11da177e4SLinus Torvalds /* 2098297b2SJeff Mahoney * Write ahead logging implementation copyright Chris Mason 2000 3098297b2SJeff Mahoney * 4098297b2SJeff Mahoney * The background commits make this code very interrelated, and 5098297b2SJeff Mahoney * overly complex. I need to rethink things a bit....The major players: 6098297b2SJeff Mahoney * 7098297b2SJeff Mahoney * journal_begin -- call with the number of blocks you expect to log. 8098297b2SJeff Mahoney * If the current transaction is too 9098297b2SJeff Mahoney * old, it will block until the current transaction is 10098297b2SJeff Mahoney * finished, and then start a new one. 11098297b2SJeff Mahoney * Usually, your transaction will get joined in with 12098297b2SJeff Mahoney * previous ones for speed. 13098297b2SJeff Mahoney * 14098297b2SJeff Mahoney * journal_join -- same as journal_begin, but won't block on the current 15098297b2SJeff Mahoney * transaction regardless of age. Don't ever call 16098297b2SJeff Mahoney * this. Ever. There are only two places it should be 17098297b2SJeff Mahoney * called from, and they are both inside this file. 18098297b2SJeff Mahoney * 19098297b2SJeff Mahoney * journal_mark_dirty -- adds blocks into this transaction. clears any flags 20098297b2SJeff Mahoney * that might make them get sent to disk 21098297b2SJeff Mahoney * and then marks them BH_JDirty. Puts the buffer head 22098297b2SJeff Mahoney * into the current transaction hash. 23098297b2SJeff Mahoney * 24098297b2SJeff Mahoney * journal_end -- if the current transaction is batchable, it does nothing 25098297b2SJeff Mahoney * otherwise, it could do an async/synchronous commit, or 26098297b2SJeff Mahoney * a full flush of all log and real blocks in the 27098297b2SJeff Mahoney * transaction. 28098297b2SJeff Mahoney * 29098297b2SJeff Mahoney * flush_old_commits -- if the current transaction is too old, it is ended and 30098297b2SJeff Mahoney * commit blocks are sent to disk. Forces commit blocks 31098297b2SJeff Mahoney * to disk for all backgrounded commits that have been 32098297b2SJeff Mahoney * around too long. 33098297b2SJeff Mahoney * -- Note, if you call this as an immediate flush from 34098297b2SJeff Mahoney * from within kupdate, it will ignore the immediate flag 351da177e4SLinus Torvalds */ 361da177e4SLinus Torvalds 371da177e4SLinus Torvalds #include <linux/time.h> 386188e10dSMatthew Wilcox #include <linux/semaphore.h> 391da177e4SLinus Torvalds #include <linux/vmalloc.h> 40f466c6fdSAl Viro #include "reiserfs.h" 411da177e4SLinus Torvalds #include <linux/kernel.h> 421da177e4SLinus Torvalds #include <linux/errno.h> 431da177e4SLinus Torvalds #include <linux/fcntl.h> 441da177e4SLinus Torvalds #include <linux/stat.h> 451da177e4SLinus Torvalds #include <linux/string.h> 461da177e4SLinus Torvalds #include <linux/buffer_head.h> 471da177e4SLinus Torvalds #include <linux/workqueue.h> 481da177e4SLinus Torvalds #include <linux/writeback.h> 491da177e4SLinus Torvalds #include <linux/blkdev.h> 503fcfab16SAndrew Morton #include <linux/backing-dev.h> 5190415deaSJeff Mahoney #include <linux/uaccess.h> 525a0e3ad6STejun Heo #include <linux/slab.h> 5390415deaSJeff Mahoney 541da177e4SLinus Torvalds 551da177e4SLinus Torvalds /* gets a struct reiserfs_journal_list * from a list head */ 561da177e4SLinus Torvalds #define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \ 571da177e4SLinus Torvalds j_list)) 581da177e4SLinus Torvalds #define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \ 591da177e4SLinus Torvalds j_working_list)) 601da177e4SLinus Torvalds 61098297b2SJeff Mahoney /* must be correct to keep the desc and commit structs at 4k */ 62098297b2SJeff Mahoney #define JOURNAL_TRANS_HALF 1018 631da177e4SLinus Torvalds #define BUFNR 64 /*read ahead */ 641da177e4SLinus Torvalds 651da177e4SLinus Torvalds /* cnode stat bits. Move these into reiserfs_fs.h */ 661da177e4SLinus Torvalds 67098297b2SJeff Mahoney /* this block was freed, and can't be written. */ 68098297b2SJeff Mahoney #define BLOCK_FREED 2 69098297b2SJeff Mahoney /* this block was freed during this transaction, and can't be written */ 70098297b2SJeff Mahoney #define BLOCK_FREED_HOLDER 3 711da177e4SLinus Torvalds 72098297b2SJeff Mahoney /* used in flush_journal_list */ 73098297b2SJeff Mahoney #define BLOCK_NEEDS_FLUSH 4 741da177e4SLinus Torvalds #define BLOCK_DIRTIED 5 751da177e4SLinus Torvalds 761da177e4SLinus Torvalds /* journal list state bits */ 771da177e4SLinus Torvalds #define LIST_TOUCHED 1 781da177e4SLinus Torvalds #define LIST_DIRTY 2 791da177e4SLinus Torvalds #define LIST_COMMIT_PENDING 4 /* someone will commit this list */ 801da177e4SLinus Torvalds 811da177e4SLinus Torvalds /* flags for do_journal_end */ 821da177e4SLinus Torvalds #define FLUSH_ALL 1 /* flush commit and real blocks */ 831da177e4SLinus Torvalds #define COMMIT_NOW 2 /* end and commit this transaction */ 841da177e4SLinus Torvalds #define WAIT 4 /* wait for the log blocks to hit the disk */ 851da177e4SLinus Torvalds 8658d85426SJeff Mahoney static int do_journal_end(struct reiserfs_transaction_handle *, int flags); 87bd4c625cSLinus Torvalds static int flush_journal_list(struct super_block *s, 88bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl, int flushall); 89bd4c625cSLinus Torvalds static int flush_commit_list(struct super_block *s, 90bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl, int flushall); 911da177e4SLinus Torvalds static int can_dirty(struct reiserfs_journal_cnode *cn); 92bd4c625cSLinus Torvalds static int journal_join(struct reiserfs_transaction_handle *th, 93b491dd17SJeff Mahoney struct super_block *sb); 944385bab1SAl Viro static void release_journal_dev(struct super_block *super, 951da177e4SLinus Torvalds struct reiserfs_journal *journal); 961da177e4SLinus Torvalds static int dirty_one_transaction(struct super_block *s, 971da177e4SLinus Torvalds struct reiserfs_journal_list *jl); 98c4028958SDavid Howells static void flush_async_commits(struct work_struct *work); 991da177e4SLinus Torvalds static void queue_log_writer(struct super_block *s); 1001da177e4SLinus Torvalds 1011da177e4SLinus Torvalds /* values for join in do_journal_begin_r */ 1021da177e4SLinus Torvalds enum { 1031da177e4SLinus Torvalds JBEGIN_REG = 0, /* regular journal begin */ 104098297b2SJeff Mahoney /* join the running transaction if at all possible */ 105098297b2SJeff Mahoney JBEGIN_JOIN = 1, 106098297b2SJeff Mahoney /* called from cleanup code, ignores aborted flag */ 107098297b2SJeff Mahoney JBEGIN_ABORT = 2, 1081da177e4SLinus Torvalds }; 1091da177e4SLinus Torvalds 1101da177e4SLinus Torvalds static int do_journal_begin_r(struct reiserfs_transaction_handle *th, 111a9dd3643SJeff Mahoney struct super_block *sb, 1121da177e4SLinus Torvalds unsigned long nblocks, int join); 1131da177e4SLinus Torvalds 114a9dd3643SJeff Mahoney static void init_journal_hash(struct super_block *sb) 115bd4c625cSLinus Torvalds { 116a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 117bd4c625cSLinus Torvalds memset(journal->j_hash_table, 0, 118bd4c625cSLinus Torvalds JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *)); 1191da177e4SLinus Torvalds } 1201da177e4SLinus Torvalds 1211da177e4SLinus Torvalds /* 122098297b2SJeff Mahoney * clears BH_Dirty and sticks the buffer on the clean list. Called because 123098297b2SJeff Mahoney * I can't allow refile_buffer to make schedule happen after I've freed a 124098297b2SJeff Mahoney * block. Look at remove_from_transaction and journal_mark_freed for 125098297b2SJeff Mahoney * more details. 1261da177e4SLinus Torvalds */ 127bd4c625cSLinus Torvalds static int reiserfs_clean_and_file_buffer(struct buffer_head *bh) 128bd4c625cSLinus Torvalds { 1291da177e4SLinus Torvalds if (bh) { 1301da177e4SLinus Torvalds clear_buffer_dirty(bh); 1311da177e4SLinus Torvalds clear_buffer_journal_test(bh); 1321da177e4SLinus Torvalds } 1331da177e4SLinus Torvalds return 0; 1341da177e4SLinus Torvalds } 1351da177e4SLinus Torvalds 136bd4c625cSLinus Torvalds static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block 137a9dd3643SJeff Mahoney *sb) 138bd4c625cSLinus Torvalds { 1391da177e4SLinus Torvalds struct reiserfs_bitmap_node *bn; 1401da177e4SLinus Torvalds static int id; 1411da177e4SLinus Torvalds 142d739b42bSPekka Enberg bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS); 1431da177e4SLinus Torvalds if (!bn) { 1441da177e4SLinus Torvalds return NULL; 1451da177e4SLinus Torvalds } 146a9dd3643SJeff Mahoney bn->data = kzalloc(sb->s_blocksize, GFP_NOFS); 1471da177e4SLinus Torvalds if (!bn->data) { 148d739b42bSPekka Enberg kfree(bn); 1491da177e4SLinus Torvalds return NULL; 1501da177e4SLinus Torvalds } 1511da177e4SLinus Torvalds bn->id = id++; 1521da177e4SLinus Torvalds INIT_LIST_HEAD(&bn->list); 1531da177e4SLinus Torvalds return bn; 1541da177e4SLinus Torvalds } 1551da177e4SLinus Torvalds 156a9dd3643SJeff Mahoney static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *sb) 157bd4c625cSLinus Torvalds { 158a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 1591da177e4SLinus Torvalds struct reiserfs_bitmap_node *bn = NULL; 1601da177e4SLinus Torvalds struct list_head *entry = journal->j_bitmap_nodes.next; 1611da177e4SLinus Torvalds 1621da177e4SLinus Torvalds journal->j_used_bitmap_nodes++; 1631da177e4SLinus Torvalds repeat: 1641da177e4SLinus Torvalds 1651da177e4SLinus Torvalds if (entry != &journal->j_bitmap_nodes) { 1661da177e4SLinus Torvalds bn = list_entry(entry, struct reiserfs_bitmap_node, list); 1671da177e4SLinus Torvalds list_del(entry); 168a9dd3643SJeff Mahoney memset(bn->data, 0, sb->s_blocksize); 1691da177e4SLinus Torvalds journal->j_free_bitmap_nodes--; 1701da177e4SLinus Torvalds return bn; 1711da177e4SLinus Torvalds } 172a9dd3643SJeff Mahoney bn = allocate_bitmap_node(sb); 1731da177e4SLinus Torvalds if (!bn) { 1741da177e4SLinus Torvalds yield(); 1751da177e4SLinus Torvalds goto repeat; 1761da177e4SLinus Torvalds } 1771da177e4SLinus Torvalds return bn; 1781da177e4SLinus Torvalds } 179a9dd3643SJeff Mahoney static inline void free_bitmap_node(struct super_block *sb, 180bd4c625cSLinus Torvalds struct reiserfs_bitmap_node *bn) 181bd4c625cSLinus Torvalds { 182a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 1831da177e4SLinus Torvalds journal->j_used_bitmap_nodes--; 1841da177e4SLinus Torvalds if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) { 185d739b42bSPekka Enberg kfree(bn->data); 186d739b42bSPekka Enberg kfree(bn); 1871da177e4SLinus Torvalds } else { 1881da177e4SLinus Torvalds list_add(&bn->list, &journal->j_bitmap_nodes); 1891da177e4SLinus Torvalds journal->j_free_bitmap_nodes++; 1901da177e4SLinus Torvalds } 1911da177e4SLinus Torvalds } 1921da177e4SLinus Torvalds 193a9dd3643SJeff Mahoney static void allocate_bitmap_nodes(struct super_block *sb) 194bd4c625cSLinus Torvalds { 1951da177e4SLinus Torvalds int i; 196a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 1971da177e4SLinus Torvalds struct reiserfs_bitmap_node *bn = NULL; 1981da177e4SLinus Torvalds for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) { 199a9dd3643SJeff Mahoney bn = allocate_bitmap_node(sb); 2001da177e4SLinus Torvalds if (bn) { 2011da177e4SLinus Torvalds list_add(&bn->list, &journal->j_bitmap_nodes); 2021da177e4SLinus Torvalds journal->j_free_bitmap_nodes++; 2031da177e4SLinus Torvalds } else { 204098297b2SJeff Mahoney /* this is ok, we'll try again when more are needed */ 205098297b2SJeff Mahoney break; 2061da177e4SLinus Torvalds } 2071da177e4SLinus Torvalds } 2081da177e4SLinus Torvalds } 2091da177e4SLinus Torvalds 210a9dd3643SJeff Mahoney static int set_bit_in_list_bitmap(struct super_block *sb, 2113ee16670SJeff Mahoney b_blocknr_t block, 212bd4c625cSLinus Torvalds struct reiserfs_list_bitmap *jb) 213bd4c625cSLinus Torvalds { 214a9dd3643SJeff Mahoney unsigned int bmap_nr = block / (sb->s_blocksize << 3); 215a9dd3643SJeff Mahoney unsigned int bit_nr = block % (sb->s_blocksize << 3); 2161da177e4SLinus Torvalds 2171da177e4SLinus Torvalds if (!jb->bitmaps[bmap_nr]) { 218a9dd3643SJeff Mahoney jb->bitmaps[bmap_nr] = get_bitmap_node(sb); 2191da177e4SLinus Torvalds } 2201da177e4SLinus Torvalds set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data); 2211da177e4SLinus Torvalds return 0; 2221da177e4SLinus Torvalds } 2231da177e4SLinus Torvalds 224a9dd3643SJeff Mahoney static void cleanup_bitmap_list(struct super_block *sb, 225bd4c625cSLinus Torvalds struct reiserfs_list_bitmap *jb) 226bd4c625cSLinus Torvalds { 2271da177e4SLinus Torvalds int i; 2281da177e4SLinus Torvalds if (jb->bitmaps == NULL) 2291da177e4SLinus Torvalds return; 2301da177e4SLinus Torvalds 231a9dd3643SJeff Mahoney for (i = 0; i < reiserfs_bmap_count(sb); i++) { 2321da177e4SLinus Torvalds if (jb->bitmaps[i]) { 233a9dd3643SJeff Mahoney free_bitmap_node(sb, jb->bitmaps[i]); 2341da177e4SLinus Torvalds jb->bitmaps[i] = NULL; 2351da177e4SLinus Torvalds } 2361da177e4SLinus Torvalds } 2371da177e4SLinus Torvalds } 2381da177e4SLinus Torvalds 2391da177e4SLinus Torvalds /* 240098297b2SJeff Mahoney * only call this on FS unmount. 2411da177e4SLinus Torvalds */ 242a9dd3643SJeff Mahoney static int free_list_bitmaps(struct super_block *sb, 243bd4c625cSLinus Torvalds struct reiserfs_list_bitmap *jb_array) 244bd4c625cSLinus Torvalds { 2451da177e4SLinus Torvalds int i; 2461da177e4SLinus Torvalds struct reiserfs_list_bitmap *jb; 2471da177e4SLinus Torvalds for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) { 2481da177e4SLinus Torvalds jb = jb_array + i; 2491da177e4SLinus Torvalds jb->journal_list = NULL; 250a9dd3643SJeff Mahoney cleanup_bitmap_list(sb, jb); 2511da177e4SLinus Torvalds vfree(jb->bitmaps); 2521da177e4SLinus Torvalds jb->bitmaps = NULL; 2531da177e4SLinus Torvalds } 2541da177e4SLinus Torvalds return 0; 2551da177e4SLinus Torvalds } 2561da177e4SLinus Torvalds 257a9dd3643SJeff Mahoney static int free_bitmap_nodes(struct super_block *sb) 258bd4c625cSLinus Torvalds { 259a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 2601da177e4SLinus Torvalds struct list_head *next = journal->j_bitmap_nodes.next; 2611da177e4SLinus Torvalds struct reiserfs_bitmap_node *bn; 2621da177e4SLinus Torvalds 2631da177e4SLinus Torvalds while (next != &journal->j_bitmap_nodes) { 2641da177e4SLinus Torvalds bn = list_entry(next, struct reiserfs_bitmap_node, list); 2651da177e4SLinus Torvalds list_del(next); 266d739b42bSPekka Enberg kfree(bn->data); 267d739b42bSPekka Enberg kfree(bn); 2681da177e4SLinus Torvalds next = journal->j_bitmap_nodes.next; 2691da177e4SLinus Torvalds journal->j_free_bitmap_nodes--; 2701da177e4SLinus Torvalds } 2711da177e4SLinus Torvalds 2721da177e4SLinus Torvalds return 0; 2731da177e4SLinus Torvalds } 2741da177e4SLinus Torvalds 2751da177e4SLinus Torvalds /* 276098297b2SJeff Mahoney * get memory for JOURNAL_NUM_BITMAPS worth of bitmaps. 277098297b2SJeff Mahoney * jb_array is the array to be filled in. 2781da177e4SLinus Torvalds */ 279a9dd3643SJeff Mahoney int reiserfs_allocate_list_bitmaps(struct super_block *sb, 2801da177e4SLinus Torvalds struct reiserfs_list_bitmap *jb_array, 2813ee16670SJeff Mahoney unsigned int bmap_nr) 282bd4c625cSLinus Torvalds { 2831da177e4SLinus Torvalds int i; 2841da177e4SLinus Torvalds int failed = 0; 2851da177e4SLinus Torvalds struct reiserfs_list_bitmap *jb; 2861da177e4SLinus Torvalds int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *); 2871da177e4SLinus Torvalds 2881da177e4SLinus Torvalds for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) { 2891da177e4SLinus Torvalds jb = jb_array + i; 2901da177e4SLinus Torvalds jb->journal_list = NULL; 291558feb08SJoe Perches jb->bitmaps = vzalloc(mem); 2921da177e4SLinus Torvalds if (!jb->bitmaps) { 293a9dd3643SJeff Mahoney reiserfs_warning(sb, "clm-2000", "unable to " 29445b03d5eSJeff Mahoney "allocate bitmaps for journal lists"); 2951da177e4SLinus Torvalds failed = 1; 2961da177e4SLinus Torvalds break; 2971da177e4SLinus Torvalds } 2981da177e4SLinus Torvalds } 2991da177e4SLinus Torvalds if (failed) { 300a9dd3643SJeff Mahoney free_list_bitmaps(sb, jb_array); 3011da177e4SLinus Torvalds return -1; 3021da177e4SLinus Torvalds } 3031da177e4SLinus Torvalds return 0; 3041da177e4SLinus Torvalds } 3051da177e4SLinus Torvalds 3061da177e4SLinus Torvalds /* 307098297b2SJeff Mahoney * find an available list bitmap. If you can't find one, flush a commit list 308098297b2SJeff Mahoney * and try again 3091da177e4SLinus Torvalds */ 310a9dd3643SJeff Mahoney static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *sb, 311bd4c625cSLinus Torvalds struct reiserfs_journal_list 312bd4c625cSLinus Torvalds *jl) 313bd4c625cSLinus Torvalds { 3141da177e4SLinus Torvalds int i, j; 315a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 3161da177e4SLinus Torvalds struct reiserfs_list_bitmap *jb = NULL; 3171da177e4SLinus Torvalds 3181da177e4SLinus Torvalds for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) { 3191da177e4SLinus Torvalds i = journal->j_list_bitmap_index; 3201da177e4SLinus Torvalds journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS; 3211da177e4SLinus Torvalds jb = journal->j_list_bitmap + i; 3221da177e4SLinus Torvalds if (journal->j_list_bitmap[i].journal_list) { 323a9dd3643SJeff Mahoney flush_commit_list(sb, 324bd4c625cSLinus Torvalds journal->j_list_bitmap[i]. 325bd4c625cSLinus Torvalds journal_list, 1); 3261da177e4SLinus Torvalds if (!journal->j_list_bitmap[i].journal_list) { 3271da177e4SLinus Torvalds break; 3281da177e4SLinus Torvalds } 3291da177e4SLinus Torvalds } else { 3301da177e4SLinus Torvalds break; 3311da177e4SLinus Torvalds } 3321da177e4SLinus Torvalds } 333098297b2SJeff Mahoney /* double check to make sure if flushed correctly */ 334098297b2SJeff Mahoney if (jb->journal_list) 3351da177e4SLinus Torvalds return NULL; 3361da177e4SLinus Torvalds jb->journal_list = jl; 3371da177e4SLinus Torvalds return jb; 3381da177e4SLinus Torvalds } 3391da177e4SLinus Torvalds 3401da177e4SLinus Torvalds /* 341098297b2SJeff Mahoney * allocates a new chunk of X nodes, and links them all together as a list. 342098297b2SJeff Mahoney * Uses the cnode->next and cnode->prev pointers 343098297b2SJeff Mahoney * returns NULL on failure 3441da177e4SLinus Torvalds */ 345bd4c625cSLinus Torvalds static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes) 346bd4c625cSLinus Torvalds { 3471da177e4SLinus Torvalds struct reiserfs_journal_cnode *head; 3481da177e4SLinus Torvalds int i; 3491da177e4SLinus Torvalds if (num_cnodes <= 0) { 3501da177e4SLinus Torvalds return NULL; 3511da177e4SLinus Torvalds } 352558feb08SJoe Perches head = vzalloc(num_cnodes * sizeof(struct reiserfs_journal_cnode)); 3531da177e4SLinus Torvalds if (!head) { 3541da177e4SLinus Torvalds return NULL; 3551da177e4SLinus Torvalds } 3561da177e4SLinus Torvalds head[0].prev = NULL; 3571da177e4SLinus Torvalds head[0].next = head + 1; 3581da177e4SLinus Torvalds for (i = 1; i < num_cnodes; i++) { 3591da177e4SLinus Torvalds head[i].prev = head + (i - 1); 3601da177e4SLinus Torvalds head[i].next = head + (i + 1); /* if last one, overwrite it after the if */ 3611da177e4SLinus Torvalds } 3621da177e4SLinus Torvalds head[num_cnodes - 1].next = NULL; 3631da177e4SLinus Torvalds return head; 3641da177e4SLinus Torvalds } 3651da177e4SLinus Torvalds 366098297b2SJeff Mahoney /* pulls a cnode off the free list, or returns NULL on failure */ 367a9dd3643SJeff Mahoney static struct reiserfs_journal_cnode *get_cnode(struct super_block *sb) 368bd4c625cSLinus Torvalds { 3691da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 370a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 3711da177e4SLinus Torvalds 372a9dd3643SJeff Mahoney reiserfs_check_lock_depth(sb, "get_cnode"); 3731da177e4SLinus Torvalds 3741da177e4SLinus Torvalds if (journal->j_cnode_free <= 0) { 3751da177e4SLinus Torvalds return NULL; 3761da177e4SLinus Torvalds } 3771da177e4SLinus Torvalds journal->j_cnode_used++; 3781da177e4SLinus Torvalds journal->j_cnode_free--; 3791da177e4SLinus Torvalds cn = journal->j_cnode_free_list; 3801da177e4SLinus Torvalds if (!cn) { 3811da177e4SLinus Torvalds return cn; 3821da177e4SLinus Torvalds } 3831da177e4SLinus Torvalds if (cn->next) { 3841da177e4SLinus Torvalds cn->next->prev = NULL; 3851da177e4SLinus Torvalds } 3861da177e4SLinus Torvalds journal->j_cnode_free_list = cn->next; 3871da177e4SLinus Torvalds memset(cn, 0, sizeof(struct reiserfs_journal_cnode)); 3881da177e4SLinus Torvalds return cn; 3891da177e4SLinus Torvalds } 3901da177e4SLinus Torvalds 3911da177e4SLinus Torvalds /* 392098297b2SJeff Mahoney * returns a cnode to the free list 3931da177e4SLinus Torvalds */ 394a9dd3643SJeff Mahoney static void free_cnode(struct super_block *sb, 395bd4c625cSLinus Torvalds struct reiserfs_journal_cnode *cn) 396bd4c625cSLinus Torvalds { 397a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 3981da177e4SLinus Torvalds 399a9dd3643SJeff Mahoney reiserfs_check_lock_depth(sb, "free_cnode"); 4001da177e4SLinus Torvalds 4011da177e4SLinus Torvalds journal->j_cnode_used--; 4021da177e4SLinus Torvalds journal->j_cnode_free++; 4031da177e4SLinus Torvalds /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */ 4041da177e4SLinus Torvalds cn->next = journal->j_cnode_free_list; 4051da177e4SLinus Torvalds if (journal->j_cnode_free_list) { 4061da177e4SLinus Torvalds journal->j_cnode_free_list->prev = cn; 4071da177e4SLinus Torvalds } 4081da177e4SLinus Torvalds cn->prev = NULL; /* not needed with the memset, but I might kill the memset, and forget to do this */ 4091da177e4SLinus Torvalds journal->j_cnode_free_list = cn; 4101da177e4SLinus Torvalds } 4111da177e4SLinus Torvalds 412bd4c625cSLinus Torvalds static void clear_prepared_bits(struct buffer_head *bh) 413bd4c625cSLinus Torvalds { 4141da177e4SLinus Torvalds clear_buffer_journal_prepared(bh); 4151da177e4SLinus Torvalds clear_buffer_journal_restore_dirty(bh); 4161da177e4SLinus Torvalds } 4171da177e4SLinus Torvalds 418098297b2SJeff Mahoney /* 419098297b2SJeff Mahoney * return a cnode with same dev, block number and size in table, 420098297b2SJeff Mahoney * or null if not found 421098297b2SJeff Mahoney */ 422bd4c625cSLinus Torvalds static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct 423bd4c625cSLinus Torvalds super_block 424bd4c625cSLinus Torvalds *sb, 425bd4c625cSLinus Torvalds struct 426bd4c625cSLinus Torvalds reiserfs_journal_cnode 427bd4c625cSLinus Torvalds **table, 4281da177e4SLinus Torvalds long bl) 4291da177e4SLinus Torvalds { 4301da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 4311da177e4SLinus Torvalds cn = journal_hash(table, sb, bl); 4321da177e4SLinus Torvalds while (cn) { 4331da177e4SLinus Torvalds if (cn->blocknr == bl && cn->sb == sb) 4341da177e4SLinus Torvalds return cn; 4351da177e4SLinus Torvalds cn = cn->hnext; 4361da177e4SLinus Torvalds } 4371da177e4SLinus Torvalds return (struct reiserfs_journal_cnode *)0; 4381da177e4SLinus Torvalds } 4391da177e4SLinus Torvalds 4401da177e4SLinus Torvalds /* 441098297b2SJeff Mahoney * this actually means 'can this block be reallocated yet?'. If you set 442098297b2SJeff Mahoney * search_all, a block can only be allocated if it is not in the current 443098297b2SJeff Mahoney * transaction, was not freed by the current transaction, and has no chance 444098297b2SJeff Mahoney * of ever being overwritten by a replay after crashing. 445098297b2SJeff Mahoney * 446098297b2SJeff Mahoney * If you don't set search_all, a block can only be allocated if it is not 447098297b2SJeff Mahoney * in the current transaction. Since deleting a block removes it from the 448098297b2SJeff Mahoney * current transaction, this case should never happen. If you don't set 449098297b2SJeff Mahoney * search_all, make sure you never write the block without logging it. 450098297b2SJeff Mahoney * 451098297b2SJeff Mahoney * next_zero_bit is a suggestion about the next block to try for find_forward. 452098297b2SJeff Mahoney * when bl is rejected because it is set in a journal list bitmap, we search 453098297b2SJeff Mahoney * for the next zero bit in the bitmap that rejected bl. Then, we return 454098297b2SJeff Mahoney * that through next_zero_bit for find_forward to try. 455098297b2SJeff Mahoney * 456098297b2SJeff Mahoney * Just because we return something in next_zero_bit does not mean we won't 457098297b2SJeff Mahoney * reject it on the next call to reiserfs_in_journal 4581da177e4SLinus Torvalds */ 459a9dd3643SJeff Mahoney int reiserfs_in_journal(struct super_block *sb, 4603ee16670SJeff Mahoney unsigned int bmap_nr, int bit_nr, int search_all, 461bd4c625cSLinus Torvalds b_blocknr_t * next_zero_bit) 462bd4c625cSLinus Torvalds { 463a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 4641da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 4651da177e4SLinus Torvalds struct reiserfs_list_bitmap *jb; 4661da177e4SLinus Torvalds int i; 4671da177e4SLinus Torvalds unsigned long bl; 4681da177e4SLinus Torvalds 4691da177e4SLinus Torvalds *next_zero_bit = 0; /* always start this at zero. */ 4701da177e4SLinus Torvalds 471a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.in_journal); 472098297b2SJeff Mahoney /* 473098297b2SJeff Mahoney * If we aren't doing a search_all, this is a metablock, and it 474098297b2SJeff Mahoney * will be logged before use. if we crash before the transaction 475098297b2SJeff Mahoney * that freed it commits, this transaction won't have committed 476098297b2SJeff Mahoney * either, and the block will never be written 4771da177e4SLinus Torvalds */ 4781da177e4SLinus Torvalds if (search_all) { 4791da177e4SLinus Torvalds for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) { 480a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.in_journal_bitmap); 4811da177e4SLinus Torvalds jb = journal->j_list_bitmap + i; 4821da177e4SLinus Torvalds if (jb->journal_list && jb->bitmaps[bmap_nr] && 483bd4c625cSLinus Torvalds test_bit(bit_nr, 484bd4c625cSLinus Torvalds (unsigned long *)jb->bitmaps[bmap_nr]-> 485bd4c625cSLinus Torvalds data)) { 486bd4c625cSLinus Torvalds *next_zero_bit = 487bd4c625cSLinus Torvalds find_next_zero_bit((unsigned long *) 488bd4c625cSLinus Torvalds (jb->bitmaps[bmap_nr]-> 489bd4c625cSLinus Torvalds data), 490a9dd3643SJeff Mahoney sb->s_blocksize << 3, 491bd4c625cSLinus Torvalds bit_nr + 1); 4921da177e4SLinus Torvalds return 1; 4931da177e4SLinus Torvalds } 4941da177e4SLinus Torvalds } 4951da177e4SLinus Torvalds } 4961da177e4SLinus Torvalds 497a9dd3643SJeff Mahoney bl = bmap_nr * (sb->s_blocksize << 3) + bit_nr; 4981da177e4SLinus Torvalds /* is it in any old transactions? */ 499bd4c625cSLinus Torvalds if (search_all 500bd4c625cSLinus Torvalds && (cn = 501a9dd3643SJeff Mahoney get_journal_hash_dev(sb, journal->j_list_hash_table, bl))) { 5021da177e4SLinus Torvalds return 1; 5031da177e4SLinus Torvalds } 5041da177e4SLinus Torvalds 5051da177e4SLinus Torvalds /* is it in the current transaction. This should never happen */ 506a9dd3643SJeff Mahoney if ((cn = get_journal_hash_dev(sb, journal->j_hash_table, bl))) { 5071da177e4SLinus Torvalds BUG(); 5081da177e4SLinus Torvalds return 1; 5091da177e4SLinus Torvalds } 5101da177e4SLinus Torvalds 511a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.in_journal_reusable); 5121da177e4SLinus Torvalds /* safe for reuse */ 5131da177e4SLinus Torvalds return 0; 5141da177e4SLinus Torvalds } 5151da177e4SLinus Torvalds 516098297b2SJeff Mahoney /* insert cn into table */ 517bd4c625cSLinus Torvalds static inline void insert_journal_hash(struct reiserfs_journal_cnode **table, 518bd4c625cSLinus Torvalds struct reiserfs_journal_cnode *cn) 519bd4c625cSLinus Torvalds { 5201da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn_orig; 5211da177e4SLinus Torvalds 5221da177e4SLinus Torvalds cn_orig = journal_hash(table, cn->sb, cn->blocknr); 5231da177e4SLinus Torvalds cn->hnext = cn_orig; 5241da177e4SLinus Torvalds cn->hprev = NULL; 5251da177e4SLinus Torvalds if (cn_orig) { 5261da177e4SLinus Torvalds cn_orig->hprev = cn; 5271da177e4SLinus Torvalds } 5281da177e4SLinus Torvalds journal_hash(table, cn->sb, cn->blocknr) = cn; 5291da177e4SLinus Torvalds } 5301da177e4SLinus Torvalds 5311da177e4SLinus Torvalds /* lock the current transaction */ 532a9dd3643SJeff Mahoney static inline void lock_journal(struct super_block *sb) 533bd4c625cSLinus Torvalds { 534a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.lock_journal); 5358ebc4232SFrederic Weisbecker 5368ebc4232SFrederic Weisbecker reiserfs_mutex_lock_safe(&SB_JOURNAL(sb)->j_mutex, sb); 5371da177e4SLinus Torvalds } 5381da177e4SLinus Torvalds 5391da177e4SLinus Torvalds /* unlock the current transaction */ 540a9dd3643SJeff Mahoney static inline void unlock_journal(struct super_block *sb) 541bd4c625cSLinus Torvalds { 542a9dd3643SJeff Mahoney mutex_unlock(&SB_JOURNAL(sb)->j_mutex); 5431da177e4SLinus Torvalds } 5441da177e4SLinus Torvalds 5451da177e4SLinus Torvalds static inline void get_journal_list(struct reiserfs_journal_list *jl) 5461da177e4SLinus Torvalds { 5471da177e4SLinus Torvalds jl->j_refcount++; 5481da177e4SLinus Torvalds } 5491da177e4SLinus Torvalds 5501da177e4SLinus Torvalds static inline void put_journal_list(struct super_block *s, 5511da177e4SLinus Torvalds struct reiserfs_journal_list *jl) 5521da177e4SLinus Torvalds { 5531da177e4SLinus Torvalds if (jl->j_refcount < 1) { 554c3a9c210SJeff Mahoney reiserfs_panic(s, "journal-2", "trans id %u, refcount at %d", 555bd4c625cSLinus Torvalds jl->j_trans_id, jl->j_refcount); 5561da177e4SLinus Torvalds } 5571da177e4SLinus Torvalds if (--jl->j_refcount == 0) 558d739b42bSPekka Enberg kfree(jl); 5591da177e4SLinus Torvalds } 5601da177e4SLinus Torvalds 5611da177e4SLinus Torvalds /* 562098297b2SJeff Mahoney * this used to be much more involved, and I'm keeping it just in case 563098297b2SJeff Mahoney * things get ugly again. it gets called by flush_commit_list, and 564098297b2SJeff Mahoney * cleans up any data stored about blocks freed during a transaction. 