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 602ea1754a0SKirill A. Shutemov * final put_page 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)) { 60809cbfeafSKirill A. Shutemov get_page(page); 609398c95bdSChris Mason put_bh(bh); 610398c95bdSChris Mason if (!page->mapping) 611398c95bdSChris Mason try_to_free_buffers(page); 612398c95bdSChris Mason unlock_page(page); 61309cbfeafSKirill A. Shutemov put_page(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 if (buffer_journaled(bh)) { 62245b03d5eSJeff Mahoney reiserfs_warning(NULL, "clm-2084", 623a1c6f057SDmitry Monakhov "pinned buffer %lu:%pg sent to disk", 624a1c6f057SDmitry Monakhov bh->b_blocknr, bh->b_bdev); 6251da177e4SLinus Torvalds } 6261da177e4SLinus Torvalds if (uptodate) 6271da177e4SLinus Torvalds set_buffer_uptodate(bh); 6281da177e4SLinus Torvalds else 6291da177e4SLinus Torvalds clear_buffer_uptodate(bh); 630398c95bdSChris Mason 6311da177e4SLinus Torvalds unlock_buffer(bh); 632398c95bdSChris Mason release_buffer_page(bh); 6331da177e4SLinus Torvalds } 6341da177e4SLinus Torvalds 635bd4c625cSLinus Torvalds static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate) 636bd4c625cSLinus Torvalds { 6371da177e4SLinus Torvalds if (uptodate) 6381da177e4SLinus Torvalds set_buffer_uptodate(bh); 6391da177e4SLinus Torvalds else 6401da177e4SLinus Torvalds clear_buffer_uptodate(bh); 6411da177e4SLinus Torvalds unlock_buffer(bh); 6421da177e4SLinus Torvalds put_bh(bh); 6431da177e4SLinus Torvalds } 6441da177e4SLinus Torvalds 645bd4c625cSLinus Torvalds static void submit_logged_buffer(struct buffer_head *bh) 646bd4c625cSLinus Torvalds { 6471da177e4SLinus Torvalds get_bh(bh); 6481da177e4SLinus Torvalds bh->b_end_io = reiserfs_end_buffer_io_sync; 6491da177e4SLinus Torvalds clear_buffer_journal_new(bh); 6501da177e4SLinus Torvalds clear_buffer_dirty(bh); 6511da177e4SLinus Torvalds if (!test_clear_buffer_journal_test(bh)) 6521da177e4SLinus Torvalds BUG(); 6531da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 6541da177e4SLinus Torvalds BUG(); 6552a222ca9SMike Christie submit_bh(REQ_OP_WRITE, 0, bh); 6561da177e4SLinus Torvalds } 6571da177e4SLinus Torvalds 658bd4c625cSLinus Torvalds static void submit_ordered_buffer(struct buffer_head *bh) 659bd4c625cSLinus Torvalds { 6601da177e4SLinus Torvalds get_bh(bh); 6611da177e4SLinus Torvalds bh->b_end_io = reiserfs_end_ordered_io; 6621da177e4SLinus Torvalds clear_buffer_dirty(bh); 6631da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 6641da177e4SLinus Torvalds BUG(); 6652a222ca9SMike Christie submit_bh(REQ_OP_WRITE, 0, bh); 6661da177e4SLinus Torvalds } 6671da177e4SLinus Torvalds 6681da177e4SLinus Torvalds #define CHUNK_SIZE 32 6691da177e4SLinus Torvalds struct buffer_chunk { 6701da177e4SLinus Torvalds struct buffer_head *bh[CHUNK_SIZE]; 6711da177e4SLinus Torvalds int nr; 6721da177e4SLinus Torvalds }; 6731da177e4SLinus Torvalds 674bd4c625cSLinus Torvalds static void write_chunk(struct buffer_chunk *chunk) 675bd4c625cSLinus Torvalds { 6761da177e4SLinus Torvalds int i; 6771da177e4SLinus Torvalds for (i = 0; i < chunk->nr; i++) { 6781da177e4SLinus Torvalds submit_logged_buffer(chunk->bh[i]); 6791da177e4SLinus Torvalds } 6801da177e4SLinus Torvalds chunk->nr = 0; 6811da177e4SLinus Torvalds } 6821da177e4SLinus Torvalds 683bd4c625cSLinus Torvalds static void write_ordered_chunk(struct buffer_chunk *chunk) 684bd4c625cSLinus Torvalds { 6851da177e4SLinus Torvalds int i; 6861da177e4SLinus Torvalds for (i = 0; i < chunk->nr; i++) { 6871da177e4SLinus Torvalds submit_ordered_buffer(chunk->bh[i]); 6881da177e4SLinus Torvalds } 6891da177e4SLinus Torvalds chunk->nr = 0; 6901da177e4SLinus Torvalds } 6911da177e4SLinus Torvalds 6921da177e4SLinus Torvalds static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh, 693bd4c625cSLinus Torvalds spinlock_t * lock, void (fn) (struct buffer_chunk *)) 6941da177e4SLinus Torvalds { 6951da177e4SLinus Torvalds int ret = 0; 69614a61442SEric Sesterhenn BUG_ON(chunk->nr >= CHUNK_SIZE); 6971da177e4SLinus Torvalds chunk->bh[chunk->nr++] = bh; 6981da177e4SLinus Torvalds if (chunk->nr >= CHUNK_SIZE) { 6991da177e4SLinus Torvalds ret = 1; 70054cc6ceaSFabian Frederick if (lock) { 7011da177e4SLinus Torvalds spin_unlock(lock); 7021da177e4SLinus Torvalds fn(chunk); 7031da177e4SLinus Torvalds spin_lock(lock); 70454cc6ceaSFabian Frederick } else { 70554cc6ceaSFabian Frederick fn(chunk); 70654cc6ceaSFabian Frederick } 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); 873dfec8a14SMike Christie ll_rw_block(REQ_OP_WRITE, 0, 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); 1060dfec8a14SMike Christie ll_rw_block(REQ_OP_WRITE, 0, 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)) 111470fd7614SChristoph Hellwig __sync_dirty_buffer(jl->j_commit_bh, 111570fd7614SChristoph Hellwig REQ_PREFLUSH | REQ_FUA); 11167cd33ad2SChristoph Hellwig else 11171da177e4SLinus Torvalds sync_dirty_buffer(jl->j_commit_bh); 1118278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 11195d5e8156SJeff Mahoney } 11201da177e4SLinus Torvalds 1121098297b2SJeff Mahoney /* 1122098297b2SJeff Mahoney * If there was a write error in the journal - we can't commit this 11231da177e4SLinus Torvalds * transaction - it will be invalid and, if successful, will just end 1124098297b2SJeff Mahoney * up propagating the write error out to the filesystem. 1125098297b2SJeff Mahoney */ 11261da177e4SLinus Torvalds if (unlikely(!buffer_uptodate(jl->j_commit_bh))) { 11271da177e4SLinus Torvalds #ifdef CONFIG_REISERFS_CHECK 112845b03d5eSJeff Mahoney reiserfs_warning(s, "journal-615", "buffer write failed"); 11291da177e4SLinus Torvalds #endif 11301da177e4SLinus Torvalds retval = -EIO; 11311da177e4SLinus Torvalds } 11321da177e4SLinus Torvalds bforget(jl->j_commit_bh); 11331da177e4SLinus Torvalds if (journal->j_last_commit_id != 0 && 11341da177e4SLinus Torvalds (jl->j_trans_id - journal->j_last_commit_id) != 1) { 113545b03d5eSJeff Mahoney reiserfs_warning(s, "clm-2200", "last commit %lu, current %lu", 1136bd4c625cSLinus Torvalds journal->j_last_commit_id, jl->j_trans_id); 11371da177e4SLinus Torvalds } 11381da177e4SLinus Torvalds journal->j_last_commit_id = jl->j_trans_id; 11391da177e4SLinus Torvalds 1140098297b2SJeff Mahoney /* 1141098297b2SJeff Mahoney * now, every commit block is on the disk. It is safe to allow 1142098297b2SJeff Mahoney * blocks freed during this transaction to be reallocated 1143098297b2SJeff Mahoney */ 11441da177e4SLinus Torvalds cleanup_freed_for_journal_list(s, jl); 11451da177e4SLinus Torvalds 11461da177e4SLinus Torvalds retval = retval ? retval : journal->j_errno; 11471da177e4SLinus Torvalds 11481da177e4SLinus Torvalds /* mark the metadata dirty */ 11491da177e4SLinus Torvalds if (!retval) 11501da177e4SLinus Torvalds dirty_one_transaction(s, jl); 1151a228bf8fSJeff Mahoney atomic_dec(&jl->j_commit_left); 11521da177e4SLinus Torvalds 11531da177e4SLinus Torvalds if (flushall) { 1154a228bf8fSJeff Mahoney atomic_set(&jl->j_older_commits_done, 1); 11551da177e4SLinus Torvalds } 115690415deaSJeff Mahoney mutex_unlock(&jl->j_commit_mutex); 11571da177e4SLinus Torvalds put_jl: 11581da177e4SLinus Torvalds put_journal_list(s, jl); 11591da177e4SLinus Torvalds 11601da177e4SLinus Torvalds if (retval) 1161bd4c625cSLinus Torvalds reiserfs_abort(s, retval, "Journal write error in %s", 1162fbe5498bSHarvey Harrison __func__); 11631da177e4SLinus Torvalds return retval; 11641da177e4SLinus Torvalds } 11651da177e4SLinus Torvalds 11661da177e4SLinus Torvalds /* 1167098297b2SJeff Mahoney * flush_journal_list frequently needs to find a newer transaction for a 1168098297b2SJeff Mahoney * given block. This does that, or returns NULL if it can't find anything 11691da177e4SLinus Torvalds */ 1170bd4c625cSLinus Torvalds static struct reiserfs_journal_list *find_newer_jl_for_cn(struct 1171bd4c625cSLinus Torvalds reiserfs_journal_cnode 1172bd4c625cSLinus Torvalds *cn) 1173bd4c625cSLinus Torvalds { 11741da177e4SLinus Torvalds struct super_block *sb = cn->sb; 11751da177e4SLinus Torvalds b_blocknr_t blocknr = cn->blocknr; 11761da177e4SLinus Torvalds 11771da177e4SLinus Torvalds cn = cn->hprev; 11781da177e4SLinus Torvalds while (cn) { 11791da177e4SLinus Torvalds if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) { 11801da177e4SLinus Torvalds return cn->jlist; 11811da177e4SLinus Torvalds } 11821da177e4SLinus Torvalds cn = cn->hprev; 11831da177e4SLinus Torvalds } 11841da177e4SLinus Torvalds return NULL; 11851da177e4SLinus Torvalds } 11861da177e4SLinus Torvalds 1187bd4c625cSLinus Torvalds static void remove_journal_hash(struct super_block *, 1188bd4c625cSLinus Torvalds struct reiserfs_journal_cnode **, 1189bd4c625cSLinus Torvalds struct reiserfs_journal_list *, unsigned long, 1190bd4c625cSLinus Torvalds int); 11911da177e4SLinus Torvalds 11921da177e4SLinus Torvalds /* 1193098297b2SJeff Mahoney * once all the real blocks have been flushed, it is safe to remove them 1194098297b2SJeff Mahoney * from the journal list for this transaction. Aside from freeing the 1195098297b2SJeff Mahoney * cnode, this also allows the block to be reallocated for data blocks 1196098297b2SJeff Mahoney * if it had been deleted. 11971da177e4SLinus Torvalds */ 1198a9dd3643SJeff Mahoney static void remove_all_from_journal_list(struct super_block *sb, 1199bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl, 1200bd4c625cSLinus Torvalds int debug) 1201bd4c625cSLinus Torvalds { 1202a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 12031da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn, *last; 12041da177e4SLinus Torvalds cn = jl->j_realblock; 12051da177e4SLinus Torvalds 1206098297b2SJeff Mahoney /* 1207098297b2SJeff Mahoney * which is better, to lock once around the whole loop, or 1208098297b2SJeff Mahoney * to lock for each call to remove_journal_hash? 12091da177e4SLinus Torvalds */ 12101da177e4SLinus Torvalds while (cn) { 12111da177e4SLinus Torvalds if (cn->blocknr != 0) { 12121da177e4SLinus Torvalds if (debug) { 1213a9dd3643SJeff Mahoney reiserfs_warning(sb, "reiserfs-2201", 1214bd4c625cSLinus Torvalds "block %u, bh is %d, state %ld", 1215bd4c625cSLinus Torvalds cn->blocknr, cn->bh ? 1 : 0, 1216bd4c625cSLinus Torvalds cn->state); 12171da177e4SLinus Torvalds } 12181da177e4SLinus Torvalds cn->state = 0; 1219a9dd3643SJeff Mahoney remove_journal_hash(sb, journal->j_list_hash_table, 1220bd4c625cSLinus Torvalds jl, cn->blocknr, 1); 12211da177e4SLinus Torvalds } 12221da177e4SLinus Torvalds last = cn; 12231da177e4SLinus Torvalds cn = cn->next; 1224a9dd3643SJeff Mahoney free_cnode(sb, last); 12251da177e4SLinus Torvalds } 12261da177e4SLinus Torvalds jl->j_realblock = NULL; 12271da177e4SLinus Torvalds } 12281da177e4SLinus Torvalds 12291da177e4SLinus Torvalds /* 1230098297b2SJeff Mahoney * if this timestamp is greater than the timestamp we wrote last to the 1231098297b2SJeff Mahoney * header block, write it to the header block. once this is done, I can 1232098297b2SJeff Mahoney * safely say the log area for this transaction won't ever be replayed, 1233098297b2SJeff Mahoney * and I can start releasing blocks in this transaction for reuse as data 1234098297b2SJeff Mahoney * blocks. called by flush_journal_list, before it calls 1235098297b2SJeff Mahoney * remove_all_from_journal_list 12361da177e4SLinus Torvalds */ 1237a9dd3643SJeff Mahoney static int _update_journal_header_block(struct super_block *sb, 1238bd4c625cSLinus Torvalds unsigned long offset, 1239600ed416SJeff Mahoney unsigned int trans_id) 1240bd4c625cSLinus Torvalds { 12411da177e4SLinus Torvalds struct reiserfs_journal_header *jh; 1242a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 1243278f6679SJeff Mahoney int depth; 12441da177e4SLinus Torvalds 12451da177e4SLinus Torvalds if (reiserfs_is_journal_aborted(journal)) 12461da177e4SLinus Torvalds return -EIO; 12471da177e4SLinus Torvalds 12481da177e4SLinus Torvalds if (trans_id >= journal->j_last_flush_trans_id) { 12491da177e4SLinus Torvalds if (buffer_locked((journal->j_header_bh))) { 1250278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(sb); 1251278f6679SJeff Mahoney __wait_on_buffer(journal->j_header_bh); 1252278f6679SJeff Mahoney reiserfs_write_lock_nested(sb, depth); 12531da177e4SLinus Torvalds if (unlikely(!buffer_uptodate(journal->j_header_bh))) { 12541da177e4SLinus Torvalds #ifdef CONFIG_REISERFS_CHECK 1255a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-699", 125645b03d5eSJeff Mahoney "buffer write failed"); 12571da177e4SLinus Torvalds #endif 12581da177e4SLinus Torvalds return -EIO; 12591da177e4SLinus Torvalds } 12601da177e4SLinus Torvalds } 12611da177e4SLinus Torvalds journal->j_last_flush_trans_id = trans_id; 12621da177e4SLinus Torvalds journal->j_first_unflushed_offset = offset; 1263bd4c625cSLinus Torvalds jh = (struct reiserfs_journal_header *)(journal->j_header_bh-> 1264bd4c625cSLinus Torvalds b_data); 12651da177e4SLinus Torvalds jh->j_last_flush_trans_id = cpu_to_le32(trans_id); 12661da177e4SLinus Torvalds jh->j_first_unflushed_offset = cpu_to_le32(offset); 12671da177e4SLinus Torvalds jh->j_mount_id = cpu_to_le32(journal->j_mount_id); 12681da177e4SLinus Torvalds 12691da177e4SLinus Torvalds set_buffer_dirty(journal->j_header_bh); 1270278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(sb); 12717cd33ad2SChristoph Hellwig 12727cd33ad2SChristoph Hellwig if (reiserfs_barrier_flush(sb)) 127370fd7614SChristoph Hellwig __sync_dirty_buffer(journal->j_header_bh, 127470fd7614SChristoph Hellwig REQ_PREFLUSH | REQ_FUA); 12757cd33ad2SChristoph Hellwig else 12761da177e4SLinus Torvalds sync_dirty_buffer(journal->j_header_bh); 12777cd33ad2SChristoph Hellwig 1278278f6679SJeff Mahoney reiserfs_write_lock_nested(sb, depth); 12791da177e4SLinus Torvalds if (!buffer_uptodate(journal->j_header_bh)) { 1280a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-837", 128145b03d5eSJeff Mahoney "IO error during journal replay"); 12821da177e4SLinus Torvalds return -EIO; 12831da177e4SLinus Torvalds } 12841da177e4SLinus Torvalds } 12851da177e4SLinus Torvalds return 0; 12861da177e4SLinus Torvalds } 12871da177e4SLinus Torvalds 1288a9dd3643SJeff Mahoney static int update_journal_header_block(struct super_block *sb, 12891da177e4SLinus Torvalds unsigned long offset, 1290600ed416SJeff Mahoney unsigned int trans_id) 1291bd4c625cSLinus Torvalds { 1292a9dd3643SJeff Mahoney return _update_journal_header_block(sb, offset, trans_id); 12931da177e4SLinus Torvalds } 1294bd4c625cSLinus Torvalds 12951da177e4SLinus Torvalds /* 12961da177e4SLinus Torvalds ** flush any and all journal lists older than you are 12971da177e4SLinus Torvalds ** can only be called from flush_journal_list 12981da177e4SLinus Torvalds */ 1299a9dd3643SJeff Mahoney static int flush_older_journal_lists(struct super_block *sb, 13001da177e4SLinus Torvalds struct reiserfs_journal_list *jl) 13011da177e4SLinus Torvalds { 13021da177e4SLinus Torvalds struct list_head *entry; 13031da177e4SLinus Torvalds struct reiserfs_journal_list *other_jl; 1304a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 1305600ed416SJeff Mahoney unsigned int trans_id = jl->j_trans_id; 13061da177e4SLinus Torvalds 1307098297b2SJeff Mahoney /* 1308098297b2SJeff Mahoney * we know we are the only ones flushing things, no extra race 13091da177e4SLinus Torvalds * protection is required. 13101da177e4SLinus Torvalds */ 13111da177e4SLinus Torvalds restart: 13121da177e4SLinus Torvalds entry = journal->j_journal_list.next; 13131da177e4SLinus Torvalds /* Did we wrap? */ 13141da177e4SLinus Torvalds if (entry == &journal->j_journal_list) 13151da177e4SLinus Torvalds return 0; 13161da177e4SLinus Torvalds other_jl = JOURNAL_LIST_ENTRY(entry); 13171da177e4SLinus Torvalds if (other_jl->j_trans_id < trans_id) { 13181da177e4SLinus Torvalds BUG_ON(other_jl->j_refcount <= 0); 13191da177e4SLinus Torvalds /* do not flush all */ 1320a9dd3643SJeff Mahoney flush_journal_list(sb, other_jl, 0); 13211da177e4SLinus Torvalds 13221da177e4SLinus Torvalds /* other_jl is now deleted from the list */ 13231da177e4SLinus Torvalds goto restart; 13241da177e4SLinus Torvalds } 13251da177e4SLinus Torvalds return 0; 13261da177e4SLinus Torvalds } 13271da177e4SLinus Torvalds 13281da177e4SLinus Torvalds static void del_from_work_list(struct super_block *s, 1329bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl) 1330bd4c625cSLinus Torvalds { 13311da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 13321da177e4SLinus Torvalds if (!list_empty(&jl->j_working_list)) { 13331da177e4SLinus Torvalds list_del_init(&jl->j_working_list); 13341da177e4SLinus Torvalds journal->j_num_work_lists--; 13351da177e4SLinus Torvalds } 13361da177e4SLinus Torvalds } 13371da177e4SLinus Torvalds 1338098297b2SJeff Mahoney /* 1339098297b2SJeff Mahoney * flush a journal list, both commit and real blocks 1340098297b2SJeff Mahoney * 1341098297b2SJeff Mahoney * always set flushall to 1, unless you are calling from inside 1342098297b2SJeff Mahoney * flush_journal_list 1343098297b2SJeff Mahoney * 1344098297b2SJeff Mahoney * IMPORTANT. This can only be called while there are no journal writers, 1345098297b2SJeff Mahoney * and the journal is locked. That means it can only be called from 1346098297b2SJeff Mahoney * do_journal_end, or by journal_release 13471da177e4SLinus Torvalds */ 13481da177e4SLinus Torvalds static int flush_journal_list(struct super_block *s, 1349bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl, int flushall) 1350bd4c625cSLinus Torvalds { 13511da177e4SLinus Torvalds struct reiserfs_journal_list *pjl; 13521da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn, *last; 13531da177e4SLinus Torvalds int count; 13541da177e4SLinus Torvalds int was_jwait = 0; 13551da177e4SLinus Torvalds int was_dirty = 0; 13561da177e4SLinus Torvalds struct buffer_head *saved_bh; 13571da177e4SLinus Torvalds unsigned long j_len_saved = jl->j_len; 13581da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 13591da177e4SLinus Torvalds int err = 0; 1360278f6679SJeff Mahoney int depth; 13611da177e4SLinus Torvalds 13621da177e4SLinus Torvalds BUG_ON(j_len_saved <= 0); 13631da177e4SLinus Torvalds 13641da177e4SLinus Torvalds if (atomic_read(&journal->j_wcount) != 0) { 136545b03d5eSJeff Mahoney reiserfs_warning(s, "clm-2048", "called with wcount %d", 13661da177e4SLinus Torvalds atomic_read(&journal->j_wcount)); 13671da177e4SLinus Torvalds } 13681da177e4SLinus Torvalds 13691da177e4SLinus Torvalds /* if flushall == 0, the lock is already held */ 13701da177e4SLinus Torvalds if (flushall) { 13718ebc4232SFrederic Weisbecker reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s); 1372afe70259SJeff Mahoney } else if (mutex_trylock(&journal->j_flush_mutex)) { 13731da177e4SLinus Torvalds BUG(); 13741da177e4SLinus Torvalds } 13751da177e4SLinus Torvalds 13761da177e4SLinus Torvalds count = 0; 13771da177e4SLinus Torvalds if (j_len_saved > journal->j_trans_max) { 1378c3a9c210SJeff Mahoney reiserfs_panic(s, "journal-715", "length is %lu, trans id %lu", 1379bd4c625cSLinus Torvalds j_len_saved, jl->j_trans_id); 13801da177e4SLinus Torvalds return 0; 13811da177e4SLinus Torvalds } 13821da177e4SLinus Torvalds 13831da177e4SLinus Torvalds /* if all the work is already done, get out of here */ 1384a228bf8fSJeff Mahoney if (atomic_read(&jl->j_nonzerolen) <= 0 && 1385a228bf8fSJeff Mahoney atomic_read(&jl->j_commit_left) <= 0) { 13861da177e4SLinus Torvalds goto flush_older_and_return; 13871da177e4SLinus Torvalds } 13881da177e4SLinus Torvalds 1389098297b2SJeff Mahoney /* 1390098297b2SJeff Mahoney * start by putting the commit list on disk. This will also flush 1391098297b2SJeff Mahoney * the commit lists of any olders transactions 13921da177e4SLinus Torvalds */ 13931da177e4SLinus Torvalds flush_commit_list(s, jl, 1); 13941da177e4SLinus Torvalds 1395bd4c625cSLinus Torvalds if (!(jl->j_state & LIST_DIRTY) 1396bd4c625cSLinus Torvalds && !reiserfs_is_journal_aborted(journal)) 13971da177e4SLinus Torvalds BUG(); 13981da177e4SLinus Torvalds 13991da177e4SLinus Torvalds /* are we done now? */ 1400a228bf8fSJeff Mahoney if (atomic_read(&jl->j_nonzerolen) <= 0 && 1401a228bf8fSJeff Mahoney atomic_read(&jl->j_commit_left) <= 0) { 14021da177e4SLinus Torvalds goto flush_older_and_return; 14031da177e4SLinus Torvalds } 14041da177e4SLinus Torvalds 1405098297b2SJeff Mahoney /* 1406098297b2SJeff Mahoney * loop through each cnode, see if we need to write it, 1407098297b2SJeff Mahoney * or wait on a more recent transaction, or just ignore it 14081da177e4SLinus Torvalds */ 1409a228bf8fSJeff Mahoney if (atomic_read(&journal->j_wcount) != 0) { 1410c3a9c210SJeff Mahoney reiserfs_panic(s, "journal-844", "journal list is flushing, " 1411c3a9c210SJeff Mahoney "wcount is not 0"); 14121da177e4SLinus Torvalds } 14131da177e4SLinus Torvalds cn = jl->j_realblock; 14141da177e4SLinus Torvalds while (cn) { 14151da177e4SLinus Torvalds was_jwait = 0; 14161da177e4SLinus Torvalds was_dirty = 0; 14171da177e4SLinus Torvalds saved_bh = NULL; 14181da177e4SLinus Torvalds /* blocknr of 0 is no longer in the hash, ignore it */ 14191da177e4SLinus Torvalds if (cn->blocknr == 0) { 14201da177e4SLinus Torvalds goto free_cnode; 14211da177e4SLinus Torvalds } 14221da177e4SLinus Torvalds 1423098297b2SJeff Mahoney /* 1424098297b2SJeff Mahoney * This transaction failed commit. 1425098297b2SJeff Mahoney * Don't write out to the disk 1426098297b2SJeff Mahoney */ 14271da177e4SLinus Torvalds if (!