1b2441318SGreg Kroah-Hartman // SPDX-License-Identifier: GPL-2.0 21da177e4SLinus Torvalds /* 3098297b2SJeff Mahoney * Write ahead logging implementation copyright Chris Mason 2000 4098297b2SJeff Mahoney * 5098297b2SJeff Mahoney * The background commits make this code very interrelated, and 6098297b2SJeff Mahoney * overly complex. I need to rethink things a bit....The major players: 7098297b2SJeff Mahoney * 8098297b2SJeff Mahoney * journal_begin -- call with the number of blocks you expect to log. 9098297b2SJeff Mahoney * If the current transaction is too 10098297b2SJeff Mahoney * old, it will block until the current transaction is 11098297b2SJeff Mahoney * finished, and then start a new one. 12098297b2SJeff Mahoney * Usually, your transaction will get joined in with 13098297b2SJeff Mahoney * previous ones for speed. 14098297b2SJeff Mahoney * 15098297b2SJeff Mahoney * journal_join -- same as journal_begin, but won't block on the current 16098297b2SJeff Mahoney * transaction regardless of age. Don't ever call 17098297b2SJeff Mahoney * this. Ever. There are only two places it should be 18098297b2SJeff Mahoney * called from, and they are both inside this file. 19098297b2SJeff Mahoney * 20098297b2SJeff Mahoney * journal_mark_dirty -- adds blocks into this transaction. clears any flags 21098297b2SJeff Mahoney * that might make them get sent to disk 22098297b2SJeff Mahoney * and then marks them BH_JDirty. Puts the buffer head 23098297b2SJeff Mahoney * into the current transaction hash. 24098297b2SJeff Mahoney * 25098297b2SJeff Mahoney * journal_end -- if the current transaction is batchable, it does nothing 26098297b2SJeff Mahoney * otherwise, it could do an async/synchronous commit, or 27098297b2SJeff Mahoney * a full flush of all log and real blocks in the 28098297b2SJeff Mahoney * transaction. 29098297b2SJeff Mahoney * 30098297b2SJeff Mahoney * flush_old_commits -- if the current transaction is too old, it is ended and 31098297b2SJeff Mahoney * commit blocks are sent to disk. Forces commit blocks 32098297b2SJeff Mahoney * to disk for all backgrounded commits that have been 33098297b2SJeff Mahoney * around too long. 34098297b2SJeff Mahoney * -- Note, if you call this as an immediate flush from 35098297b2SJeff Mahoney * from within kupdate, it will ignore the immediate flag 361da177e4SLinus Torvalds */ 371da177e4SLinus Torvalds 381da177e4SLinus Torvalds #include <linux/time.h> 396188e10dSMatthew Wilcox #include <linux/semaphore.h> 401da177e4SLinus Torvalds #include <linux/vmalloc.h> 41f466c6fdSAl Viro #include "reiserfs.h" 421da177e4SLinus Torvalds #include <linux/kernel.h> 431da177e4SLinus Torvalds #include <linux/errno.h> 441da177e4SLinus Torvalds #include <linux/fcntl.h> 451da177e4SLinus Torvalds #include <linux/stat.h> 461da177e4SLinus Torvalds #include <linux/string.h> 471da177e4SLinus Torvalds #include <linux/buffer_head.h> 481da177e4SLinus Torvalds #include <linux/workqueue.h> 491da177e4SLinus Torvalds #include <linux/writeback.h> 501da177e4SLinus Torvalds #include <linux/blkdev.h> 513fcfab16SAndrew Morton #include <linux/backing-dev.h> 5290415deaSJeff Mahoney #include <linux/uaccess.h> 535a0e3ad6STejun Heo #include <linux/slab.h> 5490415deaSJeff Mahoney 551da177e4SLinus Torvalds 561da177e4SLinus Torvalds /* gets a struct reiserfs_journal_list * from a list head */ 571da177e4SLinus Torvalds #define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \ 581da177e4SLinus Torvalds j_list)) 591da177e4SLinus Torvalds #define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \ 601da177e4SLinus Torvalds j_working_list)) 611da177e4SLinus Torvalds 62098297b2SJeff Mahoney /* must be correct to keep the desc and commit structs at 4k */ 63098297b2SJeff Mahoney #define JOURNAL_TRANS_HALF 1018 641da177e4SLinus Torvalds #define BUFNR 64 /*read ahead */ 651da177e4SLinus Torvalds 661da177e4SLinus Torvalds /* cnode stat bits. Move these into reiserfs_fs.h */ 671da177e4SLinus Torvalds 68098297b2SJeff Mahoney /* this block was freed, and can't be written. */ 69098297b2SJeff Mahoney #define BLOCK_FREED 2 70098297b2SJeff Mahoney /* this block was freed during this transaction, and can't be written */ 71098297b2SJeff Mahoney #define BLOCK_FREED_HOLDER 3 721da177e4SLinus Torvalds 73098297b2SJeff Mahoney /* used in flush_journal_list */ 74098297b2SJeff Mahoney #define BLOCK_NEEDS_FLUSH 4 751da177e4SLinus Torvalds #define BLOCK_DIRTIED 5 761da177e4SLinus Torvalds 771da177e4SLinus Torvalds /* journal list state bits */ 781da177e4SLinus Torvalds #define LIST_TOUCHED 1 791da177e4SLinus Torvalds #define LIST_DIRTY 2 801da177e4SLinus Torvalds #define LIST_COMMIT_PENDING 4 /* someone will commit this list */ 811da177e4SLinus Torvalds 821da177e4SLinus Torvalds /* flags for do_journal_end */ 831da177e4SLinus Torvalds #define FLUSH_ALL 1 /* flush commit and real blocks */ 841da177e4SLinus Torvalds #define COMMIT_NOW 2 /* end and commit this transaction */ 851da177e4SLinus Torvalds #define WAIT 4 /* wait for the log blocks to hit the disk */ 861da177e4SLinus Torvalds 8758d85426SJeff Mahoney static int do_journal_end(struct reiserfs_transaction_handle *, int flags); 88bd4c625cSLinus Torvalds static int flush_journal_list(struct super_block *s, 89bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl, int flushall); 90bd4c625cSLinus Torvalds static int flush_commit_list(struct super_block *s, 91bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl, int flushall); 921da177e4SLinus Torvalds static int can_dirty(struct reiserfs_journal_cnode *cn); 93bd4c625cSLinus Torvalds static int journal_join(struct reiserfs_transaction_handle *th, 94b491dd17SJeff Mahoney struct super_block *sb); 954385bab1SAl Viro static void release_journal_dev(struct super_block *super, 961da177e4SLinus Torvalds struct reiserfs_journal *journal); 97dc0dde61SHariprasad Kelam static void dirty_one_transaction(struct super_block *s, 981da177e4SLinus Torvalds struct reiserfs_journal_list *jl); 99c4028958SDavid Howells static void flush_async_commits(struct work_struct *work); 1001da177e4SLinus Torvalds static void queue_log_writer(struct super_block *s); 1011da177e4SLinus Torvalds 1021da177e4SLinus Torvalds /* values for join in do_journal_begin_r */ 1031da177e4SLinus Torvalds enum { 1041da177e4SLinus Torvalds JBEGIN_REG = 0, /* regular journal begin */ 105098297b2SJeff Mahoney /* join the running transaction if at all possible */ 106098297b2SJeff Mahoney JBEGIN_JOIN = 1, 107098297b2SJeff Mahoney /* called from cleanup code, ignores aborted flag */ 108098297b2SJeff Mahoney JBEGIN_ABORT = 2, 1091da177e4SLinus Torvalds }; 1101da177e4SLinus Torvalds 1111da177e4SLinus Torvalds static int do_journal_begin_r(struct reiserfs_transaction_handle *th, 112a9dd3643SJeff Mahoney struct super_block *sb, 1131da177e4SLinus Torvalds unsigned long nblocks, int join); 1141da177e4SLinus Torvalds 115a9dd3643SJeff Mahoney static void init_journal_hash(struct super_block *sb) 116bd4c625cSLinus Torvalds { 117a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 118bd4c625cSLinus Torvalds memset(journal->j_hash_table, 0, 119bd4c625cSLinus Torvalds JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *)); 1201da177e4SLinus Torvalds } 1211da177e4SLinus Torvalds 1221da177e4SLinus Torvalds /* 123098297b2SJeff Mahoney * clears BH_Dirty and sticks the buffer on the clean list. Called because 124098297b2SJeff Mahoney * I can't allow refile_buffer to make schedule happen after I've freed a 125098297b2SJeff Mahoney * block. Look at remove_from_transaction and journal_mark_freed for 126098297b2SJeff Mahoney * more details. 1271da177e4SLinus Torvalds */ 128bd4c625cSLinus Torvalds static int reiserfs_clean_and_file_buffer(struct buffer_head *bh) 129bd4c625cSLinus Torvalds { 1301da177e4SLinus Torvalds if (bh) { 1311da177e4SLinus Torvalds clear_buffer_dirty(bh); 1321da177e4SLinus Torvalds clear_buffer_journal_test(bh); 1331da177e4SLinus Torvalds } 1341da177e4SLinus Torvalds return 0; 1351da177e4SLinus Torvalds } 1361da177e4SLinus Torvalds 137bd4c625cSLinus Torvalds static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block 138a9dd3643SJeff Mahoney *sb) 139bd4c625cSLinus Torvalds { 1401da177e4SLinus Torvalds struct reiserfs_bitmap_node *bn; 1411da177e4SLinus Torvalds static int id; 1421da177e4SLinus Torvalds 143d739b42bSPekka Enberg bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS); 1441da177e4SLinus Torvalds if (!bn) { 1451da177e4SLinus Torvalds return NULL; 1461da177e4SLinus Torvalds } 147a9dd3643SJeff Mahoney bn->data = kzalloc(sb->s_blocksize, GFP_NOFS); 1481da177e4SLinus Torvalds if (!bn->data) { 149d739b42bSPekka Enberg kfree(bn); 1501da177e4SLinus Torvalds return NULL; 1511da177e4SLinus Torvalds } 1521da177e4SLinus Torvalds bn->id = id++; 1531da177e4SLinus Torvalds INIT_LIST_HEAD(&bn->list); 1541da177e4SLinus Torvalds return bn; 1551da177e4SLinus Torvalds } 1561da177e4SLinus Torvalds 157a9dd3643SJeff Mahoney static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *sb) 158bd4c625cSLinus Torvalds { 159a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 1601da177e4SLinus Torvalds struct reiserfs_bitmap_node *bn = NULL; 1611da177e4SLinus Torvalds struct list_head *entry = journal->j_bitmap_nodes.next; 1621da177e4SLinus Torvalds 1631da177e4SLinus Torvalds journal->j_used_bitmap_nodes++; 1641da177e4SLinus Torvalds repeat: 1651da177e4SLinus Torvalds 1661da177e4SLinus Torvalds if (entry != &journal->j_bitmap_nodes) { 1671da177e4SLinus Torvalds bn = list_entry(entry, struct reiserfs_bitmap_node, list); 1681da177e4SLinus Torvalds list_del(entry); 169a9dd3643SJeff Mahoney memset(bn->data, 0, sb->s_blocksize); 1701da177e4SLinus Torvalds journal->j_free_bitmap_nodes--; 1711da177e4SLinus Torvalds return bn; 1721da177e4SLinus Torvalds } 173a9dd3643SJeff Mahoney bn = allocate_bitmap_node(sb); 1741da177e4SLinus Torvalds if (!bn) { 1751da177e4SLinus Torvalds yield(); 1761da177e4SLinus Torvalds goto repeat; 1771da177e4SLinus Torvalds } 1781da177e4SLinus Torvalds return bn; 1791da177e4SLinus Torvalds } 180a9dd3643SJeff Mahoney static inline void free_bitmap_node(struct super_block *sb, 181bd4c625cSLinus Torvalds struct reiserfs_bitmap_node *bn) 182bd4c625cSLinus Torvalds { 183a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 1841da177e4SLinus Torvalds journal->j_used_bitmap_nodes--; 1851da177e4SLinus Torvalds if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) { 186d739b42bSPekka Enberg kfree(bn->data); 187d739b42bSPekka Enberg kfree(bn); 1881da177e4SLinus Torvalds } else { 1891da177e4SLinus Torvalds list_add(&bn->list, &journal->j_bitmap_nodes); 1901da177e4SLinus Torvalds journal->j_free_bitmap_nodes++; 1911da177e4SLinus Torvalds } 1921da177e4SLinus Torvalds } 1931da177e4SLinus Torvalds 194a9dd3643SJeff Mahoney static void allocate_bitmap_nodes(struct super_block *sb) 195bd4c625cSLinus Torvalds { 1961da177e4SLinus Torvalds int i; 197a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 1981da177e4SLinus Torvalds struct reiserfs_bitmap_node *bn = NULL; 1991da177e4SLinus Torvalds for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) { 200a9dd3643SJeff Mahoney bn = allocate_bitmap_node(sb); 2011da177e4SLinus Torvalds if (bn) { 2021da177e4SLinus Torvalds list_add(&bn->list, &journal->j_bitmap_nodes); 2031da177e4SLinus Torvalds journal->j_free_bitmap_nodes++; 2041da177e4SLinus Torvalds } else { 205098297b2SJeff Mahoney /* this is ok, we'll try again when more are needed */ 206098297b2SJeff Mahoney break; 2071da177e4SLinus Torvalds } 2081da177e4SLinus Torvalds } 2091da177e4SLinus Torvalds } 2101da177e4SLinus Torvalds 211a9dd3643SJeff Mahoney static int set_bit_in_list_bitmap(struct super_block *sb, 2123ee16670SJeff Mahoney b_blocknr_t block, 213bd4c625cSLinus Torvalds struct reiserfs_list_bitmap *jb) 214bd4c625cSLinus Torvalds { 215a9dd3643SJeff Mahoney unsigned int bmap_nr = block / (sb->s_blocksize << 3); 216a9dd3643SJeff Mahoney unsigned int bit_nr = block % (sb->s_blocksize << 3); 2171da177e4SLinus Torvalds 2181da177e4SLinus Torvalds if (!jb->bitmaps[bmap_nr]) { 219a9dd3643SJeff Mahoney jb->bitmaps[bmap_nr] = get_bitmap_node(sb); 2201da177e4SLinus Torvalds } 2211da177e4SLinus Torvalds set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data); 2221da177e4SLinus Torvalds return 0; 2231da177e4SLinus Torvalds } 2241da177e4SLinus Torvalds 225a9dd3643SJeff Mahoney static void cleanup_bitmap_list(struct super_block *sb, 226bd4c625cSLinus Torvalds struct reiserfs_list_bitmap *jb) 227bd4c625cSLinus Torvalds { 2281da177e4SLinus Torvalds int i; 2291da177e4SLinus Torvalds if (jb->bitmaps == NULL) 2301da177e4SLinus Torvalds return; 2311da177e4SLinus Torvalds 232a9dd3643SJeff Mahoney for (i = 0; i < reiserfs_bmap_count(sb); i++) { 2331da177e4SLinus Torvalds if (jb->bitmaps[i]) { 234a9dd3643SJeff Mahoney free_bitmap_node(sb, jb->bitmaps[i]); 2351da177e4SLinus Torvalds jb->bitmaps[i] = NULL; 2361da177e4SLinus Torvalds } 2371da177e4SLinus Torvalds } 2381da177e4SLinus Torvalds } 2391da177e4SLinus Torvalds 2401da177e4SLinus Torvalds /* 241098297b2SJeff Mahoney * only call this on FS unmount. 2421da177e4SLinus Torvalds */ 243a9dd3643SJeff Mahoney static int free_list_bitmaps(struct super_block *sb, 244bd4c625cSLinus Torvalds struct reiserfs_list_bitmap *jb_array) 245bd4c625cSLinus Torvalds { 2461da177e4SLinus Torvalds int i; 2471da177e4SLinus Torvalds struct reiserfs_list_bitmap *jb; 2481da177e4SLinus Torvalds for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) { 2491da177e4SLinus Torvalds jb = jb_array + i; 2501da177e4SLinus Torvalds jb->journal_list = NULL; 251a9dd3643SJeff Mahoney cleanup_bitmap_list(sb, jb); 2521da177e4SLinus Torvalds vfree(jb->bitmaps); 2531da177e4SLinus Torvalds jb->bitmaps = NULL; 2541da177e4SLinus Torvalds } 2551da177e4SLinus Torvalds return 0; 2561da177e4SLinus Torvalds } 2571da177e4SLinus Torvalds 258a9dd3643SJeff Mahoney static int free_bitmap_nodes(struct super_block *sb) 259bd4c625cSLinus Torvalds { 260a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 2611da177e4SLinus Torvalds struct list_head *next = journal->j_bitmap_nodes.next; 2621da177e4SLinus Torvalds struct reiserfs_bitmap_node *bn; 2631da177e4SLinus Torvalds 2641da177e4SLinus Torvalds while (next != &journal->j_bitmap_nodes) { 2651da177e4SLinus Torvalds bn = list_entry(next, struct reiserfs_bitmap_node, list); 2661da177e4SLinus Torvalds list_del(next); 267d739b42bSPekka Enberg kfree(bn->data); 268d739b42bSPekka Enberg kfree(bn); 2691da177e4SLinus Torvalds next = journal->j_bitmap_nodes.next; 2701da177e4SLinus Torvalds journal->j_free_bitmap_nodes--; 2711da177e4SLinus Torvalds } 2721da177e4SLinus Torvalds 2731da177e4SLinus Torvalds return 0; 2741da177e4SLinus Torvalds } 2751da177e4SLinus Torvalds 2761da177e4SLinus Torvalds /* 277098297b2SJeff Mahoney * get memory for JOURNAL_NUM_BITMAPS worth of bitmaps. 278098297b2SJeff Mahoney * jb_array is the array to be filled in. 2791da177e4SLinus Torvalds */ 280a9dd3643SJeff Mahoney int reiserfs_allocate_list_bitmaps(struct super_block *sb, 2811da177e4SLinus Torvalds struct reiserfs_list_bitmap *jb_array, 2823ee16670SJeff Mahoney unsigned int bmap_nr) 283bd4c625cSLinus Torvalds { 2841da177e4SLinus Torvalds int i; 2851da177e4SLinus Torvalds int failed = 0; 2861da177e4SLinus Torvalds struct reiserfs_list_bitmap *jb; 2871da177e4SLinus Torvalds int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *); 2881da177e4SLinus Torvalds 2891da177e4SLinus Torvalds for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) { 2901da177e4SLinus Torvalds jb = jb_array + i; 2911da177e4SLinus Torvalds jb->journal_list = NULL; 292558feb08SJoe Perches jb->bitmaps = vzalloc(mem); 2931da177e4SLinus Torvalds if (!jb->bitmaps) { 294a9dd3643SJeff Mahoney reiserfs_warning(sb, "clm-2000", "unable to " 29545b03d5eSJeff Mahoney "allocate bitmaps for journal lists"); 2961da177e4SLinus Torvalds failed = 1; 2971da177e4SLinus Torvalds break; 2981da177e4SLinus Torvalds } 2991da177e4SLinus Torvalds } 3001da177e4SLinus Torvalds if (failed) { 301a9dd3643SJeff Mahoney free_list_bitmaps(sb, jb_array); 3021da177e4SLinus Torvalds return -1; 3031da177e4SLinus Torvalds } 3041da177e4SLinus Torvalds return 0; 3051da177e4SLinus Torvalds } 3061da177e4SLinus Torvalds 3071da177e4SLinus Torvalds /* 308098297b2SJeff Mahoney * find an available list bitmap. If you can't find one, flush a commit list 309098297b2SJeff Mahoney * and try again 3101da177e4SLinus Torvalds */ 311a9dd3643SJeff Mahoney static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *sb, 312bd4c625cSLinus Torvalds struct reiserfs_journal_list 313bd4c625cSLinus Torvalds *jl) 314bd4c625cSLinus Torvalds { 3151da177e4SLinus Torvalds int i, j; 316a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 3171da177e4SLinus Torvalds struct reiserfs_list_bitmap *jb = NULL; 3181da177e4SLinus Torvalds 3191da177e4SLinus Torvalds for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) { 3201da177e4SLinus Torvalds i = journal->j_list_bitmap_index; 3211da177e4SLinus Torvalds journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS; 3221da177e4SLinus Torvalds jb = journal->j_list_bitmap + i; 3231da177e4SLinus Torvalds if (journal->j_list_bitmap[i].journal_list) { 324a9dd3643SJeff Mahoney flush_commit_list(sb, 325bd4c625cSLinus Torvalds journal->j_list_bitmap[i]. 326bd4c625cSLinus Torvalds journal_list, 1); 3271da177e4SLinus Torvalds if (!journal->j_list_bitmap[i].journal_list) { 3281da177e4SLinus Torvalds break; 3291da177e4SLinus Torvalds } 3301da177e4SLinus Torvalds } else { 3311da177e4SLinus Torvalds break; 3321da177e4SLinus Torvalds } 3331da177e4SLinus Torvalds } 334098297b2SJeff Mahoney /* double check to make sure if flushed correctly */ 335098297b2SJeff Mahoney if (jb->journal_list) 3361da177e4SLinus Torvalds return NULL; 3371da177e4SLinus Torvalds jb->journal_list = jl; 3381da177e4SLinus Torvalds return jb; 3391da177e4SLinus Torvalds } 3401da177e4SLinus Torvalds 3411da177e4SLinus Torvalds /* 342098297b2SJeff Mahoney * allocates a new chunk of X nodes, and links them all together as a list. 343098297b2SJeff Mahoney * Uses the cnode->next and cnode->prev pointers 344098297b2SJeff Mahoney * returns NULL on failure 3451da177e4SLinus Torvalds */ 346bd4c625cSLinus Torvalds static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes) 347bd4c625cSLinus Torvalds { 3481da177e4SLinus Torvalds struct reiserfs_journal_cnode *head; 3491da177e4SLinus Torvalds int i; 3501da177e4SLinus Torvalds if (num_cnodes <= 0) { 3511da177e4SLinus Torvalds return NULL; 3521da177e4SLinus Torvalds } 353fad953ceSKees Cook head = vzalloc(array_size(num_cnodes, 354fad953ceSKees Cook sizeof(struct reiserfs_journal_cnode))); 3551da177e4SLinus Torvalds if (!head) { 3561da177e4SLinus Torvalds return NULL; 3571da177e4SLinus Torvalds } 3581da177e4SLinus Torvalds head[0].prev = NULL; 3591da177e4SLinus Torvalds head[0].next = head + 1; 3601da177e4SLinus Torvalds for (i = 1; i < num_cnodes; i++) { 3611da177e4SLinus Torvalds head[i].prev = head + (i - 1); 3621da177e4SLinus Torvalds head[i].next = head + (i + 1); /* if last one, overwrite it after the if */ 3631da177e4SLinus Torvalds } 3641da177e4SLinus Torvalds head[num_cnodes - 1].next = NULL; 3651da177e4SLinus Torvalds return head; 3661da177e4SLinus Torvalds } 3671da177e4SLinus Torvalds 368098297b2SJeff Mahoney /* pulls a cnode off the free list, or returns NULL on failure */ 369a9dd3643SJeff Mahoney static struct reiserfs_journal_cnode *get_cnode(struct super_block *sb) 370bd4c625cSLinus Torvalds { 3711da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 372a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 3731da177e4SLinus Torvalds 374a9dd3643SJeff Mahoney reiserfs_check_lock_depth(sb, "get_cnode"); 3751da177e4SLinus Torvalds 3761da177e4SLinus Torvalds if (journal->j_cnode_free <= 0) { 3771da177e4SLinus Torvalds return NULL; 3781da177e4SLinus Torvalds } 3791da177e4SLinus Torvalds journal->j_cnode_used++; 3801da177e4SLinus Torvalds journal->j_cnode_free--; 3811da177e4SLinus Torvalds cn = journal->j_cnode_free_list; 3821da177e4SLinus Torvalds if (!cn) { 3831da177e4SLinus Torvalds return cn; 3841da177e4SLinus Torvalds } 3851da177e4SLinus Torvalds if (cn->next) { 3861da177e4SLinus Torvalds cn->next->prev = NULL; 3871da177e4SLinus Torvalds } 3881da177e4SLinus Torvalds journal->j_cnode_free_list = cn->next; 3891da177e4SLinus Torvalds memset(cn, 0, sizeof(struct reiserfs_journal_cnode)); 3901da177e4SLinus Torvalds return cn; 3911da177e4SLinus Torvalds } 3921da177e4SLinus Torvalds 3931da177e4SLinus Torvalds /* 394098297b2SJeff Mahoney * returns a cnode to the free list 3951da177e4SLinus Torvalds */ 396a9dd3643SJeff Mahoney static void free_cnode(struct super_block *sb, 397bd4c625cSLinus Torvalds struct reiserfs_journal_cnode *cn) 398bd4c625cSLinus Torvalds { 399a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 4001da177e4SLinus Torvalds 401a9dd3643SJeff Mahoney reiserfs_check_lock_depth(sb, "free_cnode"); 4021da177e4SLinus Torvalds 4031da177e4SLinus Torvalds journal->j_cnode_used--; 4041da177e4SLinus Torvalds journal->j_cnode_free++; 4051da177e4SLinus Torvalds /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */ 4061da177e4SLinus Torvalds cn->next = journal->j_cnode_free_list; 4071da177e4SLinus Torvalds if (journal->j_cnode_free_list) { 4081da177e4SLinus Torvalds journal->j_cnode_free_list->prev = cn; 4091da177e4SLinus Torvalds } 4101da177e4SLinus Torvalds cn->prev = NULL; /* not needed with the memset, but I might kill the memset, and forget to do this */ 4111da177e4SLinus Torvalds journal->j_cnode_free_list = cn; 4121da177e4SLinus Torvalds } 4131da177e4SLinus Torvalds 414bd4c625cSLinus Torvalds static void clear_prepared_bits(struct buffer_head *bh) 415bd4c625cSLinus Torvalds { 4161da177e4SLinus Torvalds clear_buffer_journal_prepared(bh); 4171da177e4SLinus Torvalds clear_buffer_journal_restore_dirty(bh); 4181da177e4SLinus Torvalds } 4191da177e4SLinus Torvalds 420098297b2SJeff Mahoney /* 421098297b2SJeff Mahoney * return a cnode with same dev, block number and size in table, 422098297b2SJeff Mahoney * or null if not found 423098297b2SJeff Mahoney */ 424bd4c625cSLinus Torvalds static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct 425bd4c625cSLinus Torvalds super_block 426bd4c625cSLinus Torvalds *sb, 427bd4c625cSLinus Torvalds struct 428bd4c625cSLinus Torvalds reiserfs_journal_cnode 429bd4c625cSLinus Torvalds **table, 4301da177e4SLinus Torvalds long bl) 4311da177e4SLinus Torvalds { 4321da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 4331da177e4SLinus Torvalds cn = journal_hash(table, sb, bl); 4341da177e4SLinus Torvalds while (cn) { 4351da177e4SLinus Torvalds if (cn->blocknr == bl && cn->sb == sb) 4361da177e4SLinus Torvalds return cn; 4371da177e4SLinus Torvalds cn = cn->hnext; 4381da177e4SLinus Torvalds } 4391da177e4SLinus Torvalds return (struct reiserfs_journal_cnode *)0; 4401da177e4SLinus Torvalds } 4411da177e4SLinus Torvalds 4421da177e4SLinus Torvalds /* 443098297b2SJeff Mahoney * this actually means 'can this block be reallocated yet?'. If you set 444098297b2SJeff Mahoney * search_all, a block can only be allocated if it is not in the current 445098297b2SJeff Mahoney * transaction, was not freed by the current transaction, and has no chance 446098297b2SJeff Mahoney * of ever being overwritten by a replay after crashing. 447098297b2SJeff Mahoney * 448098297b2SJeff Mahoney * If you don't set search_all, a block can only be allocated if it is not 449098297b2SJeff Mahoney * in the current transaction. Since deleting a block removes it from the 450098297b2SJeff Mahoney * current transaction, this case should never happen. If you don't set 451098297b2SJeff Mahoney * search_all, make sure you never write the block without logging it. 452098297b2SJeff Mahoney * 453098297b2SJeff Mahoney * next_zero_bit is a suggestion about the next block to try for find_forward. 454098297b2SJeff Mahoney * when bl is rejected because it is set in a journal list bitmap, we search 455098297b2SJeff Mahoney * for the next zero bit in the bitmap that rejected bl. Then, we return 456098297b2SJeff Mahoney * that through next_zero_bit for find_forward to try. 457098297b2SJeff Mahoney * 458098297b2SJeff Mahoney * Just because we return something in next_zero_bit does not mean we won't 459098297b2SJeff Mahoney * reject it on the next call to reiserfs_in_journal 4601da177e4SLinus Torvalds */ 461a9dd3643SJeff Mahoney int reiserfs_in_journal(struct super_block *sb, 4623ee16670SJeff Mahoney unsigned int bmap_nr, int bit_nr, int search_all, 463bd4c625cSLinus Torvalds b_blocknr_t * next_zero_bit) 464bd4c625cSLinus Torvalds { 465a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 4661da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 4671da177e4SLinus Torvalds struct reiserfs_list_bitmap *jb; 4681da177e4SLinus Torvalds int i; 4691da177e4SLinus Torvalds unsigned long bl; 4701da177e4SLinus Torvalds 4711da177e4SLinus Torvalds *next_zero_bit = 0; /* always start this at zero. */ 4721da177e4SLinus Torvalds 473a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.in_journal); 474098297b2SJeff Mahoney /* 475098297b2SJeff Mahoney * If we aren't doing a search_all, this is a metablock, and it 476098297b2SJeff Mahoney * will be logged before use. if we crash before the transaction 477098297b2SJeff Mahoney * that freed it commits, this transaction won't have committed 478098297b2SJeff Mahoney * either, and the block will never be written 4791da177e4SLinus Torvalds */ 4801da177e4SLinus Torvalds if (search_all) { 4811da177e4SLinus Torvalds for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) { 482a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.in_journal_bitmap); 4831da177e4SLinus Torvalds jb = journal->j_list_bitmap + i; 4841da177e4SLinus Torvalds if (jb->journal_list && jb->bitmaps[bmap_nr] && 485bd4c625cSLinus Torvalds test_bit(bit_nr, 486bd4c625cSLinus Torvalds (unsigned long *)jb->bitmaps[bmap_nr]-> 487bd4c625cSLinus Torvalds data)) { 488bd4c625cSLinus Torvalds *next_zero_bit = 489bd4c625cSLinus Torvalds find_next_zero_bit((unsigned long *) 490bd4c625cSLinus Torvalds (jb->bitmaps[bmap_nr]-> 491bd4c625cSLinus Torvalds data), 492a9dd3643SJeff Mahoney sb->s_blocksize << 3, 493bd4c625cSLinus Torvalds bit_nr + 1); 4941da177e4SLinus Torvalds return 1; 4951da177e4SLinus Torvalds } 4961da177e4SLinus Torvalds } 4971da177e4SLinus Torvalds } 4981da177e4SLinus Torvalds 499a9dd3643SJeff Mahoney bl = bmap_nr * (sb->s_blocksize << 3) + bit_nr; 5001da177e4SLinus Torvalds /* is it in any old transactions? */ 501bd4c625cSLinus Torvalds if (search_all 502bd4c625cSLinus Torvalds && (cn = 503a9dd3643SJeff Mahoney get_journal_hash_dev(sb, journal->j_list_hash_table, bl))) { 5041da177e4SLinus Torvalds return 1; 5051da177e4SLinus Torvalds } 5061da177e4SLinus Torvalds 5071da177e4SLinus Torvalds /* is it in the current transaction. This should never happen */ 508a9dd3643SJeff Mahoney if ((cn = get_journal_hash_dev(sb, journal->j_hash_table, bl))) { 5091da177e4SLinus Torvalds BUG(); 5101da177e4SLinus Torvalds return 1; 5111da177e4SLinus Torvalds } 5121da177e4SLinus Torvalds 513a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.in_journal_reusable); 5141da177e4SLinus Torvalds /* safe for reuse */ 5151da177e4SLinus Torvalds return 0; 5161da177e4SLinus Torvalds } 5171da177e4SLinus Torvalds 518098297b2SJeff Mahoney /* insert cn into table */ 519bd4c625cSLinus Torvalds static inline void insert_journal_hash(struct reiserfs_journal_cnode **table, 520bd4c625cSLinus Torvalds struct reiserfs_journal_cnode *cn) 521bd4c625cSLinus Torvalds { 5221da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn_orig; 5231da177e4SLinus Torvalds 5241da177e4SLinus Torvalds cn_orig = journal_hash(table, cn->sb, cn->blocknr); 5251da177e4SLinus Torvalds cn->hnext = cn_orig; 5261da177e4SLinus Torvalds cn->hprev = NULL; 5271da177e4SLinus Torvalds if (cn_orig) { 5281da177e4SLinus Torvalds cn_orig->hprev = cn; 5291da177e4SLinus Torvalds } 5301da177e4SLinus Torvalds journal_hash(table, cn->sb, cn->blocknr) = cn; 5311da177e4SLinus Torvalds } 5321da177e4SLinus Torvalds 5331da177e4SLinus Torvalds /* lock the current transaction */ 534a9dd3643SJeff Mahoney static inline void lock_journal(struct super_block *sb) 535bd4c625cSLinus Torvalds { 536a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.lock_journal); 5378ebc4232SFrederic Weisbecker 5388ebc4232SFrederic Weisbecker reiserfs_mutex_lock_safe(&SB_JOURNAL(sb)->j_mutex, sb); 5391da177e4SLinus Torvalds } 5401da177e4SLinus Torvalds 5411da177e4SLinus Torvalds /* unlock the current transaction */ 542a9dd3643SJeff Mahoney static inline void unlock_journal(struct super_block *sb) 543bd4c625cSLinus Torvalds { 544a9dd3643SJeff Mahoney mutex_unlock(&SB_JOURNAL(sb)->j_mutex); 5451da177e4SLinus Torvalds } 5461da177e4SLinus Torvalds 5471da177e4SLinus Torvalds static inline void get_journal_list(struct reiserfs_journal_list *jl) 5481da177e4SLinus Torvalds { 5491da177e4SLinus Torvalds jl->j_refcount++; 5501da177e4SLinus Torvalds } 5511da177e4SLinus Torvalds 5521da177e4SLinus Torvalds static inline void put_journal_list(struct super_block *s, 5531da177e4SLinus Torvalds struct reiserfs_journal_list *jl) 5541da177e4SLinus Torvalds { 5551da177e4SLinus Torvalds if (jl->j_refcount < 1) { 556c3a9c210SJeff Mahoney reiserfs_panic(s, "journal-2", "trans id %u, refcount at %d", 557bd4c625cSLinus Torvalds jl->j_trans_id, jl->j_refcount); 5581da177e4SLinus Torvalds } 5591da177e4SLinus Torvalds if (--jl->j_refcount == 0) 560d739b42bSPekka Enberg kfree(jl); 5611da177e4SLinus Torvalds } 5621da177e4SLinus Torvalds 5631da177e4SLinus Torvalds /* 564098297b2SJeff Mahoney * this used to be much more involved, and I'm keeping it just in case 565098297b2SJeff Mahoney * things get ugly again. it gets called by flush_commit_list, and 566098297b2SJeff Mahoney * cleans up any data stored about blocks freed during a transaction. 