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 359436fb4dSRandy Dunlap * 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 60098297b2SJeff Mahoney /* must be correct to keep the desc and commit structs at 4k */ 61098297b2SJeff Mahoney #define JOURNAL_TRANS_HALF 1018 621da177e4SLinus Torvalds #define BUFNR 64 /*read ahead */ 631da177e4SLinus Torvalds 641da177e4SLinus Torvalds /* cnode stat bits. Move these into reiserfs_fs.h */ 651da177e4SLinus Torvalds 66098297b2SJeff Mahoney /* this block was freed, and can't be written. */ 67098297b2SJeff Mahoney #define BLOCK_FREED 2 68098297b2SJeff Mahoney /* this block was freed during this transaction, and can't be written */ 69098297b2SJeff Mahoney #define BLOCK_FREED_HOLDER 3 701da177e4SLinus Torvalds 71098297b2SJeff Mahoney /* used in flush_journal_list */ 72098297b2SJeff Mahoney #define BLOCK_NEEDS_FLUSH 4 731da177e4SLinus Torvalds #define BLOCK_DIRTIED 5 741da177e4SLinus Torvalds 751da177e4SLinus Torvalds /* journal list state bits */ 761da177e4SLinus Torvalds #define LIST_TOUCHED 1 771da177e4SLinus Torvalds #define LIST_DIRTY 2 781da177e4SLinus Torvalds #define LIST_COMMIT_PENDING 4 /* someone will commit this list */ 791da177e4SLinus Torvalds 801da177e4SLinus Torvalds /* flags for do_journal_end */ 811da177e4SLinus Torvalds #define FLUSH_ALL 1 /* flush commit and real blocks */ 821da177e4SLinus Torvalds #define COMMIT_NOW 2 /* end and commit this transaction */ 831da177e4SLinus Torvalds #define WAIT 4 /* wait for the log blocks to hit the disk */ 841da177e4SLinus Torvalds 8558d85426SJeff Mahoney static int do_journal_end(struct reiserfs_transaction_handle *, int flags); 86bd4c625cSLinus Torvalds static int flush_journal_list(struct super_block *s, 87bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl, int flushall); 88bd4c625cSLinus Torvalds static int flush_commit_list(struct super_block *s, 89bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl, int flushall); 901da177e4SLinus Torvalds static int can_dirty(struct reiserfs_journal_cnode *cn); 91bd4c625cSLinus Torvalds static int journal_join(struct reiserfs_transaction_handle *th, 92b491dd17SJeff Mahoney struct super_block *sb); 934385bab1SAl Viro static void release_journal_dev(struct super_block *super, 941da177e4SLinus Torvalds struct reiserfs_journal *journal); 95dc0dde61SHariprasad Kelam static void dirty_one_transaction(struct super_block *s, 961da177e4SLinus Torvalds struct reiserfs_journal_list *jl); 97c4028958SDavid Howells static void flush_async_commits(struct work_struct *work); 981da177e4SLinus Torvalds static void queue_log_writer(struct super_block *s); 991da177e4SLinus Torvalds 1001da177e4SLinus Torvalds /* values for join in do_journal_begin_r */ 1011da177e4SLinus Torvalds enum { 1021da177e4SLinus Torvalds JBEGIN_REG = 0, /* regular journal begin */ 103098297b2SJeff Mahoney /* join the running transaction if at all possible */ 104098297b2SJeff Mahoney JBEGIN_JOIN = 1, 105098297b2SJeff Mahoney /* called from cleanup code, ignores aborted flag */ 106098297b2SJeff Mahoney JBEGIN_ABORT = 2, 1071da177e4SLinus Torvalds }; 1081da177e4SLinus Torvalds 1091da177e4SLinus Torvalds static int do_journal_begin_r(struct reiserfs_transaction_handle *th, 110a9dd3643SJeff Mahoney struct super_block *sb, 1111da177e4SLinus Torvalds unsigned long nblocks, int join); 1121da177e4SLinus Torvalds 113a9dd3643SJeff Mahoney static void init_journal_hash(struct super_block *sb) 114bd4c625cSLinus Torvalds { 115a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 116bd4c625cSLinus Torvalds memset(journal->j_hash_table, 0, 117bd4c625cSLinus Torvalds JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *)); 1181da177e4SLinus Torvalds } 1191da177e4SLinus Torvalds 1201da177e4SLinus Torvalds /* 121098297b2SJeff Mahoney * clears BH_Dirty and sticks the buffer on the clean list. Called because 122098297b2SJeff Mahoney * I can't allow refile_buffer to make schedule happen after I've freed a 123098297b2SJeff Mahoney * block. Look at remove_from_transaction and journal_mark_freed for 124098297b2SJeff Mahoney * more details. 1251da177e4SLinus Torvalds */ 126bd4c625cSLinus Torvalds static int reiserfs_clean_and_file_buffer(struct buffer_head *bh) 127bd4c625cSLinus Torvalds { 1281da177e4SLinus Torvalds if (bh) { 1291da177e4SLinus Torvalds clear_buffer_dirty(bh); 1301da177e4SLinus Torvalds clear_buffer_journal_test(bh); 1311da177e4SLinus Torvalds } 1321da177e4SLinus Torvalds return 0; 1331da177e4SLinus Torvalds } 1341da177e4SLinus Torvalds 135bd4c625cSLinus Torvalds static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block 136a9dd3643SJeff Mahoney *sb) 137bd4c625cSLinus Torvalds { 1381da177e4SLinus Torvalds struct reiserfs_bitmap_node *bn; 1391da177e4SLinus Torvalds static int id; 1401da177e4SLinus Torvalds 141d739b42bSPekka Enberg bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS); 1421da177e4SLinus Torvalds if (!bn) { 1431da177e4SLinus Torvalds return NULL; 1441da177e4SLinus Torvalds } 145a9dd3643SJeff Mahoney bn->data = kzalloc(sb->s_blocksize, GFP_NOFS); 1461da177e4SLinus Torvalds if (!bn->data) { 147d739b42bSPekka Enberg kfree(bn); 1481da177e4SLinus Torvalds return NULL; 1491da177e4SLinus Torvalds } 1501da177e4SLinus Torvalds bn->id = id++; 1511da177e4SLinus Torvalds INIT_LIST_HEAD(&bn->list); 1521da177e4SLinus Torvalds return bn; 1531da177e4SLinus Torvalds } 1541da177e4SLinus Torvalds 155a9dd3643SJeff Mahoney static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *sb) 156bd4c625cSLinus Torvalds { 157a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 1581da177e4SLinus Torvalds struct reiserfs_bitmap_node *bn = NULL; 1591da177e4SLinus Torvalds struct list_head *entry = journal->j_bitmap_nodes.next; 1601da177e4SLinus Torvalds 1611da177e4SLinus Torvalds journal->j_used_bitmap_nodes++; 1621da177e4SLinus Torvalds repeat: 1631da177e4SLinus Torvalds 1641da177e4SLinus Torvalds if (entry != &journal->j_bitmap_nodes) { 1651da177e4SLinus Torvalds bn = list_entry(entry, struct reiserfs_bitmap_node, list); 1661da177e4SLinus Torvalds list_del(entry); 167a9dd3643SJeff Mahoney memset(bn->data, 0, sb->s_blocksize); 1681da177e4SLinus Torvalds journal->j_free_bitmap_nodes--; 1691da177e4SLinus Torvalds return bn; 1701da177e4SLinus Torvalds } 171a9dd3643SJeff Mahoney bn = allocate_bitmap_node(sb); 1721da177e4SLinus Torvalds if (!bn) { 1731da177e4SLinus Torvalds yield(); 1741da177e4SLinus Torvalds goto repeat; 1751da177e4SLinus Torvalds } 1761da177e4SLinus Torvalds return bn; 1771da177e4SLinus Torvalds } 178a9dd3643SJeff Mahoney static inline void free_bitmap_node(struct super_block *sb, 179bd4c625cSLinus Torvalds struct reiserfs_bitmap_node *bn) 180bd4c625cSLinus Torvalds { 181a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 1821da177e4SLinus Torvalds journal->j_used_bitmap_nodes--; 1831da177e4SLinus Torvalds if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) { 184d739b42bSPekka Enberg kfree(bn->data); 185d739b42bSPekka Enberg kfree(bn); 1861da177e4SLinus Torvalds } else { 1871da177e4SLinus Torvalds list_add(&bn->list, &journal->j_bitmap_nodes); 1881da177e4SLinus Torvalds journal->j_free_bitmap_nodes++; 1891da177e4SLinus Torvalds } 1901da177e4SLinus Torvalds } 1911da177e4SLinus Torvalds 192a9dd3643SJeff Mahoney static void allocate_bitmap_nodes(struct super_block *sb) 193bd4c625cSLinus Torvalds { 1941da177e4SLinus Torvalds int i; 195a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 1961da177e4SLinus Torvalds struct reiserfs_bitmap_node *bn = NULL; 1971da177e4SLinus Torvalds for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) { 198a9dd3643SJeff Mahoney bn = allocate_bitmap_node(sb); 1991da177e4SLinus Torvalds if (bn) { 2001da177e4SLinus Torvalds list_add(&bn->list, &journal->j_bitmap_nodes); 2011da177e4SLinus Torvalds journal->j_free_bitmap_nodes++; 2021da177e4SLinus Torvalds } else { 203098297b2SJeff Mahoney /* this is ok, we'll try again when more are needed */ 204098297b2SJeff Mahoney break; 2051da177e4SLinus Torvalds } 2061da177e4SLinus Torvalds } 2071da177e4SLinus Torvalds } 2081da177e4SLinus Torvalds 209a9dd3643SJeff Mahoney static int set_bit_in_list_bitmap(struct super_block *sb, 2103ee16670SJeff Mahoney b_blocknr_t block, 211bd4c625cSLinus Torvalds struct reiserfs_list_bitmap *jb) 212bd4c625cSLinus Torvalds { 213a9dd3643SJeff Mahoney unsigned int bmap_nr = block / (sb->s_blocksize << 3); 214a9dd3643SJeff Mahoney unsigned int bit_nr = block % (sb->s_blocksize << 3); 2151da177e4SLinus Torvalds 2161da177e4SLinus Torvalds if (!jb->bitmaps[bmap_nr]) { 217a9dd3643SJeff Mahoney jb->bitmaps[bmap_nr] = get_bitmap_node(sb); 2181da177e4SLinus Torvalds } 2191da177e4SLinus Torvalds set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data); 2201da177e4SLinus Torvalds return 0; 2211da177e4SLinus Torvalds } 2221da177e4SLinus Torvalds 223a9dd3643SJeff Mahoney static void cleanup_bitmap_list(struct super_block *sb, 224bd4c625cSLinus Torvalds struct reiserfs_list_bitmap *jb) 225bd4c625cSLinus Torvalds { 2261da177e4SLinus Torvalds int i; 2271da177e4SLinus Torvalds if (jb->bitmaps == NULL) 2281da177e4SLinus Torvalds return; 2291da177e4SLinus Torvalds 230a9dd3643SJeff Mahoney for (i = 0; i < reiserfs_bmap_count(sb); i++) { 2311da177e4SLinus Torvalds if (jb->bitmaps[i]) { 232a9dd3643SJeff Mahoney free_bitmap_node(sb, jb->bitmaps[i]); 2331da177e4SLinus Torvalds jb->bitmaps[i] = NULL; 2341da177e4SLinus Torvalds } 2351da177e4SLinus Torvalds } 2361da177e4SLinus Torvalds } 2371da177e4SLinus Torvalds 2381da177e4SLinus Torvalds /* 239098297b2SJeff Mahoney * only call this on FS unmount. 2401da177e4SLinus Torvalds */ 241a9dd3643SJeff Mahoney static int free_list_bitmaps(struct super_block *sb, 242bd4c625cSLinus Torvalds struct reiserfs_list_bitmap *jb_array) 243bd4c625cSLinus Torvalds { 2441da177e4SLinus Torvalds int i; 2451da177e4SLinus Torvalds struct reiserfs_list_bitmap *jb; 2461da177e4SLinus Torvalds for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) { 2471da177e4SLinus Torvalds jb = jb_array + i; 2481da177e4SLinus Torvalds jb->journal_list = NULL; 249a9dd3643SJeff Mahoney cleanup_bitmap_list(sb, jb); 2501da177e4SLinus Torvalds vfree(jb->bitmaps); 2511da177e4SLinus Torvalds jb->bitmaps = NULL; 2521da177e4SLinus Torvalds } 2531da177e4SLinus Torvalds return 0; 2541da177e4SLinus Torvalds } 2551da177e4SLinus Torvalds 256a9dd3643SJeff Mahoney static int free_bitmap_nodes(struct super_block *sb) 257bd4c625cSLinus Torvalds { 258a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 2591da177e4SLinus Torvalds struct list_head *next = journal->j_bitmap_nodes.next; 2601da177e4SLinus Torvalds struct reiserfs_bitmap_node *bn; 2611da177e4SLinus Torvalds 2621da177e4SLinus Torvalds while (next != &journal->j_bitmap_nodes) { 2631da177e4SLinus Torvalds bn = list_entry(next, struct reiserfs_bitmap_node, list); 2641da177e4SLinus Torvalds list_del(next); 265d739b42bSPekka Enberg kfree(bn->data); 266d739b42bSPekka Enberg kfree(bn); 2671da177e4SLinus Torvalds next = journal->j_bitmap_nodes.next; 2681da177e4SLinus Torvalds journal->j_free_bitmap_nodes--; 2691da177e4SLinus Torvalds } 2701da177e4SLinus Torvalds 2711da177e4SLinus Torvalds return 0; 2721da177e4SLinus Torvalds } 2731da177e4SLinus Torvalds 2741da177e4SLinus Torvalds /* 275098297b2SJeff Mahoney * get memory for JOURNAL_NUM_BITMAPS worth of bitmaps. 276098297b2SJeff Mahoney * jb_array is the array to be filled in. 2771da177e4SLinus Torvalds */ 278a9dd3643SJeff Mahoney int reiserfs_allocate_list_bitmaps(struct super_block *sb, 2791da177e4SLinus Torvalds struct reiserfs_list_bitmap *jb_array, 2803ee16670SJeff Mahoney unsigned int bmap_nr) 281bd4c625cSLinus Torvalds { 2821da177e4SLinus Torvalds int i; 2831da177e4SLinus Torvalds int failed = 0; 2841da177e4SLinus Torvalds struct reiserfs_list_bitmap *jb; 2851da177e4SLinus Torvalds int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *); 2861da177e4SLinus Torvalds 2871da177e4SLinus Torvalds for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) { 2881da177e4SLinus Torvalds jb = jb_array + i; 2891da177e4SLinus Torvalds jb->journal_list = NULL; 290558feb08SJoe Perches jb->bitmaps = vzalloc(mem); 2911da177e4SLinus Torvalds if (!jb->bitmaps) { 292a9dd3643SJeff Mahoney reiserfs_warning(sb, "clm-2000", "unable to " 29345b03d5eSJeff Mahoney "allocate bitmaps for journal lists"); 2941da177e4SLinus Torvalds failed = 1; 2951da177e4SLinus Torvalds break; 2961da177e4SLinus Torvalds } 2971da177e4SLinus Torvalds } 2981da177e4SLinus Torvalds if (failed) { 299a9dd3643SJeff Mahoney free_list_bitmaps(sb, jb_array); 3001da177e4SLinus Torvalds return -1; 3011da177e4SLinus Torvalds } 3021da177e4SLinus Torvalds return 0; 3031da177e4SLinus Torvalds } 3041da177e4SLinus Torvalds 3051da177e4SLinus Torvalds /* 306098297b2SJeff Mahoney * find an available list bitmap. If you can't find one, flush a commit list 307098297b2SJeff Mahoney * and try again 3081da177e4SLinus Torvalds */ 309a9dd3643SJeff Mahoney static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *sb, 310bd4c625cSLinus Torvalds struct reiserfs_journal_list 311bd4c625cSLinus Torvalds *jl) 312bd4c625cSLinus Torvalds { 3131da177e4SLinus Torvalds int i, j; 314a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 3151da177e4SLinus Torvalds struct reiserfs_list_bitmap *jb = NULL; 3161da177e4SLinus Torvalds 3171da177e4SLinus Torvalds for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) { 3181da177e4SLinus Torvalds i = journal->j_list_bitmap_index; 3191da177e4SLinus Torvalds journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS; 3201da177e4SLinus Torvalds jb = journal->j_list_bitmap + i; 3211da177e4SLinus Torvalds if (journal->j_list_bitmap[i].journal_list) { 322a9dd3643SJeff Mahoney flush_commit_list(sb, 323bd4c625cSLinus Torvalds journal->j_list_bitmap[i]. 324bd4c625cSLinus Torvalds journal_list, 1); 3251da177e4SLinus Torvalds if (!journal->j_list_bitmap[i].journal_list) { 3261da177e4SLinus Torvalds break; 3271da177e4SLinus Torvalds } 3281da177e4SLinus Torvalds } else { 3291da177e4SLinus Torvalds break; 3301da177e4SLinus Torvalds } 3311da177e4SLinus Torvalds } 332098297b2SJeff Mahoney /* double check to make sure if flushed correctly */ 333098297b2SJeff Mahoney if (jb->journal_list) 3341da177e4SLinus Torvalds return NULL; 3351da177e4SLinus Torvalds jb->journal_list = jl; 3361da177e4SLinus Torvalds return jb; 3371da177e4SLinus Torvalds } 3381da177e4SLinus Torvalds 3391da177e4SLinus Torvalds /* 340098297b2SJeff Mahoney * allocates a new chunk of X nodes, and links them all together as a list. 341098297b2SJeff Mahoney * Uses the cnode->next and cnode->prev pointers 342098297b2SJeff Mahoney * returns NULL on failure 3431da177e4SLinus Torvalds */ 344bd4c625cSLinus Torvalds static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes) 345bd4c625cSLinus Torvalds { 3461da177e4SLinus Torvalds struct reiserfs_journal_cnode *head; 3471da177e4SLinus Torvalds int i; 3481da177e4SLinus Torvalds if (num_cnodes <= 0) { 3491da177e4SLinus Torvalds return NULL; 3501da177e4SLinus Torvalds } 351fad953ceSKees Cook head = vzalloc(array_size(num_cnodes, 352fad953ceSKees Cook sizeof(struct reiserfs_journal_cnode))); 3531da177e4SLinus Torvalds if (!head) { 3541da177e4SLinus Torvalds return NULL; 3551da177e4SLinus Torvalds } 3561da177e4SLinus Torvalds head[0].prev = NULL; 3571da177e4SLinus Torvalds head[0].next = head + 1; 3581da177e4SLinus Torvalds for (i = 1; i < num_cnodes; i++) { 3591da177e4SLinus Torvalds head[i].prev = head + (i - 1); 3601da177e4SLinus Torvalds head[i].next = head + (i + 1); /* if last one, overwrite it after the if */ 3611da177e4SLinus Torvalds } 3621da177e4SLinus Torvalds head[num_cnodes - 1].next = NULL; 3631da177e4SLinus Torvalds return head; 3641da177e4SLinus Torvalds } 3651da177e4SLinus Torvalds 366098297b2SJeff Mahoney /* pulls a cnode off the free list, or returns NULL on failure */ 367a9dd3643SJeff Mahoney static struct reiserfs_journal_cnode *get_cnode(struct super_block *sb) 368bd4c625cSLinus Torvalds { 3691da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 370a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 3711da177e4SLinus Torvalds 372a9dd3643SJeff Mahoney reiserfs_check_lock_depth(sb, "get_cnode"); 3731da177e4SLinus Torvalds 3741da177e4SLinus Torvalds if (journal->j_cnode_free <= 0) { 3751da177e4SLinus Torvalds return NULL; 3761da177e4SLinus Torvalds } 3771da177e4SLinus Torvalds journal->j_cnode_used++; 3781da177e4SLinus Torvalds journal->j_cnode_free--; 3791da177e4SLinus Torvalds cn = journal->j_cnode_free_list; 3801da177e4SLinus Torvalds if (!cn) { 3811da177e4SLinus Torvalds return cn; 3821da177e4SLinus Torvalds } 3831da177e4SLinus Torvalds if (cn->next) { 3841da177e4SLinus Torvalds cn->next->prev = NULL; 3851da177e4SLinus Torvalds } 3861da177e4SLinus Torvalds journal->j_cnode_free_list = cn->next; 3871da177e4SLinus Torvalds memset(cn, 0, sizeof(struct reiserfs_journal_cnode)); 3881da177e4SLinus Torvalds return cn; 3891da177e4SLinus Torvalds } 3901da177e4SLinus Torvalds 3911da177e4SLinus Torvalds /* 392098297b2SJeff Mahoney * returns a cnode to the free list 3931da177e4SLinus Torvalds */ 394a9dd3643SJeff Mahoney static void free_cnode(struct super_block *sb, 395bd4c625cSLinus Torvalds struct reiserfs_journal_cnode *cn) 396bd4c625cSLinus Torvalds { 397a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 3981da177e4SLinus Torvalds 399a9dd3643SJeff Mahoney reiserfs_check_lock_depth(sb, "free_cnode"); 4001da177e4SLinus Torvalds 4011da177e4SLinus Torvalds journal->j_cnode_used--; 4021da177e4SLinus Torvalds journal->j_cnode_free++; 4031da177e4SLinus Torvalds /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */ 4041da177e4SLinus Torvalds cn->next = journal->j_cnode_free_list; 4051da177e4SLinus Torvalds if (journal->j_cnode_free_list) { 4061da177e4SLinus Torvalds journal->j_cnode_free_list->prev = cn; 4071da177e4SLinus Torvalds } 4081da177e4SLinus Torvalds cn->prev = NULL; /* not needed with the memset, but I might kill the memset, and forget to do this */ 4091da177e4SLinus Torvalds journal->j_cnode_free_list = cn; 4101da177e4SLinus Torvalds } 4111da177e4SLinus Torvalds 412bd4c625cSLinus Torvalds static void clear_prepared_bits(struct buffer_head *bh) 413bd4c625cSLinus Torvalds { 4141da177e4SLinus Torvalds clear_buffer_journal_prepared(bh); 4151da177e4SLinus Torvalds clear_buffer_journal_restore_dirty(bh); 4161da177e4SLinus Torvalds } 4171da177e4SLinus Torvalds 418098297b2SJeff Mahoney /* 419098297b2SJeff Mahoney * return a cnode with same dev, block number and size in table, 420098297b2SJeff Mahoney * or null if not found 421098297b2SJeff Mahoney */ 422bd4c625cSLinus Torvalds static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct 423bd4c625cSLinus Torvalds super_block 424bd4c625cSLinus Torvalds *sb, 425bd4c625cSLinus Torvalds struct 426bd4c625cSLinus Torvalds reiserfs_journal_cnode 427bd4c625cSLinus Torvalds **table, 4281da177e4SLinus Torvalds long bl) 4291da177e4SLinus Torvalds { 4301da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 4311da177e4SLinus Torvalds cn = journal_hash(table, sb, bl); 4321da177e4SLinus Torvalds while (cn) { 4331da177e4SLinus Torvalds if (cn->blocknr == bl && cn->sb == sb) 4341da177e4SLinus Torvalds return cn; 4351da177e4SLinus Torvalds cn = cn->hnext; 4361da177e4SLinus Torvalds } 4371da177e4SLinus Torvalds return (struct reiserfs_journal_cnode *)0; 4381da177e4SLinus Torvalds } 4391da177e4SLinus Torvalds 4401da177e4SLinus Torvalds /* 441098297b2SJeff Mahoney * this actually means 'can this block be reallocated yet?'. If you set 442098297b2SJeff Mahoney * search_all, a block can only be allocated if it is not in the current 443098297b2SJeff Mahoney * transaction, was not freed by the current transaction, and has no chance 444098297b2SJeff Mahoney * of ever being overwritten by a replay after crashing. 445098297b2SJeff Mahoney * 446098297b2SJeff Mahoney * If you don't set search_all, a block can only be allocated if it is not 447098297b2SJeff Mahoney * in the current transaction. Since deleting a block removes it from the 448098297b2SJeff Mahoney * current transaction, this case should never happen. If you don't set 449098297b2SJeff Mahoney * search_all, make sure you never write the block without logging it. 450098297b2SJeff Mahoney * 451098297b2SJeff Mahoney * next_zero_bit is a suggestion about the next block to try for find_forward. 452098297b2SJeff Mahoney * when bl is rejected because it is set in a journal list bitmap, we search 453098297b2SJeff Mahoney * for the next zero bit in the bitmap that rejected bl. Then, we return 454098297b2SJeff Mahoney * that through next_zero_bit for find_forward to try. 455098297b2SJeff Mahoney * 456098297b2SJeff Mahoney * Just because we return something in next_zero_bit does not mean we won't 457098297b2SJeff Mahoney * reject it on the next call to reiserfs_in_journal 4581da177e4SLinus Torvalds */ 459a9dd3643SJeff Mahoney int reiserfs_in_journal(struct super_block *sb, 4603ee16670SJeff Mahoney unsigned int bmap_nr, int bit_nr, int search_all, 461bd4c625cSLinus Torvalds b_blocknr_t * next_zero_bit) 462bd4c625cSLinus Torvalds { 463a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 4641da177e4SLinus Torvalds struct reiserfs_list_bitmap *jb; 4651da177e4SLinus Torvalds int i; 4661da177e4SLinus Torvalds unsigned long bl; 4671da177e4SLinus Torvalds 4681da177e4SLinus Torvalds *next_zero_bit = 0; /* always start this at zero. */ 4691da177e4SLinus Torvalds 470a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.in_journal); 471098297b2SJeff Mahoney /* 472098297b2SJeff Mahoney * If we aren't doing a search_all, this is a metablock, and it 473098297b2SJeff Mahoney * will be logged before use. if we crash before the transaction 474098297b2SJeff Mahoney * that freed it commits, this transaction won't have committed 475098297b2SJeff Mahoney * either, and the block will never be written 4761da177e4SLinus Torvalds */ 4771da177e4SLinus Torvalds if (search_all) { 4781da177e4SLinus Torvalds for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) { 479a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.in_journal_bitmap); 4801da177e4SLinus Torvalds jb = journal->j_list_bitmap + i; 4811da177e4SLinus Torvalds if (jb->journal_list && jb->bitmaps[bmap_nr] && 482bd4c625cSLinus Torvalds test_bit(bit_nr, 483bd4c625cSLinus Torvalds (unsigned long *)jb->bitmaps[bmap_nr]-> 484bd4c625cSLinus Torvalds data)) { 485bd4c625cSLinus Torvalds *next_zero_bit = 486bd4c625cSLinus Torvalds find_next_zero_bit((unsigned long *) 487bd4c625cSLinus Torvalds (jb->bitmaps[bmap_nr]-> 488bd4c625cSLinus Torvalds data), 489a9dd3643SJeff Mahoney sb->s_blocksize << 3, 490bd4c625cSLinus Torvalds bit_nr + 1); 4911da177e4SLinus Torvalds return 1; 4921da177e4SLinus Torvalds } 4931da177e4SLinus Torvalds } 4941da177e4SLinus Torvalds } 4951da177e4SLinus Torvalds 496a9dd3643SJeff Mahoney bl = bmap_nr * (sb->s_blocksize << 3) + bit_nr; 4971da177e4SLinus Torvalds /* is it in any old transactions? */ 498bd4c625cSLinus Torvalds if (search_all 499*a3cc754aSTian Tao && (get_journal_hash_dev(sb, journal->j_list_hash_table, bl))) { 5001da177e4SLinus Torvalds return 1; 5011da177e4SLinus Torvalds } 5021da177e4SLinus Torvalds 5031da177e4SLinus Torvalds /* is it in the current transaction. This should never happen */ 504*a3cc754aSTian Tao if ((get_journal_hash_dev(sb, journal->j_hash_table, bl))) { 5051da177e4SLinus Torvalds BUG(); 5061da177e4SLinus Torvalds return 1; 5071da177e4SLinus Torvalds } 5081da177e4SLinus Torvalds 509a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.in_journal_reusable); 5101da177e4SLinus Torvalds /* safe for reuse */ 5111da177e4SLinus Torvalds return 0; 5121da177e4SLinus Torvalds } 5131da177e4SLinus Torvalds 514098297b2SJeff Mahoney /* insert cn into table */ 515bd4c625cSLinus Torvalds static inline void insert_journal_hash(struct reiserfs_journal_cnode **table, 516bd4c625cSLinus Torvalds struct reiserfs_journal_cnode *cn) 517bd4c625cSLinus Torvalds { 5181da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn_orig; 5191da177e4SLinus Torvalds 5201da177e4SLinus Torvalds cn_orig = journal_hash(table, cn->sb, cn->blocknr); 5211da177e4SLinus Torvalds cn->hnext = cn_orig; 5221da177e4SLinus Torvalds cn->hprev = NULL; 5231da177e4SLinus Torvalds if (cn_orig) { 5241da177e4SLinus Torvalds cn_orig->hprev = cn; 5251da177e4SLinus Torvalds } 5261da177e4SLinus Torvalds journal_hash(table, cn->sb, cn->blocknr) = cn; 5271da177e4SLinus Torvalds } 5281da177e4SLinus Torvalds 5291da177e4SLinus Torvalds /* lock the current transaction */ 530a9dd3643SJeff Mahoney static inline void lock_journal(struct super_block *sb) 531bd4c625cSLinus Torvalds { 532a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.lock_journal); 5338ebc4232SFrederic Weisbecker 5348ebc4232SFrederic Weisbecker reiserfs_mutex_lock_safe(&SB_JOURNAL(sb)->j_mutex, sb); 5351da177e4SLinus Torvalds } 5361da177e4SLinus Torvalds 5371da177e4SLinus Torvalds /* unlock the current transaction */ 538a9dd3643SJeff Mahoney static inline void unlock_journal(struct super_block *sb) 539bd4c625cSLinus Torvalds { 540a9dd3643SJeff Mahoney mutex_unlock(&SB_JOURNAL(sb)->j_mutex); 5411da177e4SLinus Torvalds } 5421da177e4SLinus Torvalds 5431da177e4SLinus Torvalds static inline void get_journal_list(struct reiserfs_journal_list *jl) 5441da177e4SLinus Torvalds { 5451da177e4SLinus Torvalds jl->j_refcount++; 5461da177e4SLinus Torvalds } 5471da177e4SLinus Torvalds 5481da177e4SLinus Torvalds static inline void put_journal_list(struct super_block *s, 5491da177e4SLinus Torvalds struct reiserfs_journal_list *jl) 5501da177e4SLinus Torvalds { 5511da177e4SLinus Torvalds if (jl->j_refcount < 1) { 552c3a9c210SJeff Mahoney reiserfs_panic(s, "journal-2", "trans id %u, refcount at %d", 553bd4c625cSLinus Torvalds jl->j_trans_id, jl->j_refcount); 5541da177e4SLinus Torvalds } 5551da177e4SLinus Torvalds if (--jl->j_refcount == 0) 556d739b42bSPekka Enberg kfree(jl); 5571da177e4SLinus Torvalds } 5581da177e4SLinus Torvalds 5591da177e4SLinus Torvalds /* 560098297b2SJeff Mahoney * this used to be much more involved, and I'm keeping it just in case 561098297b2SJeff Mahoney * things get ugly again. it gets called by flush_commit_list, and 562098297b2SJeff Mahoney * cleans up any data stored about blocks freed during a transaction. 