blob: ba98546fabbd1921beedc54074f7399115f14f54 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2** Write ahead logging implementation copyright Chris Mason 2000
3**
Jeff Mahoney0222e652009-03-30 14:02:44 -04004** The background commits make this code very interelated, and
Linus Torvalds1da177e2005-04-16 15:20:36 -07005** overly complex. I need to rethink things a bit....The major players:
6**
Jeff Mahoney0222e652009-03-30 14:02:44 -04007** journal_begin -- call with the number of blocks you expect to log.
Linus Torvalds1da177e2005-04-16 15:20:36 -07008** If the current transaction is too
Jeff Mahoney0222e652009-03-30 14:02:44 -04009** old, it will block until the current transaction is
Linus Torvalds1da177e2005-04-16 15:20:36 -070010** finished, and then start a new one.
Jeff Mahoney0222e652009-03-30 14:02:44 -040011** Usually, your transaction will get joined in with
Linus Torvalds1da177e2005-04-16 15:20:36 -070012** previous ones for speed.
13**
Jeff Mahoney0222e652009-03-30 14:02:44 -040014** journal_join -- same as journal_begin, but won't block on the current
Linus Torvalds1da177e2005-04-16 15:20:36 -070015** transaction regardless of age. Don't ever call
Jeff Mahoney0222e652009-03-30 14:02:44 -040016** this. Ever. There are only two places it should be
Linus Torvalds1da177e2005-04-16 15:20:36 -070017** called from, and they are both inside this file.
18**
Jeff Mahoney0222e652009-03-30 14:02:44 -040019** journal_mark_dirty -- adds blocks into this transaction. clears any flags
Linus Torvalds1da177e2005-04-16 15:20:36 -070020** that might make them get sent to disk
Jeff Mahoney0222e652009-03-30 14:02:44 -040021** and then marks them BH_JDirty. Puts the buffer head
22** into the current transaction hash.
Linus Torvalds1da177e2005-04-16 15:20:36 -070023**
24** journal_end -- if the current transaction is batchable, it does nothing
25** otherwise, it could do an async/synchronous commit, or
Jeff Mahoney0222e652009-03-30 14:02:44 -040026** a full flush of all log and real blocks in the
Linus Torvalds1da177e2005-04-16 15:20:36 -070027** transaction.
28**
Jeff Mahoney0222e652009-03-30 14:02:44 -040029** 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
Linus Torvalds1da177e2005-04-16 15:20:36 -070032** around too long.
Jeff Mahoney0222e652009-03-30 14:02:44 -040033** -- Note, if you call this as an immediate flush from
Linus Torvalds1da177e2005-04-16 15:20:36 -070034** 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,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400100 struct super_block *sb, unsigned long nblocks);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700101static 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,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400116 struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700117 unsigned long nblocks, int join);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400119static void init_journal_hash(struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700120{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400121 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700122 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
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400148 *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700149{
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 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400157 bn->data = kzalloc(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
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400167static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700168{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400169 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700170 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);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400179 memset(bn->data, 0, sb->s_blocksize);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700180 journal->j_free_bitmap_nodes--;
181 return bn;
182 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400183 bn = allocate_bitmap_node(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700184 if (!bn) {
185 yield();
186 goto repeat;
187 }
188 return bn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189}
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400190static inline void free_bitmap_node(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700191 struct reiserfs_bitmap_node *bn)
192{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400193 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700194 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
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400204static void allocate_bitmap_nodes(struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700205{
206 int i;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400207 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700208 struct reiserfs_bitmap_node *bn = NULL;
209 for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400210 bn = allocate_bitmap_node(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700211 if (bn) {
212 list_add(&bn->list, &journal->j_bitmap_nodes);
213 journal->j_free_bitmap_nodes++;
214 } else {
Jeff Mahoney0222e652009-03-30 14:02:44 -0400215 break; /* this is ok, we'll try again when more are needed */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700216 }
217 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218}
219
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400220static int set_bit_in_list_bitmap(struct super_block *sb,
Jeff Mahoney3ee16672007-10-18 23:39:25 -0700221 b_blocknr_t block,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700222 struct reiserfs_list_bitmap *jb)
223{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400224 unsigned int bmap_nr = block / (sb->s_blocksize << 3);
225 unsigned int bit_nr = block % (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]) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400228 jb->bitmaps[bmap_nr] = get_bitmap_node(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700229 }
230 set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
231 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232}
233
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400234static void cleanup_bitmap_list(struct super_block *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 Mahoneya9dd3642009-03-30 14:02:45 -0400241 for (i = 0; i < reiserfs_bmap_count(sb); i++) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700242 if (jb->bitmaps[i]) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400243 free_bitmap_node(sb, jb->bitmaps[i]);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700244 jb->bitmaps[i] = NULL;
245 }
246 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247}
248
249/*
250** only call this on FS unmount.
251*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400252static int free_list_bitmaps(struct super_block *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;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400260 cleanup_bitmap_list(sb, jb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700261 vfree(jb->bitmaps);
262 jb->bitmaps = NULL;
263 }
264 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265}
266
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400267static int free_bitmap_nodes(struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700268{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400269 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700270 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/*
Jeff Mahoney0222e652009-03-30 14:02:44 -0400286** get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287** jb_array is the array to be filled in.
288*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400289int reiserfs_allocate_list_bitmaps(struct super_block *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) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400303 reiserfs_warning(sb, "clm-2000", "unable to "
Jeff Mahoney45b03d52009-03-30 14:02:21 -0400304 "allocate bitmaps for journal lists");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700305 failed = 1;
306 break;
307 }
308 memset(jb->bitmaps, 0, mem);
309 }
310 if (failed) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400311 free_list_bitmaps(sb, jb_array);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700312 return -1;
313 }
314 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315}
316
317/*
Jeff Mahoney0222e652009-03-30 14:02:44 -0400318** find an available list bitmap. If you can't find one, flush a commit list
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319** and try again
320*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400321static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700322 struct reiserfs_journal_list
323 *jl)
324{
325 int i, j;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400326 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700327 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) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400334 flush_commit_list(sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700335 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
Jeff Mahoney0222e652009-03-30 14:02:44 -0400351/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352** 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/*
Jeff Mahoney0222e652009-03-30 14:02:44 -0400379** pulls a cnode off the free list, or returns NULL on failure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400381static struct reiserfs_journal_cnode *get_cnode(struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700382{
383 struct reiserfs_journal_cnode *cn;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400384 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400386 reiserfs_check_lock_depth(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/*
Jeff Mahoney0222e652009-03-30 14:02:44 -0400406** returns a cnode to the free list
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400408static void free_cnode(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700409 struct reiserfs_journal_cnode *cn)
410{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400411 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400413 reiserfs_check_lock_depth(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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432/* return a cnode with same dev, block number and size in table, or null if not found */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700433static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
434 super_block
435 *sb,
436 struct
437 reiserfs_journal_cnode
438 **table,
439 long bl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700441 struct reiserfs_journal_cnode *cn;
442 cn = journal_hash(table, sb, bl);
443 while (cn) {
444 if (cn->blocknr == bl && cn->sb == sb)
445 return cn;
446 cn = cn->hnext;
447 }
448 return (struct reiserfs_journal_cnode *)0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449}
450
451/*
452** this actually means 'can this block be reallocated yet?'. If you set search_all, a block can only be allocated
453** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever
454** being overwritten by a replay after crashing.
455**
456** If you don't set search_all, a block can only be allocated if it is not in the current transaction. Since deleting
457** a block removes it from the current transaction, this case should never happen. If you don't set search_all, make
458** sure you never write the block without logging it.
459**
460** next_zero_bit is a suggestion about the next block to try for find_forward.
461** when bl is rejected because it is set in a journal list bitmap, we search
462** for the next zero bit in the bitmap that rejected bl. Then, we return that
463** through next_zero_bit for find_forward to try.
464**
465** Just because we return something in next_zero_bit does not mean we won't
466** reject it on the next call to reiserfs_in_journal
467**
468*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400469int reiserfs_in_journal(struct super_block *sb,
Jeff Mahoney3ee16672007-10-18 23:39:25 -0700470 unsigned int bmap_nr, int bit_nr, int search_all,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700471 b_blocknr_t * next_zero_bit)
472{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400473 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700474 struct reiserfs_journal_cnode *cn;
475 struct reiserfs_list_bitmap *jb;
476 int i;
477 unsigned long bl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700479 *next_zero_bit = 0; /* always start this at zero. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400481 PROC_INFO_INC(sb, journal.in_journal);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700482 /* If we aren't doing a search_all, this is a metablock, and it will be logged before use.
483 ** if we crash before the transaction that freed it commits, this transaction won't
484 ** have committed either, and the block will never be written
485 */
486 if (search_all) {
487 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400488 PROC_INFO_INC(sb, journal.in_journal_bitmap);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700489 jb = journal->j_list_bitmap + i;
490 if (jb->journal_list && jb->bitmaps[bmap_nr] &&
491 test_bit(bit_nr,
492 (unsigned long *)jb->bitmaps[bmap_nr]->
493 data)) {
494 *next_zero_bit =
495 find_next_zero_bit((unsigned long *)
496 (jb->bitmaps[bmap_nr]->
497 data),
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400498 sb->s_blocksize << 3,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700499 bit_nr + 1);
500 return 1;
501 }
502 }
503 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400505 bl = bmap_nr * (sb->s_blocksize << 3) + bit_nr;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700506 /* is it in any old transactions? */
507 if (search_all
508 && (cn =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400509 get_journal_hash_dev(sb, journal->j_list_hash_table, bl))) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700510 return 1;
511 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700513 /* is it in the current transaction. This should never happen */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400514 if ((cn = get_journal_hash_dev(sb, journal->j_hash_table, bl))) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700515 BUG();
516 return 1;
517 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400519 PROC_INFO_INC(sb, journal.in_journal_reusable);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700520 /* safe for reuse */
521 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522}
523
524/* insert cn into table
525*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700526static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
527 struct reiserfs_journal_cnode *cn)
528{
529 struct reiserfs_journal_cnode *cn_orig;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700531 cn_orig = journal_hash(table, cn->sb, cn->blocknr);
532 cn->hnext = cn_orig;
533 cn->hprev = NULL;
534 if (cn_orig) {
535 cn_orig->hprev = cn;
536 }
537 journal_hash(table, cn->sb, cn->blocknr) = cn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538}
539
540/* lock the current transaction */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400541static inline void lock_journal(struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700542{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400543 PROC_INFO_INC(sb, journal.lock_journal);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +0200544
545 reiserfs_mutex_lock_safe(&SB_JOURNAL(sb)->j_mutex, sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546}
547
548/* unlock the current transaction */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400549static inline void unlock_journal(struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700550{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400551 mutex_unlock(&SB_JOURNAL(sb)->j_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552}
553
554static inline void get_journal_list(struct reiserfs_journal_list *jl)
555{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700556 jl->j_refcount++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557}
558
559static inline void put_journal_list(struct super_block *s,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700560 struct reiserfs_journal_list *jl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700562 if (jl->j_refcount < 1) {
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -0400563 reiserfs_panic(s, "journal-2", "trans id %u, refcount at %d",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700564 jl->j_trans_id, jl->j_refcount);
565 }
566 if (--jl->j_refcount == 0)
Pekka Enbergd739b422006-02-01 03:06:43 -0800567 kfree(jl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568}
569
570/*
571** this used to be much more involved, and I'm keeping it just in case things get ugly again.
572** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a
573** transaction.
574*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400575static void cleanup_freed_for_journal_list(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700576 struct reiserfs_journal_list *jl)
577{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700579 struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
580 if (jb) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400581 cleanup_bitmap_list(sb, jb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700582 }
583 jl->j_list_bitmap->journal_list = NULL;
584 jl->j_list_bitmap = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585}
586
587static int journal_list_still_alive(struct super_block *s,
Jeff Mahoney600ed412009-03-30 14:02:17 -0400588 unsigned int trans_id)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700590 struct reiserfs_journal *journal = SB_JOURNAL(s);
591 struct list_head *entry = &journal->j_journal_list;
592 struct reiserfs_journal_list *jl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700594 if (!list_empty(entry)) {
595 jl = JOURNAL_LIST_ENTRY(entry->next);
596 if (jl->j_trans_id <= trans_id) {
597 return 1;
598 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700600 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601}
602
Chris Mason398c95b2007-10-16 23:29:44 -0700603/*
604 * If page->mapping was null, we failed to truncate this page for
605 * some reason. Most likely because it was truncated after being
606 * logged via data=journal.
607 *
608 * This does a check to see if the buffer belongs to one of these
609 * lost pages before doing the final put_bh. If page->mapping was
610 * null, it tries to free buffers on the page, which should make the
611 * final page_cache_release drop the page from the lru.
612 */
613static void release_buffer_page(struct buffer_head *bh)
614{
615 struct page *page = bh->b_page;
Nick Piggin529ae9a2008-08-02 12:01:03 +0200616 if (!page->mapping && trylock_page(page)) {
Chris Mason398c95b2007-10-16 23:29:44 -0700617 page_cache_get(page);
618 put_bh(bh);
619 if (!page->mapping)
620 try_to_free_buffers(page);
621 unlock_page(page);
622 page_cache_release(page);
623 } else {
624 put_bh(bh);
625 }
626}
627
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700628static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
629{
630 char b[BDEVNAME_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700632 if (buffer_journaled(bh)) {
Jeff Mahoney45b03d52009-03-30 14:02:21 -0400633 reiserfs_warning(NULL, "clm-2084",
634 "pinned buffer %lu:%s sent to disk",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700635 bh->b_blocknr, bdevname(bh->b_bdev, b));
636 }
637 if (uptodate)
638 set_buffer_uptodate(bh);
639 else
640 clear_buffer_uptodate(bh);
Chris Mason398c95b2007-10-16 23:29:44 -0700641
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700642 unlock_buffer(bh);
Chris Mason398c95b2007-10-16 23:29:44 -0700643 release_buffer_page(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644}
645
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700646static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
647{
648 if (uptodate)
649 set_buffer_uptodate(bh);
650 else
651 clear_buffer_uptodate(bh);
652 unlock_buffer(bh);
653 put_bh(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654}
655
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700656static void submit_logged_buffer(struct buffer_head *bh)
657{
658 get_bh(bh);
659 bh->b_end_io = reiserfs_end_buffer_io_sync;
660 clear_buffer_journal_new(bh);
661 clear_buffer_dirty(bh);
662 if (!test_clear_buffer_journal_test(bh))
663 BUG();
664 if (!buffer_uptodate(bh))
665 BUG();
666 submit_bh(WRITE, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667}
668
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700669static void submit_ordered_buffer(struct buffer_head *bh)
670{
671 get_bh(bh);
672 bh->b_end_io = reiserfs_end_ordered_io;
673 clear_buffer_dirty(bh);
674 if (!buffer_uptodate(bh))
675 BUG();
676 submit_bh(WRITE, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677}
678
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700679static int submit_barrier_buffer(struct buffer_head *bh)
680{
681 get_bh(bh);
682 bh->b_end_io = reiserfs_end_ordered_io;
683 clear_buffer_dirty(bh);
684 if (!buffer_uptodate(bh))
685 BUG();
686 return submit_bh(WRITE_BARRIER, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687}
688
689static void check_barrier_completion(struct super_block *s,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700690 struct buffer_head *bh)
691{
692 if (buffer_eopnotsupp(bh)) {
693 clear_buffer_eopnotsupp(bh);
694 disable_barrier(s);
695 set_buffer_uptodate(bh);
696 set_buffer_dirty(bh);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +0200697 reiserfs_write_unlock(s);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700698 sync_dirty_buffer(bh);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +0200699 reiserfs_write_lock(s);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700700 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701}
702
703#define CHUNK_SIZE 32
704struct buffer_chunk {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700705 struct buffer_head *bh[CHUNK_SIZE];
706 int nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707};
708
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700709static void write_chunk(struct buffer_chunk *chunk)
710{
711 int i;
712 get_fs_excl();
713 for (i = 0; i < chunk->nr; i++) {
714 submit_logged_buffer(chunk->bh[i]);
715 }
716 chunk->nr = 0;
717 put_fs_excl();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718}
719
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700720static void write_ordered_chunk(struct buffer_chunk *chunk)
721{
722 int i;
723 get_fs_excl();
724 for (i = 0; i < chunk->nr; i++) {
725 submit_ordered_buffer(chunk->bh[i]);
726 }
727 chunk->nr = 0;
728 put_fs_excl();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729}
730
731static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700732 spinlock_t * lock, void (fn) (struct buffer_chunk *))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700734 int ret = 0;
Eric Sesterhenn14a61442006-10-03 23:36:38 +0200735 BUG_ON(chunk->nr >= CHUNK_SIZE);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700736 chunk->bh[chunk->nr++] = bh;
737 if (chunk->nr >= CHUNK_SIZE) {
738 ret = 1;
739 if (lock)
740 spin_unlock(lock);
741 fn(chunk);
742 if (lock)
743 spin_lock(lock);
744 }
745 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746}
747
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700749static struct reiserfs_jh *alloc_jh(void)
750{
751 struct reiserfs_jh *jh;
752 while (1) {
753 jh = kmalloc(sizeof(*jh), GFP_NOFS);
754 if (jh) {
755 atomic_inc(&nr_reiserfs_jh);
756 return jh;
757 }
758 yield();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760}
761
762/*
763 * we want to free the jh when the buffer has been written
764 * and waited on
765 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700766void reiserfs_free_jh(struct buffer_head *bh)
767{
768 struct reiserfs_jh *jh;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700770 jh = bh->b_private;
771 if (jh) {
772 bh->b_private = NULL;
773 jh->bh = NULL;
774 list_del_init(&jh->list);
775 kfree(jh);
776 if (atomic_read(&nr_reiserfs_jh) <= 0)
777 BUG();
778 atomic_dec(&nr_reiserfs_jh);
779 put_bh(bh);
780 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781}
782
783static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700784 int tail)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700786 struct reiserfs_jh *jh;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700788 if (bh->b_private) {
789 spin_lock(&j->j_dirty_buffers_lock);
790 if (!bh->b_private) {
791 spin_unlock(&j->j_dirty_buffers_lock);
792 goto no_jh;
793 }
794 jh = bh->b_private;
795 list_del_init(&jh->list);
796 } else {
797 no_jh:
798 get_bh(bh);
799 jh = alloc_jh();
800 spin_lock(&j->j_dirty_buffers_lock);
801 /* buffer must be locked for __add_jh, should be able to have
802 * two adds at the same time
803 */
Eric Sesterhenn14a61442006-10-03 23:36:38 +0200804 BUG_ON(bh->b_private);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700805 jh->bh = bh;
806 bh->b_private = jh;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700808 jh->jl = j->j_current_jl;
809 if (tail)
810 list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
811 else {
812 list_add_tail(&jh->list, &jh->jl->j_bh_list);
813 }
814 spin_unlock(&j->j_dirty_buffers_lock);
815 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816}
817
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700818int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
819{
820 return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821}
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700822int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
823{
824 return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825}
826
827#define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700828static int write_ordered_buffers(spinlock_t * lock,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 struct reiserfs_journal *j,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700830 struct reiserfs_journal_list *jl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 struct list_head *list)
832{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700833 struct buffer_head *bh;
834 struct reiserfs_jh *jh;
835 int ret = j->j_errno;
836 struct buffer_chunk chunk;
837 struct list_head tmp;
838 INIT_LIST_HEAD(&tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700840 chunk.nr = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 spin_lock(lock);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700842 while (!list_empty(list)) {
843 jh = JH_ENTRY(list->next);
844 bh = jh->bh;
845 get_bh(bh);
Nick Pigginca5de402008-08-02 12:02:13 +0200846 if (!trylock_buffer(bh)) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700847 if (!buffer_dirty(bh)) {
Akinobu Mitaf1166292006-06-26 00:24:46 -0700848 list_move(&jh->list, &tmp);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700849 goto loop_next;
850 }
851 spin_unlock(lock);
852 if (chunk.nr)
853 write_ordered_chunk(&chunk);
854 wait_on_buffer(bh);
855 cond_resched();
856 spin_lock(lock);
857 goto loop_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 }
Chris Mason3d4492f2006-02-01 03:06:49 -0800859 /* in theory, dirty non-uptodate buffers should never get here,
860 * but the upper layer io error paths still have a few quirks.