5671da177e4SLinus Torvalds */ 568a9dd3643SJeff Mahoney static void cleanup_freed_for_journal_list(struct super_block *sb, 569bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl) 570bd4c625cSLinus Torvalds { 5711da177e4SLinus Torvalds 5721da177e4SLinus Torvalds struct reiserfs_list_bitmap *jb = jl->j_list_bitmap; 5731da177e4SLinus Torvalds if (jb) { 574a9dd3643SJeff Mahoney cleanup_bitmap_list(sb, jb); 5751da177e4SLinus Torvalds } 5761da177e4SLinus Torvalds jl->j_list_bitmap->journal_list = NULL; 5771da177e4SLinus Torvalds jl->j_list_bitmap = NULL; 5781da177e4SLinus Torvalds } 5791da177e4SLinus Torvalds 5801da177e4SLinus Torvalds static int journal_list_still_alive(struct super_block *s, 581600ed416SJeff Mahoney unsigned int trans_id) 5821da177e4SLinus Torvalds { 5831da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 5841da177e4SLinus Torvalds struct list_head *entry = &journal->j_journal_list; 5851da177e4SLinus Torvalds struct reiserfs_journal_list *jl; 5861da177e4SLinus Torvalds 5871da177e4SLinus Torvalds if (!list_empty(entry)) { 5881da177e4SLinus Torvalds jl = JOURNAL_LIST_ENTRY(entry->next); 5891da177e4SLinus Torvalds if (jl->j_trans_id <= trans_id) { 5901da177e4SLinus Torvalds return 1; 5911da177e4SLinus Torvalds } 5921da177e4SLinus Torvalds } 5931da177e4SLinus Torvalds return 0; 5941da177e4SLinus Torvalds } 5951da177e4SLinus Torvalds 596398c95bdSChris Mason /* 597398c95bdSChris Mason * If page->mapping was null, we failed to truncate this page for 598398c95bdSChris Mason * some reason. Most likely because it was truncated after being 599398c95bdSChris Mason * logged via data=journal. 600398c95bdSChris Mason * 601398c95bdSChris Mason * This does a check to see if the buffer belongs to one of these 602398c95bdSChris Mason * lost pages before doing the final put_bh. If page->mapping was 603398c95bdSChris Mason * null, it tries to free buffers on the page, which should make the 604ea1754a0SKirill A. Shutemov * final put_page drop the page from the lru. 605398c95bdSChris Mason */ 606398c95bdSChris Mason static void release_buffer_page(struct buffer_head *bh) 607398c95bdSChris Mason { 608398c95bdSChris Mason struct page *page = bh->b_page; 609529ae9aaSNick Piggin if (!page->mapping && trylock_page(page)) { 61009cbfeafSKirill A. Shutemov get_page(page); 611398c95bdSChris Mason put_bh(bh); 612398c95bdSChris Mason if (!page->mapping) 613398c95bdSChris Mason try_to_free_buffers(page); 614398c95bdSChris Mason unlock_page(page); 61509cbfeafSKirill A. Shutemov put_page(page); 616398c95bdSChris Mason } else { 617398c95bdSChris Mason put_bh(bh); 618398c95bdSChris Mason } 619398c95bdSChris Mason } 620398c95bdSChris Mason 621bd4c625cSLinus Torvalds static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate) 622bd4c625cSLinus Torvalds { 6231da177e4SLinus Torvalds if (buffer_journaled(bh)) { 62445b03d5eSJeff Mahoney reiserfs_warning(NULL, "clm-2084", 625a1c6f057SDmitry Monakhov "pinned buffer %lu:%pg sent to disk", 626a1c6f057SDmitry Monakhov bh->b_blocknr, bh->b_bdev); 6271da177e4SLinus Torvalds } 6281da177e4SLinus Torvalds if (uptodate) 6291da177e4SLinus Torvalds set_buffer_uptodate(bh); 6301da177e4SLinus Torvalds else 6311da177e4SLinus Torvalds clear_buffer_uptodate(bh); 632398c95bdSChris Mason 6331da177e4SLinus Torvalds unlock_buffer(bh); 634398c95bdSChris Mason release_buffer_page(bh); 6351da177e4SLinus Torvalds } 6361da177e4SLinus Torvalds 637bd4c625cSLinus Torvalds static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate) 638bd4c625cSLinus Torvalds { 6391da177e4SLinus Torvalds if (uptodate) 6401da177e4SLinus Torvalds set_buffer_uptodate(bh); 6411da177e4SLinus Torvalds else 6421da177e4SLinus Torvalds clear_buffer_uptodate(bh); 6431da177e4SLinus Torvalds unlock_buffer(bh); 6441da177e4SLinus Torvalds put_bh(bh); 6451da177e4SLinus Torvalds } 6461da177e4SLinus Torvalds 647bd4c625cSLinus Torvalds static void submit_logged_buffer(struct buffer_head *bh) 648bd4c625cSLinus Torvalds { 6491da177e4SLinus Torvalds get_bh(bh); 6501da177e4SLinus Torvalds bh->b_end_io = reiserfs_end_buffer_io_sync; 6511da177e4SLinus Torvalds clear_buffer_journal_new(bh); 6521da177e4SLinus Torvalds clear_buffer_dirty(bh); 6531da177e4SLinus Torvalds if (!test_clear_buffer_journal_test(bh)) 6541da177e4SLinus Torvalds BUG(); 6551da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 6561da177e4SLinus Torvalds BUG(); 6572a222ca9SMike Christie submit_bh(REQ_OP_WRITE, 0, bh); 6581da177e4SLinus Torvalds } 6591da177e4SLinus Torvalds 660bd4c625cSLinus Torvalds static void submit_ordered_buffer(struct buffer_head *bh) 661bd4c625cSLinus Torvalds { 6621da177e4SLinus Torvalds get_bh(bh); 6631da177e4SLinus Torvalds bh->b_end_io = reiserfs_end_ordered_io; 6641da177e4SLinus Torvalds clear_buffer_dirty(bh); 6651da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 6661da177e4SLinus Torvalds BUG(); 6672a222ca9SMike Christie submit_bh(REQ_OP_WRITE, 0, bh); 6681da177e4SLinus Torvalds } 6691da177e4SLinus Torvalds 6701da177e4SLinus Torvalds #define CHUNK_SIZE 32 6711da177e4SLinus Torvalds struct buffer_chunk { 6721da177e4SLinus Torvalds struct buffer_head *bh[CHUNK_SIZE]; 6731da177e4SLinus Torvalds int nr; 6741da177e4SLinus Torvalds }; 6751da177e4SLinus Torvalds 676bd4c625cSLinus Torvalds static void write_chunk(struct buffer_chunk *chunk) 677bd4c625cSLinus Torvalds { 6781da177e4SLinus Torvalds int i; 6791da177e4SLinus Torvalds for (i = 0; i < chunk->nr; i++) { 6801da177e4SLinus Torvalds submit_logged_buffer(chunk->bh[i]); 6811da177e4SLinus Torvalds } 6821da177e4SLinus Torvalds chunk->nr = 0; 6831da177e4SLinus Torvalds } 6841da177e4SLinus Torvalds 685bd4c625cSLinus Torvalds static void write_ordered_chunk(struct buffer_chunk *chunk) 686bd4c625cSLinus Torvalds { 6871da177e4SLinus Torvalds int i; 6881da177e4SLinus Torvalds for (i = 0; i < chunk->nr; i++) { 6891da177e4SLinus Torvalds submit_ordered_buffer(chunk->bh[i]); 6901da177e4SLinus Torvalds } 6911da177e4SLinus Torvalds chunk->nr = 0; 6921da177e4SLinus Torvalds } 6931da177e4SLinus Torvalds 6941da177e4SLinus Torvalds static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh, 695bd4c625cSLinus Torvalds spinlock_t * lock, void (fn) (struct buffer_chunk *)) 6961da177e4SLinus Torvalds { 6971da177e4SLinus Torvalds int ret = 0; 69814a61442SEric Sesterhenn BUG_ON(chunk->nr >= CHUNK_SIZE); 6991da177e4SLinus Torvalds chunk->bh[chunk->nr++] = bh; 7001da177e4SLinus Torvalds if (chunk->nr >= CHUNK_SIZE) { 7011da177e4SLinus Torvalds ret = 1; 70254cc6ceaSFabian Frederick if (lock) { 7031da177e4SLinus Torvalds spin_unlock(lock); 7041da177e4SLinus Torvalds fn(chunk); 7051da177e4SLinus Torvalds spin_lock(lock); 70654cc6ceaSFabian Frederick } else { 70754cc6ceaSFabian Frederick fn(chunk); 70854cc6ceaSFabian Frederick } 7091da177e4SLinus Torvalds } 7101da177e4SLinus Torvalds return ret; 7111da177e4SLinus Torvalds } 7121da177e4SLinus Torvalds 7131da177e4SLinus Torvalds static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0); 714bd4c625cSLinus Torvalds static struct reiserfs_jh *alloc_jh(void) 715bd4c625cSLinus Torvalds { 7161da177e4SLinus Torvalds struct reiserfs_jh *jh; 7171da177e4SLinus Torvalds while (1) { 7181da177e4SLinus Torvalds jh = kmalloc(sizeof(*jh), GFP_NOFS); 7191da177e4SLinus Torvalds if (jh) { 7201da177e4SLinus Torvalds atomic_inc(&nr_reiserfs_jh); 7211da177e4SLinus Torvalds return jh; 7221da177e4SLinus Torvalds } 7231da177e4SLinus Torvalds yield(); 7241da177e4SLinus Torvalds } 7251da177e4SLinus Torvalds } 7261da177e4SLinus Torvalds 7271da177e4SLinus Torvalds /* 7281da177e4SLinus Torvalds * we want to free the jh when the buffer has been written 7291da177e4SLinus Torvalds * and waited on 7301da177e4SLinus Torvalds */ 731bd4c625cSLinus Torvalds void reiserfs_free_jh(struct buffer_head *bh) 732bd4c625cSLinus Torvalds { 7331da177e4SLinus Torvalds struct reiserfs_jh *jh; 7341da177e4SLinus Torvalds 7351da177e4SLinus Torvalds jh = bh->b_private; 7361da177e4SLinus Torvalds if (jh) { 7371da177e4SLinus Torvalds bh->b_private = NULL; 7381da177e4SLinus Torvalds jh->bh = NULL; 7391da177e4SLinus Torvalds list_del_init(&jh->list); 7401da177e4SLinus Torvalds kfree(jh); 7411da177e4SLinus Torvalds if (atomic_read(&nr_reiserfs_jh) <= 0) 7421da177e4SLinus Torvalds BUG(); 7431da177e4SLinus Torvalds atomic_dec(&nr_reiserfs_jh); 7441da177e4SLinus Torvalds put_bh(bh); 7451da177e4SLinus Torvalds } 7461da177e4SLinus Torvalds } 7471da177e4SLinus Torvalds 7481da177e4SLinus Torvalds static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh, 7491da177e4SLinus Torvalds int tail) 7501da177e4SLinus Torvalds { 7511da177e4SLinus Torvalds struct reiserfs_jh *jh; 7521da177e4SLinus Torvalds 7531da177e4SLinus Torvalds if (bh->b_private) { 7541da177e4SLinus Torvalds spin_lock(&j->j_dirty_buffers_lock); 7551da177e4SLinus Torvalds if (!bh->b_private) { 7561da177e4SLinus Torvalds spin_unlock(&j->j_dirty_buffers_lock); 7571da177e4SLinus Torvalds goto no_jh; 7581da177e4SLinus Torvalds } 7591da177e4SLinus Torvalds jh = bh->b_private; 7601da177e4SLinus Torvalds list_del_init(&jh->list); 7611da177e4SLinus Torvalds } else { 7621da177e4SLinus Torvalds no_jh: 7631da177e4SLinus Torvalds get_bh(bh); 7641da177e4SLinus Torvalds jh = alloc_jh(); 7651da177e4SLinus Torvalds spin_lock(&j->j_dirty_buffers_lock); 766098297b2SJeff Mahoney /* 767098297b2SJeff Mahoney * buffer must be locked for __add_jh, should be able to have 7681da177e4SLinus Torvalds * two adds at the same time 7691da177e4SLinus Torvalds */ 77014a61442SEric Sesterhenn BUG_ON(bh->b_private); 7711da177e4SLinus Torvalds jh->bh = bh; 7721da177e4SLinus Torvalds bh->b_private = jh; 7731da177e4SLinus Torvalds } 7741da177e4SLinus Torvalds jh->jl = j->j_current_jl; 7751da177e4SLinus Torvalds if (tail) 7761da177e4SLinus Torvalds list_add_tail(&jh->list, &jh->jl->j_tail_bh_list); 7771da177e4SLinus Torvalds else { 7781da177e4SLinus Torvalds list_add_tail(&jh->list, &jh->jl->j_bh_list); 7791da177e4SLinus Torvalds } 7801da177e4SLinus Torvalds spin_unlock(&j->j_dirty_buffers_lock); 7811da177e4SLinus Torvalds return 0; 7821da177e4SLinus Torvalds } 7831da177e4SLinus Torvalds 784bd4c625cSLinus Torvalds int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh) 785bd4c625cSLinus Torvalds { 7861da177e4SLinus Torvalds return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1); 7871da177e4SLinus Torvalds } 788bd4c625cSLinus Torvalds int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh) 789bd4c625cSLinus Torvalds { 7901da177e4SLinus Torvalds return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0); 7911da177e4SLinus Torvalds } 7921da177e4SLinus Torvalds 7931da177e4SLinus Torvalds #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list) 7941da177e4SLinus Torvalds static int write_ordered_buffers(spinlock_t * lock, 7951da177e4SLinus Torvalds struct reiserfs_journal *j, 7961da177e4SLinus Torvalds struct reiserfs_journal_list *jl, 7971da177e4SLinus Torvalds struct list_head *list) 7981da177e4SLinus Torvalds { 7991da177e4SLinus Torvalds struct buffer_head *bh; 8001da177e4SLinus Torvalds struct reiserfs_jh *jh; 8011da177e4SLinus Torvalds int ret = j->j_errno; 8021da177e4SLinus Torvalds struct buffer_chunk chunk; 8031da177e4SLinus Torvalds struct list_head tmp; 8041da177e4SLinus Torvalds INIT_LIST_HEAD(&tmp); 8051da177e4SLinus Torvalds 8061da177e4SLinus Torvalds chunk.nr = 0; 8071da177e4SLinus Torvalds spin_lock(lock); 8081da177e4SLinus Torvalds while (!list_empty(list)) { 8091da177e4SLinus Torvalds jh = JH_ENTRY(list->next); 8101da177e4SLinus Torvalds bh = jh->bh; 8111da177e4SLinus Torvalds get_bh(bh); 812ca5de404SNick Piggin if (!trylock_buffer(bh)) { 8131da177e4SLinus Torvalds if (!buffer_dirty(bh)) { 814f116629dSAkinobu Mita list_move(&jh->list, &tmp); 8151da177e4SLinus Torvalds goto loop_next; 8161da177e4SLinus Torvalds } 8171da177e4SLinus Torvalds spin_unlock(lock); 8181da177e4SLinus Torvalds if (chunk.nr) 8191da177e4SLinus Torvalds write_ordered_chunk(&chunk); 8201da177e4SLinus Torvalds wait_on_buffer(bh); 8211da177e4SLinus Torvalds cond_resched(); 8221da177e4SLinus Torvalds spin_lock(lock); 8231da177e4SLinus Torvalds goto loop_next; 8241da177e4SLinus Torvalds } 825098297b2SJeff Mahoney /* 826098297b2SJeff Mahoney * in theory, dirty non-uptodate buffers should never get here, 8273d4492f8SChris Mason * but the upper layer io error paths still have a few quirks. 8283d4492f8SChris Mason * Handle them here as gracefully as we can 8293d4492f8SChris Mason */ 8303d4492f8SChris Mason if (!buffer_uptodate(bh) && buffer_dirty(bh)) { 8313d4492f8SChris Mason clear_buffer_dirty(bh); 8323d4492f8SChris Mason ret = -EIO; 8333d4492f8SChris Mason } 8341da177e4SLinus Torvalds if (buffer_dirty(bh)) { 835f116629dSAkinobu Mita list_move(&jh->list, &tmp); 8361da177e4SLinus Torvalds add_to_chunk(&chunk, bh, lock, write_ordered_chunk); 8371da177e4SLinus Torvalds } else { 8381da177e4SLinus Torvalds reiserfs_free_jh(bh); 8391da177e4SLinus Torvalds unlock_buffer(bh); 8401da177e4SLinus Torvalds } 8411da177e4SLinus Torvalds loop_next: 8421da177e4SLinus Torvalds put_bh(bh); 8431da177e4SLinus Torvalds cond_resched_lock(lock); 8441da177e4SLinus Torvalds } 8451da177e4SLinus Torvalds if (chunk.nr) { 8461da177e4SLinus Torvalds spin_unlock(lock); 8471da177e4SLinus Torvalds write_ordered_chunk(&chunk); 8481da177e4SLinus Torvalds spin_lock(lock); 8491da177e4SLinus Torvalds } 8501da177e4SLinus Torvalds while (!list_empty(&tmp)) { 8511da177e4SLinus Torvalds jh = JH_ENTRY(tmp.prev); 8521da177e4SLinus Torvalds bh = jh->bh; 8531da177e4SLinus Torvalds get_bh(bh); 8541da177e4SLinus Torvalds reiserfs_free_jh(bh); 8551da177e4SLinus Torvalds 8561da177e4SLinus Torvalds if (buffer_locked(bh)) { 8571da177e4SLinus Torvalds spin_unlock(lock); 8581da177e4SLinus Torvalds wait_on_buffer(bh); 8591da177e4SLinus Torvalds spin_lock(lock); 8601da177e4SLinus Torvalds } 8611da177e4SLinus Torvalds if (!buffer_uptodate(bh)) { 8621da177e4SLinus Torvalds ret = -EIO; 8631da177e4SLinus Torvalds } 864098297b2SJeff Mahoney /* 865098297b2SJeff Mahoney * ugly interaction with invalidatepage here. 866098297b2SJeff Mahoney * reiserfs_invalidate_page will pin any buffer that has a 867098297b2SJeff Mahoney * valid journal head from an older transaction. If someone 868098297b2SJeff Mahoney * else sets our buffer dirty after we write it in the first 869098297b2SJeff Mahoney * loop, and then someone truncates the page away, nobody 870098297b2SJeff Mahoney * will ever write the buffer. We're safe if we write the 871098297b2SJeff Mahoney * page one last time after freeing the journal header. 872d62b1b87SChris Mason */ 873d62b1b87SChris Mason if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) { 874d62b1b87SChris Mason spin_unlock(lock); 875dfec8a14SMike Christie ll_rw_block(REQ_OP_WRITE, 0, 1, &bh); 876d62b1b87SChris Mason spin_lock(lock); 877d62b1b87SChris Mason } 8781da177e4SLinus Torvalds put_bh(bh); 8791da177e4SLinus Torvalds cond_resched_lock(lock); 8801da177e4SLinus Torvalds } 8811da177e4SLinus Torvalds spin_unlock(lock); 8821da177e4SLinus Torvalds return ret; 8831da177e4SLinus Torvalds } 8841da177e4SLinus Torvalds 885bd4c625cSLinus Torvalds static int flush_older_commits(struct super_block *s, 886bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl) 887bd4c625cSLinus Torvalds { 8881da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 8891da177e4SLinus Torvalds struct reiserfs_journal_list *other_jl; 8901da177e4SLinus Torvalds struct reiserfs_journal_list *first_jl; 8911da177e4SLinus Torvalds struct list_head *entry; 892600ed416SJeff Mahoney unsigned int trans_id = jl->j_trans_id; 893600ed416SJeff Mahoney unsigned int other_trans_id; 8941da177e4SLinus Torvalds 8951da177e4SLinus Torvalds find_first: 8961da177e4SLinus Torvalds /* 8971da177e4SLinus Torvalds * first we walk backwards to find the oldest uncommitted transation 8981da177e4SLinus Torvalds */ 8991da177e4SLinus Torvalds first_jl = jl; 9001da177e4SLinus Torvalds entry = jl->j_list.prev; 9011da177e4SLinus Torvalds while (1) { 9021da177e4SLinus Torvalds other_jl = JOURNAL_LIST_ENTRY(entry); 9031da177e4SLinus Torvalds if (entry == &journal->j_journal_list || 9041da177e4SLinus Torvalds atomic_read(&other_jl->j_older_commits_done)) 9051da177e4SLinus Torvalds break; 9061da177e4SLinus Torvalds 9071da177e4SLinus Torvalds first_jl = other_jl; 9081da177e4SLinus Torvalds entry = other_jl->j_list.prev; 9091da177e4SLinus Torvalds } 9101da177e4SLinus Torvalds 9111da177e4SLinus Torvalds /* if we didn't find any older uncommitted transactions, return now */ 9121da177e4SLinus Torvalds if (first_jl == jl) { 9131da177e4SLinus Torvalds return 0; 9141da177e4SLinus Torvalds } 9151da177e4SLinus Torvalds 9161da177e4SLinus Torvalds entry = &first_jl->j_list; 9171da177e4SLinus Torvalds while (1) { 9181da177e4SLinus Torvalds other_jl = JOURNAL_LIST_ENTRY(entry); 9191da177e4SLinus Torvalds other_trans_id = other_jl->j_trans_id; 9201da177e4SLinus Torvalds 9211da177e4SLinus Torvalds if (other_trans_id < trans_id) { 9221da177e4SLinus Torvalds if (atomic_read(&other_jl->j_commit_left) != 0) { 9231da177e4SLinus Torvalds flush_commit_list(s, other_jl, 0); 9241da177e4SLinus Torvalds 9251da177e4SLinus Torvalds /* list we were called with is gone, return */ 9261da177e4SLinus Torvalds if (!journal_list_still_alive(s, trans_id)) 9271da177e4SLinus Torvalds return 1; 9281da177e4SLinus Torvalds 929098297b2SJeff Mahoney /* 930098297b2SJeff Mahoney * the one we just flushed is gone, this means 931098297b2SJeff Mahoney * all older lists are also gone, so first_jl 932098297b2SJeff Mahoney * is no longer valid either. Go back to the 933098297b2SJeff Mahoney * beginning. 9341da177e4SLinus Torvalds */ 935bd4c625cSLinus Torvalds if (!journal_list_still_alive 936bd4c625cSLinus Torvalds (s, other_trans_id)) { 9371da177e4SLinus Torvalds goto find_first; 9381da177e4SLinus Torvalds } 9391da177e4SLinus Torvalds } 9401da177e4SLinus Torvalds entry = entry->next; 9411da177e4SLinus Torvalds if (entry == &journal->j_journal_list) 9421da177e4SLinus Torvalds return 0; 9431da177e4SLinus Torvalds } else { 9441da177e4SLinus Torvalds return 0; 9451da177e4SLinus Torvalds } 9461da177e4SLinus Torvalds } 9471da177e4SLinus Torvalds return 0; 9481da177e4SLinus Torvalds } 949deba0f49SAdrian Bunk 950deba0f49SAdrian Bunk static int reiserfs_async_progress_wait(struct super_block *s) 951bd4c625cSLinus Torvalds { 9521da177e4SLinus Torvalds struct reiserfs_journal *j = SB_JOURNAL(s); 9538ebc4232SFrederic Weisbecker 9548ebc4232SFrederic Weisbecker if (atomic_read(&j->j_async_throttle)) { 955278f6679SJeff Mahoney int depth; 956278f6679SJeff Mahoney 957278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(s); 9588aa7e847SJens Axboe congestion_wait(BLK_RW_ASYNC, HZ / 10); 959278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 9608ebc4232SFrederic Weisbecker } 9618ebc4232SFrederic Weisbecker 9621da177e4SLinus Torvalds return 0; 9631da177e4SLinus Torvalds } 9641da177e4SLinus Torvalds 9651da177e4SLinus Torvalds /* 966098297b2SJeff Mahoney * if this journal list still has commit blocks unflushed, send them to disk. 967098297b2SJeff Mahoney * 968098297b2SJeff Mahoney * log areas must be flushed in order (transaction 2 can't commit before 969098297b2SJeff Mahoney * transaction 1) Before the commit block can by written, every other log 970098297b2SJeff Mahoney * block must be safely on disk 9711da177e4SLinus Torvalds */ 972bd4c625cSLinus Torvalds static int flush_commit_list(struct super_block *s, 973bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl, int flushall) 974bd4c625cSLinus Torvalds { 9751da177e4SLinus Torvalds int i; 9763ee16670SJeff Mahoney b_blocknr_t bn; 9771da177e4SLinus Torvalds struct buffer_head *tbh = NULL; 978600ed416SJeff Mahoney unsigned int trans_id = jl->j_trans_id; 9791da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 9801da177e4SLinus Torvalds int retval = 0; 981e0e851cfSChris Mason int write_len; 982278f6679SJeff Mahoney int depth; 9831da177e4SLinus Torvalds 9841da177e4SLinus Torvalds reiserfs_check_lock_depth(s, "flush_commit_list"); 9851da177e4SLinus Torvalds 9861da177e4SLinus Torvalds if (atomic_read(&jl->j_older_commits_done)) { 9871da177e4SLinus Torvalds return 0; 9881da177e4SLinus Torvalds } 9891da177e4SLinus Torvalds 990098297b2SJeff Mahoney /* 991098297b2SJeff Mahoney * before we can put our commit blocks on disk, we have to make 992098297b2SJeff Mahoney * sure everyone older than us is on disk too 9931da177e4SLinus Torvalds */ 9941da177e4SLinus Torvalds BUG_ON(jl->j_len <= 0); 9951da177e4SLinus Torvalds BUG_ON(trans_id == journal->j_trans_id); 9961da177e4SLinus Torvalds 9971da177e4SLinus Torvalds get_journal_list(jl); 9981da177e4SLinus Torvalds if (flushall) { 9991da177e4SLinus Torvalds if (flush_older_commits(s, jl) == 1) { 1000098297b2SJeff Mahoney /* 1001098297b2SJeff Mahoney * list disappeared during flush_older_commits. 1002098297b2SJeff Mahoney * return 1003098297b2SJeff Mahoney */ 10041da177e4SLinus Torvalds goto put_jl; 10051da177e4SLinus Torvalds } 10061da177e4SLinus Torvalds } 10071da177e4SLinus Torvalds 10081da177e4SLinus Torvalds /* make sure nobody is trying to flush this one at the same time */ 10098ebc4232SFrederic Weisbecker reiserfs_mutex_lock_safe(&jl->j_commit_mutex, s); 10108ebc4232SFrederic Weisbecker 10111da177e4SLinus Torvalds if (!journal_list_still_alive(s, trans_id)) { 101290415deaSJeff Mahoney mutex_unlock(&jl->j_commit_mutex); 10131da177e4SLinus Torvalds goto put_jl; 10141da177e4SLinus Torvalds } 10151da177e4SLinus Torvalds BUG_ON(jl->j_trans_id == 0); 10161da177e4SLinus Torvalds 10171da177e4SLinus Torvalds /* this commit is done, exit */ 1018a228bf8fSJeff Mahoney if (atomic_read(&jl->j_commit_left) <= 0) { 10191da177e4SLinus Torvalds if (flushall) { 1020a228bf8fSJeff Mahoney atomic_set(&jl->j_older_commits_done, 1); 10211da177e4SLinus Torvalds } 102290415deaSJeff Mahoney mutex_unlock(&jl->j_commit_mutex); 10231da177e4SLinus Torvalds goto put_jl; 10241da177e4SLinus Torvalds } 10251da177e4SLinus Torvalds 10261da177e4SLinus Torvalds if (!list_empty(&jl->j_bh_list)) { 10273d4492f8SChris Mason int ret; 10288ebc4232SFrederic Weisbecker 10298ebc4232SFrederic Weisbecker /* 10308ebc4232SFrederic Weisbecker * We might sleep in numerous places inside 10318ebc4232SFrederic Weisbecker * write_ordered_buffers. Relax the write lock. 10328ebc4232SFrederic Weisbecker */ 1033278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(s); 10343d4492f8SChris Mason ret = write_ordered_buffers(&journal->j_dirty_buffers_lock, 10351da177e4SLinus Torvalds journal, jl, &jl->j_bh_list); 10363d4492f8SChris Mason if (ret < 0 && retval == 0) 10373d4492f8SChris Mason retval = ret; 1038278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 10391da177e4SLinus Torvalds } 10401da177e4SLinus Torvalds BUG_ON(!list_empty(&jl->j_bh_list)); 10411da177e4SLinus Torvalds /* 10421da177e4SLinus Torvalds * for the description block and all the log blocks, submit any buffers 1043e0e851cfSChris Mason * that haven't already reached the disk. Try to write at least 256 1044e0e851cfSChris Mason * log blocks. later on, we will only wait on blocks that correspond 1045e0e851cfSChris Mason * to this transaction, but while we're unplugging we might as well 1046e0e851cfSChris Mason * get a chunk of data on there. 10471da177e4SLinus Torvalds */ 10481da177e4SLinus Torvalds atomic_inc(&journal->j_async_throttle); 1049e0e851cfSChris Mason write_len = jl->j_len + 1; 1050e0e851cfSChris Mason if (write_len < 256) 1051e0e851cfSChris Mason write_len = 256; 1052e0e851cfSChris Mason for (i = 0 ; i < write_len ; i++) { 10531da177e4SLinus Torvalds bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) % 10541da177e4SLinus Torvalds SB_ONDISK_JOURNAL_SIZE(s); 10551da177e4SLinus Torvalds tbh = journal_find_get_block(s, bn); 1056e0e851cfSChris Mason if (tbh) { 10576e3647acSFrederic Weisbecker if (buffer_dirty(tbh)) { 1058278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(s); 1059dfec8a14SMike Christie ll_rw_block(REQ_OP_WRITE, 0, 1, &tbh); 1060278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 10616e3647acSFrederic Weisbecker } 10621da177e4SLinus Torvalds put_bh(tbh) ; 10631da177e4SLinus Torvalds } 1064e0e851cfSChris Mason } 10651da177e4SLinus Torvalds atomic_dec(&journal->j_async_throttle); 10661da177e4SLinus Torvalds 10671da177e4SLinus Torvalds for (i = 0; i < (jl->j_len + 1); i++) { 10681da177e4SLinus Torvalds bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + 10691da177e4SLinus Torvalds (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s); 10701da177e4SLinus Torvalds tbh = journal_find_get_block(s, bn); 10718ebc4232SFrederic Weisbecker 1072278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(s); 1073278f6679SJeff Mahoney __wait_on_buffer(tbh); 1074278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 1075098297b2SJeff Mahoney /* 1076098297b2SJeff Mahoney * since we're using ll_rw_blk above, it might have skipped 1077098297b2SJeff Mahoney * over a locked buffer. Double check here 1078098297b2SJeff Mahoney */ 10798ebc4232SFrederic Weisbecker /* redundant, sync_dirty_buffer() checks */ 10808ebc4232SFrederic Weisbecker if (buffer_dirty(tbh)) { 1081278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(s); 10821da177e4SLinus Torvalds sync_dirty_buffer(tbh); 1083278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 10848ebc4232SFrederic Weisbecker } 10851da177e4SLinus Torvalds if (unlikely(!buffer_uptodate(tbh))) { 10861da177e4SLinus Torvalds #ifdef CONFIG_REISERFS_CHECK 108745b03d5eSJeff Mahoney reiserfs_warning(s, "journal-601", 108845b03d5eSJeff Mahoney "buffer write failed"); 10891da177e4SLinus Torvalds #endif 10901da177e4SLinus Torvalds retval = -EIO; 10911da177e4SLinus Torvalds } 1092098297b2SJeff Mahoney /* once for journal_find_get_block */ 1093098297b2SJeff Mahoney put_bh(tbh); 1094098297b2SJeff Mahoney /* once due to original getblk in do_journal_end */ 1095098297b2SJeff Mahoney put_bh(tbh); 1096a228bf8fSJeff Mahoney atomic_dec(&jl->j_commit_left); 10971da177e4SLinus Torvalds } 10981da177e4SLinus Torvalds 1099a228bf8fSJeff Mahoney BUG_ON(atomic_read(&jl->j_commit_left) != 1); 11001da177e4SLinus Torvalds 1101098297b2SJeff Mahoney /* 1102098297b2SJeff Mahoney * If there was a write error in the journal - we can't commit 11035d5e8156SJeff Mahoney * this transaction - it will be invalid and, if successful, 1104beb7dd86SRobert P. J. Day * will just end up propagating the write error out to 1105098297b2SJeff Mahoney * the file system. 1106098297b2SJeff Mahoney */ 11075d5e8156SJeff Mahoney if (likely(!retval && !reiserfs_is_journal_aborted (journal))) { 11081da177e4SLinus Torvalds if (buffer_dirty(jl->j_commit_bh)) 11091da177e4SLinus Torvalds BUG(); 11101da177e4SLinus Torvalds mark_buffer_dirty(jl->j_commit_bh) ; 1111278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(s); 11127cd33ad2SChristoph Hellwig if (reiserfs_barrier_flush(s)) 111370fd7614SChristoph Hellwig __sync_dirty_buffer(jl->j_commit_bh, 1114d8747d64SJan Kara REQ_SYNC | REQ_PREFLUSH | REQ_FUA); 11157cd33ad2SChristoph Hellwig else 11161da177e4SLinus Torvalds sync_dirty_buffer(jl->j_commit_bh); 1117278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 11185d5e8156SJeff Mahoney } 11191da177e4SLinus Torvalds 1120098297b2SJeff Mahoney /* 1121098297b2SJeff Mahoney * If there was a write error in the journal - we can't commit this 11221da177e4SLinus Torvalds * transaction - it will be invalid and, if successful, will just end 1123098297b2SJeff Mahoney * up propagating the write error out to the filesystem. 