blob: 3eea859e6990ca90601ba7fef043103b87f98745 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070046#include <linux/buffer_head.h>
47#include <linux/workqueue.h>
48#include <linux/writeback.h>
49#include <linux/blkdev.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070050#include <linux/backing-dev.h>
Jeff Mahoney90415de2008-07-25 01:46:40 -070051#include <linux/uaccess.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090052#include <linux/slab.h>
Jeff Mahoney90415de2008-07-25 01:46:40 -070053
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
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700140static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400141 *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700142{
143 struct reiserfs_bitmap_node *bn;
144 static int id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145
Pekka Enbergd739b422006-02-01 03:06:43 -0800146 bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700147 if (!bn) {
148 return NULL;
149 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400150 bn->data = kzalloc(sb->s_blocksize, GFP_NOFS);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700151 if (!bn->data) {
Pekka Enbergd739b422006-02-01 03:06:43 -0800152 kfree(bn);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700153 return NULL;
154 }
155 bn->id = id++;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700156 INIT_LIST_HEAD(&bn->list);
157 return bn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158}
159
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400160static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700161{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400162 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700163 struct reiserfs_bitmap_node *bn = NULL;
164 struct list_head *entry = journal->j_bitmap_nodes.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700166 journal->j_used_bitmap_nodes++;
167 repeat:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700169 if (entry != &journal->j_bitmap_nodes) {
170 bn = list_entry(entry, struct reiserfs_bitmap_node, list);
171 list_del(entry);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400172 memset(bn->data, 0, sb->s_blocksize);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700173 journal->j_free_bitmap_nodes--;
174 return bn;
175 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400176 bn = allocate_bitmap_node(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700177 if (!bn) {
178 yield();
179 goto repeat;
180 }
181 return bn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182}
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400183static inline void free_bitmap_node(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700184 struct reiserfs_bitmap_node *bn)
185{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400186 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700187 journal->j_used_bitmap_nodes--;
188 if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
Pekka Enbergd739b422006-02-01 03:06:43 -0800189 kfree(bn->data);
190 kfree(bn);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700191 } else {
192 list_add(&bn->list, &journal->j_bitmap_nodes);
193 journal->j_free_bitmap_nodes++;
194 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195}
196
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400197static void allocate_bitmap_nodes(struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700198{
199 int i;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400200 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700201 struct reiserfs_bitmap_node *bn = NULL;
202 for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400203 bn = allocate_bitmap_node(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700204 if (bn) {
205 list_add(&bn->list, &journal->j_bitmap_nodes);
206 journal->j_free_bitmap_nodes++;
207 } else {
Jeff Mahoney0222e652009-03-30 14:02:44 -0400208 break; /* this is ok, we'll try again when more are needed */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700209 }
210 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211}
212
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400213static int set_bit_in_list_bitmap(struct super_block *sb,
Jeff Mahoney3ee16672007-10-18 23:39:25 -0700214 b_blocknr_t block,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700215 struct reiserfs_list_bitmap *jb)
216{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400217 unsigned int bmap_nr = block / (sb->s_blocksize << 3);
218 unsigned int bit_nr = block % (sb->s_blocksize << 3);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700220 if (!jb->bitmaps[bmap_nr]) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400221 jb->bitmaps[bmap_nr] = get_bitmap_node(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700222 }
223 set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
224 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225}
226
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400227static void cleanup_bitmap_list(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700228 struct reiserfs_list_bitmap *jb)
229{
230 int i;
231 if (jb->bitmaps == NULL)
232 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400234 for (i = 0; i < reiserfs_bmap_count(sb); i++) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700235 if (jb->bitmaps[i]) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400236 free_bitmap_node(sb, jb->bitmaps[i]);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700237 jb->bitmaps[i] = NULL;
238 }
239 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240}
241
242/*
243** only call this on FS unmount.
244*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400245static int free_list_bitmaps(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700246 struct reiserfs_list_bitmap *jb_array)
247{
248 int i;
249 struct reiserfs_list_bitmap *jb;
250 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
251 jb = jb_array + i;
252 jb->journal_list = NULL;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400253 cleanup_bitmap_list(sb, jb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700254 vfree(jb->bitmaps);
255 jb->bitmaps = NULL;
256 }
257 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258}
259
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400260static int free_bitmap_nodes(struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700261{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400262 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700263 struct list_head *next = journal->j_bitmap_nodes.next;
264 struct reiserfs_bitmap_node *bn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700266 while (next != &journal->j_bitmap_nodes) {
267 bn = list_entry(next, struct reiserfs_bitmap_node, list);
268 list_del(next);
Pekka Enbergd739b422006-02-01 03:06:43 -0800269 kfree(bn->data);
270 kfree(bn);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700271 next = journal->j_bitmap_nodes.next;
272 journal->j_free_bitmap_nodes--;
273 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700275 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276}
277
278/*
Jeff Mahoney0222e652009-03-30 14:02:44 -0400279** get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280** jb_array is the array to be filled in.
281*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400282int reiserfs_allocate_list_bitmaps(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700283 struct reiserfs_list_bitmap *jb_array,
Jeff Mahoney3ee16672007-10-18 23:39:25 -0700284 unsigned int bmap_nr)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700285{
286 int i;
287 int failed = 0;
288 struct reiserfs_list_bitmap *jb;
289 int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700291 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
292 jb = jb_array + i;
293 jb->journal_list = NULL;
294 jb->bitmaps = vmalloc(mem);
295 if (!jb->bitmaps) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400296 reiserfs_warning(sb, "clm-2000", "unable to "
Jeff Mahoney45b03d52009-03-30 14:02:21 -0400297 "allocate bitmaps for journal lists");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700298 failed = 1;
299 break;
300 }
301 memset(jb->bitmaps, 0, mem);
302 }
303 if (failed) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400304 free_list_bitmaps(sb, jb_array);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700305 return -1;
306 }
307 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308}
309
310/*
Jeff Mahoney0222e652009-03-30 14:02:44 -0400311** find an available list bitmap. If you can't find one, flush a commit list
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312** and try again
313*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400314static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700315 struct reiserfs_journal_list
316 *jl)
317{
318 int i, j;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400319 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700320 struct reiserfs_list_bitmap *jb = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700322 for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) {
323 i = journal->j_list_bitmap_index;
324 journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS;
325 jb = journal->j_list_bitmap + i;
326 if (journal->j_list_bitmap[i].journal_list) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400327 flush_commit_list(sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700328 journal->j_list_bitmap[i].
329 journal_list, 1);
330 if (!journal->j_list_bitmap[i].journal_list) {
331 break;
332 }
333 } else {
334 break;
335 }
336 }
337 if (jb->journal_list) { /* double check to make sure if flushed correctly */
338 return NULL;
339 }
340 jb->journal_list = jl;
341 return jb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342}
343
Jeff Mahoney0222e652009-03-30 14:02:44 -0400344/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345** allocates a new chunk of X nodes, and links them all together as a list.
346** Uses the cnode->next and cnode->prev pointers
347** returns NULL on failure
348*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700349static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
350{
351 struct reiserfs_journal_cnode *head;
352 int i;
353 if (num_cnodes <= 0) {
354 return NULL;
355 }
356 head = vmalloc(num_cnodes * sizeof(struct reiserfs_journal_cnode));
357 if (!head) {
358 return NULL;
359 }
360 memset(head, 0, num_cnodes * sizeof(struct reiserfs_journal_cnode));
361 head[0].prev = NULL;
362 head[0].next = head + 1;
363 for (i = 1; i < num_cnodes; i++) {
364 head[i].prev = head + (i - 1);
365 head[i].next = head + (i + 1); /* if last one, overwrite it after the if */
366 }
367 head[num_cnodes - 1].next = NULL;
368 return head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369}
370
371/*
Jeff Mahoney0222e652009-03-30 14:02:44 -0400372** pulls a cnode off the free list, or returns NULL on failure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400374static struct reiserfs_journal_cnode *get_cnode(struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700375{
376 struct reiserfs_journal_cnode *cn;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400377 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400379 reiserfs_check_lock_depth(sb, "get_cnode");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700381 if (journal->j_cnode_free <= 0) {
382 return NULL;
383 }
384 journal->j_cnode_used++;
385 journal->j_cnode_free--;
386 cn = journal->j_cnode_free_list;
387 if (!cn) {
388 return cn;
389 }
390 if (cn->next) {
391 cn->next->prev = NULL;
392 }
393 journal->j_cnode_free_list = cn->next;
394 memset(cn, 0, sizeof(struct reiserfs_journal_cnode));
395 return cn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396}
397
398/*
Jeff Mahoney0222e652009-03-30 14:02:44 -0400399** returns a cnode to the free list
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400401static void free_cnode(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700402 struct reiserfs_journal_cnode *cn)
403{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400404 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400406 reiserfs_check_lock_depth(sb, "free_cnode");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700408 journal->j_cnode_used--;
409 journal->j_cnode_free++;
410 /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
411 cn->next = journal->j_cnode_free_list;
412 if (journal->j_cnode_free_list) {
413 journal->j_cnode_free_list->prev = cn;
414 }
415 cn->prev = NULL; /* not needed with the memset, but I might kill the memset, and forget to do this */
416 journal->j_cnode_free_list = cn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417}
418
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700419static void clear_prepared_bits(struct buffer_head *bh)
420{
421 clear_buffer_journal_prepared(bh);
422 clear_buffer_journal_restore_dirty(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423}
424
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425/* return a cnode with same dev, block number and size in table, or null if not found */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700426static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
427 super_block
428 *sb,
429 struct
430 reiserfs_journal_cnode
431 **table,
432 long bl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700434 struct reiserfs_journal_cnode *cn;
435 cn = journal_hash(table, sb, bl);
436 while (cn) {
437 if (cn->blocknr == bl && cn->sb == sb)
438 return cn;
439 cn = cn->hnext;
440 }
441 return (struct reiserfs_journal_cnode *)0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442}
443
444/*
445** this actually means 'can this block be reallocated yet?'. If you set search_all, a block can only be allocated
446** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever
447** being overwritten by a replay after crashing.
448**
449** If you don't set search_all, a block can only be allocated if it is not in the current transaction. Since deleting
450** a block removes it from the current transaction, this case should never happen. If you don't set search_all, make
451** sure you never write the block without logging it.
452**
453** next_zero_bit is a suggestion about the next block to try for find_forward.
454** when bl is rejected because it is set in a journal list bitmap, we search
455** for the next zero bit in the bitmap that rejected bl. Then, we return that
456** through next_zero_bit for find_forward to try.
457**
458** Just because we return something in next_zero_bit does not mean we won't
459** reject it on the next call to reiserfs_in_journal
460**
461*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400462int reiserfs_in_journal(struct super_block *sb,
Jeff Mahoney3ee16672007-10-18 23:39:25 -0700463 unsigned int bmap_nr, int bit_nr, int search_all,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700464 b_blocknr_t * next_zero_bit)
465{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400466 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700467 struct reiserfs_journal_cnode *cn;
468 struct reiserfs_list_bitmap *jb;
469 int i;
470 unsigned long bl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700472 *next_zero_bit = 0; /* always start this at zero. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400474 PROC_INFO_INC(sb, journal.in_journal);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700475 /* If we aren't doing a search_all, this is a metablock, and it will be logged before use.
476 ** if we crash before the transaction that freed it commits, this transaction won't
477 ** have committed either, and the block will never be written
478 */
479 if (search_all) {
480 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400481 PROC_INFO_INC(sb, journal.in_journal_bitmap);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700482 jb = journal->j_list_bitmap + i;
483 if (jb->journal_list && jb->bitmaps[bmap_nr] &&
484 test_bit(bit_nr,
485 (unsigned long *)jb->bitmaps[bmap_nr]->
486 data)) {
487 *next_zero_bit =
488 find_next_zero_bit((unsigned long *)
489 (jb->bitmaps[bmap_nr]->
490 data),
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400491 sb->s_blocksize << 3,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700492 bit_nr + 1);
493 return 1;
494 }
495 }
496 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400498 bl = bmap_nr * (sb->s_blocksize << 3) + bit_nr;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700499 /* is it in any old transactions? */
500 if (search_all
501 && (cn =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400502 get_journal_hash_dev(sb, journal->j_list_hash_table, bl))) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700503 return 1;
504 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700506 /* is it in the current transaction. This should never happen */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400507 if ((cn = get_journal_hash_dev(sb, journal->j_hash_table, bl))) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700508 BUG();
509 return 1;
510 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400512 PROC_INFO_INC(sb, journal.in_journal_reusable);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700513 /* safe for reuse */
514 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515}
516
517/* insert cn into table
518*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700519static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
520 struct reiserfs_journal_cnode *cn)
521{
522 struct reiserfs_journal_cnode *cn_orig;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700524 cn_orig = journal_hash(table, cn->sb, cn->blocknr);
525 cn->hnext = cn_orig;
526 cn->hprev = NULL;
527 if (cn_orig) {
528 cn_orig->hprev = cn;
529 }
530 journal_hash(table, cn->sb, cn->blocknr) = cn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531}
532
533/* lock the current transaction */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400534static inline void lock_journal(struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700535{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400536 PROC_INFO_INC(sb, journal.lock_journal);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +0200537
538 reiserfs_mutex_lock_safe(&SB_JOURNAL(sb)->j_mutex, sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539}
540
541/* unlock the current transaction */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400542static inline void unlock_journal(struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700543{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400544 mutex_unlock(&SB_JOURNAL(sb)->j_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545}
546
547static inline void get_journal_list(struct reiserfs_journal_list *jl)
548{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700549 jl->j_refcount++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550}
551
552static inline void put_journal_list(struct super_block *s,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700553 struct reiserfs_journal_list *jl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700555 if (jl->j_refcount < 1) {
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -0400556 reiserfs_panic(s, "journal-2", "trans id %u, refcount at %d",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700557 jl->j_trans_id, jl->j_refcount);
558 }
559 if (--jl->j_refcount == 0)
Pekka Enbergd739b422006-02-01 03:06:43 -0800560 kfree(jl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561}
562
563/*
564** this used to be much more involved, and I'm keeping it just in case things get ugly again.
565** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a
566** transaction.
567*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400568static void cleanup_freed_for_journal_list(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700569 struct reiserfs_journal_list *jl)
570{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700572 struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
573 if (jb) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400574 cleanup_bitmap_list(sb, jb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700575 }
576 jl->j_list_bitmap->journal_list = NULL;
577 jl->j_list_bitmap = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578}
579
580static int journal_list_still_alive(struct super_block *s,
Jeff Mahoney600ed412009-03-30 14:02:17 -0400581 unsigned int trans_id)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700583 struct reiserfs_journal *journal = SB_JOURNAL(s);
584 struct list_head *entry = &journal->j_journal_list;
585 struct reiserfs_journal_list *jl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700587 if (!list_empty(entry)) {
588 jl = JOURNAL_LIST_ENTRY(entry->next);
589 if (jl->j_trans_id <= trans_id) {
590 return 1;
591 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700593 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594}
595
Chris Mason398c95b2007-10-16 23:29:44 -0700596/*
597 * If page->mapping was null, we failed to truncate this page for
598 * some reason. Most likely because it was truncated after being
599 * logged via data=journal.
600 *
601 * This does a check to see if the buffer belongs to one of these
602 * lost pages before doing the final put_bh. If page->mapping was
603 * null, it tries to free buffers on the page, which should make the
604 * final page_cache_release drop the page from the lru.
605 */
606static void release_buffer_page(struct buffer_head *bh)
607{
608 struct page *page = bh->b_page;
Nick Piggin529ae9a2008-08-02 12:01:03 +0200609 if (!page->mapping && trylock_page(page)) {
Chris Mason398c95b2007-10-16 23:29:44 -0700610 page_cache_get(page);
611 put_bh(bh);
612 if (!page->mapping)
613 try_to_free_buffers(page);
614 unlock_page(page);
615 page_cache_release(page);
616 } else {
617 put_bh(bh);
618 }
619}
620
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700621static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
622{
623 char b[BDEVNAME_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700625 if (buffer_journaled(bh)) {
Jeff Mahoney45b03d52009-03-30 14:02:21 -0400626 reiserfs_warning(NULL, "clm-2084",
627 "pinned buffer %lu:%s sent to disk",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700628 bh->b_blocknr, bdevname(bh->b_bdev, b));
629 }
630 if (uptodate)
631 set_buffer_uptodate(bh);
632 else
633 clear_buffer_uptodate(bh);
Chris Mason398c95b2007-10-16 23:29:44 -0700634
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700635 unlock_buffer(bh);
Chris Mason398c95b2007-10-16 23:29:44 -0700636 release_buffer_page(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637}
638
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700639static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
640{
641 if (uptodate)
642 set_buffer_uptodate(bh);
643 else
644 clear_buffer_uptodate(bh);
645 unlock_buffer(bh);
646 put_bh(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647}
648
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700649static void submit_logged_buffer(struct buffer_head *bh)
650{
651 get_bh(bh);
652 bh->b_end_io = reiserfs_end_buffer_io_sync;
653 clear_buffer_journal_new(bh);
654 clear_buffer_dirty(bh);
655 if (!test_clear_buffer_journal_test(bh))
656 BUG();
657 if (!buffer_uptodate(bh))
658 BUG();
659 submit_bh(WRITE, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660}
661
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700662static void submit_ordered_buffer(struct buffer_head *bh)
663{
664 get_bh(bh);
665 bh->b_end_io = reiserfs_end_ordered_io;
666 clear_buffer_dirty(bh);
667 if (!buffer_uptodate(bh))
668 BUG();
669 submit_bh(WRITE, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670}
671
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672#define CHUNK_SIZE 32
673struct buffer_chunk {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700674 struct buffer_head *bh[CHUNK_SIZE];
675 int nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676};
677
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700678static void write_chunk(struct buffer_chunk *chunk)
679{
680 int i;
681 get_fs_excl();
682 for (i = 0; i < chunk->nr; i++) {
683 submit_logged_buffer(chunk->bh[i]);
684 }
685 chunk->nr = 0;
686 put_fs_excl();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687}
688
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700689static void write_ordered_chunk(struct buffer_chunk *chunk)
690{
691 int i;
692 get_fs_excl();
693 for (i = 0; i < chunk->nr; i++) {
694 submit_ordered_buffer(chunk->bh[i]);
695 }
696 chunk->nr = 0;
697 put_fs_excl();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698}
699
700static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700701 spinlock_t * lock, void (fn) (struct buffer_chunk *))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700703 int ret = 0;
Eric Sesterhenn14a61442006-10-03 23:36:38 +0200704 BUG_ON(chunk->nr >= CHUNK_SIZE);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700705 chunk->bh[chunk->nr++] = bh;
706 if (chunk->nr >= CHUNK_SIZE) {
707 ret = 1;
708 if (lock)
709 spin_unlock(lock);
710 fn(chunk);
711 if (lock)
712 spin_lock(lock);
713 }
714 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715}
716
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700718static struct reiserfs_jh *alloc_jh(void)
719{
720 struct reiserfs_jh *jh;
721 while (1) {
722 jh = kmalloc(sizeof(*jh), GFP_NOFS);
723 if (jh) {
724 atomic_inc(&nr_reiserfs_jh);
725 return jh;
726 }
727 yield();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729}
730
731/*
732 * we want to free the jh when the buffer has been written
733 * and waited on
734 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700735void reiserfs_free_jh(struct buffer_head *bh)
736{
737 struct reiserfs_jh *jh;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700739 jh = bh->b_private;
740 if (jh) {
741 bh->b_private = NULL;
742 jh->bh = NULL;
743 list_del_init(&jh->list);
744 kfree(jh);
745 if (atomic_read(&nr_reiserfs_jh) <= 0)
746 BUG();
747 atomic_dec(&nr_reiserfs_jh);
748 put_bh(bh);
749 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750}
751
752static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700753 int tail)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700755 struct reiserfs_jh *jh;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700757 if (bh->b_private) {
758 spin_lock(&j->j_dirty_buffers_lock);
759 if (!bh->b_private) {
760 spin_unlock(&j->j_dirty_buffers_lock);
761 goto no_jh;
762 }
763 jh = bh->b_private;
764 list_del_init(&jh->list);
765 } else {
766 no_jh:
767 get_bh(bh);
768 jh = alloc_jh();
769 spin_lock(&j->j_dirty_buffers_lock);
770 /* buffer must be locked for __add_jh, should be able to have
771 * two adds at the same time
772 */
Eric Sesterhenn14a61442006-10-03 23:36:38 +0200773 BUG_ON(bh->b_private);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700774 jh->bh = bh;
775 bh->b_private = jh;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700777 jh->jl = j->j_current_jl;
778 if (tail)
779 list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
780 else {
781 list_add_tail(&jh->list, &jh->jl->j_bh_list);
782 }
783 spin_unlock(&j->j_dirty_buffers_lock);
784 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785}
786
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700787int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
788{
789 return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790}
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700791int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
792{
793 return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794}
795
796#define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700797static int write_ordered_buffers(spinlock_t * lock,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 struct reiserfs_journal *j,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700799 struct reiserfs_journal_list *jl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 struct list_head *list)
801{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700802 struct buffer_head *bh;
803 struct reiserfs_jh *jh;
804 int ret = j->j_errno;
805 struct buffer_chunk chunk;
806 struct list_head tmp;
807 INIT_LIST_HEAD(&tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700809 chunk.nr = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 spin_lock(lock);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700811 while (!list_empty(list)) {
812 jh = JH_ENTRY(list->next);
813 bh = jh->bh;
814 get_bh(bh);
Nick Pigginca5de402008-08-02 12:02:13 +0200815 if (!trylock_buffer(bh)) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700816 if (!buffer_dirty(bh)) {
Akinobu Mitaf1166292006-06-26 00:24:46 -0700817 list_move(&jh->list, &tmp);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700818 goto loop_next;
819 }
820 spin_unlock(lock);
821 if (chunk.nr)
822 write_ordered_chunk(&chunk);
823 wait_on_buffer(bh);
824 cond_resched();
825 spin_lock(lock);
826 goto loop_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 }
Chris Mason3d4492f2006-02-01 03:06:49 -0800828 /* in theory, dirty non-uptodate buffers should never get here,
829 * but the upper layer io error paths still have a few quirks.