5651da177e4SLinus Torvalds */ 566a9dd3643SJeff Mahoney static void cleanup_freed_for_journal_list(struct super_block *sb, 567bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl) 568bd4c625cSLinus Torvalds { 5691da177e4SLinus Torvalds 5701da177e4SLinus Torvalds struct reiserfs_list_bitmap *jb = jl->j_list_bitmap; 5711da177e4SLinus Torvalds if (jb) { 572a9dd3643SJeff Mahoney cleanup_bitmap_list(sb, jb); 5731da177e4SLinus Torvalds } 5741da177e4SLinus Torvalds jl->j_list_bitmap->journal_list = NULL; 5751da177e4SLinus Torvalds jl->j_list_bitmap = NULL; 5761da177e4SLinus Torvalds } 5771da177e4SLinus Torvalds 5781da177e4SLinus Torvalds static int journal_list_still_alive(struct super_block *s, 579600ed416SJeff Mahoney unsigned int trans_id) 5801da177e4SLinus Torvalds { 5811da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 5821da177e4SLinus Torvalds struct list_head *entry = &journal->j_journal_list; 5831da177e4SLinus Torvalds struct reiserfs_journal_list *jl; 5841da177e4SLinus Torvalds 5851da177e4SLinus Torvalds if (!list_empty(entry)) { 5861da177e4SLinus Torvalds jl = JOURNAL_LIST_ENTRY(entry->next); 5871da177e4SLinus Torvalds if (jl->j_trans_id <= trans_id) { 5881da177e4SLinus Torvalds return 1; 5891da177e4SLinus Torvalds } 5901da177e4SLinus Torvalds } 5911da177e4SLinus Torvalds return 0; 5921da177e4SLinus Torvalds } 5931da177e4SLinus Torvalds 594398c95bdSChris Mason /* 595398c95bdSChris Mason * If page->mapping was null, we failed to truncate this page for 596398c95bdSChris Mason * some reason. Most likely because it was truncated after being 597398c95bdSChris Mason * logged via data=journal. 598398c95bdSChris Mason * 599398c95bdSChris Mason * This does a check to see if the buffer belongs to one of these 600398c95bdSChris Mason * lost pages before doing the final put_bh. If page->mapping was 601398c95bdSChris Mason * null, it tries to free buffers on the page, which should make the 602398c95bdSChris Mason * final page_cache_release drop the page from the lru. 603398c95bdSChris Mason */ 604398c95bdSChris Mason static void release_buffer_page(struct buffer_head *bh) 605398c95bdSChris Mason { 606398c95bdSChris Mason struct page *page = bh->b_page; 607529ae9aaSNick Piggin if (!page->mapping && trylock_page(page)) { 608398c95bdSChris Mason page_cache_get(page); 609398c95bdSChris Mason put_bh(bh); 610398c95bdSChris Mason if (!page->mapping) 611398c95bdSChris Mason try_to_free_buffers(page); 612398c95bdSChris Mason unlock_page(page); 613398c95bdSChris Mason page_cache_release(page); 614398c95bdSChris Mason } else { 615398c95bdSChris Mason put_bh(bh); 616398c95bdSChris Mason } 617398c95bdSChris Mason } 618398c95bdSChris Mason 619bd4c625cSLinus Torvalds static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate) 620bd4c625cSLinus Torvalds { 6211da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 6221da177e4SLinus Torvalds 6231da177e4SLinus Torvalds if (buffer_journaled(bh)) { 62445b03d5eSJeff Mahoney reiserfs_warning(NULL, "clm-2084", 62545b03d5eSJeff Mahoney "pinned buffer %lu:%s sent to disk", 6261da177e4SLinus Torvalds bh->b_blocknr, bdevname(bh->b_bdev, b)); 6271da177e4SLinus Torvalds } 6281da177e4SLinus Torvalds if (uptodate) 6291da177e4SLinus Torvalds set_buffer_uptodate(bh); 6301da177e4SLinus Torvalds else 6311da177e4SLinus Torvalds clear_buffer_uptodate(bh); 632398c95bdSChris Mason 6331da177e4SLinus Torvalds unlock_buffer(bh); 634398c95bdSChris Mason release_buffer_page(bh); 6351da177e4SLinus Torvalds } 6361da177e4SLinus Torvalds 637bd4c625cSLinus Torvalds static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate) 638bd4c625cSLinus Torvalds { 6391da177e4SLinus Torvalds if (uptodate) 6401da177e4SLinus Torvalds set_buffer_uptodate(bh); 6411da177e4SLinus Torvalds else 6421da177e4SLinus Torvalds clear_buffer_uptodate(bh); 6431da177e4SLinus Torvalds unlock_buffer(bh); 6441da177e4SLinus Torvalds put_bh(bh); 6451da177e4SLinus Torvalds } 6461da177e4SLinus Torvalds 647bd4c625cSLinus Torvalds static void submit_logged_buffer(struct buffer_head *bh) 648bd4c625cSLinus Torvalds { 6491da177e4SLinus Torvalds get_bh(bh); 6501da177e4SLinus Torvalds bh->b_end_io = reiserfs_end_buffer_io_sync; 6511da177e4SLinus Torvalds clear_buffer_journal_new(bh); 6521da177e4SLinus Torvalds clear_buffer_dirty(bh); 6531da177e4SLinus Torvalds if (!test_clear_buffer_journal_test(bh)) 6541da177e4SLinus Torvalds BUG(); 6551da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 6561da177e4SLinus Torvalds BUG(); 6571da177e4SLinus Torvalds submit_bh(WRITE, bh); 6581da177e4SLinus Torvalds } 6591da177e4SLinus Torvalds 660bd4c625cSLinus Torvalds static void submit_ordered_buffer(struct buffer_head *bh) 661bd4c625cSLinus Torvalds { 6621da177e4SLinus Torvalds get_bh(bh); 6631da177e4SLinus Torvalds bh->b_end_io = reiserfs_end_ordered_io; 6641da177e4SLinus Torvalds clear_buffer_dirty(bh); 6651da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 6661da177e4SLinus Torvalds BUG(); 6671da177e4SLinus Torvalds submit_bh(WRITE, bh); 6681da177e4SLinus Torvalds } 6691da177e4SLinus Torvalds 6701da177e4SLinus Torvalds #define CHUNK_SIZE 32 6711da177e4SLinus Torvalds struct buffer_chunk { 6721da177e4SLinus Torvalds struct buffer_head *bh[CHUNK_SIZE]; 6731da177e4SLinus Torvalds int nr; 6741da177e4SLinus Torvalds }; 6751da177e4SLinus Torvalds 676bd4c625cSLinus Torvalds static void write_chunk(struct buffer_chunk *chunk) 677bd4c625cSLinus Torvalds { 6781da177e4SLinus Torvalds int i; 6791da177e4SLinus Torvalds for (i = 0; i < chunk->nr; i++) { 6801da177e4SLinus Torvalds submit_logged_buffer(chunk->bh[i]); 6811da177e4SLinus Torvalds } 6821da177e4SLinus Torvalds chunk->nr = 0; 6831da177e4SLinus Torvalds } 6841da177e4SLinus Torvalds 685bd4c625cSLinus Torvalds static void write_ordered_chunk(struct buffer_chunk *chunk) 686bd4c625cSLinus Torvalds { 6871da177e4SLinus Torvalds int i; 6881da177e4SLinus Torvalds for (i = 0; i < chunk->nr; i++) { 6891da177e4SLinus Torvalds submit_ordered_buffer(chunk->bh[i]); 6901da177e4SLinus Torvalds } 6911da177e4SLinus Torvalds chunk->nr = 0; 6921da177e4SLinus Torvalds } 6931da177e4SLinus Torvalds 6941da177e4SLinus Torvalds static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh, 695bd4c625cSLinus Torvalds spinlock_t * lock, void (fn) (struct buffer_chunk *)) 6961da177e4SLinus Torvalds { 6971da177e4SLinus Torvalds int ret = 0; 69814a61442SEric Sesterhenn BUG_ON(chunk->nr >= CHUNK_SIZE); 6991da177e4SLinus Torvalds chunk->bh[chunk->nr++] = bh; 7001da177e4SLinus Torvalds if (chunk->nr >= CHUNK_SIZE) { 7011da177e4SLinus Torvalds ret = 1; 7021da177e4SLinus Torvalds if (lock) 7031da177e4SLinus Torvalds spin_unlock(lock); 7041da177e4SLinus Torvalds fn(chunk); 7051da177e4SLinus Torvalds if (lock) 7061da177e4SLinus Torvalds spin_lock(lock); 7071da177e4SLinus Torvalds } 7081da177e4SLinus Torvalds return ret; 7091da177e4SLinus Torvalds } 7101da177e4SLinus Torvalds 7111da177e4SLinus Torvalds static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0); 712bd4c625cSLinus Torvalds static struct reiserfs_jh *alloc_jh(void) 713bd4c625cSLinus Torvalds { 7141da177e4SLinus Torvalds struct reiserfs_jh *jh; 7151da177e4SLinus Torvalds while (1) { 7161da177e4SLinus Torvalds jh = kmalloc(sizeof(*jh), GFP_NOFS); 7171da177e4SLinus Torvalds if (jh) { 7181da177e4SLinus Torvalds atomic_inc(&nr_reiserfs_jh); 7191da177e4SLinus Torvalds return jh; 7201da177e4SLinus Torvalds } 7211da177e4SLinus Torvalds yield(); 7221da177e4SLinus Torvalds } 7231da177e4SLinus Torvalds } 7241da177e4SLinus Torvalds 7251da177e4SLinus Torvalds /* 7261da177e4SLinus Torvalds * we want to free the jh when the buffer has been written 7271da177e4SLinus Torvalds * and waited on 7281da177e4SLinus Torvalds */ 729bd4c625cSLinus Torvalds void reiserfs_free_jh(struct buffer_head *bh) 730bd4c625cSLinus Torvalds { 7311da177e4SLinus Torvalds struct reiserfs_jh *jh; 7321da177e4SLinus Torvalds 7331da177e4SLinus Torvalds jh = bh->b_private; 7341da177e4SLinus Torvalds if (jh) { 7351da177e4SLinus Torvalds bh->b_private = NULL; 7361da177e4SLinus Torvalds jh->bh = NULL; 7371da177e4SLinus Torvalds list_del_init(&jh->list); 7381da177e4SLinus Torvalds kfree(jh); 7391da177e4SLinus Torvalds if (atomic_read(&nr_reiserfs_jh) <= 0) 7401da177e4SLinus Torvalds BUG(); 7411da177e4SLinus Torvalds atomic_dec(&nr_reiserfs_jh); 7421da177e4SLinus Torvalds put_bh(bh); 7431da177e4SLinus Torvalds } 7441da177e4SLinus Torvalds } 7451da177e4SLinus Torvalds 7461da177e4SLinus Torvalds static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh, 7471da177e4SLinus Torvalds int tail) 7481da177e4SLinus Torvalds { 7491da177e4SLinus Torvalds struct reiserfs_jh *jh; 7501da177e4SLinus Torvalds 7511da177e4SLinus Torvalds if (bh->b_private) { 7521da177e4SLinus Torvalds spin_lock(&j->j_dirty_buffers_lock); 7531da177e4SLinus Torvalds if (!bh->b_private) { 7541da177e4SLinus Torvalds spin_unlock(&j->j_dirty_buffers_lock); 7551da177e4SLinus Torvalds goto no_jh; 7561da177e4SLinus Torvalds } 7571da177e4SLinus Torvalds jh = bh->b_private; 7581da177e4SLinus Torvalds list_del_init(&jh->list); 7591da177e4SLinus Torvalds } else { 7601da177e4SLinus Torvalds no_jh: 7611da177e4SLinus Torvalds get_bh(bh); 7621da177e4SLinus Torvalds jh = alloc_jh(); 7631da177e4SLinus Torvalds spin_lock(&j->j_dirty_buffers_lock); 764098297b2SJeff Mahoney /* 765098297b2SJeff Mahoney * buffer must be locked for __add_jh, should be able to have 7661da177e4SLinus Torvalds * two adds at the same time 7671da177e4SLinus Torvalds */ 76814a61442SEric Sesterhenn BUG_ON(bh->b_private); 7691da177e4SLinus Torvalds jh->bh = bh; 7701da177e4SLinus Torvalds bh->b_private = jh; 7711da177e4SLinus Torvalds } 7721da177e4SLinus Torvalds jh->jl = j->j_current_jl; 7731da177e4SLinus Torvalds if (tail) 7741da177e4SLinus Torvalds list_add_tail(&jh->list, &jh->jl->j_tail_bh_list); 7751da177e4SLinus Torvalds else { 7761da177e4SLinus Torvalds list_add_tail(&jh->list, &jh->jl->j_bh_list); 7771da177e4SLinus Torvalds } 7781da177e4SLinus Torvalds spin_unlock(&j->j_dirty_buffers_lock); 7791da177e4SLinus Torvalds return 0; 7801da177e4SLinus Torvalds } 7811da177e4SLinus Torvalds 782bd4c625cSLinus Torvalds int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh) 783bd4c625cSLinus Torvalds { 7841da177e4SLinus Torvalds return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1); 7851da177e4SLinus Torvalds } 786bd4c625cSLinus Torvalds int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh) 787bd4c625cSLinus Torvalds { 7881da177e4SLinus Torvalds return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0); 7891da177e4SLinus Torvalds } 7901da177e4SLinus Torvalds 7911da177e4SLinus Torvalds #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list) 7921da177e4SLinus Torvalds static int write_ordered_buffers(spinlock_t * lock, 7931da177e4SLinus Torvalds struct reiserfs_journal *j, 7941da177e4SLinus Torvalds struct reiserfs_journal_list *jl, 7951da177e4SLinus Torvalds struct list_head *list) 7961da177e4SLinus Torvalds { 7971da177e4SLinus Torvalds struct buffer_head *bh; 7981da177e4SLinus Torvalds struct reiserfs_jh *jh; 7991da177e4SLinus Torvalds int ret = j->j_errno; 8001da177e4SLinus Torvalds struct buffer_chunk chunk; 8011da177e4SLinus Torvalds struct list_head tmp; 8021da177e4SLinus Torvalds INIT_LIST_HEAD(&tmp); 8031da177e4SLinus Torvalds 8041da177e4SLinus Torvalds chunk.nr = 0; 8051da177e4SLinus Torvalds spin_lock(lock); 8061da177e4SLinus Torvalds while (!list_empty(list)) { 8071da177e4SLinus Torvalds jh = JH_ENTRY(list->next); 8081da177e4SLinus Torvalds bh = jh->bh; 8091da177e4SLinus Torvalds get_bh(bh); 810ca5de404SNick Piggin if (!trylock_buffer(bh)) { 8111da177e4SLinus Torvalds if (!buffer_dirty(bh)) { 812f116629dSAkinobu Mita list_move(&jh->list, &tmp); 8131da177e4SLinus Torvalds goto loop_next; 8141da177e4SLinus Torvalds } 8151da177e4SLinus Torvalds spin_unlock(lock); 8161da177e4SLinus Torvalds if (chunk.nr) 8171da177e4SLinus Torvalds write_ordered_chunk(&chunk); 8181da177e4SLinus Torvalds wait_on_buffer(bh); 8191da177e4SLinus Torvalds cond_resched(); 8201da177e4SLinus Torvalds spin_lock(lock); 8211da177e4SLinus Torvalds goto loop_next; 8221da177e4SLinus Torvalds } 823098297b2SJeff Mahoney /* 824098297b2SJeff Mahoney * in theory, dirty non-uptodate buffers should never get here, 8253d4492f8SChris Mason * but the upper layer io error paths still have a few quirks. 8263d4492f8SChris Mason * Handle them here as gracefully as we can 8273d4492f8SChris Mason */ 8283d4492f8SChris Mason if (!buffer_uptodate(bh) && buffer_dirty(bh)) { 8293d4492f8SChris Mason clear_buffer_dirty(bh); 8303d4492f8SChris Mason ret = -EIO; 8313d4492f8SChris Mason } 8321da177e4SLinus Torvalds if (buffer_dirty(bh)) { 833f116629dSAkinobu Mita list_move(&jh->list, &tmp); 8341da177e4SLinus Torvalds add_to_chunk(&chunk, bh, lock, write_ordered_chunk); 8351da177e4SLinus Torvalds } else { 8361da177e4SLinus Torvalds reiserfs_free_jh(bh); 8371da177e4SLinus Torvalds unlock_buffer(bh); 8381da177e4SLinus Torvalds } 8391da177e4SLinus Torvalds loop_next: 8401da177e4SLinus Torvalds put_bh(bh); 8411da177e4SLinus Torvalds cond_resched_lock(lock); 8421da177e4SLinus Torvalds } 8431da177e4SLinus Torvalds if (chunk.nr) { 8441da177e4SLinus Torvalds spin_unlock(lock); 8451da177e4SLinus Torvalds write_ordered_chunk(&chunk); 8461da177e4SLinus Torvalds spin_lock(lock); 8471da177e4SLinus Torvalds } 8481da177e4SLinus Torvalds while (!list_empty(&tmp)) { 8491da177e4SLinus Torvalds jh = JH_ENTRY(tmp.prev); 8501da177e4SLinus Torvalds bh = jh->bh; 8511da177e4SLinus Torvalds get_bh(bh); 8521da177e4SLinus Torvalds reiserfs_free_jh(bh); 8531da177e4SLinus Torvalds 8541da177e4SLinus Torvalds if (buffer_locked(bh)) { 8551da177e4SLinus Torvalds spin_unlock(lock); 8561da177e4SLinus Torvalds wait_on_buffer(bh); 8571da177e4SLinus Torvalds spin_lock(lock); 8581da177e4SLinus Torvalds } 8591da177e4SLinus Torvalds if (!buffer_uptodate(bh)) { 8601da177e4SLinus Torvalds ret = -EIO; 8611da177e4SLinus Torvalds } 862098297b2SJeff Mahoney /* 863098297b2SJeff Mahoney * ugly interaction with invalidatepage here. 864098297b2SJeff Mahoney * reiserfs_invalidate_page will pin any buffer that has a 865098297b2SJeff Mahoney * valid journal head from an older transaction. If someone 866098297b2SJeff Mahoney * else sets our buffer dirty after we write it in the first 867098297b2SJeff Mahoney * loop, and then someone truncates the page away, nobody 868098297b2SJeff Mahoney * will ever write the buffer. We're safe if we write the 869098297b2SJeff Mahoney * page one last time after freeing the journal header. 870d62b1b87SChris Mason */ 871d62b1b87SChris Mason if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) { 872d62b1b87SChris Mason spin_unlock(lock); 873d62b1b87SChris Mason ll_rw_block(WRITE, 1, &bh); 874d62b1b87SChris Mason spin_lock(lock); 875d62b1b87SChris Mason } 8761da177e4SLinus Torvalds put_bh(bh); 8771da177e4SLinus Torvalds cond_resched_lock(lock); 8781da177e4SLinus Torvalds } 8791da177e4SLinus Torvalds spin_unlock(lock); 8801da177e4SLinus Torvalds return ret; 8811da177e4SLinus Torvalds } 8821da177e4SLinus Torvalds 883bd4c625cSLinus Torvalds static int flush_older_commits(struct super_block *s, 884bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl) 885bd4c625cSLinus Torvalds { 8861da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 8871da177e4SLinus Torvalds struct reiserfs_journal_list *other_jl; 8881da177e4SLinus Torvalds struct reiserfs_journal_list *first_jl; 8891da177e4SLinus Torvalds struct list_head *entry; 890600ed416SJeff Mahoney unsigned int trans_id = jl->j_trans_id; 891600ed416SJeff Mahoney unsigned int other_trans_id; 892600ed416SJeff Mahoney unsigned int first_trans_id; 8931da177e4SLinus Torvalds 8941da177e4SLinus Torvalds find_first: 8951da177e4SLinus Torvalds /* 8961da177e4SLinus Torvalds * first we walk backwards to find the oldest uncommitted transation 8971da177e4SLinus Torvalds */ 8981da177e4SLinus Torvalds first_jl = jl; 8991da177e4SLinus Torvalds entry = jl->j_list.prev; 9001da177e4SLinus Torvalds while (1) { 9011da177e4SLinus Torvalds other_jl = JOURNAL_LIST_ENTRY(entry); 9021da177e4SLinus Torvalds if (entry == &journal->j_journal_list || 9031da177e4SLinus Torvalds atomic_read(&other_jl->j_older_commits_done)) 9041da177e4SLinus Torvalds break; 9051da177e4SLinus Torvalds 9061da177e4SLinus Torvalds first_jl = other_jl; 9071da177e4SLinus Torvalds entry = other_jl->j_list.prev; 9081da177e4SLinus Torvalds } 9091da177e4SLinus Torvalds 9101da177e4SLinus Torvalds /* if we didn't find any older uncommitted transactions, return now */ 9111da177e4SLinus Torvalds if (first_jl == jl) { 9121da177e4SLinus Torvalds return 0; 9131da177e4SLinus Torvalds } 9141da177e4SLinus Torvalds 9151da177e4SLinus Torvalds first_trans_id = first_jl->j_trans_id; 9161da177e4SLinus Torvalds 9171da177e4SLinus Torvalds entry = &first_jl->j_list; 9181da177e4SLinus Torvalds while (1) { 9191da177e4SLinus Torvalds other_jl = JOURNAL_LIST_ENTRY(entry); 9201da177e4SLinus Torvalds other_trans_id = other_jl->j_trans_id; 9211da177e4SLinus Torvalds 9221da177e4SLinus Torvalds if (other_trans_id < trans_id) { 9231da177e4SLinus Torvalds if (atomic_read(&other_jl->j_commit_left) != 0) { 9241da177e4SLinus Torvalds flush_commit_list(s, other_jl, 0); 9251da177e4SLinus Torvalds 9261da177e4SLinus Torvalds /* list we were called with is gone, return */ 9271da177e4SLinus Torvalds if (!journal_list_still_alive(s, trans_id)) 9281da177e4SLinus Torvalds return 1; 9291da177e4SLinus Torvalds 930098297b2SJeff Mahoney /* 931098297b2SJeff Mahoney * the one we just flushed is gone, this means 932098297b2SJeff Mahoney * all older lists are also gone, so first_jl 933098297b2SJeff Mahoney * is no longer valid either. Go back to the 934098297b2SJeff Mahoney * beginning. 9351da177e4SLinus Torvalds */ 936bd4c625cSLinus Torvalds if (!journal_list_still_alive 937bd4c625cSLinus Torvalds (s, other_trans_id)) { 9381da177e4SLinus Torvalds goto find_first; 9391da177e4SLinus Torvalds } 9401da177e4SLinus Torvalds } 9411da177e4SLinus Torvalds entry = entry->next; 9421da177e4SLinus Torvalds if (entry == &journal->j_journal_list) 9431da177e4SLinus Torvalds return 0; 9441da177e4SLinus Torvalds } else { 9451da177e4SLinus Torvalds return 0; 9461da177e4SLinus Torvalds } 9471da177e4SLinus Torvalds } 9481da177e4SLinus Torvalds return 0; 9491da177e4SLinus Torvalds } 950deba0f49SAdrian Bunk 951deba0f49SAdrian Bunk static int reiserfs_async_progress_wait(struct super_block *s) 952bd4c625cSLinus Torvalds { 9531da177e4SLinus Torvalds struct reiserfs_journal *j = SB_JOURNAL(s); 9548ebc4232SFrederic Weisbecker 9558ebc4232SFrederic Weisbecker if (atomic_read(&j->j_async_throttle)) { 956278f6679SJeff Mahoney int depth; 957278f6679SJeff Mahoney 958278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(s); 9598aa7e847SJens Axboe congestion_wait(BLK_RW_ASYNC, HZ / 10); 960278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 9618ebc4232SFrederic Weisbecker } 9628ebc4232SFrederic Weisbecker 9631da177e4SLinus Torvalds return 0; 9641da177e4SLinus Torvalds } 9651da177e4SLinus Torvalds 9661da177e4SLinus Torvalds /* 967098297b2SJeff Mahoney * if this journal list still has commit blocks unflushed, send them to disk. 968098297b2SJeff Mahoney * 969098297b2SJeff Mahoney * log areas must be flushed in order (transaction 2 can't commit before 970098297b2SJeff Mahoney * transaction 1) Before the commit block can by written, every other log 971098297b2SJeff Mahoney * block must be safely on disk 9721da177e4SLinus Torvalds */ 973bd4c625cSLinus Torvalds static int flush_commit_list(struct super_block *s, 974bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl, int flushall) 975bd4c625cSLinus Torvalds { 9761da177e4SLinus Torvalds int i; 9773ee16670SJeff Mahoney b_blocknr_t bn; 9781da177e4SLinus Torvalds struct buffer_head *tbh = NULL; 979600ed416SJeff Mahoney unsigned int trans_id = jl->j_trans_id; 9801da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 9811da177e4SLinus Torvalds int retval = 0; 982e0e851cfSChris Mason int write_len; 983278f6679SJeff Mahoney int depth; 9841da177e4SLinus Torvalds 9851da177e4SLinus Torvalds reiserfs_check_lock_depth(s, "flush_commit_list"); 9861da177e4SLinus Torvalds 9871da177e4SLinus Torvalds if (atomic_read(&jl->j_older_commits_done)) { 9881da177e4SLinus Torvalds return 0; 9891da177e4SLinus Torvalds } 9901da177e4SLinus Torvalds 991098297b2SJeff Mahoney /* 992098297b2SJeff Mahoney * before we can put our commit blocks on disk, we have to make 993098297b2SJeff Mahoney * sure everyone older than us is on disk too 9941da177e4SLinus Torvalds */ 9951da177e4SLinus Torvalds BUG_ON(jl->j_len <= 0); 9961da177e4SLinus Torvalds BUG_ON(trans_id == journal->j_trans_id); 9971da177e4SLinus Torvalds 9981da177e4SLinus Torvalds get_journal_list(jl); 9991da177e4SLinus Torvalds if (flushall) { 10001da177e4SLinus Torvalds if (flush_older_commits(s, jl) == 1) { 1001098297b2SJeff Mahoney /* 1002098297b2SJeff Mahoney * list disappeared during flush_older_commits. 1003098297b2SJeff Mahoney * return 1004098297b2SJeff Mahoney */ 10051da177e4SLinus Torvalds goto put_jl; 10061da177e4SLinus Torvalds } 10071da177e4SLinus Torvalds } 10081da177e4SLinus Torvalds 10091da177e4SLinus Torvalds /* make sure nobody is trying to flush this one at the same time */ 10108ebc4232SFrederic Weisbecker reiserfs_mutex_lock_safe(&jl->j_commit_mutex, s); 10118ebc4232SFrederic Weisbecker 10121da177e4SLinus Torvalds if (!journal_list_still_alive(s, trans_id)) { 101390415deaSJeff Mahoney mutex_unlock(&jl->j_commit_mutex); 10141da177e4SLinus Torvalds goto put_jl; 10151da177e4SLinus Torvalds } 10161da177e4SLinus Torvalds BUG_ON(jl->j_trans_id == 0); 10171da177e4SLinus Torvalds 10181da177e4SLinus Torvalds /* this commit is done, exit */ 1019a228bf8fSJeff Mahoney if (atomic_read(&jl->j_commit_left) <= 0) { 10201da177e4SLinus Torvalds if (flushall) { 1021a228bf8fSJeff Mahoney atomic_set(&jl->j_older_commits_done, 1); 10221da177e4SLinus Torvalds } 102390415deaSJeff Mahoney mutex_unlock(&jl->j_commit_mutex); 10241da177e4SLinus Torvalds goto put_jl; 10251da177e4SLinus Torvalds } 10261da177e4SLinus Torvalds 10271da177e4SLinus Torvalds if (!list_empty(&jl->j_bh_list)) { 10283d4492f8SChris Mason int ret; 10298ebc4232SFrederic Weisbecker 10308ebc4232SFrederic Weisbecker /* 10318ebc4232SFrederic Weisbecker * We might sleep in numerous places inside 10328ebc4232SFrederic Weisbecker * write_ordered_buffers. Relax the write lock. 10338ebc4232SFrederic Weisbecker */ 1034278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(s); 10353d4492f8SChris Mason ret = write_ordered_buffers(&journal->j_dirty_buffers_lock, 10361da177e4SLinus Torvalds journal, jl, &jl->j_bh_list); 10373d4492f8SChris Mason if (ret < 0 && retval == 0) 10383d4492f8SChris Mason retval = ret; 1039278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 10401da177e4SLinus Torvalds } 10411da177e4SLinus Torvalds BUG_ON(!list_empty(&jl->j_bh_list)); 10421da177e4SLinus Torvalds /* 10431da177e4SLinus Torvalds * for the description block and all the log blocks, submit any buffers 1044e0e851cfSChris Mason * that haven't already reached the disk. Try to write at least 256 1045e0e851cfSChris Mason * log blocks. later on, we will only wait on blocks that correspond 1046e0e851cfSChris Mason * to this transaction, but while we're unplugging we might as well 1047e0e851cfSChris Mason * get a chunk of data on there. 10481da177e4SLinus Torvalds */ 10491da177e4SLinus Torvalds atomic_inc(&journal->j_async_throttle); 1050e0e851cfSChris Mason write_len = jl->j_len + 1; 1051e0e851cfSChris Mason if (write_len < 256) 1052e0e851cfSChris Mason write_len = 256; 1053e0e851cfSChris Mason for (i = 0 ; i < write_len ; i++) { 10541da177e4SLinus Torvalds bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) % 10551da177e4SLinus Torvalds SB_ONDISK_JOURNAL_SIZE(s); 10561da177e4SLinus Torvalds tbh = journal_find_get_block(s, bn); 1057e0e851cfSChris Mason if (tbh) { 10586e3647acSFrederic Weisbecker if (buffer_dirty(tbh)) { 1059278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(s); 1060e0e851cfSChris Mason ll_rw_block(WRITE, 1, &tbh); 1061278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 10626e3647acSFrederic Weisbecker } 10631da177e4SLinus Torvalds put_bh(tbh) ; 10641da177e4SLinus Torvalds } 1065e0e851cfSChris Mason } 10661da177e4SLinus Torvalds atomic_dec(&journal->j_async_throttle); 10671da177e4SLinus Torvalds 10681da177e4SLinus Torvalds for (i = 0; i < (jl->j_len + 1); i++) { 10691da177e4SLinus Torvalds bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + 10701da177e4SLinus Torvalds (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s); 10711da177e4SLinus Torvalds tbh = journal_find_get_block(s, bn); 10728ebc4232SFrederic Weisbecker 1073278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(s); 1074278f6679SJeff Mahoney __wait_on_buffer(tbh); 1075278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 1076098297b2SJeff Mahoney /* 1077098297b2SJeff Mahoney * since we're using ll_rw_blk above, it might have skipped 1078098297b2SJeff Mahoney * over a locked buffer. Double check here 1079098297b2SJeff Mahoney */ 10808ebc4232SFrederic Weisbecker /* redundant, sync_dirty_buffer() checks */ 10818ebc4232SFrederic Weisbecker if (buffer_dirty(tbh)) { 1082278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(s); 10831da177e4SLinus Torvalds sync_dirty_buffer(tbh); 1084278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 10858ebc4232SFrederic Weisbecker } 10861da177e4SLinus Torvalds if (unlikely(!buffer_uptodate(tbh))) { 10871da177e4SLinus Torvalds #ifdef CONFIG_REISERFS_CHECK 108845b03d5eSJeff Mahoney reiserfs_warning(s, "journal-601", 108945b03d5eSJeff Mahoney "buffer write failed"); 10901da177e4SLinus Torvalds #endif 10911da177e4SLinus Torvalds retval = -EIO; 10921da177e4SLinus Torvalds } 1093098297b2SJeff Mahoney /* once for journal_find_get_block */ 1094098297b2SJeff Mahoney put_bh(tbh); 1095098297b2SJeff Mahoney /* once due to original getblk in do_journal_end */ 1096098297b2SJeff Mahoney put_bh(tbh); 1097a228bf8fSJeff Mahoney atomic_dec(&jl->j_commit_left); 10981da177e4SLinus Torvalds } 10991da177e4SLinus Torvalds 1100a228bf8fSJeff Mahoney BUG_ON(atomic_read(&jl->j_commit_left) != 1); 11011da177e4SLinus Torvalds 1102098297b2SJeff Mahoney /* 1103098297b2SJeff Mahoney * If there was a write error in the journal - we can't commit 11045d5e8156SJeff Mahoney * this transaction - it will be invalid and, if successful, 1105beb7dd86SRobert P. J. Day * will just end up propagating the write error out to 1106098297b2SJeff Mahoney * the file system. 1107098297b2SJeff Mahoney */ 11085d5e8156SJeff Mahoney if (likely(!retval && !reiserfs_is_journal_aborted (journal))) { 11091da177e4SLinus Torvalds if (buffer_dirty(jl->j_commit_bh)) 11101da177e4SLinus Torvalds BUG(); 11111da177e4SLinus Torvalds mark_buffer_dirty(jl->j_commit_bh) ; 1112278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(s); 11137cd33ad2SChristoph Hellwig if (reiserfs_barrier_flush(s)) 11147cd33ad2SChristoph Hellwig __sync_dirty_buffer(jl->j_commit_bh, WRITE_FLUSH_FUA); 11157cd33ad2SChristoph Hellwig else 11161da177e4SLinus Torvalds sync_dirty_buffer(jl->j_commit_bh); 1117278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 11185d5e8156SJeff Mahoney } 11191da177e4SLinus Torvalds 1120098297b2SJeff Mahoney /* 1121098297b2SJeff Mahoney * If there was a write error in the journal - we can't commit this 11221da177e4SLinus Torvalds * transaction - it will be invalid and, if successful, will just end 1123098297b2SJeff Mahoney * up propagating the write error out to the filesystem. 