1124098297b2SJeff Mahoney */ 11251da177e4SLinus Torvalds if (unlikely(!buffer_uptodate(jl->j_commit_bh))) { 11261da177e4SLinus Torvalds #ifdef CONFIG_REISERFS_CHECK 112745b03d5eSJeff Mahoney reiserfs_warning(s, "journal-615", "buffer write failed"); 11281da177e4SLinus Torvalds #endif 11291da177e4SLinus Torvalds retval = -EIO; 11301da177e4SLinus Torvalds } 11311da177e4SLinus Torvalds bforget(jl->j_commit_bh); 11321da177e4SLinus Torvalds if (journal->j_last_commit_id != 0 && 11331da177e4SLinus Torvalds (jl->j_trans_id - journal->j_last_commit_id) != 1) { 113445b03d5eSJeff Mahoney reiserfs_warning(s, "clm-2200", "last commit %lu, current %lu", 1135bd4c625cSLinus Torvalds journal->j_last_commit_id, jl->j_trans_id); 11361da177e4SLinus Torvalds } 11371da177e4SLinus Torvalds journal->j_last_commit_id = jl->j_trans_id; 11381da177e4SLinus Torvalds 1139098297b2SJeff Mahoney /* 1140098297b2SJeff Mahoney * now, every commit block is on the disk. It is safe to allow 1141098297b2SJeff Mahoney * blocks freed during this transaction to be reallocated 1142098297b2SJeff Mahoney */ 11431da177e4SLinus Torvalds cleanup_freed_for_journal_list(s, jl); 11441da177e4SLinus Torvalds 11451da177e4SLinus Torvalds retval = retval ? retval : journal->j_errno; 11461da177e4SLinus Torvalds 11471da177e4SLinus Torvalds /* mark the metadata dirty */ 11481da177e4SLinus Torvalds if (!retval) 11491da177e4SLinus Torvalds dirty_one_transaction(s, jl); 1150a228bf8fSJeff Mahoney atomic_dec(&jl->j_commit_left); 11511da177e4SLinus Torvalds 11521da177e4SLinus Torvalds if (flushall) { 1153a228bf8fSJeff Mahoney atomic_set(&jl->j_older_commits_done, 1); 11541da177e4SLinus Torvalds } 115590415deaSJeff Mahoney mutex_unlock(&jl->j_commit_mutex); 11561da177e4SLinus Torvalds put_jl: 11571da177e4SLinus Torvalds put_journal_list(s, jl); 11581da177e4SLinus Torvalds 11591da177e4SLinus Torvalds if (retval) 1160bd4c625cSLinus Torvalds reiserfs_abort(s, retval, "Journal write error in %s", 1161fbe5498bSHarvey Harrison __func__); 11621da177e4SLinus Torvalds return retval; 11631da177e4SLinus Torvalds } 11641da177e4SLinus Torvalds 11651da177e4SLinus Torvalds /* 1166098297b2SJeff Mahoney * flush_journal_list frequently needs to find a newer transaction for a 1167098297b2SJeff Mahoney * given block. This does that, or returns NULL if it can't find anything 11681da177e4SLinus Torvalds */ 1169bd4c625cSLinus Torvalds static struct reiserfs_journal_list *find_newer_jl_for_cn(struct 1170bd4c625cSLinus Torvalds reiserfs_journal_cnode 1171bd4c625cSLinus Torvalds *cn) 1172bd4c625cSLinus Torvalds { 11731da177e4SLinus Torvalds struct super_block *sb = cn->sb; 11741da177e4SLinus Torvalds b_blocknr_t blocknr = cn->blocknr; 11751da177e4SLinus Torvalds 11761da177e4SLinus Torvalds cn = cn->hprev; 11771da177e4SLinus Torvalds while (cn) { 11781da177e4SLinus Torvalds if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) { 11791da177e4SLinus Torvalds return cn->jlist; 11801da177e4SLinus Torvalds } 11811da177e4SLinus Torvalds cn = cn->hprev; 11821da177e4SLinus Torvalds } 11831da177e4SLinus Torvalds return NULL; 11841da177e4SLinus Torvalds } 11851da177e4SLinus Torvalds 1186bd4c625cSLinus Torvalds static void remove_journal_hash(struct super_block *, 1187bd4c625cSLinus Torvalds struct reiserfs_journal_cnode **, 1188bd4c625cSLinus Torvalds struct reiserfs_journal_list *, unsigned long, 1189bd4c625cSLinus Torvalds int); 11901da177e4SLinus Torvalds 11911da177e4SLinus Torvalds /* 1192098297b2SJeff Mahoney * once all the real blocks have been flushed, it is safe to remove them 1193098297b2SJeff Mahoney * from the journal list for this transaction. Aside from freeing the 1194098297b2SJeff Mahoney * cnode, this also allows the block to be reallocated for data blocks 1195098297b2SJeff Mahoney * if it had been deleted. 11961da177e4SLinus Torvalds */ 1197a9dd3643SJeff Mahoney static void remove_all_from_journal_list(struct super_block *sb, 1198bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl, 1199bd4c625cSLinus Torvalds int debug) 1200bd4c625cSLinus Torvalds { 1201a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 12021da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn, *last; 12031da177e4SLinus Torvalds cn = jl->j_realblock; 12041da177e4SLinus Torvalds 1205098297b2SJeff Mahoney /* 1206098297b2SJeff Mahoney * which is better, to lock once around the whole loop, or 1207098297b2SJeff Mahoney * to lock for each call to remove_journal_hash? 12081da177e4SLinus Torvalds */ 12091da177e4SLinus Torvalds while (cn) { 12101da177e4SLinus Torvalds if (cn->blocknr != 0) { 12111da177e4SLinus Torvalds if (debug) { 1212a9dd3643SJeff Mahoney reiserfs_warning(sb, "reiserfs-2201", 1213bd4c625cSLinus Torvalds "block %u, bh is %d, state %ld", 1214bd4c625cSLinus Torvalds cn->blocknr, cn->bh ? 1 : 0, 1215bd4c625cSLinus Torvalds cn->state); 12161da177e4SLinus Torvalds } 12171da177e4SLinus Torvalds cn->state = 0; 1218a9dd3643SJeff Mahoney remove_journal_hash(sb, journal->j_list_hash_table, 1219bd4c625cSLinus Torvalds jl, cn->blocknr, 1); 12201da177e4SLinus Torvalds } 12211da177e4SLinus Torvalds last = cn; 12221da177e4SLinus Torvalds cn = cn->next; 1223a9dd3643SJeff Mahoney free_cnode(sb, last); 12241da177e4SLinus Torvalds } 12251da177e4SLinus Torvalds jl->j_realblock = NULL; 12261da177e4SLinus Torvalds } 12271da177e4SLinus Torvalds 12281da177e4SLinus Torvalds /* 1229098297b2SJeff Mahoney * if this timestamp is greater than the timestamp we wrote last to the 1230098297b2SJeff Mahoney * header block, write it to the header block. once this is done, I can 1231098297b2SJeff Mahoney * safely say the log area for this transaction won't ever be replayed, 1232098297b2SJeff Mahoney * and I can start releasing blocks in this transaction for reuse as data 1233098297b2SJeff Mahoney * blocks. called by flush_journal_list, before it calls 1234098297b2SJeff Mahoney * remove_all_from_journal_list 12351da177e4SLinus Torvalds */ 1236a9dd3643SJeff Mahoney static int _update_journal_header_block(struct super_block *sb, 1237bd4c625cSLinus Torvalds unsigned long offset, 1238600ed416SJeff Mahoney unsigned int trans_id) 1239bd4c625cSLinus Torvalds { 12401da177e4SLinus Torvalds struct reiserfs_journal_header *jh; 1241a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 1242278f6679SJeff Mahoney int depth; 12431da177e4SLinus Torvalds 12441da177e4SLinus Torvalds if (reiserfs_is_journal_aborted(journal)) 12451da177e4SLinus Torvalds return -EIO; 12461da177e4SLinus Torvalds 12471da177e4SLinus Torvalds if (trans_id >= journal->j_last_flush_trans_id) { 12481da177e4SLinus Torvalds if (buffer_locked((journal->j_header_bh))) { 1249278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(sb); 1250278f6679SJeff Mahoney __wait_on_buffer(journal->j_header_bh); 1251278f6679SJeff Mahoney reiserfs_write_lock_nested(sb, depth); 12521da177e4SLinus Torvalds if (unlikely(!buffer_uptodate(journal->j_header_bh))) { 12531da177e4SLinus Torvalds #ifdef CONFIG_REISERFS_CHECK 1254a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-699", 125545b03d5eSJeff Mahoney "buffer write failed"); 12561da177e4SLinus Torvalds #endif 12571da177e4SLinus Torvalds return -EIO; 12581da177e4SLinus Torvalds } 12591da177e4SLinus Torvalds } 12601da177e4SLinus Torvalds journal->j_last_flush_trans_id = trans_id; 12611da177e4SLinus Torvalds journal->j_first_unflushed_offset = offset; 1262bd4c625cSLinus Torvalds jh = (struct reiserfs_journal_header *)(journal->j_header_bh-> 1263bd4c625cSLinus Torvalds b_data); 12641da177e4SLinus Torvalds jh->j_last_flush_trans_id = cpu_to_le32(trans_id); 12651da177e4SLinus Torvalds jh->j_first_unflushed_offset = cpu_to_le32(offset); 12661da177e4SLinus Torvalds jh->j_mount_id = cpu_to_le32(journal->j_mount_id); 12671da177e4SLinus Torvalds 12681da177e4SLinus Torvalds set_buffer_dirty(journal->j_header_bh); 1269278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(sb); 12707cd33ad2SChristoph Hellwig 12717cd33ad2SChristoph Hellwig if (reiserfs_barrier_flush(sb)) 127270fd7614SChristoph Hellwig __sync_dirty_buffer(journal->j_header_bh, 1273d8747d64SJan Kara REQ_SYNC | REQ_PREFLUSH | REQ_FUA); 12747cd33ad2SChristoph Hellwig else 12751da177e4SLinus Torvalds sync_dirty_buffer(journal->j_header_bh); 12767cd33ad2SChristoph Hellwig 1277278f6679SJeff Mahoney reiserfs_write_lock_nested(sb, depth); 12781da177e4SLinus Torvalds if (!buffer_uptodate(journal->j_header_bh)) { 1279a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-837", 128045b03d5eSJeff Mahoney "IO error during journal replay"); 12811da177e4SLinus Torvalds return -EIO; 12821da177e4SLinus Torvalds } 12831da177e4SLinus Torvalds } 12841da177e4SLinus Torvalds return 0; 12851da177e4SLinus Torvalds } 12861da177e4SLinus Torvalds 1287a9dd3643SJeff Mahoney static int update_journal_header_block(struct super_block *sb, 12881da177e4SLinus Torvalds unsigned long offset, 1289600ed416SJeff Mahoney unsigned int trans_id) 1290bd4c625cSLinus Torvalds { 1291a9dd3643SJeff Mahoney return _update_journal_header_block(sb, offset, trans_id); 12921da177e4SLinus Torvalds } 1293bd4c625cSLinus Torvalds 12941da177e4SLinus Torvalds /* 12951da177e4SLinus Torvalds ** flush any and all journal lists older than you are 12961da177e4SLinus Torvalds ** can only be called from flush_journal_list 12971da177e4SLinus Torvalds */ 1298a9dd3643SJeff Mahoney static int flush_older_journal_lists(struct super_block *sb, 12991da177e4SLinus Torvalds struct reiserfs_journal_list *jl) 13001da177e4SLinus Torvalds { 13011da177e4SLinus Torvalds struct list_head *entry; 13021da177e4SLinus Torvalds struct reiserfs_journal_list *other_jl; 1303a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 1304600ed416SJeff Mahoney unsigned int trans_id = jl->j_trans_id; 13051da177e4SLinus Torvalds 1306098297b2SJeff Mahoney /* 1307098297b2SJeff Mahoney * we know we are the only ones flushing things, no extra race 13081da177e4SLinus Torvalds * protection is required. 13091da177e4SLinus Torvalds */ 13101da177e4SLinus Torvalds restart: 13111da177e4SLinus Torvalds entry = journal->j_journal_list.next; 13121da177e4SLinus Torvalds /* Did we wrap? */ 13131da177e4SLinus Torvalds if (entry == &journal->j_journal_list) 13141da177e4SLinus Torvalds return 0; 13151da177e4SLinus Torvalds other_jl = JOURNAL_LIST_ENTRY(entry); 13161da177e4SLinus Torvalds if (other_jl->j_trans_id < trans_id) { 13171da177e4SLinus Torvalds BUG_ON(other_jl->j_refcount <= 0); 13181da177e4SLinus Torvalds /* do not flush all */ 1319a9dd3643SJeff Mahoney flush_journal_list(sb, other_jl, 0); 13201da177e4SLinus Torvalds 13211da177e4SLinus Torvalds /* other_jl is now deleted from the list */ 13221da177e4SLinus Torvalds goto restart; 13231da177e4SLinus Torvalds } 13241da177e4SLinus Torvalds return 0; 13251da177e4SLinus Torvalds } 13261da177e4SLinus Torvalds 13271da177e4SLinus Torvalds static void del_from_work_list(struct super_block *s, 1328bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl) 1329bd4c625cSLinus Torvalds { 13301da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 13311da177e4SLinus Torvalds if (!list_empty(&jl->j_working_list)) { 13321da177e4SLinus Torvalds list_del_init(&jl->j_working_list); 13331da177e4SLinus Torvalds journal->j_num_work_lists--; 13341da177e4SLinus Torvalds } 13351da177e4SLinus Torvalds } 13361da177e4SLinus Torvalds 1337098297b2SJeff Mahoney /* 1338098297b2SJeff Mahoney * flush a journal list, both commit and real blocks 1339098297b2SJeff Mahoney * 1340098297b2SJeff Mahoney * always set flushall to 1, unless you are calling from inside 1341098297b2SJeff Mahoney * flush_journal_list 1342098297b2SJeff Mahoney * 1343098297b2SJeff Mahoney * IMPORTANT. This can only be called while there are no journal writers, 1344098297b2SJeff Mahoney * and the journal is locked. That means it can only be called from 1345098297b2SJeff Mahoney * do_journal_end, or by journal_release 13461da177e4SLinus Torvalds */ 13471da177e4SLinus Torvalds static int flush_journal_list(struct super_block *s, 1348bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl, int flushall) 1349bd4c625cSLinus Torvalds { 13501da177e4SLinus Torvalds struct reiserfs_journal_list *pjl; 13516e9ca45fSzhengbin struct reiserfs_journal_cnode *cn; 13521da177e4SLinus Torvalds int count; 13531da177e4SLinus Torvalds int was_jwait = 0; 13541da177e4SLinus Torvalds int was_dirty = 0; 13551da177e4SLinus Torvalds struct buffer_head *saved_bh; 13561da177e4SLinus Torvalds unsigned long j_len_saved = jl->j_len; 13571da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 13581da177e4SLinus Torvalds int err = 0; 1359278f6679SJeff Mahoney int depth; 13601da177e4SLinus Torvalds 13611da177e4SLinus Torvalds BUG_ON(j_len_saved <= 0); 13621da177e4SLinus Torvalds 13631da177e4SLinus Torvalds if (atomic_read(&journal->j_wcount) != 0) { 136445b03d5eSJeff Mahoney reiserfs_warning(s, "clm-2048", "called with wcount %d", 13651da177e4SLinus Torvalds atomic_read(&journal->j_wcount)); 13661da177e4SLinus Torvalds } 13671da177e4SLinus Torvalds 13681da177e4SLinus Torvalds /* if flushall == 0, the lock is already held */ 13691da177e4SLinus Torvalds if (flushall) { 13708ebc4232SFrederic Weisbecker reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s); 1371afe70259SJeff Mahoney } else if (mutex_trylock(&journal->j_flush_mutex)) { 13721da177e4SLinus Torvalds BUG(); 13731da177e4SLinus Torvalds } 13741da177e4SLinus Torvalds 13751da177e4SLinus Torvalds count = 0; 13761da177e4SLinus Torvalds if (j_len_saved > journal->j_trans_max) { 1377c3a9c210SJeff Mahoney reiserfs_panic(s, "journal-715", "length is %lu, trans id %lu", 1378bd4c625cSLinus Torvalds j_len_saved, jl->j_trans_id); 13791da177e4SLinus Torvalds return 0; 13801da177e4SLinus Torvalds } 13811da177e4SLinus Torvalds 13821da177e4SLinus Torvalds /* if all the work is already done, get out of here */ 1383a228bf8fSJeff Mahoney if (atomic_read(&jl->j_nonzerolen) <= 0 && 1384a228bf8fSJeff Mahoney atomic_read(&jl->j_commit_left) <= 0) { 13851da177e4SLinus Torvalds goto flush_older_and_return; 13861da177e4SLinus Torvalds } 13871da177e4SLinus Torvalds 1388098297b2SJeff Mahoney /* 1389098297b2SJeff Mahoney * start by putting the commit list on disk. This will also flush 1390098297b2SJeff Mahoney * the commit lists of any olders transactions 13911da177e4SLinus Torvalds */ 13921da177e4SLinus Torvalds flush_commit_list(s, jl, 1); 13931da177e4SLinus Torvalds 1394bd4c625cSLinus Torvalds if (!(jl->j_state & LIST_DIRTY) 1395bd4c625cSLinus Torvalds && !reiserfs_is_journal_aborted(journal)) 13961da177e4SLinus Torvalds BUG(); 13971da177e4SLinus Torvalds 13981da177e4SLinus Torvalds /* are we done now? */ 1399a228bf8fSJeff Mahoney if (atomic_read(&jl->j_nonzerolen) <= 0 && 1400a228bf8fSJeff Mahoney atomic_read(&jl->j_commit_left) <= 0) { 14011da177e4SLinus Torvalds goto flush_older_and_return; 14021da177e4SLinus Torvalds } 14031da177e4SLinus Torvalds 1404098297b2SJeff Mahoney /* 1405098297b2SJeff Mahoney * loop through each cnode, see if we need to write it, 1406098297b2SJeff Mahoney * or wait on a more recent transaction, or just ignore it 14071da177e4SLinus Torvalds */ 1408a228bf8fSJeff Mahoney if (atomic_read(&journal->j_wcount) != 0) { 1409c3a9c210SJeff Mahoney reiserfs_panic(s, "journal-844", "journal list is flushing, " 1410c3a9c210SJeff Mahoney "wcount is not 0"); 14111da177e4SLinus Torvalds } 14121da177e4SLinus Torvalds cn = jl->j_realblock; 14131da177e4SLinus Torvalds while (cn) { 14141da177e4SLinus Torvalds was_jwait = 0; 14151da177e4SLinus Torvalds was_dirty = 0; 14161da177e4SLinus Torvalds saved_bh = NULL; 14171da177e4SLinus Torvalds /* blocknr of 0 is no longer in the hash, ignore it */ 14181da177e4SLinus Torvalds if (cn->blocknr == 0) { 14191da177e4SLinus Torvalds goto free_cnode; 14201da177e4SLinus Torvalds } 14211da177e4SLinus Torvalds 1422098297b2SJeff Mahoney /* 1423098297b2SJeff Mahoney * This transaction failed commit. 1424098297b2SJeff Mahoney * Don't write out to the disk 1425098297b2SJeff Mahoney */ 14261da177e4SLinus Torvalds if (!(jl->j_state & LIST_DIRTY)) 14271da177e4SLinus Torvalds goto free_cnode; 14281da177e4SLinus Torvalds 14291da177e4SLinus Torvalds pjl = find_newer_jl_for_cn(cn); 1430098297b2SJeff Mahoney /* 1431098297b2SJeff Mahoney * the order is important here. We check pjl to make sure we 1432098297b2SJeff Mahoney * don't clear BH_JDirty_wait if we aren't the one writing this 1433098297b2SJeff Mahoney * block to disk 14341da177e4SLinus Torvalds */ 14351da177e4SLinus Torvalds if (!pjl && cn->bh) { 14361da177e4SLinus Torvalds saved_bh = cn->bh; 14371da177e4SLinus Torvalds 1438098297b2SJeff Mahoney /* 1439098297b2SJeff Mahoney * we do this to make sure nobody releases the 1440098297b2SJeff Mahoney * buffer while we are working with it 14411da177e4SLinus Torvalds */ 14421da177e4SLinus Torvalds get_bh(saved_bh); 14431da177e4SLinus Torvalds 14441da177e4SLinus Torvalds if (buffer_journal_dirty(saved_bh)) { 14451da177e4SLinus Torvalds BUG_ON(!can_dirty(cn)); 14461da177e4SLinus Torvalds was_jwait = 1; 14471da177e4SLinus Torvalds was_dirty = 1; 14481da177e4SLinus Torvalds } else if (can_dirty(cn)) { 1449098297b2SJeff Mahoney /* 1450098297b2SJeff Mahoney * everything with !pjl && jwait 1451098297b2SJeff Mahoney * should be writable 1452098297b2SJeff Mahoney */ 14531da177e4SLinus Torvalds BUG(); 14541da177e4SLinus Torvalds } 14551da177e4SLinus Torvalds } 14561da177e4SLinus Torvalds 1457098297b2SJeff Mahoney /* 1458098297b2SJeff Mahoney * if someone has this block in a newer transaction, just make 1459098297b2SJeff Mahoney * sure they are committed, and don't try writing it to disk 14601da177e4SLinus Torvalds */ 14611da177e4SLinus Torvalds if (pjl) { 14621da177e4SLinus Torvalds if (atomic_read(&pjl->j_commit_left)) 14631da177e4SLinus Torvalds flush_commit_list(s, pjl, 1); 14641da177e4SLinus Torvalds goto free_cnode; 14651da177e4SLinus Torvalds } 14661da177e4SLinus Torvalds 1467098297b2SJeff Mahoney /* 1468098297b2SJeff Mahoney * bh == NULL when the block got to disk on its own, OR, 1469098297b2SJeff Mahoney * the block got freed in a future transaction 14701da177e4SLinus Torvalds */ 14711da177e4SLinus Torvalds if (saved_bh == NULL) { 14721da177e4SLinus Torvalds goto free_cnode; 14731da177e4SLinus Torvalds } 14741da177e4SLinus Torvalds 1475098297b2SJeff Mahoney /* 1476098297b2SJeff Mahoney * this should never happen. kupdate_one_transaction has 1477098297b2SJeff Mahoney * this list locked while it works, so we should never see a 1478098297b2SJeff Mahoney * buffer here that is not marked JDirty_wait 14791da177e4SLinus Torvalds */ 14801da177e4SLinus Torvalds if ((!was_jwait) && !buffer_locked(saved_bh)) { 148145b03d5eSJeff Mahoney reiserfs_warning(s, "journal-813", 148245b03d5eSJeff Mahoney "BAD! buffer %llu %cdirty %cjwait, " 148365f2b263SColin Ian King "not in a newer transaction", 1484bd4c625cSLinus Torvalds (unsigned long long)saved_bh-> 1485bd4c625cSLinus Torvalds b_blocknr, was_dirty ? ' ' : '!', 1486bd4c625cSLinus Torvalds was_jwait ? ' ' : '!'); 14871da177e4SLinus Torvalds } 14881da177e4SLinus Torvalds if (was_dirty) { 1489098297b2SJeff Mahoney /* 1490098297b2SJeff Mahoney * we inc again because saved_bh gets decremented 1491098297b2SJeff Mahoney * at free_cnode 1492098297b2SJeff Mahoney */ 14931da177e4SLinus Torvalds get_bh(saved_bh); 14941da177e4SLinus Torvalds set_bit(BLOCK_NEEDS_FLUSH, &cn->state); 14951da177e4SLinus Torvalds lock_buffer(saved_bh); 14961da177e4SLinus Torvalds BUG_ON(cn->blocknr != saved_bh->b_blocknr); 14971da177e4SLinus Torvalds if (buffer_dirty(saved_bh)) 14981da177e4SLinus Torvalds submit_logged_buffer(saved_bh); 14991da177e4SLinus Torvalds else 15001da177e4SLinus Torvalds unlock_buffer(saved_bh); 15011da177e4SLinus Torvalds count++; 15021da177e4SLinus Torvalds } else { 150345b03d5eSJeff Mahoney reiserfs_warning(s, "clm-2082", 150445b03d5eSJeff Mahoney "Unable to flush buffer %llu in %s", 1505bd4c625cSLinus Torvalds (unsigned long long)saved_bh-> 1506fbe5498bSHarvey Harrison b_blocknr, __func__); 15071da177e4SLinus Torvalds } 15081da177e4SLinus Torvalds free_cnode: 15091da177e4SLinus Torvalds cn = cn->next; 15101da177e4SLinus Torvalds if (saved_bh) { 1511098297b2SJeff Mahoney /* 1512098297b2SJeff Mahoney * we incremented this to keep others from 1513098297b2SJeff Mahoney * taking the buffer head away 1514098297b2SJeff Mahoney */ 15151da177e4SLinus Torvalds put_bh(saved_bh); 1516a228bf8fSJeff Mahoney if (atomic_read(&saved_bh->b_count) < 0) { 151745b03d5eSJeff Mahoney reiserfs_warning(s, "journal-945", 151845b03d5eSJeff Mahoney "saved_bh->b_count < 0"); 15191da177e4SLinus Torvalds } 15201da177e4SLinus Torvalds } 15211da177e4SLinus Torvalds } 15221da177e4SLinus Torvalds if (count > 0) { 15231da177e4SLinus Torvalds cn = jl->j_realblock; 15241da177e4SLinus Torvalds while (cn) { 15251da177e4SLinus Torvalds if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) { 15261da177e4SLinus Torvalds if (!cn->bh) { 1527c3a9c210SJeff Mahoney reiserfs_panic(s, "journal-1011", 1528c3a9c210SJeff Mahoney "cn->bh is NULL"); 15291da177e4SLinus Torvalds } 15308ebc4232SFrederic Weisbecker 1531278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(s); 1532278f6679SJeff Mahoney __wait_on_buffer(cn->bh); 1533278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 15348ebc4232SFrederic Weisbecker 15351da177e4SLinus Torvalds if (!cn->bh) { 1536c3a9c210SJeff Mahoney reiserfs_panic(s, "journal-1012", 1537c3a9c210SJeff Mahoney "cn->bh is NULL"); 15381da177e4SLinus Torvalds } 15391da177e4SLinus Torvalds if (unlikely(!buffer_uptodate(cn->bh))) { 15401da177e4SLinus Torvalds #ifdef CONFIG_REISERFS_CHECK 154145b03d5eSJeff Mahoney reiserfs_warning(s, "journal-949", 154245b03d5eSJeff Mahoney "buffer write failed"); 15431da177e4SLinus Torvalds #endif 15441da177e4SLinus Torvalds err = -EIO; 15451da177e4SLinus Torvalds } 1546098297b2SJeff Mahoney /* 1547098297b2SJeff Mahoney * note, we must clear the JDirty_wait bit 1548098297b2SJeff Mahoney * after the up to date check, otherwise we 1549098297b2SJeff Mahoney * race against our flushpage routine 15501da177e4SLinus Torvalds */ 1551bd4c625cSLinus Torvalds BUG_ON(!test_clear_buffer_journal_dirty 1552bd4c625cSLinus Torvalds (cn->bh)); 15531da177e4SLinus Torvalds 1554398c95bdSChris Mason /* drop one ref for us */ 15551da177e4SLinus Torvalds put_bh(cn->bh); 1556398c95bdSChris Mason /* drop one ref for journal_mark_dirty */ 1557398c95bdSChris Mason release_buffer_page(cn->bh); 15581da177e4SLinus Torvalds } 15591da177e4SLinus Torvalds cn = cn->next; 15601da177e4SLinus Torvalds } 15611da177e4SLinus Torvalds } 15621da177e4SLinus Torvalds 15631da177e4SLinus Torvalds if (err) 1564bd4c625cSLinus Torvalds reiserfs_abort(s, -EIO, 1565bd4c625cSLinus Torvalds "Write error while pushing transaction to disk in %s", 1566fbe5498bSHarvey Harrison __func__); 15671da177e4SLinus Torvalds flush_older_and_return: 15681da177e4SLinus Torvalds 1569098297b2SJeff Mahoney /* 1570098297b2SJeff Mahoney * before we can update the journal header block, we _must_ flush all 1571098297b2SJeff Mahoney * real blocks from all older transactions to disk. This is because 1572098297b2SJeff Mahoney * once the header block is updated, this transaction will not be 1573098297b2SJeff Mahoney * replayed after a crash 15741da177e4SLinus Torvalds */ 15751da177e4SLinus Torvalds if (flushall) { 15761da177e4SLinus Torvalds flush_older_journal_lists(s, jl); 15771da177e4SLinus Torvalds } 15781da177e4SLinus Torvalds 15791da177e4SLinus Torvalds err = journal->j_errno; 1580098297b2SJeff Mahoney /* 1581098297b2SJeff Mahoney * before we can remove everything from the hash tables for this 1582098297b2SJeff Mahoney * transaction, we must make sure it can never be replayed 1583098297b2SJeff Mahoney * 1584098297b2SJeff Mahoney * since we are only called from do_journal_end, we know for sure there 1585098297b2SJeff Mahoney * are no allocations going on while we are flushing journal lists. So, 1586098297b2SJeff Mahoney * we only need to update the journal header block for the last list 1587098297b2SJeff Mahoney * being flushed 15881da177e4SLinus Torvalds */ 15891da177e4SLinus Torvalds if (!err && flushall) { 1590bd4c625cSLinus Torvalds err = 1591bd4c625cSLinus Torvalds update_journal_header_block(s, 1592bd4c625cSLinus Torvalds (jl->j_start + jl->j_len + 1593bd4c625cSLinus Torvalds 2) % SB_ONDISK_JOURNAL_SIZE(s), 1594bd4c625cSLinus Torvalds jl->j_trans_id); 15951da177e4SLinus Torvalds if (err) 1596bd4c625cSLinus Torvalds reiserfs_abort(s, -EIO, 1597bd4c625cSLinus Torvalds "Write error while updating journal header in %s", 1598fbe5498bSHarvey Harrison __func__); 15991da177e4SLinus Torvalds } 16001da177e4SLinus Torvalds remove_all_from_journal_list(s, jl, 0); 16011da177e4SLinus Torvalds list_del_init(&jl->j_list); 16021da177e4SLinus Torvalds journal->j_num_lists--; 16031da177e4SLinus Torvalds del_from_work_list(s, jl); 16041da177e4SLinus Torvalds 16051da177e4SLinus Torvalds if (journal->j_last_flush_id != 0 && 16061da177e4SLinus Torvalds (jl->j_trans_id - journal->j_last_flush_id) != 1) { 160745b03d5eSJeff Mahoney reiserfs_warning(s, "clm-2201", "last flush %lu, current %lu", 1608bd4c625cSLinus Torvalds journal->j_last_flush_id, jl->j_trans_id); 16091da177e4SLinus Torvalds } 16101da177e4SLinus Torvalds journal->j_last_flush_id = jl->j_trans_id; 16111da177e4SLinus Torvalds 1612098297b2SJeff Mahoney /* 1613098297b2SJeff Mahoney * not strictly required since we are freeing the list, but it should 16141da177e4SLinus Torvalds * help find code using dead lists later on 16151da177e4SLinus Torvalds */ 16161da177e4SLinus Torvalds jl->j_len = 0; 1617a228bf8fSJeff Mahoney atomic_set(&jl->j_nonzerolen, 0); 16181da177e4SLinus Torvalds jl->j_start = 0; 16191da177e4SLinus Torvalds jl->j_realblock = NULL; 16201da177e4SLinus Torvalds jl->j_commit_bh = NULL; 16211da177e4SLinus Torvalds jl->j_trans_id = 0; 16221da177e4SLinus Torvalds jl->j_state = 0; 16231da177e4SLinus Torvalds put_journal_list(s, jl); 16241da177e4SLinus Torvalds if (flushall) 1625afe70259SJeff Mahoney mutex_unlock(&journal->j_flush_mutex); 16261da177e4SLinus Torvalds return err; 16271da177e4SLinus Torvalds } 16281da177e4SLinus Torvalds 16291da177e4SLinus Torvalds static int write_one_transaction(struct super_block *s, 16301da177e4SLinus Torvalds struct reiserfs_journal_list *jl, 16311da177e4SLinus Torvalds struct buffer_chunk *chunk) 16321da177e4SLinus Torvalds { 16331da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 16341da177e4SLinus Torvalds int ret = 0; 16351da177e4SLinus Torvalds 16361da177e4SLinus Torvalds jl->j_state |= LIST_TOUCHED; 16371da177e4SLinus Torvalds del_from_work_list(s, jl); 16381da177e4SLinus Torvalds if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) { 16391da177e4SLinus Torvalds return 0; 16401da177e4SLinus Torvalds } 16411da177e4SLinus Torvalds 16421da177e4SLinus Torvalds cn = jl->j_realblock; 16431da177e4SLinus Torvalds while (cn) { 1644098297b2SJeff Mahoney /* 1645098297b2SJeff Mahoney * if the blocknr == 0, this has been cleared from the hash, 1646098297b2SJeff Mahoney * skip it 16471da177e4SLinus Torvalds */ 16481da177e4SLinus Torvalds if (cn->blocknr == 0) { 16491da177e4SLinus Torvalds goto next; 16501da177e4SLinus Torvalds } 16511da177e4SLinus Torvalds if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) { 16521da177e4SLinus Torvalds struct buffer_head *tmp_bh; 1653098297b2SJeff Mahoney /* 1654098297b2SJeff Mahoney * we can race against journal_mark_freed when we try 16551da177e4SLinus Torvalds * to lock_buffer(cn->bh), so we have to inc the buffer 16561da177e4SLinus Torvalds * count, and recheck things after locking 16571da177e4SLinus Torvalds */ 16581da177e4SLinus Torvalds tmp_bh = cn->bh; 16591da177e4SLinus Torvalds get_bh(tmp_bh); 16601da177e4SLinus Torvalds lock_buffer(tmp_bh); 16611da177e4SLinus Torvalds if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) { 16621da177e4SLinus Torvalds if (!buffer_journal_dirty(tmp_bh) || 16631da177e4SLinus Torvalds buffer_journal_prepared(tmp_bh)) 16641da177e4SLinus Torvalds BUG(); 16651da177e4SLinus Torvalds add_to_chunk(chunk, tmp_bh, NULL, write_chunk); 16661da177e4SLinus Torvalds ret++; 16671da177e4SLinus Torvalds } else { 16681da177e4SLinus Torvalds /* note, cn->bh might be null now */ 16691da177e4SLinus Torvalds unlock_buffer(tmp_bh); 16701da177e4SLinus Torvalds } 16711da177e4SLinus Torvalds put_bh(tmp_bh); 16721da177e4SLinus Torvalds } 16731da177e4SLinus Torvalds next: 16741da177e4SLinus Torvalds cn = cn->next; 16751da177e4SLinus Torvalds cond_resched(); 16761da177e4SLinus Torvalds } 16771da177e4SLinus Torvalds return ret; 16781da177e4SLinus Torvalds } 16791da177e4SLinus Torvalds 16801da177e4SLinus Torvalds /* used by flush_commit_list */ 1681dc0dde61SHariprasad Kelam static void dirty_one_transaction(struct super_block *s, 16821da177e4SLinus Torvalds struct reiserfs_journal_list *jl) 16831da177e4SLinus Torvalds { 16841da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 16851da177e4SLinus Torvalds struct reiserfs_journal_list *pjl; 16861da177e4SLinus Torvalds 16871da177e4SLinus Torvalds jl->j_state |= LIST_DIRTY; 16881da177e4SLinus Torvalds cn = jl->j_realblock; 16891da177e4SLinus Torvalds while (cn) { 1690098297b2SJeff Mahoney /* 1691098297b2SJeff Mahoney * look for a more recent transaction that logged this 1692098297b2SJeff Mahoney * buffer. Only the most recent transaction with a buffer in 1693098297b2SJeff Mahoney * it is allowed to send that buffer to disk 16941da177e4SLinus Torvalds */ 16951da177e4SLinus Torvalds pjl = find_newer_jl_for_cn(cn); 1696bd4c625cSLinus Torvalds if (!pjl && cn->blocknr && cn->bh 1697bd4c625cSLinus Torvalds && buffer_journal_dirty(cn->bh)) { 16981da177e4SLinus Torvalds BUG_ON(!can_dirty(cn)); 1699098297b2SJeff Mahoney /* 1700098297b2SJeff Mahoney * if the buffer is prepared, it will either be logged 17011da177e4SLinus Torvalds * or restored. If restored, we need to make sure 17021da177e4SLinus Torvalds * it actually gets marked dirty 17031da177e4SLinus Torvalds */ 17041da177e4SLinus Torvalds clear_buffer_journal_new(cn->bh); 17051da177e4SLinus Torvalds if (buffer_journal_prepared(cn->bh)) { 17061da177e4SLinus Torvalds set_buffer_journal_restore_dirty(cn->bh); 17071da177e4SLinus Torvalds } else { 17081da177e4SLinus Torvalds set_buffer_journal_test(cn->bh); 17091da177e4SLinus Torvalds mark_buffer_dirty(cn->bh); 17101da177e4SLinus Torvalds } 17111da177e4SLinus Torvalds } 17121da177e4SLinus Torvalds cn = cn->next; 17131da177e4SLinus Torvalds } 17141da177e4SLinus Torvalds } 17151da177e4SLinus Torvalds 17161da177e4SLinus Torvalds static int kupdate_transactions(struct super_block *s, 17171da177e4SLinus Torvalds struct reiserfs_journal_list *jl, 17181da177e4SLinus Torvalds struct reiserfs_journal_list **next_jl, 1719600ed416SJeff Mahoney unsigned int *next_trans_id, 1720bd4c625cSLinus Torvalds int num_blocks, int num_trans) 1721bd4c625cSLinus Torvalds { 17221da177e4SLinus Torvalds int ret = 0; 17231da177e4SLinus Torvalds int written = 0; 17241da177e4SLinus Torvalds int transactions_flushed = 0; 1725600ed416SJeff Mahoney unsigned int orig_trans_id = jl->j_trans_id; 17261da177e4SLinus Torvalds struct buffer_chunk chunk; 17271da177e4SLinus Torvalds struct list_head *entry; 17281da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 17291da177e4SLinus Torvalds chunk.nr = 0; 17301da177e4SLinus Torvalds 1731a412f9efSFrederic Weisbecker reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s); 17321da177e4SLinus Torvalds if (!journal_list_still_alive(s, orig_trans_id)) { 17331da177e4SLinus Torvalds goto done; 17341da177e4SLinus Torvalds } 17351da177e4SLinus Torvalds 1736098297b2SJeff Mahoney /* 1737098297b2SJeff Mahoney * we've got j_flush_mutex held, nobody is going to delete any 17381da177e4SLinus Torvalds * of these lists out from underneath us 17391da177e4SLinus Torvalds */ 17401da177e4SLinus Torvalds while ((num_trans && transactions_flushed < num_trans) || 17411da177e4SLinus Torvalds (!num_trans && written < num_blocks)) { 17421da177e4SLinus Torvalds 17431da177e4SLinus Torvalds if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) || 1744bd4c625cSLinus Torvalds atomic_read(&jl->j_commit_left) 1745bd4c625cSLinus Torvalds || !