5631da177e4SLinus Torvalds */ 564a9dd3643SJeff Mahoney static void cleanup_freed_for_journal_list(struct super_block *sb, 565bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl) 566bd4c625cSLinus Torvalds { 5671da177e4SLinus Torvalds 5681da177e4SLinus Torvalds struct reiserfs_list_bitmap *jb = jl->j_list_bitmap; 5691da177e4SLinus Torvalds if (jb) { 570a9dd3643SJeff Mahoney cleanup_bitmap_list(sb, jb); 5711da177e4SLinus Torvalds } 5721da177e4SLinus Torvalds jl->j_list_bitmap->journal_list = NULL; 5731da177e4SLinus Torvalds jl->j_list_bitmap = NULL; 5741da177e4SLinus Torvalds } 5751da177e4SLinus Torvalds 5761da177e4SLinus Torvalds static int journal_list_still_alive(struct super_block *s, 577600ed416SJeff Mahoney unsigned int trans_id) 5781da177e4SLinus Torvalds { 5791da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 5801da177e4SLinus Torvalds struct list_head *entry = &journal->j_journal_list; 5811da177e4SLinus Torvalds struct reiserfs_journal_list *jl; 5821da177e4SLinus Torvalds 5831da177e4SLinus Torvalds if (!list_empty(entry)) { 5841da177e4SLinus Torvalds jl = JOURNAL_LIST_ENTRY(entry->next); 5851da177e4SLinus Torvalds if (jl->j_trans_id <= trans_id) { 5861da177e4SLinus Torvalds return 1; 5871da177e4SLinus Torvalds } 5881da177e4SLinus Torvalds } 5891da177e4SLinus Torvalds return 0; 5901da177e4SLinus Torvalds } 5911da177e4SLinus Torvalds 592398c95bdSChris Mason /* 593398c95bdSChris Mason * If page->mapping was null, we failed to truncate this page for 594398c95bdSChris Mason * some reason. Most likely because it was truncated after being 595398c95bdSChris Mason * logged via data=journal. 596398c95bdSChris Mason * 597398c95bdSChris Mason * This does a check to see if the buffer belongs to one of these 598398c95bdSChris Mason * lost pages before doing the final put_bh. If page->mapping was 599398c95bdSChris Mason * null, it tries to free buffers on the page, which should make the 600ea1754a0SKirill A. Shutemov * final put_page drop the page from the lru. 601398c95bdSChris Mason */ 602398c95bdSChris Mason static void release_buffer_page(struct buffer_head *bh) 603398c95bdSChris Mason { 604398c95bdSChris Mason struct page *page = bh->b_page; 605529ae9aaSNick Piggin if (!page->mapping && trylock_page(page)) { 60609cbfeafSKirill A. Shutemov get_page(page); 607398c95bdSChris Mason put_bh(bh); 608398c95bdSChris Mason if (!page->mapping) 609398c95bdSChris Mason try_to_free_buffers(page); 610398c95bdSChris Mason unlock_page(page); 61109cbfeafSKirill A. Shutemov put_page(page); 612398c95bdSChris Mason } else { 613398c95bdSChris Mason put_bh(bh); 614398c95bdSChris Mason } 615398c95bdSChris Mason } 616398c95bdSChris Mason 617bd4c625cSLinus Torvalds static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate) 618bd4c625cSLinus Torvalds { 6191da177e4SLinus Torvalds if (buffer_journaled(bh)) { 62045b03d5eSJeff Mahoney reiserfs_warning(NULL, "clm-2084", 621a1c6f057SDmitry Monakhov "pinned buffer %lu:%pg sent to disk", 622a1c6f057SDmitry Monakhov bh->b_blocknr, bh->b_bdev); 6231da177e4SLinus Torvalds } 6241da177e4SLinus Torvalds if (uptodate) 6251da177e4SLinus Torvalds set_buffer_uptodate(bh); 6261da177e4SLinus Torvalds else 6271da177e4SLinus Torvalds clear_buffer_uptodate(bh); 628398c95bdSChris Mason 6291da177e4SLinus Torvalds unlock_buffer(bh); 630398c95bdSChris Mason release_buffer_page(bh); 6311da177e4SLinus Torvalds } 6321da177e4SLinus Torvalds 633bd4c625cSLinus Torvalds static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate) 634bd4c625cSLinus Torvalds { 6351da177e4SLinus Torvalds if (uptodate) 6361da177e4SLinus Torvalds set_buffer_uptodate(bh); 6371da177e4SLinus Torvalds else 6381da177e4SLinus Torvalds clear_buffer_uptodate(bh); 6391da177e4SLinus Torvalds unlock_buffer(bh); 6401da177e4SLinus Torvalds put_bh(bh); 6411da177e4SLinus Torvalds } 6421da177e4SLinus Torvalds 643bd4c625cSLinus Torvalds static void submit_logged_buffer(struct buffer_head *bh) 644bd4c625cSLinus Torvalds { 6451da177e4SLinus Torvalds get_bh(bh); 6461da177e4SLinus Torvalds bh->b_end_io = reiserfs_end_buffer_io_sync; 6471da177e4SLinus Torvalds clear_buffer_journal_new(bh); 6481da177e4SLinus Torvalds clear_buffer_dirty(bh); 6491da177e4SLinus Torvalds if (!test_clear_buffer_journal_test(bh)) 6501da177e4SLinus Torvalds BUG(); 6511da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 6521da177e4SLinus Torvalds BUG(); 6532a222ca9SMike Christie submit_bh(REQ_OP_WRITE, 0, bh); 6541da177e4SLinus Torvalds } 6551da177e4SLinus Torvalds 656bd4c625cSLinus Torvalds static void submit_ordered_buffer(struct buffer_head *bh) 657bd4c625cSLinus Torvalds { 6581da177e4SLinus Torvalds get_bh(bh); 6591da177e4SLinus Torvalds bh->b_end_io = reiserfs_end_ordered_io; 6601da177e4SLinus Torvalds clear_buffer_dirty(bh); 6611da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 6621da177e4SLinus Torvalds BUG(); 6632a222ca9SMike Christie submit_bh(REQ_OP_WRITE, 0, bh); 6641da177e4SLinus Torvalds } 6651da177e4SLinus Torvalds 6661da177e4SLinus Torvalds #define CHUNK_SIZE 32 6671da177e4SLinus Torvalds struct buffer_chunk { 6681da177e4SLinus Torvalds struct buffer_head *bh[CHUNK_SIZE]; 6691da177e4SLinus Torvalds int nr; 6701da177e4SLinus Torvalds }; 6711da177e4SLinus Torvalds 672bd4c625cSLinus Torvalds static void write_chunk(struct buffer_chunk *chunk) 673bd4c625cSLinus Torvalds { 6741da177e4SLinus Torvalds int i; 6751da177e4SLinus Torvalds for (i = 0; i < chunk->nr; i++) { 6761da177e4SLinus Torvalds submit_logged_buffer(chunk->bh[i]); 6771da177e4SLinus Torvalds } 6781da177e4SLinus Torvalds chunk->nr = 0; 6791da177e4SLinus Torvalds } 6801da177e4SLinus Torvalds 681bd4c625cSLinus Torvalds static void write_ordered_chunk(struct buffer_chunk *chunk) 682bd4c625cSLinus Torvalds { 6831da177e4SLinus Torvalds int i; 6841da177e4SLinus Torvalds for (i = 0; i < chunk->nr; i++) { 6851da177e4SLinus Torvalds submit_ordered_buffer(chunk->bh[i]); 6861da177e4SLinus Torvalds } 6871da177e4SLinus Torvalds chunk->nr = 0; 6881da177e4SLinus Torvalds } 6891da177e4SLinus Torvalds 6901da177e4SLinus Torvalds static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh, 691bd4c625cSLinus Torvalds spinlock_t * lock, void (fn) (struct buffer_chunk *)) 6921da177e4SLinus Torvalds { 6931da177e4SLinus Torvalds int ret = 0; 69414a61442SEric Sesterhenn BUG_ON(chunk->nr >= CHUNK_SIZE); 6951da177e4SLinus Torvalds chunk->bh[chunk->nr++] = bh; 6961da177e4SLinus Torvalds if (chunk->nr >= CHUNK_SIZE) { 6971da177e4SLinus Torvalds ret = 1; 69854cc6ceaSFabian Frederick if (lock) { 6991da177e4SLinus Torvalds spin_unlock(lock); 7001da177e4SLinus Torvalds fn(chunk); 7011da177e4SLinus Torvalds spin_lock(lock); 70254cc6ceaSFabian Frederick } else { 70354cc6ceaSFabian Frederick fn(chunk); 70454cc6ceaSFabian Frederick } 7051da177e4SLinus Torvalds } 7061da177e4SLinus Torvalds return ret; 7071da177e4SLinus Torvalds } 7081da177e4SLinus Torvalds 7091da177e4SLinus Torvalds static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0); 710bd4c625cSLinus Torvalds static struct reiserfs_jh *alloc_jh(void) 711bd4c625cSLinus Torvalds { 7121da177e4SLinus Torvalds struct reiserfs_jh *jh; 7131da177e4SLinus Torvalds while (1) { 7141da177e4SLinus Torvalds jh = kmalloc(sizeof(*jh), GFP_NOFS); 7151da177e4SLinus Torvalds if (jh) { 7161da177e4SLinus Torvalds atomic_inc(&nr_reiserfs_jh); 7171da177e4SLinus Torvalds return jh; 7181da177e4SLinus Torvalds } 7191da177e4SLinus Torvalds yield(); 7201da177e4SLinus Torvalds } 7211da177e4SLinus Torvalds } 7221da177e4SLinus Torvalds 7231da177e4SLinus Torvalds /* 7241da177e4SLinus Torvalds * we want to free the jh when the buffer has been written 7251da177e4SLinus Torvalds * and waited on 7261da177e4SLinus Torvalds */ 727bd4c625cSLinus Torvalds void reiserfs_free_jh(struct buffer_head *bh) 728bd4c625cSLinus Torvalds { 7291da177e4SLinus Torvalds struct reiserfs_jh *jh; 7301da177e4SLinus Torvalds 7311da177e4SLinus Torvalds jh = bh->b_private; 7321da177e4SLinus Torvalds if (jh) { 7331da177e4SLinus Torvalds bh->b_private = NULL; 7341da177e4SLinus Torvalds jh->bh = NULL; 7351da177e4SLinus Torvalds list_del_init(&jh->list); 7361da177e4SLinus Torvalds kfree(jh); 7371da177e4SLinus Torvalds if (atomic_read(&nr_reiserfs_jh) <= 0) 7381da177e4SLinus Torvalds BUG(); 7391da177e4SLinus Torvalds atomic_dec(&nr_reiserfs_jh); 7401da177e4SLinus Torvalds put_bh(bh); 7411da177e4SLinus Torvalds } 7421da177e4SLinus Torvalds } 7431da177e4SLinus Torvalds 7441da177e4SLinus Torvalds static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh, 7451da177e4SLinus Torvalds int tail) 7461da177e4SLinus Torvalds { 7471da177e4SLinus Torvalds struct reiserfs_jh *jh; 7481da177e4SLinus Torvalds 7491da177e4SLinus Torvalds if (bh->b_private) { 7501da177e4SLinus Torvalds spin_lock(&j->j_dirty_buffers_lock); 7511da177e4SLinus Torvalds if (!bh->b_private) { 7521da177e4SLinus Torvalds spin_unlock(&j->j_dirty_buffers_lock); 7531da177e4SLinus Torvalds goto no_jh; 7541da177e4SLinus Torvalds } 7551da177e4SLinus Torvalds jh = bh->b_private; 7561da177e4SLinus Torvalds list_del_init(&jh->list); 7571da177e4SLinus Torvalds } else { 7581da177e4SLinus Torvalds no_jh: 7591da177e4SLinus Torvalds get_bh(bh); 7601da177e4SLinus Torvalds jh = alloc_jh(); 7611da177e4SLinus Torvalds spin_lock(&j->j_dirty_buffers_lock); 762098297b2SJeff Mahoney /* 763098297b2SJeff Mahoney * buffer must be locked for __add_jh, should be able to have 7641da177e4SLinus Torvalds * two adds at the same time 7651da177e4SLinus Torvalds */ 76614a61442SEric Sesterhenn BUG_ON(bh->b_private); 7671da177e4SLinus Torvalds jh->bh = bh; 7681da177e4SLinus Torvalds bh->b_private = jh; 7691da177e4SLinus Torvalds } 7701da177e4SLinus Torvalds jh->jl = j->j_current_jl; 7711da177e4SLinus Torvalds if (tail) 7721da177e4SLinus Torvalds list_add_tail(&jh->list, &jh->jl->j_tail_bh_list); 7731da177e4SLinus Torvalds else { 7741da177e4SLinus Torvalds list_add_tail(&jh->list, &jh->jl->j_bh_list); 7751da177e4SLinus Torvalds } 7761da177e4SLinus Torvalds spin_unlock(&j->j_dirty_buffers_lock); 7771da177e4SLinus Torvalds return 0; 7781da177e4SLinus Torvalds } 7791da177e4SLinus Torvalds 780bd4c625cSLinus Torvalds int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh) 781bd4c625cSLinus Torvalds { 7821da177e4SLinus Torvalds return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1); 7831da177e4SLinus Torvalds } 784bd4c625cSLinus Torvalds int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh) 785bd4c625cSLinus Torvalds { 7861da177e4SLinus Torvalds return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0); 7871da177e4SLinus Torvalds } 7881da177e4SLinus Torvalds 7891da177e4SLinus Torvalds #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list) 7901da177e4SLinus Torvalds static int write_ordered_buffers(spinlock_t * lock, 7911da177e4SLinus Torvalds struct reiserfs_journal *j, 7921da177e4SLinus Torvalds struct reiserfs_journal_list *jl, 7931da177e4SLinus Torvalds struct list_head *list) 7941da177e4SLinus Torvalds { 7951da177e4SLinus Torvalds struct buffer_head *bh; 7961da177e4SLinus Torvalds struct reiserfs_jh *jh; 7971da177e4SLinus Torvalds int ret = j->j_errno; 7981da177e4SLinus Torvalds struct buffer_chunk chunk; 7991da177e4SLinus Torvalds struct list_head tmp; 8001da177e4SLinus Torvalds INIT_LIST_HEAD(&tmp); 8011da177e4SLinus Torvalds 8021da177e4SLinus Torvalds chunk.nr = 0; 8031da177e4SLinus Torvalds spin_lock(lock); 8041da177e4SLinus Torvalds while (!list_empty(list)) { 8051da177e4SLinus Torvalds jh = JH_ENTRY(list->next); 8061da177e4SLinus Torvalds bh = jh->bh; 8071da177e4SLinus Torvalds get_bh(bh); 808ca5de404SNick Piggin if (!trylock_buffer(bh)) { 8091da177e4SLinus Torvalds if (!buffer_dirty(bh)) { 810f116629dSAkinobu Mita list_move(&jh->list, &tmp); 8111da177e4SLinus Torvalds goto loop_next; 8121da177e4SLinus Torvalds } 8131da177e4SLinus Torvalds spin_unlock(lock); 8141da177e4SLinus Torvalds if (chunk.nr) 8151da177e4SLinus Torvalds write_ordered_chunk(&chunk); 8161da177e4SLinus Torvalds wait_on_buffer(bh); 8171da177e4SLinus Torvalds cond_resched(); 8181da177e4SLinus Torvalds spin_lock(lock); 8191da177e4SLinus Torvalds goto loop_next; 8201da177e4SLinus Torvalds } 821098297b2SJeff Mahoney /* 822098297b2SJeff Mahoney * in theory, dirty non-uptodate buffers should never get here, 8233d4492f8SChris Mason * but the upper layer io error paths still have a few quirks. 8243d4492f8SChris Mason * Handle them here as gracefully as we can 8253d4492f8SChris Mason */ 8263d4492f8SChris Mason if (!buffer_uptodate(bh) && buffer_dirty(bh)) { 8273d4492f8SChris Mason clear_buffer_dirty(bh); 8283d4492f8SChris Mason ret = -EIO; 8293d4492f8SChris Mason } 8301da177e4SLinus Torvalds if (buffer_dirty(bh)) { 831f116629dSAkinobu Mita list_move(&jh->list, &tmp); 8321da177e4SLinus Torvalds add_to_chunk(&chunk, bh, lock, write_ordered_chunk); 8331da177e4SLinus Torvalds } else { 8341da177e4SLinus Torvalds reiserfs_free_jh(bh); 8351da177e4SLinus Torvalds unlock_buffer(bh); 8361da177e4SLinus Torvalds } 8371da177e4SLinus Torvalds loop_next: 8381da177e4SLinus Torvalds put_bh(bh); 8391da177e4SLinus Torvalds cond_resched_lock(lock); 8401da177e4SLinus Torvalds } 8411da177e4SLinus Torvalds if (chunk.nr) { 8421da177e4SLinus Torvalds spin_unlock(lock); 8431da177e4SLinus Torvalds write_ordered_chunk(&chunk); 8441da177e4SLinus Torvalds spin_lock(lock); 8451da177e4SLinus Torvalds } 8461da177e4SLinus Torvalds while (!list_empty(&tmp)) { 8471da177e4SLinus Torvalds jh = JH_ENTRY(tmp.prev); 8481da177e4SLinus Torvalds bh = jh->bh; 8491da177e4SLinus Torvalds get_bh(bh); 8501da177e4SLinus Torvalds reiserfs_free_jh(bh); 8511da177e4SLinus Torvalds 8521da177e4SLinus Torvalds if (buffer_locked(bh)) { 8531da177e4SLinus Torvalds spin_unlock(lock); 8541da177e4SLinus Torvalds wait_on_buffer(bh); 8551da177e4SLinus Torvalds spin_lock(lock); 8561da177e4SLinus Torvalds } 8571da177e4SLinus Torvalds if (!buffer_uptodate(bh)) { 8581da177e4SLinus Torvalds ret = -EIO; 8591da177e4SLinus Torvalds } 860098297b2SJeff Mahoney /* 861098297b2SJeff Mahoney * ugly interaction with invalidatepage here. 862098297b2SJeff Mahoney * reiserfs_invalidate_page will pin any buffer that has a 863098297b2SJeff Mahoney * valid journal head from an older transaction. If someone 864098297b2SJeff Mahoney * else sets our buffer dirty after we write it in the first 865098297b2SJeff Mahoney * loop, and then someone truncates the page away, nobody 866098297b2SJeff Mahoney * will ever write the buffer. We're safe if we write the 867098297b2SJeff Mahoney * page one last time after freeing the journal header. 868d62b1b87SChris Mason */ 869d62b1b87SChris Mason if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) { 870d62b1b87SChris Mason spin_unlock(lock); 871dfec8a14SMike Christie ll_rw_block(REQ_OP_WRITE, 0, 1, &bh); 872d62b1b87SChris Mason spin_lock(lock); 873d62b1b87SChris Mason } 8741da177e4SLinus Torvalds put_bh(bh); 8751da177e4SLinus Torvalds cond_resched_lock(lock); 8761da177e4SLinus Torvalds } 8771da177e4SLinus Torvalds spin_unlock(lock); 8781da177e4SLinus Torvalds return ret; 8791da177e4SLinus Torvalds } 8801da177e4SLinus Torvalds 881bd4c625cSLinus Torvalds static int flush_older_commits(struct super_block *s, 882bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl) 883bd4c625cSLinus Torvalds { 8841da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 8851da177e4SLinus Torvalds struct reiserfs_journal_list *other_jl; 8861da177e4SLinus Torvalds struct reiserfs_journal_list *first_jl; 8871da177e4SLinus Torvalds struct list_head *entry; 888600ed416SJeff Mahoney unsigned int trans_id = jl->j_trans_id; 889600ed416SJeff Mahoney unsigned int other_trans_id; 8901da177e4SLinus Torvalds 8911da177e4SLinus Torvalds find_first: 8921da177e4SLinus Torvalds /* 8931da177e4SLinus Torvalds * first we walk backwards to find the oldest uncommitted transation 8941da177e4SLinus Torvalds */ 8951da177e4SLinus Torvalds first_jl = jl; 8961da177e4SLinus Torvalds entry = jl->j_list.prev; 8971da177e4SLinus Torvalds while (1) { 8981da177e4SLinus Torvalds other_jl = JOURNAL_LIST_ENTRY(entry); 8991da177e4SLinus Torvalds if (entry == &journal->j_journal_list || 9001da177e4SLinus Torvalds atomic_read(&other_jl->j_older_commits_done)) 9011da177e4SLinus Torvalds break; 9021da177e4SLinus Torvalds 9031da177e4SLinus Torvalds first_jl = other_jl; 9041da177e4SLinus Torvalds entry = other_jl->j_list.prev; 9051da177e4SLinus Torvalds } 9061da177e4SLinus Torvalds 9071da177e4SLinus Torvalds /* if we didn't find any older uncommitted transactions, return now */ 9081da177e4SLinus Torvalds if (first_jl == jl) { 9091da177e4SLinus Torvalds return 0; 9101da177e4SLinus Torvalds } 9111da177e4SLinus Torvalds 9121da177e4SLinus Torvalds entry = &first_jl->j_list; 9131da177e4SLinus Torvalds while (1) { 9141da177e4SLinus Torvalds other_jl = JOURNAL_LIST_ENTRY(entry); 9151da177e4SLinus Torvalds other_trans_id = other_jl->j_trans_id; 9161da177e4SLinus Torvalds 9171da177e4SLinus Torvalds if (other_trans_id < trans_id) { 9181da177e4SLinus Torvalds if (atomic_read(&other_jl->j_commit_left) != 0) { 9191da177e4SLinus Torvalds flush_commit_list(s, other_jl, 0); 9201da177e4SLinus Torvalds 9211da177e4SLinus Torvalds /* list we were called with is gone, return */ 9221da177e4SLinus Torvalds if (!journal_list_still_alive(s, trans_id)) 9231da177e4SLinus Torvalds return 1; 9241da177e4SLinus Torvalds 925098297b2SJeff Mahoney /* 926098297b2SJeff Mahoney * the one we just flushed is gone, this means 927098297b2SJeff Mahoney * all older lists are also gone, so first_jl 928098297b2SJeff Mahoney * is no longer valid either. Go back to the 929098297b2SJeff Mahoney * beginning. 9301da177e4SLinus Torvalds */ 931bd4c625cSLinus Torvalds if (!journal_list_still_alive 932bd4c625cSLinus Torvalds (s, other_trans_id)) { 9331da177e4SLinus Torvalds goto find_first; 9341da177e4SLinus Torvalds } 9351da177e4SLinus Torvalds } 9361da177e4SLinus Torvalds entry = entry->next; 9371da177e4SLinus Torvalds if (entry == &journal->j_journal_list) 9381da177e4SLinus Torvalds return 0; 9391da177e4SLinus Torvalds } else { 9401da177e4SLinus Torvalds return 0; 9411da177e4SLinus Torvalds } 9421da177e4SLinus Torvalds } 9431da177e4SLinus Torvalds return 0; 9441da177e4SLinus Torvalds } 945deba0f49SAdrian Bunk 946deba0f49SAdrian Bunk static int reiserfs_async_progress_wait(struct super_block *s) 947bd4c625cSLinus Torvalds { 9481da177e4SLinus Torvalds struct reiserfs_journal *j = SB_JOURNAL(s); 9498ebc4232SFrederic Weisbecker 9508ebc4232SFrederic Weisbecker if (atomic_read(&j->j_async_throttle)) { 951278f6679SJeff Mahoney int depth; 952278f6679SJeff Mahoney 953278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(s); 9548aa7e847SJens Axboe congestion_wait(BLK_RW_ASYNC, HZ / 10); 955278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 9568ebc4232SFrederic Weisbecker } 9578ebc4232SFrederic Weisbecker 9581da177e4SLinus Torvalds return 0; 9591da177e4SLinus Torvalds } 9601da177e4SLinus Torvalds 9611da177e4SLinus Torvalds /* 962098297b2SJeff Mahoney * if this journal list still has commit blocks unflushed, send them to disk. 963098297b2SJeff Mahoney * 964098297b2SJeff Mahoney * log areas must be flushed in order (transaction 2 can't commit before 965098297b2SJeff Mahoney * transaction 1) Before the commit block can by written, every other log 966098297b2SJeff Mahoney * block must be safely on disk 9671da177e4SLinus Torvalds */ 968bd4c625cSLinus Torvalds static int flush_commit_list(struct super_block *s, 969bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl, int flushall) 970bd4c625cSLinus Torvalds { 9711da177e4SLinus Torvalds int i; 9723ee16670SJeff Mahoney b_blocknr_t bn; 9731da177e4SLinus Torvalds struct buffer_head *tbh = NULL; 974600ed416SJeff Mahoney unsigned int trans_id = jl->j_trans_id; 9751da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 9761da177e4SLinus Torvalds int retval = 0; 977e0e851cfSChris Mason int write_len; 978278f6679SJeff Mahoney int depth; 9791da177e4SLinus Torvalds 9801da177e4SLinus Torvalds reiserfs_check_lock_depth(s, "flush_commit_list"); 9811da177e4SLinus Torvalds 9821da177e4SLinus Torvalds if (atomic_read(&jl->j_older_commits_done)) { 9831da177e4SLinus Torvalds return 0; 9841da177e4SLinus Torvalds } 9851da177e4SLinus Torvalds 986098297b2SJeff Mahoney /* 987098297b2SJeff Mahoney * before we can put our commit blocks on disk, we have to make 988098297b2SJeff Mahoney * sure everyone older than us is on disk too 9891da177e4SLinus Torvalds */ 9901da177e4SLinus Torvalds BUG_ON(jl->j_len <= 0); 9911da177e4SLinus Torvalds BUG_ON(trans_id == journal->j_trans_id); 9921da177e4SLinus Torvalds 9931da177e4SLinus Torvalds get_journal_list(jl); 9941da177e4SLinus Torvalds if (flushall) { 9951da177e4SLinus Torvalds if (flush_older_commits(s, jl) == 1) { 996098297b2SJeff Mahoney /* 997098297b2SJeff Mahoney * list disappeared during flush_older_commits. 