blob: a88b1b3e7db3e4f4f5c514062586dbf7a62fa21e [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Jeff Mahoney098297b2014-04-23 10:00:36 -04002 * Write ahead logging implementation copyright Chris Mason 2000
3 *
4 * The background commits make this code very interrelated, and
5 * overly complex. I need to rethink things a bit....The major players:
6 *
7 * journal_begin -- call with the number of blocks you expect to log.
8 * If the current transaction is too
9 * old, it will block until the current transaction is
10 * finished, and then start a new one.
11 * Usually, your transaction will get joined in with
12 * previous ones for speed.
13 *
14 * journal_join -- same as journal_begin, but won't block on the current
15 * transaction regardless of age. Don't ever call
16 * this. Ever. There are only two places it should be
17 * called from, and they are both inside this file.
18 *
19 * journal_mark_dirty -- adds blocks into this transaction. clears any flags
20 * that might make them get sent to disk
21 * and then marks them BH_JDirty. Puts the buffer head
22 * into the current transaction hash.
23 *
24 * journal_end -- if the current transaction is batchable, it does nothing
25 * otherwise, it could do an async/synchronous commit, or
26 * a full flush of all log and real blocks in the
27 * transaction.
28 *
29 * flush_old_commits -- if the current transaction is too old, it is ended and
30 * commit blocks are sent to disk. Forces commit blocks
31 * to disk for all backgrounded commits that have been
32 * around too long.
33 * -- Note, if you call this as an immediate flush from
34 * from within kupdate, it will ignore the immediate flag
35 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
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>
Al Virof466c6f2012-03-17 01:16:43 -040040#include "reiserfs.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
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
Linus Torvalds1da177e2005-04-16 15:20:36 -070055/* gets a struct reiserfs_journal_list * from a list head */
56#define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
57 j_list))
58#define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
59 j_working_list))
60
Jeff Mahoney098297b2014-04-23 10:00:36 -040061/* must be correct to keep the desc and commit structs at 4k */
62#define JOURNAL_TRANS_HALF 1018
Linus Torvaldsbd4c6252005-07-12 20:21:28 -070063#define BUFNR 64 /*read ahead */
Linus Torvalds1da177e2005-04-16 15:20:36 -070064
65/* cnode stat bits. Move these into reiserfs_fs.h */
66
Jeff Mahoney098297b2014-04-23 10:00:36 -040067/* this block was freed, and can't be written. */
68#define BLOCK_FREED 2
69/* this block was freed during this transaction, and can't be written */
70#define BLOCK_FREED_HOLDER 3
Linus Torvalds1da177e2005-04-16 15:20:36 -070071
Jeff Mahoney098297b2014-04-23 10:00:36 -040072/* used in flush_journal_list */
73#define BLOCK_NEEDS_FLUSH 4
Linus Torvalds1da177e2005-04-16 15:20:36 -070074#define BLOCK_DIRTIED 5
75
Linus Torvalds1da177e2005-04-16 15:20:36 -070076/* journal list state bits */
77#define LIST_TOUCHED 1
78#define LIST_DIRTY 2
Linus Torvaldsbd4c6252005-07-12 20:21:28 -070079#define LIST_COMMIT_PENDING 4 /* someone will commit this list */
Linus Torvalds1da177e2005-04-16 15:20:36 -070080
81/* flags for do_journal_end */
82#define FLUSH_ALL 1 /* flush commit and real blocks */
83#define COMMIT_NOW 2 /* end and commit this transaction */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -070084#define WAIT 4 /* wait for the log blocks to hit the disk */
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
Jeff Mahoney58d85422014-04-23 10:00:38 -040086static int do_journal_end(struct reiserfs_transaction_handle *, int flags);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -070087static int flush_journal_list(struct super_block *s,
88 struct reiserfs_journal_list *jl, int flushall);
89static int flush_commit_list(struct super_block *s,
90 struct reiserfs_journal_list *jl, int flushall);
91static int can_dirty(struct reiserfs_journal_cnode *cn);
92static int journal_join(struct reiserfs_transaction_handle *th,
Jeff Mahoneyb491dd12014-04-23 10:00:40 -040093 struct super_block *sb);
Al Viro4385bab2013-05-05 22:11:03 -040094static void release_journal_dev(struct super_block *super,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -070095 struct reiserfs_journal *journal);
Linus Torvalds1da177e2005-04-16 15:20:36 -070096static int dirty_one_transaction(struct super_block *s,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -070097 struct reiserfs_journal_list *jl);
David Howellsc4028952006-11-22 14:57:56 +000098static void flush_async_commits(struct work_struct *work);
Linus Torvalds1da177e2005-04-16 15:20:36 -070099static void queue_log_writer(struct super_block *s);
100
101/* values for join in do_journal_begin_r */
102enum {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700103 JBEGIN_REG = 0, /* regular journal begin */
Jeff Mahoney098297b2014-04-23 10:00:36 -0400104 /* join the running transaction if at all possible */
105 JBEGIN_JOIN = 1,
106 /* called from cleanup code, ignores aborted flag */
107 JBEGIN_ABORT = 2,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108};
109
110static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400111 struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700112 unsigned long nblocks, int join);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400114static void init_journal_hash(struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700115{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400116 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700117 memset(journal->j_hash_table, 0,
118 JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119}
120
121/*
Jeff Mahoney098297b2014-04-23 10:00:36 -0400122 * clears BH_Dirty and sticks the buffer on the clean list. Called because
123 * I can't allow refile_buffer to make schedule happen after I've freed a
124 * block. Look at remove_from_transaction and journal_mark_freed for
125 * more details.
126 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700127static int reiserfs_clean_and_file_buffer(struct buffer_head *bh)
128{
129 if (bh) {
130 clear_buffer_dirty(bh);
131 clear_buffer_journal_test(bh);
132 }
133 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134}
135
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700136static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400137 *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700138{
139 struct reiserfs_bitmap_node *bn;
140 static int id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141
Pekka Enbergd739b422006-02-01 03:06:43 -0800142 bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700143 if (!bn) {
144 return NULL;
145 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400146 bn->data = kzalloc(sb->s_blocksize, GFP_NOFS);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700147 if (!bn->data) {
Pekka Enbergd739b422006-02-01 03:06:43 -0800148 kfree(bn);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700149 return NULL;
150 }
151 bn->id = id++;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700152 INIT_LIST_HEAD(&bn->list);
153 return bn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154}
155
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400156static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700157{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400158 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700159 struct reiserfs_bitmap_node *bn = NULL;
160 struct list_head *entry = journal->j_bitmap_nodes.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700162 journal->j_used_bitmap_nodes++;
Jeff Mahoneycf776a72014-04-23 10:00:41 -0400163repeat:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700165 if (entry != &journal->j_bitmap_nodes) {
166 bn = list_entry(entry, struct reiserfs_bitmap_node, list);
167 list_del(entry);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400168 memset(bn->data, 0, sb->s_blocksize);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700169 journal->j_free_bitmap_nodes--;
170 return bn;
171 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400172 bn = allocate_bitmap_node(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700173 if (!bn) {
174 yield();
175 goto repeat;
176 }
177 return bn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178}
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400179static inline void free_bitmap_node(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700180 struct reiserfs_bitmap_node *bn)
181{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400182 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700183 journal->j_used_bitmap_nodes--;
184 if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
Pekka Enbergd739b422006-02-01 03:06:43 -0800185 kfree(bn->data);
186 kfree(bn);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700187 } else {
188 list_add(&bn->list, &journal->j_bitmap_nodes);
189 journal->j_free_bitmap_nodes++;
190 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191}
192
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400193static void allocate_bitmap_nodes(struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700194{
195 int i;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400196 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700197 struct reiserfs_bitmap_node *bn = NULL;
198 for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400199 bn = allocate_bitmap_node(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700200 if (bn) {
201 list_add(&bn->list, &journal->j_bitmap_nodes);
202 journal->j_free_bitmap_nodes++;
203 } else {
Jeff Mahoney098297b2014-04-23 10:00:36 -0400204 /* this is ok, we'll try again when more are needed */
205 break;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700206 }
207 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208}
209
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400210static int set_bit_in_list_bitmap(struct super_block *sb,
Jeff Mahoney3ee16672007-10-18 23:39:25 -0700211 b_blocknr_t block,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700212 struct reiserfs_list_bitmap *jb)
213{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400214 unsigned int bmap_nr = block / (sb->s_blocksize << 3);
215 unsigned int bit_nr = block % (sb->s_blocksize << 3);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700217 if (!jb->bitmaps[bmap_nr]) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400218 jb->bitmaps[bmap_nr] = get_bitmap_node(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700219 }
220 set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
221 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222}
223
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400224static void cleanup_bitmap_list(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700225 struct reiserfs_list_bitmap *jb)
226{
227 int i;
228 if (jb->bitmaps == NULL)
229 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400231 for (i = 0; i < reiserfs_bmap_count(sb); i++) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700232 if (jb->bitmaps[i]) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400233 free_bitmap_node(sb, jb->bitmaps[i]);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700234 jb->bitmaps[i] = NULL;
235 }
236 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237}
238
239/*
Jeff Mahoney098297b2014-04-23 10:00:36 -0400240 * only call this on FS unmount.
241 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400242static int free_list_bitmaps(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700243 struct reiserfs_list_bitmap *jb_array)
244{
245 int i;
246 struct reiserfs_list_bitmap *jb;
247 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
248 jb = jb_array + i;
249 jb->journal_list = NULL;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400250 cleanup_bitmap_list(sb, jb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700251 vfree(jb->bitmaps);
252 jb->bitmaps = NULL;
253 }
254 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255}
256
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400257static int free_bitmap_nodes(struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700258{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400259 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700260 struct list_head *next = journal->j_bitmap_nodes.next;
261 struct reiserfs_bitmap_node *bn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700263 while (next != &journal->j_bitmap_nodes) {
264 bn = list_entry(next, struct reiserfs_bitmap_node, list);
265 list_del(next);
Pekka Enbergd739b422006-02-01 03:06:43 -0800266 kfree(bn->data);
267 kfree(bn);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700268 next = journal->j_bitmap_nodes.next;
269 journal->j_free_bitmap_nodes--;
270 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700272 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273}
274
275/*
Jeff Mahoney098297b2014-04-23 10:00:36 -0400276 * get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
277 * jb_array is the array to be filled in.
278 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400279int reiserfs_allocate_list_bitmaps(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700280 struct reiserfs_list_bitmap *jb_array,
Jeff Mahoney3ee16672007-10-18 23:39:25 -0700281 unsigned int bmap_nr)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700282{
283 int i;
284 int failed = 0;
285 struct reiserfs_list_bitmap *jb;
286 int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700288 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
289 jb = jb_array + i;
290 jb->journal_list = NULL;
Joe Perches558feb02011-05-28 10:36:33 -0700291 jb->bitmaps = vzalloc(mem);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700292 if (!jb->bitmaps) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400293 reiserfs_warning(sb, "clm-2000", "unable to "
Jeff Mahoney45b03d52009-03-30 14:02:21 -0400294 "allocate bitmaps for journal lists");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700295 failed = 1;
296 break;
297 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700298 }
299 if (failed) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400300 free_list_bitmaps(sb, jb_array);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700301 return -1;
302 }
303 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304}
305
306/*
Jeff Mahoney098297b2014-04-23 10:00:36 -0400307 * find an available list bitmap. If you can't find one, flush a commit list
308 * and try again
309 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400310static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700311 struct reiserfs_journal_list
312 *jl)
313{
314 int i, j;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400315 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700316 struct reiserfs_list_bitmap *jb = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700318 for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) {
319 i = journal->j_list_bitmap_index;
320 journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS;
321 jb = journal->j_list_bitmap + i;
322 if (journal->j_list_bitmap[i].journal_list) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400323 flush_commit_list(sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700324 journal->j_list_bitmap[i].
325 journal_list, 1);
326 if (!journal->j_list_bitmap[i].journal_list) {
327 break;
328 }
329 } else {
330 break;
331 }
332 }
Jeff Mahoney098297b2014-04-23 10:00:36 -0400333 /* double check to make sure if flushed correctly */
334 if (jb->journal_list)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700335 return NULL;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700336 jb->journal_list = jl;
337 return jb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338}
339
Jeff Mahoney0222e652009-03-30 14:02:44 -0400340/*
Jeff Mahoney098297b2014-04-23 10:00:36 -0400341 * allocates a new chunk of X nodes, and links them all together as a list.
342 * Uses the cnode->next and cnode->prev pointers
343 * returns NULL on failure
344 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700345static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
346{
347 struct reiserfs_journal_cnode *head;
348 int i;
349 if (num_cnodes <= 0) {
350 return NULL;
351 }
Joe Perches558feb02011-05-28 10:36:33 -0700352 head = vzalloc(num_cnodes * sizeof(struct reiserfs_journal_cnode));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700353 if (!head) {
354 return NULL;
355 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700356 head[0].prev = NULL;
357 head[0].next = head + 1;
358 for (i = 1; i < num_cnodes; i++) {
359 head[i].prev = head + (i - 1);
360 head[i].next = head + (i + 1); /* if last one, overwrite it after the if */
361 }
362 head[num_cnodes - 1].next = NULL;
363 return head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364}
365
Jeff Mahoney098297b2014-04-23 10:00:36 -0400366/* pulls a cnode off the free list, or returns NULL on failure */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400367static struct reiserfs_journal_cnode *get_cnode(struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700368{
369 struct reiserfs_journal_cnode *cn;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400370 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400372 reiserfs_check_lock_depth(sb, "get_cnode");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700374 if (journal->j_cnode_free <= 0) {
375 return NULL;
376 }
377 journal->j_cnode_used++;
378 journal->j_cnode_free--;
379 cn = journal->j_cnode_free_list;
380 if (!cn) {
381 return cn;
382 }
383 if (cn->next) {
384 cn->next->prev = NULL;
385 }
386 journal->j_cnode_free_list = cn->next;
387 memset(cn, 0, sizeof(struct reiserfs_journal_cnode));
388 return cn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389}
390
391/*
Jeff Mahoney098297b2014-04-23 10:00:36 -0400392 * returns a cnode to the free list
393 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400394static void free_cnode(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700395 struct reiserfs_journal_cnode *cn)
396{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400397 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400399 reiserfs_check_lock_depth(sb, "free_cnode");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700401 journal->j_cnode_used--;
402 journal->j_cnode_free++;
403 /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
404 cn->next = journal->j_cnode_free_list;
405 if (journal->j_cnode_free_list) {
406 journal->j_cnode_free_list->prev = cn;
407 }
408 cn->prev = NULL; /* not needed with the memset, but I might kill the memset, and forget to do this */
409 journal->j_cnode_free_list = cn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410}
411
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700412static void clear_prepared_bits(struct buffer_head *bh)
413{
414 clear_buffer_journal_prepared(bh);
415 clear_buffer_journal_restore_dirty(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416}
417
Jeff Mahoney098297b2014-04-23 10:00:36 -0400418/*
419 * return a cnode with same dev, block number and size in table,
420 * or null if not found
421 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700422static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
423 super_block
424 *sb,
425 struct
426 reiserfs_journal_cnode
427 **table,
428 long bl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700430 struct reiserfs_journal_cnode *cn;
431 cn = journal_hash(table, sb, bl);
432 while (cn) {
433 if (cn->blocknr == bl && cn->sb == sb)
434 return cn;
435 cn = cn->hnext;
436 }
437 return (struct reiserfs_journal_cnode *)0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438}
439
440/*
Jeff Mahoney098297b2014-04-23 10:00:36 -0400441 * this actually means 'can this block be reallocated yet?'. If you set
442 * search_all, a block can only be allocated if it is not in the current
443 * transaction, was not freed by the current transaction, and has no chance
444 * of ever being overwritten by a replay after crashing.
445 *
446 * If you don't set search_all, a block can only be allocated if it is not
447 * in the current transaction. Since deleting a block removes it from the
448 * current transaction, this case should never happen. If you don't set
449 * search_all, make sure you never write the block without logging it.
450 *
451 * next_zero_bit is a suggestion about the next block to try for find_forward.
452 * when bl is rejected because it is set in a journal list bitmap, we search
453 * for the next zero bit in the bitmap that rejected bl. Then, we return
454 * that through next_zero_bit for find_forward to try.
455 *
456 * Just because we return something in next_zero_bit does not mean we won't
457 * reject it on the next call to reiserfs_in_journal
458 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400459int reiserfs_in_journal(struct super_block *sb,
Jeff Mahoney3ee16672007-10-18 23:39:25 -0700460 unsigned int bmap_nr, int bit_nr, int search_all,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700461 b_blocknr_t * next_zero_bit)
462{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400463 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700464 struct reiserfs_journal_cnode *cn;
465 struct reiserfs_list_bitmap *jb;
466 int i;
467 unsigned long bl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700469 *next_zero_bit = 0; /* always start this at zero. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400471 PROC_INFO_INC(sb, journal.in_journal);
Jeff Mahoney098297b2014-04-23 10:00:36 -0400472 /*
473 * If we aren't doing a search_all, this is a metablock, and it
474 * will be logged before use. if we crash before the transaction
475 * that freed it commits, this transaction won't have committed
476 * either, and the block will never be written
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700477 */
478 if (search_all) {
479 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400480 PROC_INFO_INC(sb, journal.in_journal_bitmap);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700481 jb = journal->j_list_bitmap + i;
482 if (jb->journal_list && jb->bitmaps[bmap_nr] &&
483 test_bit(bit_nr,
484 (unsigned long *)jb->bitmaps[bmap_nr]->
485 data)) {
486 *next_zero_bit =
487 find_next_zero_bit((unsigned long *)
488 (jb->bitmaps[bmap_nr]->
489 data),
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400490 sb->s_blocksize << 3,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700491 bit_nr + 1);
492 return 1;
493 }
494 }
495 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400497 bl = bmap_nr * (sb->s_blocksize << 3) + bit_nr;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700498 /* is it in any old transactions? */
499 if (search_all
500 && (cn =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400501 get_journal_hash_dev(sb, journal->j_list_hash_table, bl))) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700502 return 1;
503 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700505 /* is it in the current transaction. This should never happen */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400506 if ((cn = get_journal_hash_dev(sb, journal->j_hash_table, bl))) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700507 BUG();
508 return 1;
509 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400511 PROC_INFO_INC(sb, journal.in_journal_reusable);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700512 /* safe for reuse */
513 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514}
515
Jeff Mahoney098297b2014-04-23 10:00:36 -0400516/* insert cn into table */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700517static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
518 struct reiserfs_journal_cnode *cn)
519{
520 struct reiserfs_journal_cnode *cn_orig;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700522 cn_orig = journal_hash(table, cn->sb, cn->blocknr);
523 cn->hnext = cn_orig;
524 cn->hprev = NULL;
525 if (cn_orig) {
526 cn_orig->hprev = cn;
527 }
528 journal_hash(table, cn->sb, cn->blocknr) = cn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529}
530
531/* lock the current transaction */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400532static inline void lock_journal(struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700533{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400534 PROC_INFO_INC(sb, journal.lock_journal);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +0200535
536 reiserfs_mutex_lock_safe(&SB_JOURNAL(sb)->j_mutex, sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537}
538
539/* unlock the current transaction */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400540static inline void unlock_journal(struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700541{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400542 mutex_unlock(&SB_JOURNAL(sb)->j_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543}
544
545static inline void get_journal_list(struct reiserfs_journal_list *jl)
546{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700547 jl->j_refcount++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548}
549
550static inline void put_journal_list(struct super_block *s,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700551 struct reiserfs_journal_list *jl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700553 if (jl->j_refcount < 1) {
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -0400554 reiserfs_panic(s, "journal-2", "trans id %u, refcount at %d",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700555 jl->j_trans_id, jl->j_refcount);
556 }
557 if (--jl->j_refcount == 0)
Pekka Enbergd739b422006-02-01 03:06:43 -0800558 kfree(jl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559}
560
561/*
Jeff Mahoney098297b2014-04-23 10:00:36 -0400562 * this used to be much more involved, and I'm keeping it just in case
563 * things get ugly again. it gets called by flush_commit_list, and
564 * cleans up any data stored about blocks freed during a transaction.
565 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400566static void cleanup_freed_for_journal_list(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700567 struct reiserfs_journal_list *jl)
568{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700570 struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
571 if (jb) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400572 cleanup_bitmap_list(sb, jb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700573 }
574 jl->j_list_bitmap->journal_list = NULL;
575 jl->j_list_bitmap = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576}
577
578static int journal_list_still_alive(struct super_block *s,
Jeff Mahoney600ed412009-03-30 14:02:17 -0400579 unsigned int trans_id)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700581 struct reiserfs_journal *journal = SB_JOURNAL(s);
582 struct list_head *entry = &journal->j_journal_list;
583 struct reiserfs_journal_list *jl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700585 if (!list_empty(entry)) {
586 jl = JOURNAL_LIST_ENTRY(entry->next);
587 if (jl->j_trans_id <= trans_id) {
588 return 1;
589 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700591 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592}
593
Chris Mason398c95b2007-10-16 23:29:44 -0700594/*
595 * If page->mapping was null, we failed to truncate this page for
596 * some reason. Most likely because it was truncated after being
597 * logged via data=journal.
598 *
599 * This does a check to see if the buffer belongs to one of these
600 * lost pages before doing the final put_bh. If page->mapping was
601 * null, it tries to free buffers on the page, which should make the
602 * final page_cache_release drop the page from the lru.