830 * Handle them here as gracefully as we can
831 */
832 if (!buffer_uptodate(bh) && buffer_dirty(bh)) {
833 clear_buffer_dirty(bh);
834 ret = -EIO;
835 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700836 if (buffer_dirty(bh)) {
Akinobu Mitaf1166292006-06-26 00:24:46 -0700837 list_move(&jh->list, &tmp);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700838 add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
839 } else {
840 reiserfs_free_jh(bh);
841 unlock_buffer(bh);
842 }
843 loop_next:
844 put_bh(bh);
845 cond_resched_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700847 if (chunk.nr) {
848 spin_unlock(lock);
849 write_ordered_chunk(&chunk);
850 spin_lock(lock);
851 }
852 while (!list_empty(&tmp)) {
853 jh = JH_ENTRY(tmp.prev);
854 bh = jh->bh;
855 get_bh(bh);
856 reiserfs_free_jh(bh);
857
858 if (buffer_locked(bh)) {
859 spin_unlock(lock);
860 wait_on_buffer(bh);
861 spin_lock(lock);
862 }
863 if (!buffer_uptodate(bh)) {
864 ret = -EIO;
865 }
Chris Masond62b1b82006-02-01 03:06:47 -0800866 /* ugly interaction with invalidatepage here.
867 * reiserfs_invalidate_page will pin any buffer that has a valid
868 * journal head from an older transaction. If someone else sets
869 * our buffer dirty after we write it in the first loop, and
870 * then someone truncates the page away, nobody will ever write
871 * the buffer. We're safe if we write the page one last time
872 * after freeing the journal header.
873 */
874 if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
875 spin_unlock(lock);
876 ll_rw_block(WRITE, 1, &bh);
877 spin_lock(lock);
878 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700879 put_bh(bh);
880 cond_resched_lock(lock);
881 }
882 spin_unlock(lock);
883 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884}
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700885
886static int flush_older_commits(struct super_block *s,
887 struct reiserfs_journal_list *jl)
888{
889 struct reiserfs_journal *journal = SB_JOURNAL(s);
890 struct reiserfs_journal_list *other_jl;
891 struct reiserfs_journal_list *first_jl;
892 struct list_head *entry;
Jeff Mahoney600ed412009-03-30 14:02:17 -0400893 unsigned int trans_id = jl->j_trans_id;
894 unsigned int other_trans_id;
895 unsigned int first_trans_id;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700896
897 find_first:
898 /*
899 * first we walk backwards to find the oldest uncommitted transation
900 */
901 first_jl = jl;
902 entry = jl->j_list.prev;
903 while (1) {
904 other_jl = JOURNAL_LIST_ENTRY(entry);
905 if (entry == &journal->j_journal_list ||
906 atomic_read(&other_jl->j_older_commits_done))
907 break;
908
909 first_jl = other_jl;
910 entry = other_jl->j_list.prev;
911 }
912
913 /* if we didn't find any older uncommitted transactions, return now */
914 if (first_jl == jl) {
915 return 0;
916 }
917
918 first_trans_id = first_jl->j_trans_id;
919
920 entry = &first_jl->j_list;
921 while (1) {
922 other_jl = JOURNAL_LIST_ENTRY(entry);
923 other_trans_id = other_jl->j_trans_id;
924
925 if (other_trans_id < trans_id) {
926 if (atomic_read(&other_jl->j_commit_left) != 0) {
927 flush_commit_list(s, other_jl, 0);
928
929 /* list we were called with is gone, return */
930 if (!journal_list_still_alive(s, trans_id))
931 return 1;
932
933 /* the one we just flushed is gone, this means all
934 * older lists are also gone, so first_jl is no longer
935 * valid either. Go back to the beginning.
936 */
937 if (!journal_list_still_alive
938 (s, other_trans_id)) {
939 goto find_first;
940 }
941 }
942 entry = entry->next;
943 if (entry == &journal->j_journal_list)
944 return 0;
945 } else {
946 return 0;
947 }
948 }
949 return 0;
950}
Adrian Bunkdeba0f42007-10-16 23:26:03 -0700951
952static int reiserfs_async_progress_wait(struct super_block *s)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700953{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700954 struct reiserfs_journal *j = SB_JOURNAL(s);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +0200955
956 if (atomic_read(&j->j_async_throttle)) {
957 reiserfs_write_unlock(s);
Jens Axboe8aa7e842009-07-09 14:52:32 +0200958 congestion_wait(BLK_RW_ASYNC, HZ / 10);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +0200959 reiserfs_write_lock(s);
960 }
961
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700962 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963}
964
965/*
966** if this journal list still has commit blocks unflushed, send them to disk.
967**
968** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
969** Before the commit block can by written, every other log block must be safely on disk
970**
971*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700972static int flush_commit_list(struct super_block *s,
973 struct reiserfs_journal_list *jl, int flushall)
974{
975 int i;
Jeff Mahoney3ee16672007-10-18 23:39:25 -0700976 b_blocknr_t bn;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700977 struct buffer_head *tbh = NULL;
Jeff Mahoney600ed412009-03-30 14:02:17 -0400978 unsigned int trans_id = jl->j_trans_id;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700979 struct reiserfs_journal *journal = SB_JOURNAL(s);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700980 int retval = 0;
Chris Masone0e851c2006-02-01 03:06:49 -0800981 int write_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700983 reiserfs_check_lock_depth(s, "flush_commit_list");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700985 if (atomic_read(&jl->j_older_commits_done)) {
986 return 0;
987 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700989 get_fs_excl();
Jens Axboe22e2c502005-06-27 10:55:12 +0200990
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700991 /* before we can put our commit blocks on disk, we have to make sure everyone older than
992 ** us is on disk too
993 */
994 BUG_ON(jl->j_len <= 0);
995 BUG_ON(trans_id == journal->j_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700997 get_journal_list(jl);
998 if (flushall) {
999 if (flush_older_commits(s, jl) == 1) {
1000 /* list disappeared during flush_older_commits. return */
1001 goto put_jl;
1002 }
1003 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001005 /* make sure nobody is trying to flush this one at the same time */
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001006 reiserfs_mutex_lock_safe(&jl->j_commit_mutex, s);
1007
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001008 if (!journal_list_still_alive(s, trans_id)) {
Jeff Mahoney90415de2008-07-25 01:46:40 -07001009 mutex_unlock(&jl->j_commit_mutex);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001010 goto put_jl;
1011 }
1012 BUG_ON(jl->j_trans_id == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001014 /* this commit is done, exit */
1015 if (atomic_read(&(jl->j_commit_left)) <= 0) {
1016 if (flushall) {
1017 atomic_set(&(jl->j_older_commits_done), 1);
1018 }
Jeff Mahoney90415de2008-07-25 01:46:40 -07001019 mutex_unlock(&jl->j_commit_mutex);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001020 goto put_jl;
1021 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001023 if (!list_empty(&jl->j_bh_list)) {
Chris Mason3d4492f2006-02-01 03:06:49 -08001024 int ret;
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001025
1026 /*
1027 * We might sleep in numerous places inside
1028 * write_ordered_buffers. Relax the write lock.
1029 */
1030 reiserfs_write_unlock(s);
Chris Mason3d4492f2006-02-01 03:06:49 -08001031 ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
1032 journal, jl, &jl->j_bh_list);
1033 if (ret < 0 && retval == 0)
1034 retval = ret;
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001035 reiserfs_write_lock(s);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001036 }
1037 BUG_ON(!list_empty(&jl->j_bh_list));
1038 /*
1039 * for the description block and all the log blocks, submit any buffers
Chris Masone0e851c2006-02-01 03:06:49 -08001040 * that haven't already reached the disk. Try to write at least 256
1041 * log blocks. later on, we will only wait on blocks that correspond
1042 * to this transaction, but while we're unplugging we might as well
1043 * get a chunk of data on there.
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001044 */
1045 atomic_inc(&journal->j_async_throttle);
Chris Masone0e851c2006-02-01 03:06:49 -08001046 write_len = jl->j_len + 1;
1047 if (write_len < 256)
1048 write_len = 256;
1049 for (i = 0 ; i < write_len ; i++) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001050 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1051 SB_ONDISK_JOURNAL_SIZE(s);
1052 tbh = journal_find_get_block(s, bn);
Chris Masone0e851c2006-02-01 03:06:49 -08001053 if (tbh) {
Frederic Weisbecker6e3647a2009-05-01 02:27:39 +02001054 if (buffer_dirty(tbh)) {
1055 reiserfs_write_unlock(s);
1056 ll_rw_block(WRITE, 1, &tbh);
1057 reiserfs_write_lock(s);
1058 }
Chris Masone0e851c2006-02-01 03:06:49 -08001059 put_bh(tbh) ;
1060 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001061 }
1062 atomic_dec(&journal->j_async_throttle);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001064 for (i = 0; i < (jl->j_len + 1); i++) {
1065 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1066 (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1067 tbh = journal_find_get_block(s, bn);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001068
1069 reiserfs_write_unlock(s);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001070 wait_on_buffer(tbh);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001071 reiserfs_write_lock(s);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001072 // since we're using ll_rw_blk above, it might have skipped over
1073 // a locked buffer. Double check here
1074 //
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001075 /* redundant, sync_dirty_buffer() checks */
1076 if (buffer_dirty(tbh)) {
1077 reiserfs_write_unlock(s);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001078 sync_dirty_buffer(tbh);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001079 reiserfs_write_lock(s);
1080 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001081 if (unlikely(!buffer_uptodate(tbh))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082#ifdef CONFIG_REISERFS_CHECK
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001083 reiserfs_warning(s, "journal-601",
1084 "buffer write failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085#endif
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001086 retval = -EIO;
1087 }
1088 put_bh(tbh); /* once for journal_find_get_block */
1089 put_bh(tbh); /* once due to original getblk in do_journal_end */
1090 atomic_dec(&(jl->j_commit_left));
1091 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001093 BUG_ON(atomic_read(&(jl->j_commit_left)) != 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094
Christoph Hellwig7cd33ad2010-08-18 05:29:14 -04001095 /* If there was a write error in the journal - we can't commit
1096 * this transaction - it will be invalid and, if successful,
1097 * will just end up propagating the write error out to
1098 * the file system. */
1099 if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1100 if (buffer_dirty(jl->j_commit_bh))
1101 BUG();
1102 mark_buffer_dirty(jl->j_commit_bh) ;
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001103 reiserfs_write_unlock(s);
Christoph Hellwig7cd33ad2010-08-18 05:29:14 -04001104 if (reiserfs_barrier_flush(s))
1105 __sync_dirty_buffer(jl->j_commit_bh, WRITE_FLUSH_FUA);
1106 else
1107 sync_dirty_buffer(jl->j_commit_bh);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001108 reiserfs_write_lock(s);
1109 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001111 /* If there was a write error in the journal - we can't commit this
1112 * transaction - it will be invalid and, if successful, will just end
Robert P. J. Daybeb7dd82007-05-09 07:14:03 +02001113 * up propagating the write error out to the filesystem. */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001114 if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115#ifdef CONFIG_REISERFS_CHECK
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001116 reiserfs_warning(s, "journal-615", "buffer write failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117#endif
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001118 retval = -EIO;
1119 }
1120 bforget(jl->j_commit_bh);
1121 if (journal->j_last_commit_id != 0 &&
1122 (jl->j_trans_id - journal->j_last_commit_id) != 1) {
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001123 reiserfs_warning(s, "clm-2200", "last commit %lu, current %lu",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001124 journal->j_last_commit_id, jl->j_trans_id);
1125 }
1126 journal->j_last_commit_id = jl->j_trans_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001128 /* now, every commit block is on the disk. It is safe to allow blocks freed during this transaction to be reallocated */
1129 cleanup_freed_for_journal_list(s, jl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001131 retval = retval ? retval : journal->j_errno;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001133 /* mark the metadata dirty */
1134 if (!retval)
1135 dirty_one_transaction(s, jl);
1136 atomic_dec(&(jl->j_commit_left));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001138 if (flushall) {
1139 atomic_set(&(jl->j_older_commits_done), 1);
1140 }
Jeff Mahoney90415de2008-07-25 01:46:40 -07001141 mutex_unlock(&jl->j_commit_mutex);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001142 put_jl:
1143 put_journal_list(s, jl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001145 if (retval)
1146 reiserfs_abort(s, retval, "Journal write error in %s",
Harvey Harrisonfbe54982008-04-28 02:16:22 -07001147 __func__);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001148 put_fs_excl();
1149 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150}
1151
1152/*
Jeff Mahoney0222e652009-03-30 14:02:44 -04001153** flush_journal_list frequently needs to find a newer transaction for a given block. This does that, or
1154** returns NULL if it can't find anything
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001156static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1157 reiserfs_journal_cnode
1158 *cn)
1159{
1160 struct super_block *sb = cn->sb;
1161 b_blocknr_t blocknr = cn->blocknr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001163 cn = cn->hprev;
1164 while (cn) {
1165 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1166 return cn->jlist;
1167 }
1168 cn = cn->hprev;
1169 }
1170 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171}
1172
Chris Masona3172022006-09-29 01:59:56 -07001173static int newer_jl_done(struct reiserfs_journal_cnode *cn)
1174{
1175 struct super_block *sb = cn->sb;
1176 b_blocknr_t blocknr = cn->blocknr;
1177
1178 cn = cn->hprev;
1179 while (cn) {
1180 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist &&
1181 atomic_read(&cn->jlist->j_commit_left) != 0)
1182 return 0;
1183 cn = cn->hprev;
1184 }
1185 return 1;
1186}
1187
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001188static void remove_journal_hash(struct super_block *,
1189 struct reiserfs_journal_cnode **,
1190 struct reiserfs_journal_list *, unsigned long,
1191 int);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192
1193/*
1194** once all the real blocks have been flushed, it is safe to remove them from the
1195** journal list for this transaction. Aside from freeing the cnode, this also allows the
1196** block to be reallocated for data blocks if it had been deleted.
1197*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001198static void remove_all_from_journal_list(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001199 struct reiserfs_journal_list *jl,
1200 int debug)
1201{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001202 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001203 struct reiserfs_journal_cnode *cn, *last;
1204 cn = jl->j_realblock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001206 /* which is better, to lock once around the whole loop, or
1207 ** to lock for each call to remove_journal_hash?
1208 */
1209 while (cn) {
1210 if (cn->blocknr != 0) {
1211 if (debug) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001212 reiserfs_warning(sb, "reiserfs-2201",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001213 "block %u, bh is %d, state %ld",
1214 cn->blocknr, cn->bh ? 1 : 0,
1215 cn->state);
1216 }
1217 cn->state = 0;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001218 remove_journal_hash(sb, journal->j_list_hash_table,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001219 jl, cn->blocknr, 1);
1220 }
1221 last = cn;
1222 cn = cn->next;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001223 free_cnode(sb, last);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001224 }
1225 jl->j_realblock = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226}
1227
1228/*
1229** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block.
1230** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start
1231** releasing blocks in this transaction for reuse as data blocks.
1232** called by flush_journal_list, before it calls remove_all_from_journal_list
1233**
1234*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001235static int _update_journal_header_block(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001236 unsigned long offset,
Jeff Mahoney600ed412009-03-30 14:02:17 -04001237 unsigned int trans_id)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001238{
1239 struct reiserfs_journal_header *jh;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001240 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001242 if (reiserfs_is_journal_aborted(journal))
1243 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001245 if (trans_id >= journal->j_last_flush_trans_id) {
1246 if (buffer_locked((journal->j_header_bh))) {
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001247 reiserfs_write_unlock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001248 wait_on_buffer((journal->j_header_bh));
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001249 reiserfs_write_lock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001250 if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251#ifdef CONFIG_REISERFS_CHECK
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001252 reiserfs_warning(sb, "journal-699",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001253 "buffer write failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254#endif
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001255 return -EIO;
1256 }
1257 }
1258 journal->j_last_flush_trans_id = trans_id;
1259 journal->j_first_unflushed_offset = offset;
1260 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1261 b_data);
1262 jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1263 jh->j_first_unflushed_offset = cpu_to_le32(offset);
1264 jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265
Christoph Hellwig7cd33ad2010-08-18 05:29:14 -04001266 set_buffer_dirty(journal->j_header_bh);
1267 reiserfs_write_unlock(sb);
1268
1269 if (reiserfs_barrier_flush(sb))
1270 __sync_dirty_buffer(journal->j_header_bh, WRITE_FLUSH_FUA);
1271 else
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001272 sync_dirty_buffer(journal->j_header_bh);
Christoph Hellwig7cd33ad2010-08-18 05:29:14 -04001273
1274 reiserfs_write_lock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001275 if (!buffer_uptodate(journal->j_header_bh)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001276 reiserfs_warning(sb, "journal-837",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001277 "IO error during journal replay");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001278 return -EIO;
1279 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001281 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282}
1283
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001284static int update_journal_header_block(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001285 unsigned long offset,
Jeff Mahoney600ed412009-03-30 14:02:17 -04001286 unsigned int trans_id)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001287{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001288 return _update_journal_header_block(sb, offset, trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289}
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001290
Jeff Mahoney0222e652009-03-30 14:02:44 -04001291/*
1292** flush any and all journal lists older than you are
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293** can only be called from flush_journal_list
1294*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001295static int flush_older_journal_lists(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001296 struct reiserfs_journal_list *jl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001298 struct list_head *entry;
1299 struct reiserfs_journal_list *other_jl;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001300 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Jeff Mahoney600ed412009-03-30 14:02:17 -04001301 unsigned int trans_id = jl->j_trans_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001303 /* we know we are the only ones flushing things, no extra race
1304 * protection is required.