998098297b2SJeff Mahoney * return 999098297b2SJeff Mahoney */ 10001da177e4SLinus Torvalds goto put_jl; 10011da177e4SLinus Torvalds } 10021da177e4SLinus Torvalds } 10031da177e4SLinus Torvalds 10041da177e4SLinus Torvalds /* make sure nobody is trying to flush this one at the same time */ 10058ebc4232SFrederic Weisbecker reiserfs_mutex_lock_safe(&jl->j_commit_mutex, s); 10068ebc4232SFrederic Weisbecker 10071da177e4SLinus Torvalds if (!journal_list_still_alive(s, trans_id)) { 100890415deaSJeff Mahoney mutex_unlock(&jl->j_commit_mutex); 10091da177e4SLinus Torvalds goto put_jl; 10101da177e4SLinus Torvalds } 10111da177e4SLinus Torvalds BUG_ON(jl->j_trans_id == 0); 10121da177e4SLinus Torvalds 10131da177e4SLinus Torvalds /* this commit is done, exit */ 1014a228bf8fSJeff Mahoney if (atomic_read(&jl->j_commit_left) <= 0) { 10151da177e4SLinus Torvalds if (flushall) { 1016a228bf8fSJeff Mahoney atomic_set(&jl->j_older_commits_done, 1); 10171da177e4SLinus Torvalds } 101890415deaSJeff Mahoney mutex_unlock(&jl->j_commit_mutex); 10191da177e4SLinus Torvalds goto put_jl; 10201da177e4SLinus Torvalds } 10211da177e4SLinus Torvalds 10221da177e4SLinus Torvalds if (!list_empty(&jl->j_bh_list)) { 10233d4492f8SChris Mason int ret; 10248ebc4232SFrederic Weisbecker 10258ebc4232SFrederic Weisbecker /* 10268ebc4232SFrederic Weisbecker * We might sleep in numerous places inside 10278ebc4232SFrederic Weisbecker * write_ordered_buffers. Relax the write lock. 10288ebc4232SFrederic Weisbecker */ 1029278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(s); 10303d4492f8SChris Mason ret = write_ordered_buffers(&journal->j_dirty_buffers_lock, 10311da177e4SLinus Torvalds journal, jl, &jl->j_bh_list); 10323d4492f8SChris Mason if (ret < 0 && retval == 0) 10333d4492f8SChris Mason retval = ret; 1034278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 10351da177e4SLinus Torvalds } 10361da177e4SLinus Torvalds BUG_ON(!list_empty(&jl->j_bh_list)); 10371da177e4SLinus Torvalds /* 10381da177e4SLinus Torvalds * for the description block and all the log blocks, submit any buffers 1039e0e851cfSChris Mason * that haven't already reached the disk. Try to write at least 256 1040e0e851cfSChris Mason * log blocks. later on, we will only wait on blocks that correspond 1041e0e851cfSChris Mason * to this transaction, but while we're unplugging we might as well 1042e0e851cfSChris Mason * get a chunk of data on there. 10431da177e4SLinus Torvalds */ 10441da177e4SLinus Torvalds atomic_inc(&journal->j_async_throttle); 1045e0e851cfSChris Mason write_len = jl->j_len + 1; 1046e0e851cfSChris Mason if (write_len < 256) 1047e0e851cfSChris Mason write_len = 256; 1048e0e851cfSChris Mason for (i = 0 ; i < write_len ; i++) { 10491da177e4SLinus Torvalds bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) % 10501da177e4SLinus Torvalds SB_ONDISK_JOURNAL_SIZE(s); 10511da177e4SLinus Torvalds tbh = journal_find_get_block(s, bn); 1052e0e851cfSChris Mason if (tbh) { 10536e3647acSFrederic Weisbecker if (buffer_dirty(tbh)) { 1054278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(s); 1055dfec8a14SMike Christie ll_rw_block(REQ_OP_WRITE, 0, 1, &tbh); 1056278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 10576e3647acSFrederic Weisbecker } 10581da177e4SLinus Torvalds put_bh(tbh) ; 10591da177e4SLinus Torvalds } 1060e0e851cfSChris Mason } 10611da177e4SLinus Torvalds atomic_dec(&journal->j_async_throttle); 10621da177e4SLinus Torvalds 10631da177e4SLinus Torvalds for (i = 0; i < (jl->j_len + 1); i++) { 10641da177e4SLinus Torvalds bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + 10651da177e4SLinus Torvalds (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s); 10661da177e4SLinus Torvalds tbh = journal_find_get_block(s, bn); 10678ebc4232SFrederic Weisbecker 1068278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(s); 1069278f6679SJeff Mahoney __wait_on_buffer(tbh); 1070278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 1071098297b2SJeff Mahoney /* 1072098297b2SJeff Mahoney * since we're using ll_rw_blk above, it might have skipped 1073098297b2SJeff Mahoney * over a locked buffer. Double check here 1074098297b2SJeff Mahoney */ 10758ebc4232SFrederic Weisbecker /* redundant, sync_dirty_buffer() checks */ 10768ebc4232SFrederic Weisbecker if (buffer_dirty(tbh)) { 1077278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(s); 10781da177e4SLinus Torvalds sync_dirty_buffer(tbh); 1079278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 10808ebc4232SFrederic Weisbecker } 10811da177e4SLinus Torvalds if (unlikely(!buffer_uptodate(tbh))) { 10821da177e4SLinus Torvalds #ifdef CONFIG_REISERFS_CHECK 108345b03d5eSJeff Mahoney reiserfs_warning(s, "journal-601", 108445b03d5eSJeff Mahoney "buffer write failed"); 10851da177e4SLinus Torvalds #endif 10861da177e4SLinus Torvalds retval = -EIO; 10871da177e4SLinus Torvalds } 1088098297b2SJeff Mahoney /* once for journal_find_get_block */ 1089098297b2SJeff Mahoney put_bh(tbh); 1090098297b2SJeff Mahoney /* once due to original getblk in do_journal_end */ 1091098297b2SJeff Mahoney put_bh(tbh); 1092a228bf8fSJeff Mahoney atomic_dec(&jl->j_commit_left); 10931da177e4SLinus Torvalds } 10941da177e4SLinus Torvalds 1095a228bf8fSJeff Mahoney BUG_ON(atomic_read(&jl->j_commit_left) != 1); 10961da177e4SLinus Torvalds 1097098297b2SJeff Mahoney /* 1098098297b2SJeff Mahoney * If there was a write error in the journal - we can't commit 10995d5e8156SJeff Mahoney * this transaction - it will be invalid and, if successful, 1100beb7dd86SRobert P. J. Day * will just end up propagating the write error out to 1101098297b2SJeff Mahoney * the file system. 1102098297b2SJeff Mahoney */ 11035d5e8156SJeff Mahoney if (likely(!retval && !reiserfs_is_journal_aborted (journal))) { 11041da177e4SLinus Torvalds if (buffer_dirty(jl->j_commit_bh)) 11051da177e4SLinus Torvalds BUG(); 11061da177e4SLinus Torvalds mark_buffer_dirty(jl->j_commit_bh) ; 1107278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(s); 11087cd33ad2SChristoph Hellwig if (reiserfs_barrier_flush(s)) 110970fd7614SChristoph Hellwig __sync_dirty_buffer(jl->j_commit_bh, 1110d8747d64SJan Kara REQ_SYNC | REQ_PREFLUSH | REQ_FUA); 11117cd33ad2SChristoph Hellwig else 11121da177e4SLinus Torvalds sync_dirty_buffer(jl->j_commit_bh); 1113278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 11145d5e8156SJeff Mahoney } 11151da177e4SLinus Torvalds 1116098297b2SJeff Mahoney /* 1117098297b2SJeff Mahoney * If there was a write error in the journal - we can't commit this 11181da177e4SLinus Torvalds * transaction - it will be invalid and, if successful, will just end 1119098297b2SJeff Mahoney * up propagating the write error out to the filesystem. 1120098297b2SJeff Mahoney */ 11211da177e4SLinus Torvalds if (unlikely(!buffer_uptodate(jl->j_commit_bh))) { 11221da177e4SLinus Torvalds #ifdef CONFIG_REISERFS_CHECK 112345b03d5eSJeff Mahoney reiserfs_warning(s, "journal-615", "buffer write failed"); 11241da177e4SLinus Torvalds #endif 11251da177e4SLinus Torvalds retval = -EIO; 11261da177e4SLinus Torvalds } 11271da177e4SLinus Torvalds bforget(jl->j_commit_bh); 11281da177e4SLinus Torvalds if (journal->j_last_commit_id != 0 && 11291da177e4SLinus Torvalds (jl->j_trans_id - journal->j_last_commit_id) != 1) { 113045b03d5eSJeff Mahoney reiserfs_warning(s, "clm-2200", "last commit %lu, current %lu", 1131bd4c625cSLinus Torvalds journal->j_last_commit_id, jl->j_trans_id); 11321da177e4SLinus Torvalds } 11331da177e4SLinus Torvalds journal->j_last_commit_id = jl->j_trans_id; 11341da177e4SLinus Torvalds 1135098297b2SJeff Mahoney /* 1136098297b2SJeff Mahoney * now, every commit block is on the disk. It is safe to allow 1137098297b2SJeff Mahoney * blocks freed during this transaction to be reallocated 1138098297b2SJeff Mahoney */ 11391da177e4SLinus Torvalds cleanup_freed_for_journal_list(s, jl); 11401da177e4SLinus Torvalds 11411da177e4SLinus Torvalds retval = retval ? retval : journal->j_errno; 11421da177e4SLinus Torvalds 11431da177e4SLinus Torvalds /* mark the metadata dirty */ 11441da177e4SLinus Torvalds if (!retval) 11451da177e4SLinus Torvalds dirty_one_transaction(s, jl); 1146a228bf8fSJeff Mahoney atomic_dec(&jl->j_commit_left); 11471da177e4SLinus Torvalds 11481da177e4SLinus Torvalds if (flushall) { 1149a228bf8fSJeff Mahoney atomic_set(&jl->j_older_commits_done, 1); 11501da177e4SLinus Torvalds } 115190415deaSJeff Mahoney mutex_unlock(&jl->j_commit_mutex); 11521da177e4SLinus Torvalds put_jl: 11531da177e4SLinus Torvalds put_journal_list(s, jl); 11541da177e4SLinus Torvalds 11551da177e4SLinus Torvalds if (retval) 1156bd4c625cSLinus Torvalds reiserfs_abort(s, retval, "Journal write error in %s", 1157fbe5498bSHarvey Harrison __func__); 11581da177e4SLinus Torvalds return retval; 11591da177e4SLinus Torvalds } 11601da177e4SLinus Torvalds 11611da177e4SLinus Torvalds /* 1162098297b2SJeff Mahoney * flush_journal_list frequently needs to find a newer transaction for a 1163098297b2SJeff Mahoney * given block. This does that, or returns NULL if it can't find anything 11641da177e4SLinus Torvalds */ 1165bd4c625cSLinus Torvalds static struct reiserfs_journal_list *find_newer_jl_for_cn(struct 1166bd4c625cSLinus Torvalds reiserfs_journal_cnode 1167bd4c625cSLinus Torvalds *cn) 1168bd4c625cSLinus Torvalds { 11691da177e4SLinus Torvalds struct super_block *sb = cn->sb; 11701da177e4SLinus Torvalds b_blocknr_t blocknr = cn->blocknr; 11711da177e4SLinus Torvalds 11721da177e4SLinus Torvalds cn = cn->hprev; 11731da177e4SLinus Torvalds while (cn) { 11741da177e4SLinus Torvalds if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) { 11751da177e4SLinus Torvalds return cn->jlist; 11761da177e4SLinus Torvalds } 11771da177e4SLinus Torvalds cn = cn->hprev; 11781da177e4SLinus Torvalds } 11791da177e4SLinus Torvalds return NULL; 11801da177e4SLinus Torvalds } 11811da177e4SLinus Torvalds 1182bd4c625cSLinus Torvalds static void remove_journal_hash(struct super_block *, 1183bd4c625cSLinus Torvalds struct reiserfs_journal_cnode **, 1184bd4c625cSLinus Torvalds struct reiserfs_journal_list *, unsigned long, 1185bd4c625cSLinus Torvalds int); 11861da177e4SLinus Torvalds 11871da177e4SLinus Torvalds /* 1188098297b2SJeff Mahoney * once all the real blocks have been flushed, it is safe to remove them 1189098297b2SJeff Mahoney * from the journal list for this transaction. Aside from freeing the 1190098297b2SJeff Mahoney * cnode, this also allows the block to be reallocated for data blocks 1191098297b2SJeff Mahoney * if it had been deleted. 11921da177e4SLinus Torvalds */ 1193a9dd3643SJeff Mahoney static void remove_all_from_journal_list(struct super_block *sb, 1194bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl, 1195bd4c625cSLinus Torvalds int debug) 1196bd4c625cSLinus Torvalds { 1197a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 11981da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn, *last; 11991da177e4SLinus Torvalds cn = jl->j_realblock; 12001da177e4SLinus Torvalds 1201098297b2SJeff Mahoney /* 1202098297b2SJeff Mahoney * which is better, to lock once around the whole loop, or 1203098297b2SJeff Mahoney * to lock for each call to remove_journal_hash? 12041da177e4SLinus Torvalds */ 12051da177e4SLinus Torvalds while (cn) { 12061da177e4SLinus Torvalds if (cn->blocknr != 0) { 12071da177e4SLinus Torvalds if (debug) { 1208a9dd3643SJeff Mahoney reiserfs_warning(sb, "reiserfs-2201", 1209bd4c625cSLinus Torvalds "block %u, bh is %d, state %ld", 1210bd4c625cSLinus Torvalds cn->blocknr, cn->bh ? 1 : 0, 1211bd4c625cSLinus Torvalds cn->state); 12121da177e4SLinus Torvalds } 12131da177e4SLinus Torvalds cn->state = 0; 1214a9dd3643SJeff Mahoney remove_journal_hash(sb, journal->j_list_hash_table, 1215bd4c625cSLinus Torvalds jl, cn->blocknr, 1); 12161da177e4SLinus Torvalds } 12171da177e4SLinus Torvalds last = cn; 12181da177e4SLinus Torvalds cn = cn->next; 1219a9dd3643SJeff Mahoney free_cnode(sb, last); 12201da177e4SLinus Torvalds } 12211da177e4SLinus Torvalds jl->j_realblock = NULL; 12221da177e4SLinus Torvalds } 12231da177e4SLinus Torvalds 12241da177e4SLinus Torvalds /* 1225098297b2SJeff Mahoney * if this timestamp is greater than the timestamp we wrote last to the 1226098297b2SJeff Mahoney * header block, write it to the header block. once this is done, I can 1227098297b2SJeff Mahoney * safely say the log area for this transaction won't ever be replayed, 1228098297b2SJeff Mahoney * and I can start releasing blocks in this transaction for reuse as data 1229098297b2SJeff Mahoney * blocks. called by flush_journal_list, before it calls 1230098297b2SJeff Mahoney * remove_all_from_journal_list 12311da177e4SLinus Torvalds */ 1232a9dd3643SJeff Mahoney static int _update_journal_header_block(struct super_block *sb, 1233bd4c625cSLinus Torvalds unsigned long offset, 1234600ed416SJeff Mahoney unsigned int trans_id) 1235bd4c625cSLinus Torvalds { 12361da177e4SLinus Torvalds struct reiserfs_journal_header *jh; 1237a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 1238278f6679SJeff Mahoney int depth; 12391da177e4SLinus Torvalds 12401da177e4SLinus Torvalds if (reiserfs_is_journal_aborted(journal)) 12411da177e4SLinus Torvalds return -EIO; 12421da177e4SLinus Torvalds 12431da177e4SLinus Torvalds if (trans_id >= journal->j_last_flush_trans_id) { 12441da177e4SLinus Torvalds if (buffer_locked((journal->j_header_bh))) { 1245278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(sb); 1246278f6679SJeff Mahoney __wait_on_buffer(journal->j_header_bh); 1247278f6679SJeff Mahoney reiserfs_write_lock_nested(sb, depth); 12481da177e4SLinus Torvalds if (unlikely(!buffer_uptodate(journal->j_header_bh))) { 12491da177e4SLinus Torvalds #ifdef CONFIG_REISERFS_CHECK 1250a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-699", 125145b03d5eSJeff Mahoney "buffer write failed"); 12521da177e4SLinus Torvalds #endif 12531da177e4SLinus Torvalds return -EIO; 12541da177e4SLinus Torvalds } 12551da177e4SLinus Torvalds } 12561da177e4SLinus Torvalds journal->j_last_flush_trans_id = trans_id; 12571da177e4SLinus Torvalds journal->j_first_unflushed_offset = offset; 1258bd4c625cSLinus Torvalds jh = (struct reiserfs_journal_header *)(journal->j_header_bh-> 1259bd4c625cSLinus Torvalds b_data); 12601da177e4SLinus Torvalds jh->j_last_flush_trans_id = cpu_to_le32(trans_id); 12611da177e4SLinus Torvalds jh->j_first_unflushed_offset = cpu_to_le32(offset); 12621da177e4SLinus Torvalds jh->j_mount_id = cpu_to_le32(journal->j_mount_id); 12631da177e4SLinus Torvalds 12641da177e4SLinus Torvalds set_buffer_dirty(journal->j_header_bh); 1265278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(sb); 12667cd33ad2SChristoph Hellwig 12677cd33ad2SChristoph Hellwig if (reiserfs_barrier_flush(sb)) 126870fd7614SChristoph Hellwig __sync_dirty_buffer(journal->j_header_bh, 1269d8747d64SJan Kara REQ_SYNC | REQ_PREFLUSH | REQ_FUA); 12707cd33ad2SChristoph Hellwig else 12711da177e4SLinus Torvalds sync_dirty_buffer(journal->j_header_bh); 12727cd33ad2SChristoph Hellwig 1273278f6679SJeff Mahoney reiserfs_write_lock_nested(sb, depth); 12741da177e4SLinus Torvalds if (!buffer_uptodate(journal->j_header_bh)) { 1275a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-837", 127645b03d5eSJeff Mahoney "IO error during journal replay"); 12771da177e4SLinus Torvalds return -EIO; 12781da177e4SLinus Torvalds } 12791da177e4SLinus Torvalds } 12801da177e4SLinus Torvalds return 0; 12811da177e4SLinus Torvalds } 12821da177e4SLinus Torvalds 1283a9dd3643SJeff Mahoney static int update_journal_header_block(struct super_block *sb, 12841da177e4SLinus Torvalds unsigned long offset, 1285600ed416SJeff Mahoney unsigned int trans_id) 1286bd4c625cSLinus Torvalds { 1287a9dd3643SJeff Mahoney return _update_journal_header_block(sb, offset, trans_id); 12881da177e4SLinus Torvalds } 1289bd4c625cSLinus Torvalds 12901da177e4SLinus Torvalds /* 12911da177e4SLinus Torvalds ** flush any and all journal lists older than you are 12921da177e4SLinus Torvalds ** can only be called from flush_journal_list 12931da177e4SLinus Torvalds */ 1294a9dd3643SJeff Mahoney static int flush_older_journal_lists(struct super_block *sb, 12951da177e4SLinus Torvalds struct reiserfs_journal_list *jl) 12961da177e4SLinus Torvalds { 12971da177e4SLinus Torvalds struct list_head *entry; 12981da177e4SLinus Torvalds struct reiserfs_journal_list *other_jl; 1299a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 1300600ed416SJeff Mahoney unsigned int trans_id = jl->j_trans_id; 13011da177e4SLinus Torvalds 1302098297b2SJeff Mahoney /* 1303098297b2SJeff Mahoney * we know we are the only ones flushing things, no extra race 13041da177e4SLinus Torvalds * protection is required. 13051da177e4SLinus Torvalds */ 13061da177e4SLinus Torvalds restart: 13071da177e4SLinus Torvalds entry = journal->j_journal_list.next; 13081da177e4SLinus Torvalds /* Did we wrap? */ 13091da177e4SLinus Torvalds if (entry == &journal->j_journal_list) 13101da177e4SLinus Torvalds return 0; 13111da177e4SLinus Torvalds other_jl = JOURNAL_LIST_ENTRY(entry); 13121da177e4SLinus Torvalds if (other_jl->j_trans_id < trans_id) { 13131da177e4SLinus Torvalds BUG_ON(other_jl->j_refcount <= 0); 13141da177e4SLinus Torvalds /* do not flush all */ 1315a9dd3643SJeff Mahoney flush_journal_list(sb, other_jl, 0); 13161da177e4SLinus Torvalds 13171da177e4SLinus Torvalds /* other_jl is now deleted from the list */ 13181da177e4SLinus Torvalds goto restart; 13191da177e4SLinus Torvalds } 13201da177e4SLinus Torvalds return 0; 13211da177e4SLinus Torvalds } 13221da177e4SLinus Torvalds 13231da177e4SLinus Torvalds static void del_from_work_list(struct super_block *s, 1324bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl) 1325bd4c625cSLinus Torvalds { 13261da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 13271da177e4SLinus Torvalds if (!list_empty(&jl->j_working_list)) { 13281da177e4SLinus Torvalds list_del_init(&jl->j_working_list); 13291da177e4SLinus Torvalds journal->j_num_work_lists--; 13301da177e4SLinus Torvalds } 13311da177e4SLinus Torvalds } 13321da177e4SLinus Torvalds 1333098297b2SJeff Mahoney /* 1334098297b2SJeff Mahoney * flush a journal list, both commit and real blocks 1335098297b2SJeff Mahoney * 1336098297b2SJeff Mahoney * always set flushall to 1, unless you are calling from inside 1337098297b2SJeff Mahoney * flush_journal_list 1338098297b2SJeff Mahoney * 1339098297b2SJeff Mahoney * IMPORTANT. This can only be called while there are no journal writers, 1340098297b2SJeff Mahoney * and the journal is locked. That means it can only be called from 1341098297b2SJeff Mahoney * do_journal_end, or by journal_release 13421da177e4SLinus Torvalds */ 13431da177e4SLinus Torvalds static int flush_journal_list(struct super_block *s, 1344bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl, int flushall) 1345bd4c625cSLinus Torvalds { 13461da177e4SLinus Torvalds struct reiserfs_journal_list *pjl; 13476e9ca45fSzhengbin struct reiserfs_journal_cnode *cn; 13481da177e4SLinus Torvalds int count; 13491da177e4SLinus Torvalds int was_jwait = 0; 13501da177e4SLinus Torvalds int was_dirty = 0; 13511da177e4SLinus Torvalds struct buffer_head *saved_bh; 13521da177e4SLinus Torvalds unsigned long j_len_saved = jl->j_len; 13531da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 13541da177e4SLinus Torvalds int err = 0; 1355278f6679SJeff Mahoney int depth; 13561da177e4SLinus Torvalds 13571da177e4SLinus Torvalds BUG_ON(j_len_saved <= 0); 13581da177e4SLinus Torvalds 13591da177e4SLinus Torvalds if (atomic_read(&journal->j_wcount) != 0) { 136045b03d5eSJeff Mahoney reiserfs_warning(s, "clm-2048", "called with wcount %d", 13611da177e4SLinus Torvalds atomic_read(&journal->j_wcount)); 13621da177e4SLinus Torvalds } 13631da177e4SLinus Torvalds 13641da177e4SLinus Torvalds /* if flushall == 0, the lock is already held */ 13651da177e4SLinus Torvalds if (flushall) { 13668ebc4232SFrederic Weisbecker reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s); 1367afe70259SJeff Mahoney } else if (mutex_trylock(&journal->j_flush_mutex)) { 13681da177e4SLinus Torvalds BUG(); 13691da177e4SLinus Torvalds } 13701da177e4SLinus Torvalds 13711da177e4SLinus Torvalds count = 0; 13721da177e4SLinus Torvalds if (j_len_saved > journal->j_trans_max) { 1373c3a9c210SJeff Mahoney reiserfs_panic(s, "journal-715", "length is %lu, trans id %lu", 1374bd4c625cSLinus Torvalds j_len_saved, jl->j_trans_id); 13751da177e4SLinus Torvalds return 0; 13761da177e4SLinus Torvalds } 13771da177e4SLinus Torvalds 13781da177e4SLinus Torvalds /* if all the work is already done, get out of here */ 1379a228bf8fSJeff Mahoney if (atomic_read(&jl->j_nonzerolen) <= 0 && 1380a228bf8fSJeff Mahoney atomic_read(&jl->j_commit_left) <= 0) { 13811da177e4SLinus Torvalds goto flush_older_and_return; 13821da177e4SLinus Torvalds } 13831da177e4SLinus Torvalds 1384098297b2SJeff Mahoney /* 1385098297b2SJeff Mahoney * start by putting the commit list on disk. This will also flush 1386098297b2SJeff Mahoney * the commit lists of any olders transactions 13871da177e4SLinus Torvalds */ 13881da177e4SLinus Torvalds flush_commit_list(s, jl, 1); 13891da177e4SLinus Torvalds 1390bd4c625cSLinus Torvalds if (!(jl->j_state & LIST_DIRTY) 1391bd4c625cSLinus Torvalds && !reiserfs_is_journal_aborted(journal)) 13921da177e4SLinus Torvalds BUG(); 13931da177e4SLinus Torvalds 13941da177e4SLinus Torvalds /* are we done now? */ 1395a228bf8fSJeff Mahoney if (atomic_read(&jl->j_nonzerolen) <= 0 && 1396a228bf8fSJeff Mahoney atomic_read(&jl->j_commit_left) <= 0) { 13971da177e4SLinus Torvalds goto flush_older_and_return; 13981da177e4SLinus Torvalds } 13991da177e4SLinus Torvalds 1400098297b2SJeff Mahoney /* 1401098297b2SJeff Mahoney * loop through each cnode, see if we need to write it, 1402098297b2SJeff Mahoney * or wait on a more recent transaction, or just ignore it 14031da177e4SLinus Torvalds */ 1404a228bf8fSJeff Mahoney if (atomic_read(&journal->j_wcount) != 0) { 1405c3a9c210SJeff Mahoney reiserfs_panic(s, "journal-844", "journal list is flushing, " 1406c3a9c210SJeff Mahoney "wcount is not 0"); 14071da177e4SLinus Torvalds } 14081da177e4SLinus Torvalds cn = jl->j_realblock; 14091da177e4SLinus Torvalds while (cn) { 14101da177e4SLinus Torvalds was_jwait = 0; 14111da177e4SLinus Torvalds was_dirty = 0; 14121da177e4SLinus Torvalds saved_bh = NULL; 14131da177e4SLinus Torvalds /* blocknr of 0 is no longer in the hash, ignore it */ 14141da177e4SLinus Torvalds if (cn->blocknr == 0) { 14151da177e4SLinus Torvalds goto free_cnode; 14161da177e4SLinus Torvalds } 14171da177e4SLinus Torvalds 1418098297b2SJeff Mahoney /* 1419098297b2SJeff Mahoney * This transaction failed commit. 1420098297b2SJeff Mahoney * Don't write out to the disk 1421098297b2SJeff Mahoney */ 14221da177e4SLinus Torvalds if (!