861 * Handle them here as gracefully as we can
862 */
863 if (!buffer_uptodate(bh) && buffer_dirty(bh)) {
864 clear_buffer_dirty(bh);
865 ret = -EIO;
866 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700867 if (buffer_dirty(bh)) {
Akinobu Mitaf1166292006-06-26 00:24:46 -0700868 list_move(&jh->list, &tmp);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700869 add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
870 } else {
871 reiserfs_free_jh(bh);
872 unlock_buffer(bh);
873 }
874 loop_next:
875 put_bh(bh);
876 cond_resched_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700878 if (chunk.nr) {
879 spin_unlock(lock);
880 write_ordered_chunk(&chunk);
881 spin_lock(lock);
882 }
883 while (!list_empty(&tmp)) {
884 jh = JH_ENTRY(tmp.prev);
885 bh = jh->bh;
886 get_bh(bh);
887 reiserfs_free_jh(bh);
888
889 if (buffer_locked(bh)) {
890 spin_unlock(lock);
891 wait_on_buffer(bh);
892 spin_lock(lock);
893 }
894 if (!buffer_uptodate(bh)) {
895 ret = -EIO;
896 }
Chris Masond62b1b82006-02-01 03:06:47 -0800897 /* ugly interaction with invalidatepage here.
898 * reiserfs_invalidate_page will pin any buffer that has a valid
899 * journal head from an older transaction. If someone else sets
900 * our buffer dirty after we write it in the first loop, and
901 * then someone truncates the page away, nobody will ever write
902 * the buffer. We're safe if we write the page one last time
903 * after freeing the journal header.
904 */
905 if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
906 spin_unlock(lock);
907 ll_rw_block(WRITE, 1, &bh);
908 spin_lock(lock);
909 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700910 put_bh(bh);
911 cond_resched_lock(lock);
912 }
913 spin_unlock(lock);
914 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915}
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700916
917static int flush_older_commits(struct super_block *s,
918 struct reiserfs_journal_list *jl)
919{
920 struct reiserfs_journal *journal = SB_JOURNAL(s);
921 struct reiserfs_journal_list *other_jl;
922 struct reiserfs_journal_list *first_jl;
923 struct list_head *entry;
Jeff Mahoney600ed412009-03-30 14:02:17 -0400924 unsigned int trans_id = jl->j_trans_id;
925 unsigned int other_trans_id;
926 unsigned int first_trans_id;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700927
928 find_first:
929 /*
930 * first we walk backwards to find the oldest uncommitted transation
931 */
932 first_jl = jl;
933 entry = jl->j_list.prev;
934 while (1) {
935 other_jl = JOURNAL_LIST_ENTRY(entry);
936 if (entry == &journal->j_journal_list ||
937 atomic_read(&other_jl->j_older_commits_done))
938 break;
939
940 first_jl = other_jl;
941 entry = other_jl->j_list.prev;
942 }
943
944 /* if we didn't find any older uncommitted transactions, return now */
945 if (first_jl == jl) {
946 return 0;
947 }
948
949 first_trans_id = first_jl->j_trans_id;
950
951 entry = &first_jl->j_list;
952 while (1) {
953 other_jl = JOURNAL_LIST_ENTRY(entry);
954 other_trans_id = other_jl->j_trans_id;
955
956 if (other_trans_id < trans_id) {
957 if (atomic_read(&other_jl->j_commit_left) != 0) {
958 flush_commit_list(s, other_jl, 0);
959
960 /* list we were called with is gone, return */
961 if (!journal_list_still_alive(s, trans_id))
962 return 1;
963
964 /* the one we just flushed is gone, this means all
965 * older lists are also gone, so first_jl is no longer
966 * valid either. Go back to the beginning.
967 */
968 if (!journal_list_still_alive
969 (s, other_trans_id)) {
970 goto find_first;
971 }
972 }
973 entry = entry->next;
974 if (entry == &journal->j_journal_list)
975 return 0;
976 } else {
977 return 0;
978 }
979 }
980 return 0;
981}
Adrian Bunkdeba0f42007-10-16 23:26:03 -0700982
983static int reiserfs_async_progress_wait(struct super_block *s)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700984{
985 DEFINE_WAIT(wait);
986 struct reiserfs_journal *j = SB_JOURNAL(s);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +0200987
988 if (atomic_read(&j->j_async_throttle)) {
989 reiserfs_write_unlock(s);
Jens Axboe8aa7e842009-07-09 14:52:32 +0200990 congestion_wait(BLK_RW_ASYNC, HZ / 10);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +0200991 reiserfs_write_lock(s);
992 }
993
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700994 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995}
996
997/*
998** if this journal list still has commit blocks unflushed, send them to disk.
999**
1000** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
1001** Before the commit block can by written, every other log block must be safely on disk
1002**
1003*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001004static int flush_commit_list(struct super_block *s,
1005 struct reiserfs_journal_list *jl, int flushall)
1006{
1007 int i;
Jeff Mahoney3ee16672007-10-18 23:39:25 -07001008 b_blocknr_t bn;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001009 struct buffer_head *tbh = NULL;
Jeff Mahoney600ed412009-03-30 14:02:17 -04001010 unsigned int trans_id = jl->j_trans_id;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001011 struct reiserfs_journal *journal = SB_JOURNAL(s);
1012 int barrier = 0;
1013 int retval = 0;
Chris Masone0e851c2006-02-01 03:06:49 -08001014 int write_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001016 reiserfs_check_lock_depth(s, "flush_commit_list");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001018 if (atomic_read(&jl->j_older_commits_done)) {
1019 return 0;
1020 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001022 get_fs_excl();
Jens Axboe22e2c502005-06-27 10:55:12 +02001023
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001024 /* before we can put our commit blocks on disk, we have to make sure everyone older than
1025 ** us is on disk too
1026 */
1027 BUG_ON(jl->j_len <= 0);
1028 BUG_ON(trans_id == journal->j_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001030 get_journal_list(jl);
1031 if (flushall) {
1032 if (flush_older_commits(s, jl) == 1) {
1033 /* list disappeared during flush_older_commits. return */
1034 goto put_jl;
1035 }
1036 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001038 /* make sure nobody is trying to flush this one at the same time */
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001039 reiserfs_mutex_lock_safe(&jl->j_commit_mutex, s);
1040
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001041 if (!journal_list_still_alive(s, trans_id)) {
Jeff Mahoney90415de2008-07-25 01:46:40 -07001042 mutex_unlock(&jl->j_commit_mutex);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001043 goto put_jl;
1044 }
1045 BUG_ON(jl->j_trans_id == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001047 /* this commit is done, exit */
1048 if (atomic_read(&(jl->j_commit_left)) <= 0) {
1049 if (flushall) {
1050 atomic_set(&(jl->j_older_commits_done), 1);
1051 }
Jeff Mahoney90415de2008-07-25 01:46:40 -07001052 mutex_unlock(&jl->j_commit_mutex);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001053 goto put_jl;
1054 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001056 if (!list_empty(&jl->j_bh_list)) {
Chris Mason3d4492f2006-02-01 03:06:49 -08001057 int ret;
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001058
1059 /*
1060 * We might sleep in numerous places inside
1061 * write_ordered_buffers. Relax the write lock.
1062 */
1063 reiserfs_write_unlock(s);
Chris Mason3d4492f2006-02-01 03:06:49 -08001064 ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
1065 journal, jl, &jl->j_bh_list);
1066 if (ret < 0 && retval == 0)
1067 retval = ret;
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001068 reiserfs_write_lock(s);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001069 }
1070 BUG_ON(!list_empty(&jl->j_bh_list));
1071 /*
1072 * for the description block and all the log blocks, submit any buffers
Chris Masone0e851c2006-02-01 03:06:49 -08001073 * that haven't already reached the disk. Try to write at least 256
1074 * log blocks. later on, we will only wait on blocks that correspond
1075 * to this transaction, but while we're unplugging we might as well
1076 * get a chunk of data on there.
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001077 */
1078 atomic_inc(&journal->j_async_throttle);
Chris Masone0e851c2006-02-01 03:06:49 -08001079 write_len = jl->j_len + 1;
1080 if (write_len < 256)
1081 write_len = 256;
1082 for (i = 0 ; i < write_len ; i++) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001083 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1084 SB_ONDISK_JOURNAL_SIZE(s);
1085 tbh = journal_find_get_block(s, bn);
Chris Masone0e851c2006-02-01 03:06:49 -08001086 if (tbh) {
Frederic Weisbecker6e3647a2009-05-01 02:27:39 +02001087 if (buffer_dirty(tbh)) {
1088 reiserfs_write_unlock(s);
1089 ll_rw_block(WRITE, 1, &tbh);
1090 reiserfs_write_lock(s);
1091 }
Chris Masone0e851c2006-02-01 03:06:49 -08001092 put_bh(tbh) ;
1093 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001094 }
1095 atomic_dec(&journal->j_async_throttle);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096
Jeff Mahoney5d5e81562005-12-14 14:38:36 -05001097 /* We're skipping the commit if there's an error */
1098 if (retval || reiserfs_is_journal_aborted(journal))
1099 barrier = 0;
1100
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001101 /* wait on everything written so far before writing the commit
1102 * if we are in barrier mode, send the commit down now
1103 */
1104 barrier = reiserfs_barrier_flush(s);
1105 if (barrier) {
1106 int ret;
1107 lock_buffer(jl->j_commit_bh);
1108 ret = submit_barrier_buffer(jl->j_commit_bh);
1109 if (ret == -EOPNOTSUPP) {
1110 set_buffer_uptodate(jl->j_commit_bh);
1111 disable_barrier(s);
1112 barrier = 0;
1113 }
1114 }
1115 for (i = 0; i < (jl->j_len + 1); i++) {
1116 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1117 (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1118 tbh = journal_find_get_block(s, bn);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001119
1120 reiserfs_write_unlock(s);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001121 wait_on_buffer(tbh);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001122 reiserfs_write_lock(s);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001123 // since we're using ll_rw_blk above, it might have skipped over
1124 // a locked buffer. Double check here
1125 //
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001126 /* redundant, sync_dirty_buffer() checks */
1127 if (buffer_dirty(tbh)) {
1128 reiserfs_write_unlock(s);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001129 sync_dirty_buffer(tbh);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001130 reiserfs_write_lock(s);
1131 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001132 if (unlikely(!buffer_uptodate(tbh))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133#ifdef CONFIG_REISERFS_CHECK
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001134 reiserfs_warning(s, "journal-601",
1135 "buffer write failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136#endif
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001137 retval = -EIO;
1138 }
1139 put_bh(tbh); /* once for journal_find_get_block */
1140 put_bh(tbh); /* once due to original getblk in do_journal_end */
1141 atomic_dec(&(jl->j_commit_left));
1142 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001144 BUG_ON(atomic_read(&(jl->j_commit_left)) != 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001146 if (!barrier) {
Jeff Mahoney5d5e81562005-12-14 14:38:36 -05001147 /* If there was a write error in the journal - we can't commit
1148 * this transaction - it will be invalid and, if successful,
Robert P. J. Daybeb7dd82007-05-09 07:14:03 +02001149 * will just end up propagating the write error out to
Jeff Mahoney5d5e81562005-12-14 14:38:36 -05001150 * the file system. */
1151 if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1152 if (buffer_dirty(jl->j_commit_bh))
1153 BUG();
1154 mark_buffer_dirty(jl->j_commit_bh) ;
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001155 reiserfs_write_unlock(s);
Jeff Mahoney5d5e81562005-12-14 14:38:36 -05001156 sync_dirty_buffer(jl->j_commit_bh) ;
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001157 reiserfs_write_lock(s);
Jeff Mahoney5d5e81562005-12-14 14:38:36 -05001158 }
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001159 } else {
1160 reiserfs_write_unlock(s);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001161 wait_on_buffer(jl->j_commit_bh);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001162 reiserfs_write_lock(s);
1163 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001165 check_barrier_completion(s, jl->j_commit_bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001167 /* If there was a write error in the journal - we can't commit this
1168 * transaction - it will be invalid and, if successful, will just end
Robert P. J. Daybeb7dd82007-05-09 07:14:03 +02001169 * up propagating the write error out to the filesystem. */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001170 if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171#ifdef CONFIG_REISERFS_CHECK
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001172 reiserfs_warning(s, "journal-615", "buffer write failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173#endif
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001174 retval = -EIO;
1175 }
1176 bforget(jl->j_commit_bh);
1177 if (journal->j_last_commit_id != 0 &&
1178 (jl->j_trans_id - journal->j_last_commit_id) != 1) {
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001179 reiserfs_warning(s, "clm-2200", "last commit %lu, current %lu",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001180 journal->j_last_commit_id, jl->j_trans_id);
1181 }
1182 journal->j_last_commit_id = jl->j_trans_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001184 /* now, every commit block is on the disk. It is safe to allow blocks freed during this transaction to be reallocated */
1185 cleanup_freed_for_journal_list(s, jl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001187 retval = retval ? retval : journal->j_errno;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001189 /* mark the metadata dirty */
1190 if (!retval)
1191 dirty_one_transaction(s, jl);
1192 atomic_dec(&(jl->j_commit_left));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001194 if (flushall) {
1195 atomic_set(&(jl->j_older_commits_done), 1);
1196 }
Jeff Mahoney90415de2008-07-25 01:46:40 -07001197 mutex_unlock(&jl->j_commit_mutex);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001198 put_jl:
1199 put_journal_list(s, jl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001201 if (retval)
1202 reiserfs_abort(s, retval, "Journal write error in %s",
Harvey Harrisonfbe54982008-04-28 02:16:22 -07001203 __func__);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001204 put_fs_excl();
1205 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206}
1207
1208/*
Jeff Mahoney0222e652009-03-30 14:02:44 -04001209** flush_journal_list frequently needs to find a newer transaction for a given block. This does that, or
1210** returns NULL if it can't find anything
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001212static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1213 reiserfs_journal_cnode
1214 *cn)
1215{
1216 struct super_block *sb = cn->sb;
1217 b_blocknr_t blocknr = cn->blocknr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001219 cn = cn->hprev;
1220 while (cn) {
1221 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1222 return cn->jlist;
1223 }
1224 cn = cn->hprev;
1225 }
1226 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227}
1228
Chris Masona3172022006-09-29 01:59:56 -07001229static int newer_jl_done(struct reiserfs_journal_cnode *cn)
1230{
1231 struct super_block *sb = cn->sb;
1232 b_blocknr_t blocknr = cn->blocknr;
1233
1234 cn = cn->hprev;
1235 while (cn) {
1236 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist &&
1237 atomic_read(&cn->jlist->j_commit_left) != 0)
1238 return 0;
1239 cn = cn->hprev;
1240 }
1241 return 1;
1242}
1243
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001244static void remove_journal_hash(struct super_block *,
1245 struct reiserfs_journal_cnode **,
1246 struct reiserfs_journal_list *, unsigned long,
1247 int);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248
1249/*
1250** once all the real blocks have been flushed, it is safe to remove them from the
1251** journal list for this transaction. Aside from freeing the cnode, this also allows the
1252** block to be reallocated for data blocks if it had been deleted.
1253*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001254static void remove_all_from_journal_list(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001255 struct reiserfs_journal_list *jl,
1256 int debug)
1257{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001258 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001259 struct reiserfs_journal_cnode *cn, *last;
1260 cn = jl->j_realblock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001262 /* which is better, to lock once around the whole loop, or
1263 ** to lock for each call to remove_journal_hash?
1264 */
1265 while (cn) {
1266 if (cn->blocknr != 0) {
1267 if (debug) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001268 reiserfs_warning(sb, "reiserfs-2201",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001269 "block %u, bh is %d, state %ld",
1270 cn->blocknr, cn->bh ? 1 : 0,
1271 cn->state);
1272 }
1273 cn->state = 0;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001274 remove_journal_hash(sb, journal->j_list_hash_table,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001275 jl, cn->blocknr, 1);
1276 }
1277 last = cn;
1278 cn = cn->next;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001279 free_cnode(sb, last);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001280 }
1281 jl->j_realblock = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282}
1283
1284/*
1285** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block.
1286** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start
1287** releasing blocks in this transaction for reuse as data blocks.
1288** called by flush_journal_list, before it calls remove_all_from_journal_list
1289**
1290*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001291static int _update_journal_header_block(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001292 unsigned long offset,
Jeff Mahoney600ed412009-03-30 14:02:17 -04001293 unsigned int trans_id)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001294{
1295 struct reiserfs_journal_header *jh;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001296 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001298 if (reiserfs_is_journal_aborted(journal))
1299 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001301 if (trans_id >= journal->j_last_flush_trans_id) {
1302 if (buffer_locked((journal->j_header_bh))) {
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001303 reiserfs_write_unlock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001304 wait_on_buffer((journal->j_header_bh));
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001305 reiserfs_write_lock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001306 if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307#ifdef CONFIG_REISERFS_CHECK
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001308 reiserfs_warning(sb, "journal-699",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001309 "buffer write failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310#endif
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001311 return -EIO;
1312 }
1313 }
1314 journal->j_last_flush_trans_id = trans_id;
1315 journal->j_first_unflushed_offset = offset;
1316 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1317 b_data);
1318 jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1319 jh->j_first_unflushed_offset = cpu_to_le32(offset);
1320 jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001322 if (reiserfs_barrier_flush(sb)) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001323 int ret;
1324 lock_buffer(journal->j_header_bh);
1325 ret = submit_barrier_buffer(journal->j_header_bh);
1326 if (ret == -EOPNOTSUPP) {
1327 set_buffer_uptodate(journal->j_header_bh);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001328 disable_barrier(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001329 goto sync;
1330 }
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001331 reiserfs_write_unlock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001332 wait_on_buffer(journal->j_header_bh);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001333 reiserfs_write_lock(sb);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001334 check_barrier_completion(sb, journal->j_header_bh);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001335 } else {
1336 sync:
1337 set_buffer_dirty(journal->j_header_bh);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001338 reiserfs_write_unlock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001339 sync_dirty_buffer(journal->j_header_bh);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001340 reiserfs_write_lock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001341 }
1342 if (!buffer_uptodate(journal->j_header_bh)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001343 reiserfs_warning(sb, "journal-837",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001344 "IO error during journal replay");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001345 return -EIO;
1346 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001348 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349}
1350
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001351static int update_journal_header_block(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001352 unsigned long offset,
Jeff Mahoney600ed412009-03-30 14:02:17 -04001353 unsigned int trans_id)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001354{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001355 return _update_journal_header_block(sb, offset, trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356}
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001357
Jeff Mahoney0222e652009-03-30 14:02:44 -04001358/*
1359** flush any and all journal lists older than you are
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360** can only be called from flush_journal_list
1361*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001362static int flush_older_journal_lists(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001363 struct reiserfs_journal_list *jl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001365 struct list_head *entry;
1366 struct reiserfs_journal_list *other_jl;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001367 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Jeff Mahoney600ed412009-03-30 14:02:17 -04001368 unsigned int trans_id = jl->j_trans_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001370 /* we know we are the only ones flushing things, no extra race
1371 * protection is required.