1305 */
1306 restart:
1307 entry = journal->j_journal_list.next;
1308 /* Did we wrap? */
1309 if (entry == &journal->j_journal_list)
1310 return 0;
1311 other_jl = JOURNAL_LIST_ENTRY(entry);
1312 if (other_jl->j_trans_id < trans_id) {
1313 BUG_ON(other_jl->j_refcount <= 0);
1314 /* do not flush all */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001315 flush_journal_list(sb, other_jl, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001317 /* other_jl is now deleted from the list */
1318 goto restart;
1319 }
1320 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321}
1322
1323static void del_from_work_list(struct super_block *s,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001324 struct reiserfs_journal_list *jl)
1325{
1326 struct reiserfs_journal *journal = SB_JOURNAL(s);
1327 if (!list_empty(&jl->j_working_list)) {
1328 list_del_init(&jl->j_working_list);
1329 journal->j_num_work_lists--;
1330 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331}
1332
1333/* flush a journal list, both commit and real blocks
1334**
1335** always set flushall to 1, unless you are calling from inside
1336** flush_journal_list
1337**
Jeff Mahoney0222e652009-03-30 14:02:44 -04001338** IMPORTANT. This can only be called while there are no journal writers,
1339** and the journal is locked. That means it can only be called from
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340** do_journal_end, or by journal_release
1341*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001342static int flush_journal_list(struct super_block *s,
1343 struct reiserfs_journal_list *jl, int flushall)
1344{
1345 struct reiserfs_journal_list *pjl;
1346 struct reiserfs_journal_cnode *cn, *last;
1347 int count;
1348 int was_jwait = 0;
1349 int was_dirty = 0;
1350 struct buffer_head *saved_bh;
1351 unsigned long j_len_saved = jl->j_len;
1352 struct reiserfs_journal *journal = SB_JOURNAL(s);
1353 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001355 BUG_ON(j_len_saved <= 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001357 if (atomic_read(&journal->j_wcount) != 0) {
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001358 reiserfs_warning(s, "clm-2048", "called with wcount %d",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001359 atomic_read(&journal->j_wcount));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001361 BUG_ON(jl->j_trans_id == 0);
1362
1363 /* if flushall == 0, the lock is already held */
1364 if (flushall) {
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001365 reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
Jeff Mahoneyafe70252008-07-25 01:46:39 -07001366 } else if (mutex_trylock(&journal->j_flush_mutex)) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001367 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001369
1370 count = 0;
1371 if (j_len_saved > journal->j_trans_max) {
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04001372 reiserfs_panic(s, "journal-715", "length is %lu, trans id %lu",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001373 j_len_saved, jl->j_trans_id);
1374 return 0;
1375 }
1376
1377 get_fs_excl();
1378
1379 /* if all the work is already done, get out of here */
1380 if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1381 atomic_read(&(jl->j_commit_left)) <= 0) {
1382 goto flush_older_and_return;
1383 }
1384
Jeff Mahoney0222e652009-03-30 14:02:44 -04001385 /* start by putting the commit list on disk. This will also flush
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001386 ** the commit lists of any olders transactions
1387 */
1388 flush_commit_list(s, jl, 1);
1389
1390 if (!(jl->j_state & LIST_DIRTY)
1391 && !reiserfs_is_journal_aborted(journal))
1392 BUG();
1393
1394 /* are we done now? */
1395 if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1396 atomic_read(&(jl->j_commit_left)) <= 0) {
1397 goto flush_older_and_return;
1398 }
1399
Jeff Mahoney0222e652009-03-30 14:02:44 -04001400 /* loop through each cnode, see if we need to write it,
1401 ** or wait on a more recent transaction, or just ignore it
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001402 */
1403 if (atomic_read(&(journal->j_wcount)) != 0) {
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04001404 reiserfs_panic(s, "journal-844", "journal list is flushing, "
1405 "wcount is not 0");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001406 }
1407 cn = jl->j_realblock;
1408 while (cn) {
1409 was_jwait = 0;
1410 was_dirty = 0;
1411 saved_bh = NULL;
1412 /* blocknr of 0 is no longer in the hash, ignore it */
1413 if (cn->blocknr == 0) {
1414 goto free_cnode;
1415 }
1416
1417 /* This transaction failed commit. Don't write out to the disk */
1418 if (!(jl->j_state & LIST_DIRTY))
1419 goto free_cnode;
1420
1421 pjl = find_newer_jl_for_cn(cn);
1422 /* the order is important here. We check pjl to make sure we
1423 ** don't clear BH_JDirty_wait if we aren't the one writing this
1424 ** block to disk
1425 */
1426 if (!pjl && cn->bh) {
1427 saved_bh = cn->bh;
1428
Jeff Mahoney0222e652009-03-30 14:02:44 -04001429 /* we do this to make sure nobody releases the buffer while
1430 ** we are working with it
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001431 */
1432 get_bh(saved_bh);
1433
1434 if (buffer_journal_dirty(saved_bh)) {
1435 BUG_ON(!can_dirty(cn));
1436 was_jwait = 1;
1437 was_dirty = 1;
1438 } else if (can_dirty(cn)) {
1439 /* everything with !pjl && jwait should be writable */
1440 BUG();
1441 }
1442 }
1443
1444 /* if someone has this block in a newer transaction, just make
Matt LaPlante0779bf22006-11-30 05:24:39 +01001445 ** sure they are committed, and don't try writing it to disk
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001446 */
1447 if (pjl) {
1448 if (atomic_read(&pjl->j_commit_left))
1449 flush_commit_list(s, pjl, 1);
1450 goto free_cnode;
1451 }
1452
Jeff Mahoney0222e652009-03-30 14:02:44 -04001453 /* bh == NULL when the block got to disk on its own, OR,
1454 ** the block got freed in a future transaction
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001455 */
1456 if (saved_bh == NULL) {
1457 goto free_cnode;
1458 }
1459
1460 /* this should never happen. kupdate_one_transaction has this list
1461 ** locked while it works, so we should never see a buffer here that
1462 ** is not marked JDirty_wait
1463 */
1464 if ((!was_jwait) && !buffer_locked(saved_bh)) {
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001465 reiserfs_warning(s, "journal-813",
1466 "BAD! buffer %llu %cdirty %cjwait, "
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001467 "not in a newer tranasction",
1468 (unsigned long long)saved_bh->
1469 b_blocknr, was_dirty ? ' ' : '!',
1470 was_jwait ? ' ' : '!');
1471 }
1472 if (was_dirty) {
1473 /* we inc again because saved_bh gets decremented at free_cnode */
1474 get_bh(saved_bh);
1475 set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1476 lock_buffer(saved_bh);
1477 BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1478 if (buffer_dirty(saved_bh))
1479 submit_logged_buffer(saved_bh);
1480 else
1481 unlock_buffer(saved_bh);
1482 count++;
1483 } else {
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001484 reiserfs_warning(s, "clm-2082",
1485 "Unable to flush buffer %llu in %s",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001486 (unsigned long long)saved_bh->
Harvey Harrisonfbe54982008-04-28 02:16:22 -07001487 b_blocknr, __func__);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001488 }
1489 free_cnode:
1490 last = cn;
1491 cn = cn->next;
1492 if (saved_bh) {
1493 /* we incremented this to keep others from taking the buffer head away */
1494 put_bh(saved_bh);
1495 if (atomic_read(&(saved_bh->b_count)) < 0) {
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001496 reiserfs_warning(s, "journal-945",
1497 "saved_bh->b_count < 0");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001498 }
1499 }
1500 }
1501 if (count > 0) {
1502 cn = jl->j_realblock;
1503 while (cn) {
1504 if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1505 if (!cn->bh) {
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04001506 reiserfs_panic(s, "journal-1011",
1507 "cn->bh is NULL");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001508 }
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001509
1510 reiserfs_write_unlock(s);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001511 wait_on_buffer(cn->bh);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001512 reiserfs_write_lock(s);
1513
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001514 if (!cn->bh) {
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04001515 reiserfs_panic(s, "journal-1012",
1516 "cn->bh is NULL");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001517 }
1518 if (unlikely(!buffer_uptodate(cn->bh))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519#ifdef CONFIG_REISERFS_CHECK
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001520 reiserfs_warning(s, "journal-949",
1521 "buffer write failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522#endif
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001523 err = -EIO;
1524 }
1525 /* note, we must clear the JDirty_wait bit after the up to date
1526 ** check, otherwise we race against our flushpage routine
1527 */
1528 BUG_ON(!test_clear_buffer_journal_dirty
1529 (cn->bh));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530
Chris Mason398c95b2007-10-16 23:29:44 -07001531 /* drop one ref for us */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001532 put_bh(cn->bh);
Chris Mason398c95b2007-10-16 23:29:44 -07001533 /* drop one ref for journal_mark_dirty */
1534 release_buffer_page(cn->bh);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001535 }
1536 cn = cn->next;
1537 }
1538 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001540 if (err)
1541 reiserfs_abort(s, -EIO,
1542 "Write error while pushing transaction to disk in %s",
Harvey Harrisonfbe54982008-04-28 02:16:22 -07001543 __func__);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001544 flush_older_and_return:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545
Jeff Mahoney0222e652009-03-30 14:02:44 -04001546 /* before we can update the journal header block, we _must_ flush all
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001547 ** real blocks from all older transactions to disk. This is because
1548 ** once the header block is updated, this transaction will not be
1549 ** replayed after a crash
1550 */
1551 if (flushall) {
1552 flush_older_journal_lists(s, jl);
1553 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001555 err = journal->j_errno;
Jeff Mahoney0222e652009-03-30 14:02:44 -04001556 /* before we can remove everything from the hash tables for this
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001557 ** transaction, we must make sure it can never be replayed
1558 **
1559 ** since we are only called from do_journal_end, we know for sure there
1560 ** are no allocations going on while we are flushing journal lists. So,
1561 ** we only need to update the journal header block for the last list
1562 ** being flushed
1563 */
1564 if (!err && flushall) {
1565 err =
1566 update_journal_header_block(s,
1567 (jl->j_start + jl->j_len +
1568 2) % SB_ONDISK_JOURNAL_SIZE(s),
1569 jl->j_trans_id);
1570 if (err)
1571 reiserfs_abort(s, -EIO,
1572 "Write error while updating journal header in %s",
Harvey Harrisonfbe54982008-04-28 02:16:22 -07001573 __func__);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001574 }
1575 remove_all_from_journal_list(s, jl, 0);
1576 list_del_init(&jl->j_list);
1577 journal->j_num_lists--;
1578 del_from_work_list(s, jl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001580 if (journal->j_last_flush_id != 0 &&
1581 (jl->j_trans_id - journal->j_last_flush_id) != 1) {
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001582 reiserfs_warning(s, "clm-2201", "last flush %lu, current %lu",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001583 journal->j_last_flush_id, jl->j_trans_id);
1584 }
1585 journal->j_last_flush_id = jl->j_trans_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001587 /* not strictly required since we are freeing the list, but it should
1588 * help find code using dead lists later on
1589 */
1590 jl->j_len = 0;
1591 atomic_set(&(jl->j_nonzerolen), 0);
1592 jl->j_start = 0;
1593 jl->j_realblock = NULL;
1594 jl->j_commit_bh = NULL;
1595 jl->j_trans_id = 0;
1596 jl->j_state = 0;
1597 put_journal_list(s, jl);
1598 if (flushall)
Jeff Mahoneyafe70252008-07-25 01:46:39 -07001599 mutex_unlock(&journal->j_flush_mutex);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001600 put_fs_excl();
1601 return err;
1602}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603
Chris Masona3172022006-09-29 01:59:56 -07001604static int test_transaction(struct super_block *s,
1605 struct reiserfs_journal_list *jl)
1606{
1607 struct reiserfs_journal_cnode *cn;
1608
1609 if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0)
1610 return 1;
1611
1612 cn = jl->j_realblock;
1613 while (cn) {
1614 /* if the blocknr == 0, this has been cleared from the hash,
1615 ** skip it
1616 */
1617 if (cn->blocknr == 0) {
1618 goto next;
1619 }
1620 if (cn->bh && !newer_jl_done(cn))
1621 return 0;
1622 next:
1623 cn = cn->next;
1624 cond_resched();
1625 }
1626 return 0;
1627}
1628
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629static int write_one_transaction(struct super_block *s,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001630 struct reiserfs_journal_list *jl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631 struct buffer_chunk *chunk)
1632{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001633 struct reiserfs_journal_cnode *cn;
1634 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001636 jl->j_state |= LIST_TOUCHED;
1637 del_from_work_list(s, jl);
1638 if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1639 return 0;
1640 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001642 cn = jl->j_realblock;
1643 while (cn) {
1644 /* if the blocknr == 0, this has been cleared from the hash,
1645 ** skip it
1646 */
1647 if (cn->blocknr == 0) {
1648 goto next;
1649 }
1650 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1651 struct buffer_head *tmp_bh;
1652 /* we can race against journal_mark_freed when we try
1653 * to lock_buffer(cn->bh), so we have to inc the buffer
1654 * count, and recheck things after locking
1655 */
1656 tmp_bh = cn->bh;
1657 get_bh(tmp_bh);
1658 lock_buffer(tmp_bh);
1659 if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1660 if (!buffer_journal_dirty(tmp_bh) ||
1661 buffer_journal_prepared(tmp_bh))
1662 BUG();
1663 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1664 ret++;
1665 } else {
1666 /* note, cn->bh might be null now */
1667 unlock_buffer(tmp_bh);
1668 }
1669 put_bh(tmp_bh);
1670 }
1671 next:
1672 cn = cn->next;
1673 cond_resched();
1674 }
1675 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676}
1677
1678/* used by flush_commit_list */
1679static int dirty_one_transaction(struct super_block *s,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001680 struct reiserfs_journal_list *jl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001682 struct reiserfs_journal_cnode *cn;
1683 struct reiserfs_journal_list *pjl;
1684 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001686 jl->j_state |= LIST_DIRTY;
1687 cn = jl->j_realblock;
1688 while (cn) {
1689 /* look for a more recent transaction that logged this
1690 ** buffer. Only the most recent transaction with a buffer in
1691 ** it is allowed to send that buffer to disk
1692 */
1693 pjl = find_newer_jl_for_cn(cn);
1694 if (!pjl && cn->blocknr && cn->bh
1695 && buffer_journal_dirty(cn->bh)) {
1696 BUG_ON(!can_dirty(cn));
1697 /* if the buffer is prepared, it will either be logged
1698 * or restored. If restored, we need to make sure
1699 * it actually gets marked dirty
1700 */
1701 clear_buffer_journal_new(cn->bh);
1702 if (buffer_journal_prepared(cn->bh)) {
1703 set_buffer_journal_restore_dirty(cn->bh);
1704 } else {
1705 set_buffer_journal_test(cn->bh);
1706 mark_buffer_dirty(cn->bh);
1707 }
1708 }
1709 cn = cn->next;
1710 }
1711 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712}
1713
1714static int kupdate_transactions(struct super_block *s,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001715 struct reiserfs_journal_list *jl,
1716 struct reiserfs_journal_list **next_jl,
Jeff Mahoney600ed412009-03-30 14:02:17 -04001717 unsigned int *next_trans_id,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001718 int num_blocks, int num_trans)
1719{
1720 int ret = 0;
1721 int written = 0;
1722 int transactions_flushed = 0;
Jeff Mahoney600ed412009-03-30 14:02:17 -04001723 unsigned int orig_trans_id = jl->j_trans_id;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001724 struct buffer_chunk chunk;
1725 struct list_head *entry;
1726 struct reiserfs_journal *journal = SB_JOURNAL(s);
1727 chunk.nr = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728
Frederic Weisbeckera412f9e2009-04-14 00:10:35 +02001729 reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001730 if (!journal_list_still_alive(s, orig_trans_id)) {
1731 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733
Jeff Mahoneyafe70252008-07-25 01:46:39 -07001734 /* we've got j_flush_mutex held, nobody is going to delete any
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001735 * of these lists out from underneath us
1736 */
1737 while ((num_trans && transactions_flushed < num_trans) ||
1738 (!num_trans && written < num_blocks)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001740 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1741 atomic_read(&jl->j_commit_left)
1742 || !(jl->j_state & LIST_DIRTY)) {
1743 del_from_work_list(s, jl);
1744 break;
1745 }
1746 ret = write_one_transaction(s, jl, &chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001748 if (ret < 0)
1749 goto done;
1750 transactions_flushed++;
1751 written += ret;
1752 entry = jl->j_list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001754 /* did we wrap? */
1755 if (entry == &journal->j_journal_list) {
1756 break;
1757 }
1758 jl = JOURNAL_LIST_ENTRY(entry);
1759
1760 /* don't bother with older transactions */
1761 if (jl->j_trans_id <= orig_trans_id)
1762 break;
1763 }
1764 if (chunk.nr) {
1765 write_chunk(&chunk);
1766 }
1767
1768 done:
Jeff Mahoneyafe70252008-07-25 01:46:39 -07001769 mutex_unlock(&journal->j_flush_mutex);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001770 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771}
1772
1773/* for o_sync and fsync heavy applications, they tend to use
1774** all the journa list slots with tiny transactions. These
1775** trigger lots and lots of calls to update the header block, which
1776** adds seeks and slows things down.
1777**
1778** This function tries to clear out a large chunk of the journal lists
1779** at once, which makes everything faster since only the newest journal
1780** list updates the header block
1781*/
1782static int flush_used_journal_lists(struct super_block *s,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001783 struct reiserfs_journal_list *jl)
1784{
1785 unsigned long len = 0;
1786 unsigned long cur_len;
1787 int ret;
1788 int i;
1789 int limit = 256;
1790 struct reiserfs_journal_list *tjl;
1791 struct reiserfs_journal_list *flush_jl;
Jeff Mahoney600ed412009-03-30 14:02:17 -04001792 unsigned int trans_id;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001793 struct reiserfs_journal *journal = SB_JOURNAL(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001795 flush_jl = tjl = jl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001797 /* in data logging mode, try harder to flush a lot of blocks */
1798 if (reiserfs_data_log(s))
1799 limit = 1024;
1800 /* flush for 256 transactions or limit blocks, whichever comes first */
1801 for (i = 0; i < 256 && len < limit; i++) {
1802 if (atomic_read(&tjl->j_commit_left) ||
1803 tjl->j_trans_id < jl->j_trans_id) {
1804 break;
1805 }
1806 cur_len = atomic_read(&tjl->j_nonzerolen);
1807 if (cur_len > 0) {
1808 tjl->j_state &= ~LIST_TOUCHED;
1809 }
1810 len += cur_len;
1811 flush_jl = tjl;
1812 if (tjl->j_list.next == &journal->j_journal_list)
1813 break;
1814 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001816 /* try to find a group of blocks we can flush across all the
1817 ** transactions, but only bother if we've actually spanned
1818 ** across multiple lists
1819 */
1820 if (flush_jl != jl) {
1821 ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001823 flush_journal_list(s, flush_jl, 1);
1824 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825}
1826
1827/*
1828** removes any nodes in table with name block and dev as bh.