(jl->j_state & LIST_DIRTY)) 14231da177e4SLinus Torvalds goto free_cnode; 14241da177e4SLinus Torvalds 14251da177e4SLinus Torvalds pjl = find_newer_jl_for_cn(cn); 1426098297b2SJeff Mahoney /* 1427098297b2SJeff Mahoney * the order is important here. We check pjl to make sure we 1428098297b2SJeff Mahoney * don't clear BH_JDirty_wait if we aren't the one writing this 1429098297b2SJeff Mahoney * block to disk 14301da177e4SLinus Torvalds */ 14311da177e4SLinus Torvalds if (!pjl && cn->bh) { 14321da177e4SLinus Torvalds saved_bh = cn->bh; 14331da177e4SLinus Torvalds 1434098297b2SJeff Mahoney /* 1435098297b2SJeff Mahoney * we do this to make sure nobody releases the 1436098297b2SJeff Mahoney * buffer while we are working with it 14371da177e4SLinus Torvalds */ 14381da177e4SLinus Torvalds get_bh(saved_bh); 14391da177e4SLinus Torvalds 14401da177e4SLinus Torvalds if (buffer_journal_dirty(saved_bh)) { 14411da177e4SLinus Torvalds BUG_ON(!can_dirty(cn)); 14421da177e4SLinus Torvalds was_jwait = 1; 14431da177e4SLinus Torvalds was_dirty = 1; 14441da177e4SLinus Torvalds } else if (can_dirty(cn)) { 1445098297b2SJeff Mahoney /* 1446098297b2SJeff Mahoney * everything with !pjl && jwait 1447098297b2SJeff Mahoney * should be writable 1448098297b2SJeff Mahoney */ 14491da177e4SLinus Torvalds BUG(); 14501da177e4SLinus Torvalds } 14511da177e4SLinus Torvalds } 14521da177e4SLinus Torvalds 1453098297b2SJeff Mahoney /* 1454098297b2SJeff Mahoney * if someone has this block in a newer transaction, just make 1455098297b2SJeff Mahoney * sure they are committed, and don't try writing it to disk 14561da177e4SLinus Torvalds */ 14571da177e4SLinus Torvalds if (pjl) { 14581da177e4SLinus Torvalds if (atomic_read(&pjl->j_commit_left)) 14591da177e4SLinus Torvalds flush_commit_list(s, pjl, 1); 14601da177e4SLinus Torvalds goto free_cnode; 14611da177e4SLinus Torvalds } 14621da177e4SLinus Torvalds 1463098297b2SJeff Mahoney /* 1464098297b2SJeff Mahoney * bh == NULL when the block got to disk on its own, OR, 1465098297b2SJeff Mahoney * the block got freed in a future transaction 14661da177e4SLinus Torvalds */ 14671da177e4SLinus Torvalds if (saved_bh == NULL) { 14681da177e4SLinus Torvalds goto free_cnode; 14691da177e4SLinus Torvalds } 14701da177e4SLinus Torvalds 1471098297b2SJeff Mahoney /* 1472098297b2SJeff Mahoney * this should never happen. kupdate_one_transaction has 1473098297b2SJeff Mahoney * this list locked while it works, so we should never see a 1474098297b2SJeff Mahoney * buffer here that is not marked JDirty_wait 14751da177e4SLinus Torvalds */ 14761da177e4SLinus Torvalds if ((!was_jwait) && !buffer_locked(saved_bh)) { 147745b03d5eSJeff Mahoney reiserfs_warning(s, "journal-813", 147845b03d5eSJeff Mahoney "BAD! buffer %llu %cdirty %cjwait, " 147965f2b263SColin Ian King "not in a newer transaction", 1480bd4c625cSLinus Torvalds (unsigned long long)saved_bh-> 1481bd4c625cSLinus Torvalds b_blocknr, was_dirty ? ' ' : '!', 1482bd4c625cSLinus Torvalds was_jwait ? ' ' : '!'); 14831da177e4SLinus Torvalds } 14841da177e4SLinus Torvalds if (was_dirty) { 1485098297b2SJeff Mahoney /* 1486098297b2SJeff Mahoney * we inc again because saved_bh gets decremented 1487098297b2SJeff Mahoney * at free_cnode 1488098297b2SJeff Mahoney */ 14891da177e4SLinus Torvalds get_bh(saved_bh); 14901da177e4SLinus Torvalds set_bit(BLOCK_NEEDS_FLUSH, &cn->state); 14911da177e4SLinus Torvalds lock_buffer(saved_bh); 14921da177e4SLinus Torvalds BUG_ON(cn->blocknr != saved_bh->b_blocknr); 14931da177e4SLinus Torvalds if (buffer_dirty(saved_bh)) 14941da177e4SLinus Torvalds submit_logged_buffer(saved_bh); 14951da177e4SLinus Torvalds else 14961da177e4SLinus Torvalds unlock_buffer(saved_bh); 14971da177e4SLinus Torvalds count++; 14981da177e4SLinus Torvalds } else { 149945b03d5eSJeff Mahoney reiserfs_warning(s, "clm-2082", 150045b03d5eSJeff Mahoney "Unable to flush buffer %llu in %s", 1501bd4c625cSLinus Torvalds (unsigned long long)saved_bh-> 1502fbe5498bSHarvey Harrison b_blocknr, __func__); 15031da177e4SLinus Torvalds } 15041da177e4SLinus Torvalds free_cnode: 15051da177e4SLinus Torvalds cn = cn->next; 15061da177e4SLinus Torvalds if (saved_bh) { 1507098297b2SJeff Mahoney /* 1508098297b2SJeff Mahoney * we incremented this to keep others from 1509098297b2SJeff Mahoney * taking the buffer head away 1510098297b2SJeff Mahoney */ 15111da177e4SLinus Torvalds put_bh(saved_bh); 1512a228bf8fSJeff Mahoney if (atomic_read(&saved_bh->b_count) < 0) { 151345b03d5eSJeff Mahoney reiserfs_warning(s, "journal-945", 151445b03d5eSJeff Mahoney "saved_bh->b_count < 0"); 15151da177e4SLinus Torvalds } 15161da177e4SLinus Torvalds } 15171da177e4SLinus Torvalds } 15181da177e4SLinus Torvalds if (count > 0) { 15191da177e4SLinus Torvalds cn = jl->j_realblock; 15201da177e4SLinus Torvalds while (cn) { 15211da177e4SLinus Torvalds if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) { 15221da177e4SLinus Torvalds if (!cn->bh) { 1523c3a9c210SJeff Mahoney reiserfs_panic(s, "journal-1011", 1524c3a9c210SJeff Mahoney "cn->bh is NULL"); 15251da177e4SLinus Torvalds } 15268ebc4232SFrederic Weisbecker 1527278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(s); 1528278f6679SJeff Mahoney __wait_on_buffer(cn->bh); 1529278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 15308ebc4232SFrederic Weisbecker 15311da177e4SLinus Torvalds if (!cn->bh) { 1532c3a9c210SJeff Mahoney reiserfs_panic(s, "journal-1012", 1533c3a9c210SJeff Mahoney "cn->bh is NULL"); 15341da177e4SLinus Torvalds } 15351da177e4SLinus Torvalds if (unlikely(!buffer_uptodate(cn->bh))) { 15361da177e4SLinus Torvalds #ifdef CONFIG_REISERFS_CHECK 153745b03d5eSJeff Mahoney reiserfs_warning(s, "journal-949", 153845b03d5eSJeff Mahoney "buffer write failed"); 15391da177e4SLinus Torvalds #endif 15401da177e4SLinus Torvalds err = -EIO; 15411da177e4SLinus Torvalds } 1542098297b2SJeff Mahoney /* 1543098297b2SJeff Mahoney * note, we must clear the JDirty_wait bit 1544098297b2SJeff Mahoney * after the up to date check, otherwise we 1545098297b2SJeff Mahoney * race against our flushpage routine 15461da177e4SLinus Torvalds */ 1547bd4c625cSLinus Torvalds BUG_ON(!test_clear_buffer_journal_dirty 1548bd4c625cSLinus Torvalds (cn->bh)); 15491da177e4SLinus Torvalds 1550398c95bdSChris Mason /* drop one ref for us */ 15511da177e4SLinus Torvalds put_bh(cn->bh); 1552398c95bdSChris Mason /* drop one ref for journal_mark_dirty */ 1553398c95bdSChris Mason release_buffer_page(cn->bh); 15541da177e4SLinus Torvalds } 15551da177e4SLinus Torvalds cn = cn->next; 15561da177e4SLinus Torvalds } 15571da177e4SLinus Torvalds } 15581da177e4SLinus Torvalds 15591da177e4SLinus Torvalds if (err) 1560bd4c625cSLinus Torvalds reiserfs_abort(s, -EIO, 1561bd4c625cSLinus Torvalds "Write error while pushing transaction to disk in %s", 1562fbe5498bSHarvey Harrison __func__); 15631da177e4SLinus Torvalds flush_older_and_return: 15641da177e4SLinus Torvalds 1565098297b2SJeff Mahoney /* 1566098297b2SJeff Mahoney * before we can update the journal header block, we _must_ flush all 1567098297b2SJeff Mahoney * real blocks from all older transactions to disk. This is because 1568098297b2SJeff Mahoney * once the header block is updated, this transaction will not be 1569098297b2SJeff Mahoney * replayed after a crash 15701da177e4SLinus Torvalds */ 15711da177e4SLinus Torvalds if (flushall) { 15721da177e4SLinus Torvalds flush_older_journal_lists(s, jl); 15731da177e4SLinus Torvalds } 15741da177e4SLinus Torvalds 15751da177e4SLinus Torvalds err = journal->j_errno; 1576098297b2SJeff Mahoney /* 1577098297b2SJeff Mahoney * before we can remove everything from the hash tables for this 1578098297b2SJeff Mahoney * transaction, we must make sure it can never be replayed 1579098297b2SJeff Mahoney * 1580098297b2SJeff Mahoney * since we are only called from do_journal_end, we know for sure there 1581098297b2SJeff Mahoney * are no allocations going on while we are flushing journal lists. So, 1582098297b2SJeff Mahoney * we only need to update the journal header block for the last list 1583098297b2SJeff Mahoney * being flushed 15841da177e4SLinus Torvalds */ 15851da177e4SLinus Torvalds if (!err && flushall) { 1586bd4c625cSLinus Torvalds err = 1587bd4c625cSLinus Torvalds update_journal_header_block(s, 1588bd4c625cSLinus Torvalds (jl->j_start + jl->j_len + 1589bd4c625cSLinus Torvalds 2) % SB_ONDISK_JOURNAL_SIZE(s), 1590bd4c625cSLinus Torvalds jl->j_trans_id); 15911da177e4SLinus Torvalds if (err) 1592bd4c625cSLinus Torvalds reiserfs_abort(s, -EIO, 1593bd4c625cSLinus Torvalds "Write error while updating journal header in %s", 1594fbe5498bSHarvey Harrison __func__); 15951da177e4SLinus Torvalds } 15961da177e4SLinus Torvalds remove_all_from_journal_list(s, jl, 0); 15971da177e4SLinus Torvalds list_del_init(&jl->j_list); 15981da177e4SLinus Torvalds journal->j_num_lists--; 15991da177e4SLinus Torvalds del_from_work_list(s, jl); 16001da177e4SLinus Torvalds 16011da177e4SLinus Torvalds if (journal->j_last_flush_id != 0 && 16021da177e4SLinus Torvalds (jl->j_trans_id - journal->j_last_flush_id) != 1) { 160345b03d5eSJeff Mahoney reiserfs_warning(s, "clm-2201", "last flush %lu, current %lu", 1604bd4c625cSLinus Torvalds journal->j_last_flush_id, jl->j_trans_id); 16051da177e4SLinus Torvalds } 16061da177e4SLinus Torvalds journal->j_last_flush_id = jl->j_trans_id; 16071da177e4SLinus Torvalds 1608098297b2SJeff Mahoney /* 1609098297b2SJeff Mahoney * not strictly required since we are freeing the list, but it should 16101da177e4SLinus Torvalds * help find code using dead lists later on 16111da177e4SLinus Torvalds */ 16121da177e4SLinus Torvalds jl->j_len = 0; 1613a228bf8fSJeff Mahoney atomic_set(&jl->j_nonzerolen, 0); 16141da177e4SLinus Torvalds jl->j_start = 0; 16151da177e4SLinus Torvalds jl->j_realblock = NULL; 16161da177e4SLinus Torvalds jl->j_commit_bh = NULL; 16171da177e4SLinus Torvalds jl->j_trans_id = 0; 16181da177e4SLinus Torvalds jl->j_state = 0; 16191da177e4SLinus Torvalds put_journal_list(s, jl); 16201da177e4SLinus Torvalds if (flushall) 1621afe70259SJeff Mahoney mutex_unlock(&journal->j_flush_mutex); 16221da177e4SLinus Torvalds return err; 16231da177e4SLinus Torvalds } 16241da177e4SLinus Torvalds 16251da177e4SLinus Torvalds static int write_one_transaction(struct super_block *s, 16261da177e4SLinus Torvalds struct reiserfs_journal_list *jl, 16271da177e4SLinus Torvalds struct buffer_chunk *chunk) 16281da177e4SLinus Torvalds { 16291da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 16301da177e4SLinus Torvalds int ret = 0; 16311da177e4SLinus Torvalds 16321da177e4SLinus Torvalds jl->j_state |= LIST_TOUCHED; 16331da177e4SLinus Torvalds del_from_work_list(s, jl); 16341da177e4SLinus Torvalds if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) { 16351da177e4SLinus Torvalds return 0; 16361da177e4SLinus Torvalds } 16371da177e4SLinus Torvalds 16381da177e4SLinus Torvalds cn = jl->j_realblock; 16391da177e4SLinus Torvalds while (cn) { 1640098297b2SJeff Mahoney /* 1641098297b2SJeff Mahoney * if the blocknr == 0, this has been cleared from the hash, 1642098297b2SJeff Mahoney * skip it 16431da177e4SLinus Torvalds */ 16441da177e4SLinus Torvalds if (cn->blocknr == 0) { 16451da177e4SLinus Torvalds goto next; 16461da177e4SLinus Torvalds } 16471da177e4SLinus Torvalds if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) { 16481da177e4SLinus Torvalds struct buffer_head *tmp_bh; 1649098297b2SJeff Mahoney /* 1650098297b2SJeff Mahoney * we can race against journal_mark_freed when we try 16511da177e4SLinus Torvalds * to lock_buffer(cn->bh), so we have to inc the buffer 16521da177e4SLinus Torvalds * count, and recheck things after locking 16531da177e4SLinus Torvalds */ 16541da177e4SLinus Torvalds tmp_bh = cn->bh; 16551da177e4SLinus Torvalds get_bh(tmp_bh); 16561da177e4SLinus Torvalds lock_buffer(tmp_bh); 16571da177e4SLinus Torvalds if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) { 16581da177e4SLinus Torvalds if (!buffer_journal_dirty(tmp_bh) || 16591da177e4SLinus Torvalds buffer_journal_prepared(tmp_bh)) 16601da177e4SLinus Torvalds BUG(); 16611da177e4SLinus Torvalds add_to_chunk(chunk, tmp_bh, NULL, write_chunk); 16621da177e4SLinus Torvalds ret++; 16631da177e4SLinus Torvalds } else { 16641da177e4SLinus Torvalds /* note, cn->bh might be null now */ 16651da177e4SLinus Torvalds unlock_buffer(tmp_bh); 16661da177e4SLinus Torvalds } 16671da177e4SLinus Torvalds put_bh(tmp_bh); 16681da177e4SLinus Torvalds } 16691da177e4SLinus Torvalds next: 16701da177e4SLinus Torvalds cn = cn->next; 16711da177e4SLinus Torvalds cond_resched(); 16721da177e4SLinus Torvalds } 16731da177e4SLinus Torvalds return ret; 16741da177e4SLinus Torvalds } 16751da177e4SLinus Torvalds 16761da177e4SLinus Torvalds /* used by flush_commit_list */ 1677dc0dde61SHariprasad Kelam static void dirty_one_transaction(struct super_block *s, 16781da177e4SLinus Torvalds struct reiserfs_journal_list *jl) 16791da177e4SLinus Torvalds { 16801da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 16811da177e4SLinus Torvalds struct reiserfs_journal_list *pjl; 16821da177e4SLinus Torvalds 16831da177e4SLinus Torvalds jl->j_state |= LIST_DIRTY; 16841da177e4SLinus Torvalds cn = jl->j_realblock; 16851da177e4SLinus Torvalds while (cn) { 1686098297b2SJeff Mahoney /* 1687098297b2SJeff Mahoney * look for a more recent transaction that logged this 1688098297b2SJeff Mahoney * buffer. Only the most recent transaction with a buffer in 1689098297b2SJeff Mahoney * it is allowed to send that buffer to disk 16901da177e4SLinus Torvalds */ 16911da177e4SLinus Torvalds pjl = find_newer_jl_for_cn(cn); 1692bd4c625cSLinus Torvalds if (!pjl && cn->blocknr && cn->bh 1693bd4c625cSLinus Torvalds && buffer_journal_dirty(cn->bh)) { 16941da177e4SLinus Torvalds BUG_ON(!can_dirty(cn)); 1695098297b2SJeff Mahoney /* 1696098297b2SJeff Mahoney * if the buffer is prepared, it will either be logged 16971da177e4SLinus Torvalds * or restored. If restored, we need to make sure 16981da177e4SLinus Torvalds * it actually gets marked dirty 16991da177e4SLinus Torvalds */ 17001da177e4SLinus Torvalds clear_buffer_journal_new(cn->bh); 17011da177e4SLinus Torvalds if (buffer_journal_prepared(cn->bh)) { 17021da177e4SLinus Torvalds set_buffer_journal_restore_dirty(cn->bh); 17031da177e4SLinus Torvalds } else { 17041da177e4SLinus Torvalds set_buffer_journal_test(cn->bh); 17051da177e4SLinus Torvalds mark_buffer_dirty(cn->bh); 17061da177e4SLinus Torvalds } 17071da177e4SLinus Torvalds } 17081da177e4SLinus Torvalds cn = cn->next; 17091da177e4SLinus Torvalds } 17101da177e4SLinus Torvalds } 17111da177e4SLinus Torvalds 17121da177e4SLinus Torvalds static int kupdate_transactions(struct super_block *s, 17131da177e4SLinus Torvalds struct reiserfs_journal_list *jl, 17141da177e4SLinus Torvalds struct reiserfs_journal_list **next_jl, 1715600ed416SJeff Mahoney unsigned int *next_trans_id, 1716bd4c625cSLinus Torvalds int num_blocks, int num_trans) 1717bd4c625cSLinus Torvalds { 17181da177e4SLinus Torvalds int ret = 0; 17191da177e4SLinus Torvalds int written = 0; 17201da177e4SLinus Torvalds int transactions_flushed = 0; 1721600ed416SJeff Mahoney unsigned int orig_trans_id = jl->j_trans_id; 17221da177e4SLinus Torvalds struct buffer_chunk chunk; 17231da177e4SLinus Torvalds struct list_head *entry; 17241da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 17251da177e4SLinus Torvalds chunk.nr = 0; 17261da177e4SLinus Torvalds 1727a412f9efSFrederic Weisbecker reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s); 17281da177e4SLinus Torvalds if (!journal_list_still_alive(s, orig_trans_id)) { 17291da177e4SLinus Torvalds goto done; 17301da177e4SLinus Torvalds } 17311da177e4SLinus Torvalds 1732098297b2SJeff Mahoney /* 1733098297b2SJeff Mahoney * we've got j_flush_mutex held, nobody is going to delete any 17341da177e4SLinus Torvalds * of these lists out from underneath us 17351da177e4SLinus Torvalds */ 17361da177e4SLinus Torvalds while ((num_trans && transactions_flushed < num_trans) || 17371da177e4SLinus Torvalds (!num_trans && written < num_blocks)) { 17381da177e4SLinus Torvalds 17391da177e4SLinus Torvalds if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) || 1740bd4c625cSLinus Torvalds atomic_read(&jl->j_commit_left) 1741bd4c625cSLinus Torvalds || !(jl->j_state & LIST_DIRTY)) { 17421da177e4SLinus Torvalds del_from_work_list(s, jl); 17431da177e4SLinus Torvalds break; 17441da177e4SLinus Torvalds } 17451da177e4SLinus Torvalds ret = write_one_transaction(s, jl, &chunk); 17461da177e4SLinus Torvalds 17471da177e4SLinus Torvalds if (ret < 0) 17481da177e4SLinus Torvalds goto done; 17491da177e4SLinus Torvalds transactions_flushed++; 17501da177e4SLinus Torvalds written += ret; 17511da177e4SLinus Torvalds entry = jl->j_list.next; 17521da177e4SLinus Torvalds 17531da177e4SLinus Torvalds /* did we wrap? */ 17541da177e4SLinus Torvalds if (entry == &journal->j_journal_list) { 17551da177e4SLinus Torvalds break; 17561da177e4SLinus Torvalds } 17571da177e4SLinus Torvalds jl = JOURNAL_LIST_ENTRY(entry); 17581da177e4SLinus Torvalds 17591da177e4SLinus Torvalds /* don't bother with older transactions */ 17601da177e4SLinus Torvalds if (jl->j_trans_id <= orig_trans_id) 17611da177e4SLinus Torvalds break; 17621da177e4SLinus Torvalds } 17631da177e4SLinus Torvalds if (chunk.nr) { 17641da177e4SLinus Torvalds write_chunk(&chunk); 17651da177e4SLinus Torvalds } 17661da177e4SLinus Torvalds 17671da177e4SLinus Torvalds done: 1768afe70259SJeff Mahoney mutex_unlock(&journal->j_flush_mutex); 17691da177e4SLinus Torvalds return ret; 17701da177e4SLinus Torvalds } 17711da177e4SLinus Torvalds 1772098297b2SJeff Mahoney /* 1773098297b2SJeff Mahoney * for o_sync and fsync heavy applications, they tend to use 1774098297b2SJeff Mahoney * all the journa list slots with tiny transactions. These 1775098297b2SJeff Mahoney * trigger lots and lots of calls to update the header block, which 1776098297b2SJeff Mahoney * adds seeks and slows things down. 1777098297b2SJeff Mahoney * 1778098297b2SJeff Mahoney * This function tries to clear out a large chunk of the journal lists 1779098297b2SJeff Mahoney * at once, which makes everything faster since only the newest journal 1780098297b2SJeff Mahoney * list updates the header block 17811da177e4SLinus Torvalds */ 17821da177e4SLinus Torvalds static int flush_used_journal_lists(struct super_block *s, 1783bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl) 1784bd4c625cSLinus Torvalds { 17851da177e4SLinus Torvalds unsigned long len = 0; 17861da177e4SLinus Torvalds unsigned long cur_len; 17871da177e4SLinus Torvalds int i; 17881da177e4SLinus Torvalds int limit = 256; 17891da177e4SLinus Torvalds struct reiserfs_journal_list *tjl; 17901da177e4SLinus Torvalds struct reiserfs_journal_list *flush_jl; 1791600ed416SJeff Mahoney unsigned int trans_id; 17921da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 17931da177e4SLinus Torvalds 17941da177e4SLinus Torvalds flush_jl = tjl = jl; 17951da177e4SLinus Torvalds 17961da177e4SLinus Torvalds /* in data logging mode, try harder to flush a lot of blocks */ 17971da177e4SLinus Torvalds if (reiserfs_data_log(s)) 17981da177e4SLinus Torvalds limit = 1024; 17991da177e4SLinus Torvalds /* flush for 256 transactions or limit blocks, whichever comes first */ 18001da177e4SLinus Torvalds for (i = 0; i < 256 && len < limit; i++) { 18011da177e4SLinus Torvalds if (atomic_read(&tjl->j_commit_left) || 18021da177e4SLinus Torvalds tjl->j_trans_id < jl->j_trans_id) { 18031da177e4SLinus Torvalds break; 18041da177e4SLinus Torvalds } 18051da177e4SLinus Torvalds cur_len = atomic_read(&tjl->j_nonzerolen); 18061da177e4SLinus Torvalds if (cur_len > 0) { 18071da177e4SLinus Torvalds tjl->j_state &= ~LIST_TOUCHED; 18081da177e4SLinus Torvalds } 18091da177e4SLinus Torvalds len += cur_len; 18101da177e4SLinus Torvalds flush_jl = tjl; 18111da177e4SLinus Torvalds if (tjl->j_list.next == &journal->j_journal_list) 18121da177e4SLinus Torvalds break; 18131da177e4SLinus Torvalds tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next); 18141da177e4SLinus Torvalds } 1815721a769cSJeff Mahoney get_journal_list(jl); 1816721a769cSJeff Mahoney get_journal_list(flush_jl); 1817098297b2SJeff Mahoney 1818098297b2SJeff Mahoney /* 1819098297b2SJeff Mahoney * try to find a group of blocks we can flush across all the 1820098297b2SJeff Mahoney * transactions, but only bother if we've actually spanned 1821098297b2SJeff Mahoney * across multiple lists 18221da177e4SLinus Torvalds */ 18233e9fd5a4SJason Yan if (flush_jl != jl) 18243e9fd5a4SJason Yan kupdate_transactions(s, jl, &tjl, &trans_id, len, i); 18253e9fd5a4SJason Yan 18261da177e4SLinus Torvalds flush_journal_list(s, flush_jl, 1); 1827721a769cSJeff Mahoney put_journal_list(s, flush_jl); 1828721a769cSJeff Mahoney put_journal_list(s, jl); 18291da177e4SLinus Torvalds return 0; 18301da177e4SLinus Torvalds } 18311da177e4SLinus Torvalds 18321da177e4SLinus Torvalds /* 1833098297b2SJeff Mahoney * removes any nodes in table with name block and dev as bh. 1834098297b2SJeff Mahoney * only touchs the hnext and hprev pointers. 18351da177e4SLinus Torvalds */ 18365d552ad7SBharath Vedartham static void remove_journal_hash(struct super_block *sb, 18371da177e4SLinus Torvalds struct reiserfs_journal_cnode **table, 18381da177e4SLinus Torvalds struct reiserfs_journal_list *jl, 18391da177e4SLinus Torvalds unsigned long block, int remove_freed) 18401da177e4SLinus Torvalds { 18411da177e4SLinus Torvalds struct reiserfs_journal_cnode *cur; 18421da177e4SLinus Torvalds struct reiserfs_journal_cnode **head; 18431da177e4SLinus Torvalds 18441da177e4SLinus Torvalds head = &(journal_hash(table, sb, block)); 18451da177e4SLinus Torvalds if (!head) { 18461da177e4SLinus Torvalds return; 18471da177e4SLinus Torvalds } 18481da177e4SLinus Torvalds cur = *head; 18491da177e4SLinus Torvalds while (cur) { 1850bd4c625cSLinus Torvalds if (cur->blocknr == block && cur->sb == sb 1851bd4c625cSLinus Torvalds && (jl == NULL || jl == cur->jlist) 1852bd4c625cSLinus Torvalds && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) { 18531da177e4SLinus Torvalds if (cur->hnext) { 18541da177e4SLinus Torvalds cur->hnext->hprev = cur->hprev; 18551da177e4SLinus Torvalds } 18561da177e4SLinus Torvalds if (cur->hprev) { 18571da177e4SLinus Torvalds cur->hprev->hnext = cur->hnext; 18581da177e4SLinus Torvalds } else { 18591da177e4SLinus Torvalds *head = cur->hnext; 18601da177e4SLinus Torvalds } 18611da177e4SLinus Torvalds cur->blocknr = 0; 18621da177e4SLinus Torvalds cur->sb = NULL; 18631da177e4SLinus Torvalds cur->state = 0; 1864098297b2SJeff Mahoney /* 1865098297b2SJeff Mahoney * anybody who clears the cur->bh will also 1866098297b2SJeff Mahoney * dec the nonzerolen 1867098297b2SJeff Mahoney */ 1868098297b2SJeff Mahoney if (cur->bh && cur->jlist) 1869a228bf8fSJeff Mahoney atomic_dec(&cur->jlist->j_nonzerolen); 18701da177e4SLinus Torvalds cur->bh = NULL; 18711da177e4SLinus Torvalds cur->jlist = NULL; 18721da177e4SLinus Torvalds } 18731da177e4SLinus Torvalds cur = cur->hnext; 18741da177e4SLinus Torvalds } 18751da177e4SLinus Torvalds } 18761da177e4SLinus Torvalds 1877a9dd3643SJeff Mahoney static void free_journal_ram(struct super_block *sb) 1878bd4c625cSLinus Torvalds { 1879a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 1880d739b42bSPekka Enberg kfree(journal->j_current_jl); 18811da177e4SLinus Torvalds journal->j_num_lists--; 18821da177e4SLinus Torvalds 18831da177e4SLinus Torvalds vfree(journal->j_cnode_free_orig); 1884a9dd3643SJeff Mahoney free_list_bitmaps(sb, journal->j_list_bitmap); 1885a9dd3643SJeff Mahoney free_bitmap_nodes(sb); /* must be after free_list_bitmaps */ 18861da177e4SLinus Torvalds if (journal->j_header_bh) { 18871da177e4SLinus Torvalds brelse(journal->j_header_bh); 18881da177e4SLinus Torvalds } 1889098297b2SJeff Mahoney /* 1890098297b2SJeff Mahoney * j_header_bh is on the journal dev, make sure 1891098297b2SJeff Mahoney * not to release the journal dev until we brelse j_header_bh 18921da177e4SLinus Torvalds */ 1893a9dd3643SJeff Mahoney release_journal_dev(sb, journal); 18941da177e4SLinus Torvalds vfree(journal); 18951da177e4SLinus Torvalds } 18961da177e4SLinus Torvalds 18971da177e4SLinus Torvalds /* 1898098297b2SJeff Mahoney * call on unmount. Only set error to 1 if you haven't made your way out 1899098297b2SJeff Mahoney * of read_super() yet. Any other caller must keep error at 0. 