blob: c21df71943a635c7eedabbf36e0ae3ac6107c770 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2** Write ahead logging implementation copyright Chris Mason 2000
3**
4** The background commits make this code very interelated, and
5** overly complex. I need to rethink things a bit....The major players:
6**
7** journal_begin -- call with the number of blocks you expect to log.
8** If the current transaction is too
9** old, it will block until the current transaction is
10** finished, and then start a new one.
11** Usually, your transaction will get joined in with
12** previous ones for speed.
13**
14** journal_join -- same as journal_begin, but won't block on the current
15** transaction regardless of age. Don't ever call
16** this. Ever. There are only two places it should be
17** called from, and they are both inside this file.
18**
19** journal_mark_dirty -- adds blocks into this transaction. clears any flags
20** that might make them get sent to disk
21** and then marks them BH_JDirty. Puts the buffer head
22** into the current transaction hash.
23**
24** journal_end -- if the current transaction is batchable, it does nothing
25** otherwise, it could do an async/synchronous commit, or
26** a full flush of all log and real blocks in the
27** transaction.
28**
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
32** around too long.
33** -- Note, if you call this as an immediate flush from
34** from within kupdate, it will ignore the immediate flag
35*/
36
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include <linux/time.h>
Matthew Wilcox6188e102008-04-18 22:21:05 -040038#include <linux/semaphore.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include <linux/vmalloc.h>
40#include <linux/reiserfs_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#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>
Andrew Morton3fcfab12006-10-19 23:28:16 -070051#include <linux/backing-dev.h>
Jeff Mahoney90415de2008-07-25 01:46:40 -070052#include <linux/uaccess.h>
53
54#include <asm/system.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
Linus Torvalds1da177e2005-04-16 15:20:36 -070056/* 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*/
65static int reiserfs_mounted_fs_count;
66
67static struct workqueue_struct *commit_wq;
68
Linus Torvaldsbd4c6252005-07-12 20:21:28 -070069#define JOURNAL_TRANS_HALF 1018 /* must be correct to keep the desc and commit
70 structs at 4k */
71#define BUFNR 64 /*read ahead */
Linus Torvalds1da177e2005-04-16 15:20:36 -070072
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. */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -070076#define BLOCK_FREED_HOLDER 3 /* this block was freed during this transaction, and can't be written */
Linus Torvalds1da177e2005-04-16 15:20:36 -070077
78#define BLOCK_NEEDS_FLUSH 4 /* used in flush_journal_list */
79#define BLOCK_DIRTIED 5
80
Linus Torvalds1da177e2005-04-16 15:20:36 -070081/* journal list state bits */
82#define LIST_TOUCHED 1
83#define LIST_DIRTY 2
Linus Torvaldsbd4c6252005-07-12 20:21:28 -070084#define LIST_COMMIT_PENDING 4 /* someone will commit this list */
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
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 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -070089#define WAIT 4 /* wait for the log blocks to hit the disk */
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
Linus Torvaldsbd4c6252005-07-12 20:21:28 -070091static int do_journal_end(struct reiserfs_transaction_handle *,
92 struct super_block *, unsigned long nblocks,
93 int flags);
94static int flush_journal_list(struct super_block *s,
95 struct reiserfs_journal_list *jl, int flushall);
96static int flush_commit_list(struct super_block *s,
97 struct reiserfs_journal_list *jl, int flushall);
98static int can_dirty(struct reiserfs_journal_cnode *cn);
99static int journal_join(struct reiserfs_transaction_handle *th,
100 struct super_block *p_s_sb, unsigned long nblocks);
101static int release_journal_dev(struct super_block *super,
102 struct reiserfs_journal *journal);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103static int dirty_one_transaction(struct super_block *s,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700104 struct reiserfs_journal_list *jl);
David Howellsc4028952006-11-22 14:57:56 +0000105static void flush_async_commits(struct work_struct *work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106static void queue_log_writer(struct super_block *s);
107
108/* values for join in do_journal_begin_r */
109enum {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700110 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 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113};
114
115static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700116 struct super_block *p_s_sb,
117 unsigned long nblocks, int join);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700119static void init_journal_hash(struct super_block *p_s_sb)
120{
121 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
122 memset(journal->j_hash_table, 0,
123 JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124}
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*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700131static 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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138}
139
140static void disable_barrier(struct super_block *s)
141{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700142 REISERFS_SB(s)->s_mount_opt &= ~(1 << REISERFS_BARRIER_FLUSH);
143 printk("reiserfs: disabling flush barriers on %s\n",
144 reiserfs_bdevname(s));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145}
146
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700147static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block
148 *p_s_sb)
149{
150 struct reiserfs_bitmap_node *bn;
151 static int id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152
Pekka Enbergd739b422006-02-01 03:06:43 -0800153 bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700154 if (!bn) {
155 return NULL;
156 }
Pekka Enbergd739b422006-02-01 03:06:43 -0800157 bn->data = kzalloc(p_s_sb->s_blocksize, GFP_NOFS);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700158 if (!bn->data) {
Pekka Enbergd739b422006-02-01 03:06:43 -0800159 kfree(bn);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700160 return NULL;
161 }
162 bn->id = id++;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700163 INIT_LIST_HEAD(&bn->list);
164 return bn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165}
166
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700167static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *p_s_sb)
168{
169 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
170 struct reiserfs_bitmap_node *bn = NULL;
171 struct list_head *entry = journal->j_bitmap_nodes.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700173 journal->j_used_bitmap_nodes++;
174 repeat:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700176 if (entry != &journal->j_bitmap_nodes) {
177 bn = list_entry(entry, struct reiserfs_bitmap_node, list);
178 list_del(entry);
179 memset(bn->data, 0, p_s_sb->s_blocksize);
180 journal->j_free_bitmap_nodes--;
181 return bn;
182 }
183 bn = allocate_bitmap_node(p_s_sb);
184 if (!bn) {
185 yield();
186 goto repeat;
187 }
188 return bn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189}
190static inline void free_bitmap_node(struct super_block *p_s_sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700191 struct reiserfs_bitmap_node *bn)
192{
193 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
194 journal->j_used_bitmap_nodes--;
195 if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
Pekka Enbergd739b422006-02-01 03:06:43 -0800196 kfree(bn->data);
197 kfree(bn);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700198 } else {
199 list_add(&bn->list, &journal->j_bitmap_nodes);
200 journal->j_free_bitmap_nodes++;
201 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202}
203
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700204static void allocate_bitmap_nodes(struct super_block *p_s_sb)
205{
206 int i;
207 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
208 struct reiserfs_bitmap_node *bn = NULL;
209 for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
210 bn = allocate_bitmap_node(p_s_sb);
211 if (bn) {
212 list_add(&bn->list, &journal->j_bitmap_nodes);
213 journal->j_free_bitmap_nodes++;
214 } else {
215 break; // this is ok, we'll try again when more are needed
216 }
217 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218}
219
Jeff Mahoney3ee16672007-10-18 23:39:25 -0700220static int set_bit_in_list_bitmap(struct super_block *p_s_sb,
221 b_blocknr_t block,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700222 struct reiserfs_list_bitmap *jb)
223{
Jeff Mahoney3ee16672007-10-18 23:39:25 -0700224 unsigned int bmap_nr = block / (p_s_sb->s_blocksize << 3);
225 unsigned int bit_nr = block % (p_s_sb->s_blocksize << 3);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700227 if (!jb->bitmaps[bmap_nr]) {
228 jb->bitmaps[bmap_nr] = get_bitmap_node(p_s_sb);
229 }
230 set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
231 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232}
233
234static void cleanup_bitmap_list(struct super_block *p_s_sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700235 struct reiserfs_list_bitmap *jb)
236{
237 int i;
238 if (jb->bitmaps == NULL)
239 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240
Jeff Mahoneycb680c12007-10-18 23:39:27 -0700241 for (i = 0; i < reiserfs_bmap_count(p_s_sb); i++) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700242 if (jb->bitmaps[i]) {
243 free_bitmap_node(p_s_sb, jb->bitmaps[i]);
244 jb->bitmaps[i] = NULL;
245 }
246 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247}
248
249/*
250** only call this on FS unmount.
251*/
252static int free_list_bitmaps(struct super_block *p_s_sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700253 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;
260 cleanup_bitmap_list(p_s_sb, jb);
261 vfree(jb->bitmaps);
262 jb->bitmaps = NULL;
263 }
264 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265}
266
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700267static int free_bitmap_nodes(struct super_block *p_s_sb)
268{
269 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
270 struct list_head *next = journal->j_bitmap_nodes.next;
271 struct reiserfs_bitmap_node *bn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700273 while (next != &journal->j_bitmap_nodes) {
274 bn = list_entry(next, struct reiserfs_bitmap_node, list);
275 list_del(next);
Pekka Enbergd739b422006-02-01 03:06:43 -0800276 kfree(bn->data);
277 kfree(bn);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700278 next = journal->j_bitmap_nodes.next;
279 journal->j_free_bitmap_nodes--;
280 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700282 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283}
284
285/*
286** get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
287** jb_array is the array to be filled in.
288*/
289int reiserfs_allocate_list_bitmaps(struct super_block *p_s_sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700290 struct reiserfs_list_bitmap *jb_array,
Jeff Mahoney3ee16672007-10-18 23:39:25 -0700291 unsigned int bmap_nr)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700292{
293 int i;
294 int failed = 0;
295 struct reiserfs_list_bitmap *jb;
296 int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700298 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) {
303 reiserfs_warning(p_s_sb,
304 "clm-2000, unable to allocate bitmaps for journal lists");
305 failed = 1;
306 break;
307 }
308 memset(jb->bitmaps, 0, mem);
309 }
310 if (failed) {
311 free_list_bitmaps(p_s_sb, jb_array);
312 return -1;
313 }
314 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315}
316
317/*
318** find an available list bitmap. If you can't find one, flush a commit list
319** and try again
320*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700321static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *p_s_sb,
322 struct reiserfs_journal_list
323 *jl)
324{
325 int i, j;
326 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
327 struct reiserfs_list_bitmap *jb = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700329 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) {
334 flush_commit_list(p_s_sb,
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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349}
350
351/*
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*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700356static 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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376}
377
378/*
379** pulls a cnode off the free list, or returns NULL on failure
380*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700381static struct reiserfs_journal_cnode *get_cnode(struct super_block *p_s_sb)
382{
383 struct reiserfs_journal_cnode *cn;
384 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700386 reiserfs_check_lock_depth(p_s_sb, "get_cnode");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700388 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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403}
404
405/*
406** returns a cnode to the free list
407*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700408static void free_cnode(struct super_block *p_s_sb,
409 struct reiserfs_journal_cnode *cn)
410{
411 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700413 reiserfs_check_lock_depth(p_s_sb, "free_cnode");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700415 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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424}
425
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700426static void clear_prepared_bits(struct buffer_head *bh)
427{
428 clear_buffer_journal_prepared(bh);
429 clear_buffer_journal_restore_dirty(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430}
431
432/* utility function to force a BUG if it is called without the big
433** kernel lock held. caller is the string printed just before calling BUG()
434*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700435void reiserfs_check_lock_depth(struct super_block *sb, char *caller)
436{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437#ifdef CONFIG_SMP
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700438 if (current->lock_depth < 0) {
439 reiserfs_panic(sb, "%s called without kernel lock held",
440 caller);
441 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442#else
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700443 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444#endif
445}
446
447/* return a cnode with same dev, block number and size in table, or null if not found */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700448static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
449 super_block
450 *sb,
451 struct
452 reiserfs_journal_cnode
453 **table,
454 long bl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700456 struct reiserfs_journal_cnode *cn;
457 cn = journal_hash(table, sb, bl);
458 while (cn) {
459 if (cn->blocknr == bl && cn->sb == sb)
460 return cn;
461 cn = cn->hnext;
462 }
463 return (struct reiserfs_journal_cnode *)0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464}
465
466/*
467** this actually means 'can this block be reallocated yet?'. If you set search_all, a block can only be allocated
468** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever
469** being overwritten by a replay after crashing.
470**
471** If you don't set search_all, a block can only be allocated if it is not in the current transaction. Since deleting
472** a block removes it from the current transaction, this case should never happen. If you don't set search_all, make
473** sure you never write the block without logging it.
474**
475** next_zero_bit is a suggestion about the next block to try for find_forward.
476** when bl is rejected because it is set in a journal list bitmap, we search
477** for the next zero bit in the bitmap that rejected bl. Then, we return that
478** through next_zero_bit for find_forward to try.
479**
480** Just because we return something in next_zero_bit does not mean we won't
481** reject it on the next call to reiserfs_in_journal
482**
483*/
484int reiserfs_in_journal(struct super_block *p_s_sb,
Jeff Mahoney3ee16672007-10-18 23:39:25 -0700485 unsigned int bmap_nr, int bit_nr, int search_all,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700486 b_blocknr_t * next_zero_bit)
487{
488 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
489 struct reiserfs_journal_cnode *cn;
490 struct reiserfs_list_bitmap *jb;
491 int i;
492 unsigned long bl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700494 *next_zero_bit = 0; /* always start this at zero. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700496 PROC_INFO_INC(p_s_sb, journal.in_journal);
497 /* If we aren't doing a search_all, this is a metablock, and it will be logged before use.
498 ** if we crash before the transaction that freed it commits, this transaction won't
499 ** have committed either, and the block will never be written
500 */
501 if (search_all) {
502 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
503 PROC_INFO_INC(p_s_sb, journal.in_journal_bitmap);
504 jb = journal->j_list_bitmap + i;
505 if (jb->journal_list && jb->bitmaps[bmap_nr] &&
506 test_bit(bit_nr,
507 (unsigned long *)jb->bitmaps[bmap_nr]->
508 data)) {
509 *next_zero_bit =
510 find_next_zero_bit((unsigned long *)
511 (jb->bitmaps[bmap_nr]->
512 data),
513 p_s_sb->s_blocksize << 3,
514 bit_nr + 1);
515 return 1;
516 }
517 }
518 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700520 bl = bmap_nr * (p_s_sb->s_blocksize << 3) + bit_nr;
521 /* is it in any old transactions? */
522 if (search_all
523 && (cn =
524 get_journal_hash_dev(p_s_sb, journal->j_list_hash_table, bl))) {
525 return 1;
526 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700528 /* is it in the current transaction. This should never happen */
529 if ((cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, bl))) {
530 BUG();
531 return 1;
532 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700534 PROC_INFO_INC(p_s_sb, journal.in_journal_reusable);
535 /* safe for reuse */
536 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537}
538
539/* insert cn into table
540*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700541static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
542 struct reiserfs_journal_cnode *cn)
543{
544 struct reiserfs_journal_cnode *cn_orig;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700546 cn_orig = journal_hash(table, cn->sb, cn->blocknr);
547 cn->hnext = cn_orig;
548 cn->hprev = NULL;
549 if (cn_orig) {
550 cn_orig->hprev = cn;
551 }
552 journal_hash(table, cn->sb, cn->blocknr) = cn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553}
554
555/* lock the current transaction */
Jesper Juhl77933d72005-07-27 11:46:09 -0700556static inline void lock_journal(struct super_block *p_s_sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700557{
558 PROC_INFO_INC(p_s_sb, journal.lock_journal);
Jeff Mahoneyf68215c2008-07-25 01:46:38 -0700559 mutex_lock(&SB_JOURNAL(p_s_sb)->j_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560}
561
562/* unlock the current transaction */
Jesper Juhl77933d72005-07-27 11:46:09 -0700563static inline void unlock_journal(struct super_block *p_s_sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700564{
Jeff Mahoneyf68215c2008-07-25 01:46:38 -0700565 mutex_unlock(&SB_JOURNAL(p_s_sb)->j_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566}
567
568static inline void get_journal_list(struct reiserfs_journal_list *jl)
569{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700570 jl->j_refcount++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571}
572
573static inline void put_journal_list(struct super_block *s,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700574 struct reiserfs_journal_list *jl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700576 if (jl->j_refcount < 1) {
577 reiserfs_panic(s, "trans id %lu, refcount at %d",
578 jl->j_trans_id, jl->j_refcount);
579 }
580 if (--jl->j_refcount == 0)
Pekka Enbergd739b422006-02-01 03:06:43 -0800581 kfree(jl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582}
583
584/*
585** this used to be much more involved, and I'm keeping it just in case things get ugly again.
586** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a
587** transaction.
588*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700589static void cleanup_freed_for_journal_list(struct super_block *p_s_sb,
590 struct reiserfs_journal_list *jl)
591{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700593 struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
594 if (jb) {
595 cleanup_bitmap_list(p_s_sb, jb);
596 }
597 jl->j_list_bitmap->journal_list = NULL;
598 jl->j_list_bitmap = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599}
600
601static int journal_list_still_alive(struct super_block *s,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700602 unsigned long trans_id)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700604 struct reiserfs_journal *journal = SB_JOURNAL(s);
605 struct list_head *entry = &journal->j_journal_list;
606 struct reiserfs_journal_list *jl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700608 if (!list_empty(entry)) {
609 jl = JOURNAL_LIST_ENTRY(entry->next);
610 if (jl->j_trans_id <= trans_id) {
611 return 1;
612 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700614 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615}
616
Chris Mason398c95b2007-10-16 23:29:44 -0700617/*
618 * If page->mapping was null, we failed to truncate this page for
619 * some reason. Most likely because it was truncated after being
620 * logged via data=journal.
621 *
622 * This does a check to see if the buffer belongs to one of these
623 * lost pages before doing the final put_bh. If page->mapping was
624 * null, it tries to free buffers on the page, which should make the
625 * final page_cache_release drop the page from the lru.