(jl->j_state & LIST_DIRTY)) 14281da177e4SLinus Torvalds goto free_cnode; 14291da177e4SLinus Torvalds 14301da177e4SLinus Torvalds pjl = find_newer_jl_for_cn(cn); 1431098297b2SJeff Mahoney /* 1432098297b2SJeff Mahoney * the order is important here. We check pjl to make sure we 1433098297b2SJeff Mahoney * don't clear BH_JDirty_wait if we aren't the one writing this 1434098297b2SJeff Mahoney * block to disk 14351da177e4SLinus Torvalds */ 14361da177e4SLinus Torvalds if (!pjl && cn->bh) { 14371da177e4SLinus Torvalds saved_bh = cn->bh; 14381da177e4SLinus Torvalds 1439098297b2SJeff Mahoney /* 1440098297b2SJeff Mahoney * we do this to make sure nobody releases the 1441098297b2SJeff Mahoney * buffer while we are working with it 14421da177e4SLinus Torvalds */ 14431da177e4SLinus Torvalds get_bh(saved_bh); 14441da177e4SLinus Torvalds 14451da177e4SLinus Torvalds if (buffer_journal_dirty(saved_bh)) { 14461da177e4SLinus Torvalds BUG_ON(!can_dirty(cn)); 14471da177e4SLinus Torvalds was_jwait = 1; 14481da177e4SLinus Torvalds was_dirty = 1; 14491da177e4SLinus Torvalds } else if (can_dirty(cn)) { 1450098297b2SJeff Mahoney /* 1451098297b2SJeff Mahoney * everything with !pjl && jwait 1452098297b2SJeff Mahoney * should be writable 1453098297b2SJeff Mahoney */ 14541da177e4SLinus Torvalds BUG(); 14551da177e4SLinus Torvalds } 14561da177e4SLinus Torvalds } 14571da177e4SLinus Torvalds 1458098297b2SJeff Mahoney /* 1459098297b2SJeff Mahoney * if someone has this block in a newer transaction, just make 1460098297b2SJeff Mahoney * sure they are committed, and don't try writing it to disk 14611da177e4SLinus Torvalds */ 14621da177e4SLinus Torvalds if (pjl) { 14631da177e4SLinus Torvalds if (atomic_read(&pjl->j_commit_left)) 14641da177e4SLinus Torvalds flush_commit_list(s, pjl, 1); 14651da177e4SLinus Torvalds goto free_cnode; 14661da177e4SLinus Torvalds } 14671da177e4SLinus Torvalds 1468098297b2SJeff Mahoney /* 1469098297b2SJeff Mahoney * bh == NULL when the block got to disk on its own, OR, 1470098297b2SJeff Mahoney * the block got freed in a future transaction 14711da177e4SLinus Torvalds */ 14721da177e4SLinus Torvalds if (saved_bh == NULL) { 14731da177e4SLinus Torvalds goto free_cnode; 14741da177e4SLinus Torvalds } 14751da177e4SLinus Torvalds 1476098297b2SJeff Mahoney /* 1477098297b2SJeff Mahoney * this should never happen. kupdate_one_transaction has 1478098297b2SJeff Mahoney * this list locked while it works, so we should never see a 1479098297b2SJeff Mahoney * buffer here that is not marked JDirty_wait 14801da177e4SLinus Torvalds */ 14811da177e4SLinus Torvalds if ((!was_jwait) && !buffer_locked(saved_bh)) { 148245b03d5eSJeff Mahoney reiserfs_warning(s, "journal-813", 148345b03d5eSJeff Mahoney "BAD! buffer %llu %cdirty %cjwait, " 14841da177e4SLinus Torvalds "not in a newer tranasction", 1485bd4c625cSLinus Torvalds (unsigned long long)saved_bh-> 1486bd4c625cSLinus Torvalds b_blocknr, was_dirty ? ' ' : '!', 1487bd4c625cSLinus Torvalds was_jwait ? ' ' : '!'); 14881da177e4SLinus Torvalds } 14891da177e4SLinus Torvalds if (was_dirty) { 1490098297b2SJeff Mahoney /* 1491098297b2SJeff Mahoney * we inc again because saved_bh gets decremented 1492098297b2SJeff Mahoney * at free_cnode 1493098297b2SJeff Mahoney */ 14941da177e4SLinus Torvalds get_bh(saved_bh); 14951da177e4SLinus Torvalds set_bit(BLOCK_NEEDS_FLUSH, &cn->state); 14961da177e4SLinus Torvalds lock_buffer(saved_bh); 14971da177e4SLinus Torvalds BUG_ON(cn->blocknr != saved_bh->b_blocknr); 14981da177e4SLinus Torvalds if (buffer_dirty(saved_bh)) 14991da177e4SLinus Torvalds submit_logged_buffer(saved_bh); 15001da177e4SLinus Torvalds else 15011da177e4SLinus Torvalds unlock_buffer(saved_bh); 15021da177e4SLinus Torvalds count++; 15031da177e4SLinus Torvalds } else { 150445b03d5eSJeff Mahoney reiserfs_warning(s, "clm-2082", 150545b03d5eSJeff Mahoney "Unable to flush buffer %llu in %s", 1506bd4c625cSLinus Torvalds (unsigned long long)saved_bh-> 1507fbe5498bSHarvey Harrison b_blocknr, __func__); 15081da177e4SLinus Torvalds } 15091da177e4SLinus Torvalds free_cnode: 15101da177e4SLinus Torvalds last = cn; 15111da177e4SLinus Torvalds cn = cn->next; 15121da177e4SLinus Torvalds if (saved_bh) { 1513098297b2SJeff Mahoney /* 1514098297b2SJeff Mahoney * we incremented this to keep others from 1515098297b2SJeff Mahoney * taking the buffer head away 1516098297b2SJeff Mahoney */ 15171da177e4SLinus Torvalds put_bh(saved_bh); 1518a228bf8fSJeff Mahoney if (atomic_read(&saved_bh->b_count) < 0) { 151945b03d5eSJeff Mahoney reiserfs_warning(s, "journal-945", 152045b03d5eSJeff Mahoney "saved_bh->b_count < 0"); 15211da177e4SLinus Torvalds } 15221da177e4SLinus Torvalds } 15231da177e4SLinus Torvalds } 15241da177e4SLinus Torvalds if (count > 0) { 15251da177e4SLinus Torvalds cn = jl->j_realblock; 15261da177e4SLinus Torvalds while (cn) { 15271da177e4SLinus Torvalds if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) { 15281da177e4SLinus Torvalds if (!cn->bh) { 1529c3a9c210SJeff Mahoney reiserfs_panic(s, "journal-1011", 1530c3a9c210SJeff Mahoney "cn->bh is NULL"); 15311da177e4SLinus Torvalds } 15328ebc4232SFrederic Weisbecker 1533278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(s); 1534278f6679SJeff Mahoney __wait_on_buffer(cn->bh); 1535278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 15368ebc4232SFrederic Weisbecker 15371da177e4SLinus Torvalds if (!cn->bh) { 1538c3a9c210SJeff Mahoney reiserfs_panic(s, "journal-1012", 1539c3a9c210SJeff Mahoney "cn->bh is NULL"); 15401da177e4SLinus Torvalds } 15411da177e4SLinus Torvalds if (unlikely(!buffer_uptodate(cn->bh))) { 15421da177e4SLinus Torvalds #ifdef CONFIG_REISERFS_CHECK 154345b03d5eSJeff Mahoney reiserfs_warning(s, "journal-949", 154445b03d5eSJeff Mahoney "buffer write failed"); 15451da177e4SLinus Torvalds #endif 15461da177e4SLinus Torvalds err = -EIO; 15471da177e4SLinus Torvalds } 1548098297b2SJeff Mahoney /* 1549098297b2SJeff Mahoney * note, we must clear the JDirty_wait bit 1550098297b2SJeff Mahoney * after the up to date check, otherwise we 1551098297b2SJeff Mahoney * race against our flushpage routine 15521da177e4SLinus Torvalds */ 1553bd4c625cSLinus Torvalds BUG_ON(!test_clear_buffer_journal_dirty 1554bd4c625cSLinus Torvalds (cn->bh)); 15551da177e4SLinus Torvalds 1556398c95bdSChris Mason /* drop one ref for us */ 15571da177e4SLinus Torvalds put_bh(cn->bh); 1558398c95bdSChris Mason /* drop one ref for journal_mark_dirty */ 1559398c95bdSChris Mason release_buffer_page(cn->bh); 15601da177e4SLinus Torvalds } 15611da177e4SLinus Torvalds cn = cn->next; 15621da177e4SLinus Torvalds } 15631da177e4SLinus Torvalds } 15641da177e4SLinus Torvalds 15651da177e4SLinus Torvalds if (err) 1566bd4c625cSLinus Torvalds reiserfs_abort(s, -EIO, 1567bd4c625cSLinus Torvalds "Write error while pushing transaction to disk in %s", 1568fbe5498bSHarvey Harrison __func__); 15691da177e4SLinus Torvalds flush_older_and_return: 15701da177e4SLinus Torvalds 1571098297b2SJeff Mahoney /* 1572098297b2SJeff Mahoney * before we can update the journal header block, we _must_ flush all 1573098297b2SJeff Mahoney * real blocks from all older transactions to disk. This is because 1574098297b2SJeff Mahoney * once the header block is updated, this transaction will not be 1575098297b2SJeff Mahoney * replayed after a crash 15761da177e4SLinus Torvalds */ 15771da177e4SLinus Torvalds if (flushall) { 15781da177e4SLinus Torvalds flush_older_journal_lists(s, jl); 15791da177e4SLinus Torvalds } 15801da177e4SLinus Torvalds 15811da177e4SLinus Torvalds err = journal->j_errno; 1582098297b2SJeff Mahoney /* 1583098297b2SJeff Mahoney * before we can remove everything from the hash tables for this 1584098297b2SJeff Mahoney * transaction, we must make sure it can never be replayed 1585098297b2SJeff Mahoney * 1586098297b2SJeff Mahoney * since we are only called from do_journal_end, we know for sure there 1587098297b2SJeff Mahoney * are no allocations going on while we are flushing journal lists. So, 1588098297b2SJeff Mahoney * we only need to update the journal header block for the last list 1589098297b2SJeff Mahoney * being flushed 15901da177e4SLinus Torvalds */ 15911da177e4SLinus Torvalds if (!err && flushall) { 1592bd4c625cSLinus Torvalds err = 1593bd4c625cSLinus Torvalds update_journal_header_block(s, 1594bd4c625cSLinus Torvalds (jl->j_start + jl->j_len + 1595bd4c625cSLinus Torvalds 2) % SB_ONDISK_JOURNAL_SIZE(s), 1596bd4c625cSLinus Torvalds jl->j_trans_id); 15971da177e4SLinus Torvalds if (err) 1598bd4c625cSLinus Torvalds reiserfs_abort(s, -EIO, 1599bd4c625cSLinus Torvalds "Write error while updating journal header in %s", 1600fbe5498bSHarvey Harrison __func__); 16011da177e4SLinus Torvalds } 16021da177e4SLinus Torvalds remove_all_from_journal_list(s, jl, 0); 16031da177e4SLinus Torvalds list_del_init(&jl->j_list); 16041da177e4SLinus Torvalds journal->j_num_lists--; 16051da177e4SLinus Torvalds del_from_work_list(s, jl); 16061da177e4SLinus Torvalds 16071da177e4SLinus Torvalds if (journal->j_last_flush_id != 0 && 16081da177e4SLinus Torvalds (jl->j_trans_id - journal->j_last_flush_id) != 1) { 160945b03d5eSJeff Mahoney reiserfs_warning(s, "clm-2201", "last flush %lu, current %lu", 1610bd4c625cSLinus Torvalds journal->j_last_flush_id, jl->j_trans_id); 16111da177e4SLinus Torvalds } 16121da177e4SLinus Torvalds journal->j_last_flush_id = jl->j_trans_id; 16131da177e4SLinus Torvalds 1614098297b2SJeff Mahoney /* 1615098297b2SJeff Mahoney * not strictly required since we are freeing the list, but it should 16161da177e4SLinus Torvalds * help find code using dead lists later on 16171da177e4SLinus Torvalds */ 16181da177e4SLinus Torvalds jl->j_len = 0; 1619a228bf8fSJeff Mahoney atomic_set(&jl->j_nonzerolen, 0); 16201da177e4SLinus Torvalds jl->j_start = 0; 16211da177e4SLinus Torvalds jl->j_realblock = NULL; 16221da177e4SLinus Torvalds jl->j_commit_bh = NULL; 16231da177e4SLinus Torvalds jl->j_trans_id = 0; 16241da177e4SLinus Torvalds jl->j_state = 0; 16251da177e4SLinus Torvalds put_journal_list(s, jl); 16261da177e4SLinus Torvalds if (flushall) 1627afe70259SJeff Mahoney mutex_unlock(&journal->j_flush_mutex); 16281da177e4SLinus Torvalds return err; 16291da177e4SLinus Torvalds } 16301da177e4SLinus Torvalds 16311da177e4SLinus Torvalds static int write_one_transaction(struct super_block *s, 16321da177e4SLinus Torvalds struct reiserfs_journal_list *jl, 16331da177e4SLinus Torvalds struct buffer_chunk *chunk) 16341da177e4SLinus Torvalds { 16351da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 16361da177e4SLinus Torvalds int ret = 0; 16371da177e4SLinus Torvalds 16381da177e4SLinus Torvalds jl->j_state |= LIST_TOUCHED; 16391da177e4SLinus Torvalds del_from_work_list(s, jl); 16401da177e4SLinus Torvalds if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) { 16411da177e4SLinus Torvalds return 0; 16421da177e4SLinus Torvalds } 16431da177e4SLinus Torvalds 16441da177e4SLinus Torvalds cn = jl->j_realblock; 16451da177e4SLinus Torvalds while (cn) { 1646098297b2SJeff Mahoney /* 1647098297b2SJeff Mahoney * if the blocknr == 0, this has been cleared from the hash, 1648098297b2SJeff Mahoney * skip it 16491da177e4SLinus Torvalds */ 16501da177e4SLinus Torvalds if (cn->blocknr == 0) { 16511da177e4SLinus Torvalds goto next; 16521da177e4SLinus Torvalds } 16531da177e4SLinus Torvalds if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) { 16541da177e4SLinus Torvalds struct buffer_head *tmp_bh; 1655098297b2SJeff Mahoney /* 1656098297b2SJeff Mahoney * we can race against journal_mark_freed when we try 16571da177e4SLinus Torvalds * to lock_buffer(cn->bh), so we have to inc the buffer 16581da177e4SLinus Torvalds * count, and recheck things after locking 16591da177e4SLinus Torvalds */ 16601da177e4SLinus Torvalds tmp_bh = cn->bh; 16611da177e4SLinus Torvalds get_bh(tmp_bh); 16621da177e4SLinus Torvalds lock_buffer(tmp_bh); 16631da177e4SLinus Torvalds if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) { 16641da177e4SLinus Torvalds if (!buffer_journal_dirty(tmp_bh) || 16651da177e4SLinus Torvalds buffer_journal_prepared(tmp_bh)) 16661da177e4SLinus Torvalds BUG(); 16671da177e4SLinus Torvalds add_to_chunk(chunk, tmp_bh, NULL, write_chunk); 16681da177e4SLinus Torvalds ret++; 16691da177e4SLinus Torvalds } else { 16701da177e4SLinus Torvalds /* note, cn->bh might be null now */ 16711da177e4SLinus Torvalds unlock_buffer(tmp_bh); 16721da177e4SLinus Torvalds } 16731da177e4SLinus Torvalds put_bh(tmp_bh); 16741da177e4SLinus Torvalds } 16751da177e4SLinus Torvalds next: 16761da177e4SLinus Torvalds cn = cn->next; 16771da177e4SLinus Torvalds cond_resched(); 16781da177e4SLinus Torvalds } 16791da177e4SLinus Torvalds return ret; 16801da177e4SLinus Torvalds } 16811da177e4SLinus Torvalds 16821da177e4SLinus Torvalds /* used by flush_commit_list */ 16831da177e4SLinus Torvalds static int dirty_one_transaction(struct super_block *s, 16841da177e4SLinus Torvalds struct reiserfs_journal_list *jl) 16851da177e4SLinus Torvalds { 16861da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 16871da177e4SLinus Torvalds struct reiserfs_journal_list *pjl; 16881da177e4SLinus Torvalds int ret = 0; 16891da177e4SLinus Torvalds 16901da177e4SLinus Torvalds jl->j_state |= LIST_DIRTY; 16911da177e4SLinus Torvalds cn = jl->j_realblock; 16921da177e4SLinus Torvalds while (cn) { 1693098297b2SJeff Mahoney /* 1694098297b2SJeff Mahoney * look for a more recent transaction that logged this 1695098297b2SJeff Mahoney * buffer. Only the most recent transaction with a buffer in 1696098297b2SJeff Mahoney * it is allowed to send that buffer to disk 16971da177e4SLinus Torvalds */ 16981da177e4SLinus Torvalds pjl = find_newer_jl_for_cn(cn); 1699bd4c625cSLinus Torvalds if (!pjl && cn->blocknr && cn->bh 1700bd4c625cSLinus Torvalds && buffer_journal_dirty(cn->bh)) { 17011da177e4SLinus Torvalds BUG_ON(!can_dirty(cn)); 1702098297b2SJeff Mahoney /* 1703098297b2SJeff Mahoney * if the buffer is prepared, it will either be logged 17041da177e4SLinus Torvalds * or restored. If restored, we need to make sure 17051da177e4SLinus Torvalds * it actually gets marked dirty 17061da177e4SLinus Torvalds */ 17071da177e4SLinus Torvalds clear_buffer_journal_new(cn->bh); 17081da177e4SLinus Torvalds if (buffer_journal_prepared(cn->bh)) { 17091da177e4SLinus Torvalds set_buffer_journal_restore_dirty(cn->bh); 17101da177e4SLinus Torvalds } else { 17111da177e4SLinus Torvalds set_buffer_journal_test(cn->bh); 17121da177e4SLinus Torvalds mark_buffer_dirty(cn->bh); 17131da177e4SLinus Torvalds } 17141da177e4SLinus Torvalds } 17151da177e4SLinus Torvalds cn = cn->next; 17161da177e4SLinus Torvalds } 17171da177e4SLinus Torvalds return ret; 17181da177e4SLinus Torvalds } 17191da177e4SLinus Torvalds 17201da177e4SLinus Torvalds static int kupdate_transactions(struct super_block *s, 17211da177e4SLinus Torvalds struct reiserfs_journal_list *jl, 17221da177e4SLinus Torvalds struct reiserfs_journal_list **next_jl, 1723600ed416SJeff Mahoney unsigned int *next_trans_id, 1724bd4c625cSLinus Torvalds int num_blocks, int num_trans) 1725bd4c625cSLinus Torvalds { 17261da177e4SLinus Torvalds int ret = 0; 17271da177e4SLinus Torvalds int written = 0; 17281da177e4SLinus Torvalds int transactions_flushed = 0; 1729600ed416SJeff Mahoney unsigned int orig_trans_id = jl->j_trans_id; 17301da177e4SLinus Torvalds struct buffer_chunk chunk; 17311da177e4SLinus Torvalds struct list_head *entry; 17321da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 17331da177e4SLinus Torvalds chunk.nr = 0; 17341da177e4SLinus Torvalds 1735a412f9efSFrederic Weisbecker reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s); 17361da177e4SLinus Torvalds if (!journal_list_still_alive(s, orig_trans_id)) { 17371da177e4SLinus Torvalds goto done; 17381da177e4SLinus Torvalds } 17391da177e4SLinus Torvalds 1740098297b2SJeff Mahoney /* 1741098297b2SJeff Mahoney * we've got j_flush_mutex held, nobody is going to delete any 17421da177e4SLinus Torvalds * of these lists out from underneath us 17431da177e4SLinus Torvalds */ 17441da177e4SLinus Torvalds while ((num_trans && transactions_flushed < num_trans) || 17451da177e4SLinus Torvalds (!num_trans && written < num_blocks)) { 17461da177e4SLinus Torvalds 17471da177e4SLinus Torvalds if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) || 1748bd4c625cSLinus Torvalds atomic_read(&jl->j_commit_left) 1749bd4c625cSLinus Torvalds || !(jl->j_state & LIST_DIRTY)) { 17501da177e4SLinus Torvalds del_from_work_list(s, jl); 17511da177e4SLinus Torvalds break; 17521da177e4SLinus Torvalds } 17531da177e4SLinus Torvalds ret = write_one_transaction(s, jl, &chunk); 17541da177e4SLinus Torvalds 17551da177e4SLinus Torvalds if (ret < 0) 17561da177e4SLinus Torvalds goto done; 17571da177e4SLinus Torvalds transactions_flushed++; 17581da177e4SLinus Torvalds written += ret; 17591da177e4SLinus Torvalds entry = jl->j_list.next; 17601da177e4SLinus Torvalds 17611da177e4SLinus Torvalds /* did we wrap? */ 17621da177e4SLinus Torvalds if (entry == &journal->j_journal_list) { 17631da177e4SLinus Torvalds break; 17641da177e4SLinus Torvalds } 17651da177e4SLinus Torvalds jl = JOURNAL_LIST_ENTRY(entry); 17661da177e4SLinus Torvalds 17671da177e4SLinus Torvalds /* don't bother with older transactions */ 17681da177e4SLinus Torvalds if (jl->j_trans_id <= orig_trans_id) 17691da177e4SLinus Torvalds break; 17701da177e4SLinus Torvalds } 17711da177e4SLinus Torvalds if (chunk.nr) { 17721da177e4SLinus Torvalds write_chunk(&chunk); 17731da177e4SLinus Torvalds } 17741da177e4SLinus Torvalds 17751da177e4SLinus Torvalds done: 1776afe70259SJeff Mahoney mutex_unlock(&journal->j_flush_mutex); 17771da177e4SLinus Torvalds return ret; 17781da177e4SLinus Torvalds } 17791da177e4SLinus Torvalds 1780098297b2SJeff Mahoney /* 1781098297b2SJeff Mahoney * for o_sync and fsync heavy applications, they tend to use 1782098297b2SJeff Mahoney * all the journa list slots with tiny transactions. These 1783098297b2SJeff Mahoney * trigger lots and lots of calls to update the header block, which 1784098297b2SJeff Mahoney * adds seeks and slows things down. 1785098297b2SJeff Mahoney * 1786098297b2SJeff Mahoney * This function tries to clear out a large chunk of the journal lists 1787098297b2SJeff Mahoney * at once, which makes everything faster since only the newest journal 1788098297b2SJeff Mahoney * list updates the header block 17891da177e4SLinus Torvalds */ 17901da177e4SLinus Torvalds static int flush_used_journal_lists(struct super_block *s, 1791bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl) 1792bd4c625cSLinus Torvalds { 17931da177e4SLinus Torvalds unsigned long len = 0; 17941da177e4SLinus Torvalds unsigned long cur_len; 17951da177e4SLinus Torvalds int ret; 17961da177e4SLinus Torvalds int i; 17971da177e4SLinus Torvalds int limit = 256; 17981da177e4SLinus Torvalds struct reiserfs_journal_list *tjl; 17991da177e4SLinus Torvalds struct reiserfs_journal_list *flush_jl; 1800600ed416SJeff Mahoney unsigned int trans_id; 18011da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 18021da177e4SLinus Torvalds 18031da177e4SLinus Torvalds flush_jl = tjl = jl; 18041da177e4SLinus Torvalds 18051da177e4SLinus Torvalds /* in data logging mode, try harder to flush a lot of blocks */ 18061da177e4SLinus Torvalds if (reiserfs_data_log(s)) 18071da177e4SLinus Torvalds limit = 1024; 18081da177e4SLinus Torvalds /* flush for 256 transactions or limit blocks, whichever comes first */ 18091da177e4SLinus Torvalds for (i = 0; i < 256 && len < limit; i++) { 18101da177e4SLinus Torvalds if (atomic_read(&tjl->j_commit_left) || 18111da177e4SLinus Torvalds tjl->j_trans_id < jl->j_trans_id) { 18121da177e4SLinus Torvalds break; 18131da177e4SLinus Torvalds } 18141da177e4SLinus Torvalds cur_len = atomic_read(&tjl->j_nonzerolen); 18151da177e4SLinus Torvalds if (cur_len > 0) { 18161da177e4SLinus Torvalds tjl->j_state &= ~LIST_TOUCHED; 18171da177e4SLinus Torvalds } 18181da177e4SLinus Torvalds len += cur_len; 18191da177e4SLinus Torvalds flush_jl = tjl; 18201da177e4SLinus Torvalds if (tjl->j_list.next == &journal->j_journal_list) 18211da177e4SLinus Torvalds break; 18221da177e4SLinus Torvalds tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next); 18231da177e4SLinus Torvalds } 1824721a769cSJeff Mahoney get_journal_list(jl); 1825721a769cSJeff Mahoney get_journal_list(flush_jl); 1826098297b2SJeff Mahoney 1827098297b2SJeff Mahoney /* 1828098297b2SJeff Mahoney * try to find a group of blocks we can flush across all the 1829098297b2SJeff Mahoney * transactions, but only bother if we've actually spanned 1830098297b2SJeff Mahoney * across multiple lists 18311da177e4SLinus Torvalds */ 18321da177e4SLinus Torvalds if (flush_jl != jl) { 18331da177e4SLinus Torvalds ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i); 18341da177e4SLinus Torvalds } 18351da177e4SLinus Torvalds flush_journal_list(s, flush_jl, 1); 1836721a769cSJeff Mahoney put_journal_list(s, flush_jl); 1837721a769cSJeff Mahoney put_journal_list(s, jl); 18381da177e4SLinus Torvalds return 0; 18391da177e4SLinus Torvalds } 18401da177e4SLinus Torvalds 18411da177e4SLinus Torvalds /* 1842098297b2SJeff Mahoney * removes any nodes in table with name block and dev as bh. 1843098297b2SJeff Mahoney * only touchs the hnext and hprev pointers. 18441da177e4SLinus Torvalds */ 18451da177e4SLinus Torvalds void remove_journal_hash(struct super_block *sb, 18461da177e4SLinus Torvalds struct reiserfs_journal_cnode **table, 18471da177e4SLinus Torvalds struct reiserfs_journal_list *jl, 18481da177e4SLinus Torvalds unsigned long block, int remove_freed) 18491da177e4SLinus Torvalds { 18501da177e4SLinus Torvalds struct reiserfs_journal_cnode *cur; 18511da177e4SLinus Torvalds struct reiserfs_journal_cnode **head; 18521da177e4SLinus Torvalds 18531da177e4SLinus Torvalds head = &(journal_hash(table, sb, block)); 18541da177e4SLinus Torvalds if (!head) { 18551da177e4SLinus Torvalds return; 18561da177e4SLinus Torvalds } 18571da177e4SLinus Torvalds cur = *head; 18581da177e4SLinus Torvalds while (cur) { 1859bd4c625cSLinus Torvalds if (cur->blocknr == block && cur->sb == sb 1860bd4c625cSLinus Torvalds && (jl == NULL || jl == cur->jlist) 1861bd4c625cSLinus Torvalds && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) { 18621da177e4SLinus Torvalds if (cur->hnext) { 18631da177e4SLinus Torvalds cur->hnext->hprev = cur->hprev; 18641da177e4SLinus Torvalds } 18651da177e4SLinus Torvalds if (cur->hprev) { 18661da177e4SLinus Torvalds cur->hprev->hnext = cur->hnext; 18671da177e4SLinus Torvalds } else { 18681da177e4SLinus Torvalds *head = cur->hnext; 18691da177e4SLinus Torvalds } 18701da177e4SLinus Torvalds cur->blocknr = 0; 18711da177e4SLinus Torvalds cur->sb = NULL; 18721da177e4SLinus Torvalds cur->state = 0; 1873098297b2SJeff Mahoney /* 1874098297b2SJeff Mahoney * anybody who clears the cur->bh will also 1875098297b2SJeff Mahoney * dec the nonzerolen 1876098297b2SJeff Mahoney */ 1877098297b2SJeff Mahoney if (cur->bh && cur->jlist) 1878a228bf8fSJeff Mahoney atomic_dec(&cur->jlist->j_nonzerolen); 18791da177e4SLinus Torvalds cur->bh = NULL; 18801da177e4SLinus Torvalds cur->jlist = NULL; 18811da177e4SLinus Torvalds } 18821da177e4SLinus Torvalds cur = cur->hnext; 18831da177e4SLinus Torvalds } 18841da177e4SLinus Torvalds } 18851da177e4SLinus Torvalds 1886a9dd3643SJeff Mahoney static void free_journal_ram(struct super_block *sb) 1887bd4c625cSLinus Torvalds { 1888a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 1889d739b42bSPekka Enberg kfree(journal->j_current_jl); 18901da177e4SLinus Torvalds journal->j_num_lists--; 18911da177e4SLinus Torvalds 18921da177e4SLinus Torvalds vfree(journal->j_cnode_free_orig); 1893a9dd3643SJeff Mahoney free_list_bitmaps(sb, journal->j_list_bitmap); 1894a9dd3643SJeff Mahoney free_bitmap_nodes(sb); /* must be after free_list_bitmaps */ 18951da177e4SLinus Torvalds if (journal->j_header_bh) { 18961da177e4SLinus Torvalds brelse(journal->j_header_bh); 18971da177e4SLinus Torvalds } 1898098297b2SJeff Mahoney /* 1899098297b2SJeff Mahoney * j_header_bh is on the journal dev, make sure 1900098297b2SJeff Mahoney * not to release the journal dev until we brelse j_header_bh 19011da177e4SLinus Torvalds */ 1902a9dd3643SJeff Mahoney release_journal_dev(sb, journal); 19031da177e4SLinus Torvalds vfree(journal); 19041da177e4SLinus Torvalds } 19051da177e4SLinus Torvalds 19061da177e4SLinus Torvalds /* 1907098297b2SJeff Mahoney * call on unmount. Only set error to 1 if you haven't made your way out 1908098297b2SJeff Mahoney * of read_super() yet. Any other caller must keep error at 0. 