1124098297b2SJeff Mahoney */ 11251da177e4SLinus Torvalds if (unlikely(!buffer_uptodate(jl->j_commit_bh))) { 11261da177e4SLinus Torvalds #ifdef CONFIG_REISERFS_CHECK 112745b03d5eSJeff Mahoney reiserfs_warning(s, "journal-615", "buffer write failed"); 11281da177e4SLinus Torvalds #endif 11291da177e4SLinus Torvalds retval = -EIO; 11301da177e4SLinus Torvalds } 11311da177e4SLinus Torvalds bforget(jl->j_commit_bh); 11321da177e4SLinus Torvalds if (journal->j_last_commit_id != 0 && 11331da177e4SLinus Torvalds (jl->j_trans_id - journal->j_last_commit_id) != 1) { 113445b03d5eSJeff Mahoney reiserfs_warning(s, "clm-2200", "last commit %lu, current %lu", 1135bd4c625cSLinus Torvalds journal->j_last_commit_id, jl->j_trans_id); 11361da177e4SLinus Torvalds } 11371da177e4SLinus Torvalds journal->j_last_commit_id = jl->j_trans_id; 11381da177e4SLinus Torvalds 1139098297b2SJeff Mahoney /* 1140098297b2SJeff Mahoney * now, every commit block is on the disk. It is safe to allow 1141098297b2SJeff Mahoney * blocks freed during this transaction to be reallocated 1142098297b2SJeff Mahoney */ 11431da177e4SLinus Torvalds cleanup_freed_for_journal_list(s, jl); 11441da177e4SLinus Torvalds 11451da177e4SLinus Torvalds retval = retval ? retval : journal->j_errno; 11461da177e4SLinus Torvalds 11471da177e4SLinus Torvalds /* mark the metadata dirty */ 11481da177e4SLinus Torvalds if (!retval) 11491da177e4SLinus Torvalds dirty_one_transaction(s, jl); 1150a228bf8fSJeff Mahoney atomic_dec(&jl->j_commit_left); 11511da177e4SLinus Torvalds 11521da177e4SLinus Torvalds if (flushall) { 1153a228bf8fSJeff Mahoney atomic_set(&jl->j_older_commits_done, 1); 11541da177e4SLinus Torvalds } 115590415deaSJeff Mahoney mutex_unlock(&jl->j_commit_mutex); 11561da177e4SLinus Torvalds put_jl: 11571da177e4SLinus Torvalds put_journal_list(s, jl); 11581da177e4SLinus Torvalds 11591da177e4SLinus Torvalds if (retval) 1160bd4c625cSLinus Torvalds reiserfs_abort(s, retval, "Journal write error in %s", 1161fbe5498bSHarvey Harrison __func__); 11621da177e4SLinus Torvalds return retval; 11631da177e4SLinus Torvalds } 11641da177e4SLinus Torvalds 11651da177e4SLinus Torvalds /* 1166098297b2SJeff Mahoney * flush_journal_list frequently needs to find a newer transaction for a 1167098297b2SJeff Mahoney * given block. This does that, or returns NULL if it can't find anything 11681da177e4SLinus Torvalds */ 1169bd4c625cSLinus Torvalds static struct reiserfs_journal_list *find_newer_jl_for_cn(struct 1170bd4c625cSLinus Torvalds reiserfs_journal_cnode 1171bd4c625cSLinus Torvalds *cn) 1172bd4c625cSLinus Torvalds { 11731da177e4SLinus Torvalds struct super_block *sb = cn->sb; 11741da177e4SLinus Torvalds b_blocknr_t blocknr = cn->blocknr; 11751da177e4SLinus Torvalds 11761da177e4SLinus Torvalds cn = cn->hprev; 11771da177e4SLinus Torvalds while (cn) { 11781da177e4SLinus Torvalds if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) { 11791da177e4SLinus Torvalds return cn->jlist; 11801da177e4SLinus Torvalds } 11811da177e4SLinus Torvalds cn = cn->hprev; 11821da177e4SLinus Torvalds } 11831da177e4SLinus Torvalds return NULL; 11841da177e4SLinus Torvalds } 11851da177e4SLinus Torvalds 1186bd4c625cSLinus Torvalds static void remove_journal_hash(struct super_block *, 1187bd4c625cSLinus Torvalds struct reiserfs_journal_cnode **, 1188bd4c625cSLinus Torvalds struct reiserfs_journal_list *, unsigned long, 1189bd4c625cSLinus Torvalds int); 11901da177e4SLinus Torvalds 11911da177e4SLinus Torvalds /* 1192098297b2SJeff Mahoney * once all the real blocks have been flushed, it is safe to remove them 1193098297b2SJeff Mahoney * from the journal list for this transaction. Aside from freeing the 1194098297b2SJeff Mahoney * cnode, this also allows the block to be reallocated for data blocks 1195098297b2SJeff Mahoney * if it had been deleted. 11961da177e4SLinus Torvalds */ 1197a9dd3643SJeff Mahoney static void remove_all_from_journal_list(struct super_block *sb, 1198bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl, 1199bd4c625cSLinus Torvalds int debug) 1200bd4c625cSLinus Torvalds { 1201a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 12021da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn, *last; 12031da177e4SLinus Torvalds cn = jl->j_realblock; 12041da177e4SLinus Torvalds 1205098297b2SJeff Mahoney /* 1206098297b2SJeff Mahoney * which is better, to lock once around the whole loop, or 1207098297b2SJeff Mahoney * to lock for each call to remove_journal_hash? 12081da177e4SLinus Torvalds */ 12091da177e4SLinus Torvalds while (cn) { 12101da177e4SLinus Torvalds if (cn->blocknr != 0) { 12111da177e4SLinus Torvalds if (debug) { 1212a9dd3643SJeff Mahoney reiserfs_warning(sb, "reiserfs-2201", 1213bd4c625cSLinus Torvalds "block %u, bh is %d, state %ld", 1214bd4c625cSLinus Torvalds cn->blocknr, cn->bh ? 1 : 0, 1215bd4c625cSLinus Torvalds cn->state); 12161da177e4SLinus Torvalds } 12171da177e4SLinus Torvalds cn->state = 0; 1218a9dd3643SJeff Mahoney remove_journal_hash(sb, journal->j_list_hash_table, 1219bd4c625cSLinus Torvalds jl, cn->blocknr, 1); 12201da177e4SLinus Torvalds } 12211da177e4SLinus Torvalds last = cn; 12221da177e4SLinus Torvalds cn = cn->next; 1223a9dd3643SJeff Mahoney free_cnode(sb, last); 12241da177e4SLinus Torvalds } 12251da177e4SLinus Torvalds jl->j_realblock = NULL; 12261da177e4SLinus Torvalds } 12271da177e4SLinus Torvalds 12281da177e4SLinus Torvalds /* 1229098297b2SJeff Mahoney * if this timestamp is greater than the timestamp we wrote last to the 1230098297b2SJeff Mahoney * header block, write it to the header block. once this is done, I can 1231098297b2SJeff Mahoney * safely say the log area for this transaction won't ever be replayed, 1232098297b2SJeff Mahoney * and I can start releasing blocks in this transaction for reuse as data 1233098297b2SJeff Mahoney * blocks. called by flush_journal_list, before it calls 1234098297b2SJeff Mahoney * remove_all_from_journal_list 12351da177e4SLinus Torvalds */ 1236a9dd3643SJeff Mahoney static int _update_journal_header_block(struct super_block *sb, 1237bd4c625cSLinus Torvalds unsigned long offset, 1238600ed416SJeff Mahoney unsigned int trans_id) 1239bd4c625cSLinus Torvalds { 12401da177e4SLinus Torvalds struct reiserfs_journal_header *jh; 1241a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 1242278f6679SJeff Mahoney int depth; 12431da177e4SLinus Torvalds 12441da177e4SLinus Torvalds if (reiserfs_is_journal_aborted(journal)) 12451da177e4SLinus Torvalds return -EIO; 12461da177e4SLinus Torvalds 12471da177e4SLinus Torvalds if (trans_id >= journal->j_last_flush_trans_id) { 12481da177e4SLinus Torvalds if (buffer_locked((journal->j_header_bh))) { 1249278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(sb); 1250278f6679SJeff Mahoney __wait_on_buffer(journal->j_header_bh); 1251278f6679SJeff Mahoney reiserfs_write_lock_nested(sb, depth); 12521da177e4SLinus Torvalds if (unlikely(!buffer_uptodate(journal->j_header_bh))) { 12531da177e4SLinus Torvalds #ifdef CONFIG_REISERFS_CHECK 1254a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-699", 125545b03d5eSJeff Mahoney "buffer write failed"); 12561da177e4SLinus Torvalds #endif 12571da177e4SLinus Torvalds return -EIO; 12581da177e4SLinus Torvalds } 12591da177e4SLinus Torvalds } 12601da177e4SLinus Torvalds journal->j_last_flush_trans_id = trans_id; 12611da177e4SLinus Torvalds journal->j_first_unflushed_offset = offset; 1262bd4c625cSLinus Torvalds jh = (struct reiserfs_journal_header *)(journal->j_header_bh-> 1263bd4c625cSLinus Torvalds b_data); 12641da177e4SLinus Torvalds jh->j_last_flush_trans_id = cpu_to_le32(trans_id); 12651da177e4SLinus Torvalds jh->j_first_unflushed_offset = cpu_to_le32(offset); 12661da177e4SLinus Torvalds jh->j_mount_id = cpu_to_le32(journal->j_mount_id); 12671da177e4SLinus Torvalds 12681da177e4SLinus Torvalds set_buffer_dirty(journal->j_header_bh); 1269278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(sb); 12707cd33ad2SChristoph Hellwig 12717cd33ad2SChristoph Hellwig if (reiserfs_barrier_flush(sb)) 12727cd33ad2SChristoph Hellwig __sync_dirty_buffer(journal->j_header_bh, WRITE_FLUSH_FUA); 12737cd33ad2SChristoph Hellwig else 12741da177e4SLinus Torvalds sync_dirty_buffer(journal->j_header_bh); 12757cd33ad2SChristoph Hellwig 1276278f6679SJeff Mahoney reiserfs_write_lock_nested(sb, depth); 12771da177e4SLinus Torvalds if (!buffer_uptodate(journal->j_header_bh)) { 1278a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-837", 127945b03d5eSJeff Mahoney "IO error during journal replay"); 12801da177e4SLinus Torvalds return -EIO; 12811da177e4SLinus Torvalds } 12821da177e4SLinus Torvalds } 12831da177e4SLinus Torvalds return 0; 12841da177e4SLinus Torvalds } 12851da177e4SLinus Torvalds 1286a9dd3643SJeff Mahoney static int update_journal_header_block(struct super_block *sb, 12871da177e4SLinus Torvalds unsigned long offset, 1288600ed416SJeff Mahoney unsigned int trans_id) 1289bd4c625cSLinus Torvalds { 1290a9dd3643SJeff Mahoney return _update_journal_header_block(sb, offset, trans_id); 12911da177e4SLinus Torvalds } 1292bd4c625cSLinus Torvalds 12931da177e4SLinus Torvalds /* 12941da177e4SLinus Torvalds ** flush any and all journal lists older than you are 12951da177e4SLinus Torvalds ** can only be called from flush_journal_list 12961da177e4SLinus Torvalds */ 1297a9dd3643SJeff Mahoney static int flush_older_journal_lists(struct super_block *sb, 12981da177e4SLinus Torvalds struct reiserfs_journal_list *jl) 12991da177e4SLinus Torvalds { 13001da177e4SLinus Torvalds struct list_head *entry; 13011da177e4SLinus Torvalds struct reiserfs_journal_list *other_jl; 1302a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 1303600ed416SJeff Mahoney unsigned int trans_id = jl->j_trans_id; 13041da177e4SLinus Torvalds 1305098297b2SJeff Mahoney /* 1306098297b2SJeff Mahoney * we know we are the only ones flushing things, no extra race 13071da177e4SLinus Torvalds * protection is required. 13081da177e4SLinus Torvalds */ 13091da177e4SLinus Torvalds restart: 13101da177e4SLinus Torvalds entry = journal->j_journal_list.next; 13111da177e4SLinus Torvalds /* Did we wrap? */ 13121da177e4SLinus Torvalds if (entry == &journal->j_journal_list) 13131da177e4SLinus Torvalds return 0; 13141da177e4SLinus Torvalds other_jl = JOURNAL_LIST_ENTRY(entry); 13151da177e4SLinus Torvalds if (other_jl->j_trans_id < trans_id) { 13161da177e4SLinus Torvalds BUG_ON(other_jl->j_refcount <= 0); 13171da177e4SLinus Torvalds /* do not flush all */ 1318a9dd3643SJeff Mahoney flush_journal_list(sb, other_jl, 0); 13191da177e4SLinus Torvalds 13201da177e4SLinus Torvalds /* other_jl is now deleted from the list */ 13211da177e4SLinus Torvalds goto restart; 13221da177e4SLinus Torvalds } 13231da177e4SLinus Torvalds return 0; 13241da177e4SLinus Torvalds } 13251da177e4SLinus Torvalds 13261da177e4SLinus Torvalds static void del_from_work_list(struct super_block *s, 1327bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl) 1328bd4c625cSLinus Torvalds { 13291da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 13301da177e4SLinus Torvalds if (!list_empty(&jl->j_working_list)) { 13311da177e4SLinus Torvalds list_del_init(&jl->j_working_list); 13321da177e4SLinus Torvalds journal->j_num_work_lists--; 13331da177e4SLinus Torvalds } 13341da177e4SLinus Torvalds } 13351da177e4SLinus Torvalds 1336098297b2SJeff Mahoney /* 1337098297b2SJeff Mahoney * flush a journal list, both commit and real blocks 1338098297b2SJeff Mahoney * 1339098297b2SJeff Mahoney * always set flushall to 1, unless you are calling from inside 1340098297b2SJeff Mahoney * flush_journal_list 1341098297b2SJeff Mahoney * 1342098297b2SJeff Mahoney * IMPORTANT. This can only be called while there are no journal writers, 1343098297b2SJeff Mahoney * and the journal is locked. That means it can only be called from 1344098297b2SJeff Mahoney * do_journal_end, or by journal_release 13451da177e4SLinus Torvalds */ 13461da177e4SLinus Torvalds static int flush_journal_list(struct super_block *s, 1347bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl, int flushall) 1348bd4c625cSLinus Torvalds { 13491da177e4SLinus Torvalds struct reiserfs_journal_list *pjl; 13501da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn, *last; 13511da177e4SLinus Torvalds int count; 13521da177e4SLinus Torvalds int was_jwait = 0; 13531da177e4SLinus Torvalds int was_dirty = 0; 13541da177e4SLinus Torvalds struct buffer_head *saved_bh; 13551da177e4SLinus Torvalds unsigned long j_len_saved = jl->j_len; 13561da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 13571da177e4SLinus Torvalds int err = 0; 1358278f6679SJeff Mahoney int depth; 13591da177e4SLinus Torvalds 13601da177e4SLinus Torvalds BUG_ON(j_len_saved <= 0); 13611da177e4SLinus Torvalds 13621da177e4SLinus Torvalds if (atomic_read(&journal->j_wcount) != 0) { 136345b03d5eSJeff Mahoney reiserfs_warning(s, "clm-2048", "called with wcount %d", 13641da177e4SLinus Torvalds atomic_read(&journal->j_wcount)); 13651da177e4SLinus Torvalds } 13661da177e4SLinus Torvalds 13671da177e4SLinus Torvalds /* if flushall == 0, the lock is already held */ 13681da177e4SLinus Torvalds if (flushall) { 13698ebc4232SFrederic Weisbecker reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s); 1370afe70259SJeff Mahoney } else if (mutex_trylock(&journal->j_flush_mutex)) { 13711da177e4SLinus Torvalds BUG(); 13721da177e4SLinus Torvalds } 13731da177e4SLinus Torvalds 13741da177e4SLinus Torvalds count = 0; 13751da177e4SLinus Torvalds if (j_len_saved > journal->j_trans_max) { 1376c3a9c210SJeff Mahoney reiserfs_panic(s, "journal-715", "length is %lu, trans id %lu", 1377bd4c625cSLinus Torvalds j_len_saved, jl->j_trans_id); 13781da177e4SLinus Torvalds return 0; 13791da177e4SLinus Torvalds } 13801da177e4SLinus Torvalds 13811da177e4SLinus Torvalds /* if all the work is already done, get out of here */ 1382a228bf8fSJeff Mahoney if (atomic_read(&jl->j_nonzerolen) <= 0 && 1383a228bf8fSJeff Mahoney atomic_read(&jl->j_commit_left) <= 0) { 13841da177e4SLinus Torvalds goto flush_older_and_return; 13851da177e4SLinus Torvalds } 13861da177e4SLinus Torvalds 1387098297b2SJeff Mahoney /* 1388098297b2SJeff Mahoney * start by putting the commit list on disk. This will also flush 1389098297b2SJeff Mahoney * the commit lists of any olders transactions 13901da177e4SLinus Torvalds */ 13911da177e4SLinus Torvalds flush_commit_list(s, jl, 1); 13921da177e4SLinus Torvalds 1393bd4c625cSLinus Torvalds if (!(jl->j_state & LIST_DIRTY) 1394bd4c625cSLinus Torvalds && !reiserfs_is_journal_aborted(journal)) 13951da177e4SLinus Torvalds BUG(); 13961da177e4SLinus Torvalds 13971da177e4SLinus Torvalds /* are we done now? */ 1398a228bf8fSJeff Mahoney if (atomic_read(&jl->j_nonzerolen) <= 0 && 1399a228bf8fSJeff Mahoney atomic_read(&jl->j_commit_left) <= 0) { 14001da177e4SLinus Torvalds goto flush_older_and_return; 14011da177e4SLinus Torvalds } 14021da177e4SLinus Torvalds 1403098297b2SJeff Mahoney /* 1404098297b2SJeff Mahoney * loop through each cnode, see if we need to write it, 1405098297b2SJeff Mahoney * or wait on a more recent transaction, or just ignore it 14061da177e4SLinus Torvalds */ 1407a228bf8fSJeff Mahoney if (atomic_read(&journal->j_wcount) != 0) { 1408c3a9c210SJeff Mahoney reiserfs_panic(s, "journal-844", "journal list is flushing, " 1409c3a9c210SJeff Mahoney "wcount is not 0"); 14101da177e4SLinus Torvalds } 14111da177e4SLinus Torvalds cn = jl->j_realblock; 14121da177e4SLinus Torvalds while (cn) { 14131da177e4SLinus Torvalds was_jwait = 0; 14141da177e4SLinus Torvalds was_dirty = 0; 14151da177e4SLinus Torvalds saved_bh = NULL; 14161da177e4SLinus Torvalds /* blocknr of 0 is no longer in the hash, ignore it */ 14171da177e4SLinus Torvalds if (cn->blocknr == 0) { 14181da177e4SLinus Torvalds goto free_cnode; 14191da177e4SLinus Torvalds } 14201da177e4SLinus Torvalds 1421098297b2SJeff Mahoney /* 1422098297b2SJeff Mahoney * This transaction failed commit. 1423098297b2SJeff Mahoney * Don't write out to the disk 1424098297b2SJeff Mahoney */ 14251da177e4SLinus Torvalds if (!(jl->j_state & LIST_DIRTY)) 14261da177e4SLinus Torvalds goto free_cnode; 14271da177e4SLinus Torvalds 14281da177e4SLinus Torvalds pjl = find_newer_jl_for_cn(cn); 1429098297b2SJeff Mahoney /* 1430098297b2SJeff Mahoney * the order is important here. We check pjl to make sure we 1431098297b2SJeff Mahoney * don't clear BH_JDirty_wait if we aren't the one writing this 1432098297b2SJeff Mahoney * block to disk 14331da177e4SLinus Torvalds */ 14341da177e4SLinus Torvalds if (!pjl && cn->bh) { 14351da177e4SLinus Torvalds saved_bh = cn->bh; 14361da177e4SLinus Torvalds 1437098297b2SJeff Mahoney /* 1438098297b2SJeff Mahoney * we do this to make sure nobody releases the 1439098297b2SJeff Mahoney * buffer while we are working with it 14401da177e4SLinus Torvalds */ 14411da177e4SLinus Torvalds get_bh(saved_bh); 14421da177e4SLinus Torvalds 14431da177e4SLinus Torvalds if (buffer_journal_dirty(saved_bh)) { 14441da177e4SLinus Torvalds BUG_ON(!can_dirty(cn)); 14451da177e4SLinus Torvalds was_jwait = 1; 14461da177e4SLinus Torvalds was_dirty = 1; 14471da177e4SLinus Torvalds } else if (can_dirty(cn)) { 1448098297b2SJeff Mahoney /* 1449098297b2SJeff Mahoney * everything with !pjl && jwait 1450098297b2SJeff Mahoney * should be writable 1451098297b2SJeff Mahoney */ 14521da177e4SLinus Torvalds BUG(); 14531da177e4SLinus Torvalds } 14541da177e4SLinus Torvalds } 14551da177e4SLinus Torvalds 1456098297b2SJeff Mahoney /* 1457098297b2SJeff Mahoney * if someone has this block in a newer transaction, just make 1458098297b2SJeff Mahoney * sure they are committed, and don't try writing it to disk 14591da177e4SLinus Torvalds */ 14601da177e4SLinus Torvalds if (pjl) { 14611da177e4SLinus Torvalds if (atomic_read(&pjl->j_commit_left)) 14621da177e4SLinus Torvalds flush_commit_list(s, pjl, 1); 14631da177e4SLinus Torvalds goto free_cnode; 14641da177e4SLinus Torvalds } 14651da177e4SLinus Torvalds 1466098297b2SJeff Mahoney /* 1467098297b2SJeff Mahoney * bh == NULL when the block got to disk on its own, OR, 1468098297b2SJeff Mahoney * the block got freed in a future transaction 14691da177e4SLinus Torvalds */ 14701da177e4SLinus Torvalds if (saved_bh == NULL) { 14711da177e4SLinus Torvalds goto free_cnode; 14721da177e4SLinus Torvalds } 14731da177e4SLinus Torvalds 1474098297b2SJeff Mahoney /* 1475098297b2SJeff Mahoney * this should never happen. kupdate_one_transaction has 1476098297b2SJeff Mahoney * this list locked while it works, so we should never see a 1477098297b2SJeff Mahoney * buffer here that is not marked JDirty_wait 14781da177e4SLinus Torvalds */ 14791da177e4SLinus Torvalds if ((!was_jwait) && !buffer_locked(saved_bh)) { 148045b03d5eSJeff Mahoney reiserfs_warning(s, "journal-813", 148145b03d5eSJeff Mahoney "BAD! buffer %llu %cdirty %cjwait, " 14821da177e4SLinus Torvalds "not in a newer tranasction", 1483bd4c625cSLinus Torvalds (unsigned long long)saved_bh-> 1484bd4c625cSLinus Torvalds b_blocknr, was_dirty ? ' ' : '!', 1485bd4c625cSLinus Torvalds was_jwait ? ' ' : '!'); 14861da177e4SLinus Torvalds } 14871da177e4SLinus Torvalds if (was_dirty) { 1488098297b2SJeff Mahoney /* 1489098297b2SJeff Mahoney * we inc again because saved_bh gets decremented 1490098297b2SJeff Mahoney * at free_cnode 1491098297b2SJeff Mahoney */ 14921da177e4SLinus Torvalds get_bh(saved_bh); 14931da177e4SLinus Torvalds set_bit(BLOCK_NEEDS_FLUSH, &cn->state); 14941da177e4SLinus Torvalds lock_buffer(saved_bh); 14951da177e4SLinus Torvalds BUG_ON(cn->blocknr != saved_bh->b_blocknr); 14961da177e4SLinus Torvalds if (buffer_dirty(saved_bh)) 14971da177e4SLinus Torvalds submit_logged_buffer(saved_bh); 14981da177e4SLinus Torvalds else 14991da177e4SLinus Torvalds unlock_buffer(saved_bh); 15001da177e4SLinus Torvalds count++; 15011da177e4SLinus Torvalds } else { 150245b03d5eSJeff Mahoney reiserfs_warning(s, "clm-2082", 150345b03d5eSJeff Mahoney "Unable to flush buffer %llu in %s", 1504bd4c625cSLinus Torvalds (unsigned long long)saved_bh-> 1505fbe5498bSHarvey Harrison b_blocknr, __func__); 15061da177e4SLinus Torvalds } 15071da177e4SLinus Torvalds free_cnode: 15081da177e4SLinus Torvalds last = cn; 15091da177e4SLinus Torvalds cn = cn->next; 15101da177e4SLinus Torvalds if (saved_bh) { 1511098297b2SJeff Mahoney /* 1512098297b2SJeff Mahoney * we incremented this to keep others from 1513098297b2SJeff Mahoney * taking the buffer head away 1514098297b2SJeff Mahoney */ 15151da177e4SLinus Torvalds put_bh(saved_bh); 1516a228bf8fSJeff Mahoney if (atomic_read(&saved_bh->b_count) < 0) { 151745b03d5eSJeff Mahoney reiserfs_warning(s, "journal-945", 151845b03d5eSJeff Mahoney "saved_bh->b_count < 0"); 15191da177e4SLinus Torvalds } 15201da177e4SLinus Torvalds } 15211da177e4SLinus Torvalds } 15221da177e4SLinus Torvalds if (count > 0) { 15231da177e4SLinus Torvalds cn = jl->j_realblock; 15241da177e4SLinus Torvalds while (cn) { 15251da177e4SLinus Torvalds if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) { 15261da177e4SLinus Torvalds if (!cn->bh) { 1527c3a9c210SJeff Mahoney reiserfs_panic(s, "journal-1011", 1528c3a9c210SJeff Mahoney "cn->bh is NULL"); 15291da177e4SLinus Torvalds } 15308ebc4232SFrederic Weisbecker 1531278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(s); 1532278f6679SJeff Mahoney __wait_on_buffer(cn->bh); 1533278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 15348ebc4232SFrederic Weisbecker 15351da177e4SLinus Torvalds if (!cn->bh) { 1536c3a9c210SJeff Mahoney reiserfs_panic(s, "journal-1012", 1537c3a9c210SJeff Mahoney "cn->bh is NULL"); 15381da177e4SLinus Torvalds } 15391da177e4SLinus Torvalds if (unlikely(!buffer_uptodate(cn->bh))) { 15401da177e4SLinus Torvalds #ifdef CONFIG_REISERFS_CHECK 154145b03d5eSJeff Mahoney reiserfs_warning(s, "journal-949", 154245b03d5eSJeff Mahoney "buffer write failed"); 15431da177e4SLinus Torvalds #endif 15441da177e4SLinus Torvalds err = -EIO; 15451da177e4SLinus Torvalds } 1546098297b2SJeff Mahoney /* 1547098297b2SJeff Mahoney * note, we must clear the JDirty_wait bit 1548098297b2SJeff Mahoney * after the up to date check, otherwise we 1549098297b2SJeff Mahoney * race against our flushpage routine 15501da177e4SLinus Torvalds */ 1551bd4c625cSLinus Torvalds BUG_ON(!test_clear_buffer_journal_dirty 1552bd4c625cSLinus Torvalds (cn->bh)); 15531da177e4SLinus Torvalds 1554398c95bdSChris Mason /* drop one ref for us */ 15551da177e4SLinus Torvalds put_bh(cn->bh); 1556398c95bdSChris Mason /* drop one ref for journal_mark_dirty */ 1557398c95bdSChris Mason release_buffer_page(cn->bh); 15581da177e4SLinus Torvalds } 15591da177e4SLinus Torvalds cn = cn->next; 15601da177e4SLinus Torvalds } 15611da177e4SLinus Torvalds } 15621da177e4SLinus Torvalds 15631da177e4SLinus Torvalds if (err) 1564bd4c625cSLinus Torvalds reiserfs_abort(s, -EIO, 1565bd4c625cSLinus Torvalds "Write error while pushing transaction to disk in %s", 1566fbe5498bSHarvey Harrison __func__); 15671da177e4SLinus Torvalds flush_older_and_return: 15681da177e4SLinus Torvalds 1569098297b2SJeff Mahoney /* 1570098297b2SJeff Mahoney * before we can update the journal header block, we _must_ flush all 1571098297b2SJeff Mahoney * real blocks from all older transactions to disk. This is because 1572098297b2SJeff Mahoney * once the header block is updated, this transaction will not be 1573098297b2SJeff Mahoney * replayed after a crash 15741da177e4SLinus Torvalds */ 15751da177e4SLinus Torvalds if (flushall) { 15761da177e4SLinus Torvalds flush_older_journal_lists(s, jl); 15771da177e4SLinus Torvalds } 15781da177e4SLinus Torvalds 15791da177e4SLinus Torvalds err = journal->j_errno; 1580098297b2SJeff Mahoney /* 1581098297b2SJeff Mahoney * before we can remove everything from the hash tables for this 1582098297b2SJeff Mahoney * transaction, we must make sure it can never be replayed 1583098297b2SJeff Mahoney * 1584098297b2SJeff Mahoney * since we are only called from do_journal_end, we know for sure there 1585098297b2SJeff Mahoney * are no allocations going on while we are flushing journal lists. So, 1586098297b2SJeff Mahoney * we only need to update the journal header block for the last list 1587098297b2SJeff Mahoney * being flushed 15881da177e4SLinus Torvalds */ 15891da177e4SLinus Torvalds if (!err && flushall) { 1590bd4c625cSLinus Torvalds err = 1591bd4c625cSLinus Torvalds update_journal_header_block(s, 1592bd4c625cSLinus Torvalds (jl->j_start + jl->j_len + 1593bd4c625cSLinus Torvalds 2) % SB_ONDISK_JOURNAL_SIZE(s), 1594bd4c625cSLinus Torvalds jl->j_trans_id); 15951da177e4SLinus Torvalds if (err) 1596bd4c625cSLinus Torvalds reiserfs_abort(s, -EIO, 1597bd4c625cSLinus Torvalds "Write error while updating journal header in %s", 1598fbe5498bSHarvey Harrison __func__); 15991da177e4SLinus Torvalds } 16001da177e4SLinus Torvalds remove_all_from_journal_list(s, jl, 0); 16011da177e4SLinus Torvalds list_del_init(&jl->j_list); 16021da177e4SLinus Torvalds journal->j_num_lists--; 16031da177e4SLinus Torvalds del_from_work_list(s, jl); 16041da177e4SLinus Torvalds 16051da177e4SLinus Torvalds if (journal->j_last_flush_id != 0 && 16061da177e4SLinus Torvalds (jl->j_trans_id - journal->j_last_flush_id) != 1) { 160745b03d5eSJeff Mahoney reiserfs_warning(s, "clm-2201", "last flush %lu, current %lu", 1608bd4c625cSLinus Torvalds journal->j_last_flush_id, jl->j_trans_id); 16091da177e4SLinus Torvalds } 16101da177e4SLinus Torvalds journal->j_last_flush_id = jl->j_trans_id; 16111da177e4SLinus Torvalds 1612098297b2SJeff Mahoney /* 1613098297b2SJeff Mahoney * not strictly required since we are freeing the list, but it should 16141da177e4SLinus Torvalds * help find code using dead lists later on 16151da177e4SLinus Torvalds */ 16161da177e4SLinus Torvalds jl->j_len = 0; 1617a228bf8fSJeff Mahoney atomic_set(&jl->j_nonzerolen, 0); 16181da177e4SLinus Torvalds jl->j_start = 0; 16191da177e4SLinus Torvalds jl->j_realblock = NULL; 16201da177e4SLinus Torvalds jl->j_commit_bh = NULL; 16211da177e4SLinus Torvalds jl->j_trans_id = 0; 16221da177e4SLinus Torvalds jl->j_state = 0; 16231da177e4SLinus Torvalds put_journal_list(s, jl); 16241da177e4SLinus Torvalds if (flushall) 1625afe70259SJeff Mahoney mutex_unlock(&journal->j_flush_mutex); 16261da177e4SLinus Torvalds return err; 16271da177e4SLinus Torvalds } 16281da177e4SLinus Torvalds 16291da177e4SLinus Torvalds static int write_one_transaction(struct super_block *s, 16301da177e4SLinus Torvalds struct reiserfs_journal_list *jl, 16311da177e4SLinus Torvalds struct buffer_chunk *chunk) 16321da177e4SLinus Torvalds { 16331da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 16341da177e4SLinus Torvalds int ret = 0; 16351da177e4SLinus Torvalds 16361da177e4SLinus Torvalds jl->j_state |= LIST_TOUCHED; 16371da177e4SLinus Torvalds del_from_work_list(s, jl); 16381da177e4SLinus Torvalds if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) { 16391da177e4SLinus Torvalds return 0; 16401da177e4SLinus Torvalds } 16411da177e4SLinus Torvalds 16421da177e4SLinus Torvalds cn = jl->j_realblock; 16431da177e4SLinus Torvalds while (cn) { 1644098297b2SJeff Mahoney /* 1645098297b2SJeff Mahoney * if the blocknr == 0, this has been cleared from the hash, 1646098297b2SJeff Mahoney * skip it 16471da177e4SLinus Torvalds */ 16481da177e4SLinus Torvalds if (cn->blocknr == 0) { 16491da177e4SLinus Torvalds goto next; 16501da177e4SLinus Torvalds } 16511da177e4SLinus Torvalds if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) { 16521da177e4SLinus Torvalds struct buffer_head *tmp_bh; 1653098297b2SJeff Mahoney /* 1654098297b2SJeff Mahoney * we can race against journal_mark_freed when we try 16551da177e4SLinus Torvalds * to lock_buffer(cn->bh), so we have to inc the buffer 16561da177e4SLinus Torvalds * count, and recheck things after locking 16571da177e4SLinus Torvalds */ 16581da177e4SLinus Torvalds tmp_bh = cn->bh; 16591da177e4SLinus Torvalds get_bh(tmp_bh); 16601da177e4SLinus Torvalds lock_buffer(tmp_bh); 16611da177e4SLinus Torvalds if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) { 16621da177e4SLinus Torvalds if (!buffer_journal_dirty(tmp_bh) || 16631da177e4SLinus Torvalds buffer_journal_prepared(tmp_bh)) 16641da177e4SLinus Torvalds BUG(); 16651da177e4SLinus Torvalds add_to_chunk(chunk, tmp_bh, NULL, write_chunk); 16661da177e4SLinus Torvalds ret++; 16671da177e4SLinus Torvalds } else { 16681da177e4SLinus Torvalds /* note, cn->bh might be null now */ 16691da177e4SLinus Torvalds unlock_buffer(tmp_bh); 16701da177e4SLinus Torvalds } 16711da177e4SLinus Torvalds put_bh(tmp_bh); 16721da177e4SLinus Torvalds } 16731da177e4SLinus Torvalds next: 16741da177e4SLinus Torvalds cn = cn->next; 16751da177e4SLinus Torvalds cond_resched(); 16761da177e4SLinus Torvalds } 16771da177e4SLinus Torvalds return ret; 16781da177e4SLinus Torvalds } 16791da177e4SLinus Torvalds 16801da177e4SLinus Torvalds /* used by flush_commit_list */ 16811da177e4SLinus Torvalds static int dirty_one_transaction(struct super_block *s, 16821da177e4SLinus Torvalds struct reiserfs_journal_list *jl) 16831da177e4SLinus Torvalds { 16841da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 16851da177e4SLinus Torvalds struct reiserfs_journal_list *pjl; 16861da177e4SLinus Torvalds int ret = 0; 16871da177e4SLinus Torvalds 16881da177e4SLinus Torvalds jl->j_state |= LIST_DIRTY; 16891da177e4SLinus Torvalds cn = jl->j_realblock; 16901da177e4SLinus Torvalds while (cn) { 1691098297b2SJeff Mahoney /* 1692098297b2SJeff Mahoney * look for a more recent transaction that logged this 1693098297b2SJeff Mahoney * buffer. Only the most recent transaction with a buffer in 1694098297b2SJeff Mahoney * it is allowed to send that buffer to disk 16951da177e4SLinus Torvalds */ 16961da177e4SLinus Torvalds pjl = find_newer_jl_for_cn(cn); 1697bd4c625cSLinus Torvalds if (!pjl && cn->blocknr && cn->bh 1698bd4c625cSLinus Torvalds && buffer_journal_dirty(cn->bh)) { 16991da177e4SLinus Torvalds BUG_ON(!can_dirty(cn)); 1700098297b2SJeff Mahoney /* 1701098297b2SJeff Mahoney * if the buffer is prepared, it will either be logged 17021da177e4SLinus Torvalds * or restored. If restored, we need to make sure 17031da177e4SLinus Torvalds * it actually gets marked dirty 17041da177e4SLinus Torvalds */ 17051da177e4SLinus Torvalds clear_buffer_journal_new(cn->bh); 17061da177e4SLinus Torvalds if (buffer_journal_prepared(cn->bh)) { 17071da177e4SLinus Torvalds set_buffer_journal_restore_dirty(cn->bh); 17081da177e4SLinus Torvalds } else { 17091da177e4SLinus Torvalds set_buffer_journal_test(cn->bh); 17101da177e4SLinus Torvalds mark_buffer_dirty(cn->bh); 17111da177e4SLinus Torvalds } 17121da177e4SLinus Torvalds } 17131da177e4SLinus Torvalds cn = cn->next; 17141da177e4SLinus Torvalds } 17151da177e4SLinus Torvalds return ret; 17161da177e4SLinus Torvalds } 17171da177e4SLinus Torvalds 17181da177e4SLinus Torvalds static int kupdate_transactions(struct super_block *s, 17191da177e4SLinus Torvalds struct reiserfs_journal_list *jl, 17201da177e4SLinus Torvalds struct reiserfs_journal_list **next_jl, 1721600ed416SJeff Mahoney unsigned int *next_trans_id, 1722bd4c625cSLinus Torvalds int num_blocks, int num_trans) 1723bd4c625cSLinus Torvalds { 17241da177e4SLinus Torvalds int ret = 0; 17251da177e4SLinus Torvalds int written = 0; 17261da177e4SLinus Torvalds int transactions_flushed = 0; 1727600ed416SJeff Mahoney unsigned int orig_trans_id = jl->j_trans_id; 17281da177e4SLinus Torvalds struct buffer_chunk chunk; 17291da177e4SLinus Torvalds struct list_head *entry; 17301da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 17311da177e4SLinus Torvalds chunk.nr = 0; 17321da177e4SLinus Torvalds 1733a412f9efSFrederic Weisbecker reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s); 17341da177e4SLinus Torvalds if (!journal_list_still_alive(s, orig_trans_id)) { 17351da177e4SLinus Torvalds goto done; 17361da177e4SLinus Torvalds } 17371da177e4SLinus Torvalds 1738098297b2SJeff Mahoney /* 1739098297b2SJeff Mahoney * we've got j_flush_mutex held, nobody is going to delete any 17401da177e4SLinus Torvalds * of these lists out from underneath us 17411da177e4SLinus Torvalds */ 17421da177e4SLinus Torvalds while ((num_trans && transactions_flushed < num_trans) || 17431da177e4SLinus Torvalds (!num_trans && written < num_blocks)) { 17441da177e4SLinus Torvalds 17451da177e4SLinus Torvalds if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) || 1746bd4c625cSLinus Torvalds atomic_read(&jl->j_commit_left) 1747bd4c625cSLinus Torvalds || !