626 */
627static void release_buffer_page(struct buffer_head *bh)
628{
629 struct page *page = bh->b_page;
Nick Piggin529ae9a2008-08-02 12:01:03 +0200630 if (!page->mapping && trylock_page(page)) {
Chris Mason398c95b2007-10-16 23:29:44 -0700631 page_cache_get(page);
632 put_bh(bh);
633 if (!page->mapping)
634 try_to_free_buffers(page);
635 unlock_page(page);
636 page_cache_release(page);
637 } else {
638 put_bh(bh);
639 }
640}
641
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700642static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
643{
644 char b[BDEVNAME_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700646 if (buffer_journaled(bh)) {
647 reiserfs_warning(NULL,
648 "clm-2084: pinned buffer %lu:%s sent to disk",
649 bh->b_blocknr, bdevname(bh->b_bdev, b));
650 }
651 if (uptodate)
652 set_buffer_uptodate(bh);
653 else
654 clear_buffer_uptodate(bh);
Chris Mason398c95b2007-10-16 23:29:44 -0700655
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700656 unlock_buffer(bh);
Chris Mason398c95b2007-10-16 23:29:44 -0700657 release_buffer_page(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658}
659
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700660static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
661{
662 if (uptodate)
663 set_buffer_uptodate(bh);
664 else
665 clear_buffer_uptodate(bh);
666 unlock_buffer(bh);
667 put_bh(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668}
669
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700670static void submit_logged_buffer(struct buffer_head *bh)
671{
672 get_bh(bh);
673 bh->b_end_io = reiserfs_end_buffer_io_sync;
674 clear_buffer_journal_new(bh);
675 clear_buffer_dirty(bh);
676 if (!test_clear_buffer_journal_test(bh))
677 BUG();
678 if (!buffer_uptodate(bh))
679 BUG();
680 submit_bh(WRITE, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681}
682
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700683static void submit_ordered_buffer(struct buffer_head *bh)
684{
685 get_bh(bh);
686 bh->b_end_io = reiserfs_end_ordered_io;
687 clear_buffer_dirty(bh);
688 if (!buffer_uptodate(bh))
689 BUG();
690 submit_bh(WRITE, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691}
692
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700693static int submit_barrier_buffer(struct buffer_head *bh)
694{
695 get_bh(bh);
696 bh->b_end_io = reiserfs_end_ordered_io;
697 clear_buffer_dirty(bh);
698 if (!buffer_uptodate(bh))
699 BUG();
700 return submit_bh(WRITE_BARRIER, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701}
702
703static void check_barrier_completion(struct super_block *s,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700704 struct buffer_head *bh)
705{
706 if (buffer_eopnotsupp(bh)) {
707 clear_buffer_eopnotsupp(bh);
708 disable_barrier(s);
709 set_buffer_uptodate(bh);
710 set_buffer_dirty(bh);
711 sync_dirty_buffer(bh);
712 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713}
714
715#define CHUNK_SIZE 32
716struct buffer_chunk {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700717 struct buffer_head *bh[CHUNK_SIZE];
718 int nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719};
720
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700721static void write_chunk(struct buffer_chunk *chunk)
722{
723 int i;
724 get_fs_excl();
725 for (i = 0; i < chunk->nr; i++) {
726 submit_logged_buffer(chunk->bh[i]);
727 }
728 chunk->nr = 0;
729 put_fs_excl();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730}
731
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700732static void write_ordered_chunk(struct buffer_chunk *chunk)
733{
734 int i;
735 get_fs_excl();
736 for (i = 0; i < chunk->nr; i++) {
737 submit_ordered_buffer(chunk->bh[i]);
738 }
739 chunk->nr = 0;
740 put_fs_excl();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741}
742
743static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700744 spinlock_t * lock, void (fn) (struct buffer_chunk *))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700746 int ret = 0;
Eric Sesterhenn14a61442006-10-03 23:36:38 +0200747 BUG_ON(chunk->nr >= CHUNK_SIZE);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700748 chunk->bh[chunk->nr++] = bh;
749 if (chunk->nr >= CHUNK_SIZE) {
750 ret = 1;
751 if (lock)
752 spin_unlock(lock);
753 fn(chunk);
754 if (lock)
755 spin_lock(lock);
756 }
757 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758}
759
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700761static struct reiserfs_jh *alloc_jh(void)
762{
763 struct reiserfs_jh *jh;
764 while (1) {
765 jh = kmalloc(sizeof(*jh), GFP_NOFS);
766 if (jh) {
767 atomic_inc(&nr_reiserfs_jh);
768 return jh;
769 }
770 yield();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772}
773
774/*
775 * we want to free the jh when the buffer has been written
776 * and waited on
777 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700778void reiserfs_free_jh(struct buffer_head *bh)
779{
780 struct reiserfs_jh *jh;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700782 jh = bh->b_private;
783 if (jh) {
784 bh->b_private = NULL;
785 jh->bh = NULL;
786 list_del_init(&jh->list);
787 kfree(jh);
788 if (atomic_read(&nr_reiserfs_jh) <= 0)
789 BUG();
790 atomic_dec(&nr_reiserfs_jh);
791 put_bh(bh);
792 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793}
794
795static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700796 int tail)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700798 struct reiserfs_jh *jh;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700800 if (bh->b_private) {
801 spin_lock(&j->j_dirty_buffers_lock);
802 if (!bh->b_private) {
803 spin_unlock(&j->j_dirty_buffers_lock);
804 goto no_jh;
805 }
806 jh = bh->b_private;
807 list_del_init(&jh->list);
808 } else {
809 no_jh:
810 get_bh(bh);
811 jh = alloc_jh();
812 spin_lock(&j->j_dirty_buffers_lock);
813 /* buffer must be locked for __add_jh, should be able to have
814 * two adds at the same time
815 */
Eric Sesterhenn14a61442006-10-03 23:36:38 +0200816 BUG_ON(bh->b_private);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700817 jh->bh = bh;
818 bh->b_private = jh;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700820 jh->jl = j->j_current_jl;
821 if (tail)
822 list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
823 else {
824 list_add_tail(&jh->list, &jh->jl->j_bh_list);
825 }
826 spin_unlock(&j->j_dirty_buffers_lock);
827 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828}
829
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700830int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
831{
832 return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833}
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700834int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
835{
836 return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837}
838
839#define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700840static int write_ordered_buffers(spinlock_t * lock,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 struct reiserfs_journal *j,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700842 struct reiserfs_journal_list *jl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 struct list_head *list)
844{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700845 struct buffer_head *bh;
846 struct reiserfs_jh *jh;
847 int ret = j->j_errno;
848 struct buffer_chunk chunk;
849 struct list_head tmp;
850 INIT_LIST_HEAD(&tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700852 chunk.nr = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 spin_lock(lock);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700854 while (!list_empty(list)) {
855 jh = JH_ENTRY(list->next);
856 bh = jh->bh;
857 get_bh(bh);
Nick Pigginca5de402008-08-02 12:02:13 +0200858 if (!trylock_buffer(bh)) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700859 if (!buffer_dirty(bh)) {
Akinobu Mitaf1166292006-06-26 00:24:46 -0700860 list_move(&jh->list, &tmp);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700861 goto loop_next;
862 }
863 spin_unlock(lock);
864 if (chunk.nr)
865 write_ordered_chunk(&chunk);
866 wait_on_buffer(bh);
867 cond_resched();
868 spin_lock(lock);
869 goto loop_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 }
Chris Mason3d4492f2006-02-01 03:06:49 -0800871 /* in theory, dirty non-uptodate buffers should never get here,
872 * but the upper layer io error paths still have a few quirks.
873 * Handle them here as gracefully as we can
874 */
875 if (!buffer_uptodate(bh) && buffer_dirty(bh)) {
876 clear_buffer_dirty(bh);
877 ret = -EIO;
878 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700879 if (buffer_dirty(bh)) {
Akinobu Mitaf1166292006-06-26 00:24:46 -0700880 list_move(&jh->list, &tmp);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700881 add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
882 } else {
883 reiserfs_free_jh(bh);
884 unlock_buffer(bh);
885 }
886 loop_next:
887 put_bh(bh);
888 cond_resched_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700890 if (chunk.nr) {
891 spin_unlock(lock);
892 write_ordered_chunk(&chunk);
893 spin_lock(lock);
894 }
895 while (!list_empty(&tmp)) {
896 jh = JH_ENTRY(tmp.prev);
897 bh = jh->bh;
898 get_bh(bh);
899 reiserfs_free_jh(bh);
900
901 if (buffer_locked(bh)) {
902 spin_unlock(lock);
903 wait_on_buffer(bh);
904 spin_lock(lock);
905 }
906 if (!buffer_uptodate(bh)) {
907 ret = -EIO;
908 }
Chris Masond62b1b82006-02-01 03:06:47 -0800909 /* ugly interaction with invalidatepage here.
910 * reiserfs_invalidate_page will pin any buffer that has a valid
911 * journal head from an older transaction. If someone else sets
912 * our buffer dirty after we write it in the first loop, and
913 * then someone truncates the page away, nobody will ever write
914 * the buffer. We're safe if we write the page one last time
915 * after freeing the journal header.
916 */
917 if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
918 spin_unlock(lock);
919 ll_rw_block(WRITE, 1, &bh);
920 spin_lock(lock);
921 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700922 put_bh(bh);
923 cond_resched_lock(lock);
924 }
925 spin_unlock(lock);
926 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927}
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700928
929static int flush_older_commits(struct super_block *s,
930 struct reiserfs_journal_list *jl)
931{
932 struct reiserfs_journal *journal = SB_JOURNAL(s);
933 struct reiserfs_journal_list *other_jl;
934 struct reiserfs_journal_list *first_jl;
935 struct list_head *entry;
936 unsigned long trans_id = jl->j_trans_id;
937 unsigned long other_trans_id;
938 unsigned long first_trans_id;
939
940 find_first:
941 /*
942 * first we walk backwards to find the oldest uncommitted transation
943 */
944 first_jl = jl;
945 entry = jl->j_list.prev;
946 while (1) {
947 other_jl = JOURNAL_LIST_ENTRY(entry);
948 if (entry == &journal->j_journal_list ||
949 atomic_read(&other_jl->j_older_commits_done))
950 break;
951
952 first_jl = other_jl;
953 entry = other_jl->j_list.prev;
954 }
955
956 /* if we didn't find any older uncommitted transactions, return now */
957 if (first_jl == jl) {
958 return 0;
959 }
960
961 first_trans_id = first_jl->j_trans_id;
962
963 entry = &first_jl->j_list;
964 while (1) {
965 other_jl = JOURNAL_LIST_ENTRY(entry);
966 other_trans_id = other_jl->j_trans_id;
967
968 if (other_trans_id < trans_id) {
969 if (atomic_read(&other_jl->j_commit_left) != 0) {
970 flush_commit_list(s, other_jl, 0);
971
972 /* list we were called with is gone, return */
973 if (!journal_list_still_alive(s, trans_id))
974 return 1;
975
976 /* the one we just flushed is gone, this means all
977 * older lists are also gone, so first_jl is no longer
978 * valid either. Go back to the beginning.
979 */
980 if (!journal_list_still_alive
981 (s, other_trans_id)) {
982 goto find_first;
983 }
984 }
985 entry = entry->next;
986 if (entry == &journal->j_journal_list)
987 return 0;
988 } else {
989 return 0;
990 }
991 }
992 return 0;
993}
Adrian Bunkdeba0f42007-10-16 23:26:03 -0700994
995static int reiserfs_async_progress_wait(struct super_block *s)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700996{
997 DEFINE_WAIT(wait);
998 struct reiserfs_journal *j = SB_JOURNAL(s);
999 if (atomic_read(&j->j_async_throttle))
Andrew Morton3fcfab12006-10-19 23:28:16 -07001000 congestion_wait(WRITE, HZ / 10);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001001 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002}
1003
1004/*
1005** if this journal list still has commit blocks unflushed, send them to disk.
1006**
1007** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
1008** Before the commit block can by written, every other log block must be safely on disk
1009**
1010*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001011static int flush_commit_list(struct super_block *s,
1012 struct reiserfs_journal_list *jl, int flushall)
1013{
1014 int i;
Jeff Mahoney3ee16672007-10-18 23:39:25 -07001015 b_blocknr_t bn;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001016 struct buffer_head *tbh = NULL;
1017 unsigned long trans_id = jl->j_trans_id;
1018 struct reiserfs_journal *journal = SB_JOURNAL(s);
1019 int barrier = 0;
1020 int retval = 0;
Chris Masone0e851c2006-02-01 03:06:49 -08001021 int write_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001023 reiserfs_check_lock_depth(s, "flush_commit_list");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001025 if (atomic_read(&jl->j_older_commits_done)) {
1026 return 0;
1027 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001029 get_fs_excl();
Jens Axboe22e2c502005-06-27 10:55:12 +02001030
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001031 /* before we can put our commit blocks on disk, we have to make sure everyone older than
1032 ** us is on disk too
1033 */
1034 BUG_ON(jl->j_len <= 0);
1035 BUG_ON(trans_id == journal->j_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001037 get_journal_list(jl);
1038 if (flushall) {
1039 if (flush_older_commits(s, jl) == 1) {
1040 /* list disappeared during flush_older_commits. return */
1041 goto put_jl;
1042 }
1043 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001045 /* make sure nobody is trying to flush this one at the same time */
Jeff Mahoney90415de2008-07-25 01:46:40 -07001046 mutex_lock(&jl->j_commit_mutex);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001047 if (!journal_list_still_alive(s, trans_id)) {
Jeff Mahoney90415de2008-07-25 01:46:40 -07001048 mutex_unlock(&jl->j_commit_mutex);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001049 goto put_jl;
1050 }
1051 BUG_ON(jl->j_trans_id == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001053 /* this commit is done, exit */
1054 if (atomic_read(&(jl->j_commit_left)) <= 0) {
1055 if (flushall) {
1056 atomic_set(&(jl->j_older_commits_done), 1);
1057 }
Jeff Mahoney90415de2008-07-25 01:46:40 -07001058 mutex_unlock(&jl->j_commit_mutex);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001059 goto put_jl;
1060 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001062 if (!list_empty(&jl->j_bh_list)) {
Chris Mason3d4492f2006-02-01 03:06:49 -08001063 int ret;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001064 unlock_kernel();
Chris Mason3d4492f2006-02-01 03:06:49 -08001065 ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
1066 journal, jl, &jl->j_bh_list);
1067 if (ret < 0 && retval == 0)
1068 retval = ret;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001069 lock_kernel();
1070 }
1071 BUG_ON(!list_empty(&jl->j_bh_list));
1072 /*
1073 * for the description block and all the log blocks, submit any buffers
Chris Masone0e851c2006-02-01 03:06:49 -08001074 * that haven't already reached the disk. Try to write at least 256
1075 * log blocks. later on, we will only wait on blocks that correspond
1076 * to this transaction, but while we're unplugging we might as well
1077 * get a chunk of data on there.
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001078 */
1079 atomic_inc(&journal->j_async_throttle);
Chris Masone0e851c2006-02-01 03:06:49 -08001080 write_len = jl->j_len + 1;
1081 if (write_len < 256)
1082 write_len = 256;
1083 for (i = 0 ; i < write_len ; i++) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001084 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1085 SB_ONDISK_JOURNAL_SIZE(s);
1086 tbh = journal_find_get_block(s, bn);
Chris Masone0e851c2006-02-01 03:06:49 -08001087 if (tbh) {
1088 if (buffer_dirty(tbh))
1089 ll_rw_block(WRITE, 1, &tbh) ;
1090 put_bh(tbh) ;
1091 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001092 }
1093 atomic_dec(&journal->j_async_throttle);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094
Jeff Mahoney5d5e81562005-12-14 14:38:36 -05001095 /* We're skipping the commit if there's an error */
1096 if (retval || reiserfs_is_journal_aborted(journal))
1097 barrier = 0;
1098
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001099 /* wait on everything written so far before writing the commit
1100 * if we are in barrier mode, send the commit down now
1101 */
1102 barrier = reiserfs_barrier_flush(s);
1103 if (barrier) {
1104 int ret;
1105 lock_buffer(jl->j_commit_bh);
1106 ret = submit_barrier_buffer(jl->j_commit_bh);
1107 if (ret == -EOPNOTSUPP) {
1108 set_buffer_uptodate(jl->j_commit_bh);
1109 disable_barrier(s);
1110 barrier = 0;
1111 }
1112 }
1113 for (i = 0; i < (jl->j_len + 1); i++) {
1114 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1115 (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1116 tbh = journal_find_get_block(s, bn);
1117 wait_on_buffer(tbh);
1118 // since we're using ll_rw_blk above, it might have skipped over
1119 // a locked buffer. Double check here
1120 //
1121 if (buffer_dirty(tbh)) /* redundant, sync_dirty_buffer() checks */
1122 sync_dirty_buffer(tbh);
1123 if (unlikely(!buffer_uptodate(tbh))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124#ifdef CONFIG_REISERFS_CHECK
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001125 reiserfs_warning(s, "journal-601, buffer write failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126#endif
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001127 retval = -EIO;
1128 }
1129 put_bh(tbh); /* once for journal_find_get_block */
1130 put_bh(tbh); /* once due to original getblk in do_journal_end */
1131 atomic_dec(&(jl->j_commit_left));
1132 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001134 BUG_ON(atomic_read(&(jl->j_commit_left)) != 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001136 if (!barrier) {
Jeff Mahoney5d5e81562005-12-14 14:38:36 -05001137 /* If there was a write error in the journal - we can't commit
1138 * this transaction - it will be invalid and, if successful,
Robert P. J. Daybeb7dd82007-05-09 07:14:03 +02001139 * will just end up propagating the write error out to
Jeff Mahoney5d5e81562005-12-14 14:38:36 -05001140 * the file system. */
1141 if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1142 if (buffer_dirty(jl->j_commit_bh))
1143 BUG();
1144 mark_buffer_dirty(jl->j_commit_bh) ;
1145 sync_dirty_buffer(jl->j_commit_bh) ;
1146 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001147 } else
1148 wait_on_buffer(jl->j_commit_bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001150 check_barrier_completion(s, jl->j_commit_bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001152 /* If there was a write error in the journal - we can't commit this
1153 * transaction - it will be invalid and, if successful, will just end
Robert P. J. Daybeb7dd82007-05-09 07:14:03 +02001154 * up propagating the write error out to the filesystem. */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001155 if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156#ifdef CONFIG_REISERFS_CHECK
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001157 reiserfs_warning(s, "journal-615: buffer write failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158#endif
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001159 retval = -EIO;
1160 }
1161 bforget(jl->j_commit_bh);
1162 if (journal->j_last_commit_id != 0 &&
1163 (jl->j_trans_id - journal->j_last_commit_id) != 1) {
1164 reiserfs_warning(s, "clm-2200: last commit %lu, current %lu",
1165 journal->j_last_commit_id, jl->j_trans_id);
1166 }
1167 journal->j_last_commit_id = jl->j_trans_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001169 /* now, every commit block is on the disk. It is safe to allow blocks freed during this transaction to be reallocated */
1170 cleanup_freed_for_journal_list(s, jl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001172 retval = retval ? retval : journal->j_errno;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001174 /* mark the metadata dirty */
1175 if (!retval)
1176 dirty_one_transaction(s, jl);
1177 atomic_dec(&(jl->j_commit_left));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001179 if (flushall) {
1180 atomic_set(&(jl->j_older_commits_done), 1);
1181 }
Jeff Mahoney90415de2008-07-25 01:46:40 -07001182 mutex_unlock(&jl->j_commit_mutex);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001183 put_jl:
1184 put_journal_list(s, jl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001186 if (retval)
1187 reiserfs_abort(s, retval, "Journal write error in %s",
Harvey Harrisonfbe54982008-04-28 02:16:22 -07001188 __func__);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001189 put_fs_excl();
1190 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191}
1192
1193/*
1194** flush_journal_list frequently needs to find a newer transaction for a given block. This does that, or
1195** returns NULL if it can't find anything
1196*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001197static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1198 reiserfs_journal_cnode
1199 *cn)
1200{
1201 struct super_block *sb = cn->sb;
1202 b_blocknr_t blocknr = cn->blocknr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001204 cn = cn->hprev;
1205 while (cn) {
1206 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1207 return cn->jlist;
1208 }
1209 cn = cn->hprev;
1210 }
1211 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212}
1213
Chris Masona3172022006-09-29 01:59:56 -07001214static int newer_jl_done(struct reiserfs_journal_cnode *cn)
1215{
1216 struct super_block *sb = cn->sb;
1217 b_blocknr_t blocknr = cn->blocknr;
1218
1219 cn = cn->hprev;
1220 while (cn) {
1221 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist &&
1222 atomic_read(&cn->jlist->j_commit_left) != 0)
1223 return 0;
1224 cn = cn->hprev;
1225 }
1226 return 1;
1227}
1228
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001229static void remove_journal_hash(struct super_block *,
1230 struct reiserfs_journal_cnode **,
1231 struct reiserfs_journal_list *, unsigned long,
1232 int);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233
1234/*
1235** once all the real blocks have been flushed, it is safe to remove them from the
1236** journal list for this transaction. Aside from freeing the cnode, this also allows the
1237** block to be reallocated for data blocks if it had been deleted.