1829** only touchs the hnext and hprev pointers.
1830*/
1831void remove_journal_hash(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001832 struct reiserfs_journal_cnode **table,
1833 struct reiserfs_journal_list *jl,
1834 unsigned long block, int remove_freed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001836 struct reiserfs_journal_cnode *cur;
1837 struct reiserfs_journal_cnode **head;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001839 head = &(journal_hash(table, sb, block));
1840 if (!head) {
1841 return;
1842 }
1843 cur = *head;
1844 while (cur) {
1845 if (cur->blocknr == block && cur->sb == sb
1846 && (jl == NULL || jl == cur->jlist)
1847 && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1848 if (cur->hnext) {
1849 cur->hnext->hprev = cur->hprev;
1850 }
1851 if (cur->hprev) {
1852 cur->hprev->hnext = cur->hnext;
1853 } else {
1854 *head = cur->hnext;
1855 }
1856 cur->blocknr = 0;
1857 cur->sb = NULL;
1858 cur->state = 0;
1859 if (cur->bh && cur->jlist) /* anybody who clears the cur->bh will also dec the nonzerolen */
1860 atomic_dec(&(cur->jlist->j_nonzerolen));
1861 cur->bh = NULL;
1862 cur->jlist = NULL;
1863 }
1864 cur = cur->hnext;
1865 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866}
1867
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001868static void free_journal_ram(struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001869{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001870 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Pekka Enbergd739b422006-02-01 03:06:43 -08001871 kfree(journal->j_current_jl);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001872 journal->j_num_lists--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001874 vfree(journal->j_cnode_free_orig);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001875 free_list_bitmaps(sb, journal->j_list_bitmap);
1876 free_bitmap_nodes(sb); /* must be after free_list_bitmaps */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001877 if (journal->j_header_bh) {
1878 brelse(journal->j_header_bh);
1879 }
1880 /* j_header_bh is on the journal dev, make sure not to release the journal
1881 * dev until we brelse j_header_bh
1882 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001883 release_journal_dev(sb, journal);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001884 vfree(journal);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885}
1886
1887/*
1888** call on unmount. Only set error to 1 if you haven't made your way out
1889** of read_super() yet. Any other caller must keep error at 0.
1890*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001891static int do_journal_release(struct reiserfs_transaction_handle *th,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001892 struct super_block *sb, int error)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001893{
1894 struct reiserfs_transaction_handle myth;
1895 int flushed = 0;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001896 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001898 /* we only want to flush out transactions if we were called with error == 0
1899 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001900 if (!error && !(sb->s_flags & MS_RDONLY)) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001901 /* end the current trans */
1902 BUG_ON(!th->t_trans_id);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001903 do_journal_end(th, sb, 10, FLUSH_ALL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001905 /* make sure something gets logged to force our way into the flush code */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001906 if (!journal_join(&myth, sb, 1)) {
1907 reiserfs_prepare_for_journal(sb,
1908 SB_BUFFER_WITH_SB(sb),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001909 1);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001910 journal_mark_dirty(&myth, sb,
1911 SB_BUFFER_WITH_SB(sb));
1912 do_journal_end(&myth, sb, 1, FLUSH_ALL);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001913 flushed = 1;
1914 }
1915 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001917 /* this also catches errors during the do_journal_end above */
1918 if (!error && reiserfs_is_journal_aborted(journal)) {
1919 memset(&myth, 0, sizeof(myth));
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001920 if (!journal_join_abort(&myth, sb, 1)) {
1921 reiserfs_prepare_for_journal(sb,
1922 SB_BUFFER_WITH_SB(sb),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001923 1);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001924 journal_mark_dirty(&myth, sb,
1925 SB_BUFFER_WITH_SB(sb));
1926 do_journal_end(&myth, sb, 1, FLUSH_ALL);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001927 }
1928 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001930 reiserfs_mounted_fs_count--;
1931 /* wait for all commits to finish */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001932 cancel_delayed_work(&SB_JOURNAL(sb)->j_work);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001933
1934 /*
1935 * We must release the write lock here because
1936 * the workqueue job (flush_async_commit) needs this lock
1937 */
1938 reiserfs_write_unlock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001939 flush_workqueue(commit_wq);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001940
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001941 if (!reiserfs_mounted_fs_count) {
1942 destroy_workqueue(commit_wq);
1943 commit_wq = NULL;
1944 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001946 free_journal_ram(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947
Frederic Weisbecker05236762009-12-30 05:56:08 +01001948 reiserfs_write_lock(sb);
1949
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001950 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951}
1952
1953/*
1954** call on unmount. flush all journal trans, release all alloc'd ram
1955*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001956int journal_release(struct reiserfs_transaction_handle *th,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001957 struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001958{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001959 return do_journal_release(th, sb, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960}
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001961
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962/*
1963** only call from an error condition inside reiserfs_read_super!
1964*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001965int journal_release_error(struct reiserfs_transaction_handle *th,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001966 struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001967{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001968 return do_journal_release(th, sb, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969}
1970
1971/* compares description block with commit block. returns 1 if they differ, 0 if they are the same */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001972static int journal_compare_desc_commit(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001973 struct reiserfs_journal_desc *desc,
1974 struct reiserfs_journal_commit *commit)
1975{
1976 if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
1977 get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001978 get_commit_trans_len(commit) > SB_JOURNAL(sb)->j_trans_max ||
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001979 get_commit_trans_len(commit) <= 0) {
1980 return 1;
1981 }
1982 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983}
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001984
Jeff Mahoney0222e652009-03-30 14:02:44 -04001985/* returns 0 if it did not find a description block
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986** returns -1 if it found a corrupt commit block
Jeff Mahoney0222e652009-03-30 14:02:44 -04001987** returns 1 if both desc and commit were valid
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001989static int journal_transaction_is_valid(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001990 struct buffer_head *d_bh,
Jeff Mahoney600ed412009-03-30 14:02:17 -04001991 unsigned int *oldest_invalid_trans_id,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001992 unsigned long *newest_mount_id)
1993{
1994 struct reiserfs_journal_desc *desc;
1995 struct reiserfs_journal_commit *commit;
1996 struct buffer_head *c_bh;
1997 unsigned long offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001999 if (!d_bh)
2000 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002002 desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2003 if (get_desc_trans_len(desc) > 0
2004 && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
2005 if (oldest_invalid_trans_id && *oldest_invalid_trans_id
2006 && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002007 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002008 "journal-986: transaction "
2009 "is valid returning because trans_id %d is greater than "
2010 "oldest_invalid %lu",
2011 get_desc_trans_id(desc),
2012 *oldest_invalid_trans_id);
2013 return 0;
2014 }
2015 if (newest_mount_id
2016 && *newest_mount_id > get_desc_mount_id(desc)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002017 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002018 "journal-1087: transaction "
2019 "is valid returning because mount_id %d is less than "
2020 "newest_mount_id %lu",
2021 get_desc_mount_id(desc),
2022 *newest_mount_id);
2023 return -1;
2024 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002025 if (get_desc_trans_len(desc) > SB_JOURNAL(sb)->j_trans_max) {
2026 reiserfs_warning(sb, "journal-2018",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002027 "Bad transaction length %d "
2028 "encountered, ignoring transaction",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002029 get_desc_trans_len(desc));
2030 return -1;
2031 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002032 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002034 /* ok, we have a journal description block, lets see if the transaction was valid */
2035 c_bh =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002036 journal_bread(sb,
2037 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002038 ((offset + get_desc_trans_len(desc) +
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002039 1) % SB_ONDISK_JOURNAL_SIZE(sb)));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002040 if (!c_bh)
2041 return 0;
2042 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002043 if (journal_compare_desc_commit(sb, desc, commit)) {
2044 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002045 "journal_transaction_is_valid, commit offset %ld had bad "
2046 "time %d or length %d",
2047 c_bh->b_blocknr -
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002048 SB_ONDISK_JOURNAL_1st_BLOCK(sb),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002049 get_commit_trans_id(commit),
2050 get_commit_trans_len(commit));
2051 brelse(c_bh);
2052 if (oldest_invalid_trans_id) {
2053 *oldest_invalid_trans_id =
2054 get_desc_trans_id(desc);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002055 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002056 "journal-1004: "
2057 "transaction_is_valid setting oldest invalid trans_id "
2058 "to %d",
2059 get_desc_trans_id(desc));
2060 }
2061 return -1;
2062 }
2063 brelse(c_bh);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002064 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002065 "journal-1006: found valid "
2066 "transaction start offset %llu, len %d id %d",
2067 d_bh->b_blocknr -
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002068 SB_ONDISK_JOURNAL_1st_BLOCK(sb),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002069 get_desc_trans_len(desc),
2070 get_desc_trans_id(desc));
2071 return 1;
2072 } else {
2073 return 0;
2074 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075}
2076
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002077static void brelse_array(struct buffer_head **heads, int num)
2078{
2079 int i;
2080 for (i = 0; i < num; i++) {
2081 brelse(heads[i]);
2082 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083}
2084
2085/*
2086** given the start, and values for the oldest acceptable transactions,
2087** this either reads in a replays a transaction, or returns because the transaction
2088** is invalid, or too old.
2089*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002090static int journal_read_transaction(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002091 unsigned long cur_dblock,
2092 unsigned long oldest_start,
Jeff Mahoney600ed412009-03-30 14:02:17 -04002093 unsigned int oldest_trans_id,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002094 unsigned long newest_mount_id)
2095{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002096 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002097 struct reiserfs_journal_desc *desc;
2098 struct reiserfs_journal_commit *commit;
Jeff Mahoney600ed412009-03-30 14:02:17 -04002099 unsigned int trans_id = 0;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002100 struct buffer_head *c_bh;
2101 struct buffer_head *d_bh;
2102 struct buffer_head **log_blocks = NULL;
2103 struct buffer_head **real_blocks = NULL;
Jeff Mahoney600ed412009-03-30 14:02:17 -04002104 unsigned int trans_offset;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002105 int i;
2106 int trans_half;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002108 d_bh = journal_bread(sb, cur_dblock);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002109 if (!d_bh)
2110 return 1;
2111 desc = (struct reiserfs_journal_desc *)d_bh->b_data;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002112 trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2113 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1037: "
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002114 "journal_read_transaction, offset %llu, len %d mount_id %d",
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002115 d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002116 get_desc_trans_len(desc), get_desc_mount_id(desc));
2117 if (get_desc_trans_id(desc) < oldest_trans_id) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002118 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1039: "
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002119 "journal_read_trans skipping because %lu is too old",
2120 cur_dblock -
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002121 SB_ONDISK_JOURNAL_1st_BLOCK(sb));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002122 brelse(d_bh);
2123 return 1;
2124 }
2125 if (get_desc_mount_id(desc) != newest_mount_id) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002126 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1146: "
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002127 "journal_read_trans skipping because %d is != "
2128 "newest_mount_id %lu", get_desc_mount_id(desc),
2129 newest_mount_id);
2130 brelse(d_bh);
2131 return 1;
2132 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002133 c_bh = journal_bread(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002134 ((trans_offset + get_desc_trans_len(desc) + 1) %
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002135 SB_ONDISK_JOURNAL_SIZE(sb)));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002136 if (!c_bh) {
2137 brelse(d_bh);
2138 return 1;
2139 }
2140 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002141 if (journal_compare_desc_commit(sb, desc, commit)) {
2142 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002143 "journal_read_transaction, "
2144 "commit offset %llu had bad time %d or length %d",
2145 c_bh->b_blocknr -
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002146 SB_ONDISK_JOURNAL_1st_BLOCK(sb),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002147 get_commit_trans_id(commit),
2148 get_commit_trans_len(commit));
2149 brelse(c_bh);
2150 brelse(d_bh);
2151 return 1;
2152 }
Jeff Mahoney3f8b5ee2010-03-23 13:35:39 -07002153
2154 if (bdev_read_only(sb->s_bdev)) {
2155 reiserfs_warning(sb, "clm-2076",
2156 "device is readonly, unable to replay log");
2157 brelse(c_bh);
2158 brelse(d_bh);
2159 return -EROFS;
2160 }
2161
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002162 trans_id = get_desc_trans_id(desc);
2163 /* now we know we've got a good transaction, and it was inside the valid time ranges */
Pekka Enbergd739b422006-02-01 03:06:43 -08002164 log_blocks = kmalloc(get_desc_trans_len(desc) *
2165 sizeof(struct buffer_head *), GFP_NOFS);
2166 real_blocks = kmalloc(get_desc_trans_len(desc) *
2167 sizeof(struct buffer_head *), GFP_NOFS);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002168 if (!log_blocks || !real_blocks) {
2169 brelse(c_bh);
2170 brelse(d_bh);
Pekka Enbergd739b422006-02-01 03:06:43 -08002171 kfree(log_blocks);
2172 kfree(real_blocks);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002173 reiserfs_warning(sb, "journal-1169",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002174 "kmalloc failed, unable to mount FS");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002175 return -1;
2176 }
2177 /* get all the buffer heads */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002178 trans_half = journal_trans_half(sb->s_blocksize);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002179 for (i = 0; i < get_desc_trans_len(desc); i++) {
2180 log_blocks[i] =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002181 journal_getblk(sb,
2182 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002183 (trans_offset + 1 +
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002184 i) % SB_ONDISK_JOURNAL_SIZE(sb));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002185 if (i < trans_half) {
2186 real_blocks[i] =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002187 sb_getblk(sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002188 le32_to_cpu(desc->j_realblock[i]));
2189 } else {
2190 real_blocks[i] =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002191 sb_getblk(sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002192 le32_to_cpu(commit->
2193 j_realblock[i - trans_half]));
2194 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002195 if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(sb)) {
2196 reiserfs_warning(sb, "journal-1207",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002197 "REPLAY FAILURE fsck required! "
2198 "Block to replay is outside of "
2199 "filesystem");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002200 goto abort_replay;
2201 }
2202 /* make sure we don't try to replay onto log or reserved area */
2203 if (is_block_in_log_or_reserved_area
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002204 (sb, real_blocks[i]->b_blocknr)) {
2205 reiserfs_warning(sb, "journal-1204",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002206 "REPLAY FAILURE fsck required! "
2207 "Trying to replay onto a log block");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002208 abort_replay:
2209 brelse_array(log_blocks, i);
2210 brelse_array(real_blocks, i);
2211 brelse(c_bh);
2212 brelse(d_bh);
Pekka Enbergd739b422006-02-01 03:06:43 -08002213 kfree(log_blocks);
2214 kfree(real_blocks);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002215 return -1;
2216 }
2217 }
2218 /* read in the log blocks, memcpy to the corresponding real block */
2219 ll_rw_block(READ, get_desc_trans_len(desc), log_blocks);
2220 for (i = 0; i < get_desc_trans_len(desc); i++) {
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02002221
2222 reiserfs_write_unlock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002223 wait_on_buffer(log_blocks[i]);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02002224 reiserfs_write_lock(sb);
2225
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002226 if (!buffer_uptodate(log_blocks[i])) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002227 reiserfs_warning(sb, "journal-1212",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002228 "REPLAY FAILURE fsck required! "
2229 "buffer write failed");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002230 brelse_array(log_blocks + i,
2231 get_desc_trans_len(desc) - i);
2232 brelse_array(real_blocks, get_desc_trans_len(desc));
2233 brelse(c_bh);
2234 brelse(d_bh);
Pekka Enbergd739b422006-02-01 03:06:43 -08002235 kfree(log_blocks);
2236 kfree(real_blocks);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002237 return -1;
2238 }
2239 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2240 real_blocks[i]->b_size);
2241 set_buffer_uptodate(real_blocks[i]);
2242 brelse(log_blocks[i]);
2243 }
2244 /* flush out the real blocks */
2245 for (i = 0; i < get_desc_trans_len(desc); i++) {
2246 set_buffer_dirty(real_blocks[i]);
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02002247 write_dirty_buffer(real_blocks[i], WRITE);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002248 }
2249 for (i = 0; i < get_desc_trans_len(desc); i++) {
2250 wait_on_buffer(real_blocks[i]);
2251 if (!buffer_uptodate(real_blocks[i])) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002252 reiserfs_warning(sb, "journal-1226",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002253 "REPLAY FAILURE, fsck required! "
2254 "buffer write failed");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002255 brelse_array(real_blocks + i,
2256 get_desc_trans_len(desc) - i);
2257 brelse(c_bh);
2258 brelse(d_bh);
Pekka Enbergd739b422006-02-01 03:06:43 -08002259 kfree(log_blocks);
2260 kfree(real_blocks);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002261 return -1;
2262 }
2263 brelse(real_blocks[i]);
2264 }
2265 cur_dblock =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002266 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002267 ((trans_offset + get_desc_trans_len(desc) +
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002268 2) % SB_ONDISK_JOURNAL_SIZE(sb));
2269 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002270 "journal-1095: setting journal " "start to offset %ld",
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002271 cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002272
2273 /* init starting values for the first transaction, in case this is the last transaction to be replayed. */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002274 journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002275 journal->j_last_flush_trans_id = trans_id;
2276 journal->j_trans_id = trans_id + 1;
Alexander Zarochentseva44c94a2006-03-25 03:06:59 -08002277 /* check for trans_id overflow */
2278 if (journal->j_trans_id == 0)
2279 journal->j_trans_id = 10;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002280 brelse(c_bh);
2281 brelse(d_bh);
Pekka Enbergd739b422006-02-01 03:06:43 -08002282 kfree(log_blocks);
2283 kfree(real_blocks);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002284 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285}
2286
2287/* This function reads blocks starting from block and to max_block of bufsize
2288 size (but no more than BUFNR blocks at a time). This proved to improve
2289 mounting speed on self-rebuilding raid5 arrays at least.
2290 Right now it is only used from journal code. But later we might use it
2291 from other places.
2292 Note: Do not use journal_getblk/sb_getblk functions here! */
Jeff Mahoney3ee16672007-10-18 23:39:25 -07002293static struct buffer_head *reiserfs_breada(struct block_device *dev,
2294 b_blocknr_t block, int bufsize,
2295 b_blocknr_t max_block)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002297 struct buffer_head *bhlist[BUFNR];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298 unsigned int blocks = BUFNR;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002299 struct buffer_head *bh;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 int i, j;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002301
2302 bh = __getblk(dev, block, bufsize);
2303 if (buffer_uptodate(bh))
2304 return (bh);
2305
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306 if (block + BUFNR > max_block) {
2307 blocks = max_block - block;
2308 }
2309 bhlist[0] = bh;
2310 j = 1;
2311 for (i = 1; i < blocks; i++) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002312 bh = __getblk(dev, block + i, bufsize);
2313 if (buffer_uptodate(bh)) {
2314 brelse(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315 break;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002316 } else
2317 bhlist[j++] = bh;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002319 ll_rw_block(READ, j, bhlist);
2320 for (i = 1; i < j; i++)
2321 brelse(bhlist[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322 bh = bhlist[0];
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002323 wait_on_buffer(bh);
2324 if (buffer_uptodate(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325 return bh;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002326 brelse(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327 return NULL;
2328}
2329
2330/*
2331** read and replay the log
2332** on a clean unmount, the journal header's next unflushed pointer will be to an invalid
2333** transaction. This tests that before finding all the transactions in the log, which makes normal mount times fast.