(jl->j_state & LIST_DIRTY)) { 17481da177e4SLinus Torvalds del_from_work_list(s, jl); 17491da177e4SLinus Torvalds break; 17501da177e4SLinus Torvalds } 17511da177e4SLinus Torvalds ret = write_one_transaction(s, jl, &chunk); 17521da177e4SLinus Torvalds 17531da177e4SLinus Torvalds if (ret < 0) 17541da177e4SLinus Torvalds goto done; 17551da177e4SLinus Torvalds transactions_flushed++; 17561da177e4SLinus Torvalds written += ret; 17571da177e4SLinus Torvalds entry = jl->j_list.next; 17581da177e4SLinus Torvalds 17591da177e4SLinus Torvalds /* did we wrap? */ 17601da177e4SLinus Torvalds if (entry == &journal->j_journal_list) { 17611da177e4SLinus Torvalds break; 17621da177e4SLinus Torvalds } 17631da177e4SLinus Torvalds jl = JOURNAL_LIST_ENTRY(entry); 17641da177e4SLinus Torvalds 17651da177e4SLinus Torvalds /* don't bother with older transactions */ 17661da177e4SLinus Torvalds if (jl->j_trans_id <= orig_trans_id) 17671da177e4SLinus Torvalds break; 17681da177e4SLinus Torvalds } 17691da177e4SLinus Torvalds if (chunk.nr) { 17701da177e4SLinus Torvalds write_chunk(&chunk); 17711da177e4SLinus Torvalds } 17721da177e4SLinus Torvalds 17731da177e4SLinus Torvalds done: 1774afe70259SJeff Mahoney mutex_unlock(&journal->j_flush_mutex); 17751da177e4SLinus Torvalds return ret; 17761da177e4SLinus Torvalds } 17771da177e4SLinus Torvalds 1778098297b2SJeff Mahoney /* 1779098297b2SJeff Mahoney * for o_sync and fsync heavy applications, they tend to use 1780098297b2SJeff Mahoney * all the journa list slots with tiny transactions. These 1781098297b2SJeff Mahoney * trigger lots and lots of calls to update the header block, which 1782098297b2SJeff Mahoney * adds seeks and slows things down. 1783098297b2SJeff Mahoney * 1784098297b2SJeff Mahoney * This function tries to clear out a large chunk of the journal lists 1785098297b2SJeff Mahoney * at once, which makes everything faster since only the newest journal 1786098297b2SJeff Mahoney * list updates the header block 17871da177e4SLinus Torvalds */ 17881da177e4SLinus Torvalds static int flush_used_journal_lists(struct super_block *s, 1789bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl) 1790bd4c625cSLinus Torvalds { 17911da177e4SLinus Torvalds unsigned long len = 0; 17921da177e4SLinus Torvalds unsigned long cur_len; 17931da177e4SLinus Torvalds int ret; 17941da177e4SLinus Torvalds int i; 17951da177e4SLinus Torvalds int limit = 256; 17961da177e4SLinus Torvalds struct reiserfs_journal_list *tjl; 17971da177e4SLinus Torvalds struct reiserfs_journal_list *flush_jl; 1798600ed416SJeff Mahoney unsigned int trans_id; 17991da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 18001da177e4SLinus Torvalds 18011da177e4SLinus Torvalds flush_jl = tjl = jl; 18021da177e4SLinus Torvalds 18031da177e4SLinus Torvalds /* in data logging mode, try harder to flush a lot of blocks */ 18041da177e4SLinus Torvalds if (reiserfs_data_log(s)) 18051da177e4SLinus Torvalds limit = 1024; 18061da177e4SLinus Torvalds /* flush for 256 transactions or limit blocks, whichever comes first */ 18071da177e4SLinus Torvalds for (i = 0; i < 256 && len < limit; i++) { 18081da177e4SLinus Torvalds if (atomic_read(&tjl->j_commit_left) || 18091da177e4SLinus Torvalds tjl->j_trans_id < jl->j_trans_id) { 18101da177e4SLinus Torvalds break; 18111da177e4SLinus Torvalds } 18121da177e4SLinus Torvalds cur_len = atomic_read(&tjl->j_nonzerolen); 18131da177e4SLinus Torvalds if (cur_len > 0) { 18141da177e4SLinus Torvalds tjl->j_state &= ~LIST_TOUCHED; 18151da177e4SLinus Torvalds } 18161da177e4SLinus Torvalds len += cur_len; 18171da177e4SLinus Torvalds flush_jl = tjl; 18181da177e4SLinus Torvalds if (tjl->j_list.next == &journal->j_journal_list) 18191da177e4SLinus Torvalds break; 18201da177e4SLinus Torvalds tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next); 18211da177e4SLinus Torvalds } 1822721a769cSJeff Mahoney get_journal_list(jl); 1823721a769cSJeff Mahoney get_journal_list(flush_jl); 1824098297b2SJeff Mahoney 1825098297b2SJeff Mahoney /* 1826098297b2SJeff Mahoney * try to find a group of blocks we can flush across all the 1827098297b2SJeff Mahoney * transactions, but only bother if we've actually spanned 1828098297b2SJeff Mahoney * across multiple lists 18291da177e4SLinus Torvalds */ 18301da177e4SLinus Torvalds if (flush_jl != jl) { 18311da177e4SLinus Torvalds ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i); 18321da177e4SLinus Torvalds } 18331da177e4SLinus Torvalds flush_journal_list(s, flush_jl, 1); 1834721a769cSJeff Mahoney put_journal_list(s, flush_jl); 1835721a769cSJeff Mahoney put_journal_list(s, jl); 18361da177e4SLinus Torvalds return 0; 18371da177e4SLinus Torvalds } 18381da177e4SLinus Torvalds 18391da177e4SLinus Torvalds /* 1840098297b2SJeff Mahoney * removes any nodes in table with name block and dev as bh. 1841098297b2SJeff Mahoney * only touchs the hnext and hprev pointers. 18421da177e4SLinus Torvalds */ 18431da177e4SLinus Torvalds void remove_journal_hash(struct super_block *sb, 18441da177e4SLinus Torvalds struct reiserfs_journal_cnode **table, 18451da177e4SLinus Torvalds struct reiserfs_journal_list *jl, 18461da177e4SLinus Torvalds unsigned long block, int remove_freed) 18471da177e4SLinus Torvalds { 18481da177e4SLinus Torvalds struct reiserfs_journal_cnode *cur; 18491da177e4SLinus Torvalds struct reiserfs_journal_cnode **head; 18501da177e4SLinus Torvalds 18511da177e4SLinus Torvalds head = &(journal_hash(table, sb, block)); 18521da177e4SLinus Torvalds if (!head) { 18531da177e4SLinus Torvalds return; 18541da177e4SLinus Torvalds } 18551da177e4SLinus Torvalds cur = *head; 18561da177e4SLinus Torvalds while (cur) { 1857bd4c625cSLinus Torvalds if (cur->blocknr == block && cur->sb == sb 1858bd4c625cSLinus Torvalds && (jl == NULL || jl == cur->jlist) 1859bd4c625cSLinus Torvalds && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) { 18601da177e4SLinus Torvalds if (cur->hnext) { 18611da177e4SLinus Torvalds cur->hnext->hprev = cur->hprev; 18621da177e4SLinus Torvalds } 18631da177e4SLinus Torvalds if (cur->hprev) { 18641da177e4SLinus Torvalds cur->hprev->hnext = cur->hnext; 18651da177e4SLinus Torvalds } else { 18661da177e4SLinus Torvalds *head = cur->hnext; 18671da177e4SLinus Torvalds } 18681da177e4SLinus Torvalds cur->blocknr = 0; 18691da177e4SLinus Torvalds cur->sb = NULL; 18701da177e4SLinus Torvalds cur->state = 0; 1871098297b2SJeff Mahoney /* 1872098297b2SJeff Mahoney * anybody who clears the cur->bh will also 1873098297b2SJeff Mahoney * dec the nonzerolen 1874098297b2SJeff Mahoney */ 1875098297b2SJeff Mahoney if (cur->bh && cur->jlist) 1876a228bf8fSJeff Mahoney atomic_dec(&cur->jlist->j_nonzerolen); 18771da177e4SLinus Torvalds cur->bh = NULL; 18781da177e4SLinus Torvalds cur->jlist = NULL; 18791da177e4SLinus Torvalds } 18801da177e4SLinus Torvalds cur = cur->hnext; 18811da177e4SLinus Torvalds } 18821da177e4SLinus Torvalds } 18831da177e4SLinus Torvalds 1884a9dd3643SJeff Mahoney static void free_journal_ram(struct super_block *sb) 1885bd4c625cSLinus Torvalds { 1886a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 1887d739b42bSPekka Enberg kfree(journal->j_current_jl); 18881da177e4SLinus Torvalds journal->j_num_lists--; 18891da177e4SLinus Torvalds 18901da177e4SLinus Torvalds vfree(journal->j_cnode_free_orig); 1891a9dd3643SJeff Mahoney free_list_bitmaps(sb, journal->j_list_bitmap); 1892a9dd3643SJeff Mahoney free_bitmap_nodes(sb); /* must be after free_list_bitmaps */ 18931da177e4SLinus Torvalds if (journal->j_header_bh) { 18941da177e4SLinus Torvalds brelse(journal->j_header_bh); 18951da177e4SLinus Torvalds } 1896098297b2SJeff Mahoney /* 1897098297b2SJeff Mahoney * j_header_bh is on the journal dev, make sure 1898098297b2SJeff Mahoney * not to release the journal dev until we brelse j_header_bh 18991da177e4SLinus Torvalds */ 1900a9dd3643SJeff Mahoney release_journal_dev(sb, journal); 19011da177e4SLinus Torvalds vfree(journal); 19021da177e4SLinus Torvalds } 19031da177e4SLinus Torvalds 19041da177e4SLinus Torvalds /* 1905098297b2SJeff Mahoney * call on unmount. Only set error to 1 if you haven't made your way out 1906098297b2SJeff Mahoney * of read_super() yet. Any other caller must keep error at 0. 19071da177e4SLinus Torvalds */ 1908bd4c625cSLinus Torvalds static int do_journal_release(struct reiserfs_transaction_handle *th, 1909a9dd3643SJeff Mahoney struct super_block *sb, int error) 1910bd4c625cSLinus Torvalds { 19111da177e4SLinus Torvalds struct reiserfs_transaction_handle myth; 19121da177e4SLinus Torvalds int flushed = 0; 1913a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 19141da177e4SLinus Torvalds 1915098297b2SJeff Mahoney /* 1916098297b2SJeff Mahoney * we only want to flush out transactions if we were 1917098297b2SJeff Mahoney * called with error == 0 19181da177e4SLinus Torvalds */ 1919a9dd3643SJeff Mahoney if (!error && !(sb->s_flags & MS_RDONLY)) { 19201da177e4SLinus Torvalds /* end the current trans */ 19211da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 192258d85426SJeff Mahoney do_journal_end(th, FLUSH_ALL); 19231da177e4SLinus Torvalds 1924098297b2SJeff Mahoney /* 1925098297b2SJeff Mahoney * make sure something gets logged to force 1926098297b2SJeff Mahoney * our way into the flush code 1927098297b2SJeff Mahoney */ 1928b491dd17SJeff Mahoney if (!journal_join(&myth, sb)) { 1929a9dd3643SJeff Mahoney reiserfs_prepare_for_journal(sb, 1930a9dd3643SJeff Mahoney SB_BUFFER_WITH_SB(sb), 1931bd4c625cSLinus Torvalds 1); 193209f1b80bSJeff Mahoney journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb)); 193358d85426SJeff Mahoney do_journal_end(&myth, FLUSH_ALL); 19341da177e4SLinus Torvalds flushed = 1; 19351da177e4SLinus Torvalds } 19361da177e4SLinus Torvalds } 19371da177e4SLinus Torvalds 19381da177e4SLinus Torvalds /* this also catches errors during the do_journal_end above */ 19391da177e4SLinus Torvalds if (!error && reiserfs_is_journal_aborted(journal)) { 19401da177e4SLinus Torvalds memset(&myth, 0, sizeof(myth)); 1941b491dd17SJeff Mahoney if (!journal_join_abort(&myth, sb)) { 1942a9dd3643SJeff Mahoney reiserfs_prepare_for_journal(sb, 1943a9dd3643SJeff Mahoney SB_BUFFER_WITH_SB(sb), 1944bd4c625cSLinus Torvalds 1); 194509f1b80bSJeff Mahoney journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb)); 194658d85426SJeff Mahoney do_journal_end(&myth, FLUSH_ALL); 19471da177e4SLinus Torvalds } 19481da177e4SLinus Torvalds } 19491da177e4SLinus Torvalds 19501da177e4SLinus Torvalds /* wait for all commits to finish */ 1951a9dd3643SJeff Mahoney cancel_delayed_work(&SB_JOURNAL(sb)->j_work); 19528ebc4232SFrederic Weisbecker 19538ebc4232SFrederic Weisbecker /* 19548ebc4232SFrederic Weisbecker * We must release the write lock here because 19558ebc4232SFrederic Weisbecker * the workqueue job (flush_async_commit) needs this lock 19568ebc4232SFrederic Weisbecker */ 19578ebc4232SFrederic Weisbecker reiserfs_write_unlock(sb); 1958033369d1SArtem Bityutskiy 1959033369d1SArtem Bityutskiy cancel_delayed_work_sync(&REISERFS_SB(sb)->old_work); 1960797d9016SJeff Mahoney flush_workqueue(REISERFS_SB(sb)->commit_wq); 19611da177e4SLinus Torvalds 1962a9dd3643SJeff Mahoney free_journal_ram(sb); 19631da177e4SLinus Torvalds 19640523676dSFrederic Weisbecker reiserfs_write_lock(sb); 19650523676dSFrederic Weisbecker 19661da177e4SLinus Torvalds return 0; 19671da177e4SLinus Torvalds } 19681da177e4SLinus Torvalds 1969098297b2SJeff Mahoney /* * call on unmount. flush all journal trans, release all alloc'd ram */ 1970bd4c625cSLinus Torvalds int journal_release(struct reiserfs_transaction_handle *th, 1971a9dd3643SJeff Mahoney struct super_block *sb) 1972bd4c625cSLinus Torvalds { 1973a9dd3643SJeff Mahoney return do_journal_release(th, sb, 0); 19741da177e4SLinus Torvalds } 1975bd4c625cSLinus Torvalds 1976098297b2SJeff Mahoney /* only call from an error condition inside reiserfs_read_super! */ 1977bd4c625cSLinus Torvalds int journal_release_error(struct reiserfs_transaction_handle *th, 1978a9dd3643SJeff Mahoney struct super_block *sb) 1979bd4c625cSLinus Torvalds { 1980a9dd3643SJeff Mahoney return do_journal_release(th, sb, 1); 19811da177e4SLinus Torvalds } 19821da177e4SLinus Torvalds 1983098297b2SJeff Mahoney /* 1984098297b2SJeff Mahoney * compares description block with commit block. 1985098297b2SJeff Mahoney * returns 1 if they differ, 0 if they are the same 1986098297b2SJeff Mahoney */ 1987a9dd3643SJeff Mahoney static int journal_compare_desc_commit(struct super_block *sb, 1988bd4c625cSLinus Torvalds struct reiserfs_journal_desc *desc, 1989bd4c625cSLinus Torvalds struct reiserfs_journal_commit *commit) 1990bd4c625cSLinus Torvalds { 19911da177e4SLinus Torvalds if (get_commit_trans_id(commit) != get_desc_trans_id(desc) || 19921da177e4SLinus Torvalds get_commit_trans_len(commit) != get_desc_trans_len(desc) || 1993a9dd3643SJeff Mahoney get_commit_trans_len(commit) > SB_JOURNAL(sb)->j_trans_max || 1994bd4c625cSLinus Torvalds get_commit_trans_len(commit) <= 0) { 19951da177e4SLinus Torvalds return 1; 19961da177e4SLinus Torvalds } 19971da177e4SLinus Torvalds return 0; 19981da177e4SLinus Torvalds } 1999bd4c625cSLinus Torvalds 2000098297b2SJeff Mahoney /* 2001098297b2SJeff Mahoney * returns 0 if it did not find a description block 2002098297b2SJeff Mahoney * returns -1 if it found a corrupt commit block 2003098297b2SJeff Mahoney * returns 1 if both desc and commit were valid 2004098297b2SJeff Mahoney * NOTE: only called during fs mount 20051da177e4SLinus Torvalds */ 2006a9dd3643SJeff Mahoney static int journal_transaction_is_valid(struct super_block *sb, 2007bd4c625cSLinus Torvalds struct buffer_head *d_bh, 2008600ed416SJeff Mahoney unsigned int *oldest_invalid_trans_id, 2009bd4c625cSLinus Torvalds unsigned long *newest_mount_id) 2010bd4c625cSLinus Torvalds { 20111da177e4SLinus Torvalds struct reiserfs_journal_desc *desc; 20121da177e4SLinus Torvalds struct reiserfs_journal_commit *commit; 20131da177e4SLinus Torvalds struct buffer_head *c_bh; 20141da177e4SLinus Torvalds unsigned long offset; 20151da177e4SLinus Torvalds 20161da177e4SLinus Torvalds if (!d_bh) 20171da177e4SLinus Torvalds return 0; 20181da177e4SLinus Torvalds 20191da177e4SLinus Torvalds desc = (struct reiserfs_journal_desc *)d_bh->b_data; 2020bd4c625cSLinus Torvalds if (get_desc_trans_len(desc) > 0 2021bd4c625cSLinus Torvalds && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) { 2022bd4c625cSLinus Torvalds if (oldest_invalid_trans_id && *oldest_invalid_trans_id 2023bd4c625cSLinus Torvalds && get_desc_trans_id(desc) > *oldest_invalid_trans_id) { 2024a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2025bd4c625cSLinus Torvalds "journal-986: transaction " 20261da177e4SLinus Torvalds "is valid returning because trans_id %d is greater than " 2027bd4c625cSLinus Torvalds "oldest_invalid %lu", 2028bd4c625cSLinus Torvalds get_desc_trans_id(desc), 20291da177e4SLinus Torvalds *oldest_invalid_trans_id); 20301da177e4SLinus Torvalds return 0; 20311da177e4SLinus Torvalds } 2032bd4c625cSLinus Torvalds if (newest_mount_id 2033bd4c625cSLinus Torvalds && *newest_mount_id > get_desc_mount_id(desc)) { 2034a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2035bd4c625cSLinus Torvalds "journal-1087: transaction " 20361da177e4SLinus Torvalds "is valid returning because mount_id %d is less than " 2037bd4c625cSLinus Torvalds "newest_mount_id %lu", 2038bd4c625cSLinus Torvalds get_desc_mount_id(desc), 20391da177e4SLinus Torvalds *newest_mount_id); 20401da177e4SLinus Torvalds return -1; 20411da177e4SLinus Torvalds } 2042a9dd3643SJeff Mahoney if (get_desc_trans_len(desc) > SB_JOURNAL(sb)->j_trans_max) { 2043a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-2018", 204445b03d5eSJeff Mahoney "Bad transaction length %d " 204545b03d5eSJeff Mahoney "encountered, ignoring transaction", 2046bd4c625cSLinus Torvalds get_desc_trans_len(desc)); 20471da177e4SLinus Torvalds return -1; 20481da177e4SLinus Torvalds } 2049a9dd3643SJeff Mahoney offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb); 20501da177e4SLinus Torvalds 2051098297b2SJeff Mahoney /* 2052098297b2SJeff Mahoney * ok, we have a journal description block, 2053098297b2SJeff Mahoney * let's see if the transaction was valid 2054098297b2SJeff Mahoney */ 2055bd4c625cSLinus Torvalds c_bh = 2056a9dd3643SJeff Mahoney journal_bread(sb, 2057a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2058bd4c625cSLinus Torvalds ((offset + get_desc_trans_len(desc) + 2059a9dd3643SJeff Mahoney 1) % SB_ONDISK_JOURNAL_SIZE(sb))); 20601da177e4SLinus Torvalds if (!c_bh) 20611da177e4SLinus Torvalds return 0; 20621da177e4SLinus Torvalds commit = (struct reiserfs_journal_commit *)c_bh->b_data; 2063a9dd3643SJeff Mahoney if (journal_compare_desc_commit(sb, desc, commit)) { 2064a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 20651da177e4SLinus Torvalds "journal_transaction_is_valid, commit offset %ld had bad " 20661da177e4SLinus Torvalds "time %d or length %d", 2067bd4c625cSLinus Torvalds c_bh->b_blocknr - 2068a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb), 20691da177e4SLinus Torvalds get_commit_trans_id(commit), 20701da177e4SLinus Torvalds get_commit_trans_len(commit)); 20711da177e4SLinus Torvalds brelse(c_bh); 20721da177e4SLinus Torvalds if (oldest_invalid_trans_id) { 2073bd4c625cSLinus Torvalds *oldest_invalid_trans_id = 2074bd4c625cSLinus Torvalds get_desc_trans_id(desc); 2075a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2076bd4c625cSLinus Torvalds "journal-1004: " 20771da177e4SLinus Torvalds "transaction_is_valid setting oldest invalid trans_id " 2078bd4c625cSLinus Torvalds "to %d", 2079bd4c625cSLinus Torvalds get_desc_trans_id(desc)); 20801da177e4SLinus Torvalds } 20811da177e4SLinus Torvalds return -1; 20821da177e4SLinus Torvalds } 20831da177e4SLinus Torvalds brelse(c_bh); 2084a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2085bd4c625cSLinus Torvalds "journal-1006: found valid " 20861da177e4SLinus Torvalds "transaction start offset %llu, len %d id %d", 2087bd4c625cSLinus Torvalds d_bh->b_blocknr - 2088a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb), 2089bd4c625cSLinus Torvalds get_desc_trans_len(desc), 2090bd4c625cSLinus Torvalds get_desc_trans_id(desc)); 20911da177e4SLinus Torvalds return 1; 20921da177e4SLinus Torvalds } else { 20931da177e4SLinus Torvalds return 0; 20941da177e4SLinus Torvalds } 20951da177e4SLinus Torvalds } 20961da177e4SLinus Torvalds 2097bd4c625cSLinus Torvalds static void brelse_array(struct buffer_head **heads, int num) 2098bd4c625cSLinus Torvalds { 20991da177e4SLinus Torvalds int i; 21001da177e4SLinus Torvalds for (i = 0; i < num; i++) { 21011da177e4SLinus Torvalds brelse(heads[i]); 21021da177e4SLinus Torvalds } 21031da177e4SLinus Torvalds } 21041da177e4SLinus Torvalds 21051da177e4SLinus Torvalds /* 2106098297b2SJeff Mahoney * given the start, and values for the oldest acceptable transactions, 2107098297b2SJeff Mahoney * this either reads in a replays a transaction, or returns because the 2108098297b2SJeff Mahoney * transaction is invalid, or too old. 2109098297b2SJeff Mahoney * NOTE: only called during fs mount 21101da177e4SLinus Torvalds */ 2111a9dd3643SJeff Mahoney static int journal_read_transaction(struct super_block *sb, 2112bd4c625cSLinus Torvalds unsigned long cur_dblock, 2113bd4c625cSLinus Torvalds unsigned long oldest_start, 2114600ed416SJeff Mahoney unsigned int oldest_trans_id, 2115bd4c625cSLinus Torvalds unsigned long newest_mount_id) 2116bd4c625cSLinus Torvalds { 2117a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 21181da177e4SLinus Torvalds struct reiserfs_journal_desc *desc; 21191da177e4SLinus Torvalds struct reiserfs_journal_commit *commit; 2120600ed416SJeff Mahoney unsigned int trans_id = 0; 21211da177e4SLinus Torvalds struct buffer_head *c_bh; 21221da177e4SLinus Torvalds struct buffer_head *d_bh; 21231da177e4SLinus Torvalds struct buffer_head **log_blocks = NULL; 21241da177e4SLinus Torvalds struct buffer_head **real_blocks = NULL; 2125600ed416SJeff Mahoney unsigned int trans_offset; 21261da177e4SLinus Torvalds int i; 21271da177e4SLinus Torvalds int trans_half; 21281da177e4SLinus Torvalds 2129a9dd3643SJeff Mahoney d_bh = journal_bread(sb, cur_dblock); 21301da177e4SLinus Torvalds if (!d_bh) 21311da177e4SLinus Torvalds return 1; 21321da177e4SLinus Torvalds desc = (struct reiserfs_journal_desc *)d_bh->b_data; 2133a9dd3643SJeff Mahoney trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb); 2134a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1037: " 21351da177e4SLinus Torvalds "journal_read_transaction, offset %llu, len %d mount_id %d", 2136a9dd3643SJeff Mahoney d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb), 21371da177e4SLinus Torvalds get_desc_trans_len(desc), get_desc_mount_id(desc)); 21381da177e4SLinus Torvalds if (get_desc_trans_id(desc) < oldest_trans_id) { 2139a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1039: " 21401da177e4SLinus Torvalds "journal_read_trans skipping because %lu is too old", 2141bd4c625cSLinus Torvalds cur_dblock - 2142a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb)); 21431da177e4SLinus Torvalds brelse(d_bh); 21441da177e4SLinus Torvalds return 1; 21451da177e4SLinus Torvalds } 21461da177e4SLinus Torvalds if (get_desc_mount_id(desc) != newest_mount_id) { 2147a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1146: " 21481da177e4SLinus Torvalds "journal_read_trans skipping because %d is != " 21491da177e4SLinus Torvalds "newest_mount_id %lu", get_desc_mount_id(desc), 21501da177e4SLinus Torvalds newest_mount_id); 21511da177e4SLinus Torvalds brelse(d_bh); 21521da177e4SLinus Torvalds return 1; 21531da177e4SLinus Torvalds } 2154a9dd3643SJeff Mahoney c_bh = journal_bread(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 21551da177e4SLinus Torvalds ((trans_offset + get_desc_trans_len(desc) + 1) % 2156a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb))); 21571da177e4SLinus Torvalds if (!c_bh) { 21581da177e4SLinus Torvalds brelse(d_bh); 21591da177e4SLinus Torvalds return 1; 21601da177e4SLinus Torvalds } 21611da177e4SLinus Torvalds commit = (struct reiserfs_journal_commit *)c_bh->b_data; 2162a9dd3643SJeff Mahoney if (journal_compare_desc_commit(sb, desc, commit)) { 2163a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2164bd4c625cSLinus Torvalds "journal_read_transaction, " 21651da177e4SLinus Torvalds "commit offset %llu had bad time %d or length %d", 2166bd4c625cSLinus Torvalds c_bh->b_blocknr - 2167a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb), 2168bd4c625cSLinus Torvalds get_commit_trans_id(commit), 2169bd4c625cSLinus Torvalds get_commit_trans_len(commit)); 21701da177e4SLinus Torvalds brelse(c_bh); 21711da177e4SLinus Torvalds brelse(d_bh); 21721da177e4SLinus Torvalds return 1; 21731da177e4SLinus Torvalds } 21743f8b5ee3SJeff Mahoney 21753f8b5ee3SJeff Mahoney if (bdev_read_only(sb->s_bdev)) { 21763f8b5ee3SJeff Mahoney reiserfs_warning(sb, "clm-2076", 21773f8b5ee3SJeff Mahoney "device is readonly, unable to replay log"); 21783f8b5ee3SJeff Mahoney brelse(c_bh); 21793f8b5ee3SJeff Mahoney brelse(d_bh); 21803f8b5ee3SJeff Mahoney return -EROFS; 21813f8b5ee3SJeff Mahoney } 21823f8b5ee3SJeff Mahoney 21831da177e4SLinus Torvalds trans_id = get_desc_trans_id(desc); 2184098297b2SJeff Mahoney /* 2185098297b2SJeff Mahoney * now we know we've got a good transaction, and it was 2186098297b2SJeff Mahoney * inside the valid time ranges 2187098297b2SJeff Mahoney */ 2188d739b42bSPekka Enberg log_blocks = kmalloc(get_desc_trans_len(desc) * 2189d739b42bSPekka Enberg sizeof(struct buffer_head *), GFP_NOFS); 2190d739b42bSPekka Enberg real_blocks = kmalloc(get_desc_trans_len(desc) * 2191d739b42bSPekka Enberg sizeof(struct buffer_head *), GFP_NOFS); 21921da177e4SLinus Torvalds if (!log_blocks || !real_blocks) { 21931da177e4SLinus Torvalds brelse(c_bh); 21941da177e4SLinus Torvalds brelse(d_bh); 2195d739b42bSPekka Enberg kfree(log_blocks); 2196d739b42bSPekka Enberg kfree(real_blocks); 2197a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1169", 219845b03d5eSJeff Mahoney "kmalloc failed, unable to mount FS"); 21991da177e4SLinus Torvalds return -1; 22001da177e4SLinus Torvalds } 22011da177e4SLinus Torvalds /* get all the buffer heads */ 2202a9dd3643SJeff Mahoney trans_half = journal_trans_half(sb->s_blocksize); 22031da177e4SLinus Torvalds for (i = 0; i < get_desc_trans_len(desc); i++) { 2204bd4c625cSLinus Torvalds log_blocks[i] = 2205a9dd3643SJeff Mahoney journal_getblk(sb, 2206a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2207bd4c625cSLinus Torvalds (trans_offset + 1 + 2208a9dd3643SJeff Mahoney i) % SB_ONDISK_JOURNAL_SIZE(sb)); 22091da177e4SLinus Torvalds if (i < trans_half) { 2210bd4c625cSLinus Torvalds real_blocks[i] = 2211a9dd3643SJeff Mahoney sb_getblk(sb, 2212bd4c625cSLinus Torvalds le32_to_cpu(desc->j_realblock[i])); 22131da177e4SLinus Torvalds } else { 2214bd4c625cSLinus Torvalds real_blocks[i] = 2215a9dd3643SJeff Mahoney sb_getblk(sb, 2216bd4c625cSLinus Torvalds le32_to_cpu(commit-> 2217bd4c625cSLinus Torvalds j_realblock[i - trans_half])); 22181da177e4SLinus Torvalds } 2219a9dd3643SJeff Mahoney if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(sb)) { 2220a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1207", 222145b03d5eSJeff Mahoney "REPLAY FAILURE fsck required! " 222245b03d5eSJeff Mahoney "Block to replay is outside of " 222345b03d5eSJeff Mahoney "filesystem"); 22241da177e4SLinus Torvalds goto abort_replay; 22251da177e4SLinus Torvalds } 22261da177e4SLinus Torvalds /* make sure we don't try to replay onto log or reserved area */ 2227bd4c625cSLinus Torvalds if (is_block_in_log_or_reserved_area 2228a9dd3643SJeff Mahoney (sb, real_blocks[i]->b_blocknr)) { 2229a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1204", 223045b03d5eSJeff Mahoney "REPLAY FAILURE fsck required! " 223145b03d5eSJeff Mahoney "Trying to replay onto a log block"); 22321da177e4SLinus Torvalds abort_replay: 22331da177e4SLinus Torvalds brelse_array(log_blocks, i); 22341da177e4SLinus Torvalds brelse_array(real_blocks, i); 22351da177e4SLinus Torvalds brelse(c_bh); 22361da177e4SLinus Torvalds brelse(d_bh); 2237d739b42bSPekka Enberg kfree(log_blocks); 2238d739b42bSPekka Enberg kfree(real_blocks); 22391da177e4SLinus Torvalds return -1; 22401da177e4SLinus Torvalds } 22411da177e4SLinus Torvalds } 22421da177e4SLinus Torvalds /* read in the log blocks, memcpy to the corresponding real block */ 22431da177e4SLinus Torvalds ll_rw_block(READ, get_desc_trans_len(desc), log_blocks); 22441da177e4SLinus Torvalds for (i = 0; i < get_desc_trans_len(desc); i++) { 22458ebc4232SFrederic Weisbecker 22461da177e4SLinus Torvalds wait_on_buffer(log_blocks[i]); 22471da177e4SLinus Torvalds if (!buffer_uptodate(log_blocks[i])) { 2248a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1212", 224945b03d5eSJeff Mahoney "REPLAY FAILURE fsck required! " 225045b03d5eSJeff Mahoney "buffer write failed"); 2251bd4c625cSLinus Torvalds brelse_array(log_blocks + i, 2252bd4c625cSLinus Torvalds get_desc_trans_len(desc) - i); 22531da177e4SLinus Torvalds brelse_array(real_blocks, get_desc_trans_len(desc)); 22541da177e4SLinus Torvalds brelse(c_bh); 22551da177e4SLinus Torvalds brelse(d_bh); 2256d739b42bSPekka Enberg kfree(log_blocks); 2257d739b42bSPekka Enberg kfree(real_blocks); 22581da177e4SLinus Torvalds return -1; 22591da177e4SLinus Torvalds } 2260bd4c625cSLinus Torvalds memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data, 2261bd4c625cSLinus Torvalds real_blocks[i]->b_size); 22621da177e4SLinus Torvalds set_buffer_uptodate(real_blocks[i]); 22631da177e4SLinus Torvalds brelse(log_blocks[i]); 22641da177e4SLinus Torvalds } 22651da177e4SLinus Torvalds /* flush out the real blocks */ 22661da177e4SLinus Torvalds for (i = 0; i < get_desc_trans_len(desc); i++) { 22671da177e4SLinus Torvalds set_buffer_dirty(real_blocks[i]); 22689cb569d6SChristoph Hellwig write_dirty_buffer(real_blocks[i], WRITE); 22691da177e4SLinus Torvalds } 22701da177e4SLinus Torvalds for (i = 0; i < get_desc_trans_len(desc); i++) { 22711da177e4SLinus Torvalds wait_on_buffer(real_blocks[i]); 22721da177e4SLinus Torvalds if (!buffer_uptodate(real_blocks[i])) { 2273a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1226", 227445b03d5eSJeff Mahoney "REPLAY FAILURE, fsck required! " 227545b03d5eSJeff Mahoney "buffer write failed"); 2276bd4c625cSLinus Torvalds brelse_array(real_blocks + i, 2277bd4c625cSLinus Torvalds get_desc_trans_len(desc) - i); 22781da177e4SLinus Torvalds brelse(c_bh); 22791da177e4SLinus Torvalds brelse(d_bh); 2280d739b42bSPekka Enberg kfree(log_blocks); 2281d739b42bSPekka Enberg kfree(real_blocks); 22821da177e4SLinus Torvalds return -1; 22831da177e4SLinus Torvalds } 22841da177e4SLinus Torvalds brelse(real_blocks[i]); 22851da177e4SLinus Torvalds } 2286bd4c625cSLinus Torvalds cur_dblock = 2287a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2288bd4c625cSLinus Torvalds ((trans_offset + get_desc_trans_len(desc) + 2289a9dd3643SJeff Mahoney 2) % SB_ONDISK_JOURNAL_SIZE(sb)); 2290a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2291bd4c625cSLinus Torvalds "journal-1095: setting journal " "start to offset %ld", 2292a9dd3643SJeff Mahoney cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb)); 22931da177e4SLinus Torvalds 2294098297b2SJeff Mahoney /* 2295098297b2SJeff Mahoney * init starting values for the first transaction, in case 2296098297b2SJeff Mahoney * this is the last transaction to be replayed. 