1238*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001239static void remove_all_from_journal_list(struct super_block *p_s_sb,
1240 struct reiserfs_journal_list *jl,
1241 int debug)
1242{
1243 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1244 struct reiserfs_journal_cnode *cn, *last;
1245 cn = jl->j_realblock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001247 /* which is better, to lock once around the whole loop, or
1248 ** to lock for each call to remove_journal_hash?
1249 */
1250 while (cn) {
1251 if (cn->blocknr != 0) {
1252 if (debug) {
1253 reiserfs_warning(p_s_sb,
1254 "block %u, bh is %d, state %ld",
1255 cn->blocknr, cn->bh ? 1 : 0,
1256 cn->state);
1257 }
1258 cn->state = 0;
1259 remove_journal_hash(p_s_sb, journal->j_list_hash_table,
1260 jl, cn->blocknr, 1);
1261 }
1262 last = cn;
1263 cn = cn->next;
1264 free_cnode(p_s_sb, last);
1265 }
1266 jl->j_realblock = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267}
1268
1269/*
1270** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block.
1271** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start
1272** releasing blocks in this transaction for reuse as data blocks.
1273** called by flush_journal_list, before it calls remove_all_from_journal_list
1274**
1275*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001276static int _update_journal_header_block(struct super_block *p_s_sb,
1277 unsigned long offset,
1278 unsigned long trans_id)
1279{
1280 struct reiserfs_journal_header *jh;
1281 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001283 if (reiserfs_is_journal_aborted(journal))
1284 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001286 if (trans_id >= journal->j_last_flush_trans_id) {
1287 if (buffer_locked((journal->j_header_bh))) {
1288 wait_on_buffer((journal->j_header_bh));
1289 if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290#ifdef CONFIG_REISERFS_CHECK
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001291 reiserfs_warning(p_s_sb,
1292 "journal-699: buffer write failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293#endif
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001294 return -EIO;
1295 }
1296 }
1297 journal->j_last_flush_trans_id = trans_id;
1298 journal->j_first_unflushed_offset = offset;
1299 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1300 b_data);
1301 jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1302 jh->j_first_unflushed_offset = cpu_to_le32(offset);
1303 jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001305 if (reiserfs_barrier_flush(p_s_sb)) {
1306 int ret;
1307 lock_buffer(journal->j_header_bh);
1308 ret = submit_barrier_buffer(journal->j_header_bh);
1309 if (ret == -EOPNOTSUPP) {
1310 set_buffer_uptodate(journal->j_header_bh);
1311 disable_barrier(p_s_sb);
1312 goto sync;
1313 }
1314 wait_on_buffer(journal->j_header_bh);
1315 check_barrier_completion(p_s_sb, journal->j_header_bh);
1316 } else {
1317 sync:
1318 set_buffer_dirty(journal->j_header_bh);
1319 sync_dirty_buffer(journal->j_header_bh);
1320 }
1321 if (!buffer_uptodate(journal->j_header_bh)) {
1322 reiserfs_warning(p_s_sb,
1323 "journal-837: IO error during journal replay");
1324 return -EIO;
1325 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001327 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328}
1329
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001330static int update_journal_header_block(struct super_block *p_s_sb,
1331 unsigned long offset,
1332 unsigned long trans_id)
1333{
1334 return _update_journal_header_block(p_s_sb, offset, trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335}
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001336
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337/*
1338** flush any and all journal lists older than you are
1339** can only be called from flush_journal_list
1340*/
1341static int flush_older_journal_lists(struct super_block *p_s_sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001342 struct reiserfs_journal_list *jl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001344 struct list_head *entry;
1345 struct reiserfs_journal_list *other_jl;
1346 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1347 unsigned long trans_id = jl->j_trans_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001349 /* we know we are the only ones flushing things, no extra race
1350 * protection is required.
1351 */
1352 restart:
1353 entry = journal->j_journal_list.next;
1354 /* Did we wrap? */
1355 if (entry == &journal->j_journal_list)
1356 return 0;
1357 other_jl = JOURNAL_LIST_ENTRY(entry);
1358 if (other_jl->j_trans_id < trans_id) {
1359 BUG_ON(other_jl->j_refcount <= 0);
1360 /* do not flush all */
1361 flush_journal_list(p_s_sb, other_jl, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001363 /* other_jl is now deleted from the list */
1364 goto restart;
1365 }
1366 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367}
1368
1369static void del_from_work_list(struct super_block *s,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001370 struct reiserfs_journal_list *jl)
1371{
1372 struct reiserfs_journal *journal = SB_JOURNAL(s);
1373 if (!list_empty(&jl->j_working_list)) {
1374 list_del_init(&jl->j_working_list);
1375 journal->j_num_work_lists--;
1376 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377}
1378
1379/* flush a journal list, both commit and real blocks
1380**
1381** always set flushall to 1, unless you are calling from inside
1382** flush_journal_list
1383**
1384** IMPORTANT. This can only be called while there are no journal writers,
1385** and the journal is locked. That means it can only be called from
1386** do_journal_end, or by journal_release
1387*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001388static int flush_journal_list(struct super_block *s,
1389 struct reiserfs_journal_list *jl, int flushall)
1390{
1391 struct reiserfs_journal_list *pjl;
1392 struct reiserfs_journal_cnode *cn, *last;
1393 int count;
1394 int was_jwait = 0;
1395 int was_dirty = 0;
1396 struct buffer_head *saved_bh;
1397 unsigned long j_len_saved = jl->j_len;
1398 struct reiserfs_journal *journal = SB_JOURNAL(s);
1399 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001401 BUG_ON(j_len_saved <= 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001403 if (atomic_read(&journal->j_wcount) != 0) {
1404 reiserfs_warning(s,
1405 "clm-2048: flush_journal_list called with wcount %d",
1406 atomic_read(&journal->j_wcount));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001408 BUG_ON(jl->j_trans_id == 0);
1409
1410 /* if flushall == 0, the lock is already held */
1411 if (flushall) {
Jeff Mahoneyafe70252008-07-25 01:46:39 -07001412 mutex_lock(&journal->j_flush_mutex);
1413 } else if (mutex_trylock(&journal->j_flush_mutex)) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001414 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001416
1417 count = 0;
1418 if (j_len_saved > journal->j_trans_max) {
1419 reiserfs_panic(s,
1420 "journal-715: flush_journal_list, length is %lu, trans id %lu\n",
1421 j_len_saved, jl->j_trans_id);
1422 return 0;
1423 }
1424
1425 get_fs_excl();
1426
1427 /* if all the work is already done, get out of here */
1428 if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1429 atomic_read(&(jl->j_commit_left)) <= 0) {
1430 goto flush_older_and_return;
1431 }
1432
1433 /* start by putting the commit list on disk. This will also flush
1434 ** the commit lists of any olders transactions
1435 */
1436 flush_commit_list(s, jl, 1);
1437
1438 if (!(jl->j_state & LIST_DIRTY)
1439 && !reiserfs_is_journal_aborted(journal))
1440 BUG();
1441
1442 /* are we done now? */
1443 if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1444 atomic_read(&(jl->j_commit_left)) <= 0) {
1445 goto flush_older_and_return;
1446 }
1447
1448 /* loop through each cnode, see if we need to write it,
1449 ** or wait on a more recent transaction, or just ignore it
1450 */
1451 if (atomic_read(&(journal->j_wcount)) != 0) {
1452 reiserfs_panic(s,
1453 "journal-844: panic journal list is flushing, wcount is not 0\n");
1454 }
1455 cn = jl->j_realblock;
1456 while (cn) {
1457 was_jwait = 0;
1458 was_dirty = 0;
1459 saved_bh = NULL;
1460 /* blocknr of 0 is no longer in the hash, ignore it */
1461 if (cn->blocknr == 0) {
1462 goto free_cnode;
1463 }
1464
1465 /* This transaction failed commit. Don't write out to the disk */
1466 if (!(jl->j_state & LIST_DIRTY))
1467 goto free_cnode;
1468
1469 pjl = find_newer_jl_for_cn(cn);
1470 /* the order is important here. We check pjl to make sure we
1471 ** don't clear BH_JDirty_wait if we aren't the one writing this
1472 ** block to disk
1473 */
1474 if (!pjl && cn->bh) {
1475 saved_bh = cn->bh;
1476
1477 /* we do this to make sure nobody releases the buffer while
1478 ** we are working with it
1479 */
1480 get_bh(saved_bh);
1481
1482 if (buffer_journal_dirty(saved_bh)) {
1483 BUG_ON(!can_dirty(cn));
1484 was_jwait = 1;
1485 was_dirty = 1;
1486 } else if (can_dirty(cn)) {
1487 /* everything with !pjl && jwait should be writable */
1488 BUG();
1489 }
1490 }
1491
1492 /* if someone has this block in a newer transaction, just make
Matt LaPlante0779bf22006-11-30 05:24:39 +01001493 ** sure they are committed, and don't try writing it to disk
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001494 */
1495 if (pjl) {
1496 if (atomic_read(&pjl->j_commit_left))
1497 flush_commit_list(s, pjl, 1);
1498 goto free_cnode;
1499 }
1500
1501 /* bh == NULL when the block got to disk on its own, OR,
1502 ** the block got freed in a future transaction
1503 */
1504 if (saved_bh == NULL) {
1505 goto free_cnode;
1506 }
1507
1508 /* this should never happen. kupdate_one_transaction has this list
1509 ** locked while it works, so we should never see a buffer here that
1510 ** is not marked JDirty_wait
1511 */
1512 if ((!was_jwait) && !buffer_locked(saved_bh)) {
1513 reiserfs_warning(s,
1514 "journal-813: BAD! buffer %llu %cdirty %cjwait, "
1515 "not in a newer tranasction",
1516 (unsigned long long)saved_bh->
1517 b_blocknr, was_dirty ? ' ' : '!',
1518 was_jwait ? ' ' : '!');
1519 }
1520 if (was_dirty) {
1521 /* we inc again because saved_bh gets decremented at free_cnode */
1522 get_bh(saved_bh);
1523 set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1524 lock_buffer(saved_bh);
1525 BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1526 if (buffer_dirty(saved_bh))
1527 submit_logged_buffer(saved_bh);
1528 else
1529 unlock_buffer(saved_bh);
1530 count++;
1531 } else {
1532 reiserfs_warning(s,
1533 "clm-2082: Unable to flush buffer %llu in %s",
1534 (unsigned long long)saved_bh->
Harvey Harrisonfbe54982008-04-28 02:16:22 -07001535 b_blocknr, __func__);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001536 }
1537 free_cnode:
1538 last = cn;
1539 cn = cn->next;
1540 if (saved_bh) {
1541 /* we incremented this to keep others from taking the buffer head away */
1542 put_bh(saved_bh);
1543 if (atomic_read(&(saved_bh->b_count)) < 0) {
1544 reiserfs_warning(s,
1545 "journal-945: saved_bh->b_count < 0");
1546 }
1547 }
1548 }
1549 if (count > 0) {
1550 cn = jl->j_realblock;
1551 while (cn) {
1552 if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1553 if (!cn->bh) {
1554 reiserfs_panic(s,
1555 "journal-1011: cn->bh is NULL\n");
1556 }
1557 wait_on_buffer(cn->bh);
1558 if (!cn->bh) {
1559 reiserfs_panic(s,
1560 "journal-1012: cn->bh is NULL\n");
1561 }
1562 if (unlikely(!buffer_uptodate(cn->bh))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563#ifdef CONFIG_REISERFS_CHECK
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001564 reiserfs_warning(s,
1565 "journal-949: buffer write failed\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566#endif
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001567 err = -EIO;
1568 }
1569 /* note, we must clear the JDirty_wait bit after the up to date
1570 ** check, otherwise we race against our flushpage routine
1571 */
1572 BUG_ON(!test_clear_buffer_journal_dirty
1573 (cn->bh));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574
Chris Mason398c95b2007-10-16 23:29:44 -07001575 /* drop one ref for us */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001576 put_bh(cn->bh);
Chris Mason398c95b2007-10-16 23:29:44 -07001577 /* drop one ref for journal_mark_dirty */
1578 release_buffer_page(cn->bh);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001579 }
1580 cn = cn->next;
1581 }
1582 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001584 if (err)
1585 reiserfs_abort(s, -EIO,
1586 "Write error while pushing transaction to disk in %s",
Harvey Harrisonfbe54982008-04-28 02:16:22 -07001587 __func__);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001588 flush_older_and_return:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001590 /* before we can update the journal header block, we _must_ flush all
1591 ** real blocks from all older transactions to disk. This is because
1592 ** once the header block is updated, this transaction will not be
1593 ** replayed after a crash
1594 */
1595 if (flushall) {
1596 flush_older_journal_lists(s, jl);
1597 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001599 err = journal->j_errno;
1600 /* before we can remove everything from the hash tables for this
1601 ** transaction, we must make sure it can never be replayed
1602 **
1603 ** since we are only called from do_journal_end, we know for sure there
1604 ** are no allocations going on while we are flushing journal lists. So,
1605 ** we only need to update the journal header block for the last list
1606 ** being flushed
1607 */
1608 if (!err && flushall) {
1609 err =
1610 update_journal_header_block(s,
1611 (jl->j_start + jl->j_len +
1612 2) % SB_ONDISK_JOURNAL_SIZE(s),
1613 jl->j_trans_id);
1614 if (err)
1615 reiserfs_abort(s, -EIO,
1616 "Write error while updating journal header in %s",
Harvey Harrisonfbe54982008-04-28 02:16:22 -07001617 __func__);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001618 }
1619 remove_all_from_journal_list(s, jl, 0);
1620 list_del_init(&jl->j_list);
1621 journal->j_num_lists--;
1622 del_from_work_list(s, jl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001624 if (journal->j_last_flush_id != 0 &&
1625 (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1626 reiserfs_warning(s, "clm-2201: last flush %lu, current %lu",
1627 journal->j_last_flush_id, jl->j_trans_id);
1628 }
1629 journal->j_last_flush_id = jl->j_trans_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001631 /* not strictly required since we are freeing the list, but it should
1632 * help find code using dead lists later on
1633 */
1634 jl->j_len = 0;
1635 atomic_set(&(jl->j_nonzerolen), 0);
1636 jl->j_start = 0;
1637 jl->j_realblock = NULL;
1638 jl->j_commit_bh = NULL;
1639 jl->j_trans_id = 0;
1640 jl->j_state = 0;
1641 put_journal_list(s, jl);
1642 if (flushall)
Jeff Mahoneyafe70252008-07-25 01:46:39 -07001643 mutex_unlock(&journal->j_flush_mutex);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001644 put_fs_excl();
1645 return err;
1646}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647
Chris Masona3172022006-09-29 01:59:56 -07001648static int test_transaction(struct super_block *s,
1649 struct reiserfs_journal_list *jl)
1650{
1651 struct reiserfs_journal_cnode *cn;
1652
1653 if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0)
1654 return 1;
1655
1656 cn = jl->j_realblock;
1657 while (cn) {
1658 /* if the blocknr == 0, this has been cleared from the hash,
1659 ** skip it
1660 */
1661 if (cn->blocknr == 0) {
1662 goto next;
1663 }
1664 if (cn->bh && !newer_jl_done(cn))
1665 return 0;
1666 next:
1667 cn = cn->next;
1668 cond_resched();
1669 }
1670 return 0;
1671}
1672
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673static int write_one_transaction(struct super_block *s,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001674 struct reiserfs_journal_list *jl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 struct buffer_chunk *chunk)
1676{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001677 struct reiserfs_journal_cnode *cn;
1678 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001680 jl->j_state |= LIST_TOUCHED;
1681 del_from_work_list(s, jl);
1682 if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1683 return 0;
1684 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001686 cn = jl->j_realblock;
1687 while (cn) {
1688 /* if the blocknr == 0, this has been cleared from the hash,
1689 ** skip it
1690 */
1691 if (cn->blocknr == 0) {
1692 goto next;
1693 }
1694 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1695 struct buffer_head *tmp_bh;
1696 /* we can race against journal_mark_freed when we try
1697 * to lock_buffer(cn->bh), so we have to inc the buffer
1698 * count, and recheck things after locking
1699 */
1700 tmp_bh = cn->bh;
1701 get_bh(tmp_bh);
1702 lock_buffer(tmp_bh);
1703 if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1704 if (!buffer_journal_dirty(tmp_bh) ||
1705 buffer_journal_prepared(tmp_bh))
1706 BUG();
1707 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1708 ret++;
1709 } else {
1710 /* note, cn->bh might be null now */
1711 unlock_buffer(tmp_bh);
1712 }
1713 put_bh(tmp_bh);
1714 }
1715 next:
1716 cn = cn->next;
1717 cond_resched();
1718 }
1719 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720}
1721
1722/* used by flush_commit_list */
1723static int dirty_one_transaction(struct super_block *s,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001724 struct reiserfs_journal_list *jl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001726 struct reiserfs_journal_cnode *cn;
1727 struct reiserfs_journal_list *pjl;
1728 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001730 jl->j_state |= LIST_DIRTY;
1731 cn = jl->j_realblock;
1732 while (cn) {
1733 /* look for a more recent transaction that logged this
1734 ** buffer. Only the most recent transaction with a buffer in
1735 ** it is allowed to send that buffer to disk
1736 */
1737 pjl = find_newer_jl_for_cn(cn);
1738 if (!pjl && cn->blocknr && cn->bh
1739 && buffer_journal_dirty(cn->bh)) {
1740 BUG_ON(!can_dirty(cn));
1741 /* if the buffer is prepared, it will either be logged
1742 * or restored. If restored, we need to make sure
1743 * it actually gets marked dirty
1744 */
1745 clear_buffer_journal_new(cn->bh);
1746 if (buffer_journal_prepared(cn->bh)) {
1747 set_buffer_journal_restore_dirty(cn->bh);
1748 } else {
1749 set_buffer_journal_test(cn->bh);
1750 mark_buffer_dirty(cn->bh);
1751 }
1752 }
1753 cn = cn->next;
1754 }
1755 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756}
1757
1758static int kupdate_transactions(struct super_block *s,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001759 struct reiserfs_journal_list *jl,
1760 struct reiserfs_journal_list **next_jl,
1761 unsigned long *next_trans_id,
1762 int num_blocks, int num_trans)
1763{
1764 int ret = 0;
1765 int written = 0;
1766 int transactions_flushed = 0;
1767 unsigned long orig_trans_id = jl->j_trans_id;
1768 struct buffer_chunk chunk;
1769 struct list_head *entry;
1770 struct reiserfs_journal *journal = SB_JOURNAL(s);
1771 chunk.nr = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772
Jeff Mahoneyafe70252008-07-25 01:46:39 -07001773 mutex_lock(&journal->j_flush_mutex);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001774 if (!journal_list_still_alive(s, orig_trans_id)) {
1775 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777
Jeff Mahoneyafe70252008-07-25 01:46:39 -07001778 /* we've got j_flush_mutex held, nobody is going to delete any
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001779 * of these lists out from underneath us
1780 */
1781 while ((num_trans && transactions_flushed < num_trans) ||
1782 (!num_trans && written < num_blocks)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001784 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1785 atomic_read(&jl->j_commit_left)
1786 || !(jl->j_state & LIST_DIRTY)) {
1787 del_from_work_list(s, jl);
1788 break;
1789 }
1790 ret = write_one_transaction(s, jl, &chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001792 if (ret < 0)
1793 goto done;
1794 transactions_flushed++;
1795 written += ret;
1796 entry = jl->j_list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001798 /* did we wrap? */
1799 if (entry == &journal->j_journal_list) {
1800 break;
1801 }
1802 jl = JOURNAL_LIST_ENTRY(entry);
1803
1804 /* don't bother with older transactions */
1805 if (jl->j_trans_id <= orig_trans_id)
1806 break;
1807 }
1808 if (chunk.nr) {
1809 write_chunk(&chunk);
1810 }
1811
1812 done:
Jeff Mahoneyafe70252008-07-25 01:46:39 -07001813 mutex_unlock(&journal->j_flush_mutex);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001814 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815}
1816
1817/* for o_sync and fsync heavy applications, they tend to use
1818** all the journa list slots with tiny transactions. These
1819** trigger lots and lots of calls to update the header block, which
1820** adds seeks and slows things down.
