Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | ** Write ahead logging implementation copyright Chris Mason 2000 | |
3 | ** | |
0222e657 | 4 | ** The background commits make this code very interelated, and |
1da177e4 LT |
5 | ** overly complex. I need to rethink things a bit....The major players: |
6 | ** | |
0222e657 | 7 | ** journal_begin -- call with the number of blocks you expect to log. |
1da177e4 | 8 | ** If the current transaction is too |
0222e657 | 9 | ** old, it will block until the current transaction is |
1da177e4 | 10 | ** finished, and then start a new one. |
0222e657 | 11 | ** Usually, your transaction will get joined in with |
1da177e4 LT |
12 | ** previous ones for speed. |
13 | ** | |
0222e657 | 14 | ** journal_join -- same as journal_begin, but won't block on the current |
1da177e4 | 15 | ** transaction regardless of age. Don't ever call |
0222e657 | 16 | ** this. Ever. There are only two places it should be |
1da177e4 LT |
17 | ** called from, and they are both inside this file. |
18 | ** | |
0222e657 | 19 | ** journal_mark_dirty -- adds blocks into this transaction. clears any flags |
1da177e4 | 20 | ** that might make them get sent to disk |
0222e657 JM |
21 | ** and then marks them BH_JDirty. Puts the buffer head |
22 | ** into the current transaction hash. | |
1da177e4 LT |
23 | ** |
24 | ** journal_end -- if the current transaction is batchable, it does nothing | |
25 | ** otherwise, it could do an async/synchronous commit, or | |
0222e657 | 26 | ** a full flush of all log and real blocks in the |
1da177e4 LT |
27 | ** transaction. |
28 | ** | |
0222e657 JM |
29 | ** flush_old_commits -- if the current transaction is too old, it is ended and |
30 | ** commit blocks are sent to disk. Forces commit blocks | |
31 | ** to disk for all backgrounded commits that have been | |
1da177e4 | 32 | ** around too long. |
0222e657 | 33 | ** -- Note, if you call this as an immediate flush from |
1da177e4 LT |
34 | ** from within kupdate, it will ignore the immediate flag |
35 | */ | |
36 | ||
1da177e4 | 37 | #include <linux/time.h> |
6188e10d | 38 | #include <linux/semaphore.h> |
1da177e4 LT |
39 | #include <linux/vmalloc.h> |
40 | #include <linux/reiserfs_fs.h> | |
1da177e4 LT |
41 | #include <linux/kernel.h> |
42 | #include <linux/errno.h> | |
43 | #include <linux/fcntl.h> | |
44 | #include <linux/stat.h> | |
45 | #include <linux/string.h> | |
46 | #include <linux/smp_lock.h> | |
47 | #include <linux/buffer_head.h> | |
48 | #include <linux/workqueue.h> | |
49 | #include <linux/writeback.h> | |
50 | #include <linux/blkdev.h> | |
3fcfab16 | 51 | #include <linux/backing-dev.h> |
90415dea JM |
52 | #include <linux/uaccess.h> |
53 | ||
54 | #include <asm/system.h> | |
1da177e4 | 55 | |
1da177e4 LT |
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 | #define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \ | |
60 | j_working_list)) | |
61 | ||
62 | /* the number of mounted filesystems. This is used to decide when to | |
63 | ** start and kill the commit workqueue | |
64 | */ | |
65 | static int reiserfs_mounted_fs_count; | |
66 | ||
67 | static struct workqueue_struct *commit_wq; | |
68 | ||
bd4c625c LT |
69 | #define JOURNAL_TRANS_HALF 1018 /* must be correct to keep the desc and commit |
70 | structs at 4k */ | |
71 | #define BUFNR 64 /*read ahead */ | |
1da177e4 LT |
72 | |
73 | /* cnode stat bits. Move these into reiserfs_fs.h */ | |
74 | ||
75 | #define BLOCK_FREED 2 /* this block was freed, and can't be written. */ | |
bd4c625c | 76 | #define BLOCK_FREED_HOLDER 3 /* this block was freed during this transaction, and can't be written */ |
1da177e4 LT |
77 | |
78 | #define BLOCK_NEEDS_FLUSH 4 /* used in flush_journal_list */ | |
79 | #define BLOCK_DIRTIED 5 | |
80 | ||
1da177e4 LT |
81 | /* journal list state bits */ |
82 | #define LIST_TOUCHED 1 | |
83 | #define LIST_DIRTY 2 | |
bd4c625c | 84 | #define LIST_COMMIT_PENDING 4 /* someone will commit this list */ |
1da177e4 LT |
85 | |
86 | /* flags for do_journal_end */ | |
87 | #define FLUSH_ALL 1 /* flush commit and real blocks */ | |
88 | #define COMMIT_NOW 2 /* end and commit this transaction */ | |
bd4c625c LT |
89 | #define WAIT 4 /* wait for the log blocks to hit the disk */ |
90 | ||
91 | static int do_journal_end(struct reiserfs_transaction_handle *, | |
92 | struct super_block *, unsigned long nblocks, | |
93 | int flags); | |
94 | static int flush_journal_list(struct super_block *s, | |
95 | struct reiserfs_journal_list *jl, int flushall); | |
96 | static int flush_commit_list(struct super_block *s, | |
97 | struct reiserfs_journal_list *jl, int flushall); | |
98 | static int can_dirty(struct reiserfs_journal_cnode *cn); | |
99 | static int journal_join(struct reiserfs_transaction_handle *th, | |
a9dd3643 | 100 | struct super_block *sb, unsigned long nblocks); |
bd4c625c LT |
101 | static int release_journal_dev(struct super_block *super, |
102 | struct reiserfs_journal *journal); | |
1da177e4 | 103 | static int dirty_one_transaction(struct super_block *s, |
bd4c625c | 104 | struct reiserfs_journal_list *jl); |
c4028958 | 105 | static void flush_async_commits(struct work_struct *work); |
1da177e4 LT |
106 | static void queue_log_writer(struct super_block *s); |
107 | ||
108 | /* values for join in do_journal_begin_r */ | |
109 | enum { | |
bd4c625c LT |
110 | JBEGIN_REG = 0, /* regular journal begin */ |
111 | JBEGIN_JOIN = 1, /* join the running transaction if at all possible */ | |
112 | JBEGIN_ABORT = 2, /* called from cleanup code, ignores aborted flag */ | |
1da177e4 LT |
113 | }; |
114 | ||
115 | static int do_journal_begin_r(struct reiserfs_transaction_handle *th, | |
a9dd3643 | 116 | struct super_block *sb, |
bd4c625c | 117 | unsigned long nblocks, int join); |
1da177e4 | 118 | |
a9dd3643 | 119 | static void init_journal_hash(struct super_block *sb) |
bd4c625c | 120 | { |
a9dd3643 | 121 | struct reiserfs_journal *journal = SB_JOURNAL(sb); |
bd4c625c LT |
122 | memset(journal->j_hash_table, 0, |
123 | JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *)); | |
1da177e4 LT |
124 | } |
125 | ||
126 | /* | |
127 | ** clears BH_Dirty and sticks the buffer on the clean list. Called because I can't allow refile_buffer to | |
128 | ** make schedule happen after I've freed a block. Look at remove_from_transaction and journal_mark_freed for | |
129 | ** more details. | |
130 | */ | |
bd4c625c LT |
131 | static int reiserfs_clean_and_file_buffer(struct buffer_head *bh) |
132 | { | |
133 | if (bh) { | |
134 | clear_buffer_dirty(bh); | |
135 | clear_buffer_journal_test(bh); | |
136 | } | |
137 | return 0; | |
1da177e4 LT |
138 | } |
139 | ||
140 | static void disable_barrier(struct super_block *s) | |
141 | { | |
bd4c625c LT |
142 | REISERFS_SB(s)->s_mount_opt &= ~(1 << REISERFS_BARRIER_FLUSH); |
143 | printk("reiserfs: disabling flush barriers on %s\n", | |
144 | reiserfs_bdevname(s)); | |
145 | } | |
146 | ||
147 | static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block | |
a9dd3643 | 148 | *sb) |
bd4c625c LT |
149 | { |
150 | struct reiserfs_bitmap_node *bn; | |
151 | static int id; | |
152 | ||
d739b42b | 153 | bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS); |
bd4c625c LT |
154 | if (!bn) { |
155 | return NULL; | |
156 | } | |
a9dd3643 | 157 | bn->data = kzalloc(sb->s_blocksize, GFP_NOFS); |
bd4c625c | 158 | if (!bn->data) { |
d739b42b | 159 | kfree(bn); |
bd4c625c LT |
160 | return NULL; |
161 | } | |
162 | bn->id = id++; | |
bd4c625c LT |
163 | INIT_LIST_HEAD(&bn->list); |
164 | return bn; | |
165 | } | |
166 | ||
a9dd3643 | 167 | static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *sb) |
bd4c625c | 168 | { |
a9dd3643 | 169 | struct reiserfs_journal *journal = SB_JOURNAL(sb); |
bd4c625c LT |
170 | struct reiserfs_bitmap_node *bn = NULL; |
171 | struct list_head *entry = journal->j_bitmap_nodes.next; | |
172 | ||
173 | journal->j_used_bitmap_nodes++; | |
174 | repeat: | |
175 | ||
176 | if (entry != &journal->j_bitmap_nodes) { | |
177 | bn = list_entry(entry, struct reiserfs_bitmap_node, list); | |
178 | list_del(entry); | |
a9dd3643 | 179 | memset(bn->data, 0, sb->s_blocksize); |
bd4c625c LT |
180 | journal->j_free_bitmap_nodes--; |
181 | return bn; | |
182 | } | |
a9dd3643 | 183 | bn = allocate_bitmap_node(sb); |
bd4c625c LT |
184 | if (!bn) { |
185 | yield(); | |
186 | goto repeat; | |
187 | } | |
188 | return bn; | |
1da177e4 | 189 | } |
a9dd3643 | 190 | static inline void free_bitmap_node(struct super_block *sb, |
bd4c625c LT |
191 | struct reiserfs_bitmap_node *bn) |
192 | { | |
a9dd3643 | 193 | struct reiserfs_journal *journal = SB_JOURNAL(sb); |
bd4c625c LT |
194 | journal->j_used_bitmap_nodes--; |
195 | if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) { | |
d739b42b PE |
196 | kfree(bn->data); |
197 | kfree(bn); | |
bd4c625c LT |
198 | } else { |
199 | list_add(&bn->list, &journal->j_bitmap_nodes); | |
200 | journal->j_free_bitmap_nodes++; | |
201 | } | |
202 | } | |
203 | ||
a9dd3643 | 204 | static void allocate_bitmap_nodes(struct super_block *sb) |
bd4c625c LT |
205 | { |
206 | int i; | |
a9dd3643 | 207 | struct reiserfs_journal *journal = SB_JOURNAL(sb); |
bd4c625c LT |
208 | struct reiserfs_bitmap_node *bn = NULL; |
209 | for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) { | |
a9dd3643 | 210 | bn = allocate_bitmap_node(sb); |
bd4c625c LT |
211 | if (bn) { |
212 | list_add(&bn->list, &journal->j_bitmap_nodes); | |
213 | journal->j_free_bitmap_nodes++; | |
214 | } else { | |
0222e657 | 215 | break; /* this is ok, we'll try again when more are needed */ |
bd4c625c LT |
216 | } |
217 | } | |
1da177e4 LT |
218 | } |
219 | ||
a9dd3643 | 220 | static int set_bit_in_list_bitmap(struct super_block *sb, |
3ee16670 | 221 | b_blocknr_t block, |
bd4c625c LT |
222 | struct reiserfs_list_bitmap *jb) |
223 | { | |
a9dd3643 JM |
224 | unsigned int bmap_nr = block / (sb->s_blocksize << 3); |
225 | unsigned int bit_nr = block % (sb->s_blocksize << 3); | |
1da177e4 | 226 | |
bd4c625c | 227 | if (!jb->bitmaps[bmap_nr]) { |
a9dd3643 | 228 | jb->bitmaps[bmap_nr] = get_bitmap_node(sb); |
bd4c625c LT |
229 | } |
230 | set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data); | |
231 | return 0; | |
1da177e4 LT |
232 | } |
233 | ||
a9dd3643 | 234 | static void cleanup_bitmap_list(struct super_block *sb, |
bd4c625c LT |
235 | struct reiserfs_list_bitmap *jb) |
236 | { | |
237 | int i; | |
238 | if (jb->bitmaps == NULL) | |
239 | return; | |
240 | ||
a9dd3643 | 241 | for (i = 0; i < reiserfs_bmap_count(sb); i++) { |
bd4c625c | 242 | if (jb->bitmaps[i]) { |
a9dd3643 | 243 | free_bitmap_node(sb, jb->bitmaps[i]); |
bd4c625c LT |
244 | jb->bitmaps[i] = NULL; |
245 | } | |
246 | } | |
1da177e4 LT |
247 | } |
248 | ||
249 | /* | |
250 | ** only call this on FS unmount. | |
251 | */ | |
a9dd3643 | 252 | static int free_list_bitmaps(struct super_block *sb, |
bd4c625c LT |
253 | struct reiserfs_list_bitmap *jb_array) |
254 | { | |
255 | int i; | |
256 | struct reiserfs_list_bitmap *jb; | |
257 | for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) { | |
258 | jb = jb_array + i; | |
259 | jb->journal_list = NULL; | |
a9dd3643 | 260 | cleanup_bitmap_list(sb, jb); |
bd4c625c LT |
261 | vfree(jb->bitmaps); |
262 | jb->bitmaps = NULL; | |
263 | } | |
264 | return 0; | |
265 | } | |
266 | ||
a9dd3643 | 267 | static int free_bitmap_nodes(struct super_block *sb) |
bd4c625c | 268 | { |
a9dd3643 | 269 | struct reiserfs_journal *journal = SB_JOURNAL(sb); |
bd4c625c LT |
270 | struct list_head *next = journal->j_bitmap_nodes.next; |
271 | struct reiserfs_bitmap_node *bn; | |
272 | ||
273 | while (next != &journal->j_bitmap_nodes) { | |
274 | bn = list_entry(next, struct reiserfs_bitmap_node, list); | |
275 | list_del(next); | |
d739b42b PE |
276 | kfree(bn->data); |
277 | kfree(bn); | |
bd4c625c LT |
278 | next = journal->j_bitmap_nodes.next; |
279 | journal->j_free_bitmap_nodes--; | |
280 | } | |
281 | ||
282 | return 0; | |
1da177e4 LT |
283 | } |
284 | ||
285 | /* | |
0222e657 | 286 | ** get memory for JOURNAL_NUM_BITMAPS worth of bitmaps. |
1da177e4 LT |
287 | ** jb_array is the array to be filled in. |
288 | */ | |
a9dd3643 | 289 | int reiserfs_allocate_list_bitmaps(struct super_block *sb, |
bd4c625c | 290 | struct reiserfs_list_bitmap *jb_array, |
3ee16670 | 291 | unsigned int bmap_nr) |
bd4c625c LT |
292 | { |
293 | int i; | |
294 | int failed = 0; | |
295 | struct reiserfs_list_bitmap *jb; | |
296 | int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *); | |
297 | ||
298 | for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) { | |
299 | jb = jb_array + i; | |
300 | jb->journal_list = NULL; | |
301 | jb->bitmaps = vmalloc(mem); | |
302 | if (!jb->bitmaps) { | |
a9dd3643 | 303 | reiserfs_warning(sb, "clm-2000", "unable to " |
45b03d5e | 304 | "allocate bitmaps for journal lists"); |
bd4c625c LT |
305 | failed = 1; |
306 | break; | |
307 | } | |
308 | memset(jb->bitmaps, 0, mem); | |
309 | } | |
310 | if (failed) { | |
a9dd3643 | 311 | free_list_bitmaps(sb, jb_array); |
bd4c625c LT |
312 | return -1; |
313 | } | |
314 | return 0; | |
1da177e4 LT |
315 | } |
316 | ||
317 | /* | |
0222e657 | 318 | ** find an available list bitmap. If you can't find one, flush a commit list |
1da177e4 LT |
319 | ** and try again |
320 | */ | |
a9dd3643 | 321 | static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *sb, |
bd4c625c LT |
322 | struct reiserfs_journal_list |
323 | *jl) | |
324 | { | |
325 | int i, j; | |
a9dd3643 | 326 | struct reiserfs_journal *journal = SB_JOURNAL(sb); |
bd4c625c LT |
327 | struct reiserfs_list_bitmap *jb = NULL; |
328 | ||
329 | for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) { | |
330 | i = journal->j_list_bitmap_index; | |
331 | journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS; | |
332 | jb = journal->j_list_bitmap + i; | |
333 | if (journal->j_list_bitmap[i].journal_list) { | |
a9dd3643 | 334 | flush_commit_list(sb, |
bd4c625c LT |
335 | journal->j_list_bitmap[i]. |
336 | journal_list, 1); | |
337 | if (!journal->j_list_bitmap[i].journal_list) { | |
338 | break; | |
339 | } | |
340 | } else { | |
341 | break; | |
342 | } | |
343 | } | |
344 | if (jb->journal_list) { /* double check to make sure if flushed correctly */ | |
345 | return NULL; | |
346 | } | |
347 | jb->journal_list = jl; | |
348 | return jb; | |
1da177e4 LT |
349 | } |
350 | ||
0222e657 | 351 | /* |
1da177e4 LT |
352 | ** allocates a new chunk of X nodes, and links them all together as a list. |
353 | ** Uses the cnode->next and cnode->prev pointers | |
354 | ** returns NULL on failure | |
355 | */ | |
bd4c625c LT |
356 | static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes) |
357 | { | |
358 | struct reiserfs_journal_cnode *head; | |
359 | int i; | |
360 | if (num_cnodes <= 0) { | |
361 | return NULL; | |
362 | } | |
363 | head = vmalloc(num_cnodes * sizeof(struct reiserfs_journal_cnode)); | |
364 | if (!head) { | |
365 | return NULL; | |
366 | } | |
367 | memset(head, 0, num_cnodes * sizeof(struct reiserfs_journal_cnode)); | |
368 | head[0].prev = NULL; | |
369 | head[0].next = head + 1; | |
370 | for (i = 1; i < num_cnodes; i++) { | |
371 | head[i].prev = head + (i - 1); | |
372 | head[i].next = head + (i + 1); /* if last one, overwrite it after the if */ | |
373 | } | |
374 | head[num_cnodes - 1].next = NULL; | |
375 | return head; | |
1da177e4 LT |
376 | } |
377 | ||
378 | /* | |
0222e657 | 379 | ** pulls a cnode off the free list, or returns NULL on failure |
1da177e4 | 380 | */ |
a9dd3643 | 381 | static struct reiserfs_journal_cnode *get_cnode(struct super_block *sb) |
bd4c625c LT |
382 | { |
383 | struct reiserfs_journal_cnode *cn; | |
a9dd3643 | 384 | struct reiserfs_journal *journal = SB_JOURNAL(sb); |
bd4c625c | 385 | |
a9dd3643 | 386 | reiserfs_check_lock_depth(sb, "get_cnode"); |
bd4c625c LT |
387 | |
388 | if (journal->j_cnode_free <= 0) { | |
389 | return NULL; | |
390 | } | |
391 | journal->j_cnode_used++; | |
392 | journal->j_cnode_free--; | |
393 | cn = journal->j_cnode_free_list; | |
394 | if (!cn) { | |
395 | return cn; | |
396 | } | |
397 | if (cn->next) { | |
398 | cn->next->prev = NULL; | |
399 | } | |
400 | journal->j_cnode_free_list = cn->next; | |
401 | memset(cn, 0, sizeof(struct reiserfs_journal_cnode)); | |
402 | return cn; | |
1da177e4 LT |
403 | } |
404 | ||
405 | /* | |
0222e657 | 406 | ** returns a cnode to the free list |
1da177e4 | 407 | */ |
a9dd3643 | 408 | static void free_cnode(struct super_block *sb, |
bd4c625c LT |
409 | struct reiserfs_journal_cnode *cn) |
410 | { | |
a9dd3643 | 411 | struct reiserfs_journal *journal = SB_JOURNAL(sb); |
1da177e4 | 412 | |
a9dd3643 | 413 | reiserfs_check_lock_depth(sb, "free_cnode"); |
1da177e4 | 414 | |
bd4c625c LT |
415 | journal->j_cnode_used--; |
416 | journal->j_cnode_free++; | |
417 | /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */ | |
418 | cn->next = journal->j_cnode_free_list; | |
419 | if (journal->j_cnode_free_list) { | |
420 | journal->j_cnode_free_list->prev = cn; | |
421 | } | |
422 | cn->prev = NULL; /* not needed with the memset, but I might kill the memset, and forget to do this */ | |
423 | journal->j_cnode_free_list = cn; | |
1da177e4 LT |
424 | } |
425 | ||
bd4c625c LT |
426 | static void clear_prepared_bits(struct buffer_head *bh) |
427 | { | |
428 | clear_buffer_journal_prepared(bh); | |
429 | clear_buffer_journal_restore_dirty(bh); | |
1da177e4 LT |
430 | } |
431 | ||
1da177e4 | 432 | /* return a cnode with same dev, block number and size in table, or null if not found */ |
bd4c625c LT |
433 | static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct |
434 | super_block | |
435 | *sb, | |
436 | struct | |
437 | reiserfs_journal_cnode | |
438 | **table, | |
439 | long bl) | |
1da177e4 | 440 | { |
bd4c625c LT |
441 | struct reiserfs_journal_cnode *cn; |
442 | cn = journal_hash(table, sb, bl); | |
443 | while (cn) { | |
444 | if (cn->blocknr == bl && cn->sb == sb) | |
445 | return cn; | |
446 | cn = cn->hnext; | |
447 | } | |
448 | return (struct reiserfs_journal_cnode *)0; | |
1da177e4 LT |
449 | } |
450 | ||
451 | /* | |
452 | ** this actually means 'can this block be reallocated yet?'. If you set search_all, a block can only be allocated | |
453 | ** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever | |
454 | ** being overwritten by a replay after crashing. | |
455 | ** | |
456 | ** If you don't set search_all, a block can only be allocated if it is not in the current transaction. Since deleting | |
457 | ** a block removes it from the current transaction, this case should never happen. If you don't set search_all, make | |
458 | ** sure you never write the block without logging it. | |
459 | ** | |
460 | ** next_zero_bit is a suggestion about the next block to try for find_forward. | |
461 | ** when bl is rejected because it is set in a journal list bitmap, we search | |
462 | ** for the next zero bit in the bitmap that rejected bl. Then, we return that | |
463 | ** through next_zero_bit for find_forward to try. | |
464 | ** | |
465 | ** Just because we return something in next_zero_bit does not mean we won't | |
466 | ** reject it on the next call to reiserfs_in_journal | |
467 | ** | |
468 | */ | |
a9dd3643 | 469 | int reiserfs_in_journal(struct super_block *sb, |
3ee16670 | 470 | unsigned int bmap_nr, int bit_nr, int search_all, |
bd4c625c LT |
471 | b_blocknr_t * next_zero_bit) |
472 | { | |
a9dd3643 | 473 | struct reiserfs_journal *journal = SB_JOURNAL(sb); |
bd4c625c LT |
474 | struct reiserfs_journal_cnode *cn; |
475 | struct reiserfs_list_bitmap *jb; | |
476 | int i; | |
477 | unsigned long bl; | |
478 | ||
479 | *next_zero_bit = 0; /* always start this at zero. */ | |
480 | ||
a9dd3643 | 481 | PROC_INFO_INC(sb, journal.in_journal); |
bd4c625c LT |
482 | /* If we aren't doing a search_all, this is a metablock, and it will be logged before use. |
483 | ** if we crash before the transaction that freed it commits, this transaction won't | |
484 | ** have committed either, and the block will never be written | |
485 | */ | |
486 | if (search_all) { | |
487 | for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) { | |
a9dd3643 | 488 | PROC_INFO_INC(sb, journal.in_journal_bitmap); |
bd4c625c LT |
489 | jb = journal->j_list_bitmap + i; |
490 | if (jb->journal_list && jb->bitmaps[bmap_nr] && | |
491 | test_bit(bit_nr, | |
492 | (unsigned long *)jb->bitmaps[bmap_nr]-> | |
493 | data)) { | |
494 | *next_zero_bit = | |
495 | find_next_zero_bit((unsigned long *) | |
496 | (jb->bitmaps[bmap_nr]-> | |
497 | data), | |
a9dd3643 | 498 | sb->s_blocksize << 3, |
bd4c625c LT |
499 | bit_nr + 1); |
500 | return 1; | |
501 | } | |
502 | } | |
503 | } | |
504 | ||
a9dd3643 | 505 | bl = bmap_nr * (sb->s_blocksize << 3) + bit_nr; |
bd4c625c LT |
506 | /* is it in any old transactions? */ |
507 | if (search_all | |
508 | && (cn = | |
a9dd3643 | 509 | get_journal_hash_dev(sb, journal->j_list_hash_table, bl))) { |
bd4c625c LT |
510 | return 1; |
511 | } | |
512 | ||
513 | /* is it in the current transaction. This should never happen */ | |
a9dd3643 | 514 | if ((cn = get_journal_hash_dev(sb, journal->j_hash_table, bl))) { |
bd4c625c LT |
515 | BUG(); |
516 | return 1; | |
517 | } | |
518 | ||
a9dd3643 | 519 | PROC_INFO_INC(sb, journal.in_journal_reusable); |
bd4c625c LT |
520 | /* safe for reuse */ |
521 | return 0; | |
1da177e4 LT |
522 | } |
523 | ||
524 | /* insert cn into table | |
525 | */ | |
bd4c625c LT |
526 | static inline void insert_journal_hash(struct reiserfs_journal_cnode **table, |
527 | struct reiserfs_journal_cnode *cn) | |
528 | { | |
529 | struct reiserfs_journal_cnode *cn_orig; | |
1da177e4 | 530 | |
bd4c625c LT |
531 | cn_orig = journal_hash(table, cn->sb, cn->blocknr); |
532 | cn->hnext = cn_orig; | |
533 | cn->hprev = NULL; | |
534 | if (cn_orig) { | |
535 | cn_orig->hprev = cn; | |
536 | } | |
537 | journal_hash(table, cn->sb, cn->blocknr) = cn; | |
1da177e4 LT |
538 | } |
539 | ||
540 | /* lock the current transaction */ | |
a9dd3643 | 541 | static inline void lock_journal(struct super_block *sb) |