603 */
604static void release_buffer_page(struct buffer_head *bh)
605{
606 struct page *page = bh->b_page;
Nick Piggin529ae9a2008-08-02 12:01:03 +0200607 if (!page->mapping && trylock_page(page)) {
Chris Mason398c95b2007-10-16 23:29:44 -0700608 page_cache_get(page);
609 put_bh(bh);
610 if (!page->mapping)
611 try_to_free_buffers(page);
612 unlock_page(page);
613 page_cache_release(page);
614 } else {
615 put_bh(bh);
616 }
617}
618
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700619static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
620{
621 char b[BDEVNAME_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700623 if (buffer_journaled(bh)) {
Jeff Mahoney45b03d52009-03-30 14:02:21 -0400624 reiserfs_warning(NULL, "clm-2084",
625 "pinned buffer %lu:%s sent to disk",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700626 bh->b_blocknr, bdevname(bh->b_bdev, b));
627 }
628 if (uptodate)
629 set_buffer_uptodate(bh);
630 else
631 clear_buffer_uptodate(bh);
Chris Mason398c95b2007-10-16 23:29:44 -0700632
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700633 unlock_buffer(bh);
Chris Mason398c95b2007-10-16 23:29:44 -0700634 release_buffer_page(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635}
636
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700637static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
638{
639 if (uptodate)
640 set_buffer_uptodate(bh);
641 else
642 clear_buffer_uptodate(bh);
643 unlock_buffer(bh);
644 put_bh(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645}
646
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700647static void submit_logged_buffer(struct buffer_head *bh)
648{
649 get_bh(bh);
650 bh->b_end_io = reiserfs_end_buffer_io_sync;
651 clear_buffer_journal_new(bh);
652 clear_buffer_dirty(bh);
653 if (!test_clear_buffer_journal_test(bh))
654 BUG();
655 if (!buffer_uptodate(bh))
656 BUG();
657 submit_bh(WRITE, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658}
659
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700660static void submit_ordered_buffer(struct buffer_head *bh)
661{
662 get_bh(bh);
663 bh->b_end_io = reiserfs_end_ordered_io;
664 clear_buffer_dirty(bh);
665 if (!buffer_uptodate(bh))
666 BUG();
667 submit_bh(WRITE, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668}
669
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670#define CHUNK_SIZE 32
671struct buffer_chunk {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700672 struct buffer_head *bh[CHUNK_SIZE];
673 int nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674};
675
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700676static void write_chunk(struct buffer_chunk *chunk)
677{
678 int i;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700679 for (i = 0; i < chunk->nr; i++) {
680 submit_logged_buffer(chunk->bh[i]);
681 }
682 chunk->nr = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683}
684
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700685static void write_ordered_chunk(struct buffer_chunk *chunk)
686{
687 int i;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700688 for (i = 0; i < chunk->nr; i++) {
689 submit_ordered_buffer(chunk->bh[i]);
690 }
691 chunk->nr = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692}
693
694static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700695 spinlock_t * lock, void (fn) (struct buffer_chunk *))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700697 int ret = 0;
Eric Sesterhenn14a61442006-10-03 23:36:38 +0200698 BUG_ON(chunk->nr >= CHUNK_SIZE);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700699 chunk->bh[chunk->nr++] = bh;
700 if (chunk->nr >= CHUNK_SIZE) {
701 ret = 1;
702 if (lock)
703 spin_unlock(lock);
704 fn(chunk);
705 if (lock)
706 spin_lock(lock);
707 }
708 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709}
710
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700712static struct reiserfs_jh *alloc_jh(void)
713{
714 struct reiserfs_jh *jh;
715 while (1) {
716 jh = kmalloc(sizeof(*jh), GFP_NOFS);
717 if (jh) {
718 atomic_inc(&nr_reiserfs_jh);
719 return jh;
720 }
721 yield();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723}
724
725/*
726 * we want to free the jh when the buffer has been written
727 * and waited on
728 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700729void reiserfs_free_jh(struct buffer_head *bh)
730{
731 struct reiserfs_jh *jh;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700733 jh = bh->b_private;
734 if (jh) {
735 bh->b_private = NULL;
736 jh->bh = NULL;
737 list_del_init(&jh->list);
738 kfree(jh);
739 if (atomic_read(&nr_reiserfs_jh) <= 0)
740 BUG();
741 atomic_dec(&nr_reiserfs_jh);
742 put_bh(bh);
743 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744}
745
746static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700747 int tail)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700749 struct reiserfs_jh *jh;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700751 if (bh->b_private) {
752 spin_lock(&j->j_dirty_buffers_lock);
753 if (!bh->b_private) {
754 spin_unlock(&j->j_dirty_buffers_lock);
755 goto no_jh;
756 }
757 jh = bh->b_private;
758 list_del_init(&jh->list);
759 } else {
Jeff Mahoneycf776a72014-04-23 10:00:41 -0400760no_jh:
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700761 get_bh(bh);
762 jh = alloc_jh();
763 spin_lock(&j->j_dirty_buffers_lock);
Jeff Mahoney098297b2014-04-23 10:00:36 -0400764 /*
765 * buffer must be locked for __add_jh, should be able to have
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700766 * two adds at the same time
767 */
Eric Sesterhenn14a61442006-10-03 23:36:38 +0200768 BUG_ON(bh->b_private);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700769 jh->bh = bh;
770 bh->b_private = jh;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700772 jh->jl = j->j_current_jl;
773 if (tail)
774 list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
775 else {
776 list_add_tail(&jh->list, &jh->jl->j_bh_list);
777 }
778 spin_unlock(&j->j_dirty_buffers_lock);
779 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780}
781
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700782int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
783{
784 return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785}
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700786int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
787{
788 return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789}
790
791#define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700792static int write_ordered_buffers(spinlock_t * lock,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 struct reiserfs_journal *j,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700794 struct reiserfs_journal_list *jl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 struct list_head *list)
796{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700797 struct buffer_head *bh;
798 struct reiserfs_jh *jh;
799 int ret = j->j_errno;
800 struct buffer_chunk chunk;
801 struct list_head tmp;
802 INIT_LIST_HEAD(&tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700804 chunk.nr = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 spin_lock(lock);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700806 while (!list_empty(list)) {
807 jh = JH_ENTRY(list->next);
808 bh = jh->bh;
809 get_bh(bh);
Nick Pigginca5de402008-08-02 12:02:13 +0200810 if (!trylock_buffer(bh)) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700811 if (!buffer_dirty(bh)) {
Akinobu Mitaf1166292006-06-26 00:24:46 -0700812 list_move(&jh->list, &tmp);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700813 goto loop_next;
814 }
815 spin_unlock(lock);
816 if (chunk.nr)
817 write_ordered_chunk(&chunk);
818 wait_on_buffer(bh);
819 cond_resched();
820 spin_lock(lock);
821 goto loop_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 }
Jeff Mahoney098297b2014-04-23 10:00:36 -0400823 /*
824 * in theory, dirty non-uptodate buffers should never get here,
Chris Mason3d4492f2006-02-01 03:06:49 -0800825 * but the upper layer io error paths still have a few quirks.
826 * Handle them here as gracefully as we can
827 */
828 if (!buffer_uptodate(bh) && buffer_dirty(bh)) {
829 clear_buffer_dirty(bh);
830 ret = -EIO;
831 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700832 if (buffer_dirty(bh)) {
Akinobu Mitaf1166292006-06-26 00:24:46 -0700833 list_move(&jh->list, &tmp);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700834 add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
835 } else {
836 reiserfs_free_jh(bh);
837 unlock_buffer(bh);
838 }
Jeff Mahoneycf776a72014-04-23 10:00:41 -0400839loop_next:
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700840 put_bh(bh);
841 cond_resched_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700843 if (chunk.nr) {
844 spin_unlock(lock);
845 write_ordered_chunk(&chunk);
846 spin_lock(lock);
847 }
848 while (!list_empty(&tmp)) {
849 jh = JH_ENTRY(tmp.prev);
850 bh = jh->bh;
851 get_bh(bh);
852 reiserfs_free_jh(bh);
853
854 if (buffer_locked(bh)) {
855 spin_unlock(lock);
856 wait_on_buffer(bh);
857 spin_lock(lock);
858 }
859 if (!buffer_uptodate(bh)) {
860 ret = -EIO;
861 }
Jeff Mahoney098297b2014-04-23 10:00:36 -0400862 /*
863 * ugly interaction with invalidatepage here.
864 * reiserfs_invalidate_page will pin any buffer that has a
865 * valid journal head from an older transaction. If someone
866 * else sets our buffer dirty after we write it in the first
867 * loop, and then someone truncates the page away, nobody
868 * will ever write the buffer. We're safe if we write the
869 * page one last time after freeing the journal header.
Chris Masond62b1b82006-02-01 03:06:47 -0800870 */
871 if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
872 spin_unlock(lock);
873 ll_rw_block(WRITE, 1, &bh);
874 spin_lock(lock);
875 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700876 put_bh(bh);
877 cond_resched_lock(lock);
878 }
879 spin_unlock(lock);
880 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881}
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700882
883static int flush_older_commits(struct super_block *s,
884 struct reiserfs_journal_list *jl)
885{
886 struct reiserfs_journal *journal = SB_JOURNAL(s);
887 struct reiserfs_journal_list *other_jl;
888 struct reiserfs_journal_list *first_jl;
889 struct list_head *entry;
Jeff Mahoney600ed412009-03-30 14:02:17 -0400890 unsigned int trans_id = jl->j_trans_id;
891 unsigned int other_trans_id;
892 unsigned int first_trans_id;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700893
Jeff Mahoneycf776a72014-04-23 10:00:41 -0400894find_first:
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700895 /*
896 * first we walk backwards to find the oldest uncommitted transation
897 */
898 first_jl = jl;
899 entry = jl->j_list.prev;
900 while (1) {
901 other_jl = JOURNAL_LIST_ENTRY(entry);
902 if (entry == &journal->j_journal_list ||
903 atomic_read(&other_jl->j_older_commits_done))
904 break;
905
906 first_jl = other_jl;
907 entry = other_jl->j_list.prev;
908 }
909
910 /* if we didn't find any older uncommitted transactions, return now */
911 if (first_jl == jl) {
912 return 0;
913 }
914
915 first_trans_id = first_jl->j_trans_id;
916
917 entry = &first_jl->j_list;
918 while (1) {
919 other_jl = JOURNAL_LIST_ENTRY(entry);
920 other_trans_id = other_jl->j_trans_id;
921
922 if (other_trans_id < trans_id) {
923 if (atomic_read(&other_jl->j_commit_left) != 0) {
924 flush_commit_list(s, other_jl, 0);
925
926 /* list we were called with is gone, return */
927 if (!journal_list_still_alive(s, trans_id))
928 return 1;
929
Jeff Mahoney098297b2014-04-23 10:00:36 -0400930 /*
931 * the one we just flushed is gone, this means
932 * all older lists are also gone, so first_jl
933 * is no longer valid either. Go back to the
934 * beginning.
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700935 */
936 if (!journal_list_still_alive
937 (s, other_trans_id)) {
938 goto find_first;
939 }
940 }
941 entry = entry->next;
942 if (entry == &journal->j_journal_list)
943 return 0;
944 } else {
945 return 0;
946 }
947 }
948 return 0;
949}
Adrian Bunkdeba0f42007-10-16 23:26:03 -0700950
951static int reiserfs_async_progress_wait(struct super_block *s)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700952{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700953 struct reiserfs_journal *j = SB_JOURNAL(s);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +0200954
955 if (atomic_read(&j->j_async_throttle)) {
Jeff Mahoney278f6672013-08-08 17:34:46 -0400956 int depth;
957
958 depth = reiserfs_write_unlock_nested(s);
Jens Axboe8aa7e842009-07-09 14:52:32 +0200959 congestion_wait(BLK_RW_ASYNC, HZ / 10);
Jeff Mahoney278f6672013-08-08 17:34:46 -0400960 reiserfs_write_lock_nested(s, depth);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +0200961 }
962
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700963 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964}
965
966/*
Jeff Mahoney098297b2014-04-23 10:00:36 -0400967 * if this journal list still has commit blocks unflushed, send them to disk.
968 *
969 * log areas must be flushed in order (transaction 2 can't commit before
970 * transaction 1) Before the commit block can by written, every other log
971 * block must be safely on disk
972 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700973static int flush_commit_list(struct super_block *s,
974 struct reiserfs_journal_list *jl, int flushall)
975{
976 int i;
Jeff Mahoney3ee16672007-10-18 23:39:25 -0700977 b_blocknr_t bn;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700978 struct buffer_head *tbh = NULL;
Jeff Mahoney600ed412009-03-30 14:02:17 -0400979 unsigned int trans_id = jl->j_trans_id;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700980 struct reiserfs_journal *journal = SB_JOURNAL(s);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700981 int retval = 0;
Chris Masone0e851c2006-02-01 03:06:49 -0800982 int write_len;
Jeff Mahoney278f6672013-08-08 17:34:46 -0400983 int depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700985 reiserfs_check_lock_depth(s, "flush_commit_list");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700987 if (atomic_read(&jl->j_older_commits_done)) {
988 return 0;
989 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990
Jeff Mahoney098297b2014-04-23 10:00:36 -0400991 /*
992 * before we can put our commit blocks on disk, we have to make
993 * sure everyone older than us is on disk too
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700994 */
995 BUG_ON(jl->j_len <= 0);
996 BUG_ON(trans_id == journal->j_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700998 get_journal_list(jl);
999 if (flushall) {
1000 if (flush_older_commits(s, jl) == 1) {
Jeff Mahoney098297b2014-04-23 10:00:36 -04001001 /*
1002 * list disappeared during flush_older_commits.
1003 * return
1004 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001005 goto put_jl;
1006 }
1007 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001009 /* make sure nobody is trying to flush this one at the same time */
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001010 reiserfs_mutex_lock_safe(&jl->j_commit_mutex, s);
1011
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001012 if (!journal_list_still_alive(s, trans_id)) {
Jeff Mahoney90415de2008-07-25 01:46:40 -07001013 mutex_unlock(&jl->j_commit_mutex);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001014 goto put_jl;
1015 }
1016 BUG_ON(jl->j_trans_id == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001018 /* this commit is done, exit */
Jeff Mahoneya228bf82014-04-23 10:00:42 -04001019 if (atomic_read(&jl->j_commit_left) <= 0) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001020 if (flushall) {
Jeff Mahoneya228bf82014-04-23 10:00:42 -04001021 atomic_set(&jl->j_older_commits_done, 1);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001022 }
Jeff Mahoney90415de2008-07-25 01:46:40 -07001023 mutex_unlock(&jl->j_commit_mutex);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001024 goto put_jl;
1025 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001027 if (!list_empty(&jl->j_bh_list)) {
Chris Mason3d4492f2006-02-01 03:06:49 -08001028 int ret;
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001029
1030 /*
1031 * We might sleep in numerous places inside
1032 * write_ordered_buffers. Relax the write lock.
1033 */
Jeff Mahoney278f6672013-08-08 17:34:46 -04001034 depth = reiserfs_write_unlock_nested(s);
Chris Mason3d4492f2006-02-01 03:06:49 -08001035 ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
1036 journal, jl, &jl->j_bh_list);
1037 if (ret < 0 && retval == 0)
1038 retval = ret;
Jeff Mahoney278f6672013-08-08 17:34:46 -04001039 reiserfs_write_lock_nested(s, depth);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001040 }
1041 BUG_ON(!list_empty(&jl->j_bh_list));
1042 /*
1043 * for the description block and all the log blocks, submit any buffers
Chris Masone0e851c2006-02-01 03:06:49 -08001044 * that haven't already reached the disk. Try to write at least 256
1045 * log blocks. later on, we will only wait on blocks that correspond
1046 * to this transaction, but while we're unplugging we might as well
1047 * get a chunk of data on there.
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001048 */
1049 atomic_inc(&journal->j_async_throttle);
Chris Masone0e851c2006-02-01 03:06:49 -08001050 write_len = jl->j_len + 1;
1051 if (write_len < 256)
1052 write_len = 256;
1053 for (i = 0 ; i < write_len ; i++) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001054 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1055 SB_ONDISK_JOURNAL_SIZE(s);
1056 tbh = journal_find_get_block(s, bn);
Chris Masone0e851c2006-02-01 03:06:49 -08001057 if (tbh) {
Frederic Weisbecker6e3647a2009-05-01 02:27:39 +02001058 if (buffer_dirty(tbh)) {
Jeff Mahoney278f6672013-08-08 17:34:46 -04001059 depth = reiserfs_write_unlock_nested(s);
Frederic Weisbecker6e3647a2009-05-01 02:27:39 +02001060 ll_rw_block(WRITE, 1, &tbh);
Jeff Mahoney278f6672013-08-08 17:34:46 -04001061 reiserfs_write_lock_nested(s, depth);
Frederic Weisbecker6e3647a2009-05-01 02:27:39 +02001062 }
Chris Masone0e851c2006-02-01 03:06:49 -08001063 put_bh(tbh) ;
1064 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001065 }
1066 atomic_dec(&journal->j_async_throttle);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001068 for (i = 0; i < (jl->j_len + 1); i++) {
1069 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1070 (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1071 tbh = journal_find_get_block(s, bn);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001072
Jeff Mahoney278f6672013-08-08 17:34:46 -04001073 depth = reiserfs_write_unlock_nested(s);
1074 __wait_on_buffer(tbh);
1075 reiserfs_write_lock_nested(s, depth);
Jeff Mahoney098297b2014-04-23 10:00:36 -04001076 /*
1077 * since we're using ll_rw_blk above, it might have skipped
1078 * over a locked buffer. Double check here
1079 */
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001080 /* redundant, sync_dirty_buffer() checks */
1081 if (buffer_dirty(tbh)) {
Jeff Mahoney278f6672013-08-08 17:34:46 -04001082 depth = reiserfs_write_unlock_nested(s);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001083 sync_dirty_buffer(tbh);
Jeff Mahoney278f6672013-08-08 17:34:46 -04001084 reiserfs_write_lock_nested(s, depth);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001085 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001086 if (unlikely(!buffer_uptodate(tbh))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087#ifdef CONFIG_REISERFS_CHECK
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001088 reiserfs_warning(s, "journal-601",
1089 "buffer write failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090#endif
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001091 retval = -EIO;
1092 }
Jeff Mahoney098297b2014-04-23 10:00:36 -04001093 /* once for journal_find_get_block */
1094 put_bh(tbh);
1095 /* once due to original getblk in do_journal_end */
1096 put_bh(tbh);
Jeff Mahoneya228bf82014-04-23 10:00:42 -04001097 atomic_dec(&jl->j_commit_left);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001098 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099
Jeff Mahoneya228bf82014-04-23 10:00:42 -04001100 BUG_ON(atomic_read(&jl->j_commit_left) != 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101
Jeff Mahoney098297b2014-04-23 10:00:36 -04001102 /*
1103 * If there was a write error in the journal - we can't commit
Christoph Hellwig7cd33ad2010-08-18 05:29:14 -04001104 * this transaction - it will be invalid and, if successful,
1105 * will just end up propagating the write error out to
Jeff Mahoney098297b2014-04-23 10:00:36 -04001106 * the file system.
1107 */
Christoph Hellwig7cd33ad2010-08-18 05:29:14 -04001108 if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1109 if (buffer_dirty(jl->j_commit_bh))
1110 BUG();
1111 mark_buffer_dirty(jl->j_commit_bh) ;
Jeff Mahoney278f6672013-08-08 17:34:46 -04001112 depth = reiserfs_write_unlock_nested(s);
Christoph Hellwig7cd33ad2010-08-18 05:29:14 -04001113 if (reiserfs_barrier_flush(s))
1114 __sync_dirty_buffer(jl->j_commit_bh, WRITE_FLUSH_FUA);
1115 else
1116 sync_dirty_buffer(jl->j_commit_bh);
Jeff Mahoney278f6672013-08-08 17:34:46 -04001117 reiserfs_write_lock_nested(s, depth);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001118 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119
Jeff Mahoney098297b2014-04-23 10:00:36 -04001120 /*
1121 * If there was a write error in the journal - we can't commit this
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001122 * transaction - it will be invalid and, if successful, will just end
Jeff Mahoney098297b2014-04-23 10:00:36 -04001123 * up propagating the write error out to the filesystem.
1124 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001125 if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126#ifdef CONFIG_REISERFS_CHECK
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001127 reiserfs_warning(s, "journal-615", "buffer write failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128#endif
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001129 retval = -EIO;
1130 }
1131 bforget(jl->j_commit_bh);
1132 if (journal->j_last_commit_id != 0 &&
1133 (jl->j_trans_id - journal->j_last_commit_id) != 1) {
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001134 reiserfs_warning(s, "clm-2200", "last commit %lu, current %lu",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001135 journal->j_last_commit_id, jl->j_trans_id);
1136 }
1137 journal->j_last_commit_id = jl->j_trans_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138
Jeff Mahoney098297b2014-04-23 10:00:36 -04001139 /*
1140 * now, every commit block is on the disk. It is safe to allow
1141 * blocks freed during this transaction to be reallocated
1142 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001143 cleanup_freed_for_journal_list(s, jl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001145 retval = retval ? retval : journal->j_errno;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001147 /* mark the metadata dirty */
1148 if (!retval)
1149 dirty_one_transaction(s, jl);
Jeff Mahoneya228bf82014-04-23 10:00:42 -04001150 atomic_dec(&jl->j_commit_left);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001152 if (flushall) {
Jeff Mahoneya228bf82014-04-23 10:00:42 -04001153 atomic_set(&jl->j_older_commits_done, 1);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001154 }
Jeff Mahoney90415de2008-07-25 01:46:40 -07001155 mutex_unlock(&jl->j_commit_mutex);
Jeff Mahoneycf776a72014-04-23 10:00:41 -04001156put_jl:
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001157 put_journal_list(s, jl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001159 if (retval)
1160 reiserfs_abort(s, retval, "Journal write error in %s",
Harvey Harrisonfbe54982008-04-28 02:16:22 -07001161 __func__);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001162 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163}
1164
1165/*
Jeff Mahoney098297b2014-04-23 10:00:36 -04001166 * flush_journal_list frequently needs to find a newer transaction for a
1167 * given block. This does that, or returns NULL if it can't find anything
1168 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001169static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1170 reiserfs_journal_cnode
1171 *cn)
1172{
1173 struct super_block *sb = cn->sb;
1174 b_blocknr_t blocknr = cn->blocknr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001176 cn = cn->hprev;
1177 while (cn) {
1178 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1179 return cn->jlist;
1180 }
1181 cn = cn->hprev;
1182 }
1183 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184}
1185
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001186static void remove_journal_hash(struct super_block *,
1187 struct reiserfs_journal_cnode **,
1188 struct reiserfs_journal_list *, unsigned long,
1189 int);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190
1191/*
Jeff Mahoney098297b2014-04-23 10:00:36 -04001192 * once all the real blocks have been flushed, it is safe to remove them
1193 * from the journal list for this transaction. Aside from freeing the
1194 * cnode, this also allows the block to be reallocated for data blocks
1195 * if it had been deleted.