(jl->j_state & LIST_DIRTY)) { 17461da177e4SLinus Torvalds del_from_work_list(s, jl); 17471da177e4SLinus Torvalds break; 17481da177e4SLinus Torvalds } 17491da177e4SLinus Torvalds ret = write_one_transaction(s, jl, &chunk); 17501da177e4SLinus Torvalds 17511da177e4SLinus Torvalds if (ret < 0) 17521da177e4SLinus Torvalds goto done; 17531da177e4SLinus Torvalds transactions_flushed++; 17541da177e4SLinus Torvalds written += ret; 17551da177e4SLinus Torvalds entry = jl->j_list.next; 17561da177e4SLinus Torvalds 17571da177e4SLinus Torvalds /* did we wrap? */ 17581da177e4SLinus Torvalds if (entry == &journal->j_journal_list) { 17591da177e4SLinus Torvalds break; 17601da177e4SLinus Torvalds } 17611da177e4SLinus Torvalds jl = JOURNAL_LIST_ENTRY(entry); 17621da177e4SLinus Torvalds 17631da177e4SLinus Torvalds /* don't bother with older transactions */ 17641da177e4SLinus Torvalds if (jl->j_trans_id <= orig_trans_id) 17651da177e4SLinus Torvalds break; 17661da177e4SLinus Torvalds } 17671da177e4SLinus Torvalds if (chunk.nr) { 17681da177e4SLinus Torvalds write_chunk(&chunk); 17691da177e4SLinus Torvalds } 17701da177e4SLinus Torvalds 17711da177e4SLinus Torvalds done: 1772afe70259SJeff Mahoney mutex_unlock(&journal->j_flush_mutex); 17731da177e4SLinus Torvalds return ret; 17741da177e4SLinus Torvalds } 17751da177e4SLinus Torvalds 1776098297b2SJeff Mahoney /* 1777098297b2SJeff Mahoney * for o_sync and fsync heavy applications, they tend to use 1778098297b2SJeff Mahoney * all the journa list slots with tiny transactions. These 1779098297b2SJeff Mahoney * trigger lots and lots of calls to update the header block, which 1780098297b2SJeff Mahoney * adds seeks and slows things down. 1781098297b2SJeff Mahoney * 1782098297b2SJeff Mahoney * This function tries to clear out a large chunk of the journal lists 1783098297b2SJeff Mahoney * at once, which makes everything faster since only the newest journal 1784098297b2SJeff Mahoney * list updates the header block 17851da177e4SLinus Torvalds */ 17861da177e4SLinus Torvalds static int flush_used_journal_lists(struct super_block *s, 1787bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl) 1788bd4c625cSLinus Torvalds { 17891da177e4SLinus Torvalds unsigned long len = 0; 17901da177e4SLinus Torvalds unsigned long cur_len; 17911da177e4SLinus Torvalds int i; 17921da177e4SLinus Torvalds int limit = 256; 17931da177e4SLinus Torvalds struct reiserfs_journal_list *tjl; 17941da177e4SLinus Torvalds struct reiserfs_journal_list *flush_jl; 1795600ed416SJeff Mahoney unsigned int trans_id; 17961da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 17971da177e4SLinus Torvalds 17981da177e4SLinus Torvalds flush_jl = tjl = jl; 17991da177e4SLinus Torvalds 18001da177e4SLinus Torvalds /* in data logging mode, try harder to flush a lot of blocks */ 18011da177e4SLinus Torvalds if (reiserfs_data_log(s)) 18021da177e4SLinus Torvalds limit = 1024; 18031da177e4SLinus Torvalds /* flush for 256 transactions or limit blocks, whichever comes first */ 18041da177e4SLinus Torvalds for (i = 0; i < 256 && len < limit; i++) { 18051da177e4SLinus Torvalds if (atomic_read(&tjl->j_commit_left) || 18061da177e4SLinus Torvalds tjl->j_trans_id < jl->j_trans_id) { 18071da177e4SLinus Torvalds break; 18081da177e4SLinus Torvalds } 18091da177e4SLinus Torvalds cur_len = atomic_read(&tjl->j_nonzerolen); 18101da177e4SLinus Torvalds if (cur_len > 0) { 18111da177e4SLinus Torvalds tjl->j_state &= ~LIST_TOUCHED; 18121da177e4SLinus Torvalds } 18131da177e4SLinus Torvalds len += cur_len; 18141da177e4SLinus Torvalds flush_jl = tjl; 18151da177e4SLinus Torvalds if (tjl->j_list.next == &journal->j_journal_list) 18161da177e4SLinus Torvalds break; 18171da177e4SLinus Torvalds tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next); 18181da177e4SLinus Torvalds } 1819721a769cSJeff Mahoney get_journal_list(jl); 1820721a769cSJeff Mahoney get_journal_list(flush_jl); 1821098297b2SJeff Mahoney 1822098297b2SJeff Mahoney /* 1823098297b2SJeff Mahoney * try to find a group of blocks we can flush across all the 1824098297b2SJeff Mahoney * transactions, but only bother if we've actually spanned 1825098297b2SJeff Mahoney * across multiple lists 18261da177e4SLinus Torvalds */ 18273e9fd5a4SJason Yan if (flush_jl != jl) 18283e9fd5a4SJason Yan kupdate_transactions(s, jl, &tjl, &trans_id, len, i); 18293e9fd5a4SJason Yan 18301da177e4SLinus Torvalds flush_journal_list(s, flush_jl, 1); 1831721a769cSJeff Mahoney put_journal_list(s, flush_jl); 1832721a769cSJeff Mahoney put_journal_list(s, jl); 18331da177e4SLinus Torvalds return 0; 18341da177e4SLinus Torvalds } 18351da177e4SLinus Torvalds 18361da177e4SLinus Torvalds /* 1837098297b2SJeff Mahoney * removes any nodes in table with name block and dev as bh. 1838098297b2SJeff Mahoney * only touchs the hnext and hprev pointers. 18391da177e4SLinus Torvalds */ 18405d552ad7SBharath Vedartham static void remove_journal_hash(struct super_block *sb, 18411da177e4SLinus Torvalds struct reiserfs_journal_cnode **table, 18421da177e4SLinus Torvalds struct reiserfs_journal_list *jl, 18431da177e4SLinus Torvalds unsigned long block, int remove_freed) 18441da177e4SLinus Torvalds { 18451da177e4SLinus Torvalds struct reiserfs_journal_cnode *cur; 18461da177e4SLinus Torvalds struct reiserfs_journal_cnode **head; 18471da177e4SLinus Torvalds 18481da177e4SLinus Torvalds head = &(journal_hash(table, sb, block)); 18491da177e4SLinus Torvalds if (!head) { 18501da177e4SLinus Torvalds return; 18511da177e4SLinus Torvalds } 18521da177e4SLinus Torvalds cur = *head; 18531da177e4SLinus Torvalds while (cur) { 1854bd4c625cSLinus Torvalds if (cur->blocknr == block && cur->sb == sb 1855bd4c625cSLinus Torvalds && (jl == NULL || jl == cur->jlist) 1856bd4c625cSLinus Torvalds && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) { 18571da177e4SLinus Torvalds if (cur->hnext) { 18581da177e4SLinus Torvalds cur->hnext->hprev = cur->hprev; 18591da177e4SLinus Torvalds } 18601da177e4SLinus Torvalds if (cur->hprev) { 18611da177e4SLinus Torvalds cur->hprev->hnext = cur->hnext; 18621da177e4SLinus Torvalds } else { 18631da177e4SLinus Torvalds *head = cur->hnext; 18641da177e4SLinus Torvalds } 18651da177e4SLinus Torvalds cur->blocknr = 0; 18661da177e4SLinus Torvalds cur->sb = NULL; 18671da177e4SLinus Torvalds cur->state = 0; 1868098297b2SJeff Mahoney /* 1869098297b2SJeff Mahoney * anybody who clears the cur->bh will also 1870098297b2SJeff Mahoney * dec the nonzerolen 1871098297b2SJeff Mahoney */ 1872098297b2SJeff Mahoney if (cur->bh && cur->jlist) 1873a228bf8fSJeff Mahoney atomic_dec(&cur->jlist->j_nonzerolen); 18741da177e4SLinus Torvalds cur->bh = NULL; 18751da177e4SLinus Torvalds cur->jlist = NULL; 18761da177e4SLinus Torvalds } 18771da177e4SLinus Torvalds cur = cur->hnext; 18781da177e4SLinus Torvalds } 18791da177e4SLinus Torvalds } 18801da177e4SLinus Torvalds 1881a9dd3643SJeff Mahoney static void free_journal_ram(struct super_block *sb) 1882bd4c625cSLinus Torvalds { 1883a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 1884d739b42bSPekka Enberg kfree(journal->j_current_jl); 18851da177e4SLinus Torvalds journal->j_num_lists--; 18861da177e4SLinus Torvalds 18871da177e4SLinus Torvalds vfree(journal->j_cnode_free_orig); 1888a9dd3643SJeff Mahoney free_list_bitmaps(sb, journal->j_list_bitmap); 1889a9dd3643SJeff Mahoney free_bitmap_nodes(sb); /* must be after free_list_bitmaps */ 18901da177e4SLinus Torvalds if (journal->j_header_bh) { 18911da177e4SLinus Torvalds brelse(journal->j_header_bh); 18921da177e4SLinus Torvalds } 1893098297b2SJeff Mahoney /* 1894098297b2SJeff Mahoney * j_header_bh is on the journal dev, make sure 1895098297b2SJeff Mahoney * not to release the journal dev until we brelse j_header_bh 18961da177e4SLinus Torvalds */ 1897a9dd3643SJeff Mahoney release_journal_dev(sb, journal); 18981da177e4SLinus Torvalds vfree(journal); 18991da177e4SLinus Torvalds } 19001da177e4SLinus Torvalds 19011da177e4SLinus Torvalds /* 1902098297b2SJeff Mahoney * call on unmount. Only set error to 1 if you haven't made your way out 1903098297b2SJeff Mahoney * of read_super() yet. Any other caller must keep error at 0. 19041da177e4SLinus Torvalds */ 1905bd4c625cSLinus Torvalds static int do_journal_release(struct reiserfs_transaction_handle *th, 1906a9dd3643SJeff Mahoney struct super_block *sb, int error) 1907bd4c625cSLinus Torvalds { 19081da177e4SLinus Torvalds struct reiserfs_transaction_handle myth; 1909a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 19101da177e4SLinus Torvalds 1911098297b2SJeff Mahoney /* 1912098297b2SJeff Mahoney * we only want to flush out transactions if we were 1913098297b2SJeff Mahoney * called with error == 0 19141da177e4SLinus Torvalds */ 1915bc98a42cSDavid Howells if (!error && !sb_rdonly(sb)) { 19161da177e4SLinus Torvalds /* end the current trans */ 19171da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 191858d85426SJeff Mahoney do_journal_end(th, FLUSH_ALL); 19191da177e4SLinus Torvalds 1920098297b2SJeff Mahoney /* 1921098297b2SJeff Mahoney * make sure something gets logged to force 1922098297b2SJeff Mahoney * our way into the flush code 1923098297b2SJeff Mahoney */ 1924b491dd17SJeff Mahoney if (!journal_join(&myth, sb)) { 1925a9dd3643SJeff Mahoney reiserfs_prepare_for_journal(sb, 1926a9dd3643SJeff Mahoney SB_BUFFER_WITH_SB(sb), 1927bd4c625cSLinus Torvalds 1); 192809f1b80bSJeff Mahoney journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb)); 192958d85426SJeff Mahoney do_journal_end(&myth, FLUSH_ALL); 19301da177e4SLinus Torvalds } 19311da177e4SLinus Torvalds } 19321da177e4SLinus Torvalds 19331da177e4SLinus Torvalds /* this also catches errors during the do_journal_end above */ 19341da177e4SLinus Torvalds if (!error && reiserfs_is_journal_aborted(journal)) { 19351da177e4SLinus Torvalds memset(&myth, 0, sizeof(myth)); 1936b491dd17SJeff Mahoney if (!journal_join_abort(&myth, sb)) { 1937a9dd3643SJeff Mahoney reiserfs_prepare_for_journal(sb, 1938a9dd3643SJeff Mahoney SB_BUFFER_WITH_SB(sb), 1939bd4c625cSLinus Torvalds 1); 194009f1b80bSJeff Mahoney journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb)); 194158d85426SJeff Mahoney do_journal_end(&myth, FLUSH_ALL); 19421da177e4SLinus Torvalds } 19431da177e4SLinus Torvalds } 19441da177e4SLinus Torvalds 19458ebc4232SFrederic Weisbecker 19468ebc4232SFrederic Weisbecker /* 19478ebc4232SFrederic Weisbecker * We must release the write lock here because 19488ebc4232SFrederic Weisbecker * the workqueue job (flush_async_commit) needs this lock 19498ebc4232SFrederic Weisbecker */ 19508ebc4232SFrederic Weisbecker reiserfs_write_unlock(sb); 1951033369d1SArtem Bityutskiy 195201777836SJan Kara /* 195301777836SJan Kara * Cancel flushing of old commits. Note that neither of these works 195401777836SJan Kara * will be requeued because superblock is being shutdown and doesn't 19551751e8a6SLinus Torvalds * have SB_ACTIVE set. 195601777836SJan Kara */ 195771b0576bSJan Kara reiserfs_cancel_old_flush(sb); 195801777836SJan Kara /* wait for all commits to finish */ 195901777836SJan Kara cancel_delayed_work_sync(&SB_JOURNAL(sb)->j_work); 19601da177e4SLinus Torvalds 1961a9dd3643SJeff Mahoney free_journal_ram(sb); 19621da177e4SLinus Torvalds 19630523676dSFrederic Weisbecker reiserfs_write_lock(sb); 19640523676dSFrederic Weisbecker 19651da177e4SLinus Torvalds return 0; 19661da177e4SLinus Torvalds } 19671da177e4SLinus Torvalds 1968098297b2SJeff Mahoney /* * call on unmount. flush all journal trans, release all alloc'd ram */ 1969bd4c625cSLinus Torvalds int journal_release(struct reiserfs_transaction_handle *th, 1970a9dd3643SJeff Mahoney struct super_block *sb) 1971bd4c625cSLinus Torvalds { 1972a9dd3643SJeff Mahoney return do_journal_release(th, sb, 0); 19731da177e4SLinus Torvalds } 1974bd4c625cSLinus Torvalds 1975098297b2SJeff Mahoney /* only call from an error condition inside reiserfs_read_super! */ 1976bd4c625cSLinus Torvalds int journal_release_error(struct reiserfs_transaction_handle *th, 1977a9dd3643SJeff Mahoney struct super_block *sb) 1978bd4c625cSLinus Torvalds { 1979a9dd3643SJeff Mahoney return do_journal_release(th, sb, 1); 19801da177e4SLinus Torvalds } 19811da177e4SLinus Torvalds 1982098297b2SJeff Mahoney /* 1983098297b2SJeff Mahoney * compares description block with commit block. 1984098297b2SJeff Mahoney * returns 1 if they differ, 0 if they are the same 1985098297b2SJeff Mahoney */ 1986a9dd3643SJeff Mahoney static int journal_compare_desc_commit(struct super_block *sb, 1987bd4c625cSLinus Torvalds struct reiserfs_journal_desc *desc, 1988bd4c625cSLinus Torvalds struct reiserfs_journal_commit *commit) 1989bd4c625cSLinus Torvalds { 19901da177e4SLinus Torvalds if (get_commit_trans_id(commit) != get_desc_trans_id(desc) || 19911da177e4SLinus Torvalds get_commit_trans_len(commit) != get_desc_trans_len(desc) || 1992a9dd3643SJeff Mahoney get_commit_trans_len(commit) > SB_JOURNAL(sb)->j_trans_max || 1993bd4c625cSLinus Torvalds get_commit_trans_len(commit) <= 0) { 19941da177e4SLinus Torvalds return 1; 19951da177e4SLinus Torvalds } 19961da177e4SLinus Torvalds return 0; 19971da177e4SLinus Torvalds } 1998bd4c625cSLinus Torvalds 1999098297b2SJeff Mahoney /* 2000098297b2SJeff Mahoney * returns 0 if it did not find a description block 2001098297b2SJeff Mahoney * returns -1 if it found a corrupt commit block 2002098297b2SJeff Mahoney * returns 1 if both desc and commit were valid 2003098297b2SJeff Mahoney * NOTE: only called during fs mount 20041da177e4SLinus Torvalds */ 2005a9dd3643SJeff Mahoney static int journal_transaction_is_valid(struct super_block *sb, 2006bd4c625cSLinus Torvalds struct buffer_head *d_bh, 2007600ed416SJeff Mahoney unsigned int *oldest_invalid_trans_id, 2008bd4c625cSLinus Torvalds unsigned long *newest_mount_id) 2009bd4c625cSLinus Torvalds { 20101da177e4SLinus Torvalds struct reiserfs_journal_desc *desc; 20111da177e4SLinus Torvalds struct reiserfs_journal_commit *commit; 20121da177e4SLinus Torvalds struct buffer_head *c_bh; 20131da177e4SLinus Torvalds unsigned long offset; 20141da177e4SLinus Torvalds 20151da177e4SLinus Torvalds if (!d_bh) 20161da177e4SLinus Torvalds return 0; 20171da177e4SLinus Torvalds 20181da177e4SLinus Torvalds desc = (struct reiserfs_journal_desc *)d_bh->b_data; 2019bd4c625cSLinus Torvalds if (get_desc_trans_len(desc) > 0 2020bd4c625cSLinus Torvalds && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) { 2021bd4c625cSLinus Torvalds if (oldest_invalid_trans_id && *oldest_invalid_trans_id 2022bd4c625cSLinus Torvalds && get_desc_trans_id(desc) > *oldest_invalid_trans_id) { 2023a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2024bd4c625cSLinus Torvalds "journal-986: transaction " 20251da177e4SLinus Torvalds "is valid returning because trans_id %d is greater than " 2026bd4c625cSLinus Torvalds "oldest_invalid %lu", 2027bd4c625cSLinus Torvalds get_desc_trans_id(desc), 20281da177e4SLinus Torvalds *oldest_invalid_trans_id); 20291da177e4SLinus Torvalds return 0; 20301da177e4SLinus Torvalds } 2031bd4c625cSLinus Torvalds if (newest_mount_id 2032bd4c625cSLinus Torvalds && *newest_mount_id > get_desc_mount_id(desc)) { 2033a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2034bd4c625cSLinus Torvalds "journal-1087: transaction " 20351da177e4SLinus Torvalds "is valid returning because mount_id %d is less than " 2036bd4c625cSLinus Torvalds "newest_mount_id %lu", 2037bd4c625cSLinus Torvalds get_desc_mount_id(desc), 20381da177e4SLinus Torvalds *newest_mount_id); 20391da177e4SLinus Torvalds return -1; 20401da177e4SLinus Torvalds } 2041a9dd3643SJeff Mahoney if (get_desc_trans_len(desc) > SB_JOURNAL(sb)->j_trans_max) { 2042a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-2018", 204345b03d5eSJeff Mahoney "Bad transaction length %d " 204445b03d5eSJeff Mahoney "encountered, ignoring transaction", 2045bd4c625cSLinus Torvalds get_desc_trans_len(desc)); 20461da177e4SLinus Torvalds return -1; 20471da177e4SLinus Torvalds } 2048a9dd3643SJeff Mahoney offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb); 20491da177e4SLinus Torvalds 2050098297b2SJeff Mahoney /* 2051098297b2SJeff Mahoney * ok, we have a journal description block, 2052098297b2SJeff Mahoney * let's see if the transaction was valid 2053098297b2SJeff Mahoney */ 2054bd4c625cSLinus Torvalds c_bh = 2055a9dd3643SJeff Mahoney journal_bread(sb, 2056a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2057bd4c625cSLinus Torvalds ((offset + get_desc_trans_len(desc) + 2058a9dd3643SJeff Mahoney 1) % SB_ONDISK_JOURNAL_SIZE(sb))); 20591da177e4SLinus Torvalds if (!c_bh) 20601da177e4SLinus Torvalds return 0; 20611da177e4SLinus Torvalds commit = (struct reiserfs_journal_commit *)c_bh->b_data; 2062a9dd3643SJeff Mahoney if (journal_compare_desc_commit(sb, desc, commit)) { 2063a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 20641da177e4SLinus Torvalds "journal_transaction_is_valid, commit offset %ld had bad " 20651da177e4SLinus Torvalds "time %d or length %d", 2066bd4c625cSLinus Torvalds c_bh->b_blocknr - 2067a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb), 20681da177e4SLinus Torvalds get_commit_trans_id(commit), 20691da177e4SLinus Torvalds get_commit_trans_len(commit)); 20701da177e4SLinus Torvalds brelse(c_bh); 20711da177e4SLinus Torvalds if (oldest_invalid_trans_id) { 2072bd4c625cSLinus Torvalds *oldest_invalid_trans_id = 2073bd4c625cSLinus Torvalds get_desc_trans_id(desc); 2074a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2075bd4c625cSLinus Torvalds "journal-1004: " 20761da177e4SLinus Torvalds "transaction_is_valid setting oldest invalid trans_id " 2077bd4c625cSLinus Torvalds "to %d", 2078bd4c625cSLinus Torvalds get_desc_trans_id(desc)); 20791da177e4SLinus Torvalds } 20801da177e4SLinus Torvalds return -1; 20811da177e4SLinus Torvalds } 20821da177e4SLinus Torvalds brelse(c_bh); 2083a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2084bd4c625cSLinus Torvalds "journal-1006: found valid " 20851da177e4SLinus Torvalds "transaction start offset %llu, len %d id %d", 2086bd4c625cSLinus Torvalds d_bh->b_blocknr - 2087a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb), 2088bd4c625cSLinus Torvalds get_desc_trans_len(desc), 2089bd4c625cSLinus Torvalds get_desc_trans_id(desc)); 20901da177e4SLinus Torvalds return 1; 20911da177e4SLinus Torvalds } else { 20921da177e4SLinus Torvalds return 0; 20931da177e4SLinus Torvalds } 20941da177e4SLinus Torvalds } 20951da177e4SLinus Torvalds 2096bd4c625cSLinus Torvalds static void brelse_array(struct buffer_head **heads, int num) 2097bd4c625cSLinus Torvalds { 20981da177e4SLinus Torvalds int i; 20991da177e4SLinus Torvalds for (i = 0; i < num; i++) { 21001da177e4SLinus Torvalds brelse(heads[i]); 21011da177e4SLinus Torvalds } 21021da177e4SLinus Torvalds } 21031da177e4SLinus Torvalds 21041da177e4SLinus Torvalds /* 2105098297b2SJeff Mahoney * given the start, and values for the oldest acceptable transactions, 2106098297b2SJeff Mahoney * this either reads in a replays a transaction, or returns because the 2107098297b2SJeff Mahoney * transaction is invalid, or too old. 2108098297b2SJeff Mahoney * NOTE: only called during fs mount 21091da177e4SLinus Torvalds */ 2110a9dd3643SJeff Mahoney static int journal_read_transaction(struct super_block *sb, 2111bd4c625cSLinus Torvalds unsigned long cur_dblock, 2112bd4c625cSLinus Torvalds unsigned long oldest_start, 2113600ed416SJeff Mahoney unsigned int oldest_trans_id, 2114bd4c625cSLinus Torvalds unsigned long newest_mount_id) 2115bd4c625cSLinus Torvalds { 2116a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 21171da177e4SLinus Torvalds struct reiserfs_journal_desc *desc; 21181da177e4SLinus Torvalds struct reiserfs_journal_commit *commit; 2119600ed416SJeff Mahoney unsigned int trans_id = 0; 21201da177e4SLinus Torvalds struct buffer_head *c_bh; 21211da177e4SLinus Torvalds struct buffer_head *d_bh; 21221da177e4SLinus Torvalds struct buffer_head **log_blocks = NULL; 21231da177e4SLinus Torvalds struct buffer_head **real_blocks = NULL; 2124600ed416SJeff Mahoney unsigned int trans_offset; 21251da177e4SLinus Torvalds int i; 21261da177e4SLinus Torvalds int trans_half; 21271da177e4SLinus Torvalds 2128a9dd3643SJeff Mahoney d_bh = journal_bread(sb, cur_dblock); 21291da177e4SLinus Torvalds if (!d_bh) 21301da177e4SLinus Torvalds return 1; 21311da177e4SLinus Torvalds desc = (struct reiserfs_journal_desc *)d_bh->b_data; 2132a9dd3643SJeff Mahoney trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb); 2133a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1037: " 21341da177e4SLinus Torvalds "journal_read_transaction, offset %llu, len %d mount_id %d", 2135a9dd3643SJeff Mahoney d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb), 21361da177e4SLinus Torvalds get_desc_trans_len(desc), get_desc_mount_id(desc)); 21371da177e4SLinus Torvalds if (get_desc_trans_id(desc) < oldest_trans_id) { 2138a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1039: " 21391da177e4SLinus Torvalds "journal_read_trans skipping because %lu is too old", 2140bd4c625cSLinus Torvalds cur_dblock - 2141a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb)); 21421da177e4SLinus Torvalds brelse(d_bh); 21431da177e4SLinus Torvalds return 1; 21441da177e4SLinus Torvalds } 21451da177e4SLinus Torvalds if (get_desc_mount_id(desc) != newest_mount_id) { 2146a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1146: " 21471da177e4SLinus Torvalds "journal_read_trans skipping because %d is != " 21481da177e4SLinus Torvalds "newest_mount_id %lu", get_desc_mount_id(desc), 21491da177e4SLinus Torvalds newest_mount_id); 21501da177e4SLinus Torvalds brelse(d_bh); 21511da177e4SLinus Torvalds return 1; 21521da177e4SLinus Torvalds } 2153a9dd3643SJeff Mahoney c_bh = journal_bread(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 21541da177e4SLinus Torvalds ((trans_offset + get_desc_trans_len(desc) + 1) % 2155a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb))); 21561da177e4SLinus Torvalds if (!c_bh) { 21571da177e4SLinus Torvalds brelse(d_bh); 21581da177e4SLinus Torvalds return 1; 21591da177e4SLinus Torvalds } 21601da177e4SLinus Torvalds commit = (struct reiserfs_journal_commit *)c_bh->b_data; 2161a9dd3643SJeff Mahoney if (journal_compare_desc_commit(sb, desc, commit)) { 2162a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2163bd4c625cSLinus Torvalds "journal_read_transaction, " 21641da177e4SLinus Torvalds "commit offset %llu had bad time %d or length %d", 2165bd4c625cSLinus Torvalds c_bh->b_blocknr - 2166a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb), 2167bd4c625cSLinus Torvalds get_commit_trans_id(commit), 2168bd4c625cSLinus Torvalds get_commit_trans_len(commit)); 21691da177e4SLinus Torvalds brelse(c_bh); 21701da177e4SLinus Torvalds brelse(d_bh); 21711da177e4SLinus Torvalds return 1; 21721da177e4SLinus Torvalds } 21733f8b5ee3SJeff Mahoney 21743f8b5ee3SJeff Mahoney if (bdev_read_only(sb->s_bdev)) { 21753f8b5ee3SJeff Mahoney reiserfs_warning(sb, "clm-2076", 21763f8b5ee3SJeff Mahoney "device is readonly, unable to replay log"); 21773f8b5ee3SJeff Mahoney brelse(c_bh); 21783f8b5ee3SJeff Mahoney brelse(d_bh); 21793f8b5ee3SJeff Mahoney return -EROFS; 21803f8b5ee3SJeff Mahoney } 21813f8b5ee3SJeff Mahoney 21821da177e4SLinus Torvalds trans_id = get_desc_trans_id(desc); 2183098297b2SJeff Mahoney /* 2184098297b2SJeff Mahoney * now we know we've got a good transaction, and it was 2185098297b2SJeff Mahoney * inside the valid time ranges 2186098297b2SJeff Mahoney */ 21876da2ec56SKees Cook log_blocks = kmalloc_array(get_desc_trans_len(desc), 21886da2ec56SKees Cook sizeof(struct buffer_head *), 21896da2ec56SKees Cook GFP_NOFS); 21906da2ec56SKees Cook real_blocks = kmalloc_array(get_desc_trans_len(desc), 21916da2ec56SKees Cook sizeof(struct buffer_head *), 21926da2ec56SKees Cook GFP_NOFS); 21931da177e4SLinus Torvalds if (!log_blocks || !real_blocks) { 21941da177e4SLinus Torvalds brelse(c_bh); 21951da177e4SLinus Torvalds brelse(d_bh); 2196d739b42bSPekka Enberg kfree(log_blocks); 2197d739b42bSPekka Enberg kfree(real_blocks); 2198a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1169", 219945b03d5eSJeff Mahoney "kmalloc failed, unable to mount FS"); 22001da177e4SLinus Torvalds return -1; 22011da177e4SLinus Torvalds } 22021da177e4SLinus Torvalds /* get all the buffer heads */ 2203a9dd3643SJeff Mahoney trans_half = journal_trans_half(sb->s_blocksize); 22041da177e4SLinus Torvalds for (i = 0; i < get_desc_trans_len(desc); i++) { 2205bd4c625cSLinus Torvalds log_blocks[i] = 2206a9dd3643SJeff Mahoney journal_getblk(sb, 2207a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2208bd4c625cSLinus Torvalds (trans_offset + 1 + 2209a9dd3643SJeff Mahoney i) % SB_ONDISK_JOURNAL_SIZE(sb)); 22101da177e4SLinus Torvalds if (i < trans_half) { 2211bd4c625cSLinus Torvalds real_blocks[i] = 2212a9dd3643SJeff Mahoney sb_getblk(sb, 2213bd4c625cSLinus Torvalds le32_to_cpu(desc->j_realblock[i])); 22141da177e4SLinus Torvalds } else { 2215bd4c625cSLinus Torvalds real_blocks[i] = 2216a9dd3643SJeff Mahoney sb_getblk(sb, 2217bd4c625cSLinus Torvalds le32_to_cpu(commit-> 2218bd4c625cSLinus Torvalds j_realblock[i - trans_half])); 22191da177e4SLinus Torvalds } 2220a9dd3643SJeff Mahoney if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(sb)) { 2221a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1207", 222245b03d5eSJeff Mahoney "REPLAY FAILURE fsck required! " 222345b03d5eSJeff Mahoney "Block to replay is outside of " 222445b03d5eSJeff Mahoney "filesystem"); 22251da177e4SLinus Torvalds goto abort_replay; 22261da177e4SLinus Torvalds } 22271da177e4SLinus Torvalds /* make sure we don't try to replay onto log or reserved area */ 2228bd4c625cSLinus Torvalds if (is_block_in_log_or_reserved_area 2229a9dd3643SJeff Mahoney (sb, real_blocks[i]->b_blocknr)) { 2230a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1204", 223145b03d5eSJeff Mahoney "REPLAY FAILURE fsck required! " 223245b03d5eSJeff Mahoney "Trying to replay onto a log block"); 22331da177e4SLinus Torvalds abort_replay: 22341da177e4SLinus Torvalds brelse_array(log_blocks, i); 22351da177e4SLinus Torvalds brelse_array(real_blocks, i); 22361da177e4SLinus Torvalds brelse(c_bh); 22371da177e4SLinus Torvalds brelse(d_bh); 2238d739b42bSPekka Enberg kfree(log_blocks); 2239d739b42bSPekka Enberg kfree(real_blocks); 22401da177e4SLinus Torvalds return -1; 22411da177e4SLinus Torvalds } 22421da177e4SLinus Torvalds } 22431da177e4SLinus Torvalds /* read in the log blocks, memcpy to the corresponding real block */ 2244dfec8a14SMike Christie ll_rw_block(REQ_OP_READ, 0, get_desc_trans_len(desc), log_blocks); 22451da177e4SLinus Torvalds for (i = 0; i < get_desc_trans_len(desc); i++) { 22468ebc4232SFrederic Weisbecker 22471da177e4SLinus Torvalds wait_on_buffer(log_blocks[i]); 22481da177e4SLinus Torvalds if (!buffer_uptodate(log_blocks[i])) { 2249a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1212", 225045b03d5eSJeff Mahoney "REPLAY FAILURE fsck required! " 225145b03d5eSJeff Mahoney "buffer write failed"); 2252bd4c625cSLinus Torvalds brelse_array(log_blocks + i, 2253bd4c625cSLinus Torvalds get_desc_trans_len(desc) - i); 22541da177e4SLinus Torvalds brelse_array(real_blocks, get_desc_trans_len(desc)); 22551da177e4SLinus Torvalds brelse(c_bh); 22561da177e4SLinus Torvalds brelse(d_bh); 2257d739b42bSPekka Enberg kfree(log_blocks); 2258d739b42bSPekka Enberg kfree(real_blocks); 22591da177e4SLinus Torvalds return -1; 22601da177e4SLinus Torvalds } 2261bd4c625cSLinus Torvalds memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data, 2262bd4c625cSLinus Torvalds real_blocks[i]->b_size); 22631da177e4SLinus Torvalds set_buffer_uptodate(real_blocks[i]); 22641da177e4SLinus Torvalds brelse(log_blocks[i]); 22651da177e4SLinus Torvalds } 22661da177e4SLinus Torvalds /* flush out the real blocks */ 22671da177e4SLinus Torvalds for (i = 0; i < get_desc_trans_len(desc); i++) { 22681da177e4SLinus Torvalds set_buffer_dirty(real_blocks[i]); 22692a222ca9SMike Christie write_dirty_buffer(real_blocks[i], 0); 22701da177e4SLinus Torvalds } 22711da177e4SLinus Torvalds for (i = 0; i < get_desc_trans_len(desc); i++) { 22721da177e4SLinus Torvalds wait_on_buffer(real_blocks[i]); 22731da177e4SLinus Torvalds if (!buffer_uptodate(real_blocks[i])) { 2274a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1226", 227545b03d5eSJeff Mahoney "REPLAY FAILURE, fsck required! " 227645b03d5eSJeff Mahoney "buffer write failed"); 2277bd4c625cSLinus Torvalds brelse_array(real_blocks + i, 2278bd4c625cSLinus Torvalds get_desc_trans_len(desc) - i); 22791da177e4SLinus Torvalds brelse(c_bh); 22801da177e4SLinus Torvalds brelse(d_bh); 2281d739b42bSPekka Enberg kfree(log_blocks); 2282d739b42bSPekka Enberg kfree(real_blocks); 22831da177e4SLinus Torvalds return -1; 22841da177e4SLinus Torvalds } 22851da177e4SLinus Torvalds brelse(real_blocks[i]); 22861da177e4SLinus Torvalds } 2287bd4c625cSLinus Torvalds cur_dblock = 2288a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2289bd4c625cSLinus Torvalds ((trans_offset + get_desc_trans_len(desc) + 2290a9dd3643SJeff Mahoney 2) % SB_ONDISK_JOURNAL_SIZE(sb)); 2291a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2292bd4c625cSLinus Torvalds "journal-1095: setting journal " "start to offset %ld", 2293a9dd3643SJeff Mahoney cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb)); 22941da177e4SLinus Torvalds 2295098297b2SJeff Mahoney /* 2296098297b2SJeff Mahoney * init starting values for the first transaction, in case 2297098297b2SJeff Mahoney * this is the last transaction to be replayed. 