bd4c625c | 542 | { |
a9dd3643 | 543 | PROC_INFO_INC(sb, journal.lock_journal); |
8ebc4232 FW |
544 | |
545 | reiserfs_mutex_lock_safe(&SB_JOURNAL(sb)->j_mutex, sb); | |
1da177e4 LT |
546 | } |
547 | ||
548 | /* unlock the current transaction */ | |
a9dd3643 | 549 | static inline void unlock_journal(struct super_block *sb) |
bd4c625c | 550 | { |
a9dd3643 | 551 | mutex_unlock(&SB_JOURNAL(sb)->j_mutex); |
1da177e4 LT |
552 | } |
553 | ||
554 | static inline void get_journal_list(struct reiserfs_journal_list *jl) | |
555 | { | |
bd4c625c | 556 | jl->j_refcount++; |
1da177e4 LT |
557 | } |
558 | ||
559 | static inline void put_journal_list(struct super_block *s, | |
bd4c625c | 560 | struct reiserfs_journal_list *jl) |
1da177e4 | 561 | { |
bd4c625c | 562 | if (jl->j_refcount < 1) { |
c3a9c210 | 563 | reiserfs_panic(s, "journal-2", "trans id %u, refcount at %d", |
bd4c625c LT |
564 | jl->j_trans_id, jl->j_refcount); |
565 | } | |
566 | if (--jl->j_refcount == 0) | |
d739b42b | 567 | kfree(jl); |
1da177e4 LT |
568 | } |
569 | ||
570 | /* | |
571 | ** this used to be much more involved, and I'm keeping it just in case things get ugly again. | |
572 | ** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a | |
573 | ** transaction. | |
574 | */ | |
a9dd3643 | 575 | static void cleanup_freed_for_journal_list(struct super_block *sb, |
bd4c625c LT |
576 | struct reiserfs_journal_list *jl) |
577 | { | |
1da177e4 | 578 | |
bd4c625c LT |
579 | struct reiserfs_list_bitmap *jb = jl->j_list_bitmap; |
580 | if (jb) { | |
a9dd3643 | 581 | cleanup_bitmap_list(sb, jb); |
bd4c625c LT |
582 | } |
583 | jl->j_list_bitmap->journal_list = NULL; | |
584 | jl->j_list_bitmap = NULL; | |
1da177e4 LT |
585 | } |
586 | ||
587 | static int journal_list_still_alive(struct super_block *s, | |
600ed416 | 588 | unsigned int trans_id) |
bd4c625c LT |
589 | { |
590 | struct reiserfs_journal *journal = SB_JOURNAL(s); | |
591 | struct list_head *entry = &journal->j_journal_list; | |
592 | struct reiserfs_journal_list *jl; | |
593 | ||
594 | if (!list_empty(entry)) { | |
595 | jl = JOURNAL_LIST_ENTRY(entry->next); | |
596 | if (jl->j_trans_id <= trans_id) { | |
597 | return 1; | |
598 | } | |
599 | } | |
600 | return 0; | |
601 | } | |
602 | ||
398c95bd CM |
603 | /* |
604 | * If page->mapping was null, we failed to truncate this page for | |
605 | * some reason. Most likely because it was truncated after being | |
606 | * logged via data=journal. | |
607 | * | |
608 | * This does a check to see if the buffer belongs to one of these | |
609 | * lost pages before doing the final put_bh. If page->mapping was | |
610 | * null, it tries to free buffers on the page, which should make the | |
611 | * final page_cache_release drop the page from the lru. | |
612 | */ | |
613 | static void release_buffer_page(struct buffer_head *bh) | |
614 | { | |
615 | struct page *page = bh->b_page; | |
529ae9aa | 616 | if (!page->mapping && trylock_page(page)) { |
398c95bd CM |
617 | page_cache_get(page); |
618 | put_bh(bh); | |
619 | if (!page->mapping) | |
620 | try_to_free_buffers(page); | |
621 | unlock_page(page); | |
622 | page_cache_release(page); | |
623 | } else { | |
624 | put_bh(bh); | |
625 | } | |
626 | } | |
627 | ||
bd4c625c LT |
628 | static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate) |
629 | { | |
630 | char b[BDEVNAME_SIZE]; | |
631 | ||
632 | if (buffer_journaled(bh)) { | |
45b03d5e JM |
633 | reiserfs_warning(NULL, "clm-2084", |
634 | "pinned buffer %lu:%s sent to disk", | |
bd4c625c LT |
635 | bh->b_blocknr, bdevname(bh->b_bdev, b)); |
636 | } | |
637 | if (uptodate) | |
638 | set_buffer_uptodate(bh); | |
639 | else | |
640 | clear_buffer_uptodate(bh); | |
398c95bd | 641 | |
bd4c625c | 642 | unlock_buffer(bh); |
398c95bd | 643 | release_buffer_page(bh); |
bd4c625c LT |
644 | } |
645 | ||
646 | static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate) | |
647 | { | |
648 | if (uptodate) | |
649 | set_buffer_uptodate(bh); | |
650 | else | |
651 | clear_buffer_uptodate(bh); | |
652 | unlock_buffer(bh); | |
653 | put_bh(bh); | |
654 | } | |
655 | ||
656 | static void submit_logged_buffer(struct buffer_head *bh) | |
657 | { | |
658 | get_bh(bh); | |
659 | bh->b_end_io = reiserfs_end_buffer_io_sync; | |
660 | clear_buffer_journal_new(bh); | |
661 | clear_buffer_dirty(bh); | |
662 | if (!test_clear_buffer_journal_test(bh)) | |
663 | BUG(); | |
664 | if (!buffer_uptodate(bh)) | |
665 | BUG(); | |
666 | submit_bh(WRITE, bh); | |
667 | } | |
668 | ||
669 | static void submit_ordered_buffer(struct buffer_head *bh) | |
670 | { | |
671 | get_bh(bh); | |
672 | bh->b_end_io = reiserfs_end_ordered_io; | |
673 | clear_buffer_dirty(bh); | |
674 | if (!buffer_uptodate(bh)) | |
675 | BUG(); | |
676 | submit_bh(WRITE, bh); | |
677 | } | |
678 | ||
679 | static int submit_barrier_buffer(struct buffer_head *bh) | |
680 | { | |
681 | get_bh(bh); | |
682 | bh->b_end_io = reiserfs_end_ordered_io; | |
683 | clear_buffer_dirty(bh); | |
684 | if (!buffer_uptodate(bh)) | |
685 | BUG(); | |
686 | return submit_bh(WRITE_BARRIER, bh); | |
1da177e4 LT |
687 | } |
688 | ||
689 | static void check_barrier_completion(struct super_block *s, | |
bd4c625c LT |
690 | struct buffer_head *bh) |
691 | { | |
692 | if (buffer_eopnotsupp(bh)) { | |
693 | clear_buffer_eopnotsupp(bh); | |
694 | disable_barrier(s); | |
695 | set_buffer_uptodate(bh); | |
696 | set_buffer_dirty(bh); | |
8ebc4232 | 697 | reiserfs_write_unlock(s); |
bd4c625c | 698 | sync_dirty_buffer(bh); |
8ebc4232 | 699 | reiserfs_write_lock(s); |
bd4c625c | 700 | } |
1da177e4 LT |
701 | } |
702 | ||
703 | #define CHUNK_SIZE 32 | |
704 | struct buffer_chunk { | |
bd4c625c LT |
705 | struct buffer_head *bh[CHUNK_SIZE]; |
706 | int nr; | |
1da177e4 LT |
707 | }; |
708 | ||
bd4c625c LT |
709 | static void write_chunk(struct buffer_chunk *chunk) |
710 | { | |
711 | int i; | |
712 | get_fs_excl(); | |
713 | for (i = 0; i < chunk->nr; i++) { | |
714 | submit_logged_buffer(chunk->bh[i]); | |
715 | } | |
716 | chunk->nr = 0; | |
717 | put_fs_excl(); | |
1da177e4 LT |
718 | } |
719 | ||
bd4c625c LT |
720 | static void write_ordered_chunk(struct buffer_chunk *chunk) |
721 | { | |
722 | int i; | |
723 | get_fs_excl(); | |
724 | for (i = 0; i < chunk->nr; i++) { | |
725 | submit_ordered_buffer(chunk->bh[i]); | |
726 | } | |
727 | chunk->nr = 0; | |
728 | put_fs_excl(); | |
1da177e4 LT |
729 | } |
730 | ||
731 | static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh, | |
bd4c625c | 732 | spinlock_t * lock, void (fn) (struct buffer_chunk *)) |
1da177e4 | 733 | { |
bd4c625c | 734 | int ret = 0; |
14a61442 | 735 | BUG_ON(chunk->nr >= CHUNK_SIZE); |
bd4c625c LT |
736 | chunk->bh[chunk->nr++] = bh; |
737 | if (chunk->nr >= CHUNK_SIZE) { | |
738 | ret = 1; | |
739 | if (lock) | |
740 | spin_unlock(lock); | |
741 | fn(chunk); | |
742 | if (lock) | |
743 | spin_lock(lock); | |
744 | } | |
745 | return ret; | |
1da177e4 LT |
746 | } |
747 | ||
1da177e4 | 748 | static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0); |
bd4c625c LT |
749 | static struct reiserfs_jh *alloc_jh(void) |
750 | { | |
751 | struct reiserfs_jh *jh; | |
752 | while (1) { | |
753 | jh = kmalloc(sizeof(*jh), GFP_NOFS); | |
754 | if (jh) { | |
755 | atomic_inc(&nr_reiserfs_jh); | |
756 | return jh; | |
757 | } | |
758 | yield(); | |
1da177e4 | 759 | } |
1da177e4 LT |
760 | } |
761 | ||
762 | /* | |
763 | * we want to free the jh when the buffer has been written | |
764 | * and waited on | |
765 | */ | |
bd4c625c LT |
766 | void reiserfs_free_jh(struct buffer_head *bh) |
767 | { | |
768 | struct reiserfs_jh *jh; | |
769 | ||
770 | jh = bh->b_private; | |
771 | if (jh) { | |
772 | bh->b_private = NULL; | |
773 | jh->bh = NULL; | |
774 | list_del_init(&jh->list); | |
775 | kfree(jh); | |
776 | if (atomic_read(&nr_reiserfs_jh) <= 0) | |
777 | BUG(); | |
778 | atomic_dec(&nr_reiserfs_jh); | |
779 | put_bh(bh); | |
780 | } | |
1da177e4 LT |
781 | } |
782 | ||
783 | static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh, | |
bd4c625c | 784 | int tail) |
1da177e4 | 785 | { |
bd4c625c | 786 | struct reiserfs_jh *jh; |
1da177e4 | 787 | |
bd4c625c LT |
788 | if (bh->b_private) { |
789 | spin_lock(&j->j_dirty_buffers_lock); | |
790 | if (!bh->b_private) { | |
791 | spin_unlock(&j->j_dirty_buffers_lock); | |
792 | goto no_jh; | |
793 | } | |
794 | jh = bh->b_private; | |
795 | list_del_init(&jh->list); | |
796 | } else { | |
797 | no_jh: | |
798 | get_bh(bh); | |
799 | jh = alloc_jh(); | |
800 | spin_lock(&j->j_dirty_buffers_lock); | |
801 | /* buffer must be locked for __add_jh, should be able to have | |
802 | * two adds at the same time | |
803 | */ | |
14a61442 | 804 | BUG_ON(bh->b_private); |
bd4c625c LT |
805 | jh->bh = bh; |
806 | bh->b_private = jh; | |
1da177e4 | 807 | } |
bd4c625c LT |
808 | jh->jl = j->j_current_jl; |
809 | if (tail) | |
810 | list_add_tail(&jh->list, &jh->jl->j_tail_bh_list); | |
811 | else { | |
812 | list_add_tail(&jh->list, &jh->jl->j_bh_list); | |
813 | } | |
814 | spin_unlock(&j->j_dirty_buffers_lock); | |
815 | return 0; | |
1da177e4 LT |
816 | } |
817 | ||
bd4c625c LT |
818 | int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh) |
819 | { | |
820 | return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1); | |
1da177e4 | 821 | } |
bd4c625c LT |
822 | int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh) |
823 | { | |
824 | return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0); | |
1da177e4 LT |
825 | } |
826 | ||
827 | #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list) | |
bd4c625c | 828 | static int write_ordered_buffers(spinlock_t * lock, |
1da177e4 | 829 | struct reiserfs_journal *j, |
bd4c625c | 830 | struct reiserfs_journal_list *jl, |
1da177e4 LT |
831 | struct list_head *list) |
832 | { | |
bd4c625c LT |
833 | struct buffer_head *bh; |
834 | struct reiserfs_jh *jh; | |
835 | int ret = j->j_errno; | |
836 | struct buffer_chunk chunk; | |
837 | struct list_head tmp; | |
838 | INIT_LIST_HEAD(&tmp); | |
839 | ||
840 | chunk.nr = 0; | |
841 | spin_lock(lock); | |
842 | while (!list_empty(list)) { | |
843 | jh = JH_ENTRY(list->next); | |
844 | bh = jh->bh; | |
845 | get_bh(bh); | |
ca5de404 | 846 | if (!trylock_buffer(bh)) { |
bd4c625c | 847 | if (!buffer_dirty(bh)) { |
f116629d | 848 | list_move(&jh->list, &tmp); |
bd4c625c LT |
849 | goto loop_next; |
850 | } | |
851 | spin_unlock(lock); | |
852 | if (chunk.nr) | |
853 | write_ordered_chunk(&chunk); | |
854 | wait_on_buffer(bh); | |
855 | cond_resched(); | |
856 | spin_lock(lock); | |
857 | goto loop_next; | |
858 | } | |
3d4492f8 CM |
859 | /* in theory, dirty non-uptodate buffers should never get here, |
860 | * but the upper layer io error paths still have a few quirks. | |
861 | * Handle them here as gracefully as we can | |
862 | */ | |
863 | if (!buffer_uptodate(bh) && buffer_dirty(bh)) { | |
864 | clear_buffer_dirty(bh); | |
865 | ret = -EIO; | |
866 | } | |
bd4c625c | 867 | if (buffer_dirty(bh)) { |
f116629d | 868 | list_move(&jh->list, &tmp); |
bd4c625c LT |
869 | add_to_chunk(&chunk, bh, lock, write_ordered_chunk); |
870 | } else { | |
871 | reiserfs_free_jh(bh); | |
872 | unlock_buffer(bh); | |
873 | } | |
874 | loop_next: | |
875 | put_bh(bh); | |
876 | cond_resched_lock(lock); | |
877 | } | |
878 | if (chunk.nr) { | |
879 | spin_unlock(lock); | |
1da177e4 | 880 | write_ordered_chunk(&chunk); |
bd4c625c | 881 | spin_lock(lock); |
1da177e4 | 882 | } |
bd4c625c LT |
883 | while (!list_empty(&tmp)) { |
884 | jh = JH_ENTRY(tmp.prev); | |
885 | bh = jh->bh; | |
886 | get_bh(bh); | |
887 | reiserfs_free_jh(bh); | |
888 | ||
889 | if (buffer_locked(bh)) { | |
890 | spin_unlock(lock); | |
891 | wait_on_buffer(bh); | |
892 | spin_lock(lock); | |
893 | } | |
894 | if (!buffer_uptodate(bh)) { | |
895 | ret = -EIO; | |
896 | } | |
d62b1b87 CM |
897 | /* ugly interaction with invalidatepage here. |
898 | * reiserfs_invalidate_page will pin any buffer that has a valid | |
899 | * journal head from an older transaction. If someone else sets | |
900 | * our buffer dirty after we write it in the first loop, and | |
901 | * then someone truncates the page away, nobody will ever write | |
902 | * the buffer. We're safe if we write the page one last time | |
903 | * after freeing the journal header. | |
904 | */ | |
905 | if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) { | |
906 | spin_unlock(lock); | |
907 | ll_rw_block(WRITE, 1, &bh); | |
908 | spin_lock(lock); | |
909 | } | |
bd4c625c LT |
910 | put_bh(bh); |
911 | cond_resched_lock(lock); | |
1da177e4 | 912 | } |
bd4c625c LT |
913 | spin_unlock(lock); |
914 | return ret; | |
915 | } | |
1da177e4 | 916 | |
bd4c625c LT |
917 | static int flush_older_commits(struct super_block *s, |
918 | struct reiserfs_journal_list *jl) | |
919 | { | |
920 | struct reiserfs_journal *journal = SB_JOURNAL(s); | |
921 | struct reiserfs_journal_list *other_jl; | |
922 | struct reiserfs_journal_list *first_jl; | |
923 | struct list_head *entry; | |
600ed416 JM |
924 | unsigned int trans_id = jl->j_trans_id; |
925 | unsigned int other_trans_id; | |
926 | unsigned int first_trans_id; | |
bd4c625c LT |
927 | |
928 | find_first: | |
929 | /* | |
930 | * first we walk backwards to find the oldest uncommitted transation | |
931 | */ | |
932 | first_jl = jl; | |
933 | entry = jl->j_list.prev; | |
934 | while (1) { | |
935 | other_jl = JOURNAL_LIST_ENTRY(entry); | |
936 | if (entry == &journal->j_journal_list || | |
937 | atomic_read(&other_jl->j_older_commits_done)) | |
938 | break; | |
1da177e4 | 939 | |
bd4c625c LT |
940 | first_jl = other_jl; |
941 | entry = other_jl->j_list.prev; | |
942 | } | |
1da177e4 | 943 | |
bd4c625c LT |
944 | /* if we didn't find any older uncommitted transactions, return now */ |
945 | if (first_jl == jl) { | |
946 | return 0; | |
947 | } | |
1da177e4 | 948 | |
bd4c625c LT |
949 | first_trans_id = first_jl->j_trans_id; |
950 | ||
951 | entry = &first_jl->j_list; | |
952 | while (1) { | |
953 | other_jl = JOURNAL_LIST_ENTRY(entry); | |
954 | other_trans_id = other_jl->j_trans_id; | |
955 | ||
956 | if (other_trans_id < trans_id) { | |
957 | if (atomic_read(&other_jl->j_commit_left) != 0) { | |
958 | flush_commit_list(s, other_jl, 0); | |
959 | ||
960 | /* list we were called with is gone, return */ | |
961 | if (!journal_list_still_alive(s, trans_id)) | |
962 | return 1; | |
963 | ||
964 | /* the one we just flushed is gone, this means all | |
965 | * older lists are also gone, so first_jl is no longer | |
966 | * valid either. Go back to the beginning. | |
967 | */ | |
968 | if (!journal_list_still_alive | |
969 | (s, other_trans_id)) { | |
970 | goto find_first; | |
971 | } | |
972 | } | |
973 | entry = entry->next; | |
974 | if (entry == &journal->j_journal_list) | |
975 | return 0; | |
976 | } else { | |
977 | return 0; | |
1da177e4 | 978 | } |
1da177e4 | 979 | } |
bd4c625c | 980 | return 0; |
1da177e4 | 981 | } |
deba0f49 AB |
982 | |
983 | static int reiserfs_async_progress_wait(struct super_block *s) | |
bd4c625c LT |
984 | { |
985 | DEFINE_WAIT(wait); | |
986 | struct reiserfs_journal *j = SB_JOURNAL(s); | |
8ebc4232 FW |
987 | |
988 | if (atomic_read(&j->j_async_throttle)) { | |
989 | reiserfs_write_unlock(s); | |
8aa7e847 | 990 | congestion_wait(BLK_RW_ASYNC, HZ / 10); |
8ebc4232 FW |
991 | reiserfs_write_lock(s); |
992 | } | |
993 | ||
bd4c625c | 994 | return 0; |
1da177e4 LT |
995 | } |
996 | ||
997 | /* | |
998 | ** if this journal list still has commit blocks unflushed, send them to disk. | |
999 | ** | |
1000 | ** log areas must be flushed in order (transaction 2 can't commit before transaction 1) | |
1001 | ** Before the commit block can by written, every other log block must be safely on disk | |
1002 | ** | |
1003 | */ | |
bd4c625c LT |
1004 | static int flush_commit_list(struct super_block *s, |
1005 | struct reiserfs_journal_list *jl, int flushall) | |
1006 | { | |
1007 | int i; | |
3ee16670 | 1008 | b_blocknr_t bn; |
bd4c625c | 1009 | struct buffer_head *tbh = NULL; |
600ed416 | 1010 | unsigned int trans_id = jl->j_trans_id; |
bd4c625c LT |
1011 | struct reiserfs_journal *journal = SB_JOURNAL(s); |
1012 | int barrier = 0; | |
1013 | int retval = 0; | |
e0e851cf | 1014 | int write_len; |
bd4c625c LT |
1015 | |
1016 | reiserfs_check_lock_depth(s, "flush_commit_list"); | |
1017 | ||
1018 | if (atomic_read(&jl->j_older_commits_done)) { | |
1019 | return 0; | |
1020 | } | |
1021 | ||
1022 | get_fs_excl(); | |
1023 | ||
1024 | /* before we can put our commit blocks on disk, we have to make sure everyone older than | |
1025 | ** us is on disk too | |
1026 | */ | |
1027 | BUG_ON(jl->j_len <= 0); | |
1028 | BUG_ON(trans_id == journal->j_trans_id); | |
1029 | ||
1030 | get_journal_list(jl); | |
1031 | if (flushall) { | |
1032 | if (flush_older_commits(s, jl) == 1) { | |
1033 | /* list disappeared during flush_older_commits. return */ | |
1034 | goto put_jl; | |
1035 | } | |
1036 | } | |
1037 | ||
1038 | /* make sure nobody is trying to flush this one at the same time */ | |
8ebc4232 FW |
1039 | reiserfs_mutex_lock_safe(&jl->j_commit_mutex, s); |
1040 | ||
bd4c625c | 1041 | if (!journal_list_still_alive(s, trans_id)) { |
90415dea | 1042 | mutex_unlock(&jl->j_commit_mutex); |
bd4c625c LT |
1043 | goto put_jl; |
1044 | } | |
1045 | BUG_ON(jl->j_trans_id == 0); | |
1046 | ||
1047 | /* this commit is done, exit */ | |
1048 | if (atomic_read(&(jl->j_commit_left)) <= 0) { | |
1049 | if (flushall) { | |
1050 | atomic_set(&(jl->j_older_commits_done), 1); | |
1051 | } | |
90415dea | 1052 | mutex_unlock(&jl->j_commit_mutex); |
bd4c625c LT |
1053 | goto put_jl; |
1054 | } | |
1055 | ||
1056 | if (!list_empty(&jl->j_bh_list)) { | |
3d4492f8 | 1057 | int ret; |
8ebc4232 FW |
1058 | |
1059 | /* | |
1060 | * We might sleep in numerous places inside | |
1061 | * write_ordered_buffers. Relax the write lock. | |
1062 | */ | |
1063 | reiserfs_write_unlock(s); | |
3d4492f8 CM |
1064 | ret = write_ordered_buffers(&journal->j_dirty_buffers_lock, |
1065 | journal, jl, &jl->j_bh_list); | |
1066 | if (ret < 0 && retval == 0) | |
1067 | retval = ret; | |
8ebc4232 | 1068 | reiserfs_write_lock(s); |
bd4c625c LT |
1069 | } |
1070 | BUG_ON(!list_empty(&jl->j_bh_list)); | |
1071 | /* | |
1072 | * for the description block and all the log blocks, submit any buffers | |
e0e851cf CM |
1073 | * that haven't already reached the disk. Try to write at least 256 |
1074 | * log blocks. later on, we will only wait on blocks that correspond | |
1075 | * to this transaction, but while we're unplugging we might as well | |
1076 | * get a chunk of data on there. | |
bd4c625c LT |
1077 | */ |
1078 | atomic_inc(&journal->j_async_throttle); | |
e0e851cf CM |
1079 | write_len = jl->j_len + 1; |
1080 | if (write_len < 256) | |
1081 | write_len = 256; | |
1082 | for (i = 0 ; i < write_len ; i++) { | |
bd4c625c LT |
1083 | bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) % |
1084 | SB_ONDISK_JOURNAL_SIZE(s); | |
1085 | tbh = journal_find_get_block(s, bn); | |
e0e851cf | 1086 | if (tbh) { |
6e3647ac FW |
1087 | if (buffer_dirty(tbh)) { |
1088 | reiserfs_write_unlock(s); | |
1089 | ll_rw_block(WRITE, 1, &tbh); | |
1090 | reiserfs_write_lock(s); | |
1091 | } | |
e0e851cf CM |
1092 | put_bh(tbh) ; |
1093 | } | |
bd4c625c LT |
1094 | } |
1095 | atomic_dec(&journal->j_async_throttle); | |
1096 | ||
5d5e8156 JM |
1097 | /* We're skipping the commit if there's an error */ |
1098 | if (retval || reiserfs_is_journal_aborted(journal)) | |
1099 | barrier = 0; | |
1100 | ||
bd4c625c LT |
1101 | /* wait on everything written so far before writing the commit |
1102 | * if we are in barrier mode, send the commit down now | |
1103 | */ | |
1104 | barrier = reiserfs_barrier_flush(s); | |
1105 | if (barrier) { | |
1106 | int ret; | |
1107 | lock_buffer(jl->j_commit_bh); | |
1108 | ret = submit_barrier_buffer(jl->j_commit_bh); | |
1109 | if (ret == -EOPNOTSUPP) { | |
1110 | set_buffer_uptodate(jl->j_commit_bh); | |
1111 | disable_barrier(s); | |
1112 | barrier = 0; | |
1113 | } | |
1114 | } | |
1115 | for (i = 0; i < (jl->j_len + 1); i++) { | |
1116 | bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + | |
1117 | (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s); | |
1118 | tbh = journal_find_get_block(s, bn); | |
8ebc4232 FW |
1119 | |
1120 | reiserfs_write_unlock(s); | |
bd4c625c | 1121 | wait_on_buffer(tbh); |
8ebc4232 | 1122 | reiserfs_write_lock(s); |
bd4c625c LT |
1123 | // since we're using ll_rw_blk above, it might have skipped over |
1124 | // a locked buffer. Double check here | |
1125 | // | |
8ebc4232 FW |
1126 | /* redundant, sync_dirty_buffer() checks */ |
1127 | if (buffer_dirty(tbh)) { | |
1128 | reiserfs_write_unlock(s); | |
bd4c625c | 1129 | sync_dirty_buffer(tbh); |
8ebc4232 FW |
1130 | reiserfs_write_lock(s); |
1131 | } | |
bd4c625c | 1132 | if (unlikely(!buffer_uptodate(tbh))) { |
1da177e4 | 1133 | #ifdef CONFIG_REISERFS_CHECK |
45b03d5e JM |
1134 | reiserfs_warning(s, "journal-601", |
1135 | "buffer write failed"); | |
1da177e4 | 1136 | #endif |
bd4c625c LT |
1137 | retval = -EIO; |
1138 | } | |
1139 | put_bh(tbh); /* once for journal_find_get_block */ | |
1140 | put_bh(tbh); /* once due to original getblk in do_journal_end */ | |
1141 | atomic_dec(&(jl->j_commit_left)); | |
1142 | } | |
1143 | ||
1144 | BUG_ON(atomic_read(&(jl->j_commit_left)) != 1); | |
1145 | ||
1146 | if (!barrier) { | |
5d5e8156 JM |
1147 | /* If there was a write error in the journal - we can't commit |
1148 | * this transaction - it will be invalid and, if successful, | |
beb7dd86 | 1149 | * will just end up propagating the write error out to |
5d5e8156 JM |
1150 | * the file system. */ |
1151 | if (likely(!retval && !reiserfs_is_journal_aborted (journal))) { | |
1152 | if (buffer_dirty(jl->j_commit_bh)) | |
1153 | BUG(); | |
1154 | mark_buffer_dirty(jl->j_commit_bh) ; | |
8ebc4232 | 1155 | reiserfs_write_unlock(s); |
5d5e8156 | 1156 | sync_dirty_buffer(jl->j_commit_bh) ; |
8ebc4232 | 1157 | reiserfs_write_lock(s); |
5d5e8156 | 1158 | } |
8ebc4232 FW |
1159 | } else { |
1160 | reiserfs_write_unlock(s); | |