2334**
2335** After a crash, this starts with the next unflushed transaction, and replays until it finds one too old, or invalid.
2336**
2337** On exit, it sets things up so the first transaction will work correctly.
2338*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002339static int journal_read(struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002340{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002341 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002342 struct reiserfs_journal_desc *desc;
Jeff Mahoney600ed412009-03-30 14:02:17 -04002343 unsigned int oldest_trans_id = 0;
2344 unsigned int oldest_invalid_trans_id = 0;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002345 time_t start;
2346 unsigned long oldest_start = 0;
2347 unsigned long cur_dblock = 0;
2348 unsigned long newest_mount_id = 9;
2349 struct buffer_head *d_bh;
2350 struct reiserfs_journal_header *jh;
2351 int valid_journal_header = 0;
2352 int replay_count = 0;
2353 int continue_replay = 1;
2354 int ret;
2355 char b[BDEVNAME_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002357 cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2358 reiserfs_info(sb, "checking transaction log (%s)\n",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002359 bdevname(journal->j_dev_bd, b));
2360 start = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361
Jeff Mahoney0222e652009-03-30 14:02:44 -04002362 /* step 1, read in the journal header block. Check the transaction it says
2363 ** is the first unflushed, and if that transaction is not valid,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002364 ** replay is done
2365 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002366 journal->j_header_bh = journal_bread(sb,
2367 SB_ONDISK_JOURNAL_1st_BLOCK(sb)
2368 + SB_ONDISK_JOURNAL_SIZE(sb));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002369 if (!journal->j_header_bh) {
2370 return 1;
2371 }
2372 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
Vladimir V. Savelievc499ec22006-03-02 02:54:39 -08002373 if (le32_to_cpu(jh->j_first_unflushed_offset) <
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002374 SB_ONDISK_JOURNAL_SIZE(sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002375 && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2376 oldest_start =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002377 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002378 le32_to_cpu(jh->j_first_unflushed_offset);
2379 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2380 newest_mount_id = le32_to_cpu(jh->j_mount_id);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002381 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002382 "journal-1153: found in "
2383 "header: first_unflushed_offset %d, last_flushed_trans_id "
2384 "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2385 le32_to_cpu(jh->j_last_flush_trans_id));
2386 valid_journal_header = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387
Jeff Mahoney0222e652009-03-30 14:02:44 -04002388 /* now, we try to read the first unflushed offset. If it is not valid,
2389 ** there is nothing more we can do, and it makes no sense to read
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002390 ** through the whole log.
2391 */
2392 d_bh =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002393 journal_bread(sb,
2394 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002395 le32_to_cpu(jh->j_first_unflushed_offset));
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002396 ret = journal_transaction_is_valid(sb, d_bh, NULL, NULL);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002397 if (!ret) {
2398 continue_replay = 0;
2399 }
2400 brelse(d_bh);
2401 goto start_log_replay;
2402 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002404 /* ok, there are transactions that need to be replayed. start with the first log block, find
2405 ** all the valid transactions, and pick out the oldest.
2406 */
2407 while (continue_replay
2408 && cur_dblock <
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002409 (SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2410 SB_ONDISK_JOURNAL_SIZE(sb))) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002411 /* Note that it is required for blocksize of primary fs device and journal
2412 device to be the same */
2413 d_bh =
2414 reiserfs_breada(journal->j_dev_bd, cur_dblock,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002415 sb->s_blocksize,
2416 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2417 SB_ONDISK_JOURNAL_SIZE(sb));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002418 ret =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002419 journal_transaction_is_valid(sb, d_bh,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002420 &oldest_invalid_trans_id,
2421 &newest_mount_id);
2422 if (ret == 1) {
2423 desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2424 if (oldest_start == 0) { /* init all oldest_ values */
2425 oldest_trans_id = get_desc_trans_id(desc);
2426 oldest_start = d_bh->b_blocknr;
2427 newest_mount_id = get_desc_mount_id(desc);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002428 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002429 "journal-1179: Setting "
2430 "oldest_start to offset %llu, trans_id %lu",
2431 oldest_start -
2432 SB_ONDISK_JOURNAL_1st_BLOCK
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002433 (sb), oldest_trans_id);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002434 } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2435 /* one we just read was older */
2436 oldest_trans_id = get_desc_trans_id(desc);
2437 oldest_start = d_bh->b_blocknr;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002438 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002439 "journal-1180: Resetting "
2440 "oldest_start to offset %lu, trans_id %lu",
2441 oldest_start -
2442 SB_ONDISK_JOURNAL_1st_BLOCK
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002443 (sb), oldest_trans_id);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002444 }
2445 if (newest_mount_id < get_desc_mount_id(desc)) {
2446 newest_mount_id = get_desc_mount_id(desc);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002447 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002448 "journal-1299: Setting "
2449 "newest_mount_id to %d",
2450 get_desc_mount_id(desc));
2451 }
2452 cur_dblock += get_desc_trans_len(desc) + 2;
2453 } else {
2454 cur_dblock++;
2455 }
2456 brelse(d_bh);
2457 }
2458
2459 start_log_replay:
2460 cur_dblock = oldest_start;
2461 if (oldest_trans_id) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002462 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002463 "journal-1206: Starting replay "
2464 "from offset %llu, trans_id %lu",
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002465 cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002466 oldest_trans_id);
2467
2468 }
2469 replay_count = 0;
2470 while (continue_replay && oldest_trans_id > 0) {
2471 ret =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002472 journal_read_transaction(sb, cur_dblock, oldest_start,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002473 oldest_trans_id, newest_mount_id);
2474 if (ret < 0) {
2475 return ret;
2476 } else if (ret != 0) {
2477 break;
2478 }
2479 cur_dblock =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002480 SB_ONDISK_JOURNAL_1st_BLOCK(sb) + journal->j_start;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002481 replay_count++;
2482 if (cur_dblock == oldest_start)
2483 break;
2484 }
2485
2486 if (oldest_trans_id == 0) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002487 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002488 "journal-1225: No valid " "transactions found");
2489 }
2490 /* j_start does not get set correctly if we don't replay any transactions.
2491 ** if we had a valid journal_header, set j_start to the first unflushed transaction value,
2492 ** copy the trans_id from the header
2493 */
2494 if (valid_journal_header && replay_count == 0) {
2495 journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2496 journal->j_trans_id =
2497 le32_to_cpu(jh->j_last_flush_trans_id) + 1;
Alexander Zarochentseva44c94a2006-03-25 03:06:59 -08002498 /* check for trans_id overflow */
2499 if (journal->j_trans_id == 0)
2500 journal->j_trans_id = 10;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002501 journal->j_last_flush_trans_id =
2502 le32_to_cpu(jh->j_last_flush_trans_id);
2503 journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2504 } else {
2505 journal->j_mount_id = newest_mount_id + 1;
2506 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002507 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002508 "newest_mount_id to %lu", journal->j_mount_id);
2509 journal->j_first_unflushed_offset = journal->j_start;
2510 if (replay_count > 0) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002511 reiserfs_info(sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002512 "replayed %d transactions in %lu seconds\n",
2513 replay_count, get_seconds() - start);
2514 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002515 if (!bdev_read_only(sb->s_bdev) &&
2516 _update_journal_header_block(sb, journal->j_start,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002517 journal->j_last_flush_trans_id)) {
2518 /* replay failed, caller must call free_journal_ram and abort
2519 ** the mount
2520 */
2521 return -1;
2522 }
2523 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524}
2525
2526static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2527{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002528 struct reiserfs_journal_list *jl;
Pekka Enberg8c777cc2006-02-01 03:06:43 -08002529 jl = kzalloc(sizeof(struct reiserfs_journal_list),
2530 GFP_NOFS | __GFP_NOFAIL);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002531 INIT_LIST_HEAD(&jl->j_list);
2532 INIT_LIST_HEAD(&jl->j_working_list);
2533 INIT_LIST_HEAD(&jl->j_tail_bh_list);
2534 INIT_LIST_HEAD(&jl->j_bh_list);
Jeff Mahoney90415de2008-07-25 01:46:40 -07002535 mutex_init(&jl->j_commit_mutex);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002536 SB_JOURNAL(s)->j_num_lists++;
2537 get_journal_list(jl);
2538 return jl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002539}
2540
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002541static void journal_list_init(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002543 SB_JOURNAL(sb)->j_current_jl = alloc_journal_list(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544}
2545
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002546static int release_journal_dev(struct super_block *super,
2547 struct reiserfs_journal *journal)
2548{
2549 int result;
2550
2551 result = 0;
2552
Christoph Hellwig86098fa2008-04-30 00:54:46 -07002553 if (journal->j_dev_bd != NULL) {
Al Viroe5eb8ca2007-10-08 13:24:05 -04002554 result = blkdev_put(journal->j_dev_bd, journal->j_dev_mode);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002555 journal->j_dev_bd = NULL;
2556 }
2557
2558 if (result != 0) {
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002559 reiserfs_warning(super, "sh-457",
2560 "Cannot release journal device: %i", result);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002561 }
2562 return result;
2563}
2564
2565static int journal_init_dev(struct super_block *super,
2566 struct reiserfs_journal *journal,
2567 const char *jdev_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568{
2569 int result;
2570 dev_t jdev;
Tejun Heoe525fd82010-11-13 11:55:17 +01002571 fmode_t blkdev_mode = FMODE_READ | FMODE_WRITE | FMODE_EXCL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572 char b[BDEVNAME_SIZE];
2573
2574 result = 0;
2575
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002576 journal->j_dev_bd = NULL;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002577 jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2578 new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579
2580 if (bdev_read_only(super->s_bdev))
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002581 blkdev_mode = FMODE_READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582
2583 /* there is no "jdev" option and journal is on separate device */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002584 if ((!jdev_name || !jdev_name[0])) {
Tejun Heoe525fd82010-11-13 11:55:17 +01002585 if (jdev == super->s_dev)
2586 blkdev_mode &= ~FMODE_EXCL;
Tejun Heod4d77622010-11-13 11:55:18 +01002587 journal->j_dev_bd = blkdev_get_by_dev(jdev, blkdev_mode,
2588 journal);
Al Viroe5eb8ca2007-10-08 13:24:05 -04002589 journal->j_dev_mode = blkdev_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590 if (IS_ERR(journal->j_dev_bd)) {
2591 result = PTR_ERR(journal->j_dev_bd);
2592 journal->j_dev_bd = NULL;
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002593 reiserfs_warning(super, "sh-458",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002594 "cannot init journal device '%s': %i",
2595 __bdevname(jdev, b), result);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596 return result;
Tejun Heoe525fd82010-11-13 11:55:17 +01002597 } else if (jdev != super->s_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598 set_blocksize(journal->j_dev_bd, super->s_blocksize);
Christoph Hellwig86098fa2008-04-30 00:54:46 -07002599
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600 return 0;
2601 }
2602
Al Viroe5eb8ca2007-10-08 13:24:05 -04002603 journal->j_dev_mode = blkdev_mode;
Tejun Heod4d77622010-11-13 11:55:18 +01002604 journal->j_dev_bd = blkdev_get_by_path(jdev_name, blkdev_mode, journal);
Christoph Hellwig86098fa2008-04-30 00:54:46 -07002605 if (IS_ERR(journal->j_dev_bd)) {
2606 result = PTR_ERR(journal->j_dev_bd);
2607 journal->j_dev_bd = NULL;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002608 reiserfs_warning(super,
2609 "journal_init_dev: Cannot open '%s': %i",
2610 jdev_name, result);
Christoph Hellwig86098fa2008-04-30 00:54:46 -07002611 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612 }
Christoph Hellwig86098fa2008-04-30 00:54:46 -07002613
2614 set_blocksize(journal->j_dev_bd, super->s_blocksize);
2615 reiserfs_info(super,
2616 "journal_init_dev: journal device: %s\n",
2617 bdevname(journal->j_dev_bd, b));
2618 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619}
2620
Edward Shishkincf3d0b82007-10-16 23:30:30 -07002621/**
2622 * When creating/tuning a file system user can assign some
2623 * journal params within boundaries which depend on the ratio
2624 * blocksize/standard_blocksize.
2625 *
2626 * For blocks >= standard_blocksize transaction size should
2627 * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more
2628 * then JOURNAL_TRANS_MAX_DEFAULT.
2629 *
2630 * For blocks < standard_blocksize these boundaries should be
2631 * decreased proportionally.
2632 */
2633#define REISERFS_STANDARD_BLKSIZE (4096)
2634
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002635static int check_advise_trans_params(struct super_block *sb,
Edward Shishkincf3d0b82007-10-16 23:30:30 -07002636 struct reiserfs_journal *journal)
2637{
2638 if (journal->j_trans_max) {
2639 /* Non-default journal params.
2640 Do sanity check for them. */
2641 int ratio = 1;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002642 if (sb->s_blocksize < REISERFS_STANDARD_BLKSIZE)
2643 ratio = REISERFS_STANDARD_BLKSIZE / sb->s_blocksize;
Edward Shishkincf3d0b82007-10-16 23:30:30 -07002644
2645 if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio ||
2646 journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio ||
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002647 SB_ONDISK_JOURNAL_SIZE(sb) / journal->j_trans_max <
Edward Shishkincf3d0b82007-10-16 23:30:30 -07002648 JOURNAL_MIN_RATIO) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002649 reiserfs_warning(sb, "sh-462",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002650 "bad transaction max size (%u). "
2651 "FSCK?", journal->j_trans_max);
Edward Shishkincf3d0b82007-10-16 23:30:30 -07002652 return 1;
2653 }
2654 if (journal->j_max_batch != (journal->j_trans_max) *
2655 JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002656 reiserfs_warning(sb, "sh-463",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002657 "bad transaction max batch (%u). "
2658 "FSCK?", journal->j_max_batch);
Edward Shishkincf3d0b82007-10-16 23:30:30 -07002659 return 1;
2660 }
2661 } else {
2662 /* Default journal params.
2663 The file system was created by old version
2664 of mkreiserfs, so some fields contain zeros,
2665 and we need to advise proper values for them */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002666 if (sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) {
2667 reiserfs_warning(sb, "sh-464", "bad blocksize (%u)",
2668 sb->s_blocksize);
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002669 return 1;
2670 }
Edward Shishkincf3d0b82007-10-16 23:30:30 -07002671 journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2672 journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2673 journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2674 }
2675 return 0;
2676}
2677
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678/*
2679** must be called once on fs mount. calls journal_read for you
2680*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002681int journal_init(struct super_block *sb, const char *j_dev_name,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002682 int old_format, unsigned int commit_max_age)
2683{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002684 int num_cnodes = SB_ONDISK_JOURNAL_SIZE(sb) * 2;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002685 struct buffer_head *bhjh;
2686 struct reiserfs_super_block *rs;
2687 struct reiserfs_journal_header *jh;
2688 struct reiserfs_journal *journal;
2689 struct reiserfs_journal_list *jl;
2690 char b[BDEVNAME_SIZE];
Frederic Weisbecker98ea3f52009-12-29 21:51:15 +01002691 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692
Frederic Weisbecker98ea3f52009-12-29 21:51:15 +01002693 /*
2694 * Unlock here to avoid various RECLAIM-FS-ON <-> IN-RECLAIM-FS
2695 * dependency inversion warnings.
2696 */
2697 reiserfs_write_unlock(sb);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002698 journal = SB_JOURNAL(sb) = vmalloc(sizeof(struct reiserfs_journal));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002699 if (!journal) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002700 reiserfs_warning(sb, "journal-1256",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002701 "unable to get memory for journal structure");
Frederic Weisbecker98ea3f52009-12-29 21:51:15 +01002702 reiserfs_write_lock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002703 return 1;
2704 }
2705 memset(journal, 0, sizeof(struct reiserfs_journal));
2706 INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2707 INIT_LIST_HEAD(&journal->j_prealloc_list);
2708 INIT_LIST_HEAD(&journal->j_working_list);
2709 INIT_LIST_HEAD(&journal->j_journal_list);
2710 journal->j_persistent_trans = 0;
Frederic Weisbecker98ea3f52009-12-29 21:51:15 +01002711 ret = reiserfs_allocate_list_bitmaps(sb, journal->j_list_bitmap,
2712 reiserfs_bmap_count(sb));
2713 reiserfs_write_lock(sb);
2714 if (ret)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002715 goto free_and_return;
Frederic Weisbecker98ea3f52009-12-29 21:51:15 +01002716
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002717 allocate_bitmap_nodes(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002719 /* reserved for journal area support */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002720 SB_JOURNAL_1st_RESERVED_BLOCK(sb) = (old_format ?
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002721 REISERFS_OLD_DISK_OFFSET_IN_BYTES
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002722 / sb->s_blocksize +
2723 reiserfs_bmap_count(sb) +
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002724 1 :
2725 REISERFS_DISK_OFFSET_IN_BYTES /
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002726 sb->s_blocksize + 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002727
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002728 /* Sanity check to see is the standard journal fitting withing first bitmap
2729 (actual for small blocksizes) */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002730 if (!SB_ONDISK_JOURNAL_DEVICE(sb) &&
2731 (SB_JOURNAL_1st_RESERVED_BLOCK(sb) +
2732 SB_ONDISK_JOURNAL_SIZE(sb) > sb->s_blocksize * 8)) {
2733 reiserfs_warning(sb, "journal-1393",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002734 "journal does not fit for area addressed "
2735 "by first of bitmap blocks. It starts at "
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002736 "%u and its size is %u. Block size %ld",
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002737 SB_JOURNAL_1st_RESERVED_BLOCK(sb),
2738 SB_ONDISK_JOURNAL_SIZE(sb),
2739 sb->s_blocksize);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002740 goto free_and_return;
2741 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742
Frederic Weisbecker193be0e2009-09-17 05:31:37 +02002743 /*
2744 * We need to unlock here to avoid creating the following
2745 * dependency:
2746 * reiserfs_lock -> sysfs_mutex
2747 * Because the reiserfs mmap path creates the following dependency:
2748 * mm->mmap -> reiserfs_lock, hence we have
2749 * mm->mmap -> reiserfs_lock ->sysfs_mutex
2750 * This would ends up in a circular dependency with sysfs readdir path
2751 * which does sysfs_mutex -> mm->mmap_sem
2752 * This is fine because the reiserfs lock is useless in mount path,
2753 * at least until we call journal_begin. We keep it for paranoid
2754 * reasons.