1372 */
1373 restart:
1374 entry = journal->j_journal_list.next;
1375 /* Did we wrap? */
1376 if (entry == &journal->j_journal_list)
1377 return 0;
1378 other_jl = JOURNAL_LIST_ENTRY(entry);
1379 if (other_jl->j_trans_id < trans_id) {
1380 BUG_ON(other_jl->j_refcount <= 0);
1381 /* do not flush all */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001382 flush_journal_list(sb, other_jl, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001384 /* other_jl is now deleted from the list */
1385 goto restart;
1386 }
1387 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388}
1389
1390static void del_from_work_list(struct super_block *s,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001391 struct reiserfs_journal_list *jl)
1392{
1393 struct reiserfs_journal *journal = SB_JOURNAL(s);
1394 if (!list_empty(&jl->j_working_list)) {
1395 list_del_init(&jl->j_working_list);
1396 journal->j_num_work_lists--;
1397 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398}
1399
1400/* flush a journal list, both commit and real blocks
1401**
1402** always set flushall to 1, unless you are calling from inside
1403** flush_journal_list
1404**
Jeff Mahoney0222e652009-03-30 14:02:44 -04001405** IMPORTANT. This can only be called while there are no journal writers,
1406** and the journal is locked. That means it can only be called from
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407** do_journal_end, or by journal_release
1408*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001409static int flush_journal_list(struct super_block *s,
1410 struct reiserfs_journal_list *jl, int flushall)
1411{
1412 struct reiserfs_journal_list *pjl;
1413 struct reiserfs_journal_cnode *cn, *last;
1414 int count;
1415 int was_jwait = 0;
1416 int was_dirty = 0;
1417 struct buffer_head *saved_bh;
1418 unsigned long j_len_saved = jl->j_len;
1419 struct reiserfs_journal *journal = SB_JOURNAL(s);
1420 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001422 BUG_ON(j_len_saved <= 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001424 if (atomic_read(&journal->j_wcount) != 0) {
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001425 reiserfs_warning(s, "clm-2048", "called with wcount %d",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001426 atomic_read(&journal->j_wcount));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001428 BUG_ON(jl->j_trans_id == 0);
1429
1430 /* if flushall == 0, the lock is already held */
1431 if (flushall) {
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001432 reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
Jeff Mahoneyafe70252008-07-25 01:46:39 -07001433 } else if (mutex_trylock(&journal->j_flush_mutex)) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001434 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001436
1437 count = 0;
1438 if (j_len_saved > journal->j_trans_max) {
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04001439 reiserfs_panic(s, "journal-715", "length is %lu, trans id %lu",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001440 j_len_saved, jl->j_trans_id);
1441 return 0;
1442 }
1443
1444 get_fs_excl();
1445
1446 /* if all the work is already done, get out of here */
1447 if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1448 atomic_read(&(jl->j_commit_left)) <= 0) {
1449 goto flush_older_and_return;
1450 }
1451
Jeff Mahoney0222e652009-03-30 14:02:44 -04001452 /* start by putting the commit list on disk. This will also flush
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001453 ** the commit lists of any olders transactions
1454 */
1455 flush_commit_list(s, jl, 1);
1456
1457 if (!(jl->j_state & LIST_DIRTY)
1458 && !reiserfs_is_journal_aborted(journal))
1459 BUG();
1460
1461 /* are we done now? */
1462 if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1463 atomic_read(&(jl->j_commit_left)) <= 0) {
1464 goto flush_older_and_return;
1465 }
1466
Jeff Mahoney0222e652009-03-30 14:02:44 -04001467 /* loop through each cnode, see if we need to write it,
1468 ** or wait on a more recent transaction, or just ignore it
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001469 */
1470 if (atomic_read(&(journal->j_wcount)) != 0) {
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04001471 reiserfs_panic(s, "journal-844", "journal list is flushing, "
1472 "wcount is not 0");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001473 }
1474 cn = jl->j_realblock;
1475 while (cn) {
1476 was_jwait = 0;
1477 was_dirty = 0;
1478 saved_bh = NULL;
1479 /* blocknr of 0 is no longer in the hash, ignore it */
1480 if (cn->blocknr == 0) {
1481 goto free_cnode;
1482 }
1483
1484 /* This transaction failed commit. Don't write out to the disk */
1485 if (!(jl->j_state & LIST_DIRTY))
1486 goto free_cnode;
1487
1488 pjl = find_newer_jl_for_cn(cn);
1489 /* the order is important here. We check pjl to make sure we
1490 ** don't clear BH_JDirty_wait if we aren't the one writing this
1491 ** block to disk
1492 */
1493 if (!pjl && cn->bh) {
1494 saved_bh = cn->bh;
1495
Jeff Mahoney0222e652009-03-30 14:02:44 -04001496 /* we do this to make sure nobody releases the buffer while
1497 ** we are working with it
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001498 */
1499 get_bh(saved_bh);
1500
1501 if (buffer_journal_dirty(saved_bh)) {
1502 BUG_ON(!can_dirty(cn));
1503 was_jwait = 1;
1504 was_dirty = 1;
1505 } else if (can_dirty(cn)) {
1506 /* everything with !pjl && jwait should be writable */
1507 BUG();
1508 }
1509 }
1510
1511 /* if someone has this block in a newer transaction, just make
Matt LaPlante0779bf22006-11-30 05:24:39 +01001512 ** sure they are committed, and don't try writing it to disk
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001513 */
1514 if (pjl) {
1515 if (atomic_read(&pjl->j_commit_left))
1516 flush_commit_list(s, pjl, 1);
1517 goto free_cnode;
1518 }
1519
Jeff Mahoney0222e652009-03-30 14:02:44 -04001520 /* bh == NULL when the block got to disk on its own, OR,
1521 ** the block got freed in a future transaction
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001522 */
1523 if (saved_bh == NULL) {
1524 goto free_cnode;
1525 }
1526
1527 /* this should never happen. kupdate_one_transaction has this list
1528 ** locked while it works, so we should never see a buffer here that
1529 ** is not marked JDirty_wait
1530 */
1531 if ((!was_jwait) && !buffer_locked(saved_bh)) {
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001532 reiserfs_warning(s, "journal-813",
1533 "BAD! buffer %llu %cdirty %cjwait, "
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001534 "not in a newer tranasction",
1535 (unsigned long long)saved_bh->
1536 b_blocknr, was_dirty ? ' ' : '!',
1537 was_jwait ? ' ' : '!');
1538 }
1539 if (was_dirty) {
1540 /* we inc again because saved_bh gets decremented at free_cnode */
1541 get_bh(saved_bh);
1542 set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1543 lock_buffer(saved_bh);
1544 BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1545 if (buffer_dirty(saved_bh))
1546 submit_logged_buffer(saved_bh);
1547 else
1548 unlock_buffer(saved_bh);
1549 count++;
1550 } else {
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001551 reiserfs_warning(s, "clm-2082",
1552 "Unable to flush buffer %llu in %s",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001553 (unsigned long long)saved_bh->
Harvey Harrisonfbe54982008-04-28 02:16:22 -07001554 b_blocknr, __func__);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001555 }
1556 free_cnode:
1557 last = cn;
1558 cn = cn->next;
1559 if (saved_bh) {
1560 /* we incremented this to keep others from taking the buffer head away */
1561 put_bh(saved_bh);
1562 if (atomic_read(&(saved_bh->b_count)) < 0) {
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001563 reiserfs_warning(s, "journal-945",
1564 "saved_bh->b_count < 0");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001565 }
1566 }
1567 }
1568 if (count > 0) {
1569 cn = jl->j_realblock;
1570 while (cn) {
1571 if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1572 if (!cn->bh) {
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04001573 reiserfs_panic(s, "journal-1011",
1574 "cn->bh is NULL");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001575 }
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001576
1577 reiserfs_write_unlock(s);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001578 wait_on_buffer(cn->bh);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001579 reiserfs_write_lock(s);
1580
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001581 if (!cn->bh) {
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04001582 reiserfs_panic(s, "journal-1012",
1583 "cn->bh is NULL");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001584 }
1585 if (unlikely(!buffer_uptodate(cn->bh))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586#ifdef CONFIG_REISERFS_CHECK
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001587 reiserfs_warning(s, "journal-949",
1588 "buffer write failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589#endif
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001590 err = -EIO;
1591 }
1592 /* note, we must clear the JDirty_wait bit after the up to date
1593 ** check, otherwise we race against our flushpage routine
1594 */
1595 BUG_ON(!test_clear_buffer_journal_dirty
1596 (cn->bh));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597
Chris Mason398c95b2007-10-16 23:29:44 -07001598 /* drop one ref for us */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001599 put_bh(cn->bh);
Chris Mason398c95b2007-10-16 23:29:44 -07001600 /* drop one ref for journal_mark_dirty */
1601 release_buffer_page(cn->bh);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001602 }
1603 cn = cn->next;
1604 }
1605 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001607 if (err)
1608 reiserfs_abort(s, -EIO,
1609 "Write error while pushing transaction to disk in %s",
Harvey Harrisonfbe54982008-04-28 02:16:22 -07001610 __func__);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001611 flush_older_and_return:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612
Jeff Mahoney0222e652009-03-30 14:02:44 -04001613 /* before we can update the journal header block, we _must_ flush all
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001614 ** real blocks from all older transactions to disk. This is because
1615 ** once the header block is updated, this transaction will not be
1616 ** replayed after a crash
1617 */
1618 if (flushall) {
1619 flush_older_journal_lists(s, jl);
1620 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001622 err = journal->j_errno;
Jeff Mahoney0222e652009-03-30 14:02:44 -04001623 /* before we can remove everything from the hash tables for this
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001624 ** transaction, we must make sure it can never be replayed
1625 **
1626 ** since we are only called from do_journal_end, we know for sure there
1627 ** are no allocations going on while we are flushing journal lists. So,
1628 ** we only need to update the journal header block for the last list
1629 ** being flushed
1630 */
1631 if (!err && flushall) {
1632 err =
1633 update_journal_header_block(s,
1634 (jl->j_start + jl->j_len +
1635 2) % SB_ONDISK_JOURNAL_SIZE(s),
1636 jl->j_trans_id);
1637 if (err)
1638 reiserfs_abort(s, -EIO,
1639 "Write error while updating journal header in %s",
Harvey Harrisonfbe54982008-04-28 02:16:22 -07001640 __func__);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001641 }
1642 remove_all_from_journal_list(s, jl, 0);
1643 list_del_init(&jl->j_list);
1644 journal->j_num_lists--;
1645 del_from_work_list(s, jl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001647 if (journal->j_last_flush_id != 0 &&
1648 (jl->j_trans_id - journal->j_last_flush_id) != 1) {
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001649 reiserfs_warning(s, "clm-2201", "last flush %lu, current %lu",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001650 journal->j_last_flush_id, jl->j_trans_id);
1651 }
1652 journal->j_last_flush_id = jl->j_trans_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001654 /* not strictly required since we are freeing the list, but it should
1655 * help find code using dead lists later on
1656 */
1657 jl->j_len = 0;
1658 atomic_set(&(jl->j_nonzerolen), 0);
1659 jl->j_start = 0;
1660 jl->j_realblock = NULL;
1661 jl->j_commit_bh = NULL;
1662 jl->j_trans_id = 0;
1663 jl->j_state = 0;
1664 put_journal_list(s, jl);
1665 if (flushall)
Jeff Mahoneyafe70252008-07-25 01:46:39 -07001666 mutex_unlock(&journal->j_flush_mutex);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001667 put_fs_excl();
1668 return err;
1669}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670
Chris Masona3172022006-09-29 01:59:56 -07001671static int test_transaction(struct super_block *s,
1672 struct reiserfs_journal_list *jl)
1673{
1674 struct reiserfs_journal_cnode *cn;
1675
1676 if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0)
1677 return 1;
1678
1679 cn = jl->j_realblock;
1680 while (cn) {
1681 /* if the blocknr == 0, this has been cleared from the hash,
1682 ** skip it
1683 */
1684 if (cn->blocknr == 0) {
1685 goto next;
1686 }
1687 if (cn->bh && !newer_jl_done(cn))
1688 return 0;
1689 next:
1690 cn = cn->next;
1691 cond_resched();
1692 }
1693 return 0;
1694}
1695
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696static int write_one_transaction(struct super_block *s,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001697 struct reiserfs_journal_list *jl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698 struct buffer_chunk *chunk)
1699{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001700 struct reiserfs_journal_cnode *cn;
1701 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001703 jl->j_state |= LIST_TOUCHED;
1704 del_from_work_list(s, jl);
1705 if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1706 return 0;
1707 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001709 cn = jl->j_realblock;
1710 while (cn) {
1711 /* if the blocknr == 0, this has been cleared from the hash,
1712 ** skip it
1713 */
1714 if (cn->blocknr == 0) {
1715 goto next;
1716 }
1717 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1718 struct buffer_head *tmp_bh;
1719 /* we can race against journal_mark_freed when we try
1720 * to lock_buffer(cn->bh), so we have to inc the buffer
1721 * count, and recheck things after locking
1722 */
1723 tmp_bh = cn->bh;
1724 get_bh(tmp_bh);
1725 lock_buffer(tmp_bh);
1726 if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1727 if (!buffer_journal_dirty(tmp_bh) ||
1728 buffer_journal_prepared(tmp_bh))
1729 BUG();
1730 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1731 ret++;
1732 } else {
1733 /* note, cn->bh might be null now */
1734 unlock_buffer(tmp_bh);
1735 }
1736 put_bh(tmp_bh);
1737 }
1738 next:
1739 cn = cn->next;
1740 cond_resched();
1741 }
1742 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743}
1744
1745/* used by flush_commit_list */
1746static int dirty_one_transaction(struct super_block *s,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001747 struct reiserfs_journal_list *jl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001749 struct reiserfs_journal_cnode *cn;
1750 struct reiserfs_journal_list *pjl;
1751 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001753 jl->j_state |= LIST_DIRTY;
1754 cn = jl->j_realblock;
1755 while (cn) {
1756 /* look for a more recent transaction that logged this
1757 ** buffer. Only the most recent transaction with a buffer in
1758 ** it is allowed to send that buffer to disk
1759 */
1760 pjl = find_newer_jl_for_cn(cn);
1761 if (!pjl && cn->blocknr && cn->bh
1762 && buffer_journal_dirty(cn->bh)) {
1763 BUG_ON(!can_dirty(cn));
1764 /* if the buffer is prepared, it will either be logged
1765 * or restored. If restored, we need to make sure
1766 * it actually gets marked dirty
1767 */
1768 clear_buffer_journal_new(cn->bh);
1769 if (buffer_journal_prepared(cn->bh)) {
1770 set_buffer_journal_restore_dirty(cn->bh);
1771 } else {
1772 set_buffer_journal_test(cn->bh);
1773 mark_buffer_dirty(cn->bh);
1774 }
1775 }
1776 cn = cn->next;
1777 }
1778 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779}
1780
1781static int kupdate_transactions(struct super_block *s,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001782 struct reiserfs_journal_list *jl,
1783 struct reiserfs_journal_list **next_jl,
Jeff Mahoney600ed412009-03-30 14:02:17 -04001784 unsigned int *next_trans_id,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001785 int num_blocks, int num_trans)
1786{
1787 int ret = 0;
1788 int written = 0;
1789 int transactions_flushed = 0;
Jeff Mahoney600ed412009-03-30 14:02:17 -04001790 unsigned int orig_trans_id = jl->j_trans_id;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001791 struct buffer_chunk chunk;
1792 struct list_head *entry;
1793 struct reiserfs_journal *journal = SB_JOURNAL(s);
1794 chunk.nr = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795
Frederic Weisbeckera412f9e2009-04-14 00:10:35 +02001796 reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001797 if (!journal_list_still_alive(s, orig_trans_id)) {
1798 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800
Jeff Mahoneyafe70252008-07-25 01:46:39 -07001801 /* we've got j_flush_mutex held, nobody is going to delete any
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001802 * of these lists out from underneath us
1803 */
1804 while ((num_trans && transactions_flushed < num_trans) ||
1805 (!num_trans && written < num_blocks)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001807 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1808 atomic_read(&jl->j_commit_left)
1809 || !(jl->j_state & LIST_DIRTY)) {
1810 del_from_work_list(s, jl);
1811 break;
1812 }
1813 ret = write_one_transaction(s, jl, &chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001815 if (ret < 0)
1816 goto done;
1817 transactions_flushed++;
1818 written += ret;
1819 entry = jl->j_list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001821 /* did we wrap? */
1822 if (entry == &journal->j_journal_list) {
1823 break;
1824 }
1825 jl = JOURNAL_LIST_ENTRY(entry);
1826
1827 /* don't bother with older transactions */
1828 if (jl->j_trans_id <= orig_trans_id)
1829 break;
1830 }
1831 if (chunk.nr) {
1832 write_chunk(&chunk);
1833 }
1834
1835 done:
Jeff Mahoneyafe70252008-07-25 01:46:39 -07001836 mutex_unlock(&journal->j_flush_mutex);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001837 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838}
1839
1840/* for o_sync and fsync heavy applications, they tend to use
1841** all the journa list slots with tiny transactions. These
1842** trigger lots and lots of calls to update the header block, which
1843** adds seeks and slows things down.
1844**
1845** This function tries to clear out a large chunk of the journal lists
1846** at once, which makes everything faster since only the newest journal
1847** list updates the header block
1848*/
1849static int flush_used_journal_lists(struct super_block *s,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001850 struct reiserfs_journal_list *jl)
1851{
1852 unsigned long len = 0;
1853 unsigned long cur_len;
1854 int ret;
1855 int i;
1856 int limit = 256;
1857 struct reiserfs_journal_list *tjl;
1858 struct reiserfs_journal_list *flush_jl;
Jeff Mahoney600ed412009-03-30 14:02:17 -04001859 unsigned int trans_id;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001860 struct reiserfs_journal *journal = SB_JOURNAL(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001862 flush_jl = tjl = jl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001864 /* in data logging mode, try harder to flush a lot of blocks */
1865 if (reiserfs_data_log(s))
1866 limit = 1024;
1867 /* flush for 256 transactions or limit blocks, whichever comes first */
1868 for (i = 0; i < 256 && len < limit; i++) {
1869 if (atomic_read(&tjl->j_commit_left) ||
1870 tjl->j_trans_id < jl->j_trans_id) {
1871 break;
1872 }
1873 cur_len = atomic_read(&tjl->j_nonzerolen);
1874 if (cur_len > 0) {
1875 tjl->j_state &= ~LIST_TOUCHED;
1876 }
1877 len += cur_len;
1878 flush_jl = tjl;
1879 if (tjl->j_list.next == &journal->j_journal_list)
1880 break;
1881 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001883 /* try to find a group of blocks we can flush across all the
1884 ** transactions, but only bother if we've actually spanned
1885 ** across multiple lists
1886 */
1887 if (flush_jl != jl) {
1888 ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001890 flush_journal_list(s, flush_jl, 1);
1891 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892}
1893
1894/*
1895** removes any nodes in table with name block and dev as bh.
1896** only touchs the hnext and hprev pointers.