2298098297b2SJeff Mahoney */ 2299a9dd3643SJeff Mahoney journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb); 23001da177e4SLinus Torvalds journal->j_last_flush_trans_id = trans_id; 23011da177e4SLinus Torvalds journal->j_trans_id = trans_id + 1; 2302a44c94a7SAlexander Zarochentsev /* check for trans_id overflow */ 2303a44c94a7SAlexander Zarochentsev if (journal->j_trans_id == 0) 2304a44c94a7SAlexander Zarochentsev journal->j_trans_id = 10; 23051da177e4SLinus Torvalds brelse(c_bh); 23061da177e4SLinus Torvalds brelse(d_bh); 2307d739b42bSPekka Enberg kfree(log_blocks); 2308d739b42bSPekka Enberg kfree(real_blocks); 23091da177e4SLinus Torvalds return 0; 23101da177e4SLinus Torvalds } 23111da177e4SLinus Torvalds 2312098297b2SJeff Mahoney /* 2313098297b2SJeff Mahoney * This function reads blocks starting from block and to max_block of bufsize 2314098297b2SJeff Mahoney * size (but no more than BUFNR blocks at a time). This proved to improve 2315098297b2SJeff Mahoney * mounting speed on self-rebuilding raid5 arrays at least. 2316098297b2SJeff Mahoney * Right now it is only used from journal code. But later we might use it 2317098297b2SJeff Mahoney * from other places. 2318098297b2SJeff Mahoney * Note: Do not use journal_getblk/sb_getblk functions here! 2319098297b2SJeff Mahoney */ 23203ee16670SJeff Mahoney static struct buffer_head *reiserfs_breada(struct block_device *dev, 23213ee16670SJeff Mahoney b_blocknr_t block, int bufsize, 23223ee16670SJeff Mahoney b_blocknr_t max_block) 23231da177e4SLinus Torvalds { 23241da177e4SLinus Torvalds struct buffer_head *bhlist[BUFNR]; 23251da177e4SLinus Torvalds unsigned int blocks = BUFNR; 23261da177e4SLinus Torvalds struct buffer_head *bh; 23271da177e4SLinus Torvalds int i, j; 23281da177e4SLinus Torvalds 23291da177e4SLinus Torvalds bh = __getblk(dev, block, bufsize); 23301da177e4SLinus Torvalds if (buffer_uptodate(bh)) 23311da177e4SLinus Torvalds return (bh); 23321da177e4SLinus Torvalds 23331da177e4SLinus Torvalds if (block + BUFNR > max_block) { 23341da177e4SLinus Torvalds blocks = max_block - block; 23351da177e4SLinus Torvalds } 23361da177e4SLinus Torvalds bhlist[0] = bh; 23371da177e4SLinus Torvalds j = 1; 23381da177e4SLinus Torvalds for (i = 1; i < blocks; i++) { 23391da177e4SLinus Torvalds bh = __getblk(dev, block + i, bufsize); 23401da177e4SLinus Torvalds if (buffer_uptodate(bh)) { 23411da177e4SLinus Torvalds brelse(bh); 23421da177e4SLinus Torvalds break; 2343bd4c625cSLinus Torvalds } else 2344bd4c625cSLinus Torvalds bhlist[j++] = bh; 23451da177e4SLinus Torvalds } 2346dfec8a14SMike Christie ll_rw_block(REQ_OP_READ, 0, j, bhlist); 23471da177e4SLinus Torvalds for (i = 1; i < j; i++) 23481da177e4SLinus Torvalds brelse(bhlist[i]); 23491da177e4SLinus Torvalds bh = bhlist[0]; 23501da177e4SLinus Torvalds wait_on_buffer(bh); 23511da177e4SLinus Torvalds if (buffer_uptodate(bh)) 23521da177e4SLinus Torvalds return bh; 23531da177e4SLinus Torvalds brelse(bh); 23541da177e4SLinus Torvalds return NULL; 23551da177e4SLinus Torvalds } 23561da177e4SLinus Torvalds 23571da177e4SLinus Torvalds /* 2358098297b2SJeff Mahoney * read and replay the log 2359098297b2SJeff Mahoney * on a clean unmount, the journal header's next unflushed pointer will be 2360098297b2SJeff Mahoney * to an invalid transaction. This tests that before finding all the 2361098297b2SJeff Mahoney * transactions in the log, which makes normal mount times fast. 2362098297b2SJeff Mahoney * 2363098297b2SJeff Mahoney * After a crash, this starts with the next unflushed transaction, and 2364098297b2SJeff Mahoney * replays until it finds one too old, or invalid. 2365098297b2SJeff Mahoney * 2366098297b2SJeff Mahoney * On exit, it sets things up so the first transaction will work correctly. 2367098297b2SJeff Mahoney * NOTE: only called during fs mount 23681da177e4SLinus Torvalds */ 2369a9dd3643SJeff Mahoney static int journal_read(struct super_block *sb) 2370bd4c625cSLinus Torvalds { 2371a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 23721da177e4SLinus Torvalds struct reiserfs_journal_desc *desc; 2373600ed416SJeff Mahoney unsigned int oldest_trans_id = 0; 2374600ed416SJeff Mahoney unsigned int oldest_invalid_trans_id = 0; 237534d08260SArnd Bergmann time64_t start; 23761da177e4SLinus Torvalds unsigned long oldest_start = 0; 23771da177e4SLinus Torvalds unsigned long cur_dblock = 0; 23781da177e4SLinus Torvalds unsigned long newest_mount_id = 9; 23791da177e4SLinus Torvalds struct buffer_head *d_bh; 23801da177e4SLinus Torvalds struct reiserfs_journal_header *jh; 23811da177e4SLinus Torvalds int valid_journal_header = 0; 23821da177e4SLinus Torvalds int replay_count = 0; 23831da177e4SLinus Torvalds int continue_replay = 1; 23841da177e4SLinus Torvalds int ret; 23851da177e4SLinus Torvalds 2386a9dd3643SJeff Mahoney cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(sb); 2387a1c6f057SDmitry Monakhov reiserfs_info(sb, "checking transaction log (%pg)\n", 2388a1c6f057SDmitry Monakhov journal->j_dev_bd); 238934d08260SArnd Bergmann start = ktime_get_seconds(); 23901da177e4SLinus Torvalds 2391098297b2SJeff Mahoney /* 2392098297b2SJeff Mahoney * step 1, read in the journal header block. Check the transaction 2393098297b2SJeff Mahoney * it says is the first unflushed, and if that transaction is not 2394098297b2SJeff Mahoney * valid, replay is done 23951da177e4SLinus Torvalds */ 2396a9dd3643SJeff Mahoney journal->j_header_bh = journal_bread(sb, 2397a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) 2398a9dd3643SJeff Mahoney + SB_ONDISK_JOURNAL_SIZE(sb)); 23991da177e4SLinus Torvalds if (!journal->j_header_bh) { 24001da177e4SLinus Torvalds return 1; 24011da177e4SLinus Torvalds } 24021da177e4SLinus Torvalds jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data); 2403c499ec24SVladimir V. Saveliev if (le32_to_cpu(jh->j_first_unflushed_offset) < 2404a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb) 2405bd4c625cSLinus Torvalds && le32_to_cpu(jh->j_last_flush_trans_id) > 0) { 2406bd4c625cSLinus Torvalds oldest_start = 2407a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 24081da177e4SLinus Torvalds le32_to_cpu(jh->j_first_unflushed_offset); 24091da177e4SLinus Torvalds oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1; 24101da177e4SLinus Torvalds newest_mount_id = le32_to_cpu(jh->j_mount_id); 2411a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2412bd4c625cSLinus Torvalds "journal-1153: found in " 24131da177e4SLinus Torvalds "header: first_unflushed_offset %d, last_flushed_trans_id " 24141da177e4SLinus Torvalds "%lu", le32_to_cpu(jh->j_first_unflushed_offset), 24151da177e4SLinus Torvalds le32_to_cpu(jh->j_last_flush_trans_id)); 24161da177e4SLinus Torvalds valid_journal_header = 1; 24171da177e4SLinus Torvalds 2418098297b2SJeff Mahoney /* 2419098297b2SJeff Mahoney * now, we try to read the first unflushed offset. If it 2420098297b2SJeff Mahoney * is not valid, there is nothing more we can do, and it 2421098297b2SJeff Mahoney * makes no sense to read through the whole log. 24221da177e4SLinus Torvalds */ 2423bd4c625cSLinus Torvalds d_bh = 2424a9dd3643SJeff Mahoney journal_bread(sb, 2425a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2426bd4c625cSLinus Torvalds le32_to_cpu(jh->j_first_unflushed_offset)); 2427a9dd3643SJeff Mahoney ret = journal_transaction_is_valid(sb, d_bh, NULL, NULL); 24281da177e4SLinus Torvalds if (!ret) { 24291da177e4SLinus Torvalds continue_replay = 0; 24301da177e4SLinus Torvalds } 24311da177e4SLinus Torvalds brelse(d_bh); 24321da177e4SLinus Torvalds goto start_log_replay; 24331da177e4SLinus Torvalds } 24341da177e4SLinus Torvalds 2435098297b2SJeff Mahoney /* 2436098297b2SJeff Mahoney * ok, there are transactions that need to be replayed. start 2437098297b2SJeff Mahoney * with the first log block, find all the valid transactions, and 2438098297b2SJeff Mahoney * pick out the oldest. 24391da177e4SLinus Torvalds */ 2440bd4c625cSLinus Torvalds while (continue_replay 2441bd4c625cSLinus Torvalds && cur_dblock < 2442a9dd3643SJeff Mahoney (SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2443a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb))) { 2444098297b2SJeff Mahoney /* 2445098297b2SJeff Mahoney * Note that it is required for blocksize of primary fs 2446098297b2SJeff Mahoney * device and journal device to be the same 2447098297b2SJeff Mahoney */ 2448bd4c625cSLinus Torvalds d_bh = 2449bd4c625cSLinus Torvalds reiserfs_breada(journal->j_dev_bd, cur_dblock, 2450a9dd3643SJeff Mahoney sb->s_blocksize, 2451a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2452a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb)); 2453bd4c625cSLinus Torvalds ret = 2454a9dd3643SJeff Mahoney journal_transaction_is_valid(sb, d_bh, 2455bd4c625cSLinus Torvalds &oldest_invalid_trans_id, 2456bd4c625cSLinus Torvalds &newest_mount_id); 24571da177e4SLinus Torvalds if (ret == 1) { 24581da177e4SLinus Torvalds desc = (struct reiserfs_journal_desc *)d_bh->b_data; 24591da177e4SLinus Torvalds if (oldest_start == 0) { /* init all oldest_ values */ 24601da177e4SLinus Torvalds oldest_trans_id = get_desc_trans_id(desc); 24611da177e4SLinus Torvalds oldest_start = d_bh->b_blocknr; 24621da177e4SLinus Torvalds newest_mount_id = get_desc_mount_id(desc); 2463a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2464bd4c625cSLinus Torvalds "journal-1179: Setting " 24651da177e4SLinus Torvalds "oldest_start to offset %llu, trans_id %lu", 2466bd4c625cSLinus Torvalds oldest_start - 2467bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK 2468a9dd3643SJeff Mahoney (sb), oldest_trans_id); 24691da177e4SLinus Torvalds } else if (oldest_trans_id > get_desc_trans_id(desc)) { 24701da177e4SLinus Torvalds /* one we just read was older */ 24711da177e4SLinus Torvalds oldest_trans_id = get_desc_trans_id(desc); 24721da177e4SLinus Torvalds oldest_start = d_bh->b_blocknr; 2473a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2474bd4c625cSLinus Torvalds "journal-1180: Resetting " 24751da177e4SLinus Torvalds "oldest_start to offset %lu, trans_id %lu", 2476bd4c625cSLinus Torvalds oldest_start - 2477bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK 2478a9dd3643SJeff Mahoney (sb), oldest_trans_id); 24791da177e4SLinus Torvalds } 24801da177e4SLinus Torvalds if (newest_mount_id < get_desc_mount_id(desc)) { 24811da177e4SLinus Torvalds newest_mount_id = get_desc_mount_id(desc); 2482a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2483bd4c625cSLinus Torvalds "journal-1299: Setting " 2484bd4c625cSLinus Torvalds "newest_mount_id to %d", 2485bd4c625cSLinus Torvalds get_desc_mount_id(desc)); 24861da177e4SLinus Torvalds } 24871da177e4SLinus Torvalds cur_dblock += get_desc_trans_len(desc) + 2; 24881da177e4SLinus Torvalds } else { 24891da177e4SLinus Torvalds cur_dblock++; 24901da177e4SLinus Torvalds } 24911da177e4SLinus Torvalds brelse(d_bh); 24921da177e4SLinus Torvalds } 24931da177e4SLinus Torvalds 24941da177e4SLinus Torvalds start_log_replay: 24951da177e4SLinus Torvalds cur_dblock = oldest_start; 24961da177e4SLinus Torvalds if (oldest_trans_id) { 2497a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2498bd4c625cSLinus Torvalds "journal-1206: Starting replay " 24991da177e4SLinus Torvalds "from offset %llu, trans_id %lu", 2500a9dd3643SJeff Mahoney cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb), 25011da177e4SLinus Torvalds oldest_trans_id); 25021da177e4SLinus Torvalds 25031da177e4SLinus Torvalds } 25041da177e4SLinus Torvalds replay_count = 0; 25051da177e4SLinus Torvalds while (continue_replay && oldest_trans_id > 0) { 2506bd4c625cSLinus Torvalds ret = 2507a9dd3643SJeff Mahoney journal_read_transaction(sb, cur_dblock, oldest_start, 2508bd4c625cSLinus Torvalds oldest_trans_id, newest_mount_id); 25091da177e4SLinus Torvalds if (ret < 0) { 25101da177e4SLinus Torvalds return ret; 25111da177e4SLinus Torvalds } else if (ret != 0) { 25121da177e4SLinus Torvalds break; 25131da177e4SLinus Torvalds } 2514bd4c625cSLinus Torvalds cur_dblock = 2515a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + journal->j_start; 25161da177e4SLinus Torvalds replay_count++; 25171da177e4SLinus Torvalds if (cur_dblock == oldest_start) 25181da177e4SLinus Torvalds break; 25191da177e4SLinus Torvalds } 25201da177e4SLinus Torvalds 25211da177e4SLinus Torvalds if (oldest_trans_id == 0) { 2522a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2523bd4c625cSLinus Torvalds "journal-1225: No valid " "transactions found"); 25241da177e4SLinus Torvalds } 2525098297b2SJeff Mahoney /* 2526098297b2SJeff Mahoney * j_start does not get set correctly if we don't replay any 2527098297b2SJeff Mahoney * transactions. if we had a valid journal_header, set j_start 2528098297b2SJeff Mahoney * to the first unflushed transaction value, copy the trans_id 2529098297b2SJeff Mahoney * from the header 25301da177e4SLinus Torvalds */ 25311da177e4SLinus Torvalds if (valid_journal_header && replay_count == 0) { 25321da177e4SLinus Torvalds journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset); 2533bd4c625cSLinus Torvalds journal->j_trans_id = 2534bd4c625cSLinus Torvalds le32_to_cpu(jh->j_last_flush_trans_id) + 1; 2535a44c94a7SAlexander Zarochentsev /* check for trans_id overflow */ 2536a44c94a7SAlexander Zarochentsev if (journal->j_trans_id == 0) 2537a44c94a7SAlexander Zarochentsev journal->j_trans_id = 10; 2538bd4c625cSLinus Torvalds journal->j_last_flush_trans_id = 2539bd4c625cSLinus Torvalds le32_to_cpu(jh->j_last_flush_trans_id); 25401da177e4SLinus Torvalds journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1; 25411da177e4SLinus Torvalds } else { 25421da177e4SLinus Torvalds journal->j_mount_id = newest_mount_id + 1; 25431da177e4SLinus Torvalds } 2544a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1299: Setting " 25451da177e4SLinus Torvalds "newest_mount_id to %lu", journal->j_mount_id); 25461da177e4SLinus Torvalds journal->j_first_unflushed_offset = journal->j_start; 25471da177e4SLinus Torvalds if (replay_count > 0) { 2548a9dd3643SJeff Mahoney reiserfs_info(sb, 2549bd4c625cSLinus Torvalds "replayed %d transactions in %lu seconds\n", 255034d08260SArnd Bergmann replay_count, ktime_get_seconds() - start); 25511da177e4SLinus Torvalds } 2552278f6679SJeff Mahoney /* needed to satisfy the locking in _update_journal_header_block */ 2553278f6679SJeff Mahoney reiserfs_write_lock(sb); 2554a9dd3643SJeff Mahoney if (!bdev_read_only(sb->s_bdev) && 2555a9dd3643SJeff Mahoney _update_journal_header_block(sb, journal->j_start, 2556bd4c625cSLinus Torvalds journal->j_last_flush_trans_id)) { 2557278f6679SJeff Mahoney reiserfs_write_unlock(sb); 2558098297b2SJeff Mahoney /* 2559098297b2SJeff Mahoney * replay failed, caller must call free_journal_ram and abort 2560098297b2SJeff Mahoney * the mount 25611da177e4SLinus Torvalds */ 25621da177e4SLinus Torvalds return -1; 25631da177e4SLinus Torvalds } 2564278f6679SJeff Mahoney reiserfs_write_unlock(sb); 25651da177e4SLinus Torvalds return 0; 25661da177e4SLinus Torvalds } 25671da177e4SLinus Torvalds 25681da177e4SLinus Torvalds static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s) 25691da177e4SLinus Torvalds { 25701da177e4SLinus Torvalds struct reiserfs_journal_list *jl; 25718c777cc4SPekka Enberg jl = kzalloc(sizeof(struct reiserfs_journal_list), 25728c777cc4SPekka Enberg GFP_NOFS | __GFP_NOFAIL); 25731da177e4SLinus Torvalds INIT_LIST_HEAD(&jl->j_list); 25741da177e4SLinus Torvalds INIT_LIST_HEAD(&jl->j_working_list); 25751da177e4SLinus Torvalds INIT_LIST_HEAD(&jl->j_tail_bh_list); 25761da177e4SLinus Torvalds INIT_LIST_HEAD(&jl->j_bh_list); 257790415deaSJeff Mahoney mutex_init(&jl->j_commit_mutex); 25781da177e4SLinus Torvalds SB_JOURNAL(s)->j_num_lists++; 25791da177e4SLinus Torvalds get_journal_list(jl); 25801da177e4SLinus Torvalds return jl; 25811da177e4SLinus Torvalds } 25821da177e4SLinus Torvalds 2583a9dd3643SJeff Mahoney static void journal_list_init(struct super_block *sb) 2584bd4c625cSLinus Torvalds { 2585a9dd3643SJeff Mahoney SB_JOURNAL(sb)->j_current_jl = alloc_journal_list(sb); 25861da177e4SLinus Torvalds } 25871da177e4SLinus Torvalds 25884385bab1SAl Viro static void release_journal_dev(struct super_block *super, 25891da177e4SLinus Torvalds struct reiserfs_journal *journal) 25901da177e4SLinus Torvalds { 259186098fa0SChristoph Hellwig if (journal->j_dev_bd != NULL) { 25924385bab1SAl Viro blkdev_put(journal->j_dev_bd, journal->j_dev_mode); 25931da177e4SLinus Torvalds journal->j_dev_bd = NULL; 25941da177e4SLinus Torvalds } 25951da177e4SLinus Torvalds } 25961da177e4SLinus Torvalds 25971da177e4SLinus Torvalds static int journal_init_dev(struct super_block *super, 25981da177e4SLinus Torvalds struct reiserfs_journal *journal, 25991da177e4SLinus Torvalds const char *jdev_name) 26001da177e4SLinus Torvalds { 26011da177e4SLinus Torvalds int result; 26021da177e4SLinus Torvalds dev_t jdev; 2603e525fd89STejun Heo fmode_t blkdev_mode = FMODE_READ | FMODE_WRITE | FMODE_EXCL; 26041da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 26051da177e4SLinus Torvalds 26061da177e4SLinus Torvalds result = 0; 26071da177e4SLinus Torvalds 26081da177e4SLinus Torvalds journal->j_dev_bd = NULL; 26091da177e4SLinus Torvalds jdev = SB_ONDISK_JOURNAL_DEVICE(super) ? 26101da177e4SLinus Torvalds new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev; 26111da177e4SLinus Torvalds 26121da177e4SLinus Torvalds if (bdev_read_only(super->s_bdev)) 26131da177e4SLinus Torvalds blkdev_mode = FMODE_READ; 26141da177e4SLinus Torvalds 26151da177e4SLinus Torvalds /* there is no "jdev" option and journal is on separate device */ 26161da177e4SLinus Torvalds if ((!jdev_name || !jdev_name[0])) { 2617e525fd89STejun Heo if (jdev == super->s_dev) 2618e525fd89STejun Heo blkdev_mode &= ~FMODE_EXCL; 2619d4d77629STejun Heo journal->j_dev_bd = blkdev_get_by_dev(jdev, blkdev_mode, 2620d4d77629STejun Heo journal); 2621e5eb8caaSAl Viro journal->j_dev_mode = blkdev_mode; 26221da177e4SLinus Torvalds if (IS_ERR(journal->j_dev_bd)) { 26231da177e4SLinus Torvalds result = PTR_ERR(journal->j_dev_bd); 26241da177e4SLinus Torvalds journal->j_dev_bd = NULL; 262545b03d5eSJeff Mahoney reiserfs_warning(super, "sh-458", 26261da177e4SLinus Torvalds "cannot init journal device '%s': %i", 26271da177e4SLinus Torvalds __bdevname(jdev, b), result); 26281da177e4SLinus Torvalds return result; 2629e525fd89STejun Heo } else if (jdev != super->s_dev) 26301da177e4SLinus Torvalds set_blocksize(journal->j_dev_bd, super->s_blocksize); 263186098fa0SChristoph Hellwig 26321da177e4SLinus Torvalds return 0; 26331da177e4SLinus Torvalds } 26341da177e4SLinus Torvalds 2635e5eb8caaSAl Viro journal->j_dev_mode = blkdev_mode; 2636d4d77629STejun Heo journal->j_dev_bd = blkdev_get_by_path(jdev_name, blkdev_mode, journal); 263786098fa0SChristoph Hellwig if (IS_ERR(journal->j_dev_bd)) { 263886098fa0SChristoph Hellwig result = PTR_ERR(journal->j_dev_bd); 263986098fa0SChristoph Hellwig journal->j_dev_bd = NULL; 26409ad553abSAndrew Morton reiserfs_warning(super, "sh-457", 264186098fa0SChristoph Hellwig "journal_init_dev: Cannot open '%s': %i", 264286098fa0SChristoph Hellwig jdev_name, result); 264386098fa0SChristoph Hellwig return result; 264486098fa0SChristoph Hellwig } 264586098fa0SChristoph Hellwig 26461da177e4SLinus Torvalds set_blocksize(journal->j_dev_bd, super->s_blocksize); 2647bd4c625cSLinus Torvalds reiserfs_info(super, 2648a1c6f057SDmitry Monakhov "journal_init_dev: journal device: %pg\n", 2649a1c6f057SDmitry Monakhov journal->j_dev_bd); 265086098fa0SChristoph Hellwig return 0; 26511da177e4SLinus Torvalds } 26521da177e4SLinus Torvalds 2653098297b2SJeff Mahoney /* 2654cf3d0b81SEdward Shishkin * When creating/tuning a file system user can assign some 2655cf3d0b81SEdward Shishkin * journal params within boundaries which depend on the ratio 2656cf3d0b81SEdward Shishkin * blocksize/standard_blocksize. 2657cf3d0b81SEdward Shishkin * 2658cf3d0b81SEdward Shishkin * For blocks >= standard_blocksize transaction size should 2659cf3d0b81SEdward Shishkin * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more 2660cf3d0b81SEdward Shishkin * then JOURNAL_TRANS_MAX_DEFAULT. 2661cf3d0b81SEdward Shishkin * 2662cf3d0b81SEdward Shishkin * For blocks < standard_blocksize these boundaries should be 2663cf3d0b81SEdward Shishkin * decreased proportionally. 2664cf3d0b81SEdward Shishkin */ 2665cf3d0b81SEdward Shishkin #define REISERFS_STANDARD_BLKSIZE (4096) 2666cf3d0b81SEdward Shishkin 2667a9dd3643SJeff Mahoney static int check_advise_trans_params(struct super_block *sb, 2668cf3d0b81SEdward Shishkin struct reiserfs_journal *journal) 2669cf3d0b81SEdward Shishkin { 2670cf3d0b81SEdward Shishkin if (journal->j_trans_max) { 2671098297b2SJeff Mahoney /* Non-default journal params. Do sanity check for them. */ 2672cf3d0b81SEdward Shishkin int ratio = 1; 2673a9dd3643SJeff Mahoney if (sb->s_blocksize < REISERFS_STANDARD_BLKSIZE) 2674a9dd3643SJeff Mahoney ratio = REISERFS_STANDARD_BLKSIZE / sb->s_blocksize; 2675cf3d0b81SEdward Shishkin 2676cf3d0b81SEdward Shishkin if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio || 2677cf3d0b81SEdward Shishkin journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio || 2678a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb) / journal->j_trans_max < 2679cf3d0b81SEdward Shishkin JOURNAL_MIN_RATIO) { 2680a9dd3643SJeff Mahoney reiserfs_warning(sb, "sh-462", 268145b03d5eSJeff Mahoney "bad transaction max size (%u). " 268245b03d5eSJeff Mahoney "FSCK?", journal->j_trans_max); 2683cf3d0b81SEdward Shishkin return 1; 2684cf3d0b81SEdward Shishkin } 2685cf3d0b81SEdward Shishkin if (journal->j_max_batch != (journal->j_trans_max) * 2686cf3d0b81SEdward Shishkin JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) { 2687a9dd3643SJeff Mahoney reiserfs_warning(sb, "sh-463", 268845b03d5eSJeff Mahoney "bad transaction max batch (%u). " 268945b03d5eSJeff Mahoney "FSCK?", journal->j_max_batch); 2690cf3d0b81SEdward Shishkin return 1; 2691cf3d0b81SEdward Shishkin } 2692cf3d0b81SEdward Shishkin } else { 2693098297b2SJeff Mahoney /* 2694098297b2SJeff Mahoney * Default journal params. 2695098297b2SJeff Mahoney * The file system was created by old version 2696098297b2SJeff Mahoney * of mkreiserfs, so some fields contain zeros, 2697098297b2SJeff Mahoney * and we need to advise proper values for them 2698098297b2SJeff Mahoney */ 2699a9dd3643SJeff Mahoney if (sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) { 2700a9dd3643SJeff Mahoney reiserfs_warning(sb, "sh-464", "bad blocksize (%u)", 2701a9dd3643SJeff Mahoney sb->s_blocksize); 270245b03d5eSJeff Mahoney return 1; 270345b03d5eSJeff Mahoney } 2704cf3d0b81SEdward Shishkin journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT; 2705cf3d0b81SEdward Shishkin journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT; 2706cf3d0b81SEdward Shishkin journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE; 2707cf3d0b81SEdward Shishkin } 2708cf3d0b81SEdward Shishkin return 0; 2709cf3d0b81SEdward Shishkin } 2710cf3d0b81SEdward Shishkin 2711098297b2SJeff Mahoney /* must be called once on fs mount. calls journal_read for you */ 2712a9dd3643SJeff Mahoney int journal_init(struct super_block *sb, const char *j_dev_name, 2713bd4c625cSLinus Torvalds int old_format, unsigned int commit_max_age) 2714bd4c625cSLinus Torvalds { 2715a9dd3643SJeff Mahoney int num_cnodes = SB_ONDISK_JOURNAL_SIZE(sb) * 2; 27161da177e4SLinus Torvalds struct buffer_head *bhjh; 27171da177e4SLinus Torvalds struct reiserfs_super_block *rs; 27181da177e4SLinus Torvalds struct reiserfs_journal_header *jh; 27191da177e4SLinus Torvalds struct reiserfs_journal *journal; 27201da177e4SLinus Torvalds struct reiserfs_journal_list *jl; 272198ea3f50SFrederic Weisbecker int ret; 27221da177e4SLinus Torvalds 2723558feb08SJoe Perches journal = SB_JOURNAL(sb) = vzalloc(sizeof(struct reiserfs_journal)); 27241da177e4SLinus Torvalds if (!journal) { 2725a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1256", 272645b03d5eSJeff Mahoney "unable to get memory for journal structure"); 27271da177e4SLinus Torvalds return 1; 27281da177e4SLinus Torvalds } 27291da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_bitmap_nodes); 27301da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_prealloc_list); 27311da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_working_list); 27321da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_journal_list); 27331da177e4SLinus Torvalds journal->j_persistent_trans = 0; 273437c69b98SFrederic Weisbecker if (reiserfs_allocate_list_bitmaps(sb, journal->j_list_bitmap, 273537c69b98SFrederic Weisbecker reiserfs_bmap_count(sb))) 27361da177e4SLinus Torvalds goto free_and_return; 273798ea3f50SFrederic Weisbecker 2738a9dd3643SJeff Mahoney allocate_bitmap_nodes(sb); 27391da177e4SLinus Torvalds 27401da177e4SLinus Torvalds /* reserved for journal area support */ 2741a9dd3643SJeff Mahoney SB_JOURNAL_1st_RESERVED_BLOCK(sb) = (old_format ? 