1821**
1822** This function tries to clear out a large chunk of the journal lists
1823** at once, which makes everything faster since only the newest journal
1824** list updates the header block
1825*/
1826static int flush_used_journal_lists(struct super_block *s,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001827 struct reiserfs_journal_list *jl)
1828{
1829 unsigned long len = 0;
1830 unsigned long cur_len;
1831 int ret;
1832 int i;
1833 int limit = 256;
1834 struct reiserfs_journal_list *tjl;
1835 struct reiserfs_journal_list *flush_jl;
1836 unsigned long trans_id;
1837 struct reiserfs_journal *journal = SB_JOURNAL(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001839 flush_jl = tjl = jl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001841 /* in data logging mode, try harder to flush a lot of blocks */
1842 if (reiserfs_data_log(s))
1843 limit = 1024;
1844 /* flush for 256 transactions or limit blocks, whichever comes first */
1845 for (i = 0; i < 256 && len < limit; i++) {
1846 if (atomic_read(&tjl->j_commit_left) ||
1847 tjl->j_trans_id < jl->j_trans_id) {
1848 break;
1849 }
1850 cur_len = atomic_read(&tjl->j_nonzerolen);
1851 if (cur_len > 0) {
1852 tjl->j_state &= ~LIST_TOUCHED;
1853 }
1854 len += cur_len;
1855 flush_jl = tjl;
1856 if (tjl->j_list.next == &journal->j_journal_list)
1857 break;
1858 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001860 /* try to find a group of blocks we can flush across all the
1861 ** transactions, but only bother if we've actually spanned
1862 ** across multiple lists
1863 */
1864 if (flush_jl != jl) {
1865 ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001867 flush_journal_list(s, flush_jl, 1);
1868 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869}
1870
1871/*
1872** removes any nodes in table with name block and dev as bh.
1873** only touchs the hnext and hprev pointers.
1874*/
1875void remove_journal_hash(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001876 struct reiserfs_journal_cnode **table,
1877 struct reiserfs_journal_list *jl,
1878 unsigned long block, int remove_freed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001880 struct reiserfs_journal_cnode *cur;
1881 struct reiserfs_journal_cnode **head;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001883 head = &(journal_hash(table, sb, block));
1884 if (!head) {
1885 return;
1886 }
1887 cur = *head;
1888 while (cur) {
1889 if (cur->blocknr == block && cur->sb == sb
1890 && (jl == NULL || jl == cur->jlist)
1891 && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1892 if (cur->hnext) {
1893 cur->hnext->hprev = cur->hprev;
1894 }
1895 if (cur->hprev) {
1896 cur->hprev->hnext = cur->hnext;
1897 } else {
1898 *head = cur->hnext;
1899 }
1900 cur->blocknr = 0;
1901 cur->sb = NULL;
1902 cur->state = 0;
1903 if (cur->bh && cur->jlist) /* anybody who clears the cur->bh will also dec the nonzerolen */
1904 atomic_dec(&(cur->jlist->j_nonzerolen));
1905 cur->bh = NULL;
1906 cur->jlist = NULL;
1907 }
1908 cur = cur->hnext;
1909 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910}
1911
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001912static void free_journal_ram(struct super_block *p_s_sb)
1913{
1914 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
Pekka Enbergd739b422006-02-01 03:06:43 -08001915 kfree(journal->j_current_jl);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001916 journal->j_num_lists--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001918 vfree(journal->j_cnode_free_orig);
1919 free_list_bitmaps(p_s_sb, journal->j_list_bitmap);
1920 free_bitmap_nodes(p_s_sb); /* must be after free_list_bitmaps */
1921 if (journal->j_header_bh) {
1922 brelse(journal->j_header_bh);
1923 }
1924 /* j_header_bh is on the journal dev, make sure not to release the journal
1925 * dev until we brelse j_header_bh
1926 */
1927 release_journal_dev(p_s_sb, journal);
1928 vfree(journal);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929}
1930
1931/*
1932** call on unmount. Only set error to 1 if you haven't made your way out
1933** of read_super() yet. Any other caller must keep error at 0.
1934*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001935static int do_journal_release(struct reiserfs_transaction_handle *th,
1936 struct super_block *p_s_sb, int error)
1937{
1938 struct reiserfs_transaction_handle myth;
1939 int flushed = 0;
1940 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001942 /* we only want to flush out transactions if we were called with error == 0
1943 */
1944 if (!error && !(p_s_sb->s_flags & MS_RDONLY)) {
1945 /* end the current trans */
1946 BUG_ON(!th->t_trans_id);
1947 do_journal_end(th, p_s_sb, 10, FLUSH_ALL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001949 /* make sure something gets logged to force our way into the flush code */
1950 if (!journal_join(&myth, p_s_sb, 1)) {
1951 reiserfs_prepare_for_journal(p_s_sb,
1952 SB_BUFFER_WITH_SB(p_s_sb),
1953 1);
1954 journal_mark_dirty(&myth, p_s_sb,
1955 SB_BUFFER_WITH_SB(p_s_sb));
1956 do_journal_end(&myth, p_s_sb, 1, FLUSH_ALL);
1957 flushed = 1;
1958 }
1959 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001961 /* this also catches errors during the do_journal_end above */
1962 if (!error && reiserfs_is_journal_aborted(journal)) {
1963 memset(&myth, 0, sizeof(myth));
1964 if (!journal_join_abort(&myth, p_s_sb, 1)) {
1965 reiserfs_prepare_for_journal(p_s_sb,
1966 SB_BUFFER_WITH_SB(p_s_sb),
1967 1);
1968 journal_mark_dirty(&myth, p_s_sb,
1969 SB_BUFFER_WITH_SB(p_s_sb));
1970 do_journal_end(&myth, p_s_sb, 1, FLUSH_ALL);
1971 }
1972 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001974 reiserfs_mounted_fs_count--;
1975 /* wait for all commits to finish */
1976 cancel_delayed_work(&SB_JOURNAL(p_s_sb)->j_work);
1977 flush_workqueue(commit_wq);
1978 if (!reiserfs_mounted_fs_count) {
1979 destroy_workqueue(commit_wq);
1980 commit_wq = NULL;
1981 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001983 free_journal_ram(p_s_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001985 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986}
1987
1988/*
1989** call on unmount. flush all journal trans, release all alloc'd ram
1990*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001991int journal_release(struct reiserfs_transaction_handle *th,
1992 struct super_block *p_s_sb)
1993{
1994 return do_journal_release(th, p_s_sb, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995}
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001996
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997/*
1998** only call from an error condition inside reiserfs_read_super!
1999*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002000int journal_release_error(struct reiserfs_transaction_handle *th,
2001 struct super_block *p_s_sb)
2002{
2003 return do_journal_release(th, p_s_sb, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004}
2005
2006/* compares description block with commit block. returns 1 if they differ, 0 if they are the same */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002007static int journal_compare_desc_commit(struct super_block *p_s_sb,
2008 struct reiserfs_journal_desc *desc,
2009 struct reiserfs_journal_commit *commit)
2010{
2011 if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
2012 get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
2013 get_commit_trans_len(commit) > SB_JOURNAL(p_s_sb)->j_trans_max ||
2014 get_commit_trans_len(commit) <= 0) {
2015 return 1;
2016 }
2017 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018}
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002019
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020/* returns 0 if it did not find a description block
2021** returns -1 if it found a corrupt commit block
2022** returns 1 if both desc and commit were valid
2023*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002024static int journal_transaction_is_valid(struct super_block *p_s_sb,
2025 struct buffer_head *d_bh,
2026 unsigned long *oldest_invalid_trans_id,
2027 unsigned long *newest_mount_id)
2028{
2029 struct reiserfs_journal_desc *desc;
2030 struct reiserfs_journal_commit *commit;
2031 struct buffer_head *c_bh;
2032 unsigned long offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002034 if (!d_bh)
2035 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002037 desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2038 if (get_desc_trans_len(desc) > 0
2039 && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
2040 if (oldest_invalid_trans_id && *oldest_invalid_trans_id
2041 && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
2042 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2043 "journal-986: transaction "
2044 "is valid returning because trans_id %d is greater than "
2045 "oldest_invalid %lu",
2046 get_desc_trans_id(desc),
2047 *oldest_invalid_trans_id);
2048 return 0;
2049 }
2050 if (newest_mount_id
2051 && *newest_mount_id > get_desc_mount_id(desc)) {
2052 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2053 "journal-1087: transaction "
2054 "is valid returning because mount_id %d is less than "
2055 "newest_mount_id %lu",
2056 get_desc_mount_id(desc),
2057 *newest_mount_id);
2058 return -1;
2059 }
2060 if (get_desc_trans_len(desc) > SB_JOURNAL(p_s_sb)->j_trans_max) {
2061 reiserfs_warning(p_s_sb,
2062 "journal-2018: Bad transaction length %d encountered, ignoring transaction",
2063 get_desc_trans_len(desc));
2064 return -1;
2065 }
2066 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002068 /* ok, we have a journal description block, lets see if the transaction was valid */
2069 c_bh =
2070 journal_bread(p_s_sb,
2071 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2072 ((offset + get_desc_trans_len(desc) +
2073 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
2074 if (!c_bh)
2075 return 0;
2076 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2077 if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
2078 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2079 "journal_transaction_is_valid, commit offset %ld had bad "
2080 "time %d or length %d",
2081 c_bh->b_blocknr -
2082 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2083 get_commit_trans_id(commit),
2084 get_commit_trans_len(commit));
2085 brelse(c_bh);
2086 if (oldest_invalid_trans_id) {
2087 *oldest_invalid_trans_id =
2088 get_desc_trans_id(desc);
2089 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2090 "journal-1004: "
2091 "transaction_is_valid setting oldest invalid trans_id "
2092 "to %d",
2093 get_desc_trans_id(desc));
2094 }
2095 return -1;
2096 }
2097 brelse(c_bh);
2098 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2099 "journal-1006: found valid "
2100 "transaction start offset %llu, len %d id %d",
2101 d_bh->b_blocknr -
2102 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2103 get_desc_trans_len(desc),
2104 get_desc_trans_id(desc));
2105 return 1;
2106 } else {
2107 return 0;
2108 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109}
2110
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002111static void brelse_array(struct buffer_head **heads, int num)
2112{
2113 int i;
2114 for (i = 0; i < num; i++) {
2115 brelse(heads[i]);
2116 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117}
2118
2119/*
2120** given the start, and values for the oldest acceptable transactions,
2121** this either reads in a replays a transaction, or returns because the transaction
2122** is invalid, or too old.
2123*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002124static int journal_read_transaction(struct super_block *p_s_sb,
2125 unsigned long cur_dblock,
2126 unsigned long oldest_start,
2127 unsigned long oldest_trans_id,
2128 unsigned long newest_mount_id)
2129{
2130 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2131 struct reiserfs_journal_desc *desc;
2132 struct reiserfs_journal_commit *commit;
2133 unsigned long trans_id = 0;
2134 struct buffer_head *c_bh;
2135 struct buffer_head *d_bh;
2136 struct buffer_head **log_blocks = NULL;
2137 struct buffer_head **real_blocks = NULL;
2138 unsigned long trans_offset;
2139 int i;
2140 int trans_half;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002142 d_bh = journal_bread(p_s_sb, cur_dblock);
2143 if (!d_bh)
2144 return 1;
2145 desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2146 trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2147 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1037: "
2148 "journal_read_transaction, offset %llu, len %d mount_id %d",
2149 d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2150 get_desc_trans_len(desc), get_desc_mount_id(desc));
2151 if (get_desc_trans_id(desc) < oldest_trans_id) {
2152 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1039: "
2153 "journal_read_trans skipping because %lu is too old",
2154 cur_dblock -
2155 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb));
2156 brelse(d_bh);
2157 return 1;
2158 }
2159 if (get_desc_mount_id(desc) != newest_mount_id) {
2160 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1146: "
2161 "journal_read_trans skipping because %d is != "
2162 "newest_mount_id %lu", get_desc_mount_id(desc),
2163 newest_mount_id);
2164 brelse(d_bh);
2165 return 1;
2166 }
2167 c_bh = journal_bread(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2168 ((trans_offset + get_desc_trans_len(desc) + 1) %
2169 SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
2170 if (!c_bh) {
2171 brelse(d_bh);
2172 return 1;
2173 }
2174 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2175 if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
2176 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2177 "journal_read_transaction, "
2178 "commit offset %llu had bad time %d or length %d",
2179 c_bh->b_blocknr -
2180 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2181 get_commit_trans_id(commit),
2182 get_commit_trans_len(commit));
2183 brelse(c_bh);
2184 brelse(d_bh);
2185 return 1;
2186 }
2187 trans_id = get_desc_trans_id(desc);
2188 /* now we know we've got a good transaction, and it was inside the valid time ranges */
Pekka Enbergd739b422006-02-01 03:06:43 -08002189 log_blocks = kmalloc(get_desc_trans_len(desc) *
2190 sizeof(struct buffer_head *), GFP_NOFS);
2191 real_blocks = kmalloc(get_desc_trans_len(desc) *
2192 sizeof(struct buffer_head *), GFP_NOFS);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002193 if (!log_blocks || !real_blocks) {
2194 brelse(c_bh);
2195 brelse(d_bh);
Pekka Enbergd739b422006-02-01 03:06:43 -08002196 kfree(log_blocks);
2197 kfree(real_blocks);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002198 reiserfs_warning(p_s_sb,
2199 "journal-1169: kmalloc failed, unable to mount FS");
2200 return -1;
2201 }
2202 /* get all the buffer heads */
2203 trans_half = journal_trans_half(p_s_sb->s_blocksize);
2204 for (i = 0; i < get_desc_trans_len(desc); i++) {
2205 log_blocks[i] =
2206 journal_getblk(p_s_sb,
2207 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2208 (trans_offset + 1 +
2209 i) % SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2210 if (i < trans_half) {
2211 real_blocks[i] =
2212 sb_getblk(p_s_sb,
2213 le32_to_cpu(desc->j_realblock[i]));
2214 } else {
2215 real_blocks[i] =
2216 sb_getblk(p_s_sb,
2217 le32_to_cpu(commit->
2218 j_realblock[i - trans_half]));
2219 }
2220 if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(p_s_sb)) {
2221 reiserfs_warning(p_s_sb,
2222 "journal-1207: REPLAY FAILURE fsck required! Block to replay is outside of filesystem");
2223 goto abort_replay;
2224 }
2225 /* make sure we don't try to replay onto log or reserved area */
2226 if (is_block_in_log_or_reserved_area
2227 (p_s_sb, real_blocks[i]->b_blocknr)) {
2228 reiserfs_warning(p_s_sb,
2229 "journal-1204: REPLAY FAILURE fsck required! Trying to replay onto a log block");
2230 abort_replay:
2231 brelse_array(log_blocks, i);
2232 brelse_array(real_blocks, i);
2233 brelse(c_bh);
2234 brelse(d_bh);
Pekka Enbergd739b422006-02-01 03:06:43 -08002235 kfree(log_blocks);
2236 kfree(real_blocks);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002237 return -1;
2238 }
2239 }
2240 /* read in the log blocks, memcpy to the corresponding real block */
2241 ll_rw_block(READ, get_desc_trans_len(desc), log_blocks);
2242 for (i = 0; i < get_desc_trans_len(desc); i++) {
2243 wait_on_buffer(log_blocks[i]);
2244 if (!buffer_uptodate(log_blocks[i])) {
2245 reiserfs_warning(p_s_sb,
2246 "journal-1212: REPLAY FAILURE fsck required! buffer write failed");
2247 brelse_array(log_blocks + i,
2248 get_desc_trans_len(desc) - i);
2249 brelse_array(real_blocks, get_desc_trans_len(desc));
2250 brelse(c_bh);
2251 brelse(d_bh);
Pekka Enbergd739b422006-02-01 03:06:43 -08002252 kfree(log_blocks);
2253 kfree(real_blocks);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002254 return -1;
2255 }
2256 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2257 real_blocks[i]->b_size);
2258 set_buffer_uptodate(real_blocks[i]);
2259 brelse(log_blocks[i]);
2260 }
2261 /* flush out the real blocks */
2262 for (i = 0; i < get_desc_trans_len(desc); i++) {
2263 set_buffer_dirty(real_blocks[i]);
Jan Kara53778ff2005-09-06 15:19:14 -07002264 ll_rw_block(SWRITE, 1, real_blocks + i);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002265 }
2266 for (i = 0; i < get_desc_trans_len(desc); i++) {
2267 wait_on_buffer(real_blocks[i]);
2268 if (!buffer_uptodate(real_blocks[i])) {
2269 reiserfs_warning(p_s_sb,
2270 "journal-1226: REPLAY FAILURE, fsck required! buffer write failed");
2271 brelse_array(real_blocks + i,
2272 get_desc_trans_len(desc) - i);
2273 brelse(c_bh);
2274 brelse(d_bh);
Pekka Enbergd739b422006-02-01 03:06:43 -08002275 kfree(log_blocks);
2276 kfree(real_blocks);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002277 return -1;
2278 }
2279 brelse(real_blocks[i]);
2280 }
2281 cur_dblock =
2282 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2283 ((trans_offset + get_desc_trans_len(desc) +
2284 2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2285 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2286 "journal-1095: setting journal " "start to offset %ld",
2287 cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb));
2288
2289 /* init starting values for the first transaction, in case this is the last transaction to be replayed. */
2290 journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2291 journal->j_last_flush_trans_id = trans_id;
2292 journal->j_trans_id = trans_id + 1;
Alexander Zarochentseva44c94a2006-03-25 03:06:59 -08002293 /* check for trans_id overflow */
2294 if (journal->j_trans_id == 0)
2295 journal->j_trans_id = 10;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002296 brelse(c_bh);
2297 brelse(d_bh);
Pekka Enbergd739b422006-02-01 03:06:43 -08002298 kfree(log_blocks);
2299 kfree(real_blocks);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002300 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301}
2302
2303/* This function reads blocks starting from block and to max_block of bufsize
2304 size (but no more than BUFNR blocks at a time). This proved to improve
2305 mounting speed on self-rebuilding raid5 arrays at least.