1897*/
1898void remove_journal_hash(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001899 struct reiserfs_journal_cnode **table,
1900 struct reiserfs_journal_list *jl,
1901 unsigned long block, int remove_freed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001903 struct reiserfs_journal_cnode *cur;
1904 struct reiserfs_journal_cnode **head;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001906 head = &(journal_hash(table, sb, block));
1907 if (!head) {
1908 return;
1909 }
1910 cur = *head;
1911 while (cur) {
1912 if (cur->blocknr == block && cur->sb == sb
1913 && (jl == NULL || jl == cur->jlist)
1914 && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1915 if (cur->hnext) {
1916 cur->hnext->hprev = cur->hprev;
1917 }
1918 if (cur->hprev) {
1919 cur->hprev->hnext = cur->hnext;
1920 } else {
1921 *head = cur->hnext;
1922 }
1923 cur->blocknr = 0;
1924 cur->sb = NULL;
1925 cur->state = 0;
1926 if (cur->bh && cur->jlist) /* anybody who clears the cur->bh will also dec the nonzerolen */
1927 atomic_dec(&(cur->jlist->j_nonzerolen));
1928 cur->bh = NULL;
1929 cur->jlist = NULL;
1930 }
1931 cur = cur->hnext;
1932 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933}
1934
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001935static void free_journal_ram(struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001936{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001937 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Pekka Enbergd739b422006-02-01 03:06:43 -08001938 kfree(journal->j_current_jl);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001939 journal->j_num_lists--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001941 vfree(journal->j_cnode_free_orig);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001942 free_list_bitmaps(sb, journal->j_list_bitmap);
1943 free_bitmap_nodes(sb); /* must be after free_list_bitmaps */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001944 if (journal->j_header_bh) {
1945 brelse(journal->j_header_bh);
1946 }
1947 /* j_header_bh is on the journal dev, make sure not to release the journal
1948 * dev until we brelse j_header_bh
1949 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001950 release_journal_dev(sb, journal);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001951 vfree(journal);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952}
1953
1954/*
1955** call on unmount. Only set error to 1 if you haven't made your way out
1956** of read_super() yet. Any other caller must keep error at 0.
1957*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001958static int do_journal_release(struct reiserfs_transaction_handle *th,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001959 struct super_block *sb, int error)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001960{
1961 struct reiserfs_transaction_handle myth;
1962 int flushed = 0;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001963 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001965 /* we only want to flush out transactions if we were called with error == 0
1966 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001967 if (!error && !(sb->s_flags & MS_RDONLY)) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001968 /* end the current trans */
1969 BUG_ON(!th->t_trans_id);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001970 do_journal_end(th, sb, 10, FLUSH_ALL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001972 /* make sure something gets logged to force our way into the flush code */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001973 if (!journal_join(&myth, sb, 1)) {
1974 reiserfs_prepare_for_journal(sb,
1975 SB_BUFFER_WITH_SB(sb),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001976 1);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001977 journal_mark_dirty(&myth, sb,
1978 SB_BUFFER_WITH_SB(sb));
1979 do_journal_end(&myth, sb, 1, FLUSH_ALL);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001980 flushed = 1;
1981 }
1982 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001984 /* this also catches errors during the do_journal_end above */
1985 if (!error && reiserfs_is_journal_aborted(journal)) {
1986 memset(&myth, 0, sizeof(myth));
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001987 if (!journal_join_abort(&myth, sb, 1)) {
1988 reiserfs_prepare_for_journal(sb,
1989 SB_BUFFER_WITH_SB(sb),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001990 1);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001991 journal_mark_dirty(&myth, sb,
1992 SB_BUFFER_WITH_SB(sb));
1993 do_journal_end(&myth, sb, 1, FLUSH_ALL);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001994 }
1995 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001997 reiserfs_mounted_fs_count--;
1998 /* wait for all commits to finish */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001999 cancel_delayed_work(&SB_JOURNAL(sb)->j_work);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02002000
2001 /*
2002 * We must release the write lock here because
2003 * the workqueue job (flush_async_commit) needs this lock
2004 */
2005 reiserfs_write_unlock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002006 flush_workqueue(commit_wq);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02002007
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002008 if (!reiserfs_mounted_fs_count) {
2009 destroy_workqueue(commit_wq);
2010 commit_wq = NULL;
2011 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002013 free_journal_ram(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014
Frederic Weisbecker05236762009-12-30 05:56:08 +01002015 reiserfs_write_lock(sb);
2016
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002017 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018}
2019
2020/*
2021** call on unmount. flush all journal trans, release all alloc'd ram
2022*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002023int journal_release(struct reiserfs_transaction_handle *th,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002024 struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002025{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002026 return do_journal_release(th, sb, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027}
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002028
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029/*
2030** only call from an error condition inside reiserfs_read_super!
2031*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002032int journal_release_error(struct reiserfs_transaction_handle *th,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002033 struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002034{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002035 return do_journal_release(th, sb, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036}
2037
2038/* compares description block with commit block. returns 1 if they differ, 0 if they are the same */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002039static int journal_compare_desc_commit(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002040 struct reiserfs_journal_desc *desc,
2041 struct reiserfs_journal_commit *commit)
2042{
2043 if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
2044 get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002045 get_commit_trans_len(commit) > SB_JOURNAL(sb)->j_trans_max ||
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002046 get_commit_trans_len(commit) <= 0) {
2047 return 1;
2048 }
2049 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050}
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002051
Jeff Mahoney0222e652009-03-30 14:02:44 -04002052/* returns 0 if it did not find a description block
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053** returns -1 if it found a corrupt commit block
Jeff Mahoney0222e652009-03-30 14:02:44 -04002054** returns 1 if both desc and commit were valid
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002056static int journal_transaction_is_valid(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002057 struct buffer_head *d_bh,
Jeff Mahoney600ed412009-03-30 14:02:17 -04002058 unsigned int *oldest_invalid_trans_id,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002059 unsigned long *newest_mount_id)
2060{
2061 struct reiserfs_journal_desc *desc;
2062 struct reiserfs_journal_commit *commit;
2063 struct buffer_head *c_bh;
2064 unsigned long offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002066 if (!d_bh)
2067 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002069 desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2070 if (get_desc_trans_len(desc) > 0
2071 && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
2072 if (oldest_invalid_trans_id && *oldest_invalid_trans_id
2073 && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002074 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002075 "journal-986: transaction "
2076 "is valid returning because trans_id %d is greater than "
2077 "oldest_invalid %lu",
2078 get_desc_trans_id(desc),
2079 *oldest_invalid_trans_id);
2080 return 0;
2081 }
2082 if (newest_mount_id
2083 && *newest_mount_id > get_desc_mount_id(desc)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002084 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002085 "journal-1087: transaction "
2086 "is valid returning because mount_id %d is less than "
2087 "newest_mount_id %lu",
2088 get_desc_mount_id(desc),
2089 *newest_mount_id);
2090 return -1;
2091 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002092 if (get_desc_trans_len(desc) > SB_JOURNAL(sb)->j_trans_max) {
2093 reiserfs_warning(sb, "journal-2018",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002094 "Bad transaction length %d "
2095 "encountered, ignoring transaction",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002096 get_desc_trans_len(desc));
2097 return -1;
2098 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002099 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002101 /* ok, we have a journal description block, lets see if the transaction was valid */
2102 c_bh =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002103 journal_bread(sb,
2104 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002105 ((offset + get_desc_trans_len(desc) +
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002106 1) % SB_ONDISK_JOURNAL_SIZE(sb)));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002107 if (!c_bh)
2108 return 0;
2109 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002110 if (journal_compare_desc_commit(sb, desc, commit)) {
2111 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002112 "journal_transaction_is_valid, commit offset %ld had bad "
2113 "time %d or length %d",
2114 c_bh->b_blocknr -
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002115 SB_ONDISK_JOURNAL_1st_BLOCK(sb),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002116 get_commit_trans_id(commit),
2117 get_commit_trans_len(commit));
2118 brelse(c_bh);
2119 if (oldest_invalid_trans_id) {
2120 *oldest_invalid_trans_id =
2121 get_desc_trans_id(desc);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002122 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002123 "journal-1004: "
2124 "transaction_is_valid setting oldest invalid trans_id "
2125 "to %d",
2126 get_desc_trans_id(desc));
2127 }
2128 return -1;
2129 }
2130 brelse(c_bh);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002131 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002132 "journal-1006: found valid "
2133 "transaction start offset %llu, len %d id %d",
2134 d_bh->b_blocknr -
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002135 SB_ONDISK_JOURNAL_1st_BLOCK(sb),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002136 get_desc_trans_len(desc),
2137 get_desc_trans_id(desc));
2138 return 1;
2139 } else {
2140 return 0;
2141 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142}
2143
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002144static void brelse_array(struct buffer_head **heads, int num)
2145{
2146 int i;
2147 for (i = 0; i < num; i++) {
2148 brelse(heads[i]);
2149 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150}
2151
2152/*
2153** given the start, and values for the oldest acceptable transactions,
2154** this either reads in a replays a transaction, or returns because the transaction
2155** is invalid, or too old.
2156*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002157static int journal_read_transaction(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002158 unsigned long cur_dblock,
2159 unsigned long oldest_start,
Jeff Mahoney600ed412009-03-30 14:02:17 -04002160 unsigned int oldest_trans_id,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002161 unsigned long newest_mount_id)
2162{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002163 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002164 struct reiserfs_journal_desc *desc;
2165 struct reiserfs_journal_commit *commit;
Jeff Mahoney600ed412009-03-30 14:02:17 -04002166 unsigned int trans_id = 0;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002167 struct buffer_head *c_bh;
2168 struct buffer_head *d_bh;
2169 struct buffer_head **log_blocks = NULL;
2170 struct buffer_head **real_blocks = NULL;
Jeff Mahoney600ed412009-03-30 14:02:17 -04002171 unsigned int trans_offset;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002172 int i;
2173 int trans_half;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002175 d_bh = journal_bread(sb, cur_dblock);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002176 if (!d_bh)
2177 return 1;
2178 desc = (struct reiserfs_journal_desc *)d_bh->b_data;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002179 trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2180 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1037: "
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002181 "journal_read_transaction, offset %llu, len %d mount_id %d",
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002182 d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002183 get_desc_trans_len(desc), get_desc_mount_id(desc));
2184 if (get_desc_trans_id(desc) < oldest_trans_id) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002185 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1039: "
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002186 "journal_read_trans skipping because %lu is too old",
2187 cur_dblock -
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002188 SB_ONDISK_JOURNAL_1st_BLOCK(sb));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002189 brelse(d_bh);
2190 return 1;
2191 }
2192 if (get_desc_mount_id(desc) != newest_mount_id) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002193 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1146: "
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002194 "journal_read_trans skipping because %d is != "
2195 "newest_mount_id %lu", get_desc_mount_id(desc),
2196 newest_mount_id);
2197 brelse(d_bh);
2198 return 1;
2199 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002200 c_bh = journal_bread(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002201 ((trans_offset + get_desc_trans_len(desc) + 1) %
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002202 SB_ONDISK_JOURNAL_SIZE(sb)));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002203 if (!c_bh) {
2204 brelse(d_bh);
2205 return 1;
2206 }
2207 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002208 if (journal_compare_desc_commit(sb, desc, commit)) {
2209 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002210 "journal_read_transaction, "
2211 "commit offset %llu had bad time %d or length %d",
2212 c_bh->b_blocknr -
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002213 SB_ONDISK_JOURNAL_1st_BLOCK(sb),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002214 get_commit_trans_id(commit),
2215 get_commit_trans_len(commit));
2216 brelse(c_bh);
2217 brelse(d_bh);
2218 return 1;
2219 }
2220 trans_id = get_desc_trans_id(desc);
2221 /* now we know we've got a good transaction, and it was inside the valid time ranges */
Pekka Enbergd739b422006-02-01 03:06:43 -08002222 log_blocks = kmalloc(get_desc_trans_len(desc) *
2223 sizeof(struct buffer_head *), GFP_NOFS);
2224 real_blocks = kmalloc(get_desc_trans_len(desc) *
2225 sizeof(struct buffer_head *), GFP_NOFS);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002226 if (!log_blocks || !real_blocks) {
2227 brelse(c_bh);
2228 brelse(d_bh);
Pekka Enbergd739b422006-02-01 03:06:43 -08002229 kfree(log_blocks);
2230 kfree(real_blocks);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002231 reiserfs_warning(sb, "journal-1169",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002232 "kmalloc failed, unable to mount FS");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002233 return -1;
2234 }
2235 /* get all the buffer heads */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002236 trans_half = journal_trans_half(sb->s_blocksize);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002237 for (i = 0; i < get_desc_trans_len(desc); i++) {
2238 log_blocks[i] =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002239 journal_getblk(sb,
2240 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002241 (trans_offset + 1 +
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002242 i) % SB_ONDISK_JOURNAL_SIZE(sb));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002243 if (i < trans_half) {
2244 real_blocks[i] =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002245 sb_getblk(sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002246 le32_to_cpu(desc->j_realblock[i]));
2247 } else {
2248 real_blocks[i] =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002249 sb_getblk(sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002250 le32_to_cpu(commit->
2251 j_realblock[i - trans_half]));
2252 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002253 if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(sb)) {
2254 reiserfs_warning(sb, "journal-1207",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002255 "REPLAY FAILURE fsck required! "
2256 "Block to replay is outside of "
2257 "filesystem");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002258 goto abort_replay;
2259 }
2260 /* make sure we don't try to replay onto log or reserved area */
2261 if (is_block_in_log_or_reserved_area
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002262 (sb, real_blocks[i]->b_blocknr)) {
2263 reiserfs_warning(sb, "journal-1204",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002264 "REPLAY FAILURE fsck required! "
2265 "Trying to replay onto a log block");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002266 abort_replay:
2267 brelse_array(log_blocks, i);
2268 brelse_array(real_blocks, i);
2269 brelse(c_bh);
2270 brelse(d_bh);
Pekka Enbergd739b422006-02-01 03:06:43 -08002271 kfree(log_blocks);
2272 kfree(real_blocks);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002273 return -1;
2274 }
2275 }
2276 /* read in the log blocks, memcpy to the corresponding real block */
2277 ll_rw_block(READ, get_desc_trans_len(desc), log_blocks);
2278 for (i = 0; i < get_desc_trans_len(desc); i++) {
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02002279
2280 reiserfs_write_unlock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002281 wait_on_buffer(log_blocks[i]);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02002282 reiserfs_write_lock(sb);
2283
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002284 if (!buffer_uptodate(log_blocks[i])) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002285 reiserfs_warning(sb, "journal-1212",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002286 "REPLAY FAILURE fsck required! "
2287 "buffer write failed");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002288 brelse_array(log_blocks + i,
2289 get_desc_trans_len(desc) - i);
2290 brelse_array(real_blocks, get_desc_trans_len(desc));
2291 brelse(c_bh);
2292 brelse(d_bh);
Pekka Enbergd739b422006-02-01 03:06:43 -08002293 kfree(log_blocks);
2294 kfree(real_blocks);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002295 return -1;
2296 }
2297 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2298 real_blocks[i]->b_size);
2299 set_buffer_uptodate(real_blocks[i]);
2300 brelse(log_blocks[i]);
2301 }
2302 /* flush out the real blocks */
2303 for (i = 0; i < get_desc_trans_len(desc); i++) {
2304 set_buffer_dirty(real_blocks[i]);
Jan Kara53778ff2005-09-06 15:19:14 -07002305 ll_rw_block(SWRITE, 1, real_blocks + i);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002306 }
2307 for (i = 0; i < get_desc_trans_len(desc); i++) {
2308 wait_on_buffer(real_blocks[i]);
2309 if (!buffer_uptodate(real_blocks[i])) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002310 reiserfs_warning(sb, "journal-1226",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002311 "REPLAY FAILURE, fsck required! "
2312 "buffer write failed");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002313 brelse_array(real_blocks + i,
2314 get_desc_trans_len(desc) - i);
2315 brelse(c_bh);
2316 brelse(d_bh);
Pekka Enbergd739b422006-02-01 03:06:43 -08002317 kfree(log_blocks);
2318 kfree(real_blocks);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002319 return -1;
2320 }
2321 brelse(real_blocks[i]);
2322 }
2323 cur_dblock =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002324 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002325 ((trans_offset + get_desc_trans_len(desc) +
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002326 2) % SB_ONDISK_JOURNAL_SIZE(sb));
2327 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002328 "journal-1095: setting journal " "start to offset %ld",
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002329 cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002330
2331 /* init starting values for the first transaction, in case this is the last transaction to be replayed. */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002332 journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002333 journal->j_last_flush_trans_id = trans_id;
2334 journal->j_trans_id = trans_id + 1;
Alexander Zarochentseva44c94a2006-03-25 03:06:59 -08002335 /* check for trans_id overflow */
2336 if (journal->j_trans_id == 0)
2337 journal->j_trans_id = 10;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002338 brelse(c_bh);
2339 brelse(d_bh);
Pekka Enbergd739b422006-02-01 03:06:43 -08002340 kfree(log_blocks);
2341 kfree(real_blocks);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002342 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343}
2344
2345/* This function reads blocks starting from block and to max_block of bufsize
2346 size (but no more than BUFNR blocks at a time). This proved to improve
2347 mounting speed on self-rebuilding raid5 arrays at least.
2348 Right now it is only used from journal code. But later we might use it
2349 from other places.
2350 Note: Do not use journal_getblk/sb_getblk functions here! */
Jeff Mahoney3ee16672007-10-18 23:39:25 -07002351static struct buffer_head *reiserfs_breada(struct block_device *dev,
2352 b_blocknr_t block, int bufsize,
2353 b_blocknr_t max_block)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002355 struct buffer_head *bhlist[BUFNR];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356 unsigned int blocks = BUFNR;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002357 struct buffer_head *bh;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358 int i, j;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002359
2360 bh = __getblk(dev, block, bufsize);
2361 if (buffer_uptodate(bh))
2362 return (bh);
2363
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364 if (block + BUFNR > max_block) {
2365 blocks = max_block - block;
2366 }
2367 bhlist[0] = bh;
2368 j = 1;
2369 for (i = 1; i < blocks; i++) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002370 bh = __getblk(dev, block + i, bufsize);
2371 if (buffer_uptodate(bh)) {
2372 brelse(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373 break;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002374 } else
2375 bhlist[j++] = bh;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002377 ll_rw_block(READ, j, bhlist);
2378 for (i = 1; i < j; i++)
2379 brelse(bhlist[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380 bh = bhlist[0];
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002381 wait_on_buffer(bh);
2382 if (buffer_uptodate(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383 return bh;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002384 brelse(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 return NULL;
2386}
2387
2388/*
2389** read and replay the log
2390** on a clean unmount, the journal header's next unflushed pointer will be to an invalid
2391** transaction. This tests that before finding all the transactions in the log, which makes normal mount times fast.
2392**
2393** After a crash, this starts with the next unflushed transaction, and replays until it finds one too old, or invalid.
2394**
2395** On exit, it sets things up so the first transaction will work correctly.