2742bd4c625cSLinus Torvalds REISERFS_OLD_DISK_OFFSET_IN_BYTES 2743a9dd3643SJeff Mahoney / sb->s_blocksize + 2744a9dd3643SJeff Mahoney reiserfs_bmap_count(sb) + 2745bd4c625cSLinus Torvalds 1 : 2746bd4c625cSLinus Torvalds REISERFS_DISK_OFFSET_IN_BYTES / 2747a9dd3643SJeff Mahoney sb->s_blocksize + 2); 27481da177e4SLinus Torvalds 2749098297b2SJeff Mahoney /* 2750098297b2SJeff Mahoney * Sanity check to see is the standard journal fitting 2751098297b2SJeff Mahoney * within first bitmap (actual for small blocksizes) 2752098297b2SJeff Mahoney */ 2753a9dd3643SJeff Mahoney if (!SB_ONDISK_JOURNAL_DEVICE(sb) && 2754a9dd3643SJeff Mahoney (SB_JOURNAL_1st_RESERVED_BLOCK(sb) + 2755a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb) > sb->s_blocksize * 8)) { 2756a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1393", 275745b03d5eSJeff Mahoney "journal does not fit for area addressed " 275845b03d5eSJeff Mahoney "by first of bitmap blocks. It starts at " 27591da177e4SLinus Torvalds "%u and its size is %u. Block size %ld", 2760a9dd3643SJeff Mahoney SB_JOURNAL_1st_RESERVED_BLOCK(sb), 2761a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb), 2762a9dd3643SJeff Mahoney sb->s_blocksize); 27631da177e4SLinus Torvalds goto free_and_return; 27641da177e4SLinus Torvalds } 27651da177e4SLinus Torvalds 2766a9dd3643SJeff Mahoney if (journal_init_dev(sb, journal, j_dev_name) != 0) { 2767a9dd3643SJeff Mahoney reiserfs_warning(sb, "sh-462", 27686774def6SMasanari Iida "unable to initialize journal device"); 27691da177e4SLinus Torvalds goto free_and_return; 27701da177e4SLinus Torvalds } 27711da177e4SLinus Torvalds 2772a9dd3643SJeff Mahoney rs = SB_DISK_SUPER_BLOCK(sb); 27731da177e4SLinus Torvalds 27741da177e4SLinus Torvalds /* read journal header */ 2775a9dd3643SJeff Mahoney bhjh = journal_bread(sb, 2776a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2777a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb)); 27781da177e4SLinus Torvalds if (!bhjh) { 2779a9dd3643SJeff Mahoney reiserfs_warning(sb, "sh-459", 278045b03d5eSJeff Mahoney "unable to read journal header"); 27811da177e4SLinus Torvalds goto free_and_return; 27821da177e4SLinus Torvalds } 27831da177e4SLinus Torvalds jh = (struct reiserfs_journal_header *)(bhjh->b_data); 27841da177e4SLinus Torvalds 27851da177e4SLinus Torvalds /* make sure that journal matches to the super block */ 2786bd4c625cSLinus Torvalds if (is_reiserfs_jr(rs) 2787bd4c625cSLinus Torvalds && (le32_to_cpu(jh->jh_journal.jp_journal_magic) != 2788bd4c625cSLinus Torvalds sb_jp_journal_magic(rs))) { 2789a9dd3643SJeff Mahoney reiserfs_warning(sb, "sh-460", 2790a1c6f057SDmitry Monakhov "journal header magic %x (device %pg) does " 279145b03d5eSJeff Mahoney "not match to magic found in super block %x", 279245b03d5eSJeff Mahoney jh->jh_journal.jp_journal_magic, 2793a1c6f057SDmitry Monakhov journal->j_dev_bd, 27941da177e4SLinus Torvalds sb_jp_journal_magic(rs)); 27951da177e4SLinus Torvalds brelse(bhjh); 27961da177e4SLinus Torvalds goto free_and_return; 27971da177e4SLinus Torvalds } 27981da177e4SLinus Torvalds 27991da177e4SLinus Torvalds journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max); 28001da177e4SLinus Torvalds journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch); 2801bd4c625cSLinus Torvalds journal->j_max_commit_age = 2802bd4c625cSLinus Torvalds le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age); 28031da177e4SLinus Torvalds journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE; 28041da177e4SLinus Torvalds 2805a9dd3643SJeff Mahoney if (check_advise_trans_params(sb, journal) != 0) 2806cf3d0b81SEdward Shishkin goto free_and_return; 28071da177e4SLinus Torvalds journal->j_default_max_commit_age = journal->j_max_commit_age; 28081da177e4SLinus Torvalds 28091da177e4SLinus Torvalds if (commit_max_age != 0) { 28101da177e4SLinus Torvalds journal->j_max_commit_age = commit_max_age; 28111da177e4SLinus Torvalds journal->j_max_trans_age = commit_max_age; 28121da177e4SLinus Torvalds } 28131da177e4SLinus Torvalds 2814a1c6f057SDmitry Monakhov reiserfs_info(sb, "journal params: device %pg, size %u, " 28151da177e4SLinus Torvalds "journal first block %u, max trans len %u, max batch %u, " 28161da177e4SLinus Torvalds "max commit age %u, max trans age %u\n", 2817a1c6f057SDmitry Monakhov journal->j_dev_bd, 2818a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb), 2819a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb), 28201da177e4SLinus Torvalds journal->j_trans_max, 28211da177e4SLinus Torvalds journal->j_max_batch, 2822bd4c625cSLinus Torvalds journal->j_max_commit_age, journal->j_max_trans_age); 28231da177e4SLinus Torvalds 28241da177e4SLinus Torvalds brelse(bhjh); 28251da177e4SLinus Torvalds 28261da177e4SLinus Torvalds journal->j_list_bitmap_index = 0; 2827a9dd3643SJeff Mahoney journal_list_init(sb); 28281da177e4SLinus Torvalds 2829bd4c625cSLinus Torvalds memset(journal->j_list_hash_table, 0, 2830bd4c625cSLinus Torvalds JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *)); 28311da177e4SLinus Torvalds 28321da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_dirty_buffers); 28331da177e4SLinus Torvalds spin_lock_init(&journal->j_dirty_buffers_lock); 28341da177e4SLinus Torvalds 28351da177e4SLinus Torvalds journal->j_start = 0; 28361da177e4SLinus Torvalds journal->j_len = 0; 28371da177e4SLinus Torvalds journal->j_len_alloc = 0; 2838a228bf8fSJeff Mahoney atomic_set(&journal->j_wcount, 0); 2839a228bf8fSJeff Mahoney atomic_set(&journal->j_async_throttle, 0); 28401da177e4SLinus Torvalds journal->j_bcount = 0; 28411da177e4SLinus Torvalds journal->j_trans_start_time = 0; 28421da177e4SLinus Torvalds journal->j_last = NULL; 28431da177e4SLinus Torvalds journal->j_first = NULL; 2844a228bf8fSJeff Mahoney init_waitqueue_head(&journal->j_join_wait); 2845f68215c4SJeff Mahoney mutex_init(&journal->j_mutex); 2846afe70259SJeff Mahoney mutex_init(&journal->j_flush_mutex); 28471da177e4SLinus Torvalds 28481da177e4SLinus Torvalds journal->j_trans_id = 10; 28491da177e4SLinus Torvalds journal->j_mount_id = 10; 28501da177e4SLinus Torvalds journal->j_state = 0; 2851a228bf8fSJeff Mahoney atomic_set(&journal->j_jlock, 0); 28521da177e4SLinus Torvalds journal->j_cnode_free_list = allocate_cnodes(num_cnodes); 28531da177e4SLinus Torvalds journal->j_cnode_free_orig = journal->j_cnode_free_list; 28541da177e4SLinus Torvalds journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0; 28551da177e4SLinus Torvalds journal->j_cnode_used = 0; 28561da177e4SLinus Torvalds journal->j_must_wait = 0; 28571da177e4SLinus Torvalds 2858576f6d79SJeff Mahoney if (journal->j_cnode_free == 0) { 2859a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-2004", "Journal cnode memory " 2860576f6d79SJeff Mahoney "allocation failed (%ld bytes). Journal is " 2861576f6d79SJeff Mahoney "too large for available memory. Usually " 2862576f6d79SJeff Mahoney "this is due to a journal that is too large.", 2863576f6d79SJeff Mahoney sizeof (struct reiserfs_journal_cnode) * num_cnodes); 2864576f6d79SJeff Mahoney goto free_and_return; 2865576f6d79SJeff Mahoney } 2866576f6d79SJeff Mahoney 2867a9dd3643SJeff Mahoney init_journal_hash(sb); 28681da177e4SLinus Torvalds jl = journal->j_current_jl; 286937c69b98SFrederic Weisbecker 287037c69b98SFrederic Weisbecker /* 287137c69b98SFrederic Weisbecker * get_list_bitmap() may call flush_commit_list() which 287237c69b98SFrederic Weisbecker * requires the lock. Calling flush_commit_list() shouldn't happen 287337c69b98SFrederic Weisbecker * this early but I like to be paranoid. 287437c69b98SFrederic Weisbecker */ 287537c69b98SFrederic Weisbecker reiserfs_write_lock(sb); 2876a9dd3643SJeff Mahoney jl->j_list_bitmap = get_list_bitmap(sb, jl); 287737c69b98SFrederic Weisbecker reiserfs_write_unlock(sb); 28781da177e4SLinus Torvalds if (!jl->j_list_bitmap) { 2879a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-2005", 288045b03d5eSJeff Mahoney "get_list_bitmap failed for journal list 0"); 28811da177e4SLinus Torvalds goto free_and_return; 28821da177e4SLinus Torvalds } 288337c69b98SFrederic Weisbecker 288437c69b98SFrederic Weisbecker ret = journal_read(sb); 288537c69b98SFrederic Weisbecker if (ret < 0) { 2886a9dd3643SJeff Mahoney reiserfs_warning(sb, "reiserfs-2006", 288745b03d5eSJeff Mahoney "Replay Failure, unable to mount"); 28881da177e4SLinus Torvalds goto free_and_return; 28891da177e4SLinus Torvalds } 28901da177e4SLinus Torvalds 2891c4028958SDavid Howells INIT_DELAYED_WORK(&journal->j_work, flush_async_commits); 2892a9dd3643SJeff Mahoney journal->j_work_sb = sb; 28931da177e4SLinus Torvalds return 0; 28941da177e4SLinus Torvalds free_and_return: 2895a9dd3643SJeff Mahoney free_journal_ram(sb); 28961da177e4SLinus Torvalds return 1; 28971da177e4SLinus Torvalds } 28981da177e4SLinus Torvalds 28991da177e4SLinus Torvalds /* 2900098297b2SJeff Mahoney * test for a polite end of the current transaction. Used by file_write, 2901098297b2SJeff Mahoney * and should be used by delete to make sure they don't write more than 2902098297b2SJeff Mahoney * can fit inside a single transaction 29031da177e4SLinus Torvalds */ 2904bd4c625cSLinus Torvalds int journal_transaction_should_end(struct reiserfs_transaction_handle *th, 2905bd4c625cSLinus Torvalds int new_alloc) 2906bd4c625cSLinus Torvalds { 29071da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(th->t_super); 290834d08260SArnd Bergmann time64_t now = ktime_get_seconds(); 29091da177e4SLinus Torvalds /* cannot restart while nested */ 29101da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 29111da177e4SLinus Torvalds if (th->t_refcount > 1) 29121da177e4SLinus Torvalds return 0; 29131da177e4SLinus Torvalds if (journal->j_must_wait > 0 || 29141da177e4SLinus Torvalds (journal->j_len_alloc + new_alloc) >= journal->j_max_batch || 2915a228bf8fSJeff Mahoney atomic_read(&journal->j_jlock) || 29161da177e4SLinus Torvalds (now - journal->j_trans_start_time) > journal->j_max_trans_age || 29171da177e4SLinus Torvalds journal->j_cnode_free < (journal->j_trans_max * 3)) { 29181da177e4SLinus Torvalds return 1; 29191da177e4SLinus Torvalds } 2920b18c1c6eSDavidlohr Bueso 29216ae1ea44SChris Mason journal->j_len_alloc += new_alloc; 29226ae1ea44SChris Mason th->t_blocks_allocated += new_alloc ; 29231da177e4SLinus Torvalds return 0; 29241da177e4SLinus Torvalds } 29251da177e4SLinus Torvalds 2926098297b2SJeff Mahoney /* this must be called inside a transaction */ 2927bd4c625cSLinus Torvalds void reiserfs_block_writes(struct reiserfs_transaction_handle *th) 2928bd4c625cSLinus Torvalds { 29291da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(th->t_super); 29301da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 29311da177e4SLinus Torvalds journal->j_must_wait = 1; 29321da177e4SLinus Torvalds set_bit(J_WRITERS_BLOCKED, &journal->j_state); 29331da177e4SLinus Torvalds return; 29341da177e4SLinus Torvalds } 29351da177e4SLinus Torvalds 2936098297b2SJeff Mahoney /* this must be called without a transaction started */ 2937bd4c625cSLinus Torvalds void reiserfs_allow_writes(struct super_block *s) 2938bd4c625cSLinus Torvalds { 29391da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 29401da177e4SLinus Torvalds clear_bit(J_WRITERS_BLOCKED, &journal->j_state); 29411da177e4SLinus Torvalds wake_up(&journal->j_join_wait); 29421da177e4SLinus Torvalds } 29431da177e4SLinus Torvalds 2944098297b2SJeff Mahoney /* this must be called without a transaction started */ 2945bd4c625cSLinus Torvalds void reiserfs_wait_on_write_block(struct super_block *s) 2946bd4c625cSLinus Torvalds { 29471da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 29481da177e4SLinus Torvalds wait_event(journal->j_join_wait, 29491da177e4SLinus Torvalds !test_bit(J_WRITERS_BLOCKED, &journal->j_state)); 29501da177e4SLinus Torvalds } 29511da177e4SLinus Torvalds 2952bd4c625cSLinus Torvalds static void queue_log_writer(struct super_block *s) 2953bd4c625cSLinus Torvalds { 2954ac6424b9SIngo Molnar wait_queue_entry_t wait; 29551da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 29561da177e4SLinus Torvalds set_bit(J_WRITERS_QUEUED, &journal->j_state); 29571da177e4SLinus Torvalds 29581da177e4SLinus Torvalds /* 29591da177e4SLinus Torvalds * we don't want to use wait_event here because 29601da177e4SLinus Torvalds * we only want to wait once. 29611da177e4SLinus Torvalds */ 29621da177e4SLinus Torvalds init_waitqueue_entry(&wait, current); 29631da177e4SLinus Torvalds add_wait_queue(&journal->j_join_wait, &wait); 29641da177e4SLinus Torvalds set_current_state(TASK_UNINTERRUPTIBLE); 29658ebc4232SFrederic Weisbecker if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) { 2966278f6679SJeff Mahoney int depth = reiserfs_write_unlock_nested(s); 29671da177e4SLinus Torvalds schedule(); 2968278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 29698ebc4232SFrederic Weisbecker } 29705ab2f7e0SMilind Arun Choudhary __set_current_state(TASK_RUNNING); 29711da177e4SLinus Torvalds remove_wait_queue(&journal->j_join_wait, &wait); 29721da177e4SLinus Torvalds } 29731da177e4SLinus Torvalds 2974bd4c625cSLinus Torvalds static void wake_queued_writers(struct super_block *s) 2975bd4c625cSLinus Torvalds { 29761da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 29771da177e4SLinus Torvalds if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state)) 29781da177e4SLinus Torvalds wake_up(&journal->j_join_wait); 29791da177e4SLinus Torvalds } 29801da177e4SLinus Torvalds 2981600ed416SJeff Mahoney static void let_transaction_grow(struct super_block *sb, unsigned int trans_id) 29821da177e4SLinus Torvalds { 29831da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(sb); 29841da177e4SLinus Torvalds unsigned long bcount = journal->j_bcount; 29851da177e4SLinus Torvalds while (1) { 2986278f6679SJeff Mahoney int depth; 2987278f6679SJeff Mahoney 2988278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(sb); 2989041e0e3bSNishanth Aravamudan schedule_timeout_uninterruptible(1); 2990278f6679SJeff Mahoney reiserfs_write_lock_nested(sb, depth); 2991278f6679SJeff Mahoney 29921da177e4SLinus Torvalds journal->j_current_jl->j_state |= LIST_COMMIT_PENDING; 29931da177e4SLinus Torvalds while ((atomic_read(&journal->j_wcount) > 0 || 29941da177e4SLinus Torvalds atomic_read(&journal->j_jlock)) && 29951da177e4SLinus Torvalds journal->j_trans_id == trans_id) { 29961da177e4SLinus Torvalds queue_log_writer(sb); 29971da177e4SLinus Torvalds } 29981da177e4SLinus Torvalds if (journal->j_trans_id != trans_id) 29991da177e4SLinus Torvalds break; 30001da177e4SLinus Torvalds if (bcount == journal->j_bcount) 30011da177e4SLinus Torvalds break; 30021da177e4SLinus Torvalds bcount = journal->j_bcount; 30031da177e4SLinus Torvalds } 30041da177e4SLinus Torvalds } 30051da177e4SLinus Torvalds 3006098297b2SJeff Mahoney /* 3007098297b2SJeff Mahoney * join == true if you must join an existing transaction. 3008098297b2SJeff Mahoney * join == false if you can deal with waiting for others to finish 3009098297b2SJeff Mahoney * 3010098297b2SJeff Mahoney * this will block until the transaction is joinable. send the number of 3011098297b2SJeff Mahoney * blocks you expect to use in nblocks. 30121da177e4SLinus Torvalds */ 3013bd4c625cSLinus Torvalds static int do_journal_begin_r(struct reiserfs_transaction_handle *th, 3014a9dd3643SJeff Mahoney struct super_block *sb, unsigned long nblocks, 3015bd4c625cSLinus Torvalds int join) 3016bd4c625cSLinus Torvalds { 301734d08260SArnd Bergmann time64_t now = ktime_get_seconds(); 3018600ed416SJeff Mahoney unsigned int old_trans_id; 3019a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 30201da177e4SLinus Torvalds struct reiserfs_transaction_handle myth; 30211da177e4SLinus Torvalds int sched_count = 0; 30221da177e4SLinus Torvalds int retval; 3023278f6679SJeff Mahoney int depth; 30241da177e4SLinus Torvalds 3025a9dd3643SJeff Mahoney reiserfs_check_lock_depth(sb, "journal_begin"); 302614a61442SEric Sesterhenn BUG_ON(nblocks > journal->j_trans_max); 30271da177e4SLinus Torvalds 3028a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.journal_being); 30291da177e4SLinus Torvalds /* set here for journal_join */ 30301da177e4SLinus Torvalds th->t_refcount = 1; 3031a9dd3643SJeff Mahoney th->t_super = sb; 30321da177e4SLinus Torvalds 30331da177e4SLinus Torvalds relock: 3034a9dd3643SJeff Mahoney lock_journal(sb); 30351da177e4SLinus Torvalds if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) { 3036a9dd3643SJeff Mahoney unlock_journal(sb); 30371da177e4SLinus Torvalds retval = journal->j_errno; 30381da177e4SLinus Torvalds goto out_fail; 30391da177e4SLinus Torvalds } 30401da177e4SLinus Torvalds journal->j_bcount++; 30411da177e4SLinus Torvalds 30421da177e4SLinus Torvalds if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) { 3043a9dd3643SJeff Mahoney unlock_journal(sb); 3044278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(sb); 3045a9dd3643SJeff Mahoney reiserfs_wait_on_write_block(sb); 3046278f6679SJeff Mahoney reiserfs_write_lock_nested(sb, depth); 3047a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.journal_relock_writers); 30481da177e4SLinus Torvalds goto relock; 30491da177e4SLinus Torvalds } 305034d08260SArnd Bergmann now = ktime_get_seconds(); 30511da177e4SLinus Torvalds 3052098297b2SJeff Mahoney /* 3053098297b2SJeff Mahoney * if there is no room in the journal OR 3054098297b2SJeff Mahoney * if this transaction is too old, and we weren't called joinable, 3055098297b2SJeff Mahoney * wait for it to finish before beginning we don't sleep if there 3056098297b2SJeff Mahoney * aren't other writers 30571da177e4SLinus Torvalds */ 30581da177e4SLinus Torvalds 30591da177e4SLinus Torvalds if ((!join && journal->j_must_wait > 0) || 3060bd4c625cSLinus Torvalds (!join 3061bd4c625cSLinus Torvalds && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch) 3062bd4c625cSLinus Torvalds || (!join && atomic_read(&journal->j_wcount) > 0 3063bd4c625cSLinus Torvalds && journal->j_trans_start_time > 0 3064bd4c625cSLinus Torvalds && (now - journal->j_trans_start_time) > 3065bd4c625cSLinus Torvalds journal->j_max_trans_age) || (!join 3066bd4c625cSLinus Torvalds && atomic_read(&journal->j_jlock)) 3067bd4c625cSLinus Torvalds || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) { 30681da177e4SLinus Torvalds 30691da177e4SLinus Torvalds old_trans_id = journal->j_trans_id; 3070098297b2SJeff Mahoney /* allow others to finish this transaction */ 3071098297b2SJeff Mahoney unlock_journal(sb); 30721da177e4SLinus Torvalds 30731da177e4SLinus Torvalds if (!join && (journal->j_len_alloc + nblocks + 2) >= 30741da177e4SLinus Torvalds journal->j_max_batch && 3075bd4c625cSLinus Torvalds ((journal->j_len + nblocks + 2) * 100) < 3076bd4c625cSLinus Torvalds (journal->j_len_alloc * 75)) { 30771da177e4SLinus Torvalds if (atomic_read(&journal->j_wcount) > 10) { 30781da177e4SLinus Torvalds sched_count++; 3079a9dd3643SJeff Mahoney queue_log_writer(sb); 30801da177e4SLinus Torvalds goto relock; 30811da177e4SLinus Torvalds } 30821da177e4SLinus Torvalds } 3083098297b2SJeff Mahoney /* 3084098297b2SJeff Mahoney * don't mess with joining the transaction if all we 3085098297b2SJeff Mahoney * have to do is wait for someone else to do a commit 30861da177e4SLinus Torvalds */ 30871da177e4SLinus Torvalds if (atomic_read(&journal->j_jlock)) { 30881da177e4SLinus Torvalds while (journal->j_trans_id == old_trans_id && 30891da177e4SLinus Torvalds atomic_read(&journal->j_jlock)) { 3090a9dd3643SJeff Mahoney queue_log_writer(sb); 30911da177e4SLinus Torvalds } 30921da177e4SLinus Torvalds goto relock; 30931da177e4SLinus Torvalds } 3094b491dd17SJeff Mahoney retval = journal_join(&myth, sb); 30951da177e4SLinus Torvalds if (retval) 30961da177e4SLinus Torvalds goto out_fail; 30971da177e4SLinus Torvalds 30981da177e4SLinus Torvalds /* someone might have ended the transaction while we joined */ 30991da177e4SLinus Torvalds if (old_trans_id != journal->j_trans_id) { 310058d85426SJeff Mahoney retval = do_journal_end(&myth, 0); 31011da177e4SLinus Torvalds } else { 310258d85426SJeff Mahoney retval = do_journal_end(&myth, COMMIT_NOW); 31031da177e4SLinus Torvalds } 31041da177e4SLinus Torvalds 31051da177e4SLinus Torvalds if (retval) 31061da177e4SLinus Torvalds goto out_fail; 31071da177e4SLinus Torvalds 3108a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.journal_relock_wcount); 31091da177e4SLinus Torvalds goto relock; 31101da177e4SLinus Torvalds } 31111da177e4SLinus Torvalds /* we are the first writer, set trans_id */ 31121da177e4SLinus Torvalds if (journal->j_trans_start_time == 0) { 311334d08260SArnd Bergmann journal->j_trans_start_time = ktime_get_seconds(); 31141da177e4SLinus Torvalds } 3115a228bf8fSJeff Mahoney atomic_inc(&journal->j_wcount); 31161da177e4SLinus Torvalds journal->j_len_alloc += nblocks; 31171da177e4SLinus Torvalds th->t_blocks_logged = 0; 31181da177e4SLinus Torvalds th->t_blocks_allocated = nblocks; 31191da177e4SLinus Torvalds th->t_trans_id = journal->j_trans_id; 3120a9dd3643SJeff Mahoney unlock_journal(sb); 31211da177e4SLinus Torvalds INIT_LIST_HEAD(&th->t_list); 31221da177e4SLinus Torvalds return 0; 31231da177e4SLinus Torvalds 31241da177e4SLinus Torvalds out_fail: 31251da177e4SLinus Torvalds memset(th, 0, sizeof(*th)); 3126098297b2SJeff Mahoney /* 3127098297b2SJeff Mahoney * Re-set th->t_super, so we can properly keep track of how many 31281da177e4SLinus Torvalds * persistent transactions there are. We need to do this so if this 3129098297b2SJeff Mahoney * call is part of a failed restart_transaction, we can free it later 3130098297b2SJeff Mahoney */ 3131a9dd3643SJeff Mahoney th->t_super = sb; 31321da177e4SLinus Torvalds return retval; 31331da177e4SLinus Torvalds } 31341da177e4SLinus Torvalds 3135bd4c625cSLinus Torvalds struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct 3136bd4c625cSLinus Torvalds super_block 3137bd4c625cSLinus Torvalds *s, 3138bd4c625cSLinus Torvalds int nblocks) 3139bd4c625cSLinus Torvalds { 31401da177e4SLinus Torvalds int ret; 31411da177e4SLinus Torvalds struct reiserfs_transaction_handle *th; 31421da177e4SLinus Torvalds 3143098297b2SJeff Mahoney /* 3144098297b2SJeff Mahoney * if we're nesting into an existing transaction. It will be 3145098297b2SJeff Mahoney * persistent on its own 31461da177e4SLinus Torvalds */ 31471da177e4SLinus Torvalds if (reiserfs_transaction_running(s)) { 31481da177e4SLinus Torvalds th = current->journal_info; 31491da177e4SLinus Torvalds th->t_refcount++; 315014a61442SEric Sesterhenn BUG_ON(th->t_refcount < 2); 315114a61442SEric Sesterhenn 31521da177e4SLinus Torvalds return th; 31531da177e4SLinus Torvalds } 3154d739b42bSPekka Enberg th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS); 31551da177e4SLinus Torvalds if (!th) 31561da177e4SLinus Torvalds return NULL; 31571da177e4SLinus Torvalds ret = journal_begin(th, s, nblocks); 31581da177e4SLinus Torvalds if (ret) { 3159d739b42bSPekka Enberg kfree(th); 31601da177e4SLinus Torvalds return NULL; 31611da177e4SLinus Torvalds } 31621da177e4SLinus Torvalds 31631da177e4SLinus Torvalds SB_JOURNAL(s)->j_persistent_trans++; 31641da177e4SLinus Torvalds return th; 31651da177e4SLinus Torvalds } 31661da177e4SLinus Torvalds 3167bd4c625cSLinus Torvalds int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th) 3168bd4c625cSLinus Torvalds { 31691da177e4SLinus Torvalds struct super_block *s = th->t_super; 31701da177e4SLinus Torvalds int ret = 0; 31711da177e4SLinus Torvalds if (th->t_trans_id) 317258d85426SJeff Mahoney ret = journal_end(th); 31731da177e4SLinus Torvalds else 31741da177e4SLinus Torvalds ret = -EIO; 31751da177e4SLinus Torvalds if (th->t_refcount == 0) { 31761da177e4SLinus Torvalds SB_JOURNAL(s)->j_persistent_trans--; 3177d739b42bSPekka Enberg kfree(th); 31781da177e4SLinus Torvalds } 31791da177e4SLinus Torvalds return ret; 31801da177e4SLinus Torvalds } 31811da177e4SLinus Torvalds 3182bd4c625cSLinus Torvalds static int journal_join(struct reiserfs_transaction_handle *th, 3183b491dd17SJeff Mahoney struct super_block *sb) 3184bd4c625cSLinus Torvalds { 31851da177e4SLinus Torvalds struct reiserfs_transaction_handle *cur_th = current->journal_info; 31861da177e4SLinus Torvalds 3187098297b2SJeff Mahoney /* 3188098297b2SJeff Mahoney * this keeps do_journal_end from NULLing out the 3189098297b2SJeff Mahoney * current->journal_info pointer 31901da177e4SLinus Torvalds */ 31911da177e4SLinus Torvalds th->t_handle_save = cur_th; 319214a61442SEric Sesterhenn BUG_ON(cur_th && cur_th->t_refcount > 1); 3193b491dd17SJeff Mahoney return do_journal_begin_r(th, sb, 1, JBEGIN_JOIN); 31941da177e4SLinus Torvalds } 31951da177e4SLinus Torvalds 3196bd4c625cSLinus Torvalds int journal_join_abort(struct reiserfs_transaction_handle *th, 3197b491dd17SJeff Mahoney struct super_block *sb) 3198bd4c625cSLinus Torvalds { 31991da177e4SLinus Torvalds struct reiserfs_transaction_handle *cur_th = current->journal_info; 32001da177e4SLinus Torvalds 3201098297b2SJeff Mahoney /* 3202098297b2SJeff Mahoney * this keeps do_journal_end from NULLing out the 3203098297b2SJeff Mahoney * current->journal_info pointer 32041da177e4SLinus Torvalds */ 32051da177e4SLinus Torvalds th->t_handle_save = cur_th; 320614a61442SEric Sesterhenn BUG_ON(cur_th && cur_th->t_refcount > 1); 3207b491dd17SJeff Mahoney return do_journal_begin_r(th, sb, 1, JBEGIN_ABORT); 32081da177e4SLinus Torvalds } 32091da177e4SLinus Torvalds 3210bd4c625cSLinus Torvalds int journal_begin(struct reiserfs_transaction_handle *th, 3211a9dd3643SJeff Mahoney struct super_block *sb, unsigned long nblocks) 3212bd4c625cSLinus Torvalds { 32131da177e4SLinus Torvalds struct reiserfs_transaction_handle *cur_th = current->journal_info; 32141da177e4SLinus Torvalds int ret; 32151da177e4SLinus Torvalds 32161da177e4SLinus Torvalds th->t_handle_save = NULL; 32171da177e4SLinus Torvalds if (cur_th) { 32181da177e4SLinus Torvalds /* we are nesting into the current transaction */ 3219a9dd3643SJeff Mahoney if (cur_th->t_super == sb) { 32201da177e4SLinus Torvalds BUG_ON(!cur_th->t_refcount); 32211da177e4SLinus Torvalds cur_th->t_refcount++; 32221da177e4SLinus Torvalds memcpy(th, cur_th, sizeof(*th)); 32231da177e4SLinus Torvalds if (th->t_refcount <= 1) 3224a9dd3643SJeff Mahoney reiserfs_warning(sb, "reiserfs-2005", 322545b03d5eSJeff Mahoney "BAD: refcount <= 1, but " 322645b03d5eSJeff Mahoney "journal_info != 0"); 32271da177e4SLinus Torvalds return 0; 32281da177e4SLinus Torvalds } else { 3229098297b2SJeff Mahoney /* 3230098297b2SJeff Mahoney * we've ended up with a handle from a different 3231098297b2SJeff Mahoney * filesystem. save it and restore on journal_end. 3232098297b2SJeff Mahoney * This should never really happen... 32331da177e4SLinus Torvalds */ 3234a9dd3643SJeff Mahoney reiserfs_warning(sb, "clm-2100", 323545b03d5eSJeff Mahoney "nesting info a different FS"); 32361da177e4SLinus Torvalds th->t_handle_save = current->journal_info; 32371da177e4SLinus Torvalds current->journal_info = th; 32381da177e4SLinus Torvalds } 32391da177e4SLinus Torvalds } else { 32401da177e4SLinus Torvalds current->journal_info = th; 32411da177e4SLinus Torvalds } 3242a9dd3643SJeff Mahoney ret = do_journal_begin_r(th, sb, nblocks, JBEGIN_REG); 324314a61442SEric Sesterhenn BUG_ON(current->journal_info != th); 32441da177e4SLinus Torvalds 3245098297b2SJeff Mahoney /* 3246098297b2SJeff Mahoney * I guess this boils down to being the reciprocal of clm-2100 above. 3247098297b2SJeff Mahoney * If do_journal_begin_r fails, we need to put it back, since 3248098297b2SJeff Mahoney * journal_end won't be called to do it. */ 32491da177e4SLinus Torvalds if (ret) 32501da177e4SLinus Torvalds current->journal_info = th->t_handle_save; 32511da177e4SLinus Torvalds else 32521da177e4SLinus Torvalds BUG_ON(!th->t_refcount); 32531da177e4SLinus Torvalds 32541da177e4SLinus Torvalds return ret; 32551da177e4SLinus Torvalds } 32561da177e4SLinus Torvalds 32571da177e4SLinus Torvalds /* 3258098297b2SJeff Mahoney * puts bh into the current transaction. If it was already there, reorders 3259098297b2SJeff Mahoney * removes the old pointers from the hash, and puts new ones in (to make 3260098297b2SJeff Mahoney * sure replay happen in the right order). 3261098297b2SJeff Mahoney * 3262098297b2SJeff Mahoney * if it was dirty, cleans and files onto the clean list. I can't let it 3263098297b2SJeff Mahoney * be dirty again until the transaction is committed. 3264098297b2SJeff Mahoney * 3265098297b2SJeff Mahoney * if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len. 32661da177e4SLinus Torvalds */ 3267bd4c625cSLinus Torvalds int journal_mark_dirty(struct reiserfs_transaction_handle *th, 326809f1b80bSJeff Mahoney struct buffer_head *bh) 3269bd4c625cSLinus Torvalds { 327009f1b80bSJeff Mahoney struct super_block *sb = th->t_super; 3271a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 32721da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn = NULL; 32731da177e4SLinus Torvalds int count_already_incd = 0; 32741da177e4SLinus Torvalds int prepared = 0; 32751da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 32761da177e4SLinus Torvalds 3277a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.mark_dirty); 32781da177e4SLinus Torvalds if (th->t_trans_id != journal->j_trans_id) { 3279c3a9c210SJeff Mahoney reiserfs_panic(th->t_super, "journal-1577", 3280c3a9c210SJeff Mahoney "handle trans id %ld != current trans id %ld", 32811da177e4SLinus Torvalds th->t_trans_id, journal->j_trans_id); 32821da177e4SLinus Torvalds } 32831da177e4SLinus Torvalds 32841da177e4SLinus Torvalds prepared = test_clear_buffer_journal_prepared(bh); 32851da177e4SLinus Torvalds clear_buffer_journal_restore_dirty(bh); 32861da177e4SLinus Torvalds /* already in this transaction, we are done */ 32871da177e4SLinus Torvalds if (buffer_journaled(bh)) { 3288a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.mark_dirty_already); 32891da177e4SLinus Torvalds return 0; 32901da177e4SLinus Torvalds } 32911da177e4SLinus Torvalds 3292098297b2SJeff Mahoney /* 3293098297b2SJeff Mahoney * this must be turned into a panic instead of a warning. We can't 3294098297b2SJeff Mahoney * allow a dirty or journal_dirty or locked buffer to be logged, as 3295098297b2SJeff Mahoney * some changes could get to disk too early. NOT GOOD. 32961da177e4SLinus Torvalds */ 32971da177e4SLinus Torvalds if (!prepared || buffer_dirty(bh)) { 3298a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1777", 329945b03d5eSJeff Mahoney "buffer %llu bad state " 33001da177e4SLinus Torvalds "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT", 3301bd4c625cSLinus Torvalds (unsigned long long)bh->b_blocknr, 3302bd4c625cSLinus Torvalds prepared ? ' ' : '!', 33031da177e4SLinus Torvalds buffer_locked(bh) ? ' ' : '!', 33041da177e4SLinus Torvalds buffer_dirty(bh) ? ' ' : '!', 33051da177e4SLinus Torvalds buffer_journal_dirty(bh) ? ' ' : '!'); 33061da177e4SLinus Torvalds } 33071da177e4SLinus Torvalds 3308a228bf8fSJeff Mahoney if (atomic_read(&journal->j_wcount) <= 0) { 3309a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1409", 331045b03d5eSJeff Mahoney "returning because j_wcount was %d", 3311a228bf8fSJeff Mahoney atomic_read(&journal->j_wcount)); 33121da177e4SLinus Torvalds return 1; 33131da177e4SLinus Torvalds } 3314098297b2SJeff Mahoney /* 3315098297b2SJeff Mahoney * this error means I've screwed up, and we've overflowed 3316098297b2SJeff Mahoney * the transaction. Nothing can be done here, except make the 3317098297b2SJeff Mahoney * FS readonly or panic. 