2755 */
2756 reiserfs_write_unlock(sb);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002757 if (journal_init_dev(sb, journal, j_dev_name) != 0) {
Frederic Weisbecker193be0e2009-09-17 05:31:37 +02002758 reiserfs_write_lock(sb);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002759 reiserfs_warning(sb, "sh-462",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002760 "unable to initialize jornal device");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002761 goto free_and_return;
2762 }
Frederic Weisbecker193be0e2009-09-17 05:31:37 +02002763 reiserfs_write_lock(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002764
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002765 rs = SB_DISK_SUPER_BLOCK(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002766
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002767 /* read journal header */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002768 bhjh = journal_bread(sb,
2769 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2770 SB_ONDISK_JOURNAL_SIZE(sb));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002771 if (!bhjh) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002772 reiserfs_warning(sb, "sh-459",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002773 "unable to read journal header");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002774 goto free_and_return;
2775 }
2776 jh = (struct reiserfs_journal_header *)(bhjh->b_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002778 /* make sure that journal matches to the super block */
2779 if (is_reiserfs_jr(rs)
2780 && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2781 sb_jp_journal_magic(rs))) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002782 reiserfs_warning(sb, "sh-460",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002783 "journal header magic %x (device %s) does "
2784 "not match to magic found in super block %x",
2785 jh->jh_journal.jp_journal_magic,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002786 bdevname(journal->j_dev_bd, b),
2787 sb_jp_journal_magic(rs));
2788 brelse(bhjh);
2789 goto free_and_return;
2790 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002792 journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2793 journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2794 journal->j_max_commit_age =
2795 le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2796 journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002798 if (check_advise_trans_params(sb, journal) != 0)
Edward Shishkincf3d0b82007-10-16 23:30:30 -07002799 goto free_and_return;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002800 journal->j_default_max_commit_age = journal->j_max_commit_age;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002801
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002802 if (commit_max_age != 0) {
2803 journal->j_max_commit_age = commit_max_age;
2804 journal->j_max_trans_age = commit_max_age;
2805 }
2806
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002807 reiserfs_info(sb, "journal params: device %s, size %u, "
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002808 "journal first block %u, max trans len %u, max batch %u, "
2809 "max commit age %u, max trans age %u\n",
2810 bdevname(journal->j_dev_bd, b),
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002811 SB_ONDISK_JOURNAL_SIZE(sb),
2812 SB_ONDISK_JOURNAL_1st_BLOCK(sb),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002813 journal->j_trans_max,
2814 journal->j_max_batch,
2815 journal->j_max_commit_age, journal->j_max_trans_age);
2816
2817 brelse(bhjh);
2818
2819 journal->j_list_bitmap_index = 0;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002820 journal_list_init(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002821
2822 memset(journal->j_list_hash_table, 0,
2823 JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2824
2825 INIT_LIST_HEAD(&journal->j_dirty_buffers);
2826 spin_lock_init(&journal->j_dirty_buffers_lock);
2827
2828 journal->j_start = 0;
2829 journal->j_len = 0;
2830 journal->j_len_alloc = 0;
2831 atomic_set(&(journal->j_wcount), 0);
2832 atomic_set(&(journal->j_async_throttle), 0);
2833 journal->j_bcount = 0;
2834 journal->j_trans_start_time = 0;
2835 journal->j_last = NULL;
2836 journal->j_first = NULL;
2837 init_waitqueue_head(&(journal->j_join_wait));
Jeff Mahoneyf68215c2008-07-25 01:46:38 -07002838 mutex_init(&journal->j_mutex);
Jeff Mahoneyafe70252008-07-25 01:46:39 -07002839 mutex_init(&journal->j_flush_mutex);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002840
2841 journal->j_trans_id = 10;
2842 journal->j_mount_id = 10;
2843 journal->j_state = 0;
2844 atomic_set(&(journal->j_jlock), 0);
Frederic Weisbeckerbbec9192010-01-28 13:43:50 +01002845 reiserfs_write_unlock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002846 journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
Frederic Weisbeckerbbec9192010-01-28 13:43:50 +01002847 reiserfs_write_lock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002848 journal->j_cnode_free_orig = journal->j_cnode_free_list;
2849 journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2850 journal->j_cnode_used = 0;
2851 journal->j_must_wait = 0;
2852
Jeff Mahoney576f6d72005-11-29 19:34:39 -08002853 if (journal->j_cnode_free == 0) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002854 reiserfs_warning(sb, "journal-2004", "Journal cnode memory "
Jeff Mahoney576f6d72005-11-29 19:34:39 -08002855 "allocation failed (%ld bytes). Journal is "
2856 "too large for available memory. Usually "
2857 "this is due to a journal that is too large.",
2858 sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2859 goto free_and_return;
2860 }
2861
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002862 init_journal_hash(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002863 jl = journal->j_current_jl;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002864 jl->j_list_bitmap = get_list_bitmap(sb, jl);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002865 if (!jl->j_list_bitmap) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002866 reiserfs_warning(sb, "journal-2005",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002867 "get_list_bitmap failed for journal list 0");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002868 goto free_and_return;
2869 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002870 if (journal_read(sb) < 0) {
2871 reiserfs_warning(sb, "reiserfs-2006",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002872 "Replay Failure, unable to mount");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002873 goto free_and_return;
2874 }
2875
2876 reiserfs_mounted_fs_count++;
Frederic Weisbecker48f6ba52009-10-05 16:31:37 +02002877 if (reiserfs_mounted_fs_count <= 1) {
2878 reiserfs_write_unlock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002879 commit_wq = create_workqueue("reiserfs");
Frederic Weisbecker48f6ba52009-10-05 16:31:37 +02002880 reiserfs_write_lock(sb);
2881 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002882
David Howellsc4028952006-11-22 14:57:56 +00002883 INIT_DELAYED_WORK(&journal->j_work, flush_async_commits);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002884 journal->j_work_sb = sb;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002885 return 0;
2886 free_and_return:
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002887 free_journal_ram(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002888 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002889}
2890
2891/*
2892** test for a polite end of the current transaction. Used by file_write, and should
2893** be used by delete to make sure they don't write more than can fit inside a single
2894** transaction
2895*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002896int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2897 int new_alloc)
2898{
2899 struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2900 time_t now = get_seconds();
2901 /* cannot restart while nested */
2902 BUG_ON(!th->t_trans_id);
2903 if (th->t_refcount > 1)
2904 return 0;
2905 if (journal->j_must_wait > 0 ||
2906 (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2907 atomic_read(&(journal->j_jlock)) ||
2908 (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2909 journal->j_cnode_free < (journal->j_trans_max * 3)) {
2910 return 1;
2911 }
Chris Mason6ae1ea42006-02-01 03:06:50 -08002912 /* protected by the BKL here */
2913 journal->j_len_alloc += new_alloc;
2914 th->t_blocks_allocated += new_alloc ;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002915 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002916}
2917
Jeff Mahoney0222e652009-03-30 14:02:44 -04002918/* this must be called inside a transaction, and requires the
Linus Torvalds1da177e2005-04-16 15:20:36 -07002919** kernel_lock to be held
2920*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002921void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002922{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002923 struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2924 BUG_ON(!th->t_trans_id);
2925 journal->j_must_wait = 1;
2926 set_bit(J_WRITERS_BLOCKED, &journal->j_state);
2927 return;
2928}
2929
2930/* this must be called without a transaction started, and does not
2931** require BKL
2932*/
2933void reiserfs_allow_writes(struct super_block *s)
2934{
2935 struct reiserfs_journal *journal = SB_JOURNAL(s);
2936 clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
2937 wake_up(&journal->j_join_wait);
2938}
2939
2940/* this must be called without a transaction started, and does not
2941** require BKL
2942*/
2943void reiserfs_wait_on_write_block(struct super_block *s)
2944{
2945 struct reiserfs_journal *journal = SB_JOURNAL(s);
2946 wait_event(journal->j_join_wait,
2947 !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
2948}
2949
2950static void queue_log_writer(struct super_block *s)
2951{
2952 wait_queue_t wait;
2953 struct reiserfs_journal *journal = SB_JOURNAL(s);
2954 set_bit(J_WRITERS_QUEUED, &journal->j_state);
2955
2956 /*
2957 * we don't want to use wait_event here because
2958 * we only want to wait once.
2959 */
2960 init_waitqueue_entry(&wait, current);
2961 add_wait_queue(&journal->j_join_wait, &wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002962 set_current_state(TASK_UNINTERRUPTIBLE);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02002963 if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) {
2964 reiserfs_write_unlock(s);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002965 schedule();
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02002966 reiserfs_write_lock(s);
2967 }
Milind Arun Choudhary5ab2f7e2007-05-08 00:30:51 -07002968 __set_current_state(TASK_RUNNING);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002969 remove_wait_queue(&journal->j_join_wait, &wait);
2970}
2971
2972static void wake_queued_writers(struct super_block *s)
2973{
2974 struct reiserfs_journal *journal = SB_JOURNAL(s);
2975 if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
2976 wake_up(&journal->j_join_wait);
2977}
2978
Jeff Mahoney600ed412009-03-30 14:02:17 -04002979static void let_transaction_grow(struct super_block *sb, unsigned int trans_id)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002980{
2981 struct reiserfs_journal *journal = SB_JOURNAL(sb);
2982 unsigned long bcount = journal->j_bcount;
2983 while (1) {
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02002984 reiserfs_write_unlock(sb);
Nishanth Aravamudan041e0e32005-09-10 00:27:23 -07002985 schedule_timeout_uninterruptible(1);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02002986 reiserfs_write_lock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002987 journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
2988 while ((atomic_read(&journal->j_wcount) > 0 ||
2989 atomic_read(&journal->j_jlock)) &&
2990 journal->j_trans_id == trans_id) {
2991 queue_log_writer(sb);
2992 }
2993 if (journal->j_trans_id != trans_id)
2994 break;
2995 if (bcount == journal->j_bcount)
2996 break;
2997 bcount = journal->j_bcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002998 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002999}
3000
3001/* join == true if you must join an existing transaction.
3002** join == false if you can deal with waiting for others to finish
3003**
3004** this will block until the transaction is joinable. send the number of blocks you
3005** expect to use in nblocks.
3006*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003007static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003008 struct super_block *sb, unsigned long nblocks,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003009 int join)
3010{
3011 time_t now = get_seconds();
Jeff Mahoney600ed412009-03-30 14:02:17 -04003012 unsigned int old_trans_id;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003013 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003014 struct reiserfs_transaction_handle myth;
3015 int sched_count = 0;
3016 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003017
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003018 reiserfs_check_lock_depth(sb, "journal_begin");
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003019 BUG_ON(nblocks > journal->j_trans_max);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003020
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003021 PROC_INFO_INC(sb, journal.journal_being);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003022 /* set here for journal_join */
3023 th->t_refcount = 1;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003024 th->t_super = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003025
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003026 relock:
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003027 lock_journal(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003028 if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003029 unlock_journal(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003030 retval = journal->j_errno;
3031 goto out_fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003032 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003033 journal->j_bcount++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003034
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003035 if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003036 unlock_journal(sb);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02003037 reiserfs_write_unlock(sb);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003038 reiserfs_wait_on_write_block(sb);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02003039 reiserfs_write_lock(sb);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003040 PROC_INFO_INC(sb, journal.journal_relock_writers);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003041 goto relock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003042 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003043 now = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003044
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003045 /* if there is no room in the journal OR
Jeff Mahoney0222e652009-03-30 14:02:44 -04003046 ** 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 -07003047 ** we don't sleep if there aren't other writers
3048 */
3049
3050 if ((!join && journal->j_must_wait > 0) ||
3051 (!join
3052 && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
3053 || (!join && atomic_read(&journal->j_wcount) > 0
3054 && journal->j_trans_start_time > 0
3055 && (now - journal->j_trans_start_time) >
3056 journal->j_max_trans_age) || (!join
3057 && atomic_read(&journal->j_jlock))
3058 || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
3059
3060 old_trans_id = journal->j_trans_id;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003061 unlock_journal(sb); /* allow others to finish this transaction */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003062
3063 if (!join && (journal->j_len_alloc + nblocks + 2) >=
3064 journal->j_max_batch &&
3065 ((journal->j_len + nblocks + 2) * 100) <
3066 (journal->j_len_alloc * 75)) {
3067 if (atomic_read(&journal->j_wcount) > 10) {
3068 sched_count++;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003069 queue_log_writer(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003070 goto relock;
3071 }
3072 }
3073 /* don't mess with joining the transaction if all we have to do is
3074 * wait for someone else to do a commit
3075 */
3076 if (atomic_read(&journal->j_jlock)) {
3077 while (journal->j_trans_id == old_trans_id &&
3078 atomic_read(&journal->j_jlock)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003079 queue_log_writer(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003080 }
3081 goto relock;
3082 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003083 retval = journal_join(&myth, sb, 1);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003084 if (retval)
3085 goto out_fail;
3086
3087 /* someone might have ended the transaction while we joined */
3088 if (old_trans_id != journal->j_trans_id) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003089 retval = do_journal_end(&myth, sb, 1, 0);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003090 } else {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003091 retval = do_journal_end(&myth, sb, 1, COMMIT_NOW);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003092 }
3093
3094 if (retval)
3095 goto out_fail;
3096
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003097 PROC_INFO_INC(sb, journal.journal_relock_wcount);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003098 goto relock;
3099 }
3100 /* we are the first writer, set trans_id */
3101 if (journal->j_trans_start_time == 0) {
3102 journal->j_trans_start_time = get_seconds();
3103 }
3104 atomic_inc(&(journal->j_wcount));
3105 journal->j_len_alloc += nblocks;
3106 th->t_blocks_logged = 0;
3107 th->t_blocks_allocated = nblocks;
3108 th->t_trans_id = journal->j_trans_id;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003109 unlock_journal(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003110 INIT_LIST_HEAD(&th->t_list);
3111 get_fs_excl();
3112 return 0;
3113
3114 out_fail:
3115 memset(th, 0, sizeof(*th));
3116 /* Re-set th->t_super, so we can properly keep track of how many
3117 * persistent transactions there are. We need to do this so if this
3118 * call is part of a failed restart_transaction, we can free it later */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003119 th->t_super = sb;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003120 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003121}
3122
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003123struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3124 super_block
3125 *s,
3126 int nblocks)
3127{
3128 int ret;
3129 struct reiserfs_transaction_handle *th;
3130
3131 /* if we're nesting into an existing transaction. It will be
3132 ** persistent on its own
3133 */
3134 if (reiserfs_transaction_running(s)) {
3135 th = current->journal_info;
3136 th->t_refcount++;
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003137 BUG_ON(th->t_refcount < 2);
3138
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003139 return th;
3140 }
Pekka Enbergd739b422006-02-01 03:06:43 -08003141 th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003142 if (!th)
3143 return NULL;
3144 ret = journal_begin(th, s, nblocks);
3145 if (ret) {
Pekka Enbergd739b422006-02-01 03:06:43 -08003146 kfree(th);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003147 return NULL;
3148 }
3149
3150 SB_JOURNAL(s)->j_persistent_trans++;
3151 return th;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003152}
3153
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003154int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3155{
3156 struct super_block *s = th->t_super;
3157 int ret = 0;
3158 if (th->t_trans_id)
3159 ret = journal_end(th, th->t_super, th->t_blocks_allocated);
3160 else
3161 ret = -EIO;
3162 if (th->t_refcount == 0) {
3163 SB_JOURNAL(s)->j_persistent_trans--;
Pekka Enbergd739b422006-02-01 03:06:43 -08003164 kfree(th);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003165 }
3166 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003167}
3168
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003169static int journal_join(struct reiserfs_transaction_handle *th,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003170 struct super_block *sb, unsigned long nblocks)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003171{
3172 struct reiserfs_transaction_handle *cur_th = current->journal_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003173
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003174 /* this keeps do_journal_end from NULLing out the current->journal_info
3175 ** pointer
3176 */
3177 th->t_handle_save = cur_th;
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003178 BUG_ON(cur_th && cur_th->t_refcount > 1);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003179 return do_journal_begin_r(th, sb, nblocks, JBEGIN_JOIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003180}
3181
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003182int journal_join_abort(struct reiserfs_transaction_handle *th,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003183 struct super_block *sb, unsigned long nblocks)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003184{
3185 struct reiserfs_transaction_handle *cur_th = current->journal_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003186
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003187 /* this keeps do_journal_end from NULLing out the current->journal_info
3188 ** pointer
3189 */
3190 th->t_handle_save = cur_th;
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003191 BUG_ON(cur_th && cur_th->t_refcount > 1);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003192 return do_journal_begin_r(th, sb, nblocks, JBEGIN_ABORT);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003193}
3194
3195int journal_begin(struct reiserfs_transaction_handle *th,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003196 struct super_block *sb, unsigned long nblocks)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003197{
3198 struct reiserfs_transaction_handle *cur_th = current->journal_info;
3199 int ret;
3200
3201 th->t_handle_save = NULL;
3202 if (cur_th) {
3203 /* we are nesting into the current transaction */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003204 if (cur_th->t_super == sb) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003205 BUG_ON(!cur_th->t_refcount);
3206 cur_th->t_refcount++;
3207 memcpy(th, cur_th, sizeof(*th));
3208 if (th->t_refcount <= 1)
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003209 reiserfs_warning(sb, "reiserfs-2005",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04003210 "BAD: refcount <= 1, but "
3211 "journal_info != 0");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003212 return 0;
3213 } else {
3214 /* we've ended up with a handle from a different filesystem.
3215 ** save it and restore on journal_end. This should never
3216 ** really happen...
3217 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003218 reiserfs_warning(sb, "clm-2100",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04003219 "nesting info a different FS");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003220 th->t_handle_save = current->journal_info;
3221 current->journal_info = th;
3222 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003223 } else {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003224 current->journal_info = th;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003225 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003226 ret = do_journal_begin_r(th, sb, nblocks, JBEGIN_REG);
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003227 BUG_ON(current->journal_info != th);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003228
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003229 /* I guess this boils down to being the reciprocal of clm-2100 above.
3230 * If do_journal_begin_r fails, we need to put it back, since journal_end
3231 * won't be called to do it. */
3232 if (ret)
3233 current->journal_info = th->t_handle_save;
3234 else
3235 BUG_ON(!th->t_refcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003236
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003237 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003238}
3239
3240/*
3241** puts bh into the current transaction. If it was already there, reorders removes the
3242** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order).
3243**
3244** if it was dirty, cleans and files onto the clean list. I can't let it be dirty again until the
3245** transaction is committed.
Jeff Mahoney0222e652009-03-30 14:02:44 -04003246**
Linus Torvalds1da177e2005-04-16 15:20:36 -07003247** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3248*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003249int journal_mark_dirty(struct reiserfs_transaction_handle *th,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003250 struct super_block *sb, struct buffer_head *bh)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003251{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003252 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003253 struct reiserfs_journal_cnode *cn = NULL;
3254 int count_already_incd = 0;
3255 int prepared = 0;
3256 BUG_ON(!th->t_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003257
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003258 PROC_INFO_INC(sb, journal.mark_dirty);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003259 if (th->t_trans_id != journal->j_trans_id) {
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04003260 reiserfs_panic(th->t_super, "journal-1577",
3261 "handle trans id %ld != current trans id %ld",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003262 th->t_trans_id, journal->j_trans_id);
3263 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003264
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003265 sb->s_dirt = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003266
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003267 prepared = test_clear_buffer_journal_prepared(bh);
3268 clear_buffer_journal_restore_dirty(bh);
3269 /* already in this transaction, we are done */
3270 if (buffer_journaled(bh)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003271 PROC_INFO_INC(sb, journal.mark_dirty_already);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003272 return 0;
3273 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003274
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003275 /* this must be turned into a panic instead of a warning. We can't allow
3276 ** a dirty or journal_dirty or locked buffer to be logged, as some changes
3277 ** could get to disk too early. NOT GOOD.
3278 */
3279 if (!prepared || buffer_dirty(bh)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003280 reiserfs_warning(sb, "journal-1777",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04003281 "buffer %llu bad state "
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003282 "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3283 (unsigned long long)bh->b_blocknr,
3284 prepared ? ' ' : '!',
3285 buffer_locked(bh) ? ' ' : '!',
3286 buffer_dirty(bh) ? ' ' : '!',
3287 buffer_journal_dirty(bh) ? ' ' : '!');
3288 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003289
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003290 if (atomic_read(&(journal->j_wcount)) <= 0) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003291 reiserfs_warning(sb, "journal-1409",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04003292 "returning because j_wcount was %d",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003293 atomic_read(&(journal->j_wcount)));
3294 return 1;
3295 }
Jeff Mahoney0222e652009-03-30 14:02:44 -04003296 /* this error means I've screwed up, and we've overflowed the transaction.