19001da177e4SLinus Torvalds */ 1901bd4c625cSLinus Torvalds static int do_journal_release(struct reiserfs_transaction_handle *th, 1902a9dd3643SJeff Mahoney struct super_block *sb, int error) 1903bd4c625cSLinus Torvalds { 19041da177e4SLinus Torvalds struct reiserfs_transaction_handle myth; 1905a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 19061da177e4SLinus Torvalds 1907098297b2SJeff Mahoney /* 1908098297b2SJeff Mahoney * we only want to flush out transactions if we were 1909098297b2SJeff Mahoney * called with error == 0 19101da177e4SLinus Torvalds */ 1911bc98a42cSDavid Howells if (!error && !sb_rdonly(sb)) { 19121da177e4SLinus Torvalds /* end the current trans */ 19131da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 191458d85426SJeff Mahoney do_journal_end(th, FLUSH_ALL); 19151da177e4SLinus Torvalds 1916098297b2SJeff Mahoney /* 1917098297b2SJeff Mahoney * make sure something gets logged to force 1918098297b2SJeff Mahoney * our way into the flush code 1919098297b2SJeff Mahoney */ 1920b491dd17SJeff Mahoney if (!journal_join(&myth, sb)) { 1921a9dd3643SJeff Mahoney reiserfs_prepare_for_journal(sb, 1922a9dd3643SJeff Mahoney SB_BUFFER_WITH_SB(sb), 1923bd4c625cSLinus Torvalds 1); 192409f1b80bSJeff Mahoney journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb)); 192558d85426SJeff Mahoney do_journal_end(&myth, FLUSH_ALL); 19261da177e4SLinus Torvalds } 19271da177e4SLinus Torvalds } 19281da177e4SLinus Torvalds 19291da177e4SLinus Torvalds /* this also catches errors during the do_journal_end above */ 19301da177e4SLinus Torvalds if (!error && reiserfs_is_journal_aborted(journal)) { 19311da177e4SLinus Torvalds memset(&myth, 0, sizeof(myth)); 1932b491dd17SJeff Mahoney if (!journal_join_abort(&myth, sb)) { 1933a9dd3643SJeff Mahoney reiserfs_prepare_for_journal(sb, 1934a9dd3643SJeff Mahoney SB_BUFFER_WITH_SB(sb), 1935bd4c625cSLinus Torvalds 1); 193609f1b80bSJeff Mahoney journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb)); 193758d85426SJeff Mahoney do_journal_end(&myth, FLUSH_ALL); 19381da177e4SLinus Torvalds } 19391da177e4SLinus Torvalds } 19401da177e4SLinus Torvalds 19418ebc4232SFrederic Weisbecker 19428ebc4232SFrederic Weisbecker /* 19438ebc4232SFrederic Weisbecker * We must release the write lock here because 19448ebc4232SFrederic Weisbecker * the workqueue job (flush_async_commit) needs this lock 19458ebc4232SFrederic Weisbecker */ 19468ebc4232SFrederic Weisbecker reiserfs_write_unlock(sb); 1947033369d1SArtem Bityutskiy 194801777836SJan Kara /* 194901777836SJan Kara * Cancel flushing of old commits. Note that neither of these works 195001777836SJan Kara * will be requeued because superblock is being shutdown and doesn't 19511751e8a6SLinus Torvalds * have SB_ACTIVE set. 195201777836SJan Kara */ 195371b0576bSJan Kara reiserfs_cancel_old_flush(sb); 195401777836SJan Kara /* wait for all commits to finish */ 195501777836SJan Kara cancel_delayed_work_sync(&SB_JOURNAL(sb)->j_work); 19561da177e4SLinus Torvalds 1957a9dd3643SJeff Mahoney free_journal_ram(sb); 19581da177e4SLinus Torvalds 19590523676dSFrederic Weisbecker reiserfs_write_lock(sb); 19600523676dSFrederic Weisbecker 19611da177e4SLinus Torvalds return 0; 19621da177e4SLinus Torvalds } 19631da177e4SLinus Torvalds 1964098297b2SJeff Mahoney /* * call on unmount. flush all journal trans, release all alloc'd ram */ 1965bd4c625cSLinus Torvalds int journal_release(struct reiserfs_transaction_handle *th, 1966a9dd3643SJeff Mahoney struct super_block *sb) 1967bd4c625cSLinus Torvalds { 1968a9dd3643SJeff Mahoney return do_journal_release(th, sb, 0); 19691da177e4SLinus Torvalds } 1970bd4c625cSLinus Torvalds 1971098297b2SJeff Mahoney /* only call from an error condition inside reiserfs_read_super! */ 1972bd4c625cSLinus Torvalds int journal_release_error(struct reiserfs_transaction_handle *th, 1973a9dd3643SJeff Mahoney struct super_block *sb) 1974bd4c625cSLinus Torvalds { 1975a9dd3643SJeff Mahoney return do_journal_release(th, sb, 1); 19761da177e4SLinus Torvalds } 19771da177e4SLinus Torvalds 1978098297b2SJeff Mahoney /* 1979098297b2SJeff Mahoney * compares description block with commit block. 1980098297b2SJeff Mahoney * returns 1 if they differ, 0 if they are the same 1981098297b2SJeff Mahoney */ 1982a9dd3643SJeff Mahoney static int journal_compare_desc_commit(struct super_block *sb, 1983bd4c625cSLinus Torvalds struct reiserfs_journal_desc *desc, 1984bd4c625cSLinus Torvalds struct reiserfs_journal_commit *commit) 1985bd4c625cSLinus Torvalds { 19861da177e4SLinus Torvalds if (get_commit_trans_id(commit) != get_desc_trans_id(desc) || 19871da177e4SLinus Torvalds get_commit_trans_len(commit) != get_desc_trans_len(desc) || 1988a9dd3643SJeff Mahoney get_commit_trans_len(commit) > SB_JOURNAL(sb)->j_trans_max || 1989bd4c625cSLinus Torvalds get_commit_trans_len(commit) <= 0) { 19901da177e4SLinus Torvalds return 1; 19911da177e4SLinus Torvalds } 19921da177e4SLinus Torvalds return 0; 19931da177e4SLinus Torvalds } 1994bd4c625cSLinus Torvalds 1995098297b2SJeff Mahoney /* 1996098297b2SJeff Mahoney * returns 0 if it did not find a description block 1997098297b2SJeff Mahoney * returns -1 if it found a corrupt commit block 1998098297b2SJeff Mahoney * returns 1 if both desc and commit were valid 1999098297b2SJeff Mahoney * NOTE: only called during fs mount 20001da177e4SLinus Torvalds */ 2001a9dd3643SJeff Mahoney static int journal_transaction_is_valid(struct super_block *sb, 2002bd4c625cSLinus Torvalds struct buffer_head *d_bh, 2003600ed416SJeff Mahoney unsigned int *oldest_invalid_trans_id, 2004bd4c625cSLinus Torvalds unsigned long *newest_mount_id) 2005bd4c625cSLinus Torvalds { 20061da177e4SLinus Torvalds struct reiserfs_journal_desc *desc; 20071da177e4SLinus Torvalds struct reiserfs_journal_commit *commit; 20081da177e4SLinus Torvalds struct buffer_head *c_bh; 20091da177e4SLinus Torvalds unsigned long offset; 20101da177e4SLinus Torvalds 20111da177e4SLinus Torvalds if (!d_bh) 20121da177e4SLinus Torvalds return 0; 20131da177e4SLinus Torvalds 20141da177e4SLinus Torvalds desc = (struct reiserfs_journal_desc *)d_bh->b_data; 2015bd4c625cSLinus Torvalds if (get_desc_trans_len(desc) > 0 2016bd4c625cSLinus Torvalds && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) { 2017bd4c625cSLinus Torvalds if (oldest_invalid_trans_id && *oldest_invalid_trans_id 2018bd4c625cSLinus Torvalds && get_desc_trans_id(desc) > *oldest_invalid_trans_id) { 2019a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2020bd4c625cSLinus Torvalds "journal-986: transaction " 20211da177e4SLinus Torvalds "is valid returning because trans_id %d is greater than " 2022bd4c625cSLinus Torvalds "oldest_invalid %lu", 2023bd4c625cSLinus Torvalds get_desc_trans_id(desc), 20241da177e4SLinus Torvalds *oldest_invalid_trans_id); 20251da177e4SLinus Torvalds return 0; 20261da177e4SLinus Torvalds } 2027bd4c625cSLinus Torvalds if (newest_mount_id 2028bd4c625cSLinus Torvalds && *newest_mount_id > get_desc_mount_id(desc)) { 2029a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2030bd4c625cSLinus Torvalds "journal-1087: transaction " 20311da177e4SLinus Torvalds "is valid returning because mount_id %d is less than " 2032bd4c625cSLinus Torvalds "newest_mount_id %lu", 2033bd4c625cSLinus Torvalds get_desc_mount_id(desc), 20341da177e4SLinus Torvalds *newest_mount_id); 20351da177e4SLinus Torvalds return -1; 20361da177e4SLinus Torvalds } 2037a9dd3643SJeff Mahoney if (get_desc_trans_len(desc) > SB_JOURNAL(sb)->j_trans_max) { 2038a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-2018", 203945b03d5eSJeff Mahoney "Bad transaction length %d " 204045b03d5eSJeff Mahoney "encountered, ignoring transaction", 2041bd4c625cSLinus Torvalds get_desc_trans_len(desc)); 20421da177e4SLinus Torvalds return -1; 20431da177e4SLinus Torvalds } 2044a9dd3643SJeff Mahoney offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb); 20451da177e4SLinus Torvalds 2046098297b2SJeff Mahoney /* 2047098297b2SJeff Mahoney * ok, we have a journal description block, 2048098297b2SJeff Mahoney * let's see if the transaction was valid 2049098297b2SJeff Mahoney */ 2050bd4c625cSLinus Torvalds c_bh = 2051a9dd3643SJeff Mahoney journal_bread(sb, 2052a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2053bd4c625cSLinus Torvalds ((offset + get_desc_trans_len(desc) + 2054a9dd3643SJeff Mahoney 1) % SB_ONDISK_JOURNAL_SIZE(sb))); 20551da177e4SLinus Torvalds if (!c_bh) 20561da177e4SLinus Torvalds return 0; 20571da177e4SLinus Torvalds commit = (struct reiserfs_journal_commit *)c_bh->b_data; 2058a9dd3643SJeff Mahoney if (journal_compare_desc_commit(sb, desc, commit)) { 2059a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 20601da177e4SLinus Torvalds "journal_transaction_is_valid, commit offset %ld had bad " 20611da177e4SLinus Torvalds "time %d or length %d", 2062bd4c625cSLinus Torvalds c_bh->b_blocknr - 2063a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb), 20641da177e4SLinus Torvalds get_commit_trans_id(commit), 20651da177e4SLinus Torvalds get_commit_trans_len(commit)); 20661da177e4SLinus Torvalds brelse(c_bh); 20671da177e4SLinus Torvalds if (oldest_invalid_trans_id) { 2068bd4c625cSLinus Torvalds *oldest_invalid_trans_id = 2069bd4c625cSLinus Torvalds get_desc_trans_id(desc); 2070a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2071bd4c625cSLinus Torvalds "journal-1004: " 20721da177e4SLinus Torvalds "transaction_is_valid setting oldest invalid trans_id " 2073bd4c625cSLinus Torvalds "to %d", 2074bd4c625cSLinus Torvalds get_desc_trans_id(desc)); 20751da177e4SLinus Torvalds } 20761da177e4SLinus Torvalds return -1; 20771da177e4SLinus Torvalds } 20781da177e4SLinus Torvalds brelse(c_bh); 2079a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2080bd4c625cSLinus Torvalds "journal-1006: found valid " 20811da177e4SLinus Torvalds "transaction start offset %llu, len %d id %d", 2082bd4c625cSLinus Torvalds d_bh->b_blocknr - 2083a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb), 2084bd4c625cSLinus Torvalds get_desc_trans_len(desc), 2085bd4c625cSLinus Torvalds get_desc_trans_id(desc)); 20861da177e4SLinus Torvalds return 1; 20871da177e4SLinus Torvalds } else { 20881da177e4SLinus Torvalds return 0; 20891da177e4SLinus Torvalds } 20901da177e4SLinus Torvalds } 20911da177e4SLinus Torvalds 2092bd4c625cSLinus Torvalds static void brelse_array(struct buffer_head **heads, int num) 2093bd4c625cSLinus Torvalds { 20941da177e4SLinus Torvalds int i; 20951da177e4SLinus Torvalds for (i = 0; i < num; i++) { 20961da177e4SLinus Torvalds brelse(heads[i]); 20971da177e4SLinus Torvalds } 20981da177e4SLinus Torvalds } 20991da177e4SLinus Torvalds 21001da177e4SLinus Torvalds /* 2101098297b2SJeff Mahoney * given the start, and values for the oldest acceptable transactions, 2102098297b2SJeff Mahoney * this either reads in a replays a transaction, or returns because the 2103098297b2SJeff Mahoney * transaction is invalid, or too old. 2104098297b2SJeff Mahoney * NOTE: only called during fs mount 21051da177e4SLinus Torvalds */ 2106a9dd3643SJeff Mahoney static int journal_read_transaction(struct super_block *sb, 2107bd4c625cSLinus Torvalds unsigned long cur_dblock, 2108bd4c625cSLinus Torvalds unsigned long oldest_start, 2109600ed416SJeff Mahoney unsigned int oldest_trans_id, 2110bd4c625cSLinus Torvalds unsigned long newest_mount_id) 2111bd4c625cSLinus Torvalds { 2112a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 21131da177e4SLinus Torvalds struct reiserfs_journal_desc *desc; 21141da177e4SLinus Torvalds struct reiserfs_journal_commit *commit; 2115600ed416SJeff Mahoney unsigned int trans_id = 0; 21161da177e4SLinus Torvalds struct buffer_head *c_bh; 21171da177e4SLinus Torvalds struct buffer_head *d_bh; 21181da177e4SLinus Torvalds struct buffer_head **log_blocks = NULL; 21191da177e4SLinus Torvalds struct buffer_head **real_blocks = NULL; 2120600ed416SJeff Mahoney unsigned int trans_offset; 21211da177e4SLinus Torvalds int i; 21221da177e4SLinus Torvalds int trans_half; 21231da177e4SLinus Torvalds 2124a9dd3643SJeff Mahoney d_bh = journal_bread(sb, cur_dblock); 21251da177e4SLinus Torvalds if (!d_bh) 21261da177e4SLinus Torvalds return 1; 21271da177e4SLinus Torvalds desc = (struct reiserfs_journal_desc *)d_bh->b_data; 2128a9dd3643SJeff Mahoney trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb); 2129a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1037: " 21301da177e4SLinus Torvalds "journal_read_transaction, offset %llu, len %d mount_id %d", 2131a9dd3643SJeff Mahoney d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb), 21321da177e4SLinus Torvalds get_desc_trans_len(desc), get_desc_mount_id(desc)); 21331da177e4SLinus Torvalds if (get_desc_trans_id(desc) < oldest_trans_id) { 2134a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1039: " 21351da177e4SLinus Torvalds "journal_read_trans skipping because %lu is too old", 2136bd4c625cSLinus Torvalds cur_dblock - 2137a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb)); 21381da177e4SLinus Torvalds brelse(d_bh); 21391da177e4SLinus Torvalds return 1; 21401da177e4SLinus Torvalds } 21411da177e4SLinus Torvalds if (get_desc_mount_id(desc) != newest_mount_id) { 2142a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1146: " 21431da177e4SLinus Torvalds "journal_read_trans skipping because %d is != " 21441da177e4SLinus Torvalds "newest_mount_id %lu", get_desc_mount_id(desc), 21451da177e4SLinus Torvalds newest_mount_id); 21461da177e4SLinus Torvalds brelse(d_bh); 21471da177e4SLinus Torvalds return 1; 21481da177e4SLinus Torvalds } 2149a9dd3643SJeff Mahoney c_bh = journal_bread(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 21501da177e4SLinus Torvalds ((trans_offset + get_desc_trans_len(desc) + 1) % 2151a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb))); 21521da177e4SLinus Torvalds if (!c_bh) { 21531da177e4SLinus Torvalds brelse(d_bh); 21541da177e4SLinus Torvalds return 1; 21551da177e4SLinus Torvalds } 21561da177e4SLinus Torvalds commit = (struct reiserfs_journal_commit *)c_bh->b_data; 2157a9dd3643SJeff Mahoney if (journal_compare_desc_commit(sb, desc, commit)) { 2158a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2159bd4c625cSLinus Torvalds "journal_read_transaction, " 21601da177e4SLinus Torvalds "commit offset %llu had bad time %d or length %d", 2161bd4c625cSLinus Torvalds c_bh->b_blocknr - 2162a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb), 2163bd4c625cSLinus Torvalds get_commit_trans_id(commit), 2164bd4c625cSLinus Torvalds get_commit_trans_len(commit)); 21651da177e4SLinus Torvalds brelse(c_bh); 21661da177e4SLinus Torvalds brelse(d_bh); 21671da177e4SLinus Torvalds return 1; 21681da177e4SLinus Torvalds } 21693f8b5ee3SJeff Mahoney 21703f8b5ee3SJeff Mahoney if (bdev_read_only(sb->s_bdev)) { 21713f8b5ee3SJeff Mahoney reiserfs_warning(sb, "clm-2076", 21723f8b5ee3SJeff Mahoney "device is readonly, unable to replay log"); 21733f8b5ee3SJeff Mahoney brelse(c_bh); 21743f8b5ee3SJeff Mahoney brelse(d_bh); 21753f8b5ee3SJeff Mahoney return -EROFS; 21763f8b5ee3SJeff Mahoney } 21773f8b5ee3SJeff Mahoney 21781da177e4SLinus Torvalds trans_id = get_desc_trans_id(desc); 2179098297b2SJeff Mahoney /* 2180098297b2SJeff Mahoney * now we know we've got a good transaction, and it was 2181098297b2SJeff Mahoney * inside the valid time ranges 2182098297b2SJeff Mahoney */ 21836da2ec56SKees Cook log_blocks = kmalloc_array(get_desc_trans_len(desc), 21846da2ec56SKees Cook sizeof(struct buffer_head *), 21856da2ec56SKees Cook GFP_NOFS); 21866da2ec56SKees Cook real_blocks = kmalloc_array(get_desc_trans_len(desc), 21876da2ec56SKees Cook sizeof(struct buffer_head *), 21886da2ec56SKees Cook GFP_NOFS); 21891da177e4SLinus Torvalds if (!log_blocks || !real_blocks) { 21901da177e4SLinus Torvalds brelse(c_bh); 21911da177e4SLinus Torvalds brelse(d_bh); 2192d739b42bSPekka Enberg kfree(log_blocks); 2193d739b42bSPekka Enberg kfree(real_blocks); 2194a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1169", 219545b03d5eSJeff Mahoney "kmalloc failed, unable to mount FS"); 21961da177e4SLinus Torvalds return -1; 21971da177e4SLinus Torvalds } 21981da177e4SLinus Torvalds /* get all the buffer heads */ 2199a9dd3643SJeff Mahoney trans_half = journal_trans_half(sb->s_blocksize); 22001da177e4SLinus Torvalds for (i = 0; i < get_desc_trans_len(desc); i++) { 2201bd4c625cSLinus Torvalds log_blocks[i] = 2202a9dd3643SJeff Mahoney journal_getblk(sb, 2203a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2204bd4c625cSLinus Torvalds (trans_offset + 1 + 2205a9dd3643SJeff Mahoney i) % SB_ONDISK_JOURNAL_SIZE(sb)); 22061da177e4SLinus Torvalds if (i < trans_half) { 2207bd4c625cSLinus Torvalds real_blocks[i] = 2208a9dd3643SJeff Mahoney sb_getblk(sb, 2209bd4c625cSLinus Torvalds le32_to_cpu(desc->j_realblock[i])); 22101da177e4SLinus Torvalds } else { 2211bd4c625cSLinus Torvalds real_blocks[i] = 2212a9dd3643SJeff Mahoney sb_getblk(sb, 2213bd4c625cSLinus Torvalds le32_to_cpu(commit-> 2214bd4c625cSLinus Torvalds j_realblock[i - trans_half])); 22151da177e4SLinus Torvalds } 2216a9dd3643SJeff Mahoney if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(sb)) { 2217a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1207", 221845b03d5eSJeff Mahoney "REPLAY FAILURE fsck required! " 221945b03d5eSJeff Mahoney "Block to replay is outside of " 222045b03d5eSJeff Mahoney "filesystem"); 22211da177e4SLinus Torvalds goto abort_replay; 22221da177e4SLinus Torvalds } 22231da177e4SLinus Torvalds /* make sure we don't try to replay onto log or reserved area */ 2224bd4c625cSLinus Torvalds if (is_block_in_log_or_reserved_area 2225a9dd3643SJeff Mahoney (sb, real_blocks[i]->b_blocknr)) { 2226a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1204", 222745b03d5eSJeff Mahoney "REPLAY FAILURE fsck required! " 222845b03d5eSJeff Mahoney "Trying to replay onto a log block"); 22291da177e4SLinus Torvalds abort_replay: 22301da177e4SLinus Torvalds brelse_array(log_blocks, i); 22311da177e4SLinus Torvalds brelse_array(real_blocks, i); 22321da177e4SLinus Torvalds brelse(c_bh); 22331da177e4SLinus Torvalds brelse(d_bh); 2234d739b42bSPekka Enberg kfree(log_blocks); 2235d739b42bSPekka Enberg kfree(real_blocks); 22361da177e4SLinus Torvalds return -1; 22371da177e4SLinus Torvalds } 22381da177e4SLinus Torvalds } 22391da177e4SLinus Torvalds /* read in the log blocks, memcpy to the corresponding real block */ 2240dfec8a14SMike Christie ll_rw_block(REQ_OP_READ, 0, get_desc_trans_len(desc), log_blocks); 22411da177e4SLinus Torvalds for (i = 0; i < get_desc_trans_len(desc); i++) { 22428ebc4232SFrederic Weisbecker 22431da177e4SLinus Torvalds wait_on_buffer(log_blocks[i]); 22441da177e4SLinus Torvalds if (!buffer_uptodate(log_blocks[i])) { 2245a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1212", 224645b03d5eSJeff Mahoney "REPLAY FAILURE fsck required! " 224745b03d5eSJeff Mahoney "buffer write failed"); 2248bd4c625cSLinus Torvalds brelse_array(log_blocks + i, 2249bd4c625cSLinus Torvalds get_desc_trans_len(desc) - i); 22501da177e4SLinus Torvalds brelse_array(real_blocks, get_desc_trans_len(desc)); 22511da177e4SLinus Torvalds brelse(c_bh); 22521da177e4SLinus Torvalds brelse(d_bh); 2253d739b42bSPekka Enberg kfree(log_blocks); 2254d739b42bSPekka Enberg kfree(real_blocks); 22551da177e4SLinus Torvalds return -1; 22561da177e4SLinus Torvalds } 2257bd4c625cSLinus Torvalds memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data, 2258bd4c625cSLinus Torvalds real_blocks[i]->b_size); 22591da177e4SLinus Torvalds set_buffer_uptodate(real_blocks[i]); 22601da177e4SLinus Torvalds brelse(log_blocks[i]); 22611da177e4SLinus Torvalds } 22621da177e4SLinus Torvalds /* flush out the real blocks */ 22631da177e4SLinus Torvalds for (i = 0; i < get_desc_trans_len(desc); i++) { 22641da177e4SLinus Torvalds set_buffer_dirty(real_blocks[i]); 22652a222ca9SMike Christie write_dirty_buffer(real_blocks[i], 0); 22661da177e4SLinus Torvalds } 22671da177e4SLinus Torvalds for (i = 0; i < get_desc_trans_len(desc); i++) { 22681da177e4SLinus Torvalds wait_on_buffer(real_blocks[i]); 22691da177e4SLinus Torvalds if (!buffer_uptodate(real_blocks[i])) { 2270a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1226", 227145b03d5eSJeff Mahoney "REPLAY FAILURE, fsck required! " 227245b03d5eSJeff Mahoney "buffer write failed"); 2273bd4c625cSLinus Torvalds brelse_array(real_blocks + i, 2274bd4c625cSLinus Torvalds get_desc_trans_len(desc) - i); 22751da177e4SLinus Torvalds brelse(c_bh); 22761da177e4SLinus Torvalds brelse(d_bh); 2277d739b42bSPekka Enberg kfree(log_blocks); 2278d739b42bSPekka Enberg kfree(real_blocks); 22791da177e4SLinus Torvalds return -1; 22801da177e4SLinus Torvalds } 22811da177e4SLinus Torvalds brelse(real_blocks[i]); 22821da177e4SLinus Torvalds } 2283bd4c625cSLinus Torvalds cur_dblock = 2284a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2285bd4c625cSLinus Torvalds ((trans_offset + get_desc_trans_len(desc) + 2286a9dd3643SJeff Mahoney 2) % SB_ONDISK_JOURNAL_SIZE(sb)); 2287a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2288bd4c625cSLinus Torvalds "journal-1095: setting journal " "start to offset %ld", 2289a9dd3643SJeff Mahoney cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb)); 22901da177e4SLinus Torvalds 2291098297b2SJeff Mahoney /* 2292098297b2SJeff Mahoney * init starting values for the first transaction, in case 2293098297b2SJeff Mahoney * this is the last transaction to be replayed. 2294098297b2SJeff Mahoney */ 2295a9dd3643SJeff Mahoney journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb); 22961da177e4SLinus Torvalds journal->j_last_flush_trans_id = trans_id; 22971da177e4SLinus Torvalds journal->j_trans_id = trans_id + 1; 2298a44c94a7SAlexander Zarochentsev /* check for trans_id overflow */ 2299a44c94a7SAlexander Zarochentsev if (journal->j_trans_id == 0) 2300a44c94a7SAlexander Zarochentsev journal->j_trans_id = 10; 23011da177e4SLinus Torvalds brelse(c_bh); 23021da177e4SLinus Torvalds brelse(d_bh); 2303d739b42bSPekka Enberg kfree(log_blocks); 2304d739b42bSPekka Enberg kfree(real_blocks); 23051da177e4SLinus Torvalds return 0; 23061da177e4SLinus Torvalds } 23071da177e4SLinus Torvalds 2308098297b2SJeff Mahoney /* 2309098297b2SJeff Mahoney * This function reads blocks starting from block and to max_block of bufsize 2310098297b2SJeff Mahoney * size (but no more than BUFNR blocks at a time). This proved to improve 2311098297b2SJeff Mahoney * mounting speed on self-rebuilding raid5 arrays at least. 2312098297b2SJeff Mahoney * Right now it is only used from journal code. But later we might use it 2313098297b2SJeff Mahoney * from other places. 2314098297b2SJeff Mahoney * Note: Do not use journal_getblk/sb_getblk functions here! 2315098297b2SJeff Mahoney */ 23163ee16670SJeff Mahoney static struct buffer_head *reiserfs_breada(struct block_device *dev, 23173ee16670SJeff Mahoney b_blocknr_t block, int bufsize, 23183ee16670SJeff Mahoney b_blocknr_t max_block) 23191da177e4SLinus Torvalds { 23201da177e4SLinus Torvalds struct buffer_head *bhlist[BUFNR]; 23211da177e4SLinus Torvalds unsigned int blocks = BUFNR; 23221da177e4SLinus Torvalds struct buffer_head *bh; 23231da177e4SLinus Torvalds int i, j; 23241da177e4SLinus Torvalds 23251da177e4SLinus Torvalds bh = __getblk(dev, block, bufsize); 23261da177e4SLinus Torvalds if (buffer_uptodate(bh)) 23271da177e4SLinus Torvalds return (bh); 23281da177e4SLinus Torvalds 23291da177e4SLinus Torvalds if (block + BUFNR > max_block) { 23301da177e4SLinus Torvalds blocks = max_block - block; 23311da177e4SLinus Torvalds } 23321da177e4SLinus Torvalds bhlist[0] = bh; 23331da177e4SLinus Torvalds j = 1; 23341da177e4SLinus Torvalds for (i = 1; i < blocks; i++) { 23351da177e4SLinus Torvalds bh = __getblk(dev, block + i, bufsize); 23361da177e4SLinus Torvalds if (buffer_uptodate(bh)) { 23371da177e4SLinus Torvalds brelse(bh); 23381da177e4SLinus Torvalds break; 2339bd4c625cSLinus Torvalds } else 2340bd4c625cSLinus Torvalds bhlist[j++] = bh; 23411da177e4SLinus Torvalds } 2342dfec8a14SMike Christie ll_rw_block(REQ_OP_READ, 0, j, bhlist); 23431da177e4SLinus Torvalds for (i = 1; i < j; i++) 23441da177e4SLinus Torvalds brelse(bhlist[i]); 23451da177e4SLinus Torvalds bh = bhlist[0]; 23461da177e4SLinus Torvalds wait_on_buffer(bh); 23471da177e4SLinus Torvalds if (buffer_uptodate(bh)) 23481da177e4SLinus Torvalds return bh; 23491da177e4SLinus Torvalds brelse(bh); 23501da177e4SLinus Torvalds return NULL; 23511da177e4SLinus Torvalds } 23521da177e4SLinus Torvalds 23531da177e4SLinus Torvalds /* 2354098297b2SJeff Mahoney * read and replay the log 2355098297b2SJeff Mahoney * on a clean unmount, the journal header's next unflushed pointer will be 2356098297b2SJeff Mahoney * to an invalid transaction. This tests that before finding all the 2357098297b2SJeff Mahoney * transactions in the log, which makes normal mount times fast. 2358098297b2SJeff Mahoney * 2359098297b2SJeff Mahoney * After a crash, this starts with the next unflushed transaction, and 2360098297b2SJeff Mahoney * replays until it finds one too old, or invalid. 