2306 Right now it is only used from journal code. But later we might use it
2307 from other places.
2308 Note: Do not use journal_getblk/sb_getblk functions here! */
Jeff Mahoney3ee16672007-10-18 23:39:25 -07002309static struct buffer_head *reiserfs_breada(struct block_device *dev,
2310 b_blocknr_t block, int bufsize,
2311 b_blocknr_t max_block)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002313 struct buffer_head *bhlist[BUFNR];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314 unsigned int blocks = BUFNR;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002315 struct buffer_head *bh;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 int i, j;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002317
2318 bh = __getblk(dev, block, bufsize);
2319 if (buffer_uptodate(bh))
2320 return (bh);
2321
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322 if (block + BUFNR > max_block) {
2323 blocks = max_block - block;
2324 }
2325 bhlist[0] = bh;
2326 j = 1;
2327 for (i = 1; i < blocks; i++) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002328 bh = __getblk(dev, block + i, bufsize);
2329 if (buffer_uptodate(bh)) {
2330 brelse(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331 break;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002332 } else
2333 bhlist[j++] = bh;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002335 ll_rw_block(READ, j, bhlist);
2336 for (i = 1; i < j; i++)
2337 brelse(bhlist[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338 bh = bhlist[0];
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002339 wait_on_buffer(bh);
2340 if (buffer_uptodate(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341 return bh;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002342 brelse(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343 return NULL;
2344}
2345
2346/*
2347** read and replay the log
2348** on a clean unmount, the journal header's next unflushed pointer will be to an invalid
2349** transaction. This tests that before finding all the transactions in the log, which makes normal mount times fast.
2350**
2351** After a crash, this starts with the next unflushed transaction, and replays until it finds one too old, or invalid.
2352**
2353** On exit, it sets things up so the first transaction will work correctly.
2354*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002355static int journal_read(struct super_block *p_s_sb)
2356{
2357 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2358 struct reiserfs_journal_desc *desc;
2359 unsigned long oldest_trans_id = 0;
2360 unsigned long oldest_invalid_trans_id = 0;
2361 time_t start;
2362 unsigned long oldest_start = 0;
2363 unsigned long cur_dblock = 0;
2364 unsigned long newest_mount_id = 9;
2365 struct buffer_head *d_bh;
2366 struct reiserfs_journal_header *jh;
2367 int valid_journal_header = 0;
2368 int replay_count = 0;
2369 int continue_replay = 1;
2370 int ret;
2371 char b[BDEVNAME_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002373 cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2374 reiserfs_info(p_s_sb, "checking transaction log (%s)\n",
2375 bdevname(journal->j_dev_bd, b));
2376 start = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002378 /* step 1, read in the journal header block. Check the transaction it says
2379 ** is the first unflushed, and if that transaction is not valid,
2380 ** replay is done
2381 */
2382 journal->j_header_bh = journal_bread(p_s_sb,
2383 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb)
2384 + SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2385 if (!journal->j_header_bh) {
2386 return 1;
2387 }
2388 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
Vladimir V. Savelievc499ec22006-03-02 02:54:39 -08002389 if (le32_to_cpu(jh->j_first_unflushed_offset) <
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002390 SB_ONDISK_JOURNAL_SIZE(p_s_sb)
2391 && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2392 oldest_start =
2393 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2394 le32_to_cpu(jh->j_first_unflushed_offset);
2395 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2396 newest_mount_id = le32_to_cpu(jh->j_mount_id);
2397 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2398 "journal-1153: found in "
2399 "header: first_unflushed_offset %d, last_flushed_trans_id "
2400 "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2401 le32_to_cpu(jh->j_last_flush_trans_id));
2402 valid_journal_header = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002404 /* now, we try to read the first unflushed offset. If it is not valid,
2405 ** there is nothing more we can do, and it makes no sense to read
2406 ** through the whole log.
2407 */
2408 d_bh =
2409 journal_bread(p_s_sb,
2410 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2411 le32_to_cpu(jh->j_first_unflushed_offset));
2412 ret = journal_transaction_is_valid(p_s_sb, d_bh, NULL, NULL);
2413 if (!ret) {
2414 continue_replay = 0;
2415 }
2416 brelse(d_bh);
2417 goto start_log_replay;
2418 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002420 if (continue_replay && bdev_read_only(p_s_sb->s_bdev)) {
2421 reiserfs_warning(p_s_sb,
2422 "clm-2076: device is readonly, unable to replay log");
2423 return -1;
2424 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002426 /* ok, there are transactions that need to be replayed. start with the first log block, find
2427 ** all the valid transactions, and pick out the oldest.
2428 */
2429 while (continue_replay
2430 && cur_dblock <
2431 (SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2432 SB_ONDISK_JOURNAL_SIZE(p_s_sb))) {
2433 /* Note that it is required for blocksize of primary fs device and journal
2434 device to be the same */
2435 d_bh =
2436 reiserfs_breada(journal->j_dev_bd, cur_dblock,
2437 p_s_sb->s_blocksize,
2438 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2439 SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2440 ret =
2441 journal_transaction_is_valid(p_s_sb, d_bh,
2442 &oldest_invalid_trans_id,
2443 &newest_mount_id);
2444 if (ret == 1) {
2445 desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2446 if (oldest_start == 0) { /* init all oldest_ values */
2447 oldest_trans_id = get_desc_trans_id(desc);
2448 oldest_start = d_bh->b_blocknr;
2449 newest_mount_id = get_desc_mount_id(desc);
2450 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2451 "journal-1179: Setting "
2452 "oldest_start to offset %llu, trans_id %lu",
2453 oldest_start -
2454 SB_ONDISK_JOURNAL_1st_BLOCK
2455 (p_s_sb), oldest_trans_id);
2456 } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2457 /* one we just read was older */
2458 oldest_trans_id = get_desc_trans_id(desc);
2459 oldest_start = d_bh->b_blocknr;
2460 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2461 "journal-1180: Resetting "
2462 "oldest_start to offset %lu, trans_id %lu",
2463 oldest_start -
2464 SB_ONDISK_JOURNAL_1st_BLOCK
2465 (p_s_sb), oldest_trans_id);
2466 }
2467 if (newest_mount_id < get_desc_mount_id(desc)) {
2468 newest_mount_id = get_desc_mount_id(desc);
2469 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2470 "journal-1299: Setting "
2471 "newest_mount_id to %d",
2472 get_desc_mount_id(desc));
2473 }
2474 cur_dblock += get_desc_trans_len(desc) + 2;
2475 } else {
2476 cur_dblock++;
2477 }
2478 brelse(d_bh);
2479 }
2480
2481 start_log_replay:
2482 cur_dblock = oldest_start;
2483 if (oldest_trans_id) {
2484 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2485 "journal-1206: Starting replay "
2486 "from offset %llu, trans_id %lu",
2487 cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2488 oldest_trans_id);
2489
2490 }
2491 replay_count = 0;
2492 while (continue_replay && oldest_trans_id > 0) {
2493 ret =
2494 journal_read_transaction(p_s_sb, cur_dblock, oldest_start,
2495 oldest_trans_id, newest_mount_id);
2496 if (ret < 0) {
2497 return ret;
2498 } else if (ret != 0) {
2499 break;
2500 }
2501 cur_dblock =
2502 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + journal->j_start;
2503 replay_count++;
2504 if (cur_dblock == oldest_start)
2505 break;
2506 }
2507
2508 if (oldest_trans_id == 0) {
2509 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2510 "journal-1225: No valid " "transactions found");
2511 }
2512 /* j_start does not get set correctly if we don't replay any transactions.
2513 ** if we had a valid journal_header, set j_start to the first unflushed transaction value,
2514 ** copy the trans_id from the header
2515 */
2516 if (valid_journal_header && replay_count == 0) {
2517 journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2518 journal->j_trans_id =
2519 le32_to_cpu(jh->j_last_flush_trans_id) + 1;
Alexander Zarochentseva44c94a2006-03-25 03:06:59 -08002520 /* check for trans_id overflow */
2521 if (journal->j_trans_id == 0)
2522 journal->j_trans_id = 10;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002523 journal->j_last_flush_trans_id =
2524 le32_to_cpu(jh->j_last_flush_trans_id);
2525 journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2526 } else {
2527 journal->j_mount_id = newest_mount_id + 1;
2528 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002530 "newest_mount_id to %lu", journal->j_mount_id);
2531 journal->j_first_unflushed_offset = journal->j_start;
2532 if (replay_count > 0) {
2533 reiserfs_info(p_s_sb,
2534 "replayed %d transactions in %lu seconds\n",
2535 replay_count, get_seconds() - start);
2536 }
2537 if (!bdev_read_only(p_s_sb->s_bdev) &&
2538 _update_journal_header_block(p_s_sb, journal->j_start,
2539 journal->j_last_flush_trans_id)) {
2540 /* replay failed, caller must call free_journal_ram and abort
2541 ** the mount
2542 */
2543 return -1;
2544 }
2545 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546}
2547
2548static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2549{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002550 struct reiserfs_journal_list *jl;
Pekka Enberg8c777cc2006-02-01 03:06:43 -08002551 jl = kzalloc(sizeof(struct reiserfs_journal_list),
2552 GFP_NOFS | __GFP_NOFAIL);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002553 INIT_LIST_HEAD(&jl->j_list);
2554 INIT_LIST_HEAD(&jl->j_working_list);
2555 INIT_LIST_HEAD(&jl->j_tail_bh_list);
2556 INIT_LIST_HEAD(&jl->j_bh_list);
Jeff Mahoney90415de2008-07-25 01:46:40 -07002557 mutex_init(&jl->j_commit_mutex);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002558 SB_JOURNAL(s)->j_num_lists++;
2559 get_journal_list(jl);
2560 return jl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561}
2562
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002563static void journal_list_init(struct super_block *p_s_sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002565 SB_JOURNAL(p_s_sb)->j_current_jl = alloc_journal_list(p_s_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566}
2567
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002568static int release_journal_dev(struct super_block *super,
2569 struct reiserfs_journal *journal)
2570{
2571 int result;
2572
2573 result = 0;
2574
Christoph Hellwig86098fa2008-04-30 00:54:46 -07002575 if (journal->j_dev_bd != NULL) {
2576 if (journal->j_dev_bd->bd_dev != super->s_dev)
2577 bd_release(journal->j_dev_bd);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002578 result = blkdev_put(journal->j_dev_bd);
2579 journal->j_dev_bd = NULL;
2580 }
2581
2582 if (result != 0) {
2583 reiserfs_warning(super,
2584 "sh-457: release_journal_dev: Cannot release journal device: %i",
2585 result);
2586 }
2587 return result;
2588}
2589
2590static int journal_init_dev(struct super_block *super,
2591 struct reiserfs_journal *journal,
2592 const char *jdev_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593{
2594 int result;
2595 dev_t jdev;
2596 int blkdev_mode = FMODE_READ | FMODE_WRITE;
2597 char b[BDEVNAME_SIZE];
2598
2599 result = 0;
2600
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002601 journal->j_dev_bd = NULL;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002602 jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2603 new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002604
2605 if (bdev_read_only(super->s_bdev))
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002606 blkdev_mode = FMODE_READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607
2608 /* there is no "jdev" option and journal is on separate device */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002609 if ((!jdev_name || !jdev_name[0])) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610 journal->j_dev_bd = open_by_devnum(jdev, blkdev_mode);
2611 if (IS_ERR(journal->j_dev_bd)) {
2612 result = PTR_ERR(journal->j_dev_bd);
2613 journal->j_dev_bd = NULL;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002614 reiserfs_warning(super, "sh-458: journal_init_dev: "
2615 "cannot init journal device '%s': %i",
2616 __bdevname(jdev, b), result);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617 return result;
Christoph Hellwig86098fa2008-04-30 00:54:46 -07002618 } else if (jdev != super->s_dev) {
2619 result = bd_claim(journal->j_dev_bd, journal);
2620 if (result) {
2621 blkdev_put(journal->j_dev_bd);
2622 return result;
2623 }
2624
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625 set_blocksize(journal->j_dev_bd, super->s_blocksize);
Christoph Hellwig86098fa2008-04-30 00:54:46 -07002626 }
2627
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628 return 0;
2629 }
2630
Christoph Hellwig86098fa2008-04-30 00:54:46 -07002631 journal->j_dev_bd = open_bdev_excl(jdev_name, 0, journal);
2632 if (IS_ERR(journal->j_dev_bd)) {
2633 result = PTR_ERR(journal->j_dev_bd);
2634 journal->j_dev_bd = NULL;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002635 reiserfs_warning(super,
2636 "journal_init_dev: Cannot open '%s': %i",
2637 jdev_name, result);
Christoph Hellwig86098fa2008-04-30 00:54:46 -07002638 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639 }
Christoph Hellwig86098fa2008-04-30 00:54:46 -07002640
2641 set_blocksize(journal->j_dev_bd, super->s_blocksize);
2642 reiserfs_info(super,
2643 "journal_init_dev: journal device: %s\n",
2644 bdevname(journal->j_dev_bd, b));
2645 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002646}
2647
Edward Shishkincf3d0b82007-10-16 23:30:30 -07002648/**
2649 * When creating/tuning a file system user can assign some
2650 * journal params within boundaries which depend on the ratio
2651 * blocksize/standard_blocksize.
2652 *
2653 * For blocks >= standard_blocksize transaction size should
2654 * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more
2655 * then JOURNAL_TRANS_MAX_DEFAULT.
2656 *
2657 * For blocks < standard_blocksize these boundaries should be
2658 * decreased proportionally.
2659 */
2660#define REISERFS_STANDARD_BLKSIZE (4096)
2661
2662static int check_advise_trans_params(struct super_block *p_s_sb,
2663 struct reiserfs_journal *journal)
2664{
2665 if (journal->j_trans_max) {
2666 /* Non-default journal params.
2667 Do sanity check for them. */
2668 int ratio = 1;
2669 if (p_s_sb->s_blocksize < REISERFS_STANDARD_BLKSIZE)
2670 ratio = REISERFS_STANDARD_BLKSIZE / p_s_sb->s_blocksize;
2671
2672 if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio ||
2673 journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio ||
2674 SB_ONDISK_JOURNAL_SIZE(p_s_sb) / journal->j_trans_max <
2675 JOURNAL_MIN_RATIO) {
2676 reiserfs_warning(p_s_sb,
2677 "sh-462: bad transaction max size (%u). FSCK?",
2678 journal->j_trans_max);
2679 return 1;
2680 }
2681 if (journal->j_max_batch != (journal->j_trans_max) *
2682 JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) {
2683 reiserfs_warning(p_s_sb,
2684 "sh-463: bad transaction max batch (%u). FSCK?",
2685 journal->j_max_batch);
2686 return 1;
2687 }
2688 } else {
2689 /* Default journal params.
2690 The file system was created by old version
2691 of mkreiserfs, so some fields contain zeros,
2692 and we need to advise proper values for them */
2693 if (p_s_sb->s_blocksize != REISERFS_STANDARD_BLKSIZE)
2694 reiserfs_panic(p_s_sb, "sh-464: bad blocksize (%u)",
2695 p_s_sb->s_blocksize);
2696 journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2697 journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2698 journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2699 }
2700 return 0;
2701}
2702
Linus Torvalds1da177e2005-04-16 15:20:36 -07002703/*
2704** must be called once on fs mount. calls journal_read for you
2705*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002706int journal_init(struct super_block *p_s_sb, const char *j_dev_name,
2707 int old_format, unsigned int commit_max_age)
2708{
2709 int num_cnodes = SB_ONDISK_JOURNAL_SIZE(p_s_sb) * 2;
2710 struct buffer_head *bhjh;
2711 struct reiserfs_super_block *rs;
2712 struct reiserfs_journal_header *jh;
2713 struct reiserfs_journal *journal;
2714 struct reiserfs_journal_list *jl;
2715 char b[BDEVNAME_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002717 journal = SB_JOURNAL(p_s_sb) = vmalloc(sizeof(struct reiserfs_journal));
2718 if (!journal) {
2719 reiserfs_warning(p_s_sb,
2720 "journal-1256: unable to get memory for journal structure");
2721 return 1;
2722 }
2723 memset(journal, 0, sizeof(struct reiserfs_journal));
2724 INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2725 INIT_LIST_HEAD(&journal->j_prealloc_list);
2726 INIT_LIST_HEAD(&journal->j_working_list);
2727 INIT_LIST_HEAD(&journal->j_journal_list);
2728 journal->j_persistent_trans = 0;
2729 if (reiserfs_allocate_list_bitmaps(p_s_sb,
2730 journal->j_list_bitmap,
Jeff Mahoneycb680c12007-10-18 23:39:27 -07002731 reiserfs_bmap_count(p_s_sb)))
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002732 goto free_and_return;
2733 allocate_bitmap_nodes(p_s_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002734
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002735 /* reserved for journal area support */