bd4c625c | 1161 | wait_on_buffer(jl->j_commit_bh); |
8ebc4232 FW |
1162 | reiserfs_write_lock(s); |
1163 | } | |
bd4c625c LT |
1164 | |
1165 | check_barrier_completion(s, jl->j_commit_bh); | |
1166 | ||
1167 | /* If there was a write error in the journal - we can't commit this | |
1168 | * transaction - it will be invalid and, if successful, will just end | |
beb7dd86 | 1169 | * up propagating the write error out to the filesystem. */ |
bd4c625c | 1170 | if (unlikely(!buffer_uptodate(jl->j_commit_bh))) { |
1da177e4 | 1171 | #ifdef CONFIG_REISERFS_CHECK |
45b03d5e | 1172 | reiserfs_warning(s, "journal-615", "buffer write failed"); |
1da177e4 | 1173 | #endif |
bd4c625c LT |
1174 | retval = -EIO; |
1175 | } | |
1176 | bforget(jl->j_commit_bh); | |
1177 | if (journal->j_last_commit_id != 0 && | |
1178 | (jl->j_trans_id - journal->j_last_commit_id) != 1) { | |
45b03d5e | 1179 | reiserfs_warning(s, "clm-2200", "last commit %lu, current %lu", |
bd4c625c LT |
1180 | journal->j_last_commit_id, jl->j_trans_id); |
1181 | } | |
1182 | journal->j_last_commit_id = jl->j_trans_id; | |
1183 | ||
1184 | /* now, every commit block is on the disk. It is safe to allow blocks freed during this transaction to be reallocated */ | |
1185 | cleanup_freed_for_journal_list(s, jl); | |
1186 | ||
1187 | retval = retval ? retval : journal->j_errno; | |
1188 | ||
1189 | /* mark the metadata dirty */ | |
1190 | if (!retval) | |
1191 | dirty_one_transaction(s, jl); | |
1192 | atomic_dec(&(jl->j_commit_left)); | |
1193 | ||
1194 | if (flushall) { | |
1195 | atomic_set(&(jl->j_older_commits_done), 1); | |
1196 | } | |
90415dea | 1197 | mutex_unlock(&jl->j_commit_mutex); |
bd4c625c LT |
1198 | put_jl: |
1199 | put_journal_list(s, jl); | |
1200 | ||
1201 | if (retval) | |
1202 | reiserfs_abort(s, retval, "Journal write error in %s", | |
fbe5498b | 1203 | __func__); |
bd4c625c LT |
1204 | put_fs_excl(); |
1205 | return retval; | |
1da177e4 LT |
1206 | } |
1207 | ||
1208 | /* | |
0222e657 JM |
1209 | ** flush_journal_list frequently needs to find a newer transaction for a given block. This does that, or |
1210 | ** returns NULL if it can't find anything | |
1da177e4 | 1211 | */ |
bd4c625c LT |
1212 | static struct reiserfs_journal_list *find_newer_jl_for_cn(struct |
1213 | reiserfs_journal_cnode | |
1214 | *cn) | |
1215 | { | |
1216 | struct super_block *sb = cn->sb; | |
1217 | b_blocknr_t blocknr = cn->blocknr; | |
1da177e4 | 1218 | |
bd4c625c LT |
1219 | cn = cn->hprev; |
1220 | while (cn) { | |
1221 | if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) { | |
1222 | return cn->jlist; | |
1223 | } | |
1224 | cn = cn->hprev; | |
1225 | } | |
1226 | return NULL; | |
1da177e4 LT |
1227 | } |
1228 | ||
a3172027 CM |
1229 | static int newer_jl_done(struct reiserfs_journal_cnode *cn) |
1230 | { | |
1231 | struct super_block *sb = cn->sb; | |
1232 | b_blocknr_t blocknr = cn->blocknr; | |
1233 | ||
1234 | cn = cn->hprev; | |
1235 | while (cn) { | |
1236 | if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist && | |
1237 | atomic_read(&cn->jlist->j_commit_left) != 0) | |
1238 | return 0; | |
1239 | cn = cn->hprev; | |
1240 | } | |
1241 | return 1; | |
1242 | } | |
1243 | ||
bd4c625c LT |
1244 | static void remove_journal_hash(struct super_block *, |
1245 | struct reiserfs_journal_cnode **, | |
1246 | struct reiserfs_journal_list *, unsigned long, | |
1247 | int); | |
1da177e4 LT |
1248 | |
1249 | /* | |
1250 | ** once all the real blocks have been flushed, it is safe to remove them from the | |
1251 | ** journal list for this transaction. Aside from freeing the cnode, this also allows the | |
1252 | ** block to be reallocated for data blocks if it had been deleted. | |
1253 | */ | |
a9dd3643 | 1254 | static void remove_all_from_journal_list(struct super_block *sb, |
bd4c625c LT |
1255 | struct reiserfs_journal_list *jl, |
1256 | int debug) | |
1257 | { | |
a9dd3643 | 1258 | struct reiserfs_journal *journal = SB_JOURNAL(sb); |
bd4c625c LT |
1259 | struct reiserfs_journal_cnode *cn, *last; |
1260 | cn = jl->j_realblock; | |
1261 | ||
1262 | /* which is better, to lock once around the whole loop, or | |
1263 | ** to lock for each call to remove_journal_hash? | |
1264 | */ | |
1265 | while (cn) { | |
1266 | if (cn->blocknr != 0) { | |
1267 | if (debug) { | |
a9dd3643 | 1268 | reiserfs_warning(sb, "reiserfs-2201", |
bd4c625c LT |
1269 | "block %u, bh is %d, state %ld", |
1270 | cn->blocknr, cn->bh ? 1 : 0, | |
1271 | cn->state); | |
1272 | } | |
1273 | cn->state = 0; | |
a9dd3643 | 1274 | remove_journal_hash(sb, journal->j_list_hash_table, |
bd4c625c LT |
1275 | jl, cn->blocknr, 1); |
1276 | } | |
1277 | last = cn; | |
1278 | cn = cn->next; | |
a9dd3643 | 1279 | free_cnode(sb, last); |
bd4c625c LT |
1280 | } |
1281 | jl->j_realblock = NULL; | |
1da177e4 LT |
1282 | } |
1283 | ||
1284 | /* | |
1285 | ** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block. | |
1286 | ** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start | |
1287 | ** releasing blocks in this transaction for reuse as data blocks. | |
1288 | ** called by flush_journal_list, before it calls remove_all_from_journal_list | |
1289 | ** | |
1290 | */ | |
a9dd3643 | 1291 | static int _update_journal_header_block(struct super_block *sb, |
bd4c625c | 1292 | unsigned long offset, |
600ed416 | 1293 | unsigned int trans_id) |
bd4c625c LT |
1294 | { |
1295 | struct reiserfs_journal_header *jh; | |
a9dd3643 | 1296 | struct reiserfs_journal *journal = SB_JOURNAL(sb); |
1da177e4 | 1297 | |
bd4c625c LT |
1298 | if (reiserfs_is_journal_aborted(journal)) |
1299 | return -EIO; | |
1da177e4 | 1300 | |
bd4c625c LT |
1301 | if (trans_id >= journal->j_last_flush_trans_id) { |
1302 | if (buffer_locked((journal->j_header_bh))) { | |
8ebc4232 | 1303 | reiserfs_write_unlock(sb); |
bd4c625c | 1304 | wait_on_buffer((journal->j_header_bh)); |
8ebc4232 | 1305 | reiserfs_write_lock(sb); |
bd4c625c | 1306 | if (unlikely(!buffer_uptodate(journal->j_header_bh))) { |
1da177e4 | 1307 | #ifdef CONFIG_REISERFS_CHECK |
a9dd3643 | 1308 | reiserfs_warning(sb, "journal-699", |
45b03d5e | 1309 | "buffer write failed"); |
1da177e4 | 1310 | #endif |
bd4c625c LT |
1311 | return -EIO; |
1312 | } | |
1313 | } | |
1314 | journal->j_last_flush_trans_id = trans_id; | |
1315 | journal->j_first_unflushed_offset = offset; | |
1316 | jh = (struct reiserfs_journal_header *)(journal->j_header_bh-> | |
1317 | b_data); | |
1318 | jh->j_last_flush_trans_id = cpu_to_le32(trans_id); | |
1319 | jh->j_first_unflushed_offset = cpu_to_le32(offset); | |
1320 | jh->j_mount_id = cpu_to_le32(journal->j_mount_id); | |
1321 | ||
a9dd3643 | 1322 | if (reiserfs_barrier_flush(sb)) { |
bd4c625c LT |
1323 | int ret; |
1324 | lock_buffer(journal->j_header_bh); | |
1325 | ret = submit_barrier_buffer(journal->j_header_bh); | |
1326 | if (ret == -EOPNOTSUPP) { | |
1327 | set_buffer_uptodate(journal->j_header_bh); | |
a9dd3643 | 1328 | disable_barrier(sb); |
bd4c625c LT |
1329 | goto sync; |
1330 | } | |
8ebc4232 | 1331 | reiserfs_write_unlock(sb); |
bd4c625c | 1332 | wait_on_buffer(journal->j_header_bh); |
8ebc4232 | 1333 | reiserfs_write_lock(sb); |
a9dd3643 | 1334 | check_barrier_completion(sb, journal->j_header_bh); |
bd4c625c LT |
1335 | } else { |
1336 | sync: | |
1337 | set_buffer_dirty(journal->j_header_bh); | |
8ebc4232 | 1338 | reiserfs_write_unlock(sb); |
bd4c625c | 1339 | sync_dirty_buffer(journal->j_header_bh); |
8ebc4232 | 1340 | reiserfs_write_lock(sb); |
bd4c625c LT |
1341 | } |
1342 | if (!buffer_uptodate(journal->j_header_bh)) { | |
a9dd3643 | 1343 | reiserfs_warning(sb, "journal-837", |
45b03d5e | 1344 | "IO error during journal replay"); |
bd4c625c LT |
1345 | return -EIO; |
1346 | } | |
1347 | } | |
1348 | return 0; | |
1349 | } | |
1350 | ||
a9dd3643 | 1351 | static int update_journal_header_block(struct super_block *sb, |
bd4c625c | 1352 | unsigned long offset, |
600ed416 | 1353 | unsigned int trans_id) |
bd4c625c | 1354 | { |
a9dd3643 | 1355 | return _update_journal_header_block(sb, offset, trans_id); |
1da177e4 | 1356 | } |
bd4c625c | 1357 | |
0222e657 JM |
1358 | /* |
1359 | ** flush any and all journal lists older than you are | |
1da177e4 LT |
1360 | ** can only be called from flush_journal_list |
1361 | */ | |
a9dd3643 | 1362 | static int flush_older_journal_lists(struct super_block *sb, |
bd4c625c LT |
1363 | struct reiserfs_journal_list *jl) |
1364 | { | |
1365 | struct list_head *entry; | |
1366 | struct reiserfs_journal_list *other_jl; | |
a9dd3643 | 1367 | struct reiserfs_journal *journal = SB_JOURNAL(sb); |
600ed416 | 1368 | unsigned int trans_id = jl->j_trans_id; |
bd4c625c LT |
1369 | |
1370 | /* we know we are the only ones flushing things, no extra race | |
1371 | * protection is required. | |
1372 | */ | |
1373 | restart: | |
1374 | entry = journal->j_journal_list.next; | |
1375 | /* Did we wrap? */ | |
1376 | if (entry == &journal->j_journal_list) | |
1377 | return 0; | |
1378 | other_jl = JOURNAL_LIST_ENTRY(entry); | |
1379 | if (other_jl->j_trans_id < trans_id) { | |
1380 | BUG_ON(other_jl->j_refcount <= 0); | |
1381 | /* do not flush all */ | |
a9dd3643 | 1382 | flush_journal_list(sb, other_jl, 0); |
bd4c625c LT |
1383 | |
1384 | /* other_jl is now deleted from the list */ | |
1385 | goto restart; | |
1386 | } | |
1387 | return 0; | |
1da177e4 LT |
1388 | } |
1389 | ||
1390 | static void del_from_work_list(struct super_block *s, | |
bd4c625c LT |
1391 | struct reiserfs_journal_list *jl) |
1392 | { | |
1393 | struct reiserfs_journal *journal = SB_JOURNAL(s); | |
1394 | if (!list_empty(&jl->j_working_list)) { | |
1395 | list_del_init(&jl->j_working_list); | |
1396 | journal->j_num_work_lists--; | |
1397 | } | |
1da177e4 LT |
1398 | } |
1399 | ||
1400 | /* flush a journal list, both commit and real blocks | |
1401 | ** | |
1402 | ** always set flushall to 1, unless you are calling from inside | |
1403 | ** flush_journal_list | |
1404 | ** | |
0222e657 JM |
1405 | ** IMPORTANT. This can only be called while there are no journal writers, |
1406 | ** and the journal is locked. That means it can only be called from | |
1da177e4 LT |
1407 | ** do_journal_end, or by journal_release |
1408 | */ | |
bd4c625c LT |
1409 | static int flush_journal_list(struct super_block *s, |
1410 | struct reiserfs_journal_list *jl, int flushall) | |
1da177e4 | 1411 | { |
bd4c625c LT |
1412 | struct reiserfs_journal_list *pjl; |
1413 | struct reiserfs_journal_cnode *cn, *last; | |
1414 | int count; | |
1415 | int was_jwait = 0; | |
1416 | int was_dirty = 0; | |
1417 | struct buffer_head *saved_bh; | |
1418 | unsigned long j_len_saved = jl->j_len; | |
1419 | struct reiserfs_journal *journal = SB_JOURNAL(s); | |
1420 | int err = 0; | |
1421 | ||
1422 | BUG_ON(j_len_saved <= 0); | |
1423 | ||
1424 | if (atomic_read(&journal->j_wcount) != 0) { | |
45b03d5e | 1425 | reiserfs_warning(s, "clm-2048", "called with wcount %d", |
bd4c625c LT |
1426 | atomic_read(&journal->j_wcount)); |
1427 | } | |
1428 | BUG_ON(jl->j_trans_id == 0); | |
1da177e4 | 1429 | |
bd4c625c LT |
1430 | /* if flushall == 0, the lock is already held */ |
1431 | if (flushall) { | |
8ebc4232 | 1432 | reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s); |
afe70259 | 1433 | } else if (mutex_trylock(&journal->j_flush_mutex)) { |
bd4c625c LT |
1434 | BUG(); |
1435 | } | |
1da177e4 | 1436 | |
bd4c625c LT |
1437 | count = 0; |
1438 | if (j_len_saved > journal->j_trans_max) { | |
c3a9c210 | 1439 | reiserfs_panic(s, "journal-715", "length is %lu, trans id %lu", |
bd4c625c LT |
1440 | j_len_saved, jl->j_trans_id); |
1441 | return 0; | |
1442 | } | |
1da177e4 | 1443 | |
bd4c625c LT |
1444 | get_fs_excl(); |
1445 | ||
1446 | /* if all the work is already done, get out of here */ | |
1447 | if (atomic_read(&(jl->j_nonzerolen)) <= 0 && | |
1448 | atomic_read(&(jl->j_commit_left)) <= 0) { | |
1449 | goto flush_older_and_return; | |
1450 | } | |
1451 | ||
0222e657 | 1452 | /* start by putting the commit list on disk. This will also flush |
bd4c625c LT |
1453 | ** the commit lists of any olders transactions |
1454 | */ | |
1455 | flush_commit_list(s, jl, 1); | |
1456 | ||
1457 | if (!(jl->j_state & LIST_DIRTY) | |
1458 | && !reiserfs_is_journal_aborted(journal)) | |
1459 | BUG(); | |
1460 | ||
1461 | /* are we done now? */ | |
1462 | if (atomic_read(&(jl->j_nonzerolen)) <= 0 && | |
1463 | atomic_read(&(jl->j_commit_left)) <= 0) { | |
1464 | goto flush_older_and_return; | |
1465 | } | |
1466 | ||
0222e657 JM |
1467 | /* loop through each cnode, see if we need to write it, |
1468 | ** or wait on a more recent transaction, or just ignore it | |
bd4c625c LT |
1469 | */ |
1470 | if (atomic_read(&(journal->j_wcount)) != 0) { | |
c3a9c210 JM |
1471 | reiserfs_panic(s, "journal-844", "journal list is flushing, " |
1472 | "wcount is not 0"); | |
bd4c625c LT |
1473 | } |
1474 | cn = jl->j_realblock; | |
1475 | while (cn) { | |
1476 | was_jwait = 0; | |
1477 | was_dirty = 0; | |
1478 | saved_bh = NULL; | |
1479 | /* blocknr of 0 is no longer in the hash, ignore it */ | |
1480 | if (cn->blocknr == 0) { | |
1481 | goto free_cnode; | |
1482 | } | |
1483 | ||
1484 | /* This transaction failed commit. Don't write out to the disk */ | |
1485 | if (!(jl->j_state & LIST_DIRTY)) | |
1486 | goto free_cnode; | |
1487 | ||
1488 | pjl = find_newer_jl_for_cn(cn); | |
1489 | /* the order is important here. We check pjl to make sure we | |
1490 | ** don't clear BH_JDirty_wait if we aren't the one writing this | |
1491 | ** block to disk | |
1492 | */ | |
1493 | if (!pjl && cn->bh) { | |
1494 | saved_bh = cn->bh; | |
1495 | ||
0222e657 JM |
1496 | /* we do this to make sure nobody releases the buffer while |
1497 | ** we are working with it | |
bd4c625c LT |
1498 | */ |
1499 | get_bh(saved_bh); | |
1500 | ||
1501 | if (buffer_journal_dirty(saved_bh)) { | |
1502 | BUG_ON(!can_dirty(cn)); | |
1503 | was_jwait = 1; | |
1504 | was_dirty = 1; | |
1505 | } else if (can_dirty(cn)) { | |
1506 | /* everything with !pjl && jwait should be writable */ | |
1507 | BUG(); | |
1508 | } | |
1509 | } | |
1510 | ||
1511 | /* if someone has this block in a newer transaction, just make | |
0779bf2d | 1512 | ** sure they are committed, and don't try writing it to disk |
bd4c625c LT |
1513 | */ |
1514 | if (pjl) { | |
1515 | if (atomic_read(&pjl->j_commit_left)) | |
1516 | flush_commit_list(s, pjl, 1); | |
1517 | goto free_cnode; | |
1518 | } | |
1519 | ||
0222e657 JM |
1520 | /* bh == NULL when the block got to disk on its own, OR, |
1521 | ** the block got freed in a future transaction | |
bd4c625c LT |
1522 | */ |
1523 | if (saved_bh == NULL) { | |
1524 | goto free_cnode; | |
1525 | } | |
1526 | ||
1527 | /* this should never happen. kupdate_one_transaction has this list | |
1528 | ** locked while it works, so we should never see a buffer here that | |
1529 | ** is not marked JDirty_wait | |
1530 | */ | |
1531 | if ((!was_jwait) && !buffer_locked(saved_bh)) { | |
45b03d5e JM |
1532 | reiserfs_warning(s, "journal-813", |
1533 | "BAD! buffer %llu %cdirty %cjwait, " | |
bd4c625c LT |
1534 | "not in a newer tranasction", |
1535 | (unsigned long long)saved_bh-> | |
1536 | b_blocknr, was_dirty ? ' ' : '!', | |
1537 | was_jwait ? ' ' : '!'); | |
1538 | } | |
1539 | if (was_dirty) { | |
1540 | /* we inc again because saved_bh gets decremented at free_cnode */ | |
1541 | get_bh(saved_bh); | |
1542 | set_bit(BLOCK_NEEDS_FLUSH, &cn->state); | |
1543 | lock_buffer(saved_bh); | |
1544 | BUG_ON(cn->blocknr != saved_bh->b_blocknr); | |
1545 | if (buffer_dirty(saved_bh)) | |
1546 | submit_logged_buffer(saved_bh); | |
1547 | else | |
1548 | unlock_buffer(saved_bh); | |
1549 | count++; | |
1550 | } else { | |
45b03d5e JM |
1551 | reiserfs_warning(s, "clm-2082", |
1552 | "Unable to flush buffer %llu in %s", | |
bd4c625c | 1553 | (unsigned long long)saved_bh-> |
fbe5498b | 1554 | b_blocknr, __func__); |
bd4c625c LT |
1555 | } |
1556 | free_cnode: | |
1557 | last = cn; | |
1558 | cn = cn->next; | |
1559 | if (saved_bh) { | |
1560 | /* we incremented this to keep others from taking the buffer head away */ | |
1561 | put_bh(saved_bh); | |
1562 | if (atomic_read(&(saved_bh->b_count)) < 0) { | |
45b03d5e JM |
1563 | reiserfs_warning(s, "journal-945", |
1564 | "saved_bh->b_count < 0"); | |
bd4c625c LT |
1565 | } |
1566 | } | |
1567 | } | |
1568 | if (count > 0) { | |
1569 | cn = jl->j_realblock; | |
1570 | while (cn) { | |
1571 | if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) { | |
1572 | if (!cn->bh) { | |
c3a9c210 JM |
1573 | reiserfs_panic(s, "journal-1011", |
1574 | "cn->bh is NULL"); | |
bd4c625c | 1575 | } |
8ebc4232 FW |
1576 | |
1577 | reiserfs_write_unlock(s); | |
bd4c625c | 1578 | wait_on_buffer(cn->bh); |
8ebc4232 FW |
1579 | reiserfs_write_lock(s); |
1580 | ||
bd4c625c | 1581 | if (!cn->bh) { |
c3a9c210 JM |
1582 | reiserfs_panic(s, "journal-1012", |
1583 | "cn->bh is NULL"); | |
bd4c625c LT |
1584 | } |
1585 | if (unlikely(!buffer_uptodate(cn->bh))) { | |
1586 | #ifdef CONFIG_REISERFS_CHECK | |
45b03d5e JM |
1587 | reiserfs_warning(s, "journal-949", |
1588 | "buffer write failed"); | |
bd4c625c LT |
1589 | #endif |
1590 | err = -EIO; | |
1591 | } | |
1592 | /* note, we must clear the JDirty_wait bit after the up to date | |
1593 | ** check, otherwise we race against our flushpage routine | |
1594 | */ | |
1595 | BUG_ON(!test_clear_buffer_journal_dirty | |
1596 | (cn->bh)); | |
1597 | ||
398c95bd | 1598 | /* drop one ref for us */ |
bd4c625c | 1599 | put_bh(cn->bh); |
398c95bd CM |
1600 | /* drop one ref for journal_mark_dirty */ |
1601 | release_buffer_page(cn->bh); | |
bd4c625c LT |
1602 | } |
1603 | cn = cn->next; | |
1604 | } | |
1605 | } | |
1606 | ||
1607 | if (err) | |
1608 | reiserfs_abort(s, -EIO, | |
1609 | "Write error while pushing transaction to disk in %s", | |
fbe5498b | 1610 | __func__); |
bd4c625c LT |
1611 | flush_older_and_return: |
1612 | ||
0222e657 | 1613 | /* before we can update the journal header block, we _must_ flush all |
bd4c625c LT |
1614 | ** real blocks from all older transactions to disk. This is because |
1615 | ** once the header block is updated, this transaction will not be | |
1616 | ** replayed after a crash | |
1617 | */ | |
1618 | if (flushall) { | |
1619 | flush_older_journal_lists(s, jl); | |
1620 | } | |
1621 | ||
1622 | err = journal->j_errno; | |
0222e657 | 1623 | /* before we can remove everything from the hash tables for this |
bd4c625c LT |
1624 | ** transaction, we must make sure it can never be replayed |
1625 | ** | |
1626 | ** since we are only called from do_journal_end, we know for sure there | |
1627 | ** are no allocations going on while we are flushing journal lists. So, | |
1628 | ** we only need to update the journal header block for the last list | |
1629 | ** being flushed | |
1630 | */ | |
1631 | if (!err && flushall) { | |
1632 | err = | |
1633 | update_journal_header_block(s, | |
1634 | (jl->j_start + jl->j_len + | |
1635 | 2) % SB_ONDISK_JOURNAL_SIZE(s), | |
1636 | jl->j_trans_id); | |
1637 | if (err) | |
1638 | reiserfs_abort(s, -EIO, | |
1639 | "Write error while updating journal header in %s", | |
fbe5498b | 1640 | __func__); |
bd4c625c LT |
1641 | } |
1642 | remove_all_from_journal_list(s, jl, 0); | |
1643 | list_del_init(&jl->j_list); | |
1644 | journal->j_num_lists--; | |
1645 | del_from_work_list(s, jl); | |
1646 | ||
1647 | if (journal->j_last_flush_id != 0 && | |
1648 | (jl->j_trans_id - journal->j_last_flush_id) != 1) { | |
45b03d5e | 1649 | reiserfs_warning(s, "clm-2201", "last flush %lu, current %lu", |
bd4c625c LT |
1650 | journal->j_last_flush_id, jl->j_trans_id); |
1651 | } | |
1652 | journal->j_last_flush_id = jl->j_trans_id; | |
1653 | ||
1654 | /* not strictly required since we are freeing the list, but it should | |
1655 | * help find code using dead lists later on | |
1656 | */ | |
1657 | jl->j_len = 0; | |
1658 | atomic_set(&(jl->j_nonzerolen), 0); | |
1659 | jl->j_start = 0; | |
1660 | jl->j_realblock = NULL; | |
1661 | jl->j_commit_bh = NULL; | |
1662 | jl->j_trans_id = 0; | |
1663 | jl->j_state = 0; | |
1664 | put_journal_list(s, jl); | |
1665 | if (flushall) | |
afe70259 | 1666 | mutex_unlock(&journal->j_flush_mutex); |
bd4c625c LT |
1667 | put_fs_excl(); |
1668 | return err; | |
1669 | } | |
1670 | ||
a3172027 CM |
1671 | static int test_transaction(struct super_block *s, |
1672 | struct reiserfs_journal_list *jl) | |
1673 | { | |
1674 | struct reiserfs_journal_cnode *cn; | |
1675 | ||
1676 | if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) | |
1677 | return 1; | |
1678 | ||
1679 | cn = jl->j_realblock; | |
1680 | while (cn) { | |
1681 | /* if the blocknr == 0, this has been cleared from the hash, | |
1682 | ** skip it | |
1683 | */ | |
1684 | if (cn->blocknr == 0) { | |
1685 | goto next; | |
1686 | } | |
1687 | if (cn->bh && !newer_jl_done(cn)) | |
1688 | return 0; | |
1689 | next: | |
1690 | cn = cn->next; | |
1691 | cond_resched(); | |
1692 | } | |
1693 | return 0; | |
1694 | } | |
1695 | ||
bd4c625c LT |
1696 | static int write_one_transaction(struct super_block *s, |
1697 | struct reiserfs_journal_list *jl, | |
1698 | struct buffer_chunk *chunk) | |
1699 | { | |
1700 | struct reiserfs_journal_cnode *cn; | |
1701 | int ret = 0; | |
1702 | ||
1703 | jl->j_state |= LIST_TOUCHED; | |
1704 | del_from_work_list(s, jl); | |
1705 | if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) { | |
1706 | return 0; | |
1707 | } | |
1708 | ||
1709 | cn = jl->j_realblock; | |
1710 | while (cn) { | |
1711 | /* if the blocknr == 0, this has been cleared from the hash, | |
1712 | ** skip it | |
1713 | */ | |
1714 | if (cn->blocknr == 0) { | |
1715 | goto next; | |
1716 | } | |
1717 | if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) { | |
1718 | struct buffer_head *tmp_bh; | |
1719 | /* we can race against journal_mark_freed when we try | |
1720 | * to lock_buffer(cn->bh), so we have to inc the buffer | |
1721 | * count, and recheck things after locking | |
1722 | */ | |
1723 | tmp_bh = cn->bh; | |
1724 | get_bh(tmp_bh); | |
1725 | lock_buffer(tmp_bh); | |