2361098297b2SJeff Mahoney * 2362098297b2SJeff Mahoney * On exit, it sets things up so the first transaction will work correctly. 2363098297b2SJeff Mahoney * NOTE: only called during fs mount 23641da177e4SLinus Torvalds */ 2365a9dd3643SJeff Mahoney static int journal_read(struct super_block *sb) 2366bd4c625cSLinus Torvalds { 2367a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 23681da177e4SLinus Torvalds struct reiserfs_journal_desc *desc; 2369600ed416SJeff Mahoney unsigned int oldest_trans_id = 0; 2370600ed416SJeff Mahoney unsigned int oldest_invalid_trans_id = 0; 237134d08260SArnd Bergmann time64_t start; 23721da177e4SLinus Torvalds unsigned long oldest_start = 0; 23731da177e4SLinus Torvalds unsigned long cur_dblock = 0; 23741da177e4SLinus Torvalds unsigned long newest_mount_id = 9; 23751da177e4SLinus Torvalds struct buffer_head *d_bh; 23761da177e4SLinus Torvalds struct reiserfs_journal_header *jh; 23771da177e4SLinus Torvalds int valid_journal_header = 0; 23781da177e4SLinus Torvalds int replay_count = 0; 23791da177e4SLinus Torvalds int continue_replay = 1; 23801da177e4SLinus Torvalds int ret; 23811da177e4SLinus Torvalds 2382a9dd3643SJeff Mahoney cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(sb); 2383a1c6f057SDmitry Monakhov reiserfs_info(sb, "checking transaction log (%pg)\n", 2384a1c6f057SDmitry Monakhov journal->j_dev_bd); 238534d08260SArnd Bergmann start = ktime_get_seconds(); 23861da177e4SLinus Torvalds 2387098297b2SJeff Mahoney /* 2388098297b2SJeff Mahoney * step 1, read in the journal header block. Check the transaction 2389098297b2SJeff Mahoney * it says is the first unflushed, and if that transaction is not 2390098297b2SJeff Mahoney * valid, replay is done 23911da177e4SLinus Torvalds */ 2392a9dd3643SJeff Mahoney journal->j_header_bh = journal_bread(sb, 2393a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) 2394a9dd3643SJeff Mahoney + SB_ONDISK_JOURNAL_SIZE(sb)); 23951da177e4SLinus Torvalds if (!journal->j_header_bh) { 23961da177e4SLinus Torvalds return 1; 23971da177e4SLinus Torvalds } 23981da177e4SLinus Torvalds jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data); 2399c499ec24SVladimir V. Saveliev if (le32_to_cpu(jh->j_first_unflushed_offset) < 2400a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb) 2401bd4c625cSLinus Torvalds && le32_to_cpu(jh->j_last_flush_trans_id) > 0) { 2402bd4c625cSLinus Torvalds oldest_start = 2403a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 24041da177e4SLinus Torvalds le32_to_cpu(jh->j_first_unflushed_offset); 24051da177e4SLinus Torvalds oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1; 24061da177e4SLinus Torvalds newest_mount_id = le32_to_cpu(jh->j_mount_id); 2407a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2408bd4c625cSLinus Torvalds "journal-1153: found in " 24091da177e4SLinus Torvalds "header: first_unflushed_offset %d, last_flushed_trans_id " 24101da177e4SLinus Torvalds "%lu", le32_to_cpu(jh->j_first_unflushed_offset), 24111da177e4SLinus Torvalds le32_to_cpu(jh->j_last_flush_trans_id)); 24121da177e4SLinus Torvalds valid_journal_header = 1; 24131da177e4SLinus Torvalds 2414098297b2SJeff Mahoney /* 2415098297b2SJeff Mahoney * now, we try to read the first unflushed offset. If it 2416098297b2SJeff Mahoney * is not valid, there is nothing more we can do, and it 2417098297b2SJeff Mahoney * makes no sense to read through the whole log. 24181da177e4SLinus Torvalds */ 2419bd4c625cSLinus Torvalds d_bh = 2420a9dd3643SJeff Mahoney journal_bread(sb, 2421a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2422bd4c625cSLinus Torvalds le32_to_cpu(jh->j_first_unflushed_offset)); 2423a9dd3643SJeff Mahoney ret = journal_transaction_is_valid(sb, d_bh, NULL, NULL); 24241da177e4SLinus Torvalds if (!ret) { 24251da177e4SLinus Torvalds continue_replay = 0; 24261da177e4SLinus Torvalds } 24271da177e4SLinus Torvalds brelse(d_bh); 24281da177e4SLinus Torvalds goto start_log_replay; 24291da177e4SLinus Torvalds } 24301da177e4SLinus Torvalds 2431098297b2SJeff Mahoney /* 2432098297b2SJeff Mahoney * ok, there are transactions that need to be replayed. start 2433098297b2SJeff Mahoney * with the first log block, find all the valid transactions, and 2434098297b2SJeff Mahoney * pick out the oldest. 24351da177e4SLinus Torvalds */ 2436bd4c625cSLinus Torvalds while (continue_replay 2437bd4c625cSLinus Torvalds && cur_dblock < 2438a9dd3643SJeff Mahoney (SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2439a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb))) { 2440098297b2SJeff Mahoney /* 2441098297b2SJeff Mahoney * Note that it is required for blocksize of primary fs 2442098297b2SJeff Mahoney * device and journal device to be the same 2443098297b2SJeff Mahoney */ 2444bd4c625cSLinus Torvalds d_bh = 2445bd4c625cSLinus Torvalds reiserfs_breada(journal->j_dev_bd, cur_dblock, 2446a9dd3643SJeff Mahoney sb->s_blocksize, 2447a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2448a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb)); 2449bd4c625cSLinus Torvalds ret = 2450a9dd3643SJeff Mahoney journal_transaction_is_valid(sb, d_bh, 2451bd4c625cSLinus Torvalds &oldest_invalid_trans_id, 2452bd4c625cSLinus Torvalds &newest_mount_id); 24531da177e4SLinus Torvalds if (ret == 1) { 24541da177e4SLinus Torvalds desc = (struct reiserfs_journal_desc *)d_bh->b_data; 24551da177e4SLinus Torvalds if (oldest_start == 0) { /* init all oldest_ values */ 24561da177e4SLinus Torvalds oldest_trans_id = get_desc_trans_id(desc); 24571da177e4SLinus Torvalds oldest_start = d_bh->b_blocknr; 24581da177e4SLinus Torvalds newest_mount_id = get_desc_mount_id(desc); 2459a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2460bd4c625cSLinus Torvalds "journal-1179: Setting " 24611da177e4SLinus Torvalds "oldest_start to offset %llu, trans_id %lu", 2462bd4c625cSLinus Torvalds oldest_start - 2463bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK 2464a9dd3643SJeff Mahoney (sb), oldest_trans_id); 24651da177e4SLinus Torvalds } else if (oldest_trans_id > get_desc_trans_id(desc)) { 24661da177e4SLinus Torvalds /* one we just read was older */ 24671da177e4SLinus Torvalds oldest_trans_id = get_desc_trans_id(desc); 24681da177e4SLinus Torvalds oldest_start = d_bh->b_blocknr; 2469a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2470bd4c625cSLinus Torvalds "journal-1180: Resetting " 24711da177e4SLinus Torvalds "oldest_start to offset %lu, trans_id %lu", 2472bd4c625cSLinus Torvalds oldest_start - 2473bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK 2474a9dd3643SJeff Mahoney (sb), oldest_trans_id); 24751da177e4SLinus Torvalds } 24761da177e4SLinus Torvalds if (newest_mount_id < get_desc_mount_id(desc)) { 24771da177e4SLinus Torvalds newest_mount_id = get_desc_mount_id(desc); 2478a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2479bd4c625cSLinus Torvalds "journal-1299: Setting " 2480bd4c625cSLinus Torvalds "newest_mount_id to %d", 2481bd4c625cSLinus Torvalds get_desc_mount_id(desc)); 24821da177e4SLinus Torvalds } 24831da177e4SLinus Torvalds cur_dblock += get_desc_trans_len(desc) + 2; 24841da177e4SLinus Torvalds } else { 24851da177e4SLinus Torvalds cur_dblock++; 24861da177e4SLinus Torvalds } 24871da177e4SLinus Torvalds brelse(d_bh); 24881da177e4SLinus Torvalds } 24891da177e4SLinus Torvalds 24901da177e4SLinus Torvalds start_log_replay: 24911da177e4SLinus Torvalds cur_dblock = oldest_start; 24921da177e4SLinus Torvalds if (oldest_trans_id) { 2493a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2494bd4c625cSLinus Torvalds "journal-1206: Starting replay " 24951da177e4SLinus Torvalds "from offset %llu, trans_id %lu", 2496a9dd3643SJeff Mahoney cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb), 24971da177e4SLinus Torvalds oldest_trans_id); 24981da177e4SLinus Torvalds 24991da177e4SLinus Torvalds } 25001da177e4SLinus Torvalds replay_count = 0; 25011da177e4SLinus Torvalds while (continue_replay && oldest_trans_id > 0) { 2502bd4c625cSLinus Torvalds ret = 2503a9dd3643SJeff Mahoney journal_read_transaction(sb, cur_dblock, oldest_start, 2504bd4c625cSLinus Torvalds oldest_trans_id, newest_mount_id); 25051da177e4SLinus Torvalds if (ret < 0) { 25061da177e4SLinus Torvalds return ret; 25071da177e4SLinus Torvalds } else if (ret != 0) { 25081da177e4SLinus Torvalds break; 25091da177e4SLinus Torvalds } 2510bd4c625cSLinus Torvalds cur_dblock = 2511a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + journal->j_start; 25121da177e4SLinus Torvalds replay_count++; 25131da177e4SLinus Torvalds if (cur_dblock == oldest_start) 25141da177e4SLinus Torvalds break; 25151da177e4SLinus Torvalds } 25161da177e4SLinus Torvalds 25171da177e4SLinus Torvalds if (oldest_trans_id == 0) { 2518a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2519bd4c625cSLinus Torvalds "journal-1225: No valid " "transactions found"); 25201da177e4SLinus Torvalds } 2521098297b2SJeff Mahoney /* 2522098297b2SJeff Mahoney * j_start does not get set correctly if we don't replay any 2523098297b2SJeff Mahoney * transactions. if we had a valid journal_header, set j_start 2524098297b2SJeff Mahoney * to the first unflushed transaction value, copy the trans_id 2525098297b2SJeff Mahoney * from the header 25261da177e4SLinus Torvalds */ 25271da177e4SLinus Torvalds if (valid_journal_header && replay_count == 0) { 25281da177e4SLinus Torvalds journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset); 2529bd4c625cSLinus Torvalds journal->j_trans_id = 2530bd4c625cSLinus Torvalds le32_to_cpu(jh->j_last_flush_trans_id) + 1; 2531a44c94a7SAlexander Zarochentsev /* check for trans_id overflow */ 2532a44c94a7SAlexander Zarochentsev if (journal->j_trans_id == 0) 2533a44c94a7SAlexander Zarochentsev journal->j_trans_id = 10; 2534bd4c625cSLinus Torvalds journal->j_last_flush_trans_id = 2535bd4c625cSLinus Torvalds le32_to_cpu(jh->j_last_flush_trans_id); 25361da177e4SLinus Torvalds journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1; 25371da177e4SLinus Torvalds } else { 25381da177e4SLinus Torvalds journal->j_mount_id = newest_mount_id + 1; 25391da177e4SLinus Torvalds } 2540a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1299: Setting " 25411da177e4SLinus Torvalds "newest_mount_id to %lu", journal->j_mount_id); 25421da177e4SLinus Torvalds journal->j_first_unflushed_offset = journal->j_start; 25431da177e4SLinus Torvalds if (replay_count > 0) { 2544a9dd3643SJeff Mahoney reiserfs_info(sb, 2545bd4c625cSLinus Torvalds "replayed %d transactions in %lu seconds\n", 254634d08260SArnd Bergmann replay_count, ktime_get_seconds() - start); 25471da177e4SLinus Torvalds } 2548278f6679SJeff Mahoney /* needed to satisfy the locking in _update_journal_header_block */ 2549278f6679SJeff Mahoney reiserfs_write_lock(sb); 2550a9dd3643SJeff Mahoney if (!bdev_read_only(sb->s_bdev) && 2551a9dd3643SJeff Mahoney _update_journal_header_block(sb, journal->j_start, 2552bd4c625cSLinus Torvalds journal->j_last_flush_trans_id)) { 2553278f6679SJeff Mahoney reiserfs_write_unlock(sb); 2554098297b2SJeff Mahoney /* 2555098297b2SJeff Mahoney * replay failed, caller must call free_journal_ram and abort 2556098297b2SJeff Mahoney * the mount 25571da177e4SLinus Torvalds */ 25581da177e4SLinus Torvalds return -1; 25591da177e4SLinus Torvalds } 2560278f6679SJeff Mahoney reiserfs_write_unlock(sb); 25611da177e4SLinus Torvalds return 0; 25621da177e4SLinus Torvalds } 25631da177e4SLinus Torvalds 25641da177e4SLinus Torvalds static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s) 25651da177e4SLinus Torvalds { 25661da177e4SLinus Torvalds struct reiserfs_journal_list *jl; 25678c777cc4SPekka Enberg jl = kzalloc(sizeof(struct reiserfs_journal_list), 25688c777cc4SPekka Enberg GFP_NOFS | __GFP_NOFAIL); 25691da177e4SLinus Torvalds INIT_LIST_HEAD(&jl->j_list); 25701da177e4SLinus Torvalds INIT_LIST_HEAD(&jl->j_working_list); 25711da177e4SLinus Torvalds INIT_LIST_HEAD(&jl->j_tail_bh_list); 25721da177e4SLinus Torvalds INIT_LIST_HEAD(&jl->j_bh_list); 257390415deaSJeff Mahoney mutex_init(&jl->j_commit_mutex); 25741da177e4SLinus Torvalds SB_JOURNAL(s)->j_num_lists++; 25751da177e4SLinus Torvalds get_journal_list(jl); 25761da177e4SLinus Torvalds return jl; 25771da177e4SLinus Torvalds } 25781da177e4SLinus Torvalds 2579a9dd3643SJeff Mahoney static void journal_list_init(struct super_block *sb) 2580bd4c625cSLinus Torvalds { 2581a9dd3643SJeff Mahoney SB_JOURNAL(sb)->j_current_jl = alloc_journal_list(sb); 25821da177e4SLinus Torvalds } 25831da177e4SLinus Torvalds 25844385bab1SAl Viro static void release_journal_dev(struct super_block *super, 25851da177e4SLinus Torvalds struct reiserfs_journal *journal) 25861da177e4SLinus Torvalds { 258786098fa0SChristoph Hellwig if (journal->j_dev_bd != NULL) { 25884385bab1SAl Viro blkdev_put(journal->j_dev_bd, journal->j_dev_mode); 25891da177e4SLinus Torvalds journal->j_dev_bd = NULL; 25901da177e4SLinus Torvalds } 25911da177e4SLinus Torvalds } 25921da177e4SLinus Torvalds 25931da177e4SLinus Torvalds static int journal_init_dev(struct super_block *super, 25941da177e4SLinus Torvalds struct reiserfs_journal *journal, 25951da177e4SLinus Torvalds const char *jdev_name) 25961da177e4SLinus Torvalds { 25971da177e4SLinus Torvalds int result; 25981da177e4SLinus Torvalds dev_t jdev; 2599e525fd89STejun Heo fmode_t blkdev_mode = FMODE_READ | FMODE_WRITE | FMODE_EXCL; 26001da177e4SLinus Torvalds 26011da177e4SLinus Torvalds result = 0; 26021da177e4SLinus Torvalds 26031da177e4SLinus Torvalds journal->j_dev_bd = NULL; 26041da177e4SLinus Torvalds jdev = SB_ONDISK_JOURNAL_DEVICE(super) ? 26051da177e4SLinus Torvalds new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev; 26061da177e4SLinus Torvalds 26071da177e4SLinus Torvalds if (bdev_read_only(super->s_bdev)) 26081da177e4SLinus Torvalds blkdev_mode = FMODE_READ; 26091da177e4SLinus Torvalds 26101da177e4SLinus Torvalds /* there is no "jdev" option and journal is on separate device */ 26111da177e4SLinus Torvalds if ((!jdev_name || !jdev_name[0])) { 2612e525fd89STejun Heo if (jdev == super->s_dev) 2613e525fd89STejun Heo blkdev_mode &= ~FMODE_EXCL; 2614d4d77629STejun Heo journal->j_dev_bd = blkdev_get_by_dev(jdev, blkdev_mode, 2615d4d77629STejun Heo journal); 2616e5eb8caaSAl Viro journal->j_dev_mode = blkdev_mode; 26171da177e4SLinus Torvalds if (IS_ERR(journal->j_dev_bd)) { 26181da177e4SLinus Torvalds result = PTR_ERR(journal->j_dev_bd); 26191da177e4SLinus Torvalds journal->j_dev_bd = NULL; 262045b03d5eSJeff Mahoney reiserfs_warning(super, "sh-458", 2621ea3edd4dSChristoph Hellwig "cannot init journal device unknown-block(%u,%u): %i", 2622ea3edd4dSChristoph Hellwig MAJOR(jdev), MINOR(jdev), result); 26231da177e4SLinus Torvalds return result; 2624e525fd89STejun Heo } else if (jdev != super->s_dev) 26251da177e4SLinus Torvalds set_blocksize(journal->j_dev_bd, super->s_blocksize); 262686098fa0SChristoph Hellwig 26271da177e4SLinus Torvalds return 0; 26281da177e4SLinus Torvalds } 26291da177e4SLinus Torvalds 2630e5eb8caaSAl Viro journal->j_dev_mode = blkdev_mode; 2631d4d77629STejun Heo journal->j_dev_bd = blkdev_get_by_path(jdev_name, blkdev_mode, journal); 263286098fa0SChristoph Hellwig if (IS_ERR(journal->j_dev_bd)) { 263386098fa0SChristoph Hellwig result = PTR_ERR(journal->j_dev_bd); 263486098fa0SChristoph Hellwig journal->j_dev_bd = NULL; 26359ad553abSAndrew Morton reiserfs_warning(super, "sh-457", 263686098fa0SChristoph Hellwig "journal_init_dev: Cannot open '%s': %i", 263786098fa0SChristoph Hellwig jdev_name, result); 263886098fa0SChristoph Hellwig return result; 263986098fa0SChristoph Hellwig } 264086098fa0SChristoph Hellwig 26411da177e4SLinus Torvalds set_blocksize(journal->j_dev_bd, super->s_blocksize); 2642bd4c625cSLinus Torvalds reiserfs_info(super, 2643a1c6f057SDmitry Monakhov "journal_init_dev: journal device: %pg\n", 2644a1c6f057SDmitry Monakhov journal->j_dev_bd); 264586098fa0SChristoph Hellwig return 0; 26461da177e4SLinus Torvalds } 26471da177e4SLinus Torvalds 2648098297b2SJeff Mahoney /* 2649cf3d0b81SEdward Shishkin * When creating/tuning a file system user can assign some 2650cf3d0b81SEdward Shishkin * journal params within boundaries which depend on the ratio 2651cf3d0b81SEdward Shishkin * blocksize/standard_blocksize. 2652cf3d0b81SEdward Shishkin * 2653cf3d0b81SEdward Shishkin * For blocks >= standard_blocksize transaction size should 2654cf3d0b81SEdward Shishkin * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more 2655cf3d0b81SEdward Shishkin * then JOURNAL_TRANS_MAX_DEFAULT. 2656cf3d0b81SEdward Shishkin * 2657cf3d0b81SEdward Shishkin * For blocks < standard_blocksize these boundaries should be 2658cf3d0b81SEdward Shishkin * decreased proportionally. 2659cf3d0b81SEdward Shishkin */ 2660cf3d0b81SEdward Shishkin #define REISERFS_STANDARD_BLKSIZE (4096) 2661cf3d0b81SEdward Shishkin 2662a9dd3643SJeff Mahoney static int check_advise_trans_params(struct super_block *sb, 2663cf3d0b81SEdward Shishkin struct reiserfs_journal *journal) 2664cf3d0b81SEdward Shishkin { 2665cf3d0b81SEdward Shishkin if (journal->j_trans_max) { 2666098297b2SJeff Mahoney /* Non-default journal params. Do sanity check for them. */ 2667cf3d0b81SEdward Shishkin int ratio = 1; 2668a9dd3643SJeff Mahoney if (sb->s_blocksize < REISERFS_STANDARD_BLKSIZE) 2669a9dd3643SJeff Mahoney ratio = REISERFS_STANDARD_BLKSIZE / sb->s_blocksize; 2670cf3d0b81SEdward Shishkin 2671cf3d0b81SEdward Shishkin if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio || 2672cf3d0b81SEdward Shishkin journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio || 2673a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb) / journal->j_trans_max < 2674cf3d0b81SEdward Shishkin JOURNAL_MIN_RATIO) { 2675a9dd3643SJeff Mahoney reiserfs_warning(sb, "sh-462", 267645b03d5eSJeff Mahoney "bad transaction max size (%u). " 267745b03d5eSJeff Mahoney "FSCK?", journal->j_trans_max); 2678cf3d0b81SEdward Shishkin return 1; 2679cf3d0b81SEdward Shishkin } 2680cf3d0b81SEdward Shishkin if (journal->j_max_batch != (journal->j_trans_max) * 2681cf3d0b81SEdward Shishkin JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) { 2682a9dd3643SJeff Mahoney reiserfs_warning(sb, "sh-463", 268345b03d5eSJeff Mahoney "bad transaction max batch (%u). " 268445b03d5eSJeff Mahoney "FSCK?", journal->j_max_batch); 2685cf3d0b81SEdward Shishkin return 1; 2686cf3d0b81SEdward Shishkin } 2687cf3d0b81SEdward Shishkin } else { 2688098297b2SJeff Mahoney /* 2689098297b2SJeff Mahoney * Default journal params. 2690098297b2SJeff Mahoney * The file system was created by old version 2691098297b2SJeff Mahoney * of mkreiserfs, so some fields contain zeros, 2692098297b2SJeff Mahoney * and we need to advise proper values for them 2693098297b2SJeff Mahoney */ 2694a9dd3643SJeff Mahoney if (sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) { 2695a9dd3643SJeff Mahoney reiserfs_warning(sb, "sh-464", "bad blocksize (%u)", 2696a9dd3643SJeff Mahoney sb->s_blocksize); 269745b03d5eSJeff Mahoney return 1; 269845b03d5eSJeff Mahoney } 2699cf3d0b81SEdward Shishkin journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT; 2700cf3d0b81SEdward Shishkin journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT; 2701cf3d0b81SEdward Shishkin journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE; 2702cf3d0b81SEdward Shishkin } 2703cf3d0b81SEdward Shishkin return 0; 2704cf3d0b81SEdward Shishkin } 2705cf3d0b81SEdward Shishkin 2706098297b2SJeff Mahoney /* must be called once on fs mount. calls journal_read for you */ 2707a9dd3643SJeff Mahoney int journal_init(struct super_block *sb, const char *j_dev_name, 2708bd4c625cSLinus Torvalds int old_format, unsigned int commit_max_age) 2709bd4c625cSLinus Torvalds { 2710a9dd3643SJeff Mahoney int num_cnodes = SB_ONDISK_JOURNAL_SIZE(sb) * 2; 27111da177e4SLinus Torvalds struct buffer_head *bhjh; 27121da177e4SLinus Torvalds struct reiserfs_super_block *rs; 27131da177e4SLinus Torvalds struct reiserfs_journal_header *jh; 27141da177e4SLinus Torvalds struct reiserfs_journal *journal; 27151da177e4SLinus Torvalds struct reiserfs_journal_list *jl; 271698ea3f50SFrederic Weisbecker int ret; 27171da177e4SLinus Torvalds 2718558feb08SJoe Perches journal = SB_JOURNAL(sb) = vzalloc(sizeof(struct reiserfs_journal)); 27191da177e4SLinus Torvalds if (!journal) { 2720a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1256", 272145b03d5eSJeff Mahoney "unable to get memory for journal structure"); 27221da177e4SLinus Torvalds return 1; 27231da177e4SLinus Torvalds } 27241da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_bitmap_nodes); 27251da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_prealloc_list); 27261da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_working_list); 27271da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_journal_list); 27281da177e4SLinus Torvalds journal->j_persistent_trans = 0; 272937c69b98SFrederic Weisbecker if (reiserfs_allocate_list_bitmaps(sb, journal->j_list_bitmap, 273037c69b98SFrederic Weisbecker reiserfs_bmap_count(sb))) 27311da177e4SLinus Torvalds goto free_and_return; 273298ea3f50SFrederic Weisbecker 2733a9dd3643SJeff Mahoney allocate_bitmap_nodes(sb); 27341da177e4SLinus Torvalds 27351da177e4SLinus Torvalds /* reserved for journal area support */ 2736a9dd3643SJeff Mahoney SB_JOURNAL_1st_RESERVED_BLOCK(sb) = (old_format ? 