2297098297b2SJeff Mahoney */ 2298a9dd3643SJeff Mahoney journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb); 22991da177e4SLinus Torvalds journal->j_last_flush_trans_id = trans_id; 23001da177e4SLinus Torvalds journal->j_trans_id = trans_id + 1; 2301a44c94a7SAlexander Zarochentsev /* check for trans_id overflow */ 2302a44c94a7SAlexander Zarochentsev if (journal->j_trans_id == 0) 2303a44c94a7SAlexander Zarochentsev journal->j_trans_id = 10; 23041da177e4SLinus Torvalds brelse(c_bh); 23051da177e4SLinus Torvalds brelse(d_bh); 2306d739b42bSPekka Enberg kfree(log_blocks); 2307d739b42bSPekka Enberg kfree(real_blocks); 23081da177e4SLinus Torvalds return 0; 23091da177e4SLinus Torvalds } 23101da177e4SLinus Torvalds 2311098297b2SJeff Mahoney /* 2312098297b2SJeff Mahoney * This function reads blocks starting from block and to max_block of bufsize 2313098297b2SJeff Mahoney * size (but no more than BUFNR blocks at a time). This proved to improve 2314098297b2SJeff Mahoney * mounting speed on self-rebuilding raid5 arrays at least. 2315098297b2SJeff Mahoney * Right now it is only used from journal code. But later we might use it 2316098297b2SJeff Mahoney * from other places. 2317098297b2SJeff Mahoney * Note: Do not use journal_getblk/sb_getblk functions here! 2318098297b2SJeff Mahoney */ 23193ee16670SJeff Mahoney static struct buffer_head *reiserfs_breada(struct block_device *dev, 23203ee16670SJeff Mahoney b_blocknr_t block, int bufsize, 23213ee16670SJeff Mahoney b_blocknr_t max_block) 23221da177e4SLinus Torvalds { 23231da177e4SLinus Torvalds struct buffer_head *bhlist[BUFNR]; 23241da177e4SLinus Torvalds unsigned int blocks = BUFNR; 23251da177e4SLinus Torvalds struct buffer_head *bh; 23261da177e4SLinus Torvalds int i, j; 23271da177e4SLinus Torvalds 23281da177e4SLinus Torvalds bh = __getblk(dev, block, bufsize); 23291da177e4SLinus Torvalds if (buffer_uptodate(bh)) 23301da177e4SLinus Torvalds return (bh); 23311da177e4SLinus Torvalds 23321da177e4SLinus Torvalds if (block + BUFNR > max_block) { 23331da177e4SLinus Torvalds blocks = max_block - block; 23341da177e4SLinus Torvalds } 23351da177e4SLinus Torvalds bhlist[0] = bh; 23361da177e4SLinus Torvalds j = 1; 23371da177e4SLinus Torvalds for (i = 1; i < blocks; i++) { 23381da177e4SLinus Torvalds bh = __getblk(dev, block + i, bufsize); 23391da177e4SLinus Torvalds if (buffer_uptodate(bh)) { 23401da177e4SLinus Torvalds brelse(bh); 23411da177e4SLinus Torvalds break; 2342bd4c625cSLinus Torvalds } else 2343bd4c625cSLinus Torvalds bhlist[j++] = bh; 23441da177e4SLinus Torvalds } 23451da177e4SLinus Torvalds ll_rw_block(READ, j, bhlist); 23461da177e4SLinus Torvalds for (i = 1; i < j; i++) 23471da177e4SLinus Torvalds brelse(bhlist[i]); 23481da177e4SLinus Torvalds bh = bhlist[0]; 23491da177e4SLinus Torvalds wait_on_buffer(bh); 23501da177e4SLinus Torvalds if (buffer_uptodate(bh)) 23511da177e4SLinus Torvalds return bh; 23521da177e4SLinus Torvalds brelse(bh); 23531da177e4SLinus Torvalds return NULL; 23541da177e4SLinus Torvalds } 23551da177e4SLinus Torvalds 23561da177e4SLinus Torvalds /* 2357098297b2SJeff Mahoney * read and replay the log 2358098297b2SJeff Mahoney * on a clean unmount, the journal header's next unflushed pointer will be 2359098297b2SJeff Mahoney * to an invalid transaction. This tests that before finding all the 2360098297b2SJeff Mahoney * transactions in the log, which makes normal mount times fast. 2361098297b2SJeff Mahoney * 2362098297b2SJeff Mahoney * After a crash, this starts with the next unflushed transaction, and 2363098297b2SJeff Mahoney * replays until it finds one too old, or invalid. 2364098297b2SJeff Mahoney * 2365098297b2SJeff Mahoney * On exit, it sets things up so the first transaction will work correctly. 2366098297b2SJeff Mahoney * NOTE: only called during fs mount 23671da177e4SLinus Torvalds */ 2368a9dd3643SJeff Mahoney static int journal_read(struct super_block *sb) 2369bd4c625cSLinus Torvalds { 2370a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 23711da177e4SLinus Torvalds struct reiserfs_journal_desc *desc; 2372600ed416SJeff Mahoney unsigned int oldest_trans_id = 0; 2373600ed416SJeff Mahoney unsigned int oldest_invalid_trans_id = 0; 23741da177e4SLinus Torvalds time_t start; 23751da177e4SLinus Torvalds unsigned long oldest_start = 0; 23761da177e4SLinus Torvalds unsigned long cur_dblock = 0; 23771da177e4SLinus Torvalds unsigned long newest_mount_id = 9; 23781da177e4SLinus Torvalds struct buffer_head *d_bh; 23791da177e4SLinus Torvalds struct reiserfs_journal_header *jh; 23801da177e4SLinus Torvalds int valid_journal_header = 0; 23811da177e4SLinus Torvalds int replay_count = 0; 23821da177e4SLinus Torvalds int continue_replay = 1; 23831da177e4SLinus Torvalds int ret; 23841da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 23851da177e4SLinus Torvalds 2386a9dd3643SJeff Mahoney cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(sb); 2387a9dd3643SJeff Mahoney reiserfs_info(sb, "checking transaction log (%s)\n", 23881da177e4SLinus Torvalds bdevname(journal->j_dev_bd, b)); 23891da177e4SLinus Torvalds start = get_seconds(); 23901da177e4SLinus Torvalds 2391098297b2SJeff Mahoney /* 2392098297b2SJeff Mahoney * step 1, read in the journal header block. Check the transaction 2393098297b2SJeff Mahoney * it says is the first unflushed, and if that transaction is not 2394098297b2SJeff Mahoney * valid, replay is done 23951da177e4SLinus Torvalds */ 2396a9dd3643SJeff Mahoney journal->j_header_bh = journal_bread(sb, 2397a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) 2398a9dd3643SJeff Mahoney + SB_ONDISK_JOURNAL_SIZE(sb)); 23991da177e4SLinus Torvalds if (!journal->j_header_bh) { 24001da177e4SLinus Torvalds return 1; 24011da177e4SLinus Torvalds } 24021da177e4SLinus Torvalds jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data); 2403c499ec24SVladimir V. Saveliev if (le32_to_cpu(jh->j_first_unflushed_offset) < 2404a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb) 2405bd4c625cSLinus Torvalds && le32_to_cpu(jh->j_last_flush_trans_id) > 0) { 2406bd4c625cSLinus Torvalds oldest_start = 2407a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 24081da177e4SLinus Torvalds le32_to_cpu(jh->j_first_unflushed_offset); 24091da177e4SLinus Torvalds oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1; 24101da177e4SLinus Torvalds newest_mount_id = le32_to_cpu(jh->j_mount_id); 2411a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2412bd4c625cSLinus Torvalds "journal-1153: found in " 24131da177e4SLinus Torvalds "header: first_unflushed_offset %d, last_flushed_trans_id " 24141da177e4SLinus Torvalds "%lu", le32_to_cpu(jh->j_first_unflushed_offset), 24151da177e4SLinus Torvalds le32_to_cpu(jh->j_last_flush_trans_id)); 24161da177e4SLinus Torvalds valid_journal_header = 1; 24171da177e4SLinus Torvalds 2418098297b2SJeff Mahoney /* 2419098297b2SJeff Mahoney * now, we try to read the first unflushed offset. If it 2420098297b2SJeff Mahoney * is not valid, there is nothing more we can do, and it 2421098297b2SJeff Mahoney * makes no sense to read through the whole log. 24221da177e4SLinus Torvalds */ 2423bd4c625cSLinus Torvalds d_bh = 2424a9dd3643SJeff Mahoney journal_bread(sb, 2425a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2426bd4c625cSLinus Torvalds le32_to_cpu(jh->j_first_unflushed_offset)); 2427a9dd3643SJeff Mahoney ret = journal_transaction_is_valid(sb, d_bh, NULL, NULL); 24281da177e4SLinus Torvalds if (!ret) { 24291da177e4SLinus Torvalds continue_replay = 0; 24301da177e4SLinus Torvalds } 24311da177e4SLinus Torvalds brelse(d_bh); 24321da177e4SLinus Torvalds goto start_log_replay; 24331da177e4SLinus Torvalds } 24341da177e4SLinus Torvalds 2435098297b2SJeff Mahoney /* 2436098297b2SJeff Mahoney * ok, there are transactions that need to be replayed. start 2437098297b2SJeff Mahoney * with the first log block, find all the valid transactions, and 2438098297b2SJeff Mahoney * pick out the oldest. 24391da177e4SLinus Torvalds */ 2440bd4c625cSLinus Torvalds while (continue_replay 2441bd4c625cSLinus Torvalds && cur_dblock < 2442a9dd3643SJeff Mahoney (SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2443a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb))) { 2444098297b2SJeff Mahoney /* 2445098297b2SJeff Mahoney * Note that it is required for blocksize of primary fs 2446098297b2SJeff Mahoney * device and journal device to be the same 2447098297b2SJeff Mahoney */ 2448bd4c625cSLinus Torvalds d_bh = 2449bd4c625cSLinus Torvalds reiserfs_breada(journal->j_dev_bd, cur_dblock, 2450a9dd3643SJeff Mahoney sb->s_blocksize, 2451a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2452a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb)); 2453bd4c625cSLinus Torvalds ret = 2454a9dd3643SJeff Mahoney journal_transaction_is_valid(sb, d_bh, 2455bd4c625cSLinus Torvalds &oldest_invalid_trans_id, 2456bd4c625cSLinus Torvalds &newest_mount_id); 24571da177e4SLinus Torvalds if (ret == 1) { 24581da177e4SLinus Torvalds desc = (struct reiserfs_journal_desc *)d_bh->b_data; 24591da177e4SLinus Torvalds if (oldest_start == 0) { /* init all oldest_ values */ 24601da177e4SLinus Torvalds oldest_trans_id = get_desc_trans_id(desc); 24611da177e4SLinus Torvalds oldest_start = d_bh->b_blocknr; 24621da177e4SLinus Torvalds newest_mount_id = get_desc_mount_id(desc); 2463a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2464bd4c625cSLinus Torvalds "journal-1179: Setting " 24651da177e4SLinus Torvalds "oldest_start to offset %llu, trans_id %lu", 2466bd4c625cSLinus Torvalds oldest_start - 2467bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK 2468a9dd3643SJeff Mahoney (sb), oldest_trans_id); 24691da177e4SLinus Torvalds } else if (oldest_trans_id > get_desc_trans_id(desc)) { 24701da177e4SLinus Torvalds /* one we just read was older */ 24711da177e4SLinus Torvalds oldest_trans_id = get_desc_trans_id(desc); 24721da177e4SLinus Torvalds oldest_start = d_bh->b_blocknr; 2473a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2474bd4c625cSLinus Torvalds "journal-1180: Resetting " 24751da177e4SLinus Torvalds "oldest_start to offset %lu, trans_id %lu", 2476bd4c625cSLinus Torvalds oldest_start - 2477bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK 2478a9dd3643SJeff Mahoney (sb), oldest_trans_id); 24791da177e4SLinus Torvalds } 24801da177e4SLinus Torvalds if (newest_mount_id < get_desc_mount_id(desc)) { 24811da177e4SLinus Torvalds newest_mount_id = get_desc_mount_id(desc); 2482a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2483bd4c625cSLinus Torvalds "journal-1299: Setting " 2484bd4c625cSLinus Torvalds "newest_mount_id to %d", 2485bd4c625cSLinus Torvalds get_desc_mount_id(desc)); 24861da177e4SLinus Torvalds } 24871da177e4SLinus Torvalds cur_dblock += get_desc_trans_len(desc) + 2; 24881da177e4SLinus Torvalds } else { 24891da177e4SLinus Torvalds cur_dblock++; 24901da177e4SLinus Torvalds } 24911da177e4SLinus Torvalds brelse(d_bh); 24921da177e4SLinus Torvalds } 24931da177e4SLinus Torvalds 24941da177e4SLinus Torvalds start_log_replay: 24951da177e4SLinus Torvalds cur_dblock = oldest_start; 24961da177e4SLinus Torvalds if (oldest_trans_id) { 2497a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2498bd4c625cSLinus Torvalds "journal-1206: Starting replay " 24991da177e4SLinus Torvalds "from offset %llu, trans_id %lu", 2500a9dd3643SJeff Mahoney cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb), 25011da177e4SLinus Torvalds oldest_trans_id); 25021da177e4SLinus Torvalds 25031da177e4SLinus Torvalds } 25041da177e4SLinus Torvalds replay_count = 0; 25051da177e4SLinus Torvalds while (continue_replay && oldest_trans_id > 0) { 2506bd4c625cSLinus Torvalds ret = 2507a9dd3643SJeff Mahoney journal_read_transaction(sb, cur_dblock, oldest_start, 2508bd4c625cSLinus Torvalds oldest_trans_id, newest_mount_id); 25091da177e4SLinus Torvalds if (ret < 0) { 25101da177e4SLinus Torvalds return ret; 25111da177e4SLinus Torvalds } else if (ret != 0) { 25121da177e4SLinus Torvalds break; 25131da177e4SLinus Torvalds } 2514bd4c625cSLinus Torvalds cur_dblock = 2515a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + journal->j_start; 25161da177e4SLinus Torvalds replay_count++; 25171da177e4SLinus Torvalds if (cur_dblock == oldest_start) 25181da177e4SLinus Torvalds break; 25191da177e4SLinus Torvalds } 25201da177e4SLinus Torvalds 25211da177e4SLinus Torvalds if (oldest_trans_id == 0) { 2522a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2523bd4c625cSLinus Torvalds "journal-1225: No valid " "transactions found"); 25241da177e4SLinus Torvalds } 2525098297b2SJeff Mahoney /* 2526098297b2SJeff Mahoney * j_start does not get set correctly if we don't replay any 2527098297b2SJeff Mahoney * transactions. if we had a valid journal_header, set j_start 2528098297b2SJeff Mahoney * to the first unflushed transaction value, copy the trans_id 2529098297b2SJeff Mahoney * from the header 25301da177e4SLinus Torvalds */ 25311da177e4SLinus Torvalds if (valid_journal_header && replay_count == 0) { 25321da177e4SLinus Torvalds journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset); 2533bd4c625cSLinus Torvalds journal->j_trans_id = 2534bd4c625cSLinus Torvalds le32_to_cpu(jh->j_last_flush_trans_id) + 1; 2535a44c94a7SAlexander Zarochentsev /* check for trans_id overflow */ 2536a44c94a7SAlexander Zarochentsev if (journal->j_trans_id == 0) 2537a44c94a7SAlexander Zarochentsev journal->j_trans_id = 10; 2538bd4c625cSLinus Torvalds journal->j_last_flush_trans_id = 2539bd4c625cSLinus Torvalds le32_to_cpu(jh->j_last_flush_trans_id); 25401da177e4SLinus Torvalds journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1; 25411da177e4SLinus Torvalds } else { 25421da177e4SLinus Torvalds journal->j_mount_id = newest_mount_id + 1; 25431da177e4SLinus Torvalds } 2544a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1299: Setting " 25451da177e4SLinus Torvalds "newest_mount_id to %lu", journal->j_mount_id); 25461da177e4SLinus Torvalds journal->j_first_unflushed_offset = journal->j_start; 25471da177e4SLinus Torvalds if (replay_count > 0) { 2548a9dd3643SJeff Mahoney reiserfs_info(sb, 2549bd4c625cSLinus Torvalds "replayed %d transactions in %lu seconds\n", 25501da177e4SLinus Torvalds replay_count, get_seconds() - start); 25511da177e4SLinus Torvalds } 2552278f6679SJeff Mahoney /* needed to satisfy the locking in _update_journal_header_block */ 2553278f6679SJeff Mahoney reiserfs_write_lock(sb); 2554a9dd3643SJeff Mahoney if (!bdev_read_only(sb->s_bdev) && 2555a9dd3643SJeff Mahoney _update_journal_header_block(sb, journal->j_start, 2556bd4c625cSLinus Torvalds journal->j_last_flush_trans_id)) { 2557278f6679SJeff Mahoney reiserfs_write_unlock(sb); 2558098297b2SJeff Mahoney /* 2559098297b2SJeff Mahoney * replay failed, caller must call free_journal_ram and abort 2560098297b2SJeff Mahoney * the mount 25611da177e4SLinus Torvalds */ 25621da177e4SLinus Torvalds return -1; 25631da177e4SLinus Torvalds } 2564278f6679SJeff Mahoney reiserfs_write_unlock(sb); 25651da177e4SLinus Torvalds return 0; 25661da177e4SLinus Torvalds } 25671da177e4SLinus Torvalds 25681da177e4SLinus Torvalds static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s) 25691da177e4SLinus Torvalds { 25701da177e4SLinus Torvalds struct reiserfs_journal_list *jl; 25718c777cc4SPekka Enberg jl = kzalloc(sizeof(struct reiserfs_journal_list), 25728c777cc4SPekka Enberg GFP_NOFS | __GFP_NOFAIL); 25731da177e4SLinus Torvalds INIT_LIST_HEAD(&jl->j_list); 25741da177e4SLinus Torvalds INIT_LIST_HEAD(&jl->j_working_list); 25751da177e4SLinus Torvalds INIT_LIST_HEAD(&jl->j_tail_bh_list); 25761da177e4SLinus Torvalds INIT_LIST_HEAD(&jl->j_bh_list); 257790415deaSJeff Mahoney mutex_init(&jl->j_commit_mutex); 25781da177e4SLinus Torvalds SB_JOURNAL(s)->j_num_lists++; 25791da177e4SLinus Torvalds get_journal_list(jl); 25801da177e4SLinus Torvalds return jl; 25811da177e4SLinus Torvalds } 25821da177e4SLinus Torvalds 2583a9dd3643SJeff Mahoney static void journal_list_init(struct super_block *sb) 2584bd4c625cSLinus Torvalds { 2585a9dd3643SJeff Mahoney SB_JOURNAL(sb)->j_current_jl = alloc_journal_list(sb); 25861da177e4SLinus Torvalds } 25871da177e4SLinus Torvalds 25884385bab1SAl Viro static void release_journal_dev(struct super_block *super, 25891da177e4SLinus Torvalds struct reiserfs_journal *journal) 25901da177e4SLinus Torvalds { 259186098fa0SChristoph Hellwig if (journal->j_dev_bd != NULL) { 25924385bab1SAl Viro blkdev_put(journal->j_dev_bd, journal->j_dev_mode); 25931da177e4SLinus Torvalds journal->j_dev_bd = NULL; 25941da177e4SLinus Torvalds } 25951da177e4SLinus Torvalds } 25961da177e4SLinus Torvalds 25971da177e4SLinus Torvalds static int journal_init_dev(struct super_block *super, 25981da177e4SLinus Torvalds struct reiserfs_journal *journal, 25991da177e4SLinus Torvalds const char *jdev_name) 26001da177e4SLinus Torvalds { 26011da177e4SLinus Torvalds int result; 26021da177e4SLinus Torvalds dev_t jdev; 2603e525fd89STejun Heo fmode_t blkdev_mode = FMODE_READ | FMODE_WRITE | FMODE_EXCL; 26041da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 26051da177e4SLinus Torvalds 26061da177e4SLinus Torvalds result = 0; 26071da177e4SLinus Torvalds 26081da177e4SLinus Torvalds journal->j_dev_bd = NULL; 26091da177e4SLinus Torvalds jdev = SB_ONDISK_JOURNAL_DEVICE(super) ? 26101da177e4SLinus Torvalds new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev; 26111da177e4SLinus Torvalds 26121da177e4SLinus Torvalds if (bdev_read_only(super->s_bdev)) 26131da177e4SLinus Torvalds blkdev_mode = FMODE_READ; 26141da177e4SLinus Torvalds 26151da177e4SLinus Torvalds /* there is no "jdev" option and journal is on separate device */ 26161da177e4SLinus Torvalds if ((!jdev_name || !jdev_name[0])) { 2617e525fd89STejun Heo if (jdev == super->s_dev) 2618e525fd89STejun Heo blkdev_mode &= ~FMODE_EXCL; 2619d4d77629STejun Heo journal->j_dev_bd = blkdev_get_by_dev(jdev, blkdev_mode, 2620d4d77629STejun Heo journal); 2621e5eb8caaSAl Viro journal->j_dev_mode = blkdev_mode; 26221da177e4SLinus Torvalds if (IS_ERR(journal->j_dev_bd)) { 26231da177e4SLinus Torvalds result = PTR_ERR(journal->j_dev_bd); 26241da177e4SLinus Torvalds journal->j_dev_bd = NULL; 262545b03d5eSJeff Mahoney reiserfs_warning(super, "sh-458", 26261da177e4SLinus Torvalds "cannot init journal device '%s': %i", 26271da177e4SLinus Torvalds __bdevname(jdev, b), result); 26281da177e4SLinus Torvalds return result; 2629e525fd89STejun Heo } else if (jdev != super->s_dev) 26301da177e4SLinus Torvalds set_blocksize(journal->j_dev_bd, super->s_blocksize); 263186098fa0SChristoph Hellwig 26321da177e4SLinus Torvalds return 0; 26331da177e4SLinus Torvalds } 26341da177e4SLinus Torvalds 2635e5eb8caaSAl Viro journal->j_dev_mode = blkdev_mode; 2636d4d77629STejun Heo journal->j_dev_bd = blkdev_get_by_path(jdev_name, blkdev_mode, journal); 263786098fa0SChristoph Hellwig if (IS_ERR(journal->j_dev_bd)) { 263886098fa0SChristoph Hellwig result = PTR_ERR(journal->j_dev_bd); 263986098fa0SChristoph Hellwig journal->j_dev_bd = NULL; 264086098fa0SChristoph Hellwig reiserfs_warning(super, 264186098fa0SChristoph Hellwig "journal_init_dev: Cannot open '%s': %i", 264286098fa0SChristoph Hellwig jdev_name, result); 264386098fa0SChristoph Hellwig return result; 264486098fa0SChristoph Hellwig } 264586098fa0SChristoph Hellwig 26461da177e4SLinus Torvalds set_blocksize(journal->j_dev_bd, super->s_blocksize); 2647bd4c625cSLinus Torvalds reiserfs_info(super, 2648bd4c625cSLinus Torvalds "journal_init_dev: journal device: %s\n", 264974f9f974SEdward Shishkin bdevname(journal->j_dev_bd, b)); 265086098fa0SChristoph Hellwig return 0; 26511da177e4SLinus Torvalds } 26521da177e4SLinus Torvalds 2653098297b2SJeff Mahoney /* 2654cf3d0b81SEdward Shishkin * When creating/tuning a file system user can assign some 2655cf3d0b81SEdward Shishkin * journal params within boundaries which depend on the ratio 2656cf3d0b81SEdward Shishkin * blocksize/standard_blocksize. 2657cf3d0b81SEdward Shishkin * 2658cf3d0b81SEdward Shishkin * For blocks >= standard_blocksize transaction size should 2659cf3d0b81SEdward Shishkin * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more 2660cf3d0b81SEdward Shishkin * then JOURNAL_TRANS_MAX_DEFAULT. 2661cf3d0b81SEdward Shishkin * 2662cf3d0b81SEdward Shishkin * For blocks < standard_blocksize these boundaries should be 2663cf3d0b81SEdward Shishkin * decreased proportionally. 2664cf3d0b81SEdward Shishkin */ 2665cf3d0b81SEdward Shishkin #define REISERFS_STANDARD_BLKSIZE (4096) 2666cf3d0b81SEdward Shishkin 2667a9dd3643SJeff Mahoney static int check_advise_trans_params(struct super_block *sb, 2668cf3d0b81SEdward Shishkin struct reiserfs_journal *journal) 2669cf3d0b81SEdward Shishkin { 2670cf3d0b81SEdward Shishkin if (journal->j_trans_max) { 2671098297b2SJeff Mahoney /* Non-default journal params. Do sanity check for them. */ 2672cf3d0b81SEdward Shishkin int ratio = 1; 2673a9dd3643SJeff Mahoney if (sb->s_blocksize < REISERFS_STANDARD_BLKSIZE) 2674a9dd3643SJeff Mahoney ratio = REISERFS_STANDARD_BLKSIZE / sb->s_blocksize; 2675cf3d0b81SEdward Shishkin 2676cf3d0b81SEdward Shishkin if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio || 2677cf3d0b81SEdward Shishkin journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio || 2678a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb) / journal->j_trans_max < 2679cf3d0b81SEdward Shishkin JOURNAL_MIN_RATIO) { 2680a9dd3643SJeff Mahoney reiserfs_warning(sb, "sh-462", 268145b03d5eSJeff Mahoney "bad transaction max size (%u). " 268245b03d5eSJeff Mahoney "FSCK?", journal->j_trans_max); 2683cf3d0b81SEdward Shishkin return 1; 2684cf3d0b81SEdward Shishkin } 2685cf3d0b81SEdward Shishkin if (journal->j_max_batch != (journal->j_trans_max) * 2686cf3d0b81SEdward Shishkin JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) { 2687a9dd3643SJeff Mahoney reiserfs_warning(sb, "sh-463", 268845b03d5eSJeff Mahoney "bad transaction max batch (%u). " 268945b03d5eSJeff Mahoney "FSCK?", journal->j_max_batch); 2690cf3d0b81SEdward Shishkin return 1; 2691cf3d0b81SEdward Shishkin } 2692cf3d0b81SEdward Shishkin } else { 2693098297b2SJeff Mahoney /* 2694098297b2SJeff Mahoney * Default journal params. 2695098297b2SJeff Mahoney * The file system was created by old version 2696098297b2SJeff Mahoney * of mkreiserfs, so some fields contain zeros, 2697098297b2SJeff Mahoney * and we need to advise proper values for them 2698098297b2SJeff Mahoney */ 2699a9dd3643SJeff Mahoney if (sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) { 2700a9dd3643SJeff Mahoney reiserfs_warning(sb, "sh-464", "bad blocksize (%u)", 2701a9dd3643SJeff Mahoney sb->s_blocksize); 270245b03d5eSJeff Mahoney return 1; 270345b03d5eSJeff Mahoney } 2704cf3d0b81SEdward Shishkin journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT; 2705cf3d0b81SEdward Shishkin journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT; 2706cf3d0b81SEdward Shishkin journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE; 2707cf3d0b81SEdward Shishkin } 2708cf3d0b81SEdward Shishkin return 0; 2709cf3d0b81SEdward Shishkin } 2710cf3d0b81SEdward Shishkin 2711098297b2SJeff Mahoney /* must be called once on fs mount. calls journal_read for you */ 2712a9dd3643SJeff Mahoney int journal_init(struct super_block *sb, const char *j_dev_name, 2713bd4c625cSLinus Torvalds int old_format, unsigned int commit_max_age) 2714bd4c625cSLinus Torvalds { 2715a9dd3643SJeff Mahoney int num_cnodes = SB_ONDISK_JOURNAL_SIZE(sb) * 2; 27161da177e4SLinus Torvalds struct buffer_head *bhjh; 27171da177e4SLinus Torvalds struct reiserfs_super_block *rs; 27181da177e4SLinus Torvalds struct reiserfs_journal_header *jh; 27191da177e4SLinus Torvalds struct reiserfs_journal *journal; 27201da177e4SLinus Torvalds struct reiserfs_journal_list *jl; 27211da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 272298ea3f50SFrederic Weisbecker int ret; 27231da177e4SLinus Torvalds 2724558feb08SJoe Perches journal = SB_JOURNAL(sb) = vzalloc(sizeof(struct reiserfs_journal)); 27251da177e4SLinus Torvalds if (!journal) { 2726a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1256", 272745b03d5eSJeff Mahoney "unable to get memory for journal structure"); 27281da177e4SLinus Torvalds return 1; 27291da177e4SLinus Torvalds } 27301da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_bitmap_nodes); 27311da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_prealloc_list); 27321da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_working_list); 27331da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_journal_list); 27341da177e4SLinus Torvalds journal->j_persistent_trans = 0; 273537c69b98SFrederic Weisbecker if (reiserfs_allocate_list_bitmaps(sb, journal->j_list_bitmap, 273637c69b98SFrederic Weisbecker reiserfs_bmap_count(sb))) 27371da177e4SLinus Torvalds goto free_and_return; 273898ea3f50SFrederic Weisbecker 2739a9dd3643SJeff Mahoney allocate_bitmap_nodes(sb); 27401da177e4SLinus Torvalds 27411da177e4SLinus Torvalds /* reserved for journal area support */ 2742a9dd3643SJeff Mahoney SB_JOURNAL_1st_RESERVED_BLOCK(sb) = (old_format ? 