1726 | if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) { | |
1727 | if (!buffer_journal_dirty(tmp_bh) || | |
1728 | buffer_journal_prepared(tmp_bh)) | |
1729 | BUG(); | |
1730 | add_to_chunk(chunk, tmp_bh, NULL, write_chunk); | |
1731 | ret++; | |
1732 | } else { | |
1733 | /* note, cn->bh might be null now */ | |
1734 | unlock_buffer(tmp_bh); | |
1735 | } | |
1736 | put_bh(tmp_bh); | |
1737 | } | |
1738 | next: | |
1739 | cn = cn->next; | |
1740 | cond_resched(); | |
1741 | } | |
1742 | return ret; | |
1743 | } | |
1744 | ||
1745 | /* used by flush_commit_list */ | |
1746 | static int dirty_one_transaction(struct super_block *s, | |
1747 | struct reiserfs_journal_list *jl) | |
1748 | { | |
1749 | struct reiserfs_journal_cnode *cn; | |
1750 | struct reiserfs_journal_list *pjl; | |
1751 | int ret = 0; | |
1752 | ||
1753 | jl->j_state |= LIST_DIRTY; | |
1754 | cn = jl->j_realblock; | |
1755 | while (cn) { | |
1756 | /* look for a more recent transaction that logged this | |
1757 | ** buffer. Only the most recent transaction with a buffer in | |
1758 | ** it is allowed to send that buffer to disk | |
1759 | */ | |
1760 | pjl = find_newer_jl_for_cn(cn); | |
1761 | if (!pjl && cn->blocknr && cn->bh | |
1762 | && buffer_journal_dirty(cn->bh)) { | |
1763 | BUG_ON(!can_dirty(cn)); | |
1764 | /* if the buffer is prepared, it will either be logged | |
1765 | * or restored. If restored, we need to make sure | |
1766 | * it actually gets marked dirty | |
1767 | */ | |
1768 | clear_buffer_journal_new(cn->bh); | |
1769 | if (buffer_journal_prepared(cn->bh)) { | |
1770 | set_buffer_journal_restore_dirty(cn->bh); | |
1771 | } else { | |
1772 | set_buffer_journal_test(cn->bh); | |
1773 | mark_buffer_dirty(cn->bh); | |
1774 | } | |
1775 | } | |
1776 | cn = cn->next; | |
1777 | } | |
1778 | return ret; | |
1779 | } | |
1780 | ||
1781 | static int kupdate_transactions(struct super_block *s, | |
1782 | struct reiserfs_journal_list *jl, | |
1783 | struct reiserfs_journal_list **next_jl, | |
600ed416 | 1784 | unsigned int *next_trans_id, |
bd4c625c LT |
1785 | int num_blocks, int num_trans) |
1786 | { | |
1787 | int ret = 0; | |
1788 | int written = 0; | |
1789 | int transactions_flushed = 0; | |
600ed416 | 1790 | unsigned int orig_trans_id = jl->j_trans_id; |
bd4c625c LT |
1791 | struct buffer_chunk chunk; |
1792 | struct list_head *entry; | |
1793 | struct reiserfs_journal *journal = SB_JOURNAL(s); | |
1794 | chunk.nr = 0; | |
1795 | ||
a412f9ef | 1796 | reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s); |
bd4c625c LT |
1797 | if (!journal_list_still_alive(s, orig_trans_id)) { |
1798 | goto done; | |
1799 | } | |
1800 | ||
afe70259 | 1801 | /* we've got j_flush_mutex held, nobody is going to delete any |
bd4c625c LT |
1802 | * of these lists out from underneath us |
1803 | */ | |
1804 | while ((num_trans && transactions_flushed < num_trans) || | |
1805 | (!num_trans && written < num_blocks)) { | |
1806 | ||
1807 | if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) || | |
1808 | atomic_read(&jl->j_commit_left) | |
1809 | || !(jl->j_state & LIST_DIRTY)) { | |
1810 | del_from_work_list(s, jl); | |
1811 | break; | |
1812 | } | |
1813 | ret = write_one_transaction(s, jl, &chunk); | |
1814 | ||
1815 | if (ret < 0) | |
1816 | goto done; | |
1817 | transactions_flushed++; | |
1818 | written += ret; | |
1819 | entry = jl->j_list.next; | |
1820 | ||
1821 | /* did we wrap? */ | |
1822 | if (entry == &journal->j_journal_list) { | |
1823 | break; | |
1824 | } | |
1825 | jl = JOURNAL_LIST_ENTRY(entry); | |
1826 | ||
1827 | /* don't bother with older transactions */ | |
1828 | if (jl->j_trans_id <= orig_trans_id) | |
1829 | break; | |
1830 | } | |
1831 | if (chunk.nr) { | |
1832 | write_chunk(&chunk); | |
1833 | } | |
1834 | ||
1835 | done: | |
afe70259 | 1836 | mutex_unlock(&journal->j_flush_mutex); |
bd4c625c LT |
1837 | return ret; |
1838 | } | |
1839 | ||
1840 | /* for o_sync and fsync heavy applications, they tend to use | |
1841 | ** all the journa list slots with tiny transactions. These | |
1842 | ** trigger lots and lots of calls to update the header block, which | |
1843 | ** adds seeks and slows things down. | |
1844 | ** | |
1845 | ** This function tries to clear out a large chunk of the journal lists | |
1846 | ** at once, which makes everything faster since only the newest journal | |
1da177e4 LT |
1847 | ** list updates the header block |
1848 | */ | |
1849 | static int flush_used_journal_lists(struct super_block *s, | |
bd4c625c LT |
1850 | struct reiserfs_journal_list *jl) |
1851 | { | |
1852 | unsigned long len = 0; | |
1853 | unsigned long cur_len; | |
1854 | int ret; | |
1855 | int i; | |
1856 | int limit = 256; | |
1857 | struct reiserfs_journal_list *tjl; | |
1858 | struct reiserfs_journal_list *flush_jl; | |
600ed416 | 1859 | unsigned int trans_id; |
bd4c625c LT |
1860 | struct reiserfs_journal *journal = SB_JOURNAL(s); |
1861 | ||
1862 | flush_jl = tjl = jl; | |
1863 | ||
1864 | /* in data logging mode, try harder to flush a lot of blocks */ | |
1865 | if (reiserfs_data_log(s)) | |
1866 | limit = 1024; | |
1867 | /* flush for 256 transactions or limit blocks, whichever comes first */ | |
1868 | for (i = 0; i < 256 && len < limit; i++) { | |
1869 | if (atomic_read(&tjl->j_commit_left) || | |
1870 | tjl->j_trans_id < jl->j_trans_id) { | |
1871 | break; | |
1872 | } | |
1873 | cur_len = atomic_read(&tjl->j_nonzerolen); | |
1874 | if (cur_len > 0) { | |
1875 | tjl->j_state &= ~LIST_TOUCHED; | |
1876 | } | |
1877 | len += cur_len; | |
1878 | flush_jl = tjl; | |
1879 | if (tjl->j_list.next == &journal->j_journal_list) | |
1880 | break; | |
1881 | tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next); | |
1882 | } | |
1883 | /* try to find a group of blocks we can flush across all the | |
1884 | ** transactions, but only bother if we've actually spanned | |
1885 | ** across multiple lists | |
1886 | */ | |
1887 | if (flush_jl != jl) { | |
1888 | ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i); | |
1889 | } | |
1890 | flush_journal_list(s, flush_jl, 1); | |
1891 | return 0; | |
1da177e4 LT |
1892 | } |
1893 | ||
1894 | /* | |
1895 | ** removes any nodes in table with name block and dev as bh. | |
1896 | ** only touchs the hnext and hprev pointers. | |
1897 | */ | |
1898 | void remove_journal_hash(struct super_block *sb, | |
bd4c625c LT |
1899 | struct reiserfs_journal_cnode **table, |
1900 | struct reiserfs_journal_list *jl, | |
1901 | unsigned long block, int remove_freed) | |
1902 | { | |
1903 | struct reiserfs_journal_cnode *cur; | |
1904 | struct reiserfs_journal_cnode **head; | |
1905 | ||
1906 | head = &(journal_hash(table, sb, block)); | |
1907 | if (!head) { | |
1908 | return; | |
1909 | } | |
1910 | cur = *head; | |
1911 | while (cur) { | |
1912 | if (cur->blocknr == block && cur->sb == sb | |
1913 | && (jl == NULL || jl == cur->jlist) | |
1914 | && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) { | |
1915 | if (cur->hnext) { | |
1916 | cur->hnext->hprev = cur->hprev; | |
1917 | } | |
1918 | if (cur->hprev) { | |
1919 | cur->hprev->hnext = cur->hnext; | |
1920 | } else { | |
1921 | *head = cur->hnext; | |
1922 | } | |
1923 | cur->blocknr = 0; | |
1924 | cur->sb = NULL; | |
1925 | cur->state = 0; | |
1926 | if (cur->bh && cur->jlist) /* anybody who clears the cur->bh will also dec the nonzerolen */ | |
1927 | atomic_dec(&(cur->jlist->j_nonzerolen)); | |
1928 | cur->bh = NULL; | |
1929 | cur->jlist = NULL; | |
1930 | } | |
1931 | cur = cur->hnext; | |
1932 | } | |
1933 | } | |
1934 | ||
a9dd3643 | 1935 | static void free_journal_ram(struct super_block *sb) |
bd4c625c | 1936 | { |
a9dd3643 | 1937 | struct reiserfs_journal *journal = SB_JOURNAL(sb); |
d739b42b | 1938 | kfree(journal->j_current_jl); |
bd4c625c LT |
1939 | journal->j_num_lists--; |
1940 | ||
1941 | vfree(journal->j_cnode_free_orig); | |
a9dd3643 JM |
1942 | free_list_bitmaps(sb, journal->j_list_bitmap); |
1943 | free_bitmap_nodes(sb); /* must be after free_list_bitmaps */ | |
bd4c625c LT |
1944 | if (journal->j_header_bh) { |
1945 | brelse(journal->j_header_bh); | |
1946 | } | |
1947 | /* j_header_bh is on the journal dev, make sure not to release the journal | |
1948 | * dev until we brelse j_header_bh | |
1949 | */ | |
a9dd3643 | 1950 | release_journal_dev(sb, journal); |
bd4c625c | 1951 | vfree(journal); |
1da177e4 LT |
1952 | } |
1953 | ||
1954 | /* | |
1955 | ** call on unmount. Only set error to 1 if you haven't made your way out | |
1956 | ** of read_super() yet. Any other caller must keep error at 0. | |
1957 | */ | |
bd4c625c | 1958 | static int do_journal_release(struct reiserfs_transaction_handle *th, |
a9dd3643 | 1959 | struct super_block *sb, int error) |
bd4c625c LT |
1960 | { |
1961 | struct reiserfs_transaction_handle myth; | |
1962 | int flushed = 0; | |
a9dd3643 | 1963 | struct reiserfs_journal *journal = SB_JOURNAL(sb); |
bd4c625c LT |
1964 | |
1965 | /* we only want to flush out transactions if we were called with error == 0 | |
1966 | */ | |
a9dd3643 | 1967 | if (!error && !(sb->s_flags & MS_RDONLY)) { |
bd4c625c LT |
1968 | /* end the current trans */ |
1969 | BUG_ON(!th->t_trans_id); | |
a9dd3643 | 1970 | do_journal_end(th, sb, 10, FLUSH_ALL); |
bd4c625c LT |
1971 | |
1972 | /* make sure something gets logged to force our way into the flush code */ | |
a9dd3643 JM |
1973 | if (!journal_join(&myth, sb, 1)) { |
1974 | reiserfs_prepare_for_journal(sb, | |
1975 | SB_BUFFER_WITH_SB(sb), | |
bd4c625c | 1976 | 1); |
a9dd3643 JM |
1977 | journal_mark_dirty(&myth, sb, |
1978 | SB_BUFFER_WITH_SB(sb)); | |
1979 | do_journal_end(&myth, sb, 1, FLUSH_ALL); | |
bd4c625c LT |
1980 | flushed = 1; |
1981 | } | |
1982 | } | |
1983 | ||
1984 | /* this also catches errors during the do_journal_end above */ | |
1985 | if (!error && reiserfs_is_journal_aborted(journal)) { | |
1986 | memset(&myth, 0, sizeof(myth)); | |
a9dd3643 JM |
1987 | if (!journal_join_abort(&myth, sb, 1)) { |
1988 | reiserfs_prepare_for_journal(sb, | |
1989 | SB_BUFFER_WITH_SB(sb), | |
bd4c625c | 1990 | 1); |
a9dd3643 JM |
1991 | journal_mark_dirty(&myth, sb, |
1992 | SB_BUFFER_WITH_SB(sb)); | |
1993 | do_journal_end(&myth, sb, 1, FLUSH_ALL); | |
bd4c625c LT |
1994 | } |
1995 | } | |
1996 | ||
1997 | reiserfs_mounted_fs_count--; | |
1998 | /* wait for all commits to finish */ | |
a9dd3643 | 1999 | cancel_delayed_work(&SB_JOURNAL(sb)->j_work); |
8ebc4232 FW |
2000 | |
2001 | /* | |
2002 | * We must release the write lock here because | |
2003 | * the workqueue job (flush_async_commit) needs this lock | |
2004 | */ | |
2005 | reiserfs_write_unlock(sb); | |
bd4c625c | 2006 | flush_workqueue(commit_wq); |
8ebc4232 | 2007 | |
bd4c625c LT |
2008 | if (!reiserfs_mounted_fs_count) { |
2009 | destroy_workqueue(commit_wq); | |
2010 | commit_wq = NULL; | |
2011 | } | |
8ebc4232 | 2012 | reiserfs_write_lock(sb); |
bd4c625c | 2013 | |
a9dd3643 | 2014 | free_journal_ram(sb); |
bd4c625c LT |
2015 | |
2016 | return 0; | |
1da177e4 LT |
2017 | } |
2018 | ||
2019 | /* | |
2020 | ** call on unmount. flush all journal trans, release all alloc'd ram | |
2021 | */ | |
bd4c625c | 2022 | int journal_release(struct reiserfs_transaction_handle *th, |
a9dd3643 | 2023 | struct super_block *sb) |
bd4c625c | 2024 | { |
a9dd3643 | 2025 | return do_journal_release(th, sb, 0); |
1da177e4 | 2026 | } |
bd4c625c | 2027 | |
1da177e4 LT |
2028 | /* |
2029 | ** only call from an error condition inside reiserfs_read_super! | |
2030 | */ | |
bd4c625c | 2031 | int journal_release_error(struct reiserfs_transaction_handle *th, |
a9dd3643 | 2032 | struct super_block *sb) |
bd4c625c | 2033 | { |
a9dd3643 | 2034 | return do_journal_release(th, sb, 1); |
1da177e4 LT |
2035 | } |
2036 | ||
2037 | /* compares description block with commit block. returns 1 if they differ, 0 if they are the same */ | |
a9dd3643 | 2038 | static int journal_compare_desc_commit(struct super_block *sb, |
bd4c625c LT |
2039 | struct reiserfs_journal_desc *desc, |
2040 | struct reiserfs_journal_commit *commit) | |
2041 | { | |
2042 | if (get_commit_trans_id(commit) != get_desc_trans_id(desc) || | |
2043 | get_commit_trans_len(commit) != get_desc_trans_len(desc) || | |
a9dd3643 | 2044 | get_commit_trans_len(commit) > SB_JOURNAL(sb)->j_trans_max || |
bd4c625c LT |
2045 | get_commit_trans_len(commit) <= 0) { |
2046 | return 1; | |
2047 | } | |
2048 | return 0; | |
1da177e4 | 2049 | } |
bd4c625c | 2050 | |
0222e657 | 2051 | /* returns 0 if it did not find a description block |
1da177e4 | 2052 | ** returns -1 if it found a corrupt commit block |
0222e657 | 2053 | ** returns 1 if both desc and commit were valid |
1da177e4 | 2054 | */ |
a9dd3643 | 2055 | static int journal_transaction_is_valid(struct super_block *sb, |
bd4c625c | 2056 | struct buffer_head *d_bh, |
600ed416 | 2057 | unsigned int *oldest_invalid_trans_id, |
bd4c625c LT |
2058 | unsigned long *newest_mount_id) |
2059 | { | |
2060 | struct reiserfs_journal_desc *desc; | |
2061 | struct reiserfs_journal_commit *commit; | |
2062 | struct buffer_head *c_bh; | |
2063 | unsigned long offset; | |
2064 | ||
2065 | if (!d_bh) | |
2066 | return 0; | |
2067 | ||
2068 | desc = (struct reiserfs_journal_desc *)d_bh->b_data; | |
2069 | if (get_desc_trans_len(desc) > 0 | |
2070 | && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) { | |
2071 | if (oldest_invalid_trans_id && *oldest_invalid_trans_id | |
2072 | && get_desc_trans_id(desc) > *oldest_invalid_trans_id) { | |
a9dd3643 | 2073 | reiserfs_debug(sb, REISERFS_DEBUG_CODE, |
bd4c625c LT |
2074 | "journal-986: transaction " |
2075 | "is valid returning because trans_id %d is greater than " | |
2076 | "oldest_invalid %lu", | |
2077 | get_desc_trans_id(desc), | |
2078 | *oldest_invalid_trans_id); | |
2079 | return 0; | |
2080 | } | |
2081 | if (newest_mount_id | |
2082 | && *newest_mount_id > get_desc_mount_id(desc)) { | |
a9dd3643 | 2083 | reiserfs_debug(sb, REISERFS_DEBUG_CODE, |
bd4c625c LT |
2084 | "journal-1087: transaction " |
2085 | "is valid returning because mount_id %d is less than " | |
2086 | "newest_mount_id %lu", | |
2087 | get_desc_mount_id(desc), | |
2088 | *newest_mount_id); | |
2089 | return -1; | |
2090 | } | |
a9dd3643 JM |
2091 | if (get_desc_trans_len(desc) > SB_JOURNAL(sb)->j_trans_max) { |
2092 | reiserfs_warning(sb, "journal-2018", | |
45b03d5e JM |
2093 | "Bad transaction length %d " |
2094 | "encountered, ignoring transaction", | |
bd4c625c LT |
2095 | get_desc_trans_len(desc)); |
2096 | return -1; | |
2097 | } | |
a9dd3643 | 2098 | offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb); |
bd4c625c LT |
2099 | |
2100 | /* ok, we have a journal description block, lets see if the transaction was valid */ | |
2101 | c_bh = | |
a9dd3643 JM |
2102 | journal_bread(sb, |
2103 | SB_ONDISK_JOURNAL_1st_BLOCK(sb) + | |
bd4c625c | 2104 | ((offset + get_desc_trans_len(desc) + |
a9dd3643 | 2105 | 1) % SB_ONDISK_JOURNAL_SIZE(sb))); |
bd4c625c LT |
2106 | if (!c_bh) |
2107 | return 0; | |
2108 | commit = (struct reiserfs_journal_commit *)c_bh->b_data; | |
a9dd3643 JM |
2109 | if (journal_compare_desc_commit(sb, desc, commit)) { |
2110 | reiserfs_debug(sb, REISERFS_DEBUG_CODE, | |
bd4c625c LT |
2111 | "journal_transaction_is_valid, commit offset %ld had bad " |
2112 | "time %d or length %d", | |
2113 | c_bh->b_blocknr - | |
a9dd3643 | 2114 | SB_ONDISK_JOURNAL_1st_BLOCK(sb), |
bd4c625c LT |
2115 | get_commit_trans_id(commit), |
2116 | get_commit_trans_len(commit)); | |
2117 | brelse(c_bh); | |
2118 | if (oldest_invalid_trans_id) { | |
2119 | *oldest_invalid_trans_id = | |
2120 | get_desc_trans_id(desc); | |
a9dd3643 | 2121 | reiserfs_debug(sb, REISERFS_DEBUG_CODE, |
bd4c625c LT |
2122 | "journal-1004: " |
2123 | "transaction_is_valid setting oldest invalid trans_id " | |
2124 | "to %d", | |
2125 | get_desc_trans_id(desc)); | |
2126 | } | |
2127 | return -1; | |
2128 | } | |
2129 | brelse(c_bh); | |
a9dd3643 | 2130 | reiserfs_debug(sb, REISERFS_DEBUG_CODE, |
bd4c625c LT |
2131 | "journal-1006: found valid " |
2132 | "transaction start offset %llu, len %d id %d", | |
2133 | d_bh->b_blocknr - | |
a9dd3643 | 2134 | SB_ONDISK_JOURNAL_1st_BLOCK(sb), |
bd4c625c LT |
2135 | get_desc_trans_len(desc), |
2136 | get_desc_trans_id(desc)); | |
2137 | return 1; | |
2138 | } else { | |
2139 | return 0; | |
2140 | } | |
2141 | } | |
2142 | ||
2143 | static void brelse_array(struct buffer_head **heads, int num) | |
2144 | { | |
2145 | int i; | |
2146 | for (i = 0; i < num; i++) { | |
2147 | brelse(heads[i]); | |
2148 | } | |
1da177e4 LT |
2149 | } |
2150 | ||
2151 | /* | |
2152 | ** given the start, and values for the oldest acceptable transactions, | |
2153 | ** this either reads in a replays a transaction, or returns because the transaction | |
2154 | ** is invalid, or too old. | |
2155 | */ | |
a9dd3643 | 2156 | static int journal_read_transaction(struct super_block *sb, |
bd4c625c LT |
2157 | unsigned long cur_dblock, |
2158 | unsigned long oldest_start, | |
600ed416 | 2159 | unsigned int oldest_trans_id, |
bd4c625c LT |
2160 | unsigned long newest_mount_id) |
2161 | { | |
a9dd3643 | 2162 | struct reiserfs_journal *journal = SB_JOURNAL(sb); |
bd4c625c LT |
2163 | struct reiserfs_journal_desc *desc; |
2164 | struct reiserfs_journal_commit *commit; | |
600ed416 | 2165 | unsigned int trans_id = 0; |
bd4c625c LT |
2166 | struct buffer_head *c_bh; |
2167 | struct buffer_head *d_bh; | |
2168 | struct buffer_head **log_blocks = NULL; | |
2169 | struct buffer_head **real_blocks = NULL; | |
600ed416 | 2170 | unsigned int trans_offset; |
bd4c625c LT |
2171 | int i; |
2172 | int trans_half; | |
2173 | ||
a9dd3643 | 2174 | d_bh = journal_bread(sb, cur_dblock); |
bd4c625c LT |
2175 | if (!d_bh) |
2176 | return 1; | |
2177 | desc = (struct reiserfs_journal_desc *)d_bh->b_data; | |
a9dd3643 JM |
2178 | trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb); |
2179 | reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1037: " | |
bd4c625c | 2180 | "journal_read_transaction, offset %llu, len %d mount_id %d", |
a9dd3643 | 2181 | d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb), |
bd4c625c LT |
2182 | get_desc_trans_len(desc), get_desc_mount_id(desc)); |
2183 | if (get_desc_trans_id(desc) < oldest_trans_id) { | |
a9dd3643 | 2184 | reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1039: " |
bd4c625c LT |
2185 | "journal_read_trans skipping because %lu is too old", |
2186 | cur_dblock - | |
a9dd3643 | 2187 | SB_ONDISK_JOURNAL_1st_BLOCK(sb)); |
bd4c625c LT |
2188 | brelse(d_bh); |
2189 | return 1; | |
2190 | } | |
2191 | if (get_desc_mount_id(desc) != newest_mount_id) { | |
a9dd3643 | 2192 | reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1146: " |
bd4c625c LT |
2193 | "journal_read_trans skipping because %d is != " |
2194 | "newest_mount_id %lu", get_desc_mount_id(desc), | |
2195 | newest_mount_id); | |
2196 | brelse(d_bh); | |
2197 | return 1; | |
2198 | } | |
a9dd3643 | 2199 | c_bh = journal_bread(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) + |
bd4c625c | 2200 | ((trans_offset + get_desc_trans_len(desc) + 1) % |
a9dd3643 | 2201 | SB_ONDISK_JOURNAL_SIZE(sb))); |
bd4c625c LT |
2202 | if (!c_bh) { |
2203 | brelse(d_bh); | |
2204 | return 1; | |
2205 | } | |
2206 | commit = (struct reiserfs_journal_commit *)c_bh->b_data; | |
a9dd3643 JM |
2207 | if (journal_compare_desc_commit(sb, desc, commit)) { |
2208 | reiserfs_debug(sb, REISERFS_DEBUG_CODE, | |
bd4c625c LT |
2209 | "journal_read_transaction, " |
2210 | "commit offset %llu had bad time %d or length %d", | |
2211 | c_bh->b_blocknr - | |
a9dd3643 | 2212 | SB_ONDISK_JOURNAL_1st_BLOCK(sb), |
bd4c625c LT |
2213 | get_commit_trans_id(commit), |
2214 | get_commit_trans_len(commit)); | |
2215 | brelse(c_bh); | |
2216 | brelse(d_bh); | |
2217 | return 1; | |
2218 | } | |
2219 | trans_id = get_desc_trans_id(desc); | |
2220 | /* now we know we've got a good transaction, and it was inside the valid time ranges */ | |
d739b42b PE |
2221 | log_blocks = kmalloc(get_desc_trans_len(desc) * |
2222 | sizeof(struct buffer_head *), GFP_NOFS); | |
2223 | real_blocks = kmalloc(get_desc_trans_len(desc) * | |
2224 | sizeof(struct buffer_head *), GFP_NOFS); | |
bd4c625c LT |
2225 | if (!log_blocks || !real_blocks) { |
2226 | brelse(c_bh); | |
2227 | brelse(d_bh); | |
d739b42b PE |
2228 | kfree(log_blocks); |
2229 | kfree(real_blocks); | |
a9dd3643 | 2230 | reiserfs_warning(sb, "journal-1169", |
45b03d5e | 2231 | "kmalloc failed, unable to mount FS"); |
bd4c625c LT |
2232 | return -1; |
2233 | } | |
2234 | /* get all the buffer heads */ | |
a9dd3643 | 2235 | trans_half = journal_trans_half(sb->s_blocksize); |
bd4c625c LT |
2236 | for (i = 0; i < get_desc_trans_len(desc); i++) { |
2237 | log_blocks[i] = | |
a9dd3643 JM |
2238 | journal_getblk(sb, |
2239 | SB_ONDISK_JOURNAL_1st_BLOCK(sb) + | |
bd4c625c | 2240 | (trans_offset + 1 + |
a9dd3643 | 2241 | i) % SB_ONDISK_JOURNAL_SIZE(sb)); |
bd4c625c LT |
2242 | if (i < trans_half) { |
2243 | real_blocks[i] = | |
a9dd3643 | 2244 | sb_getblk(sb, |
bd4c625c LT |
2245 | le32_to_cpu(desc->j_realblock[i])); |
2246 | } else { | |
2247 | real_blocks[i] = | |
a9dd3643 | 2248 | sb_getblk(sb, |
bd4c625c LT |
2249 | le32_to_cpu(commit-> |
2250 | j_realblock[i - trans_half])); | |
2251 | } | |
a9dd3643 JM |
2252 | if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(sb)) { |
2253 | reiserfs_warning(sb, "journal-1207", | |
45b03d5e JM |
2254 | "REPLAY FAILURE fsck required! " |
2255 | "Block to replay is outside of " | |
2256 | "filesystem"); | |
bd4c625c LT |
2257 | goto abort_replay; |
2258 | } | |
2259 | /* make sure we don't try to replay onto log or reserved area */ | |
2260 | if (is_block_in_log_or_reserved_area | |
a9dd3643 JM |
2261 | (sb, real_blocks[i]->b_blocknr)) { |
2262 | reiserfs_warning(sb, "journal-1204", | |
45b03d5e JM |
2263 | "REPLAY FAILURE fsck required! " |
2264 | "Trying to replay onto a log block"); | |
bd4c625c LT |
2265 | abort_replay: |
2266 | brelse_array(log_blocks, i); | |
2267 | brelse_array(real_blocks, i); | |
2268 | brelse(c_bh); | |
2269 | brelse(d_bh); | |
d739b42b PE |
2270 | kfree(log_blocks); |