2737bd4c625cSLinus Torvalds REISERFS_OLD_DISK_OFFSET_IN_BYTES 2738a9dd3643SJeff Mahoney / sb->s_blocksize + 2739a9dd3643SJeff Mahoney reiserfs_bmap_count(sb) + 2740bd4c625cSLinus Torvalds 1 : 2741bd4c625cSLinus Torvalds REISERFS_DISK_OFFSET_IN_BYTES / 2742a9dd3643SJeff Mahoney sb->s_blocksize + 2); 27431da177e4SLinus Torvalds 2744098297b2SJeff Mahoney /* 2745098297b2SJeff Mahoney * Sanity check to see is the standard journal fitting 2746098297b2SJeff Mahoney * within first bitmap (actual for small blocksizes) 2747098297b2SJeff Mahoney */ 2748a9dd3643SJeff Mahoney if (!SB_ONDISK_JOURNAL_DEVICE(sb) && 2749a9dd3643SJeff Mahoney (SB_JOURNAL_1st_RESERVED_BLOCK(sb) + 2750a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb) > sb->s_blocksize * 8)) { 2751a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1393", 275245b03d5eSJeff Mahoney "journal does not fit for area addressed " 275345b03d5eSJeff Mahoney "by first of bitmap blocks. It starts at " 27541da177e4SLinus Torvalds "%u and its size is %u. Block size %ld", 2755a9dd3643SJeff Mahoney SB_JOURNAL_1st_RESERVED_BLOCK(sb), 2756a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb), 2757a9dd3643SJeff Mahoney sb->s_blocksize); 27581da177e4SLinus Torvalds goto free_and_return; 27591da177e4SLinus Torvalds } 27601da177e4SLinus Torvalds 2761a9dd3643SJeff Mahoney if (journal_init_dev(sb, journal, j_dev_name) != 0) { 2762a9dd3643SJeff Mahoney reiserfs_warning(sb, "sh-462", 27636774def6SMasanari Iida "unable to initialize journal device"); 27641da177e4SLinus Torvalds goto free_and_return; 27651da177e4SLinus Torvalds } 27661da177e4SLinus Torvalds 2767a9dd3643SJeff Mahoney rs = SB_DISK_SUPER_BLOCK(sb); 27681da177e4SLinus Torvalds 27691da177e4SLinus Torvalds /* read journal header */ 2770a9dd3643SJeff Mahoney bhjh = journal_bread(sb, 2771a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2772a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb)); 27731da177e4SLinus Torvalds if (!bhjh) { 2774a9dd3643SJeff Mahoney reiserfs_warning(sb, "sh-459", 277545b03d5eSJeff Mahoney "unable to read journal header"); 27761da177e4SLinus Torvalds goto free_and_return; 27771da177e4SLinus Torvalds } 27781da177e4SLinus Torvalds jh = (struct reiserfs_journal_header *)(bhjh->b_data); 27791da177e4SLinus Torvalds 27801da177e4SLinus Torvalds /* make sure that journal matches to the super block */ 2781bd4c625cSLinus Torvalds if (is_reiserfs_jr(rs) 2782bd4c625cSLinus Torvalds && (le32_to_cpu(jh->jh_journal.jp_journal_magic) != 2783bd4c625cSLinus Torvalds sb_jp_journal_magic(rs))) { 2784a9dd3643SJeff Mahoney reiserfs_warning(sb, "sh-460", 2785a1c6f057SDmitry Monakhov "journal header magic %x (device %pg) does " 278645b03d5eSJeff Mahoney "not match to magic found in super block %x", 278745b03d5eSJeff Mahoney jh->jh_journal.jp_journal_magic, 2788a1c6f057SDmitry Monakhov journal->j_dev_bd, 27891da177e4SLinus Torvalds sb_jp_journal_magic(rs)); 27901da177e4SLinus Torvalds brelse(bhjh); 27911da177e4SLinus Torvalds goto free_and_return; 27921da177e4SLinus Torvalds } 27931da177e4SLinus Torvalds 27941da177e4SLinus Torvalds journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max); 27951da177e4SLinus Torvalds journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch); 2796bd4c625cSLinus Torvalds journal->j_max_commit_age = 2797bd4c625cSLinus Torvalds le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age); 27981da177e4SLinus Torvalds journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE; 27991da177e4SLinus Torvalds 2800a9dd3643SJeff Mahoney if (check_advise_trans_params(sb, journal) != 0) 2801cf3d0b81SEdward Shishkin goto free_and_return; 28021da177e4SLinus Torvalds journal->j_default_max_commit_age = journal->j_max_commit_age; 28031da177e4SLinus Torvalds 28041da177e4SLinus Torvalds if (commit_max_age != 0) { 28051da177e4SLinus Torvalds journal->j_max_commit_age = commit_max_age; 28061da177e4SLinus Torvalds journal->j_max_trans_age = commit_max_age; 28071da177e4SLinus Torvalds } 28081da177e4SLinus Torvalds 2809a1c6f057SDmitry Monakhov reiserfs_info(sb, "journal params: device %pg, size %u, " 28101da177e4SLinus Torvalds "journal first block %u, max trans len %u, max batch %u, " 28111da177e4SLinus Torvalds "max commit age %u, max trans age %u\n", 2812a1c6f057SDmitry Monakhov journal->j_dev_bd, 2813a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb), 2814a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb), 28151da177e4SLinus Torvalds journal->j_trans_max, 28161da177e4SLinus Torvalds journal->j_max_batch, 2817bd4c625cSLinus Torvalds journal->j_max_commit_age, journal->j_max_trans_age); 28181da177e4SLinus Torvalds 28191da177e4SLinus Torvalds brelse(bhjh); 28201da177e4SLinus Torvalds 28211da177e4SLinus Torvalds journal->j_list_bitmap_index = 0; 2822a9dd3643SJeff Mahoney journal_list_init(sb); 28231da177e4SLinus Torvalds 2824bd4c625cSLinus Torvalds memset(journal->j_list_hash_table, 0, 2825bd4c625cSLinus Torvalds JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *)); 28261da177e4SLinus Torvalds 28271da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_dirty_buffers); 28281da177e4SLinus Torvalds spin_lock_init(&journal->j_dirty_buffers_lock); 28291da177e4SLinus Torvalds 28301da177e4SLinus Torvalds journal->j_start = 0; 28311da177e4SLinus Torvalds journal->j_len = 0; 28321da177e4SLinus Torvalds journal->j_len_alloc = 0; 2833a228bf8fSJeff Mahoney atomic_set(&journal->j_wcount, 0); 2834a228bf8fSJeff Mahoney atomic_set(&journal->j_async_throttle, 0); 28351da177e4SLinus Torvalds journal->j_bcount = 0; 28361da177e4SLinus Torvalds journal->j_trans_start_time = 0; 28371da177e4SLinus Torvalds journal->j_last = NULL; 28381da177e4SLinus Torvalds journal->j_first = NULL; 2839a228bf8fSJeff Mahoney init_waitqueue_head(&journal->j_join_wait); 2840f68215c4SJeff Mahoney mutex_init(&journal->j_mutex); 2841afe70259SJeff Mahoney mutex_init(&journal->j_flush_mutex); 28421da177e4SLinus Torvalds 28431da177e4SLinus Torvalds journal->j_trans_id = 10; 28441da177e4SLinus Torvalds journal->j_mount_id = 10; 28451da177e4SLinus Torvalds journal->j_state = 0; 2846a228bf8fSJeff Mahoney atomic_set(&journal->j_jlock, 0); 28471da177e4SLinus Torvalds journal->j_cnode_free_list = allocate_cnodes(num_cnodes); 28481da177e4SLinus Torvalds journal->j_cnode_free_orig = journal->j_cnode_free_list; 28491da177e4SLinus Torvalds journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0; 28501da177e4SLinus Torvalds journal->j_cnode_used = 0; 28511da177e4SLinus Torvalds journal->j_must_wait = 0; 28521da177e4SLinus Torvalds 2853576f6d79SJeff Mahoney if (journal->j_cnode_free == 0) { 2854a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-2004", "Journal cnode memory " 2855576f6d79SJeff Mahoney "allocation failed (%ld bytes). Journal is " 2856576f6d79SJeff Mahoney "too large for available memory. Usually " 2857576f6d79SJeff Mahoney "this is due to a journal that is too large.", 2858576f6d79SJeff Mahoney sizeof (struct reiserfs_journal_cnode) * num_cnodes); 2859576f6d79SJeff Mahoney goto free_and_return; 2860576f6d79SJeff Mahoney } 2861576f6d79SJeff Mahoney 2862a9dd3643SJeff Mahoney init_journal_hash(sb); 28631da177e4SLinus Torvalds jl = journal->j_current_jl; 286437c69b98SFrederic Weisbecker 286537c69b98SFrederic Weisbecker /* 286637c69b98SFrederic Weisbecker * get_list_bitmap() may call flush_commit_list() which 286737c69b98SFrederic Weisbecker * requires the lock. Calling flush_commit_list() shouldn't happen 286837c69b98SFrederic Weisbecker * this early but I like to be paranoid. 286937c69b98SFrederic Weisbecker */ 287037c69b98SFrederic Weisbecker reiserfs_write_lock(sb); 2871a9dd3643SJeff Mahoney jl->j_list_bitmap = get_list_bitmap(sb, jl); 287237c69b98SFrederic Weisbecker reiserfs_write_unlock(sb); 28731da177e4SLinus Torvalds if (!jl->j_list_bitmap) { 2874a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-2005", 287545b03d5eSJeff Mahoney "get_list_bitmap failed for journal list 0"); 28761da177e4SLinus Torvalds goto free_and_return; 28771da177e4SLinus Torvalds } 287837c69b98SFrederic Weisbecker 287937c69b98SFrederic Weisbecker ret = journal_read(sb); 288037c69b98SFrederic Weisbecker if (ret < 0) { 2881a9dd3643SJeff Mahoney reiserfs_warning(sb, "reiserfs-2006", 288245b03d5eSJeff Mahoney "Replay Failure, unable to mount"); 28831da177e4SLinus Torvalds goto free_and_return; 28841da177e4SLinus Torvalds } 28851da177e4SLinus Torvalds 2886c4028958SDavid Howells INIT_DELAYED_WORK(&journal->j_work, flush_async_commits); 2887a9dd3643SJeff Mahoney journal->j_work_sb = sb; 28881da177e4SLinus Torvalds return 0; 28891da177e4SLinus Torvalds free_and_return: 2890a9dd3643SJeff Mahoney free_journal_ram(sb); 28911da177e4SLinus Torvalds return 1; 28921da177e4SLinus Torvalds } 28931da177e4SLinus Torvalds 28941da177e4SLinus Torvalds /* 2895098297b2SJeff Mahoney * test for a polite end of the current transaction. Used by file_write, 2896098297b2SJeff Mahoney * and should be used by delete to make sure they don't write more than 2897098297b2SJeff Mahoney * can fit inside a single transaction 28981da177e4SLinus Torvalds */ 2899bd4c625cSLinus Torvalds int journal_transaction_should_end(struct reiserfs_transaction_handle *th, 2900bd4c625cSLinus Torvalds int new_alloc) 2901bd4c625cSLinus Torvalds { 29021da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(th->t_super); 290334d08260SArnd Bergmann time64_t now = ktime_get_seconds(); 29041da177e4SLinus Torvalds /* cannot restart while nested */ 29051da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 29061da177e4SLinus Torvalds if (th->t_refcount > 1) 29071da177e4SLinus Torvalds return 0; 29081da177e4SLinus Torvalds if (journal->j_must_wait > 0 || 29091da177e4SLinus Torvalds (journal->j_len_alloc + new_alloc) >= journal->j_max_batch || 2910a228bf8fSJeff Mahoney atomic_read(&journal->j_jlock) || 29111da177e4SLinus Torvalds (now - journal->j_trans_start_time) > journal->j_max_trans_age || 29121da177e4SLinus Torvalds journal->j_cnode_free < (journal->j_trans_max * 3)) { 29131da177e4SLinus Torvalds return 1; 29141da177e4SLinus Torvalds } 2915b18c1c6eSDavidlohr Bueso 29166ae1ea44SChris Mason journal->j_len_alloc += new_alloc; 29176ae1ea44SChris Mason th->t_blocks_allocated += new_alloc ; 29181da177e4SLinus Torvalds return 0; 29191da177e4SLinus Torvalds } 29201da177e4SLinus Torvalds 2921098297b2SJeff Mahoney /* this must be called inside a transaction */ 2922bd4c625cSLinus Torvalds void reiserfs_block_writes(struct reiserfs_transaction_handle *th) 2923bd4c625cSLinus Torvalds { 29241da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(th->t_super); 29251da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 29261da177e4SLinus Torvalds journal->j_must_wait = 1; 29271da177e4SLinus Torvalds set_bit(J_WRITERS_BLOCKED, &journal->j_state); 29281da177e4SLinus Torvalds return; 29291da177e4SLinus Torvalds } 29301da177e4SLinus Torvalds 2931098297b2SJeff Mahoney /* this must be called without a transaction started */ 2932bd4c625cSLinus Torvalds void reiserfs_allow_writes(struct super_block *s) 2933bd4c625cSLinus Torvalds { 29341da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 29351da177e4SLinus Torvalds clear_bit(J_WRITERS_BLOCKED, &journal->j_state); 29361da177e4SLinus Torvalds wake_up(&journal->j_join_wait); 29371da177e4SLinus Torvalds } 29381da177e4SLinus Torvalds 2939098297b2SJeff Mahoney /* this must be called without a transaction started */ 2940bd4c625cSLinus Torvalds void reiserfs_wait_on_write_block(struct super_block *s) 2941bd4c625cSLinus Torvalds { 29421da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 29431da177e4SLinus Torvalds wait_event(journal->j_join_wait, 29441da177e4SLinus Torvalds !test_bit(J_WRITERS_BLOCKED, &journal->j_state)); 29451da177e4SLinus Torvalds } 29461da177e4SLinus Torvalds 2947bd4c625cSLinus Torvalds static void queue_log_writer(struct super_block *s) 2948bd4c625cSLinus Torvalds { 2949ac6424b9SIngo Molnar wait_queue_entry_t wait; 29501da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 29511da177e4SLinus Torvalds set_bit(J_WRITERS_QUEUED, &journal->j_state); 29521da177e4SLinus Torvalds 29531da177e4SLinus Torvalds /* 29541da177e4SLinus Torvalds * we don't want to use wait_event here because 29551da177e4SLinus Torvalds * we only want to wait once. 29561da177e4SLinus Torvalds */ 29571da177e4SLinus Torvalds init_waitqueue_entry(&wait, current); 29581da177e4SLinus Torvalds add_wait_queue(&journal->j_join_wait, &wait); 29591da177e4SLinus Torvalds set_current_state(TASK_UNINTERRUPTIBLE); 29608ebc4232SFrederic Weisbecker if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) { 2961278f6679SJeff Mahoney int depth = reiserfs_write_unlock_nested(s); 29621da177e4SLinus Torvalds schedule(); 2963278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 29648ebc4232SFrederic Weisbecker } 29655ab2f7e0SMilind Arun Choudhary __set_current_state(TASK_RUNNING); 29661da177e4SLinus Torvalds remove_wait_queue(&journal->j_join_wait, &wait); 29671da177e4SLinus Torvalds } 29681da177e4SLinus Torvalds 2969bd4c625cSLinus Torvalds static void wake_queued_writers(struct super_block *s) 2970bd4c625cSLinus Torvalds { 29711da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 29721da177e4SLinus Torvalds if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state)) 29731da177e4SLinus Torvalds wake_up(&journal->j_join_wait); 29741da177e4SLinus Torvalds } 29751da177e4SLinus Torvalds 2976600ed416SJeff Mahoney static void let_transaction_grow(struct super_block *sb, unsigned int trans_id) 29771da177e4SLinus Torvalds { 29781da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(sb); 29791da177e4SLinus Torvalds unsigned long bcount = journal->j_bcount; 29801da177e4SLinus Torvalds while (1) { 2981278f6679SJeff Mahoney int depth; 2982278f6679SJeff Mahoney 2983278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(sb); 2984041e0e3bSNishanth Aravamudan schedule_timeout_uninterruptible(1); 2985278f6679SJeff Mahoney reiserfs_write_lock_nested(sb, depth); 2986278f6679SJeff Mahoney 29871da177e4SLinus Torvalds journal->j_current_jl->j_state |= LIST_COMMIT_PENDING; 29881da177e4SLinus Torvalds while ((atomic_read(&journal->j_wcount) > 0 || 29891da177e4SLinus Torvalds atomic_read(&journal->j_jlock)) && 29901da177e4SLinus Torvalds journal->j_trans_id == trans_id) { 29911da177e4SLinus Torvalds queue_log_writer(sb); 29921da177e4SLinus Torvalds } 29931da177e4SLinus Torvalds if (journal->j_trans_id != trans_id) 29941da177e4SLinus Torvalds break; 29951da177e4SLinus Torvalds if (bcount == journal->j_bcount) 29961da177e4SLinus Torvalds break; 29971da177e4SLinus Torvalds bcount = journal->j_bcount; 29981da177e4SLinus Torvalds } 29991da177e4SLinus Torvalds } 30001da177e4SLinus Torvalds 3001098297b2SJeff Mahoney /* 3002098297b2SJeff Mahoney * join == true if you must join an existing transaction. 3003098297b2SJeff Mahoney * join == false if you can deal with waiting for others to finish 3004098297b2SJeff Mahoney * 3005098297b2SJeff Mahoney * this will block until the transaction is joinable. send the number of 3006098297b2SJeff Mahoney * blocks you expect to use in nblocks. 30071da177e4SLinus Torvalds */ 3008bd4c625cSLinus Torvalds static int do_journal_begin_r(struct reiserfs_transaction_handle *th, 3009a9dd3643SJeff Mahoney struct super_block *sb, unsigned long nblocks, 3010bd4c625cSLinus Torvalds int join) 3011bd4c625cSLinus Torvalds { 301234d08260SArnd Bergmann time64_t now = ktime_get_seconds(); 3013600ed416SJeff Mahoney unsigned int old_trans_id; 3014a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 30151da177e4SLinus Torvalds struct reiserfs_transaction_handle myth; 30161da177e4SLinus Torvalds int sched_count = 0; 30171da177e4SLinus Torvalds int retval; 3018278f6679SJeff Mahoney int depth; 30191da177e4SLinus Torvalds 3020a9dd3643SJeff Mahoney reiserfs_check_lock_depth(sb, "journal_begin"); 302114a61442SEric Sesterhenn BUG_ON(nblocks > journal->j_trans_max); 30221da177e4SLinus Torvalds 3023a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.journal_being); 30241da177e4SLinus Torvalds /* set here for journal_join */ 30251da177e4SLinus Torvalds th->t_refcount = 1; 3026a9dd3643SJeff Mahoney th->t_super = sb; 30271da177e4SLinus Torvalds 30281da177e4SLinus Torvalds relock: 3029a9dd3643SJeff Mahoney lock_journal(sb); 30301da177e4SLinus Torvalds if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) { 3031a9dd3643SJeff Mahoney unlock_journal(sb); 30321da177e4SLinus Torvalds retval = journal->j_errno; 30331da177e4SLinus Torvalds goto out_fail; 30341da177e4SLinus Torvalds } 30351da177e4SLinus Torvalds journal->j_bcount++; 30361da177e4SLinus Torvalds 30371da177e4SLinus Torvalds if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) { 3038a9dd3643SJeff Mahoney unlock_journal(sb); 3039278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(sb); 3040a9dd3643SJeff Mahoney reiserfs_wait_on_write_block(sb); 3041278f6679SJeff Mahoney reiserfs_write_lock_nested(sb, depth); 3042a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.journal_relock_writers); 30431da177e4SLinus Torvalds goto relock; 30441da177e4SLinus Torvalds } 304534d08260SArnd Bergmann now = ktime_get_seconds(); 30461da177e4SLinus Torvalds 3047098297b2SJeff Mahoney /* 3048098297b2SJeff Mahoney * if there is no room in the journal OR 3049098297b2SJeff Mahoney * if this transaction is too old, and we weren't called joinable, 3050098297b2SJeff Mahoney * wait for it to finish before beginning we don't sleep if there 3051098297b2SJeff Mahoney * aren't other writers 30521da177e4SLinus Torvalds */ 30531da177e4SLinus Torvalds 30541da177e4SLinus Torvalds if ((!join && journal->j_must_wait > 0) || 3055bd4c625cSLinus Torvalds (!join 3056bd4c625cSLinus Torvalds && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch) 3057bd4c625cSLinus Torvalds || (!join && atomic_read(&journal->j_wcount) > 0 3058bd4c625cSLinus Torvalds && journal->j_trans_start_time > 0 3059bd4c625cSLinus Torvalds && (now - journal->j_trans_start_time) > 3060bd4c625cSLinus Torvalds journal->j_max_trans_age) || (!join 3061bd4c625cSLinus Torvalds && atomic_read(&journal->j_jlock)) 3062bd4c625cSLinus Torvalds || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) { 30631da177e4SLinus Torvalds 30641da177e4SLinus Torvalds old_trans_id = journal->j_trans_id; 3065098297b2SJeff Mahoney /* allow others to finish this transaction */ 3066098297b2SJeff Mahoney unlock_journal(sb); 30671da177e4SLinus Torvalds 30681da177e4SLinus Torvalds if (!join && (journal->j_len_alloc + nblocks + 2) >= 30691da177e4SLinus Torvalds journal->j_max_batch && 3070bd4c625cSLinus Torvalds ((journal->j_len + nblocks + 2) * 100) < 3071bd4c625cSLinus Torvalds (journal->j_len_alloc * 75)) { 30721da177e4SLinus Torvalds if (atomic_read(&journal->j_wcount) > 10) { 30731da177e4SLinus Torvalds sched_count++; 3074a9dd3643SJeff Mahoney queue_log_writer(sb); 30751da177e4SLinus Torvalds goto relock; 30761da177e4SLinus Torvalds } 30771da177e4SLinus Torvalds } 3078098297b2SJeff Mahoney /* 3079098297b2SJeff Mahoney * don't mess with joining the transaction if all we 3080098297b2SJeff Mahoney * have to do is wait for someone else to do a commit 30811da177e4SLinus Torvalds */ 30821da177e4SLinus Torvalds if (atomic_read(&journal->j_jlock)) { 30831da177e4SLinus Torvalds while (journal->j_trans_id == old_trans_id && 30841da177e4SLinus Torvalds atomic_read(&journal->j_jlock)) { 3085a9dd3643SJeff Mahoney queue_log_writer(sb); 30861da177e4SLinus Torvalds } 30871da177e4SLinus Torvalds goto relock; 30881da177e4SLinus Torvalds } 3089b491dd17SJeff Mahoney retval = journal_join(&myth, sb); 30901da177e4SLinus Torvalds if (retval) 30911da177e4SLinus Torvalds goto out_fail; 30921da177e4SLinus Torvalds 30931da177e4SLinus Torvalds /* someone might have ended the transaction while we joined */ 30941da177e4SLinus Torvalds if (old_trans_id != journal->j_trans_id) { 309558d85426SJeff Mahoney retval = do_journal_end(&myth, 0); 30961da177e4SLinus Torvalds } else { 309758d85426SJeff Mahoney retval = do_journal_end(&myth, COMMIT_NOW); 30981da177e4SLinus Torvalds } 30991da177e4SLinus Torvalds 31001da177e4SLinus Torvalds if (retval) 31011da177e4SLinus Torvalds goto out_fail; 31021da177e4SLinus Torvalds 3103a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.journal_relock_wcount); 31041da177e4SLinus Torvalds goto relock; 31051da177e4SLinus Torvalds } 31061da177e4SLinus Torvalds /* we are the first writer, set trans_id */ 31071da177e4SLinus Torvalds if (journal->j_trans_start_time == 0) { 310834d08260SArnd Bergmann journal->j_trans_start_time = ktime_get_seconds(); 31091da177e4SLinus Torvalds } 3110a228bf8fSJeff Mahoney atomic_inc(&journal->j_wcount); 31111da177e4SLinus Torvalds journal->j_len_alloc += nblocks; 31121da177e4SLinus Torvalds th->t_blocks_logged = 0; 31131da177e4SLinus Torvalds th->t_blocks_allocated = nblocks; 31141da177e4SLinus Torvalds th->t_trans_id = journal->j_trans_id; 3115a9dd3643SJeff Mahoney unlock_journal(sb); 31161da177e4SLinus Torvalds INIT_LIST_HEAD(&th->t_list); 31171da177e4SLinus Torvalds return 0; 31181da177e4SLinus Torvalds 31191da177e4SLinus Torvalds out_fail: 31201da177e4SLinus Torvalds memset(th, 0, sizeof(*th)); 3121098297b2SJeff Mahoney /* 3122098297b2SJeff Mahoney * Re-set th->t_super, so we can properly keep track of how many 31231da177e4SLinus Torvalds * persistent transactions there are. We need to do this so if this 3124098297b2SJeff Mahoney * call is part of a failed restart_transaction, we can free it later 3125098297b2SJeff Mahoney */ 3126a9dd3643SJeff Mahoney th->t_super = sb; 31271da177e4SLinus Torvalds return retval; 31281da177e4SLinus Torvalds } 31291da177e4SLinus Torvalds 3130bd4c625cSLinus Torvalds struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct 3131bd4c625cSLinus Torvalds super_block 3132bd4c625cSLinus Torvalds *s, 3133bd4c625cSLinus Torvalds int nblocks) 3134bd4c625cSLinus Torvalds { 31351da177e4SLinus Torvalds int ret; 31361da177e4SLinus Torvalds struct reiserfs_transaction_handle *th; 31371da177e4SLinus Torvalds 3138098297b2SJeff Mahoney /* 3139098297b2SJeff Mahoney * if we're nesting into an existing transaction. It will be 3140098297b2SJeff Mahoney * persistent on its own 31411da177e4SLinus Torvalds */ 31421da177e4SLinus Torvalds if (reiserfs_transaction_running(s)) { 31431da177e4SLinus Torvalds th = current->journal_info; 31441da177e4SLinus Torvalds th->t_refcount++; 314514a61442SEric Sesterhenn BUG_ON(th->t_refcount < 2); 314614a61442SEric Sesterhenn 31471da177e4SLinus Torvalds return th; 31481da177e4SLinus Torvalds } 3149d739b42bSPekka Enberg th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS); 31501da177e4SLinus Torvalds if (!th) 31511da177e4SLinus Torvalds return NULL; 31521da177e4SLinus Torvalds ret = journal_begin(th, s, nblocks); 31531da177e4SLinus Torvalds if (ret) { 3154d739b42bSPekka Enberg kfree(th); 31551da177e4SLinus Torvalds return NULL; 31561da177e4SLinus Torvalds } 31571da177e4SLinus Torvalds 31581da177e4SLinus Torvalds SB_JOURNAL(s)->j_persistent_trans++; 31591da177e4SLinus Torvalds return th; 31601da177e4SLinus Torvalds } 31611da177e4SLinus Torvalds 3162bd4c625cSLinus Torvalds int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th) 3163bd4c625cSLinus Torvalds { 31641da177e4SLinus Torvalds struct super_block *s = th->t_super; 31651da177e4SLinus Torvalds int ret = 0; 31661da177e4SLinus Torvalds if (th->t_trans_id) 316758d85426SJeff Mahoney ret = journal_end(th); 31681da177e4SLinus Torvalds else 31691da177e4SLinus Torvalds ret = -EIO; 31701da177e4SLinus Torvalds if (th->t_refcount == 0) { 31711da177e4SLinus Torvalds SB_JOURNAL(s)->j_persistent_trans--; 3172d739b42bSPekka Enberg kfree(th); 31731da177e4SLinus Torvalds } 31741da177e4SLinus Torvalds return ret; 31751da177e4SLinus Torvalds } 31761da177e4SLinus Torvalds 3177bd4c625cSLinus Torvalds static int journal_join(struct reiserfs_transaction_handle *th, 3178b491dd17SJeff Mahoney struct super_block *sb) 3179bd4c625cSLinus Torvalds { 31801da177e4SLinus Torvalds struct reiserfs_transaction_handle *cur_th = current->journal_info; 31811da177e4SLinus Torvalds 3182098297b2SJeff Mahoney /* 3183098297b2SJeff Mahoney * this keeps do_journal_end from NULLing out the 3184098297b2SJeff Mahoney * current->journal_info pointer 31851da177e4SLinus Torvalds */ 31861da177e4SLinus Torvalds th->t_handle_save = cur_th; 318714a61442SEric Sesterhenn BUG_ON(cur_th && cur_th->t_refcount > 1); 3188b491dd17SJeff Mahoney return do_journal_begin_r(th, sb, 1, JBEGIN_JOIN); 31891da177e4SLinus Torvalds } 31901da177e4SLinus Torvalds 3191bd4c625cSLinus Torvalds int journal_join_abort(struct reiserfs_transaction_handle *th, 3192b491dd17SJeff Mahoney struct super_block *sb) 3193bd4c625cSLinus Torvalds { 31941da177e4SLinus Torvalds struct reiserfs_transaction_handle *cur_th = current->journal_info; 31951da177e4SLinus Torvalds 3196098297b2SJeff Mahoney /* 3197098297b2SJeff Mahoney * this keeps do_journal_end from NULLing out the 3198098297b2SJeff Mahoney * current->journal_info pointer 31991da177e4SLinus Torvalds */ 32001da177e4SLinus Torvalds th->t_handle_save = cur_th; 320114a61442SEric Sesterhenn BUG_ON(cur_th && cur_th->t_refcount > 1); 3202b491dd17SJeff Mahoney return do_journal_begin_r(th, sb, 1, JBEGIN_ABORT); 32031da177e4SLinus Torvalds } 32041da177e4SLinus Torvalds 3205bd4c625cSLinus Torvalds int journal_begin(struct reiserfs_transaction_handle *th, 3206a9dd3643SJeff Mahoney struct super_block *sb, unsigned long nblocks) 3207bd4c625cSLinus Torvalds { 32081da177e4SLinus Torvalds struct reiserfs_transaction_handle *cur_th = current->journal_info; 32091da177e4SLinus Torvalds int ret; 32101da177e4SLinus Torvalds 32111da177e4SLinus Torvalds th->t_handle_save = NULL; 32121da177e4SLinus Torvalds if (cur_th) { 32131da177e4SLinus Torvalds /* we are nesting into the current transaction */ 3214a9dd3643SJeff Mahoney if (cur_th->t_super == sb) { 32151da177e4SLinus Torvalds BUG_ON(!cur_th->t_refcount); 32161da177e4SLinus Torvalds cur_th->t_refcount++; 32171da177e4SLinus Torvalds memcpy(th, cur_th, sizeof(*th)); 32181da177e4SLinus Torvalds if (th->t_refcount <= 1) 3219a9dd3643SJeff Mahoney reiserfs_warning(sb, "reiserfs-2005", 322045b03d5eSJeff Mahoney "BAD: refcount <= 1, but " 322145b03d5eSJeff Mahoney "journal_info != 0"); 32221da177e4SLinus Torvalds return 0; 32231da177e4SLinus Torvalds } else { 3224098297b2SJeff Mahoney /* 3225098297b2SJeff Mahoney * we've ended up with a handle from a different 3226098297b2SJeff Mahoney * filesystem. save it and restore on journal_end. 3227098297b2SJeff Mahoney * This should never really happen... 