19091da177e4SLinus Torvalds */ 1910bd4c625cSLinus Torvalds static int do_journal_release(struct reiserfs_transaction_handle *th, 1911a9dd3643SJeff Mahoney struct super_block *sb, int error) 1912bd4c625cSLinus Torvalds { 19131da177e4SLinus Torvalds struct reiserfs_transaction_handle myth; 19141da177e4SLinus Torvalds int flushed = 0; 1915a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 19161da177e4SLinus Torvalds 1917098297b2SJeff Mahoney /* 1918098297b2SJeff Mahoney * we only want to flush out transactions if we were 1919098297b2SJeff Mahoney * called with error == 0 19201da177e4SLinus Torvalds */ 1921a9dd3643SJeff Mahoney if (!error && !(sb->s_flags & MS_RDONLY)) { 19221da177e4SLinus Torvalds /* end the current trans */ 19231da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 192458d85426SJeff Mahoney do_journal_end(th, FLUSH_ALL); 19251da177e4SLinus Torvalds 1926098297b2SJeff Mahoney /* 1927098297b2SJeff Mahoney * make sure something gets logged to force 1928098297b2SJeff Mahoney * our way into the flush code 1929098297b2SJeff Mahoney */ 1930b491dd17SJeff Mahoney if (!journal_join(&myth, sb)) { 1931a9dd3643SJeff Mahoney reiserfs_prepare_for_journal(sb, 1932a9dd3643SJeff Mahoney SB_BUFFER_WITH_SB(sb), 1933bd4c625cSLinus Torvalds 1); 193409f1b80bSJeff Mahoney journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb)); 193558d85426SJeff Mahoney do_journal_end(&myth, FLUSH_ALL); 19361da177e4SLinus Torvalds flushed = 1; 19371da177e4SLinus Torvalds } 19381da177e4SLinus Torvalds } 19391da177e4SLinus Torvalds 19401da177e4SLinus Torvalds /* this also catches errors during the do_journal_end above */ 19411da177e4SLinus Torvalds if (!error && reiserfs_is_journal_aborted(journal)) { 19421da177e4SLinus Torvalds memset(&myth, 0, sizeof(myth)); 1943b491dd17SJeff Mahoney if (!journal_join_abort(&myth, sb)) { 1944a9dd3643SJeff Mahoney reiserfs_prepare_for_journal(sb, 1945a9dd3643SJeff Mahoney SB_BUFFER_WITH_SB(sb), 1946bd4c625cSLinus Torvalds 1); 194709f1b80bSJeff Mahoney journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb)); 194858d85426SJeff Mahoney do_journal_end(&myth, FLUSH_ALL); 19491da177e4SLinus Torvalds } 19501da177e4SLinus Torvalds } 19511da177e4SLinus Torvalds 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 195901777836SJan Kara /* 196001777836SJan Kara * Cancel flushing of old commits. Note that neither of these works 196101777836SJan Kara * will be requeued because superblock is being shutdown and doesn't 196201777836SJan Kara * have MS_ACTIVE set. 196301777836SJan Kara */ 196471b0576bSJan Kara reiserfs_cancel_old_flush(sb); 196501777836SJan Kara /* wait for all commits to finish */ 196601777836SJan Kara cancel_delayed_work_sync(&SB_JOURNAL(sb)->j_work); 19671da177e4SLinus Torvalds 1968a9dd3643SJeff Mahoney free_journal_ram(sb); 19691da177e4SLinus Torvalds 19700523676dSFrederic Weisbecker reiserfs_write_lock(sb); 19710523676dSFrederic Weisbecker 19721da177e4SLinus Torvalds return 0; 19731da177e4SLinus Torvalds } 19741da177e4SLinus Torvalds 1975098297b2SJeff Mahoney /* * call on unmount. flush all journal trans, release all alloc'd ram */ 1976bd4c625cSLinus Torvalds int journal_release(struct reiserfs_transaction_handle *th, 1977a9dd3643SJeff Mahoney struct super_block *sb) 1978bd4c625cSLinus Torvalds { 1979a9dd3643SJeff Mahoney return do_journal_release(th, sb, 0); 19801da177e4SLinus Torvalds } 1981bd4c625cSLinus Torvalds 1982098297b2SJeff Mahoney /* only call from an error condition inside reiserfs_read_super! */ 1983bd4c625cSLinus Torvalds int journal_release_error(struct reiserfs_transaction_handle *th, 1984a9dd3643SJeff Mahoney struct super_block *sb) 1985bd4c625cSLinus Torvalds { 1986a9dd3643SJeff Mahoney return do_journal_release(th, sb, 1); 19871da177e4SLinus Torvalds } 19881da177e4SLinus Torvalds 1989098297b2SJeff Mahoney /* 1990098297b2SJeff Mahoney * compares description block with commit block. 1991098297b2SJeff Mahoney * returns 1 if they differ, 0 if they are the same 1992098297b2SJeff Mahoney */ 1993a9dd3643SJeff Mahoney static int journal_compare_desc_commit(struct super_block *sb, 1994bd4c625cSLinus Torvalds struct reiserfs_journal_desc *desc, 1995bd4c625cSLinus Torvalds struct reiserfs_journal_commit *commit) 1996bd4c625cSLinus Torvalds { 19971da177e4SLinus Torvalds if (get_commit_trans_id(commit) != get_desc_trans_id(desc) || 19981da177e4SLinus Torvalds get_commit_trans_len(commit) != get_desc_trans_len(desc) || 1999a9dd3643SJeff Mahoney get_commit_trans_len(commit) > SB_JOURNAL(sb)->j_trans_max || 2000bd4c625cSLinus Torvalds get_commit_trans_len(commit) <= 0) { 20011da177e4SLinus Torvalds return 1; 20021da177e4SLinus Torvalds } 20031da177e4SLinus Torvalds return 0; 20041da177e4SLinus Torvalds } 2005bd4c625cSLinus Torvalds 2006098297b2SJeff Mahoney /* 2007098297b2SJeff Mahoney * returns 0 if it did not find a description block 2008098297b2SJeff Mahoney * returns -1 if it found a corrupt commit block 2009098297b2SJeff Mahoney * returns 1 if both desc and commit were valid 2010098297b2SJeff Mahoney * NOTE: only called during fs mount 20111da177e4SLinus Torvalds */ 2012a9dd3643SJeff Mahoney static int journal_transaction_is_valid(struct super_block *sb, 2013bd4c625cSLinus Torvalds struct buffer_head *d_bh, 2014600ed416SJeff Mahoney unsigned int *oldest_invalid_trans_id, 2015bd4c625cSLinus Torvalds unsigned long *newest_mount_id) 2016bd4c625cSLinus Torvalds { 20171da177e4SLinus Torvalds struct reiserfs_journal_desc *desc; 20181da177e4SLinus Torvalds struct reiserfs_journal_commit *commit; 20191da177e4SLinus Torvalds struct buffer_head *c_bh; 20201da177e4SLinus Torvalds unsigned long offset; 20211da177e4SLinus Torvalds 20221da177e4SLinus Torvalds if (!d_bh) 20231da177e4SLinus Torvalds return 0; 20241da177e4SLinus Torvalds 20251da177e4SLinus Torvalds desc = (struct reiserfs_journal_desc *)d_bh->b_data; 2026bd4c625cSLinus Torvalds if (get_desc_trans_len(desc) > 0 2027bd4c625cSLinus Torvalds && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) { 2028bd4c625cSLinus Torvalds if (oldest_invalid_trans_id && *oldest_invalid_trans_id 2029bd4c625cSLinus Torvalds && get_desc_trans_id(desc) > *oldest_invalid_trans_id) { 2030a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2031bd4c625cSLinus Torvalds "journal-986: transaction " 20321da177e4SLinus Torvalds "is valid returning because trans_id %d is greater than " 2033bd4c625cSLinus Torvalds "oldest_invalid %lu", 2034bd4c625cSLinus Torvalds get_desc_trans_id(desc), 20351da177e4SLinus Torvalds *oldest_invalid_trans_id); 20361da177e4SLinus Torvalds return 0; 20371da177e4SLinus Torvalds } 2038bd4c625cSLinus Torvalds if (newest_mount_id 2039bd4c625cSLinus Torvalds && *newest_mount_id > get_desc_mount_id(desc)) { 2040a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2041bd4c625cSLinus Torvalds "journal-1087: transaction " 20421da177e4SLinus Torvalds "is valid returning because mount_id %d is less than " 2043bd4c625cSLinus Torvalds "newest_mount_id %lu", 2044bd4c625cSLinus Torvalds get_desc_mount_id(desc), 20451da177e4SLinus Torvalds *newest_mount_id); 20461da177e4SLinus Torvalds return -1; 20471da177e4SLinus Torvalds } 2048a9dd3643SJeff Mahoney if (get_desc_trans_len(desc) > SB_JOURNAL(sb)->j_trans_max) { 2049a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-2018", 205045b03d5eSJeff Mahoney "Bad transaction length %d " 205145b03d5eSJeff Mahoney "encountered, ignoring transaction", 2052bd4c625cSLinus Torvalds get_desc_trans_len(desc)); 20531da177e4SLinus Torvalds return -1; 20541da177e4SLinus Torvalds } 2055a9dd3643SJeff Mahoney offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb); 20561da177e4SLinus Torvalds 2057098297b2SJeff Mahoney /* 2058098297b2SJeff Mahoney * ok, we have a journal description block, 2059098297b2SJeff Mahoney * let's see if the transaction was valid 2060098297b2SJeff Mahoney */ 2061bd4c625cSLinus Torvalds c_bh = 2062a9dd3643SJeff Mahoney journal_bread(sb, 2063a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2064bd4c625cSLinus Torvalds ((offset + get_desc_trans_len(desc) + 2065a9dd3643SJeff Mahoney 1) % SB_ONDISK_JOURNAL_SIZE(sb))); 20661da177e4SLinus Torvalds if (!c_bh) 20671da177e4SLinus Torvalds return 0; 20681da177e4SLinus Torvalds commit = (struct reiserfs_journal_commit *)c_bh->b_data; 2069a9dd3643SJeff Mahoney if (journal_compare_desc_commit(sb, desc, commit)) { 2070a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 20711da177e4SLinus Torvalds "journal_transaction_is_valid, commit offset %ld had bad " 20721da177e4SLinus Torvalds "time %d or length %d", 2073bd4c625cSLinus Torvalds c_bh->b_blocknr - 2074a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb), 20751da177e4SLinus Torvalds get_commit_trans_id(commit), 20761da177e4SLinus Torvalds get_commit_trans_len(commit)); 20771da177e4SLinus Torvalds brelse(c_bh); 20781da177e4SLinus Torvalds if (oldest_invalid_trans_id) { 2079bd4c625cSLinus Torvalds *oldest_invalid_trans_id = 2080bd4c625cSLinus Torvalds get_desc_trans_id(desc); 2081a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2082bd4c625cSLinus Torvalds "journal-1004: " 20831da177e4SLinus Torvalds "transaction_is_valid setting oldest invalid trans_id " 2084bd4c625cSLinus Torvalds "to %d", 2085bd4c625cSLinus Torvalds get_desc_trans_id(desc)); 20861da177e4SLinus Torvalds } 20871da177e4SLinus Torvalds return -1; 20881da177e4SLinus Torvalds } 20891da177e4SLinus Torvalds brelse(c_bh); 2090a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2091bd4c625cSLinus Torvalds "journal-1006: found valid " 20921da177e4SLinus Torvalds "transaction start offset %llu, len %d id %d", 2093bd4c625cSLinus Torvalds d_bh->b_blocknr - 2094a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb), 2095bd4c625cSLinus Torvalds get_desc_trans_len(desc), 2096bd4c625cSLinus Torvalds get_desc_trans_id(desc)); 20971da177e4SLinus Torvalds return 1; 20981da177e4SLinus Torvalds } else { 20991da177e4SLinus Torvalds return 0; 21001da177e4SLinus Torvalds } 21011da177e4SLinus Torvalds } 21021da177e4SLinus Torvalds 2103bd4c625cSLinus Torvalds static void brelse_array(struct buffer_head **heads, int num) 2104bd4c625cSLinus Torvalds { 21051da177e4SLinus Torvalds int i; 21061da177e4SLinus Torvalds for (i = 0; i < num; i++) { 21071da177e4SLinus Torvalds brelse(heads[i]); 21081da177e4SLinus Torvalds } 21091da177e4SLinus Torvalds } 21101da177e4SLinus Torvalds 21111da177e4SLinus Torvalds /* 2112098297b2SJeff Mahoney * given the start, and values for the oldest acceptable transactions, 2113098297b2SJeff Mahoney * this either reads in a replays a transaction, or returns because the 2114098297b2SJeff Mahoney * transaction is invalid, or too old. 2115098297b2SJeff Mahoney * NOTE: only called during fs mount 21161da177e4SLinus Torvalds */ 2117a9dd3643SJeff Mahoney static int journal_read_transaction(struct super_block *sb, 2118bd4c625cSLinus Torvalds unsigned long cur_dblock, 2119bd4c625cSLinus Torvalds unsigned long oldest_start, 2120600ed416SJeff Mahoney unsigned int oldest_trans_id, 2121bd4c625cSLinus Torvalds unsigned long newest_mount_id) 2122bd4c625cSLinus Torvalds { 2123a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 21241da177e4SLinus Torvalds struct reiserfs_journal_desc *desc; 21251da177e4SLinus Torvalds struct reiserfs_journal_commit *commit; 2126600ed416SJeff Mahoney unsigned int trans_id = 0; 21271da177e4SLinus Torvalds struct buffer_head *c_bh; 21281da177e4SLinus Torvalds struct buffer_head *d_bh; 21291da177e4SLinus Torvalds struct buffer_head **log_blocks = NULL; 21301da177e4SLinus Torvalds struct buffer_head **real_blocks = NULL; 2131600ed416SJeff Mahoney unsigned int trans_offset; 21321da177e4SLinus Torvalds int i; 21331da177e4SLinus Torvalds int trans_half; 21341da177e4SLinus Torvalds 2135a9dd3643SJeff Mahoney d_bh = journal_bread(sb, cur_dblock); 21361da177e4SLinus Torvalds if (!d_bh) 21371da177e4SLinus Torvalds return 1; 21381da177e4SLinus Torvalds desc = (struct reiserfs_journal_desc *)d_bh->b_data; 2139a9dd3643SJeff Mahoney trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb); 2140a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1037: " 21411da177e4SLinus Torvalds "journal_read_transaction, offset %llu, len %d mount_id %d", 2142a9dd3643SJeff Mahoney d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb), 21431da177e4SLinus Torvalds get_desc_trans_len(desc), get_desc_mount_id(desc)); 21441da177e4SLinus Torvalds if (get_desc_trans_id(desc) < oldest_trans_id) { 2145a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1039: " 21461da177e4SLinus Torvalds "journal_read_trans skipping because %lu is too old", 2147bd4c625cSLinus Torvalds cur_dblock - 2148a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb)); 21491da177e4SLinus Torvalds brelse(d_bh); 21501da177e4SLinus Torvalds return 1; 21511da177e4SLinus Torvalds } 21521da177e4SLinus Torvalds if (get_desc_mount_id(desc) != newest_mount_id) { 2153a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1146: " 21541da177e4SLinus Torvalds "journal_read_trans skipping because %d is != " 21551da177e4SLinus Torvalds "newest_mount_id %lu", get_desc_mount_id(desc), 21561da177e4SLinus Torvalds newest_mount_id); 21571da177e4SLinus Torvalds brelse(d_bh); 21581da177e4SLinus Torvalds return 1; 21591da177e4SLinus Torvalds } 2160a9dd3643SJeff Mahoney c_bh = journal_bread(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 21611da177e4SLinus Torvalds ((trans_offset + get_desc_trans_len(desc) + 1) % 2162a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb))); 21631da177e4SLinus Torvalds if (!c_bh) { 21641da177e4SLinus Torvalds brelse(d_bh); 21651da177e4SLinus Torvalds return 1; 21661da177e4SLinus Torvalds } 21671da177e4SLinus Torvalds commit = (struct reiserfs_journal_commit *)c_bh->b_data; 2168a9dd3643SJeff Mahoney if (journal_compare_desc_commit(sb, desc, commit)) { 2169a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2170bd4c625cSLinus Torvalds "journal_read_transaction, " 21711da177e4SLinus Torvalds "commit offset %llu had bad time %d or length %d", 2172bd4c625cSLinus Torvalds c_bh->b_blocknr - 2173a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb), 2174bd4c625cSLinus Torvalds get_commit_trans_id(commit), 2175bd4c625cSLinus Torvalds get_commit_trans_len(commit)); 21761da177e4SLinus Torvalds brelse(c_bh); 21771da177e4SLinus Torvalds brelse(d_bh); 21781da177e4SLinus Torvalds return 1; 21791da177e4SLinus Torvalds } 21803f8b5ee3SJeff Mahoney 21813f8b5ee3SJeff Mahoney if (bdev_read_only(sb->s_bdev)) { 21823f8b5ee3SJeff Mahoney reiserfs_warning(sb, "clm-2076", 21833f8b5ee3SJeff Mahoney "device is readonly, unable to replay log"); 21843f8b5ee3SJeff Mahoney brelse(c_bh); 21853f8b5ee3SJeff Mahoney brelse(d_bh); 21863f8b5ee3SJeff Mahoney return -EROFS; 21873f8b5ee3SJeff Mahoney } 21883f8b5ee3SJeff Mahoney 21891da177e4SLinus Torvalds trans_id = get_desc_trans_id(desc); 2190098297b2SJeff Mahoney /* 2191098297b2SJeff Mahoney * now we know we've got a good transaction, and it was 2192098297b2SJeff Mahoney * inside the valid time ranges 2193098297b2SJeff Mahoney */ 2194d739b42bSPekka Enberg log_blocks = kmalloc(get_desc_trans_len(desc) * 2195d739b42bSPekka Enberg sizeof(struct buffer_head *), GFP_NOFS); 2196d739b42bSPekka Enberg real_blocks = kmalloc(get_desc_trans_len(desc) * 2197d739b42bSPekka Enberg sizeof(struct buffer_head *), GFP_NOFS); 21981da177e4SLinus Torvalds if (!log_blocks || !real_blocks) { 21991da177e4SLinus Torvalds brelse(c_bh); 22001da177e4SLinus Torvalds brelse(d_bh); 2201d739b42bSPekka Enberg kfree(log_blocks); 2202d739b42bSPekka Enberg kfree(real_blocks); 2203a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1169", 220445b03d5eSJeff Mahoney "kmalloc failed, unable to mount FS"); 22051da177e4SLinus Torvalds return -1; 22061da177e4SLinus Torvalds } 22071da177e4SLinus Torvalds /* get all the buffer heads */ 2208a9dd3643SJeff Mahoney trans_half = journal_trans_half(sb->s_blocksize); 22091da177e4SLinus Torvalds for (i = 0; i < get_desc_trans_len(desc); i++) { 2210bd4c625cSLinus Torvalds log_blocks[i] = 2211a9dd3643SJeff Mahoney journal_getblk(sb, 2212a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2213bd4c625cSLinus Torvalds (trans_offset + 1 + 2214a9dd3643SJeff Mahoney i) % SB_ONDISK_JOURNAL_SIZE(sb)); 22151da177e4SLinus Torvalds if (i < trans_half) { 2216bd4c625cSLinus Torvalds real_blocks[i] = 2217a9dd3643SJeff Mahoney sb_getblk(sb, 2218bd4c625cSLinus Torvalds le32_to_cpu(desc->j_realblock[i])); 22191da177e4SLinus Torvalds } else { 2220bd4c625cSLinus Torvalds real_blocks[i] = 2221a9dd3643SJeff Mahoney sb_getblk(sb, 2222bd4c625cSLinus Torvalds le32_to_cpu(commit-> 2223bd4c625cSLinus Torvalds j_realblock[i - trans_half])); 22241da177e4SLinus Torvalds } 2225a9dd3643SJeff Mahoney if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(sb)) { 2226a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1207", 222745b03d5eSJeff Mahoney "REPLAY FAILURE fsck required! " 222845b03d5eSJeff Mahoney "Block to replay is outside of " 222945b03d5eSJeff Mahoney "filesystem"); 22301da177e4SLinus Torvalds goto abort_replay; 22311da177e4SLinus Torvalds } 22321da177e4SLinus Torvalds /* make sure we don't try to replay onto log or reserved area */ 2233bd4c625cSLinus Torvalds if (is_block_in_log_or_reserved_area 2234a9dd3643SJeff Mahoney (sb, real_blocks[i]->b_blocknr)) { 2235a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1204", 223645b03d5eSJeff Mahoney "REPLAY FAILURE fsck required! " 223745b03d5eSJeff Mahoney "Trying to replay onto a log block"); 22381da177e4SLinus Torvalds abort_replay: 22391da177e4SLinus Torvalds brelse_array(log_blocks, i); 22401da177e4SLinus Torvalds brelse_array(real_blocks, i); 22411da177e4SLinus Torvalds brelse(c_bh); 22421da177e4SLinus Torvalds brelse(d_bh); 2243d739b42bSPekka Enberg kfree(log_blocks); 2244d739b42bSPekka Enberg kfree(real_blocks); 22451da177e4SLinus Torvalds return -1; 22461da177e4SLinus Torvalds } 22471da177e4SLinus Torvalds } 22481da177e4SLinus Torvalds /* read in the log blocks, memcpy to the corresponding real block */ 2249dfec8a14SMike Christie ll_rw_block(REQ_OP_READ, 0, get_desc_trans_len(desc), log_blocks); 22501da177e4SLinus Torvalds for (i = 0; i < get_desc_trans_len(desc); i++) { 22518ebc4232SFrederic Weisbecker 22521da177e4SLinus Torvalds wait_on_buffer(log_blocks[i]); 22531da177e4SLinus Torvalds if (!buffer_uptodate(log_blocks[i])) { 2254a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1212", 225545b03d5eSJeff Mahoney "REPLAY FAILURE fsck required! " 225645b03d5eSJeff Mahoney "buffer write failed"); 2257bd4c625cSLinus Torvalds brelse_array(log_blocks + i, 2258bd4c625cSLinus Torvalds get_desc_trans_len(desc) - i); 22591da177e4SLinus Torvalds brelse_array(real_blocks, get_desc_trans_len(desc)); 22601da177e4SLinus Torvalds brelse(c_bh); 22611da177e4SLinus Torvalds brelse(d_bh); 2262d739b42bSPekka Enberg kfree(log_blocks); 2263d739b42bSPekka Enberg kfree(real_blocks); 22641da177e4SLinus Torvalds return -1; 22651da177e4SLinus Torvalds } 2266bd4c625cSLinus Torvalds memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data, 2267bd4c625cSLinus Torvalds real_blocks[i]->b_size); 22681da177e4SLinus Torvalds set_buffer_uptodate(real_blocks[i]); 22691da177e4SLinus Torvalds brelse(log_blocks[i]); 22701da177e4SLinus Torvalds } 22711da177e4SLinus Torvalds /* flush out the real blocks */ 22721da177e4SLinus Torvalds for (i = 0; i < get_desc_trans_len(desc); i++) { 22731da177e4SLinus Torvalds set_buffer_dirty(real_blocks[i]); 22742a222ca9SMike Christie write_dirty_buffer(real_blocks[i], 0); 22751da177e4SLinus Torvalds } 22761da177e4SLinus Torvalds for (i = 0; i < get_desc_trans_len(desc); i++) { 22771da177e4SLinus Torvalds wait_on_buffer(real_blocks[i]); 22781da177e4SLinus Torvalds if (!buffer_uptodate(real_blocks[i])) { 2279a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1226", 228045b03d5eSJeff Mahoney "REPLAY FAILURE, fsck required! " 228145b03d5eSJeff Mahoney "buffer write failed"); 2282bd4c625cSLinus Torvalds brelse_array(real_blocks + i, 2283bd4c625cSLinus Torvalds get_desc_trans_len(desc) - i); 22841da177e4SLinus Torvalds brelse(c_bh); 22851da177e4SLinus Torvalds brelse(d_bh); 2286d739b42bSPekka Enberg kfree(log_blocks); 2287d739b42bSPekka Enberg kfree(real_blocks); 22881da177e4SLinus Torvalds return -1; 22891da177e4SLinus Torvalds } 22901da177e4SLinus Torvalds brelse(real_blocks[i]); 22911da177e4SLinus Torvalds } 2292bd4c625cSLinus Torvalds cur_dblock = 2293a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2294bd4c625cSLinus Torvalds ((trans_offset + get_desc_trans_len(desc) + 2295a9dd3643SJeff Mahoney 2) % SB_ONDISK_JOURNAL_SIZE(sb)); 2296a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2297bd4c625cSLinus Torvalds "journal-1095: setting journal " "start to offset %ld", 2298a9dd3643SJeff Mahoney cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb)); 22991da177e4SLinus Torvalds 2300098297b2SJeff Mahoney /* 2301098297b2SJeff Mahoney * init starting values for the first transaction, in case 2302098297b2SJeff Mahoney * this is the last transaction to be replayed. 2303098297b2SJeff Mahoney */ 2304a9dd3643SJeff Mahoney journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb); 23051da177e4SLinus Torvalds journal->j_last_flush_trans_id = trans_id; 23061da177e4SLinus Torvalds journal->j_trans_id = trans_id + 1; 2307a44c94a7SAlexander Zarochentsev /* check for trans_id overflow */ 2308a44c94a7SAlexander Zarochentsev if (journal->j_trans_id == 0) 2309a44c94a7SAlexander Zarochentsev journal->j_trans_id = 10; 23101da177e4SLinus Torvalds brelse(c_bh); 23111da177e4SLinus Torvalds brelse(d_bh); 2312d739b42bSPekka Enberg kfree(log_blocks); 2313d739b42bSPekka Enberg kfree(real_blocks); 23141da177e4SLinus Torvalds return 0; 23151da177e4SLinus Torvalds } 23161da177e4SLinus Torvalds 2317098297b2SJeff Mahoney /* 2318098297b2SJeff Mahoney * This function reads blocks starting from block and to max_block of bufsize 2319098297b2SJeff Mahoney * size (but no more than BUFNR blocks at a time). This proved to improve 2320098297b2SJeff Mahoney * mounting speed on self-rebuilding raid5 arrays at least. 