2271 | kfree(real_blocks); | |
bd4c625c LT |
2272 | return -1; |
2273 | } | |
2274 | } | |
2275 | /* read in the log blocks, memcpy to the corresponding real block */ | |
2276 | ll_rw_block(READ, get_desc_trans_len(desc), log_blocks); | |
2277 | for (i = 0; i < get_desc_trans_len(desc); i++) { | |
8ebc4232 FW |
2278 | |
2279 | reiserfs_write_unlock(sb); | |
bd4c625c | 2280 | wait_on_buffer(log_blocks[i]); |
8ebc4232 FW |
2281 | reiserfs_write_lock(sb); |
2282 | ||
bd4c625c | 2283 | if (!buffer_uptodate(log_blocks[i])) { |
a9dd3643 | 2284 | reiserfs_warning(sb, "journal-1212", |
45b03d5e JM |
2285 | "REPLAY FAILURE fsck required! " |
2286 | "buffer write failed"); | |
bd4c625c LT |
2287 | brelse_array(log_blocks + i, |
2288 | get_desc_trans_len(desc) - i); | |
2289 | brelse_array(real_blocks, get_desc_trans_len(desc)); | |
2290 | brelse(c_bh); | |
2291 | brelse(d_bh); | |
d739b42b PE |
2292 | kfree(log_blocks); |
2293 | kfree(real_blocks); | |
bd4c625c LT |
2294 | return -1; |
2295 | } | |
2296 | memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data, | |
2297 | real_blocks[i]->b_size); | |
2298 | set_buffer_uptodate(real_blocks[i]); | |
2299 | brelse(log_blocks[i]); | |
2300 | } | |
2301 | /* flush out the real blocks */ | |
2302 | for (i = 0; i < get_desc_trans_len(desc); i++) { | |
2303 | set_buffer_dirty(real_blocks[i]); | |
53778ffd | 2304 | ll_rw_block(SWRITE, 1, real_blocks + i); |
bd4c625c LT |
2305 | } |
2306 | for (i = 0; i < get_desc_trans_len(desc); i++) { | |
2307 | wait_on_buffer(real_blocks[i]); | |
2308 | if (!buffer_uptodate(real_blocks[i])) { | |
a9dd3643 | 2309 | reiserfs_warning(sb, "journal-1226", |
45b03d5e JM |
2310 | "REPLAY FAILURE, fsck required! " |
2311 | "buffer write failed"); | |
bd4c625c LT |
2312 | brelse_array(real_blocks + i, |
2313 | get_desc_trans_len(desc) - i); | |
2314 | brelse(c_bh); | |
2315 | brelse(d_bh); | |
d739b42b PE |
2316 | kfree(log_blocks); |
2317 | kfree(real_blocks); | |
bd4c625c LT |
2318 | return -1; |
2319 | } | |
2320 | brelse(real_blocks[i]); | |
2321 | } | |
2322 | cur_dblock = | |
a9dd3643 | 2323 | SB_ONDISK_JOURNAL_1st_BLOCK(sb) + |
bd4c625c | 2324 | ((trans_offset + get_desc_trans_len(desc) + |
a9dd3643 JM |
2325 | 2) % SB_ONDISK_JOURNAL_SIZE(sb)); |
2326 | reiserfs_debug(sb, REISERFS_DEBUG_CODE, | |
bd4c625c | 2327 | "journal-1095: setting journal " "start to offset %ld", |
a9dd3643 | 2328 | cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb)); |
bd4c625c LT |
2329 | |
2330 | /* init starting values for the first transaction, in case this is the last transaction to be replayed. */ | |
a9dd3643 | 2331 | journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb); |
bd4c625c LT |
2332 | journal->j_last_flush_trans_id = trans_id; |
2333 | journal->j_trans_id = trans_id + 1; | |
a44c94a7 AZ |
2334 | /* check for trans_id overflow */ |
2335 | if (journal->j_trans_id == 0) | |
2336 | journal->j_trans_id = 10; | |
bd4c625c LT |
2337 | brelse(c_bh); |
2338 | brelse(d_bh); | |
d739b42b PE |
2339 | kfree(log_blocks); |
2340 | kfree(real_blocks); | |
bd4c625c | 2341 | return 0; |
1da177e4 LT |
2342 | } |
2343 | ||
2344 | /* This function reads blocks starting from block and to max_block of bufsize | |
2345 | size (but no more than BUFNR blocks at a time). This proved to improve | |
2346 | mounting speed on self-rebuilding raid5 arrays at least. | |
2347 | Right now it is only used from journal code. But later we might use it | |
2348 | from other places. | |
2349 | Note: Do not use journal_getblk/sb_getblk functions here! */ | |
3ee16670 JM |
2350 | static struct buffer_head *reiserfs_breada(struct block_device *dev, |
2351 | b_blocknr_t block, int bufsize, | |
2352 | b_blocknr_t max_block) | |
1da177e4 | 2353 | { |
bd4c625c | 2354 | struct buffer_head *bhlist[BUFNR]; |
1da177e4 | 2355 | unsigned int blocks = BUFNR; |
bd4c625c | 2356 | struct buffer_head *bh; |
1da177e4 | 2357 | int i, j; |
bd4c625c LT |
2358 | |
2359 | bh = __getblk(dev, block, bufsize); | |
2360 | if (buffer_uptodate(bh)) | |
2361 | return (bh); | |
2362 | ||
1da177e4 LT |
2363 | if (block + BUFNR > max_block) { |
2364 | blocks = max_block - block; | |
2365 | } | |
2366 | bhlist[0] = bh; | |
2367 | j = 1; | |
2368 | for (i = 1; i < blocks; i++) { | |
bd4c625c LT |
2369 | bh = __getblk(dev, block + i, bufsize); |
2370 | if (buffer_uptodate(bh)) { | |
2371 | brelse(bh); | |
1da177e4 | 2372 | break; |
bd4c625c LT |
2373 | } else |
2374 | bhlist[j++] = bh; | |
1da177e4 | 2375 | } |
bd4c625c LT |
2376 | ll_rw_block(READ, j, bhlist); |
2377 | for (i = 1; i < j; i++) | |
2378 | brelse(bhlist[i]); | |
1da177e4 | 2379 | bh = bhlist[0]; |
bd4c625c LT |
2380 | wait_on_buffer(bh); |
2381 | if (buffer_uptodate(bh)) | |
1da177e4 | 2382 | return bh; |
bd4c625c | 2383 | brelse(bh); |
1da177e4 LT |
2384 | return NULL; |
2385 | } | |
2386 | ||
2387 | /* | |
2388 | ** read and replay the log | |
2389 | ** on a clean unmount, the journal header's next unflushed pointer will be to an invalid | |
2390 | ** transaction. This tests that before finding all the transactions in the log, which makes normal mount times fast. | |
2391 | ** | |
2392 | ** After a crash, this starts with the next unflushed transaction, and replays until it finds one too old, or invalid. | |
2393 | ** | |
2394 | ** On exit, it sets things up so the first transaction will work correctly. | |
2395 | */ | |
a9dd3643 | 2396 | static int journal_read(struct super_block *sb) |
bd4c625c | 2397 | { |
a9dd3643 | 2398 | struct reiserfs_journal *journal = SB_JOURNAL(sb); |
bd4c625c | 2399 | struct reiserfs_journal_desc *desc; |
600ed416 JM |
2400 | unsigned int oldest_trans_id = 0; |
2401 | unsigned int oldest_invalid_trans_id = 0; | |
bd4c625c LT |
2402 | time_t start; |
2403 | unsigned long oldest_start = 0; | |
2404 | unsigned long cur_dblock = 0; | |
2405 | unsigned long newest_mount_id = 9; | |
2406 | struct buffer_head *d_bh; | |
2407 | struct reiserfs_journal_header *jh; | |
2408 | int valid_journal_header = 0; | |
2409 | int replay_count = 0; | |
2410 | int continue_replay = 1; | |
2411 | int ret; | |
2412 | char b[BDEVNAME_SIZE]; | |
2413 | ||
a9dd3643 JM |
2414 | cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(sb); |
2415 | reiserfs_info(sb, "checking transaction log (%s)\n", | |
bd4c625c LT |
2416 | bdevname(journal->j_dev_bd, b)); |
2417 | start = get_seconds(); | |
2418 | ||
0222e657 JM |
2419 | /* step 1, read in the journal header block. Check the transaction it says |
2420 | ** is the first unflushed, and if that transaction is not valid, | |
bd4c625c LT |
2421 | ** replay is done |
2422 | */ | |
a9dd3643 JM |
2423 | journal->j_header_bh = journal_bread(sb, |
2424 | SB_ONDISK_JOURNAL_1st_BLOCK(sb) | |
2425 | + SB_ONDISK_JOURNAL_SIZE(sb)); | |
bd4c625c LT |
2426 | if (!journal->j_header_bh) { |
2427 | return 1; | |
2428 | } | |
2429 | jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data); | |
c499ec24 | 2430 | if (le32_to_cpu(jh->j_first_unflushed_offset) < |
a9dd3643 | 2431 | SB_ONDISK_JOURNAL_SIZE(sb) |
bd4c625c LT |
2432 | && le32_to_cpu(jh->j_last_flush_trans_id) > 0) { |
2433 | oldest_start = | |
a9dd3643 | 2434 | SB_ONDISK_JOURNAL_1st_BLOCK(sb) + |
bd4c625c LT |
2435 | le32_to_cpu(jh->j_first_unflushed_offset); |
2436 | oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1; | |
2437 | newest_mount_id = le32_to_cpu(jh->j_mount_id); | |
a9dd3643 | 2438 | reiserfs_debug(sb, REISERFS_DEBUG_CODE, |
bd4c625c LT |
2439 | "journal-1153: found in " |
2440 | "header: first_unflushed_offset %d, last_flushed_trans_id " | |
2441 | "%lu", le32_to_cpu(jh->j_first_unflushed_offset), | |
2442 | le32_to_cpu(jh->j_last_flush_trans_id)); | |
2443 | valid_journal_header = 1; | |
2444 | ||
0222e657 JM |
2445 | /* now, we try to read the first unflushed offset. If it is not valid, |
2446 | ** there is nothing more we can do, and it makes no sense to read | |
bd4c625c LT |
2447 | ** through the whole log. |
2448 | */ | |
2449 | d_bh = | |
a9dd3643 JM |
2450 | journal_bread(sb, |
2451 | SB_ONDISK_JOURNAL_1st_BLOCK(sb) + | |
bd4c625c | 2452 | le32_to_cpu(jh->j_first_unflushed_offset)); |
a9dd3643 | 2453 | ret = journal_transaction_is_valid(sb, d_bh, NULL, NULL); |
bd4c625c LT |
2454 | if (!ret) { |
2455 | continue_replay = 0; | |
2456 | } | |
2457 | brelse(d_bh); | |
2458 | goto start_log_replay; | |
2459 | } | |
2460 | ||
a9dd3643 JM |
2461 | if (continue_replay && bdev_read_only(sb->s_bdev)) { |
2462 | reiserfs_warning(sb, "clm-2076", | |
45b03d5e | 2463 | "device is readonly, unable to replay log"); |
bd4c625c LT |
2464 | return -1; |
2465 | } | |
2466 | ||
2467 | /* ok, there are transactions that need to be replayed. start with the first log block, find | |
2468 | ** all the valid transactions, and pick out the oldest. | |
2469 | */ | |
2470 | while (continue_replay | |
2471 | && cur_dblock < | |
a9dd3643 JM |
2472 | (SB_ONDISK_JOURNAL_1st_BLOCK(sb) + |
2473 | SB_ONDISK_JOURNAL_SIZE(sb))) { | |
bd4c625c LT |
2474 | /* Note that it is required for blocksize of primary fs device and journal |
2475 | device to be the same */ | |
2476 | d_bh = | |
2477 | reiserfs_breada(journal->j_dev_bd, cur_dblock, | |
a9dd3643 JM |
2478 | sb->s_blocksize, |
2479 | SB_ONDISK_JOURNAL_1st_BLOCK(sb) + | |
2480 | SB_ONDISK_JOURNAL_SIZE(sb)); | |
bd4c625c | 2481 | ret = |
a9dd3643 | 2482 | journal_transaction_is_valid(sb, d_bh, |
bd4c625c LT |
2483 | &oldest_invalid_trans_id, |
2484 | &newest_mount_id); | |
2485 | if (ret == 1) { | |
2486 | desc = (struct reiserfs_journal_desc *)d_bh->b_data; | |
2487 | if (oldest_start == 0) { /* init all oldest_ values */ | |
2488 | oldest_trans_id = get_desc_trans_id(desc); | |
2489 | oldest_start = d_bh->b_blocknr; | |
2490 | newest_mount_id = get_desc_mount_id(desc); | |
a9dd3643 | 2491 | reiserfs_debug(sb, REISERFS_DEBUG_CODE, |
bd4c625c LT |
2492 | "journal-1179: Setting " |
2493 | "oldest_start to offset %llu, trans_id %lu", | |
2494 | oldest_start - | |
2495 | SB_ONDISK_JOURNAL_1st_BLOCK | |
a9dd3643 | 2496 | (sb), oldest_trans_id); |
bd4c625c LT |
2497 | } else if (oldest_trans_id > get_desc_trans_id(desc)) { |
2498 | /* one we just read was older */ | |
2499 | oldest_trans_id = get_desc_trans_id(desc); | |
2500 | oldest_start = d_bh->b_blocknr; | |
a9dd3643 | 2501 | reiserfs_debug(sb, REISERFS_DEBUG_CODE, |
bd4c625c LT |
2502 | "journal-1180: Resetting " |
2503 | "oldest_start to offset %lu, trans_id %lu", | |
2504 | oldest_start - | |
2505 | SB_ONDISK_JOURNAL_1st_BLOCK | |
a9dd3643 | 2506 | (sb), oldest_trans_id); |
bd4c625c LT |
2507 | } |
2508 | if (newest_mount_id < get_desc_mount_id(desc)) { | |
2509 | newest_mount_id = get_desc_mount_id(desc); | |
a9dd3643 | 2510 | reiserfs_debug(sb, REISERFS_DEBUG_CODE, |
bd4c625c LT |
2511 | "journal-1299: Setting " |
2512 | "newest_mount_id to %d", | |
2513 | get_desc_mount_id(desc)); | |
2514 | } | |
2515 | cur_dblock += get_desc_trans_len(desc) + 2; | |
2516 | } else { | |
2517 | cur_dblock++; | |
2518 | } | |
2519 | brelse(d_bh); | |
2520 | } | |
2521 | ||
2522 | start_log_replay: | |
2523 | cur_dblock = oldest_start; | |
2524 | if (oldest_trans_id) { | |
a9dd3643 | 2525 | reiserfs_debug(sb, REISERFS_DEBUG_CODE, |
bd4c625c LT |
2526 | "journal-1206: Starting replay " |
2527 | "from offset %llu, trans_id %lu", | |
a9dd3643 | 2528 | cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb), |
bd4c625c LT |
2529 | oldest_trans_id); |
2530 | ||
2531 | } | |
2532 | replay_count = 0; | |
2533 | while (continue_replay && oldest_trans_id > 0) { | |
2534 | ret = | |
a9dd3643 | 2535 | journal_read_transaction(sb, cur_dblock, oldest_start, |
bd4c625c LT |
2536 | oldest_trans_id, newest_mount_id); |
2537 | if (ret < 0) { | |
2538 | return ret; | |
2539 | } else if (ret != 0) { | |
2540 | break; | |
2541 | } | |
2542 | cur_dblock = | |
a9dd3643 | 2543 | SB_ONDISK_JOURNAL_1st_BLOCK(sb) + journal->j_start; |
bd4c625c LT |
2544 | replay_count++; |
2545 | if (cur_dblock == oldest_start) | |
2546 | break; | |
2547 | } | |
2548 | ||
2549 | if (oldest_trans_id == 0) { | |
a9dd3643 | 2550 | reiserfs_debug(sb, REISERFS_DEBUG_CODE, |
bd4c625c LT |
2551 | "journal-1225: No valid " "transactions found"); |
2552 | } | |
2553 | /* j_start does not get set correctly if we don't replay any transactions. | |
2554 | ** if we had a valid journal_header, set j_start to the first unflushed transaction value, | |
2555 | ** copy the trans_id from the header | |
2556 | */ | |
2557 | if (valid_journal_header && replay_count == 0) { | |
2558 | journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset); | |
2559 | journal->j_trans_id = | |
2560 | le32_to_cpu(jh->j_last_flush_trans_id) + 1; | |
a44c94a7 AZ |
2561 | /* check for trans_id overflow */ |
2562 | if (journal->j_trans_id == 0) | |
2563 | journal->j_trans_id = 10; | |
bd4c625c LT |
2564 | journal->j_last_flush_trans_id = |
2565 | le32_to_cpu(jh->j_last_flush_trans_id); | |
2566 | journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1; | |
2567 | } else { | |
2568 | journal->j_mount_id = newest_mount_id + 1; | |
2569 | } | |
a9dd3643 | 2570 | reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1299: Setting " |
bd4c625c LT |
2571 | "newest_mount_id to %lu", journal->j_mount_id); |
2572 | journal->j_first_unflushed_offset = journal->j_start; | |
2573 | if (replay_count > 0) { | |
a9dd3643 | 2574 | reiserfs_info(sb, |
bd4c625c LT |
2575 | "replayed %d transactions in %lu seconds\n", |
2576 | replay_count, get_seconds() - start); | |
2577 | } | |
a9dd3643 JM |
2578 | if (!bdev_read_only(sb->s_bdev) && |
2579 | _update_journal_header_block(sb, journal->j_start, | |
bd4c625c LT |
2580 | journal->j_last_flush_trans_id)) { |
2581 | /* replay failed, caller must call free_journal_ram and abort | |
2582 | ** the mount | |
2583 | */ | |
2584 | return -1; | |
2585 | } | |
2586 | return 0; | |
1da177e4 LT |
2587 | } |
2588 | ||
2589 | static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s) | |
2590 | { | |
bd4c625c | 2591 | struct reiserfs_journal_list *jl; |
8c777cc4 PE |
2592 | jl = kzalloc(sizeof(struct reiserfs_journal_list), |
2593 | GFP_NOFS | __GFP_NOFAIL); | |
bd4c625c LT |
2594 | INIT_LIST_HEAD(&jl->j_list); |
2595 | INIT_LIST_HEAD(&jl->j_working_list); | |
2596 | INIT_LIST_HEAD(&jl->j_tail_bh_list); | |
2597 | INIT_LIST_HEAD(&jl->j_bh_list); | |
90415dea | 2598 | mutex_init(&jl->j_commit_mutex); |
bd4c625c LT |
2599 | SB_JOURNAL(s)->j_num_lists++; |
2600 | get_journal_list(jl); | |
2601 | return jl; | |
2602 | } | |
2603 | ||
a9dd3643 | 2604 | static void journal_list_init(struct super_block *sb) |
bd4c625c | 2605 | { |
a9dd3643 | 2606 | SB_JOURNAL(sb)->j_current_jl = alloc_journal_list(sb); |
bd4c625c LT |
2607 | } |
2608 | ||
2609 | static int release_journal_dev(struct super_block *super, | |
2610 | struct reiserfs_journal *journal) | |
2611 | { | |
2612 | int result; | |
2613 | ||
2614 | result = 0; | |
2615 | ||
86098fa0 CH |
2616 | if (journal->j_dev_bd != NULL) { |
2617 | if (journal->j_dev_bd->bd_dev != super->s_dev) | |
2618 | bd_release(journal->j_dev_bd); | |
e5eb8caa | 2619 | result = blkdev_put(journal->j_dev_bd, journal->j_dev_mode); |
bd4c625c LT |
2620 | journal->j_dev_bd = NULL; |
2621 | } | |
2622 | ||
2623 | if (result != 0) { | |
45b03d5e JM |
2624 | reiserfs_warning(super, "sh-457", |
2625 | "Cannot release journal device: %i", result); | |
bd4c625c LT |
2626 | } |
2627 | return result; | |
2628 | } | |
2629 | ||
2630 | static int journal_init_dev(struct super_block *super, | |
2631 | struct reiserfs_journal *journal, | |
2632 | const char *jdev_name) | |
1da177e4 LT |
2633 | { |
2634 | int result; | |
2635 | dev_t jdev; | |
aeb5d727 | 2636 | fmode_t blkdev_mode = FMODE_READ | FMODE_WRITE; |
1da177e4 LT |
2637 | char b[BDEVNAME_SIZE]; |
2638 | ||
2639 | result = 0; | |
2640 | ||
bd4c625c | 2641 | journal->j_dev_bd = NULL; |
bd4c625c LT |
2642 | jdev = SB_ONDISK_JOURNAL_DEVICE(super) ? |
2643 | new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev; | |
1da177e4 LT |
2644 | |
2645 | if (bdev_read_only(super->s_bdev)) | |
bd4c625c | 2646 | blkdev_mode = FMODE_READ; |
1da177e4 LT |
2647 | |
2648 | /* there is no "jdev" option and journal is on separate device */ | |
bd4c625c | 2649 | if ((!jdev_name || !jdev_name[0])) { |
1da177e4 | 2650 | journal->j_dev_bd = open_by_devnum(jdev, blkdev_mode); |
e5eb8caa | 2651 | journal->j_dev_mode = blkdev_mode; |
1da177e4 LT |
2652 | if (IS_ERR(journal->j_dev_bd)) { |
2653 | result = PTR_ERR(journal->j_dev_bd); | |
2654 | journal->j_dev_bd = NULL; | |
45b03d5e | 2655 | reiserfs_warning(super, "sh-458", |
bd4c625c LT |
2656 | "cannot init journal device '%s': %i", |
2657 | __bdevname(jdev, b), result); | |
1da177e4 | 2658 | return result; |
86098fa0 CH |
2659 | } else if (jdev != super->s_dev) { |
2660 | result = bd_claim(journal->j_dev_bd, journal); | |
2661 | if (result) { | |
9a1c3542 | 2662 | blkdev_put(journal->j_dev_bd, blkdev_mode); |
86098fa0 CH |
2663 | return result; |
2664 | } | |
2665 | ||
1da177e4 | 2666 | set_blocksize(journal->j_dev_bd, super->s_blocksize); |
86098fa0 CH |
2667 | } |
2668 | ||
1da177e4 LT |
2669 | return 0; |
2670 | } | |
2671 | ||
e5eb8caa | 2672 | journal->j_dev_mode = blkdev_mode; |
30c40d2c | 2673 | journal->j_dev_bd = open_bdev_exclusive(jdev_name, |
e5eb8caa | 2674 | blkdev_mode, journal); |
86098fa0 CH |
2675 | if (IS_ERR(journal->j_dev_bd)) { |
2676 | result = PTR_ERR(journal->j_dev_bd); | |
2677 | journal->j_dev_bd = NULL; | |
bd4c625c LT |
2678 | reiserfs_warning(super, |
2679 | "journal_init_dev: Cannot open '%s': %i", | |
2680 | jdev_name, result); | |
86098fa0 | 2681 | return result; |
1da177e4 | 2682 | } |
86098fa0 CH |
2683 | |
2684 | set_blocksize(journal->j_dev_bd, super->s_blocksize); | |
2685 | reiserfs_info(super, | |
2686 | "journal_init_dev: journal device: %s\n", | |
2687 | bdevname(journal->j_dev_bd, b)); | |
2688 | return 0; | |
1da177e4 LT |
2689 | } |
2690 | ||
cf3d0b81 ES |
2691 | /** |
2692 | * When creating/tuning a file system user can assign some | |
2693 | * journal params within boundaries which depend on the ratio | |
2694 | * blocksize/standard_blocksize. | |
2695 | * | |
2696 | * For blocks >= standard_blocksize transaction size should | |
2697 | * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more | |
2698 | * then JOURNAL_TRANS_MAX_DEFAULT. | |
2699 | * | |
2700 | * For blocks < standard_blocksize these boundaries should be | |
2701 | * decreased proportionally. | |
2702 | */ | |
2703 | #define REISERFS_STANDARD_BLKSIZE (4096) | |
2704 | ||
a9dd3643 | 2705 | static int check_advise_trans_params(struct super_block *sb, |
cf3d0b81 ES |
2706 | struct reiserfs_journal *journal) |
2707 | { | |
2708 | if (journal->j_trans_max) { | |
2709 | /* Non-default journal params. | |
2710 | Do sanity check for them. */ | |
2711 | int ratio = 1; | |
a9dd3643 JM |
2712 | if (sb->s_blocksize < REISERFS_STANDARD_BLKSIZE) |
2713 | ratio = REISERFS_STANDARD_BLKSIZE / sb->s_blocksize; | |
cf3d0b81 ES |
2714 | |
2715 | if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio || | |
2716 | journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio || | |
a9dd3643 | 2717 | SB_ONDISK_JOURNAL_SIZE(sb) / journal->j_trans_max < |
cf3d0b81 | 2718 | JOURNAL_MIN_RATIO) { |
a9dd3643 | 2719 | reiserfs_warning(sb, "sh-462", |
45b03d5e JM |
2720 | "bad transaction max size (%u). " |
2721 | "FSCK?", journal->j_trans_max); | |
cf3d0b81 ES |
2722 | return 1; |
2723 | } | |
2724 | if (journal->j_max_batch != (journal->j_trans_max) * | |
2725 | JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) { | |
a9dd3643 | 2726 | reiserfs_warning(sb, "sh-463", |
45b03d5e JM |
2727 | "bad transaction max batch (%u). " |
2728 | "FSCK?", journal->j_max_batch); | |
cf3d0b81 ES |
2729 | return 1; |
2730 | } | |
2731 | } else { | |
2732 | /* Default journal params. | |
2733 | The file system was created by old version | |
2734 | of mkreiserfs, so some fields contain zeros, | |
2735 | and we need to advise proper values for them */ | |
a9dd3643 JM |
2736 | if (sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) { |
2737 | reiserfs_warning(sb, "sh-464", "bad blocksize (%u)", | |
2738 | sb->s_blocksize); | |
45b03d5e JM |
2739 | return 1; |
2740 | } | |
cf3d0b81 ES |
2741 | journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT; |
2742 | journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT; | |
2743 | journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE; | |
2744 | } | |
2745 | return 0; | |
2746 | } | |
2747 | ||
1da177e4 LT |
2748 | /* |
2749 | ** must be called once on fs mount. calls journal_read for you | |
2750 | */ | |
a9dd3643 | 2751 | int journal_init(struct super_block *sb, const char *j_dev_name, |
bd4c625c LT |
2752 | int old_format, unsigned int commit_max_age) |
2753 | { | |
a9dd3643 | 2754 | int num_cnodes = SB_ONDISK_JOURNAL_SIZE(sb) * 2; |
bd4c625c LT |
2755 | struct buffer_head *bhjh; |
2756 | struct reiserfs_super_block *rs; | |
2757 | struct reiserfs_journal_header *jh; | |
2758 | struct reiserfs_journal *journal; | |
2759 | struct reiserfs_journal_list *jl; | |
2760 | char b[BDEVNAME_SIZE]; | |
2761 | ||
a9dd3643 | 2762 | journal = SB_JOURNAL(sb) = vmalloc(sizeof(struct reiserfs_journal)); |
bd4c625c | 2763 | if (!journal) { |
a9dd3643 | 2764 | reiserfs_warning(sb, "journal-1256", |
45b03d5e | 2765 | "unable to get memory for journal structure"); |
bd4c625c LT |
2766 | return 1; |
2767 | } | |
2768 | memset(journal, 0, sizeof(struct reiserfs_journal)); | |
2769 | INIT_LIST_HEAD(&journal->j_bitmap_nodes); | |
2770 | INIT_LIST_HEAD(&journal->j_prealloc_list); | |
2771 | INIT_LIST_HEAD(&journal->j_working_list); | |
2772 | INIT_LIST_HEAD(&journal->j_journal_list); | |
2773 | journal->j_persistent_trans = 0; | |
a9dd3643 | 2774 | if (reiserfs_allocate_list_bitmaps(sb, |
bd4c625c | 2775 | journal->j_list_bitmap, |
a9dd3643 | 2776 | reiserfs_bmap_count(sb))) |
bd4c625c | 2777 | goto free_and_return; |
a9dd3643 | 2778 | allocate_bitmap_nodes(sb); |
bd4c625c LT |
2779 | |
2780 | /* reserved for journal area support */ | |
a9dd3643 | 2781 | SB_JOURNAL_1st_RESERVED_BLOCK(sb) = (old_format ? |