1196 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001197static void remove_all_from_journal_list(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001198 struct reiserfs_journal_list *jl,
1199 int debug)
1200{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001201 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001202 struct reiserfs_journal_cnode *cn, *last;
1203 cn = jl->j_realblock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204
Jeff Mahoney098297b2014-04-23 10:00:36 -04001205 /*
1206 * which is better, to lock once around the whole loop, or
1207 * to lock for each call to remove_journal_hash?
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001208 */
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/*
Jeff Mahoney098297b2014-04-23 10:00:36 -04001229 * if this timestamp is greater than the timestamp we wrote last to the
1230 * header block, write it to the header block. once this is done, I can
1231 * safely say the log area for this transaction won't ever be replayed,
1232 * and I can start releasing blocks in this transaction for reuse as data
1233 * blocks. called by flush_journal_list, before it calls
1234 * remove_all_from_journal_list
1235 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001236static int _update_journal_header_block(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001237 unsigned long offset,
Jeff Mahoney600ed412009-03-30 14:02:17 -04001238 unsigned int trans_id)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001239{
1240 struct reiserfs_journal_header *jh;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001241 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Jeff Mahoney278f6672013-08-08 17:34:46 -04001242 int depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001244 if (reiserfs_is_journal_aborted(journal))
1245 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001247 if (trans_id >= journal->j_last_flush_trans_id) {
1248 if (buffer_locked((journal->j_header_bh))) {
Jeff Mahoney278f6672013-08-08 17:34:46 -04001249 depth = reiserfs_write_unlock_nested(sb);
1250 __wait_on_buffer(journal->j_header_bh);
1251 reiserfs_write_lock_nested(sb, depth);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001252 if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253#ifdef CONFIG_REISERFS_CHECK
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001254 reiserfs_warning(sb, "journal-699",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001255 "buffer write failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256#endif
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001257 return -EIO;
1258 }
1259 }
1260 journal->j_last_flush_trans_id = trans_id;
1261 journal->j_first_unflushed_offset = offset;
1262 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1263 b_data);
1264 jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1265 jh->j_first_unflushed_offset = cpu_to_le32(offset);
1266 jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267
Christoph Hellwig7cd33ad2010-08-18 05:29:14 -04001268 set_buffer_dirty(journal->j_header_bh);
Jeff Mahoney278f6672013-08-08 17:34:46 -04001269 depth = reiserfs_write_unlock_nested(sb);
Christoph Hellwig7cd33ad2010-08-18 05:29:14 -04001270
1271 if (reiserfs_barrier_flush(sb))
1272 __sync_dirty_buffer(journal->j_header_bh, WRITE_FLUSH_FUA);
1273 else
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001274 sync_dirty_buffer(journal->j_header_bh);
Christoph Hellwig7cd33ad2010-08-18 05:29:14 -04001275
Jeff Mahoney278f6672013-08-08 17:34:46 -04001276 reiserfs_write_lock_nested(sb, depth);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001277 if (!buffer_uptodate(journal->j_header_bh)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001278 reiserfs_warning(sb, "journal-837",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001279 "IO error during journal replay");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001280 return -EIO;
1281 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001283 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284}
1285
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001286static int update_journal_header_block(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001287 unsigned long offset,
Jeff Mahoney600ed412009-03-30 14:02:17 -04001288 unsigned int trans_id)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001289{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001290 return _update_journal_header_block(sb, offset, trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291}
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001292
Jeff Mahoney0222e652009-03-30 14:02:44 -04001293/*
1294** flush any and all journal lists older than you are
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295** can only be called from flush_journal_list
1296*/
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001297static int flush_older_journal_lists(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001298 struct reiserfs_journal_list *jl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001300 struct list_head *entry;
1301 struct reiserfs_journal_list *other_jl;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001302 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Jeff Mahoney600ed412009-03-30 14:02:17 -04001303 unsigned int trans_id = jl->j_trans_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304
Jeff Mahoney098297b2014-04-23 10:00:36 -04001305 /*
1306 * we know we are the only ones flushing things, no extra race
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001307 * protection is required.
1308 */
Jeff Mahoneycf776a72014-04-23 10:00:41 -04001309restart:
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001310 entry = journal->j_journal_list.next;
1311 /* Did we wrap? */
1312 if (entry == &journal->j_journal_list)
1313 return 0;
1314 other_jl = JOURNAL_LIST_ENTRY(entry);
1315 if (other_jl->j_trans_id < trans_id) {
1316 BUG_ON(other_jl->j_refcount <= 0);
1317 /* do not flush all */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001318 flush_journal_list(sb, other_jl, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001320 /* other_jl is now deleted from the list */
1321 goto restart;
1322 }
1323 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324}
1325
1326static void del_from_work_list(struct super_block *s,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001327 struct reiserfs_journal_list *jl)
1328{
1329 struct reiserfs_journal *journal = SB_JOURNAL(s);
1330 if (!list_empty(&jl->j_working_list)) {
1331 list_del_init(&jl->j_working_list);
1332 journal->j_num_work_lists--;
1333 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334}
1335
Jeff Mahoney098297b2014-04-23 10:00:36 -04001336/*
1337 * flush a journal list, both commit and real blocks
1338 *
1339 * always set flushall to 1, unless you are calling from inside
1340 * flush_journal_list
1341 *
1342 * IMPORTANT. This can only be called while there are no journal writers,
1343 * and the journal is locked. That means it can only be called from
1344 * do_journal_end, or by journal_release
1345 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001346static int flush_journal_list(struct super_block *s,
1347 struct reiserfs_journal_list *jl, int flushall)
1348{
1349 struct reiserfs_journal_list *pjl;
1350 struct reiserfs_journal_cnode *cn, *last;
1351 int count;
1352 int was_jwait = 0;
1353 int was_dirty = 0;
1354 struct buffer_head *saved_bh;
1355 unsigned long j_len_saved = jl->j_len;
1356 struct reiserfs_journal *journal = SB_JOURNAL(s);
1357 int err = 0;
Jeff Mahoney278f6672013-08-08 17:34:46 -04001358 int depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001360 BUG_ON(j_len_saved <= 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001362 if (atomic_read(&journal->j_wcount) != 0) {
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001363 reiserfs_warning(s, "clm-2048", "called with wcount %d",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001364 atomic_read(&journal->j_wcount));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001366
1367 /* if flushall == 0, the lock is already held */
1368 if (flushall) {
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001369 reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
Jeff Mahoneyafe70252008-07-25 01:46:39 -07001370 } else if (mutex_trylock(&journal->j_flush_mutex)) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001371 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001373
1374 count = 0;
1375 if (j_len_saved > journal->j_trans_max) {
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04001376 reiserfs_panic(s, "journal-715", "length is %lu, trans id %lu",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001377 j_len_saved, jl->j_trans_id);
1378 return 0;
1379 }
1380
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001381 /* if all the work is already done, get out of here */
Jeff Mahoneya228bf82014-04-23 10:00:42 -04001382 if (atomic_read(&jl->j_nonzerolen) <= 0 &&
1383 atomic_read(&jl->j_commit_left) <= 0) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001384 goto flush_older_and_return;
1385 }
1386
Jeff Mahoney098297b2014-04-23 10:00:36 -04001387 /*
1388 * start by putting the commit list on disk. This will also flush
1389 * the commit lists of any olders transactions
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001390 */
1391 flush_commit_list(s, jl, 1);
1392
1393 if (!(jl->j_state & LIST_DIRTY)
1394 && !reiserfs_is_journal_aborted(journal))
1395 BUG();
1396
1397 /* are we done now? */
Jeff Mahoneya228bf82014-04-23 10:00:42 -04001398 if (atomic_read(&jl->j_nonzerolen) <= 0 &&
1399 atomic_read(&jl->j_commit_left) <= 0) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001400 goto flush_older_and_return;
1401 }
1402
Jeff Mahoney098297b2014-04-23 10:00:36 -04001403 /*
1404 * loop through each cnode, see if we need to write it,
1405 * or wait on a more recent transaction, or just ignore it
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001406 */
Jeff Mahoneya228bf82014-04-23 10:00:42 -04001407 if (atomic_read(&journal->j_wcount) != 0) {
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04001408 reiserfs_panic(s, "journal-844", "journal list is flushing, "
1409 "wcount is not 0");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001410 }
1411 cn = jl->j_realblock;
1412 while (cn) {
1413 was_jwait = 0;
1414 was_dirty = 0;
1415 saved_bh = NULL;
1416 /* blocknr of 0 is no longer in the hash, ignore it */
1417 if (cn->blocknr == 0) {
1418 goto free_cnode;
1419 }
1420
Jeff Mahoney098297b2014-04-23 10:00:36 -04001421 /*
1422 * This transaction failed commit.
1423 * Don't write out to the disk
1424 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001425 if (!(jl->j_state & LIST_DIRTY))
1426 goto free_cnode;
1427
1428 pjl = find_newer_jl_for_cn(cn);
Jeff Mahoney098297b2014-04-23 10:00:36 -04001429 /*
1430 * the order is important here. We check pjl to make sure we
1431 * don't clear BH_JDirty_wait if we aren't the one writing this
1432 * block to disk
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001433 */
1434 if (!pjl && cn->bh) {
1435 saved_bh = cn->bh;
1436
Jeff Mahoney098297b2014-04-23 10:00:36 -04001437 /*
1438 * we do this to make sure nobody releases the
1439 * buffer while we are working with it
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001440 */
1441 get_bh(saved_bh);
1442
1443 if (buffer_journal_dirty(saved_bh)) {
1444 BUG_ON(!can_dirty(cn));
1445 was_jwait = 1;
1446 was_dirty = 1;
1447 } else if (can_dirty(cn)) {
Jeff Mahoney098297b2014-04-23 10:00:36 -04001448 /*
1449 * everything with !pjl && jwait
1450 * should be writable
1451 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001452 BUG();
1453 }
1454 }
1455
Jeff Mahoney098297b2014-04-23 10:00:36 -04001456 /*
1457 * if someone has this block in a newer transaction, just make
1458 * sure they are committed, and don't try writing it to disk
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001459 */
1460 if (pjl) {
1461 if (atomic_read(&pjl->j_commit_left))
1462 flush_commit_list(s, pjl, 1);
1463 goto free_cnode;
1464 }
1465
Jeff Mahoney098297b2014-04-23 10:00:36 -04001466 /*
1467 * bh == NULL when the block got to disk on its own, OR,
1468 * the block got freed in a future transaction
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001469 */
1470 if (saved_bh == NULL) {
1471 goto free_cnode;
1472 }
1473
Jeff Mahoney098297b2014-04-23 10:00:36 -04001474 /*
1475 * this should never happen. kupdate_one_transaction has
1476 * this list locked while it works, so we should never see a
1477 * buffer here that is not marked JDirty_wait
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001478 */
1479 if ((!was_jwait) && !buffer_locked(saved_bh)) {
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001480 reiserfs_warning(s, "journal-813",
1481 "BAD! buffer %llu %cdirty %cjwait, "
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001482 "not in a newer tranasction",
1483 (unsigned long long)saved_bh->
1484 b_blocknr, was_dirty ? ' ' : '!',
1485 was_jwait ? ' ' : '!');
1486 }
1487 if (was_dirty) {
Jeff Mahoney098297b2014-04-23 10:00:36 -04001488 /*
1489 * we inc again because saved_bh gets decremented
1490 * at free_cnode
1491 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001492 get_bh(saved_bh);
1493 set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1494 lock_buffer(saved_bh);
1495 BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1496 if (buffer_dirty(saved_bh))
1497 submit_logged_buffer(saved_bh);
1498 else
1499 unlock_buffer(saved_bh);
1500 count++;
1501 } else {
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001502 reiserfs_warning(s, "clm-2082",
1503 "Unable to flush buffer %llu in %s",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001504 (unsigned long long)saved_bh->
Harvey Harrisonfbe54982008-04-28 02:16:22 -07001505 b_blocknr, __func__);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001506 }
Jeff Mahoneycf776a72014-04-23 10:00:41 -04001507free_cnode:
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001508 last = cn;
1509 cn = cn->next;
1510 if (saved_bh) {
Jeff Mahoney098297b2014-04-23 10:00:36 -04001511 /*
1512 * we incremented this to keep others from
1513 * taking the buffer head away
1514 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001515 put_bh(saved_bh);
Jeff Mahoneya228bf82014-04-23 10:00:42 -04001516 if (atomic_read(&saved_bh->b_count) < 0) {
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001517 reiserfs_warning(s, "journal-945",
1518 "saved_bh->b_count < 0");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001519 }
1520 }
1521 }
1522 if (count > 0) {
1523 cn = jl->j_realblock;
1524 while (cn) {
1525 if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1526 if (!cn->bh) {
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04001527 reiserfs_panic(s, "journal-1011",
1528 "cn->bh is NULL");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001529 }
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001530
Jeff Mahoney278f6672013-08-08 17:34:46 -04001531 depth = reiserfs_write_unlock_nested(s);
1532 __wait_on_buffer(cn->bh);
1533 reiserfs_write_lock_nested(s, depth);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001534
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001535 if (!cn->bh) {
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04001536 reiserfs_panic(s, "journal-1012",
1537 "cn->bh is NULL");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001538 }
1539 if (unlikely(!buffer_uptodate(cn->bh))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540#ifdef CONFIG_REISERFS_CHECK
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001541 reiserfs_warning(s, "journal-949",
1542 "buffer write failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543#endif
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001544 err = -EIO;
1545 }
Jeff Mahoney098297b2014-04-23 10:00:36 -04001546 /*
1547 * note, we must clear the JDirty_wait bit
1548 * after the up to date check, otherwise we
1549 * race against our flushpage routine
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001550 */
1551 BUG_ON(!test_clear_buffer_journal_dirty
1552 (cn->bh));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553
Chris Mason398c95b2007-10-16 23:29:44 -07001554 /* drop one ref for us */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001555 put_bh(cn->bh);
Chris Mason398c95b2007-10-16 23:29:44 -07001556 /* drop one ref for journal_mark_dirty */
1557 release_buffer_page(cn->bh);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001558 }
1559 cn = cn->next;
1560 }
1561 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001563 if (err)
1564 reiserfs_abort(s, -EIO,
1565 "Write error while pushing transaction to disk in %s",
Harvey Harrisonfbe54982008-04-28 02:16:22 -07001566 __func__);
Jeff Mahoneycf776a72014-04-23 10:00:41 -04001567flush_older_and_return:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568
Jeff Mahoney098297b2014-04-23 10:00:36 -04001569 /*
1570 * before we can update the journal header block, we _must_ flush all
1571 * real blocks from all older transactions to disk. This is because
1572 * once the header block is updated, this transaction will not be
1573 * replayed after a crash
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001574 */
1575 if (flushall) {
1576 flush_older_journal_lists(s, jl);
1577 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001579 err = journal->j_errno;
Jeff Mahoney098297b2014-04-23 10:00:36 -04001580 /*
1581 * before we can remove everything from the hash tables for this
1582 * transaction, we must make sure it can never be replayed
1583 *
1584 * since we are only called from do_journal_end, we know for sure there
1585 * are no allocations going on while we are flushing journal lists. So,
1586 * we only need to update the journal header block for the last list
1587 * being flushed
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001588 */
1589 if (!err && flushall) {
1590 err =
1591 update_journal_header_block(s,
1592 (jl->j_start + jl->j_len +
1593 2) % SB_ONDISK_JOURNAL_SIZE(s),
1594 jl->j_trans_id);
1595 if (err)
1596 reiserfs_abort(s, -EIO,
1597 "Write error while updating journal header in %s",
Harvey Harrisonfbe54982008-04-28 02:16:22 -07001598 __func__);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001599 }
1600 remove_all_from_journal_list(s, jl, 0);
1601 list_del_init(&jl->j_list);
1602 journal->j_num_lists--;
1603 del_from_work_list(s, jl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001605 if (journal->j_last_flush_id != 0 &&
1606 (jl->j_trans_id - journal->j_last_flush_id) != 1) {
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001607 reiserfs_warning(s, "clm-2201", "last flush %lu, current %lu",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001608 journal->j_last_flush_id, jl->j_trans_id);
1609 }
1610 journal->j_last_flush_id = jl->j_trans_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611
Jeff Mahoney098297b2014-04-23 10:00:36 -04001612 /*
1613 * not strictly required since we are freeing the list, but it should
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001614 * help find code using dead lists later on
1615 */
1616 jl->j_len = 0;
Jeff Mahoneya228bf82014-04-23 10:00:42 -04001617 atomic_set(&jl->j_nonzerolen, 0);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001618 jl->j_start = 0;
1619 jl->j_realblock = NULL;
1620 jl->j_commit_bh = NULL;
1621 jl->j_trans_id = 0;
1622 jl->j_state = 0;
1623 put_journal_list(s, jl);
1624 if (flushall)
Jeff Mahoneyafe70252008-07-25 01:46:39 -07001625 mutex_unlock(&journal->j_flush_mutex);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001626 return err;
1627}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628
1629static 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) {
Jeff Mahoney098297b2014-04-23 10:00:36 -04001644 /*
1645 * if the blocknr == 0, this has been cleared from the hash,
1646 * skip it
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001647 */
1648 if (cn->blocknr == 0) {
1649 goto next;
1650 }
1651 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1652 struct buffer_head *tmp_bh;
Jeff Mahoney098297b2014-04-23 10:00:36 -04001653 /*
1654 * we can race against journal_mark_freed when we try
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001655 * to lock_buffer(cn->bh), so we have to inc the buffer
1656 * count, and recheck things after locking
1657 */
1658 tmp_bh = cn->bh;
1659 get_bh(tmp_bh);
1660 lock_buffer(tmp_bh);
1661 if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1662 if (!buffer_journal_dirty(tmp_bh) ||
1663 buffer_journal_prepared(tmp_bh))
1664 BUG();
1665 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1666 ret++;
1667 } else {
1668 /* note, cn->bh might be null now */
1669 unlock_buffer(tmp_bh);
1670 }
1671 put_bh(tmp_bh);
1672 }
Jeff Mahoneycf776a72014-04-23 10:00:41 -04001673next:
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001674 cn = cn->next;
1675 cond_resched();
1676 }
1677 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678}
1679
1680/* used by flush_commit_list */
1681static int dirty_one_transaction(struct super_block *s,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001682 struct reiserfs_journal_list *jl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001684 struct reiserfs_journal_cnode *cn;
1685 struct reiserfs_journal_list *pjl;
1686 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001688 jl->j_state |= LIST_DIRTY;
1689 cn = jl->j_realblock;
1690 while (cn) {
Jeff Mahoney098297b2014-04-23 10:00:36 -04001691 /*
1692 * look for a more recent transaction that logged this
1693 * buffer. Only the most recent transaction with a buffer in
1694 * it is allowed to send that buffer to disk
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001695 */
1696 pjl = find_newer_jl_for_cn(cn);
1697 if (!pjl && cn->blocknr && cn->bh
1698 && buffer_journal_dirty(cn->bh)) {
1699 BUG_ON(!can_dirty(cn));
Jeff Mahoney098297b2014-04-23 10:00:36 -04001700 /*
1701 * if the buffer is prepared, it will either be logged
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001702 * or restored. If restored, we need to make sure
1703 * it actually gets marked dirty
1704 */
1705 clear_buffer_journal_new(cn->bh);
1706 if (buffer_journal_prepared(cn->bh)) {
1707 set_buffer_journal_restore_dirty(cn->bh);
1708 } else {
1709 set_buffer_journal_test(cn->bh);
1710 mark_buffer_dirty(cn->bh);
1711 }
1712 }
1713 cn = cn->next;
1714 }
1715 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716}
1717
1718static int kupdate_transactions(struct super_block *s,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001719 struct reiserfs_journal_list *jl,
1720 struct reiserfs_journal_list **next_jl,
Jeff Mahoney600ed412009-03-30 14:02:17 -04001721 unsigned int *next_trans_id,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001722 int num_blocks, int num_trans)
1723{
1724 int ret = 0;
1725 int written = 0;
1726 int transactions_flushed = 0;
Jeff Mahoney600ed412009-03-30 14:02:17 -04001727 unsigned int orig_trans_id = jl->j_trans_id;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001728 struct buffer_chunk chunk;
1729 struct list_head *entry;
1730 struct reiserfs_journal *journal = SB_JOURNAL(s);
1731 chunk.nr = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732
Frederic Weisbeckera412f9e2009-04-14 00:10:35 +02001733 reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001734 if (!journal_list_still_alive(s, orig_trans_id)) {
1735 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737
Jeff Mahoney098297b2014-04-23 10:00:36 -04001738 /*
1739 * we've got j_flush_mutex held, nobody is going to delete any
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001740 * of these lists out from underneath us
1741 */
1742 while ((num_trans && transactions_flushed < num_trans) ||
1743 (!num_trans && written < num_blocks)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001745 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1746 atomic_read(&jl->j_commit_left)
1747 || !(jl->j_state & LIST_DIRTY)) {
1748 del_from_work_list(s, jl);
1749 break;
1750 }
1751 ret = write_one_transaction(s, jl, &chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001753 if (ret < 0)
1754 goto done;
1755 transactions_flushed++;
1756 written += ret;
1757 entry = jl->j_list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001759 /* did we wrap? */
1760 if (entry == &journal->j_journal_list) {
1761 break;
1762 }
1763 jl = JOURNAL_LIST_ENTRY(entry);
1764
1765 /* don't bother with older transactions */
1766 if (jl->j_trans_id <= orig_trans_id)
1767 break;
1768 }
1769 if (chunk.nr) {
1770 write_chunk(&chunk);
1771 }
1772
Jeff Mahoneycf776a72014-04-23 10:00:41 -04001773done:
Jeff Mahoneyafe70252008-07-25 01:46:39 -07001774 mutex_unlock(&journal->j_flush_mutex);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001775 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776}
1777
Jeff Mahoney098297b2014-04-23 10:00:36 -04001778/*
1779 * for o_sync and fsync heavy applications, they tend to use
1780 * all the journa list slots with tiny transactions. These
1781 * trigger lots and lots of calls to update the header block, which
1782 * adds seeks and slows things down.