2743bd4c625cSLinus Torvalds REISERFS_OLD_DISK_OFFSET_IN_BYTES 2744a9dd3643SJeff Mahoney / sb->s_blocksize + 2745a9dd3643SJeff Mahoney reiserfs_bmap_count(sb) + 2746bd4c625cSLinus Torvalds 1 : 2747bd4c625cSLinus Torvalds REISERFS_DISK_OFFSET_IN_BYTES / 2748a9dd3643SJeff Mahoney sb->s_blocksize + 2); 27491da177e4SLinus Torvalds 2750098297b2SJeff Mahoney /* 2751098297b2SJeff Mahoney * Sanity check to see is the standard journal fitting 2752098297b2SJeff Mahoney * within first bitmap (actual for small blocksizes) 2753098297b2SJeff Mahoney */ 2754a9dd3643SJeff Mahoney if (!SB_ONDISK_JOURNAL_DEVICE(sb) && 2755a9dd3643SJeff Mahoney (SB_JOURNAL_1st_RESERVED_BLOCK(sb) + 2756a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb) > sb->s_blocksize * 8)) { 2757a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1393", 275845b03d5eSJeff Mahoney "journal does not fit for area addressed " 275945b03d5eSJeff Mahoney "by first of bitmap blocks. It starts at " 27601da177e4SLinus Torvalds "%u and its size is %u. Block size %ld", 2761a9dd3643SJeff Mahoney SB_JOURNAL_1st_RESERVED_BLOCK(sb), 2762a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb), 2763a9dd3643SJeff Mahoney sb->s_blocksize); 27641da177e4SLinus Torvalds goto free_and_return; 27651da177e4SLinus Torvalds } 27661da177e4SLinus Torvalds 2767a9dd3643SJeff Mahoney if (journal_init_dev(sb, journal, j_dev_name) != 0) { 2768a9dd3643SJeff Mahoney reiserfs_warning(sb, "sh-462", 276945b03d5eSJeff Mahoney "unable to initialize jornal device"); 27701da177e4SLinus Torvalds goto free_and_return; 27711da177e4SLinus Torvalds } 27721da177e4SLinus Torvalds 2773a9dd3643SJeff Mahoney rs = SB_DISK_SUPER_BLOCK(sb); 27741da177e4SLinus Torvalds 27751da177e4SLinus Torvalds /* read journal header */ 2776a9dd3643SJeff Mahoney bhjh = journal_bread(sb, 2777a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2778a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb)); 27791da177e4SLinus Torvalds if (!bhjh) { 2780a9dd3643SJeff Mahoney reiserfs_warning(sb, "sh-459", 278145b03d5eSJeff Mahoney "unable to read journal header"); 27821da177e4SLinus Torvalds goto free_and_return; 27831da177e4SLinus Torvalds } 27841da177e4SLinus Torvalds jh = (struct reiserfs_journal_header *)(bhjh->b_data); 27851da177e4SLinus Torvalds 27861da177e4SLinus Torvalds /* make sure that journal matches to the super block */ 2787bd4c625cSLinus Torvalds if (is_reiserfs_jr(rs) 2788bd4c625cSLinus Torvalds && (le32_to_cpu(jh->jh_journal.jp_journal_magic) != 2789bd4c625cSLinus Torvalds sb_jp_journal_magic(rs))) { 2790a9dd3643SJeff Mahoney reiserfs_warning(sb, "sh-460", 279145b03d5eSJeff Mahoney "journal header magic %x (device %s) does " 279245b03d5eSJeff Mahoney "not match to magic found in super block %x", 279345b03d5eSJeff Mahoney jh->jh_journal.jp_journal_magic, 27941da177e4SLinus Torvalds bdevname(journal->j_dev_bd, b), 27951da177e4SLinus Torvalds sb_jp_journal_magic(rs)); 27961da177e4SLinus Torvalds brelse(bhjh); 27971da177e4SLinus Torvalds goto free_and_return; 27981da177e4SLinus Torvalds } 27991da177e4SLinus Torvalds 28001da177e4SLinus Torvalds journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max); 28011da177e4SLinus Torvalds journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch); 2802bd4c625cSLinus Torvalds journal->j_max_commit_age = 2803bd4c625cSLinus Torvalds le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age); 28041da177e4SLinus Torvalds journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE; 28051da177e4SLinus Torvalds 2806a9dd3643SJeff Mahoney if (check_advise_trans_params(sb, journal) != 0) 2807cf3d0b81SEdward Shishkin goto free_and_return; 28081da177e4SLinus Torvalds journal->j_default_max_commit_age = journal->j_max_commit_age; 28091da177e4SLinus Torvalds 28101da177e4SLinus Torvalds if (commit_max_age != 0) { 28111da177e4SLinus Torvalds journal->j_max_commit_age = commit_max_age; 28121da177e4SLinus Torvalds journal->j_max_trans_age = commit_max_age; 28131da177e4SLinus Torvalds } 28141da177e4SLinus Torvalds 2815a9dd3643SJeff Mahoney reiserfs_info(sb, "journal params: device %s, size %u, " 28161da177e4SLinus Torvalds "journal first block %u, max trans len %u, max batch %u, " 28171da177e4SLinus Torvalds "max commit age %u, max trans age %u\n", 28181da177e4SLinus Torvalds bdevname(journal->j_dev_bd, b), 2819a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb), 2820a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb), 28211da177e4SLinus Torvalds journal->j_trans_max, 28221da177e4SLinus Torvalds journal->j_max_batch, 2823bd4c625cSLinus Torvalds journal->j_max_commit_age, journal->j_max_trans_age); 28241da177e4SLinus Torvalds 28251da177e4SLinus Torvalds brelse(bhjh); 28261da177e4SLinus Torvalds 28271da177e4SLinus Torvalds journal->j_list_bitmap_index = 0; 2828a9dd3643SJeff Mahoney journal_list_init(sb); 28291da177e4SLinus Torvalds 2830bd4c625cSLinus Torvalds memset(journal->j_list_hash_table, 0, 2831bd4c625cSLinus Torvalds JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *)); 28321da177e4SLinus Torvalds 28331da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_dirty_buffers); 28341da177e4SLinus Torvalds spin_lock_init(&journal->j_dirty_buffers_lock); 28351da177e4SLinus Torvalds 28361da177e4SLinus Torvalds journal->j_start = 0; 28371da177e4SLinus Torvalds journal->j_len = 0; 28381da177e4SLinus Torvalds journal->j_len_alloc = 0; 2839a228bf8fSJeff Mahoney atomic_set(&journal->j_wcount, 0); 2840a228bf8fSJeff Mahoney atomic_set(&journal->j_async_throttle, 0); 28411da177e4SLinus Torvalds journal->j_bcount = 0; 28421da177e4SLinus Torvalds journal->j_trans_start_time = 0; 28431da177e4SLinus Torvalds journal->j_last = NULL; 28441da177e4SLinus Torvalds journal->j_first = NULL; 2845a228bf8fSJeff Mahoney init_waitqueue_head(&journal->j_join_wait); 2846f68215c4SJeff Mahoney mutex_init(&journal->j_mutex); 2847afe70259SJeff Mahoney mutex_init(&journal->j_flush_mutex); 28481da177e4SLinus Torvalds 28491da177e4SLinus Torvalds journal->j_trans_id = 10; 28501da177e4SLinus Torvalds journal->j_mount_id = 10; 28511da177e4SLinus Torvalds journal->j_state = 0; 2852a228bf8fSJeff Mahoney atomic_set(&journal->j_jlock, 0); 28531da177e4SLinus Torvalds journal->j_cnode_free_list = allocate_cnodes(num_cnodes); 28541da177e4SLinus Torvalds journal->j_cnode_free_orig = journal->j_cnode_free_list; 28551da177e4SLinus Torvalds journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0; 28561da177e4SLinus Torvalds journal->j_cnode_used = 0; 28571da177e4SLinus Torvalds journal->j_must_wait = 0; 28581da177e4SLinus Torvalds 2859576f6d79SJeff Mahoney if (journal->j_cnode_free == 0) { 2860a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-2004", "Journal cnode memory " 2861576f6d79SJeff Mahoney "allocation failed (%ld bytes). Journal is " 2862576f6d79SJeff Mahoney "too large for available memory. Usually " 2863576f6d79SJeff Mahoney "this is due to a journal that is too large.", 2864576f6d79SJeff Mahoney sizeof (struct reiserfs_journal_cnode) * num_cnodes); 2865576f6d79SJeff Mahoney goto free_and_return; 2866576f6d79SJeff Mahoney } 2867576f6d79SJeff Mahoney 2868a9dd3643SJeff Mahoney init_journal_hash(sb); 28691da177e4SLinus Torvalds jl = journal->j_current_jl; 287037c69b98SFrederic Weisbecker 287137c69b98SFrederic Weisbecker /* 287237c69b98SFrederic Weisbecker * get_list_bitmap() may call flush_commit_list() which 287337c69b98SFrederic Weisbecker * requires the lock. Calling flush_commit_list() shouldn't happen 287437c69b98SFrederic Weisbecker * this early but I like to be paranoid. 287537c69b98SFrederic Weisbecker */ 287637c69b98SFrederic Weisbecker reiserfs_write_lock(sb); 2877a9dd3643SJeff Mahoney jl->j_list_bitmap = get_list_bitmap(sb, jl); 287837c69b98SFrederic Weisbecker reiserfs_write_unlock(sb); 28791da177e4SLinus Torvalds if (!jl->j_list_bitmap) { 2880a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-2005", 288145b03d5eSJeff Mahoney "get_list_bitmap failed for journal list 0"); 28821da177e4SLinus Torvalds goto free_and_return; 28831da177e4SLinus Torvalds } 288437c69b98SFrederic Weisbecker 288537c69b98SFrederic Weisbecker ret = journal_read(sb); 288637c69b98SFrederic Weisbecker if (ret < 0) { 2887a9dd3643SJeff Mahoney reiserfs_warning(sb, "reiserfs-2006", 288845b03d5eSJeff Mahoney "Replay Failure, unable to mount"); 28891da177e4SLinus Torvalds goto free_and_return; 28901da177e4SLinus Torvalds } 28911da177e4SLinus Torvalds 2892c4028958SDavid Howells INIT_DELAYED_WORK(&journal->j_work, flush_async_commits); 2893a9dd3643SJeff Mahoney journal->j_work_sb = sb; 28941da177e4SLinus Torvalds return 0; 28951da177e4SLinus Torvalds free_and_return: 2896a9dd3643SJeff Mahoney free_journal_ram(sb); 28971da177e4SLinus Torvalds return 1; 28981da177e4SLinus Torvalds } 28991da177e4SLinus Torvalds 29001da177e4SLinus Torvalds /* 2901098297b2SJeff Mahoney * test for a polite end of the current transaction. Used by file_write, 2902098297b2SJeff Mahoney * and should be used by delete to make sure they don't write more than 2903098297b2SJeff Mahoney * can fit inside a single transaction 29041da177e4SLinus Torvalds */ 2905bd4c625cSLinus Torvalds int journal_transaction_should_end(struct reiserfs_transaction_handle *th, 2906bd4c625cSLinus Torvalds int new_alloc) 2907bd4c625cSLinus Torvalds { 29081da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(th->t_super); 29091da177e4SLinus Torvalds time_t now = get_seconds(); 29101da177e4SLinus Torvalds /* cannot restart while nested */ 29111da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 29121da177e4SLinus Torvalds if (th->t_refcount > 1) 29131da177e4SLinus Torvalds return 0; 29141da177e4SLinus Torvalds if (journal->j_must_wait > 0 || 29151da177e4SLinus Torvalds (journal->j_len_alloc + new_alloc) >= journal->j_max_batch || 2916a228bf8fSJeff Mahoney atomic_read(&journal->j_jlock) || 29171da177e4SLinus Torvalds (now - journal->j_trans_start_time) > journal->j_max_trans_age || 29181da177e4SLinus Torvalds journal->j_cnode_free < (journal->j_trans_max * 3)) { 29191da177e4SLinus Torvalds return 1; 29201da177e4SLinus Torvalds } 2921b18c1c6eSDavidlohr Bueso 29226ae1ea44SChris Mason journal->j_len_alloc += new_alloc; 29236ae1ea44SChris Mason th->t_blocks_allocated += new_alloc ; 29241da177e4SLinus Torvalds return 0; 29251da177e4SLinus Torvalds } 29261da177e4SLinus Torvalds 2927098297b2SJeff Mahoney /* this must be called inside a transaction */ 2928bd4c625cSLinus Torvalds void reiserfs_block_writes(struct reiserfs_transaction_handle *th) 2929bd4c625cSLinus Torvalds { 29301da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(th->t_super); 29311da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 29321da177e4SLinus Torvalds journal->j_must_wait = 1; 29331da177e4SLinus Torvalds set_bit(J_WRITERS_BLOCKED, &journal->j_state); 29341da177e4SLinus Torvalds return; 29351da177e4SLinus Torvalds } 29361da177e4SLinus Torvalds 2937098297b2SJeff Mahoney /* this must be called without a transaction started */ 2938bd4c625cSLinus Torvalds void reiserfs_allow_writes(struct super_block *s) 2939bd4c625cSLinus Torvalds { 29401da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 29411da177e4SLinus Torvalds clear_bit(J_WRITERS_BLOCKED, &journal->j_state); 29421da177e4SLinus Torvalds wake_up(&journal->j_join_wait); 29431da177e4SLinus Torvalds } 29441da177e4SLinus Torvalds 2945098297b2SJeff Mahoney /* this must be called without a transaction started */ 2946bd4c625cSLinus Torvalds void reiserfs_wait_on_write_block(struct super_block *s) 2947bd4c625cSLinus Torvalds { 29481da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 29491da177e4SLinus Torvalds wait_event(journal->j_join_wait, 29501da177e4SLinus Torvalds !test_bit(J_WRITERS_BLOCKED, &journal->j_state)); 29511da177e4SLinus Torvalds } 29521da177e4SLinus Torvalds 2953bd4c625cSLinus Torvalds static void queue_log_writer(struct super_block *s) 2954bd4c625cSLinus Torvalds { 29551da177e4SLinus Torvalds wait_queue_t wait; 29561da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 29571da177e4SLinus Torvalds set_bit(J_WRITERS_QUEUED, &journal->j_state); 29581da177e4SLinus Torvalds 29591da177e4SLinus Torvalds /* 29601da177e4SLinus Torvalds * we don't want to use wait_event here because 29611da177e4SLinus Torvalds * we only want to wait once. 29621da177e4SLinus Torvalds */ 29631da177e4SLinus Torvalds init_waitqueue_entry(&wait, current); 29641da177e4SLinus Torvalds add_wait_queue(&journal->j_join_wait, &wait); 29651da177e4SLinus Torvalds set_current_state(TASK_UNINTERRUPTIBLE); 29668ebc4232SFrederic Weisbecker if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) { 2967278f6679SJeff Mahoney int depth = reiserfs_write_unlock_nested(s); 29681da177e4SLinus Torvalds schedule(); 2969278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 29708ebc4232SFrederic Weisbecker } 29715ab2f7e0SMilind Arun Choudhary __set_current_state(TASK_RUNNING); 29721da177e4SLinus Torvalds remove_wait_queue(&journal->j_join_wait, &wait); 29731da177e4SLinus Torvalds } 29741da177e4SLinus Torvalds 2975bd4c625cSLinus Torvalds static void wake_queued_writers(struct super_block *s) 2976bd4c625cSLinus Torvalds { 29771da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 29781da177e4SLinus Torvalds if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state)) 29791da177e4SLinus Torvalds wake_up(&journal->j_join_wait); 29801da177e4SLinus Torvalds } 29811da177e4SLinus Torvalds 2982600ed416SJeff Mahoney static void let_transaction_grow(struct super_block *sb, unsigned int trans_id) 29831da177e4SLinus Torvalds { 29841da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(sb); 29851da177e4SLinus Torvalds unsigned long bcount = journal->j_bcount; 29861da177e4SLinus Torvalds while (1) { 2987278f6679SJeff Mahoney int depth; 2988278f6679SJeff Mahoney 2989278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(sb); 2990041e0e3bSNishanth Aravamudan schedule_timeout_uninterruptible(1); 2991278f6679SJeff Mahoney reiserfs_write_lock_nested(sb, depth); 2992278f6679SJeff Mahoney 29931da177e4SLinus Torvalds journal->j_current_jl->j_state |= LIST_COMMIT_PENDING; 29941da177e4SLinus Torvalds while ((atomic_read(&journal->j_wcount) > 0 || 29951da177e4SLinus Torvalds atomic_read(&journal->j_jlock)) && 29961da177e4SLinus Torvalds journal->j_trans_id == trans_id) { 29971da177e4SLinus Torvalds queue_log_writer(sb); 29981da177e4SLinus Torvalds } 29991da177e4SLinus Torvalds if (journal->j_trans_id != trans_id) 30001da177e4SLinus Torvalds break; 30011da177e4SLinus Torvalds if (bcount == journal->j_bcount) 30021da177e4SLinus Torvalds break; 30031da177e4SLinus Torvalds bcount = journal->j_bcount; 30041da177e4SLinus Torvalds } 30051da177e4SLinus Torvalds } 30061da177e4SLinus Torvalds 3007098297b2SJeff Mahoney /* 3008098297b2SJeff Mahoney * join == true if you must join an existing transaction. 3009098297b2SJeff Mahoney * join == false if you can deal with waiting for others to finish 3010098297b2SJeff Mahoney * 3011098297b2SJeff Mahoney * this will block until the transaction is joinable. send the number of 3012098297b2SJeff Mahoney * blocks you expect to use in nblocks. 30131da177e4SLinus Torvalds */ 3014bd4c625cSLinus Torvalds static int do_journal_begin_r(struct reiserfs_transaction_handle *th, 3015a9dd3643SJeff Mahoney struct super_block *sb, unsigned long nblocks, 3016bd4c625cSLinus Torvalds int join) 3017bd4c625cSLinus Torvalds { 30181da177e4SLinus Torvalds time_t now = get_seconds(); 3019600ed416SJeff Mahoney unsigned int old_trans_id; 3020a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 30211da177e4SLinus Torvalds struct reiserfs_transaction_handle myth; 30221da177e4SLinus Torvalds int sched_count = 0; 30231da177e4SLinus Torvalds int retval; 3024278f6679SJeff Mahoney int depth; 30251da177e4SLinus Torvalds 3026a9dd3643SJeff Mahoney reiserfs_check_lock_depth(sb, "journal_begin"); 302714a61442SEric Sesterhenn BUG_ON(nblocks > journal->j_trans_max); 30281da177e4SLinus Torvalds 3029a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.journal_being); 30301da177e4SLinus Torvalds /* set here for journal_join */ 30311da177e4SLinus Torvalds th->t_refcount = 1; 3032a9dd3643SJeff Mahoney th->t_super = sb; 30331da177e4SLinus Torvalds 30341da177e4SLinus Torvalds relock: 3035a9dd3643SJeff Mahoney lock_journal(sb); 30361da177e4SLinus Torvalds if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) { 3037a9dd3643SJeff Mahoney unlock_journal(sb); 30381da177e4SLinus Torvalds retval = journal->j_errno; 30391da177e4SLinus Torvalds goto out_fail; 30401da177e4SLinus Torvalds } 30411da177e4SLinus Torvalds journal->j_bcount++; 30421da177e4SLinus Torvalds 30431da177e4SLinus Torvalds if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) { 3044a9dd3643SJeff Mahoney unlock_journal(sb); 3045278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(sb); 3046a9dd3643SJeff Mahoney reiserfs_wait_on_write_block(sb); 3047278f6679SJeff Mahoney reiserfs_write_lock_nested(sb, depth); 3048a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.journal_relock_writers); 30491da177e4SLinus Torvalds goto relock; 30501da177e4SLinus Torvalds } 30511da177e4SLinus Torvalds now = get_seconds(); 30521da177e4SLinus Torvalds 3053098297b2SJeff Mahoney /* 3054098297b2SJeff Mahoney * if there is no room in the journal OR 3055098297b2SJeff Mahoney * if this transaction is too old, and we weren't called joinable, 3056098297b2SJeff Mahoney * wait for it to finish before beginning we don't sleep if there 3057098297b2SJeff Mahoney * aren't other writers 30581da177e4SLinus Torvalds */ 30591da177e4SLinus Torvalds 30601da177e4SLinus Torvalds if ((!join && journal->j_must_wait > 0) || 3061bd4c625cSLinus Torvalds (!join 3062bd4c625cSLinus Torvalds && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch) 3063bd4c625cSLinus Torvalds || (!join && atomic_read(&journal->j_wcount) > 0 3064bd4c625cSLinus Torvalds && journal->j_trans_start_time > 0 3065bd4c625cSLinus Torvalds && (now - journal->j_trans_start_time) > 3066bd4c625cSLinus Torvalds journal->j_max_trans_age) || (!join 3067bd4c625cSLinus Torvalds && atomic_read(&journal->j_jlock)) 3068bd4c625cSLinus Torvalds || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) { 30691da177e4SLinus Torvalds 30701da177e4SLinus Torvalds old_trans_id = journal->j_trans_id; 3071098297b2SJeff Mahoney /* allow others to finish this transaction */ 3072098297b2SJeff Mahoney unlock_journal(sb); 30731da177e4SLinus Torvalds 30741da177e4SLinus Torvalds if (!join && (journal->j_len_alloc + nblocks + 2) >= 30751da177e4SLinus Torvalds journal->j_max_batch && 3076bd4c625cSLinus Torvalds ((journal->j_len + nblocks + 2) * 100) < 3077bd4c625cSLinus Torvalds (journal->j_len_alloc * 75)) { 30781da177e4SLinus Torvalds if (atomic_read(&journal->j_wcount) > 10) { 30791da177e4SLinus Torvalds sched_count++; 3080a9dd3643SJeff Mahoney queue_log_writer(sb); 30811da177e4SLinus Torvalds goto relock; 30821da177e4SLinus Torvalds } 30831da177e4SLinus Torvalds } 3084098297b2SJeff Mahoney /* 3085098297b2SJeff Mahoney * don't mess with joining the transaction if all we 3086098297b2SJeff Mahoney * have to do is wait for someone else to do a commit 30871da177e4SLinus Torvalds */ 30881da177e4SLinus Torvalds if (atomic_read(&journal->j_jlock)) { 30891da177e4SLinus Torvalds while (journal->j_trans_id == old_trans_id && 30901da177e4SLinus Torvalds atomic_read(&journal->j_jlock)) { 3091a9dd3643SJeff Mahoney queue_log_writer(sb); 30921da177e4SLinus Torvalds } 30931da177e4SLinus Torvalds goto relock; 30941da177e4SLinus Torvalds } 3095b491dd17SJeff Mahoney retval = journal_join(&myth, sb); 30961da177e4SLinus Torvalds if (retval) 30971da177e4SLinus Torvalds goto out_fail; 30981da177e4SLinus Torvalds 30991da177e4SLinus Torvalds /* someone might have ended the transaction while we joined */ 31001da177e4SLinus Torvalds if (old_trans_id != journal->j_trans_id) { 310158d85426SJeff Mahoney retval = do_journal_end(&myth, 0); 31021da177e4SLinus Torvalds } else { 310358d85426SJeff Mahoney retval = do_journal_end(&myth, COMMIT_NOW); 31041da177e4SLinus Torvalds } 31051da177e4SLinus Torvalds 31061da177e4SLinus Torvalds if (retval) 31071da177e4SLinus Torvalds goto out_fail; 31081da177e4SLinus Torvalds 3109a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.journal_relock_wcount); 31101da177e4SLinus Torvalds goto relock; 31111da177e4SLinus Torvalds } 31121da177e4SLinus Torvalds /* we are the first writer, set trans_id */ 31131da177e4SLinus Torvalds if (journal->j_trans_start_time == 0) { 31141da177e4SLinus Torvalds journal->j_trans_start_time = get_seconds(); 31151da177e4SLinus Torvalds } 3116a228bf8fSJeff Mahoney atomic_inc(&journal->j_wcount); 31171da177e4SLinus Torvalds journal->j_len_alloc += nblocks; 31181da177e4SLinus Torvalds th->t_blocks_logged = 0; 31191da177e4SLinus Torvalds th->t_blocks_allocated = nblocks; 31201da177e4SLinus Torvalds th->t_trans_id = journal->j_trans_id; 3121a9dd3643SJeff Mahoney unlock_journal(sb); 31221da177e4SLinus Torvalds INIT_LIST_HEAD(&th->t_list); 31231da177e4SLinus Torvalds return 0; 31241da177e4SLinus Torvalds 31251da177e4SLinus Torvalds out_fail: 31261da177e4SLinus Torvalds memset(th, 0, sizeof(*th)); 3127098297b2SJeff Mahoney /* 3128098297b2SJeff Mahoney * Re-set th->t_super, so we can properly keep track of how many 31291da177e4SLinus Torvalds * persistent transactions there are. We need to do this so if this 3130098297b2SJeff Mahoney * call is part of a failed restart_transaction, we can free it later 3131098297b2SJeff Mahoney */ 3132a9dd3643SJeff Mahoney th->t_super = sb; 31331da177e4SLinus Torvalds return retval; 31341da177e4SLinus Torvalds } 31351da177e4SLinus Torvalds 3136bd4c625cSLinus Torvalds struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct 3137bd4c625cSLinus Torvalds super_block 3138bd4c625cSLinus Torvalds *s, 3139bd4c625cSLinus Torvalds int nblocks) 3140bd4c625cSLinus Torvalds { 31411da177e4SLinus Torvalds int ret; 31421da177e4SLinus Torvalds struct reiserfs_transaction_handle *th; 31431da177e4SLinus Torvalds 3144098297b2SJeff Mahoney /* 3145098297b2SJeff Mahoney * if we're nesting into an existing transaction. It will be 3146098297b2SJeff Mahoney * persistent on its own 31471da177e4SLinus Torvalds */ 31481da177e4SLinus Torvalds if (reiserfs_transaction_running(s)) { 31491da177e4SLinus Torvalds th = current->journal_info; 31501da177e4SLinus Torvalds th->t_refcount++; 315114a61442SEric Sesterhenn BUG_ON(th->t_refcount < 2); 315214a61442SEric Sesterhenn 31531da177e4SLinus Torvalds return th; 31541da177e4SLinus Torvalds } 3155d739b42bSPekka Enberg th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS); 31561da177e4SLinus Torvalds if (!th) 31571da177e4SLinus Torvalds return NULL; 31581da177e4SLinus Torvalds ret = journal_begin(th, s, nblocks); 31591da177e4SLinus Torvalds if (ret) { 3160d739b42bSPekka Enberg kfree(th); 31611da177e4SLinus Torvalds return NULL; 31621da177e4SLinus Torvalds } 31631da177e4SLinus Torvalds 31641da177e4SLinus Torvalds SB_JOURNAL(s)->j_persistent_trans++; 31651da177e4SLinus Torvalds return th; 31661da177e4SLinus Torvalds } 31671da177e4SLinus Torvalds 3168bd4c625cSLinus Torvalds int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th) 3169bd4c625cSLinus Torvalds { 31701da177e4SLinus Torvalds struct super_block *s = th->t_super; 31711da177e4SLinus Torvalds int ret = 0; 31721da177e4SLinus Torvalds if (th->t_trans_id) 317358d85426SJeff Mahoney ret = journal_end(th); 31741da177e4SLinus Torvalds else 31751da177e4SLinus Torvalds ret = -EIO; 31761da177e4SLinus Torvalds if (th->t_refcount == 0) { 31771da177e4SLinus Torvalds SB_JOURNAL(s)->j_persistent_trans--; 3178d739b42bSPekka Enberg kfree(th); 31791da177e4SLinus Torvalds } 31801da177e4SLinus Torvalds return ret; 31811da177e4SLinus Torvalds } 31821da177e4SLinus Torvalds 3183bd4c625cSLinus Torvalds static int journal_join(struct reiserfs_transaction_handle *th, 3184b491dd17SJeff Mahoney struct super_block *sb) 3185bd4c625cSLinus Torvalds { 31861da177e4SLinus Torvalds struct reiserfs_transaction_handle *cur_th = current->journal_info; 31871da177e4SLinus Torvalds 3188098297b2SJeff Mahoney /* 3189098297b2SJeff Mahoney * this keeps do_journal_end from NULLing out the 3190098297b2SJeff Mahoney * current->journal_info pointer 31911da177e4SLinus Torvalds */ 31921da177e4SLinus Torvalds th->t_handle_save = cur_th; 319314a61442SEric Sesterhenn BUG_ON(cur_th && cur_th->t_refcount > 1); 3194b491dd17SJeff Mahoney return do_journal_begin_r(th, sb, 1, JBEGIN_JOIN); 31951da177e4SLinus Torvalds } 31961da177e4SLinus Torvalds 3197bd4c625cSLinus Torvalds int journal_join_abort(struct reiserfs_transaction_handle *th, 3198b491dd17SJeff Mahoney struct super_block *sb) 3199bd4c625cSLinus Torvalds { 32001da177e4SLinus Torvalds struct reiserfs_transaction_handle *cur_th = current->journal_info; 32011da177e4SLinus Torvalds 3202098297b2SJeff Mahoney /* 3203098297b2SJeff Mahoney * this keeps do_journal_end from NULLing out the 3204098297b2SJeff Mahoney * current->journal_info pointer 32051da177e4SLinus Torvalds */ 32061da177e4SLinus Torvalds th->t_handle_save = cur_th; 320714a61442SEric Sesterhenn BUG_ON(cur_th && cur_th->t_refcount > 1); 3208b491dd17SJeff Mahoney return do_journal_begin_r(th, sb, 1, JBEGIN_ABORT); 32091da177e4SLinus Torvalds } 32101da177e4SLinus Torvalds 3211bd4c625cSLinus Torvalds int journal_begin(struct reiserfs_transaction_handle *th, 3212a9dd3643SJeff Mahoney struct super_block *sb, unsigned long nblocks) 3213bd4c625cSLinus Torvalds { 32141da177e4SLinus Torvalds struct reiserfs_transaction_handle *cur_th = current->journal_info; 32151da177e4SLinus Torvalds int ret; 32161da177e4SLinus Torvalds 32171da177e4SLinus Torvalds th->t_handle_save = NULL; 32181da177e4SLinus Torvalds if (cur_th) { 32191da177e4SLinus Torvalds /* we are nesting into the current transaction */ 3220a9dd3643SJeff Mahoney if (cur_th->t_super == sb) { 32211da177e4SLinus Torvalds BUG_ON(!cur_th->t_refcount); 32221da177e4SLinus Torvalds cur_th->t_refcount++; 32231da177e4SLinus Torvalds memcpy(th, cur_th, sizeof(*th)); 32241da177e4SLinus Torvalds if (th->t_refcount <= 1) 3225a9dd3643SJeff Mahoney reiserfs_warning(sb, "reiserfs-2005", 322645b03d5eSJeff Mahoney "BAD: refcount <= 1, but " 322745b03d5eSJeff Mahoney "journal_info != 0"); 32281da177e4SLinus Torvalds return 0; 32291da177e4SLinus Torvalds } else { 3230098297b2SJeff Mahoney /* 3231098297b2SJeff Mahoney * we've ended up with a handle from a different 3232098297b2SJeff Mahoney * filesystem. save it and restore on journal_end. 3233098297b2SJeff Mahoney * This should never really happen... 32341da177e4SLinus Torvalds */ 3235a9dd3643SJeff Mahoney reiserfs_warning(sb, "clm-2100", 323645b03d5eSJeff Mahoney "nesting info a different FS"); 32371da177e4SLinus Torvalds th->t_handle_save = current->journal_info; 32381da177e4SLinus Torvalds current->journal_info = th; 32391da177e4SLinus Torvalds } 32401da177e4SLinus Torvalds } else { 32411da177e4SLinus Torvalds current->journal_info = th; 32421da177e4SLinus Torvalds } 3243a9dd3643SJeff Mahoney ret = do_journal_begin_r(th, sb, nblocks, JBEGIN_REG); 324414a61442SEric Sesterhenn BUG_ON(current->journal_info != th); 32451da177e4SLinus Torvalds 3246098297b2SJeff Mahoney /* 3247098297b2SJeff Mahoney * I guess this boils down to being the reciprocal of clm-2100 above. 3248098297b2SJeff Mahoney * If do_journal_begin_r fails, we need to put it back, since 3249098297b2SJeff Mahoney * journal_end won't be called to do it. */ 32501da177e4SLinus Torvalds if (ret) 32511da177e4SLinus Torvalds current->journal_info = th->t_handle_save; 32521da177e4SLinus Torvalds else 32531da177e4SLinus Torvalds BUG_ON(!th->t_refcount); 32541da177e4SLinus Torvalds 32551da177e4SLinus Torvalds return ret; 32561da177e4SLinus Torvalds } 32571da177e4SLinus Torvalds 32581da177e4SLinus Torvalds /* 3259098297b2SJeff Mahoney * puts bh into the current transaction. If it was already there, reorders 3260098297b2SJeff Mahoney * removes the old pointers from the hash, and puts new ones in (to make 3261098297b2SJeff Mahoney * sure replay happen in the right order). 3262098297b2SJeff Mahoney * 3263098297b2SJeff Mahoney * if it was dirty, cleans and files onto the clean list. I can't let it 3264098297b2SJeff Mahoney * be dirty again until the transaction is committed. 3265098297b2SJeff Mahoney * 3266098297b2SJeff Mahoney * if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len. 32671da177e4SLinus Torvalds */ 3268bd4c625cSLinus Torvalds int journal_mark_dirty(struct reiserfs_transaction_handle *th, 326909f1b80bSJeff Mahoney struct buffer_head *bh) 3270bd4c625cSLinus Torvalds { 327109f1b80bSJeff Mahoney struct super_block *sb = th->t_super; 3272a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 32731da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn = NULL; 32741da177e4SLinus Torvalds int count_already_incd = 0; 32751da177e4SLinus Torvalds int prepared = 0; 32761da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 32771da177e4SLinus Torvalds 3278a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.mark_dirty); 32791da177e4SLinus Torvalds if (th->t_trans_id != journal->j_trans_id) { 3280c3a9c210SJeff Mahoney reiserfs_panic(th->t_super, "journal-1577", 3281c3a9c210SJeff Mahoney "handle trans id %ld != current trans id %ld", 32821da177e4SLinus Torvalds th->t_trans_id, journal->j_trans_id); 32831da177e4SLinus Torvalds } 32841da177e4SLinus Torvalds 32851da177e4SLinus Torvalds prepared = test_clear_buffer_journal_prepared(bh); 32861da177e4SLinus Torvalds clear_buffer_journal_restore_dirty(bh); 32871da177e4SLinus Torvalds /* already in this transaction, we are done */ 32881da177e4SLinus Torvalds if (buffer_journaled(bh)) { 3289a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.mark_dirty_already); 32901da177e4SLinus Torvalds return 0; 32911da177e4SLinus Torvalds } 32921da177e4SLinus Torvalds 3293098297b2SJeff Mahoney /* 3294098297b2SJeff Mahoney * this must be turned into a panic instead of a warning. We can't 3295098297b2SJeff Mahoney * allow a dirty or journal_dirty or locked buffer to be logged, as 3296098297b2SJeff Mahoney * some changes could get to disk too early. NOT GOOD. 32971da177e4SLinus Torvalds */ 32981da177e4SLinus Torvalds if (!prepared || buffer_dirty(bh)) { 3299a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1777", 330045b03d5eSJeff Mahoney "buffer %llu bad state " 33011da177e4SLinus Torvalds "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT", 3302bd4c625cSLinus Torvalds (unsigned long long)bh->b_blocknr, 3303bd4c625cSLinus Torvalds prepared ? ' ' : '!', 33041da177e4SLinus Torvalds buffer_locked(bh) ? ' ' : '!', 33051da177e4SLinus Torvalds buffer_dirty(bh) ? ' ' : '!', 33061da177e4SLinus Torvalds buffer_journal_dirty(bh) ? ' ' : '!'); 33071da177e4SLinus Torvalds } 33081da177e4SLinus Torvalds 3309a228bf8fSJeff Mahoney if (atomic_read(&journal->j_wcount) <= 0) { 3310a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1409", 331145b03d5eSJeff Mahoney "returning because j_wcount was %d", 3312a228bf8fSJeff Mahoney atomic_read(&journal->j_wcount)); 33131da177e4SLinus Torvalds return 1; 33141da177e4SLinus Torvalds } 3315098297b2SJeff Mahoney /* 3316098297b2SJeff Mahoney * this error means I've screwed up, and we've overflowed 3317098297b2SJeff Mahoney * the transaction. Nothing can be done here, except make the 3318098297b2SJeff Mahoney * FS readonly or panic. 