bd4c625c | 2782 | REISERFS_OLD_DISK_OFFSET_IN_BYTES |
a9dd3643 JM |
2783 | / sb->s_blocksize + |
2784 | reiserfs_bmap_count(sb) + | |
bd4c625c LT |
2785 | 1 : |
2786 | REISERFS_DISK_OFFSET_IN_BYTES / | |
a9dd3643 | 2787 | sb->s_blocksize + 2); |
bd4c625c LT |
2788 | |
2789 | /* Sanity check to see is the standard journal fitting withing first bitmap | |
2790 | (actual for small blocksizes) */ | |
a9dd3643 JM |
2791 | if (!SB_ONDISK_JOURNAL_DEVICE(sb) && |
2792 | (SB_JOURNAL_1st_RESERVED_BLOCK(sb) + | |
2793 | SB_ONDISK_JOURNAL_SIZE(sb) > sb->s_blocksize * 8)) { | |
2794 | reiserfs_warning(sb, "journal-1393", | |
45b03d5e JM |
2795 | "journal does not fit for area addressed " |
2796 | "by first of bitmap blocks. It starts at " | |
bd4c625c | 2797 | "%u and its size is %u. Block size %ld", |
a9dd3643 JM |
2798 | SB_JOURNAL_1st_RESERVED_BLOCK(sb), |
2799 | SB_ONDISK_JOURNAL_SIZE(sb), | |
2800 | sb->s_blocksize); | |
bd4c625c LT |
2801 | goto free_and_return; |
2802 | } | |
2803 | ||
193be0ee FW |
2804 | /* |
2805 | * We need to unlock here to avoid creating the following | |
2806 | * dependency: | |
2807 | * reiserfs_lock -> sysfs_mutex | |
2808 | * Because the reiserfs mmap path creates the following dependency: | |
2809 | * mm->mmap -> reiserfs_lock, hence we have | |
2810 | * mm->mmap -> reiserfs_lock ->sysfs_mutex | |
2811 | * This would ends up in a circular dependency with sysfs readdir path | |
2812 | * which does sysfs_mutex -> mm->mmap_sem | |
2813 | * This is fine because the reiserfs lock is useless in mount path, | |
2814 | * at least until we call journal_begin. We keep it for paranoid | |
2815 | * reasons. | |
2816 | */ | |
2817 | reiserfs_write_unlock(sb); | |
a9dd3643 | 2818 | if (journal_init_dev(sb, journal, j_dev_name) != 0) { |
193be0ee | 2819 | reiserfs_write_lock(sb); |
a9dd3643 | 2820 | reiserfs_warning(sb, "sh-462", |
45b03d5e | 2821 | "unable to initialize jornal device"); |
bd4c625c LT |
2822 | goto free_and_return; |
2823 | } | |
193be0ee | 2824 | reiserfs_write_lock(sb); |
bd4c625c | 2825 | |
a9dd3643 | 2826 | rs = SB_DISK_SUPER_BLOCK(sb); |
bd4c625c LT |
2827 | |
2828 | /* read journal header */ | |
a9dd3643 JM |
2829 | bhjh = journal_bread(sb, |
2830 | SB_ONDISK_JOURNAL_1st_BLOCK(sb) + | |
2831 | SB_ONDISK_JOURNAL_SIZE(sb)); | |
bd4c625c | 2832 | if (!bhjh) { |
a9dd3643 | 2833 | reiserfs_warning(sb, "sh-459", |
45b03d5e | 2834 | "unable to read journal header"); |
bd4c625c LT |
2835 | goto free_and_return; |
2836 | } | |
2837 | jh = (struct reiserfs_journal_header *)(bhjh->b_data); | |
2838 | ||
2839 | /* make sure that journal matches to the super block */ | |
2840 | if (is_reiserfs_jr(rs) | |
2841 | && (le32_to_cpu(jh->jh_journal.jp_journal_magic) != | |
2842 | sb_jp_journal_magic(rs))) { | |
a9dd3643 | 2843 | reiserfs_warning(sb, "sh-460", |
45b03d5e JM |
2844 | "journal header magic %x (device %s) does " |
2845 | "not match to magic found in super block %x", | |
2846 | jh->jh_journal.jp_journal_magic, | |
bd4c625c LT |
2847 | bdevname(journal->j_dev_bd, b), |
2848 | sb_jp_journal_magic(rs)); | |
2849 | brelse(bhjh); | |
2850 | goto free_and_return; | |
2851 | } | |
2852 | ||
2853 | journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max); | |
2854 | journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch); | |
2855 | journal->j_max_commit_age = | |
2856 | le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age); | |
2857 | journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE; | |
2858 | ||
a9dd3643 | 2859 | if (check_advise_trans_params(sb, journal) != 0) |
cf3d0b81 | 2860 | goto free_and_return; |
bd4c625c LT |
2861 | journal->j_default_max_commit_age = journal->j_max_commit_age; |
2862 | ||
2863 | if (commit_max_age != 0) { | |
2864 | journal->j_max_commit_age = commit_max_age; | |
2865 | journal->j_max_trans_age = commit_max_age; | |
2866 | } | |
2867 | ||
a9dd3643 | 2868 | reiserfs_info(sb, "journal params: device %s, size %u, " |
bd4c625c LT |
2869 | "journal first block %u, max trans len %u, max batch %u, " |
2870 | "max commit age %u, max trans age %u\n", | |
2871 | bdevname(journal->j_dev_bd, b), | |
a9dd3643 JM |
2872 | SB_ONDISK_JOURNAL_SIZE(sb), |
2873 | SB_ONDISK_JOURNAL_1st_BLOCK(sb), | |
bd4c625c LT |
2874 | journal->j_trans_max, |
2875 | journal->j_max_batch, | |
2876 | journal->j_max_commit_age, journal->j_max_trans_age); | |
2877 | ||
2878 | brelse(bhjh); | |
2879 | ||
2880 | journal->j_list_bitmap_index = 0; | |
a9dd3643 | 2881 | journal_list_init(sb); |
bd4c625c LT |
2882 | |
2883 | memset(journal->j_list_hash_table, 0, | |
2884 | JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *)); | |
2885 | ||
2886 | INIT_LIST_HEAD(&journal->j_dirty_buffers); | |
2887 | spin_lock_init(&journal->j_dirty_buffers_lock); | |
2888 | ||
2889 | journal->j_start = 0; | |
2890 | journal->j_len = 0; | |
2891 | journal->j_len_alloc = 0; | |
2892 | atomic_set(&(journal->j_wcount), 0); | |
2893 | atomic_set(&(journal->j_async_throttle), 0); | |
2894 | journal->j_bcount = 0; | |
2895 | journal->j_trans_start_time = 0; | |
2896 | journal->j_last = NULL; | |
2897 | journal->j_first = NULL; | |
2898 | init_waitqueue_head(&(journal->j_join_wait)); | |
f68215c4 | 2899 | mutex_init(&journal->j_mutex); |
afe70259 | 2900 | mutex_init(&journal->j_flush_mutex); |
bd4c625c LT |
2901 | |
2902 | journal->j_trans_id = 10; | |
2903 | journal->j_mount_id = 10; | |
2904 | journal->j_state = 0; | |
2905 | atomic_set(&(journal->j_jlock), 0); | |
2906 | journal->j_cnode_free_list = allocate_cnodes(num_cnodes); | |
2907 | journal->j_cnode_free_orig = journal->j_cnode_free_list; | |
2908 | journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0; | |
2909 | journal->j_cnode_used = 0; | |
2910 | journal->j_must_wait = 0; | |
2911 | ||
576f6d79 | 2912 | if (journal->j_cnode_free == 0) { |
a9dd3643 | 2913 | reiserfs_warning(sb, "journal-2004", "Journal cnode memory " |
576f6d79 JM |
2914 | "allocation failed (%ld bytes). Journal is " |
2915 | "too large for available memory. Usually " | |
2916 | "this is due to a journal that is too large.", | |
2917 | sizeof (struct reiserfs_journal_cnode) * num_cnodes); | |
2918 | goto free_and_return; | |
2919 | } | |
2920 | ||
a9dd3643 | 2921 | init_journal_hash(sb); |
bd4c625c | 2922 | jl = journal->j_current_jl; |
a9dd3643 | 2923 | jl->j_list_bitmap = get_list_bitmap(sb, jl); |
bd4c625c | 2924 | if (!jl->j_list_bitmap) { |
a9dd3643 | 2925 | reiserfs_warning(sb, "journal-2005", |
45b03d5e | 2926 | "get_list_bitmap failed for journal list 0"); |
bd4c625c LT |
2927 | goto free_and_return; |
2928 | } | |
a9dd3643 JM |
2929 | if (journal_read(sb) < 0) { |
2930 | reiserfs_warning(sb, "reiserfs-2006", | |
45b03d5e | 2931 | "Replay Failure, unable to mount"); |
bd4c625c LT |
2932 | goto free_and_return; |
2933 | } | |
2934 | ||
2935 | reiserfs_mounted_fs_count++; | |
48f6ba5e FW |
2936 | if (reiserfs_mounted_fs_count <= 1) { |
2937 | reiserfs_write_unlock(sb); | |
bd4c625c | 2938 | commit_wq = create_workqueue("reiserfs"); |
48f6ba5e FW |
2939 | reiserfs_write_lock(sb); |
2940 | } | |
bd4c625c | 2941 | |
c4028958 | 2942 | INIT_DELAYED_WORK(&journal->j_work, flush_async_commits); |
a9dd3643 | 2943 | journal->j_work_sb = sb; |
bd4c625c LT |
2944 | return 0; |
2945 | free_and_return: | |
a9dd3643 | 2946 | free_journal_ram(sb); |
bd4c625c | 2947 | return 1; |
1da177e4 LT |
2948 | } |
2949 | ||
2950 | /* | |
2951 | ** test for a polite end of the current transaction. Used by file_write, and should | |
2952 | ** be used by delete to make sure they don't write more than can fit inside a single | |
2953 | ** transaction | |
2954 | */ | |
bd4c625c LT |
2955 | int journal_transaction_should_end(struct reiserfs_transaction_handle *th, |
2956 | int new_alloc) | |
2957 | { | |
2958 | struct reiserfs_journal *journal = SB_JOURNAL(th->t_super); | |
2959 | time_t now = get_seconds(); | |
2960 | /* cannot restart while nested */ | |
2961 | BUG_ON(!th->t_trans_id); | |
2962 | if (th->t_refcount > 1) | |
2963 | return 0; | |
2964 | if (journal->j_must_wait > 0 || | |
2965 | (journal->j_len_alloc + new_alloc) >= journal->j_max_batch || | |
2966 | atomic_read(&(journal->j_jlock)) || | |
2967 | (now - journal->j_trans_start_time) > journal->j_max_trans_age || | |
2968 | journal->j_cnode_free < (journal->j_trans_max * 3)) { | |
2969 | return 1; | |
2970 | } | |
6ae1ea44 CM |
2971 | /* protected by the BKL here */ |
2972 | journal->j_len_alloc += new_alloc; | |
2973 | th->t_blocks_allocated += new_alloc ; | |
bd4c625c | 2974 | return 0; |
1da177e4 LT |
2975 | } |
2976 | ||
0222e657 | 2977 | /* this must be called inside a transaction, and requires the |
1da177e4 LT |
2978 | ** kernel_lock to be held |
2979 | */ | |
bd4c625c LT |
2980 | void reiserfs_block_writes(struct reiserfs_transaction_handle *th) |
2981 | { | |
2982 | struct reiserfs_journal *journal = SB_JOURNAL(th->t_super); | |
2983 | BUG_ON(!th->t_trans_id); | |
2984 | journal->j_must_wait = 1; | |
2985 | set_bit(J_WRITERS_BLOCKED, &journal->j_state); | |
2986 | return; | |
1da177e4 LT |
2987 | } |
2988 | ||
2989 | /* this must be called without a transaction started, and does not | |
2990 | ** require BKL | |
2991 | */ | |
bd4c625c LT |
2992 | void reiserfs_allow_writes(struct super_block *s) |
2993 | { | |
2994 | struct reiserfs_journal *journal = SB_JOURNAL(s); | |
2995 | clear_bit(J_WRITERS_BLOCKED, &journal->j_state); | |
2996 | wake_up(&journal->j_join_wait); | |
1da177e4 LT |
2997 | } |
2998 | ||
2999 | /* this must be called without a transaction started, and does not | |
3000 | ** require BKL | |
3001 | */ | |
bd4c625c LT |
3002 | void reiserfs_wait_on_write_block(struct super_block *s) |
3003 | { | |
3004 | struct reiserfs_journal *journal = SB_JOURNAL(s); | |
3005 | wait_event(journal->j_join_wait, | |
3006 | !test_bit(J_WRITERS_BLOCKED, &journal->j_state)); | |
3007 | } | |
3008 | ||
3009 | static void queue_log_writer(struct super_block *s) | |
3010 | { | |
3011 | wait_queue_t wait; | |
3012 | struct reiserfs_journal *journal = SB_JOURNAL(s); | |
3013 | set_bit(J_WRITERS_QUEUED, &journal->j_state); | |
3014 | ||
3015 | /* | |
3016 | * we don't want to use wait_event here because | |
3017 | * we only want to wait once. | |
3018 | */ | |
3019 | init_waitqueue_entry(&wait, current); | |
3020 | add_wait_queue(&journal->j_join_wait, &wait); | |
1da177e4 | 3021 | set_current_state(TASK_UNINTERRUPTIBLE); |
8ebc4232 FW |
3022 | if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) { |
3023 | reiserfs_write_unlock(s); | |
bd4c625c | 3024 | schedule(); |
8ebc4232 FW |
3025 | reiserfs_write_lock(s); |
3026 | } | |
5ab2f7e0 | 3027 | __set_current_state(TASK_RUNNING); |
bd4c625c LT |
3028 | remove_wait_queue(&journal->j_join_wait, &wait); |
3029 | } | |
3030 | ||
3031 | static void wake_queued_writers(struct super_block *s) | |
3032 | { | |
3033 | struct reiserfs_journal *journal = SB_JOURNAL(s); | |
3034 | if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state)) | |
3035 | wake_up(&journal->j_join_wait); | |
3036 | } | |
3037 | ||
600ed416 | 3038 | static void let_transaction_grow(struct super_block *sb, unsigned int trans_id) |
bd4c625c LT |
3039 | { |
3040 | struct reiserfs_journal *journal = SB_JOURNAL(sb); | |
3041 | unsigned long bcount = journal->j_bcount; | |
3042 | while (1) { | |
8ebc4232 | 3043 | reiserfs_write_unlock(sb); |
041e0e3b | 3044 | schedule_timeout_uninterruptible(1); |
8ebc4232 | 3045 | reiserfs_write_lock(sb); |
bd4c625c LT |
3046 | journal->j_current_jl->j_state |= LIST_COMMIT_PENDING; |
3047 | while ((atomic_read(&journal->j_wcount) > 0 || | |
3048 | atomic_read(&journal->j_jlock)) && | |
3049 | journal->j_trans_id == trans_id) { | |
3050 | queue_log_writer(sb); | |
3051 | } | |
3052 | if (journal->j_trans_id != trans_id) | |
3053 | break; | |
3054 | if (bcount == journal->j_bcount) | |
3055 | break; | |
3056 | bcount = journal->j_bcount; | |
1da177e4 | 3057 | } |
1da177e4 LT |
3058 | } |
3059 | ||
3060 | /* join == true if you must join an existing transaction. | |
3061 | ** join == false if you can deal with waiting for others to finish | |
3062 | ** | |
3063 | ** this will block until the transaction is joinable. send the number of blocks you | |
3064 | ** expect to use in nblocks. | |
3065 | */ | |
bd4c625c | 3066 | static int do_journal_begin_r(struct reiserfs_transaction_handle *th, |
a9dd3643 | 3067 | struct super_block *sb, unsigned long nblocks, |
bd4c625c LT |
3068 | int join) |
3069 | { | |
3070 | time_t now = get_seconds(); | |
600ed416 | 3071 | unsigned int old_trans_id; |
a9dd3643 | 3072 | struct reiserfs_journal *journal = SB_JOURNAL(sb); |
bd4c625c LT |
3073 | struct reiserfs_transaction_handle myth; |
3074 | int sched_count = 0; | |
3075 | int retval; | |
3076 | ||
a9dd3643 | 3077 | reiserfs_check_lock_depth(sb, "journal_begin"); |
14a61442 | 3078 | BUG_ON(nblocks > journal->j_trans_max); |
bd4c625c | 3079 | |
a9dd3643 | 3080 | PROC_INFO_INC(sb, journal.journal_being); |
bd4c625c LT |
3081 | /* set here for journal_join */ |
3082 | th->t_refcount = 1; | |
a9dd3643 | 3083 | th->t_super = sb; |
bd4c625c LT |
3084 | |
3085 | relock: | |
a9dd3643 | 3086 | lock_journal(sb); |
bd4c625c | 3087 | if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) { |
a9dd3643 | 3088 | unlock_journal(sb); |
bd4c625c LT |
3089 | retval = journal->j_errno; |
3090 | goto out_fail; | |
3091 | } | |
3092 | journal->j_bcount++; | |
3093 | ||
3094 | if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) { | |
a9dd3643 | 3095 | unlock_journal(sb); |
8ebc4232 | 3096 | reiserfs_write_unlock(sb); |
a9dd3643 | 3097 | reiserfs_wait_on_write_block(sb); |
8ebc4232 | 3098 | reiserfs_write_lock(sb); |
a9dd3643 | 3099 | PROC_INFO_INC(sb, journal.journal_relock_writers); |
bd4c625c LT |
3100 | goto relock; |
3101 | } | |
3102 | now = get_seconds(); | |
3103 | ||
3104 | /* if there is no room in the journal OR | |
0222e657 | 3105 | ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning |
bd4c625c LT |
3106 | ** we don't sleep if there aren't other writers |
3107 | */ | |
3108 | ||
3109 | if ((!join && journal->j_must_wait > 0) || | |
3110 | (!join | |
3111 | && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch) | |
3112 | || (!join && atomic_read(&journal->j_wcount) > 0 | |
3113 | && journal->j_trans_start_time > 0 | |
3114 | && (now - journal->j_trans_start_time) > | |
3115 | journal->j_max_trans_age) || (!join | |
3116 | && atomic_read(&journal->j_jlock)) | |
3117 | || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) { | |
3118 | ||
3119 | old_trans_id = journal->j_trans_id; | |
a9dd3643 | 3120 | unlock_journal(sb); /* allow others to finish this transaction */ |
bd4c625c LT |
3121 | |
3122 | if (!join && (journal->j_len_alloc + nblocks + 2) >= | |
3123 | journal->j_max_batch && | |
3124 | ((journal->j_len + nblocks + 2) * 100) < | |
3125 | (journal->j_len_alloc * 75)) { | |
3126 | if (atomic_read(&journal->j_wcount) > 10) { | |
3127 | sched_count++; | |
a9dd3643 | 3128 | queue_log_writer(sb); |
bd4c625c LT |
3129 | goto relock; |
3130 | } | |
3131 | } | |
3132 | /* don't mess with joining the transaction if all we have to do is | |
3133 | * wait for someone else to do a commit | |
3134 | */ | |
3135 | if (atomic_read(&journal->j_jlock)) { | |
3136 | while (journal->j_trans_id == old_trans_id && | |
3137 | atomic_read(&journal->j_jlock)) { | |
a9dd3643 | 3138 | queue_log_writer(sb); |
bd4c625c LT |
3139 | } |
3140 | goto relock; | |
3141 | } | |
a9dd3643 | 3142 | retval = journal_join(&myth, sb, 1); |
bd4c625c LT |
3143 | if (retval) |
3144 | goto out_fail; | |
3145 | ||
3146 | /* someone might have ended the transaction while we joined */ | |
3147 | if (old_trans_id != journal->j_trans_id) { | |
a9dd3643 | 3148 | retval = do_journal_end(&myth, sb, 1, 0); |
bd4c625c | 3149 | } else { |
a9dd3643 | 3150 | retval = do_journal_end(&myth, sb, 1, COMMIT_NOW); |
bd4c625c LT |
3151 | } |
3152 | ||
3153 | if (retval) | |
3154 | goto out_fail; | |
3155 | ||
a9dd3643 | 3156 | PROC_INFO_INC(sb, journal.journal_relock_wcount); |
bd4c625c LT |
3157 | goto relock; |
3158 | } | |
3159 | /* we are the first writer, set trans_id */ | |
3160 | if (journal->j_trans_start_time == 0) { | |
3161 | journal->j_trans_start_time = get_seconds(); | |
3162 | } | |
3163 | atomic_inc(&(journal->j_wcount)); | |
3164 | journal->j_len_alloc += nblocks; | |
3165 | th->t_blocks_logged = 0; | |
3166 | th->t_blocks_allocated = nblocks; | |
3167 | th->t_trans_id = journal->j_trans_id; | |
a9dd3643 | 3168 | unlock_journal(sb); |
bd4c625c LT |
3169 | INIT_LIST_HEAD(&th->t_list); |
3170 | get_fs_excl(); | |
3171 | return 0; | |
3172 | ||
3173 | out_fail: | |
3174 | memset(th, 0, sizeof(*th)); | |
3175 | /* Re-set th->t_super, so we can properly keep track of how many | |
3176 | * persistent transactions there are. We need to do this so if this | |
3177 | * call is part of a failed restart_transaction, we can free it later */ | |
a9dd3643 | 3178 | th->t_super = sb; |
bd4c625c LT |
3179 | return retval; |
3180 | } | |
3181 | ||
3182 | struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct | |
3183 | super_block | |
3184 | *s, | |
3185 | int nblocks) | |
3186 | { | |
3187 | int ret; | |
3188 | struct reiserfs_transaction_handle *th; | |
3189 | ||
3190 | /* if we're nesting into an existing transaction. It will be | |
3191 | ** persistent on its own | |
3192 | */ | |
3193 | if (reiserfs_transaction_running(s)) { | |
3194 | th = current->journal_info; | |
3195 | th->t_refcount++; | |
14a61442 ES |
3196 | BUG_ON(th->t_refcount < 2); |
3197 | ||
bd4c625c LT |
3198 | return th; |
3199 | } | |
d739b42b | 3200 | th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS); |
bd4c625c LT |
3201 | if (!th) |
3202 | return NULL; | |
3203 | ret = journal_begin(th, s, nblocks); | |
3204 | if (ret) { | |
d739b42b | 3205 | kfree(th); |
bd4c625c LT |
3206 | return NULL; |
3207 | } | |
3208 | ||
3209 | SB_JOURNAL(s)->j_persistent_trans++; | |
3210 | return th; | |
3211 | } | |
3212 | ||
3213 | int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th) | |
3214 | { | |
3215 | struct super_block *s = th->t_super; | |
3216 | int ret = 0; | |
3217 | if (th->t_trans_id) | |
3218 | ret = journal_end(th, th->t_super, th->t_blocks_allocated); | |
3219 | else | |
3220 | ret = -EIO; | |
3221 | if (th->t_refcount == 0) { | |
3222 | SB_JOURNAL(s)->j_persistent_trans--; | |
d739b42b | 3223 | kfree(th); |
bd4c625c LT |
3224 | } |
3225 | return ret; | |
3226 | } | |
3227 | ||
3228 | static int journal_join(struct reiserfs_transaction_handle *th, | |
a9dd3643 | 3229 | struct super_block *sb, unsigned long nblocks) |
bd4c625c LT |
3230 | { |
3231 | struct reiserfs_transaction_handle *cur_th = current->journal_info; | |
3232 | ||
3233 | /* this keeps do_journal_end from NULLing out the current->journal_info | |
3234 | ** pointer | |
3235 | */ | |
3236 | th->t_handle_save = cur_th; | |
14a61442 | 3237 | BUG_ON(cur_th && cur_th->t_refcount > 1); |
a9dd3643 | 3238 | return do_journal_begin_r(th, sb, nblocks, JBEGIN_JOIN); |
bd4c625c LT |
3239 | } |
3240 | ||
3241 | int journal_join_abort(struct reiserfs_transaction_handle *th, | |
a9dd3643 | 3242 | struct super_block *sb, unsigned long nblocks) |
bd4c625c LT |
3243 | { |
3244 | struct reiserfs_transaction_handle *cur_th = current->journal_info; | |
3245 | ||
3246 | /* this keeps do_journal_end from NULLing out the current->journal_info | |
3247 | ** pointer | |
3248 | */ | |
3249 | th->t_handle_save = cur_th; | |
14a61442 | 3250 | BUG_ON(cur_th && cur_th->t_refcount > 1); |
a9dd3643 | 3251 | return do_journal_begin_r(th, sb, nblocks, JBEGIN_ABORT); |
bd4c625c LT |
3252 | } |
3253 | ||
3254 | int journal_begin(struct reiserfs_transaction_handle *th, | |
a9dd3643 | 3255 | struct super_block *sb, unsigned long nblocks) |
bd4c625c LT |
3256 | { |
3257 | struct reiserfs_transaction_handle *cur_th = current->journal_info; | |
3258 | int ret; | |
3259 | ||
3260 | th->t_handle_save = NULL; | |
3261 | if (cur_th) { | |
3262 | /* we are nesting into the current transaction */ | |
a9dd3643 | 3263 | if (cur_th->t_super == sb) { |
bd4c625c LT |
3264 | BUG_ON(!cur_th->t_refcount); |
3265 | cur_th->t_refcount++; | |
3266 | memcpy(th, cur_th, sizeof(*th)); | |
3267 | if (th->t_refcount <= 1) | |
a9dd3643 | 3268 | reiserfs_warning(sb, "reiserfs-2005", |
45b03d5e JM |
3269 | "BAD: refcount <= 1, but " |
3270 | "journal_info != 0"); | |
bd4c625c LT |
3271 | return 0; |
3272 | } else { | |
3273 | /* we've ended up with a handle from a different filesystem. | |
3274 | ** save it and restore on journal_end. This should never | |
3275 | ** really happen... | |
3276 | */ | |
a9dd3643 | 3277 | reiserfs_warning(sb, "clm-2100", |
45b03d5e | 3278 | "nesting info a different FS"); |
bd4c625c LT |
3279 | th->t_handle_save = current->journal_info; |
3280 | current->journal_info = th; | |
3281 | } | |
1da177e4 | 3282 | } else { |
bd4c625c LT |
3283 | current->journal_info = th; |
3284 | } | |
a9dd3643 | 3285 | ret = do_journal_begin_r(th, sb, nblocks, JBEGIN_REG); |
14a61442 | 3286 | BUG_ON(current->journal_info != th); |
1da177e4 | 3287 | |
bd4c625c LT |
3288 | /* I guess this boils down to being the reciprocal of clm-2100 above. |
3289 | * If do_journal_begin_r fails, we need to put it back, since journal_end | |
3290 | * won't be called to do it. */ | |
3291 | if (ret) | |
3292 | current->journal_info = th->t_handle_save; | |
3293 | else | |
3294 | BUG_ON(!th->t_refcount); | |
1da177e4 | 3295 | |
bd4c625c | 3296 | return ret; |
1da177e4 LT |
3297 | } |
3298 | ||
3299 | /* | |
3300 | ** puts bh into the current transaction. If it was already there, reorders removes the | |
3301 | ** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order). | |
3302 | ** | |
3303 | ** if it was dirty, cleans and files onto the clean list. I can't let it be dirty again until the | |
3304 | ** transaction is committed. | |
0222e657 | 3305 | ** |
1da177e4 LT |
3306 | ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len. |
3307 | */ | |
bd4c625c | 3308 | int journal_mark_dirty(struct reiserfs_transaction_handle *th, |