1783 *
1784 * This function tries to clear out a large chunk of the journal lists
1785 * at once, which makes everything faster since only the newest journal
1786 * list updates the header block
1787 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788static int flush_used_journal_lists(struct super_block *s,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001789 struct reiserfs_journal_list *jl)
1790{
1791 unsigned long len = 0;
1792 unsigned long cur_len;
1793 int ret;
1794 int i;
1795 int limit = 256;
1796 struct reiserfs_journal_list *tjl;
1797 struct reiserfs_journal_list *flush_jl;
Jeff Mahoney600ed412009-03-30 14:02:17 -04001798 unsigned int trans_id;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001799 struct reiserfs_journal *journal = SB_JOURNAL(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001801 flush_jl = tjl = jl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001803 /* in data logging mode, try harder to flush a lot of blocks */
1804 if (reiserfs_data_log(s))
1805 limit = 1024;
1806 /* flush for 256 transactions or limit blocks, whichever comes first */
1807 for (i = 0; i < 256 && len < limit; i++) {
1808 if (atomic_read(&tjl->j_commit_left) ||
1809 tjl->j_trans_id < jl->j_trans_id) {
1810 break;
1811 }
1812 cur_len = atomic_read(&tjl->j_nonzerolen);
1813 if (cur_len > 0) {
1814 tjl->j_state &= ~LIST_TOUCHED;
1815 }
1816 len += cur_len;
1817 flush_jl = tjl;
1818 if (tjl->j_list.next == &journal->j_journal_list)
1819 break;
1820 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 }
Jeff Mahoney721a7692013-09-23 16:50:42 -04001822 get_journal_list(jl);
1823 get_journal_list(flush_jl);
Jeff Mahoney098297b2014-04-23 10:00:36 -04001824
1825 /*
1826 * try to find a group of blocks we can flush across all the
1827 * transactions, but only bother if we've actually spanned
1828 * across multiple lists
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001829 */
1830 if (flush_jl != jl) {
1831 ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001833 flush_journal_list(s, flush_jl, 1);
Jeff Mahoney721a7692013-09-23 16:50:42 -04001834 put_journal_list(s, flush_jl);
1835 put_journal_list(s, jl);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001836 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837}
1838
1839/*
Jeff Mahoney098297b2014-04-23 10:00:36 -04001840 * removes any nodes in table with name block and dev as bh.
1841 * only touchs the hnext and hprev pointers.
1842 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843void remove_journal_hash(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001844 struct reiserfs_journal_cnode **table,
1845 struct reiserfs_journal_list *jl,
1846 unsigned long block, int remove_freed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001848 struct reiserfs_journal_cnode *cur;
1849 struct reiserfs_journal_cnode **head;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001851 head = &(journal_hash(table, sb, block));
1852 if (!head) {
1853 return;
1854 }
1855 cur = *head;
1856 while (cur) {
1857 if (cur->blocknr == block && cur->sb == sb
1858 && (jl == NULL || jl == cur->jlist)
1859 && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1860 if (cur->hnext) {
1861 cur->hnext->hprev = cur->hprev;
1862 }
1863 if (cur->hprev) {
1864 cur->hprev->hnext = cur->hnext;
1865 } else {
1866 *head = cur->hnext;
1867 }
1868 cur->blocknr = 0;
1869 cur->sb = NULL;
1870 cur->state = 0;
Jeff Mahoney098297b2014-04-23 10:00:36 -04001871 /*
1872 * anybody who clears the cur->bh will also
1873 * dec the nonzerolen
1874 */
1875 if (cur->bh && cur->jlist)
Jeff Mahoneya228bf82014-04-23 10:00:42 -04001876 atomic_dec(&cur->jlist->j_nonzerolen);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001877 cur->bh = NULL;
1878 cur->jlist = NULL;
1879 }
1880 cur = cur->hnext;
1881 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882}
1883
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001884static void free_journal_ram(struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001885{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001886 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Pekka Enbergd739b422006-02-01 03:06:43 -08001887 kfree(journal->j_current_jl);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001888 journal->j_num_lists--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001890 vfree(journal->j_cnode_free_orig);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001891 free_list_bitmaps(sb, journal->j_list_bitmap);
1892 free_bitmap_nodes(sb); /* must be after free_list_bitmaps */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001893 if (journal->j_header_bh) {
1894 brelse(journal->j_header_bh);
1895 }
Jeff Mahoney098297b2014-04-23 10:00:36 -04001896 /*
1897 * j_header_bh is on the journal dev, make sure
1898 * not to release the journal dev until we brelse j_header_bh
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001899 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001900 release_journal_dev(sb, journal);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001901 vfree(journal);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902}
1903
1904/*
Jeff Mahoney098297b2014-04-23 10:00:36 -04001905 * call on unmount. Only set error to 1 if you haven't made your way out
1906 * of read_super() yet. Any other caller must keep error at 0.
1907 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001908static int do_journal_release(struct reiserfs_transaction_handle *th,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001909 struct super_block *sb, int error)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001910{
1911 struct reiserfs_transaction_handle myth;
1912 int flushed = 0;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001913 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914
Jeff Mahoney098297b2014-04-23 10:00:36 -04001915 /*
1916 * we only want to flush out transactions if we were
1917 * called with error == 0
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001918 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001919 if (!error && !(sb->s_flags & MS_RDONLY)) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001920 /* end the current trans */
1921 BUG_ON(!th->t_trans_id);
Jeff Mahoney58d85422014-04-23 10:00:38 -04001922 do_journal_end(th, FLUSH_ALL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923
Jeff Mahoney098297b2014-04-23 10:00:36 -04001924 /*
1925 * make sure something gets logged to force
1926 * our way into the flush code
1927 */
Jeff Mahoneyb491dd12014-04-23 10:00:40 -04001928 if (!journal_join(&myth, sb)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001929 reiserfs_prepare_for_journal(sb,
1930 SB_BUFFER_WITH_SB(sb),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001931 1);
Jeff Mahoney09f1b802014-04-23 10:00:39 -04001932 journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb));
Jeff Mahoney58d85422014-04-23 10:00:38 -04001933 do_journal_end(&myth, FLUSH_ALL);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001934 flushed = 1;
1935 }
1936 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001938 /* this also catches errors during the do_journal_end above */
1939 if (!error && reiserfs_is_journal_aborted(journal)) {
1940 memset(&myth, 0, sizeof(myth));
Jeff Mahoneyb491dd12014-04-23 10:00:40 -04001941 if (!journal_join_abort(&myth, sb)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001942 reiserfs_prepare_for_journal(sb,
1943 SB_BUFFER_WITH_SB(sb),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001944 1);
Jeff Mahoney09f1b802014-04-23 10:00:39 -04001945 journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb));
Jeff Mahoney58d85422014-04-23 10:00:38 -04001946 do_journal_end(&myth, FLUSH_ALL);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001947 }
1948 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02001950
1951 /*
1952 * We must release the write lock here because
1953 * the workqueue job (flush_async_commit) needs this lock
1954 */
1955 reiserfs_write_unlock(sb);
Artem Bityutskiy033369d2012-06-01 17:18:08 +03001956
Jan Kara01777832014-08-06 19:43:56 +02001957 /*
1958 * Cancel flushing of old commits. Note that neither of these works
1959 * will be requeued because superblock is being shutdown and doesn't
1960 * have MS_ACTIVE set.
1961 */
Artem Bityutskiy033369d2012-06-01 17:18:08 +03001962 cancel_delayed_work_sync(&REISERFS_SB(sb)->old_work);
Jan Kara01777832014-08-06 19:43:56 +02001963 /* wait for all commits to finish */
1964 cancel_delayed_work_sync(&SB_JOURNAL(sb)->j_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001966 free_journal_ram(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967
Frederic Weisbecker05236762009-12-30 05:56:08 +01001968 reiserfs_write_lock(sb);
1969
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001970 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971}
1972
Jeff Mahoney098297b2014-04-23 10:00:36 -04001973/* * call on unmount. flush all journal trans, release all alloc'd ram */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001974int journal_release(struct reiserfs_transaction_handle *th,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001975 struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001976{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001977 return do_journal_release(th, sb, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978}
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001979
Jeff Mahoney098297b2014-04-23 10:00:36 -04001980/* only call from an error condition inside reiserfs_read_super! */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001981int journal_release_error(struct reiserfs_transaction_handle *th,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001982 struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001983{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001984 return do_journal_release(th, sb, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985}
1986
Jeff Mahoney098297b2014-04-23 10:00:36 -04001987/*
1988 * compares description block with commit block.
1989 * returns 1 if they differ, 0 if they are the same
1990 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001991static int journal_compare_desc_commit(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001992 struct reiserfs_journal_desc *desc,
1993 struct reiserfs_journal_commit *commit)
1994{
1995 if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
1996 get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001997 get_commit_trans_len(commit) > SB_JOURNAL(sb)->j_trans_max ||
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001998 get_commit_trans_len(commit) <= 0) {
1999 return 1;
2000 }
2001 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002}
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002003
Jeff Mahoney098297b2014-04-23 10:00:36 -04002004/*
2005 * returns 0 if it did not find a description block
2006 * returns -1 if it found a corrupt commit block
2007 * returns 1 if both desc and commit were valid
2008 * NOTE: only called during fs mount
2009 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002010static int journal_transaction_is_valid(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002011 struct buffer_head *d_bh,
Jeff Mahoney600ed412009-03-30 14:02:17 -04002012 unsigned int *oldest_invalid_trans_id,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002013 unsigned long *newest_mount_id)
2014{
2015 struct reiserfs_journal_desc *desc;
2016 struct reiserfs_journal_commit *commit;
2017 struct buffer_head *c_bh;
2018 unsigned long offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002020 if (!d_bh)
2021 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002023 desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2024 if (get_desc_trans_len(desc) > 0
2025 && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
2026 if (oldest_invalid_trans_id && *oldest_invalid_trans_id
2027 && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002028 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002029 "journal-986: transaction "
2030 "is valid returning because trans_id %d is greater than "
2031 "oldest_invalid %lu",
2032 get_desc_trans_id(desc),
2033 *oldest_invalid_trans_id);
2034 return 0;
2035 }
2036 if (newest_mount_id
2037 && *newest_mount_id > get_desc_mount_id(desc)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002038 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002039 "journal-1087: transaction "
2040 "is valid returning because mount_id %d is less than "
2041 "newest_mount_id %lu",
2042 get_desc_mount_id(desc),
2043 *newest_mount_id);
2044 return -1;
2045 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002046 if (get_desc_trans_len(desc) > SB_JOURNAL(sb)->j_trans_max) {
2047 reiserfs_warning(sb, "journal-2018",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002048 "Bad transaction length %d "
2049 "encountered, ignoring transaction",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002050 get_desc_trans_len(desc));
2051 return -1;
2052 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002053 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054
Jeff Mahoney098297b2014-04-23 10:00:36 -04002055 /*
2056 * ok, we have a journal description block,
2057 * let's see if the transaction was valid
2058 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002059 c_bh =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002060 journal_bread(sb,
2061 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002062 ((offset + get_desc_trans_len(desc) +
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002063 1) % SB_ONDISK_JOURNAL_SIZE(sb)));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002064 if (!c_bh)
2065 return 0;
2066 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002067 if (journal_compare_desc_commit(sb, desc, commit)) {
2068 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002069 "journal_transaction_is_valid, commit offset %ld had bad "
2070 "time %d or length %d",
2071 c_bh->b_blocknr -
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002072 SB_ONDISK_JOURNAL_1st_BLOCK(sb),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002073 get_commit_trans_id(commit),
2074 get_commit_trans_len(commit));
2075 brelse(c_bh);
2076 if (oldest_invalid_trans_id) {
2077 *oldest_invalid_trans_id =
2078 get_desc_trans_id(desc);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002079 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002080 "journal-1004: "
2081 "transaction_is_valid setting oldest invalid trans_id "
2082 "to %d",
2083 get_desc_trans_id(desc));
2084 }
2085 return -1;
2086 }
2087 brelse(c_bh);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002088 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002089 "journal-1006: found valid "
2090 "transaction start offset %llu, len %d id %d",
2091 d_bh->b_blocknr -
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002092 SB_ONDISK_JOURNAL_1st_BLOCK(sb),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002093 get_desc_trans_len(desc),
2094 get_desc_trans_id(desc));
2095 return 1;
2096 } else {
2097 return 0;
2098 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099}
2100
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002101static void brelse_array(struct buffer_head **heads, int num)
2102{
2103 int i;
2104 for (i = 0; i < num; i++) {
2105 brelse(heads[i]);
2106 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107}
2108
2109/*
Jeff Mahoney098297b2014-04-23 10:00:36 -04002110 * given the start, and values for the oldest acceptable transactions,
2111 * this either reads in a replays a transaction, or returns because the
2112 * transaction is invalid, or too old.
2113 * NOTE: only called during fs mount
2114 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002115static int journal_read_transaction(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002116 unsigned long cur_dblock,
2117 unsigned long oldest_start,
Jeff Mahoney600ed412009-03-30 14:02:17 -04002118 unsigned int oldest_trans_id,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002119 unsigned long newest_mount_id)
2120{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002121 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002122 struct reiserfs_journal_desc *desc;
2123 struct reiserfs_journal_commit *commit;
Jeff Mahoney600ed412009-03-30 14:02:17 -04002124 unsigned int trans_id = 0;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002125 struct buffer_head *c_bh;
2126 struct buffer_head *d_bh;
2127 struct buffer_head **log_blocks = NULL;
2128 struct buffer_head **real_blocks = NULL;
Jeff Mahoney600ed412009-03-30 14:02:17 -04002129 unsigned int trans_offset;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002130 int i;
2131 int trans_half;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002133 d_bh = journal_bread(sb, cur_dblock);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002134 if (!d_bh)
2135 return 1;
2136 desc = (struct reiserfs_journal_desc *)d_bh->b_data;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002137 trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2138 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1037: "
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002139 "journal_read_transaction, offset %llu, len %d mount_id %d",
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002140 d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002141 get_desc_trans_len(desc), get_desc_mount_id(desc));
2142 if (get_desc_trans_id(desc) < oldest_trans_id) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002143 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1039: "
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002144 "journal_read_trans skipping because %lu is too old",
2145 cur_dblock -
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002146 SB_ONDISK_JOURNAL_1st_BLOCK(sb));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002147 brelse(d_bh);
2148 return 1;
2149 }
2150 if (get_desc_mount_id(desc) != newest_mount_id) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002151 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1146: "
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002152 "journal_read_trans skipping because %d is != "
2153 "newest_mount_id %lu", get_desc_mount_id(desc),
2154 newest_mount_id);
2155 brelse(d_bh);
2156 return 1;
2157 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002158 c_bh = journal_bread(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002159 ((trans_offset + get_desc_trans_len(desc) + 1) %
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002160 SB_ONDISK_JOURNAL_SIZE(sb)));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002161 if (!c_bh) {
2162 brelse(d_bh);
2163 return 1;
2164 }
2165 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002166 if (journal_compare_desc_commit(sb, desc, commit)) {
2167 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002168 "journal_read_transaction, "
2169 "commit offset %llu had bad time %d or length %d",
2170 c_bh->b_blocknr -
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002171 SB_ONDISK_JOURNAL_1st_BLOCK(sb),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002172 get_commit_trans_id(commit),
2173 get_commit_trans_len(commit));
2174 brelse(c_bh);
2175 brelse(d_bh);
2176 return 1;
2177 }
Jeff Mahoney3f8b5ee2010-03-23 13:35:39 -07002178
2179 if (bdev_read_only(sb->s_bdev)) {
2180 reiserfs_warning(sb, "clm-2076",
2181 "device is readonly, unable to replay log");
2182 brelse(c_bh);
2183 brelse(d_bh);
2184 return -EROFS;
2185 }
2186
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002187 trans_id = get_desc_trans_id(desc);
Jeff Mahoney098297b2014-04-23 10:00:36 -04002188 /*
2189 * now we know we've got a good transaction, and it was
2190 * inside the valid time ranges
2191 */
Pekka Enbergd739b422006-02-01 03:06:43 -08002192 log_blocks = kmalloc(get_desc_trans_len(desc) *
2193 sizeof(struct buffer_head *), GFP_NOFS);
2194 real_blocks = kmalloc(get_desc_trans_len(desc) *
2195 sizeof(struct buffer_head *), GFP_NOFS);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002196 if (!log_blocks || !real_blocks) {
2197 brelse(c_bh);
2198 brelse(d_bh);
Pekka Enbergd739b422006-02-01 03:06:43 -08002199 kfree(log_blocks);
2200 kfree(real_blocks);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002201 reiserfs_warning(sb, "journal-1169",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002202 "kmalloc failed, unable to mount FS");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002203 return -1;
2204 }
2205 /* get all the buffer heads */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002206 trans_half = journal_trans_half(sb->s_blocksize);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002207 for (i = 0; i < get_desc_trans_len(desc); i++) {
2208 log_blocks[i] =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002209 journal_getblk(sb,
2210 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002211 (trans_offset + 1 +
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002212 i) % SB_ONDISK_JOURNAL_SIZE(sb));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002213 if (i < trans_half) {
2214 real_blocks[i] =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002215 sb_getblk(sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002216 le32_to_cpu(desc->j_realblock[i]));
2217 } else {
2218 real_blocks[i] =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002219 sb_getblk(sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002220 le32_to_cpu(commit->
2221 j_realblock[i - trans_half]));
2222 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002223 if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(sb)) {
2224 reiserfs_warning(sb, "journal-1207",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002225 "REPLAY FAILURE fsck required! "
2226 "Block to replay is outside of "
2227 "filesystem");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002228 goto abort_replay;
2229 }
2230 /* make sure we don't try to replay onto log or reserved area */
2231 if (is_block_in_log_or_reserved_area
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002232 (sb, real_blocks[i]->b_blocknr)) {
2233 reiserfs_warning(sb, "journal-1204",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002234 "REPLAY FAILURE fsck required! "
2235 "Trying to replay onto a log block");
Jeff Mahoneycf776a72014-04-23 10:00:41 -04002236abort_replay:
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002237 brelse_array(log_blocks, i);
2238 brelse_array(real_blocks, i);
2239 brelse(c_bh);
2240 brelse(d_bh);
Pekka Enbergd739b422006-02-01 03:06:43 -08002241 kfree(log_blocks);
2242 kfree(real_blocks);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002243 return -1;
2244 }
2245 }
2246 /* read in the log blocks, memcpy to the corresponding real block */
2247 ll_rw_block(READ, get_desc_trans_len(desc), log_blocks);
2248 for (i = 0; i < get_desc_trans_len(desc); i++) {
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02002249
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002250 wait_on_buffer(log_blocks[i]);
2251 if (!buffer_uptodate(log_blocks[i])) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002252 reiserfs_warning(sb, "journal-1212",
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(log_blocks + i,
2256 get_desc_trans_len(desc) - i);
2257 brelse_array(real_blocks, get_desc_trans_len(desc));
2258 brelse(c_bh);
2259 brelse(d_bh);
Pekka Enbergd739b422006-02-01 03:06:43 -08002260 kfree(log_blocks);
2261 kfree(real_blocks);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002262 return -1;
2263 }
2264 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2265 real_blocks[i]->b_size);
2266 set_buffer_uptodate(real_blocks[i]);
2267 brelse(log_blocks[i]);
2268 }
2269 /* flush out the real blocks */
2270 for (i = 0; i < get_desc_trans_len(desc); i++) {
2271 set_buffer_dirty(real_blocks[i]);
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02002272 write_dirty_buffer(real_blocks[i], WRITE);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002273 }
2274 for (i = 0; i < get_desc_trans_len(desc); i++) {
2275 wait_on_buffer(real_blocks[i]);
2276 if (!buffer_uptodate(real_blocks[i])) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002277 reiserfs_warning(sb, "journal-1226",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002278 "REPLAY FAILURE, fsck required! "
2279 "buffer write failed");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002280 brelse_array(real_blocks + i,
2281 get_desc_trans_len(desc) - i);
2282 brelse(c_bh);
2283 brelse(d_bh);
Pekka Enbergd739b422006-02-01 03:06:43 -08002284 kfree(log_blocks);
2285 kfree(real_blocks);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002286 return -1;
2287 }
2288 brelse(real_blocks[i]);
2289 }
2290 cur_dblock =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002291 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002292 ((trans_offset + get_desc_trans_len(desc) +
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002293 2) % SB_ONDISK_JOURNAL_SIZE(sb));
2294 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002295 "journal-1095: setting journal " "start to offset %ld",
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002296 cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002297
Jeff Mahoney098297b2014-04-23 10:00:36 -04002298 /*
2299 * init starting values for the first transaction, in case
2300 * this is the last transaction to be replayed.
2301 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002302 journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002303 journal->j_last_flush_trans_id = trans_id;
2304 journal->j_trans_id = trans_id + 1;
Alexander Zarochentseva44c94a2006-03-25 03:06:59 -08002305 /* check for trans_id overflow */
2306 if (journal->j_trans_id == 0)
2307 journal->j_trans_id = 10;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002308 brelse(c_bh);
2309 brelse(d_bh);
Pekka Enbergd739b422006-02-01 03:06:43 -08002310 kfree(log_blocks);
2311 kfree(real_blocks);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002312 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313}
2314
Jeff Mahoney098297b2014-04-23 10:00:36 -04002315/*
2316 * This function reads blocks starting from block and to max_block of bufsize
2317 * size (but no more than BUFNR blocks at a time). This proved to improve
2318 * mounting speed on self-rebuilding raid5 arrays at least.
2319 * Right now it is only used from journal code. But later we might use it
2320 * from other places.