2321098297b2SJeff Mahoney * Right now it is only used from journal code. But later we might use it 2322098297b2SJeff Mahoney * from other places. 2323098297b2SJeff Mahoney * Note: Do not use journal_getblk/sb_getblk functions here! 2324098297b2SJeff Mahoney */ 23253ee16670SJeff Mahoney static struct buffer_head *reiserfs_breada(struct block_device *dev, 23263ee16670SJeff Mahoney b_blocknr_t block, int bufsize, 23273ee16670SJeff Mahoney b_blocknr_t max_block) 23281da177e4SLinus Torvalds { 23291da177e4SLinus Torvalds struct buffer_head *bhlist[BUFNR]; 23301da177e4SLinus Torvalds unsigned int blocks = BUFNR; 23311da177e4SLinus Torvalds struct buffer_head *bh; 23321da177e4SLinus Torvalds int i, j; 23331da177e4SLinus Torvalds 23341da177e4SLinus Torvalds bh = __getblk(dev, block, bufsize); 23351da177e4SLinus Torvalds if (buffer_uptodate(bh)) 23361da177e4SLinus Torvalds return (bh); 23371da177e4SLinus Torvalds 23381da177e4SLinus Torvalds if (block + BUFNR > max_block) { 23391da177e4SLinus Torvalds blocks = max_block - block; 23401da177e4SLinus Torvalds } 23411da177e4SLinus Torvalds bhlist[0] = bh; 23421da177e4SLinus Torvalds j = 1; 23431da177e4SLinus Torvalds for (i = 1; i < blocks; i++) { 23441da177e4SLinus Torvalds bh = __getblk(dev, block + i, bufsize); 23451da177e4SLinus Torvalds if (buffer_uptodate(bh)) { 23461da177e4SLinus Torvalds brelse(bh); 23471da177e4SLinus Torvalds break; 2348bd4c625cSLinus Torvalds } else 2349bd4c625cSLinus Torvalds bhlist[j++] = bh; 23501da177e4SLinus Torvalds } 2351dfec8a14SMike Christie ll_rw_block(REQ_OP_READ, 0, j, bhlist); 23521da177e4SLinus Torvalds for (i = 1; i < j; i++) 23531da177e4SLinus Torvalds brelse(bhlist[i]); 23541da177e4SLinus Torvalds bh = bhlist[0]; 23551da177e4SLinus Torvalds wait_on_buffer(bh); 23561da177e4SLinus Torvalds if (buffer_uptodate(bh)) 23571da177e4SLinus Torvalds return bh; 23581da177e4SLinus Torvalds brelse(bh); 23591da177e4SLinus Torvalds return NULL; 23601da177e4SLinus Torvalds } 23611da177e4SLinus Torvalds 23621da177e4SLinus Torvalds /* 2363098297b2SJeff Mahoney * read and replay the log 2364098297b2SJeff Mahoney * on a clean unmount, the journal header's next unflushed pointer will be 2365098297b2SJeff Mahoney * to an invalid transaction. This tests that before finding all the 2366098297b2SJeff Mahoney * transactions in the log, which makes normal mount times fast. 2367098297b2SJeff Mahoney * 2368098297b2SJeff Mahoney * After a crash, this starts with the next unflushed transaction, and 2369098297b2SJeff Mahoney * replays until it finds one too old, or invalid. 2370098297b2SJeff Mahoney * 2371098297b2SJeff Mahoney * On exit, it sets things up so the first transaction will work correctly. 2372098297b2SJeff Mahoney * NOTE: only called during fs mount 23731da177e4SLinus Torvalds */ 2374a9dd3643SJeff Mahoney static int journal_read(struct super_block *sb) 2375bd4c625cSLinus Torvalds { 2376a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 23771da177e4SLinus Torvalds struct reiserfs_journal_desc *desc; 2378600ed416SJeff Mahoney unsigned int oldest_trans_id = 0; 2379600ed416SJeff Mahoney unsigned int oldest_invalid_trans_id = 0; 23801da177e4SLinus Torvalds time_t start; 23811da177e4SLinus Torvalds unsigned long oldest_start = 0; 23821da177e4SLinus Torvalds unsigned long cur_dblock = 0; 23831da177e4SLinus Torvalds unsigned long newest_mount_id = 9; 23841da177e4SLinus Torvalds struct buffer_head *d_bh; 23851da177e4SLinus Torvalds struct reiserfs_journal_header *jh; 23861da177e4SLinus Torvalds int valid_journal_header = 0; 23871da177e4SLinus Torvalds int replay_count = 0; 23881da177e4SLinus Torvalds int continue_replay = 1; 23891da177e4SLinus Torvalds int ret; 23901da177e4SLinus Torvalds 2391a9dd3643SJeff Mahoney cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(sb); 2392a1c6f057SDmitry Monakhov reiserfs_info(sb, "checking transaction log (%pg)\n", 2393a1c6f057SDmitry Monakhov journal->j_dev_bd); 23941da177e4SLinus Torvalds start = get_seconds(); 23951da177e4SLinus Torvalds 2396098297b2SJeff Mahoney /* 2397098297b2SJeff Mahoney * step 1, read in the journal header block. Check the transaction 2398098297b2SJeff Mahoney * it says is the first unflushed, and if that transaction is not 2399098297b2SJeff Mahoney * valid, replay is done 24001da177e4SLinus Torvalds */ 2401a9dd3643SJeff Mahoney journal->j_header_bh = journal_bread(sb, 2402a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) 2403a9dd3643SJeff Mahoney + SB_ONDISK_JOURNAL_SIZE(sb)); 24041da177e4SLinus Torvalds if (!journal->j_header_bh) { 24051da177e4SLinus Torvalds return 1; 24061da177e4SLinus Torvalds } 24071da177e4SLinus Torvalds jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data); 2408c499ec24SVladimir V. Saveliev if (le32_to_cpu(jh->j_first_unflushed_offset) < 2409a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb) 2410bd4c625cSLinus Torvalds && le32_to_cpu(jh->j_last_flush_trans_id) > 0) { 2411bd4c625cSLinus Torvalds oldest_start = 2412a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 24131da177e4SLinus Torvalds le32_to_cpu(jh->j_first_unflushed_offset); 24141da177e4SLinus Torvalds oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1; 24151da177e4SLinus Torvalds newest_mount_id = le32_to_cpu(jh->j_mount_id); 2416a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2417bd4c625cSLinus Torvalds "journal-1153: found in " 24181da177e4SLinus Torvalds "header: first_unflushed_offset %d, last_flushed_trans_id " 24191da177e4SLinus Torvalds "%lu", le32_to_cpu(jh->j_first_unflushed_offset), 24201da177e4SLinus Torvalds le32_to_cpu(jh->j_last_flush_trans_id)); 24211da177e4SLinus Torvalds valid_journal_header = 1; 24221da177e4SLinus Torvalds 2423098297b2SJeff Mahoney /* 2424098297b2SJeff Mahoney * now, we try to read the first unflushed offset. If it 2425098297b2SJeff Mahoney * is not valid, there is nothing more we can do, and it 2426098297b2SJeff Mahoney * makes no sense to read through the whole log. 24271da177e4SLinus Torvalds */ 2428bd4c625cSLinus Torvalds d_bh = 2429a9dd3643SJeff Mahoney journal_bread(sb, 2430a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2431bd4c625cSLinus Torvalds le32_to_cpu(jh->j_first_unflushed_offset)); 2432a9dd3643SJeff Mahoney ret = journal_transaction_is_valid(sb, d_bh, NULL, NULL); 24331da177e4SLinus Torvalds if (!ret) { 24341da177e4SLinus Torvalds continue_replay = 0; 24351da177e4SLinus Torvalds } 24361da177e4SLinus Torvalds brelse(d_bh); 24371da177e4SLinus Torvalds goto start_log_replay; 24381da177e4SLinus Torvalds } 24391da177e4SLinus Torvalds 2440098297b2SJeff Mahoney /* 2441098297b2SJeff Mahoney * ok, there are transactions that need to be replayed. start 2442098297b2SJeff Mahoney * with the first log block, find all the valid transactions, and 2443098297b2SJeff Mahoney * pick out the oldest. 24441da177e4SLinus Torvalds */ 2445bd4c625cSLinus Torvalds while (continue_replay 2446bd4c625cSLinus Torvalds && cur_dblock < 2447a9dd3643SJeff Mahoney (SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2448a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb))) { 2449098297b2SJeff Mahoney /* 2450098297b2SJeff Mahoney * Note that it is required for blocksize of primary fs 2451098297b2SJeff Mahoney * device and journal device to be the same 2452098297b2SJeff Mahoney */ 2453bd4c625cSLinus Torvalds d_bh = 2454bd4c625cSLinus Torvalds reiserfs_breada(journal->j_dev_bd, cur_dblock, 2455a9dd3643SJeff Mahoney sb->s_blocksize, 2456a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2457a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb)); 2458bd4c625cSLinus Torvalds ret = 2459a9dd3643SJeff Mahoney journal_transaction_is_valid(sb, d_bh, 2460bd4c625cSLinus Torvalds &oldest_invalid_trans_id, 2461bd4c625cSLinus Torvalds &newest_mount_id); 24621da177e4SLinus Torvalds if (ret == 1) { 24631da177e4SLinus Torvalds desc = (struct reiserfs_journal_desc *)d_bh->b_data; 24641da177e4SLinus Torvalds if (oldest_start == 0) { /* init all oldest_ values */ 24651da177e4SLinus Torvalds oldest_trans_id = get_desc_trans_id(desc); 24661da177e4SLinus Torvalds oldest_start = d_bh->b_blocknr; 24671da177e4SLinus Torvalds newest_mount_id = get_desc_mount_id(desc); 2468a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2469bd4c625cSLinus Torvalds "journal-1179: Setting " 24701da177e4SLinus Torvalds "oldest_start to offset %llu, trans_id %lu", 2471bd4c625cSLinus Torvalds oldest_start - 2472bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK 2473a9dd3643SJeff Mahoney (sb), oldest_trans_id); 24741da177e4SLinus Torvalds } else if (oldest_trans_id > get_desc_trans_id(desc)) { 24751da177e4SLinus Torvalds /* one we just read was older */ 24761da177e4SLinus Torvalds oldest_trans_id = get_desc_trans_id(desc); 24771da177e4SLinus Torvalds oldest_start = d_bh->b_blocknr; 2478a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2479bd4c625cSLinus Torvalds "journal-1180: Resetting " 24801da177e4SLinus Torvalds "oldest_start to offset %lu, trans_id %lu", 2481bd4c625cSLinus Torvalds oldest_start - 2482bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK 2483a9dd3643SJeff Mahoney (sb), oldest_trans_id); 24841da177e4SLinus Torvalds } 24851da177e4SLinus Torvalds if (newest_mount_id < get_desc_mount_id(desc)) { 24861da177e4SLinus Torvalds newest_mount_id = get_desc_mount_id(desc); 2487a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2488bd4c625cSLinus Torvalds "journal-1299: Setting " 2489bd4c625cSLinus Torvalds "newest_mount_id to %d", 2490bd4c625cSLinus Torvalds get_desc_mount_id(desc)); 24911da177e4SLinus Torvalds } 24921da177e4SLinus Torvalds cur_dblock += get_desc_trans_len(desc) + 2; 24931da177e4SLinus Torvalds } else { 24941da177e4SLinus Torvalds cur_dblock++; 24951da177e4SLinus Torvalds } 24961da177e4SLinus Torvalds brelse(d_bh); 24971da177e4SLinus Torvalds } 24981da177e4SLinus Torvalds 24991da177e4SLinus Torvalds start_log_replay: 25001da177e4SLinus Torvalds cur_dblock = oldest_start; 25011da177e4SLinus Torvalds if (oldest_trans_id) { 2502a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2503bd4c625cSLinus Torvalds "journal-1206: Starting replay " 25041da177e4SLinus Torvalds "from offset %llu, trans_id %lu", 2505a9dd3643SJeff Mahoney cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb), 25061da177e4SLinus Torvalds oldest_trans_id); 25071da177e4SLinus Torvalds 25081da177e4SLinus Torvalds } 25091da177e4SLinus Torvalds replay_count = 0; 25101da177e4SLinus Torvalds while (continue_replay && oldest_trans_id > 0) { 2511bd4c625cSLinus Torvalds ret = 2512a9dd3643SJeff Mahoney journal_read_transaction(sb, cur_dblock, oldest_start, 2513bd4c625cSLinus Torvalds oldest_trans_id, newest_mount_id); 25141da177e4SLinus Torvalds if (ret < 0) { 25151da177e4SLinus Torvalds return ret; 25161da177e4SLinus Torvalds } else if (ret != 0) { 25171da177e4SLinus Torvalds break; 25181da177e4SLinus Torvalds } 2519bd4c625cSLinus Torvalds cur_dblock = 2520a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + journal->j_start; 25211da177e4SLinus Torvalds replay_count++; 25221da177e4SLinus Torvalds if (cur_dblock == oldest_start) 25231da177e4SLinus Torvalds break; 25241da177e4SLinus Torvalds } 25251da177e4SLinus Torvalds 25261da177e4SLinus Torvalds if (oldest_trans_id == 0) { 2527a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2528bd4c625cSLinus Torvalds "journal-1225: No valid " "transactions found"); 25291da177e4SLinus Torvalds } 2530098297b2SJeff Mahoney /* 2531098297b2SJeff Mahoney * j_start does not get set correctly if we don't replay any 2532098297b2SJeff Mahoney * transactions. if we had a valid journal_header, set j_start 2533098297b2SJeff Mahoney * to the first unflushed transaction value, copy the trans_id 2534098297b2SJeff Mahoney * from the header 25351da177e4SLinus Torvalds */ 25361da177e4SLinus Torvalds if (valid_journal_header && replay_count == 0) { 25371da177e4SLinus Torvalds journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset); 2538bd4c625cSLinus Torvalds journal->j_trans_id = 2539bd4c625cSLinus Torvalds le32_to_cpu(jh->j_last_flush_trans_id) + 1; 2540a44c94a7SAlexander Zarochentsev /* check for trans_id overflow */ 2541a44c94a7SAlexander Zarochentsev if (journal->j_trans_id == 0) 2542a44c94a7SAlexander Zarochentsev journal->j_trans_id = 10; 2543bd4c625cSLinus Torvalds journal->j_last_flush_trans_id = 2544bd4c625cSLinus Torvalds le32_to_cpu(jh->j_last_flush_trans_id); 25451da177e4SLinus Torvalds journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1; 25461da177e4SLinus Torvalds } else { 25471da177e4SLinus Torvalds journal->j_mount_id = newest_mount_id + 1; 25481da177e4SLinus Torvalds } 2549a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1299: Setting " 25501da177e4SLinus Torvalds "newest_mount_id to %lu", journal->j_mount_id); 25511da177e4SLinus Torvalds journal->j_first_unflushed_offset = journal->j_start; 25521da177e4SLinus Torvalds if (replay_count > 0) { 2553a9dd3643SJeff Mahoney reiserfs_info(sb, 2554bd4c625cSLinus Torvalds "replayed %d transactions in %lu seconds\n", 25551da177e4SLinus Torvalds replay_count, get_seconds() - start); 25561da177e4SLinus Torvalds } 2557278f6679SJeff Mahoney /* needed to satisfy the locking in _update_journal_header_block */ 2558278f6679SJeff Mahoney reiserfs_write_lock(sb); 2559a9dd3643SJeff Mahoney if (!bdev_read_only(sb->s_bdev) && 2560a9dd3643SJeff Mahoney _update_journal_header_block(sb, journal->j_start, 2561bd4c625cSLinus Torvalds journal->j_last_flush_trans_id)) { 2562278f6679SJeff Mahoney reiserfs_write_unlock(sb); 2563098297b2SJeff Mahoney /* 2564098297b2SJeff Mahoney * replay failed, caller must call free_journal_ram and abort 2565098297b2SJeff Mahoney * the mount 25661da177e4SLinus Torvalds */ 25671da177e4SLinus Torvalds return -1; 25681da177e4SLinus Torvalds } 2569278f6679SJeff Mahoney reiserfs_write_unlock(sb); 25701da177e4SLinus Torvalds return 0; 25711da177e4SLinus Torvalds } 25721da177e4SLinus Torvalds 25731da177e4SLinus Torvalds static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s) 25741da177e4SLinus Torvalds { 25751da177e4SLinus Torvalds struct reiserfs_journal_list *jl; 25768c777cc4SPekka Enberg jl = kzalloc(sizeof(struct reiserfs_journal_list), 25778c777cc4SPekka Enberg GFP_NOFS | __GFP_NOFAIL); 25781da177e4SLinus Torvalds INIT_LIST_HEAD(&jl->j_list); 25791da177e4SLinus Torvalds INIT_LIST_HEAD(&jl->j_working_list); 25801da177e4SLinus Torvalds INIT_LIST_HEAD(&jl->j_tail_bh_list); 25811da177e4SLinus Torvalds INIT_LIST_HEAD(&jl->j_bh_list); 258290415deaSJeff Mahoney mutex_init(&jl->j_commit_mutex); 25831da177e4SLinus Torvalds SB_JOURNAL(s)->j_num_lists++; 25841da177e4SLinus Torvalds get_journal_list(jl); 25851da177e4SLinus Torvalds return jl; 25861da177e4SLinus Torvalds } 25871da177e4SLinus Torvalds 2588a9dd3643SJeff Mahoney static void journal_list_init(struct super_block *sb) 2589bd4c625cSLinus Torvalds { 2590a9dd3643SJeff Mahoney SB_JOURNAL(sb)->j_current_jl = alloc_journal_list(sb); 25911da177e4SLinus Torvalds } 25921da177e4SLinus Torvalds 25934385bab1SAl Viro static void release_journal_dev(struct super_block *super, 25941da177e4SLinus Torvalds struct reiserfs_journal *journal) 25951da177e4SLinus Torvalds { 259686098fa0SChristoph Hellwig if (journal->j_dev_bd != NULL) { 25974385bab1SAl Viro blkdev_put(journal->j_dev_bd, journal->j_dev_mode); 25981da177e4SLinus Torvalds journal->j_dev_bd = NULL; 25991da177e4SLinus Torvalds } 26001da177e4SLinus Torvalds } 26011da177e4SLinus Torvalds 26021da177e4SLinus Torvalds static int journal_init_dev(struct super_block *super, 26031da177e4SLinus Torvalds struct reiserfs_journal *journal, 26041da177e4SLinus Torvalds const char *jdev_name) 26051da177e4SLinus Torvalds { 26061da177e4SLinus Torvalds int result; 26071da177e4SLinus Torvalds dev_t jdev; 2608e525fd89STejun Heo fmode_t blkdev_mode = FMODE_READ | FMODE_WRITE | FMODE_EXCL; 26091da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 26101da177e4SLinus Torvalds 26111da177e4SLinus Torvalds result = 0; 26121da177e4SLinus Torvalds 26131da177e4SLinus Torvalds journal->j_dev_bd = NULL; 26141da177e4SLinus Torvalds jdev = SB_ONDISK_JOURNAL_DEVICE(super) ? 26151da177e4SLinus Torvalds new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev; 26161da177e4SLinus Torvalds 26171da177e4SLinus Torvalds if (bdev_read_only(super->s_bdev)) 26181da177e4SLinus Torvalds blkdev_mode = FMODE_READ; 26191da177e4SLinus Torvalds 26201da177e4SLinus Torvalds /* there is no "jdev" option and journal is on separate device */ 26211da177e4SLinus Torvalds if ((!jdev_name || !jdev_name[0])) { 2622e525fd89STejun Heo if (jdev == super->s_dev) 2623e525fd89STejun Heo blkdev_mode &= ~FMODE_EXCL; 2624d4d77629STejun Heo journal->j_dev_bd = blkdev_get_by_dev(jdev, blkdev_mode, 2625d4d77629STejun Heo journal); 2626e5eb8caaSAl Viro journal->j_dev_mode = blkdev_mode; 26271da177e4SLinus Torvalds if (IS_ERR(journal->j_dev_bd)) { 26281da177e4SLinus Torvalds result = PTR_ERR(journal->j_dev_bd); 26291da177e4SLinus Torvalds journal->j_dev_bd = NULL; 263045b03d5eSJeff Mahoney reiserfs_warning(super, "sh-458", 26311da177e4SLinus Torvalds "cannot init journal device '%s': %i", 26321da177e4SLinus Torvalds __bdevname(jdev, b), result); 26331da177e4SLinus Torvalds return result; 2634e525fd89STejun Heo } else if (jdev != super->s_dev) 26351da177e4SLinus Torvalds set_blocksize(journal->j_dev_bd, super->s_blocksize); 263686098fa0SChristoph Hellwig 26371da177e4SLinus Torvalds return 0; 26381da177e4SLinus Torvalds } 26391da177e4SLinus Torvalds 2640e5eb8caaSAl Viro journal->j_dev_mode = blkdev_mode; 2641d4d77629STejun Heo journal->j_dev_bd = blkdev_get_by_path(jdev_name, blkdev_mode, journal); 264286098fa0SChristoph Hellwig if (IS_ERR(journal->j_dev_bd)) { 264386098fa0SChristoph Hellwig result = PTR_ERR(journal->j_dev_bd); 264486098fa0SChristoph Hellwig journal->j_dev_bd = NULL; 264586098fa0SChristoph Hellwig reiserfs_warning(super, 264686098fa0SChristoph Hellwig "journal_init_dev: Cannot open '%s': %i", 264786098fa0SChristoph Hellwig jdev_name, result); 264886098fa0SChristoph Hellwig return result; 264986098fa0SChristoph Hellwig } 265086098fa0SChristoph Hellwig 26511da177e4SLinus Torvalds set_blocksize(journal->j_dev_bd, super->s_blocksize); 2652bd4c625cSLinus Torvalds reiserfs_info(super, 2653a1c6f057SDmitry Monakhov "journal_init_dev: journal device: %pg\n", 2654a1c6f057SDmitry Monakhov journal->j_dev_bd); 265586098fa0SChristoph Hellwig return 0; 26561da177e4SLinus Torvalds } 26571da177e4SLinus Torvalds 2658098297b2SJeff Mahoney /* 2659cf3d0b81SEdward Shishkin * When creating/tuning a file system user can assign some 2660cf3d0b81SEdward Shishkin * journal params within boundaries which depend on the ratio 2661cf3d0b81SEdward Shishkin * blocksize/standard_blocksize. 2662cf3d0b81SEdward Shishkin * 2663cf3d0b81SEdward Shishkin * For blocks >= standard_blocksize transaction size should 2664cf3d0b81SEdward Shishkin * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more 2665cf3d0b81SEdward Shishkin * then JOURNAL_TRANS_MAX_DEFAULT. 2666cf3d0b81SEdward Shishkin * 2667cf3d0b81SEdward Shishkin * For blocks < standard_blocksize these boundaries should be 2668cf3d0b81SEdward Shishkin * decreased proportionally. 2669cf3d0b81SEdward Shishkin */ 2670cf3d0b81SEdward Shishkin #define REISERFS_STANDARD_BLKSIZE (4096) 2671cf3d0b81SEdward Shishkin 2672a9dd3643SJeff Mahoney static int check_advise_trans_params(struct super_block *sb, 2673cf3d0b81SEdward Shishkin struct reiserfs_journal *journal) 2674cf3d0b81SEdward Shishkin { 2675cf3d0b81SEdward Shishkin if (journal->j_trans_max) { 2676098297b2SJeff Mahoney /* Non-default journal params. Do sanity check for them. */ 2677cf3d0b81SEdward Shishkin int ratio = 1; 2678a9dd3643SJeff Mahoney if (sb->s_blocksize < REISERFS_STANDARD_BLKSIZE) 2679a9dd3643SJeff Mahoney ratio = REISERFS_STANDARD_BLKSIZE / sb->s_blocksize; 2680cf3d0b81SEdward Shishkin 2681cf3d0b81SEdward Shishkin if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio || 2682cf3d0b81SEdward Shishkin journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio || 2683a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb) / journal->j_trans_max < 2684cf3d0b81SEdward Shishkin JOURNAL_MIN_RATIO) { 2685a9dd3643SJeff Mahoney reiserfs_warning(sb, "sh-462", 268645b03d5eSJeff Mahoney "bad transaction max size (%u). " 268745b03d5eSJeff Mahoney "FSCK?", journal->j_trans_max); 2688cf3d0b81SEdward Shishkin return 1; 2689cf3d0b81SEdward Shishkin } 2690cf3d0b81SEdward Shishkin if (journal->j_max_batch != (journal->j_trans_max) * 2691cf3d0b81SEdward Shishkin JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) { 2692a9dd3643SJeff Mahoney reiserfs_warning(sb, "sh-463", 269345b03d5eSJeff Mahoney "bad transaction max batch (%u). " 269445b03d5eSJeff Mahoney "FSCK?", journal->j_max_batch); 2695cf3d0b81SEdward Shishkin return 1; 2696cf3d0b81SEdward Shishkin } 2697cf3d0b81SEdward Shishkin } else { 2698098297b2SJeff Mahoney /* 2699098297b2SJeff Mahoney * Default journal params. 