a9dd3643 | 3309 | struct super_block *sb, struct buffer_head *bh) |
bd4c625c | 3310 | { |
a9dd3643 | 3311 | struct reiserfs_journal *journal = SB_JOURNAL(sb); |
bd4c625c LT |
3312 | struct reiserfs_journal_cnode *cn = NULL; |
3313 | int count_already_incd = 0; | |
3314 | int prepared = 0; | |
3315 | BUG_ON(!th->t_trans_id); | |
3316 | ||
a9dd3643 | 3317 | PROC_INFO_INC(sb, journal.mark_dirty); |
bd4c625c | 3318 | if (th->t_trans_id != journal->j_trans_id) { |
c3a9c210 JM |
3319 | reiserfs_panic(th->t_super, "journal-1577", |
3320 | "handle trans id %ld != current trans id %ld", | |
bd4c625c LT |
3321 | th->t_trans_id, journal->j_trans_id); |
3322 | } | |
3323 | ||
a9dd3643 | 3324 | sb->s_dirt = 1; |
bd4c625c LT |
3325 | |
3326 | prepared = test_clear_buffer_journal_prepared(bh); | |
3327 | clear_buffer_journal_restore_dirty(bh); | |
3328 | /* already in this transaction, we are done */ | |
3329 | if (buffer_journaled(bh)) { | |
a9dd3643 | 3330 | PROC_INFO_INC(sb, journal.mark_dirty_already); |
bd4c625c LT |
3331 | return 0; |
3332 | } | |
3333 | ||
3334 | /* this must be turned into a panic instead of a warning. We can't allow | |
3335 | ** a dirty or journal_dirty or locked buffer to be logged, as some changes | |
3336 | ** could get to disk too early. NOT GOOD. | |
3337 | */ | |
3338 | if (!prepared || buffer_dirty(bh)) { | |
a9dd3643 | 3339 | reiserfs_warning(sb, "journal-1777", |
45b03d5e | 3340 | "buffer %llu bad state " |
bd4c625c LT |
3341 | "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT", |
3342 | (unsigned long long)bh->b_blocknr, | |
3343 | prepared ? ' ' : '!', | |
3344 | buffer_locked(bh) ? ' ' : '!', | |
3345 | buffer_dirty(bh) ? ' ' : '!', | |
3346 | buffer_journal_dirty(bh) ? ' ' : '!'); | |
3347 | } | |
3348 | ||
3349 | if (atomic_read(&(journal->j_wcount)) <= 0) { | |
a9dd3643 | 3350 | reiserfs_warning(sb, "journal-1409", |
45b03d5e | 3351 | "returning because j_wcount was %d", |
bd4c625c LT |
3352 | atomic_read(&(journal->j_wcount))); |
3353 | return 1; | |
3354 | } | |
0222e657 | 3355 | /* this error means I've screwed up, and we've overflowed the transaction. |
bd4c625c LT |
3356 | ** Nothing can be done here, except make the FS readonly or panic. |
3357 | */ | |
3358 | if (journal->j_len >= journal->j_trans_max) { | |
c3a9c210 JM |
3359 | reiserfs_panic(th->t_super, "journal-1413", |
3360 | "j_len (%lu) is too big", | |
bd4c625c LT |
3361 | journal->j_len); |
3362 | } | |
3363 | ||
3364 | if (buffer_journal_dirty(bh)) { | |
3365 | count_already_incd = 1; | |
a9dd3643 | 3366 | PROC_INFO_INC(sb, journal.mark_dirty_notjournal); |
bd4c625c LT |
3367 | clear_buffer_journal_dirty(bh); |
3368 | } | |
3369 | ||
3370 | if (journal->j_len > journal->j_len_alloc) { | |
3371 | journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT; | |
3372 | } | |
3373 | ||
3374 | set_buffer_journaled(bh); | |
3375 | ||
3376 | /* now put this guy on the end */ | |
3377 | if (!cn) { | |
a9dd3643 | 3378 | cn = get_cnode(sb); |
bd4c625c | 3379 | if (!cn) { |
a9dd3643 | 3380 | reiserfs_panic(sb, "journal-4", "get_cnode failed!"); |
bd4c625c LT |
3381 | } |
3382 | ||
3383 | if (th->t_blocks_logged == th->t_blocks_allocated) { | |
3384 | th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT; | |
3385 | journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT; | |
3386 | } | |
3387 | th->t_blocks_logged++; | |
3388 | journal->j_len++; | |
3389 | ||
3390 | cn->bh = bh; | |
3391 | cn->blocknr = bh->b_blocknr; | |
a9dd3643 | 3392 | cn->sb = sb; |
bd4c625c LT |
3393 | cn->jlist = NULL; |
3394 | insert_journal_hash(journal->j_hash_table, cn); | |
3395 | if (!count_already_incd) { | |
3396 | get_bh(bh); | |
3397 | } | |
3398 | } | |
3399 | cn->next = NULL; | |
3400 | cn->prev = journal->j_last; | |
3401 | cn->bh = bh; | |
3402 | if (journal->j_last) { | |
3403 | journal->j_last->next = cn; | |
3404 | journal->j_last = cn; | |
3405 | } else { | |
3406 | journal->j_first = cn; | |
3407 | journal->j_last = cn; | |
3408 | } | |
3409 | return 0; | |
3410 | } | |
3411 | ||
3412 | int journal_end(struct reiserfs_transaction_handle *th, | |
a9dd3643 | 3413 | struct super_block *sb, unsigned long nblocks) |
bd4c625c LT |
3414 | { |
3415 | if (!current->journal_info && th->t_refcount > 1) | |
a9dd3643 | 3416 | reiserfs_warning(sb, "REISER-NESTING", |
45b03d5e | 3417 | "th NULL, refcount %d", th->t_refcount); |
bd4c625c LT |
3418 | |
3419 | if (!th->t_trans_id) { | |
3420 | WARN_ON(1); | |
3421 | return -EIO; | |
3422 | } | |
3423 | ||
3424 | th->t_refcount--; | |
3425 | if (th->t_refcount > 0) { | |
3426 | struct reiserfs_transaction_handle *cur_th = | |
3427 | current->journal_info; | |
3428 | ||
3429 | /* we aren't allowed to close a nested transaction on a different | |
3430 | ** filesystem from the one in the task struct | |
3431 | */ | |
14a61442 | 3432 | BUG_ON(cur_th->t_super != th->t_super); |
bd4c625c LT |
3433 | |
3434 | if (th != cur_th) { | |
3435 | memcpy(current->journal_info, th, sizeof(*th)); | |
3436 | th->t_trans_id = 0; | |
3437 | } | |
3438 | return 0; | |
3439 | } else { | |
a9dd3643 | 3440 | return do_journal_end(th, sb, nblocks, 0); |
bd4c625c | 3441 | } |
1da177e4 LT |
3442 | } |
3443 | ||
0222e657 | 3444 | /* removes from the current transaction, relsing and descrementing any counters. |
1da177e4 LT |
3445 | ** also files the removed buffer directly onto the clean list |
3446 | ** | |
3447 | ** called by journal_mark_freed when a block has been deleted | |
3448 | ** | |
3449 | ** returns 1 if it cleaned and relsed the buffer. 0 otherwise | |
3450 | */ | |
a9dd3643 | 3451 | static int remove_from_transaction(struct super_block *sb, |
bd4c625c LT |
3452 | b_blocknr_t blocknr, int already_cleaned) |
3453 | { | |
3454 | struct buffer_head *bh; | |
3455 | struct reiserfs_journal_cnode *cn; | |
a9dd3643 | 3456 | struct reiserfs_journal *journal = SB_JOURNAL(sb); |
bd4c625c LT |
3457 | int ret = 0; |
3458 | ||
a9dd3643 | 3459 | cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr); |
bd4c625c LT |
3460 | if (!cn || !cn->bh) { |
3461 | return ret; | |
3462 | } | |
3463 | bh = cn->bh; | |
3464 | if (cn->prev) { | |
3465 | cn->prev->next = cn->next; | |
3466 | } | |
3467 | if (cn->next) { | |
3468 | cn->next->prev = cn->prev; | |
3469 | } | |
3470 | if (cn == journal->j_first) { | |
3471 | journal->j_first = cn->next; | |
3472 | } | |
3473 | if (cn == journal->j_last) { | |
3474 | journal->j_last = cn->prev; | |
3475 | } | |
3476 | if (bh) | |
a9dd3643 | 3477 | remove_journal_hash(sb, journal->j_hash_table, NULL, |
bd4c625c LT |
3478 | bh->b_blocknr, 0); |
3479 | clear_buffer_journaled(bh); /* don't log this one */ | |
3480 | ||
3481 | if (!already_cleaned) { | |
3482 | clear_buffer_journal_dirty(bh); | |
3483 | clear_buffer_dirty(bh); | |
3484 | clear_buffer_journal_test(bh); | |
3485 | put_bh(bh); | |
3486 | if (atomic_read(&(bh->b_count)) < 0) { | |
a9dd3643 | 3487 | reiserfs_warning(sb, "journal-1752", |
45b03d5e | 3488 | "b_count < 0"); |
bd4c625c LT |
3489 | } |
3490 | ret = 1; | |
3491 | } | |
3492 | journal->j_len--; | |
3493 | journal->j_len_alloc--; | |
a9dd3643 | 3494 | free_cnode(sb, cn); |
bd4c625c | 3495 | return ret; |
1da177e4 LT |
3496 | } |
3497 | ||
3498 | /* | |
3499 | ** for any cnode in a journal list, it can only be dirtied of all the | |
0779bf2d | 3500 | ** transactions that include it are committed to disk. |
1da177e4 LT |
3501 | ** this checks through each transaction, and returns 1 if you are allowed to dirty, |
3502 | ** and 0 if you aren't | |
3503 | ** | |
3504 | ** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log | |
3505 | ** blocks for a given transaction on disk | |
3506 | ** | |
3507 | */ | |
bd4c625c LT |
3508 | static int can_dirty(struct reiserfs_journal_cnode *cn) |
3509 | { | |
3510 | struct super_block *sb = cn->sb; | |
3511 | b_blocknr_t blocknr = cn->blocknr; | |
3512 | struct reiserfs_journal_cnode *cur = cn->hprev; | |
3513 | int can_dirty = 1; | |
3514 | ||
3515 | /* first test hprev. These are all newer than cn, so any node here | |
3516 | ** with the same block number and dev means this node can't be sent | |
3517 | ** to disk right now. | |
3518 | */ | |
3519 | while (cur && can_dirty) { | |
3520 | if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb && | |
3521 | cur->blocknr == blocknr) { | |
3522 | can_dirty = 0; | |
3523 | } | |
3524 | cur = cur->hprev; | |
3525 | } | |
3526 | /* then test hnext. These are all older than cn. As long as they | |
3527 | ** are committed to the log, it is safe to write cn to disk | |
3528 | */ | |
3529 | cur = cn->hnext; | |
3530 | while (cur && can_dirty) { | |
3531 | if (cur->jlist && cur->jlist->j_len > 0 && | |
3532 | atomic_read(&(cur->jlist->j_commit_left)) > 0 && cur->bh && | |
3533 | cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) { | |
3534 | can_dirty = 0; | |
3535 | } | |
3536 | cur = cur->hnext; | |
3537 | } | |
3538 | return can_dirty; | |
1da177e4 LT |
3539 | } |
3540 | ||
3541 | /* syncs the commit blocks, but does not force the real buffers to disk | |
0222e657 | 3542 | ** will wait until the current transaction is done/committed before returning |
1da177e4 | 3543 | */ |
bd4c625c | 3544 | int journal_end_sync(struct reiserfs_transaction_handle *th, |
a9dd3643 | 3545 | struct super_block *sb, unsigned long nblocks) |
bd4c625c | 3546 | { |
a9dd3643 | 3547 | struct reiserfs_journal *journal = SB_JOURNAL(sb); |
1da177e4 | 3548 | |
bd4c625c LT |
3549 | BUG_ON(!th->t_trans_id); |
3550 | /* you can sync while nested, very, very bad */ | |
14a61442 | 3551 | BUG_ON(th->t_refcount > 1); |
bd4c625c | 3552 | if (journal->j_len == 0) { |
a9dd3643 | 3553 | reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb), |
bd4c625c | 3554 | 1); |
a9dd3643 | 3555 | journal_mark_dirty(th, sb, SB_BUFFER_WITH_SB(sb)); |
bd4c625c | 3556 | } |
a9dd3643 | 3557 | return do_journal_end(th, sb, nblocks, COMMIT_NOW | WAIT); |
1da177e4 LT |
3558 | } |
3559 | ||
3560 | /* | |
3561 | ** writeback the pending async commits to disk | |
3562 | */ | |
c4028958 | 3563 | static void flush_async_commits(struct work_struct *work) |
bd4c625c | 3564 | { |
c4028958 DH |
3565 | struct reiserfs_journal *journal = |
3566 | container_of(work, struct reiserfs_journal, j_work.work); | |
a9dd3643 | 3567 | struct super_block *sb = journal->j_work_sb; |
bd4c625c LT |
3568 | struct reiserfs_journal_list *jl; |
3569 | struct list_head *entry; | |
3570 | ||
8ebc4232 | 3571 | reiserfs_write_lock(sb); |
bd4c625c LT |
3572 | if (!list_empty(&journal->j_journal_list)) { |
3573 | /* last entry is the youngest, commit it and you get everything */ | |
3574 | entry = journal->j_journal_list.prev; | |
3575 | jl = JOURNAL_LIST_ENTRY(entry); | |
a9dd3643 | 3576 | flush_commit_list(sb, jl, 1); |
bd4c625c | 3577 | } |
8ebc4232 | 3578 | reiserfs_write_unlock(sb); |
1da177e4 LT |
3579 | } |
3580 | ||
3581 | /* | |
3582 | ** flushes any old transactions to disk | |
3583 | ** ends the current transaction if it is too old | |
3584 | */ | |
a9dd3643 | 3585 | int reiserfs_flush_old_commits(struct super_block *sb) |
bd4c625c LT |
3586 | { |
3587 | time_t now; | |
3588 | struct reiserfs_transaction_handle th; | |
a9dd3643 | 3589 | struct reiserfs_journal *journal = SB_JOURNAL(sb); |
bd4c625c LT |
3590 | |
3591 | now = get_seconds(); | |
3592 | /* safety check so we don't flush while we are replaying the log during | |
3593 | * mount | |
3594 | */ | |
3595 | if (list_empty(&journal->j_journal_list)) { | |
3596 | return 0; | |
3597 | } | |
3598 | ||
3599 | /* check the current transaction. If there are no writers, and it is | |
3600 | * too old, finish it, and force the commit blocks to disk | |
3601 | */ | |
3602 | if (atomic_read(&journal->j_wcount) <= 0 && | |
3603 | journal->j_trans_start_time > 0 && | |
3604 | journal->j_len > 0 && | |
3605 | (now - journal->j_trans_start_time) > journal->j_max_trans_age) { | |
a9dd3643 JM |
3606 | if (!journal_join(&th, sb, 1)) { |
3607 | reiserfs_prepare_for_journal(sb, | |
3608 | SB_BUFFER_WITH_SB(sb), | |
bd4c625c | 3609 | 1); |
a9dd3643 JM |
3610 | journal_mark_dirty(&th, sb, |
3611 | SB_BUFFER_WITH_SB(sb)); | |
bd4c625c LT |
3612 | |
3613 | /* we're only being called from kreiserfsd, it makes no sense to do | |
3614 | ** an async commit so that kreiserfsd can do it later | |
3615 | */ | |
a9dd3643 | 3616 | do_journal_end(&th, sb, 1, COMMIT_NOW | WAIT); |
bd4c625c LT |
3617 | } |
3618 | } | |
a9dd3643 | 3619 | return sb->s_dirt; |
1da177e4 LT |
3620 | } |
3621 | ||
3622 | /* | |
3623 | ** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit | |
0222e657 JM |
3624 | ** |
3625 | ** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all | |
1da177e4 LT |
3626 | ** the writers are done. By the time it wakes up, the transaction it was called has already ended, so it just |
3627 | ** flushes the commit list and returns 0. | |
3628 | ** | |
3629 | ** Won't batch when flush or commit_now is set. Also won't batch when others are waiting on j_join_wait. | |
0222e657 | 3630 | ** |
1da177e4 LT |
3631 | ** Note, we can't allow the journal_end to proceed while there are still writers in the log. |
3632 | */ | |
bd4c625c | 3633 | static int check_journal_end(struct reiserfs_transaction_handle *th, |
a9dd3643 | 3634 | struct super_block *sb, unsigned long nblocks, |
bd4c625c LT |
3635 | int flags) |
3636 | { | |
3637 | ||
3638 | time_t now; | |
3639 | int flush = flags & FLUSH_ALL; | |
3640 | int commit_now = flags & COMMIT_NOW; | |
3641 | int wait_on_commit = flags & WAIT; | |
3642 | struct reiserfs_journal_list *jl; | |
a9dd3643 | 3643 | struct reiserfs_journal *journal = SB_JOURNAL(sb); |
bd4c625c LT |
3644 | |
3645 | BUG_ON(!th->t_trans_id); | |
3646 | ||
3647 | if (th->t_trans_id != journal->j_trans_id) { | |
c3a9c210 JM |
3648 | reiserfs_panic(th->t_super, "journal-1577", |
3649 | "handle trans id %ld != current trans id %ld", | |
bd4c625c LT |
3650 | th->t_trans_id, journal->j_trans_id); |
3651 | } | |
3652 | ||
3653 | journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged); | |
3654 | if (atomic_read(&(journal->j_wcount)) > 0) { /* <= 0 is allowed. unmounting might not call begin */ | |
3655 | atomic_dec(&(journal->j_wcount)); | |
3656 | } | |
3657 | ||
0222e657 | 3658 | /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released |
bd4c625c LT |
3659 | ** will be dealt with by next transaction that actually writes something, but should be taken |
3660 | ** care of in this trans | |
3661 | */ | |
14a61442 ES |
3662 | BUG_ON(journal->j_len == 0); |
3663 | ||
bd4c625c LT |
3664 | /* if wcount > 0, and we are called to with flush or commit_now, |
3665 | ** we wait on j_join_wait. We will wake up when the last writer has | |
3666 | ** finished the transaction, and started it on its way to the disk. | |
0222e657 | 3667 | ** Then, we flush the commit or journal list, and just return 0 |
bd4c625c LT |
3668 | ** because the rest of journal end was already done for this transaction. |
3669 | */ | |
3670 | if (atomic_read(&(journal->j_wcount)) > 0) { | |
3671 | if (flush || commit_now) { | |
3672 | unsigned trans_id; | |
3673 | ||
3674 | jl = journal->j_current_jl; | |
3675 | trans_id = jl->j_trans_id; | |
3676 | if (wait_on_commit) | |
3677 | jl->j_state |= LIST_COMMIT_PENDING; | |
3678 | atomic_set(&(journal->j_jlock), 1); | |
3679 | if (flush) { | |
3680 | journal->j_next_full_flush = 1; | |
3681 | } | |
a9dd3643 | 3682 | unlock_journal(sb); |
bd4c625c LT |
3683 | |
3684 | /* sleep while the current transaction is still j_jlocked */ | |
3685 | while (journal->j_trans_id == trans_id) { | |
3686 | if (atomic_read(&journal->j_jlock)) { | |
a9dd3643 | 3687 | queue_log_writer(sb); |
bd4c625c | 3688 | } else { |
a9dd3643 | 3689 | lock_journal(sb); |
bd4c625c LT |
3690 | if (journal->j_trans_id == trans_id) { |
3691 | atomic_set(&(journal->j_jlock), | |
3692 | 1); | |
3693 | } | |
a9dd3643 | 3694 | unlock_journal(sb); |
bd4c625c LT |
3695 | } |
3696 | } | |
14a61442 ES |
3697 | BUG_ON(journal->j_trans_id == trans_id); |
3698 | ||
bd4c625c | 3699 | if (commit_now |
a9dd3643 | 3700 | && journal_list_still_alive(sb, trans_id) |
bd4c625c | 3701 | && wait_on_commit) { |
a9dd3643 | 3702 | flush_commit_list(sb, jl, 1); |
bd4c625c LT |
3703 | } |
3704 | return 0; | |
3705 | } | |
a9dd3643 | 3706 | unlock_journal(sb); |
bd4c625c LT |
3707 | return 0; |
3708 | } | |
3709 | ||
3710 | /* deal with old transactions where we are the last writers */ | |
3711 | now = get_seconds(); | |
3712 | if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) { | |
3713 | commit_now = 1; | |
3714 | journal->j_next_async_flush = 1; | |
3715 | } | |
3716 | /* don't batch when someone is waiting on j_join_wait */ | |
3717 | /* don't batch when syncing the commit or flushing the whole trans */ | |
3718 | if (!(journal->j_must_wait > 0) && !(atomic_read(&(journal->j_jlock))) | |
3719 | && !flush && !commit_now && (journal->j_len < journal->j_max_batch) | |
3720 | && journal->j_len_alloc < journal->j_max_batch | |
3721 | && journal->j_cnode_free > (journal->j_trans_max * 3)) { | |
3722 | journal->j_bcount++; | |
a9dd3643 | 3723 | unlock_journal(sb); |
bd4c625c LT |
3724 | return 0; |
3725 | } | |
3726 | ||
a9dd3643 JM |
3727 | if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(sb)) { |
3728 | reiserfs_panic(sb, "journal-003", | |
c3a9c210 | 3729 | "j_start (%ld) is too high", |
bd4c625c LT |
3730 | journal->j_start); |
3731 | } | |
3732 | return 1; | |
1da177e4 LT |
3733 | } |
3734 | ||
3735 | /* | |
3736 | ** Does all the work that makes deleting blocks safe. | |
3737 | ** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on. | |
0222e657 | 3738 | ** |
1da177e4 LT |
3739 | ** otherwise: |
3740 | ** set a bit for the block in the journal bitmap. That will prevent it from being allocated for unformatted nodes | |
3741 | ** before this transaction has finished. | |
3742 | ** | |
3743 | ** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers. That will prevent any old transactions with | |
3744 | ** this block from trying to flush to the real location. Since we aren't removing the cnode from the journal_list_hash, | |
3745 | ** the block can't be reallocated yet. | |
3746 | ** | |
3747 | ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list. | |
3748 | */ | |
bd4c625c | 3749 | int journal_mark_freed(struct reiserfs_transaction_handle *th, |
a9dd3643 | 3750 | struct super_block *sb, b_blocknr_t blocknr) |
bd4c625c | 3751 | { |
a9dd3643 | 3752 | struct reiserfs_journal *journal = SB_JOURNAL(sb); |
bd4c625c LT |
3753 | struct reiserfs_journal_cnode *cn = NULL; |
3754 | struct buffer_head *bh = NULL; | |
3755 | struct reiserfs_list_bitmap *jb = NULL; | |
3756 | int cleaned = 0; | |
3757 | BUG_ON(!th->t_trans_id); | |
3758 | ||
a9dd3643 | 3759 | cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr); |
bd4c625c LT |
3760 | if (cn && cn->bh) { |
3761 | bh = cn->bh; | |
3762 | get_bh(bh); | |
3763 | } | |
3764 | /* if it is journal new, we just remove it from this transaction */ | |
3765 | if (bh && buffer_journal_new(bh)) { | |
3766 | clear_buffer_journal_new(bh); | |
3767 | clear_prepared_bits(bh); | |
3768 | reiserfs_clean_and_file_buffer(bh); | |
a9dd3643 | 3769 | cleaned = remove_from_transaction(sb, blocknr, cleaned); |
bd4c625c LT |
3770 | } else { |
3771 | /* set the bit for this block in the journal bitmap for this transaction */ | |
3772 | jb = journal->j_current_jl->j_list_bitmap; | |
3773 | if (!jb) { | |
a9dd3643 | 3774 | reiserfs_panic(sb, "journal-1702", |
c3a9c210 | 3775 | "journal_list_bitmap is NULL"); |
bd4c625c | 3776 | } |
a9dd3643 | 3777 | set_bit_in_list_bitmap(sb, blocknr, jb); |
bd4c625c LT |
3778 | |
3779 | /* Note, the entire while loop is not allowed to schedule. */ | |
3780 | ||
3781 | if (bh) { | |
3782 | clear_prepared_bits(bh); | |
3783 | reiserfs_clean_and_file_buffer(bh); | |
3784 | } | |
a9dd3643 | 3785 | cleaned = remove_from_transaction(sb, blocknr, cleaned); |
bd4c625c LT |
3786 | |
3787 | /* find all older transactions with this block, make sure they don't try to write it out */ | |
a9dd3643 | 3788 | cn = get_journal_hash_dev(sb, journal->j_list_hash_table, |
bd4c625c LT |
3789 | blocknr); |
3790 | while (cn) { | |
a9dd3643 | 3791 | if (sb == cn->sb && blocknr == cn->blocknr) { |
bd4c625c LT |
3792 | set_bit(BLOCK_FREED, &cn->state); |
3793 | if (cn->bh) { | |
3794 | if (!cleaned) { | |
3795 | /* remove_from_transaction will brelse the buffer if it was | |
3796 | ** in the current trans | |
3797 | */ | |
3798 | clear_buffer_journal_dirty(cn-> | |
3799 | bh); | |
3800 | clear_buffer_dirty(cn->bh); | |
3801 | clear_buffer_journal_test(cn-> | |
3802 | bh); | |
3803 | cleaned = 1; | |
3804 | put_bh(cn->bh); | |
3805 | if (atomic_read | |
3806 | (&(cn->bh->b_count)) < 0) { | |
a9dd3643 | 3807 | reiserfs_warning(sb, |
45b03d5e JM |
3808 | "journal-2138", |
3809 | "cn->bh->b_count < 0"); | |
bd4c625c LT |
3810 | } |
3811 | } | |
3812 | if (cn->jlist) { /* since we are clearing the bh, we MUST dec nonzerolen */ | |
3813 | atomic_dec(& | |
3814 | (cn->jlist-> | |
3815 | j_nonzerolen)); | |
3816 | } | |
3817 | cn->bh = NULL; | |
3818 | } | |
3819 | } | |
3820 | cn = cn->hnext; | |
3821 | } | |
3822 | } | |
3823 | ||
398c95bd CM |
3824 | if (bh) |
3825 | release_buffer_page(bh); /* get_hash grabs the buffer */ | |
bd4c625c LT |
3826 | return 0; |
3827 | } | |
3828 | ||
3829 | void reiserfs_update_inode_transaction(struct inode *inode) | |
3830 | { | |
3831 | struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb); | |
3832 | REISERFS_I(inode)->i_jl = journal->j_current_jl; | |
3833 | REISERFS_I(inode)->i_trans_id = journal->j_trans_id; | |
1da177e4 LT |
3834 | } |
3835 | ||
3836 | /* | |
3837 | * returns -1 on error, 0 if no commits/barriers were done and 1 | |
3838 | * if a transaction was actually committed and the barrier was done | |
3839 | */ | |
3840 | static int __commit_trans_jl(struct inode *inode, unsigned long id, | |
bd4c625c | 3841 | struct reiserfs_journal_list *jl) |
1da177e4 | 3842 | { |
bd4c625c LT |