2321 * Note: Do not use journal_getblk/sb_getblk functions here!
2322 */
Jeff Mahoney3ee16672007-10-18 23:39:25 -07002323static struct buffer_head *reiserfs_breada(struct block_device *dev,
2324 b_blocknr_t block, int bufsize,
2325 b_blocknr_t max_block)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002327 struct buffer_head *bhlist[BUFNR];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328 unsigned int blocks = BUFNR;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002329 struct buffer_head *bh;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330 int i, j;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002331
2332 bh = __getblk(dev, block, bufsize);
2333 if (buffer_uptodate(bh))
2334 return (bh);
2335
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 if (block + BUFNR > max_block) {
2337 blocks = max_block - block;
2338 }
2339 bhlist[0] = bh;
2340 j = 1;
2341 for (i = 1; i < blocks; i++) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002342 bh = __getblk(dev, block + i, bufsize);
2343 if (buffer_uptodate(bh)) {
2344 brelse(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345 break;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002346 } else
2347 bhlist[j++] = bh;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002349 ll_rw_block(READ, j, bhlist);
2350 for (i = 1; i < j; i++)
2351 brelse(bhlist[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352 bh = bhlist[0];
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002353 wait_on_buffer(bh);
2354 if (buffer_uptodate(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355 return bh;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002356 brelse(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357 return NULL;
2358}
2359
2360/*
Jeff Mahoney098297b2014-04-23 10:00:36 -04002361 * read and replay the log
2362 * on a clean unmount, the journal header's next unflushed pointer will be
2363 * to an invalid transaction. This tests that before finding all the
2364 * transactions in the log, which makes normal mount times fast.
2365 *
2366 * After a crash, this starts with the next unflushed transaction, and
2367 * replays until it finds one too old, or invalid.
2368 *
2369 * On exit, it sets things up so the first transaction will work correctly.
2370 * NOTE: only called during fs mount
2371 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002372static int journal_read(struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002373{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002374 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002375 struct reiserfs_journal_desc *desc;
Jeff Mahoney600ed412009-03-30 14:02:17 -04002376 unsigned int oldest_trans_id = 0;
2377 unsigned int oldest_invalid_trans_id = 0;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002378 time_t start;
2379 unsigned long oldest_start = 0;
2380 unsigned long cur_dblock = 0;
2381 unsigned long newest_mount_id = 9;
2382 struct buffer_head *d_bh;
2383 struct reiserfs_journal_header *jh;
2384 int valid_journal_header = 0;
2385 int replay_count = 0;
2386 int continue_replay = 1;
2387 int ret;
2388 char b[BDEVNAME_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002390 cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2391 reiserfs_info(sb, "checking transaction log (%s)\n",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002392 bdevname(journal->j_dev_bd, b));
2393 start = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002394
Jeff Mahoney098297b2014-04-23 10:00:36 -04002395 /*
2396 * step 1, read in the journal header block. Check the transaction
2397 * it says is the first unflushed, and if that transaction is not
2398 * valid, replay is done
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002399 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002400 journal->j_header_bh = journal_bread(sb,
2401 SB_ONDISK_JOURNAL_1st_BLOCK(sb)
2402 + SB_ONDISK_JOURNAL_SIZE(sb));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002403 if (!journal->j_header_bh) {
2404 return 1;
2405 }
2406 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
Vladimir V. Savelievc499ec22006-03-02 02:54:39 -08002407 if (le32_to_cpu(jh->j_first_unflushed_offset) <
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002408 SB_ONDISK_JOURNAL_SIZE(sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002409 && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2410 oldest_start =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002411 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002412 le32_to_cpu(jh->j_first_unflushed_offset);
2413 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2414 newest_mount_id = le32_to_cpu(jh->j_mount_id);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002415 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002416 "journal-1153: found in "
2417 "header: first_unflushed_offset %d, last_flushed_trans_id "
2418 "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2419 le32_to_cpu(jh->j_last_flush_trans_id));
2420 valid_journal_header = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421
Jeff Mahoney098297b2014-04-23 10:00:36 -04002422 /*
2423 * now, we try to read the first unflushed offset. If it
2424 * is not valid, there is nothing more we can do, and it
2425 * makes no sense to read through the whole log.
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002426 */
2427 d_bh =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002428 journal_bread(sb,
2429 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002430 le32_to_cpu(jh->j_first_unflushed_offset));
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002431 ret = journal_transaction_is_valid(sb, d_bh, NULL, NULL);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002432 if (!ret) {
2433 continue_replay = 0;
2434 }
2435 brelse(d_bh);
2436 goto start_log_replay;
2437 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438
Jeff Mahoney098297b2014-04-23 10:00:36 -04002439 /*
2440 * ok, there are transactions that need to be replayed. start
2441 * with the first log block, find all the valid transactions, and
2442 * pick out the oldest.
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002443 */
2444 while (continue_replay
2445 && cur_dblock <
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002446 (SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2447 SB_ONDISK_JOURNAL_SIZE(sb))) {
Jeff Mahoney098297b2014-04-23 10:00:36 -04002448 /*
2449 * Note that it is required for blocksize of primary fs
2450 * device and journal device to be the same
2451 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002452 d_bh =
2453 reiserfs_breada(journal->j_dev_bd, cur_dblock,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002454 sb->s_blocksize,
2455 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2456 SB_ONDISK_JOURNAL_SIZE(sb));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002457 ret =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002458 journal_transaction_is_valid(sb, d_bh,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002459 &oldest_invalid_trans_id,
2460 &newest_mount_id);
2461 if (ret == 1) {
2462 desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2463 if (oldest_start == 0) { /* init all oldest_ values */
2464 oldest_trans_id = get_desc_trans_id(desc);
2465 oldest_start = d_bh->b_blocknr;
2466 newest_mount_id = get_desc_mount_id(desc);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002467 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002468 "journal-1179: Setting "
2469 "oldest_start to offset %llu, trans_id %lu",
2470 oldest_start -
2471 SB_ONDISK_JOURNAL_1st_BLOCK
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002472 (sb), oldest_trans_id);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002473 } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2474 /* one we just read was older */
2475 oldest_trans_id = get_desc_trans_id(desc);
2476 oldest_start = d_bh->b_blocknr;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002477 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002478 "journal-1180: Resetting "
2479 "oldest_start to offset %lu, trans_id %lu",
2480 oldest_start -
2481 SB_ONDISK_JOURNAL_1st_BLOCK
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002482 (sb), oldest_trans_id);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002483 }
2484 if (newest_mount_id < get_desc_mount_id(desc)) {
2485 newest_mount_id = get_desc_mount_id(desc);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002486 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002487 "journal-1299: Setting "
2488 "newest_mount_id to %d",
2489 get_desc_mount_id(desc));
2490 }
2491 cur_dblock += get_desc_trans_len(desc) + 2;
2492 } else {
2493 cur_dblock++;
2494 }
2495 brelse(d_bh);
2496 }
2497
Jeff Mahoneycf776a72014-04-23 10:00:41 -04002498start_log_replay:
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002499 cur_dblock = oldest_start;
2500 if (oldest_trans_id) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002501 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002502 "journal-1206: Starting replay "
2503 "from offset %llu, trans_id %lu",
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002504 cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002505 oldest_trans_id);
2506
2507 }
2508 replay_count = 0;
2509 while (continue_replay && oldest_trans_id > 0) {
2510 ret =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002511 journal_read_transaction(sb, cur_dblock, oldest_start,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002512 oldest_trans_id, newest_mount_id);
2513 if (ret < 0) {
2514 return ret;
2515 } else if (ret != 0) {
2516 break;
2517 }
2518 cur_dblock =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002519 SB_ONDISK_JOURNAL_1st_BLOCK(sb) + journal->j_start;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002520 replay_count++;
2521 if (cur_dblock == oldest_start)
2522 break;
2523 }
2524
2525 if (oldest_trans_id == 0) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002526 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002527 "journal-1225: No valid " "transactions found");
2528 }
Jeff Mahoney098297b2014-04-23 10:00:36 -04002529 /*
2530 * j_start does not get set correctly if we don't replay any
2531 * transactions. if we had a valid journal_header, set j_start
2532 * to the first unflushed transaction value, copy the trans_id
2533 * from the header
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002534 */
2535 if (valid_journal_header && replay_count == 0) {
2536 journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2537 journal->j_trans_id =
2538 le32_to_cpu(jh->j_last_flush_trans_id) + 1;
Alexander Zarochentseva44c94a2006-03-25 03:06:59 -08002539 /* check for trans_id overflow */
2540 if (journal->j_trans_id == 0)
2541 journal->j_trans_id = 10;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002542 journal->j_last_flush_trans_id =
2543 le32_to_cpu(jh->j_last_flush_trans_id);
2544 journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2545 } else {
2546 journal->j_mount_id = newest_mount_id + 1;
2547 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002548 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002549 "newest_mount_id to %lu", journal->j_mount_id);
2550 journal->j_first_unflushed_offset = journal->j_start;
2551 if (replay_count > 0) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002552 reiserfs_info(sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002553 "replayed %d transactions in %lu seconds\n",
2554 replay_count, get_seconds() - start);
2555 }
Jeff Mahoney278f6672013-08-08 17:34:46 -04002556 /* needed to satisfy the locking in _update_journal_header_block */
2557 reiserfs_write_lock(sb);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002558 if (!bdev_read_only(sb->s_bdev) &&
2559 _update_journal_header_block(sb, journal->j_start,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002560 journal->j_last_flush_trans_id)) {
Jeff Mahoney278f6672013-08-08 17:34:46 -04002561 reiserfs_write_unlock(sb);
Jeff Mahoney098297b2014-04-23 10:00:36 -04002562 /*
2563 * replay failed, caller must call free_journal_ram and abort
2564 * the mount
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002565 */
2566 return -1;
2567 }
Jeff Mahoney278f6672013-08-08 17:34:46 -04002568 reiserfs_write_unlock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002569 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570}
2571
2572static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2573{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002574 struct reiserfs_journal_list *jl;
Pekka Enberg8c777cc2006-02-01 03:06:43 -08002575 jl = kzalloc(sizeof(struct reiserfs_journal_list),
2576 GFP_NOFS | __GFP_NOFAIL);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002577 INIT_LIST_HEAD(&jl->j_list);
2578 INIT_LIST_HEAD(&jl->j_working_list);
2579 INIT_LIST_HEAD(&jl->j_tail_bh_list);
2580 INIT_LIST_HEAD(&jl->j_bh_list);
Jeff Mahoney90415de2008-07-25 01:46:40 -07002581 mutex_init(&jl->j_commit_mutex);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002582 SB_JOURNAL(s)->j_num_lists++;
2583 get_journal_list(jl);
2584 return jl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585}
2586
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002587static void journal_list_init(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002589 SB_JOURNAL(sb)->j_current_jl = alloc_journal_list(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590}
2591
Al Viro4385bab2013-05-05 22:11:03 -04002592static void release_journal_dev(struct super_block *super,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002593 struct reiserfs_journal *journal)
2594{
Christoph Hellwig86098fa2008-04-30 00:54:46 -07002595 if (journal->j_dev_bd != NULL) {
Al Viro4385bab2013-05-05 22:11:03 -04002596 blkdev_put(journal->j_dev_bd, journal->j_dev_mode);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002597 journal->j_dev_bd = NULL;
2598 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002599}
2600
2601static int journal_init_dev(struct super_block *super,
2602 struct reiserfs_journal *journal,
2603 const char *jdev_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002604{
2605 int result;
2606 dev_t jdev;
Tejun Heoe525fd82010-11-13 11:55:17 +01002607 fmode_t blkdev_mode = FMODE_READ | FMODE_WRITE | FMODE_EXCL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608 char b[BDEVNAME_SIZE];
2609
2610 result = 0;
2611
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002612 journal->j_dev_bd = NULL;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002613 jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2614 new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615
2616 if (bdev_read_only(super->s_bdev))
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002617 blkdev_mode = FMODE_READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618
2619 /* there is no "jdev" option and journal is on separate device */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002620 if ((!jdev_name || !jdev_name[0])) {
Tejun Heoe525fd82010-11-13 11:55:17 +01002621 if (jdev == super->s_dev)
2622 blkdev_mode &= ~FMODE_EXCL;
Tejun Heod4d77622010-11-13 11:55:18 +01002623 journal->j_dev_bd = blkdev_get_by_dev(jdev, blkdev_mode,
2624 journal);
Al Viroe5eb8ca2007-10-08 13:24:05 -04002625 journal->j_dev_mode = blkdev_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626 if (IS_ERR(journal->j_dev_bd)) {
2627 result = PTR_ERR(journal->j_dev_bd);
2628 journal->j_dev_bd = NULL;
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002629 reiserfs_warning(super, "sh-458",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002630 "cannot init journal device '%s': %i",
2631 __bdevname(jdev, b), result);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632 return result;
Tejun Heoe525fd82010-11-13 11:55:17 +01002633 } else if (jdev != super->s_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634 set_blocksize(journal->j_dev_bd, super->s_blocksize);
Christoph Hellwig86098fa2008-04-30 00:54:46 -07002635
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636 return 0;
2637 }
2638
Al Viroe5eb8ca2007-10-08 13:24:05 -04002639 journal->j_dev_mode = blkdev_mode;
Tejun Heod4d77622010-11-13 11:55:18 +01002640 journal->j_dev_bd = blkdev_get_by_path(jdev_name, blkdev_mode, journal);
Christoph Hellwig86098fa2008-04-30 00:54:46 -07002641 if (IS_ERR(journal->j_dev_bd)) {
2642 result = PTR_ERR(journal->j_dev_bd);
2643 journal->j_dev_bd = NULL;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002644 reiserfs_warning(super,
2645 "journal_init_dev: Cannot open '%s': %i",
2646 jdev_name, result);
Christoph Hellwig86098fa2008-04-30 00:54:46 -07002647 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648 }
Christoph Hellwig86098fa2008-04-30 00:54:46 -07002649
2650 set_blocksize(journal->j_dev_bd, super->s_blocksize);
2651 reiserfs_info(super,
2652 "journal_init_dev: journal device: %s\n",
2653 bdevname(journal->j_dev_bd, b));
2654 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655}
2656
Jeff Mahoney098297b2014-04-23 10:00:36 -04002657/*
Edward Shishkincf3d0b82007-10-16 23:30:30 -07002658 * When creating/tuning a file system user can assign some
2659 * journal params within boundaries which depend on the ratio
2660 * blocksize/standard_blocksize.
2661 *
2662 * For blocks >= standard_blocksize transaction size should
2663 * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more
2664 * then JOURNAL_TRANS_MAX_DEFAULT.
2665 *
2666 * For blocks < standard_blocksize these boundaries should be
2667 * decreased proportionally.
2668 */
2669#define REISERFS_STANDARD_BLKSIZE (4096)
2670
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002671static int check_advise_trans_params(struct super_block *sb,
Edward Shishkincf3d0b82007-10-16 23:30:30 -07002672 struct reiserfs_journal *journal)
2673{
2674 if (journal->j_trans_max) {
Jeff Mahoney098297b2014-04-23 10:00:36 -04002675 /* Non-default journal params. Do sanity check for them. */
Edward Shishkincf3d0b82007-10-16 23:30:30 -07002676 int ratio = 1;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002677 if (sb->s_blocksize < REISERFS_STANDARD_BLKSIZE)
2678 ratio = REISERFS_STANDARD_BLKSIZE / sb->s_blocksize;
Edward Shishkincf3d0b82007-10-16 23:30:30 -07002679
2680 if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio ||
2681 journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio ||
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002682 SB_ONDISK_JOURNAL_SIZE(sb) / journal->j_trans_max <
Edward Shishkincf3d0b82007-10-16 23:30:30 -07002683 JOURNAL_MIN_RATIO) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002684 reiserfs_warning(sb, "sh-462",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002685 "bad transaction max size (%u). "
2686 "FSCK?", journal->j_trans_max);
Edward Shishkincf3d0b82007-10-16 23:30:30 -07002687 return 1;
2688 }
2689 if (journal->j_max_batch != (journal->j_trans_max) *
2690 JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002691 reiserfs_warning(sb, "sh-463",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002692 "bad transaction max batch (%u). "
2693 "FSCK?", journal->j_max_batch);
Edward Shishkincf3d0b82007-10-16 23:30:30 -07002694 return 1;
2695 }
2696 } else {
Jeff Mahoney098297b2014-04-23 10:00:36 -04002697 /*
2698 * Default journal params.
2699 * The file system was created by old version
2700 * of mkreiserfs, so some fields contain zeros,
2701 * and we need to advise proper values for them
2702 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002703 if (sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) {
2704 reiserfs_warning(sb, "sh-464", "bad blocksize (%u)",
2705 sb->s_blocksize);
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002706 return 1;
2707 }
Edward Shishkincf3d0b82007-10-16 23:30:30 -07002708 journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2709 journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2710 journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2711 }
2712 return 0;
2713}
2714
Jeff Mahoney098297b2014-04-23 10:00:36 -04002715/* must be called once on fs mount. calls journal_read for you */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002716int journal_init(struct super_block *sb, const char *j_dev_name,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002717 int old_format, unsigned int commit_max_age)
2718{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002719 int num_cnodes = SB_ONDISK_JOURNAL_SIZE(sb) * 2;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002720 struct buffer_head *bhjh;
2721 struct reiserfs_super_block *rs;
2722 struct reiserfs_journal_header *jh;
2723 struct reiserfs_journal *journal;
2724 struct reiserfs_journal_list *jl;
2725 char b[BDEVNAME_SIZE];
Frederic Weisbecker98ea3f52009-12-29 21:51:15 +01002726 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002727
Joe Perches558feb02011-05-28 10:36:33 -07002728 journal = SB_JOURNAL(sb) = vzalloc(sizeof(struct reiserfs_journal));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002729 if (!journal) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002730 reiserfs_warning(sb, "journal-1256",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002731 "unable to get memory for journal structure");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002732 return 1;
2733 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002734 INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2735 INIT_LIST_HEAD(&journal->j_prealloc_list);
2736 INIT_LIST_HEAD(&journal->j_working_list);
2737 INIT_LIST_HEAD(&journal->j_journal_list);
2738 journal->j_persistent_trans = 0;
Frederic Weisbecker37c69b92012-01-10 15:11:09 -08002739 if (reiserfs_allocate_list_bitmaps(sb, journal->j_list_bitmap,
2740 reiserfs_bmap_count(sb)))
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002741 goto free_and_return;
Frederic Weisbecker98ea3f52009-12-29 21:51:15 +01002742
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002743 allocate_bitmap_nodes(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002745 /* reserved for journal area support */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002746 SB_JOURNAL_1st_RESERVED_BLOCK(sb) = (old_format ?
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002747 REISERFS_OLD_DISK_OFFSET_IN_BYTES
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002748 / sb->s_blocksize +
2749 reiserfs_bmap_count(sb) +
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002750 1 :
2751 REISERFS_DISK_OFFSET_IN_BYTES /
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002752 sb->s_blocksize + 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002753
Jeff Mahoney098297b2014-04-23 10:00:36 -04002754 /*
2755 * Sanity check to see is the standard journal fitting
2756 * within first bitmap (actual for small blocksizes)
2757 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002758 if (!SB_ONDISK_JOURNAL_DEVICE(sb) &&
2759 (SB_JOURNAL_1st_RESERVED_BLOCK(sb) +
2760 SB_ONDISK_JOURNAL_SIZE(sb) > sb->s_blocksize * 8)) {
2761 reiserfs_warning(sb, "journal-1393",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002762 "journal does not fit for area addressed "
2763 "by first of bitmap blocks. It starts at "
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002764 "%u and its size is %u. Block size %ld",
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002765 SB_JOURNAL_1st_RESERVED_BLOCK(sb),
2766 SB_ONDISK_JOURNAL_SIZE(sb),
2767 sb->s_blocksize);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002768 goto free_and_return;
2769 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002770
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002771 if (journal_init_dev(sb, journal, j_dev_name) != 0) {
2772 reiserfs_warning(sb, "sh-462",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002773 "unable to initialize jornal device");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002774 goto free_and_return;
2775 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002777 rs = SB_DISK_SUPER_BLOCK(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002778
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002779 /* read journal header */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002780 bhjh = journal_bread(sb,
2781 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2782 SB_ONDISK_JOURNAL_SIZE(sb));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002783 if (!bhjh) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002784 reiserfs_warning(sb, "sh-459",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002785 "unable to read journal header");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002786 goto free_and_return;
2787 }
2788 jh = (struct reiserfs_journal_header *)(bhjh->b_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002789
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002790 /* make sure that journal matches to the super block */
2791 if (is_reiserfs_jr(rs)
2792 && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2793 sb_jp_journal_magic(rs))) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002794 reiserfs_warning(sb, "sh-460",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002795 "journal header magic %x (device %s) does "
2796 "not match to magic found in super block %x",
2797 jh->jh_journal.jp_journal_magic,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002798 bdevname(journal->j_dev_bd, b),
2799 sb_jp_journal_magic(rs));
2800 brelse(bhjh);
2801 goto free_and_return;
2802 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002803
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002804 journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2805 journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2806 journal->j_max_commit_age =
2807 le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2808 journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002809
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002810 if (check_advise_trans_params(sb, journal) != 0)
Edward Shishkincf3d0b82007-10-16 23:30:30 -07002811 goto free_and_return;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002812 journal->j_default_max_commit_age = journal->j_max_commit_age;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002813
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002814 if (commit_max_age != 0) {
2815 journal->j_max_commit_age = commit_max_age;
2816 journal->j_max_trans_age = commit_max_age;
2817 }
2818
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002819 reiserfs_info(sb, "journal params: device %s, size %u, "
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002820 "journal first block %u, max trans len %u, max batch %u, "
2821 "max commit age %u, max trans age %u\n",
2822 bdevname(journal->j_dev_bd, b),
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002823 SB_ONDISK_JOURNAL_SIZE(sb),
2824 SB_ONDISK_JOURNAL_1st_BLOCK(sb),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002825 journal->j_trans_max,
2826 journal->j_max_batch,
2827 journal->j_max_commit_age, journal->j_max_trans_age);
2828
2829 brelse(bhjh);
2830
2831 journal->j_list_bitmap_index = 0;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002832 journal_list_init(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002833
2834 memset(journal->j_list_hash_table, 0,
2835 JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2836
2837 INIT_LIST_HEAD(&journal->j_dirty_buffers);
2838 spin_lock_init(&journal->j_dirty_buffers_lock);
2839
2840 journal->j_start = 0;
2841 journal->j_len = 0;
2842 journal->j_len_alloc = 0;
Jeff Mahoneya228bf82014-04-23 10:00:42 -04002843 atomic_set(&journal->j_wcount, 0);
2844 atomic_set(&journal->j_async_throttle, 0);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002845 journal->j_bcount = 0;
2846 journal->j_trans_start_time = 0;
2847 journal->j_last = NULL;
2848 journal->j_first = NULL;
Jeff Mahoneya228bf82014-04-23 10:00:42 -04002849 init_waitqueue_head(&journal->j_join_wait);
Jeff Mahoneyf68215c2008-07-25 01:46:38 -07002850 mutex_init(&journal->j_mutex);
Jeff Mahoneyafe70252008-07-25 01:46:39 -07002851 mutex_init(&journal->j_flush_mutex);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002852
2853 journal->j_trans_id = 10;
2854 journal->j_mount_id = 10;
2855 journal->j_state = 0;
Jeff Mahoneya228bf82014-04-23 10:00:42 -04002856 atomic_set(&journal->j_jlock, 0);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002857 journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2858 journal->j_cnode_free_orig = journal->j_cnode_free_list;
2859 journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2860 journal->j_cnode_used = 0;
2861 journal->j_must_wait = 0;
2862
Jeff Mahoney576f6d72005-11-29 19:34:39 -08002863 if (journal->j_cnode_free == 0) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002864 reiserfs_warning(sb, "journal-2004", "Journal cnode memory "
Jeff Mahoney576f6d72005-11-29 19:34:39 -08002865 "allocation failed (%ld bytes). Journal is "
2866 "too large for available memory. Usually "
2867 "this is due to a journal that is too large.",
2868 sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2869 goto free_and_return;
2870 }
2871
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002872 init_journal_hash(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002873 jl = journal->j_current_jl;
Frederic Weisbecker37c69b92012-01-10 15:11:09 -08002874
2875 /*
2876 * get_list_bitmap() may call flush_commit_list() which
2877 * requires the lock. Calling flush_commit_list() shouldn't happen
2878 * this early but I like to be paranoid.