2396*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002397static int journal_read(struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002398{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002399 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002400 struct reiserfs_journal_desc *desc;
Jeff Mahoney600ed412009-03-30 14:02:17 -04002401 unsigned int oldest_trans_id = 0;
2402 unsigned int oldest_invalid_trans_id = 0;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002403 time_t start;
2404 unsigned long oldest_start = 0;
2405 unsigned long cur_dblock = 0;
2406 unsigned long newest_mount_id = 9;
2407 struct buffer_head *d_bh;
2408 struct reiserfs_journal_header *jh;
2409 int valid_journal_header = 0;
2410 int replay_count = 0;
2411 int continue_replay = 1;
2412 int ret;
2413 char b[BDEVNAME_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002415 cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2416 reiserfs_info(sb, "checking transaction log (%s)\n",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002417 bdevname(journal->j_dev_bd, b));
2418 start = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419
Jeff Mahoney0222e652009-03-30 14:02:44 -04002420 /* step 1, read in the journal header block. Check the transaction it says
2421 ** is the first unflushed, and if that transaction is not valid,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002422 ** replay is done
2423 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002424 journal->j_header_bh = journal_bread(sb,
2425 SB_ONDISK_JOURNAL_1st_BLOCK(sb)
2426 + SB_ONDISK_JOURNAL_SIZE(sb));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002427 if (!journal->j_header_bh) {
2428 return 1;
2429 }
2430 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
Vladimir V. Savelievc499ec22006-03-02 02:54:39 -08002431 if (le32_to_cpu(jh->j_first_unflushed_offset) <
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002432 SB_ONDISK_JOURNAL_SIZE(sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002433 && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2434 oldest_start =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002435 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002436 le32_to_cpu(jh->j_first_unflushed_offset);
2437 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2438 newest_mount_id = le32_to_cpu(jh->j_mount_id);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002439 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002440 "journal-1153: found in "
2441 "header: first_unflushed_offset %d, last_flushed_trans_id "
2442 "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2443 le32_to_cpu(jh->j_last_flush_trans_id));
2444 valid_journal_header = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445
Jeff Mahoney0222e652009-03-30 14:02:44 -04002446 /* now, we try to read the first unflushed offset. If it is not valid,
2447 ** there is nothing more we can do, and it makes no sense to read
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002448 ** through the whole log.
2449 */
2450 d_bh =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002451 journal_bread(sb,
2452 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002453 le32_to_cpu(jh->j_first_unflushed_offset));
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002454 ret = journal_transaction_is_valid(sb, d_bh, NULL, NULL);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002455 if (!ret) {
2456 continue_replay = 0;
2457 }
2458 brelse(d_bh);
2459 goto start_log_replay;
2460 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002462 if (continue_replay && bdev_read_only(sb->s_bdev)) {
2463 reiserfs_warning(sb, "clm-2076",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002464 "device is readonly, unable to replay log");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002465 return -1;
2466 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002468 /* ok, there are transactions that need to be replayed. start with the first log block, find
2469 ** all the valid transactions, and pick out the oldest.
2470 */
2471 while (continue_replay
2472 && cur_dblock <
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002473 (SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2474 SB_ONDISK_JOURNAL_SIZE(sb))) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002475 /* Note that it is required for blocksize of primary fs device and journal
2476 device to be the same */
2477 d_bh =
2478 reiserfs_breada(journal->j_dev_bd, cur_dblock,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002479 sb->s_blocksize,
2480 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2481 SB_ONDISK_JOURNAL_SIZE(sb));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002482 ret =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002483 journal_transaction_is_valid(sb, d_bh,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002484 &oldest_invalid_trans_id,
2485 &newest_mount_id);
2486 if (ret == 1) {
2487 desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2488 if (oldest_start == 0) { /* init all oldest_ values */
2489 oldest_trans_id = get_desc_trans_id(desc);
2490 oldest_start = d_bh->b_blocknr;
2491 newest_mount_id = get_desc_mount_id(desc);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002492 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002493 "journal-1179: Setting "
2494 "oldest_start to offset %llu, trans_id %lu",
2495 oldest_start -
2496 SB_ONDISK_JOURNAL_1st_BLOCK
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002497 (sb), oldest_trans_id);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002498 } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2499 /* one we just read was older */
2500 oldest_trans_id = get_desc_trans_id(desc);
2501 oldest_start = d_bh->b_blocknr;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002502 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002503 "journal-1180: Resetting "
2504 "oldest_start to offset %lu, trans_id %lu",
2505 oldest_start -
2506 SB_ONDISK_JOURNAL_1st_BLOCK
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002507 (sb), oldest_trans_id);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002508 }
2509 if (newest_mount_id < get_desc_mount_id(desc)) {
2510 newest_mount_id = get_desc_mount_id(desc);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002511 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002512 "journal-1299: Setting "
2513 "newest_mount_id to %d",
2514 get_desc_mount_id(desc));
2515 }
2516 cur_dblock += get_desc_trans_len(desc) + 2;
2517 } else {
2518 cur_dblock++;
2519 }
2520 brelse(d_bh);
2521 }
2522
2523 start_log_replay:
2524 cur_dblock = oldest_start;
2525 if (oldest_trans_id) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002526 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002527 "journal-1206: Starting replay "
2528 "from offset %llu, trans_id %lu",
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002529 cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002530 oldest_trans_id);
2531
2532 }
2533 replay_count = 0;
2534 while (continue_replay && oldest_trans_id > 0) {
2535 ret =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002536 journal_read_transaction(sb, cur_dblock, oldest_start,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002537 oldest_trans_id, newest_mount_id);
2538 if (ret < 0) {
2539 return ret;
2540 } else if (ret != 0) {
2541 break;
2542 }
2543 cur_dblock =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002544 SB_ONDISK_JOURNAL_1st_BLOCK(sb) + journal->j_start;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002545 replay_count++;
2546 if (cur_dblock == oldest_start)
2547 break;
2548 }
2549
2550 if (oldest_trans_id == 0) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002551 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002552 "journal-1225: No valid " "transactions found");
2553 }
2554 /* j_start does not get set correctly if we don't replay any transactions.
2555 ** if we had a valid journal_header, set j_start to the first unflushed transaction value,
2556 ** copy the trans_id from the header
2557 */
2558 if (valid_journal_header && replay_count == 0) {
2559 journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2560 journal->j_trans_id =
2561 le32_to_cpu(jh->j_last_flush_trans_id) + 1;
Alexander Zarochentseva44c94a2006-03-25 03:06:59 -08002562 /* check for trans_id overflow */
2563 if (journal->j_trans_id == 0)
2564 journal->j_trans_id = 10;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002565 journal->j_last_flush_trans_id =
2566 le32_to_cpu(jh->j_last_flush_trans_id);
2567 journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2568 } else {
2569 journal->j_mount_id = newest_mount_id + 1;
2570 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002571 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002572 "newest_mount_id to %lu", journal->j_mount_id);
2573 journal->j_first_unflushed_offset = journal->j_start;
2574 if (replay_count > 0) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002575 reiserfs_info(sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002576 "replayed %d transactions in %lu seconds\n",
2577 replay_count, get_seconds() - start);
2578 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002579 if (!bdev_read_only(sb->s_bdev) &&
2580 _update_journal_header_block(sb, journal->j_start,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002581 journal->j_last_flush_trans_id)) {
2582 /* replay failed, caller must call free_journal_ram and abort
2583 ** the mount
2584 */
2585 return -1;
2586 }
2587 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588}
2589
2590static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2591{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002592 struct reiserfs_journal_list *jl;
Pekka Enberg8c777cc2006-02-01 03:06:43 -08002593 jl = kzalloc(sizeof(struct reiserfs_journal_list),
2594 GFP_NOFS | __GFP_NOFAIL);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002595 INIT_LIST_HEAD(&jl->j_list);
2596 INIT_LIST_HEAD(&jl->j_working_list);
2597 INIT_LIST_HEAD(&jl->j_tail_bh_list);
2598 INIT_LIST_HEAD(&jl->j_bh_list);
Jeff Mahoney90415de2008-07-25 01:46:40 -07002599 mutex_init(&jl->j_commit_mutex);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002600 SB_JOURNAL(s)->j_num_lists++;
2601 get_journal_list(jl);
2602 return jl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603}
2604
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002605static void journal_list_init(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002606{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002607 SB_JOURNAL(sb)->j_current_jl = alloc_journal_list(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608}
2609
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002610static int release_journal_dev(struct super_block *super,
2611 struct reiserfs_journal *journal)
2612{
2613 int result;
2614
2615 result = 0;
2616
Christoph Hellwig86098fa2008-04-30 00:54:46 -07002617 if (journal->j_dev_bd != NULL) {
2618 if (journal->j_dev_bd->bd_dev != super->s_dev)
2619 bd_release(journal->j_dev_bd);
Al Viroe5eb8ca2007-10-08 13:24:05 -04002620 result = blkdev_put(journal->j_dev_bd, journal->j_dev_mode);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002621 journal->j_dev_bd = NULL;
2622 }
2623
2624 if (result != 0) {
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002625 reiserfs_warning(super, "sh-457",
2626 "Cannot release journal device: %i", result);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002627 }
2628 return result;
2629}
2630
2631static int journal_init_dev(struct super_block *super,
2632 struct reiserfs_journal *journal,
2633 const char *jdev_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634{
2635 int result;
2636 dev_t jdev;
Al Viroaeb5d722008-09-02 15:28:45 -04002637 fmode_t blkdev_mode = FMODE_READ | FMODE_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638 char b[BDEVNAME_SIZE];
2639
2640 result = 0;
2641
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002642 journal->j_dev_bd = NULL;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002643 jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2644 new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645
2646 if (bdev_read_only(super->s_bdev))
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002647 blkdev_mode = FMODE_READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648
2649 /* there is no "jdev" option and journal is on separate device */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002650 if ((!jdev_name || !jdev_name[0])) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651 journal->j_dev_bd = open_by_devnum(jdev, blkdev_mode);
Al Viroe5eb8ca2007-10-08 13:24:05 -04002652 journal->j_dev_mode = blkdev_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002653 if (IS_ERR(journal->j_dev_bd)) {
2654 result = PTR_ERR(journal->j_dev_bd);
2655 journal->j_dev_bd = NULL;
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002656 reiserfs_warning(super, "sh-458",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002657 "cannot init journal device '%s': %i",
2658 __bdevname(jdev, b), result);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659 return result;
Christoph Hellwig86098fa2008-04-30 00:54:46 -07002660 } else if (jdev != super->s_dev) {
2661 result = bd_claim(journal->j_dev_bd, journal);
2662 if (result) {
Al Viro9a1c3542008-02-22 20:40:24 -05002663 blkdev_put(journal->j_dev_bd, blkdev_mode);
Christoph Hellwig86098fa2008-04-30 00:54:46 -07002664 return result;
2665 }
2666
Linus Torvalds1da177e2005-04-16 15:20:36 -07002667 set_blocksize(journal->j_dev_bd, super->s_blocksize);
Christoph Hellwig86098fa2008-04-30 00:54:46 -07002668 }
2669
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670 return 0;
2671 }
2672
Al Viroe5eb8ca2007-10-08 13:24:05 -04002673 journal->j_dev_mode = blkdev_mode;
Al Viro30c40d22008-02-22 19:50:45 -05002674 journal->j_dev_bd = open_bdev_exclusive(jdev_name,
Al Viroe5eb8ca2007-10-08 13:24:05 -04002675 blkdev_mode, journal);
Christoph Hellwig86098fa2008-04-30 00:54:46 -07002676 if (IS_ERR(journal->j_dev_bd)) {
2677 result = PTR_ERR(journal->j_dev_bd);
2678 journal->j_dev_bd = NULL;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002679 reiserfs_warning(super,
2680 "journal_init_dev: Cannot open '%s': %i",
2681 jdev_name, result);
Christoph Hellwig86098fa2008-04-30 00:54:46 -07002682 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002683 }
Christoph Hellwig86098fa2008-04-30 00:54:46 -07002684
2685 set_blocksize(journal->j_dev_bd, super->s_blocksize);
2686 reiserfs_info(super,
2687 "journal_init_dev: journal device: %s\n",
2688 bdevname(journal->j_dev_bd, b));
2689 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002690}
2691
Edward Shishkincf3d0b82007-10-16 23:30:30 -07002692/**
2693 * When creating/tuning a file system user can assign some
2694 * journal params within boundaries which depend on the ratio
2695 * blocksize/standard_blocksize.
2696 *
2697 * For blocks >= standard_blocksize transaction size should
2698 * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more
2699 * then JOURNAL_TRANS_MAX_DEFAULT.
2700 *
2701 * For blocks < standard_blocksize these boundaries should be
2702 * decreased proportionally.
2703 */
2704#define REISERFS_STANDARD_BLKSIZE (4096)
2705
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002706static int check_advise_trans_params(struct super_block *sb,
Edward Shishkincf3d0b82007-10-16 23:30:30 -07002707 struct reiserfs_journal *journal)
2708{
2709 if (journal->j_trans_max) {
2710 /* Non-default journal params.
2711 Do sanity check for them. */
2712 int ratio = 1;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002713 if (sb->s_blocksize < REISERFS_STANDARD_BLKSIZE)
2714 ratio = REISERFS_STANDARD_BLKSIZE / sb->s_blocksize;
Edward Shishkincf3d0b82007-10-16 23:30:30 -07002715
2716 if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio ||
2717 journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio ||
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002718 SB_ONDISK_JOURNAL_SIZE(sb) / journal->j_trans_max <
Edward Shishkincf3d0b82007-10-16 23:30:30 -07002719 JOURNAL_MIN_RATIO) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002720 reiserfs_warning(sb, "sh-462",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002721 "bad transaction max size (%u). "
2722 "FSCK?", journal->j_trans_max);
Edward Shishkincf3d0b82007-10-16 23:30:30 -07002723 return 1;
2724 }
2725 if (journal->j_max_batch != (journal->j_trans_max) *
2726 JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002727 reiserfs_warning(sb, "sh-463",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002728 "bad transaction max batch (%u). "
2729 "FSCK?", journal->j_max_batch);
Edward Shishkincf3d0b82007-10-16 23:30:30 -07002730 return 1;
2731 }
2732 } else {
2733 /* Default journal params.
2734 The file system was created by old version
2735 of mkreiserfs, so some fields contain zeros,
2736 and we need to advise proper values for them */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002737 if (sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) {
2738 reiserfs_warning(sb, "sh-464", "bad blocksize (%u)",
2739 sb->s_blocksize);
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002740 return 1;
2741 }
Edward Shishkincf3d0b82007-10-16 23:30:30 -07002742 journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2743 journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2744 journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2745 }
2746 return 0;
2747}
2748
Linus Torvalds1da177e2005-04-16 15:20:36 -07002749/*
2750** must be called once on fs mount. calls journal_read for you
2751*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002752int journal_init(struct super_block *sb, const char *j_dev_name,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002753 int old_format, unsigned int commit_max_age)
2754{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002755 int num_cnodes = SB_ONDISK_JOURNAL_SIZE(sb) * 2;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002756 struct buffer_head *bhjh;
2757 struct reiserfs_super_block *rs;
2758 struct reiserfs_journal_header *jh;
2759 struct reiserfs_journal *journal;
2760 struct reiserfs_journal_list *jl;
2761 char b[BDEVNAME_SIZE];
Frederic Weisbecker98ea3f52009-12-29 21:51:15 +01002762 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002763
Frederic Weisbecker98ea3f52009-12-29 21:51:15 +01002764 /*
2765 * Unlock here to avoid various RECLAIM-FS-ON <-> IN-RECLAIM-FS
2766 * dependency inversion warnings.
2767 */
2768 reiserfs_write_unlock(sb);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002769 journal = SB_JOURNAL(sb) = vmalloc(sizeof(struct reiserfs_journal));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002770 if (!journal) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002771 reiserfs_warning(sb, "journal-1256",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002772 "unable to get memory for journal structure");
Frederic Weisbecker98ea3f52009-12-29 21:51:15 +01002773 reiserfs_write_lock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002774 return 1;
2775 }
2776 memset(journal, 0, sizeof(struct reiserfs_journal));
2777 INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2778 INIT_LIST_HEAD(&journal->j_prealloc_list);
2779 INIT_LIST_HEAD(&journal->j_working_list);
2780 INIT_LIST_HEAD(&journal->j_journal_list);
2781 journal->j_persistent_trans = 0;
Frederic Weisbecker98ea3f52009-12-29 21:51:15 +01002782 ret = reiserfs_allocate_list_bitmaps(sb, journal->j_list_bitmap,
2783 reiserfs_bmap_count(sb));
2784 reiserfs_write_lock(sb);
2785 if (ret)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002786 goto free_and_return;
Frederic Weisbecker98ea3f52009-12-29 21:51:15 +01002787
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002788 allocate_bitmap_nodes(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002789
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002790 /* reserved for journal area support */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002791 SB_JOURNAL_1st_RESERVED_BLOCK(sb) = (old_format ?
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002792 REISERFS_OLD_DISK_OFFSET_IN_BYTES
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002793 / sb->s_blocksize +
2794 reiserfs_bmap_count(sb) +
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002795 1 :
2796 REISERFS_DISK_OFFSET_IN_BYTES /
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002797 sb->s_blocksize + 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002798
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002799 /* Sanity check to see is the standard journal fitting withing first bitmap
2800 (actual for small blocksizes) */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002801 if (!SB_ONDISK_JOURNAL_DEVICE(sb) &&
2802 (SB_JOURNAL_1st_RESERVED_BLOCK(sb) +
2803 SB_ONDISK_JOURNAL_SIZE(sb) > sb->s_blocksize * 8)) {
2804 reiserfs_warning(sb, "journal-1393",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002805 "journal does not fit for area addressed "
2806 "by first of bitmap blocks. It starts at "
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002807 "%u and its size is %u. Block size %ld",
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002808 SB_JOURNAL_1st_RESERVED_BLOCK(sb),
2809 SB_ONDISK_JOURNAL_SIZE(sb),
2810 sb->s_blocksize);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002811 goto free_and_return;
2812 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002813
Frederic Weisbecker193be0e2009-09-17 05:31:37 +02002814 /*
2815 * We need to unlock here to avoid creating the following
2816 * dependency:
2817 * reiserfs_lock -> sysfs_mutex
2818 * Because the reiserfs mmap path creates the following dependency:
2819 * mm->mmap -> reiserfs_lock, hence we have
2820 * mm->mmap -> reiserfs_lock ->sysfs_mutex
2821 * This would ends up in a circular dependency with sysfs readdir path
2822 * which does sysfs_mutex -> mm->mmap_sem
2823 * This is fine because the reiserfs lock is useless in mount path,
2824 * at least until we call journal_begin. We keep it for paranoid
2825 * reasons.