2700098297b2SJeff Mahoney * The file system was created by old version 2701098297b2SJeff Mahoney * of mkreiserfs, so some fields contain zeros, 2702098297b2SJeff Mahoney * and we need to advise proper values for them 2703098297b2SJeff Mahoney */ 2704a9dd3643SJeff Mahoney if (sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) { 2705a9dd3643SJeff Mahoney reiserfs_warning(sb, "sh-464", "bad blocksize (%u)", 2706a9dd3643SJeff Mahoney sb->s_blocksize); 270745b03d5eSJeff Mahoney return 1; 270845b03d5eSJeff Mahoney } 2709cf3d0b81SEdward Shishkin journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT; 2710cf3d0b81SEdward Shishkin journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT; 2711cf3d0b81SEdward Shishkin journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE; 2712cf3d0b81SEdward Shishkin } 2713cf3d0b81SEdward Shishkin return 0; 2714cf3d0b81SEdward Shishkin } 2715cf3d0b81SEdward Shishkin 2716098297b2SJeff Mahoney /* must be called once on fs mount. calls journal_read for you */ 2717a9dd3643SJeff Mahoney int journal_init(struct super_block *sb, const char *j_dev_name, 2718bd4c625cSLinus Torvalds int old_format, unsigned int commit_max_age) 2719bd4c625cSLinus Torvalds { 2720a9dd3643SJeff Mahoney int num_cnodes = SB_ONDISK_JOURNAL_SIZE(sb) * 2; 27211da177e4SLinus Torvalds struct buffer_head *bhjh; 27221da177e4SLinus Torvalds struct reiserfs_super_block *rs; 27231da177e4SLinus Torvalds struct reiserfs_journal_header *jh; 27241da177e4SLinus Torvalds struct reiserfs_journal *journal; 27251da177e4SLinus Torvalds struct reiserfs_journal_list *jl; 272698ea3f50SFrederic Weisbecker int ret; 27271da177e4SLinus Torvalds 2728558feb08SJoe Perches journal = SB_JOURNAL(sb) = vzalloc(sizeof(struct reiserfs_journal)); 27291da177e4SLinus Torvalds if (!journal) { 2730a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1256", 273145b03d5eSJeff Mahoney "unable to get memory for journal structure"); 27321da177e4SLinus Torvalds return 1; 27331da177e4SLinus Torvalds } 27341da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_bitmap_nodes); 27351da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_prealloc_list); 27361da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_working_list); 27371da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_journal_list); 27381da177e4SLinus Torvalds journal->j_persistent_trans = 0; 273937c69b98SFrederic Weisbecker if (reiserfs_allocate_list_bitmaps(sb, journal->j_list_bitmap, 274037c69b98SFrederic Weisbecker reiserfs_bmap_count(sb))) 27411da177e4SLinus Torvalds goto free_and_return; 274298ea3f50SFrederic Weisbecker 2743a9dd3643SJeff Mahoney allocate_bitmap_nodes(sb); 27441da177e4SLinus Torvalds 27451da177e4SLinus Torvalds /* reserved for journal area support */ 2746a9dd3643SJeff Mahoney SB_JOURNAL_1st_RESERVED_BLOCK(sb) = (old_format ? 2747bd4c625cSLinus Torvalds REISERFS_OLD_DISK_OFFSET_IN_BYTES 2748a9dd3643SJeff Mahoney / sb->s_blocksize + 2749a9dd3643SJeff Mahoney reiserfs_bmap_count(sb) + 2750bd4c625cSLinus Torvalds 1 : 2751bd4c625cSLinus Torvalds REISERFS_DISK_OFFSET_IN_BYTES / 2752a9dd3643SJeff Mahoney sb->s_blocksize + 2); 27531da177e4SLinus Torvalds 2754098297b2SJeff Mahoney /* 2755098297b2SJeff Mahoney * Sanity check to see is the standard journal fitting 2756098297b2SJeff Mahoney * within first bitmap (actual for small blocksizes) 2757098297b2SJeff Mahoney */ 2758a9dd3643SJeff Mahoney if (!SB_ONDISK_JOURNAL_DEVICE(sb) && 2759a9dd3643SJeff Mahoney (SB_JOURNAL_1st_RESERVED_BLOCK(sb) + 2760a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb) > sb->s_blocksize * 8)) { 2761a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1393", 276245b03d5eSJeff Mahoney "journal does not fit for area addressed " 276345b03d5eSJeff Mahoney "by first of bitmap blocks. It starts at " 27641da177e4SLinus Torvalds "%u and its size is %u. Block size %ld", 2765a9dd3643SJeff Mahoney SB_JOURNAL_1st_RESERVED_BLOCK(sb), 2766a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb), 2767a9dd3643SJeff Mahoney sb->s_blocksize); 27681da177e4SLinus Torvalds goto free_and_return; 27691da177e4SLinus Torvalds } 27701da177e4SLinus Torvalds 2771a9dd3643SJeff Mahoney if (journal_init_dev(sb, journal, j_dev_name) != 0) { 2772a9dd3643SJeff Mahoney reiserfs_warning(sb, "sh-462", 27736774def6SMasanari Iida "unable to initialize journal device"); 27741da177e4SLinus Torvalds goto free_and_return; 27751da177e4SLinus Torvalds } 27761da177e4SLinus Torvalds 2777a9dd3643SJeff Mahoney rs = SB_DISK_SUPER_BLOCK(sb); 27781da177e4SLinus Torvalds 27791da177e4SLinus Torvalds /* read journal header */ 2780a9dd3643SJeff Mahoney bhjh = journal_bread(sb, 2781a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2782a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb)); 27831da177e4SLinus Torvalds if (!bhjh) { 2784a9dd3643SJeff Mahoney reiserfs_warning(sb, "sh-459", 278545b03d5eSJeff Mahoney "unable to read journal header"); 27861da177e4SLinus Torvalds goto free_and_return; 27871da177e4SLinus Torvalds } 27881da177e4SLinus Torvalds jh = (struct reiserfs_journal_header *)(bhjh->b_data); 27891da177e4SLinus Torvalds 27901da177e4SLinus Torvalds /* make sure that journal matches to the super block */ 2791bd4c625cSLinus Torvalds if (is_reiserfs_jr(rs) 2792bd4c625cSLinus Torvalds && (le32_to_cpu(jh->jh_journal.jp_journal_magic) != 2793bd4c625cSLinus Torvalds sb_jp_journal_magic(rs))) { 2794a9dd3643SJeff Mahoney reiserfs_warning(sb, "sh-460", 2795a1c6f057SDmitry Monakhov "journal header magic %x (device %pg) does " 279645b03d5eSJeff Mahoney "not match to magic found in super block %x", 279745b03d5eSJeff Mahoney jh->jh_journal.jp_journal_magic, 2798a1c6f057SDmitry Monakhov journal->j_dev_bd, 27991da177e4SLinus Torvalds sb_jp_journal_magic(rs)); 28001da177e4SLinus Torvalds brelse(bhjh); 28011da177e4SLinus Torvalds goto free_and_return; 28021da177e4SLinus Torvalds } 28031da177e4SLinus Torvalds 28041da177e4SLinus Torvalds journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max); 28051da177e4SLinus Torvalds journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch); 2806bd4c625cSLinus Torvalds journal->j_max_commit_age = 2807bd4c625cSLinus Torvalds le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age); 28081da177e4SLinus Torvalds journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE; 28091da177e4SLinus Torvalds 2810a9dd3643SJeff Mahoney if (check_advise_trans_params(sb, journal) != 0) 2811cf3d0b81SEdward Shishkin goto free_and_return; 28121da177e4SLinus Torvalds journal->j_default_max_commit_age = journal->j_max_commit_age; 28131da177e4SLinus Torvalds 28141da177e4SLinus Torvalds if (commit_max_age != 0) { 28151da177e4SLinus Torvalds journal->j_max_commit_age = commit_max_age; 28161da177e4SLinus Torvalds journal->j_max_trans_age = commit_max_age; 28171da177e4SLinus Torvalds } 28181da177e4SLinus Torvalds 2819a1c6f057SDmitry Monakhov reiserfs_info(sb, "journal params: device %pg, size %u, " 28201da177e4SLinus Torvalds "journal first block %u, max trans len %u, max batch %u, " 28211da177e4SLinus Torvalds "max commit age %u, max trans age %u\n", 2822a1c6f057SDmitry Monakhov journal->j_dev_bd, 2823a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb), 2824a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb), 28251da177e4SLinus Torvalds journal->j_trans_max, 28261da177e4SLinus Torvalds journal->j_max_batch, 2827bd4c625cSLinus Torvalds journal->j_max_commit_age, journal->j_max_trans_age); 28281da177e4SLinus Torvalds 28291da177e4SLinus Torvalds brelse(bhjh); 28301da177e4SLinus Torvalds 28311da177e4SLinus Torvalds journal->j_list_bitmap_index = 0; 2832a9dd3643SJeff Mahoney journal_list_init(sb); 28331da177e4SLinus Torvalds 2834bd4c625cSLinus Torvalds memset(journal->j_list_hash_table, 0, 2835bd4c625cSLinus Torvalds JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *)); 28361da177e4SLinus Torvalds 28371da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_dirty_buffers); 28381da177e4SLinus Torvalds spin_lock_init(&journal->j_dirty_buffers_lock); 28391da177e4SLinus Torvalds 28401da177e4SLinus Torvalds journal->j_start = 0; 28411da177e4SLinus Torvalds journal->j_len = 0; 28421da177e4SLinus Torvalds journal->j_len_alloc = 0; 2843a228bf8fSJeff Mahoney atomic_set(&journal->j_wcount, 0); 2844a228bf8fSJeff Mahoney atomic_set(&journal->j_async_throttle, 0); 28451da177e4SLinus Torvalds journal->j_bcount = 0; 28461da177e4SLinus Torvalds journal->j_trans_start_time = 0; 28471da177e4SLinus Torvalds journal->j_last = NULL; 28481da177e4SLinus Torvalds journal->j_first = NULL; 2849a228bf8fSJeff Mahoney init_waitqueue_head(&journal->j_join_wait); 2850f68215c4SJeff Mahoney mutex_init(&journal->j_mutex); 2851afe70259SJeff Mahoney mutex_init(&journal->j_flush_mutex); 28521da177e4SLinus Torvalds 28531da177e4SLinus Torvalds journal->j_trans_id = 10; 28541da177e4SLinus Torvalds journal->j_mount_id = 10; 28551da177e4SLinus Torvalds journal->j_state = 0; 2856a228bf8fSJeff Mahoney atomic_set(&journal->j_jlock, 0); 28571da177e4SLinus Torvalds journal->j_cnode_free_list = allocate_cnodes(num_cnodes); 28581da177e4SLinus Torvalds journal->j_cnode_free_orig = journal->j_cnode_free_list; 28591da177e4SLinus Torvalds journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0; 28601da177e4SLinus Torvalds journal->j_cnode_used = 0; 28611da177e4SLinus Torvalds journal->j_must_wait = 0; 28621da177e4SLinus Torvalds 2863576f6d79SJeff Mahoney if (journal->j_cnode_free == 0) { 2864a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-2004", "Journal cnode memory " 2865576f6d79SJeff Mahoney "allocation failed (%ld bytes). Journal is " 2866576f6d79SJeff Mahoney "too large for available memory. Usually " 2867576f6d79SJeff Mahoney "this is due to a journal that is too large.", 2868576f6d79SJeff Mahoney sizeof (struct reiserfs_journal_cnode) * num_cnodes); 2869576f6d79SJeff Mahoney goto free_and_return; 2870576f6d79SJeff Mahoney } 2871576f6d79SJeff Mahoney 2872a9dd3643SJeff Mahoney init_journal_hash(sb); 28731da177e4SLinus Torvalds jl = journal->j_current_jl; 287437c69b98SFrederic Weisbecker 287537c69b98SFrederic Weisbecker /* 287637c69b98SFrederic Weisbecker * get_list_bitmap() may call flush_commit_list() which 287737c69b98SFrederic Weisbecker * requires the lock. Calling flush_commit_list() shouldn't happen 287837c69b98SFrederic Weisbecker * this early but I like to be paranoid. 287937c69b98SFrederic Weisbecker */ 288037c69b98SFrederic Weisbecker reiserfs_write_lock(sb); 2881a9dd3643SJeff Mahoney jl->j_list_bitmap = get_list_bitmap(sb, jl); 288237c69b98SFrederic Weisbecker reiserfs_write_unlock(sb); 28831da177e4SLinus Torvalds if (!jl->j_list_bitmap) { 2884a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-2005", 288545b03d5eSJeff Mahoney "get_list_bitmap failed for journal list 0"); 28861da177e4SLinus Torvalds goto free_and_return; 28871da177e4SLinus Torvalds } 288837c69b98SFrederic Weisbecker 288937c69b98SFrederic Weisbecker ret = journal_read(sb); 289037c69b98SFrederic Weisbecker if (ret < 0) { 2891a9dd3643SJeff Mahoney reiserfs_warning(sb, "reiserfs-2006", 289245b03d5eSJeff Mahoney "Replay Failure, unable to mount"); 28931da177e4SLinus Torvalds goto free_and_return; 28941da177e4SLinus Torvalds } 28951da177e4SLinus Torvalds 2896c4028958SDavid Howells INIT_DELAYED_WORK(&journal->j_work, flush_async_commits); 2897a9dd3643SJeff Mahoney journal->j_work_sb = sb; 28981da177e4SLinus Torvalds return 0; 28991da177e4SLinus Torvalds free_and_return: 2900a9dd3643SJeff Mahoney free_journal_ram(sb); 29011da177e4SLinus Torvalds return 1; 29021da177e4SLinus Torvalds } 29031da177e4SLinus Torvalds 29041da177e4SLinus Torvalds /* 2905098297b2SJeff Mahoney * test for a polite end of the current transaction. Used by file_write, 2906098297b2SJeff Mahoney * and should be used by delete to make sure they don't write more than 2907098297b2SJeff Mahoney * can fit inside a single transaction 29081da177e4SLinus Torvalds */ 2909bd4c625cSLinus Torvalds int journal_transaction_should_end(struct reiserfs_transaction_handle *th, 2910bd4c625cSLinus Torvalds int new_alloc) 2911bd4c625cSLinus Torvalds { 29121da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(th->t_super); 29131da177e4SLinus Torvalds time_t now = get_seconds(); 29141da177e4SLinus Torvalds /* cannot restart while nested */ 29151da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 29161da177e4SLinus Torvalds if (th->t_refcount > 1) 29171da177e4SLinus Torvalds return 0; 29181da177e4SLinus Torvalds if (journal->j_must_wait > 0 || 29191da177e4SLinus Torvalds (journal->j_len_alloc + new_alloc) >= journal->j_max_batch || 2920a228bf8fSJeff Mahoney atomic_read(&journal->j_jlock) || 29211da177e4SLinus Torvalds (now - journal->j_trans_start_time) > journal->j_max_trans_age || 29221da177e4SLinus Torvalds journal->j_cnode_free < (journal->j_trans_max * 3)) { 29231da177e4SLinus Torvalds return 1; 29241da177e4SLinus Torvalds } 2925b18c1c6eSDavidlohr Bueso 29266ae1ea44SChris Mason journal->j_len_alloc += new_alloc; 29276ae1ea44SChris Mason th->t_blocks_allocated += new_alloc ; 29281da177e4SLinus Torvalds return 0; 29291da177e4SLinus Torvalds } 29301da177e4SLinus Torvalds 2931098297b2SJeff Mahoney /* this must be called inside a transaction */ 2932bd4c625cSLinus Torvalds void reiserfs_block_writes(struct reiserfs_transaction_handle *th) 2933bd4c625cSLinus Torvalds { 29341da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(th->t_super); 29351da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 29361da177e4SLinus Torvalds journal->j_must_wait = 1; 29371da177e4SLinus Torvalds set_bit(J_WRITERS_BLOCKED, &journal->j_state); 29381da177e4SLinus Torvalds return; 29391da177e4SLinus Torvalds } 29401da177e4SLinus Torvalds 2941098297b2SJeff Mahoney /* this must be called without a transaction started */ 2942bd4c625cSLinus Torvalds void reiserfs_allow_writes(struct super_block *s) 2943bd4c625cSLinus Torvalds { 29441da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 29451da177e4SLinus Torvalds clear_bit(J_WRITERS_BLOCKED, &journal->j_state); 29461da177e4SLinus Torvalds wake_up(&journal->j_join_wait); 29471da177e4SLinus Torvalds } 29481da177e4SLinus Torvalds 2949098297b2SJeff Mahoney /* this must be called without a transaction started */ 2950bd4c625cSLinus Torvalds void reiserfs_wait_on_write_block(struct super_block *s) 2951bd4c625cSLinus Torvalds { 29521da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 29531da177e4SLinus Torvalds wait_event(journal->j_join_wait, 29541da177e4SLinus Torvalds !test_bit(J_WRITERS_BLOCKED, &journal->j_state)); 29551da177e4SLinus Torvalds } 29561da177e4SLinus Torvalds 2957bd4c625cSLinus Torvalds static void queue_log_writer(struct super_block *s) 2958bd4c625cSLinus Torvalds { 29591da177e4SLinus Torvalds wait_queue_t wait; 29601da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 29611da177e4SLinus Torvalds set_bit(J_WRITERS_QUEUED, &journal->j_state); 29621da177e4SLinus Torvalds 29631da177e4SLinus Torvalds /* 29641da177e4SLinus Torvalds * we don't want to use wait_event here because 29651da177e4SLinus Torvalds * we only want to wait once. 29661da177e4SLinus Torvalds */ 29671da177e4SLinus Torvalds init_waitqueue_entry(&wait, current); 29681da177e4SLinus Torvalds add_wait_queue(&journal->j_join_wait, &wait); 29691da177e4SLinus Torvalds set_current_state(TASK_UNINTERRUPTIBLE); 29708ebc4232SFrederic Weisbecker if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) { 2971278f6679SJeff Mahoney int depth = reiserfs_write_unlock_nested(s); 29721da177e4SLinus Torvalds schedule(); 2973278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 29748ebc4232SFrederic Weisbecker } 29755ab2f7e0SMilind Arun Choudhary __set_current_state(TASK_RUNNING); 29761da177e4SLinus Torvalds remove_wait_queue(&journal->j_join_wait, &wait); 29771da177e4SLinus Torvalds } 29781da177e4SLinus Torvalds 2979bd4c625cSLinus Torvalds static void wake_queued_writers(struct super_block *s) 2980bd4c625cSLinus Torvalds { 29811da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 29821da177e4SLinus Torvalds if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state)) 29831da177e4SLinus Torvalds wake_up(&journal->j_join_wait); 29841da177e4SLinus Torvalds } 29851da177e4SLinus Torvalds 2986600ed416SJeff Mahoney static void let_transaction_grow(struct super_block *sb, unsigned int trans_id) 29871da177e4SLinus Torvalds { 29881da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(sb); 29891da177e4SLinus Torvalds unsigned long bcount = journal->j_bcount; 29901da177e4SLinus Torvalds while (1) { 2991278f6679SJeff Mahoney int depth; 2992278f6679SJeff Mahoney 2993278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(sb); 2994041e0e3bSNishanth Aravamudan schedule_timeout_uninterruptible(1); 2995278f6679SJeff Mahoney reiserfs_write_lock_nested(sb, depth); 2996278f6679SJeff Mahoney 29971da177e4SLinus Torvalds journal->j_current_jl->j_state |= LIST_COMMIT_PENDING; 29981da177e4SLinus Torvalds while ((atomic_read(&journal->j_wcount) > 0 || 29991da177e4SLinus Torvalds atomic_read(&journal->j_jlock)) && 30001da177e4SLinus Torvalds journal->j_trans_id == trans_id) { 30011da177e4SLinus Torvalds queue_log_writer(sb); 30021da177e4SLinus Torvalds } 30031da177e4SLinus Torvalds if (journal->j_trans_id != trans_id) 30041da177e4SLinus Torvalds break; 30051da177e4SLinus Torvalds if (bcount == journal->j_bcount) 30061da177e4SLinus Torvalds break; 30071da177e4SLinus Torvalds bcount = journal->j_bcount; 30081da177e4SLinus Torvalds } 30091da177e4SLinus Torvalds } 30101da177e4SLinus Torvalds 3011098297b2SJeff Mahoney /* 3012098297b2SJeff Mahoney * join == true if you must join an existing transaction. 3013098297b2SJeff Mahoney * join == false if you can deal with waiting for others to finish 3014098297b2SJeff Mahoney * 3015098297b2SJeff Mahoney * this will block until the transaction is joinable. send the number of 3016098297b2SJeff Mahoney * blocks you expect to use in nblocks. 30171da177e4SLinus Torvalds */ 3018bd4c625cSLinus Torvalds static int do_journal_begin_r(struct reiserfs_transaction_handle *th, 3019a9dd3643SJeff Mahoney struct super_block *sb, unsigned long nblocks, 3020bd4c625cSLinus Torvalds int join) 3021bd4c625cSLinus Torvalds { 30221da177e4SLinus Torvalds time_t now = get_seconds(); 3023600ed416SJeff Mahoney unsigned int old_trans_id; 3024a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 30251da177e4SLinus Torvalds struct reiserfs_transaction_handle myth; 30261da177e4SLinus Torvalds int sched_count = 0; 30271da177e4SLinus Torvalds int retval; 3028278f6679SJeff Mahoney int depth; 30291da177e4SLinus Torvalds 3030a9dd3643SJeff Mahoney reiserfs_check_lock_depth(sb, "journal_begin"); 303114a61442SEric Sesterhenn BUG_ON(nblocks > journal->j_trans_max); 30321da177e4SLinus Torvalds 3033a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.journal_being); 30341da177e4SLinus Torvalds /* set here for journal_join */ 30351da177e4SLinus Torvalds th->t_refcount = 1; 3036a9dd3643SJeff Mahoney th->t_super = sb; 30371da177e4SLinus Torvalds 30381da177e4SLinus Torvalds relock: 3039a9dd3643SJeff Mahoney lock_journal(sb); 30401da177e4SLinus Torvalds if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) { 3041a9dd3643SJeff Mahoney unlock_journal(sb); 30421da177e4SLinus Torvalds retval = journal->j_errno; 30431da177e4SLinus Torvalds goto out_fail; 30441da177e4SLinus Torvalds } 30451da177e4SLinus Torvalds journal->j_bcount++; 30461da177e4SLinus Torvalds 30471da177e4SLinus Torvalds if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) { 3048a9dd3643SJeff Mahoney unlock_journal(sb); 3049278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(sb); 3050a9dd3643SJeff Mahoney reiserfs_wait_on_write_block(sb); 3051278f6679SJeff Mahoney reiserfs_write_lock_nested(sb, depth); 3052a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.journal_relock_writers); 30531da177e4SLinus Torvalds goto relock; 30541da177e4SLinus Torvalds } 30551da177e4SLinus Torvalds now = get_seconds(); 30561da177e4SLinus Torvalds 3057098297b2SJeff Mahoney /* 3058098297b2SJeff Mahoney * if there is no room in the journal OR 3059098297b2SJeff Mahoney * if this transaction is too old, and we weren't called joinable, 3060098297b2SJeff Mahoney * wait for it to finish before beginning we don't sleep if there 3061098297b2SJeff Mahoney * aren't other writers 30621da177e4SLinus Torvalds */ 30631da177e4SLinus Torvalds 30641da177e4SLinus Torvalds if ((!join && journal->j_must_wait > 0) || 3065bd4c625cSLinus Torvalds (!join 3066bd4c625cSLinus Torvalds && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch) 3067bd4c625cSLinus Torvalds || (!join && atomic_read(&journal->j_wcount) > 0 3068bd4c625cSLinus Torvalds && journal->j_trans_start_time > 0 3069bd4c625cSLinus Torvalds && (now - journal->j_trans_start_time) > 3070bd4c625cSLinus Torvalds journal->j_max_trans_age) || (!join 3071bd4c625cSLinus Torvalds && atomic_read(&journal->j_jlock)) 3072bd4c625cSLinus Torvalds || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) { 30731da177e4SLinus Torvalds 30741da177e4SLinus Torvalds old_trans_id = journal->j_trans_id; 3075098297b2SJeff Mahoney /* allow others to finish this transaction */ 3076098297b2SJeff Mahoney unlock_journal(sb); 30771da177e4SLinus Torvalds 30781da177e4SLinus Torvalds if (!join && (journal->j_len_alloc + nblocks + 2) >= 30791da177e4SLinus Torvalds journal->j_max_batch && 3080bd4c625cSLinus Torvalds ((journal->j_len + nblocks + 2) * 100) < 3081bd4c625cSLinus Torvalds (journal->j_len_alloc * 75)) { 30821da177e4SLinus Torvalds if (atomic_read(&journal->j_wcount) > 10) { 30831da177e4SLinus Torvalds sched_count++; 3084a9dd3643SJeff Mahoney queue_log_writer(sb); 30851da177e4SLinus Torvalds goto relock; 30861da177e4SLinus Torvalds } 30871da177e4SLinus Torvalds } 3088098297b2SJeff Mahoney /* 3089098297b2SJeff Mahoney * don't mess with joining the transaction if all we 3090098297b2SJeff Mahoney * have to do is wait for someone else to do a commit 30911da177e4SLinus Torvalds */ 30921da177e4SLinus Torvalds if (atomic_read(&journal->j_jlock)) { 30931da177e4SLinus Torvalds while (journal->j_trans_id == old_trans_id && 30941da177e4SLinus Torvalds atomic_read(&journal->j_jlock)) { 3095a9dd3643SJeff Mahoney queue_log_writer(sb); 30961da177e4SLinus Torvalds } 30971da177e4SLinus Torvalds goto relock; 30981da177e4SLinus Torvalds } 3099b491dd17SJeff Mahoney retval = journal_join(&myth, sb); 31001da177e4SLinus Torvalds if (retval) 31011da177e4SLinus Torvalds goto out_fail; 31021da177e4SLinus Torvalds 31031da177e4SLinus Torvalds /* someone might have ended the transaction while we joined */ 31041da177e4SLinus Torvalds if (old_trans_id != journal->j_trans_id) { 310558d85426SJeff Mahoney retval = do_journal_end(&myth, 0); 31061da177e4SLinus Torvalds } else { 310758d85426SJeff Mahoney retval = do_journal_end(&myth, COMMIT_NOW); 31081da177e4SLinus Torvalds } 31091da177e4SLinus Torvalds 31101da177e4SLinus Torvalds if (retval) 31111da177e4SLinus Torvalds goto out_fail; 31121da177e4SLinus Torvalds 3113a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.journal_relock_wcount); 31141da177e4SLinus Torvalds goto relock; 31151da177e4SLinus Torvalds } 31161da177e4SLinus Torvalds /* we are the first writer, set trans_id */ 31171da177e4SLinus Torvalds if (journal->j_trans_start_time == 0) { 31181da177e4SLinus Torvalds journal->j_trans_start_time = get_seconds(); 31191da177e4SLinus Torvalds } 3120a228bf8fSJeff Mahoney atomic_inc(&journal->j_wcount); 31211da177e4SLinus Torvalds journal->j_len_alloc += nblocks; 31221da177e4SLinus Torvalds th->t_blocks_logged = 0; 31231da177e4SLinus Torvalds th->t_blocks_allocated = nblocks; 31241da177e4SLinus Torvalds th->t_trans_id = journal->j_trans_id; 3125a9dd3643SJeff Mahoney unlock_journal(sb); 31261da177e4SLinus Torvalds INIT_LIST_HEAD(&th->t_list); 31271da177e4SLinus Torvalds return 0; 31281da177e4SLinus Torvalds 31291da177e4SLinus Torvalds out_fail: 31301da177e4SLinus