33181da177e4SLinus Torvalds */ 33191da177e4SLinus Torvalds if (journal->j_len >= journal->j_trans_max) { 3320c3a9c210SJeff Mahoney reiserfs_panic(th->t_super, "journal-1413", 3321c3a9c210SJeff Mahoney "j_len (%lu) is too big", 3322bd4c625cSLinus Torvalds journal->j_len); 33231da177e4SLinus Torvalds } 33241da177e4SLinus Torvalds 33251da177e4SLinus Torvalds if (buffer_journal_dirty(bh)) { 33261da177e4SLinus Torvalds count_already_incd = 1; 3327a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.mark_dirty_notjournal); 33281da177e4SLinus Torvalds clear_buffer_journal_dirty(bh); 33291da177e4SLinus Torvalds } 33301da177e4SLinus Torvalds 33311da177e4SLinus Torvalds if (journal->j_len > journal->j_len_alloc) { 33321da177e4SLinus Torvalds journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT; 33331da177e4SLinus Torvalds } 33341da177e4SLinus Torvalds 33351da177e4SLinus Torvalds set_buffer_journaled(bh); 33361da177e4SLinus Torvalds 33371da177e4SLinus Torvalds /* now put this guy on the end */ 33381da177e4SLinus Torvalds if (!cn) { 3339a9dd3643SJeff Mahoney cn = get_cnode(sb); 33401da177e4SLinus Torvalds if (!cn) { 3341a9dd3643SJeff Mahoney reiserfs_panic(sb, "journal-4", "get_cnode failed!"); 33421da177e4SLinus Torvalds } 33431da177e4SLinus Torvalds 33441da177e4SLinus Torvalds if (th->t_blocks_logged == th->t_blocks_allocated) { 33451da177e4SLinus Torvalds th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT; 33461da177e4SLinus Torvalds journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT; 33471da177e4SLinus Torvalds } 33481da177e4SLinus Torvalds th->t_blocks_logged++; 33491da177e4SLinus Torvalds journal->j_len++; 33501da177e4SLinus Torvalds 33511da177e4SLinus Torvalds cn->bh = bh; 33521da177e4SLinus Torvalds cn->blocknr = bh->b_blocknr; 3353a9dd3643SJeff Mahoney cn->sb = sb; 33541da177e4SLinus Torvalds cn->jlist = NULL; 33551da177e4SLinus Torvalds insert_journal_hash(journal->j_hash_table, cn); 33561da177e4SLinus Torvalds if (!count_already_incd) { 33571da177e4SLinus Torvalds get_bh(bh); 33581da177e4SLinus Torvalds } 33591da177e4SLinus Torvalds } 33601da177e4SLinus Torvalds cn->next = NULL; 33611da177e4SLinus Torvalds cn->prev = journal->j_last; 33621da177e4SLinus Torvalds cn->bh = bh; 33631da177e4SLinus Torvalds if (journal->j_last) { 33641da177e4SLinus Torvalds journal->j_last->next = cn; 33651da177e4SLinus Torvalds journal->j_last = cn; 33661da177e4SLinus Torvalds } else { 33671da177e4SLinus Torvalds journal->j_first = cn; 33681da177e4SLinus Torvalds journal->j_last = cn; 33691da177e4SLinus Torvalds } 3370033369d1SArtem Bityutskiy reiserfs_schedule_old_flush(sb); 33711da177e4SLinus Torvalds return 0; 33721da177e4SLinus Torvalds } 33731da177e4SLinus Torvalds 337458d85426SJeff Mahoney int journal_end(struct reiserfs_transaction_handle *th) 3375bd4c625cSLinus Torvalds { 337658d85426SJeff Mahoney struct super_block *sb = th->t_super; 33771da177e4SLinus Torvalds if (!current->journal_info && th->t_refcount > 1) 3378a9dd3643SJeff Mahoney reiserfs_warning(sb, "REISER-NESTING", 337945b03d5eSJeff Mahoney "th NULL, refcount %d", th->t_refcount); 33801da177e4SLinus Torvalds 33811da177e4SLinus Torvalds if (!th->t_trans_id) { 33821da177e4SLinus Torvalds WARN_ON(1); 33831da177e4SLinus Torvalds return -EIO; 33841da177e4SLinus Torvalds } 33851da177e4SLinus Torvalds 33861da177e4SLinus Torvalds th->t_refcount--; 33871da177e4SLinus Torvalds if (th->t_refcount > 0) { 3388bd4c625cSLinus Torvalds struct reiserfs_transaction_handle *cur_th = 3389bd4c625cSLinus Torvalds current->journal_info; 33901da177e4SLinus Torvalds 3391098297b2SJeff Mahoney /* 3392098297b2SJeff Mahoney * we aren't allowed to close a nested transaction on a 3393098297b2SJeff Mahoney * different filesystem from the one in the task struct 33941da177e4SLinus Torvalds */ 339514a61442SEric Sesterhenn BUG_ON(cur_th->t_super != th->t_super); 33961da177e4SLinus Torvalds 33971da177e4SLinus Torvalds if (th != cur_th) { 33981da177e4SLinus Torvalds memcpy(current->journal_info, th, sizeof(*th)); 33991da177e4SLinus Torvalds th->t_trans_id = 0; 34001da177e4SLinus Torvalds } 34011da177e4SLinus Torvalds return 0; 34021da177e4SLinus Torvalds } else { 340358d85426SJeff Mahoney return do_journal_end(th, 0); 34041da177e4SLinus Torvalds } 34051da177e4SLinus Torvalds } 34061da177e4SLinus Torvalds 3407098297b2SJeff Mahoney /* 3408098297b2SJeff Mahoney * removes from the current transaction, relsing and descrementing any counters. 3409098297b2SJeff Mahoney * also files the removed buffer directly onto the clean list 3410098297b2SJeff Mahoney * 3411098297b2SJeff Mahoney * called by journal_mark_freed when a block has been deleted 3412098297b2SJeff Mahoney * 3413098297b2SJeff Mahoney * returns 1 if it cleaned and relsed the buffer. 0 otherwise 34141da177e4SLinus Torvalds */ 3415a9dd3643SJeff Mahoney static int remove_from_transaction(struct super_block *sb, 3416bd4c625cSLinus Torvalds b_blocknr_t blocknr, int already_cleaned) 3417bd4c625cSLinus Torvalds { 34181da177e4SLinus Torvalds struct buffer_head *bh; 34191da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 3420a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 34211da177e4SLinus Torvalds int ret = 0; 34221da177e4SLinus Torvalds 3423a9dd3643SJeff Mahoney cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr); 34241da177e4SLinus Torvalds if (!cn || !cn->bh) { 34251da177e4SLinus Torvalds return ret; 34261da177e4SLinus Torvalds } 34271da177e4SLinus Torvalds bh = cn->bh; 34281da177e4SLinus Torvalds if (cn->prev) { 34291da177e4SLinus Torvalds cn->prev->next = cn->next; 34301da177e4SLinus Torvalds } 34311da177e4SLinus Torvalds if (cn->next) { 34321da177e4SLinus Torvalds cn->next->prev = cn->prev; 34331da177e4SLinus Torvalds } 34341da177e4SLinus Torvalds if (cn == journal->j_first) { 34351da177e4SLinus Torvalds journal->j_first = cn->next; 34361da177e4SLinus Torvalds } 34371da177e4SLinus Torvalds if (cn == journal->j_last) { 34381da177e4SLinus Torvalds journal->j_last = cn->prev; 34391da177e4SLinus Torvalds } 3440a9dd3643SJeff Mahoney remove_journal_hash(sb, journal->j_hash_table, NULL, 3441bd4c625cSLinus Torvalds bh->b_blocknr, 0); 34421da177e4SLinus Torvalds clear_buffer_journaled(bh); /* don't log this one */ 34431da177e4SLinus Torvalds 34441da177e4SLinus Torvalds if (!already_cleaned) { 34451da177e4SLinus Torvalds clear_buffer_journal_dirty(bh); 34461da177e4SLinus Torvalds clear_buffer_dirty(bh); 34471da177e4SLinus Torvalds clear_buffer_journal_test(bh); 34481da177e4SLinus Torvalds put_bh(bh); 3449a228bf8fSJeff Mahoney if (atomic_read(&bh->b_count) < 0) { 3450a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1752", 345145b03d5eSJeff Mahoney "b_count < 0"); 34521da177e4SLinus Torvalds } 34531da177e4SLinus Torvalds ret = 1; 34541da177e4SLinus Torvalds } 34551da177e4SLinus Torvalds journal->j_len--; 34561da177e4SLinus Torvalds journal->j_len_alloc--; 3457a9dd3643SJeff Mahoney free_cnode(sb, cn); 34581da177e4SLinus Torvalds return ret; 34591da177e4SLinus Torvalds } 34601da177e4SLinus Torvalds 34611da177e4SLinus Torvalds /* 3462098297b2SJeff Mahoney * for any cnode in a journal list, it can only be dirtied of all the 3463098297b2SJeff Mahoney * transactions that include it are committed to disk. 3464098297b2SJeff Mahoney * this checks through each transaction, and returns 1 if you are allowed 3465098297b2SJeff Mahoney * to dirty, and 0 if you aren't 3466098297b2SJeff Mahoney * 3467098297b2SJeff Mahoney * it is called by dirty_journal_list, which is called after 3468098297b2SJeff Mahoney * flush_commit_list has gotten all the log blocks for a given 3469098297b2SJeff Mahoney * transaction on disk 3470098297b2SJeff Mahoney * 34711da177e4SLinus Torvalds */ 3472bd4c625cSLinus Torvalds static int can_dirty(struct reiserfs_journal_cnode *cn) 3473bd4c625cSLinus Torvalds { 34741da177e4SLinus Torvalds struct super_block *sb = cn->sb; 34751da177e4SLinus Torvalds b_blocknr_t blocknr = cn->blocknr; 34761da177e4SLinus Torvalds struct reiserfs_journal_cnode *cur = cn->hprev; 34771da177e4SLinus Torvalds int can_dirty = 1; 34781da177e4SLinus Torvalds 3479098297b2SJeff Mahoney /* 3480098297b2SJeff Mahoney * first test hprev. These are all newer than cn, so any node here 3481098297b2SJeff Mahoney * with the same block number and dev means this node can't be sent 3482098297b2SJeff Mahoney * to disk right now. 34831da177e4SLinus Torvalds */ 34841da177e4SLinus Torvalds while (cur && can_dirty) { 34851da177e4SLinus Torvalds if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb && 34861da177e4SLinus Torvalds cur->blocknr == blocknr) { 34871da177e4SLinus Torvalds can_dirty = 0; 34881da177e4SLinus Torvalds } 34891da177e4SLinus Torvalds cur = cur->hprev; 34901da177e4SLinus Torvalds } 3491098297b2SJeff Mahoney /* 3492098297b2SJeff Mahoney * then test hnext. These are all older than cn. As long as they 3493098297b2SJeff Mahoney * are committed to the log, it is safe to write cn to disk 34941da177e4SLinus Torvalds */ 34951da177e4SLinus Torvalds cur = cn->hnext; 34961da177e4SLinus Torvalds while (cur && can_dirty) { 34971da177e4SLinus Torvalds if (cur->jlist && cur->jlist->j_len > 0 && 3498a228bf8fSJeff Mahoney atomic_read(&cur->jlist->j_commit_left) > 0 && cur->bh && 34991da177e4SLinus Torvalds cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) { 35001da177e4SLinus Torvalds can_dirty = 0; 35011da177e4SLinus Torvalds } 35021da177e4SLinus Torvalds cur = cur->hnext; 35031da177e4SLinus Torvalds } 35041da177e4SLinus Torvalds return can_dirty; 35051da177e4SLinus Torvalds } 35061da177e4SLinus Torvalds 3507098297b2SJeff Mahoney /* 3508098297b2SJeff Mahoney * syncs the commit blocks, but does not force the real buffers to disk 3509098297b2SJeff Mahoney * will wait until the current transaction is done/committed before returning 35101da177e4SLinus Torvalds */ 351158d85426SJeff Mahoney int journal_end_sync(struct reiserfs_transaction_handle *th) 3512bd4c625cSLinus Torvalds { 351358d85426SJeff Mahoney struct super_block *sb = th->t_super; 3514a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 35151da177e4SLinus Torvalds 35161da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 35171da177e4SLinus Torvalds /* you can sync while nested, very, very bad */ 351814a61442SEric Sesterhenn BUG_ON(th->t_refcount > 1); 35191da177e4SLinus Torvalds if (journal->j_len == 0) { 3520a9dd3643SJeff Mahoney reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb), 3521bd4c625cSLinus Torvalds 1); 352209f1b80bSJeff Mahoney journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb)); 35231da177e4SLinus Torvalds } 352458d85426SJeff Mahoney return do_journal_end(th, COMMIT_NOW | WAIT); 35251da177e4SLinus Torvalds } 35261da177e4SLinus Torvalds 3527098297b2SJeff Mahoney /* writeback the pending async commits to disk */ 3528c4028958SDavid Howells static void flush_async_commits(struct work_struct *work) 3529bd4c625cSLinus Torvalds { 3530c4028958SDavid Howells struct reiserfs_journal *journal = 3531c4028958SDavid Howells container_of(work, struct reiserfs_journal, j_work.work); 3532a9dd3643SJeff Mahoney struct super_block *sb = journal->j_work_sb; 35331da177e4SLinus Torvalds struct reiserfs_journal_list *jl; 35341da177e4SLinus Torvalds struct list_head *entry; 35351da177e4SLinus Torvalds 35368ebc4232SFrederic Weisbecker reiserfs_write_lock(sb); 35371da177e4SLinus Torvalds if (!list_empty(&journal->j_journal_list)) { 35381da177e4SLinus Torvalds /* last entry is the youngest, commit it and you get everything */ 35391da177e4SLinus Torvalds entry = journal->j_journal_list.prev; 35401da177e4SLinus Torvalds jl = JOURNAL_LIST_ENTRY(entry); 3541a9dd3643SJeff Mahoney flush_commit_list(sb, jl, 1); 35421da177e4SLinus Torvalds } 35438ebc4232SFrederic Weisbecker reiserfs_write_unlock(sb); 35441da177e4SLinus Torvalds } 35451da177e4SLinus Torvalds 35461da177e4SLinus Torvalds /* 3547098297b2SJeff Mahoney * flushes any old transactions to disk 3548098297b2SJeff Mahoney * ends the current transaction if it is too old 35491da177e4SLinus Torvalds */ 355025729b0eSArtem Bityutskiy void reiserfs_flush_old_commits(struct super_block *sb) 3551bd4c625cSLinus Torvalds { 355234d08260SArnd Bergmann time64_t now; 35531da177e4SLinus Torvalds struct reiserfs_transaction_handle th; 3554a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 35551da177e4SLinus Torvalds 355634d08260SArnd Bergmann now = ktime_get_seconds(); 3557098297b2SJeff Mahoney /* 3558098297b2SJeff Mahoney * safety check so we don't flush while we are replaying the log during 35591da177e4SLinus Torvalds * mount 35601da177e4SLinus Torvalds */ 356125729b0eSArtem Bityutskiy if (list_empty(&journal->j_journal_list)) 356225729b0eSArtem Bityutskiy return; 35631da177e4SLinus Torvalds 3564098297b2SJeff Mahoney /* 3565098297b2SJeff Mahoney * check the current transaction. If there are no writers, and it is 35661da177e4SLinus Torvalds * too old, finish it, and force the commit blocks to disk 35671da177e4SLinus Torvalds */ 35681da177e4SLinus Torvalds if (atomic_read(&journal->j_wcount) <= 0 && 35691da177e4SLinus Torvalds journal->j_trans_start_time > 0 && 35701da177e4SLinus Torvalds journal->j_len > 0 && 3571bd4c625cSLinus Torvalds (now - journal->j_trans_start_time) > journal->j_max_trans_age) { 3572b491dd17SJeff Mahoney if (!journal_join(&th, sb)) { 3573a9dd3643SJeff Mahoney reiserfs_prepare_for_journal(sb, 3574a9dd3643SJeff Mahoney SB_BUFFER_WITH_SB(sb), 3575bd4c625cSLinus Torvalds 1); 357609f1b80bSJeff Mahoney journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb)); 35771da177e4SLinus Torvalds 3578098297b2SJeff Mahoney /* 3579098297b2SJeff Mahoney * we're only being called from kreiserfsd, it makes 3580098297b2SJeff Mahoney * no sense to do an async commit so that kreiserfsd 3581098297b2SJeff Mahoney * can do it later 35821da177e4SLinus Torvalds */ 358358d85426SJeff Mahoney do_journal_end(&th, COMMIT_NOW | WAIT); 35841da177e4SLinus Torvalds } 35851da177e4SLinus Torvalds } 35861da177e4SLinus Torvalds } 35871da177e4SLinus Torvalds 35881da177e4SLinus Torvalds /* 3589098297b2SJeff Mahoney * returns 0 if do_journal_end should return right away, returns 1 if 3590098297b2SJeff Mahoney * do_journal_end should finish the commit 3591098297b2SJeff Mahoney * 3592098297b2SJeff Mahoney * if the current transaction is too old, but still has writers, this will 3593098297b2SJeff Mahoney * wait on j_join_wait until all the writers are done. By the time it 3594098297b2SJeff Mahoney * wakes up, the transaction it was called has already ended, so it just 3595098297b2SJeff Mahoney * flushes the commit list and returns 0. 3596098297b2SJeff Mahoney * 3597098297b2SJeff Mahoney * Won't batch when flush or commit_now is set. Also won't batch when 3598098297b2SJeff Mahoney * others are waiting on j_join_wait. 3599098297b2SJeff Mahoney * 3600098297b2SJeff Mahoney * Note, we can't allow the journal_end to proceed while there are still 3601098297b2SJeff Mahoney * writers in the log. 36021da177e4SLinus Torvalds */ 360358d85426SJeff Mahoney static int check_journal_end(struct reiserfs_transaction_handle *th, int flags) 3604bd4c625cSLinus Torvalds { 36051da177e4SLinus Torvalds 360634d08260SArnd Bergmann time64_t now; 36071da177e4SLinus Torvalds int flush = flags & FLUSH_ALL; 36081da177e4SLinus Torvalds int commit_now = flags & COMMIT_NOW; 36091da177e4SLinus Torvalds int wait_on_commit = flags & WAIT; 36101da177e4SLinus Torvalds struct reiserfs_journal_list *jl; 361158d85426SJeff Mahoney struct super_block *sb = th->t_super; 3612a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 36131da177e4SLinus Torvalds 36141da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 36151da177e4SLinus Torvalds 36161da177e4SLinus Torvalds if (th->t_trans_id != journal->j_trans_id) { 3617c3a9c210SJeff Mahoney reiserfs_panic(th->t_super, "journal-1577", 3618c3a9c210SJeff Mahoney "handle trans id %ld != current trans id %ld", 36191da177e4SLinus Torvalds th->t_trans_id, journal->j_trans_id); 36201da177e4SLinus Torvalds } 36211da177e4SLinus Torvalds 36221da177e4SLinus Torvalds journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged); 3623098297b2SJeff Mahoney /* <= 0 is allowed. unmounting might not call begin */ 3624a228bf8fSJeff Mahoney if (atomic_read(&journal->j_wcount) > 0) 3625a228bf8fSJeff Mahoney atomic_dec(&journal->j_wcount); 36261da177e4SLinus Torvalds 3627098297b2SJeff Mahoney /* 3628098297b2SJeff Mahoney * BUG, deal with case where j_len is 0, but people previously 3629098297b2SJeff Mahoney * freed blocks need to be released will be dealt with by next 3630098297b2SJeff Mahoney * transaction that actually writes something, but should be taken 3631098297b2SJeff Mahoney * care of in this trans 36321da177e4SLinus Torvalds */ 363314a61442SEric Sesterhenn BUG_ON(journal->j_len == 0); 363414a61442SEric Sesterhenn 3635098297b2SJeff Mahoney /* 3636098297b2SJeff Mahoney * if wcount > 0, and we are called to with flush or commit_now, 3637098297b2SJeff Mahoney * we wait on j_join_wait. We will wake up when the last writer has 3638098297b2SJeff Mahoney * finished the transaction, and started it on its way to the disk. 3639098297b2SJeff Mahoney * Then, we flush the commit or journal list, and just return 0 3640098297b2SJeff Mahoney * because the rest of journal end was already done for this 3641098297b2SJeff Mahoney * transaction. 36421da177e4SLinus Torvalds */ 3643a228bf8fSJeff Mahoney if (atomic_read(&journal->j_wcount) > 0) { 36441da177e4SLinus Torvalds if (flush || commit_now) { 36451da177e4SLinus Torvalds unsigned trans_id; 36461da177e4SLinus Torvalds 36471da177e4SLinus Torvalds jl = journal->j_current_jl; 36481da177e4SLinus Torvalds trans_id = jl->j_trans_id; 36491da177e4SLinus Torvalds if (wait_on_commit) 36501da177e4SLinus Torvalds jl->j_state |= LIST_COMMIT_PENDING; 3651a228bf8fSJeff Mahoney atomic_set(&journal->j_jlock, 1); 36521da177e4SLinus Torvalds if (flush) { 36531da177e4SLinus Torvalds journal->j_next_full_flush = 1; 36541da177e4SLinus Torvalds } 3655a9dd3643SJeff Mahoney unlock_journal(sb); 36561da177e4SLinus Torvalds 3657098297b2SJeff Mahoney /* 3658098297b2SJeff Mahoney * sleep while the current transaction is 3659098297b2SJeff Mahoney * still j_jlocked 3660098297b2SJeff Mahoney */ 36611da177e4SLinus Torvalds while (journal->j_trans_id == trans_id) { 36621da177e4SLinus Torvalds if (atomic_read(&journal->j_jlock)) { 3663a9dd3643SJeff Mahoney queue_log_writer(sb); 36641da177e4SLinus Torvalds } else { 3665a9dd3643SJeff Mahoney lock_journal(sb); 36661da177e4SLinus Torvalds if (journal->j_trans_id == trans_id) { 3667a228bf8fSJeff Mahoney atomic_set(&journal->j_jlock, 3668bd4c625cSLinus Torvalds 1); 36691da177e4SLinus Torvalds } 3670a9dd3643SJeff Mahoney unlock_journal(sb); 36711da177e4SLinus Torvalds } 36721da177e4SLinus Torvalds } 367314a61442SEric Sesterhenn BUG_ON(journal->j_trans_id == trans_id); 367414a61442SEric Sesterhenn 3675bd4c625cSLinus Torvalds if (commit_now 3676a9dd3643SJeff Mahoney && journal_list_still_alive(sb, trans_id) 3677bd4c625cSLinus Torvalds && wait_on_commit) { 3678a9dd3643SJeff Mahoney flush_commit_list(sb, jl, 1); 36791da177e4SLinus Torvalds } 36801da177e4SLinus Torvalds return 0; 36811da177e4SLinus Torvalds } 3682a9dd3643SJeff Mahoney unlock_journal(sb); 36831da177e4SLinus Torvalds return 0; 36841da177e4SLinus Torvalds } 36851da177e4SLinus Torvalds 36861da177e4SLinus Torvalds /* deal with old transactions where we are the last writers */ 368734d08260SArnd Bergmann now = ktime_get_seconds(); 36881da177e4SLinus Torvalds if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) { 36891da177e4SLinus Torvalds commit_now = 1; 36901da177e4SLinus Torvalds journal->j_next_async_flush = 1; 36911da177e4SLinus Torvalds } 36921da177e4SLinus Torvalds /* don't batch when someone is waiting on j_join_wait */ 36931da177e4SLinus Torvalds /* don't batch when syncing the commit or flushing the whole trans */ 3694a228bf8fSJeff Mahoney if (!(journal->j_must_wait > 0) && !(atomic_read(&journal->j_jlock)) 3695bd4c625cSLinus Torvalds && !flush && !commit_now && (journal->j_len < journal->j_max_batch) 3696bd4c625cSLinus Torvalds && journal->j_len_alloc < journal->j_max_batch 3697bd4c625cSLinus Torvalds && journal->j_cnode_free > (journal->j_trans_max * 3)) { 36981da177e4SLinus Torvalds journal->j_bcount++; 3699a9dd3643SJeff Mahoney unlock_journal(sb); 37001da177e4SLinus Torvalds return 0; 37011da177e4SLinus Torvalds } 37021da177e4SLinus Torvalds 3703a9dd3643SJeff Mahoney if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(sb)) { 3704a9dd3643SJeff Mahoney reiserfs_panic(sb, "journal-003", 3705c3a9c210SJeff Mahoney "j_start (%ld) is too high", 3706bd4c625cSLinus Torvalds journal->j_start); 37071da177e4SLinus Torvalds } 37081da177e4SLinus Torvalds return 1; 37091da177e4SLinus Torvalds } 37101da177e4SLinus Torvalds 37111da177e4SLinus Torvalds /* 3712098297b2SJeff Mahoney * Does all the work that makes deleting blocks safe. 3713098297b2SJeff Mahoney * when deleting a block mark BH_JNew, just remove it from the current 3714098297b2SJeff Mahoney * transaction, clean it's buffer_head and move on. 3715098297b2SJeff Mahoney * 3716098297b2SJeff Mahoney * otherwise: 3717098297b2SJeff Mahoney * set a bit for the block in the journal bitmap. That will prevent it from 3718098297b2SJeff Mahoney * being allocated for unformatted nodes before this transaction has finished. 3719098297b2SJeff Mahoney * 3720098297b2SJeff Mahoney * mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers. 3721098297b2SJeff Mahoney * That will prevent any old transactions with this block from trying to flush 3722098297b2SJeff Mahoney * to the real location. Since we aren't removing the cnode from the 3723098297b2SJeff Mahoney * journal_list_hash, *the block can't be reallocated yet. 3724098297b2SJeff Mahoney * 3725098297b2SJeff Mahoney * Then remove it from the current transaction, decrementing any counters and 3726098297b2SJeff Mahoney * filing it on the clean list. 37271da177e4SLinus Torvalds */ 3728bd4c625cSLinus Torvalds int journal_mark_freed(struct reiserfs_transaction_handle *th, 3729a9dd3643SJeff Mahoney struct super_block *sb, b_blocknr_t blocknr) 3730bd4c625cSLinus Torvalds { 3731a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 37321da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn = NULL; 37331da177e4SLinus Torvalds struct buffer_head *bh = NULL; 37341da177e4SLinus Torvalds struct reiserfs_list_bitmap *jb = NULL; 37351da177e4SLinus Torvalds int cleaned = 0; 37361da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 37371da177e4SLinus Torvalds 3738a9dd3643SJeff Mahoney cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr); 37391da177e4SLinus Torvalds if (cn && cn->bh) { 37401da177e4SLinus Torvalds bh = cn->bh; 37411da177e4SLinus Torvalds get_bh(bh); 37421da177e4SLinus Torvalds } 37431da177e4SLinus Torvalds /* if it is journal new, we just remove it from this transaction */ 37441da177e4SLinus Torvalds if (bh && buffer_journal_new(bh)) { 37451da177e4SLinus Torvalds clear_buffer_journal_new(bh); 37461da177e4SLinus Torvalds clear_prepared_bits(bh); 37471da177e4SLinus Torvalds reiserfs_clean_and_file_buffer(bh); 3748a9dd3643SJeff Mahoney cleaned = remove_from_transaction(sb, blocknr, cleaned); 37491da177e4SLinus Torvalds } else { 3750098297b2SJeff Mahoney /* 3751098297b2SJeff Mahoney * set the bit for this block in the journal bitmap 3752098297b2SJeff Mahoney * for this transaction 3753098297b2SJeff Mahoney */ 37541da177e4SLinus Torvalds jb = journal->j_current_jl->j_list_bitmap; 37551da177e4SLinus Torvalds if (!jb) { 3756a9dd3643SJeff Mahoney reiserfs_panic(sb, "journal-1702", 3757c3a9c210SJeff Mahoney "journal_list_bitmap is NULL"); 37581da177e4SLinus Torvalds } 3759a9dd3643SJeff Mahoney set_bit_in_list_bitmap(sb, blocknr, jb); 37601da177e4SLinus Torvalds 37611da177e4SLinus Torvalds /* Note, the entire while loop is not allowed to schedule. */ 37621da177e4SLinus Torvalds 37631da177e4SLinus Torvalds if (bh) { 37641da177e4SLinus Torvalds clear_prepared_bits(bh); 37651da177e4SLinus Torvalds reiserfs_clean_and_file_buffer(bh); 37661da177e4SLinus Torvalds } 3767a9dd3643SJeff Mahoney cleaned = remove_from_transaction(sb, blocknr, cleaned); 37681da177e4SLinus Torvalds 3769098297b2SJeff Mahoney /* 3770098297b2SJeff Mahoney * find all older transactions with this block, 3771098297b2SJeff Mahoney * make sure they don't try to write it out 3772098297b2SJeff Mahoney */ 3773a9dd3643SJeff Mahoney cn = get_journal_hash_dev(sb, journal->j_list_hash_table, 3774bd4c625cSLinus Torvalds blocknr); 37751da177e4SLinus Torvalds while (cn) { 3776a9dd3643SJeff Mahoney if (sb == cn->sb && blocknr == cn->blocknr) { 37771da177e4SLinus Torvalds set_bit(BLOCK_FREED, &cn->state); 37781da177e4SLinus Torvalds if (cn->bh) { 3779098297b2SJeff Mahoney /* 3780098297b2SJeff Mahoney * remove_from_transaction will brelse 3781098297b2SJeff Mahoney * the buffer if it was in the current 3782098297b2SJeff Mahoney * trans 37831da177e4SLinus Torvalds */ 3784098297b2SJeff Mahoney if (!cleaned) { 3785bd4c625cSLinus Torvalds clear_buffer_journal_dirty(cn-> 3786bd4c625cSLinus Torvalds bh); 37871da177e4SLinus Torvalds clear_buffer_dirty(cn->bh); 3788bd4c625cSLinus Torvalds clear_buffer_journal_test(cn-> 3789bd4c625cSLinus Torvalds bh); 37901da177e4SLinus Torvalds cleaned = 1; 37911da177e4SLinus Torvalds put_bh(cn->bh); 3792bd4c625cSLinus Torvalds if (atomic_read 3793a228bf8fSJeff Mahoney (&cn->bh->b_count) < 0) { 3794a9dd3643SJeff Mahoney reiserfs_warning(sb, 379545b03d5eSJeff Mahoney "journal-2138", 379645b03d5eSJeff Mahoney "cn->bh->b_count < 0"); 37971da177e4SLinus Torvalds } 37981da177e4SLinus Torvalds } 3799098297b2SJeff Mahoney /* 3800098297b2SJeff Mahoney * since we are clearing the bh, 3801098297b2SJeff Mahoney * we MUST dec nonzerolen 3802098297b2SJeff Mahoney */ 3803098297b2SJeff Mahoney if (cn->jlist) { 3804a228bf8fSJeff Mahoney atomic_dec(&cn->jlist-> 3805a228bf8fSJeff Mahoney j_nonzerolen); 38061da177e4SLinus Torvalds } 38071da177e4SLinus Torvalds cn->bh = NULL; 38081da177e4SLinus Torvalds } 38091da177e4SLinus Torvalds } 38101da177e4SLinus Torvalds cn = cn->hnext; 38111da177e4SLinus Torvalds } 38121da177e4SLinus Torvalds } 38131da177e4SLinus Torvalds 3814398c95bdSChris Mason if (bh) 3815398c95bdSChris Mason release_buffer_page(bh); /* get_hash grabs the buffer */ 38161da177e4SLinus Torvalds return 0; 38171da177e4SLinus Torvalds } 38181da177e4SLinus Torvalds 3819bd4c625cSLinus Torvalds void reiserfs_update_inode_transaction(struct inode *inode) 3820bd4c625cSLinus Torvalds { 38211da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb); 38221da177e4SLinus Torvalds REISERFS_I(inode)->i_jl = journal->j_current_jl; 38231da177e4SLinus Torvalds REISERFS_I(inode)->i_trans_id = journal->j_trans_id; 38241da177e4SLinus Torvalds } 38251da177e4SLinus Torvalds 38261da177e4SLinus Torvalds /* 38271da177e4SLinus Torvalds * returns -1 on error, 0 if no commits/barriers were done and 1 38281da177e4SLinus Torvalds * if a transaction was actually committed and the barrier was done 38291da177e4SLinus Torvalds */ 38301da177e4SLinus Torvalds static int __commit_trans_jl(struct inode *inode, unsigned long id, 38311da177e4SLinus Torvalds struct reiserfs_journal_list *jl) 38321da177e4SLinus Torvalds { 38331da177e4SLinus Torvalds struct reiserfs_transaction_handle th; 38341da177e4SLinus Torvalds struct super_block *sb = inode->i_sb; 38351da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(sb); 38361da177e4SLinus Torvalds int ret = 0; 38371da177e4SLinus Torvalds 3838098297b2SJeff Mahoney /* 3839098297b2SJeff Mahoney * is it from the current transaction, 3840098297b2SJeff Mahoney * or from an unknown transaction? 