2826 */
2827 reiserfs_write_unlock(sb);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002828 if (journal_init_dev(sb, journal, j_dev_name) != 0) {
Frederic Weisbecker193be0e2009-09-17 05:31:37 +02002829 reiserfs_write_lock(sb);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002830 reiserfs_warning(sb, "sh-462",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002831 "unable to initialize jornal device");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002832 goto free_and_return;
2833 }
Frederic Weisbecker193be0e2009-09-17 05:31:37 +02002834 reiserfs_write_lock(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002835
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002836 rs = SB_DISK_SUPER_BLOCK(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002838 /* read journal header */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002839 bhjh = journal_bread(sb,
2840 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2841 SB_ONDISK_JOURNAL_SIZE(sb));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002842 if (!bhjh) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002843 reiserfs_warning(sb, "sh-459",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002844 "unable to read journal header");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002845 goto free_and_return;
2846 }
2847 jh = (struct reiserfs_journal_header *)(bhjh->b_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002848
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002849 /* make sure that journal matches to the super block */
2850 if (is_reiserfs_jr(rs)
2851 && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2852 sb_jp_journal_magic(rs))) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002853 reiserfs_warning(sb, "sh-460",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002854 "journal header magic %x (device %s) does "
2855 "not match to magic found in super block %x",
2856 jh->jh_journal.jp_journal_magic,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002857 bdevname(journal->j_dev_bd, b),
2858 sb_jp_journal_magic(rs));
2859 brelse(bhjh);
2860 goto free_and_return;
2861 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002863 journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2864 journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2865 journal->j_max_commit_age =
2866 le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2867 journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002869 if (check_advise_trans_params(sb, journal) != 0)
Edward Shishkincf3d0b82007-10-16 23:30:30 -07002870 goto free_and_return;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002871 journal->j_default_max_commit_age = journal->j_max_commit_age;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002873 if (commit_max_age != 0) {
2874 journal->j_max_commit_age = commit_max_age;
2875 journal->j_max_trans_age = commit_max_age;
2876 }
2877
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002878 reiserfs_info(sb, "journal params: device %s, size %u, "
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002879 "journal first block %u, max trans len %u, max batch %u, "
2880 "max commit age %u, max trans age %u\n",
2881 bdevname(journal->j_dev_bd, b),
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002882 SB_ONDISK_JOURNAL_SIZE(sb),
2883 SB_ONDISK_JOURNAL_1st_BLOCK(sb),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002884 journal->j_trans_max,
2885 journal->j_max_batch,
2886 journal->j_max_commit_age, journal->j_max_trans_age);
2887
2888 brelse(bhjh);
2889
2890 journal->j_list_bitmap_index = 0;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002891 journal_list_init(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002892
2893 memset(journal->j_list_hash_table, 0,
2894 JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2895
2896 INIT_LIST_HEAD(&journal->j_dirty_buffers);
2897 spin_lock_init(&journal->j_dirty_buffers_lock);
2898
2899 journal->j_start = 0;
2900 journal->j_len = 0;
2901 journal->j_len_alloc = 0;
2902 atomic_set(&(journal->j_wcount), 0);
2903 atomic_set(&(journal->j_async_throttle), 0);
2904 journal->j_bcount = 0;
2905 journal->j_trans_start_time = 0;
2906 journal->j_last = NULL;
2907 journal->j_first = NULL;
2908 init_waitqueue_head(&(journal->j_join_wait));
Jeff Mahoneyf68215c2008-07-25 01:46:38 -07002909 mutex_init(&journal->j_mutex);
Jeff Mahoneyafe70252008-07-25 01:46:39 -07002910 mutex_init(&journal->j_flush_mutex);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002911
2912 journal->j_trans_id = 10;
2913 journal->j_mount_id = 10;
2914 journal->j_state = 0;
2915 atomic_set(&(journal->j_jlock), 0);
Frederic Weisbeckerbbec9192010-01-28 13:43:50 +01002916 reiserfs_write_unlock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002917 journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
Frederic Weisbeckerbbec9192010-01-28 13:43:50 +01002918 reiserfs_write_lock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002919 journal->j_cnode_free_orig = journal->j_cnode_free_list;
2920 journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2921 journal->j_cnode_used = 0;
2922 journal->j_must_wait = 0;
2923
Jeff Mahoney576f6d72005-11-29 19:34:39 -08002924 if (journal->j_cnode_free == 0) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002925 reiserfs_warning(sb, "journal-2004", "Journal cnode memory "
Jeff Mahoney576f6d72005-11-29 19:34:39 -08002926 "allocation failed (%ld bytes). Journal is "
2927 "too large for available memory. Usually "
2928 "this is due to a journal that is too large.",
2929 sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2930 goto free_and_return;
2931 }
2932
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002933 init_journal_hash(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002934 jl = journal->j_current_jl;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002935 jl->j_list_bitmap = get_list_bitmap(sb, jl);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002936 if (!jl->j_list_bitmap) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002937 reiserfs_warning(sb, "journal-2005",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002938 "get_list_bitmap failed for journal list 0");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002939 goto free_and_return;
2940 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002941 if (journal_read(sb) < 0) {
2942 reiserfs_warning(sb, "reiserfs-2006",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002943 "Replay Failure, unable to mount");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002944 goto free_and_return;
2945 }
2946
2947 reiserfs_mounted_fs_count++;
Frederic Weisbecker48f6ba52009-10-05 16:31:37 +02002948 if (reiserfs_mounted_fs_count <= 1) {
2949 reiserfs_write_unlock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002950 commit_wq = create_workqueue("reiserfs");
Frederic Weisbecker48f6ba52009-10-05 16:31:37 +02002951 reiserfs_write_lock(sb);
2952 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002953
David Howellsc4028952006-11-22 14:57:56 +00002954 INIT_DELAYED_WORK(&journal->j_work, flush_async_commits);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002955 journal->j_work_sb = sb;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002956 return 0;
2957 free_and_return:
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002958 free_journal_ram(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002959 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002960}
2961
2962/*
2963** test for a polite end of the current transaction. Used by file_write, and should
2964** be used by delete to make sure they don't write more than can fit inside a single
2965** transaction
2966*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002967int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2968 int new_alloc)
2969{
2970 struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2971 time_t now = get_seconds();
2972 /* cannot restart while nested */
2973 BUG_ON(!th->t_trans_id);
2974 if (th->t_refcount > 1)
2975 return 0;
2976 if (journal->j_must_wait > 0 ||
2977 (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2978 atomic_read(&(journal->j_jlock)) ||
2979 (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2980 journal->j_cnode_free < (journal->j_trans_max * 3)) {
2981 return 1;
2982 }
Chris Mason6ae1ea42006-02-01 03:06:50 -08002983 /* protected by the BKL here */
2984 journal->j_len_alloc += new_alloc;
2985 th->t_blocks_allocated += new_alloc ;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002986 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002987}
2988
Jeff Mahoney0222e652009-03-30 14:02:44 -04002989/* this must be called inside a transaction, and requires the
Linus Torvalds1da177e2005-04-16 15:20:36 -07002990** kernel_lock to be held
2991*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002992void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002993{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002994 struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2995 BUG_ON(!th->t_trans_id);
2996 journal->j_must_wait = 1;
2997 set_bit(J_WRITERS_BLOCKED, &journal->j_state);
2998 return;
2999}
3000
3001/* this must be called without a transaction started, and does not
3002** require BKL
3003*/
3004void reiserfs_allow_writes(struct super_block *s)
3005{
3006 struct reiserfs_journal *journal = SB_JOURNAL(s);
3007 clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
3008 wake_up(&journal->j_join_wait);
3009}
3010
3011/* this must be called without a transaction started, and does not
3012** require BKL
3013*/
3014void reiserfs_wait_on_write_block(struct super_block *s)
3015{
3016 struct reiserfs_journal *journal = SB_JOURNAL(s);
3017 wait_event(journal->j_join_wait,
3018 !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
3019}
3020
3021static void queue_log_writer(struct super_block *s)
3022{
3023 wait_queue_t wait;
3024 struct reiserfs_journal *journal = SB_JOURNAL(s);
3025 set_bit(J_WRITERS_QUEUED, &journal->j_state);
3026
3027 /*
3028 * we don't want to use wait_event here because
3029 * we only want to wait once.
3030 */
3031 init_waitqueue_entry(&wait, current);
3032 add_wait_queue(&journal->j_join_wait, &wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003033 set_current_state(TASK_UNINTERRUPTIBLE);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02003034 if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) {
3035 reiserfs_write_unlock(s);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003036 schedule();
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02003037 reiserfs_write_lock(s);
3038 }
Milind Arun Choudhary5ab2f7e2007-05-08 00:30:51 -07003039 __set_current_state(TASK_RUNNING);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003040 remove_wait_queue(&journal->j_join_wait, &wait);
3041}
3042
3043static void wake_queued_writers(struct super_block *s)
3044{
3045 struct reiserfs_journal *journal = SB_JOURNAL(s);
3046 if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
3047 wake_up(&journal->j_join_wait);
3048}
3049
Jeff Mahoney600ed412009-03-30 14:02:17 -04003050static void let_transaction_grow(struct super_block *sb, unsigned int trans_id)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003051{
3052 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3053 unsigned long bcount = journal->j_bcount;
3054 while (1) {
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02003055 reiserfs_write_unlock(sb);
Nishanth Aravamudan041e0e32005-09-10 00:27:23 -07003056 schedule_timeout_uninterruptible(1);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02003057 reiserfs_write_lock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003058 journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
3059 while ((atomic_read(&journal->j_wcount) > 0 ||
3060 atomic_read(&journal->j_jlock)) &&
3061 journal->j_trans_id == trans_id) {
3062 queue_log_writer(sb);
3063 }
3064 if (journal->j_trans_id != trans_id)
3065 break;
3066 if (bcount == journal->j_bcount)
3067 break;
3068 bcount = journal->j_bcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003069 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003070}
3071
3072/* join == true if you must join an existing transaction.
3073** join == false if you can deal with waiting for others to finish
3074**
3075** this will block until the transaction is joinable. send the number of blocks you
3076** expect to use in nblocks.
3077*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003078static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003079 struct super_block *sb, unsigned long nblocks,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003080 int join)
3081{
3082 time_t now = get_seconds();
Jeff Mahoney600ed412009-03-30 14:02:17 -04003083 unsigned int old_trans_id;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003084 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003085 struct reiserfs_transaction_handle myth;
3086 int sched_count = 0;
3087 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003088
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003089 reiserfs_check_lock_depth(sb, "journal_begin");
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003090 BUG_ON(nblocks > journal->j_trans_max);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003091
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003092 PROC_INFO_INC(sb, journal.journal_being);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003093 /* set here for journal_join */
3094 th->t_refcount = 1;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003095 th->t_super = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003096
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003097 relock:
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003098 lock_journal(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003099 if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003100 unlock_journal(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003101 retval = journal->j_errno;
3102 goto out_fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003103 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003104 journal->j_bcount++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003105
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003106 if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003107 unlock_journal(sb);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02003108 reiserfs_write_unlock(sb);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003109 reiserfs_wait_on_write_block(sb);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02003110 reiserfs_write_lock(sb);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003111 PROC_INFO_INC(sb, journal.journal_relock_writers);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003112 goto relock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003113 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003114 now = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003115
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003116 /* if there is no room in the journal OR
Jeff Mahoney0222e652009-03-30 14:02:44 -04003117 ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003118 ** we don't sleep if there aren't other writers
3119 */
3120
3121 if ((!join && journal->j_must_wait > 0) ||
3122 (!join
3123 && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
3124 || (!join && atomic_read(&journal->j_wcount) > 0
3125 && journal->j_trans_start_time > 0
3126 && (now - journal->j_trans_start_time) >
3127 journal->j_max_trans_age) || (!join
3128 && atomic_read(&journal->j_jlock))
3129 || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
3130
3131 old_trans_id = journal->j_trans_id;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003132 unlock_journal(sb); /* allow others to finish this transaction */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003133
3134 if (!join && (journal->j_len_alloc + nblocks + 2) >=
3135 journal->j_max_batch &&
3136 ((journal->j_len + nblocks + 2) * 100) <
3137 (journal->j_len_alloc * 75)) {
3138 if (atomic_read(&journal->j_wcount) > 10) {
3139 sched_count++;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003140 queue_log_writer(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003141 goto relock;
3142 }
3143 }
3144 /* don't mess with joining the transaction if all we have to do is
3145 * wait for someone else to do a commit
3146 */
3147 if (atomic_read(&journal->j_jlock)) {
3148 while (journal->j_trans_id == old_trans_id &&
3149 atomic_read(&journal->j_jlock)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003150 queue_log_writer(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003151 }
3152 goto relock;
3153 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003154 retval = journal_join(&myth, sb, 1);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003155 if (retval)
3156 goto out_fail;
3157
3158 /* someone might have ended the transaction while we joined */
3159 if (old_trans_id != journal->j_trans_id) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003160 retval = do_journal_end(&myth, sb, 1, 0);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003161 } else {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003162 retval = do_journal_end(&myth, sb, 1, COMMIT_NOW);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003163 }
3164
3165 if (retval)
3166 goto out_fail;
3167
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003168 PROC_INFO_INC(sb, journal.journal_relock_wcount);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003169 goto relock;
3170 }
3171 /* we are the first writer, set trans_id */
3172 if (journal->j_trans_start_time == 0) {
3173 journal->j_trans_start_time = get_seconds();
3174 }
3175 atomic_inc(&(journal->j_wcount));
3176 journal->j_len_alloc += nblocks;
3177 th->t_blocks_logged = 0;
3178 th->t_blocks_allocated = nblocks;
3179 th->t_trans_id = journal->j_trans_id;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003180 unlock_journal(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003181 INIT_LIST_HEAD(&th->t_list);
3182 get_fs_excl();
3183 return 0;
3184
3185 out_fail:
3186 memset(th, 0, sizeof(*th));
3187 /* Re-set th->t_super, so we can properly keep track of how many
3188 * persistent transactions there are. We need to do this so if this
3189 * call is part of a failed restart_transaction, we can free it later */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003190 th->t_super = sb;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003191 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003192}
3193
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003194struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3195 super_block
3196 *s,
3197 int nblocks)
3198{
3199 int ret;
3200 struct reiserfs_transaction_handle *th;
3201
3202 /* if we're nesting into an existing transaction. It will be
3203 ** persistent on its own
3204 */
3205 if (reiserfs_transaction_running(s)) {
3206 th = current->journal_info;
3207 th->t_refcount++;
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003208 BUG_ON(th->t_refcount < 2);
3209
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003210 return th;
3211 }
Pekka Enbergd739b422006-02-01 03:06:43 -08003212 th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003213 if (!th)
3214 return NULL;
3215 ret = journal_begin(th, s, nblocks);
3216 if (ret) {
Pekka Enbergd739b422006-02-01 03:06:43 -08003217 kfree(th);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003218 return NULL;
3219 }
3220
3221 SB_JOURNAL(s)->j_persistent_trans++;
3222 return th;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003223}
3224
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003225int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3226{
3227 struct super_block *s = th->t_super;
3228 int ret = 0;
3229 if (th->t_trans_id)
3230 ret = journal_end(th, th->t_super, th->t_blocks_allocated);
3231 else
3232 ret = -EIO;
3233 if (th->t_refcount == 0) {
3234 SB_JOURNAL(s)->j_persistent_trans--;
Pekka Enbergd739b422006-02-01 03:06:43 -08003235 kfree(th);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003236 }
3237 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003238}
3239
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003240static int journal_join(struct reiserfs_transaction_handle *th,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003241 struct super_block *sb, unsigned long nblocks)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003242{
3243 struct reiserfs_transaction_handle *cur_th = current->journal_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003244
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003245 /* this keeps do_journal_end from NULLing out the current->journal_info
3246 ** pointer
3247 */
3248 th->t_handle_save = cur_th;
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003249 BUG_ON(cur_th && cur_th->t_refcount > 1);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003250 return do_journal_begin_r(th, sb, nblocks, JBEGIN_JOIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003251}
3252
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003253int journal_join_abort(struct reiserfs_transaction_handle *th,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003254 struct super_block *sb, unsigned long nblocks)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003255{
3256 struct reiserfs_transaction_handle *cur_th = current->journal_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003257
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003258 /* this keeps do_journal_end from NULLing out the current->journal_info
3259 ** pointer
3260 */
3261 th->t_handle_save = cur_th;
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003262 BUG_ON(cur_th && cur_th->t_refcount > 1);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003263 return do_journal_begin_r(th, sb, nblocks, JBEGIN_ABORT);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003264}
3265
3266int journal_begin(struct reiserfs_transaction_handle *th,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003267 struct super_block *sb, unsigned long nblocks)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003268{
3269 struct reiserfs_transaction_handle *cur_th = current->journal_info;
3270 int ret;
3271
3272 th->t_handle_save = NULL;
3273 if (cur_th) {
3274 /* we are nesting into the current transaction */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003275 if (cur_th->t_super == sb) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003276 BUG_ON(!cur_th->t_refcount);
3277 cur_th->t_refcount++;
3278 memcpy(th, cur_th, sizeof(*th));
3279 if (th->t_refcount <= 1)
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003280 reiserfs_warning(sb, "reiserfs-2005",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04003281 "BAD: refcount <= 1, but "
3282 "journal_info != 0");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003283 return 0;
3284 } else {
3285 /* we've ended up with a handle from a different filesystem.
3286 ** save it and restore on journal_end. This should never
3287 ** really happen...
3288 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003289 reiserfs_warning(sb, "clm-2100",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04003290 "nesting info a different FS");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003291 th->t_handle_save = current->journal_info;
3292 current->journal_info = th;
3293 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003294 } else {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003295 current->journal_info = th;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003296 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003297 ret = do_journal_begin_r(th, sb, nblocks, JBEGIN_REG);
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003298 BUG_ON(current->journal_info != th);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003299
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003300 /* I guess this boils down to being the reciprocal of clm-2100 above.
3301 * If do_journal_begin_r fails, we need to put it back, since journal_end
3302 * won't be called to do it. */
3303 if (ret)
3304 current->journal_info = th->t_handle_save;
3305 else
3306 BUG_ON(!th->t_refcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003307
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003308 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003309}
3310
3311/*
3312** puts bh into the current transaction. If it was already there, reorders removes the
3313** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order).
3314**
3315** if it was dirty, cleans and files onto the clean list. I can't let it be dirty again until the
3316** transaction is committed.
Jeff Mahoney0222e652009-03-30 14:02:44 -04003317**
Linus Torvalds1da177e2005-04-16 15:20:36 -07003318** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3319*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003320int journal_mark_dirty(struct reiserfs_transaction_handle *th,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003321 struct super_block *sb, struct buffer_head *bh)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003322{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003323 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003324 struct reiserfs_journal_cnode *cn = NULL;
3325 int count_already_incd = 0;
3326 int prepared = 0;
3327 BUG_ON(!th->t_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003328
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003329 PROC_INFO_INC(sb, journal.mark_dirty);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003330 if (th->t_trans_id != journal->j_trans_id) {
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04003331 reiserfs_panic(th->t_super, "journal-1577",
3332 "handle trans id %ld != current trans id %ld",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003333 th->t_trans_id, journal->j_trans_id);
3334 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003335
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003336 sb->s_dirt = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003337
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003338 prepared = test_clear_buffer_journal_prepared(bh);
3339 clear_buffer_journal_restore_dirty(bh);
3340 /* already in this transaction, we are done */
3341 if (buffer_journaled(bh)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003342 PROC_INFO_INC(sb, journal.mark_dirty_already);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003343 return 0;
3344 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003345
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003346 /* this must be turned into a panic instead of a warning. We can't allow
3347 ** a dirty or journal_dirty or locked buffer to be logged, as some changes
3348 ** could get to disk too early. NOT GOOD.
3349 */
3350 if (!prepared || buffer_dirty(bh)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003351 reiserfs_warning(sb, "journal-1777",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04003352 "buffer %llu bad state "
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003353 "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3354 (unsigned long long)bh->b_blocknr,
3355 prepared ? ' ' : '!',
3356 buffer_locked(bh) ? ' ' : '!',
3357 buffer_dirty(bh) ? ' ' : '!',
3358 buffer_journal_dirty(bh) ? ' ' : '!');
3359 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003360
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003361 if (atomic_read(&(journal->j_wcount)) <= 0) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003362 reiserfs_warning(sb, "journal-1409",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04003363 "returning because j_wcount was %d",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003364 atomic_read(&(journal->j_wcount)));
3365 return 1;
3366 }
Jeff Mahoney0222e652009-03-30 14:02:44 -04003367 /* this error means I've screwed up, and we've overflowed the transaction.
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003368 ** Nothing can be done here, except make the FS readonly or panic.
