size); | |
4201 | kunmap(page); | |
4202 | mark_buffer_dirty(tmp_bh); | |
4203 | jindex++; | |
4204 | set_buffer_journal_dirty(cn->bh); | |
4205 | clear_buffer_journaled(cn->bh); | |
4206 | } else { | |
4207 | /* | |
4208 | * JDirty cleared sometime during transaction. | |
4209 | * don't log this one | |
4210 | */ | |
4211 | reiserfs_warning(sb, "journal-2048", | |
4212 | "BAD, buffer in journal hash, " | |
4213 | "but not JDirty!"); | |
4214 | brelse(cn->bh); | |
4215 | } | |
4216 | next = cn->next; | |
4217 | free_cnode(sb, cn); | |
4218 | cn = next; | |
4219 | reiserfs_cond_resched(sb); | |
4220 | } | |
4221 | ||
4222 | /* | |
4223 | * we are done with both the c_bh and d_bh, but | |
4224 | * c_bh must be written after all other commit blocks, | |
4225 | * so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1. | |
4226 | */ | |
4227 | ||
4228 | journal->j_current_jl = alloc_journal_list(sb); | |
4229 | ||
4230 | /* now it is safe to insert this transaction on the main list */ | |
4231 | list_add_tail(&jl->j_list, &journal->j_journal_list); | |
4232 | list_add_tail(&jl->j_working_list, &journal->j_working_list); | |
4233 | journal->j_num_work_lists++; | |
4234 | ||
4235 | /* reset journal values for the next transaction */ | |
4236 | journal->j_start = | |
4237 | (journal->j_start + journal->j_len + | |
4238 | 2) % SB_ONDISK_JOURNAL_SIZE(sb); | |
4239 | atomic_set(&journal->j_wcount, 0); | |
4240 | journal->j_bcount = 0; | |
4241 | journal->j_last = NULL; | |
4242 | journal->j_first = NULL; | |
4243 | journal->j_len = 0; | |
4244 | journal->j_trans_start_time = 0; | |
4245 | /* check for trans_id overflow */ | |
4246 | if (++journal->j_trans_id == 0) | |
4247 | journal->j_trans_id = 10; | |
4248 | journal->j_current_jl->j_trans_id = journal->j_trans_id; | |
4249 | journal->j_must_wait = 0; | |
4250 | journal->j_len_alloc = 0; | |
4251 | journal->j_next_full_flush = 0; | |
4252 | journal->j_next_async_flush = 0; | |
4253 | init_journal_hash(sb); | |
4254 | ||
4255 | /* | |
4256 | * make sure reiserfs_add_jh sees the new current_jl before we | |
4257 | * write out the tails | |
4258 | */ | |
4259 | smp_mb(); | |
4260 | ||
4261 | /* | |
4262 | * tail conversion targets have to hit the disk before we end the | |
4263 | * transaction. Otherwise a later transaction might repack the tail | |
4264 | * before this transaction commits, leaving the data block unflushed | |
4265 | * and clean, if we crash before the later transaction commits, the | |
4266 | * data block is lost. | |
4267 | */ | |
4268 | if (!list_empty(&jl->j_tail_bh_list)) { | |
4269 | depth = reiserfs_write_unlock_nested(sb); | |
4270 | write_ordered_buffers(&journal->j_dirty_buffers_lock, | |
4271 | journal, jl, &jl->j_tail_bh_list); | |
4272 | reiserfs_write_lock_nested(sb, depth); | |
4273 | } | |
4274 | BUG_ON(!list_empty(&jl->j_tail_bh_list)); | |
4275 | mutex_unlock(&jl->j_commit_mutex); | |
4276 | ||
4277 | /* | |
4278 | * honor the flush wishes from the caller, simple commits can | |
4279 | * be done outside the journal lock, they are done below | |
4280 | * | |
4281 | * if we don't flush the commit list right now, we put it into | |
4282 | * the work queue so the people waiting on the async progress work | |
4283 | * queue don't wait for this proc to flush journal lists and such. | |
4284 | */ | |
4285 | if (flush) { | |
4286 | flush_commit_list(sb, jl, 1); | |
4287 | flush_journal_list(sb, jl, 1); | |
4288 | } else if (!(jl->j_state & LIST_COMMIT_PENDING)) { | |
4289 | /* | |
4290 | * Avoid queueing work when sb is being shut down. Transaction | |