3841098297b2SJeff Mahoney */ 38421da177e4SLinus Torvalds if (id == journal->j_trans_id) { 38431da177e4SLinus Torvalds jl = journal->j_current_jl; 3844098297b2SJeff Mahoney /* 3845098297b2SJeff Mahoney * try to let other writers come in and 3846098297b2SJeff Mahoney * grow this transaction 3847098297b2SJeff Mahoney */ 38481da177e4SLinus Torvalds let_transaction_grow(sb, id); 38491da177e4SLinus Torvalds if (journal->j_trans_id != id) { 38501da177e4SLinus Torvalds goto flush_commit_only; 38511da177e4SLinus Torvalds } 38521da177e4SLinus Torvalds 38531da177e4SLinus Torvalds ret = journal_begin(&th, sb, 1); 38541da177e4SLinus Torvalds if (ret) 38551da177e4SLinus Torvalds return ret; 38561da177e4SLinus Torvalds 38571da177e4SLinus Torvalds /* someone might have ended this transaction while we joined */ 38581da177e4SLinus Torvalds if (journal->j_trans_id != id) { 3859bd4c625cSLinus Torvalds reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb), 3860bd4c625cSLinus Torvalds 1); 386109f1b80bSJeff Mahoney journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb)); 386258d85426SJeff Mahoney ret = journal_end(&th); 38631da177e4SLinus Torvalds goto flush_commit_only; 38641da177e4SLinus Torvalds } 38651da177e4SLinus Torvalds 386658d85426SJeff Mahoney ret = journal_end_sync(&th); 38671da177e4SLinus Torvalds if (!ret) 38681da177e4SLinus Torvalds ret = 1; 38691da177e4SLinus Torvalds 38701da177e4SLinus Torvalds } else { 3871098297b2SJeff Mahoney /* 3872098297b2SJeff Mahoney * this gets tricky, we have to make sure the journal list in 38731da177e4SLinus Torvalds * the inode still exists. We know the list is still around 38741da177e4SLinus Torvalds * if we've got a larger transaction id than the oldest list 38751da177e4SLinus Torvalds */ 38761da177e4SLinus Torvalds flush_commit_only: 38771da177e4SLinus Torvalds if (journal_list_still_alive(inode->i_sb, id)) { 38781da177e4SLinus Torvalds /* 38791da177e4SLinus Torvalds * we only set ret to 1 when we know for sure 38801da177e4SLinus Torvalds * the barrier hasn't been started yet on the commit 38811da177e4SLinus Torvalds * block. 38821da177e4SLinus Torvalds */ 38831da177e4SLinus Torvalds if (atomic_read(&jl->j_commit_left) > 1) 38841da177e4SLinus Torvalds ret = 1; 38851da177e4SLinus Torvalds flush_commit_list(sb, jl, 1); 38861da177e4SLinus Torvalds if (journal->j_errno) 38871da177e4SLinus Torvalds ret = journal->j_errno; 38881da177e4SLinus Torvalds } 38891da177e4SLinus Torvalds } 38901da177e4SLinus Torvalds /* otherwise the list is gone, and long since committed */ 38911da177e4SLinus Torvalds return ret; 38921da177e4SLinus Torvalds } 38931da177e4SLinus Torvalds 3894bd4c625cSLinus Torvalds int reiserfs_commit_for_inode(struct inode *inode) 3895bd4c625cSLinus Torvalds { 3896600ed416SJeff Mahoney unsigned int id = REISERFS_I(inode)->i_trans_id; 38971da177e4SLinus Torvalds struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl; 38981da177e4SLinus Torvalds 3899098297b2SJeff Mahoney /* 3900098297b2SJeff Mahoney * for the whole inode, assume unset id means it was 39011da177e4SLinus Torvalds * changed in the current transaction. More conservative 39021da177e4SLinus Torvalds */ 39031da177e4SLinus Torvalds if (!id || !jl) { 39041da177e4SLinus Torvalds reiserfs_update_inode_transaction(inode); 39051da177e4SLinus Torvalds id = REISERFS_I(inode)->i_trans_id; 39061da177e4SLinus Torvalds /* jl will be updated in __commit_trans_jl */ 39071da177e4SLinus Torvalds } 39081da177e4SLinus Torvalds 39091da177e4SLinus Torvalds return __commit_trans_jl(inode, id, jl); 39101da177e4SLinus Torvalds } 39111da177e4SLinus Torvalds 3912a9dd3643SJeff Mahoney void reiserfs_restore_prepared_buffer(struct super_block *sb, 3913bd4c625cSLinus Torvalds struct buffer_head *bh) 3914bd4c625cSLinus Torvalds { 3915a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 3916a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.restore_prepared); 39171da177e4SLinus Torvalds if (!bh) { 39181da177e4SLinus Torvalds return; 39191da177e4SLinus Torvalds } 39201da177e4SLinus Torvalds if (test_clear_buffer_journal_restore_dirty(bh) && 39211da177e4SLinus Torvalds buffer_journal_dirty(bh)) { 39221da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 3923278f6679SJeff Mahoney reiserfs_write_lock(sb); 3924a9dd3643SJeff Mahoney cn = get_journal_hash_dev(sb, 39251da177e4SLinus Torvalds journal->j_list_hash_table, 39261da177e4SLinus Torvalds bh->b_blocknr); 39271da177e4SLinus Torvalds if (cn && can_dirty(cn)) { 39281da177e4SLinus Torvalds set_buffer_journal_test(bh); 39291da177e4SLinus Torvalds mark_buffer_dirty(bh); 39301da177e4SLinus Torvalds } 3931278f6679SJeff Mahoney reiserfs_write_unlock(sb); 39321da177e4SLinus Torvalds } 39331da177e4SLinus Torvalds clear_buffer_journal_prepared(bh); 39341da177e4SLinus Torvalds } 39351da177e4SLinus Torvalds 39361da177e4SLinus Torvalds extern struct tree_balance *cur_tb; 39371da177e4SLinus Torvalds /* 3938098297b2SJeff Mahoney * before we can change a metadata block, we have to make sure it won't 3939098297b2SJeff Mahoney * be written to disk while we are altering it. So, we must: 3940098297b2SJeff Mahoney * clean it 3941098297b2SJeff Mahoney * wait on it. 39421da177e4SLinus Torvalds */ 3943a9dd3643SJeff Mahoney int reiserfs_prepare_for_journal(struct super_block *sb, 3944bd4c625cSLinus Torvalds struct buffer_head *bh, int wait) 3945bd4c625cSLinus Torvalds { 3946a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.prepare); 39471da177e4SLinus Torvalds 3948ca5de404SNick Piggin if (!trylock_buffer(bh)) { 39491da177e4SLinus Torvalds if (!wait) 39501da177e4SLinus Torvalds return 0; 39511da177e4SLinus Torvalds lock_buffer(bh); 39521da177e4SLinus Torvalds } 39531da177e4SLinus Torvalds set_buffer_journal_prepared(bh); 39541da177e4SLinus Torvalds if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) { 39551da177e4SLinus Torvalds clear_buffer_journal_test(bh); 39561da177e4SLinus Torvalds set_buffer_journal_restore_dirty(bh); 39571da177e4SLinus Torvalds } 39581da177e4SLinus Torvalds unlock_buffer(bh); 39591da177e4SLinus Torvalds return 1; 39601da177e4SLinus Torvalds } 39611da177e4SLinus Torvalds 39621da177e4SLinus Torvalds /* 3963098297b2SJeff Mahoney * long and ugly. If flush, will not return until all commit 3964098297b2SJeff Mahoney * blocks and all real buffers in the trans are on disk. 3965098297b2SJeff Mahoney * If no_async, won't return until all commit blocks are on disk. 3966098297b2SJeff Mahoney * 3967098297b2SJeff Mahoney * keep reading, there are comments as you go along 3968098297b2SJeff Mahoney * 3969098297b2SJeff Mahoney * If the journal is aborted, we just clean up. Things like flushing 3970098297b2SJeff Mahoney * journal lists, etc just won't happen. 39711da177e4SLinus Torvalds */ 397258d85426SJeff Mahoney static int do_journal_end(struct reiserfs_transaction_handle *th, int flags) 3973bd4c625cSLinus Torvalds { 397458d85426SJeff Mahoney struct super_block *sb = th->t_super; 3975a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 39761da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn, *next, *jl_cn; 39771da177e4SLinus Torvalds struct reiserfs_journal_cnode *last_cn = NULL; 39781da177e4SLinus Torvalds struct reiserfs_journal_desc *desc; 39791da177e4SLinus Torvalds struct reiserfs_journal_commit *commit; 39801da177e4SLinus Torvalds struct buffer_head *c_bh; /* commit bh */ 39811da177e4SLinus Torvalds struct buffer_head *d_bh; /* desc bh */ 39821da177e4SLinus Torvalds int cur_write_start = 0; /* start index of current log write */ 39831da177e4SLinus Torvalds int i; 3984a44c94a7SAlexander Zarochentsev int flush; 3985a44c94a7SAlexander Zarochentsev int wait_on_commit; 39861da177e4SLinus Torvalds struct reiserfs_journal_list *jl, *temp_jl; 39871da177e4SLinus Torvalds struct list_head *entry, *safe; 39881da177e4SLinus Torvalds unsigned long jindex; 3989600ed416SJeff Mahoney unsigned int commit_trans_id; 39901da177e4SLinus Torvalds int trans_half; 3991278f6679SJeff Mahoney int depth; 39921da177e4SLinus Torvalds 39931da177e4SLinus Torvalds BUG_ON(th->t_refcount > 1); 39941da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 399558d85426SJeff Mahoney BUG_ON(!th->t_super); 39961da177e4SLinus Torvalds 3997098297b2SJeff Mahoney /* 3998098297b2SJeff Mahoney * protect flush_older_commits from doing mistakes if the 3999098297b2SJeff Mahoney * transaction ID counter gets overflowed. 4000098297b2SJeff Mahoney */ 4001600ed416SJeff Mahoney if (th->t_trans_id == ~0U) 4002a44c94a7SAlexander Zarochentsev flags |= FLUSH_ALL | COMMIT_NOW | WAIT; 4003a44c94a7SAlexander Zarochentsev flush = flags & FLUSH_ALL; 4004a44c94a7SAlexander Zarochentsev wait_on_commit = flags & WAIT; 4005a44c94a7SAlexander Zarochentsev 40061da177e4SLinus Torvalds current->journal_info = th->t_handle_save; 4007a9dd3643SJeff Mahoney reiserfs_check_lock_depth(sb, "journal end"); 40081da177e4SLinus Torvalds if (journal->j_len == 0) { 4009a9dd3643SJeff Mahoney reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb), 4010bd4c625cSLinus Torvalds 1); 401109f1b80bSJeff Mahoney journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb)); 40121da177e4SLinus Torvalds } 40131da177e4SLinus Torvalds 4014a9dd3643SJeff Mahoney lock_journal(sb); 40151da177e4SLinus Torvalds if (journal->j_next_full_flush) { 40161da177e4SLinus Torvalds flags |= FLUSH_ALL; 40171da177e4SLinus Torvalds flush = 1; 40181da177e4SLinus Torvalds } 40191da177e4SLinus Torvalds if (journal->j_next_async_flush) { 40201da177e4SLinus Torvalds flags |= COMMIT_NOW | WAIT; 40211da177e4SLinus Torvalds wait_on_commit = 1; 40221da177e4SLinus Torvalds } 40231da177e4SLinus Torvalds 4024098297b2SJeff Mahoney /* 4025098297b2SJeff Mahoney * check_journal_end locks the journal, and unlocks if it does 4026098297b2SJeff Mahoney * not return 1 it tells us if we should continue with the 4027098297b2SJeff Mahoney * journal_end, or just return 40281da177e4SLinus Torvalds */ 402958d85426SJeff Mahoney if (!check_journal_end(th, flags)) { 4030033369d1SArtem Bityutskiy reiserfs_schedule_old_flush(sb); 4031a9dd3643SJeff Mahoney wake_queued_writers(sb); 4032a9dd3643SJeff Mahoney reiserfs_async_progress_wait(sb); 40331da177e4SLinus Torvalds goto out; 40341da177e4SLinus Torvalds } 40351da177e4SLinus Torvalds 40361da177e4SLinus Torvalds /* check_journal_end might set these, check again */ 40371da177e4SLinus Torvalds if (journal->j_next_full_flush) { 40381da177e4SLinus Torvalds flush = 1; 40391da177e4SLinus Torvalds } 40401da177e4SLinus Torvalds 40411da177e4SLinus Torvalds /* 4042098297b2SJeff Mahoney * j must wait means we have to flush the log blocks, and the 4043098297b2SJeff Mahoney * real blocks for this transaction 40441da177e4SLinus Torvalds */ 40451da177e4SLinus Torvalds if (journal->j_must_wait > 0) { 40461da177e4SLinus Torvalds flush = 1; 40471da177e4SLinus Torvalds } 40481da177e4SLinus Torvalds #ifdef REISERFS_PREALLOCATE 4049098297b2SJeff Mahoney /* 4050098297b2SJeff Mahoney * quota ops might need to nest, setup the journal_info pointer 4051098297b2SJeff Mahoney * for them and raise the refcount so that it is > 0. 4052098297b2SJeff Mahoney */ 40531da177e4SLinus Torvalds current->journal_info = th; 4054ef43bc4fSJan Kara th->t_refcount++; 4055098297b2SJeff Mahoney 4056098297b2SJeff Mahoney /* it should not involve new blocks into the transaction */ 4057098297b2SJeff Mahoney reiserfs_discard_all_prealloc(th); 4058098297b2SJeff Mahoney 4059ef43bc4fSJan Kara th->t_refcount--; 40601da177e4SLinus Torvalds current->journal_info = th->t_handle_save; 40611da177e4SLinus Torvalds #endif 40621da177e4SLinus Torvalds 40631da177e4SLinus Torvalds /* setup description block */ 4064bd4c625cSLinus Torvalds d_bh = 4065a9dd3643SJeff Mahoney journal_getblk(sb, 4066a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 4067bd4c625cSLinus Torvalds journal->j_start); 40681da177e4SLinus Torvalds set_buffer_uptodate(d_bh); 40691da177e4SLinus Torvalds desc = (struct reiserfs_journal_desc *)(d_bh)->b_data; 40701da177e4SLinus Torvalds memset(d_bh->b_data, 0, d_bh->b_size); 40711da177e4SLinus Torvalds memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8); 40721da177e4SLinus Torvalds set_desc_trans_id(desc, journal->j_trans_id); 40731da177e4SLinus Torvalds 4074098297b2SJeff Mahoney /* 4075098297b2SJeff Mahoney * setup commit block. Don't write (keep it clean too) this one 4076098297b2SJeff Mahoney * until after everyone else is written 4077098297b2SJeff Mahoney */ 4078a9dd3643SJeff Mahoney c_bh = journal_getblk(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 4079bd4c625cSLinus Torvalds ((journal->j_start + journal->j_len + 4080a9dd3643SJeff Mahoney 1) % SB_ONDISK_JOURNAL_SIZE(sb))); 40811da177e4SLinus Torvalds commit = (struct reiserfs_journal_commit *)c_bh->b_data; 40821da177e4SLinus Torvalds memset(c_bh->b_data, 0, c_bh->b_size); 40831da177e4SLinus Torvalds set_commit_trans_id(commit, journal->j_trans_id); 40841da177e4SLinus Torvalds set_buffer_uptodate(c_bh); 40851da177e4SLinus Torvalds 40861da177e4SLinus Torvalds /* init this journal list */ 40871da177e4SLinus Torvalds jl = journal->j_current_jl; 40881da177e4SLinus Torvalds 4089098297b2SJeff Mahoney /* 4090098297b2SJeff Mahoney * we lock the commit before doing anything because 40911da177e4SLinus Torvalds * we want to make sure nobody tries to run flush_commit_list until 40921da177e4SLinus Torvalds * the new transaction is fully setup, and we've already flushed the 40931da177e4SLinus Torvalds * ordered bh list 40941da177e4SLinus Torvalds */ 40958ebc4232SFrederic Weisbecker reiserfs_mutex_lock_safe(&jl->j_commit_mutex, sb); 40961da177e4SLinus Torvalds 40971da177e4SLinus Torvalds /* save the transaction id in case we need to commit it later */ 40981da177e4SLinus Torvalds commit_trans_id = jl->j_trans_id; 40991da177e4SLinus Torvalds 41001da177e4SLinus Torvalds atomic_set(&jl->j_older_commits_done, 0); 41011da177e4SLinus Torvalds jl->j_trans_id = journal->j_trans_id; 41021da177e4SLinus Torvalds jl->j_timestamp = journal->j_trans_start_time; 41031da177e4SLinus Torvalds jl->j_commit_bh = c_bh; 41041da177e4SLinus Torvalds jl->j_start = journal->j_start; 41051da177e4SLinus Torvalds jl->j_len = journal->j_len; 41061da177e4SLinus Torvalds atomic_set(&jl->j_nonzerolen, journal->j_len); 41071da177e4SLinus Torvalds atomic_set(&jl->j_commit_left, journal->j_len + 2); 41081da177e4SLinus Torvalds jl->j_realblock = NULL; 41091da177e4SLinus Torvalds 4110098297b2SJeff Mahoney /* 4111098297b2SJeff Mahoney * The ENTIRE FOR LOOP MUST not cause schedule to occur. 4112098297b2SJeff Mahoney * for each real block, add it to the journal list hash, 4113098297b2SJeff Mahoney * copy into real block index array in the commit or desc block 41141da177e4SLinus Torvalds */ 4115a9dd3643SJeff Mahoney trans_half = journal_trans_half(sb->s_blocksize); 41161da177e4SLinus Torvalds for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) { 41171da177e4SLinus Torvalds if (buffer_journaled(cn->bh)) { 4118a9dd3643SJeff Mahoney jl_cn = get_cnode(sb); 41191da177e4SLinus Torvalds if (!jl_cn) { 4120a9dd3643SJeff Mahoney reiserfs_panic(sb, "journal-1676", 4121c3a9c210SJeff Mahoney "get_cnode returned NULL"); 41221da177e4SLinus Torvalds } 41231da177e4SLinus Torvalds if (i == 0) { 41241da177e4SLinus Torvalds jl->j_realblock = jl_cn; 41251da177e4SLinus Torvalds } 41261da177e4SLinus Torvalds jl_cn->prev = last_cn; 41271da177e4SLinus Torvalds jl_cn->next = NULL; 41281da177e4SLinus Torvalds if (last_cn) { 41291da177e4SLinus Torvalds last_cn->next = jl_cn; 41301da177e4SLinus Torvalds } 41311da177e4SLinus Torvalds last_cn = jl_cn; 4132098297b2SJeff Mahoney /* 4133098297b2SJeff Mahoney * make sure the block we are trying to log 4134098297b2SJeff Mahoney * is not a block of journal or reserved area 4135098297b2SJeff Mahoney */ 4136bd4c625cSLinus Torvalds if (is_block_in_log_or_reserved_area 4137a9dd3643SJeff Mahoney (sb, cn->bh->b_blocknr)) { 4138a9dd3643SJeff Mahoney reiserfs_panic(sb, "journal-2332", 4139c3a9c210SJeff Mahoney "Trying to log block %lu, " 4140c3a9c210SJeff Mahoney "which is a log block", 4141bd4c625cSLinus Torvalds cn->bh->b_blocknr); 41421da177e4SLinus Torvalds } 41431da177e4SLinus Torvalds jl_cn->blocknr = cn->bh->b_blocknr; 41441da177e4SLinus Torvalds jl_cn->state = 0; 4145a9dd3643SJeff Mahoney jl_cn->sb = sb; 41461da177e4SLinus Torvalds jl_cn->bh = cn->bh; 41471da177e4SLinus Torvalds jl_cn->jlist = jl; 41481da177e4SLinus Torvalds insert_journal_hash(journal->j_list_hash_table, jl_cn); 41491da177e4SLinus Torvalds if (i < trans_half) { 4150bd4c625cSLinus Torvalds desc->j_realblock[i] = 4151bd4c625cSLinus Torvalds cpu_to_le32(cn->bh->b_blocknr); 41521da177e4SLinus Torvalds } else { 4153bd4c625cSLinus Torvalds commit->j_realblock[i - trans_half] = 4154bd4c625cSLinus Torvalds cpu_to_le32(cn->bh->b_blocknr); 41551da177e4SLinus Torvalds } 41561da177e4SLinus Torvalds } else { 41571da177e4SLinus Torvalds i--; 41581da177e4SLinus Torvalds } 41591da177e4SLinus Torvalds } 41601da177e4SLinus Torvalds set_desc_trans_len(desc, journal->j_len); 41611da177e4SLinus Torvalds set_desc_mount_id(desc, journal->j_mount_id); 41621da177e4SLinus Torvalds set_desc_trans_id(desc, journal->j_trans_id); 41631da177e4SLinus Torvalds set_commit_trans_len(commit, journal->j_len); 41641da177e4SLinus Torvalds 4165098297b2SJeff Mahoney /* 4166098297b2SJeff Mahoney * special check in case all buffers in the journal 4167098297b2SJeff Mahoney * were marked for not logging 4168098297b2SJeff Mahoney */ 416914a61442SEric Sesterhenn BUG_ON(journal->j_len == 0); 41701da177e4SLinus Torvalds 4171098297b2SJeff Mahoney /* 4172098297b2SJeff Mahoney * we're about to dirty all the log blocks, mark the description block 41731da177e4SLinus Torvalds * dirty now too. Don't mark the commit block dirty until all the 41741da177e4SLinus Torvalds * others are on disk 41751da177e4SLinus Torvalds */ 41761da177e4SLinus Torvalds mark_buffer_dirty(d_bh); 41771da177e4SLinus Torvalds 4178098297b2SJeff Mahoney /* 4179098297b2SJeff Mahoney * first data block is j_start + 1, so add one to 4180098297b2SJeff Mahoney * cur_write_start wherever you use it 4181098297b2SJeff Mahoney */ 41821da177e4SLinus Torvalds cur_write_start = journal->j_start; 41831da177e4SLinus Torvalds cn = journal->j_first; 41841da177e4SLinus Torvalds jindex = 1; /* start at one so we don't get the desc again */ 41851da177e4SLinus Torvalds while (cn) { 41861da177e4SLinus Torvalds clear_buffer_journal_new(cn->bh); 41871da177e4SLinus Torvalds /* copy all the real blocks into log area. dirty log blocks */ 41881da177e4SLinus Torvalds if (buffer_journaled(cn->bh)) { 41891da177e4SLinus Torvalds struct buffer_head *tmp_bh; 41901da177e4SLinus Torvalds char *addr; 41911da177e4SLinus Torvalds struct page *page; 4192bd4c625cSLinus Torvalds tmp_bh = 4193a9dd3643SJeff Mahoney journal_getblk(sb, 4194a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 4195bd4c625cSLinus Torvalds ((cur_write_start + 4196bd4c625cSLinus Torvalds jindex) % 4197a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb))); 41981da177e4SLinus Torvalds set_buffer_uptodate(tmp_bh); 41991da177e4SLinus Torvalds page = cn->bh->b_page; 42001da177e4SLinus Torvalds addr = kmap(page); 4201bd4c625cSLinus Torvalds memcpy(tmp_bh->b_data, 4202bd4c625cSLinus Torvalds addr + offset_in_page(cn->bh->b_data), 42031da177e4SLinus Torvalds cn->bh->b_size); 42041da177e4SLinus Torvalds kunmap(page); 42051da177e4SLinus Torvalds mark_buffer_dirty(tmp_bh); 42061da177e4SLinus Torvalds jindex++; 42071da177e4SLinus Torvalds set_buffer_journal_dirty(cn->bh); 42081da177e4SLinus Torvalds clear_buffer_journaled(cn->bh); 42091da177e4SLinus Torvalds } else { 4210098297b2SJeff Mahoney /* 4211098297b2SJeff Mahoney * JDirty cleared sometime during transaction. 4212098297b2SJeff Mahoney * don't log this one 4213098297b2SJeff Mahoney */ 4214a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-2048", 421545b03d5eSJeff Mahoney "BAD, buffer in journal hash, " 421645b03d5eSJeff Mahoney "but not JDirty!"); 42171da177e4SLinus Torvalds brelse(cn->bh); 42181da177e4SLinus Torvalds } 42191da177e4SLinus Torvalds next = cn->next; 4220a9dd3643SJeff Mahoney free_cnode(sb, cn); 42211da177e4SLinus Torvalds cn = next; 4222278f6679SJeff Mahoney reiserfs_cond_resched(sb); 42231da177e4SLinus Torvalds } 42241da177e4SLinus Torvalds 4225098297b2SJeff Mahoney /* 4226098297b2SJeff Mahoney * we are done with both the c_bh and d_bh, but 4227098297b2SJeff Mahoney * c_bh must be written after all other commit blocks, 4228098297b2SJeff Mahoney * so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1. 42291da177e4SLinus Torvalds */ 42301da177e4SLinus Torvalds 4231a9dd3643SJeff Mahoney journal->j_current_jl = alloc_journal_list(sb); 42321da177e4SLinus Torvalds 42331da177e4SLinus Torvalds /* now it is safe to insert this transaction on the main list */ 42341da177e4SLinus Torvalds list_add_tail(&jl->j_list, &journal->j_journal_list); 42351da177e4SLinus Torvalds list_add_tail(&jl->j_working_list, &journal->j_working_list); 42361da177e4SLinus Torvalds journal->j_num_work_lists++; 42371da177e4SLinus Torvalds 42381da177e4SLinus Torvalds /* reset journal values for the next transaction */ 4239bd4c625cSLinus Torvalds journal->j_start = 4240bd4c625cSLinus Torvalds (journal->j_start + journal->j_len + 4241a9dd3643SJeff Mahoney 2) % SB_ONDISK_JOURNAL_SIZE(sb); 4242a228bf8fSJeff Mahoney atomic_set(&journal->j_wcount, 0); 42431da177e4SLinus Torvalds journal->j_bcount = 0; 42441da177e4SLinus Torvalds journal->j_last = NULL; 42451da177e4SLinus Torvalds journal->j_first = NULL; 42461da177e4SLinus Torvalds journal->j_len = 0; 42471da177e4SLinus Torvalds journal->j_trans_start_time = 0; 4248a44c94a7SAlexander Zarochentsev /* check for trans_id overflow */ 4249a44c94a7SAlexander Zarochentsev if (++journal->j_trans_id == 0) 4250a44c94a7SAlexander Zarochentsev journal->j_trans_id = 10; 42511da177e4SLinus Torvalds journal->j_current_jl->j_trans_id = journal->j_trans_id; 42521da177e4SLinus Torvalds journal->j_must_wait = 0; 42531da177e4SLinus Torvalds journal->j_len_alloc = 0; 42541da177e4SLinus Torvalds journal->j_next_full_flush = 0; 42551da177e4SLinus Torvalds journal->j_next_async_flush = 0; 4256a9dd3643SJeff Mahoney init_journal_hash(sb); 42571da177e4SLinus Torvalds 4258098297b2SJeff Mahoney /* 4259098297b2SJeff Mahoney * make sure reiserfs_add_jh sees the new current_jl before we 4260098297b2SJeff Mahoney * write out the tails 4261098297b2SJeff Mahoney */ 42621da177e4SLinus Torvalds smp_mb(); 42631da177e4SLinus Torvalds 4264098297b2SJeff Mahoney /* 4265098297b2SJeff Mahoney * tail conversion targets have to hit the disk before we end the 42661da177e4SLinus Torvalds * transaction. Otherwise a later transaction might repack the tail 4267098297b2SJeff Mahoney * before this transaction commits, leaving the data block unflushed 4268098297b2SJeff Mahoney * and clean, if we crash before the later transaction commits, the 4269098297b2SJeff Mahoney * data block is lost. 42701da177e4SLinus Torvalds */ 42711da177e4SLinus Torvalds if (!list_empty(&jl->j_tail_bh_list)) { 4272278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(sb); 42731da177e4SLinus Torvalds write_ordered_buffers(&journal->j_dirty_buffers_lock, 42741da177e4SLinus Torvalds journal, jl, &jl->j_tail_bh_list); 4275278f6679SJeff Mahoney reiserfs_write_lock_nested(sb, depth); 42761da177e4SLinus Torvalds } 427714a61442SEric Sesterhenn BUG_ON(!list_empty(&jl->j_tail_bh_list)); 427890415deaSJeff Mahoney mutex_unlock(&jl->j_commit_mutex); 42791da177e4SLinus Torvalds 4280098297b2SJeff Mahoney /* 4281098297b2SJeff Mahoney * honor the flush wishes from the caller, simple commits can 4282098297b2SJeff Mahoney * be done outside the journal lock, they are done below 4283098297b2SJeff Mahoney * 4284098297b2SJeff Mahoney * if we don't flush the commit list right now, we put it into 4285098297b2SJeff Mahoney * the work queue so the people waiting on the async progress work 4286098297b2SJeff Mahoney * queue don't wait for this proc to flush journal lists and such. 42871da177e4SLinus Torvalds */ 42881da177e4SLinus Torvalds if (flush) { 4289a9dd3643SJeff Mahoney flush_commit_list(sb, jl, 1); 4290a9dd3643SJeff Mahoney flush_journal_list(sb, jl, 1); 429101777836SJan Kara } else if (!(jl->j_state & LIST_COMMIT_PENDING)) { 429201777836SJan Kara /* 429301777836SJan Kara * Avoid queueing work when sb is being shut down. Transaction 429401777836SJan Kara * will be flushed on journal shutdown. 429501777836SJan Kara */ 42961751e8a6SLinus Torvalds if (sb->s_flags & SB_ACTIVE) 4297797d9016SJeff Mahoney queue_delayed_work(REISERFS_SB(sb)->commit_wq, 4298797d9016SJeff Mahoney &journal->j_work, HZ / 10); 429901777836SJan Kara } 43001da177e4SLinus Torvalds 4301098297b2SJeff Mahoney /* 4302098297b2SJeff Mahoney * if the next transaction has any chance of wrapping, flush 4303098297b2SJeff Mahoney * transactions that might get overwritten. If any journal lists 4304098297b2SJeff Mahoney * are very old flush them as well. 43051da177e4SLinus Torvalds */ 43061da177e4SLinus Torvalds first_jl: 43071da177e4SLinus Torvalds list_for_each_safe(entry, safe, &journal->j_journal_list) { 43081da177e4SLinus Torvalds temp_jl = JOURNAL_LIST_ENTRY(entry); 43091da177e4SLinus Torvalds if (journal->j_start <= temp_jl->j_start) { 43101da177e4SLinus Torvalds if ((journal->j_start + journal->j_trans_max + 1) >= 4311bd4c625cSLinus Torvalds temp_jl->j_start) { 4312a9dd3643SJeff Mahoney flush_used_journal_lists(sb, temp_jl); 43131da177e4SLinus Torvalds goto first_jl; 43141da177e4SLinus Torvalds } else if ((journal->j_start + 43151da177e4SLinus Torvalds journal->j_trans_max + 1) < 4316a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb)) { 4317098297b2SJeff Mahoney /* 4318098297b2SJeff Mahoney * if we don't cross into the next 4319098297b2SJeff Mahoney * transaction and we don't wrap, there is 4320098297b2SJeff Mahoney * no way we can overlap any later transactions 43211da177e4SLinus Torvalds * break now 43221da177e4SLinus Torvalds */ 43231da177e4SLinus Torvalds break; 43241da177e4SLinus Torvalds } 43251da177e4SLinus Torvalds } else if ((journal->j_start + 43261da177e4SLinus Torvalds journal->j_trans_max + 1) > 4327a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb)) { 43281da177e4SLinus Torvalds if (((journal->j_start + journal->j_trans_max + 1) % 4329a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb)) >= 4330bd4c625cSLinus Torvalds temp_jl->j_start) { 4331a9dd3643SJeff Mahoney flush_used_journal_lists(sb, temp_jl); 43321da177e4SLinus Torvalds goto first_jl; 43331da177e4SLinus Torvalds } else { 4334098297b2SJeff Mahoney /* 4335098297b2SJeff Mahoney * we don't overlap anything from out start 4336098297b2SJeff Mahoney * to the end of the log, and our wrapped 4337098297b2SJeff Mahoney * portion doesn't overlap anything at 43381da177e4SLinus Torvalds * the start of the log. We can break 43391da177e4SLinus Torvalds */ 43401da177e4SLinus Torvalds break; 43411da177e4SLinus Torvalds } 43421da177e4SLinus Torvalds } 43431da177e4SLinus Torvalds } 43441da177e4SLinus Torvalds 4345bd4c625cSLinus Torvalds journal->j_current_jl->j_list_bitmap = 4346a9dd3643SJeff Mahoney get_list_bitmap(sb, journal->j_current_jl); 43471da177e4SLinus Torvalds 43481da177e4SLinus Torvalds if (!(journal->j_current_jl->j_list_bitmap)) { 4349a9dd3643SJeff Mahoney reiserfs_panic(sb, "journal-1996", 4350c3a9c210SJeff Mahoney "could not get a list bitmap"); 43511da177e4SLinus Torvalds } 43521da177e4SLinus Torvalds 4353a228bf8fSJeff Mahoney atomic_set(&journal->j_jlock, 0); 4354a9dd3643SJeff Mahoney unlock_journal(sb); 43551da177e4SLinus Torvalds /* wake up any body waiting to join. */ 43561da177e4SLinus Torvalds clear_bit(J_WRITERS_QUEUED, &journal->j_state); 4357a228bf8fSJeff Mahoney wake_up(&journal->j_join_wait); 43581da177e4SLinus Torvalds 43591da177e4SLinus Torvalds if (!flush && wait_on_commit && 4360a9dd3643SJeff Mahoney journal_list_still_alive(sb, commit_trans_id)) { 4361a9dd3643SJeff Mahoney flush_commit_list(sb, jl, 1); 43621da177e4SLinus Torvalds } 43631da177e4SLinus Torvalds out: 4364a9dd3643SJeff Mahoney reiserfs_check_lock_depth(sb, "journal end2"); 43651da177e4SLinus Torvalds 43661da177e4SLinus Torvalds memset(th, 0, sizeof(*th)); 4367098297b2SJeff Mahoney /* 4368098297b2SJeff Mahoney * Re-set th->t_super, so we can properly keep track of how many 43691da177e4SLinus Torvalds * persistent transactions there are. We need to do this so if this 4370098297b2SJeff Mahoney * call is part of a failed restart_transaction, we can free it later 4371098297b2SJeff Mahoney */ 4372a9dd3643SJeff Mahoney th->t_super = sb; 43731da177e4SLinus Torvalds 43741da177e4SLinus Torvalds return journal->j_errno; 43751da177e4SLinus Torvalds } 43761da177e4SLinus Torvalds 437732e8b106SJeff Mahoney /* Send the file system read only and refuse new transactions */ 437832e8b106SJeff Mahoney void reiserfs_abort_journal(struct super_block *sb, int errno) 43791da177e4SLinus Torvalds { 43801da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(sb); 43811da177e4SLinus Torvalds if (test_bit(J_ABORTED, &journal->j_state)) 43821da177e4SLinus Torvalds return; 43831da177e4SLinus Torvalds 438432e8b106SJeff Mahoney if (!journal->j_errno) 438532e8b106SJeff Mahoney journal->j_errno = errno; 43861da177e4SLinus Torvalds 43871751e8a6SLinus Torvalds sb->s_flags |= SB_RDONLY; 43881da177e4SLinus Torvalds set_bit(J_ABORTED, &journal->j_state); 43891da177e4SLinus Torvalds 43901da177e4SLinus Torvalds #ifdef CONFIG_REISERFS_CHECK 43911da177e4SLinus Torvalds dump_stack(); 43921da177e4SLinus Torvalds #endif 43931da177e4SLinus Torvalds } 4394