3369 */
3370 if (journal->j_len >= journal->j_trans_max) {
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04003371 reiserfs_panic(th->t_super, "journal-1413",
3372 "j_len (%lu) is too big",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003373 journal->j_len);
3374 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003375
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003376 if (buffer_journal_dirty(bh)) {
3377 count_already_incd = 1;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003378 PROC_INFO_INC(sb, journal.mark_dirty_notjournal);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003379 clear_buffer_journal_dirty(bh);
3380 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003381
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003382 if (journal->j_len > journal->j_len_alloc) {
3383 journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3384 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003385
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003386 set_buffer_journaled(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003387
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003388 /* now put this guy on the end */
3389 if (!cn) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003390 cn = get_cnode(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003391 if (!cn) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003392 reiserfs_panic(sb, "journal-4", "get_cnode failed!");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003393 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003394
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003395 if (th->t_blocks_logged == th->t_blocks_allocated) {
3396 th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3397 journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3398 }
3399 th->t_blocks_logged++;
3400 journal->j_len++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003401
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003402 cn->bh = bh;
3403 cn->blocknr = bh->b_blocknr;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003404 cn->sb = sb;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003405 cn->jlist = NULL;
3406 insert_journal_hash(journal->j_hash_table, cn);
3407 if (!count_already_incd) {
3408 get_bh(bh);
3409 }
3410 }
3411 cn->next = NULL;
3412 cn->prev = journal->j_last;
3413 cn->bh = bh;
3414 if (journal->j_last) {
3415 journal->j_last->next = cn;
3416 journal->j_last = cn;
3417 } else {
3418 journal->j_first = cn;
3419 journal->j_last = cn;
3420 }
3421 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003422}
3423
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003424int journal_end(struct reiserfs_transaction_handle *th,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003425 struct super_block *sb, unsigned long nblocks)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003426{
3427 if (!current->journal_info && th->t_refcount > 1)
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003428 reiserfs_warning(sb, "REISER-NESTING",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04003429 "th NULL, refcount %d", th->t_refcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003430
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003431 if (!th->t_trans_id) {
3432 WARN_ON(1);
3433 return -EIO;
3434 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003435
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003436 th->t_refcount--;
3437 if (th->t_refcount > 0) {
3438 struct reiserfs_transaction_handle *cur_th =
3439 current->journal_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003440
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003441 /* we aren't allowed to close a nested transaction on a different
3442 ** filesystem from the one in the task struct
3443 */
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003444 BUG_ON(cur_th->t_super != th->t_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003445
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003446 if (th != cur_th) {
3447 memcpy(current->journal_info, th, sizeof(*th));
3448 th->t_trans_id = 0;
3449 }
3450 return 0;
3451 } else {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003452 return do_journal_end(th, sb, nblocks, 0);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003453 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003454}
3455
Jeff Mahoney0222e652009-03-30 14:02:44 -04003456/* removes from the current transaction, relsing and descrementing any counters.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003457** also files the removed buffer directly onto the clean list
3458**
3459** called by journal_mark_freed when a block has been deleted
3460**
3461** returns 1 if it cleaned and relsed the buffer. 0 otherwise
3462*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003463static int remove_from_transaction(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003464 b_blocknr_t blocknr, int already_cleaned)
3465{
3466 struct buffer_head *bh;
3467 struct reiserfs_journal_cnode *cn;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003468 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003469 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003470
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003471 cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003472 if (!cn || !cn->bh) {
3473 return ret;
3474 }
3475 bh = cn->bh;
3476 if (cn->prev) {
3477 cn->prev->next = cn->next;
3478 }
3479 if (cn->next) {
3480 cn->next->prev = cn->prev;
3481 }
3482 if (cn == journal->j_first) {
3483 journal->j_first = cn->next;
3484 }
3485 if (cn == journal->j_last) {
3486 journal->j_last = cn->prev;
3487 }
3488 if (bh)
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003489 remove_journal_hash(sb, journal->j_hash_table, NULL,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003490 bh->b_blocknr, 0);
3491 clear_buffer_journaled(bh); /* don't log this one */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003492
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003493 if (!already_cleaned) {
3494 clear_buffer_journal_dirty(bh);
3495 clear_buffer_dirty(bh);
3496 clear_buffer_journal_test(bh);
3497 put_bh(bh);
3498 if (atomic_read(&(bh->b_count)) < 0) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003499 reiserfs_warning(sb, "journal-1752",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04003500 "b_count < 0");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003501 }
3502 ret = 1;
3503 }
3504 journal->j_len--;
3505 journal->j_len_alloc--;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003506 free_cnode(sb, cn);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003507 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003508}
3509
3510/*
3511** for any cnode in a journal list, it can only be dirtied of all the
Matt LaPlante0779bf22006-11-30 05:24:39 +01003512** transactions that include it are committed to disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003513** this checks through each transaction, and returns 1 if you are allowed to dirty,
3514** and 0 if you aren't
3515**
3516** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log
3517** blocks for a given transaction on disk
3518**
3519*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003520static int can_dirty(struct reiserfs_journal_cnode *cn)
3521{
3522 struct super_block *sb = cn->sb;
3523 b_blocknr_t blocknr = cn->blocknr;
3524 struct reiserfs_journal_cnode *cur = cn->hprev;
3525 int can_dirty = 1;
3526
3527 /* first test hprev. These are all newer than cn, so any node here
3528 ** with the same block number and dev means this node can't be sent
3529 ** to disk right now.
3530 */
3531 while (cur && can_dirty) {
3532 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3533 cur->blocknr == blocknr) {
3534 can_dirty = 0;
3535 }
3536 cur = cur->hprev;
3537 }
3538 /* then test hnext. These are all older than cn. As long as they
3539 ** are committed to the log, it is safe to write cn to disk
3540 */
3541 cur = cn->hnext;
3542 while (cur && can_dirty) {
3543 if (cur->jlist && cur->jlist->j_len > 0 &&
3544 atomic_read(&(cur->jlist->j_commit_left)) > 0 && cur->bh &&
3545 cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3546 can_dirty = 0;
3547 }
3548 cur = cur->hnext;
3549 }
3550 return can_dirty;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003551}
3552
3553/* syncs the commit blocks, but does not force the real buffers to disk
Jeff Mahoney0222e652009-03-30 14:02:44 -04003554** will wait until the current transaction is done/committed before returning
Linus Torvalds1da177e2005-04-16 15:20:36 -07003555*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003556int journal_end_sync(struct reiserfs_transaction_handle *th,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003557 struct super_block *sb, unsigned long nblocks)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003558{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003559 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003560
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003561 BUG_ON(!th->t_trans_id);
3562 /* you can sync while nested, very, very bad */
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003563 BUG_ON(th->t_refcount > 1);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003564 if (journal->j_len == 0) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003565 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003566 1);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003567 journal_mark_dirty(th, sb, SB_BUFFER_WITH_SB(sb));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003568 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003569 return do_journal_end(th, sb, nblocks, COMMIT_NOW | WAIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003570}
3571
3572/*
3573** writeback the pending async commits to disk
3574*/
David Howellsc4028952006-11-22 14:57:56 +00003575static void flush_async_commits(struct work_struct *work)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003576{
David Howellsc4028952006-11-22 14:57:56 +00003577 struct reiserfs_journal *journal =
3578 container_of(work, struct reiserfs_journal, j_work.work);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003579 struct super_block *sb = journal->j_work_sb;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003580 struct reiserfs_journal_list *jl;
3581 struct list_head *entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003582
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02003583 reiserfs_write_lock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003584 if (!list_empty(&journal->j_journal_list)) {
3585 /* last entry is the youngest, commit it and you get everything */
3586 entry = journal->j_journal_list.prev;
3587 jl = JOURNAL_LIST_ENTRY(entry);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003588 flush_commit_list(sb, jl, 1);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003589 }
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02003590 reiserfs_write_unlock(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003591}
3592
3593/*
3594** flushes any old transactions to disk
3595** ends the current transaction if it is too old
3596*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003597int reiserfs_flush_old_commits(struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003598{
3599 time_t now;
3600 struct reiserfs_transaction_handle th;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003601 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003602
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003603 now = get_seconds();
3604 /* safety check so we don't flush while we are replaying the log during
3605 * mount
3606 */
3607 if (list_empty(&journal->j_journal_list)) {
3608 return 0;
3609 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003610
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003611 /* check the current transaction. If there are no writers, and it is
3612 * too old, finish it, and force the commit blocks to disk
3613 */
3614 if (atomic_read(&journal->j_wcount) <= 0 &&
3615 journal->j_trans_start_time > 0 &&
3616 journal->j_len > 0 &&
3617 (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003618 if (!journal_join(&th, sb, 1)) {
3619 reiserfs_prepare_for_journal(sb,
3620 SB_BUFFER_WITH_SB(sb),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003621 1);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003622 journal_mark_dirty(&th, sb,
3623 SB_BUFFER_WITH_SB(sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003624
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003625 /* we're only being called from kreiserfsd, it makes no sense to do
3626 ** an async commit so that kreiserfsd can do it later
3627 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003628 do_journal_end(&th, sb, 1, COMMIT_NOW | WAIT);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003629 }
3630 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003631 return sb->s_dirt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003632}
3633
3634/*
3635** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit
Jeff Mahoney0222e652009-03-30 14:02:44 -04003636**
3637** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all
Linus Torvalds1da177e2005-04-16 15:20:36 -07003638** the writers are done. By the time it wakes up, the transaction it was called has already ended, so it just
3639** flushes the commit list and returns 0.
3640**
3641** Won't batch when flush or commit_now is set. Also won't batch when others are waiting on j_join_wait.
Jeff Mahoney0222e652009-03-30 14:02:44 -04003642**
Linus Torvalds1da177e2005-04-16 15:20:36 -07003643** Note, we can't allow the journal_end to proceed while there are still writers in the log.
3644*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003645static int check_journal_end(struct reiserfs_transaction_handle *th,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003646 struct super_block *sb, unsigned long nblocks,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003647 int flags)
3648{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003649
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003650 time_t now;
3651 int flush = flags & FLUSH_ALL;
3652 int commit_now = flags & COMMIT_NOW;
3653 int wait_on_commit = flags & WAIT;
3654 struct reiserfs_journal_list *jl;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003655 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003656
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003657 BUG_ON(!th->t_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003658
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003659 if (th->t_trans_id != journal->j_trans_id) {
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04003660 reiserfs_panic(th->t_super, "journal-1577",
3661 "handle trans id %ld != current trans id %ld",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003662 th->t_trans_id, journal->j_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003663 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003664
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003665 journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3666 if (atomic_read(&(journal->j_wcount)) > 0) { /* <= 0 is allowed. unmounting might not call begin */
3667 atomic_dec(&(journal->j_wcount));
3668 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003669
Jeff Mahoney0222e652009-03-30 14:02:44 -04003670 /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003671 ** will be dealt with by next transaction that actually writes something, but should be taken
3672 ** care of in this trans
3673 */
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003674 BUG_ON(journal->j_len == 0);
3675
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003676 /* if wcount > 0, and we are called to with flush or commit_now,
3677 ** we wait on j_join_wait. We will wake up when the last writer has
3678 ** finished the transaction, and started it on its way to the disk.
Jeff Mahoney0222e652009-03-30 14:02:44 -04003679 ** Then, we flush the commit or journal list, and just return 0
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003680 ** because the rest of journal end was already done for this transaction.
3681 */
3682 if (atomic_read(&(journal->j_wcount)) > 0) {
3683 if (flush || commit_now) {
3684 unsigned trans_id;
3685
3686 jl = journal->j_current_jl;
3687 trans_id = jl->j_trans_id;
3688 if (wait_on_commit)
3689 jl->j_state |= LIST_COMMIT_PENDING;
3690 atomic_set(&(journal->j_jlock), 1);
3691 if (flush) {
3692 journal->j_next_full_flush = 1;
3693 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003694 unlock_journal(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003695
3696 /* sleep while the current transaction is still j_jlocked */
3697 while (journal->j_trans_id == trans_id) {
3698 if (atomic_read(&journal->j_jlock)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003699 queue_log_writer(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003700 } else {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003701 lock_journal(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003702 if (journal->j_trans_id == trans_id) {
3703 atomic_set(&(journal->j_jlock),
3704 1);
3705 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003706 unlock_journal(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003707 }
3708 }
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003709 BUG_ON(journal->j_trans_id == trans_id);
3710
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003711 if (commit_now
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003712 && journal_list_still_alive(sb, trans_id)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003713 && wait_on_commit) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003714 flush_commit_list(sb, jl, 1);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003715 }
3716 return 0;
3717 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003718 unlock_journal(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003719 return 0;
3720 }
3721
3722 /* deal with old transactions where we are the last writers */
3723 now = get_seconds();
3724 if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3725 commit_now = 1;
3726 journal->j_next_async_flush = 1;
3727 }
3728 /* don't batch when someone is waiting on j_join_wait */
3729 /* don't batch when syncing the commit or flushing the whole trans */
3730 if (!(journal->j_must_wait > 0) && !(atomic_read(&(journal->j_jlock)))
3731 && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3732 && journal->j_len_alloc < journal->j_max_batch
3733 && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3734 journal->j_bcount++;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003735 unlock_journal(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003736 return 0;
3737 }
3738
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003739 if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(sb)) {
3740 reiserfs_panic(sb, "journal-003",
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04003741 "j_start (%ld) is too high",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003742 journal->j_start);
3743 }
3744 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003745}
3746
3747/*
3748** Does all the work that makes deleting blocks safe.
3749** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on.
Jeff Mahoney0222e652009-03-30 14:02:44 -04003750**
Linus Torvalds1da177e2005-04-16 15:20:36 -07003751** otherwise:
3752** set a bit for the block in the journal bitmap. That will prevent it from being allocated for unformatted nodes
3753** before this transaction has finished.
3754**
3755** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers. That will prevent any old transactions with
3756** this block from trying to flush to the real location. Since we aren't removing the cnode from the journal_list_hash,
3757** the block can't be reallocated yet.
3758**
3759** Then remove it from the current transaction, decrementing any counters and filing it on the clean list.
3760*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003761int journal_mark_freed(struct reiserfs_transaction_handle *th,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003762 struct super_block *sb, b_blocknr_t blocknr)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003763{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003764 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003765 struct reiserfs_journal_cnode *cn = NULL;
3766 struct buffer_head *bh = NULL;
3767 struct reiserfs_list_bitmap *jb = NULL;
3768 int cleaned = 0;
3769 BUG_ON(!th->t_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003770
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003771 cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003772 if (cn && cn->bh) {
3773 bh = cn->bh;
3774 get_bh(bh);
3775 }
3776 /* if it is journal new, we just remove it from this transaction */
3777 if (bh && buffer_journal_new(bh)) {
3778 clear_buffer_journal_new(bh);
3779 clear_prepared_bits(bh);
3780 reiserfs_clean_and_file_buffer(bh);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003781 cleaned = remove_from_transaction(sb, blocknr, cleaned);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003782 } else {
3783 /* set the bit for this block in the journal bitmap for this transaction */
3784 jb = journal->j_current_jl->j_list_bitmap;
3785 if (!jb) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003786 reiserfs_panic(sb, "journal-1702",
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04003787 "journal_list_bitmap is NULL");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003788 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003789 set_bit_in_list_bitmap(sb, blocknr, jb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003790
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003791 /* Note, the entire while loop is not allowed to schedule. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003792
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003793 if (bh) {
3794 clear_prepared_bits(bh);
3795 reiserfs_clean_and_file_buffer(bh);
3796 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003797 cleaned = remove_from_transaction(sb, blocknr, cleaned);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003798
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003799 /* find all older transactions with this block, make sure they don't try to write it out */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003800 cn = get_journal_hash_dev(sb, journal->j_list_hash_table,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003801 blocknr);
3802 while (cn) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003803 if (sb == cn->sb && blocknr == cn->blocknr) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003804 set_bit(BLOCK_FREED, &cn->state);
3805 if (cn->bh) {
3806 if (!cleaned) {
3807 /* remove_from_transaction will brelse the buffer if it was
3808 ** in the current trans
3809 */
3810 clear_buffer_journal_dirty(cn->
3811 bh);
3812 clear_buffer_dirty(cn->bh);
3813 clear_buffer_journal_test(cn->
3814 bh);
3815 cleaned = 1;
3816 put_bh(cn->bh);
3817 if (atomic_read
3818 (&(cn->bh->b_count)) < 0) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003819 reiserfs_warning(sb,
Jeff Mahoney45b03d52009-03-30 14:02:21 -04003820 "journal-2138",
3821 "cn->bh->b_count < 0");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003822 }
3823 }
3824 if (cn->jlist) { /* since we are clearing the bh, we MUST dec nonzerolen */
3825 atomic_dec(&
3826 (cn->jlist->
3827 j_nonzerolen));
3828 }
3829 cn->bh = NULL;
3830 }
3831 }
3832 cn = cn->hnext;
3833 }
3834 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003835
Chris Mason398c95b2007-10-16 23:29:44 -07003836 if (bh)
3837 release_buffer_page(bh); /* get_hash grabs the buffer */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003838 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003839}
3840
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003841void reiserfs_update_inode_transaction(struct inode *inode)
3842{
3843 struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3844 REISERFS_I(inode)->i_jl = journal->j_current_jl;
3845 REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003846}
3847
3848/*
3849 * returns -1 on error, 0 if no commits/barriers were done and 1
3850 * if a transaction was actually committed and the barrier was done
3851 */
3852static int __commit_trans_jl(struct inode *inode, unsigned long id,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003853 struct reiserfs_journal_list *jl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003854{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003855 struct reiserfs_transaction_handle th;
3856 struct super_block *sb = inode->i_sb;
3857 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3858 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003859
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003860 /* is it from the current transaction, or from an unknown transaction? */
3861 if (id == journal->j_trans_id) {
3862 jl = journal->j_current_jl;
3863 /* try to let other writers come in and grow this transaction */
3864 let_transaction_grow(sb, id);
3865 if (journal->j_trans_id != id) {
3866 goto flush_commit_only;
3867 }
3868
3869 ret = journal_begin(&th, sb, 1);
3870 if (ret)
3871 return ret;
3872
3873 /* someone might have ended this transaction while we joined */
3874 if (journal->j_trans_id != id) {
3875 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3876 1);
3877 journal_mark_dirty(&th, sb, SB_BUFFER_WITH_SB(sb));
3878 ret = journal_end(&th, sb, 1);
3879 goto flush_commit_only;
3880 }
3881
3882 ret = journal_end_sync(&th, sb, 1);
3883 if (!ret)
3884 ret = 1;
3885
3886 } else {
3887 /* this gets tricky, we have to make sure the journal list in
3888 * the inode still exists. We know the list is still around
3889 * if we've got a larger transaction id than the oldest list
3890 */
3891 flush_commit_only:
3892 if (journal_list_still_alive(inode->i_sb, id)) {
3893 /*
3894 * we only set ret to 1 when we know for sure
3895 * the barrier hasn't been started yet on the commit
3896 * block.