Torvalds memset(th, 0, sizeof(*th)); 3131098297b2SJeff Mahoney /* 3132098297b2SJeff Mahoney * Re-set th->t_super, so we can properly keep track of how many 31331da177e4SLinus Torvalds * persistent transactions there are. We need to do this so if this 3134098297b2SJeff Mahoney * call is part of a failed restart_transaction, we can free it later 3135098297b2SJeff Mahoney */ 3136a9dd3643SJeff Mahoney th->t_super = sb; 31371da177e4SLinus Torvalds return retval; 31381da177e4SLinus Torvalds } 31391da177e4SLinus Torvalds 3140bd4c625cSLinus Torvalds struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct 3141bd4c625cSLinus Torvalds super_block 3142bd4c625cSLinus Torvalds *s, 3143bd4c625cSLinus Torvalds int nblocks) 3144bd4c625cSLinus Torvalds { 31451da177e4SLinus Torvalds int ret; 31461da177e4SLinus Torvalds struct reiserfs_transaction_handle *th; 31471da177e4SLinus Torvalds 3148098297b2SJeff Mahoney /* 3149098297b2SJeff Mahoney * if we're nesting into an existing transaction. It will be 3150098297b2SJeff Mahoney * persistent on its own 31511da177e4SLinus Torvalds */ 31521da177e4SLinus Torvalds if (reiserfs_transaction_running(s)) { 31531da177e4SLinus Torvalds th = current->journal_info; 31541da177e4SLinus Torvalds th->t_refcount++; 315514a61442SEric Sesterhenn BUG_ON(th->t_refcount < 2); 315614a61442SEric Sesterhenn 31571da177e4SLinus Torvalds return th; 31581da177e4SLinus Torvalds } 3159d739b42bSPekka Enberg th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS); 31601da177e4SLinus Torvalds if (!th) 31611da177e4SLinus Torvalds return NULL; 31621da177e4SLinus Torvalds ret = journal_begin(th, s, nblocks); 31631da177e4SLinus Torvalds if (ret) { 3164d739b42bSPekka Enberg kfree(th); 31651da177e4SLinus Torvalds return NULL; 31661da177e4SLinus Torvalds } 31671da177e4SLinus Torvalds 31681da177e4SLinus Torvalds SB_JOURNAL(s)->j_persistent_trans++; 31691da177e4SLinus Torvalds return th; 31701da177e4SLinus Torvalds } 31711da177e4SLinus Torvalds 3172bd4c625cSLinus Torvalds int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th) 3173bd4c625cSLinus Torvalds { 31741da177e4SLinus Torvalds struct super_block *s = th->t_super; 31751da177e4SLinus Torvalds int ret = 0; 31761da177e4SLinus Torvalds if (th->t_trans_id) 317758d85426SJeff Mahoney ret = journal_end(th); 31781da177e4SLinus Torvalds else 31791da177e4SLinus Torvalds ret = -EIO; 31801da177e4SLinus Torvalds if (th->t_refcount == 0) { 31811da177e4SLinus Torvalds SB_JOURNAL(s)->j_persistent_trans--; 3182d739b42bSPekka Enberg kfree(th); 31831da177e4SLinus Torvalds } 31841da177e4SLinus Torvalds return ret; 31851da177e4SLinus Torvalds } 31861da177e4SLinus Torvalds 3187bd4c625cSLinus Torvalds static int journal_join(struct reiserfs_transaction_handle *th, 3188b491dd17SJeff Mahoney struct super_block *sb) 3189bd4c625cSLinus Torvalds { 31901da177e4SLinus Torvalds struct reiserfs_transaction_handle *cur_th = current->journal_info; 31911da177e4SLinus Torvalds 3192098297b2SJeff Mahoney /* 3193098297b2SJeff Mahoney * this keeps do_journal_end from NULLing out the 3194098297b2SJeff Mahoney * current->journal_info pointer 31951da177e4SLinus Torvalds */ 31961da177e4SLinus Torvalds th->t_handle_save = cur_th; 319714a61442SEric Sesterhenn BUG_ON(cur_th && cur_th->t_refcount > 1); 3198b491dd17SJeff Mahoney return do_journal_begin_r(th, sb, 1, JBEGIN_JOIN); 31991da177e4SLinus Torvalds } 32001da177e4SLinus Torvalds 3201bd4c625cSLinus Torvalds int journal_join_abort(struct reiserfs_transaction_handle *th, 3202b491dd17SJeff Mahoney struct super_block *sb) 3203bd4c625cSLinus Torvalds { 32041da177e4SLinus Torvalds struct reiserfs_transaction_handle *cur_th = current->journal_info; 32051da177e4SLinus Torvalds 3206098297b2SJeff Mahoney /* 3207098297b2SJeff Mahoney * this keeps do_journal_end from NULLing out the 3208098297b2SJeff Mahoney * current->journal_info pointer 32091da177e4SLinus Torvalds */ 32101da177e4SLinus Torvalds th->t_handle_save = cur_th; 321114a61442SEric Sesterhenn BUG_ON(cur_th && cur_th->t_refcount > 1); 3212b491dd17SJeff Mahoney return do_journal_begin_r(th, sb, 1, JBEGIN_ABORT); 32131da177e4SLinus Torvalds } 32141da177e4SLinus Torvalds 3215bd4c625cSLinus Torvalds int journal_begin(struct reiserfs_transaction_handle *th, 3216a9dd3643SJeff Mahoney struct super_block *sb, unsigned long nblocks) 3217bd4c625cSLinus Torvalds { 32181da177e4SLinus Torvalds struct reiserfs_transaction_handle *cur_th = current->journal_info; 32191da177e4SLinus Torvalds int ret; 32201da177e4SLinus Torvalds 32211da177e4SLinus Torvalds th->t_handle_save = NULL; 32221da177e4SLinus Torvalds if (cur_th) { 32231da177e4SLinus Torvalds /* we are nesting into the current transaction */ 3224a9dd3643SJeff Mahoney if (cur_th->t_super == sb) { 32251da177e4SLinus Torvalds BUG_ON(!cur_th->t_refcount); 32261da177e4SLinus Torvalds cur_th->t_refcount++; 32271da177e4SLinus Torvalds memcpy(th, cur_th, sizeof(*th)); 32281da177e4SLinus Torvalds if (th->t_refcount <= 1) 3229a9dd3643SJeff Mahoney reiserfs_warning(sb, "reiserfs-2005", 323045b03d5eSJeff Mahoney "BAD: refcount <= 1, but " 323145b03d5eSJeff Mahoney "journal_info != 0"); 32321da177e4SLinus Torvalds return 0; 32331da177e4SLinus Torvalds } else { 3234098297b2SJeff Mahoney /* 3235098297b2SJeff Mahoney * we've ended up with a handle from a different 3236098297b2SJeff Mahoney * filesystem. save it and restore on journal_end. 3237098297b2SJeff Mahoney * This should never really happen... 32381da177e4SLinus Torvalds */ 3239a9dd3643SJeff Mahoney reiserfs_warning(sb, "clm-2100", 324045b03d5eSJeff Mahoney "nesting info a different FS"); 32411da177e4SLinus Torvalds th->t_handle_save = current->journal_info; 32421da177e4SLinus Torvalds current->journal_info = th; 32431da177e4SLinus Torvalds } 32441da177e4SLinus Torvalds } else { 32451da177e4SLinus Torvalds current->journal_info = th; 32461da177e4SLinus Torvalds } 3247a9dd3643SJeff Mahoney ret = do_journal_begin_r(th, sb, nblocks, JBEGIN_REG); 324814a61442SEric Sesterhenn BUG_ON(current->journal_info != th); 32491da177e4SLinus Torvalds 3250098297b2SJeff Mahoney /* 3251098297b2SJeff Mahoney * I guess this boils down to being the reciprocal of clm-2100 above. 3252098297b2SJeff Mahoney * If do_journal_begin_r fails, we need to put it back, since 3253098297b2SJeff Mahoney * journal_end won't be called to do it. */ 32541da177e4SLinus Torvalds if (ret) 32551da177e4SLinus Torvalds current->journal_info = th->t_handle_save; 32561da177e4SLinus Torvalds else 32571da177e4SLinus Torvalds BUG_ON(!th->t_refcount); 32581da177e4SLinus Torvalds 32591da177e4SLinus Torvalds return ret; 32601da177e4SLinus Torvalds } 32611da177e4SLinus Torvalds 32621da177e4SLinus Torvalds /* 3263098297b2SJeff Mahoney * puts bh into the current transaction. If it was already there, reorders 3264098297b2SJeff Mahoney * removes the old pointers from the hash, and puts new ones in (to make 3265098297b2SJeff Mahoney * sure replay happen in the right order). 3266098297b2SJeff Mahoney * 3267098297b2SJeff Mahoney * if it was dirty, cleans and files onto the clean list. I can't let it 3268098297b2SJeff Mahoney * be dirty again until the transaction is committed. 3269098297b2SJeff Mahoney * 3270098297b2SJeff Mahoney * if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len. 32711da177e4SLinus Torvalds */ 3272bd4c625cSLinus Torvalds int journal_mark_dirty(struct reiserfs_transaction_handle *th, 327309f1b80bSJeff Mahoney struct buffer_head *bh) 3274bd4c625cSLinus Torvalds { 327509f1b80bSJeff Mahoney struct super_block *sb = th->t_super; 3276a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 32771da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn = NULL; 32781da177e4SLinus Torvalds int count_already_incd = 0; 32791da177e4SLinus Torvalds int prepared = 0; 32801da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 32811da177e4SLinus Torvalds 3282a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.mark_dirty); 32831da177e4SLinus Torvalds if (th->t_trans_id != journal->j_trans_id) { 3284c3a9c210SJeff Mahoney reiserfs_panic(th->t_super, "journal-1577", 3285c3a9c210SJeff Mahoney "handle trans id %ld != current trans id %ld", 32861da177e4SLinus Torvalds th->t_trans_id, journal->j_trans_id); 32871da177e4SLinus Torvalds } 32881da177e4SLinus Torvalds 32891da177e4SLinus Torvalds prepared = test_clear_buffer_journal_prepared(bh); 32901da177e4SLinus Torvalds clear_buffer_journal_restore_dirty(bh); 32911da177e4SLinus Torvalds /* already in this transaction, we are done */ 32921da177e4SLinus Torvalds if (buffer_journaled(bh)) { 3293a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.mark_dirty_already); 32941da177e4SLinus Torvalds return 0; 32951da177e4SLinus Torvalds } 32961da177e4SLinus Torvalds 3297098297b2SJeff Mahoney /* 3298098297b2SJeff Mahoney * this must be turned into a panic instead of a warning. We can't 3299098297b2SJeff Mahoney * allow a dirty or journal_dirty or locked buffer to be logged, as 3300098297b2SJeff Mahoney * some changes could get to disk too early. NOT GOOD. 33011da177e4SLinus Torvalds */ 33021da177e4SLinus Torvalds if (!prepared || buffer_dirty(bh)) { 3303a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1777", 330445b03d5eSJeff Mahoney "buffer %llu bad state " 33051da177e4SLinus Torvalds "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT", 3306bd4c625cSLinus Torvalds (unsigned long long)bh->b_blocknr, 3307bd4c625cSLinus Torvalds prepared ? ' ' : '!', 33081da177e4SLinus Torvalds buffer_locked(bh) ? ' ' : '!', 33091da177e4SLinus Torvalds buffer_dirty(bh) ? ' ' : '!', 33101da177e4SLinus Torvalds buffer_journal_dirty(bh) ? ' ' : '!'); 33111da177e4SLinus Torvalds } 33121da177e4SLinus Torvalds 3313a228bf8fSJeff Mahoney if (atomic_read(&journal->j_wcount) <= 0) { 3314a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1409", 331545b03d5eSJeff Mahoney "returning because j_wcount was %d", 3316a228bf8fSJeff Mahoney atomic_read(&journal->j_wcount)); 33171da177e4SLinus Torvalds return 1; 33181da177e4SLinus Torvalds } 3319098297b2SJeff Mahoney /* 3320098297b2SJeff Mahoney * this error means I've screwed up, and we've overflowed 3321098297b2SJeff Mahoney * the transaction. Nothing can be done here, except make the 3322098297b2SJeff Mahoney * FS readonly or panic. 33231da177e4SLinus Torvalds */ 33241da177e4SLinus Torvalds if (journal->j_len >= journal->j_trans_max) { 3325c3a9c210SJeff Mahoney reiserfs_panic(th->t_super, "journal-1413", 3326c3a9c210SJeff Mahoney "j_len (%lu) is too big", 3327bd4c625cSLinus Torvalds journal->j_len); 33281da177e4SLinus Torvalds } 33291da177e4SLinus Torvalds 33301da177e4SLinus Torvalds if (buffer_journal_dirty(bh)) { 33311da177e4SLinus Torvalds count_already_incd = 1; 3332a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.mark_dirty_notjournal); 33331da177e4SLinus Torvalds clear_buffer_journal_dirty(bh); 33341da177e4SLinus Torvalds } 33351da177e4SLinus Torvalds 33361da177e4SLinus Torvalds if (journal->j_len > journal->j_len_alloc) { 33371da177e4SLinus Torvalds journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT; 33381da177e4SLinus Torvalds } 33391da177e4SLinus Torvalds 33401da177e4SLinus Torvalds set_buffer_journaled(bh); 33411da177e4SLinus Torvalds 33421da177e4SLinus Torvalds /* now put this guy on the end */ 33431da177e4SLinus Torvalds if (!cn) { 3344a9dd3643SJeff Mahoney cn = get_cnode(sb); 33451da177e4SLinus Torvalds if (!cn) { 3346a9dd3643SJeff Mahoney reiserfs_panic(sb, "journal-4", "get_cnode failed!"); 33471da177e4SLinus Torvalds } 33481da177e4SLinus Torvalds 33491da177e4SLinus Torvalds if (th->t_blocks_logged == th->t_blocks_allocated) { 33501da177e4SLinus Torvalds th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT; 33511da177e4SLinus Torvalds journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT; 33521da177e4SLinus Torvalds } 33531da177e4SLinus Torvalds th->t_blocks_logged++; 33541da177e4SLinus Torvalds journal->j_len++; 33551da177e4SLinus Torvalds 33561da177e4SLinus Torvalds cn->bh = bh; 33571da177e4SLinus Torvalds cn->blocknr = bh->b_blocknr; 3358a9dd3643SJeff Mahoney cn->sb = sb; 33591da177e4SLinus Torvalds cn->jlist = NULL; 33601da177e4SLinus Torvalds insert_journal_hash(journal->j_hash_table, cn); 33611da177e4SLinus Torvalds if (!count_already_incd) { 33621da177e4SLinus Torvalds get_bh(bh); 33631da177e4SLinus Torvalds } 33641da177e4SLinus Torvalds } 33651da177e4SLinus Torvalds cn->next = NULL; 33661da177e4SLinus Torvalds cn->prev = journal->j_last; 33671da177e4SLinus Torvalds cn->bh = bh; 33681da177e4SLinus Torvalds if (journal->j_last) { 33691da177e4SLinus Torvalds journal->j_last->next = cn; 33701da177e4SLinus Torvalds journal->j_last = cn; 33711da177e4SLinus Torvalds } else { 33721da177e4SLinus Torvalds journal->j_first = cn; 33731da177e4SLinus Torvalds journal->j_last = cn; 33741da177e4SLinus Torvalds } 3375033369d1SArtem Bityutskiy reiserfs_schedule_old_flush(sb); 33761da177e4SLinus Torvalds return 0; 33771da177e4SLinus Torvalds } 33781da177e4SLinus Torvalds 337958d85426SJeff Mahoney int journal_end(struct reiserfs_transaction_handle *th) 3380bd4c625cSLinus Torvalds { 338158d85426SJeff Mahoney struct super_block *sb = th->t_super; 33821da177e4SLinus Torvalds if (!current->journal_info && th->t_refcount > 1) 3383a9dd3643SJeff Mahoney reiserfs_warning(sb, "REISER-NESTING", 338445b03d5eSJeff Mahoney "th NULL, refcount %d", th->t_refcount); 33851da177e4SLinus Torvalds 33861da177e4SLinus Torvalds if (!th->t_trans_id) { 33871da177e4SLinus Torvalds WARN_ON(1); 33881da177e4SLinus Torvalds return -EIO; 33891da177e4SLinus Torvalds } 33901da177e4SLinus Torvalds 33911da177e4SLinus Torvalds th->t_refcount--; 33921da177e4SLinus Torvalds if (th->t_refcount > 0) { 3393bd4c625cSLinus Torvalds struct reiserfs_transaction_handle *cur_th = 3394bd4c625cSLinus Torvalds current->journal_info; 33951da177e4SLinus Torvalds 3396098297b2SJeff Mahoney /* 3397098297b2SJeff Mahoney * we aren't allowed to close a nested transaction on a 3398098297b2SJeff Mahoney * different filesystem from the one in the task struct 33991da177e4SLinus Torvalds */ 340014a61442SEric Sesterhenn BUG_ON(cur_th->t_super != th->t_super); 34011da177e4SLinus Torvalds 34021da177e4SLinus Torvalds if (th != cur_th) { 34031da177e4SLinus Torvalds memcpy(current->journal_info, th, sizeof(*th)); 34041da177e4SLinus Torvalds th->t_trans_id = 0; 34051da177e4SLinus Torvalds } 34061da177e4SLinus Torvalds return 0; 34071da177e4SLinus Torvalds } else { 340858d85426SJeff Mahoney return do_journal_end(th, 0); 34091da177e4SLinus Torvalds } 34101da177e4SLinus Torvalds } 34111da177e4SLinus Torvalds 3412098297b2SJeff Mahoney /* 3413098297b2SJeff Mahoney * removes from the current transaction, relsing and descrementing any counters. 3414098297b2SJeff Mahoney * also files the removed buffer directly onto the clean list 3415098297b2SJeff Mahoney * 3416098297b2SJeff Mahoney * called by journal_mark_freed when a block has been deleted 3417098297b2SJeff Mahoney * 3418098297b2SJeff Mahoney * returns 1 if it cleaned and relsed the buffer. 0 otherwise 34191da177e4SLinus Torvalds */ 3420a9dd3643SJeff Mahoney static int remove_from_transaction(struct super_block *sb, 3421bd4c625cSLinus Torvalds b_blocknr_t blocknr, int already_cleaned) 3422bd4c625cSLinus Torvalds { 34231da177e4SLinus Torvalds struct buffer_head *bh; 34241da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 3425a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 34261da177e4SLinus Torvalds int ret = 0; 34271da177e4SLinus Torvalds 3428a9dd3643SJeff Mahoney cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr); 34291da177e4SLinus Torvalds if (!cn || !cn->bh) { 34301da177e4SLinus Torvalds return ret; 34311da177e4SLinus Torvalds } 34321da177e4SLinus Torvalds bh = cn->bh; 34331da177e4SLinus Torvalds if (cn->prev) { 34341da177e4SLinus Torvalds cn->prev->next = cn->next; 34351da177e4SLinus Torvalds } 34361da177e4SLinus Torvalds if (cn->next) { 34371da177e4SLinus Torvalds cn->next->prev = cn->prev; 34381da177e4SLinus Torvalds } 34391da177e4SLinus Torvalds if (cn == journal->j_first) { 34401da177e4SLinus Torvalds journal->j_first = cn->next; 34411da177e4SLinus Torvalds } 34421da177e4SLinus Torvalds if (cn == journal->j_last) { 34431da177e4SLinus Torvalds journal->j_last = cn->prev; 34441da177e4SLinus Torvalds } 34451da177e4SLinus Torvalds if (bh) 3446a9dd3643SJeff Mahoney remove_journal_hash(sb, journal->j_hash_table, NULL, 3447bd4c625cSLinus Torvalds bh->b_blocknr, 0); 34481da177e4SLinus Torvalds clear_buffer_journaled(bh); /* don't log this one */ 34491da177e4SLinus Torvalds 34501da177e4SLinus Torvalds if (!already_cleaned) { 34511da177e4SLinus Torvalds clear_buffer_journal_dirty(bh); 34521da177e4SLinus Torvalds clear_buffer_dirty(bh); 34531da177e4SLinus Torvalds clear_buffer_journal_test(bh); 34541da177e4SLinus Torvalds put_bh(bh); 3455a228bf8fSJeff Mahoney if (atomic_read(&bh->b_count) < 0) { 3456a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1752", 345745b03d5eSJeff Mahoney "b_count < 0"); 34581da177e4SLinus Torvalds } 34591da177e4SLinus Torvalds ret = 1; 34601da177e4SLinus Torvalds } 34611da177e4SLinus Torvalds journal->j_len--; 34621da177e4SLinus Torvalds journal->j_len_alloc--; 3463a9dd3643SJeff Mahoney free_cnode(sb, cn); 34641da177e4SLinus Torvalds return ret; 34651da177e4SLinus Torvalds } 34661da177e4SLinus Torvalds 34671da177e4SLinus Torvalds /* 3468098297b2SJeff Mahoney * for any cnode in a journal list, it can only be dirtied of all the 3469098297b2SJeff Mahoney * transactions that include it are committed to disk. 3470098297b2SJeff Mahoney * this checks through each transaction, and returns 1 if you are allowed 3471098297b2SJeff Mahoney * to dirty, and 0 if you aren't 3472098297b2SJeff Mahoney * 3473098297b2SJeff Mahoney * it is called by dirty_journal_list, which is called after 3474098297b2SJeff Mahoney * flush_commit_list has gotten all the log blocks for a given 3475098297b2SJeff Mahoney * transaction on disk 3476098297b2SJeff Mahoney * 34771da177e4SLinus Torvalds */ 3478bd4c625cSLinus Torvalds static int can_dirty(struct reiserfs_journal_cnode *cn) 3479bd4c625cSLinus Torvalds { 34801da177e4SLinus Torvalds struct super_block *sb = cn->sb; 34811da177e4SLinus Torvalds b_blocknr_t blocknr = cn->blocknr; 34821da177e4SLinus Torvalds struct reiserfs_journal_cnode *cur = cn->hprev; 34831da177e4SLinus Torvalds int can_dirty = 1; 34841da177e4SLinus Torvalds 3485098297b2SJeff Mahoney /* 3486098297b2SJeff Mahoney * first test hprev. These are all newer than cn, so any node here 3487098297b2SJeff Mahoney * with the same block number and dev means this node can't be sent 3488098297b2SJeff Mahoney * to disk right now. 34891da177e4SLinus Torvalds */ 34901da177e4SLinus Torvalds while (cur && can_dirty) { 34911da177e4SLinus Torvalds if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb && 34921da177e4SLinus Torvalds cur->blocknr == blocknr) { 34931da177e4SLinus Torvalds can_dirty = 0; 34941da177e4SLinus Torvalds } 34951da177e4SLinus Torvalds cur = cur->hprev; 34961da177e4SLinus Torvalds } 3497098297b2SJeff Mahoney /* 3498098297b2SJeff Mahoney * then test hnext. These are all older than cn. As long as they 3499098297b2SJeff Mahoney * are committed to the log, it is safe to write cn to disk 35001da177e4SLinus Torvalds */ 35011da177e4SLinus Torvalds cur = cn->hnext; 35021da177e4SLinus Torvalds while (cur && can_dirty) { 35031da177e4SLinus Torvalds if (cur->jlist && cur->jlist->j_len > 0 && 3504a228bf8fSJeff Mahoney atomic_read(&cur->jlist->j_commit_left) > 0 && cur->bh && 35051da177e4SLinus Torvalds cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) { 35061da177e4SLinus Torvalds can_dirty = 0; 35071da177e4SLinus Torvalds } 35081da177e4SLinus Torvalds cur = cur->hnext; 35091da177e4SLinus Torvalds } 35101da177e4SLinus Torvalds return can_dirty; 35111da177e4SLinus Torvalds } 35121da177e4SLinus Torvalds 3513098297b2SJeff Mahoney /* 3514098297b2SJeff Mahoney * syncs the commit blocks, but does not force the real buffers to disk 3515098297b2SJeff Mahoney * will wait until the current transaction is done/committed before returning 35161da177e4SLinus Torvalds */ 351758d85426SJeff Mahoney int journal_end_sync(struct reiserfs_transaction_handle *th) 3518bd4c625cSLinus Torvalds { 351958d85426SJeff Mahoney struct super_block *sb = th->t_super; 3520a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 35211da177e4SLinus Torvalds 35221da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 35231da177e4SLinus Torvalds /* you can sync while nested, very, very bad */ 352414a61442SEric Sesterhenn BUG_ON(th->t_refcount > 1); 35251da177e4SLinus Torvalds if (journal->j_len == 0) { 3526a9dd3643SJeff Mahoney reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb), 3527bd4c625cSLinus Torvalds 1); 352809f1b80bSJeff Mahoney journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb)); 35291da177e4SLinus Torvalds } 353058d85426SJeff Mahoney return do_journal_end(th, COMMIT_NOW | WAIT); 35311da177e4SLinus Torvalds } 35321da177e4SLinus Torvalds 3533098297b2SJeff Mahoney /* writeback the pending async commits to disk */ 3534c4028958SDavid Howells static void flush_async_commits(struct work_struct *work) 3535bd4c625cSLinus Torvalds { 3536c4028958SDavid Howells struct reiserfs_journal *journal = 3537c4028958SDavid Howells container_of(work, struct reiserfs_journal, j_work.work); 3538a9dd3643SJeff Mahoney struct super_block *sb = journal->j_work_sb; 35391da177e4SLinus Torvalds struct reiserfs_journal_list *jl; 35401da177e4SLinus Torvalds struct list_head *entry; 35411da177e4SLinus Torvalds 35428ebc4232SFrederic Weisbecker reiserfs_write_lock(sb); 35431da177e4SLinus Torvalds if (!list_empty(&journal->j_journal_list)) { 35441da177e4SLinus Torvalds /* last entry is the youngest, commit it and you get everything */ 35451da177e4SLinus Torvalds entry = journal->j_journal_list.prev; 35461da177e4SLinus Torvalds jl = JOURNAL_LIST_ENTRY(entry); 3547a9dd3643SJeff Mahoney flush_commit_list(sb, jl, 1); 35481da177e4SLinus Torvalds } 35498ebc4232SFrederic Weisbecker reiserfs_write_unlock(sb); 35501da177e4SLinus Torvalds } 35511da177e4SLinus Torvalds 35521da177e4SLinus Torvalds /* 3553098297b2SJeff Mahoney * flushes any old transactions to disk 3554098297b2SJeff Mahoney * ends the current transaction if it is too old 35551da177e4SLinus Torvalds */ 355625729b0eSArtem Bityutskiy void reiserfs_flush_old_commits(struct super_block *sb) 3557bd4c625cSLinus Torvalds { 35581da177e4SLinus Torvalds time_t now; 35591da177e4SLinus Torvalds struct reiserfs_transaction_handle th; 3560a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 35611da177e4SLinus Torvalds 35621da177e4SLinus Torvalds now = get_seconds(); 3563098297b2SJeff Mahoney /* 3564098297b2SJeff Mahoney * safety check so we don't flush while we are replaying the log during 35651da177e4SLinus Torvalds * mount 35661da177e4SLinus Torvalds */ 356725729b0eSArtem Bityutskiy if (list_empty(&journal->j_journal_list)) 356825729b0eSArtem Bityutskiy return; 35691da177e4SLinus Torvalds 3570098297b2SJeff Mahoney /* 3571098297b2SJeff Mahoney * check the current transaction. If there are no writers, and it is 35721da177e4SLinus Torvalds * too old, finish it, and force the commit blocks to disk 35731da177e4SLinus Torvalds */ 35741da177e4SLinus Torvalds if (atomic_read(&journal->j_wcount) <= 0 && 35751da177e4SLinus Torvalds journal->j_trans_start_time > 0 && 35761da177e4SLinus Torvalds journal->j_len > 0 && 3577bd4c625cSLinus Torvalds (now - journal->j_trans_start_time) > journal->j_max_trans_age) { 3578b491dd17SJeff Mahoney if (!journal_join(&th, sb)) { 3579a9dd3643SJeff Mahoney reiserfs_prepare_for_journal(sb, 3580a9dd3643SJeff Mahoney SB_BUFFER_WITH_SB(sb), 3581bd4c625cSLinus Torvalds 1); 358209f1b80bSJeff Mahoney journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb)); 35831da177e4SLinus Torvalds 3584098297b2SJeff Mahoney /* 3585098297b2SJeff Mahoney * we're only being called from kreiserfsd, it makes 3586098297b2SJeff Mahoney * no sense to do an async commit so that kreiserfsd 3587098297b2SJeff Mahoney * can do it later 35881da177e4SLinus Torvalds */ 358958d85426SJeff Mahoney do_journal_end(&th, COMMIT_NOW | WAIT); 35901da177e4SLinus Torvalds } 35911da177e4SLinus Torvalds } 35921da177e4SLinus Torvalds } 35931da177e4SLinus Torvalds 35941da177e4SLinus Torvalds /* 3595098297b2SJeff Mahoney * returns 0 if do_journal_end should return right away, returns 1 if 3596098297b2SJeff Mahoney * do_journal_end should finish the commit 3597098297b2SJeff Mahoney * 3598098297b2SJeff Mahoney * if the current transaction is too old, but still has writers, this will 3599098297b2SJeff Mahoney * wait on j_join_wait until all the writers are done. By the time it 3600098297b2SJeff Mahoney * wakes up, the transaction it was called has already ended, so it just 3601098297b2SJeff Mahoney * flushes the commit list and returns 0. 