2736 SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) = (old_format ?
2737 REISERFS_OLD_DISK_OFFSET_IN_BYTES
2738 / p_s_sb->s_blocksize +
Jeff Mahoneycb680c12007-10-18 23:39:27 -07002739 reiserfs_bmap_count(p_s_sb) +
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002740 1 :
2741 REISERFS_DISK_OFFSET_IN_BYTES /
2742 p_s_sb->s_blocksize + 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002743
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002744 /* Sanity check to see is the standard journal fitting withing first bitmap
2745 (actual for small blocksizes) */
2746 if (!SB_ONDISK_JOURNAL_DEVICE(p_s_sb) &&
2747 (SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) +
2748 SB_ONDISK_JOURNAL_SIZE(p_s_sb) > p_s_sb->s_blocksize * 8)) {
2749 reiserfs_warning(p_s_sb,
2750 "journal-1393: journal does not fit for area "
2751 "addressed by first of bitmap blocks. It starts at "
2752 "%u and its size is %u. Block size %ld",
2753 SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb),
2754 SB_ONDISK_JOURNAL_SIZE(p_s_sb),
2755 p_s_sb->s_blocksize);
2756 goto free_and_return;
2757 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002758
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002759 if (journal_init_dev(p_s_sb, journal, j_dev_name) != 0) {
2760 reiserfs_warning(p_s_sb,
2761 "sh-462: unable to initialize jornal device");
2762 goto free_and_return;
2763 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002764
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002765 rs = SB_DISK_SUPER_BLOCK(p_s_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002766
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002767 /* read journal header */
2768 bhjh = journal_bread(p_s_sb,
2769 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2770 SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2771 if (!bhjh) {
2772 reiserfs_warning(p_s_sb,
2773 "sh-459: unable to read journal header");
2774 goto free_and_return;
2775 }
2776 jh = (struct reiserfs_journal_header *)(bhjh->b_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002778 /* make sure that journal matches to the super block */
2779 if (is_reiserfs_jr(rs)
2780 && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2781 sb_jp_journal_magic(rs))) {
2782 reiserfs_warning(p_s_sb,
2783 "sh-460: journal header magic %x "
2784 "(device %s) does not match to magic found in super "
2785 "block %x", jh->jh_journal.jp_journal_magic,
2786 bdevname(journal->j_dev_bd, b),
2787 sb_jp_journal_magic(rs));
2788 brelse(bhjh);
2789 goto free_and_return;
2790 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002792 journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2793 journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2794 journal->j_max_commit_age =
2795 le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2796 journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797
Edward Shishkincf3d0b82007-10-16 23:30:30 -07002798 if (check_advise_trans_params(p_s_sb, journal) != 0)
2799 goto free_and_return;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002800 journal->j_default_max_commit_age = journal->j_max_commit_age;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002801
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002802 if (commit_max_age != 0) {
2803 journal->j_max_commit_age = commit_max_age;
2804 journal->j_max_trans_age = commit_max_age;
2805 }
2806
2807 reiserfs_info(p_s_sb, "journal params: device %s, size %u, "
2808 "journal first block %u, max trans len %u, max batch %u, "
2809 "max commit age %u, max trans age %u\n",
2810 bdevname(journal->j_dev_bd, b),
2811 SB_ONDISK_JOURNAL_SIZE(p_s_sb),
2812 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2813 journal->j_trans_max,
2814 journal->j_max_batch,
2815 journal->j_max_commit_age, journal->j_max_trans_age);
2816
2817 brelse(bhjh);
2818
2819 journal->j_list_bitmap_index = 0;
2820 journal_list_init(p_s_sb);
2821
2822 memset(journal->j_list_hash_table, 0,
2823 JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2824
2825 INIT_LIST_HEAD(&journal->j_dirty_buffers);
2826 spin_lock_init(&journal->j_dirty_buffers_lock);
2827
2828 journal->j_start = 0;
2829 journal->j_len = 0;
2830 journal->j_len_alloc = 0;
2831 atomic_set(&(journal->j_wcount), 0);
2832 atomic_set(&(journal->j_async_throttle), 0);
2833 journal->j_bcount = 0;
2834 journal->j_trans_start_time = 0;
2835 journal->j_last = NULL;
2836 journal->j_first = NULL;
2837 init_waitqueue_head(&(journal->j_join_wait));
Jeff Mahoneyf68215c2008-07-25 01:46:38 -07002838 mutex_init(&journal->j_mutex);
Jeff Mahoneyafe70252008-07-25 01:46:39 -07002839 mutex_init(&journal->j_flush_mutex);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002840
2841 journal->j_trans_id = 10;
2842 journal->j_mount_id = 10;
2843 journal->j_state = 0;
2844 atomic_set(&(journal->j_jlock), 0);
2845 journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2846 journal->j_cnode_free_orig = journal->j_cnode_free_list;
2847 journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2848 journal->j_cnode_used = 0;
2849 journal->j_must_wait = 0;
2850
Jeff Mahoney576f6d72005-11-29 19:34:39 -08002851 if (journal->j_cnode_free == 0) {
2852 reiserfs_warning(p_s_sb, "journal-2004: Journal cnode memory "
2853 "allocation failed (%ld bytes). Journal is "
2854 "too large for available memory. Usually "
2855 "this is due to a journal that is too large.",
2856 sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2857 goto free_and_return;
2858 }
2859
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002860 init_journal_hash(p_s_sb);
2861 jl = journal->j_current_jl;
2862 jl->j_list_bitmap = get_list_bitmap(p_s_sb, jl);
2863 if (!jl->j_list_bitmap) {
2864 reiserfs_warning(p_s_sb,
2865 "journal-2005, get_list_bitmap failed for journal list 0");
2866 goto free_and_return;
2867 }
2868 if (journal_read(p_s_sb) < 0) {
2869 reiserfs_warning(p_s_sb, "Replay Failure, unable to mount");
2870 goto free_and_return;
2871 }
2872
2873 reiserfs_mounted_fs_count++;
2874 if (reiserfs_mounted_fs_count <= 1)
2875 commit_wq = create_workqueue("reiserfs");
2876
David Howellsc4028952006-11-22 14:57:56 +00002877 INIT_DELAYED_WORK(&journal->j_work, flush_async_commits);
2878 journal->j_work_sb = p_s_sb;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002879 return 0;
2880 free_and_return:
2881 free_journal_ram(p_s_sb);
2882 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883}
2884
2885/*
2886** test for a polite end of the current transaction. Used by file_write, and should
2887** be used by delete to make sure they don't write more than can fit inside a single
2888** transaction
2889*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002890int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2891 int new_alloc)
2892{
2893 struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2894 time_t now = get_seconds();
2895 /* cannot restart while nested */
2896 BUG_ON(!th->t_trans_id);
2897 if (th->t_refcount > 1)
2898 return 0;
2899 if (journal->j_must_wait > 0 ||
2900 (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2901 atomic_read(&(journal->j_jlock)) ||
2902 (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2903 journal->j_cnode_free < (journal->j_trans_max * 3)) {
2904 return 1;
2905 }
Chris Mason6ae1ea42006-02-01 03:06:50 -08002906 /* protected by the BKL here */
2907 journal->j_len_alloc += new_alloc;
2908 th->t_blocks_allocated += new_alloc ;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002909 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002910}
2911
2912/* this must be called inside a transaction, and requires the
2913** kernel_lock to be held
2914*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002915void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002916{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002917 struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2918 BUG_ON(!th->t_trans_id);
2919 journal->j_must_wait = 1;
2920 set_bit(J_WRITERS_BLOCKED, &journal->j_state);
2921 return;
2922}
2923
2924/* this must be called without a transaction started, and does not
2925** require BKL
2926*/
2927void reiserfs_allow_writes(struct super_block *s)
2928{
2929 struct reiserfs_journal *journal = SB_JOURNAL(s);
2930 clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
2931 wake_up(&journal->j_join_wait);
2932}
2933
2934/* this must be called without a transaction started, and does not
2935** require BKL
2936*/
2937void reiserfs_wait_on_write_block(struct super_block *s)
2938{
2939 struct reiserfs_journal *journal = SB_JOURNAL(s);
2940 wait_event(journal->j_join_wait,
2941 !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
2942}
2943
2944static void queue_log_writer(struct super_block *s)
2945{
2946 wait_queue_t wait;
2947 struct reiserfs_journal *journal = SB_JOURNAL(s);
2948 set_bit(J_WRITERS_QUEUED, &journal->j_state);
2949
2950 /*
2951 * we don't want to use wait_event here because
2952 * we only want to wait once.
2953 */
2954 init_waitqueue_entry(&wait, current);
2955 add_wait_queue(&journal->j_join_wait, &wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002956 set_current_state(TASK_UNINTERRUPTIBLE);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002957 if (test_bit(J_WRITERS_QUEUED, &journal->j_state))
2958 schedule();
Milind Arun Choudhary5ab2f7e2007-05-08 00:30:51 -07002959 __set_current_state(TASK_RUNNING);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002960 remove_wait_queue(&journal->j_join_wait, &wait);
2961}
2962
2963static void wake_queued_writers(struct super_block *s)
2964{
2965 struct reiserfs_journal *journal = SB_JOURNAL(s);
2966 if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
2967 wake_up(&journal->j_join_wait);
2968}
2969
2970static void let_transaction_grow(struct super_block *sb, unsigned long trans_id)
2971{
2972 struct reiserfs_journal *journal = SB_JOURNAL(sb);
2973 unsigned long bcount = journal->j_bcount;
2974 while (1) {
Nishanth Aravamudan041e0e32005-09-10 00:27:23 -07002975 schedule_timeout_uninterruptible(1);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002976 journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
2977 while ((atomic_read(&journal->j_wcount) > 0 ||
2978 atomic_read(&journal->j_jlock)) &&
2979 journal->j_trans_id == trans_id) {
2980 queue_log_writer(sb);
2981 }
2982 if (journal->j_trans_id != trans_id)
2983 break;
2984 if (bcount == journal->j_bcount)
2985 break;
2986 bcount = journal->j_bcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002987 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002988}
2989
2990/* join == true if you must join an existing transaction.
2991** join == false if you can deal with waiting for others to finish
2992**
2993** this will block until the transaction is joinable. send the number of blocks you
2994** expect to use in nblocks.
2995*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002996static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
2997 struct super_block *p_s_sb, unsigned long nblocks,
2998 int join)
2999{
3000 time_t now = get_seconds();
3001 int old_trans_id;
3002 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3003 struct reiserfs_transaction_handle myth;
3004 int sched_count = 0;
3005 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003006
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003007 reiserfs_check_lock_depth(p_s_sb, "journal_begin");
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003008 BUG_ON(nblocks > journal->j_trans_max);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003009
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003010 PROC_INFO_INC(p_s_sb, journal.journal_being);
3011 /* set here for journal_join */
3012 th->t_refcount = 1;
3013 th->t_super = p_s_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003014
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003015 relock:
3016 lock_journal(p_s_sb);
3017 if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
3018 unlock_journal(p_s_sb);
3019 retval = journal->j_errno;
3020 goto out_fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003022 journal->j_bcount++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003023
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003024 if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
3025 unlock_journal(p_s_sb);
3026 reiserfs_wait_on_write_block(p_s_sb);
3027 PROC_INFO_INC(p_s_sb, journal.journal_relock_writers);
3028 goto relock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003029 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003030 now = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003031
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003032 /* if there is no room in the journal OR
3033 ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning
3034 ** we don't sleep if there aren't other writers
3035 */
3036
3037 if ((!join && journal->j_must_wait > 0) ||
3038 (!join
3039 && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
3040 || (!join && atomic_read(&journal->j_wcount) > 0
3041 && journal->j_trans_start_time > 0
3042 && (now - journal->j_trans_start_time) >
3043 journal->j_max_trans_age) || (!join
3044 && atomic_read(&journal->j_jlock))
3045 || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
3046
3047 old_trans_id = journal->j_trans_id;
3048 unlock_journal(p_s_sb); /* allow others to finish this transaction */
3049
3050 if (!join && (journal->j_len_alloc + nblocks + 2) >=
3051 journal->j_max_batch &&
3052 ((journal->j_len + nblocks + 2) * 100) <
3053 (journal->j_len_alloc * 75)) {
3054 if (atomic_read(&journal->j_wcount) > 10) {
3055 sched_count++;
3056 queue_log_writer(p_s_sb);
3057 goto relock;
3058 }
3059 }
3060 /* don't mess with joining the transaction if all we have to do is
3061 * wait for someone else to do a commit
3062 */
3063 if (atomic_read(&journal->j_jlock)) {
3064 while (journal->j_trans_id == old_trans_id &&
3065 atomic_read(&journal->j_jlock)) {
3066 queue_log_writer(p_s_sb);
3067 }
3068 goto relock;
3069 }
3070 retval = journal_join(&myth, p_s_sb, 1);
3071 if (retval)
3072 goto out_fail;
3073
3074 /* someone might have ended the transaction while we joined */
3075 if (old_trans_id != journal->j_trans_id) {
3076 retval = do_journal_end(&myth, p_s_sb, 1, 0);
3077 } else {
3078 retval = do_journal_end(&myth, p_s_sb, 1, COMMIT_NOW);
3079 }
3080
3081 if (retval)
3082 goto out_fail;
3083
3084 PROC_INFO_INC(p_s_sb, journal.journal_relock_wcount);
3085 goto relock;
3086 }
3087 /* we are the first writer, set trans_id */
3088 if (journal->j_trans_start_time == 0) {
3089 journal->j_trans_start_time = get_seconds();
3090 }
3091 atomic_inc(&(journal->j_wcount));
3092 journal->j_len_alloc += nblocks;
3093 th->t_blocks_logged = 0;
3094 th->t_blocks_allocated = nblocks;
3095 th->t_trans_id = journal->j_trans_id;
3096 unlock_journal(p_s_sb);
3097 INIT_LIST_HEAD(&th->t_list);
3098 get_fs_excl();
3099 return 0;
3100
3101 out_fail:
3102 memset(th, 0, sizeof(*th));
3103 /* Re-set th->t_super, so we can properly keep track of how many
3104 * persistent transactions there are. We need to do this so if this
3105 * call is part of a failed restart_transaction, we can free it later */
3106 th->t_super = p_s_sb;
3107 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003108}
3109
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003110struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3111 super_block
3112 *s,
3113 int nblocks)
3114{
3115 int ret;
3116 struct reiserfs_transaction_handle *th;
3117
3118 /* if we're nesting into an existing transaction. It will be
3119 ** persistent on its own
3120 */
3121 if (reiserfs_transaction_running(s)) {
3122 th = current->journal_info;
3123 th->t_refcount++;
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003124 BUG_ON(th->t_refcount < 2);
3125
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003126 return th;
3127 }
Pekka Enbergd739b422006-02-01 03:06:43 -08003128 th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003129 if (!th)
3130 return NULL;
3131 ret = journal_begin(th, s, nblocks);
3132 if (ret) {
Pekka Enbergd739b422006-02-01 03:06:43 -08003133 kfree(th);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003134 return NULL;
3135 }
3136
3137 SB_JOURNAL(s)->j_persistent_trans++;
3138 return th;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003139}
3140
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003141int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3142{
3143 struct super_block *s = th->t_super;
3144 int ret = 0;
3145 if (th->t_trans_id)
3146 ret = journal_end(th, th->t_super, th->t_blocks_allocated);
3147 else
3148 ret = -EIO;
3149 if (th->t_refcount == 0) {
3150 SB_JOURNAL(s)->j_persistent_trans--;
Pekka Enbergd739b422006-02-01 03:06:43 -08003151 kfree(th);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003152 }
3153 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003154}
3155
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003156static int journal_join(struct reiserfs_transaction_handle *th,
3157 struct super_block *p_s_sb, unsigned long nblocks)
3158{
3159 struct reiserfs_transaction_handle *cur_th = current->journal_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003160
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003161 /* this keeps do_journal_end from NULLing out the current->journal_info
3162 ** pointer
3163 */
3164 th->t_handle_save = cur_th;
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003165 BUG_ON(cur_th && cur_th->t_refcount > 1);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003166 return do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_JOIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003167}
3168
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003169int journal_join_abort(struct reiserfs_transaction_handle *th,
3170 struct super_block *p_s_sb, unsigned long nblocks)
3171{
3172 struct reiserfs_transaction_handle *cur_th = current->journal_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003173
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003174 /* this keeps do_journal_end from NULLing out the current->journal_info
3175 ** pointer
3176 */
3177 th->t_handle_save = cur_th;
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003178 BUG_ON(cur_th && cur_th->t_refcount > 1);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003179 return do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_ABORT);
3180}
3181
3182int journal_begin(struct reiserfs_transaction_handle *th,
3183 struct super_block *p_s_sb, unsigned long nblocks)
3184{
3185 struct reiserfs_transaction_handle *cur_th = current->journal_info;
3186 int ret;
3187
3188 th->t_handle_save = NULL;
3189 if (cur_th) {
3190 /* we are nesting into the current transaction */
3191 if (cur_th->t_super == p_s_sb) {
3192 BUG_ON(!cur_th->t_refcount);
3193 cur_th->t_refcount++;
3194 memcpy(th, cur_th, sizeof(*th));
3195 if (th->t_refcount <= 1)
3196 reiserfs_warning(p_s_sb,
3197 "BAD: refcount <= 1, but journal_info != 0");
3198 return 0;
3199 } else {
3200 /* we've ended up with a handle from a different filesystem.
3201 ** save it and restore on journal_end. This should never
3202 ** really happen...
3203 */
3204 reiserfs_warning(p_s_sb,
3205 "clm-2100: nesting info a different FS");
3206 th->t_handle_save = current->journal_info;
3207 current->journal_info = th;
3208 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003209 } else {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003210 current->journal_info = th;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003211 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003212 ret = do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_REG);
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003213 BUG_ON(current->journal_info != th);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003214
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003215 /* I guess this boils down to being the reciprocal of clm-2100 above.
3216 * If do_journal_begin_r fails, we need to put it back, since journal_end
3217 * won't be called to do it. */
3218 if (ret)
3219 current->journal_info = th->t_handle_save;
3220 else
3221 BUG_ON(!th->t_refcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003222
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003223 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003224}
3225
3226/*
3227** puts bh into the current transaction. If it was already there, reorders removes the
3228** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order).
3229**
3230** if it was dirty, cleans and files onto the clean list. I can't let it be dirty again until the
3231** transaction is committed.
3232**
3233** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3234*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003235int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3236 struct super_block *p_s_sb, struct buffer_head *bh)
3237{
3238 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3239 struct reiserfs_journal_cnode *cn = NULL;
3240 int count_already_incd = 0;
3241 int prepared = 0;
3242 BUG_ON(!th->t_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003243
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003244 PROC_INFO_INC(p_s_sb, journal.mark_dirty);
3245 if (th->t_trans_id != journal->j_trans_id) {
3246 reiserfs_panic(th->t_super,
3247 "journal-1577: handle trans id %ld != current trans id %ld\n",
3248 th->t_trans_id, journal->j_trans_id);
3249 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003250
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003251 p_s_sb->s_dirt = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003252
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003253 prepared = test_clear_buffer_journal_prepared(bh);
3254 clear_buffer_journal_restore_dirty(bh);
3255 /* already in this transaction, we are done */
3256 if (buffer_journaled(bh)) {
3257 PROC_INFO_INC(p_s_sb, journal.mark_dirty_already);
3258 return 0;
3259 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003260
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003261 /* this must be turned into a panic instead of a warning. We can't allow
3262 ** a dirty or journal_dirty or locked buffer to be logged, as some changes
3263 ** could get to disk too early. NOT GOOD.
3264 */
3265 if (!prepared || buffer_dirty(bh)) {
3266 reiserfs_warning(p_s_sb, "journal-1777: buffer %llu bad state "
3267 "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3268 (unsigned long long)bh->b_blocknr,
3269 prepared ? ' ' : '!',
3270 buffer_locked(bh) ? ' ' : '!',
3271 buffer_dirty(bh) ? ' ' : '!',
3272 buffer_journal_dirty(bh) ? ' ' : '!');
3273 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003274
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003275 if (atomic_read(&(journal->j_wcount)) <= 0) {
3276 reiserfs_warning(p_s_sb,
3277 "journal-1409: journal_mark_dirty returning because j_wcount was %d",
3278 atomic_read(&(journal->j_wcount)));
3279 return 1;
3280 }
3281 /* this error means I've screwed up, and we've overflowed the transaction.
3282 ** Nothing can be done here, except make the FS readonly or panic.