3897 */
3898 if (atomic_read(&jl->j_commit_left) > 1)
3899 ret = 1;
3900 flush_commit_list(sb, jl, 1);
3901 if (journal->j_errno)
3902 ret = journal->j_errno;
3903 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003904 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003905 /* otherwise the list is gone, and long since committed */
3906 return ret;
3907}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003908
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003909int reiserfs_commit_for_inode(struct inode *inode)
3910{
Jeff Mahoney600ed412009-03-30 14:02:17 -04003911 unsigned int id = REISERFS_I(inode)->i_trans_id;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003912 struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003913
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003914 /* for the whole inode, assume unset id means it was
3915 * changed in the current transaction. More conservative
Linus Torvalds1da177e2005-04-16 15:20:36 -07003916 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003917 if (!id || !jl) {
3918 reiserfs_update_inode_transaction(inode);
3919 id = REISERFS_I(inode)->i_trans_id;
3920 /* jl will be updated in __commit_trans_jl */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003921 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003922
3923 return __commit_trans_jl(inode, id, jl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003924}
3925
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003926void reiserfs_restore_prepared_buffer(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003927 struct buffer_head *bh)
3928{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003929 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3930 PROC_INFO_INC(sb, journal.restore_prepared);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003931 if (!bh) {
3932 return;
3933 }
3934 if (test_clear_buffer_journal_restore_dirty(bh) &&
3935 buffer_journal_dirty(bh)) {
3936 struct reiserfs_journal_cnode *cn;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003937 cn = get_journal_hash_dev(sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003938 journal->j_list_hash_table,
3939 bh->b_blocknr);
3940 if (cn && can_dirty(cn)) {
3941 set_buffer_journal_test(bh);
3942 mark_buffer_dirty(bh);
3943 }
3944 }
3945 clear_buffer_journal_prepared(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003946}
3947
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003948extern struct tree_balance *cur_tb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003949/*
3950** before we can change a metadata block, we have to make sure it won't
3951** be written to disk while we are altering it. So, we must:
3952** clean it
3953** wait on it.
Jeff Mahoney0222e652009-03-30 14:02:44 -04003954**
Linus Torvalds1da177e2005-04-16 15:20:36 -07003955*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003956int reiserfs_prepare_for_journal(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003957 struct buffer_head *bh, int wait)
3958{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003959 PROC_INFO_INC(sb, journal.prepare);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003960
Nick Pigginca5de402008-08-02 12:02:13 +02003961 if (!trylock_buffer(bh)) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003962 if (!wait)
3963 return 0;
3964 lock_buffer(bh);
3965 }
3966 set_buffer_journal_prepared(bh);
3967 if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3968 clear_buffer_journal_test(bh);
3969 set_buffer_journal_restore_dirty(bh);
3970 }
3971 unlock_buffer(bh);
3972 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003973}
3974
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003975static void flush_old_journal_lists(struct super_block *s)
3976{
3977 struct reiserfs_journal *journal = SB_JOURNAL(s);
3978 struct reiserfs_journal_list *jl;
3979 struct list_head *entry;
3980 time_t now = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003981
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003982 while (!list_empty(&journal->j_journal_list)) {
3983 entry = journal->j_journal_list.next;
3984 jl = JOURNAL_LIST_ENTRY(entry);
3985 /* this check should always be run, to send old lists to disk */
Chris Masona3172022006-09-29 01:59:56 -07003986 if (jl->j_timestamp < (now - (JOURNAL_MAX_TRANS_AGE * 4)) &&
3987 atomic_read(&jl->j_commit_left) == 0 &&
3988 test_transaction(s, jl)) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003989 flush_used_journal_lists(s, jl);
3990 } else {
3991 break;
3992 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003993 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003994}
3995
Jeff Mahoney0222e652009-03-30 14:02:44 -04003996/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003997** long and ugly. If flush, will not return until all commit
3998** blocks and all real buffers in the trans are on disk.
3999** If no_async, won't return until all commit blocks are on disk.
4000**
4001** keep reading, there are comments as you go along
4002**
4003** If the journal is aborted, we just clean up. Things like flushing
4004** journal lists, etc just won't happen.
4005*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004006static int do_journal_end(struct reiserfs_transaction_handle *th,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004007 struct super_block *sb, unsigned long nblocks,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004008 int flags)
4009{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004010 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004011 struct reiserfs_journal_cnode *cn, *next, *jl_cn;
4012 struct reiserfs_journal_cnode *last_cn = NULL;
4013 struct reiserfs_journal_desc *desc;
4014 struct reiserfs_journal_commit *commit;
4015 struct buffer_head *c_bh; /* commit bh */
4016 struct buffer_head *d_bh; /* desc bh */
4017 int cur_write_start = 0; /* start index of current log write */
4018 int old_start;
4019 int i;
Alexander Zarochentseva44c94a2006-03-25 03:06:59 -08004020 int flush;
4021 int wait_on_commit;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004022 struct reiserfs_journal_list *jl, *temp_jl;
4023 struct list_head *entry, *safe;
4024 unsigned long jindex;
Jeff Mahoney600ed412009-03-30 14:02:17 -04004025 unsigned int commit_trans_id;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004026 int trans_half;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004027
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004028 BUG_ON(th->t_refcount > 1);
4029 BUG_ON(!th->t_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004030
Alexander Zarochentseva44c94a2006-03-25 03:06:59 -08004031 /* protect flush_older_commits from doing mistakes if the
4032 transaction ID counter gets overflowed. */
Jeff Mahoney600ed412009-03-30 14:02:17 -04004033 if (th->t_trans_id == ~0U)
Alexander Zarochentseva44c94a2006-03-25 03:06:59 -08004034 flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
4035 flush = flags & FLUSH_ALL;
4036 wait_on_commit = flags & WAIT;
4037
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004038 put_fs_excl();
4039 current->journal_info = th->t_handle_save;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004040 reiserfs_check_lock_depth(sb, "journal end");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004041 if (journal->j_len == 0) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004042 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004043 1);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004044 journal_mark_dirty(th, sb, SB_BUFFER_WITH_SB(sb));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004045 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004046
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004047 lock_journal(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004048 if (journal->j_next_full_flush) {
4049 flags |= FLUSH_ALL;
4050 flush = 1;
4051 }
4052 if (journal->j_next_async_flush) {
4053 flags |= COMMIT_NOW | WAIT;
4054 wait_on_commit = 1;
4055 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004056
Jeff Mahoney0222e652009-03-30 14:02:44 -04004057 /* check_journal_end locks the journal, and unlocks if it does not return 1
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004058 ** it tells us if we should continue with the journal_end, or just return
4059 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004060 if (!check_journal_end(th, sb, nblocks, flags)) {
4061 sb->s_dirt = 1;
4062 wake_queued_writers(sb);
4063 reiserfs_async_progress_wait(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004064 goto out;
4065 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004066
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004067 /* check_journal_end might set these, check again */
4068 if (journal->j_next_full_flush) {
4069 flush = 1;
4070 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004071
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004072 /*
4073 ** j must wait means we have to flush the log blocks, and the real blocks for
4074 ** this transaction
4075 */
4076 if (journal->j_must_wait > 0) {
4077 flush = 1;
4078 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004079#ifdef REISERFS_PREALLOCATE
Jan Karaef43bc42006-01-11 12:17:40 -08004080 /* quota ops might need to nest, setup the journal_info pointer for them
4081 * and raise the refcount so that it is > 0. */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004082 current->journal_info = th;
Jan Karaef43bc42006-01-11 12:17:40 -08004083 th->t_refcount++;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004084 reiserfs_discard_all_prealloc(th); /* it should not involve new blocks into
4085 * the transaction */
Jan Karaef43bc42006-01-11 12:17:40 -08004086 th->t_refcount--;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004087 current->journal_info = th->t_handle_save;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004088#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004089
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004090 /* setup description block */
4091 d_bh =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004092 journal_getblk(sb,
4093 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004094 journal->j_start);
4095 set_buffer_uptodate(d_bh);
4096 desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
4097 memset(d_bh->b_data, 0, d_bh->b_size);
4098 memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
4099 set_desc_trans_id(desc, journal->j_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004100
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004101 /* setup commit block. Don't write (keep it clean too) this one until after everyone else is written */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004102 c_bh = journal_getblk(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004103 ((journal->j_start + journal->j_len +
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004104 1) % SB_ONDISK_JOURNAL_SIZE(sb)));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004105 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
4106 memset(c_bh->b_data, 0, c_bh->b_size);
4107 set_commit_trans_id(commit, journal->j_trans_id);
4108 set_buffer_uptodate(c_bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004109
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004110 /* init this journal list */
4111 jl = journal->j_current_jl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004112
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004113 /* we lock the commit before doing anything because
4114 * we want to make sure nobody tries to run flush_commit_list until
4115 * the new transaction is fully setup, and we've already flushed the
4116 * ordered bh list
4117 */
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02004118 reiserfs_mutex_lock_safe(&jl->j_commit_mutex, sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004119
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004120 /* save the transaction id in case we need to commit it later */
4121 commit_trans_id = jl->j_trans_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004122
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004123 atomic_set(&jl->j_older_commits_done, 0);
4124 jl->j_trans_id = journal->j_trans_id;
4125 jl->j_timestamp = journal->j_trans_start_time;
4126 jl->j_commit_bh = c_bh;
4127 jl->j_start = journal->j_start;
4128 jl->j_len = journal->j_len;
4129 atomic_set(&jl->j_nonzerolen, journal->j_len);
4130 atomic_set(&jl->j_commit_left, journal->j_len + 2);
4131 jl->j_realblock = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004132
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004133 /* The ENTIRE FOR LOOP MUST not cause schedule to occur.
4134 ** for each real block, add it to the journal list hash,
4135 ** copy into real block index array in the commit or desc block
4136 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004137 trans_half = journal_trans_half(sb->s_blocksize);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004138 for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
4139 if (buffer_journaled(cn->bh)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004140 jl_cn = get_cnode(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004141 if (!jl_cn) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004142 reiserfs_panic(sb, "journal-1676",
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04004143 "get_cnode returned NULL");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004144 }
4145 if (i == 0) {
4146 jl->j_realblock = jl_cn;
4147 }
4148 jl_cn->prev = last_cn;
4149 jl_cn->next = NULL;
4150 if (last_cn) {
4151 last_cn->next = jl_cn;
4152 }
4153 last_cn = jl_cn;
Jeff Mahoney0222e652009-03-30 14:02:44 -04004154 /* make sure the block we are trying to log is not a block
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004155 of journal or reserved area */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004156
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004157 if (is_block_in_log_or_reserved_area
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004158 (sb, cn->bh->b_blocknr)) {
4159 reiserfs_panic(sb, "journal-2332",
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04004160 "Trying to log block %lu, "
4161 "which is a log block",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004162 cn->bh->b_blocknr);
4163 }
4164 jl_cn->blocknr = cn->bh->b_blocknr;
4165 jl_cn->state = 0;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004166 jl_cn->sb = sb;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004167 jl_cn->bh = cn->bh;
4168 jl_cn->jlist = jl;
4169 insert_journal_hash(journal->j_list_hash_table, jl_cn);
4170 if (i < trans_half) {
4171 desc->j_realblock[i] =
4172 cpu_to_le32(cn->bh->b_blocknr);
4173 } else {
4174 commit->j_realblock[i - trans_half] =
4175 cpu_to_le32(cn->bh->b_blocknr);
4176 }
4177 } else {
4178 i--;
4179 }
4180 }
4181 set_desc_trans_len(desc, journal->j_len);
4182 set_desc_mount_id(desc, journal->j_mount_id);
4183 set_desc_trans_id(desc, journal->j_trans_id);
4184 set_commit_trans_len(commit, journal->j_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004185
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004186 /* special check in case all buffers in the journal were marked for not logging */
Eric Sesterhenn14a61442006-10-03 23:36:38 +02004187 BUG_ON(journal->j_len == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004188
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004189 /* we're about to dirty all the log blocks, mark the description block
4190 * dirty now too. Don't mark the commit block dirty until all the
4191 * others are on disk
4192 */
4193 mark_buffer_dirty(d_bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004194
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004195 /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */
4196 cur_write_start = journal->j_start;
4197 cn = journal->j_first;
4198 jindex = 1; /* start at one so we don't get the desc again */
4199 while (cn) {
4200 clear_buffer_journal_new(cn->bh);
4201 /* copy all the real blocks into log area. dirty log blocks */
4202 if (buffer_journaled(cn->bh)) {
4203 struct buffer_head *tmp_bh;
4204 char *addr;
4205 struct page *page;
4206 tmp_bh =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004207 journal_getblk(sb,
4208 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004209 ((cur_write_start +
4210 jindex) %
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004211 SB_ONDISK_JOURNAL_SIZE(sb)));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004212 set_buffer_uptodate(tmp_bh);
4213 page = cn->bh->b_page;
4214 addr = kmap(page);
4215 memcpy(tmp_bh->b_data,
4216 addr + offset_in_page(cn->bh->b_data),
4217 cn->bh->b_size);
4218 kunmap(page);
4219 mark_buffer_dirty(tmp_bh);
4220 jindex++;
4221 set_buffer_journal_dirty(cn->bh);
4222 clear_buffer_journaled(cn->bh);
4223 } else {
4224 /* JDirty cleared sometime during transaction. don't log this one */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004225 reiserfs_warning(sb, "journal-2048",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04004226 "BAD, buffer in journal hash, "
4227 "but not JDirty!");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004228 brelse(cn->bh);
4229 }
4230 next = cn->next;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004231 free_cnode(sb, cn);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004232 cn = next;
Frederic Weisbeckere6950a42009-04-30 23:04:32 +02004233 reiserfs_write_unlock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004234 cond_resched();
Frederic Weisbeckere6950a42009-04-30 23:04:32 +02004235 reiserfs_write_lock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004236 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004237
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004238 /* we are done with both the c_bh and d_bh, but
4239 ** c_bh must be written after all other commit blocks,
4240 ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4241 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004242
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004243 journal->j_current_jl = alloc_journal_list(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004244
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004245 /* now it is safe to insert this transaction on the main list */
4246 list_add_tail(&jl->j_list, &journal->j_journal_list);
4247 list_add_tail(&jl->j_working_list, &journal->j_working_list);
4248 journal->j_num_work_lists++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004249
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004250 /* reset journal values for the next transaction */
4251 old_start = journal->j_start;
4252 journal->j_start =
4253 (journal->j_start + journal->j_len +
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004254 2) % SB_ONDISK_JOURNAL_SIZE(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004255 atomic_set(&(journal->j_wcount), 0);
4256 journal->j_bcount = 0;
4257 journal->j_last = NULL;
4258 journal->j_first = NULL;
4259 journal->j_len = 0;
4260 journal->j_trans_start_time = 0;
Alexander Zarochentseva44c94a2006-03-25 03:06:59 -08004261 /* check for trans_id overflow */
4262 if (++journal->j_trans_id == 0)
4263 journal->j_trans_id = 10;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004264 journal->j_current_jl->j_trans_id = journal->j_trans_id;
4265 journal->j_must_wait = 0;
4266 journal->j_len_alloc = 0;
4267 journal->j_next_full_flush = 0;
4268 journal->j_next_async_flush = 0;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004269 init_journal_hash(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004270
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004271 // make sure reiserfs_add_jh sees the new current_jl before we
4272 // write out the tails
4273 smp_mb();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004274
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004275 /* tail conversion targets have to hit the disk before we end the
4276 * transaction. Otherwise a later transaction might repack the tail
4277 * before this transaction commits, leaving the data block unflushed and
4278 * clean, if we crash before the later transaction commits, the data block
4279 * is lost.
4280 */
4281 if (!list_empty(&jl->j_tail_bh_list)) {
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02004282 reiserfs_write_unlock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004283 write_ordered_buffers(&journal->j_dirty_buffers_lock,
4284 journal, jl, &jl->j_tail_bh_list);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02004285 reiserfs_write_lock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004286 }
Eric Sesterhenn14a61442006-10-03 23:36:38 +02004287 BUG_ON(!list_empty(&jl->j_tail_bh_list));
Jeff Mahoney90415de2008-07-25 01:46:40 -07004288 mutex_unlock(&jl->j_commit_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004289
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004290 /* honor the flush wishes from the caller, simple commits can
4291 ** be done outside the journal lock, they are done below
4292 **
4293 ** if we don't flush the commit list right now, we put it into
4294 ** the work queue so the people waiting on the async progress work
4295 ** queue don't wait for this proc to flush journal lists and such.
4296 */
4297 if (flush) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004298 flush_commit_list(sb, jl, 1);
4299 flush_journal_list(sb, jl, 1);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004300 } else if (!(jl->j_state & LIST_COMMIT_PENDING))
4301 queue_delayed_work(commit_wq, &journal->j_work, HZ / 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004302
Jeff Mahoney0222e652009-03-30 14:02:44 -04004303 /* if the next transaction has any chance of wrapping, flush
4304 ** transactions that might get overwritten. If any journal lists are very
4305 ** old flush them as well.
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004306 */
4307 first_jl:
4308 list_for_each_safe(entry, safe, &journal->j_journal_list) {
4309 temp_jl = JOURNAL_LIST_ENTRY(entry);
4310 if (journal->j_start <= temp_jl->j_start) {
4311 if ((journal->j_start + journal->j_trans_max + 1) >=
4312 temp_jl->j_start) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004313 flush_used_journal_lists(sb, temp_jl);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004314 goto first_jl;
4315 } else if ((journal->j_start +
4316 journal->j_trans_max + 1) <
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004317 SB_ONDISK_JOURNAL_SIZE(sb)) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004318 /* if we don't cross into the next transaction and we don't
4319 * wrap, there is no way we can overlap any later transactions
4320 * break now
4321 */
4322 break;
4323 }
4324 } else if ((journal->j_start +
4325 journal->j_trans_max + 1) >
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004326 SB_ONDISK_JOURNAL_SIZE(sb)) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004327 if (((journal->j_start + journal->j_trans_max + 1) %
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004328 SB_ONDISK_JOURNAL_SIZE(sb)) >=
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004329 temp_jl->j_start) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004330 flush_used_journal_lists(sb, temp_jl);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004331 goto first_jl;
4332 } else {
4333 /* we don't overlap anything from out start to the end of the
4334 * log, and our wrapped portion doesn't overlap anything at
4335 * the start of the log. We can break
4336 */
4337 break;
4338 }
4339 }
4340 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004341 flush_old_journal_lists(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004342
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004343 journal->j_current_jl->j_list_bitmap =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004344 get_list_bitmap(sb, journal->j_current_jl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004345
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004346 if (!(journal->j_current_jl->j_list_bitmap)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004347 reiserfs_panic(sb, "journal-1996",
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04004348 "could not get a list bitmap");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004349 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004350
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004351 atomic_set(&(journal->j_jlock), 0);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004352 unlock_journal(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004353 /* wake up any body waiting to join. */
4354 clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4355 wake_up(&(journal->j_join_wait));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004356
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004357 if (!flush && wait_on_commit &&
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004358 journal_list_still_alive(sb, commit_trans_id)) {
4359 flush_commit_list(sb, jl, 1);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004360 }
4361 out:
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004362 reiserfs_check_lock_depth(sb, "journal end2");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004363
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004364 memset(th, 0, sizeof(*th));
4365 /* Re-set th->t_super, so we can properly keep track of how many
4366 * persistent transactions there are. We need to do this so if this
4367 * call is part of a failed restart_transaction, we can free it later */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004368 th->t_super = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004369
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004370 return journal->j_errno;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004371}
4372
Jeff Mahoney32e8b102009-03-30 14:02:26 -04004373/* Send the file system read only and refuse new transactions */
4374void reiserfs_abort_journal(struct super_block *sb, int errno)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004375{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004376 struct reiserfs_journal *journal = SB_JOURNAL(sb);
4377 if (test_bit(J_ABORTED, &journal->j_state))
4378 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004379
Jeff Mahoney32e8b102009-03-30 14:02:26 -04004380 if (!journal->j_errno)
4381 journal->j_errno = errno;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004382
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004383 sb->s_flags |= MS_RDONLY;
4384 set_bit(J_ABORTED, &journal->j_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004385
4386#ifdef CONFIG_REISERFS_CHECK
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004387 dump_stack();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004388#endif
4389}
4390