33191da177e4SLinus Torvalds */ 33201da177e4SLinus Torvalds if (journal->j_len >= journal->j_trans_max) { 3321c3a9c210SJeff Mahoney reiserfs_panic(th->t_super, "journal-1413", 3322c3a9c210SJeff Mahoney "j_len (%lu) is too big", 3323bd4c625cSLinus Torvalds journal->j_len); 33241da177e4SLinus Torvalds } 33251da177e4SLinus Torvalds 33261da177e4SLinus Torvalds if (buffer_journal_dirty(bh)) { 33271da177e4SLinus Torvalds count_already_incd = 1; 3328a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.mark_dirty_notjournal); 33291da177e4SLinus Torvalds clear_buffer_journal_dirty(bh); 33301da177e4SLinus Torvalds } 33311da177e4SLinus Torvalds 33321da177e4SLinus Torvalds if (journal->j_len > journal->j_len_alloc) { 33331da177e4SLinus Torvalds journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT; 33341da177e4SLinus Torvalds } 33351da177e4SLinus Torvalds 33361da177e4SLinus Torvalds set_buffer_journaled(bh); 33371da177e4SLinus Torvalds 33381da177e4SLinus Torvalds /* now put this guy on the end */ 33391da177e4SLinus Torvalds if (!cn) { 3340a9dd3643SJeff Mahoney cn = get_cnode(sb); 33411da177e4SLinus Torvalds if (!cn) { 3342a9dd3643SJeff Mahoney reiserfs_panic(sb, "journal-4", "get_cnode failed!"); 33431da177e4SLinus Torvalds } 33441da177e4SLinus Torvalds 33451da177e4SLinus Torvalds if (th->t_blocks_logged == th->t_blocks_allocated) { 33461da177e4SLinus Torvalds th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT; 33471da177e4SLinus Torvalds journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT; 33481da177e4SLinus Torvalds } 33491da177e4SLinus Torvalds th->t_blocks_logged++; 33501da177e4SLinus Torvalds journal->j_len++; 33511da177e4SLinus Torvalds 33521da177e4SLinus Torvalds cn->bh = bh; 33531da177e4SLinus Torvalds cn->blocknr = bh->b_blocknr; 3354a9dd3643SJeff Mahoney cn->sb = sb; 33551da177e4SLinus Torvalds cn->jlist = NULL; 33561da177e4SLinus Torvalds insert_journal_hash(journal->j_hash_table, cn); 33571da177e4SLinus Torvalds if (!count_already_incd) { 33581da177e4SLinus Torvalds get_bh(bh); 33591da177e4SLinus Torvalds } 33601da177e4SLinus Torvalds } 33611da177e4SLinus Torvalds cn->next = NULL; 33621da177e4SLinus Torvalds cn->prev = journal->j_last; 33631da177e4SLinus Torvalds cn->bh = bh; 33641da177e4SLinus Torvalds if (journal->j_last) { 33651da177e4SLinus Torvalds journal->j_last->next = cn; 33661da177e4SLinus Torvalds journal->j_last = cn; 33671da177e4SLinus Torvalds } else { 33681da177e4SLinus Torvalds journal->j_first = cn; 33691da177e4SLinus Torvalds journal->j_last = cn; 33701da177e4SLinus Torvalds } 3371033369d1SArtem Bityutskiy reiserfs_schedule_old_flush(sb); 33721da177e4SLinus Torvalds return 0; 33731da177e4SLinus Torvalds } 33741da177e4SLinus Torvalds 337558d85426SJeff Mahoney int journal_end(struct reiserfs_transaction_handle *th) 3376bd4c625cSLinus Torvalds { 337758d85426SJeff Mahoney struct super_block *sb = th->t_super; 33781da177e4SLinus Torvalds if (!current->journal_info && th->t_refcount > 1) 3379a9dd3643SJeff Mahoney reiserfs_warning(sb, "REISER-NESTING", 338045b03d5eSJeff Mahoney "th NULL, refcount %d", th->t_refcount); 33811da177e4SLinus Torvalds 33821da177e4SLinus Torvalds if (!th->t_trans_id) { 33831da177e4SLinus Torvalds WARN_ON(1); 33841da177e4SLinus Torvalds return -EIO; 33851da177e4SLinus Torvalds } 33861da177e4SLinus Torvalds 33871da177e4SLinus Torvalds th->t_refcount--; 33881da177e4SLinus Torvalds if (th->t_refcount > 0) { 3389bd4c625cSLinus Torvalds struct reiserfs_transaction_handle *cur_th = 3390bd4c625cSLinus Torvalds current->journal_info; 33911da177e4SLinus Torvalds 3392098297b2SJeff Mahoney /* 3393098297b2SJeff Mahoney * we aren't allowed to close a nested transaction on a 3394098297b2SJeff Mahoney * different filesystem from the one in the task struct 33951da177e4SLinus Torvalds */ 339614a61442SEric Sesterhenn BUG_ON(cur_th->t_super != th->t_super); 33971da177e4SLinus Torvalds 33981da177e4SLinus Torvalds if (th != cur_th) { 33991da177e4SLinus Torvalds memcpy(current->journal_info, th, sizeof(*th)); 34001da177e4SLinus Torvalds th->t_trans_id = 0; 34011da177e4SLinus Torvalds } 34021da177e4SLinus Torvalds return 0; 34031da177e4SLinus Torvalds } else { 340458d85426SJeff Mahoney return do_journal_end(th, 0); 34051da177e4SLinus Torvalds } 34061da177e4SLinus Torvalds } 34071da177e4SLinus Torvalds 3408098297b2SJeff Mahoney /* 3409098297b2SJeff Mahoney * removes from the current transaction, relsing and descrementing any counters. 3410098297b2SJeff Mahoney * also files the removed buffer directly onto the clean list 3411098297b2SJeff Mahoney * 3412098297b2SJeff Mahoney * called by journal_mark_freed when a block has been deleted 3413098297b2SJeff Mahoney * 3414098297b2SJeff Mahoney * returns 1 if it cleaned and relsed the buffer. 0 otherwise 34151da177e4SLinus Torvalds */ 3416a9dd3643SJeff Mahoney static int remove_from_transaction(struct super_block *sb, 3417bd4c625cSLinus Torvalds b_blocknr_t blocknr, int already_cleaned) 3418bd4c625cSLinus Torvalds { 34191da177e4SLinus Torvalds struct buffer_head *bh; 34201da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 3421a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 34221da177e4SLinus Torvalds int ret = 0; 34231da177e4SLinus Torvalds 3424a9dd3643SJeff Mahoney cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr); 34251da177e4SLinus Torvalds if (!cn || !cn->bh) { 34261da177e4SLinus Torvalds return ret; 34271da177e4SLinus Torvalds } 34281da177e4SLinus Torvalds bh = cn->bh; 34291da177e4SLinus Torvalds if (cn->prev) { 34301da177e4SLinus Torvalds cn->prev->next = cn->next; 34311da177e4SLinus Torvalds } 34321da177e4SLinus Torvalds if (cn->next) { 34331da177e4SLinus Torvalds cn->next->prev = cn->prev; 34341da177e4SLinus Torvalds } 34351da177e4SLinus Torvalds if (cn == journal->j_first) { 34361da177e4SLinus Torvalds journal->j_first = cn->next; 34371da177e4SLinus Torvalds } 34381da177e4SLinus Torvalds if (cn == journal->j_last) { 34391da177e4SLinus Torvalds journal->j_last = cn->prev; 34401da177e4SLinus Torvalds } 34411da177e4SLinus Torvalds if (bh) 3442a9dd3643SJeff Mahoney remove_journal_hash(sb, journal->j_hash_table, NULL, 3443bd4c625cSLinus Torvalds bh->b_blocknr, 0); 34441da177e4SLinus Torvalds clear_buffer_journaled(bh); /* don't log this one */ 34451da177e4SLinus Torvalds 34461da177e4SLinus Torvalds if (!already_cleaned) { 34471da177e4SLinus Torvalds clear_buffer_journal_dirty(bh); 34481da177e4SLinus Torvalds clear_buffer_dirty(bh); 34491da177e4SLinus Torvalds clear_buffer_journal_test(bh); 34501da177e4SLinus Torvalds put_bh(bh); 3451a228bf8fSJeff Mahoney if (atomic_read(&bh->b_count) < 0) { 3452a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1752", 345345b03d5eSJeff Mahoney "b_count < 0"); 34541da177e4SLinus Torvalds } 34551da177e4SLinus Torvalds ret = 1; 34561da177e4SLinus Torvalds } 34571da177e4SLinus Torvalds journal->j_len--; 34581da177e4SLinus Torvalds journal->j_len_alloc--; 3459a9dd3643SJeff Mahoney free_cnode(sb, cn); 34601da177e4SLinus Torvalds return ret; 34611da177e4SLinus Torvalds } 34621da177e4SLinus Torvalds 34631da177e4SLinus Torvalds /* 3464098297b2SJeff Mahoney * for any cnode in a journal list, it can only be dirtied of all the 3465098297b2SJeff Mahoney * transactions that include it are committed to disk. 3466098297b2SJeff Mahoney * this checks through each transaction, and returns 1 if you are allowed 3467098297b2SJeff Mahoney * to dirty, and 0 if you aren't 3468098297b2SJeff Mahoney * 3469098297b2SJeff Mahoney * it is called by dirty_journal_list, which is called after 3470098297b2SJeff Mahoney * flush_commit_list has gotten all the log blocks for a given 3471098297b2SJeff Mahoney * transaction on disk 3472098297b2SJeff Mahoney * 34731da177e4SLinus Torvalds */ 3474bd4c625cSLinus Torvalds static int can_dirty(struct reiserfs_journal_cnode *cn) 3475bd4c625cSLinus Torvalds { 34761da177e4SLinus Torvalds struct super_block *sb = cn->sb; 34771da177e4SLinus Torvalds b_blocknr_t blocknr = cn->blocknr; 34781da177e4SLinus Torvalds struct reiserfs_journal_cnode *cur = cn->hprev; 34791da177e4SLinus Torvalds int can_dirty = 1; 34801da177e4SLinus Torvalds 3481098297b2SJeff Mahoney /* 3482098297b2SJeff Mahoney * first test hprev. These are all newer than cn, so any node here 3483098297b2SJeff Mahoney * with the same block number and dev means this node can't be sent 3484098297b2SJeff Mahoney * to disk right now. 34851da177e4SLinus Torvalds */ 34861da177e4SLinus Torvalds while (cur && can_dirty) { 34871da177e4SLinus Torvalds if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb && 34881da177e4SLinus Torvalds cur->blocknr == blocknr) { 34891da177e4SLinus Torvalds can_dirty = 0; 34901da177e4SLinus Torvalds } 34911da177e4SLinus Torvalds cur = cur->hprev; 34921da177e4SLinus Torvalds } 3493098297b2SJeff Mahoney /* 3494098297b2SJeff Mahoney * then test hnext. These are all older than cn. As long as they 3495098297b2SJeff Mahoney * are committed to the log, it is safe to write cn to disk 34961da177e4SLinus Torvalds */ 34971da177e4SLinus Torvalds cur = cn->hnext; 34981da177e4SLinus Torvalds while (cur && can_dirty) { 34991da177e4SLinus Torvalds if (cur->jlist && cur->jlist->j_len > 0 && 3500a228bf8fSJeff Mahoney atomic_read(&cur->jlist->j_commit_left) > 0 && cur->bh && 35011da177e4SLinus Torvalds cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) { 35021da177e4SLinus Torvalds can_dirty = 0; 35031da177e4SLinus Torvalds } 35041da177e4SLinus Torvalds cur = cur->hnext; 35051da177e4SLinus Torvalds } 35061da177e4SLinus Torvalds return can_dirty; 35071da177e4SLinus Torvalds } 35081da177e4SLinus Torvalds 3509098297b2SJeff Mahoney /* 3510098297b2SJeff Mahoney * syncs the commit blocks, but does not force the real buffers to disk 3511098297b2SJeff Mahoney * will wait until the current transaction is done/committed before returning 35121da177e4SLinus Torvalds */ 351358d85426SJeff Mahoney int journal_end_sync(struct reiserfs_transaction_handle *th) 3514bd4c625cSLinus Torvalds { 351558d85426SJeff Mahoney struct super_block *sb = th->t_super; 3516a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 35171da177e4SLinus Torvalds 35181da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 35191da177e4SLinus Torvalds /* you can sync while nested, very, very bad */ 352014a61442SEric Sesterhenn BUG_ON(th->t_refcount > 1); 35211da177e4SLinus Torvalds if (journal->j_len == 0) { 3522a9dd3643SJeff Mahoney reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb), 3523bd4c625cSLinus Torvalds 1); 352409f1b80bSJeff Mahoney journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb)); 35251da177e4SLinus Torvalds } 352658d85426SJeff Mahoney return do_journal_end(th, COMMIT_NOW | WAIT); 35271da177e4SLinus Torvalds } 35281da177e4SLinus Torvalds 3529098297b2SJeff Mahoney /* writeback the pending async commits to disk */ 3530c4028958SDavid Howells static void flush_async_commits(struct work_struct *work) 3531bd4c625cSLinus Torvalds { 3532c4028958SDavid Howells struct reiserfs_journal *journal = 3533c4028958SDavid Howells container_of(work, struct reiserfs_journal, j_work.work); 3534a9dd3643SJeff Mahoney struct super_block *sb = journal->j_work_sb; 35351da177e4SLinus Torvalds struct reiserfs_journal_list *jl; 35361da177e4SLinus Torvalds struct list_head *entry; 35371da177e4SLinus Torvalds 35388ebc4232SFrederic Weisbecker reiserfs_write_lock(sb); 35391da177e4SLinus Torvalds if (!list_empty(&journal->j_journal_list)) { 35401da177e4SLinus Torvalds /* last entry is the youngest, commit it and you get everything */ 35411da177e4SLinus Torvalds entry = journal->j_journal_list.prev; 35421da177e4SLinus Torvalds jl = JOURNAL_LIST_ENTRY(entry); 3543a9dd3643SJeff Mahoney flush_commit_list(sb, jl, 1); 35441da177e4SLinus Torvalds } 35458ebc4232SFrederic Weisbecker reiserfs_write_unlock(sb); 35461da177e4SLinus Torvalds } 35471da177e4SLinus Torvalds 35481da177e4SLinus Torvalds /* 3549098297b2SJeff Mahoney * flushes any old transactions to disk 3550098297b2SJeff Mahoney * ends the current transaction if it is too old 35511da177e4SLinus Torvalds */ 355225729b0eSArtem Bityutskiy void reiserfs_flush_old_commits(struct super_block *sb) 3553bd4c625cSLinus Torvalds { 35541da177e4SLinus Torvalds time_t now; 35551da177e4SLinus Torvalds struct reiserfs_transaction_handle th; 3556a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 35571da177e4SLinus Torvalds 35581da177e4SLinus Torvalds now = get_seconds(); 3559098297b2SJeff Mahoney /* 3560098297b2SJeff Mahoney * safety check so we don't flush while we are replaying the log during 35611da177e4SLinus Torvalds * mount 35621da177e4SLinus Torvalds */ 356325729b0eSArtem Bityutskiy if (list_empty(&journal->j_journal_list)) 356425729b0eSArtem Bityutskiy return; 35651da177e4SLinus Torvalds 3566098297b2SJeff Mahoney /* 3567098297b2SJeff Mahoney * check the current transaction. If there are no writers, and it is 35681da177e4SLinus Torvalds * too old, finish it, and force the commit blocks to disk 35691da177e4SLinus Torvalds */ 35701da177e4SLinus Torvalds if (atomic_read(&journal->j_wcount) <= 0 && 35711da177e4SLinus Torvalds journal->j_trans_start_time > 0 && 35721da177e4SLinus Torvalds journal->j_len > 0 && 3573bd4c625cSLinus Torvalds (now - journal->j_trans_start_time) > journal->j_max_trans_age) { 3574b491dd17SJeff Mahoney if (!journal_join(&th, sb)) { 3575a9dd3643SJeff Mahoney reiserfs_prepare_for_journal(sb, 3576a9dd3643SJeff Mahoney SB_BUFFER_WITH_SB(sb), 3577bd4c625cSLinus Torvalds 1); 357809f1b80bSJeff Mahoney journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb)); 35791da177e4SLinus Torvalds 3580098297b2SJeff Mahoney /* 3581098297b2SJeff Mahoney * we're only being called from kreiserfsd, it makes 3582098297b2SJeff Mahoney * no sense to do an async commit so that kreiserfsd 3583098297b2SJeff Mahoney * can do it later 35841da177e4SLinus Torvalds */ 358558d85426SJeff Mahoney do_journal_end(&th, COMMIT_NOW | WAIT); 35861da177e4SLinus Torvalds } 35871da177e4SLinus Torvalds } 35881da177e4SLinus Torvalds } 35891da177e4SLinus Torvalds 35901da177e4SLinus Torvalds /* 3591098297b2SJeff Mahoney * returns 0 if do_journal_end should return right away, returns 1 if 3592098297b2SJeff Mahoney * do_journal_end should finish the commit 3593098297b2SJeff Mahoney * 3594098297b2SJeff Mahoney * if the current transaction is too old, but still has writers, this will 3595098297b2SJeff Mahoney * wait on j_join_wait until all the writers are done. By the time it 3596098297b2SJeff Mahoney * wakes up, the transaction it was called has already ended, so it just 3597098297b2SJeff Mahoney * flushes the commit list and returns 0. 3598098297b2SJeff Mahoney * 3599098297b2SJeff Mahoney * Won't batch when flush or commit_now is set. Also won't batch when 3600098297b2SJeff Mahoney * others are waiting on j_join_wait. 3601098297b2SJeff Mahoney * 3602098297b2SJeff Mahoney * Note, we can't allow the journal_end to proceed while there are still 3603098297b2SJeff Mahoney * writers in the log. 36041da177e4SLinus Torvalds */ 360558d85426SJeff Mahoney static int check_journal_end(struct reiserfs_transaction_handle *th, int flags) 3606bd4c625cSLinus Torvalds { 36071da177e4SLinus Torvalds 36081da177e4SLinus Torvalds time_t now; 36091da177e4SLinus Torvalds int flush = flags & FLUSH_ALL; 36101da177e4SLinus Torvalds int commit_now = flags & COMMIT_NOW; 36111da177e4SLinus Torvalds int wait_on_commit = flags & WAIT; 36121da177e4SLinus Torvalds struct reiserfs_journal_list *jl; 361358d85426SJeff Mahoney struct super_block *sb = th->t_super; 3614a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 36151da177e4SLinus Torvalds 36161da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 36171da177e4SLinus Torvalds 36181da177e4SLinus Torvalds if (th->t_trans_id != journal->j_trans_id) { 3619c3a9c210SJeff Mahoney reiserfs_panic(th->t_super, "journal-1577", 3620c3a9c210SJeff Mahoney "handle trans id %ld != current trans id %ld", 36211da177e4SLinus Torvalds th->t_trans_id, journal->j_trans_id); 36221da177e4SLinus Torvalds } 36231da177e4SLinus Torvalds 36241da177e4SLinus Torvalds journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged); 3625098297b2SJeff Mahoney /* <= 0 is allowed. unmounting might not call begin */ 3626a228bf8fSJeff Mahoney if (atomic_read(&journal->j_wcount) > 0) 3627a228bf8fSJeff Mahoney atomic_dec(&journal->j_wcount); 36281da177e4SLinus Torvalds 3629098297b2SJeff Mahoney /* 3630098297b2SJeff Mahoney * BUG, deal with case where j_len is 0, but people previously 3631098297b2SJeff Mahoney * freed blocks need to be released will be dealt with by next 3632098297b2SJeff Mahoney * transaction that actually writes something, but should be taken 3633098297b2SJeff Mahoney * care of in this trans 36341da177e4SLinus Torvalds */ 363514a61442SEric Sesterhenn BUG_ON(journal->j_len == 0); 363614a61442SEric Sesterhenn 3637098297b2SJeff Mahoney /* 3638098297b2SJeff Mahoney * if wcount > 0, and we are called to with flush or commit_now, 3639098297b2SJeff Mahoney * we wait on j_join_wait. We will wake up when the last writer has 3640098297b2SJeff Mahoney * finished the transaction, and started it on its way to the disk. 3641098297b2SJeff Mahoney * Then, we flush the commit or journal list, and just return 0 3642098297b2SJeff Mahoney * because the rest of journal end was already done for this 3643098297b2SJeff Mahoney * transaction. 36441da177e4SLinus Torvalds */ 3645a228bf8fSJeff Mahoney if (atomic_read(&journal->j_wcount) > 0) { 36461da177e4SLinus Torvalds if (flush || commit_now) { 36471da177e4SLinus Torvalds unsigned trans_id; 36481da177e4SLinus Torvalds 36491da177e4SLinus Torvalds jl = journal->j_current_jl; 36501da177e4SLinus Torvalds trans_id = jl->j_trans_id; 36511da177e4SLinus Torvalds if (wait_on_commit) 36521da177e4SLinus Torvalds jl->j_state |= LIST_COMMIT_PENDING; 3653a228bf8fSJeff Mahoney atomic_set(&journal->j_jlock, 1); 36541da177e4SLinus Torvalds if (flush) { 36551da177e4SLinus Torvalds journal->j_next_full_flush = 1; 36561da177e4SLinus Torvalds } 3657a9dd3643SJeff Mahoney unlock_journal(sb); 36581da177e4SLinus Torvalds 3659098297b2SJeff Mahoney /* 3660098297b2SJeff Mahoney * sleep while the current transaction is 3661098297b2SJeff Mahoney * still j_jlocked 3662098297b2SJeff Mahoney */ 36631da177e4SLinus Torvalds while (journal->j_trans_id == trans_id) { 36641da177e4SLinus Torvalds if (atomic_read(&journal->j_jlock)) { 3665a9dd3643SJeff Mahoney queue_log_writer(sb); 36661da177e4SLinus Torvalds } else { 3667a9dd3643SJeff Mahoney lock_journal(sb); 36681da177e4SLinus Torvalds if (journal->j_trans_id == trans_id) { 3669a228bf8fSJeff Mahoney atomic_set(&journal->j_jlock, 3670bd4c625cSLinus Torvalds 1); 36711da177e4SLinus Torvalds } 3672a9dd3643SJeff Mahoney unlock_journal(sb); 36731da177e4SLinus Torvalds } 36741da177e4SLinus Torvalds } 367514a61442SEric Sesterhenn BUG_ON(journal->j_trans_id == trans_id); 367614a61442SEric Sesterhenn 3677bd4c625cSLinus Torvalds if (commit_now 3678a9dd3643SJeff Mahoney && journal_list_still_alive(sb, trans_id) 3679bd4c625cSLinus Torvalds && wait_on_commit) { 3680a9dd3643SJeff Mahoney flush_commit_list(sb, jl, 1); 36811da177e4SLinus Torvalds } 36821da177e4SLinus Torvalds return 0; 36831da177e4SLinus Torvalds } 3684a9dd3643SJeff Mahoney unlock_journal(sb); 36851da177e4SLinus Torvalds return 0; 36861da177e4SLinus Torvalds } 36871da177e4SLinus Torvalds 36881da177e4SLinus Torvalds /* deal with old transactions where we are the last writers */ 36891da177e4SLinus Torvalds now = get_seconds(); 36901da177e4SLinus Torvalds if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) { 36911da177e4SLinus Torvalds commit_now = 1; 36921da177e4SLinus Torvalds journal->j_next_async_flush = 1; 36931da177e4SLinus Torvalds } 36941da177e4SLinus Torvalds /* don't batch when someone is waiting on j_join_wait */ 36951da177e4SLinus Torvalds /* don't batch when syncing the commit or flushing the whole trans */ 3696a228bf8fSJeff Mahoney if (!(journal->j_must_wait > 0) && !(atomic_read(&journal->j_jlock)) 3697bd4c625cSLinus Torvalds && !flush && !commit_now && (journal->j_len < journal->j_max_batch) 3698bd4c625cSLinus Torvalds && journal->j_len_alloc < journal->j_max_batch 3699bd4c625cSLinus Torvalds && journal->j_cnode_free > (journal->j_trans_max * 3)) { 37001da177e4SLinus Torvalds journal->j_bcount++; 3701a9dd3643SJeff Mahoney unlock_journal(sb); 37021da177e4SLinus Torvalds return 0; 37031da177e4SLinus Torvalds } 37041da177e4SLinus Torvalds 3705a9dd3643SJeff Mahoney if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(sb)) { 3706a9dd3643SJeff Mahoney reiserfs_panic(sb, "journal-003", 3707c3a9c210SJeff Mahoney "j_start (%ld) is too high", 3708bd4c625cSLinus Torvalds journal->j_start); 37091da177e4SLinus Torvalds } 37101da177e4SLinus Torvalds return 1; 37111da177e4SLinus Torvalds } 37121da177e4SLinus Torvalds 37131da177e4SLinus Torvalds /* 3714098297b2SJeff Mahoney * Does all the work that makes deleting blocks safe. 3715098297b2SJeff Mahoney * when deleting a block mark BH_JNew, just remove it from the current 3716098297b2SJeff Mahoney * transaction, clean it's buffer_head and move on. 3717098297b2SJeff Mahoney * 3718098297b2SJeff Mahoney * otherwise: 3719098297b2SJeff Mahoney * set a bit for the block in the journal bitmap. That will prevent it from 3720098297b2SJeff Mahoney * being allocated for unformatted nodes before this transaction has finished. 3721098297b2SJeff Mahoney * 3722098297b2SJeff Mahoney * mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers. 3723098297b2SJeff Mahoney * That will prevent any old transactions with this block from trying to flush 3724098297b2SJeff Mahoney * to the real location. Since we aren't removing the cnode from the 3725098297b2SJeff Mahoney * journal_list_hash, *the block can't be reallocated yet. 3726098297b2SJeff Mahoney * 3727098297b2SJeff Mahoney * Then remove it from the current transaction, decrementing any counters and 3728098297b2SJeff Mahoney * filing it on the clean list. 37291da177e4SLinus Torvalds */ 3730bd4c625cSLinus Torvalds int journal_mark_freed(struct reiserfs_transaction_handle *th, 3731a9dd3643SJeff Mahoney struct super_block *sb, b_blocknr_t blocknr) 3732bd4c625cSLinus Torvalds { 3733a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 37341da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn = NULL; 37351da177e4SLinus Torvalds struct buffer_head *bh = NULL; 37361da177e4SLinus Torvalds struct reiserfs_list_bitmap *jb = NULL; 37371da177e4SLinus Torvalds int cleaned = 0; 37381da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 37391da177e4SLinus Torvalds 3740a9dd3643SJeff Mahoney cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr); 37411da177e4SLinus Torvalds if (cn && cn->bh) { 37421da177e4SLinus Torvalds bh = cn->bh; 37431da177e4SLinus Torvalds get_bh(bh); 37441da177e4SLinus Torvalds } 37451da177e4SLinus Torvalds /* if it is journal new, we just remove it from this transaction */ 37461da177e4SLinus Torvalds if (bh && buffer_journal_new(bh)) { 37471da177e4SLinus Torvalds clear_buffer_journal_new(bh); 37481da177e4SLinus Torvalds clear_prepared_bits(bh); 37491da177e4SLinus Torvalds reiserfs_clean_and_file_buffer(bh); 3750a9dd3643SJeff Mahoney cleaned = remove_from_transaction(sb, blocknr, cleaned); 37511da177e4SLinus Torvalds } else { 3752098297b2SJeff Mahoney /* 3753098297b2SJeff Mahoney * set the bit for this block in the journal bitmap 3754098297b2SJeff Mahoney * for this transaction 3755098297b2SJeff Mahoney */ 37561da177e4SLinus Torvalds jb = journal->j_current_jl->j_list_bitmap; 37571da177e4SLinus Torvalds if (!jb) { 3758a9dd3643SJeff Mahoney reiserfs_panic(sb, "journal-1702", 3759c3a9c210SJeff Mahoney "journal_list_bitmap is NULL"); 37601da177e4SLinus Torvalds } 3761a9dd3643SJeff Mahoney set_bit_in_list_bitmap(sb, blocknr, jb); 37621da177e4SLinus Torvalds 37631da177e4SLinus Torvalds /* Note, the entire while loop is not allowed to schedule. */ 37641da177e4SLinus Torvalds 37651da177e4SLinus Torvalds if (bh) { 37661da177e4SLinus Torvalds clear_prepared_bits(bh); 37671da177e4SLinus Torvalds reiserfs_clean_and_file_buffer(bh); 37681da177e4SLinus Torvalds } 3769a9dd3643SJeff Mahoney cleaned = remove_from_transaction(sb, blocknr, cleaned); 37701da177e4SLinus Torvalds 3771098297b2SJeff Mahoney /* 3772098297b2SJeff Mahoney * find all older transactions with this block, 3773098297b2SJeff Mahoney * make sure they don't try to write it out 3774098297b2SJeff Mahoney */ 3775a9dd3643SJeff Mahoney cn = get_journal_hash_dev(sb, journal->j_list_hash_table, 3776bd4c625cSLinus Torvalds blocknr); 37771da177e4SLinus Torvalds while (cn) { 3778a9dd3643SJeff Mahoney if (sb == cn->sb && blocknr == cn->blocknr) { 37791da177e4SLinus Torvalds set_bit(BLOCK_FREED, &cn->state); 37801da177e4SLinus Torvalds if (cn->bh) { 3781098297b2SJeff Mahoney /* 3782098297b2SJeff Mahoney * remove_from_transaction will brelse 3783098297b2SJeff Mahoney * the buffer if it was in the current 3784098297b2SJeff Mahoney * trans 37851da177e4SLinus Torvalds */ 3786098297b2SJeff Mahoney if (!cleaned) { 3787bd4c625cSLinus Torvalds clear_buffer_journal_dirty(cn-> 3788bd4c625cSLinus Torvalds bh); 37891da177e4SLinus Torvalds clear_buffer_dirty(cn->bh); 3790bd4c625cSLinus Torvalds clear_buffer_journal_test(cn-> 3791bd4c625cSLinus Torvalds bh); 37921da177e4SLinus Torvalds cleaned = 1; 37931da177e4SLinus Torvalds put_bh(cn->bh); 3794bd4c625cSLinus Torvalds if (atomic_read 3795a228bf8fSJeff Mahoney (&cn->bh->b_count) < 0) { 3796a9dd3643SJeff Mahoney reiserfs_warning(sb, 379745b03d5eSJeff Mahoney "journal-2138", 379845b03d5eSJeff Mahoney "cn->bh->b_count < 0"); 37991da177e4SLinus Torvalds } 38001da177e4SLinus Torvalds } 3801098297b2SJeff Mahoney /* 3802098297b2SJeff Mahoney * since we are clearing the bh, 3803098297b2SJeff Mahoney * we MUST dec nonzerolen 3804098297b2SJeff Mahoney */ 3805098297b2SJeff Mahoney if (cn->jlist) { 3806a228bf8fSJeff Mahoney atomic_dec(&cn->jlist-> 3807a228bf8fSJeff Mahoney j_nonzerolen); 38081da177e4SLinus Torvalds } 38091da177e4SLinus Torvalds cn->bh = NULL; 38101da177e4SLinus Torvalds } 38111da177e4SLinus Torvalds } 38121da177e4SLinus Torvalds cn = cn->hnext; 38131da177e4SLinus Torvalds } 38141da177e4SLinus Torvalds } 38151da177e4SLinus Torvalds 3816398c95bdSChris Mason if (bh) 3817398c95bdSChris Mason release_buffer_page(bh); /* get_hash grabs the buffer */ 38181da177e4SLinus Torvalds return 0; 38191da177e4SLinus Torvalds } 38201da177e4SLinus Torvalds 3821bd4c625cSLinus Torvalds void reiserfs_update_inode_transaction(struct inode *inode) 3822bd4c625cSLinus Torvalds { 38231da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb); 38241da177e4SLinus Torvalds REISERFS_I(inode)->i_jl = journal->j_current_jl; 38251da177e4SLinus Torvalds REISERFS_I(inode)->i_trans_id = journal->j_trans_id; 38261da177e4SLinus Torvalds } 38271da177e4SLinus Torvalds 38281da177e4SLinus Torvalds /* 38291da177e4SLinus Torvalds * returns -1 on error, 0 if no commits/barriers were done and 1 38301da177e4SLinus Torvalds * if a transaction was actually committed and the barrier was done 38311da177e4SLinus Torvalds */ 38321da177e4SLinus Torvalds static int __commit_trans_jl(struct inode *inode, unsigned long id, 38331da177e4SLinus Torvalds struct reiserfs_journal_list *jl) 38341da177e4SLinus Torvalds { 38351da177e4SLinus Torvalds struct reiserfs_transaction_handle th; 38361da177e4SLinus Torvalds struct super_block *sb = inode->i_sb; 38371da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(sb); 38381da177e4SLinus Torvalds int ret = 0; 38391da177e4SLinus Torvalds 3840098297b2SJeff Mahoney /* 3841098297b2SJeff Mahoney * is it from the current transaction, 3842098297b2SJeff Mahoney * or from an unknown transaction? 