3283 */
3284 if (journal->j_len >= journal->j_trans_max) {
3285 reiserfs_panic(th->t_super,
3286 "journal-1413: journal_mark_dirty: j_len (%lu) is too big\n",
3287 journal->j_len);
3288 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003289
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003290 if (buffer_journal_dirty(bh)) {
3291 count_already_incd = 1;
3292 PROC_INFO_INC(p_s_sb, journal.mark_dirty_notjournal);
3293 clear_buffer_journal_dirty(bh);
3294 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003295
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003296 if (journal->j_len > journal->j_len_alloc) {
3297 journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3298 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003299
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003300 set_buffer_journaled(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003301
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003302 /* now put this guy on the end */
3303 if (!cn) {
3304 cn = get_cnode(p_s_sb);
3305 if (!cn) {
3306 reiserfs_panic(p_s_sb, "get_cnode failed!\n");
3307 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003308
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003309 if (th->t_blocks_logged == th->t_blocks_allocated) {
3310 th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3311 journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3312 }
3313 th->t_blocks_logged++;
3314 journal->j_len++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003315
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003316 cn->bh = bh;
3317 cn->blocknr = bh->b_blocknr;
3318 cn->sb = p_s_sb;
3319 cn->jlist = NULL;
3320 insert_journal_hash(journal->j_hash_table, cn);
3321 if (!count_already_incd) {
3322 get_bh(bh);
3323 }
3324 }
3325 cn->next = NULL;
3326 cn->prev = journal->j_last;
3327 cn->bh = bh;
3328 if (journal->j_last) {
3329 journal->j_last->next = cn;
3330 journal->j_last = cn;
3331 } else {
3332 journal->j_first = cn;
3333 journal->j_last = cn;
3334 }
3335 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003336}
3337
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003338int journal_end(struct reiserfs_transaction_handle *th,
3339 struct super_block *p_s_sb, unsigned long nblocks)
3340{
3341 if (!current->journal_info && th->t_refcount > 1)
3342 reiserfs_warning(p_s_sb, "REISER-NESTING: th NULL, refcount %d",
3343 th->t_refcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003344
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003345 if (!th->t_trans_id) {
3346 WARN_ON(1);
3347 return -EIO;
3348 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003349
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003350 th->t_refcount--;
3351 if (th->t_refcount > 0) {
3352 struct reiserfs_transaction_handle *cur_th =
3353 current->journal_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003354
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003355 /* we aren't allowed to close a nested transaction on a different
3356 ** filesystem from the one in the task struct
3357 */
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003358 BUG_ON(cur_th->t_super != th->t_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003359
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003360 if (th != cur_th) {
3361 memcpy(current->journal_info, th, sizeof(*th));
3362 th->t_trans_id = 0;
3363 }
3364 return 0;
3365 } else {
3366 return do_journal_end(th, p_s_sb, nblocks, 0);
3367 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003368}
3369
3370/* removes from the current transaction, relsing and descrementing any counters.
3371** also files the removed buffer directly onto the clean list
3372**
3373** called by journal_mark_freed when a block has been deleted
3374**
3375** returns 1 if it cleaned and relsed the buffer. 0 otherwise
3376*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003377static int remove_from_transaction(struct super_block *p_s_sb,
3378 b_blocknr_t blocknr, int already_cleaned)
3379{
3380 struct buffer_head *bh;
3381 struct reiserfs_journal_cnode *cn;
3382 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3383 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003384
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003385 cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, blocknr);
3386 if (!cn || !cn->bh) {
3387 return ret;
3388 }
3389 bh = cn->bh;
3390 if (cn->prev) {
3391 cn->prev->next = cn->next;
3392 }
3393 if (cn->next) {
3394 cn->next->prev = cn->prev;
3395 }
3396 if (cn == journal->j_first) {
3397 journal->j_first = cn->next;
3398 }
3399 if (cn == journal->j_last) {
3400 journal->j_last = cn->prev;
3401 }
3402 if (bh)
3403 remove_journal_hash(p_s_sb, journal->j_hash_table, NULL,
3404 bh->b_blocknr, 0);
3405 clear_buffer_journaled(bh); /* don't log this one */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003406
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003407 if (!already_cleaned) {
3408 clear_buffer_journal_dirty(bh);
3409 clear_buffer_dirty(bh);
3410 clear_buffer_journal_test(bh);
3411 put_bh(bh);
3412 if (atomic_read(&(bh->b_count)) < 0) {
3413 reiserfs_warning(p_s_sb,
3414 "journal-1752: remove from trans, b_count < 0");
3415 }
3416 ret = 1;
3417 }
3418 journal->j_len--;
3419 journal->j_len_alloc--;
3420 free_cnode(p_s_sb, cn);
3421 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003422}
3423
3424/*
3425** for any cnode in a journal list, it can only be dirtied of all the
Matt LaPlante0779bf22006-11-30 05:24:39 +01003426** transactions that include it are committed to disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003427** this checks through each transaction, and returns 1 if you are allowed to dirty,
3428** and 0 if you aren't
3429**
3430** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log
3431** blocks for a given transaction on disk
3432**
3433*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003434static int can_dirty(struct reiserfs_journal_cnode *cn)
3435{
3436 struct super_block *sb = cn->sb;
3437 b_blocknr_t blocknr = cn->blocknr;
3438 struct reiserfs_journal_cnode *cur = cn->hprev;
3439 int can_dirty = 1;
3440
3441 /* first test hprev. These are all newer than cn, so any node here
3442 ** with the same block number and dev means this node can't be sent
3443 ** to disk right now.
3444 */
3445 while (cur && can_dirty) {
3446 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3447 cur->blocknr == blocknr) {
3448 can_dirty = 0;
3449 }
3450 cur = cur->hprev;
3451 }
3452 /* then test hnext. These are all older than cn. As long as they
3453 ** are committed to the log, it is safe to write cn to disk
3454 */
3455 cur = cn->hnext;
3456 while (cur && can_dirty) {
3457 if (cur->jlist && cur->jlist->j_len > 0 &&
3458 atomic_read(&(cur->jlist->j_commit_left)) > 0 && cur->bh &&
3459 cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3460 can_dirty = 0;
3461 }
3462 cur = cur->hnext;
3463 }
3464 return can_dirty;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003465}
3466
3467/* syncs the commit blocks, but does not force the real buffers to disk
Matt LaPlante0779bf22006-11-30 05:24:39 +01003468** will wait until the current transaction is done/committed before returning
Linus Torvalds1da177e2005-04-16 15:20:36 -07003469*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003470int journal_end_sync(struct reiserfs_transaction_handle *th,
3471 struct super_block *p_s_sb, unsigned long nblocks)
3472{
3473 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003474
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003475 BUG_ON(!th->t_trans_id);
3476 /* you can sync while nested, very, very bad */
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003477 BUG_ON(th->t_refcount > 1);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003478 if (journal->j_len == 0) {
3479 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb),
3480 1);
3481 journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb));
3482 }
3483 return do_journal_end(th, p_s_sb, nblocks, COMMIT_NOW | WAIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003484}
3485
3486/*
3487** writeback the pending async commits to disk
3488*/
David Howellsc4028952006-11-22 14:57:56 +00003489static void flush_async_commits(struct work_struct *work)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003490{
David Howellsc4028952006-11-22 14:57:56 +00003491 struct reiserfs_journal *journal =
3492 container_of(work, struct reiserfs_journal, j_work.work);
3493 struct super_block *p_s_sb = journal->j_work_sb;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003494 struct reiserfs_journal_list *jl;
3495 struct list_head *entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003496
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003497 lock_kernel();
3498 if (!list_empty(&journal->j_journal_list)) {
3499 /* last entry is the youngest, commit it and you get everything */
3500 entry = journal->j_journal_list.prev;
3501 jl = JOURNAL_LIST_ENTRY(entry);
3502 flush_commit_list(p_s_sb, jl, 1);
3503 }
3504 unlock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003505}
3506
3507/*
3508** flushes any old transactions to disk
3509** ends the current transaction if it is too old
3510*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003511int reiserfs_flush_old_commits(struct super_block *p_s_sb)
3512{
3513 time_t now;
3514 struct reiserfs_transaction_handle th;
3515 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003516
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003517 now = get_seconds();
3518 /* safety check so we don't flush while we are replaying the log during
3519 * mount
3520 */
3521 if (list_empty(&journal->j_journal_list)) {
3522 return 0;
3523 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003524
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003525 /* check the current transaction. If there are no writers, and it is
3526 * too old, finish it, and force the commit blocks to disk
3527 */
3528 if (atomic_read(&journal->j_wcount) <= 0 &&
3529 journal->j_trans_start_time > 0 &&
3530 journal->j_len > 0 &&
3531 (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3532 if (!journal_join(&th, p_s_sb, 1)) {
3533 reiserfs_prepare_for_journal(p_s_sb,
3534 SB_BUFFER_WITH_SB(p_s_sb),
3535 1);
3536 journal_mark_dirty(&th, p_s_sb,
3537 SB_BUFFER_WITH_SB(p_s_sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003538
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003539 /* we're only being called from kreiserfsd, it makes no sense to do
3540 ** an async commit so that kreiserfsd can do it later
3541 */
3542 do_journal_end(&th, p_s_sb, 1, COMMIT_NOW | WAIT);
3543 }
3544 }
3545 return p_s_sb->s_dirt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003546}
3547
3548/*
3549** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit
3550**
3551** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all
3552** the writers are done. By the time it wakes up, the transaction it was called has already ended, so it just
3553** flushes the commit list and returns 0.
3554**
3555** Won't batch when flush or commit_now is set. Also won't batch when others are waiting on j_join_wait.
3556**
3557** Note, we can't allow the journal_end to proceed while there are still writers in the log.
3558*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003559static int check_journal_end(struct reiserfs_transaction_handle *th,
3560 struct super_block *p_s_sb, unsigned long nblocks,
3561 int flags)
3562{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003563
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003564 time_t now;
3565 int flush = flags & FLUSH_ALL;
3566 int commit_now = flags & COMMIT_NOW;
3567 int wait_on_commit = flags & WAIT;
3568 struct reiserfs_journal_list *jl;
3569 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003570
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003571 BUG_ON(!th->t_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003572
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003573 if (th->t_trans_id != journal->j_trans_id) {
3574 reiserfs_panic(th->t_super,
3575 "journal-1577: handle trans id %ld != current trans id %ld\n",
3576 th->t_trans_id, journal->j_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003577 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003578
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003579 journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3580 if (atomic_read(&(journal->j_wcount)) > 0) { /* <= 0 is allowed. unmounting might not call begin */
3581 atomic_dec(&(journal->j_wcount));
3582 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003583
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003584 /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released
3585 ** will be dealt with by next transaction that actually writes something, but should be taken
3586 ** care of in this trans
3587 */
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003588 BUG_ON(journal->j_len == 0);
3589
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003590 /* if wcount > 0, and we are called to with flush or commit_now,
3591 ** we wait on j_join_wait. We will wake up when the last writer has
3592 ** finished the transaction, and started it on its way to the disk.
3593 ** Then, we flush the commit or journal list, and just return 0
3594 ** because the rest of journal end was already done for this transaction.
3595 */
3596 if (atomic_read(&(journal->j_wcount)) > 0) {
3597 if (flush || commit_now) {
3598 unsigned trans_id;
3599
3600 jl = journal->j_current_jl;
3601 trans_id = jl->j_trans_id;
3602 if (wait_on_commit)
3603 jl->j_state |= LIST_COMMIT_PENDING;
3604 atomic_set(&(journal->j_jlock), 1);
3605 if (flush) {
3606 journal->j_next_full_flush = 1;
3607 }
3608 unlock_journal(p_s_sb);
3609
3610 /* sleep while the current transaction is still j_jlocked */
3611 while (journal->j_trans_id == trans_id) {
3612 if (atomic_read(&journal->j_jlock)) {
3613 queue_log_writer(p_s_sb);
3614 } else {
3615 lock_journal(p_s_sb);
3616 if (journal->j_trans_id == trans_id) {
3617 atomic_set(&(journal->j_jlock),
3618 1);
3619 }
3620 unlock_journal(p_s_sb);
3621 }
3622 }
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003623 BUG_ON(journal->j_trans_id == trans_id);
3624
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003625 if (commit_now
3626 && journal_list_still_alive(p_s_sb, trans_id)
3627 && wait_on_commit) {
3628 flush_commit_list(p_s_sb, jl, 1);
3629 }
3630 return 0;
3631 }
3632 unlock_journal(p_s_sb);
3633 return 0;
3634 }
3635
3636 /* deal with old transactions where we are the last writers */
3637 now = get_seconds();
3638 if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3639 commit_now = 1;
3640 journal->j_next_async_flush = 1;
3641 }
3642 /* don't batch when someone is waiting on j_join_wait */
3643 /* don't batch when syncing the commit or flushing the whole trans */
3644 if (!(journal->j_must_wait > 0) && !(atomic_read(&(journal->j_jlock)))
3645 && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3646 && journal->j_len_alloc < journal->j_max_batch
3647 && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3648 journal->j_bcount++;
3649 unlock_journal(p_s_sb);
3650 return 0;
3651 }
3652
3653 if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
3654 reiserfs_panic(p_s_sb,
3655 "journal-003: journal_end: j_start (%ld) is too high\n",
3656 journal->j_start);
3657 }
3658 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003659}
3660
3661/*
3662** Does all the work that makes deleting blocks safe.
3663** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on.
3664**
3665** otherwise:
3666** set a bit for the block in the journal bitmap. That will prevent it from being allocated for unformatted nodes
3667** before this transaction has finished.
3668**
3669** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers. That will prevent any old transactions with
3670** this block from trying to flush to the real location. Since we aren't removing the cnode from the journal_list_hash,
3671** the block can't be reallocated yet.
3672**
3673** Then remove it from the current transaction, decrementing any counters and filing it on the clean list.
3674*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003675int journal_mark_freed(struct reiserfs_transaction_handle *th,
3676 struct super_block *p_s_sb, b_blocknr_t blocknr)
3677{
3678 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3679 struct reiserfs_journal_cnode *cn = NULL;
3680 struct buffer_head *bh = NULL;
3681 struct reiserfs_list_bitmap *jb = NULL;
3682 int cleaned = 0;
3683 BUG_ON(!th->t_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003684
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003685 cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, blocknr);
3686 if (cn && cn->bh) {
3687 bh = cn->bh;
3688 get_bh(bh);
3689 }
3690 /* if it is journal new, we just remove it from this transaction */
3691 if (bh && buffer_journal_new(bh)) {
3692 clear_buffer_journal_new(bh);
3693 clear_prepared_bits(bh);
3694 reiserfs_clean_and_file_buffer(bh);
3695 cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned);
3696 } else {
3697 /* set the bit for this block in the journal bitmap for this transaction */
3698 jb = journal->j_current_jl->j_list_bitmap;
3699 if (!jb) {
3700 reiserfs_panic(p_s_sb,
3701 "journal-1702: journal_mark_freed, journal_list_bitmap is NULL\n");
3702 }
3703 set_bit_in_list_bitmap(p_s_sb, blocknr, jb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003704
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003705 /* Note, the entire while loop is not allowed to schedule. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003706
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003707 if (bh) {
3708 clear_prepared_bits(bh);
3709 reiserfs_clean_and_file_buffer(bh);
3710 }
3711 cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003712
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003713 /* find all older transactions with this block, make sure they don't try to write it out */
3714 cn = get_journal_hash_dev(p_s_sb, journal->j_list_hash_table,
3715 blocknr);
3716 while (cn) {
3717 if (p_s_sb == cn->sb && blocknr == cn->blocknr) {
3718 set_bit(BLOCK_FREED, &cn->state);
3719 if (cn->bh) {
3720 if (!cleaned) {
3721 /* remove_from_transaction will brelse the buffer if it was
3722 ** in the current trans
3723 */
3724 clear_buffer_journal_dirty(cn->
3725 bh);
3726 clear_buffer_dirty(cn->bh);
3727 clear_buffer_journal_test(cn->
3728 bh);
3729 cleaned = 1;
3730 put_bh(cn->bh);
3731 if (atomic_read
3732 (&(cn->bh->b_count)) < 0) {
3733 reiserfs_warning(p_s_sb,
3734 "journal-2138: cn->bh->b_count < 0");
3735 }
3736 }
3737 if (cn->jlist) { /* since we are clearing the bh, we MUST dec nonzerolen */
3738 atomic_dec(&
3739 (cn->jlist->
3740 j_nonzerolen));
3741 }
3742 cn->bh = NULL;
3743 }
3744 }
3745 cn = cn->hnext;
3746 }
3747 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003748
Chris Mason398c95b2007-10-16 23:29:44 -07003749 if (bh)
3750 release_buffer_page(bh); /* get_hash grabs the buffer */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003751 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003752}
3753
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003754void reiserfs_update_inode_transaction(struct inode *inode)
3755{
3756 struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3757 REISERFS_I(inode)->i_jl = journal->j_current_jl;
3758 REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003759}
3760
3761/*
3762 * returns -1 on error, 0 if no commits/barriers were done and 1
3763 * if a transaction was actually committed and the barrier was done
3764 */
3765static int __commit_trans_jl(struct inode *inode, unsigned long id,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003766 struct reiserfs_journal_list *jl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003767{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003768 struct reiserfs_transaction_handle th;
3769 struct super_block *sb = inode->i_sb;
3770 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3771 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003772
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003773 /* is it from the current transaction, or from an unknown transaction? */
3774 if (id == journal->j_trans_id) {
3775 jl = journal->j_current_jl;
3776 /* try to let other writers come in and grow this transaction */
3777 let_transaction_grow(sb, id);
3778 if (journal->j_trans_id != id) {
3779 goto flush_commit_only;
3780 }
3781
3782 ret = journal_begin(&th, sb, 1);
3783 if (ret)
3784 return ret;
3785
3786 /* someone might have ended this transaction while we joined */
3787 if (journal->j_trans_id != id) {
3788 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3789 1);
3790 journal_mark_dirty(&th, sb, SB_BUFFER_WITH_SB(sb));
3791 ret = journal_end(&th, sb, 1);
3792 goto flush_commit_only;
3793 }
3794
3795 ret = journal_end_sync(&th, sb, 1);
3796 if (!ret)
3797 ret = 1;
3798
3799 } else {
3800 /* this gets tricky, we have to make sure the journal list in
3801 * the inode still exists. We know the list is still around
3802 * if we've got a larger transaction id than the oldest list
3803 */
3804 flush_commit_only:
3805 if (journal_list_still_alive(inode->i_sb, id)) {
3806 /*
3807 * we only set ret to 1 when we know for sure
3808 * the barrier hasn't been started yet on the commit
3809 * block.