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003297 ** Nothing can be done here, except make the FS readonly or panic.
3298 */
3299 if (journal->j_len >= journal->j_trans_max) {
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04003300 reiserfs_panic(th->t_super, "journal-1413",
3301 "j_len (%lu) is too big",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003302 journal->j_len);
3303 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003304
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003305 if (buffer_journal_dirty(bh)) {
3306 count_already_incd = 1;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003307 PROC_INFO_INC(sb, journal.mark_dirty_notjournal);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003308 clear_buffer_journal_dirty(bh);
3309 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003310
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003311 if (journal->j_len > journal->j_len_alloc) {
3312 journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3313 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003314
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003315 set_buffer_journaled(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003316
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003317 /* now put this guy on the end */
3318 if (!cn) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003319 cn = get_cnode(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003320 if (!cn) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003321 reiserfs_panic(sb, "journal-4", "get_cnode failed!");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003322 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003323
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003324 if (th->t_blocks_logged == th->t_blocks_allocated) {
3325 th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3326 journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3327 }
3328 th->t_blocks_logged++;
3329 journal->j_len++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003330
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003331 cn->bh = bh;
3332 cn->blocknr = bh->b_blocknr;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003333 cn->sb = sb;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003334 cn->jlist = NULL;
3335 insert_journal_hash(journal->j_hash_table, cn);
3336 if (!count_already_incd) {
3337 get_bh(bh);
3338 }
3339 }
3340 cn->next = NULL;
3341 cn->prev = journal->j_last;
3342 cn->bh = bh;
3343 if (journal->j_last) {
3344 journal->j_last->next = cn;
3345 journal->j_last = cn;
3346 } else {
3347 journal->j_first = cn;
3348 journal->j_last = cn;
3349 }
3350 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003351}
3352
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003353int journal_end(struct reiserfs_transaction_handle *th,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003354 struct super_block *sb, unsigned long nblocks)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003355{
3356 if (!current->journal_info && th->t_refcount > 1)
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003357 reiserfs_warning(sb, "REISER-NESTING",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04003358 "th NULL, refcount %d", th->t_refcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003359
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003360 if (!th->t_trans_id) {
3361 WARN_ON(1);
3362 return -EIO;
3363 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003364
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003365 th->t_refcount--;
3366 if (th->t_refcount > 0) {
3367 struct reiserfs_transaction_handle *cur_th =
3368 current->journal_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003369
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003370 /* we aren't allowed to close a nested transaction on a different
3371 ** filesystem from the one in the task struct
3372 */
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003373 BUG_ON(cur_th->t_super != th->t_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003374
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003375 if (th != cur_th) {
3376 memcpy(current->journal_info, th, sizeof(*th));
3377 th->t_trans_id = 0;
3378 }
3379 return 0;
3380 } else {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003381 return do_journal_end(th, sb, nblocks, 0);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003382 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003383}
3384
Jeff Mahoney0222e652009-03-30 14:02:44 -04003385/* removes from the current transaction, relsing and descrementing any counters.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003386** also files the removed buffer directly onto the clean list
3387**
3388** called by journal_mark_freed when a block has been deleted
3389**
3390** returns 1 if it cleaned and relsed the buffer. 0 otherwise
3391*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003392static int remove_from_transaction(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003393 b_blocknr_t blocknr, int already_cleaned)
3394{
3395 struct buffer_head *bh;
3396 struct reiserfs_journal_cnode *cn;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003397 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003398 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003399
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003400 cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003401 if (!cn || !cn->bh) {
3402 return ret;
3403 }
3404 bh = cn->bh;
3405 if (cn->prev) {
3406 cn->prev->next = cn->next;
3407 }
3408 if (cn->next) {
3409 cn->next->prev = cn->prev;
3410 }
3411 if (cn == journal->j_first) {
3412 journal->j_first = cn->next;
3413 }
3414 if (cn == journal->j_last) {
3415 journal->j_last = cn->prev;
3416 }
3417 if (bh)
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003418 remove_journal_hash(sb, journal->j_hash_table, NULL,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003419 bh->b_blocknr, 0);
3420 clear_buffer_journaled(bh); /* don't log this one */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003421
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003422 if (!already_cleaned) {
3423 clear_buffer_journal_dirty(bh);
3424 clear_buffer_dirty(bh);
3425 clear_buffer_journal_test(bh);
3426 put_bh(bh);
3427 if (atomic_read(&(bh->b_count)) < 0) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003428 reiserfs_warning(sb, "journal-1752",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04003429 "b_count < 0");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003430 }
3431 ret = 1;
3432 }
3433 journal->j_len--;
3434 journal->j_len_alloc--;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003435 free_cnode(sb, cn);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003436 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003437}
3438
3439/*
3440** for any cnode in a journal list, it can only be dirtied of all the
Matt LaPlante0779bf22006-11-30 05:24:39 +01003441** transactions that include it are committed to disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003442** this checks through each transaction, and returns 1 if you are allowed to dirty,
3443** and 0 if you aren't
3444**
3445** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log
3446** blocks for a given transaction on disk
3447**
3448*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003449static int can_dirty(struct reiserfs_journal_cnode *cn)
3450{
3451 struct super_block *sb = cn->sb;
3452 b_blocknr_t blocknr = cn->blocknr;
3453 struct reiserfs_journal_cnode *cur = cn->hprev;
3454 int can_dirty = 1;
3455
3456 /* first test hprev. These are all newer than cn, so any node here
3457 ** with the same block number and dev means this node can't be sent
3458 ** to disk right now.
3459 */
3460 while (cur && can_dirty) {
3461 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3462 cur->blocknr == blocknr) {
3463 can_dirty = 0;
3464 }
3465 cur = cur->hprev;
3466 }
3467 /* then test hnext. These are all older than cn. As long as they
3468 ** are committed to the log, it is safe to write cn to disk
3469 */
3470 cur = cn->hnext;
3471 while (cur && can_dirty) {
3472 if (cur->jlist && cur->jlist->j_len > 0 &&
3473 atomic_read(&(cur->jlist->j_commit_left)) > 0 && cur->bh &&
3474 cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3475 can_dirty = 0;
3476 }
3477 cur = cur->hnext;
3478 }
3479 return can_dirty;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003480}
3481
3482/* syncs the commit blocks, but does not force the real buffers to disk
Jeff Mahoney0222e652009-03-30 14:02:44 -04003483** will wait until the current transaction is done/committed before returning
Linus Torvalds1da177e2005-04-16 15:20:36 -07003484*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003485int journal_end_sync(struct reiserfs_transaction_handle *th,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003486 struct super_block *sb, unsigned long nblocks)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003487{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003488 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003489
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003490 BUG_ON(!th->t_trans_id);
3491 /* you can sync while nested, very, very bad */
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003492 BUG_ON(th->t_refcount > 1);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003493 if (journal->j_len == 0) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003494 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003495 1);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003496 journal_mark_dirty(th, sb, SB_BUFFER_WITH_SB(sb));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003497 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003498 return do_journal_end(th, sb, nblocks, COMMIT_NOW | WAIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003499}
3500
3501/*
3502** writeback the pending async commits to disk
3503*/
David Howellsc4028952006-11-22 14:57:56 +00003504static void flush_async_commits(struct work_struct *work)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003505{
David Howellsc4028952006-11-22 14:57:56 +00003506 struct reiserfs_journal *journal =
3507 container_of(work, struct reiserfs_journal, j_work.work);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003508 struct super_block *sb = journal->j_work_sb;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003509 struct reiserfs_journal_list *jl;
3510 struct list_head *entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003511
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02003512 reiserfs_write_lock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003513 if (!list_empty(&journal->j_journal_list)) {
3514 /* last entry is the youngest, commit it and you get everything */
3515 entry = journal->j_journal_list.prev;
3516 jl = JOURNAL_LIST_ENTRY(entry);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003517 flush_commit_list(sb, jl, 1);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003518 }
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02003519 reiserfs_write_unlock(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003520}
3521
3522/*
3523** flushes any old transactions to disk
3524** ends the current transaction if it is too old
3525*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003526int reiserfs_flush_old_commits(struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003527{
3528 time_t now;
3529 struct reiserfs_transaction_handle th;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003530 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003531
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003532 now = get_seconds();
3533 /* safety check so we don't flush while we are replaying the log during
3534 * mount
3535 */
3536 if (list_empty(&journal->j_journal_list)) {
3537 return 0;
3538 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003539
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003540 /* check the current transaction. If there are no writers, and it is
3541 * too old, finish it, and force the commit blocks to disk
3542 */
3543 if (atomic_read(&journal->j_wcount) <= 0 &&
3544 journal->j_trans_start_time > 0 &&
3545 journal->j_len > 0 &&
3546 (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003547 if (!journal_join(&th, sb, 1)) {
3548 reiserfs_prepare_for_journal(sb,
3549 SB_BUFFER_WITH_SB(sb),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003550 1);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003551 journal_mark_dirty(&th, sb,
3552 SB_BUFFER_WITH_SB(sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003553
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003554 /* we're only being called from kreiserfsd, it makes no sense to do
3555 ** an async commit so that kreiserfsd can do it later
3556 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003557 do_journal_end(&th, sb, 1, COMMIT_NOW | WAIT);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003558 }
3559 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003560 return sb->s_dirt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003561}
3562
3563/*
3564** 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 -04003565**
3566** 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 -07003567** the writers are done. By the time it wakes up, the transaction it was called has already ended, so it just
3568** flushes the commit list and returns 0.
3569**
3570** 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 -04003571**
Linus Torvalds1da177e2005-04-16 15:20:36 -07003572** Note, we can't allow the journal_end to proceed while there are still writers in the log.
3573*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003574static int check_journal_end(struct reiserfs_transaction_handle *th,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003575 struct super_block *sb, unsigned long nblocks,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003576 int flags)
3577{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003578
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003579 time_t now;
3580 int flush = flags & FLUSH_ALL;
3581 int commit_now = flags & COMMIT_NOW;
3582 int wait_on_commit = flags & WAIT;
3583 struct reiserfs_journal_list *jl;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003584 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003585
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003586 BUG_ON(!th->t_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003587
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003588 if (th->t_trans_id != journal->j_trans_id) {
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04003589 reiserfs_panic(th->t_super, "journal-1577",
3590 "handle trans id %ld != current trans id %ld",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003591 th->t_trans_id, journal->j_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003592 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003593
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003594 journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3595 if (atomic_read(&(journal->j_wcount)) > 0) { /* <= 0 is allowed. unmounting might not call begin */
3596 atomic_dec(&(journal->j_wcount));
3597 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003598
Jeff Mahoney0222e652009-03-30 14:02:44 -04003599 /* 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 -07003600 ** will be dealt with by next transaction that actually writes something, but should be taken
3601 ** care of in this trans
3602 */
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003603 BUG_ON(journal->j_len == 0);
3604
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003605 /* if wcount > 0, and we are called to with flush or commit_now,
3606 ** we wait on j_join_wait. We will wake up when the last writer has
3607 ** finished the transaction, and started it on its way to the disk.
Jeff Mahoney0222e652009-03-30 14:02:44 -04003608 ** Then, we flush the commit or journal list, and just return 0
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003609 ** because the rest of journal end was already done for this transaction.
3610 */
3611 if (atomic_read(&(journal->j_wcount)) > 0) {
3612 if (flush || commit_now) {
3613 unsigned trans_id;
3614
3615 jl = journal->j_current_jl;
3616 trans_id = jl->j_trans_id;
3617 if (wait_on_commit)
3618 jl->j_state |= LIST_COMMIT_PENDING;
3619 atomic_set(&(journal->j_jlock), 1);
3620 if (flush) {
3621 journal->j_next_full_flush = 1;
3622 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003623 unlock_journal(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003624
3625 /* sleep while the current transaction is still j_jlocked */
3626 while (journal->j_trans_id == trans_id) {
3627 if (atomic_read(&journal->j_jlock)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003628 queue_log_writer(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003629 } else {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003630 lock_journal(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003631 if (journal->j_trans_id == trans_id) {
3632 atomic_set(&(journal->j_jlock),
3633 1);
3634 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003635 unlock_journal(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003636 }
3637 }
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003638 BUG_ON(journal->j_trans_id == trans_id);
3639
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003640 if (commit_now
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003641 && journal_list_still_alive(sb, trans_id)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003642 && wait_on_commit) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003643 flush_commit_list(sb, jl, 1);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003644 }
3645 return 0;
3646 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003647 unlock_journal(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003648 return 0;
3649 }
3650
3651 /* deal with old transactions where we are the last writers */
3652 now = get_seconds();
3653 if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3654 commit_now = 1;
3655 journal->j_next_async_flush = 1;
3656 }
3657 /* don't batch when someone is waiting on j_join_wait */
3658 /* don't batch when syncing the commit or flushing the whole trans */
3659 if (!(journal->j_must_wait > 0) && !(atomic_read(&(journal->j_jlock)))
3660 && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3661 && journal->j_len_alloc < journal->j_max_batch
3662 && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3663 journal->j_bcount++;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003664 unlock_journal(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003665 return 0;
3666 }
3667
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003668 if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(sb)) {
3669 reiserfs_panic(sb, "journal-003",
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04003670 "j_start (%ld) is too high",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003671 journal->j_start);
3672 }
3673 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003674}
3675
3676/*
3677** Does all the work that makes deleting blocks safe.
3678** 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 -04003679**
Linus Torvalds1da177e2005-04-16 15:20:36 -07003680** otherwise:
3681** set a bit for the block in the journal bitmap. That will prevent it from being allocated for unformatted nodes
3682** before this transaction has finished.
3683**
3684** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers. That will prevent any old transactions with
3685** this block from trying to flush to the real location. Since we aren't removing the cnode from the journal_list_hash,
3686** the block can't be reallocated yet.
3687**
3688** Then remove it from the current transaction, decrementing any counters and filing it on the clean list.
3689*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003690int journal_mark_freed(struct reiserfs_transaction_handle *th,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003691 struct super_block *sb, b_blocknr_t blocknr)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003692{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003693 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003694 struct reiserfs_journal_cnode *cn = NULL;
3695 struct buffer_head *bh = NULL;
3696 struct reiserfs_list_bitmap *jb = NULL;
3697 int cleaned = 0;
3698 BUG_ON(!th->t_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003699
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003700 cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003701 if (cn && cn->bh) {
3702 bh = cn->bh;
3703 get_bh(bh);
3704 }
3705 /* if it is journal new, we just remove it from this transaction */
3706 if (bh && buffer_journal_new(bh)) {
3707 clear_buffer_journal_new(bh);
3708 clear_prepared_bits(bh);
3709 reiserfs_clean_and_file_buffer(bh);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003710 cleaned = remove_from_transaction(sb, blocknr, cleaned);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003711 } else {
3712 /* set the bit for this block in the journal bitmap for this transaction */
3713 jb = journal->j_current_jl->j_list_bitmap;
3714 if (!jb) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003715 reiserfs_panic(sb, "journal-1702",
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04003716 "journal_list_bitmap is NULL");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003717 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003718 set_bit_in_list_bitmap(sb, blocknr, jb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003719
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003720 /* Note, the entire while loop is not allowed to schedule. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003721
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003722 if (bh) {
3723 clear_prepared_bits(bh);
3724 reiserfs_clean_and_file_buffer(bh);
3725 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003726 cleaned = remove_from_transaction(sb, blocknr, cleaned);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003727
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003728 /* find all older transactions with this block, make sure they don't try to write it out */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003729 cn = get_journal_hash_dev(sb, journal->j_list_hash_table,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003730 blocknr);
3731 while (cn) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003732 if (sb == cn->sb && blocknr == cn->blocknr) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003733 set_bit(BLOCK_FREED, &cn->state);
3734 if (cn->bh) {
3735 if (!cleaned) {
3736 /* remove_from_transaction will brelse the buffer if it was
3737 ** in the current trans
3738 */
3739 clear_buffer_journal_dirty(cn->
3740 bh);
3741 clear_buffer_dirty(cn->bh);
3742 clear_buffer_journal_test(cn->
3743 bh);
3744 cleaned = 1;
3745 put_bh(cn->bh);
3746 if (atomic_read
3747 (&(cn->bh->b_count)) < 0) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003748 reiserfs_warning(sb,
Jeff Mahoney45b03d52009-03-30 14:02:21 -04003749 "journal-2138",
3750 "cn->bh->b_count < 0");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003751 }
3752 }
3753 if (cn->jlist) { /* since we are clearing the bh, we MUST dec nonzerolen */
3754 atomic_dec(&
3755 (cn->jlist->
3756 j_nonzerolen));
3757 }
3758 cn->bh = NULL;
3759 }
3760 }
3761 cn = cn->hnext;
3762 }
3763 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003764
Chris Mason398c95b2007-10-16 23:29:44 -07003765 if (bh)
3766 release_buffer_page(bh); /* get_hash grabs the buffer */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003767 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003768}
3769
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003770void reiserfs_update_inode_transaction(struct inode *inode)
3771{
3772 struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3773 REISERFS_I(inode)->i_jl = journal->j_current_jl;
3774 REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003775}
3776
3777/*
3778 * returns -1 on error, 0 if no commits/barriers were done and 1
3779 * if a transaction was actually committed and the barrier was done
3780 */
3781static int __commit_trans_jl(struct inode *inode, unsigned long id,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003782 struct reiserfs_journal_list *jl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003783{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003784 struct reiserfs_transaction_handle th;
3785 struct super_block *sb = inode->i_sb;
3786 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3787 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003788
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003789 /* is it from the current transaction, or from an unknown transaction? */
3790 if (id == journal->j_trans_id) {
3791 jl = journal->j_current_jl;
3792 /* try to let other writers come in and grow this transaction */
3793 let_transaction_grow(sb, id);
3794 if (journal->j_trans_id != id) {
3795 goto flush_commit_only;
3796 }
3797
3798 ret = journal_begin(&th, sb, 1);
3799 if (ret)
3800 return ret;
3801
3802 /* someone might have ended this transaction while we joined */
3803 if (journal->j_trans_id != id) {
3804 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3805 1);
3806 journal_mark_dirty(&th, sb, SB_BUFFER_WITH_SB(sb));
3807 ret = journal_end(&th, sb, 1);
3808 goto flush_commit_only;
3809 }
3810
3811 ret = journal_end_sync(&th, sb, 1);
3812 if (!ret)
3813 ret = 1;
3814
3815 } else {
3816 /* this gets tricky, we have to make sure the journal list in
3817 * the inode still exists. We know the list is still around
3818 * if we've got a larger transaction id than the oldest list
3819 */
3820 flush_commit_only:
3821 if (journal_list_still_alive(inode->i_sb, id)) {
3822 /*
3823 * we only set ret to 1 when we know for sure
3824 * the barrier hasn't been started yet on the commit
3825 * block.