3843098297b2SJeff Mahoney */ 38441da177e4SLinus Torvalds if (id == journal->j_trans_id) { 38451da177e4SLinus Torvalds jl = journal->j_current_jl; 3846098297b2SJeff Mahoney /* 3847098297b2SJeff Mahoney * try to let other writers come in and 3848098297b2SJeff Mahoney * grow this transaction 3849098297b2SJeff Mahoney */ 38501da177e4SLinus Torvalds let_transaction_grow(sb, id); 38511da177e4SLinus Torvalds if (journal->j_trans_id != id) { 38521da177e4SLinus Torvalds goto flush_commit_only; 38531da177e4SLinus Torvalds } 38541da177e4SLinus Torvalds 38551da177e4SLinus Torvalds ret = journal_begin(&th, sb, 1); 38561da177e4SLinus Torvalds if (ret) 38571da177e4SLinus Torvalds return ret; 38581da177e4SLinus Torvalds 38591da177e4SLinus Torvalds /* someone might have ended this transaction while we joined */ 38601da177e4SLinus Torvalds if (journal->j_trans_id != id) { 3861bd4c625cSLinus Torvalds reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb), 3862bd4c625cSLinus Torvalds 1); 386309f1b80bSJeff Mahoney journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb)); 386458d85426SJeff Mahoney ret = journal_end(&th); 38651da177e4SLinus Torvalds goto flush_commit_only; 38661da177e4SLinus Torvalds } 38671da177e4SLinus Torvalds 386858d85426SJeff Mahoney ret = journal_end_sync(&th); 38691da177e4SLinus Torvalds if (!ret) 38701da177e4SLinus Torvalds ret = 1; 38711da177e4SLinus Torvalds 38721da177e4SLinus Torvalds } else { 3873098297b2SJeff Mahoney /* 3874098297b2SJeff Mahoney * this gets tricky, we have to make sure the journal list in 38751da177e4SLinus Torvalds * the inode still exists. We know the list is still around 38761da177e4SLinus Torvalds * if we've got a larger transaction id than the oldest list 38771da177e4SLinus Torvalds */ 38781da177e4SLinus Torvalds flush_commit_only: 38791da177e4SLinus Torvalds if (journal_list_still_alive(inode->i_sb, id)) { 38801da177e4SLinus Torvalds /* 38811da177e4SLinus Torvalds * we only set ret to 1 when we know for sure 38821da177e4SLinus Torvalds * the barrier hasn't been started yet on the commit 38831da177e4SLinus Torvalds * block. 38841da177e4SLinus Torvalds */ 38851da177e4SLinus Torvalds if (atomic_read(&jl->j_commit_left) > 1) 38861da177e4SLinus Torvalds ret = 1; 38871da177e4SLinus Torvalds flush_commit_list(sb, jl, 1); 38881da177e4SLinus Torvalds if (journal->j_errno) 38891da177e4SLinus Torvalds ret = journal->j_errno; 38901da177e4SLinus Torvalds } 38911da177e4SLinus Torvalds } 38921da177e4SLinus Torvalds /* otherwise the list is gone, and long since committed */ 38931da177e4SLinus Torvalds return ret; 38941da177e4SLinus Torvalds } 38951da177e4SLinus Torvalds 3896bd4c625cSLinus Torvalds int reiserfs_commit_for_inode(struct inode *inode) 3897bd4c625cSLinus Torvalds { 3898600ed416SJeff Mahoney unsigned int id = REISERFS_I(inode)->i_trans_id; 38991da177e4SLinus Torvalds struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl; 39001da177e4SLinus Torvalds 3901098297b2SJeff Mahoney /* 3902098297b2SJeff Mahoney * for the whole inode, assume unset id means it was 39031da177e4SLinus Torvalds * changed in the current transaction. More conservative 39041da177e4SLinus Torvalds */ 39051da177e4SLinus Torvalds if (!id || !jl) { 39061da177e4SLinus Torvalds reiserfs_update_inode_transaction(inode); 39071da177e4SLinus Torvalds id = REISERFS_I(inode)->i_trans_id; 39081da177e4SLinus Torvalds /* jl will be updated in __commit_trans_jl */ 39091da177e4SLinus Torvalds } 39101da177e4SLinus Torvalds 39111da177e4SLinus Torvalds return __commit_trans_jl(inode, id, jl); 39121da177e4SLinus Torvalds } 39131da177e4SLinus Torvalds 3914a9dd3643SJeff Mahoney void reiserfs_restore_prepared_buffer(struct super_block *sb, 3915bd4c625cSLinus Torvalds struct buffer_head *bh) 3916bd4c625cSLinus Torvalds { 3917a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 3918a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.restore_prepared); 39191da177e4SLinus Torvalds if (!bh) { 39201da177e4SLinus Torvalds return; 39211da177e4SLinus Torvalds } 39221da177e4SLinus Torvalds if (test_clear_buffer_journal_restore_dirty(bh) && 39231da177e4SLinus Torvalds buffer_journal_dirty(bh)) { 39241da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 3925278f6679SJeff Mahoney reiserfs_write_lock(sb); 3926a9dd3643SJeff Mahoney cn = get_journal_hash_dev(sb, 39271da177e4SLinus Torvalds journal->j_list_hash_table, 39281da177e4SLinus Torvalds bh->b_blocknr); 39291da177e4SLinus Torvalds if (cn && can_dirty(cn)) { 39301da177e4SLinus Torvalds set_buffer_journal_test(bh); 39311da177e4SLinus Torvalds mark_buffer_dirty(bh); 39321da177e4SLinus Torvalds } 3933278f6679SJeff Mahoney reiserfs_write_unlock(sb); 39341da177e4SLinus Torvalds } 39351da177e4SLinus Torvalds clear_buffer_journal_prepared(bh); 39361da177e4SLinus Torvalds } 39371da177e4SLinus Torvalds 39381da177e4SLinus Torvalds extern struct tree_balance *cur_tb; 39391da177e4SLinus Torvalds /* 3940098297b2SJeff Mahoney * before we can change a metadata block, we have to make sure it won't 3941098297b2SJeff Mahoney * be written to disk while we are altering it. So, we must: 3942098297b2SJeff Mahoney * clean it 3943098297b2SJeff Mahoney * wait on it. 39441da177e4SLinus Torvalds */ 3945a9dd3643SJeff Mahoney int reiserfs_prepare_for_journal(struct super_block *sb, 3946bd4c625cSLinus Torvalds struct buffer_head *bh, int wait) 3947bd4c625cSLinus Torvalds { 3948a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.prepare); 39491da177e4SLinus Torvalds 3950ca5de404SNick Piggin if (!trylock_buffer(bh)) { 39511da177e4SLinus Torvalds if (!wait) 39521da177e4SLinus Torvalds return 0; 39531da177e4SLinus Torvalds lock_buffer(bh); 39541da177e4SLinus Torvalds } 39551da177e4SLinus Torvalds set_buffer_journal_prepared(bh); 39561da177e4SLinus Torvalds if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) { 39571da177e4SLinus Torvalds clear_buffer_journal_test(bh); 39581da177e4SLinus Torvalds set_buffer_journal_restore_dirty(bh); 39591da177e4SLinus Torvalds } 39601da177e4SLinus Torvalds unlock_buffer(bh); 39611da177e4SLinus Torvalds return 1; 39621da177e4SLinus Torvalds } 39631da177e4SLinus Torvalds 39641da177e4SLinus Torvalds /* 3965098297b2SJeff Mahoney * long and ugly. If flush, will not return until all commit 3966098297b2SJeff Mahoney * blocks and all real buffers in the trans are on disk. 3967098297b2SJeff Mahoney * If no_async, won't return until all commit blocks are on disk. 3968098297b2SJeff Mahoney * 3969098297b2SJeff Mahoney * keep reading, there are comments as you go along 3970098297b2SJeff Mahoney * 3971098297b2SJeff Mahoney * If the journal is aborted, we just clean up. Things like flushing 3972098297b2SJeff Mahoney * journal lists, etc just won't happen. 39731da177e4SLinus Torvalds */ 397458d85426SJeff Mahoney static int do_journal_end(struct reiserfs_transaction_handle *th, int flags) 3975bd4c625cSLinus Torvalds { 397658d85426SJeff Mahoney struct super_block *sb = th->t_super; 3977a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 39781da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn, *next, *jl_cn; 39791da177e4SLinus Torvalds struct reiserfs_journal_cnode *last_cn = NULL; 39801da177e4SLinus Torvalds struct reiserfs_journal_desc *desc; 39811da177e4SLinus Torvalds struct reiserfs_journal_commit *commit; 39821da177e4SLinus Torvalds struct buffer_head *c_bh; /* commit bh */ 39831da177e4SLinus Torvalds struct buffer_head *d_bh; /* desc bh */ 39841da177e4SLinus Torvalds int cur_write_start = 0; /* start index of current log write */ 39851da177e4SLinus Torvalds int old_start; 39861da177e4SLinus Torvalds int i; 3987a44c94a7SAlexander Zarochentsev int flush; 3988a44c94a7SAlexander Zarochentsev int wait_on_commit; 39891da177e4SLinus Torvalds struct reiserfs_journal_list *jl, *temp_jl; 39901da177e4SLinus Torvalds struct list_head *entry, *safe; 39911da177e4SLinus Torvalds unsigned long jindex; 3992600ed416SJeff Mahoney unsigned int commit_trans_id; 39931da177e4SLinus Torvalds int trans_half; 3994278f6679SJeff Mahoney int depth; 39951da177e4SLinus Torvalds 39961da177e4SLinus Torvalds BUG_ON(th->t_refcount > 1); 39971da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 399858d85426SJeff Mahoney BUG_ON(!th->t_super); 39991da177e4SLinus Torvalds 4000098297b2SJeff Mahoney /* 4001098297b2SJeff Mahoney * protect flush_older_commits from doing mistakes if the 4002098297b2SJeff Mahoney * transaction ID counter gets overflowed. 4003098297b2SJeff Mahoney */ 4004600ed416SJeff Mahoney if (th->t_trans_id == ~0U) 4005a44c94a7SAlexander Zarochentsev flags |= FLUSH_ALL | COMMIT_NOW | WAIT; 4006a44c94a7SAlexander Zarochentsev flush = flags & FLUSH_ALL; 4007a44c94a7SAlexander Zarochentsev wait_on_commit = flags & WAIT; 4008a44c94a7SAlexander Zarochentsev 40091da177e4SLinus Torvalds current->journal_info = th->t_handle_save; 4010a9dd3643SJeff Mahoney reiserfs_check_lock_depth(sb, "journal end"); 40111da177e4SLinus Torvalds if (journal->j_len == 0) { 4012a9dd3643SJeff Mahoney reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb), 4013bd4c625cSLinus Torvalds 1); 401409f1b80bSJeff Mahoney journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb)); 40151da177e4SLinus Torvalds } 40161da177e4SLinus Torvalds 4017a9dd3643SJeff Mahoney lock_journal(sb); 40181da177e4SLinus Torvalds if (journal->j_next_full_flush) { 40191da177e4SLinus Torvalds flags |= FLUSH_ALL; 40201da177e4SLinus Torvalds flush = 1; 40211da177e4SLinus Torvalds } 40221da177e4SLinus Torvalds if (journal->j_next_async_flush) { 40231da177e4SLinus Torvalds flags |= COMMIT_NOW | WAIT; 40241da177e4SLinus Torvalds wait_on_commit = 1; 40251da177e4SLinus Torvalds } 40261da177e4SLinus Torvalds 4027098297b2SJeff Mahoney /* 4028098297b2SJeff Mahoney * check_journal_end locks the journal, and unlocks if it does 4029098297b2SJeff Mahoney * not return 1 it tells us if we should continue with the 4030098297b2SJeff Mahoney * journal_end, or just return 40311da177e4SLinus Torvalds */ 403258d85426SJeff Mahoney if (!check_journal_end(th, flags)) { 4033033369d1SArtem Bityutskiy reiserfs_schedule_old_flush(sb); 4034a9dd3643SJeff Mahoney wake_queued_writers(sb); 4035a9dd3643SJeff Mahoney reiserfs_async_progress_wait(sb); 40361da177e4SLinus Torvalds goto out; 40371da177e4SLinus Torvalds } 40381da177e4SLinus Torvalds 40391da177e4SLinus Torvalds /* check_journal_end might set these, check again */ 40401da177e4SLinus Torvalds if (journal->j_next_full_flush) { 40411da177e4SLinus Torvalds flush = 1; 40421da177e4SLinus Torvalds } 40431da177e4SLinus Torvalds 40441da177e4SLinus Torvalds /* 4045098297b2SJeff Mahoney * j must wait means we have to flush the log blocks, and the 4046098297b2SJeff Mahoney * real blocks for this transaction 40471da177e4SLinus Torvalds */ 40481da177e4SLinus Torvalds if (journal->j_must_wait > 0) { 40491da177e4SLinus Torvalds flush = 1; 40501da177e4SLinus Torvalds } 40511da177e4SLinus Torvalds #ifdef REISERFS_PREALLOCATE 4052098297b2SJeff Mahoney /* 4053098297b2SJeff Mahoney * quota ops might need to nest, setup the journal_info pointer 4054098297b2SJeff Mahoney * for them and raise the refcount so that it is > 0. 4055098297b2SJeff Mahoney */ 40561da177e4SLinus Torvalds current->journal_info = th; 4057ef43bc4fSJan Kara th->t_refcount++; 4058098297b2SJeff Mahoney 4059098297b2SJeff Mahoney /* it should not involve new blocks into the transaction */ 4060098297b2SJeff Mahoney reiserfs_discard_all_prealloc(th); 4061098297b2SJeff Mahoney 4062ef43bc4fSJan Kara th->t_refcount--; 40631da177e4SLinus Torvalds current->journal_info = th->t_handle_save; 40641da177e4SLinus Torvalds #endif 40651da177e4SLinus Torvalds 40661da177e4SLinus Torvalds /* setup description block */ 4067bd4c625cSLinus Torvalds d_bh = 4068a9dd3643SJeff Mahoney journal_getblk(sb, 4069a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 4070bd4c625cSLinus Torvalds journal->j_start); 40711da177e4SLinus Torvalds set_buffer_uptodate(d_bh); 40721da177e4SLinus Torvalds desc = (struct reiserfs_journal_desc *)(d_bh)->b_data; 40731da177e4SLinus Torvalds memset(d_bh->b_data, 0, d_bh->b_size); 40741da177e4SLinus Torvalds memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8); 40751da177e4SLinus Torvalds set_desc_trans_id(desc, journal->j_trans_id); 40761da177e4SLinus Torvalds 4077098297b2SJeff Mahoney /* 4078098297b2SJeff Mahoney * setup commit block. Don't write (keep it clean too) this one 4079098297b2SJeff Mahoney * until after everyone else is written 4080098297b2SJeff Mahoney */ 4081a9dd3643SJeff Mahoney c_bh = journal_getblk(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 4082bd4c625cSLinus Torvalds ((journal->j_start + journal->j_len + 4083a9dd3643SJeff Mahoney 1) % SB_ONDISK_JOURNAL_SIZE(sb))); 40841da177e4SLinus Torvalds commit = (struct reiserfs_journal_commit *)c_bh->b_data; 40851da177e4SLinus Torvalds memset(c_bh->b_data, 0, c_bh->b_size); 40861da177e4SLinus Torvalds set_commit_trans_id(commit, journal->j_trans_id); 40871da177e4SLinus Torvalds set_buffer_uptodate(c_bh); 40881da177e4SLinus Torvalds 40891da177e4SLinus Torvalds /* init this journal list */ 40901da177e4SLinus Torvalds jl = journal->j_current_jl; 40911da177e4SLinus Torvalds 4092098297b2SJeff Mahoney /* 4093098297b2SJeff Mahoney * we lock the commit before doing anything because 40941da177e4SLinus Torvalds * we want to make sure nobody tries to run flush_commit_list until 40951da177e4SLinus Torvalds * the new transaction is fully setup, and we've already flushed the 40961da177e4SLinus Torvalds * ordered bh list 40971da177e4SLinus Torvalds */ 40988ebc4232SFrederic Weisbecker reiserfs_mutex_lock_safe(&jl->j_commit_mutex, sb); 40991da177e4SLinus Torvalds 41001da177e4SLinus Torvalds /* save the transaction id in case we need to commit it later */ 41011da177e4SLinus Torvalds commit_trans_id = jl->j_trans_id; 41021da177e4SLinus Torvalds 41031da177e4SLinus Torvalds atomic_set(&jl->j_older_commits_done, 0); 41041da177e4SLinus Torvalds jl->j_trans_id = journal->j_trans_id; 41051da177e4SLinus Torvalds jl->j_timestamp = journal->j_trans_start_time; 41061da177e4SLinus Torvalds jl->j_commit_bh = c_bh; 41071da177e4SLinus Torvalds jl->j_start = journal->j_start; 41081da177e4SLinus Torvalds jl->j_len = journal->j_len; 41091da177e4SLinus Torvalds atomic_set(&jl->j_nonzerolen, journal->j_len); 41101da177e4SLinus Torvalds atomic_set(&jl->j_commit_left, journal->j_len + 2); 41111da177e4SLinus Torvalds jl->j_realblock = NULL; 41121da177e4SLinus Torvalds 4113098297b2SJeff Mahoney /* 4114098297b2SJeff Mahoney * The ENTIRE FOR LOOP MUST not cause schedule to occur. 4115098297b2SJeff Mahoney * for each real block, add it to the journal list hash, 4116098297b2SJeff Mahoney * copy into real block index array in the commit or desc block 41171da177e4SLinus Torvalds */ 4118a9dd3643SJeff Mahoney trans_half = journal_trans_half(sb->s_blocksize); 41191da177e4SLinus Torvalds for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) { 41201da177e4SLinus Torvalds if (buffer_journaled(cn->bh)) { 4121a9dd3643SJeff Mahoney jl_cn = get_cnode(sb); 41221da177e4SLinus Torvalds if (!jl_cn) { 4123a9dd3643SJeff Mahoney reiserfs_panic(sb, "journal-1676", 4124c3a9c210SJeff Mahoney "get_cnode returned NULL"); 41251da177e4SLinus Torvalds } 41261da177e4SLinus Torvalds if (i == 0) { 41271da177e4SLinus Torvalds jl->j_realblock = jl_cn; 41281da177e4SLinus Torvalds } 41291da177e4SLinus Torvalds jl_cn->prev = last_cn; 41301da177e4SLinus Torvalds jl_cn->next = NULL; 41311da177e4SLinus Torvalds if (last_cn) { 41321da177e4SLinus Torvalds last_cn->next = jl_cn; 41331da177e4SLinus Torvalds } 41341da177e4SLinus Torvalds last_cn = jl_cn; 4135098297b2SJeff Mahoney /* 4136098297b2SJeff Mahoney * make sure the block we are trying to log 4137098297b2SJeff Mahoney * is not a block of journal or reserved area 4138098297b2SJeff Mahoney */ 4139bd4c625cSLinus Torvalds if (is_block_in_log_or_reserved_area 4140a9dd3643SJeff Mahoney (sb, cn->bh->b_blocknr)) { 4141a9dd3643SJeff Mahoney reiserfs_panic(sb, "journal-2332", 4142c3a9c210SJeff Mahoney "Trying to log block %lu, " 4143c3a9c210SJeff Mahoney "which is a log block", 4144bd4c625cSLinus Torvalds cn->bh->b_blocknr); 41451da177e4SLinus Torvalds } 41461da177e4SLinus Torvalds jl_cn->blocknr = cn->bh->b_blocknr; 41471da177e4SLinus Torvalds jl_cn->state = 0; 4148a9dd3643SJeff Mahoney jl_cn->sb = sb; 41491da177e4SLinus Torvalds jl_cn->bh = cn->bh; 41501da177e4SLinus Torvalds jl_cn->jlist = jl; 41511da177e4SLinus Torvalds insert_journal_hash(journal->j_list_hash_table, jl_cn); 41521da177e4SLinus Torvalds if (i < trans_half) { 4153bd4c625cSLinus Torvalds desc->j_realblock[i] = 4154bd4c625cSLinus Torvalds cpu_to_le32(cn->bh->b_blocknr); 41551da177e4SLinus Torvalds } else { 4156bd4c625cSLinus Torvalds commit->j_realblock[i - trans_half] = 4157bd4c625cSLinus Torvalds cpu_to_le32(cn->bh->b_blocknr); 41581da177e4SLinus Torvalds } 41591da177e4SLinus Torvalds } else { 41601da177e4SLinus Torvalds i--; 41611da177e4SLinus Torvalds } 41621da177e4SLinus Torvalds } 41631da177e4SLinus Torvalds set_desc_trans_len(desc, journal->j_len); 41641da177e4SLinus Torvalds set_desc_mount_id(desc, journal->j_mount_id); 41651da177e4SLinus Torvalds set_desc_trans_id(desc, journal->j_trans_id); 41661da177e4SLinus Torvalds set_commit_trans_len(commit, journal->j_len); 41671da177e4SLinus Torvalds 4168098297b2SJeff Mahoney /* 4169098297b2SJeff Mahoney * special check in case all buffers in the journal 4170098297b2SJeff Mahoney * were marked for not logging 4171098297b2SJeff Mahoney */ 417214a61442SEric Sesterhenn BUG_ON(journal->j_len == 0); 41731da177e4SLinus Torvalds 4174098297b2SJeff Mahoney /* 4175098297b2SJeff Mahoney * we're about to dirty all the log blocks, mark the description block 41761da177e4SLinus Torvalds * dirty now too. Don't mark the commit block dirty until all the 41771da177e4SLinus Torvalds * others are on disk 41781da177e4SLinus Torvalds */ 41791da177e4SLinus Torvalds mark_buffer_dirty(d_bh); 41801da177e4SLinus Torvalds 4181098297b2SJeff Mahoney /* 4182098297b2SJeff Mahoney * first data block is j_start + 1, so add one to 4183098297b2SJeff Mahoney * cur_write_start wherever you use it 4184098297b2SJeff Mahoney */ 41851da177e4SLinus Torvalds cur_write_start = journal->j_start; 41861da177e4SLinus Torvalds cn = journal->j_first; 41871da177e4SLinus Torvalds jindex = 1; /* start at one so we don't get the desc again */ 41881da177e4SLinus Torvalds while (cn) { 41891da177e4SLinus Torvalds clear_buffer_journal_new(cn->bh); 41901da177e4SLinus Torvalds /* copy all the real blocks into log area. dirty log blocks */ 41911da177e4SLinus Torvalds if (buffer_journaled(cn->bh)) { 41921da177e4SLinus Torvalds struct buffer_head *tmp_bh; 41931da177e4SLinus Torvalds char *addr; 41941da177e4SLinus Torvalds struct page *page; 4195bd4c625cSLinus Torvalds tmp_bh = 4196a9dd3643SJeff Mahoney journal_getblk(sb, 4197a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 4198bd4c625cSLinus Torvalds ((cur_write_start + 4199bd4c625cSLinus Torvalds jindex) % 4200a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb))); 42011da177e4SLinus Torvalds set_buffer_uptodate(tmp_bh); 42021da177e4SLinus Torvalds page = cn->bh->b_page; 42031da177e4SLinus Torvalds addr = kmap(page); 4204bd4c625cSLinus Torvalds memcpy(tmp_bh->b_data, 4205bd4c625cSLinus Torvalds addr + offset_in_page(cn->bh->b_data), 42061da177e4SLinus Torvalds cn->bh->b_size); 42071da177e4SLinus Torvalds kunmap(page); 42081da177e4SLinus Torvalds mark_buffer_dirty(tmp_bh); 42091da177e4SLinus Torvalds jindex++; 42101da177e4SLinus Torvalds set_buffer_journal_dirty(cn->bh); 42111da177e4SLinus Torvalds clear_buffer_journaled(cn->bh); 42121da177e4SLinus Torvalds } else { 4213098297b2SJeff Mahoney /* 4214098297b2SJeff Mahoney * JDirty cleared sometime during transaction. 4215098297b2SJeff Mahoney * don't log this one 4216098297b2SJeff Mahoney */ 4217a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-2048", 421845b03d5eSJeff Mahoney "BAD, buffer in journal hash, " 421945b03d5eSJeff Mahoney "but not JDirty!"); 42201da177e4SLinus Torvalds brelse(cn->bh); 42211da177e4SLinus Torvalds } 42221da177e4SLinus Torvalds next = cn->next; 4223a9dd3643SJeff Mahoney free_cnode(sb, cn); 42241da177e4SLinus Torvalds cn = next; 4225278f6679SJeff Mahoney reiserfs_cond_resched(sb); 42261da177e4SLinus Torvalds } 42271da177e4SLinus Torvalds 4228098297b2SJeff Mahoney /* 4229098297b2SJeff Mahoney * we are done with both the c_bh and d_bh, but 4230098297b2SJeff Mahoney * c_bh must be written after all other commit blocks, 4231098297b2SJeff Mahoney * so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1. 42321da177e4SLinus Torvalds */ 42331da177e4SLinus Torvalds 4234a9dd3643SJeff Mahoney journal->j_current_jl = alloc_journal_list(sb); 42351da177e4SLinus Torvalds 42361da177e4SLinus Torvalds /* now it is safe to insert this transaction on the main list */ 42371da177e4SLinus Torvalds list_add_tail(&jl->j_list, &journal->j_journal_list); 42381da177e4SLinus Torvalds list_add_tail(&jl->j_working_list, &journal->j_working_list); 42391da177e4SLinus Torvalds journal->j_num_work_lists++; 42401da177e4SLinus Torvalds 42411da177e4SLinus Torvalds /* reset journal values for the next transaction */ 42421da177e4SLinus Torvalds old_start = journal->j_start; 4243bd4c625cSLinus Torvalds journal->j_start = 4244bd4c625cSLinus Torvalds (journal->j_start + journal->j_len + 4245a9dd3643SJeff Mahoney 2) % SB_ONDISK_JOURNAL_SIZE(sb); 4246a228bf8fSJeff Mahoney atomic_set(&journal->j_wcount, 0); 42471da177e4SLinus Torvalds journal->j_bcount = 0; 42481da177e4SLinus Torvalds journal->j_last = NULL; 42491da177e4SLinus Torvalds journal->j_first = NULL; 42501da177e4SLinus Torvalds journal->j_len = 0; 42511da177e4SLinus Torvalds journal->j_trans_start_time = 0; 4252a44c94a7SAlexander Zarochentsev /* check for trans_id overflow */ 4253a44c94a7SAlexander Zarochentsev if (++journal->j_trans_id == 0) 4254a44c94a7SAlexander Zarochentsev journal->j_trans_id = 10; 42551da177e4SLinus Torvalds journal->j_current_jl->j_trans_id = journal->j_trans_id; 42561da177e4SLinus Torvalds journal->j_must_wait = 0; 42571da177e4SLinus Torvalds journal->j_len_alloc = 0; 42581da177e4SLinus Torvalds journal->j_next_full_flush = 0; 42591da177e4SLinus Torvalds journal->j_next_async_flush = 0; 4260a9dd3643SJeff Mahoney init_journal_hash(sb); 42611da177e4SLinus Torvalds 4262098297b2SJeff Mahoney /* 4263098297b2SJeff Mahoney * make sure reiserfs_add_jh sees the new current_jl before we 4264098297b2SJeff Mahoney * write out the tails 4265098297b2SJeff Mahoney */ 42661da177e4SLinus Torvalds smp_mb(); 42671da177e4SLinus Torvalds 4268098297b2SJeff Mahoney /* 4269098297b2SJeff Mahoney * tail conversion targets have to hit the disk before we end the 42701da177e4SLinus Torvalds * transaction. Otherwise a later transaction might repack the tail 4271098297b2SJeff Mahoney * before this transaction commits, leaving the data block unflushed 4272098297b2SJeff Mahoney * and clean, if we crash before the later transaction commits, the 4273098297b2SJeff Mahoney * data block is lost. 42741da177e4SLinus Torvalds */ 42751da177e4SLinus Torvalds if (!list_empty(&jl->j_tail_bh_list)) { 4276278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(sb); 42771da177e4SLinus Torvalds write_ordered_buffers(&journal->j_dirty_buffers_lock, 42781da177e4SLinus Torvalds journal, jl, &jl->j_tail_bh_list); 4279278f6679SJeff Mahoney reiserfs_write_lock_nested(sb, depth); 42801da177e4SLinus Torvalds } 428114a61442SEric Sesterhenn BUG_ON(!list_empty(&jl->j_tail_bh_list)); 428290415deaSJeff Mahoney mutex_unlock(&jl->j_commit_mutex); 42831da177e4SLinus Torvalds 4284098297b2SJeff Mahoney /* 4285098297b2SJeff Mahoney * honor the flush wishes from the caller, simple commits can 4286098297b2SJeff Mahoney * be done outside the journal lock, they are done below 4287098297b2SJeff Mahoney * 4288098297b2SJeff Mahoney * if we don't flush the commit list right now, we put it into 4289098297b2SJeff Mahoney * the work queue so the people waiting on the async progress work 4290098297b2SJeff Mahoney * queue don't wait for this proc to flush journal lists and such. 42911da177e4SLinus Torvalds */ 42921da177e4SLinus Torvalds if (flush) { 4293a9dd3643SJeff Mahoney flush_commit_list(sb, jl, 1); 4294a9dd3643SJeff Mahoney flush_journal_list(sb, jl, 1); 42951da177e4SLinus Torvalds } else if (!(jl->j_state & LIST_COMMIT_PENDING)) 4296797d9016SJeff Mahoney queue_delayed_work(REISERFS_SB(sb)->commit_wq, 4297797d9016SJeff Mahoney &journal->j_work, HZ / 10); 42981da177e4SLinus Torvalds 4299098297b2SJeff Mahoney /* 4300098297b2SJeff Mahoney * if the next transaction has any chance of wrapping, flush 4301098297b2SJeff Mahoney * transactions that might get overwritten. If any journal lists 4302098297b2SJeff Mahoney * are very old flush them as well. 43031da177e4SLinus Torvalds */ 43041da177e4SLinus Torvalds first_jl: 43051da177e4SLinus Torvalds list_for_each_safe(entry, safe, &journal->j_journal_list) { 43061da177e4SLinus Torvalds temp_jl = JOURNAL_LIST_ENTRY(entry); 43071da177e4SLinus Torvalds if (journal->j_start <= temp_jl->j_start) { 43081da177e4SLinus Torvalds if ((journal->j_start + journal->j_trans_max + 1) >= 4309bd4c625cSLinus Torvalds temp_jl->j_start) { 4310a9dd3643SJeff Mahoney flush_used_journal_lists(sb, temp_jl); 43111da177e4SLinus Torvalds goto first_jl; 43121da177e4SLinus Torvalds } else if ((journal->j_start + 43131da177e4SLinus Torvalds journal->j_trans_max + 1) < 4314a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb)) { 4315098297b2SJeff Mahoney /* 4316098297b2SJeff Mahoney * if we don't cross into the next 4317098297b2SJeff Mahoney * transaction and we don't wrap, there is 4318098297b2SJeff Mahoney * no way we can overlap any later transactions 43191da177e4SLinus Torvalds * break now 43201da177e4SLinus Torvalds */ 43211da177e4SLinus Torvalds break; 43221da177e4SLinus Torvalds } 43231da177e4SLinus Torvalds } else if ((journal->j_start + 43241da177e4SLinus Torvalds journal->j_trans_max + 1) > 4325a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb)) { 43261da177e4SLinus Torvalds if (((journal->j_start + journal->j_trans_max + 1) % 4327a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb)) >= 4328bd4c625cSLinus Torvalds temp_jl->j_start) { 4329a9dd3643SJeff Mahoney flush_used_journal_lists(sb, temp_jl); 43301da177e4SLinus Torvalds goto first_jl; 43311da177e4SLinus Torvalds } else { 4332098297b2SJeff Mahoney /* 4333098297b2SJeff Mahoney * we don't overlap anything from out start 4334098297b2SJeff Mahoney * to the end of the log, and our wrapped 4335098297b2SJeff Mahoney * portion doesn't overlap anything at 43361da177e4SLinus Torvalds * the start of the log. We can break 43371da177e4SLinus Torvalds */ 43381da177e4SLinus Torvalds break; 43391da177e4SLinus Torvalds } 43401da177e4SLinus Torvalds } 43411da177e4SLinus Torvalds } 43421da177e4SLinus Torvalds 4343bd4c625cSLinus Torvalds journal->j_current_jl->j_list_bitmap = 4344a9dd3643SJeff Mahoney get_list_bitmap(sb, journal->j_current_jl); 43451da177e4SLinus Torvalds 43461da177e4SLinus Torvalds if (!(journal->j_current_jl->j_list_bitmap)) { 4347a9dd3643SJeff Mahoney reiserfs_panic(sb, "journal-1996", 4348c3a9c210SJeff Mahoney "could not get a list bitmap"); 43491da177e4SLinus Torvalds } 43501da177e4SLinus Torvalds 4351a228bf8fSJeff Mahoney atomic_set(&journal->j_jlock, 0); 4352a9dd3643SJeff Mahoney unlock_journal(sb); 43531da177e4SLinus Torvalds /* wake up any body waiting to join. */ 43541da177e4SLinus Torvalds clear_bit(J_WRITERS_QUEUED, &journal->j_state); 4355a228bf8fSJeff Mahoney wake_up(&journal->j_join_wait); 43561da177e4SLinus Torvalds 43571da177e4SLinus Torvalds if (!flush && wait_on_commit && 4358a9dd3643SJeff Mahoney journal_list_still_alive(sb, commit_trans_id)) { 4359a9dd3643SJeff Mahoney flush_commit_list(sb, jl, 1); 43601da177e4SLinus Torvalds } 43611da177e4SLinus Torvalds out: 4362a9dd3643SJeff Mahoney reiserfs_check_lock_depth(sb, "journal end2"); 43631da177e4SLinus Torvalds 43641da177e4SLinus Torvalds memset(th, 0, sizeof(*th)); 4365098297b2SJeff Mahoney /* 4366098297b2SJeff Mahoney * Re-set th->t_super, so we can properly keep track of how many 43671da177e4SLinus Torvalds * persistent transactions there are. We need to do this so if this 4368098297b2SJeff Mahoney * call is part of a failed restart_transaction, we can free it later 4369098297b2SJeff Mahoney */ 4370a9dd3643SJeff Mahoney th->t_super = sb; 43711da177e4SLinus Torvalds 43721da177e4SLinus Torvalds return journal->j_errno; 43731da177e4SLinus Torvalds } 43741da177e4SLinus Torvalds 437532e8b106SJeff Mahoney /* Send the file system read only and refuse new transactions */ 437632e8b106SJeff Mahoney void reiserfs_abort_journal(struct super_block *sb, int errno) 43771da177e4SLinus Torvalds { 43781da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(sb); 43791da177e4SLinus Torvalds if (test_bit(J_ABORTED, &journal->j_state)) 43801da177e4SLinus Torvalds return; 43811da177e4SLinus Torvalds 438232e8b106SJeff Mahoney if (!journal->j_errno) 438332e8b106SJeff Mahoney journal->j_errno = errno; 43841da177e4SLinus Torvalds 43851da177e4SLinus Torvalds sb->s_flags |= MS_RDONLY; 43861da177e4SLinus Torvalds set_bit(J_ABORTED, &journal->j_state); 43871da177e4SLinus Torvalds 43881da177e4SLinus Torvalds #ifdef CONFIG_REISERFS_CHECK 43891da177e4SLinus Torvalds dump_stack(); 43901da177e4SLinus Torvalds #endif 43911da177e4SLinus Torvalds } 4392