3810 */
3811 if (atomic_read(&jl->j_commit_left) > 1)
3812 ret = 1;
3813 flush_commit_list(sb, jl, 1);
3814 if (journal->j_errno)
3815 ret = journal->j_errno;
3816 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003817 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003818 /* otherwise the list is gone, and long since committed */
3819 return ret;
3820}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003821
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003822int reiserfs_commit_for_inode(struct inode *inode)
3823{
3824 unsigned long id = REISERFS_I(inode)->i_trans_id;
3825 struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003826
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003827 /* for the whole inode, assume unset id means it was
3828 * changed in the current transaction. More conservative
Linus Torvalds1da177e2005-04-16 15:20:36 -07003829 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003830 if (!id || !jl) {
3831 reiserfs_update_inode_transaction(inode);
3832 id = REISERFS_I(inode)->i_trans_id;
3833 /* jl will be updated in __commit_trans_jl */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003834 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003835
3836 return __commit_trans_jl(inode, id, jl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003837}
3838
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003839void reiserfs_restore_prepared_buffer(struct super_block *p_s_sb,
3840 struct buffer_head *bh)
3841{
3842 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3843 PROC_INFO_INC(p_s_sb, journal.restore_prepared);
3844 if (!bh) {
3845 return;
3846 }
3847 if (test_clear_buffer_journal_restore_dirty(bh) &&
3848 buffer_journal_dirty(bh)) {
3849 struct reiserfs_journal_cnode *cn;
3850 cn = get_journal_hash_dev(p_s_sb,
3851 journal->j_list_hash_table,
3852 bh->b_blocknr);
3853 if (cn && can_dirty(cn)) {
3854 set_buffer_journal_test(bh);
3855 mark_buffer_dirty(bh);
3856 }
3857 }
3858 clear_buffer_journal_prepared(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003859}
3860
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003861extern struct tree_balance *cur_tb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003862/*
3863** before we can change a metadata block, we have to make sure it won't
3864** be written to disk while we are altering it. So, we must:
3865** clean it
3866** wait on it.
3867**
3868*/
3869int reiserfs_prepare_for_journal(struct super_block *p_s_sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003870 struct buffer_head *bh, int wait)
3871{
3872 PROC_INFO_INC(p_s_sb, journal.prepare);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003873
Nick Pigginca5de402008-08-02 12:02:13 +02003874 if (!trylock_buffer(bh)) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003875 if (!wait)
3876 return 0;
3877 lock_buffer(bh);
3878 }
3879 set_buffer_journal_prepared(bh);
3880 if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3881 clear_buffer_journal_test(bh);
3882 set_buffer_journal_restore_dirty(bh);
3883 }
3884 unlock_buffer(bh);
3885 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003886}
3887
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003888static void flush_old_journal_lists(struct super_block *s)
3889{
3890 struct reiserfs_journal *journal = SB_JOURNAL(s);
3891 struct reiserfs_journal_list *jl;
3892 struct list_head *entry;
3893 time_t now = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003894
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003895 while (!list_empty(&journal->j_journal_list)) {
3896 entry = journal->j_journal_list.next;
3897 jl = JOURNAL_LIST_ENTRY(entry);
3898 /* this check should always be run, to send old lists to disk */
Chris Masona3172022006-09-29 01:59:56 -07003899 if (jl->j_timestamp < (now - (JOURNAL_MAX_TRANS_AGE * 4)) &&
3900 atomic_read(&jl->j_commit_left) == 0 &&
3901 test_transaction(s, jl)) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003902 flush_used_journal_lists(s, jl);
3903 } else {
3904 break;
3905 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003906 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003907}
3908
3909/*
3910** long and ugly. If flush, will not return until all commit
3911** blocks and all real buffers in the trans are on disk.
3912** If no_async, won't return until all commit blocks are on disk.
3913**
3914** keep reading, there are comments as you go along
3915**
3916** If the journal is aborted, we just clean up. Things like flushing
3917** journal lists, etc just won't happen.
3918*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003919static int do_journal_end(struct reiserfs_transaction_handle *th,
3920 struct super_block *p_s_sb, unsigned long nblocks,
3921 int flags)
3922{
3923 struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3924 struct reiserfs_journal_cnode *cn, *next, *jl_cn;
3925 struct reiserfs_journal_cnode *last_cn = NULL;
3926 struct reiserfs_journal_desc *desc;
3927 struct reiserfs_journal_commit *commit;
3928 struct buffer_head *c_bh; /* commit bh */
3929 struct buffer_head *d_bh; /* desc bh */
3930 int cur_write_start = 0; /* start index of current log write */
3931 int old_start;
3932 int i;
Alexander Zarochentseva44c94a2006-03-25 03:06:59 -08003933 int flush;
3934 int wait_on_commit;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003935 struct reiserfs_journal_list *jl, *temp_jl;
3936 struct list_head *entry, *safe;
3937 unsigned long jindex;
3938 unsigned long commit_trans_id;
3939 int trans_half;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003940
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003941 BUG_ON(th->t_refcount > 1);
3942 BUG_ON(!th->t_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003943
Alexander Zarochentseva44c94a2006-03-25 03:06:59 -08003944 /* protect flush_older_commits from doing mistakes if the
3945 transaction ID counter gets overflowed. */
3946 if (th->t_trans_id == ~0UL)
3947 flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
3948 flush = flags & FLUSH_ALL;
3949 wait_on_commit = flags & WAIT;
3950
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003951 put_fs_excl();
3952 current->journal_info = th->t_handle_save;
3953 reiserfs_check_lock_depth(p_s_sb, "journal end");
3954 if (journal->j_len == 0) {
3955 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb),
3956 1);
3957 journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb));
3958 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003959
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003960 lock_journal(p_s_sb);
3961 if (journal->j_next_full_flush) {
3962 flags |= FLUSH_ALL;
3963 flush = 1;
3964 }
3965 if (journal->j_next_async_flush) {
3966 flags |= COMMIT_NOW | WAIT;
3967 wait_on_commit = 1;
3968 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003969
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003970 /* check_journal_end locks the journal, and unlocks if it does not return 1
3971 ** it tells us if we should continue with the journal_end, or just return
3972 */
3973 if (!check_journal_end(th, p_s_sb, nblocks, flags)) {
3974 p_s_sb->s_dirt = 1;
3975 wake_queued_writers(p_s_sb);
3976 reiserfs_async_progress_wait(p_s_sb);
3977 goto out;
3978 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003979
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003980 /* check_journal_end might set these, check again */
3981 if (journal->j_next_full_flush) {
3982 flush = 1;
3983 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003984
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003985 /*
3986 ** j must wait means we have to flush the log blocks, and the real blocks for
3987 ** this transaction
3988 */
3989 if (journal->j_must_wait > 0) {
3990 flush = 1;
3991 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003992#ifdef REISERFS_PREALLOCATE
Jan Karaef43bc42006-01-11 12:17:40 -08003993 /* quota ops might need to nest, setup the journal_info pointer for them
3994 * and raise the refcount so that it is > 0. */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003995 current->journal_info = th;
Jan Karaef43bc42006-01-11 12:17:40 -08003996 th->t_refcount++;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003997 reiserfs_discard_all_prealloc(th); /* it should not involve new blocks into
3998 * the transaction */
Jan Karaef43bc42006-01-11 12:17:40 -08003999 th->t_refcount--;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004000 current->journal_info = th->t_handle_save;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004001#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004002
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004003 /* setup description block */
4004 d_bh =
4005 journal_getblk(p_s_sb,
4006 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
4007 journal->j_start);
4008 set_buffer_uptodate(d_bh);
4009 desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
4010 memset(d_bh->b_data, 0, d_bh->b_size);
4011 memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
4012 set_desc_trans_id(desc, journal->j_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004013
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004014 /* setup commit block. Don't write (keep it clean too) this one until after everyone else is written */
4015 c_bh = journal_getblk(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
4016 ((journal->j_start + journal->j_len +
4017 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
4018 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
4019 memset(c_bh->b_data, 0, c_bh->b_size);
4020 set_commit_trans_id(commit, journal->j_trans_id);
4021 set_buffer_uptodate(c_bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004022
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004023 /* init this journal list */
4024 jl = journal->j_current_jl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004025
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004026 /* we lock the commit before doing anything because
4027 * we want to make sure nobody tries to run flush_commit_list until
4028 * the new transaction is fully setup, and we've already flushed the
4029 * ordered bh list
4030 */
Jeff Mahoney90415de2008-07-25 01:46:40 -07004031 mutex_lock(&jl->j_commit_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004032
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004033 /* save the transaction id in case we need to commit it later */
4034 commit_trans_id = jl->j_trans_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004035
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004036 atomic_set(&jl->j_older_commits_done, 0);
4037 jl->j_trans_id = journal->j_trans_id;
4038 jl->j_timestamp = journal->j_trans_start_time;
4039 jl->j_commit_bh = c_bh;
4040 jl->j_start = journal->j_start;
4041 jl->j_len = journal->j_len;
4042 atomic_set(&jl->j_nonzerolen, journal->j_len);
4043 atomic_set(&jl->j_commit_left, journal->j_len + 2);
4044 jl->j_realblock = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004045
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004046 /* The ENTIRE FOR LOOP MUST not cause schedule to occur.
4047 ** for each real block, add it to the journal list hash,
4048 ** copy into real block index array in the commit or desc block
4049 */
4050 trans_half = journal_trans_half(p_s_sb->s_blocksize);
4051 for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
4052 if (buffer_journaled(cn->bh)) {
4053 jl_cn = get_cnode(p_s_sb);
4054 if (!jl_cn) {
4055 reiserfs_panic(p_s_sb,
4056 "journal-1676, get_cnode returned NULL\n");
4057 }
4058 if (i == 0) {
4059 jl->j_realblock = jl_cn;
4060 }
4061 jl_cn->prev = last_cn;
4062 jl_cn->next = NULL;
4063 if (last_cn) {
4064 last_cn->next = jl_cn;
4065 }
4066 last_cn = jl_cn;
4067 /* make sure the block we are trying to log is not a block
4068 of journal or reserved area */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004069
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004070 if (is_block_in_log_or_reserved_area
4071 (p_s_sb, cn->bh->b_blocknr)) {
4072 reiserfs_panic(p_s_sb,
4073 "journal-2332: Trying to log block %lu, which is a log block\n",
4074 cn->bh->b_blocknr);
4075 }
4076 jl_cn->blocknr = cn->bh->b_blocknr;
4077 jl_cn->state = 0;
4078 jl_cn->sb = p_s_sb;
4079 jl_cn->bh = cn->bh;
4080 jl_cn->jlist = jl;
4081 insert_journal_hash(journal->j_list_hash_table, jl_cn);
4082 if (i < trans_half) {
4083 desc->j_realblock[i] =
4084 cpu_to_le32(cn->bh->b_blocknr);
4085 } else {
4086 commit->j_realblock[i - trans_half] =
4087 cpu_to_le32(cn->bh->b_blocknr);
4088 }
4089 } else {
4090 i--;
4091 }
4092 }
4093 set_desc_trans_len(desc, journal->j_len);
4094 set_desc_mount_id(desc, journal->j_mount_id);
4095 set_desc_trans_id(desc, journal->j_trans_id);
4096 set_commit_trans_len(commit, journal->j_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004097
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004098 /* special check in case all buffers in the journal were marked for not logging */
Eric Sesterhenn14a61442006-10-03 23:36:38 +02004099 BUG_ON(journal->j_len == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004100
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004101 /* we're about to dirty all the log blocks, mark the description block
4102 * dirty now too. Don't mark the commit block dirty until all the
4103 * others are on disk
4104 */
4105 mark_buffer_dirty(d_bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004106
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004107 /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */
4108 cur_write_start = journal->j_start;
4109 cn = journal->j_first;
4110 jindex = 1; /* start at one so we don't get the desc again */
4111 while (cn) {
4112 clear_buffer_journal_new(cn->bh);
4113 /* copy all the real blocks into log area. dirty log blocks */
4114 if (buffer_journaled(cn->bh)) {
4115 struct buffer_head *tmp_bh;
4116 char *addr;
4117 struct page *page;
4118 tmp_bh =
4119 journal_getblk(p_s_sb,
4120 SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
4121 ((cur_write_start +
4122 jindex) %
4123 SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
4124 set_buffer_uptodate(tmp_bh);
4125 page = cn->bh->b_page;
4126 addr = kmap(page);
4127 memcpy(tmp_bh->b_data,
4128 addr + offset_in_page(cn->bh->b_data),
4129 cn->bh->b_size);
4130 kunmap(page);
4131 mark_buffer_dirty(tmp_bh);
4132 jindex++;
4133 set_buffer_journal_dirty(cn->bh);
4134 clear_buffer_journaled(cn->bh);
4135 } else {
4136 /* JDirty cleared sometime during transaction. don't log this one */
4137 reiserfs_warning(p_s_sb,
4138 "journal-2048: do_journal_end: BAD, buffer in journal hash, but not JDirty!");
4139 brelse(cn->bh);
4140 }
4141 next = cn->next;
4142 free_cnode(p_s_sb, cn);
4143 cn = next;
4144 cond_resched();
4145 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004146
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004147 /* we are done with both the c_bh and d_bh, but
4148 ** c_bh must be written after all other commit blocks,
4149 ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4150 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004151
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004152 journal->j_current_jl = alloc_journal_list(p_s_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004153
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004154 /* now it is safe to insert this transaction on the main list */
4155 list_add_tail(&jl->j_list, &journal->j_journal_list);
4156 list_add_tail(&jl->j_working_list, &journal->j_working_list);
4157 journal->j_num_work_lists++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004158
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004159 /* reset journal values for the next transaction */
4160 old_start = journal->j_start;
4161 journal->j_start =
4162 (journal->j_start + journal->j_len +
4163 2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb);
4164 atomic_set(&(journal->j_wcount), 0);
4165 journal->j_bcount = 0;
4166 journal->j_last = NULL;
4167 journal->j_first = NULL;
4168 journal->j_len = 0;
4169 journal->j_trans_start_time = 0;
Alexander Zarochentseva44c94a2006-03-25 03:06:59 -08004170 /* check for trans_id overflow */
4171 if (++journal->j_trans_id == 0)
4172 journal->j_trans_id = 10;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004173 journal->j_current_jl->j_trans_id = journal->j_trans_id;
4174 journal->j_must_wait = 0;
4175 journal->j_len_alloc = 0;
4176 journal->j_next_full_flush = 0;
4177 journal->j_next_async_flush = 0;
4178 init_journal_hash(p_s_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004179
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004180 // make sure reiserfs_add_jh sees the new current_jl before we
4181 // write out the tails
4182 smp_mb();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004183
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004184 /* tail conversion targets have to hit the disk before we end the
4185 * transaction. Otherwise a later transaction might repack the tail
4186 * before this transaction commits, leaving the data block unflushed and
4187 * clean, if we crash before the later transaction commits, the data block
4188 * is lost.
4189 */
4190 if (!list_empty(&jl->j_tail_bh_list)) {
4191 unlock_kernel();
4192 write_ordered_buffers(&journal->j_dirty_buffers_lock,
4193 journal, jl, &jl->j_tail_bh_list);
4194 lock_kernel();
4195 }
Eric Sesterhenn14a61442006-10-03 23:36:38 +02004196 BUG_ON(!list_empty(&jl->j_tail_bh_list));
Jeff Mahoney90415de2008-07-25 01:46:40 -07004197 mutex_unlock(&jl->j_commit_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004198
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004199 /* honor the flush wishes from the caller, simple commits can
4200 ** be done outside the journal lock, they are done below
4201 **
4202 ** if we don't flush the commit list right now, we put it into
4203 ** the work queue so the people waiting on the async progress work
4204 ** queue don't wait for this proc to flush journal lists and such.
4205 */
4206 if (flush) {
4207 flush_commit_list(p_s_sb, jl, 1);
4208 flush_journal_list(p_s_sb, jl, 1);
4209 } else if (!(jl->j_state & LIST_COMMIT_PENDING))
4210 queue_delayed_work(commit_wq, &journal->j_work, HZ / 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004211
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004212 /* if the next transaction has any chance of wrapping, flush
4213 ** transactions that might get overwritten. If any journal lists are very
4214 ** old flush them as well.
4215 */
4216 first_jl:
4217 list_for_each_safe(entry, safe, &journal->j_journal_list) {
4218 temp_jl = JOURNAL_LIST_ENTRY(entry);
4219 if (journal->j_start <= temp_jl->j_start) {
4220 if ((journal->j_start + journal->j_trans_max + 1) >=
4221 temp_jl->j_start) {
4222 flush_used_journal_lists(p_s_sb, temp_jl);
4223 goto first_jl;
4224 } else if ((journal->j_start +
4225 journal->j_trans_max + 1) <
4226 SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
4227 /* if we don't cross into the next transaction and we don't
4228 * wrap, there is no way we can overlap any later transactions
4229 * break now
4230 */
4231 break;
4232 }
4233 } else if ((journal->j_start +
4234 journal->j_trans_max + 1) >
4235 SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
4236 if (((journal->j_start + journal->j_trans_max + 1) %
4237 SB_ONDISK_JOURNAL_SIZE(p_s_sb)) >=
4238 temp_jl->j_start) {
4239 flush_used_journal_lists(p_s_sb, temp_jl);
4240 goto first_jl;
4241 } else {
4242 /* we don't overlap anything from out start to the end of the
4243 * log, and our wrapped portion doesn't overlap anything at
4244 * the start of the log. We can break
4245 */
4246 break;
4247 }
4248 }
4249 }
4250 flush_old_journal_lists(p_s_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004251
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004252 journal->j_current_jl->j_list_bitmap =
4253 get_list_bitmap(p_s_sb, journal->j_current_jl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004254
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004255 if (!(journal->j_current_jl->j_list_bitmap)) {
4256 reiserfs_panic(p_s_sb,
4257 "journal-1996: do_journal_end, could not get a list bitmap\n");
4258 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004259
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004260 atomic_set(&(journal->j_jlock), 0);
4261 unlock_journal(p_s_sb);
4262 /* wake up any body waiting to join. */
4263 clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4264 wake_up(&(journal->j_join_wait));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004265
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004266 if (!flush && wait_on_commit &&
4267 journal_list_still_alive(p_s_sb, commit_trans_id)) {
4268 flush_commit_list(p_s_sb, jl, 1);
4269 }
4270 out:
4271 reiserfs_check_lock_depth(p_s_sb, "journal end2");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004272
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004273 memset(th, 0, sizeof(*th));
4274 /* Re-set th->t_super, so we can properly keep track of how many
4275 * persistent transactions there are. We need to do this so if this
4276 * call is part of a failed restart_transaction, we can free it later */
4277 th->t_super = p_s_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004278
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004279 return journal->j_errno;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004280}
4281
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004282static void __reiserfs_journal_abort_hard(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004283{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004284 struct reiserfs_journal *journal = SB_JOURNAL(sb);
4285 if (test_bit(J_ABORTED, &journal->j_state))
4286 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004287
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004288 printk(KERN_CRIT "REISERFS: Aborting journal for filesystem on %s\n",
4289 reiserfs_bdevname(sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004290
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004291 sb->s_flags |= MS_RDONLY;
4292 set_bit(J_ABORTED, &journal->j_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004293
4294#ifdef CONFIG_REISERFS_CHECK
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004295 dump_stack();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004296#endif
4297}
4298
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004299static void __reiserfs_journal_abort_soft(struct super_block *sb, int errno)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004300{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004301 struct reiserfs_journal *journal = SB_JOURNAL(sb);
4302 if (test_bit(J_ABORTED, &journal->j_state))
4303 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004304
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004305 if (!journal->j_errno)
4306 journal->j_errno = errno;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004307
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004308 __reiserfs_journal_abort_hard(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004309}
4310
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004311void reiserfs_journal_abort(struct super_block *sb, int errno)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004312{
Harvey Harrisone13601b2008-04-28 02:16:21 -07004313 __reiserfs_journal_abort_soft(sb, errno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004314}