3826 */
3827 if (atomic_read(&jl->j_commit_left) > 1)
3828 ret = 1;
3829 flush_commit_list(sb, jl, 1);
3830 if (journal->j_errno)
3831 ret = journal->j_errno;
3832 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003833 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003834 /* otherwise the list is gone, and long since committed */
3835 return ret;
3836}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003837
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003838int reiserfs_commit_for_inode(struct inode *inode)
3839{
Jeff Mahoney600ed412009-03-30 14:02:17 -04003840 unsigned int id = REISERFS_I(inode)->i_trans_id;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003841 struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003842
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003843 /* for the whole inode, assume unset id means it was
3844 * changed in the current transaction. More conservative
Linus Torvalds1da177e2005-04-16 15:20:36 -07003845 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003846 if (!id || !jl) {
3847 reiserfs_update_inode_transaction(inode);
3848 id = REISERFS_I(inode)->i_trans_id;
3849 /* jl will be updated in __commit_trans_jl */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003850 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003851
3852 return __commit_trans_jl(inode, id, jl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003853}
3854
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003855void reiserfs_restore_prepared_buffer(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003856 struct buffer_head *bh)
3857{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003858 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3859 PROC_INFO_INC(sb, journal.restore_prepared);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003860 if (!bh) {
3861 return;
3862 }
3863 if (test_clear_buffer_journal_restore_dirty(bh) &&
3864 buffer_journal_dirty(bh)) {
3865 struct reiserfs_journal_cnode *cn;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003866 cn = get_journal_hash_dev(sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003867 journal->j_list_hash_table,
3868 bh->b_blocknr);
3869 if (cn && can_dirty(cn)) {
3870 set_buffer_journal_test(bh);
3871 mark_buffer_dirty(bh);
3872 }
3873 }
3874 clear_buffer_journal_prepared(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003875}
3876
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003877extern struct tree_balance *cur_tb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003878/*
3879** before we can change a metadata block, we have to make sure it won't
3880** be written to disk while we are altering it. So, we must:
3881** clean it
3882** wait on it.
Jeff Mahoney0222e652009-03-30 14:02:44 -04003883**
Linus Torvalds1da177e2005-04-16 15:20:36 -07003884*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003885int reiserfs_prepare_for_journal(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003886 struct buffer_head *bh, int wait)
3887{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003888 PROC_INFO_INC(sb, journal.prepare);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003889
Nick Pigginca5de402008-08-02 12:02:13 +02003890 if (!trylock_buffer(bh)) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003891 if (!wait)
3892 return 0;
3893 lock_buffer(bh);
3894 }
3895 set_buffer_journal_prepared(bh);
3896 if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3897 clear_buffer_journal_test(bh);
3898 set_buffer_journal_restore_dirty(bh);
3899 }
3900 unlock_buffer(bh);
3901 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003902}
3903
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003904static void flush_old_journal_lists(struct super_block *s)
3905{
3906 struct reiserfs_journal *journal = SB_JOURNAL(s);
3907 struct reiserfs_journal_list *jl;
3908 struct list_head *entry;
3909 time_t now = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003910
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003911 while (!list_empty(&journal->j_journal_list)) {
3912 entry = journal->j_journal_list.next;
3913 jl = JOURNAL_LIST_ENTRY(entry);
3914 /* this check should always be run, to send old lists to disk */
Chris Masona3172022006-09-29 01:59:56 -07003915 if (jl->j_timestamp < (now - (JOURNAL_MAX_TRANS_AGE * 4)) &&
3916 atomic_read(&jl->j_commit_left) == 0 &&
3917 test_transaction(s, jl)) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003918 flush_used_journal_lists(s, jl);
3919 } else {
3920 break;
3921 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003922 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003923}
3924
Jeff Mahoney0222e652009-03-30 14:02:44 -04003925/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003926** long and ugly. If flush, will not return until all commit
3927** blocks and all real buffers in the trans are on disk.
3928** If no_async, won't return until all commit blocks are on disk.
3929**
3930** keep reading, there are comments as you go along
3931**
3932** If the journal is aborted, we just clean up. Things like flushing
3933** journal lists, etc just won't happen.
3934*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003935static int do_journal_end(struct reiserfs_transaction_handle *th,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003936 struct super_block *sb, unsigned long nblocks,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003937 int flags)
3938{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003939 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003940 struct reiserfs_journal_cnode *cn, *next, *jl_cn;
3941 struct reiserfs_journal_cnode *last_cn = NULL;
3942 struct reiserfs_journal_desc *desc;
3943 struct reiserfs_journal_commit *commit;
3944 struct buffer_head *c_bh; /* commit bh */
3945 struct buffer_head *d_bh; /* desc bh */
3946 int cur_write_start = 0; /* start index of current log write */
3947 int old_start;
3948 int i;
Alexander Zarochentseva44c94a2006-03-25 03:06:59 -08003949 int flush;
3950 int wait_on_commit;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003951 struct reiserfs_journal_list *jl, *temp_jl;
3952 struct list_head *entry, *safe;
3953 unsigned long jindex;
Jeff Mahoney600ed412009-03-30 14:02:17 -04003954 unsigned int commit_trans_id;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003955 int trans_half;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003956
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003957 BUG_ON(th->t_refcount > 1);
3958 BUG_ON(!th->t_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003959
Alexander Zarochentseva44c94a2006-03-25 03:06:59 -08003960 /* protect flush_older_commits from doing mistakes if the
3961 transaction ID counter gets overflowed. */
Jeff Mahoney600ed412009-03-30 14:02:17 -04003962 if (th->t_trans_id == ~0U)
Alexander Zarochentseva44c94a2006-03-25 03:06:59 -08003963 flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
3964 flush = flags & FLUSH_ALL;
3965 wait_on_commit = flags & WAIT;
3966
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003967 put_fs_excl();
3968 current->journal_info = th->t_handle_save;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003969 reiserfs_check_lock_depth(sb, "journal end");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003970 if (journal->j_len == 0) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003971 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003972 1);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003973 journal_mark_dirty(th, sb, SB_BUFFER_WITH_SB(sb));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003974 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003975
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003976 lock_journal(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003977 if (journal->j_next_full_flush) {
3978 flags |= FLUSH_ALL;
3979 flush = 1;
3980 }
3981 if (journal->j_next_async_flush) {
3982 flags |= COMMIT_NOW | WAIT;
3983 wait_on_commit = 1;
3984 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003985
Jeff Mahoney0222e652009-03-30 14:02:44 -04003986 /* check_journal_end locks the journal, and unlocks if it does not return 1
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003987 ** it tells us if we should continue with the journal_end, or just return
3988 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003989 if (!check_journal_end(th, sb, nblocks, flags)) {
3990 sb->s_dirt = 1;
3991 wake_queued_writers(sb);
3992 reiserfs_async_progress_wait(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003993 goto out;
3994 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003995
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003996 /* check_journal_end might set these, check again */
3997 if (journal->j_next_full_flush) {
3998 flush = 1;
3999 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004000
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004001 /*
4002 ** j must wait means we have to flush the log blocks, and the real blocks for
4003 ** this transaction
4004 */
4005 if (journal->j_must_wait > 0) {
4006 flush = 1;
4007 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004008#ifdef REISERFS_PREALLOCATE
Jan Karaef43bc42006-01-11 12:17:40 -08004009 /* quota ops might need to nest, setup the journal_info pointer for them
4010 * and raise the refcount so that it is > 0. */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004011 current->journal_info = th;
Jan Karaef43bc42006-01-11 12:17:40 -08004012 th->t_refcount++;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004013 reiserfs_discard_all_prealloc(th); /* it should not involve new blocks into
4014 * the transaction */
Jan Karaef43bc42006-01-11 12:17:40 -08004015 th->t_refcount--;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004016 current->journal_info = th->t_handle_save;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004017#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004018
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004019 /* setup description block */
4020 d_bh =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004021 journal_getblk(sb,
4022 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004023 journal->j_start);
4024 set_buffer_uptodate(d_bh);
4025 desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
4026 memset(d_bh->b_data, 0, d_bh->b_size);
4027 memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
4028 set_desc_trans_id(desc, journal->j_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004029
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004030 /* 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 -04004031 c_bh = journal_getblk(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004032 ((journal->j_start + journal->j_len +
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004033 1) % SB_ONDISK_JOURNAL_SIZE(sb)));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004034 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
4035 memset(c_bh->b_data, 0, c_bh->b_size);
4036 set_commit_trans_id(commit, journal->j_trans_id);
4037 set_buffer_uptodate(c_bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004038
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004039 /* init this journal list */
4040 jl = journal->j_current_jl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004041
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004042 /* we lock the commit before doing anything because
4043 * we want to make sure nobody tries to run flush_commit_list until
4044 * the new transaction is fully setup, and we've already flushed the
4045 * ordered bh list
4046 */
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02004047 reiserfs_mutex_lock_safe(&jl->j_commit_mutex, sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004048
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004049 /* save the transaction id in case we need to commit it later */
4050 commit_trans_id = jl->j_trans_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004051
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004052 atomic_set(&jl->j_older_commits_done, 0);
4053 jl->j_trans_id = journal->j_trans_id;
4054 jl->j_timestamp = journal->j_trans_start_time;
4055 jl->j_commit_bh = c_bh;
4056 jl->j_start = journal->j_start;
4057 jl->j_len = journal->j_len;
4058 atomic_set(&jl->j_nonzerolen, journal->j_len);
4059 atomic_set(&jl->j_commit_left, journal->j_len + 2);
4060 jl->j_realblock = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004061
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004062 /* The ENTIRE FOR LOOP MUST not cause schedule to occur.
4063 ** for each real block, add it to the journal list hash,
4064 ** copy into real block index array in the commit or desc block
4065 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004066 trans_half = journal_trans_half(sb->s_blocksize);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004067 for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
4068 if (buffer_journaled(cn->bh)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004069 jl_cn = get_cnode(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004070 if (!jl_cn) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004071 reiserfs_panic(sb, "journal-1676",
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04004072 "get_cnode returned NULL");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004073 }
4074 if (i == 0) {
4075 jl->j_realblock = jl_cn;
4076 }
4077 jl_cn->prev = last_cn;
4078 jl_cn->next = NULL;
4079 if (last_cn) {
4080 last_cn->next = jl_cn;
4081 }
4082 last_cn = jl_cn;
Jeff Mahoney0222e652009-03-30 14:02:44 -04004083 /* make sure the block we are trying to log is not a block
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004084 of journal or reserved area */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004085
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004086 if (is_block_in_log_or_reserved_area
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004087 (sb, cn->bh->b_blocknr)) {
4088 reiserfs_panic(sb, "journal-2332",
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04004089 "Trying to log block %lu, "
4090 "which is a log block",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004091 cn->bh->b_blocknr);
4092 }
4093 jl_cn->blocknr = cn->bh->b_blocknr;
4094 jl_cn->state = 0;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004095 jl_cn->sb = sb;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004096 jl_cn->bh = cn->bh;
4097 jl_cn->jlist = jl;
4098 insert_journal_hash(journal->j_list_hash_table, jl_cn);
4099 if (i < trans_half) {
4100 desc->j_realblock[i] =
4101 cpu_to_le32(cn->bh->b_blocknr);
4102 } else {
4103 commit->j_realblock[i - trans_half] =
4104 cpu_to_le32(cn->bh->b_blocknr);
4105 }
4106 } else {
4107 i--;
4108 }
4109 }
4110 set_desc_trans_len(desc, journal->j_len);
4111 set_desc_mount_id(desc, journal->j_mount_id);
4112 set_desc_trans_id(desc, journal->j_trans_id);
4113 set_commit_trans_len(commit, journal->j_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004114
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004115 /* special check in case all buffers in the journal were marked for not logging */
Eric Sesterhenn14a61442006-10-03 23:36:38 +02004116 BUG_ON(journal->j_len == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004117
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004118 /* we're about to dirty all the log blocks, mark the description block
4119 * dirty now too. Don't mark the commit block dirty until all the
4120 * others are on disk
4121 */
4122 mark_buffer_dirty(d_bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004123
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004124 /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */
4125 cur_write_start = journal->j_start;
4126 cn = journal->j_first;
4127 jindex = 1; /* start at one so we don't get the desc again */
4128 while (cn) {
4129 clear_buffer_journal_new(cn->bh);
4130 /* copy all the real blocks into log area. dirty log blocks */
4131 if (buffer_journaled(cn->bh)) {
4132 struct buffer_head *tmp_bh;
4133 char *addr;
4134 struct page *page;
4135 tmp_bh =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004136 journal_getblk(sb,
4137 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004138 ((cur_write_start +
4139 jindex) %
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004140 SB_ONDISK_JOURNAL_SIZE(sb)));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004141 set_buffer_uptodate(tmp_bh);
4142 page = cn->bh->b_page;
4143 addr = kmap(page);
4144 memcpy(tmp_bh->b_data,
4145 addr + offset_in_page(cn->bh->b_data),
4146 cn->bh->b_size);
4147 kunmap(page);
4148 mark_buffer_dirty(tmp_bh);
4149 jindex++;
4150 set_buffer_journal_dirty(cn->bh);
4151 clear_buffer_journaled(cn->bh);
4152 } else {
4153 /* JDirty cleared sometime during transaction. don't log this one */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004154 reiserfs_warning(sb, "journal-2048",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04004155 "BAD, buffer in journal hash, "
4156 "but not JDirty!");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004157 brelse(cn->bh);
4158 }
4159 next = cn->next;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004160 free_cnode(sb, cn);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004161 cn = next;
Frederic Weisbeckere6950a42009-04-30 23:04:32 +02004162 reiserfs_write_unlock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004163 cond_resched();
Frederic Weisbeckere6950a42009-04-30 23:04:32 +02004164 reiserfs_write_lock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004165 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004166
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004167 /* we are done with both the c_bh and d_bh, but
4168 ** c_bh must be written after all other commit blocks,
4169 ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4170 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004171
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004172 journal->j_current_jl = alloc_journal_list(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004173
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004174 /* now it is safe to insert this transaction on the main list */
4175 list_add_tail(&jl->j_list, &journal->j_journal_list);
4176 list_add_tail(&jl->j_working_list, &journal->j_working_list);
4177 journal->j_num_work_lists++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004178
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004179 /* reset journal values for the next transaction */
4180 old_start = journal->j_start;
4181 journal->j_start =
4182 (journal->j_start + journal->j_len +
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004183 2) % SB_ONDISK_JOURNAL_SIZE(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004184 atomic_set(&(journal->j_wcount), 0);
4185 journal->j_bcount = 0;
4186 journal->j_last = NULL;
4187 journal->j_first = NULL;
4188 journal->j_len = 0;
4189 journal->j_trans_start_time = 0;
Alexander Zarochentseva44c94a2006-03-25 03:06:59 -08004190 /* check for trans_id overflow */
4191 if (++journal->j_trans_id == 0)
4192 journal->j_trans_id = 10;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004193 journal->j_current_jl->j_trans_id = journal->j_trans_id;
4194 journal->j_must_wait = 0;
4195 journal->j_len_alloc = 0;
4196 journal->j_next_full_flush = 0;
4197 journal->j_next_async_flush = 0;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004198 init_journal_hash(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004199
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004200 // make sure reiserfs_add_jh sees the new current_jl before we
4201 // write out the tails
4202 smp_mb();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004203
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004204 /* tail conversion targets have to hit the disk before we end the
4205 * transaction. Otherwise a later transaction might repack the tail
4206 * before this transaction commits, leaving the data block unflushed and
4207 * clean, if we crash before the later transaction commits, the data block
4208 * is lost.
4209 */
4210 if (!list_empty(&jl->j_tail_bh_list)) {
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02004211 reiserfs_write_unlock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004212 write_ordered_buffers(&journal->j_dirty_buffers_lock,
4213 journal, jl, &jl->j_tail_bh_list);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02004214 reiserfs_write_lock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004215 }
Eric Sesterhenn14a61442006-10-03 23:36:38 +02004216 BUG_ON(!list_empty(&jl->j_tail_bh_list));
Jeff Mahoney90415de2008-07-25 01:46:40 -07004217 mutex_unlock(&jl->j_commit_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004218
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004219 /* honor the flush wishes from the caller, simple commits can
4220 ** be done outside the journal lock, they are done below
4221 **
4222 ** if we don't flush the commit list right now, we put it into
4223 ** the work queue so the people waiting on the async progress work
4224 ** queue don't wait for this proc to flush journal lists and such.
4225 */
4226 if (flush) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004227 flush_commit_list(sb, jl, 1);
4228 flush_journal_list(sb, jl, 1);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004229 } else if (!(jl->j_state & LIST_COMMIT_PENDING))
4230 queue_delayed_work(commit_wq, &journal->j_work, HZ / 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004231
Jeff Mahoney0222e652009-03-30 14:02:44 -04004232 /* if the next transaction has any chance of wrapping, flush
4233 ** transactions that might get overwritten. If any journal lists are very
4234 ** old flush them as well.
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004235 */
4236 first_jl:
4237 list_for_each_safe(entry, safe, &journal->j_journal_list) {
4238 temp_jl = JOURNAL_LIST_ENTRY(entry);
4239 if (journal->j_start <= temp_jl->j_start) {
4240 if ((journal->j_start + journal->j_trans_max + 1) >=
4241 temp_jl->j_start) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004242 flush_used_journal_lists(sb, temp_jl);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004243 goto first_jl;
4244 } else if ((journal->j_start +
4245 journal->j_trans_max + 1) <
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004246 SB_ONDISK_JOURNAL_SIZE(sb)) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004247 /* if we don't cross into the next transaction and we don't
4248 * wrap, there is no way we can overlap any later transactions
4249 * break now
4250 */
4251 break;
4252 }
4253 } else if ((journal->j_start +
4254 journal->j_trans_max + 1) >
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004255 SB_ONDISK_JOURNAL_SIZE(sb)) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004256 if (((journal->j_start + journal->j_trans_max + 1) %
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004257 SB_ONDISK_JOURNAL_SIZE(sb)) >=
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004258 temp_jl->j_start) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004259 flush_used_journal_lists(sb, temp_jl);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004260 goto first_jl;
4261 } else {
4262 /* we don't overlap anything from out start to the end of the
4263 * log, and our wrapped portion doesn't overlap anything at
4264 * the start of the log. We can break
4265 */
4266 break;
4267 }
4268 }
4269 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004270 flush_old_journal_lists(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004271
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004272 journal->j_current_jl->j_list_bitmap =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004273 get_list_bitmap(sb, journal->j_current_jl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004274
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004275 if (!(journal->j_current_jl->j_list_bitmap)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004276 reiserfs_panic(sb, "journal-1996",
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04004277 "could not get a list bitmap");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004278 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004279
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004280 atomic_set(&(journal->j_jlock), 0);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004281 unlock_journal(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004282 /* wake up any body waiting to join. */
4283 clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4284 wake_up(&(journal->j_join_wait));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004285
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004286 if (!flush && wait_on_commit &&
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004287 journal_list_still_alive(sb, commit_trans_id)) {
4288 flush_commit_list(sb, jl, 1);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004289 }
4290 out:
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004291 reiserfs_check_lock_depth(sb, "journal end2");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004292
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004293 memset(th, 0, sizeof(*th));
4294 /* Re-set th->t_super, so we can properly keep track of how many
4295 * persistent transactions there are. We need to do this so if this
4296 * call is part of a failed restart_transaction, we can free it later */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004297 th->t_super = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004298
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004299 return journal->j_errno;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004300}
4301
Jeff Mahoney32e8b102009-03-30 14:02:26 -04004302/* Send the file system read only and refuse new transactions */
4303void reiserfs_abort_journal(struct super_block *sb, int errno)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004304{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004305 struct reiserfs_journal *journal = SB_JOURNAL(sb);
4306 if (test_bit(J_ABORTED, &journal->j_state))
4307 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004308
Jeff Mahoney32e8b102009-03-30 14:02:26 -04004309 if (!journal->j_errno)
4310 journal->j_errno = errno;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004311
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004312 sb->s_flags |= MS_RDONLY;
4313 set_bit(J_ABORTED, &journal->j_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004314
4315#ifdef CONFIG_REISERFS_CHECK
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004316 dump_stack();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004317#endif
4318}
4319