2879 */
2880 reiserfs_write_lock(sb);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002881 jl->j_list_bitmap = get_list_bitmap(sb, jl);
Frederic Weisbecker37c69b92012-01-10 15:11:09 -08002882 reiserfs_write_unlock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002883 if (!jl->j_list_bitmap) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002884 reiserfs_warning(sb, "journal-2005",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002885 "get_list_bitmap failed for journal list 0");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002886 goto free_and_return;
2887 }
Frederic Weisbecker37c69b92012-01-10 15:11:09 -08002888
Frederic Weisbecker37c69b92012-01-10 15:11:09 -08002889 ret = journal_read(sb);
Frederic Weisbecker37c69b92012-01-10 15:11:09 -08002890 if (ret < 0) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002891 reiserfs_warning(sb, "reiserfs-2006",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002892 "Replay Failure, unable to mount");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002893 goto free_and_return;
2894 }
2895
David Howellsc4028952006-11-22 14:57:56 +00002896 INIT_DELAYED_WORK(&journal->j_work, flush_async_commits);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002897 journal->j_work_sb = sb;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002898 return 0;
Jeff Mahoneycf776a72014-04-23 10:00:41 -04002899free_and_return:
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04002900 free_journal_ram(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002901 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902}
2903
2904/*
Jeff Mahoney098297b2014-04-23 10:00:36 -04002905 * test for a polite end of the current transaction. Used by file_write,
2906 * and should be used by delete to make sure they don't write more than
2907 * can fit inside a single transaction
2908 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002909int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2910 int new_alloc)
2911{
2912 struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2913 time_t now = get_seconds();
2914 /* cannot restart while nested */
2915 BUG_ON(!th->t_trans_id);
2916 if (th->t_refcount > 1)
2917 return 0;
2918 if (journal->j_must_wait > 0 ||
2919 (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
Jeff Mahoneya228bf82014-04-23 10:00:42 -04002920 atomic_read(&journal->j_jlock) ||
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002921 (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2922 journal->j_cnode_free < (journal->j_trans_max * 3)) {
2923 return 1;
2924 }
Davidlohr Buesob18c1c62012-01-10 15:11:05 -08002925
Chris Mason6ae1ea42006-02-01 03:06:50 -08002926 journal->j_len_alloc += new_alloc;
2927 th->t_blocks_allocated += new_alloc ;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002928 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002929}
2930
Jeff Mahoney098297b2014-04-23 10:00:36 -04002931/* this must be called inside a transaction */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002932void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002933{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002934 struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2935 BUG_ON(!th->t_trans_id);
2936 journal->j_must_wait = 1;
2937 set_bit(J_WRITERS_BLOCKED, &journal->j_state);
2938 return;
2939}
2940
Jeff Mahoney098297b2014-04-23 10:00:36 -04002941/* this must be called without a transaction started */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002942void reiserfs_allow_writes(struct super_block *s)
2943{
2944 struct reiserfs_journal *journal = SB_JOURNAL(s);
2945 clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
2946 wake_up(&journal->j_join_wait);
2947}
2948
Jeff Mahoney098297b2014-04-23 10:00:36 -04002949/* this must be called without a transaction started */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002950void reiserfs_wait_on_write_block(struct super_block *s)
2951{
2952 struct reiserfs_journal *journal = SB_JOURNAL(s);
2953 wait_event(journal->j_join_wait,
2954 !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
2955}
2956
2957static void queue_log_writer(struct super_block *s)
2958{
2959 wait_queue_t wait;
2960 struct reiserfs_journal *journal = SB_JOURNAL(s);
2961 set_bit(J_WRITERS_QUEUED, &journal->j_state);
2962
2963 /*
2964 * we don't want to use wait_event here because
2965 * we only want to wait once.
2966 */
2967 init_waitqueue_entry(&wait, current);
2968 add_wait_queue(&journal->j_join_wait, &wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002969 set_current_state(TASK_UNINTERRUPTIBLE);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02002970 if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) {
Jeff Mahoney278f6672013-08-08 17:34:46 -04002971 int depth = reiserfs_write_unlock_nested(s);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002972 schedule();
Jeff Mahoney278f6672013-08-08 17:34:46 -04002973 reiserfs_write_lock_nested(s, depth);
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02002974 }
Milind Arun Choudhary5ab2f7e2007-05-08 00:30:51 -07002975 __set_current_state(TASK_RUNNING);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002976 remove_wait_queue(&journal->j_join_wait, &wait);
2977}
2978
2979static void wake_queued_writers(struct super_block *s)
2980{
2981 struct reiserfs_journal *journal = SB_JOURNAL(s);
2982 if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
2983 wake_up(&journal->j_join_wait);
2984}
2985
Jeff Mahoney600ed412009-03-30 14:02:17 -04002986static void let_transaction_grow(struct super_block *sb, unsigned int trans_id)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002987{
2988 struct reiserfs_journal *journal = SB_JOURNAL(sb);
2989 unsigned long bcount = journal->j_bcount;
2990 while (1) {
Jeff Mahoney278f6672013-08-08 17:34:46 -04002991 int depth;
2992
2993 depth = reiserfs_write_unlock_nested(sb);
Nishanth Aravamudan041e0e32005-09-10 00:27:23 -07002994 schedule_timeout_uninterruptible(1);
Jeff Mahoney278f6672013-08-08 17:34:46 -04002995 reiserfs_write_lock_nested(sb, depth);
2996
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002997 journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
2998 while ((atomic_read(&journal->j_wcount) > 0 ||
2999 atomic_read(&journal->j_jlock)) &&
3000 journal->j_trans_id == trans_id) {
3001 queue_log_writer(sb);
3002 }
3003 if (journal->j_trans_id != trans_id)
3004 break;
3005 if (bcount == journal->j_bcount)
3006 break;
3007 bcount = journal->j_bcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003008 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003009}
3010
Jeff Mahoney098297b2014-04-23 10:00:36 -04003011/*
3012 * join == true if you must join an existing transaction.
3013 * join == false if you can deal with waiting for others to finish
3014 *
3015 * this will block until the transaction is joinable. send the number of
3016 * blocks you expect to use in nblocks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003017*/
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003018static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003019 struct super_block *sb, unsigned long nblocks,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003020 int join)
3021{
3022 time_t now = get_seconds();
Jeff Mahoney600ed412009-03-30 14:02:17 -04003023 unsigned int old_trans_id;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003024 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003025 struct reiserfs_transaction_handle myth;
3026 int sched_count = 0;
3027 int retval;
Jeff Mahoney278f6672013-08-08 17:34:46 -04003028 int depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003029
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003030 reiserfs_check_lock_depth(sb, "journal_begin");
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003031 BUG_ON(nblocks > journal->j_trans_max);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003032
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003033 PROC_INFO_INC(sb, journal.journal_being);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003034 /* set here for journal_join */
3035 th->t_refcount = 1;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003036 th->t_super = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003037
Jeff Mahoneycf776a72014-04-23 10:00:41 -04003038relock:
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003039 lock_journal(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003040 if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003041 unlock_journal(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003042 retval = journal->j_errno;
3043 goto out_fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003044 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003045 journal->j_bcount++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003046
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003047 if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003048 unlock_journal(sb);
Jeff Mahoney278f6672013-08-08 17:34:46 -04003049 depth = reiserfs_write_unlock_nested(sb);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003050 reiserfs_wait_on_write_block(sb);
Jeff Mahoney278f6672013-08-08 17:34:46 -04003051 reiserfs_write_lock_nested(sb, depth);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003052 PROC_INFO_INC(sb, journal.journal_relock_writers);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003053 goto relock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003054 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003055 now = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003056
Jeff Mahoney098297b2014-04-23 10:00:36 -04003057 /*
3058 * if there is no room in the journal OR
3059 * if this transaction is too old, and we weren't called joinable,
3060 * wait for it to finish before beginning we don't sleep if there
3061 * aren't other writers
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003062 */
3063
3064 if ((!join && journal->j_must_wait > 0) ||
3065 (!join
3066 && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
3067 || (!join && atomic_read(&journal->j_wcount) > 0
3068 && journal->j_trans_start_time > 0
3069 && (now - journal->j_trans_start_time) >
3070 journal->j_max_trans_age) || (!join
3071 && atomic_read(&journal->j_jlock))
3072 || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
3073
3074 old_trans_id = journal->j_trans_id;
Jeff Mahoney098297b2014-04-23 10:00:36 -04003075 /* allow others to finish this transaction */
3076 unlock_journal(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003077
3078 if (!join && (journal->j_len_alloc + nblocks + 2) >=
3079 journal->j_max_batch &&
3080 ((journal->j_len + nblocks + 2) * 100) <
3081 (journal->j_len_alloc * 75)) {
3082 if (atomic_read(&journal->j_wcount) > 10) {
3083 sched_count++;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003084 queue_log_writer(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003085 goto relock;
3086 }
3087 }
Jeff Mahoney098297b2014-04-23 10:00:36 -04003088 /*
3089 * don't mess with joining the transaction if all we
3090 * have to do is wait for someone else to do a commit
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003091 */
3092 if (atomic_read(&journal->j_jlock)) {
3093 while (journal->j_trans_id == old_trans_id &&
3094 atomic_read(&journal->j_jlock)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003095 queue_log_writer(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003096 }
3097 goto relock;
3098 }
Jeff Mahoneyb491dd12014-04-23 10:00:40 -04003099 retval = journal_join(&myth, sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003100 if (retval)
3101 goto out_fail;
3102
3103 /* someone might have ended the transaction while we joined */
3104 if (old_trans_id != journal->j_trans_id) {
Jeff Mahoney58d85422014-04-23 10:00:38 -04003105 retval = do_journal_end(&myth, 0);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003106 } else {
Jeff Mahoney58d85422014-04-23 10:00:38 -04003107 retval = do_journal_end(&myth, COMMIT_NOW);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003108 }
3109
3110 if (retval)
3111 goto out_fail;
3112
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003113 PROC_INFO_INC(sb, journal.journal_relock_wcount);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003114 goto relock;
3115 }
3116 /* we are the first writer, set trans_id */
3117 if (journal->j_trans_start_time == 0) {
3118 journal->j_trans_start_time = get_seconds();
3119 }
Jeff Mahoneya228bf82014-04-23 10:00:42 -04003120 atomic_inc(&journal->j_wcount);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003121 journal->j_len_alloc += nblocks;
3122 th->t_blocks_logged = 0;
3123 th->t_blocks_allocated = nblocks;
3124 th->t_trans_id = journal->j_trans_id;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003125 unlock_journal(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003126 INIT_LIST_HEAD(&th->t_list);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003127 return 0;
3128
Jeff Mahoneycf776a72014-04-23 10:00:41 -04003129out_fail:
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003130 memset(th, 0, sizeof(*th));
Jeff Mahoney098297b2014-04-23 10:00:36 -04003131 /*
3132 * Re-set th->t_super, so we can properly keep track of how many
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003133 * persistent transactions there are. We need to do this so if this
Jeff Mahoney098297b2014-04-23 10:00:36 -04003134 * call is part of a failed restart_transaction, we can free it later
3135 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003136 th->t_super = sb;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003137 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003138}
3139
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003140struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3141 super_block
3142 *s,
3143 int nblocks)
3144{
3145 int ret;
3146 struct reiserfs_transaction_handle *th;
3147
Jeff Mahoney098297b2014-04-23 10:00:36 -04003148 /*
3149 * if we're nesting into an existing transaction. It will be
3150 * persistent on its own
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003151 */
3152 if (reiserfs_transaction_running(s)) {
3153 th = current->journal_info;
3154 th->t_refcount++;
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003155 BUG_ON(th->t_refcount < 2);
Jeff Mahoney098297b2014-04-23 10:00:36 -04003156
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003157 return th;
3158 }
Pekka Enbergd739b422006-02-01 03:06:43 -08003159 th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003160 if (!th)
3161 return NULL;
3162 ret = journal_begin(th, s, nblocks);
3163 if (ret) {
Pekka Enbergd739b422006-02-01 03:06:43 -08003164 kfree(th);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003165 return NULL;
3166 }
3167
3168 SB_JOURNAL(s)->j_persistent_trans++;
3169 return th;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003170}
3171
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003172int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3173{
3174 struct super_block *s = th->t_super;
3175 int ret = 0;
3176 if (th->t_trans_id)
Jeff Mahoney58d85422014-04-23 10:00:38 -04003177 ret = journal_end(th);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003178 else
3179 ret = -EIO;
3180 if (th->t_refcount == 0) {
3181 SB_JOURNAL(s)->j_persistent_trans--;
Pekka Enbergd739b422006-02-01 03:06:43 -08003182 kfree(th);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003183 }
3184 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003185}
3186
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003187static int journal_join(struct reiserfs_transaction_handle *th,
Jeff Mahoneyb491dd12014-04-23 10:00:40 -04003188 struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003189{
3190 struct reiserfs_transaction_handle *cur_th = current->journal_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003191
Jeff Mahoney098297b2014-04-23 10:00:36 -04003192 /*
3193 * this keeps do_journal_end from NULLing out the
3194 * current->journal_info pointer
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003195 */
3196 th->t_handle_save = cur_th;
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003197 BUG_ON(cur_th && cur_th->t_refcount > 1);
Jeff Mahoneyb491dd12014-04-23 10:00:40 -04003198 return do_journal_begin_r(th, sb, 1, JBEGIN_JOIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003199}
3200
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003201int journal_join_abort(struct reiserfs_transaction_handle *th,
Jeff Mahoneyb491dd12014-04-23 10:00:40 -04003202 struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003203{
3204 struct reiserfs_transaction_handle *cur_th = current->journal_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003205
Jeff Mahoney098297b2014-04-23 10:00:36 -04003206 /*
3207 * this keeps do_journal_end from NULLing out the
3208 * current->journal_info pointer
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003209 */
3210 th->t_handle_save = cur_th;
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003211 BUG_ON(cur_th && cur_th->t_refcount > 1);
Jeff Mahoneyb491dd12014-04-23 10:00:40 -04003212 return do_journal_begin_r(th, sb, 1, JBEGIN_ABORT);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003213}
3214
3215int journal_begin(struct reiserfs_transaction_handle *th,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003216 struct super_block *sb, unsigned long nblocks)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003217{
3218 struct reiserfs_transaction_handle *cur_th = current->journal_info;
3219 int ret;
3220
3221 th->t_handle_save = NULL;
3222 if (cur_th) {
3223 /* we are nesting into the current transaction */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003224 if (cur_th->t_super == sb) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003225 BUG_ON(!cur_th->t_refcount);
3226 cur_th->t_refcount++;
3227 memcpy(th, cur_th, sizeof(*th));
3228 if (th->t_refcount <= 1)
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003229 reiserfs_warning(sb, "reiserfs-2005",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04003230 "BAD: refcount <= 1, but "
3231 "journal_info != 0");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003232 return 0;
3233 } else {
Jeff Mahoney098297b2014-04-23 10:00:36 -04003234 /*
3235 * we've ended up with a handle from a different
3236 * filesystem. save it and restore on journal_end.
3237 * This should never really happen...
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003238 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003239 reiserfs_warning(sb, "clm-2100",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04003240 "nesting info a different FS");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003241 th->t_handle_save = current->journal_info;
3242 current->journal_info = th;
3243 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003244 } else {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003245 current->journal_info = th;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003246 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003247 ret = do_journal_begin_r(th, sb, nblocks, JBEGIN_REG);
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003248 BUG_ON(current->journal_info != th);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003249
Jeff Mahoney098297b2014-04-23 10:00:36 -04003250 /*
3251 * I guess this boils down to being the reciprocal of clm-2100 above.
3252 * If do_journal_begin_r fails, we need to put it back, since
3253 * journal_end won't be called to do it. */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003254 if (ret)
3255 current->journal_info = th->t_handle_save;
3256 else
3257 BUG_ON(!th->t_refcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003258
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003259 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003260}
3261
3262/*
Jeff Mahoney098297b2014-04-23 10:00:36 -04003263 * puts bh into the current transaction. If it was already there, reorders
3264 * removes the old pointers from the hash, and puts new ones in (to make
3265 * sure replay happen in the right order).
3266 *
3267 * if it was dirty, cleans and files onto the clean list. I can't let it
3268 * be dirty again until the transaction is committed.
3269 *
3270 * if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3271 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003272int journal_mark_dirty(struct reiserfs_transaction_handle *th,
Jeff Mahoney09f1b802014-04-23 10:00:39 -04003273 struct buffer_head *bh)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003274{
Jeff Mahoney09f1b802014-04-23 10:00:39 -04003275 struct super_block *sb = th->t_super;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003276 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003277 struct reiserfs_journal_cnode *cn = NULL;
3278 int count_already_incd = 0;
3279 int prepared = 0;
3280 BUG_ON(!th->t_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003281
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003282 PROC_INFO_INC(sb, journal.mark_dirty);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003283 if (th->t_trans_id != journal->j_trans_id) {
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04003284 reiserfs_panic(th->t_super, "journal-1577",
3285 "handle trans id %ld != current trans id %ld",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003286 th->t_trans_id, journal->j_trans_id);
3287 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003288
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003289 prepared = test_clear_buffer_journal_prepared(bh);
3290 clear_buffer_journal_restore_dirty(bh);
3291 /* already in this transaction, we are done */
3292 if (buffer_journaled(bh)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003293 PROC_INFO_INC(sb, journal.mark_dirty_already);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003294 return 0;
3295 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003296
Jeff Mahoney098297b2014-04-23 10:00:36 -04003297 /*
3298 * this must be turned into a panic instead of a warning. We can't
3299 * allow a dirty or journal_dirty or locked buffer to be logged, as
3300 * some changes could get to disk too early. NOT GOOD.
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003301 */
3302 if (!prepared || buffer_dirty(bh)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003303 reiserfs_warning(sb, "journal-1777",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04003304 "buffer %llu bad state "
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003305 "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3306 (unsigned long long)bh->b_blocknr,
3307 prepared ? ' ' : '!',
3308 buffer_locked(bh) ? ' ' : '!',
3309 buffer_dirty(bh) ? ' ' : '!',
3310 buffer_journal_dirty(bh) ? ' ' : '!');
3311 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003312
Jeff Mahoneya228bf82014-04-23 10:00:42 -04003313 if (atomic_read(&journal->j_wcount) <= 0) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003314 reiserfs_warning(sb, "journal-1409",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04003315 "returning because j_wcount was %d",
Jeff Mahoneya228bf82014-04-23 10:00:42 -04003316 atomic_read(&journal->j_wcount));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003317 return 1;
3318 }
Jeff Mahoney098297b2014-04-23 10:00:36 -04003319 /*
3320 * this error means I've screwed up, and we've overflowed
3321 * the transaction. Nothing can be done here, except make the
3322 * FS readonly or panic.