3602098297b2SJeff Mahoney * 3603098297b2SJeff Mahoney * Won't batch when flush or commit_now is set. Also won't batch when 3604098297b2SJeff Mahoney * others are waiting on j_join_wait. 3605098297b2SJeff Mahoney * 3606098297b2SJeff Mahoney * Note, we can't allow the journal_end to proceed while there are still 3607098297b2SJeff Mahoney * writers in the log. 36081da177e4SLinus Torvalds */ 360958d85426SJeff Mahoney static int check_journal_end(struct reiserfs_transaction_handle *th, int flags) 3610bd4c625cSLinus Torvalds { 36111da177e4SLinus Torvalds 36121da177e4SLinus Torvalds time_t now; 36131da177e4SLinus Torvalds int flush = flags & FLUSH_ALL; 36141da177e4SLinus Torvalds int commit_now = flags & COMMIT_NOW; 36151da177e4SLinus Torvalds int wait_on_commit = flags & WAIT; 36161da177e4SLinus Torvalds struct reiserfs_journal_list *jl; 361758d85426SJeff Mahoney struct super_block *sb = th->t_super; 3618a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 36191da177e4SLinus Torvalds 36201da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 36211da177e4SLinus Torvalds 36221da177e4SLinus Torvalds if (th->t_trans_id != journal->j_trans_id) { 3623c3a9c210SJeff Mahoney reiserfs_panic(th->t_super, "journal-1577", 3624c3a9c210SJeff Mahoney "handle trans id %ld != current trans id %ld", 36251da177e4SLinus Torvalds th->t_trans_id, journal->j_trans_id); 36261da177e4SLinus Torvalds } 36271da177e4SLinus Torvalds 36281da177e4SLinus Torvalds journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged); 3629098297b2SJeff Mahoney /* <= 0 is allowed. unmounting might not call begin */ 3630a228bf8fSJeff Mahoney if (atomic_read(&journal->j_wcount) > 0) 3631a228bf8fSJeff Mahoney atomic_dec(&journal->j_wcount); 36321da177e4SLinus Torvalds 3633098297b2SJeff Mahoney /* 3634098297b2SJeff Mahoney * BUG, deal with case where j_len is 0, but people previously 3635098297b2SJeff Mahoney * freed blocks need to be released will be dealt with by next 3636098297b2SJeff Mahoney * transaction that actually writes something, but should be taken 3637098297b2SJeff Mahoney * care of in this trans 36381da177e4SLinus Torvalds */ 363914a61442SEric Sesterhenn BUG_ON(journal->j_len == 0); 364014a61442SEric Sesterhenn 3641098297b2SJeff Mahoney /* 3642098297b2SJeff Mahoney * if wcount > 0, and we are called to with flush or commit_now, 3643098297b2SJeff Mahoney * we wait on j_join_wait. We will wake up when the last writer has 3644098297b2SJeff Mahoney * finished the transaction, and started it on its way to the disk. 3645098297b2SJeff Mahoney * Then, we flush the commit or journal list, and just return 0 3646098297b2SJeff Mahoney * because the rest of journal end was already done for this 3647098297b2SJeff Mahoney * transaction. 36481da177e4SLinus Torvalds */ 3649a228bf8fSJeff Mahoney if (atomic_read(&journal->j_wcount) > 0) { 36501da177e4SLinus Torvalds if (flush || commit_now) { 36511da177e4SLinus Torvalds unsigned trans_id; 36521da177e4SLinus Torvalds 36531da177e4SLinus Torvalds jl = journal->j_current_jl; 36541da177e4SLinus Torvalds trans_id = jl->j_trans_id; 36551da177e4SLinus Torvalds if (wait_on_commit) 36561da177e4SLinus Torvalds jl->j_state |= LIST_COMMIT_PENDING; 3657a228bf8fSJeff Mahoney atomic_set(&journal->j_jlock, 1); 36581da177e4SLinus Torvalds if (flush) { 36591da177e4SLinus Torvalds journal->j_next_full_flush = 1; 36601da177e4SLinus Torvalds } 3661a9dd3643SJeff Mahoney unlock_journal(sb); 36621da177e4SLinus Torvalds 3663098297b2SJeff Mahoney /* 3664098297b2SJeff Mahoney * sleep while the current transaction is 3665098297b2SJeff Mahoney * still j_jlocked 3666098297b2SJeff Mahoney */ 36671da177e4SLinus Torvalds while (journal->j_trans_id == trans_id) { 36681da177e4SLinus Torvalds if (atomic_read(&journal->j_jlock)) { 3669a9dd3643SJeff Mahoney queue_log_writer(sb); 36701da177e4SLinus Torvalds } else { 3671a9dd3643SJeff Mahoney lock_journal(sb); 36721da177e4SLinus Torvalds if (journal->j_trans_id == trans_id) { 3673a228bf8fSJeff Mahoney atomic_set(&journal->j_jlock, 3674bd4c625cSLinus Torvalds 1); 36751da177e4SLinus Torvalds } 3676a9dd3643SJeff Mahoney unlock_journal(sb); 36771da177e4SLinus Torvalds } 36781da177e4SLinus Torvalds } 367914a61442SEric Sesterhenn BUG_ON(journal->j_trans_id == trans_id); 368014a61442SEric Sesterhenn 3681bd4c625cSLinus Torvalds if (commit_now 3682a9dd3643SJeff Mahoney && journal_list_still_alive(sb, trans_id) 3683bd4c625cSLinus Torvalds && wait_on_commit) { 3684a9dd3643SJeff Mahoney flush_commit_list(sb, jl, 1); 36851da177e4SLinus Torvalds } 36861da177e4SLinus Torvalds return 0; 36871da177e4SLinus Torvalds } 3688a9dd3643SJeff Mahoney unlock_journal(sb); 36891da177e4SLinus Torvalds return 0; 36901da177e4SLinus Torvalds } 36911da177e4SLinus Torvalds 36921da177e4SLinus Torvalds /* deal with old transactions where we are the last writers */ 36931da177e4SLinus Torvalds now = get_seconds(); 36941da177e4SLinus Torvalds if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) { 36951da177e4SLinus Torvalds commit_now = 1; 36961da177e4SLinus Torvalds journal->j_next_async_flush = 1; 36971da177e4SLinus Torvalds } 36981da177e4SLinus Torvalds /* don't batch when someone is waiting on j_join_wait */ 36991da177e4SLinus Torvalds /* don't batch when syncing the commit or flushing the whole trans */ 3700a228bf8fSJeff Mahoney if (!(journal->j_must_wait > 0) && !(atomic_read(&journal->j_jlock)) 3701bd4c625cSLinus Torvalds && !flush && !commit_now && (journal->j_len < journal->j_max_batch) 3702bd4c625cSLinus Torvalds && journal->j_len_alloc < journal->j_max_batch 3703bd4c625cSLinus Torvalds && journal->j_cnode_free > (journal->j_trans_max * 3)) { 37041da177e4SLinus Torvalds journal->j_bcount++; 3705a9dd3643SJeff Mahoney unlock_journal(sb); 37061da177e4SLinus Torvalds return 0; 37071da177e4SLinus Torvalds } 37081da177e4SLinus Torvalds 3709a9dd3643SJeff Mahoney if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(sb)) { 3710a9dd3643SJeff Mahoney reiserfs_panic(sb, "journal-003", 3711c3a9c210SJeff Mahoney "j_start (%ld) is too high", 3712bd4c625cSLinus Torvalds journal->j_start); 37131da177e4SLinus Torvalds } 37141da177e4SLinus Torvalds return 1; 37151da177e4SLinus Torvalds } 37161da177e4SLinus Torvalds 37171da177e4SLinus Torvalds /* 3718098297b2SJeff Mahoney * Does all the work that makes deleting blocks safe. 3719098297b2SJeff Mahoney * when deleting a block mark BH_JNew, just remove it from the current 3720098297b2SJeff Mahoney * transaction, clean it's buffer_head and move on. 3721098297b2SJeff Mahoney * 3722098297b2SJeff Mahoney * otherwise: 3723098297b2SJeff Mahoney * set a bit for the block in the journal bitmap. That will prevent it from 3724098297b2SJeff Mahoney * being allocated for unformatted nodes before this transaction has finished. 3725098297b2SJeff Mahoney * 3726098297b2SJeff Mahoney * mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers. 3727098297b2SJeff Mahoney * That will prevent any old transactions with this block from trying to flush 3728098297b2SJeff Mahoney * to the real location. Since we aren't removing the cnode from the 3729098297b2SJeff Mahoney * journal_list_hash, *the block can't be reallocated yet. 3730098297b2SJeff Mahoney * 3731098297b2SJeff Mahoney * Then remove it from the current transaction, decrementing any counters and 3732098297b2SJeff Mahoney * filing it on the clean list. 37331da177e4SLinus Torvalds */ 3734bd4c625cSLinus Torvalds int journal_mark_freed(struct reiserfs_transaction_handle *th, 3735a9dd3643SJeff Mahoney struct super_block *sb, b_blocknr_t blocknr) 3736bd4c625cSLinus Torvalds { 3737a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 37381da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn = NULL; 37391da177e4SLinus Torvalds struct buffer_head *bh = NULL; 37401da177e4SLinus Torvalds struct reiserfs_list_bitmap *jb = NULL; 37411da177e4SLinus Torvalds int cleaned = 0; 37421da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 37431da177e4SLinus Torvalds 3744a9dd3643SJeff Mahoney cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr); 37451da177e4SLinus Torvalds if (cn && cn->bh) { 37461da177e4SLinus Torvalds bh = cn->bh; 37471da177e4SLinus Torvalds get_bh(bh); 37481da177e4SLinus Torvalds } 37491da177e4SLinus Torvalds /* if it is journal new, we just remove it from this transaction */ 37501da177e4SLinus Torvalds if (bh && buffer_journal_new(bh)) { 37511da177e4SLinus Torvalds clear_buffer_journal_new(bh); 37521da177e4SLinus Torvalds clear_prepared_bits(bh); 37531da177e4SLinus Torvalds reiserfs_clean_and_file_buffer(bh); 3754a9dd3643SJeff Mahoney cleaned = remove_from_transaction(sb, blocknr, cleaned); 37551da177e4SLinus Torvalds } else { 3756098297b2SJeff Mahoney /* 3757098297b2SJeff Mahoney * set the bit for this block in the journal bitmap 3758098297b2SJeff Mahoney * for this transaction 3759098297b2SJeff Mahoney */ 37601da177e4SLinus Torvalds jb = journal->j_current_jl->j_list_bitmap; 37611da177e4SLinus Torvalds if (!jb) { 3762a9dd3643SJeff Mahoney reiserfs_panic(sb, "journal-1702", 3763c3a9c210SJeff Mahoney "journal_list_bitmap is NULL"); 37641da177e4SLinus Torvalds } 3765a9dd3643SJeff Mahoney set_bit_in_list_bitmap(sb, blocknr, jb); 37661da177e4SLinus Torvalds 37671da177e4SLinus Torvalds /* Note, the entire while loop is not allowed to schedule. */ 37681da177e4SLinus Torvalds 37691da177e4SLinus Torvalds if (bh) { 37701da177e4SLinus Torvalds clear_prepared_bits(bh); 37711da177e4SLinus Torvalds reiserfs_clean_and_file_buffer(bh); 37721da177e4SLinus Torvalds } 3773a9dd3643SJeff Mahoney cleaned = remove_from_transaction(sb, blocknr, cleaned); 37741da177e4SLinus Torvalds 3775098297b2SJeff Mahoney /* 3776098297b2SJeff Mahoney * find all older transactions with this block, 3777098297b2SJeff Mahoney * make sure they don't try to write it out 3778098297b2SJeff Mahoney */ 3779a9dd3643SJeff Mahoney cn = get_journal_hash_dev(sb, journal->j_list_hash_table, 3780bd4c625cSLinus Torvalds blocknr); 37811da177e4SLinus Torvalds while (cn) { 3782a9dd3643SJeff Mahoney if (sb == cn->sb && blocknr == cn->blocknr) { 37831da177e4SLinus Torvalds set_bit(BLOCK_FREED, &cn->state); 37841da177e4SLinus Torvalds if (cn->bh) { 3785098297b2SJeff Mahoney /* 3786098297b2SJeff Mahoney * remove_from_transaction will brelse 3787098297b2SJeff Mahoney * the buffer if it was in the current 3788098297b2SJeff Mahoney * trans 37891da177e4SLinus Torvalds */ 3790098297b2SJeff Mahoney if (!cleaned) { 3791bd4c625cSLinus Torvalds clear_buffer_journal_dirty(cn-> 3792bd4c625cSLinus Torvalds bh); 37931da177e4SLinus Torvalds clear_buffer_dirty(cn->bh); 3794bd4c625cSLinus Torvalds clear_buffer_journal_test(cn-> 3795bd4c625cSLinus Torvalds bh); 37961da177e4SLinus Torvalds cleaned = 1; 37971da177e4SLinus Torvalds put_bh(cn->bh); 3798bd4c625cSLinus Torvalds if (atomic_read 3799a228bf8fSJeff Mahoney (&cn->bh->b_count) < 0) { 3800a9dd3643SJeff Mahoney reiserfs_warning(sb, 380145b03d5eSJeff Mahoney "journal-2138", 380245b03d5eSJeff Mahoney "cn->bh->b_count < 0"); 38031da177e4SLinus Torvalds } 38041da177e4SLinus Torvalds } 3805098297b2SJeff Mahoney /* 3806098297b2SJeff Mahoney * since we are clearing the bh, 3807098297b2SJeff Mahoney * we MUST dec nonzerolen 3808098297b2SJeff Mahoney */ 3809098297b2SJeff Mahoney if (cn->jlist) { 3810a228bf8fSJeff Mahoney atomic_dec(&cn->jlist-> 3811a228bf8fSJeff Mahoney j_nonzerolen); 38121da177e4SLinus Torvalds } 38131da177e4SLinus Torvalds cn->bh = NULL; 38141da177e4SLinus Torvalds } 38151da177e4SLinus Torvalds } 38161da177e4SLinus Torvalds cn = cn->hnext; 38171da177e4SLinus Torvalds } 38181da177e4SLinus Torvalds } 38191da177e4SLinus Torvalds 3820398c95bdSChris Mason if (bh) 3821398c95bdSChris Mason release_buffer_page(bh); /* get_hash grabs the buffer */ 38221da177e4SLinus Torvalds return 0; 38231da177e4SLinus Torvalds } 38241da177e4SLinus Torvalds 3825bd4c625cSLinus Torvalds void reiserfs_update_inode_transaction(struct inode *inode) 3826bd4c625cSLinus Torvalds { 38271da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb); 38281da177e4SLinus Torvalds REISERFS_I(inode)->i_jl = journal->j_current_jl; 38291da177e4SLinus Torvalds REISERFS_I(inode)->i_trans_id = journal->j_trans_id; 38301da177e4SLinus Torvalds } 38311da177e4SLinus Torvalds 38321da177e4SLinus Torvalds /* 38331da177e4SLinus Torvalds * returns -1 on error, 0 if no commits/barriers were done and 1 38341da177e4SLinus Torvalds * if a transaction was actually committed and the barrier was done 38351da177e4SLinus Torvalds */ 38361da177e4SLinus Torvalds static int __commit_trans_jl(struct inode *inode, unsigned long id, 38371da177e4SLinus Torvalds struct reiserfs_journal_list *jl) 38381da177e4SLinus Torvalds { 38391da177e4SLinus Torvalds struct reiserfs_transaction_handle th; 38401da177e4SLinus Torvalds struct super_block *sb = inode->i_sb; 38411da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(sb); 38421da177e4SLinus Torvalds int ret = 0; 38431da177e4SLinus Torvalds 3844098297b2SJeff Mahoney /* 3845098297b2SJeff Mahoney * is it from the current transaction, 3846098297b2SJeff Mahoney * or from an unknown transaction? 3847098297b2SJeff Mahoney */ 38481da177e4SLinus Torvalds if (id == journal->j_trans_id) { 38491da177e4SLinus Torvalds jl = journal->j_current_jl; 3850098297b2SJeff Mahoney /* 3851098297b2SJeff Mahoney * try to let other writers come in and 3852098297b2SJeff Mahoney * grow this transaction 3853098297b2SJeff Mahoney */ 38541da177e4SLinus Torvalds let_transaction_grow(sb, id); 38551da177e4SLinus Torvalds if (journal->j_trans_id != id) { 38561da177e4SLinus Torvalds goto flush_commit_only; 38571da177e4SLinus Torvalds } 38581da177e4SLinus Torvalds 38591da177e4SLinus Torvalds ret = journal_begin(&th, sb, 1); 38601da177e4SLinus Torvalds if (ret) 38611da177e4SLinus Torvalds return ret; 38621da177e4SLinus Torvalds 38631da177e4SLinus Torvalds /* someone might have ended this transaction while we joined */ 38641da177e4SLinus Torvalds if (journal->j_trans_id != id) { 3865bd4c625cSLinus Torvalds reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb), 3866bd4c625cSLinus Torvalds 1); 386709f1b80bSJeff Mahoney journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb)); 386858d85426SJeff Mahoney ret = journal_end(&th); 38691da177e4SLinus Torvalds goto flush_commit_only; 38701da177e4SLinus Torvalds } 38711da177e4SLinus Torvalds 387258d85426SJeff Mahoney ret = journal_end_sync(&th); 38731da177e4SLinus Torvalds if (!ret) 38741da177e4SLinus Torvalds ret = 1; 38751da177e4SLinus Torvalds 38761da177e4SLinus Torvalds } else { 3877098297b2SJeff Mahoney /* 3878098297b2SJeff Mahoney * this gets tricky, we have to make sure the journal list in 38791da177e4SLinus Torvalds * the inode still exists. We know the list is still around 38801da177e4SLinus Torvalds * if we've got a larger transaction id than the oldest list 38811da177e4SLinus Torvalds */ 38821da177e4SLinus Torvalds flush_commit_only: 38831da177e4SLinus Torvalds if (journal_list_still_alive(inode->i_sb, id)) { 38841da177e4SLinus Torvalds /* 38851da177e4SLinus Torvalds * we only set ret to 1 when we know for sure 38861da177e4SLinus Torvalds * the barrier hasn't been started yet on the commit 38871da177e4SLinus Torvalds * block. 38881da177e4SLinus Torvalds */ 38891da177e4SLinus Torvalds if (atomic_read(&jl->j_commit_left) > 1) 38901da177e4SLinus Torvalds ret = 1; 38911da177e4SLinus Torvalds flush_commit_list(sb, jl, 1); 38921da177e4SLinus Torvalds if (journal->j_errno) 38931da177e4SLinus Torvalds ret = journal->j_errno; 38941da177e4SLinus Torvalds } 38951da177e4SLinus Torvalds } 38961da177e4SLinus Torvalds /* otherwise the list is gone, and long since committed */ 38971da177e4SLinus Torvalds return ret; 38981da177e4SLinus Torvalds } 38991da177e4SLinus Torvalds 3900bd4c625cSLinus Torvalds int reiserfs_commit_for_inode(struct inode *inode) 3901bd4c625cSLinus Torvalds { 3902600ed416SJeff Mahoney unsigned int id = REISERFS_I(inode)->i_trans_id; 39031da177e4SLinus Torvalds struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl; 39041da177e4SLinus Torvalds 3905098297b2SJeff Mahoney /* 3906098297b2SJeff Mahoney * for the whole inode, assume unset id means it was 39071da177e4SLinus Torvalds * changed in the current transaction. More conservative 39081da177e4SLinus Torvalds */ 39091da177e4SLinus Torvalds if (!id || !jl) { 39101da177e4SLinus Torvalds reiserfs_update_inode_transaction(inode); 39111da177e4SLinus Torvalds id = REISERFS_I(inode)->i_trans_id; 39121da177e4SLinus Torvalds /* jl will be updated in __commit_trans_jl */ 39131da177e4SLinus Torvalds } 39141da177e4SLinus Torvalds 39151da177e4SLinus Torvalds return __commit_trans_jl(inode, id, jl); 39161da177e4SLinus Torvalds } 39171da177e4SLinus Torvalds 3918a9dd3643SJeff Mahoney void reiserfs_restore_prepared_buffer(struct super_block *sb, 3919bd4c625cSLinus Torvalds struct buffer_head *bh) 3920bd4c625cSLinus Torvalds { 3921a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 3922a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.restore_prepared); 39231da177e4SLinus Torvalds if (!bh) { 39241da177e4SLinus Torvalds return; 39251da177e4SLinus Torvalds } 39261da177e4SLinus Torvalds if (test_clear_buffer_journal_restore_dirty(bh) && 39271da177e4SLinus Torvalds buffer_journal_dirty(bh)) { 39281da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 3929278f6679SJeff Mahoney reiserfs_write_lock(sb); 3930a9dd3643SJeff Mahoney cn = get_journal_hash_dev(sb, 39311da177e4SLinus Torvalds journal->j_list_hash_table, 39321da177e4SLinus Torvalds bh->b_blocknr); 39331da177e4SLinus Torvalds if (cn && can_dirty(cn)) { 39341da177e4SLinus Torvalds set_buffer_journal_test(bh); 39351da177e4SLinus Torvalds mark_buffer_dirty(bh); 39361da177e4SLinus Torvalds } 3937278f6679SJeff Mahoney reiserfs_write_unlock(sb); 39381da177e4SLinus Torvalds } 39391da177e4SLinus Torvalds clear_buffer_journal_prepared(bh); 39401da177e4SLinus Torvalds } 39411da177e4SLinus Torvalds 39421da177e4SLinus Torvalds extern struct tree_balance *cur_tb; 39431da177e4SLinus Torvalds /* 3944098297b2SJeff Mahoney * before we can change a metadata block, we have to make sure it won't 3945098297b2SJeff Mahoney * be written to disk while we are altering it. So, we must: 3946098297b2SJeff Mahoney * clean it 3947098297b2SJeff Mahoney * wait on it. 39481da177e4SLinus Torvalds */ 3949a9dd3643SJeff Mahoney int reiserfs_prepare_for_journal(struct super_block *sb, 3950bd4c625cSLinus Torvalds struct buffer_head *bh, int wait) 3951bd4c625cSLinus Torvalds { 3952a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.prepare); 39531da177e4SLinus Torvalds 3954ca5de404SNick Piggin if (!trylock_buffer(bh)) { 39551da177e4SLinus Torvalds if (!wait) 39561da177e4SLinus Torvalds return 0; 39571da177e4SLinus Torvalds lock_buffer(bh); 39581da177e4SLinus Torvalds } 39591da177e4SLinus Torvalds set_buffer_journal_prepared(bh); 39601da177e4SLinus Torvalds if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) { 39611da177e4SLinus Torvalds clear_buffer_journal_test(bh); 39621da177e4SLinus Torvalds set_buffer_journal_restore_dirty(bh); 39631da177e4SLinus Torvalds } 39641da177e4SLinus Torvalds unlock_buffer(bh); 39651da177e4SLinus Torvalds return 1; 39661da177e4SLinus Torvalds } 39671da177e4SLinus Torvalds 39681da177e4SLinus Torvalds /* 3969098297b2SJeff Mahoney * long and ugly. If flush, will not return until all commit 3970098297b2SJeff Mahoney * blocks and all real buffers in the trans are on disk. 3971098297b2SJeff Mahoney * If no_async, won't return until all commit blocks are on disk. 3972098297b2SJeff Mahoney * 3973098297b2SJeff Mahoney * keep reading, there are comments as you go along 3974098297b2SJeff Mahoney * 3975098297b2SJeff Mahoney * If the journal is aborted, we just clean up. Things like flushing 3976098297b2SJeff Mahoney * journal lists, etc just won't happen. 