4291 | * will be flushed on journal shutdown. | |
4292 | */ | |
4293 | if (sb->s_flags & SB_ACTIVE) | |
4294 | queue_delayed_work(REISERFS_SB(sb)->commit_wq, | |
4295 | &journal->j_work, HZ / 10); | |
4296 | } | |
4297 | ||
4298 | /* | |
4299 | * if the next transaction has any chance of wrapping, flush | |
4300 | * transactions that might get overwritten. If any journal lists | |
4301 | * are very old flush them as well. | |
4302 | */ | |
4303 | first_jl: | |
4304 | list_for_each_safe(entry, safe, &journal->j_journal_list) { | |
4305 | temp_jl = JOURNAL_LIST_ENTRY(entry); | |
4306 | if (journal->j_start <= temp_jl->j_start) { | |
4307 | if ((journal->j_start + journal->j_trans_max + 1) >= | |
4308 | temp_jl->j_start) { | |
4309 | flush_used_journal_lists(sb, temp_jl); | |
4310 | goto first_jl; | |
4311 | } else if ((journal->j_start + | |
4312 | journal->j_trans_max + 1) < | |
4313 | SB_ONDISK_JOURNAL_SIZE(sb)) { | |
4314 | /* | |
4315 | * if we don't cross into the next | |
4316 | * transaction and we don't wrap, there is | |
4317 | * no way we can overlap any later transactions | |
4318 | * break now | |
4319 | */ | |
4320 | break; | |
4321 | } | |
4322 | } else if ((journal->j_start + | |
4323 | journal->j_trans_max + 1) > | |
4324 | SB_ONDISK_JOURNAL_SIZE(sb)) { | |
4325 | if (((journal->j_start + journal->j_trans_max + 1) % | |
4326 | SB_ONDISK_JOURNAL_SIZE(sb)) >= | |
4327 | temp_jl->j_start) { | |
4328 | flush_used_journal_lists(sb, temp_jl); | |
4329 | goto first_jl; | |
4330 | } else { | |
4331 | /* | |
4332 | * we don't overlap anything from out start | |
4333 | * to the end of the log, and our wrapped | |
4334 | * portion doesn't overlap anything at | |
4335 | * the start of the log. We can break | |
4336 | */ | |
4337 | break; | |
4338 | } | |
4339 | } | |
4340 | } | |
4341 | ||
4342 | journal->j_current_jl->j_list_bitmap = | |
4343 | get_list_bitmap(sb, journal->j_current_jl); | |
4344 | ||
4345 | if (!(journal->j_current_jl->j_list_bitmap)) { | |
4346 | reiserfs_panic(sb, "journal-1996", | |
4347 | "could not get a list bitmap"); | |
4348 | } | |
4349 | ||
4350 | atomic_set(&journal->j_jlock, 0); | |
4351 | unlock_journal(sb); | |
4352 | /* wake up any body waiting to join. */ | |
4353 | clear_bit(J_WRITERS_QUEUED, &journal->j_state); | |
4354 | wake_up(&journal->j_join_wait); | |
4355 | ||
4356 | if (!flush && wait_on_commit && | |
4357 | journal_list_still_alive(sb, commit_trans_id)) { | |
4358 | flush_commit_list(sb, jl, 1); | |
4359 | } | |
4360 | out: | |
4361 | reiserfs_check_lock_depth(sb, "journal end2"); | |
4362 | ||
4363 | memset(th, 0, sizeof(*th)); | |
4364 | /* | |
4365 | * Re-set th->t_super, so we can properly keep track of how many | |
4366 | * persistent transactions there are. We need to do this so if this | |
4367 | * call is part of a failed restart_transaction, we can free it later | |
4368 | */ | |
4369 | th->t_super = sb; | |
4370 | ||
4371 | return journal->j_errno; | |
4372 | } | |
4373 | ||
4374 | /* Send the file system read only and refuse new transactions */ | |
4375 | void reiserfs_abort_journal(struct super_block *sb, int errno) | |
4376 | { | |
4377 | struct reiserfs_journal *journal = SB_JOURNAL(sb); | |
4378 | if (test_bit(J_ABORTED, &journal->j_state)) | |
4379 | return; | |
4380 | ||
4381 | if (!journal->j_errno) | |
4382 | journal->j_errno = errno; | |
4383 | ||
4384 | sb->s_flags |= SB_RDONLY; | |
4385 | set_bit(J_ABORTED, &journal->j_state); | |
4386 | ||
4387 | #ifdef CONFIG_REISERFS_CHECK | |
4388 | dump_stack(); | |
4389 | #endif | |
4390 | } |