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003323 */
3324 if (journal->j_len >= journal->j_trans_max) {
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04003325 reiserfs_panic(th->t_super, "journal-1413",
3326 "j_len (%lu) is too big",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003327 journal->j_len);
3328 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003329
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003330 if (buffer_journal_dirty(bh)) {
3331 count_already_incd = 1;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003332 PROC_INFO_INC(sb, journal.mark_dirty_notjournal);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003333 clear_buffer_journal_dirty(bh);
3334 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003335
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003336 if (journal->j_len > journal->j_len_alloc) {
3337 journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3338 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003339
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003340 set_buffer_journaled(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003341
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003342 /* now put this guy on the end */
3343 if (!cn) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003344 cn = get_cnode(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003345 if (!cn) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003346 reiserfs_panic(sb, "journal-4", "get_cnode failed!");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003347 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003348
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003349 if (th->t_blocks_logged == th->t_blocks_allocated) {
3350 th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3351 journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3352 }
3353 th->t_blocks_logged++;
3354 journal->j_len++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003355
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003356 cn->bh = bh;
3357 cn->blocknr = bh->b_blocknr;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003358 cn->sb = sb;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003359 cn->jlist = NULL;
3360 insert_journal_hash(journal->j_hash_table, cn);
3361 if (!count_already_incd) {
3362 get_bh(bh);
3363 }
3364 }
3365 cn->next = NULL;
3366 cn->prev = journal->j_last;
3367 cn->bh = bh;
3368 if (journal->j_last) {
3369 journal->j_last->next = cn;
3370 journal->j_last = cn;
3371 } else {
3372 journal->j_first = cn;
3373 journal->j_last = cn;
3374 }
Artem Bityutskiy033369d2012-06-01 17:18:08 +03003375 reiserfs_schedule_old_flush(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003376 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003377}
3378
Jeff Mahoney58d85422014-04-23 10:00:38 -04003379int journal_end(struct reiserfs_transaction_handle *th)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003380{
Jeff Mahoney58d85422014-04-23 10:00:38 -04003381 struct super_block *sb = th->t_super;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003382 if (!current->journal_info && th->t_refcount > 1)
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003383 reiserfs_warning(sb, "REISER-NESTING",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04003384 "th NULL, refcount %d", th->t_refcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003385
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003386 if (!th->t_trans_id) {
3387 WARN_ON(1);
3388 return -EIO;
3389 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003390
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003391 th->t_refcount--;
3392 if (th->t_refcount > 0) {
3393 struct reiserfs_transaction_handle *cur_th =
3394 current->journal_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003395
Jeff Mahoney098297b2014-04-23 10:00:36 -04003396 /*
3397 * we aren't allowed to close a nested transaction on a
3398 * different filesystem from the one in the task struct
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003399 */
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003400 BUG_ON(cur_th->t_super != th->t_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003401
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003402 if (th != cur_th) {
3403 memcpy(current->journal_info, th, sizeof(*th));
3404 th->t_trans_id = 0;
3405 }
3406 return 0;
3407 } else {
Jeff Mahoney58d85422014-04-23 10:00:38 -04003408 return do_journal_end(th, 0);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003409 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003410}
3411
Jeff Mahoney098297b2014-04-23 10:00:36 -04003412/*
3413 * removes from the current transaction, relsing and descrementing any counters.
3414 * also files the removed buffer directly onto the clean list
3415 *
3416 * called by journal_mark_freed when a block has been deleted
3417 *
3418 * returns 1 if it cleaned and relsed the buffer. 0 otherwise
3419 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003420static int remove_from_transaction(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003421 b_blocknr_t blocknr, int already_cleaned)
3422{
3423 struct buffer_head *bh;
3424 struct reiserfs_journal_cnode *cn;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003425 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003426 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003427
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003428 cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003429 if (!cn || !cn->bh) {
3430 return ret;
3431 }
3432 bh = cn->bh;
3433 if (cn->prev) {
3434 cn->prev->next = cn->next;
3435 }
3436 if (cn->next) {
3437 cn->next->prev = cn->prev;
3438 }
3439 if (cn == journal->j_first) {
3440 journal->j_first = cn->next;
3441 }
3442 if (cn == journal->j_last) {
3443 journal->j_last = cn->prev;
3444 }
3445 if (bh)
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003446 remove_journal_hash(sb, journal->j_hash_table, NULL,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003447 bh->b_blocknr, 0);
3448 clear_buffer_journaled(bh); /* don't log this one */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003449
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003450 if (!already_cleaned) {
3451 clear_buffer_journal_dirty(bh);
3452 clear_buffer_dirty(bh);
3453 clear_buffer_journal_test(bh);
3454 put_bh(bh);
Jeff Mahoneya228bf82014-04-23 10:00:42 -04003455 if (atomic_read(&bh->b_count) < 0) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003456 reiserfs_warning(sb, "journal-1752",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04003457 "b_count < 0");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003458 }
3459 ret = 1;
3460 }
3461 journal->j_len--;
3462 journal->j_len_alloc--;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003463 free_cnode(sb, cn);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003464 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003465}
3466
3467/*
Jeff Mahoney098297b2014-04-23 10:00:36 -04003468 * for any cnode in a journal list, it can only be dirtied of all the
3469 * transactions that include it are committed to disk.
3470 * this checks through each transaction, and returns 1 if you are allowed
3471 * to dirty, and 0 if you aren't
3472 *
3473 * it is called by dirty_journal_list, which is called after
3474 * flush_commit_list has gotten all the log blocks for a given
3475 * transaction on disk
3476 *
3477 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003478static int can_dirty(struct reiserfs_journal_cnode *cn)
3479{
3480 struct super_block *sb = cn->sb;
3481 b_blocknr_t blocknr = cn->blocknr;
3482 struct reiserfs_journal_cnode *cur = cn->hprev;
3483 int can_dirty = 1;
3484
Jeff Mahoney098297b2014-04-23 10:00:36 -04003485 /*
3486 * first test hprev. These are all newer than cn, so any node here
3487 * with the same block number and dev means this node can't be sent
3488 * to disk right now.
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003489 */
3490 while (cur && can_dirty) {
3491 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3492 cur->blocknr == blocknr) {
3493 can_dirty = 0;
3494 }
3495 cur = cur->hprev;
3496 }
Jeff Mahoney098297b2014-04-23 10:00:36 -04003497 /*
3498 * then test hnext. These are all older than cn. As long as they
3499 * are committed to the log, it is safe to write cn to disk
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003500 */
3501 cur = cn->hnext;
3502 while (cur && can_dirty) {
3503 if (cur->jlist && cur->jlist->j_len > 0 &&
Jeff Mahoneya228bf82014-04-23 10:00:42 -04003504 atomic_read(&cur->jlist->j_commit_left) > 0 && cur->bh &&
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003505 cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3506 can_dirty = 0;
3507 }
3508 cur = cur->hnext;
3509 }
3510 return can_dirty;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003511}
3512
Jeff Mahoney098297b2014-04-23 10:00:36 -04003513/*
3514 * syncs the commit blocks, but does not force the real buffers to disk
3515 * will wait until the current transaction is done/committed before returning
3516 */
Jeff Mahoney58d85422014-04-23 10:00:38 -04003517int journal_end_sync(struct reiserfs_transaction_handle *th)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003518{
Jeff Mahoney58d85422014-04-23 10:00:38 -04003519 struct super_block *sb = th->t_super;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003520 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003521
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003522 BUG_ON(!th->t_trans_id);
3523 /* you can sync while nested, very, very bad */
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003524 BUG_ON(th->t_refcount > 1);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003525 if (journal->j_len == 0) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003526 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003527 1);
Jeff Mahoney09f1b802014-04-23 10:00:39 -04003528 journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003529 }
Jeff Mahoney58d85422014-04-23 10:00:38 -04003530 return do_journal_end(th, COMMIT_NOW | WAIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003531}
3532
Jeff Mahoney098297b2014-04-23 10:00:36 -04003533/* writeback the pending async commits to disk */
David Howellsc4028952006-11-22 14:57:56 +00003534static void flush_async_commits(struct work_struct *work)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003535{
David Howellsc4028952006-11-22 14:57:56 +00003536 struct reiserfs_journal *journal =
3537 container_of(work, struct reiserfs_journal, j_work.work);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003538 struct super_block *sb = journal->j_work_sb;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003539 struct reiserfs_journal_list *jl;
3540 struct list_head *entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003541
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02003542 reiserfs_write_lock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003543 if (!list_empty(&journal->j_journal_list)) {
3544 /* last entry is the youngest, commit it and you get everything */
3545 entry = journal->j_journal_list.prev;
3546 jl = JOURNAL_LIST_ENTRY(entry);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003547 flush_commit_list(sb, jl, 1);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003548 }
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02003549 reiserfs_write_unlock(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003550}
3551
3552/*
Jeff Mahoney098297b2014-04-23 10:00:36 -04003553 * flushes any old transactions to disk
3554 * ends the current transaction if it is too old
3555 */
Artem Bityutskiy25729b02012-06-01 17:18:05 +03003556void reiserfs_flush_old_commits(struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003557{
3558 time_t now;
3559 struct reiserfs_transaction_handle th;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003560 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003561
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003562 now = get_seconds();
Jeff Mahoney098297b2014-04-23 10:00:36 -04003563 /*
3564 * safety check so we don't flush while we are replaying the log during
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003565 * mount
3566 */
Artem Bityutskiy25729b02012-06-01 17:18:05 +03003567 if (list_empty(&journal->j_journal_list))
3568 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003569
Jeff Mahoney098297b2014-04-23 10:00:36 -04003570 /*
3571 * check the current transaction. If there are no writers, and it is
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003572 * too old, finish it, and force the commit blocks to disk
3573 */
3574 if (atomic_read(&journal->j_wcount) <= 0 &&
3575 journal->j_trans_start_time > 0 &&
3576 journal->j_len > 0 &&
3577 (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
Jeff Mahoneyb491dd12014-04-23 10:00:40 -04003578 if (!journal_join(&th, sb)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003579 reiserfs_prepare_for_journal(sb,
3580 SB_BUFFER_WITH_SB(sb),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003581 1);
Jeff Mahoney09f1b802014-04-23 10:00:39 -04003582 journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003583
Jeff Mahoney098297b2014-04-23 10:00:36 -04003584 /*
3585 * we're only being called from kreiserfsd, it makes
3586 * no sense to do an async commit so that kreiserfsd
3587 * can do it later
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003588 */
Jeff Mahoney58d85422014-04-23 10:00:38 -04003589 do_journal_end(&th, COMMIT_NOW | WAIT);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003590 }
3591 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003592}
3593
3594/*
Jeff Mahoney098297b2014-04-23 10:00:36 -04003595 * returns 0 if do_journal_end should return right away, returns 1 if
3596 * do_journal_end should finish the commit
3597 *
3598 * if the current transaction is too old, but still has writers, this will
3599 * wait on j_join_wait until all the writers are done. By the time it
3600 * wakes up, the transaction it was called has already ended, so it just
3601 * flushes the commit list and returns 0.
3602 *
3603 * Won't batch when flush or commit_now is set. Also won't batch when
3604 * others are waiting on j_join_wait.
3605 *
3606 * Note, we can't allow the journal_end to proceed while there are still
3607 * writers in the log.
3608 */
Jeff Mahoney58d85422014-04-23 10:00:38 -04003609static int check_journal_end(struct reiserfs_transaction_handle *th, int flags)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003610{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003611
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003612 time_t now;
3613 int flush = flags & FLUSH_ALL;
3614 int commit_now = flags & COMMIT_NOW;
3615 int wait_on_commit = flags & WAIT;
3616 struct reiserfs_journal_list *jl;
Jeff Mahoney58d85422014-04-23 10:00:38 -04003617 struct super_block *sb = th->t_super;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003618 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003619
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003620 BUG_ON(!th->t_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003621
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003622 if (th->t_trans_id != journal->j_trans_id) {
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04003623 reiserfs_panic(th->t_super, "journal-1577",
3624 "handle trans id %ld != current trans id %ld",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003625 th->t_trans_id, journal->j_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003626 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003627
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003628 journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
Jeff Mahoney098297b2014-04-23 10:00:36 -04003629 /* <= 0 is allowed. unmounting might not call begin */
Jeff Mahoneya228bf82014-04-23 10:00:42 -04003630 if (atomic_read(&journal->j_wcount) > 0)
3631 atomic_dec(&journal->j_wcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003632
Jeff Mahoney098297b2014-04-23 10:00:36 -04003633 /*
3634 * BUG, deal with case where j_len is 0, but people previously
3635 * freed blocks need to be released will be dealt with by next
3636 * transaction that actually writes something, but should be taken
3637 * care of in this trans
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003638 */
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003639 BUG_ON(journal->j_len == 0);
3640
Jeff Mahoney098297b2014-04-23 10:00:36 -04003641 /*
3642 * if wcount > 0, and we are called to with flush or commit_now,
3643 * we wait on j_join_wait. We will wake up when the last writer has
3644 * finished the transaction, and started it on its way to the disk.
3645 * Then, we flush the commit or journal list, and just return 0
3646 * because the rest of journal end was already done for this
3647 * transaction.
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003648 */
Jeff Mahoneya228bf82014-04-23 10:00:42 -04003649 if (atomic_read(&journal->j_wcount) > 0) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003650 if (flush || commit_now) {
3651 unsigned trans_id;
3652
3653 jl = journal->j_current_jl;
3654 trans_id = jl->j_trans_id;
3655 if (wait_on_commit)
3656 jl->j_state |= LIST_COMMIT_PENDING;
Jeff Mahoneya228bf82014-04-23 10:00:42 -04003657 atomic_set(&journal->j_jlock, 1);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003658 if (flush) {
3659 journal->j_next_full_flush = 1;
3660 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003661 unlock_journal(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003662
Jeff Mahoney098297b2014-04-23 10:00:36 -04003663 /*
3664 * sleep while the current transaction is
3665 * still j_jlocked
3666 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003667 while (journal->j_trans_id == trans_id) {
3668 if (atomic_read(&journal->j_jlock)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003669 queue_log_writer(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003670 } else {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003671 lock_journal(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003672 if (journal->j_trans_id == trans_id) {
Jeff Mahoneya228bf82014-04-23 10:00:42 -04003673 atomic_set(&journal->j_jlock,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003674 1);
3675 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003676 unlock_journal(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003677 }
3678 }
Eric Sesterhenn14a61442006-10-03 23:36:38 +02003679 BUG_ON(journal->j_trans_id == trans_id);
Jeff Mahoney098297b2014-04-23 10:00:36 -04003680
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003681 if (commit_now
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003682 && journal_list_still_alive(sb, trans_id)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003683 && wait_on_commit) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003684 flush_commit_list(sb, jl, 1);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003685 }
3686 return 0;
3687 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003688 unlock_journal(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003689 return 0;
3690 }
3691
3692 /* deal with old transactions where we are the last writers */
3693 now = get_seconds();
3694 if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3695 commit_now = 1;
3696 journal->j_next_async_flush = 1;
3697 }
3698 /* don't batch when someone is waiting on j_join_wait */
3699 /* don't batch when syncing the commit or flushing the whole trans */
Jeff Mahoneya228bf82014-04-23 10:00:42 -04003700 if (!(journal->j_must_wait > 0) && !(atomic_read(&journal->j_jlock))
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003701 && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3702 && journal->j_len_alloc < journal->j_max_batch
3703 && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3704 journal->j_bcount++;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003705 unlock_journal(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003706 return 0;
3707 }
3708
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003709 if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(sb)) {
3710 reiserfs_panic(sb, "journal-003",
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04003711 "j_start (%ld) is too high",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003712 journal->j_start);
3713 }
3714 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003715}
3716
3717/*
Jeff Mahoney098297b2014-04-23 10:00:36 -04003718 * Does all the work that makes deleting blocks safe.
3719 * when deleting a block mark BH_JNew, just remove it from the current
3720 * transaction, clean it's buffer_head and move on.
3721 *
3722 * otherwise:
3723 * set a bit for the block in the journal bitmap. That will prevent it from
3724 * being allocated for unformatted nodes before this transaction has finished.
3725 *
3726 * mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers.
3727 * That will prevent any old transactions with this block from trying to flush
3728 * to the real location. Since we aren't removing the cnode from the
3729 * journal_list_hash, *the block can't be reallocated yet.
3730 *
3731 * Then remove it from the current transaction, decrementing any counters and
3732 * filing it on the clean list.
3733 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003734int journal_mark_freed(struct reiserfs_transaction_handle *th,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003735 struct super_block *sb, b_blocknr_t blocknr)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003736{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003737 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003738 struct reiserfs_journal_cnode *cn = NULL;
3739 struct buffer_head *bh = NULL;
3740 struct reiserfs_list_bitmap *jb = NULL;
3741 int cleaned = 0;
3742 BUG_ON(!th->t_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003743
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003744 cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003745 if (cn && cn->bh) {
3746 bh = cn->bh;
3747 get_bh(bh);
3748 }
3749 /* if it is journal new, we just remove it from this transaction */
3750 if (bh && buffer_journal_new(bh)) {
3751 clear_buffer_journal_new(bh);
3752 clear_prepared_bits(bh);
3753 reiserfs_clean_and_file_buffer(bh);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003754 cleaned = remove_from_transaction(sb, blocknr, cleaned);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003755 } else {
Jeff Mahoney098297b2014-04-23 10:00:36 -04003756 /*
3757 * set the bit for this block in the journal bitmap
3758 * for this transaction
3759 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003760 jb = journal->j_current_jl->j_list_bitmap;
3761 if (!jb) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003762 reiserfs_panic(sb, "journal-1702",
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04003763 "journal_list_bitmap is NULL");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003764 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003765 set_bit_in_list_bitmap(sb, blocknr, jb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003766
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003767 /* Note, the entire while loop is not allowed to schedule. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003768
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003769 if (bh) {
3770 clear_prepared_bits(bh);
3771 reiserfs_clean_and_file_buffer(bh);
3772 }
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003773 cleaned = remove_from_transaction(sb, blocknr, cleaned);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003774
Jeff Mahoney098297b2014-04-23 10:00:36 -04003775 /*
3776 * find all older transactions with this block,
3777 * make sure they don't try to write it out
3778 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003779 cn = get_journal_hash_dev(sb, journal->j_list_hash_table,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003780 blocknr);
3781 while (cn) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003782 if (sb == cn->sb && blocknr == cn->blocknr) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003783 set_bit(BLOCK_FREED, &cn->state);
3784 if (cn->bh) {
Jeff Mahoney098297b2014-04-23 10:00:36 -04003785 /*
3786 * remove_from_transaction will brelse
3787 * the buffer if it was in the current
3788 * trans
3789 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003790 if (!cleaned) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003791 clear_buffer_journal_dirty(cn->
3792 bh);
3793 clear_buffer_dirty(cn->bh);
3794 clear_buffer_journal_test(cn->
3795 bh);
3796 cleaned = 1;
3797 put_bh(cn->bh);
3798 if (atomic_read
Jeff Mahoneya228bf82014-04-23 10:00:42 -04003799 (&cn->bh->b_count) < 0) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003800 reiserfs_warning(sb,
Jeff Mahoney45b03d52009-03-30 14:02:21 -04003801 "journal-2138",
3802 "cn->bh->b_count < 0");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003803 }
3804 }
Jeff Mahoney098297b2014-04-23 10:00:36 -04003805 /*
3806 * since we are clearing the bh,
3807 * we MUST dec nonzerolen
3808 */
3809 if (cn->jlist) {
Jeff Mahoneya228bf82014-04-23 10:00:42 -04003810 atomic_dec(&cn->jlist->
3811 j_nonzerolen);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003812 }
3813 cn->bh = NULL;
3814 }
3815 }
3816 cn = cn->hnext;
3817 }
3818 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003819
Chris Mason398c95b2007-10-16 23:29:44 -07003820 if (bh)
3821 release_buffer_page(bh); /* get_hash grabs the buffer */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003822 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003823}
3824
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003825void reiserfs_update_inode_transaction(struct inode *inode)
3826{
3827 struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3828 REISERFS_I(inode)->i_jl = journal->j_current_jl;
3829 REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003830}
3831
3832/*
3833 * returns -1 on error, 0 if no commits/barriers were done and 1
3834 * if a transaction was actually committed and the barrier was done
3835 */
3836static int __commit_trans_jl(struct inode *inode, unsigned long id,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003837 struct reiserfs_journal_list *jl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003838{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003839 struct reiserfs_transaction_handle th;
3840 struct super_block *sb = inode->i_sb;
3841 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3842 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003843
Jeff Mahoney098297b2014-04-23 10:00:36 -04003844 /*
3845 * is it from the current transaction,
3846 * or from an unknown transaction?
3847 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003848 if (id == journal->j_trans_id) {
3849 jl = journal->j_current_jl;
Jeff Mahoney098297b2014-04-23 10:00:36 -04003850 /*
3851 * try to let other writers come in and
3852 * grow this transaction
3853 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003854 let_transaction_grow(sb, id);
3855 if (journal->j_trans_id != id) {
3856 goto flush_commit_only;
3857 }
3858
3859 ret = journal_begin(&th, sb, 1);
3860 if (ret)
3861 return ret;
3862
3863 /* someone might have ended this transaction while we joined */
3864 if (journal->j_trans_id != id) {
3865 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3866 1);
Jeff Mahoney09f1b802014-04-23 10:00:39 -04003867 journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb));
Jeff Mahoney58d85422014-04-23 10:00:38 -04003868 ret = journal_end(&th);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003869 goto flush_commit_only;
3870 }
3871
Jeff Mahoney58d85422014-04-23 10:00:38 -04003872 ret = journal_end_sync(&th);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003873 if (!ret)
3874 ret = 1;
3875
3876 } else {
Jeff Mahoney098297b2014-04-23 10:00:36 -04003877 /*
3878 * this gets tricky, we have to make sure the journal list in
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003879 * the inode still exists. We know the list is still around
3880 * if we've got a larger transaction id than the oldest list
3881 */
Jeff Mahoneycf776a72014-04-23 10:00:41 -04003882flush_commit_only:
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003883 if (journal_list_still_alive(inode->i_sb, id)) {
3884 /*
3885 * we only set ret to 1 when we know for sure
3886 * the barrier hasn't been started yet on the commit
3887 * block.