3843 | struct reiserfs_transaction_handle th; |
3844 | struct super_block *sb = inode->i_sb; | |
3845 | struct reiserfs_journal *journal = SB_JOURNAL(sb); | |
3846 | int ret = 0; | |
3847 | ||
3848 | /* is it from the current transaction, or from an unknown transaction? */ | |
3849 | if (id == journal->j_trans_id) { | |
3850 | jl = journal->j_current_jl; | |
3851 | /* try to let other writers come in and grow this transaction */ | |
3852 | let_transaction_grow(sb, id); | |
3853 | if (journal->j_trans_id != id) { | |
3854 | goto flush_commit_only; | |
3855 | } | |
1da177e4 | 3856 | |
bd4c625c LT |
3857 | ret = journal_begin(&th, sb, 1); |
3858 | if (ret) | |
3859 | return ret; | |
3860 | ||
3861 | /* someone might have ended this transaction while we joined */ | |
3862 | if (journal->j_trans_id != id) { | |
3863 | reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb), | |
3864 | 1); | |
3865 | journal_mark_dirty(&th, sb, SB_BUFFER_WITH_SB(sb)); | |
3866 | ret = journal_end(&th, sb, 1); | |
3867 | goto flush_commit_only; | |
3868 | } | |
1da177e4 | 3869 | |
bd4c625c LT |
3870 | ret = journal_end_sync(&th, sb, 1); |
3871 | if (!ret) | |
3872 | ret = 1; | |
1da177e4 | 3873 | |
bd4c625c LT |
3874 | } else { |
3875 | /* this gets tricky, we have to make sure the journal list in | |
3876 | * the inode still exists. We know the list is still around | |
3877 | * if we've got a larger transaction id than the oldest list | |
3878 | */ | |
3879 | flush_commit_only: | |
3880 | if (journal_list_still_alive(inode->i_sb, id)) { | |
3881 | /* | |
3882 | * we only set ret to 1 when we know for sure | |
3883 | * the barrier hasn't been started yet on the commit | |
3884 | * block. | |
3885 | */ | |
3886 | if (atomic_read(&jl->j_commit_left) > 1) | |
3887 | ret = 1; | |
3888 | flush_commit_list(sb, jl, 1); | |
3889 | if (journal->j_errno) | |
3890 | ret = journal->j_errno; | |
3891 | } | |
1da177e4 | 3892 | } |
bd4c625c LT |
3893 | /* otherwise the list is gone, and long since committed */ |
3894 | return ret; | |
3895 | } | |
1da177e4 | 3896 | |
bd4c625c LT |
3897 | int reiserfs_commit_for_inode(struct inode *inode) |
3898 | { | |
600ed416 | 3899 | unsigned int id = REISERFS_I(inode)->i_trans_id; |
bd4c625c | 3900 | struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl; |
1da177e4 | 3901 | |
bd4c625c LT |
3902 | /* for the whole inode, assume unset id means it was |
3903 | * changed in the current transaction. More conservative | |
1da177e4 | 3904 | */ |
bd4c625c LT |
3905 | if (!id || !jl) { |
3906 | reiserfs_update_inode_transaction(inode); | |
3907 | id = REISERFS_I(inode)->i_trans_id; | |
3908 | /* jl will be updated in __commit_trans_jl */ | |
3909 | } | |
3910 | ||
3911 | return __commit_trans_jl(inode, id, jl); | |
3912 | } | |
3913 | ||
a9dd3643 | 3914 | void reiserfs_restore_prepared_buffer(struct super_block *sb, |
bd4c625c LT |
3915 | struct buffer_head *bh) |
3916 | { | |
a9dd3643 JM |
3917 | struct reiserfs_journal *journal = SB_JOURNAL(sb); |
3918 | PROC_INFO_INC(sb, journal.restore_prepared); | |
bd4c625c LT |
3919 | if (!bh) { |
3920 | return; | |
3921 | } | |
3922 | if (test_clear_buffer_journal_restore_dirty(bh) && | |
3923 | buffer_journal_dirty(bh)) { | |
3924 | struct reiserfs_journal_cnode *cn; | |
a9dd3643 | 3925 | cn = get_journal_hash_dev(sb, |
bd4c625c LT |
3926 | journal->j_list_hash_table, |
3927 | bh->b_blocknr); | |
3928 | if (cn && can_dirty(cn)) { | |
3929 | set_buffer_journal_test(bh); | |
3930 | mark_buffer_dirty(bh); | |
3931 | } | |
3932 | } | |
3933 | clear_buffer_journal_prepared(bh); | |
3934 | } | |
3935 | ||
3936 | extern struct tree_balance *cur_tb; | |
1da177e4 LT |
3937 | /* |
3938 | ** before we can change a metadata block, we have to make sure it won't | |
3939 | ** be written to disk while we are altering it. So, we must: | |
3940 | ** clean it | |
3941 | ** wait on it. | |
0222e657 | 3942 | ** |
1da177e4 | 3943 | */ |
a9dd3643 | 3944 | int reiserfs_prepare_for_journal(struct super_block *sb, |
bd4c625c LT |
3945 | struct buffer_head *bh, int wait) |
3946 | { | |
a9dd3643 | 3947 | PROC_INFO_INC(sb, journal.prepare); |
bd4c625c | 3948 | |
ca5de404 | 3949 | if (!trylock_buffer(bh)) { |
bd4c625c LT |
3950 | if (!wait) |
3951 | return 0; | |
3952 | lock_buffer(bh); | |
3953 | } | |
3954 | set_buffer_journal_prepared(bh); | |
3955 | if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) { | |
3956 | clear_buffer_journal_test(bh); | |
3957 | set_buffer_journal_restore_dirty(bh); | |
3958 | } | |
3959 | unlock_buffer(bh); | |
3960 | return 1; | |
3961 | } | |
3962 | ||
3963 | static void flush_old_journal_lists(struct super_block *s) | |
3964 | { | |
3965 | struct reiserfs_journal *journal = SB_JOURNAL(s); | |
3966 | struct reiserfs_journal_list *jl; | |
3967 | struct list_head *entry; | |
3968 | time_t now = get_seconds(); | |
3969 | ||
3970 | while (!list_empty(&journal->j_journal_list)) { | |
3971 | entry = journal->j_journal_list.next; | |
3972 | jl = JOURNAL_LIST_ENTRY(entry); | |
3973 | /* this check should always be run, to send old lists to disk */ | |
a3172027 CM |
3974 | if (jl->j_timestamp < (now - (JOURNAL_MAX_TRANS_AGE * 4)) && |
3975 | atomic_read(&jl->j_commit_left) == 0 && | |
3976 | test_transaction(s, jl)) { | |
bd4c625c LT |
3977 | flush_used_journal_lists(s, jl); |
3978 | } else { | |
3979 | break; | |
3980 | } | |
1da177e4 | 3981 | } |
1da177e4 LT |
3982 | } |
3983 | ||
0222e657 | 3984 | /* |
1da177e4 LT |
3985 | ** long and ugly. If flush, will not return until all commit |
3986 | ** blocks and all real buffers in the trans are on disk. | |
3987 | ** If no_async, won't return until all commit blocks are on disk. | |
3988 | ** | |
3989 | ** keep reading, there are comments as you go along | |
3990 | ** | |
3991 | ** If the journal is aborted, we just clean up. Things like flushing | |
3992 | ** journal lists, etc just won't happen. | |
3993 | */ | |
bd4c625c | 3994 | static int do_journal_end(struct reiserfs_transaction_handle *th, |
a9dd3643 | 3995 | struct super_block *sb, unsigned long nblocks, |
bd4c625c LT |
3996 | int flags) |
3997 | { | |
a9dd3643 | 3998 | struct reiserfs_journal *journal = SB_JOURNAL(sb); |
bd4c625c LT |
3999 | struct reiserfs_journal_cnode *cn, *next, *jl_cn; |
4000 | struct reiserfs_journal_cnode *last_cn = NULL; | |
4001 | struct reiserfs_journal_desc *desc; | |
4002 | struct reiserfs_journal_commit *commit; | |
4003 | struct buffer_head *c_bh; /* commit bh */ | |
4004 | struct buffer_head *d_bh; /* desc bh */ | |
4005 | int cur_write_start = 0; /* start index of current log write */ | |
4006 | int old_start; | |
4007 | int i; | |
a44c94a7 AZ |
4008 | int flush; |
4009 | int wait_on_commit; | |
bd4c625c LT |
4010 | struct reiserfs_journal_list *jl, *temp_jl; |
4011 | struct list_head *entry, *safe; | |
4012 | unsigned long jindex; | |
600ed416 | 4013 | unsigned int commit_trans_id; |
bd4c625c LT |
4014 | int trans_half; |
4015 | ||
4016 | BUG_ON(th->t_refcount > 1); | |
4017 | BUG_ON(!th->t_trans_id); | |
4018 | ||
a44c94a7 AZ |
4019 | /* protect flush_older_commits from doing mistakes if the |
4020 | transaction ID counter gets overflowed. */ | |
600ed416 | 4021 | if (th->t_trans_id == ~0U) |
a44c94a7 AZ |
4022 | flags |= FLUSH_ALL | COMMIT_NOW | WAIT; |
4023 | flush = flags & FLUSH_ALL; | |
4024 | wait_on_commit = flags & WAIT; | |
4025 | ||
bd4c625c LT |
4026 | put_fs_excl(); |
4027 | current->journal_info = th->t_handle_save; | |
a9dd3643 | 4028 | reiserfs_check_lock_depth(sb, "journal end"); |
bd4c625c | 4029 | if (journal->j_len == 0) { |
a9dd3643 | 4030 | reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb), |
bd4c625c | 4031 | 1); |
a9dd3643 | 4032 | journal_mark_dirty(th, sb, SB_BUFFER_WITH_SB(sb)); |
bd4c625c | 4033 | } |
1da177e4 | 4034 | |
a9dd3643 | 4035 | lock_journal(sb); |
bd4c625c LT |
4036 | if (journal->j_next_full_flush) { |
4037 | flags |= FLUSH_ALL; | |
4038 | flush = 1; | |
4039 | } | |
4040 | if (journal->j_next_async_flush) { | |
4041 | flags |= COMMIT_NOW | WAIT; | |
4042 | wait_on_commit = 1; | |
4043 | } | |
4044 | ||
0222e657 | 4045 | /* check_journal_end locks the journal, and unlocks if it does not return 1 |
bd4c625c LT |
4046 | ** it tells us if we should continue with the journal_end, or just return |
4047 | */ | |
a9dd3643 JM |
4048 | if (!check_journal_end(th, sb, nblocks, flags)) { |
4049 | sb->s_dirt = 1; | |
4050 | wake_queued_writers(sb); | |
4051 | reiserfs_async_progress_wait(sb); | |
bd4c625c LT |
4052 | goto out; |
4053 | } | |
4054 | ||
4055 | /* check_journal_end might set these, check again */ | |
4056 | if (journal->j_next_full_flush) { | |
4057 | flush = 1; | |
4058 | } | |
4059 | ||
4060 | /* | |
4061 | ** j must wait means we have to flush the log blocks, and the real blocks for | |
4062 | ** this transaction | |
4063 | */ | |
4064 | if (journal->j_must_wait > 0) { | |
4065 | flush = 1; | |
4066 | } | |
1da177e4 | 4067 | #ifdef REISERFS_PREALLOCATE |
ef43bc4f JK |
4068 | /* quota ops might need to nest, setup the journal_info pointer for them |
4069 | * and raise the refcount so that it is > 0. */ | |
bd4c625c | 4070 | current->journal_info = th; |
ef43bc4f | 4071 | th->t_refcount++; |
bd4c625c LT |
4072 | reiserfs_discard_all_prealloc(th); /* it should not involve new blocks into |
4073 | * the transaction */ | |
ef43bc4f | 4074 | th->t_refcount--; |
bd4c625c | 4075 | current->journal_info = th->t_handle_save; |
1da177e4 | 4076 | #endif |
bd4c625c LT |
4077 | |
4078 | /* setup description block */ | |
4079 | d_bh = | |
a9dd3643 JM |
4080 | journal_getblk(sb, |
4081 | SB_ONDISK_JOURNAL_1st_BLOCK(sb) + | |
bd4c625c LT |
4082 | journal->j_start); |
4083 | set_buffer_uptodate(d_bh); | |
4084 | desc = (struct reiserfs_journal_desc *)(d_bh)->b_data; | |
4085 | memset(d_bh->b_data, 0, d_bh->b_size); | |
4086 | memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8); | |
4087 | set_desc_trans_id(desc, journal->j_trans_id); | |
4088 | ||
4089 | /* setup commit block. Don't write (keep it clean too) this one until after everyone else is written */ | |
a9dd3643 | 4090 | c_bh = journal_getblk(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) + |
bd4c625c | 4091 | ((journal->j_start + journal->j_len + |
a9dd3643 | 4092 | 1) % SB_ONDISK_JOURNAL_SIZE(sb))); |
bd4c625c LT |
4093 | commit = (struct reiserfs_journal_commit *)c_bh->b_data; |
4094 | memset(c_bh->b_data, 0, c_bh->b_size); | |
4095 | set_commit_trans_id(commit, journal->j_trans_id); | |
4096 | set_buffer_uptodate(c_bh); | |
4097 | ||
4098 | /* init this journal list */ | |
4099 | jl = journal->j_current_jl; | |
4100 | ||
4101 | /* we lock the commit before doing anything because | |
4102 | * we want to make sure nobody tries to run flush_commit_list until | |
4103 | * the new transaction is fully setup, and we've already flushed the | |
4104 | * ordered bh list | |
4105 | */ | |
8ebc4232 | 4106 | reiserfs_mutex_lock_safe(&jl->j_commit_mutex, sb); |
bd4c625c LT |
4107 | |
4108 | /* save the transaction id in case we need to commit it later */ | |
4109 | commit_trans_id = jl->j_trans_id; | |
4110 | ||
4111 | atomic_set(&jl->j_older_commits_done, 0); | |
4112 | jl->j_trans_id = journal->j_trans_id; | |
4113 | jl->j_timestamp = journal->j_trans_start_time; | |
4114 | jl->j_commit_bh = c_bh; | |
4115 | jl->j_start = journal->j_start; | |
4116 | jl->j_len = journal->j_len; | |
4117 | atomic_set(&jl->j_nonzerolen, journal->j_len); | |
4118 | atomic_set(&jl->j_commit_left, journal->j_len + 2); | |
4119 | jl->j_realblock = NULL; | |
4120 | ||
4121 | /* The ENTIRE FOR LOOP MUST not cause schedule to occur. | |
4122 | ** for each real block, add it to the journal list hash, | |
4123 | ** copy into real block index array in the commit or desc block | |
4124 | */ | |
a9dd3643 | 4125 | trans_half = journal_trans_half(sb->s_blocksize); |
bd4c625c LT |
4126 | for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) { |
4127 | if (buffer_journaled(cn->bh)) { | |
a9dd3643 | 4128 | jl_cn = get_cnode(sb); |
bd4c625c | 4129 | if (!jl_cn) { |
a9dd3643 | 4130 | reiserfs_panic(sb, "journal-1676", |
c3a9c210 | 4131 | "get_cnode returned NULL"); |
bd4c625c LT |
4132 | } |
4133 | if (i == 0) { | |
4134 | jl->j_realblock = jl_cn; | |
4135 | } | |
4136 | jl_cn->prev = last_cn; | |
4137 | jl_cn->next = NULL; | |
4138 | if (last_cn) { | |
4139 | last_cn->next = jl_cn; | |
4140 | } | |
4141 | last_cn = jl_cn; | |
0222e657 | 4142 | /* make sure the block we are trying to log is not a block |
bd4c625c LT |
4143 | of journal or reserved area */ |
4144 | ||
4145 | if (is_block_in_log_or_reserved_area | |
a9dd3643 JM |
4146 | (sb, cn->bh->b_blocknr)) { |
4147 | reiserfs_panic(sb, "journal-2332", | |
c3a9c210 JM |
4148 | "Trying to log block %lu, " |
4149 | "which is a log block", | |
bd4c625c LT |
4150 | cn->bh->b_blocknr); |
4151 | } | |
4152 | jl_cn->blocknr = cn->bh->b_blocknr; | |
4153 | jl_cn->state = 0; | |
a9dd3643 | 4154 | jl_cn->sb = sb; |
bd4c625c LT |
4155 | jl_cn->bh = cn->bh; |
4156 | jl_cn->jlist = jl; | |
4157 | insert_journal_hash(journal->j_list_hash_table, jl_cn); | |
4158 | if (i < trans_half) { | |
4159 | desc->j_realblock[i] = | |
4160 | cpu_to_le32(cn->bh->b_blocknr); | |
4161 | } else { | |
4162 | commit->j_realblock[i - trans_half] = | |
4163 | cpu_to_le32(cn->bh->b_blocknr); | |
4164 | } | |
4165 | } else { | |
4166 | i--; | |
4167 | } | |
4168 | } | |
4169 | set_desc_trans_len(desc, journal->j_len); | |
4170 | set_desc_mount_id(desc, journal->j_mount_id); | |
4171 | set_desc_trans_id(desc, journal->j_trans_id); | |
4172 | set_commit_trans_len(commit, journal->j_len); | |
4173 | ||
4174 | /* special check in case all buffers in the journal were marked for not logging */ | |
14a61442 | 4175 | BUG_ON(journal->j_len == 0); |
bd4c625c LT |
4176 | |
4177 | /* we're about to dirty all the log blocks, mark the description block | |
4178 | * dirty now too. Don't mark the commit block dirty until all the | |
4179 | * others are on disk | |
4180 | */ | |
4181 | mark_buffer_dirty(d_bh); | |
4182 | ||
4183 | /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */ | |
4184 | cur_write_start = journal->j_start; | |
4185 | cn = journal->j_first; | |
4186 | jindex = 1; /* start at one so we don't get the desc again */ | |
4187 | while (cn) { | |
4188 | clear_buffer_journal_new(cn->bh); | |
4189 | /* copy all the real blocks into log area. dirty log blocks */ | |
4190 | if (buffer_journaled(cn->bh)) { | |
4191 | struct buffer_head *tmp_bh; | |
4192 | char *addr; | |
4193 | struct page *page; | |
4194 | tmp_bh = | |
a9dd3643 JM |
4195 | journal_getblk(sb, |
4196 | SB_ONDISK_JOURNAL_1st_BLOCK(sb) + | |
bd4c625c LT |
4197 | ((cur_write_start + |
4198 | jindex) % | |
a9dd3643 | 4199 | SB_ONDISK_JOURNAL_SIZE(sb))); |
bd4c625c LT |
4200 | set_buffer_uptodate(tmp_bh); |
4201 | page = cn->bh->b_page; | |
4202 | addr = kmap(page); | |
4203 | memcpy(tmp_bh->b_data, | |
4204 | addr + offset_in_page(cn->bh->b_data), | |
4205 | cn->bh->b_size); | |
4206 | kunmap(page); | |
4207 | mark_buffer_dirty(tmp_bh); | |
4208 | jindex++; | |
4209 | set_buffer_journal_dirty(cn->bh); | |
4210 | clear_buffer_journaled(cn->bh); | |
4211 | } else { | |
4212 | /* JDirty cleared sometime during transaction. don't log this one */ | |
a9dd3643 | 4213 | reiserfs_warning(sb, "journal-2048", |
45b03d5e JM |
4214 | "BAD, buffer in journal hash, " |
4215 | "but not JDirty!"); | |
bd4c625c LT |
4216 | brelse(cn->bh); |
4217 | } | |
4218 | next = cn->next; | |
a9dd3643 | 4219 | free_cnode(sb, cn); |
bd4c625c | 4220 | cn = next; |
e6950a4d | 4221 | reiserfs_write_unlock(sb); |
bd4c625c | 4222 | cond_resched(); |
e6950a4d | 4223 | reiserfs_write_lock(sb); |
bd4c625c LT |
4224 | } |
4225 | ||
4226 | /* we are done with both the c_bh and d_bh, but | |
4227 | ** c_bh must be written after all other commit blocks, | |
4228 | ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1. | |
4229 | */ | |
4230 | ||
a9dd3643 | 4231 | journal->j_current_jl = alloc_journal_list(sb); |
bd4c625c LT |
4232 | |
4233 | /* now it is safe to insert this transaction on the main list */ | |
4234 | list_add_tail(&jl->j_list, &journal->j_journal_list); | |
4235 | list_add_tail(&jl->j_working_list, &journal->j_working_list); | |
4236 | journal->j_num_work_lists++; | |
4237 | ||
4238 | /* reset journal values for the next transaction */ | |
4239 | old_start = journal->j_start; | |
4240 | journal->j_start = | |
4241 | (journal->j_start + journal->j_len + | |
a9dd3643 | 4242 | 2) % SB_ONDISK_JOURNAL_SIZE(sb); |
bd4c625c LT |
4243 | atomic_set(&(journal->j_wcount), 0); |
4244 | journal->j_bcount = 0; | |
4245 | journal->j_last = NULL; | |
4246 | journal->j_first = NULL; | |
4247 | journal->j_len = 0; | |
4248 | journal->j_trans_start_time = 0; | |
a44c94a7 AZ |
4249 | /* check for trans_id overflow */ |
4250 | if (++journal->j_trans_id == 0) | |
4251 | journal->j_trans_id = 10; | |
bd4c625c LT |
4252 | journal->j_current_jl->j_trans_id = journal->j_trans_id; |
4253 | journal->j_must_wait = 0; | |
4254 | journal->j_len_alloc = 0; | |
4255 | journal->j_next_full_flush = 0; | |
4256 | journal->j_next_async_flush = 0; | |
a9dd3643 | 4257 | init_journal_hash(sb); |
bd4c625c LT |
4258 | |
4259 | // make sure reiserfs_add_jh sees the new current_jl before we | |
4260 | // write out the tails | |
4261 | smp_mb(); | |
4262 | ||
4263 | /* tail conversion targets have to hit the disk before we end the | |
4264 | * transaction. Otherwise a later transaction might repack the tail | |
4265 | * before this transaction commits, leaving the data block unflushed and | |
4266 | * clean, if we crash before the later transaction commits, the data block | |
4267 | * is lost. | |
4268 | */ | |
4269 | if (!list_empty(&jl->j_tail_bh_list)) { | |
8ebc4232 | 4270 | reiserfs_write_unlock(sb); |
bd4c625c LT |
4271 | write_ordered_buffers(&journal->j_dirty_buffers_lock, |
4272 | journal, jl, &jl->j_tail_bh_list); | |
8ebc4232 | 4273 | reiserfs_write_lock(sb); |
bd4c625c | 4274 | } |
14a61442 | 4275 | BUG_ON(!list_empty(&jl->j_tail_bh_list)); |
90415dea | 4276 | mutex_unlock(&jl->j_commit_mutex); |
bd4c625c LT |
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) { | |
a9dd3643 JM |
4286 | flush_commit_list(sb, jl, 1); |
4287 | flush_journal_list(sb, jl, 1); | |
bd4c625c LT |
4288 | } else if (!(jl->j_state & LIST_COMMIT_PENDING)) |
4289 | queue_delayed_work(commit_wq, &journal->j_work, HZ / 10); | |
4290 | ||
0222e657 JM |
4291 | /* if the next transaction has any chance of wrapping, flush |
4292 | ** transactions that might get overwritten. If any journal lists are very | |
4293 | ** old flush them as well. | |
bd4c625c LT |
4294 | */ |
4295 | first_jl: | |
4296 | list_for_each_safe(entry, safe, &journal->j_journal_list) { | |
4297 | temp_jl = JOURNAL_LIST_ENTRY(entry); | |
4298 | if (journal->j_start <= temp_jl->j_start) { | |
4299 | if ((journal->j_start + journal->j_trans_max + 1) >= | |
4300 | temp_jl->j_start) { | |
a9dd3643 | 4301 | flush_used_journal_lists(sb, temp_jl); |
bd4c625c LT |
4302 | goto first_jl; |
4303 | } else if ((journal->j_start + | |
4304 | journal->j_trans_max + 1) < | |
a9dd3643 | 4305 | SB_ONDISK_JOURNAL_SIZE(sb)) { |
bd4c625c LT |
4306 | /* if we don't cross into the next transaction and we don't |
4307 | * wrap, there is no way we can overlap any later transactions | |
4308 | * break now | |
4309 | */ | |
4310 | break; | |
4311 | } | |
4312 | } else if ((journal->j_start + | |
4313 | journal->j_trans_max + 1) > | |
a9dd3643 | 4314 | SB_ONDISK_JOURNAL_SIZE(sb)) { |
bd4c625c | 4315 | if (((journal->j_start + journal->j_trans_max + 1) % |
a9dd3643 | 4316 | SB_ONDISK_JOURNAL_SIZE(sb)) >= |
bd4c625c | 4317 | temp_jl->j_start) { |
a9dd3643 | 4318 | flush_used_journal_lists(sb, temp_jl); |
bd4c625c LT |
4319 | goto first_jl; |
4320 | } else { | |
4321 | /* we don't overlap anything from out start to the end of the | |
4322 | * log, and our wrapped portion doesn't overlap anything at | |
4323 | * the start of the log. We can break | |
4324 | */ | |
4325 | break; | |
4326 | } | |
4327 | } | |
4328 | } | |
a9dd3643 | 4329 | flush_old_journal_lists(sb); |
bd4c625c LT |
4330 | |
4331 | journal->j_current_jl->j_list_bitmap = | |
a9dd3643 | 4332 | get_list_bitmap(sb, journal->j_current_jl); |
bd4c625c LT |
4333 | |
4334 | if (!(journal->j_current_jl->j_list_bitmap)) { | |
a9dd3643 | 4335 | reiserfs_panic(sb, "journal-1996", |
c3a9c210 | 4336 | "could not get a list bitmap"); |
bd4c625c LT |
4337 | } |
4338 | ||
4339 | atomic_set(&(journal->j_jlock), 0); | |
a9dd3643 | 4340 | unlock_journal(sb); |
bd4c625c LT |
4341 | /* wake up any body waiting to join. */ |
4342 | clear_bit(J_WRITERS_QUEUED, &journal->j_state); | |
4343 | wake_up(&(journal->j_join_wait)); | |
4344 | ||
4345 | if (!flush && wait_on_commit && | |
a9dd3643 JM |
4346 | journal_list_still_alive(sb, commit_trans_id)) { |
4347 | flush_commit_list(sb, jl, 1); | |
bd4c625c LT |
4348 | } |
4349 | out: | |
a9dd3643 | 4350 | reiserfs_check_lock_depth(sb, "journal end2"); |
bd4c625c LT |
4351 | |
4352 | memset(th, 0, sizeof(*th)); | |
4353 | /* Re-set th->t_super, so we can properly keep track of how many | |
4354 | * persistent transactions there are. We need to do this so if this | |
4355 | * call is part of a failed restart_transaction, we can free it later */ | |
a9dd3643 | 4356 | th->t_super = sb; |
bd4c625c LT |
4357 | |
4358 | return journal->j_errno; | |
4359 | } | |
4360 | ||
32e8b106 JM |
4361 | /* Send the file system read only and refuse new transactions */ |
4362 | void reiserfs_abort_journal(struct super_block *sb, int errno) | |
bd4c625c LT |
4363 | { |
4364 | struct reiserfs_journal *journal = SB_JOURNAL(sb); | |
4365 | if (test_bit(J_ABORTED, &journal->j_state)) | |
4366 | return; | |
4367 | ||
32e8b106 JM |
4368 | if (!journal->j_errno) |
4369 | journal->j_errno = errno; | |
bd4c625c LT |
4370 | |
4371 | sb->s_flags |= MS_RDONLY; | |
4372 | set_bit(J_ABORTED, &journal->j_state); | |
1da177e4 LT |
4373 | |
4374 | #ifdef CONFIG_REISERFS_CHECK | |
bd4c625c | 4375 | dump_stack(); |
1da177e4 LT |
4376 | #endif |
4377 | } | |
4378 |