39771da177e4SLinus Torvalds */ 397858d85426SJeff Mahoney static int do_journal_end(struct reiserfs_transaction_handle *th, int flags) 3979bd4c625cSLinus Torvalds { 398058d85426SJeff Mahoney struct super_block *sb = th->t_super; 3981a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 39821da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn, *next, *jl_cn; 39831da177e4SLinus Torvalds struct reiserfs_journal_cnode *last_cn = NULL; 39841da177e4SLinus Torvalds struct reiserfs_journal_desc *desc; 39851da177e4SLinus Torvalds struct reiserfs_journal_commit *commit; 39861da177e4SLinus Torvalds struct buffer_head *c_bh; /* commit bh */ 39871da177e4SLinus Torvalds struct buffer_head *d_bh; /* desc bh */ 39881da177e4SLinus Torvalds int cur_write_start = 0; /* start index of current log write */ 39891da177e4SLinus Torvalds int old_start; 39901da177e4SLinus Torvalds int i; 3991a44c94a7SAlexander Zarochentsev int flush; 3992a44c94a7SAlexander Zarochentsev int wait_on_commit; 39931da177e4SLinus Torvalds struct reiserfs_journal_list *jl, *temp_jl; 39941da177e4SLinus Torvalds struct list_head *entry, *safe; 39951da177e4SLinus Torvalds unsigned long jindex; 3996600ed416SJeff Mahoney unsigned int commit_trans_id; 39971da177e4SLinus Torvalds int trans_half; 3998278f6679SJeff Mahoney int depth; 39991da177e4SLinus Torvalds 40001da177e4SLinus Torvalds BUG_ON(th->t_refcount > 1); 40011da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 400258d85426SJeff Mahoney BUG_ON(!th->t_super); 40031da177e4SLinus Torvalds 4004098297b2SJeff Mahoney /* 4005098297b2SJeff Mahoney * protect flush_older_commits from doing mistakes if the 4006098297b2SJeff Mahoney * transaction ID counter gets overflowed. 4007098297b2SJeff Mahoney */ 4008600ed416SJeff Mahoney if (th->t_trans_id == ~0U) 4009a44c94a7SAlexander Zarochentsev flags |= FLUSH_ALL | COMMIT_NOW | WAIT; 4010a44c94a7SAlexander Zarochentsev flush = flags & FLUSH_ALL; 4011a44c94a7SAlexander Zarochentsev wait_on_commit = flags & WAIT; 4012a44c94a7SAlexander Zarochentsev 40131da177e4SLinus Torvalds current->journal_info = th->t_handle_save; 4014a9dd3643SJeff Mahoney reiserfs_check_lock_depth(sb, "journal end"); 40151da177e4SLinus Torvalds if (journal->j_len == 0) { 4016a9dd3643SJeff Mahoney reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb), 4017bd4c625cSLinus Torvalds 1); 401809f1b80bSJeff Mahoney journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb)); 40191da177e4SLinus Torvalds } 40201da177e4SLinus Torvalds 4021a9dd3643SJeff Mahoney lock_journal(sb); 40221da177e4SLinus Torvalds if (journal->j_next_full_flush) { 40231da177e4SLinus Torvalds flags |= FLUSH_ALL; 40241da177e4SLinus Torvalds flush = 1; 40251da177e4SLinus Torvalds } 40261da177e4SLinus Torvalds if (journal->j_next_async_flush) { 40271da177e4SLinus Torvalds flags |= COMMIT_NOW | WAIT; 40281da177e4SLinus Torvalds wait_on_commit = 1; 40291da177e4SLinus Torvalds } 40301da177e4SLinus Torvalds 4031098297b2SJeff Mahoney /* 4032098297b2SJeff Mahoney * check_journal_end locks the journal, and unlocks if it does 4033098297b2SJeff Mahoney * not return 1 it tells us if we should continue with the 4034098297b2SJeff Mahoney * journal_end, or just return 40351da177e4SLinus Torvalds */ 403658d85426SJeff Mahoney if (!check_journal_end(th, flags)) { 4037033369d1SArtem Bityutskiy reiserfs_schedule_old_flush(sb); 4038a9dd3643SJeff Mahoney wake_queued_writers(sb); 4039a9dd3643SJeff Mahoney reiserfs_async_progress_wait(sb); 40401da177e4SLinus Torvalds goto out; 40411da177e4SLinus Torvalds } 40421da177e4SLinus Torvalds 40431da177e4SLinus Torvalds /* check_journal_end might set these, check again */ 40441da177e4SLinus Torvalds if (journal->j_next_full_flush) { 40451da177e4SLinus Torvalds flush = 1; 40461da177e4SLinus Torvalds } 40471da177e4SLinus Torvalds 40481da177e4SLinus Torvalds /* 4049098297b2SJeff Mahoney * j must wait means we have to flush the log blocks, and the 4050098297b2SJeff Mahoney * real blocks for this transaction 40511da177e4SLinus Torvalds */ 40521da177e4SLinus Torvalds if (journal->j_must_wait > 0) { 40531da177e4SLinus Torvalds flush = 1; 40541da177e4SLinus Torvalds } 40551da177e4SLinus Torvalds #ifdef REISERFS_PREALLOCATE 4056098297b2SJeff Mahoney /* 4057098297b2SJeff Mahoney * quota ops might need to nest, setup the journal_info pointer 4058098297b2SJeff Mahoney * for them and raise the refcount so that it is > 0. 4059098297b2SJeff Mahoney */ 40601da177e4SLinus Torvalds current->journal_info = th; 4061ef43bc4fSJan Kara th->t_refcount++; 4062098297b2SJeff Mahoney 4063098297b2SJeff Mahoney /* it should not involve new blocks into the transaction */ 4064098297b2SJeff Mahoney reiserfs_discard_all_prealloc(th); 4065098297b2SJeff Mahoney 4066ef43bc4fSJan Kara th->t_refcount--; 40671da177e4SLinus Torvalds current->journal_info = th->t_handle_save; 40681da177e4SLinus Torvalds #endif 40691da177e4SLinus Torvalds 40701da177e4SLinus Torvalds /* setup description block */ 4071bd4c625cSLinus Torvalds d_bh = 4072a9dd3643SJeff Mahoney journal_getblk(sb, 4073a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 4074bd4c625cSLinus Torvalds journal->j_start); 40751da177e4SLinus Torvalds set_buffer_uptodate(d_bh); 40761da177e4SLinus Torvalds desc = (struct reiserfs_journal_desc *)(d_bh)->b_data; 40771da177e4SLinus Torvalds memset(d_bh->b_data, 0, d_bh->b_size); 40781da177e4SLinus Torvalds memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8); 40791da177e4SLinus Torvalds set_desc_trans_id(desc, journal->j_trans_id); 40801da177e4SLinus Torvalds 4081098297b2SJeff Mahoney /* 4082098297b2SJeff Mahoney * setup commit block. Don't write (keep it clean too) this one 4083098297b2SJeff Mahoney * until after everyone else is written 4084098297b2SJeff Mahoney */ 4085a9dd3643SJeff Mahoney c_bh = journal_getblk(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 4086bd4c625cSLinus Torvalds ((journal->j_start + journal->j_len + 4087a9dd3643SJeff Mahoney 1) % SB_ONDISK_JOURNAL_SIZE(sb))); 40881da177e4SLinus Torvalds commit = (struct reiserfs_journal_commit *)c_bh->b_data; 40891da177e4SLinus Torvalds memset(c_bh->b_data, 0, c_bh->b_size); 40901da177e4SLinus Torvalds set_commit_trans_id(commit, journal->j_trans_id); 40911da177e4SLinus Torvalds set_buffer_uptodate(c_bh); 40921da177e4SLinus Torvalds 40931da177e4SLinus Torvalds /* init this journal list */ 40941da177e4SLinus Torvalds jl = journal->j_current_jl; 40951da177e4SLinus Torvalds 4096098297b2SJeff Mahoney /* 4097098297b2SJeff Mahoney * we lock the commit before doing anything because 40981da177e4SLinus Torvalds * we want to make sure nobody tries to run flush_commit_list until 40991da177e4SLinus Torvalds * the new transaction is fully setup, and we've already flushed the 41001da177e4SLinus Torvalds * ordered bh list 41011da177e4SLinus Torvalds */ 41028ebc4232SFrederic Weisbecker reiserfs_mutex_lock_safe(&jl->j_commit_mutex, sb); 41031da177e4SLinus Torvalds 41041da177e4SLinus Torvalds /* save the transaction id in case we need to commit it later */ 41051da177e4SLinus Torvalds commit_trans_id = jl->j_trans_id; 41061da177e4SLinus Torvalds 41071da177e4SLinus Torvalds atomic_set(&jl->j_older_commits_done, 0); 41081da177e4SLinus Torvalds jl->j_trans_id = journal->j_trans_id; 41091da177e4SLinus Torvalds jl->j_timestamp = journal->j_trans_start_time; 41101da177e4SLinus Torvalds jl->j_commit_bh = c_bh; 41111da177e4SLinus Torvalds jl->j_start = journal->j_start; 41121da177e4SLinus Torvalds jl->j_len = journal->j_len; 41131da177e4SLinus Torvalds atomic_set(&jl->j_nonzerolen, journal->j_len); 41141da177e4SLinus Torvalds atomic_set(&jl->j_commit_left, journal->j_len + 2); 41151da177e4SLinus Torvalds jl->j_realblock = NULL; 41161da177e4SLinus Torvalds 4117098297b2SJeff Mahoney /* 4118098297b2SJeff Mahoney * The ENTIRE FOR LOOP MUST not cause schedule to occur. 4119098297b2SJeff Mahoney * for each real block, add it to the journal list hash, 4120098297b2SJeff Mahoney * copy into real block index array in the commit or desc block 41211da177e4SLinus Torvalds */ 4122a9dd3643SJeff Mahoney trans_half = journal_trans_half(sb->s_blocksize); 41231da177e4SLinus Torvalds for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) { 41241da177e4SLinus Torvalds if (buffer_journaled(cn->bh)) { 4125a9dd3643SJeff Mahoney jl_cn = get_cnode(sb); 41261da177e4SLinus Torvalds if (!jl_cn) { 4127a9dd3643SJeff Mahoney reiserfs_panic(sb, "journal-1676", 4128c3a9c210SJeff Mahoney "get_cnode returned NULL"); 41291da177e4SLinus Torvalds } 41301da177e4SLinus Torvalds if (i == 0) { 41311da177e4SLinus Torvalds jl->j_realblock = jl_cn; 41321da177e4SLinus Torvalds } 41331da177e4SLinus Torvalds jl_cn->prev = last_cn; 41341da177e4SLinus Torvalds jl_cn->next = NULL; 41351da177e4SLinus Torvalds if (last_cn) { 41361da177e4SLinus Torvalds last_cn->next = jl_cn; 41371da177e4SLinus Torvalds } 41381da177e4SLinus Torvalds last_cn = jl_cn; 4139098297b2SJeff Mahoney /* 4140098297b2SJeff Mahoney * make sure the block we are trying to log 4141098297b2SJeff Mahoney * is not a block of journal or reserved area 4142098297b2SJeff Mahoney */ 4143bd4c625cSLinus Torvalds if (is_block_in_log_or_reserved_area 4144a9dd3643SJeff Mahoney (sb, cn->bh->b_blocknr)) { 4145a9dd3643SJeff Mahoney reiserfs_panic(sb, "journal-2332", 4146c3a9c210SJeff Mahoney "Trying to log block %lu, " 4147c3a9c210SJeff Mahoney "which is a log block", 4148bd4c625cSLinus Torvalds cn->bh->b_blocknr); 41491da177e4SLinus Torvalds } 41501da177e4SLinus Torvalds jl_cn->blocknr = cn->bh->b_blocknr; 41511da177e4SLinus Torvalds jl_cn->state = 0; 4152a9dd3643SJeff Mahoney jl_cn->sb = sb; 41531da177e4SLinus Torvalds jl_cn->bh = cn->bh; 41541da177e4SLinus Torvalds jl_cn->jlist = jl; 41551da177e4SLinus Torvalds insert_journal_hash(journal->j_list_hash_table, jl_cn); 41561da177e4SLinus Torvalds if (i < trans_half) { 4157bd4c625cSLinus Torvalds desc->j_realblock[i] = 4158bd4c625cSLinus Torvalds cpu_to_le32(cn->bh->b_blocknr); 41591da177e4SLinus Torvalds } else { 4160bd4c625cSLinus Torvalds commit->j_realblock[i - trans_half] = 4161bd4c625cSLinus Torvalds cpu_to_le32(cn->bh->b_blocknr); 41621da177e4SLinus Torvalds } 41631da177e4SLinus Torvalds } else { 41641da177e4SLinus Torvalds i--; 41651da177e4SLinus Torvalds } 41661da177e4SLinus Torvalds } 41671da177e4SLinus Torvalds set_desc_trans_len(desc, journal->j_len); 41681da177e4SLinus Torvalds set_desc_mount_id(desc, journal->j_mount_id); 41691da177e4SLinus Torvalds set_desc_trans_id(desc, journal->j_trans_id); 41701da177e4SLinus Torvalds set_commit_trans_len(commit, journal->j_len); 41711da177e4SLinus Torvalds 4172098297b2SJeff Mahoney /* 4173098297b2SJeff Mahoney * special check in case all buffers in the journal 4174098297b2SJeff Mahoney * were marked for not logging 4175098297b2SJeff Mahoney */ 417614a61442SEric Sesterhenn BUG_ON(journal->j_len == 0); 41771da177e4SLinus Torvalds 4178098297b2SJeff Mahoney /* 4179098297b2SJeff Mahoney * we're about to dirty all the log blocks, mark the description block 41801da177e4SLinus Torvalds * dirty now too. Don't mark the commit block dirty until all the 41811da177e4SLinus Torvalds * others are on disk 41821da177e4SLinus Torvalds */ 41831da177e4SLinus Torvalds mark_buffer_dirty(d_bh); 41841da177e4SLinus Torvalds 4185098297b2SJeff Mahoney /* 4186098297b2SJeff Mahoney * first data block is j_start + 1, so add one to 4187098297b2SJeff Mahoney * cur_write_start wherever you use it 4188098297b2SJeff Mahoney */ 41891da177e4SLinus Torvalds cur_write_start = journal->j_start; 41901da177e4SLinus Torvalds cn = journal->j_first; 41911da177e4SLinus Torvalds jindex = 1; /* start at one so we don't get the desc again */ 41921da177e4SLinus Torvalds while (cn) { 41931da177e4SLinus Torvalds clear_buffer_journal_new(cn->bh); 41941da177e4SLinus Torvalds /* copy all the real blocks into log area. dirty log blocks */ 41951da177e4SLinus Torvalds if (buffer_journaled(cn->bh)) { 41961da177e4SLinus Torvalds struct buffer_head *tmp_bh; 41971da177e4SLinus Torvalds char *addr; 41981da177e4SLinus Torvalds struct page *page; 4199bd4c625cSLinus Torvalds tmp_bh = 4200a9dd3643SJeff Mahoney journal_getblk(sb, 4201a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 4202bd4c625cSLinus Torvalds ((cur_write_start + 4203bd4c625cSLinus Torvalds jindex) % 4204a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb))); 42051da177e4SLinus Torvalds set_buffer_uptodate(tmp_bh); 42061da177e4SLinus Torvalds page = cn->bh->b_page; 42071da177e4SLinus Torvalds addr = kmap(page); 4208bd4c625cSLinus Torvalds memcpy(tmp_bh->b_data, 4209bd4c625cSLinus Torvalds addr + offset_in_page(cn->bh->b_data), 42101da177e4SLinus Torvalds cn->bh->b_size); 42111da177e4SLinus Torvalds kunmap(page); 42121da177e4SLinus Torvalds mark_buffer_dirty(tmp_bh); 42131da177e4SLinus Torvalds jindex++; 42141da177e4SLinus Torvalds set_buffer_journal_dirty(cn->bh); 42151da177e4SLinus Torvalds clear_buffer_journaled(cn->bh); 42161da177e4SLinus Torvalds } else { 4217098297b2SJeff Mahoney /* 4218098297b2SJeff Mahoney * JDirty cleared sometime during transaction. 4219098297b2SJeff Mahoney * don't log this one 4220098297b2SJeff Mahoney */ 4221a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-2048", 422245b03d5eSJeff Mahoney "BAD, buffer in journal hash, " 422345b03d5eSJeff Mahoney "but not JDirty!"); 42241da177e4SLinus Torvalds brelse(cn->bh); 42251da177e4SLinus Torvalds } 42261da177e4SLinus Torvalds next = cn->next; 4227a9dd3643SJeff Mahoney free_cnode(sb, cn); 42281da177e4SLinus Torvalds cn = next; 4229278f6679SJeff Mahoney reiserfs_cond_resched(sb); 42301da177e4SLinus Torvalds } 42311da177e4SLinus Torvalds 4232098297b2SJeff Mahoney /* 4233098297b2SJeff Mahoney * we are done with both the c_bh and d_bh, but 4234098297b2SJeff Mahoney * c_bh must be written after all other commit blocks, 4235098297b2SJeff Mahoney * so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1. 42361da177e4SLinus Torvalds */ 42371da177e4SLinus Torvalds 4238a9dd3643SJeff Mahoney journal->j_current_jl = alloc_journal_list(sb); 42391da177e4SLinus Torvalds 42401da177e4SLinus Torvalds /* now it is safe to insert this transaction on the main list */ 42411da177e4SLinus Torvalds list_add_tail(&jl->j_list, &journal->j_journal_list); 42421da177e4SLinus Torvalds list_add_tail(&jl->j_working_list, &journal->j_working_list); 42431da177e4SLinus Torvalds journal->j_num_work_lists++; 42441da177e4SLinus Torvalds 42451da177e4SLinus Torvalds /* reset journal values for the next transaction */ 42461da177e4SLinus Torvalds old_start = journal->j_start; 4247bd4c625cSLinus Torvalds journal->j_start = 4248bd4c625cSLinus Torvalds (journal->j_start + journal->j_len + 4249a9dd3643SJeff Mahoney 2) % SB_ONDISK_JOURNAL_SIZE(sb); 4250a228bf8fSJeff Mahoney atomic_set(&journal->j_wcount, 0); 42511da177e4SLinus Torvalds journal->j_bcount = 0; 42521da177e4SLinus Torvalds journal->j_last = NULL; 42531da177e4SLinus Torvalds journal->j_first = NULL; 42541da177e4SLinus Torvalds journal->j_len = 0; 42551da177e4SLinus Torvalds journal->j_trans_start_time = 0; 4256a44c94a7SAlexander Zarochentsev /* check for trans_id overflow */ 4257a44c94a7SAlexander Zarochentsev if (++journal->j_trans_id == 0) 4258a44c94a7SAlexander Zarochentsev journal->j_trans_id = 10; 42591da177e4SLinus Torvalds journal->j_current_jl->j_trans_id = journal->j_trans_id; 42601da177e4SLinus Torvalds journal->j_must_wait = 0; 42611da177e4SLinus Torvalds journal->j_len_alloc = 0; 42621da177e4SLinus Torvalds journal->j_next_full_flush = 0; 42631da177e4SLinus Torvalds journal->j_next_async_flush = 0; 4264a9dd3643SJeff Mahoney init_journal_hash(sb); 42651da177e4SLinus Torvalds 4266098297b2SJeff Mahoney /* 4267098297b2SJeff Mahoney * make sure reiserfs_add_jh sees the new current_jl before we 4268098297b2SJeff Mahoney * write out the tails 4269098297b2SJeff Mahoney */ 42701da177e4SLinus Torvalds smp_mb(); 42711da177e4SLinus Torvalds 4272098297b2SJeff Mahoney /* 4273098297b2SJeff Mahoney * tail conversion targets have to hit the disk before we end the 42741da177e4SLinus Torvalds * transaction. Otherwise a later transaction might repack the tail 4275098297b2SJeff Mahoney * before this transaction commits, leaving the data block unflushed 4276098297b2SJeff Mahoney * and clean, if we crash before the later transaction commits, the 4277098297b2SJeff Mahoney * data block is lost. 42781da177e4SLinus Torvalds */ 42791da177e4SLinus Torvalds if (!list_empty(&jl->j_tail_bh_list)) { 4280278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(sb); 42811da177e4SLinus Torvalds write_ordered_buffers(&journal->j_dirty_buffers_lock, 42821da177e4SLinus Torvalds journal, jl, &jl->j_tail_bh_list); 4283278f6679SJeff Mahoney reiserfs_write_lock_nested(sb, depth); 42841da177e4SLinus Torvalds } 428514a61442SEric Sesterhenn BUG_ON(!list_empty(&jl->j_tail_bh_list)); 428690415deaSJeff Mahoney mutex_unlock(&jl->j_commit_mutex); 42871da177e4SLinus Torvalds 4288098297b2SJeff Mahoney /* 4289098297b2SJeff Mahoney * honor the flush wishes from the caller, simple commits can 4290098297b2SJeff Mahoney * be done outside the journal lock, they are done below 4291098297b2SJeff Mahoney * 4292098297b2SJeff Mahoney * if we don't flush the commit list right now, we put it into 4293098297b2SJeff Mahoney * the work queue so the people waiting on the async progress work 4294098297b2SJeff Mahoney * queue don't wait for this proc to flush journal lists and such. 42951da177e4SLinus Torvalds */ 42961da177e4SLinus Torvalds if (flush) { 4297a9dd3643SJeff Mahoney flush_commit_list(sb, jl, 1); 4298a9dd3643SJeff Mahoney flush_journal_list(sb, jl, 1); 429901777836SJan Kara } else if (!(jl->j_state & LIST_COMMIT_PENDING)) { 430001777836SJan Kara /* 430101777836SJan Kara * Avoid queueing work when sb is being shut down. Transaction 430201777836SJan Kara * will be flushed on journal shutdown. 430301777836SJan Kara */ 430401777836SJan Kara if (sb->s_flags & MS_ACTIVE) 4305797d9016SJeff Mahoney queue_delayed_work(REISERFS_SB(sb)->commit_wq, 4306797d9016SJeff Mahoney &journal->j_work, HZ / 10); 430701777836SJan Kara } 43081da177e4SLinus Torvalds 4309098297b2SJeff Mahoney /* 4310098297b2SJeff Mahoney * if the next transaction has any chance of wrapping, flush 4311098297b2SJeff Mahoney * transactions that might get overwritten. If any journal lists 4312098297b2SJeff Mahoney * are very old flush them as well. 43131da177e4SLinus Torvalds */ 43141da177e4SLinus Torvalds first_jl: 43151da177e4SLinus Torvalds list_for_each_safe(entry, safe, &journal->j_journal_list) { 43161da177e4SLinus Torvalds temp_jl = JOURNAL_LIST_ENTRY(entry); 43171da177e4SLinus Torvalds if (journal->j_start <= temp_jl->j_start) { 43181da177e4SLinus Torvalds if ((journal->j_start + journal->j_trans_max + 1) >= 4319bd4c625cSLinus Torvalds temp_jl->j_start) { 4320a9dd3643SJeff Mahoney flush_used_journal_lists(sb, temp_jl); 43211da177e4SLinus Torvalds goto first_jl; 43221da177e4SLinus Torvalds } else if ((journal->j_start + 43231da177e4SLinus Torvalds journal->j_trans_max + 1) < 4324a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb)) { 4325098297b2SJeff Mahoney /* 4326098297b2SJeff Mahoney * if we don't cross into the next 4327098297b2SJeff Mahoney * transaction and we don't wrap, there is 4328098297b2SJeff Mahoney * no way we can overlap any later transactions 43291da177e4SLinus Torvalds * break now 43301da177e4SLinus Torvalds */ 43311da177e4SLinus Torvalds break; 43321da177e4SLinus Torvalds } 43331da177e4SLinus Torvalds } else if ((journal->j_start + 43341da177e4SLinus Torvalds journal->j_trans_max + 1) > 4335a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb)) { 43361da177e4SLinus Torvalds if (((journal->j_start + journal->j_trans_max + 1) % 4337a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb)) >= 4338bd4c625cSLinus Torvalds temp_jl->j_start) { 4339a9dd3643SJeff Mahoney flush_used_journal_lists(sb, temp_jl); 43401da177e4SLinus Torvalds goto first_jl; 43411da177e4SLinus Torvalds } else { 4342098297b2SJeff Mahoney /* 4343098297b2SJeff Mahoney * we don't overlap anything from out start 4344098297b2SJeff Mahoney * to the end of the log, and our wrapped 4345098297b2SJeff Mahoney * portion doesn't overlap anything at 43461da177e4SLinus Torvalds * the start of the log. We can break 43471da177e4SLinus Torvalds */ 43481da177e4SLinus Torvalds break; 43491da177e4SLinus Torvalds } 43501da177e4SLinus Torvalds } 43511da177e4SLinus Torvalds } 43521da177e4SLinus Torvalds 4353bd4c625cSLinus Torvalds journal->j_current_jl->j_list_bitmap = 4354a9dd3643SJeff Mahoney get_list_bitmap(sb, journal->j_current_jl); 43551da177e4SLinus Torvalds 43561da177e4SLinus Torvalds if (!(journal->j_current_jl->j_list_bitmap)) { 4357a9dd3643SJeff Mahoney reiserfs_panic(sb, "journal-1996", 4358c3a9c210SJeff Mahoney "could not get a list bitmap"); 43591da177e4SLinus Torvalds } 43601da177e4SLinus Torvalds 4361a228bf8fSJeff Mahoney atomic_set(&journal->j_jlock, 0); 4362a9dd3643SJeff Mahoney unlock_journal(sb); 43631da177e4SLinus Torvalds /* wake up any body waiting to join. */ 43641da177e4SLinus Torvalds clear_bit(J_WRITERS_QUEUED, &journal->j_state); 4365a228bf8fSJeff Mahoney wake_up(&journal->j_join_wait); 43661da177e4SLinus Torvalds 43671da177e4SLinus Torvalds if (!flush && wait_on_commit && 4368a9dd3643SJeff Mahoney journal_list_still_alive(sb, commit_trans_id)) { 4369a9dd3643SJeff Mahoney flush_commit_list(sb, jl, 1); 43701da177e4SLinus Torvalds } 43711da177e4SLinus Torvalds out: 4372a9dd3643SJeff Mahoney reiserfs_check_lock_depth(sb, "journal end2"); 43731da177e4SLinus Torvalds 43741da177e4SLinus Torvalds memset(th, 0, sizeof(*th)); 4375098297b2SJeff Mahoney /* 4376098297b2SJeff Mahoney * Re-set th->t_super, so we can properly keep track of how many 43771da177e4SLinus Torvalds * persistent transactions there are. We need to do this so if this 4378098297b2SJeff Mahoney * call is part of a failed restart_transaction, we can free it later 4379098297b2SJeff Mahoney */ 4380a9dd3643SJeff Mahoney th->t_super = sb; 43811da177e4SLinus Torvalds 43821da177e4SLinus Torvalds return journal->j_errno; 43831da177e4SLinus Torvalds } 43841da177e4SLinus Torvalds 438532e8b106SJeff Mahoney /* Send the file system read only and refuse new transactions */ 438632e8b106SJeff Mahoney void reiserfs_abort_journal(struct super_block *sb, int errno) 43871da177e4SLinus Torvalds { 43881da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(sb); 43891da177e4SLinus Torvalds if (test_bit(J_ABORTED, &journal->j_state)) 43901da177e4SLinus Torvalds return; 43911da177e4SLinus Torvalds 439232e8b106SJeff Mahoney if (!journal->j_errno) 439332e8b106SJeff Mahoney journal->j_errno = errno; 43941da177e4SLinus Torvalds 43951da177e4SLinus Torvalds sb->s_flags |= MS_RDONLY; 43961da177e4SLinus Torvalds set_bit(J_ABORTED, &journal->j_state); 43971da177e4SLinus Torvalds 43981da177e4SLinus Torvalds #ifdef CONFIG_REISERFS_CHECK 43991da177e4SLinus Torvalds dump_stack(); 44001da177e4SLinus Torvalds #endif 44011da177e4SLinus Torvalds } 4402