3888 */
3889 if (atomic_read(&jl->j_commit_left) > 1)
3890 ret = 1;
3891 flush_commit_list(sb, jl, 1);
3892 if (journal->j_errno)
3893 ret = journal->j_errno;
3894 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003895 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003896 /* otherwise the list is gone, and long since committed */
3897 return ret;
3898}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003899
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003900int reiserfs_commit_for_inode(struct inode *inode)
3901{
Jeff Mahoney600ed412009-03-30 14:02:17 -04003902 unsigned int id = REISERFS_I(inode)->i_trans_id;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003903 struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003904
Jeff Mahoney098297b2014-04-23 10:00:36 -04003905 /*
3906 * for the whole inode, assume unset id means it was
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003907 * changed in the current transaction. More conservative
Linus Torvalds1da177e2005-04-16 15:20:36 -07003908 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003909 if (!id || !jl) {
3910 reiserfs_update_inode_transaction(inode);
3911 id = REISERFS_I(inode)->i_trans_id;
3912 /* jl will be updated in __commit_trans_jl */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003913 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003914
3915 return __commit_trans_jl(inode, id, jl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003916}
3917
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003918void reiserfs_restore_prepared_buffer(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003919 struct buffer_head *bh)
3920{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003921 struct reiserfs_journal *journal = SB_JOURNAL(sb);
3922 PROC_INFO_INC(sb, journal.restore_prepared);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003923 if (!bh) {
3924 return;
3925 }
3926 if (test_clear_buffer_journal_restore_dirty(bh) &&
3927 buffer_journal_dirty(bh)) {
3928 struct reiserfs_journal_cnode *cn;
Jeff Mahoney278f6672013-08-08 17:34:46 -04003929 reiserfs_write_lock(sb);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003930 cn = get_journal_hash_dev(sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003931 journal->j_list_hash_table,
3932 bh->b_blocknr);
3933 if (cn && can_dirty(cn)) {
3934 set_buffer_journal_test(bh);
3935 mark_buffer_dirty(bh);
3936 }
Jeff Mahoney278f6672013-08-08 17:34:46 -04003937 reiserfs_write_unlock(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003938 }
3939 clear_buffer_journal_prepared(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003940}
3941
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003942extern struct tree_balance *cur_tb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003943/*
Jeff Mahoney098297b2014-04-23 10:00:36 -04003944 * before we can change a metadata block, we have to make sure it won't
3945 * be written to disk while we are altering it. So, we must:
3946 * clean it
3947 * wait on it.
3948 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003949int reiserfs_prepare_for_journal(struct super_block *sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003950 struct buffer_head *bh, int wait)
3951{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003952 PROC_INFO_INC(sb, journal.prepare);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003953
Nick Pigginca5de402008-08-02 12:02:13 +02003954 if (!trylock_buffer(bh)) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003955 if (!wait)
3956 return 0;
3957 lock_buffer(bh);
3958 }
3959 set_buffer_journal_prepared(bh);
3960 if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3961 clear_buffer_journal_test(bh);
3962 set_buffer_journal_restore_dirty(bh);
3963 }
3964 unlock_buffer(bh);
3965 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003966}
3967
Jeff Mahoney0222e652009-03-30 14:02:44 -04003968/*
Jeff Mahoney098297b2014-04-23 10:00:36 -04003969 * long and ugly. If flush, will not return until all commit
3970 * blocks and all real buffers in the trans are on disk.
3971 * If no_async, won't return until all commit blocks are on disk.
3972 *
3973 * keep reading, there are comments as you go along
3974 *
3975 * If the journal is aborted, we just clean up. Things like flushing
3976 * journal lists, etc just won't happen.
3977 */
Jeff Mahoney58d85422014-04-23 10:00:38 -04003978static int do_journal_end(struct reiserfs_transaction_handle *th, int flags)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003979{
Jeff Mahoney58d85422014-04-23 10:00:38 -04003980 struct super_block *sb = th->t_super;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04003981 struct reiserfs_journal *journal = SB_JOURNAL(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003982 struct reiserfs_journal_cnode *cn, *next, *jl_cn;
3983 struct reiserfs_journal_cnode *last_cn = NULL;
3984 struct reiserfs_journal_desc *desc;
3985 struct reiserfs_journal_commit *commit;
3986 struct buffer_head *c_bh; /* commit bh */
3987 struct buffer_head *d_bh; /* desc bh */
3988 int cur_write_start = 0; /* start index of current log write */
3989 int old_start;
3990 int i;
Alexander Zarochentseva44c94a2006-03-25 03:06:59 -08003991 int flush;
3992 int wait_on_commit;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003993 struct reiserfs_journal_list *jl, *temp_jl;
3994 struct list_head *entry, *safe;
3995 unsigned long jindex;
Jeff Mahoney600ed412009-03-30 14:02:17 -04003996 unsigned int commit_trans_id;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07003997 int trans_half;
Jeff Mahoney278f6672013-08-08 17:34:46 -04003998 int depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003999
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004000 BUG_ON(th->t_refcount > 1);
4001 BUG_ON(!th->t_trans_id);
Jeff Mahoney58d85422014-04-23 10:00:38 -04004002 BUG_ON(!th->t_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004003
Jeff Mahoney098297b2014-04-23 10:00:36 -04004004 /*
4005 * protect flush_older_commits from doing mistakes if the
4006 * transaction ID counter gets overflowed.
4007 */
Jeff Mahoney600ed412009-03-30 14:02:17 -04004008 if (th->t_trans_id == ~0U)
Alexander Zarochentseva44c94a2006-03-25 03:06:59 -08004009 flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
4010 flush = flags & FLUSH_ALL;
4011 wait_on_commit = flags & WAIT;
4012
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004013 current->journal_info = th->t_handle_save;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004014 reiserfs_check_lock_depth(sb, "journal end");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004015 if (journal->j_len == 0) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004016 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004017 1);
Jeff Mahoney09f1b802014-04-23 10:00:39 -04004018 journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004019 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004020
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004021 lock_journal(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004022 if (journal->j_next_full_flush) {
4023 flags |= FLUSH_ALL;
4024 flush = 1;
4025 }
4026 if (journal->j_next_async_flush) {
4027 flags |= COMMIT_NOW | WAIT;
4028 wait_on_commit = 1;
4029 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004030
Jeff Mahoney098297b2014-04-23 10:00:36 -04004031 /*
4032 * check_journal_end locks the journal, and unlocks if it does
4033 * not return 1 it tells us if we should continue with the
4034 * journal_end, or just return
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004035 */
Jeff Mahoney58d85422014-04-23 10:00:38 -04004036 if (!check_journal_end(th, flags)) {
Artem Bityutskiy033369d2012-06-01 17:18:08 +03004037 reiserfs_schedule_old_flush(sb);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004038 wake_queued_writers(sb);
4039 reiserfs_async_progress_wait(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004040 goto out;
4041 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004042
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004043 /* check_journal_end might set these, check again */
4044 if (journal->j_next_full_flush) {
4045 flush = 1;
4046 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004047
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004048 /*
Jeff Mahoney098297b2014-04-23 10:00:36 -04004049 * j must wait means we have to flush the log blocks, and the
4050 * real blocks for this transaction
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004051 */
4052 if (journal->j_must_wait > 0) {
4053 flush = 1;
4054 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004055#ifdef REISERFS_PREALLOCATE
Jeff Mahoney098297b2014-04-23 10:00:36 -04004056 /*
4057 * quota ops might need to nest, setup the journal_info pointer
4058 * for them and raise the refcount so that it is > 0.
4059 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004060 current->journal_info = th;
Jan Karaef43bc42006-01-11 12:17:40 -08004061 th->t_refcount++;
Jeff Mahoney098297b2014-04-23 10:00:36 -04004062
4063 /* it should not involve new blocks into the transaction */
4064 reiserfs_discard_all_prealloc(th);
4065
Jan Karaef43bc42006-01-11 12:17:40 -08004066 th->t_refcount--;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004067 current->journal_info = th->t_handle_save;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004068#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004069
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004070 /* setup description block */
4071 d_bh =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004072 journal_getblk(sb,
4073 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004074 journal->j_start);
4075 set_buffer_uptodate(d_bh);
4076 desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
4077 memset(d_bh->b_data, 0, d_bh->b_size);
4078 memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
4079 set_desc_trans_id(desc, journal->j_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004080
Jeff Mahoney098297b2014-04-23 10:00:36 -04004081 /*
4082 * setup commit block. Don't write (keep it clean too) this one
4083 * until after everyone else is written
4084 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004085 c_bh = journal_getblk(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004086 ((journal->j_start + journal->j_len +
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004087 1) % SB_ONDISK_JOURNAL_SIZE(sb)));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004088 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
4089 memset(c_bh->b_data, 0, c_bh->b_size);
4090 set_commit_trans_id(commit, journal->j_trans_id);
4091 set_buffer_uptodate(c_bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004092
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004093 /* init this journal list */
4094 jl = journal->j_current_jl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004095
Jeff Mahoney098297b2014-04-23 10:00:36 -04004096 /*
4097 * we lock the commit before doing anything because
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004098 * we want to make sure nobody tries to run flush_commit_list until
4099 * the new transaction is fully setup, and we've already flushed the
4100 * ordered bh list
4101 */
Frederic Weisbecker8ebc4232009-04-07 04:19:49 +02004102 reiserfs_mutex_lock_safe(&jl->j_commit_mutex, sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004103
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004104 /* save the transaction id in case we need to commit it later */
4105 commit_trans_id = jl->j_trans_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004106
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004107 atomic_set(&jl->j_older_commits_done, 0);
4108 jl->j_trans_id = journal->j_trans_id;
4109 jl->j_timestamp = journal->j_trans_start_time;
4110 jl->j_commit_bh = c_bh;
4111 jl->j_start = journal->j_start;
4112 jl->j_len = journal->j_len;
4113 atomic_set(&jl->j_nonzerolen, journal->j_len);
4114 atomic_set(&jl->j_commit_left, journal->j_len + 2);
4115 jl->j_realblock = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004116
Jeff Mahoney098297b2014-04-23 10:00:36 -04004117 /*
4118 * The ENTIRE FOR LOOP MUST not cause schedule to occur.
4119 * for each real block, add it to the journal list hash,
4120 * copy into real block index array in the commit or desc block
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004121 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004122 trans_half = journal_trans_half(sb->s_blocksize);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004123 for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
4124 if (buffer_journaled(cn->bh)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004125 jl_cn = get_cnode(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004126 if (!jl_cn) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004127 reiserfs_panic(sb, "journal-1676",
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04004128 "get_cnode returned NULL");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004129 }
4130 if (i == 0) {
4131 jl->j_realblock = jl_cn;
4132 }
4133 jl_cn->prev = last_cn;
4134 jl_cn->next = NULL;
4135 if (last_cn) {
4136 last_cn->next = jl_cn;
4137 }
4138 last_cn = jl_cn;
Jeff Mahoney098297b2014-04-23 10:00:36 -04004139 /*
4140 * make sure the block we are trying to log
4141 * is not a block of journal or reserved area
4142 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004143 if (is_block_in_log_or_reserved_area
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004144 (sb, cn->bh->b_blocknr)) {
4145 reiserfs_panic(sb, "journal-2332",
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04004146 "Trying to log block %lu, "
4147 "which is a log block",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004148 cn->bh->b_blocknr);
4149 }
4150 jl_cn->blocknr = cn->bh->b_blocknr;
4151 jl_cn->state = 0;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004152 jl_cn->sb = sb;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004153 jl_cn->bh = cn->bh;
4154 jl_cn->jlist = jl;
4155 insert_journal_hash(journal->j_list_hash_table, jl_cn);
4156 if (i < trans_half) {
4157 desc->j_realblock[i] =
4158 cpu_to_le32(cn->bh->b_blocknr);
4159 } else {
4160 commit->j_realblock[i - trans_half] =
4161 cpu_to_le32(cn->bh->b_blocknr);
4162 }
4163 } else {
4164 i--;
4165 }
4166 }
4167 set_desc_trans_len(desc, journal->j_len);
4168 set_desc_mount_id(desc, journal->j_mount_id);
4169 set_desc_trans_id(desc, journal->j_trans_id);
4170 set_commit_trans_len(commit, journal->j_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004171
Jeff Mahoney098297b2014-04-23 10:00:36 -04004172 /*
4173 * special check in case all buffers in the journal
4174 * were marked for not logging
4175 */
Eric Sesterhenn14a61442006-10-03 23:36:38 +02004176 BUG_ON(journal->j_len == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004177
Jeff Mahoney098297b2014-04-23 10:00:36 -04004178 /*
4179 * we're about to dirty all the log blocks, mark the description block
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004180 * dirty now too. Don't mark the commit block dirty until all the
4181 * others are on disk
4182 */
4183 mark_buffer_dirty(d_bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004184
Jeff Mahoney098297b2014-04-23 10:00:36 -04004185 /*
4186 * first data block is j_start + 1, so add one to
4187 * cur_write_start wherever you use it
4188 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004189 cur_write_start = journal->j_start;
4190 cn = journal->j_first;
Jeff Mahoney098297b2014-04-23 10:00:36 -04004191 jindex = 1; /* start at one so we don't get the desc again */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004192 while (cn) {
4193 clear_buffer_journal_new(cn->bh);
4194 /* copy all the real blocks into log area. dirty log blocks */
4195 if (buffer_journaled(cn->bh)) {
4196 struct buffer_head *tmp_bh;
4197 char *addr;
4198 struct page *page;
4199 tmp_bh =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004200 journal_getblk(sb,
4201 SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004202 ((cur_write_start +
4203 jindex) %
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004204 SB_ONDISK_JOURNAL_SIZE(sb)));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004205 set_buffer_uptodate(tmp_bh);
4206 page = cn->bh->b_page;
4207 addr = kmap(page);
4208 memcpy(tmp_bh->b_data,
4209 addr + offset_in_page(cn->bh->b_data),
4210 cn->bh->b_size);
4211 kunmap(page);
4212 mark_buffer_dirty(tmp_bh);
4213 jindex++;
4214 set_buffer_journal_dirty(cn->bh);
4215 clear_buffer_journaled(cn->bh);
4216 } else {
Jeff Mahoney098297b2014-04-23 10:00:36 -04004217 /*
4218 * JDirty cleared sometime during transaction.
4219 * don't log this one
4220 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004221 reiserfs_warning(sb, "journal-2048",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04004222 "BAD, buffer in journal hash, "
4223 "but not JDirty!");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004224 brelse(cn->bh);
4225 }
4226 next = cn->next;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004227 free_cnode(sb, cn);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004228 cn = next;
Jeff Mahoney278f6672013-08-08 17:34:46 -04004229 reiserfs_cond_resched(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004230 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004231
Jeff Mahoney098297b2014-04-23 10:00:36 -04004232 /*
4233 * we are done with both the c_bh and d_bh, but
4234 * c_bh must be written after all other commit blocks,
4235 * so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004236 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004237
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004238 journal->j_current_jl = alloc_journal_list(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004239
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004240 /* now it is safe to insert this transaction on the main list */
4241 list_add_tail(&jl->j_list, &journal->j_journal_list);
4242 list_add_tail(&jl->j_working_list, &journal->j_working_list);
4243 journal->j_num_work_lists++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004244
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004245 /* reset journal values for the next transaction */
4246 old_start = journal->j_start;
4247 journal->j_start =
4248 (journal->j_start + journal->j_len +
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004249 2) % SB_ONDISK_JOURNAL_SIZE(sb);
Jeff Mahoneya228bf82014-04-23 10:00:42 -04004250 atomic_set(&journal->j_wcount, 0);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004251 journal->j_bcount = 0;
4252 journal->j_last = NULL;
4253 journal->j_first = NULL;
4254 journal->j_len = 0;
4255 journal->j_trans_start_time = 0;
Alexander Zarochentseva44c94a2006-03-25 03:06:59 -08004256 /* check for trans_id overflow */
4257 if (++journal->j_trans_id == 0)
4258 journal->j_trans_id = 10;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004259 journal->j_current_jl->j_trans_id = journal->j_trans_id;
4260 journal->j_must_wait = 0;
4261 journal->j_len_alloc = 0;
4262 journal->j_next_full_flush = 0;
4263 journal->j_next_async_flush = 0;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004264 init_journal_hash(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004265
Jeff Mahoney098297b2014-04-23 10:00:36 -04004266 /*
4267 * make sure reiserfs_add_jh sees the new current_jl before we
4268 * write out the tails
4269 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004270 smp_mb();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004271
Jeff Mahoney098297b2014-04-23 10:00:36 -04004272 /*
4273 * tail conversion targets have to hit the disk before we end the
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004274 * transaction. Otherwise a later transaction might repack the tail
Jeff Mahoney098297b2014-04-23 10:00:36 -04004275 * before this transaction commits, leaving the data block unflushed
4276 * and clean, if we crash before the later transaction commits, the
4277 * data block is lost.
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004278 */
4279 if (!list_empty(&jl->j_tail_bh_list)) {
Jeff Mahoney278f6672013-08-08 17:34:46 -04004280 depth = reiserfs_write_unlock_nested(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004281 write_ordered_buffers(&journal->j_dirty_buffers_lock,
4282 journal, jl, &jl->j_tail_bh_list);
Jeff Mahoney278f6672013-08-08 17:34:46 -04004283 reiserfs_write_lock_nested(sb, depth);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004284 }
Eric Sesterhenn14a61442006-10-03 23:36:38 +02004285 BUG_ON(!list_empty(&jl->j_tail_bh_list));
Jeff Mahoney90415de2008-07-25 01:46:40 -07004286 mutex_unlock(&jl->j_commit_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004287
Jeff Mahoney098297b2014-04-23 10:00:36 -04004288 /*
4289 * honor the flush wishes from the caller, simple commits can
4290 * be done outside the journal lock, they are done below
4291 *
4292 * if we don't flush the commit list right now, we put it into
4293 * the work queue so the people waiting on the async progress work
4294 * queue don't wait for this proc to flush journal lists and such.
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004295 */
4296 if (flush) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004297 flush_commit_list(sb, jl, 1);
4298 flush_journal_list(sb, jl, 1);
Jan Kara01777832014-08-06 19:43:56 +02004299 } else if (!(jl->j_state & LIST_COMMIT_PENDING)) {
4300 /*
4301 * Avoid queueing work when sb is being shut down. Transaction
4302 * will be flushed on journal shutdown.
4303 */
4304 if (sb->s_flags & MS_ACTIVE)
4305 queue_delayed_work(REISERFS_SB(sb)->commit_wq,
4306 &journal->j_work, HZ / 10);
4307 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004308
Jeff Mahoney098297b2014-04-23 10:00:36 -04004309 /*
4310 * if the next transaction has any chance of wrapping, flush
4311 * transactions that might get overwritten. If any journal lists
4312 * are very old flush them as well.
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004313 */
Jeff Mahoneycf776a72014-04-23 10:00:41 -04004314first_jl:
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004315 list_for_each_safe(entry, safe, &journal->j_journal_list) {
4316 temp_jl = JOURNAL_LIST_ENTRY(entry);
4317 if (journal->j_start <= temp_jl->j_start) {
4318 if ((journal->j_start + journal->j_trans_max + 1) >=
4319 temp_jl->j_start) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004320 flush_used_journal_lists(sb, temp_jl);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004321 goto first_jl;
4322 } else if ((journal->j_start +
4323 journal->j_trans_max + 1) <
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004324 SB_ONDISK_JOURNAL_SIZE(sb)) {
Jeff Mahoney098297b2014-04-23 10:00:36 -04004325 /*
4326 * if we don't cross into the next
4327 * transaction and we don't wrap, there is
4328 * no way we can overlap any later transactions
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004329 * break now
4330 */
4331 break;
4332 }
4333 } else if ((journal->j_start +
4334 journal->j_trans_max + 1) >
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004335 SB_ONDISK_JOURNAL_SIZE(sb)) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004336 if (((journal->j_start + journal->j_trans_max + 1) %
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004337 SB_ONDISK_JOURNAL_SIZE(sb)) >=
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004338 temp_jl->j_start) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004339 flush_used_journal_lists(sb, temp_jl);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004340 goto first_jl;
4341 } else {
Jeff Mahoney098297b2014-04-23 10:00:36 -04004342 /*
4343 * we don't overlap anything from out start
4344 * to the end of the log, and our wrapped
4345 * portion doesn't overlap anything at
4346 * the start of the log. We can break
4347 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004348 break;
4349 }
4350 }
4351 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004352
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004353 journal->j_current_jl->j_list_bitmap =
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004354 get_list_bitmap(sb, journal->j_current_jl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004355
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004356 if (!(journal->j_current_jl->j_list_bitmap)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004357 reiserfs_panic(sb, "journal-1996",
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04004358 "could not get a list bitmap");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004359 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004360
Jeff Mahoneya228bf82014-04-23 10:00:42 -04004361 atomic_set(&journal->j_jlock, 0);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004362 unlock_journal(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004363 /* wake up any body waiting to join. */
4364 clear_bit(J_WRITERS_QUEUED, &journal->j_state);
Jeff Mahoneya228bf82014-04-23 10:00:42 -04004365 wake_up(&journal->j_join_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004366
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004367 if (!flush && wait_on_commit &&
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004368 journal_list_still_alive(sb, commit_trans_id)) {
4369 flush_commit_list(sb, jl, 1);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004370 }
Jeff Mahoneycf776a72014-04-23 10:00:41 -04004371out:
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004372 reiserfs_check_lock_depth(sb, "journal end2");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004373
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004374 memset(th, 0, sizeof(*th));
Jeff Mahoney098297b2014-04-23 10:00:36 -04004375 /*
4376 * Re-set th->t_super, so we can properly keep track of how many
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004377 * persistent transactions there are. We need to do this so if this
Jeff Mahoney098297b2014-04-23 10:00:36 -04004378 * call is part of a failed restart_transaction, we can free it later
4379 */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04004380 th->t_super = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004381
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004382 return journal->j_errno;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004383}
4384
Jeff Mahoney32e8b102009-03-30 14:02:26 -04004385/* Send the file system read only and refuse new transactions */
4386void reiserfs_abort_journal(struct super_block *sb, int errno)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004387{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004388 struct reiserfs_journal *journal = SB_JOURNAL(sb);
4389 if (test_bit(J_ABORTED, &journal->j_state))
4390 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004391
Jeff Mahoney32e8b102009-03-30 14:02:26 -04004392 if (!journal->j_errno)
4393 journal->j_errno = errno;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004394
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004395 sb->s_flags |= MS_RDONLY;
4396 set_bit(J_ABORTED, &journal->j_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004397
4398#ifdef CONFIG_REISERFS_CHECK
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07004399 dump_stack();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004400#endif
4401}