32281da177e4SLinus Torvalds */ 3229a9dd3643SJeff Mahoney reiserfs_warning(sb, "clm-2100", 323045b03d5eSJeff Mahoney "nesting info a different FS"); 32311da177e4SLinus Torvalds th->t_handle_save = current->journal_info; 32321da177e4SLinus Torvalds current->journal_info = th; 32331da177e4SLinus Torvalds } 32341da177e4SLinus Torvalds } else { 32351da177e4SLinus Torvalds current->journal_info = th; 32361da177e4SLinus Torvalds } 3237a9dd3643SJeff Mahoney ret = do_journal_begin_r(th, sb, nblocks, JBEGIN_REG); 323814a61442SEric Sesterhenn BUG_ON(current->journal_info != th); 32391da177e4SLinus Torvalds 3240098297b2SJeff Mahoney /* 3241098297b2SJeff Mahoney * I guess this boils down to being the reciprocal of clm-2100 above. 3242098297b2SJeff Mahoney * If do_journal_begin_r fails, we need to put it back, since 3243098297b2SJeff Mahoney * journal_end won't be called to do it. */ 32441da177e4SLinus Torvalds if (ret) 32451da177e4SLinus Torvalds current->journal_info = th->t_handle_save; 32461da177e4SLinus Torvalds else 32471da177e4SLinus Torvalds BUG_ON(!th->t_refcount); 32481da177e4SLinus Torvalds 32491da177e4SLinus Torvalds return ret; 32501da177e4SLinus Torvalds } 32511da177e4SLinus Torvalds 32521da177e4SLinus Torvalds /* 3253098297b2SJeff Mahoney * puts bh into the current transaction. If it was already there, reorders 3254098297b2SJeff Mahoney * removes the old pointers from the hash, and puts new ones in (to make 3255098297b2SJeff Mahoney * sure replay happen in the right order). 3256098297b2SJeff Mahoney * 3257098297b2SJeff Mahoney * if it was dirty, cleans and files onto the clean list. I can't let it 3258098297b2SJeff Mahoney * be dirty again until the transaction is committed. 3259098297b2SJeff Mahoney * 3260098297b2SJeff Mahoney * if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len. 32611da177e4SLinus Torvalds */ 3262bd4c625cSLinus Torvalds int journal_mark_dirty(struct reiserfs_transaction_handle *th, 326309f1b80bSJeff Mahoney struct buffer_head *bh) 3264bd4c625cSLinus Torvalds { 326509f1b80bSJeff Mahoney struct super_block *sb = th->t_super; 3266a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 32671da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn = NULL; 32681da177e4SLinus Torvalds int count_already_incd = 0; 32691da177e4SLinus Torvalds int prepared = 0; 32701da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 32711da177e4SLinus Torvalds 3272a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.mark_dirty); 32731da177e4SLinus Torvalds if (th->t_trans_id != journal->j_trans_id) { 3274c3a9c210SJeff Mahoney reiserfs_panic(th->t_super, "journal-1577", 3275c3a9c210SJeff Mahoney "handle trans id %ld != current trans id %ld", 32761da177e4SLinus Torvalds th->t_trans_id, journal->j_trans_id); 32771da177e4SLinus Torvalds } 32781da177e4SLinus Torvalds 32791da177e4SLinus Torvalds prepared = test_clear_buffer_journal_prepared(bh); 32801da177e4SLinus Torvalds clear_buffer_journal_restore_dirty(bh); 32811da177e4SLinus Torvalds /* already in this transaction, we are done */ 32821da177e4SLinus Torvalds if (buffer_journaled(bh)) { 3283a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.mark_dirty_already); 32841da177e4SLinus Torvalds return 0; 32851da177e4SLinus Torvalds } 32861da177e4SLinus Torvalds 3287098297b2SJeff Mahoney /* 3288098297b2SJeff Mahoney * this must be turned into a panic instead of a warning. We can't 3289098297b2SJeff Mahoney * allow a dirty or journal_dirty or locked buffer to be logged, as 3290098297b2SJeff Mahoney * some changes could get to disk too early. NOT GOOD. 32911da177e4SLinus Torvalds */ 32921da177e4SLinus Torvalds if (!prepared || buffer_dirty(bh)) { 3293a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1777", 329445b03d5eSJeff Mahoney "buffer %llu bad state " 32951da177e4SLinus Torvalds "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT", 3296bd4c625cSLinus Torvalds (unsigned long long)bh->b_blocknr, 3297bd4c625cSLinus Torvalds prepared ? ' ' : '!', 32981da177e4SLinus Torvalds buffer_locked(bh) ? ' ' : '!', 32991da177e4SLinus Torvalds buffer_dirty(bh) ? ' ' : '!', 33001da177e4SLinus Torvalds buffer_journal_dirty(bh) ? ' ' : '!'); 33011da177e4SLinus Torvalds } 33021da177e4SLinus Torvalds 3303a228bf8fSJeff Mahoney if (atomic_read(&journal->j_wcount) <= 0) { 3304a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1409", 330545b03d5eSJeff Mahoney "returning because j_wcount was %d", 3306a228bf8fSJeff Mahoney atomic_read(&journal->j_wcount)); 33071da177e4SLinus Torvalds return 1; 33081da177e4SLinus Torvalds } 3309098297b2SJeff Mahoney /* 3310098297b2SJeff Mahoney * this error means I've screwed up, and we've overflowed 3311098297b2SJeff Mahoney * the transaction. Nothing can be done here, except make the 3312098297b2SJeff Mahoney * FS readonly or panic. 33131da177e4SLinus Torvalds */ 33141da177e4SLinus Torvalds if (journal->j_len >= journal->j_trans_max) { 3315c3a9c210SJeff Mahoney reiserfs_panic(th->t_super, "journal-1413", 3316c3a9c210SJeff Mahoney "j_len (%lu) is too big", 3317bd4c625cSLinus Torvalds journal->j_len); 33181da177e4SLinus Torvalds } 33191da177e4SLinus Torvalds 33201da177e4SLinus Torvalds if (buffer_journal_dirty(bh)) { 33211da177e4SLinus Torvalds count_already_incd = 1; 3322a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.mark_dirty_notjournal); 33231da177e4SLinus Torvalds clear_buffer_journal_dirty(bh); 33241da177e4SLinus Torvalds } 33251da177e4SLinus Torvalds 33261da177e4SLinus Torvalds if (journal->j_len > journal->j_len_alloc) { 33271da177e4SLinus Torvalds journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT; 33281da177e4SLinus Torvalds } 33291da177e4SLinus Torvalds 33301da177e4SLinus Torvalds set_buffer_journaled(bh); 33311da177e4SLinus Torvalds 33321da177e4SLinus Torvalds /* now put this guy on the end */ 33331da177e4SLinus Torvalds if (!cn) { 3334a9dd3643SJeff Mahoney cn = get_cnode(sb); 33351da177e4SLinus Torvalds if (!cn) { 3336a9dd3643SJeff Mahoney reiserfs_panic(sb, "journal-4", "get_cnode failed!"); 33371da177e4SLinus Torvalds } 33381da177e4SLinus Torvalds 33391da177e4SLinus Torvalds if (th->t_blocks_logged == th->t_blocks_allocated) { 33401da177e4SLinus Torvalds th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT; 33411da177e4SLinus Torvalds journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT; 33421da177e4SLinus Torvalds } 33431da177e4SLinus Torvalds th->t_blocks_logged++; 33441da177e4SLinus Torvalds journal->j_len++; 33451da177e4SLinus Torvalds 33461da177e4SLinus Torvalds cn->bh = bh; 33471da177e4SLinus Torvalds cn->blocknr = bh->b_blocknr; 3348a9dd3643SJeff Mahoney cn->sb = sb; 33491da177e4SLinus Torvalds cn->jlist = NULL; 33501da177e4SLinus Torvalds insert_journal_hash(journal->j_hash_table, cn); 33511da177e4SLinus Torvalds if (!count_already_incd) { 33521da177e4SLinus Torvalds get_bh(bh); 33531da177e4SLinus Torvalds } 33541da177e4SLinus Torvalds } 33551da177e4SLinus Torvalds cn->next = NULL; 33561da177e4SLinus Torvalds cn->prev = journal->j_last; 33571da177e4SLinus Torvalds cn->bh = bh; 33581da177e4SLinus Torvalds if (journal->j_last) { 33591da177e4SLinus Torvalds journal->j_last->next = cn; 33601da177e4SLinus Torvalds journal->j_last = cn; 33611da177e4SLinus Torvalds } else { 33621da177e4SLinus Torvalds journal->j_first = cn; 33631da177e4SLinus Torvalds journal->j_last = cn; 33641da177e4SLinus Torvalds } 3365033369d1SArtem Bityutskiy reiserfs_schedule_old_flush(sb); 33661da177e4SLinus Torvalds return 0; 33671da177e4SLinus Torvalds } 33681da177e4SLinus Torvalds 336958d85426SJeff Mahoney int journal_end(struct reiserfs_transaction_handle *th) 3370bd4c625cSLinus Torvalds { 337158d85426SJeff Mahoney struct super_block *sb = th->t_super; 33721da177e4SLinus Torvalds if (!current->journal_info && th->t_refcount > 1) 3373a9dd3643SJeff Mahoney reiserfs_warning(sb, "REISER-NESTING", 337445b03d5eSJeff Mahoney "th NULL, refcount %d", th->t_refcount); 33751da177e4SLinus Torvalds 33761da177e4SLinus Torvalds if (!th->t_trans_id) { 33771da177e4SLinus Torvalds WARN_ON(1); 33781da177e4SLinus Torvalds return -EIO; 33791da177e4SLinus Torvalds } 33801da177e4SLinus Torvalds 33811da177e4SLinus Torvalds th->t_refcount--; 33821da177e4SLinus Torvalds if (th->t_refcount > 0) { 3383bd4c625cSLinus Torvalds struct reiserfs_transaction_handle *cur_th = 3384bd4c625cSLinus Torvalds current->journal_info; 33851da177e4SLinus Torvalds 3386098297b2SJeff Mahoney /* 3387098297b2SJeff Mahoney * we aren't allowed to close a nested transaction on a 3388098297b2SJeff Mahoney * different filesystem from the one in the task struct 33891da177e4SLinus Torvalds */ 339014a61442SEric Sesterhenn BUG_ON(cur_th->t_super != th->t_super); 33911da177e4SLinus Torvalds 33921da177e4SLinus Torvalds if (th != cur_th) { 33931da177e4SLinus Torvalds memcpy(current->journal_info, th, sizeof(*th)); 33941da177e4SLinus Torvalds th->t_trans_id = 0; 33951da177e4SLinus Torvalds } 33961da177e4SLinus Torvalds return 0; 33971da177e4SLinus Torvalds } else { 339858d85426SJeff Mahoney return do_journal_end(th, 0); 33991da177e4SLinus Torvalds } 34001da177e4SLinus Torvalds } 34011da177e4SLinus Torvalds 3402098297b2SJeff Mahoney /* 3403098297b2SJeff Mahoney * removes from the current transaction, relsing and descrementing any counters. 3404098297b2SJeff Mahoney * also files the removed buffer directly onto the clean list 3405098297b2SJeff Mahoney * 3406098297b2SJeff Mahoney * called by journal_mark_freed when a block has been deleted 3407098297b2SJeff Mahoney * 3408098297b2SJeff Mahoney * returns 1 if it cleaned and relsed the buffer. 0 otherwise 34091da177e4SLinus Torvalds */ 3410a9dd3643SJeff Mahoney static int remove_from_transaction(struct super_block *sb, 3411bd4c625cSLinus Torvalds b_blocknr_t blocknr, int already_cleaned) 3412bd4c625cSLinus Torvalds { 34131da177e4SLinus Torvalds struct buffer_head *bh; 34141da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 3415a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 34161da177e4SLinus Torvalds int ret = 0; 34171da177e4SLinus Torvalds 3418a9dd3643SJeff Mahoney cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr); 34191da177e4SLinus Torvalds if (!cn || !cn->bh) { 34201da177e4SLinus Torvalds return ret; 34211da177e4SLinus Torvalds } 34221da177e4SLinus Torvalds bh = cn->bh; 34231da177e4SLinus Torvalds if (cn->prev) { 34241da177e4SLinus Torvalds cn->prev->next = cn->next; 34251da177e4SLinus Torvalds } 34261da177e4SLinus Torvalds if (cn->next) { 34271da177e4SLinus Torvalds cn->next->prev = cn->prev; 34281da177e4SLinus Torvalds } 34291da177e4SLinus Torvalds if (cn == journal->j_first) { 34301da177e4SLinus Torvalds journal->j_first = cn->next; 34311da177e4SLinus Torvalds } 34321da177e4SLinus Torvalds if (cn == journal->j_last) { 34331da177e4SLinus Torvalds journal->j_last = cn->prev; 34341da177e4SLinus Torvalds } 3435a9dd3643SJeff Mahoney remove_journal_hash(sb, journal->j_hash_table, NULL, 3436bd4c625cSLinus Torvalds bh->b_blocknr, 0); 34371da177e4SLinus Torvalds clear_buffer_journaled(bh); /* don't log this one */ 34381da177e4SLinus Torvalds 34391da177e4SLinus Torvalds if (!already_cleaned) { 34401da177e4SLinus Torvalds clear_buffer_journal_dirty(bh); 34411da177e4SLinus Torvalds clear_buffer_dirty(bh); 34421da177e4SLinus Torvalds clear_buffer_journal_test(bh); 34431da177e4SLinus Torvalds put_bh(bh); 3444a228bf8fSJeff Mahoney if (atomic_read(&bh->b_count) < 0) { 3445a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1752", 344645b03d5eSJeff Mahoney "b_count < 0"); 34471da177e4SLinus Torvalds } 34481da177e4SLinus Torvalds ret = 1; 34491da177e4SLinus Torvalds } 34501da177e4SLinus Torvalds journal->j_len--; 34511da177e4SLinus Torvalds journal->j_len_alloc--; 3452a9dd3643SJeff Mahoney free_cnode(sb, cn); 34531da177e4SLinus Torvalds return ret; 34541da177e4SLinus Torvalds } 34551da177e4SLinus Torvalds 34561da177e4SLinus Torvalds /* 3457098297b2SJeff Mahoney * for any cnode in a journal list, it can only be dirtied of all the 3458098297b2SJeff Mahoney * transactions that include it are committed to disk. 3459098297b2SJeff Mahoney * this checks through each transaction, and returns 1 if you are allowed 3460098297b2SJeff Mahoney * to dirty, and 0 if you aren't 3461098297b2SJeff Mahoney * 3462098297b2SJeff Mahoney * it is called by dirty_journal_list, which is called after 3463098297b2SJeff Mahoney * flush_commit_list has gotten all the log blocks for a given 3464098297b2SJeff Mahoney * transaction on disk 3465098297b2SJeff Mahoney * 34661da177e4SLinus Torvalds */ 3467bd4c625cSLinus Torvalds static int can_dirty(struct reiserfs_journal_cnode *cn) 3468bd4c625cSLinus Torvalds { 34691da177e4SLinus Torvalds struct super_block *sb = cn->sb; 34701da177e4SLinus Torvalds b_blocknr_t blocknr = cn->blocknr; 34711da177e4SLinus Torvalds struct reiserfs_journal_cnode *cur = cn->hprev; 34721da177e4SLinus Torvalds int can_dirty = 1; 34731da177e4SLinus Torvalds 3474098297b2SJeff Mahoney /* 3475098297b2SJeff Mahoney * first test hprev. These are all newer than cn, so any node here 3476098297b2SJeff Mahoney * with the same block number and dev means this node can't be sent 3477098297b2SJeff Mahoney * to disk right now. 34781da177e4SLinus Torvalds */ 34791da177e4SLinus Torvalds while (cur && can_dirty) { 34801da177e4SLinus Torvalds if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb && 34811da177e4SLinus Torvalds cur->blocknr == blocknr) { 34821da177e4SLinus Torvalds can_dirty = 0; 34831da177e4SLinus Torvalds } 34841da177e4SLinus Torvalds cur = cur->hprev; 34851da177e4SLinus Torvalds } 3486098297b2SJeff Mahoney /* 3487098297b2SJeff Mahoney * then test hnext. These are all older than cn. As long as they 3488098297b2SJeff Mahoney * are committed to the log, it is safe to write cn to disk 34891da177e4SLinus Torvalds */ 34901da177e4SLinus Torvalds cur = cn->hnext; 34911da177e4SLinus Torvalds while (cur && can_dirty) { 34921da177e4SLinus Torvalds if (cur->jlist && cur->jlist->j_len > 0 && 3493a228bf8fSJeff Mahoney atomic_read(&cur->jlist->j_commit_left) > 0 && cur->bh && 34941da177e4SLinus Torvalds cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) { 34951da177e4SLinus Torvalds can_dirty = 0; 34961da177e4SLinus Torvalds } 34971da177e4SLinus Torvalds cur = cur->hnext; 34981da177e4SLinus Torvalds } 34991da177e4SLinus Torvalds return can_dirty; 35001da177e4SLinus Torvalds } 35011da177e4SLinus Torvalds 3502098297b2SJeff Mahoney /* 3503098297b2SJeff Mahoney * syncs the commit blocks, but does not force the real buffers to disk 3504098297b2SJeff Mahoney * will wait until the current transaction is done/committed before returning 35051da177e4SLinus Torvalds */ 350658d85426SJeff Mahoney int journal_end_sync(struct reiserfs_transaction_handle *th) 3507bd4c625cSLinus Torvalds { 350858d85426SJeff Mahoney struct super_block *sb = th->t_super; 3509a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 35101da177e4SLinus Torvalds 35111da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 35121da177e4SLinus Torvalds /* you can sync while nested, very, very bad */ 351314a61442SEric Sesterhenn BUG_ON(th->t_refcount > 1); 35141da177e4SLinus Torvalds if (journal->j_len == 0) { 3515a9dd3643SJeff Mahoney reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb), 3516bd4c625cSLinus Torvalds 1); 351709f1b80bSJeff Mahoney journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb)); 35181da177e4SLinus Torvalds } 351958d85426SJeff Mahoney return do_journal_end(th, COMMIT_NOW | WAIT); 35201da177e4SLinus Torvalds } 35211da177e4SLinus Torvalds 3522098297b2SJeff Mahoney /* writeback the pending async commits to disk */ 3523c4028958SDavid Howells static void flush_async_commits(struct work_struct *work) 3524bd4c625cSLinus Torvalds { 3525c4028958SDavid Howells struct reiserfs_journal *journal = 3526c4028958SDavid Howells container_of(work, struct reiserfs_journal, j_work.work); 3527a9dd3643SJeff Mahoney struct super_block *sb = journal->j_work_sb; 35281da177e4SLinus Torvalds struct reiserfs_journal_list *jl; 35291da177e4SLinus Torvalds struct list_head *entry; 35301da177e4SLinus Torvalds 35318ebc4232SFrederic Weisbecker reiserfs_write_lock(sb); 35321da177e4SLinus Torvalds if (!list_empty(&journal->j_journal_list)) { 35331da177e4SLinus Torvalds /* last entry is the youngest, commit it and you get everything */ 35341da177e4SLinus Torvalds entry = journal->j_journal_list.prev; 35351da177e4SLinus Torvalds jl = JOURNAL_LIST_ENTRY(entry); 3536a9dd3643SJeff Mahoney flush_commit_list(sb, jl, 1); 35371da177e4SLinus Torvalds } 35388ebc4232SFrederic Weisbecker reiserfs_write_unlock(sb); 35391da177e4SLinus Torvalds } 35401da177e4SLinus Torvalds 35411da177e4SLinus Torvalds /* 3542098297b2SJeff Mahoney * flushes any old transactions to disk 3543098297b2SJeff Mahoney * ends the current transaction if it is too old 35441da177e4SLinus Torvalds */ 354525729b0eSArtem Bityutskiy void reiserfs_flush_old_commits(struct super_block *sb) 3546bd4c625cSLinus Torvalds { 354734d08260SArnd Bergmann time64_t now; 35481da177e4SLinus Torvalds struct reiserfs_transaction_handle th; 3549a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 35501da177e4SLinus Torvalds 355134d08260SArnd Bergmann now = ktime_get_seconds(); 3552098297b2SJeff Mahoney /* 3553098297b2SJeff Mahoney * safety check so we don't flush while we are replaying the log during 35541da177e4SLinus Torvalds * mount 35551da177e4SLinus Torvalds */ 355625729b0eSArtem Bityutskiy if (list_empty(&journal->j_journal_list)) 355725729b0eSArtem Bityutskiy return; 35581da177e4SLinus Torvalds 3559098297b2SJeff Mahoney /* 3560098297b2SJeff Mahoney * check the current transaction. If there are no writers, and it is 35611da177e4SLinus Torvalds * too old, finish it, and force the commit blocks to disk 35621da177e4SLinus Torvalds */ 35631da177e4SLinus Torvalds if (atomic_read(&journal->j_wcount) <= 0 && 35641da177e4SLinus Torvalds journal->j_trans_start_time > 0 && 35651da177e4SLinus Torvalds journal->j_len > 0 && 3566bd4c625cSLinus Torvalds (now - journal->j_trans_start_time) > journal->j_max_trans_age) { 3567b491dd17SJeff Mahoney if (!journal_join(&th, sb)) { 3568a9dd3643SJeff Mahoney reiserfs_prepare_for_journal(sb, 3569a9dd3643SJeff Mahoney SB_BUFFER_WITH_SB(sb), 3570bd4c625cSLinus Torvalds 1); 357109f1b80bSJeff Mahoney journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb)); 35721da177e4SLinus Torvalds 3573098297b2SJeff Mahoney /* 3574098297b2SJeff Mahoney * we're only being called from kreiserfsd, it makes 3575098297b2SJeff Mahoney * no sense to do an async commit so that kreiserfsd 3576098297b2SJeff Mahoney * can do it later 35771da177e4SLinus Torvalds */ 357858d85426SJeff Mahoney do_journal_end(&th, COMMIT_NOW | WAIT); 35791da177e4SLinus Torvalds } 35801da177e4SLinus Torvalds } 35811da177e4SLinus Torvalds } 35821da177e4SLinus Torvalds 35831da177e4SLinus Torvalds /* 3584098297b2SJeff Mahoney * returns 0 if do_journal_end should return right away, returns 1 if 3585098297b2SJeff Mahoney * do_journal_end should finish the commit 3586098297b2SJeff Mahoney * 3587098297b2SJeff Mahoney * if the current transaction is too old, but still has writers, this will 3588098297b2SJeff Mahoney * wait on j_join_wait until all the writers are done. By the time it 3589098297b2SJeff Mahoney * wakes up, the transaction it was called has already ended, so it just 3590098297b2SJeff Mahoney * flushes the commit list and returns 0. 3591098297b2SJeff Mahoney * 3592098297b2SJeff Mahoney * Won't batch when flush or commit_now is set. Also won't batch when 3593098297b2SJeff Mahoney * others are waiting on j_join_wait. 3594098297b2SJeff Mahoney * 3595098297b2SJeff Mahoney * Note, we can't allow the journal_end to proceed while there are still 3596098297b2SJeff Mahoney * writers in the log. 35971da177e4SLinus Torvalds */ 359858d85426SJeff Mahoney static int check_journal_end(struct reiserfs_transaction_handle *th, int flags) 3599bd4c625cSLinus Torvalds { 36001da177e4SLinus Torvalds 360134d08260SArnd Bergmann time64_t now; 36021da177e4SLinus Torvalds int flush = flags & FLUSH_ALL; 36031da177e4SLinus Torvalds int commit_now = flags & COMMIT_NOW; 36041da177e4SLinus Torvalds int wait_on_commit = flags & WAIT; 36051da177e4SLinus Torvalds struct reiserfs_journal_list *jl; 360658d85426SJeff Mahoney struct super_block *sb = th->t_super; 3607a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 36081da177e4SLinus Torvalds 36091da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 36101da177e4SLinus Torvalds 36111da177e4SLinus Torvalds if (th->t_trans_id != journal->j_trans_id) { 3612c3a9c210SJeff Mahoney reiserfs_panic(th->t_super, "journal-1577", 3613c3a9c210SJeff Mahoney "handle trans id %ld != current trans id %ld", 36141da177e4SLinus Torvalds th->t_trans_id, journal->j_trans_id); 36151da177e4SLinus Torvalds } 36161da177e4SLinus Torvalds 36171da177e4SLinus Torvalds journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged); 3618098297b2SJeff Mahoney /* <= 0 is allowed. unmounting might not call begin */ 3619a228bf8fSJeff Mahoney if (atomic_read(&journal->j_wcount) > 0) 3620a228bf8fSJeff Mahoney atomic_dec(&journal->j_wcount); 36211da177e4SLinus Torvalds 3622098297b2SJeff Mahoney /* 3623098297b2SJeff Mahoney * BUG, deal with case where j_len is 0, but people previously 3624098297b2SJeff Mahoney * freed blocks need to be released will be dealt with by next 3625098297b2SJeff Mahoney * transaction that actually writes something, but should be taken 3626098297b2SJeff Mahoney * care of in this trans 36271da177e4SLinus Torvalds */ 362814a61442SEric Sesterhenn BUG_ON(journal->j_len == 0); 362914a61442SEric Sesterhenn 3630098297b2SJeff Mahoney /* 3631098297b2SJeff Mahoney * if wcount > 0, and we are called to with flush or commit_now, 3632098297b2SJeff Mahoney * we wait on j_join_wait. We will wake up when the last writer has 3633098297b2SJeff Mahoney * finished the transaction, and started it on its way to the disk. 3634098297b2SJeff Mahoney * Then, we flush the commit or journal list, and just return 0 3635098297b2SJeff Mahoney * because the rest of journal end was already done for this 3636098297b2SJeff Mahoney * transaction. 36371da177e4SLinus Torvalds */ 3638a228bf8fSJeff Mahoney if (atomic_read(&journal->j_wcount) > 0) { 36391da177e4SLinus Torvalds if (flush || commit_now) { 36401da177e4SLinus Torvalds unsigned trans_id; 36411da177e4SLinus Torvalds 36421da177e4SLinus Torvalds jl = journal->j_current_jl; 36431da177e4SLinus Torvalds trans_id = jl->j_trans_id; 36441da177e4SLinus Torvalds if (wait_on_commit) 36451da177e4SLinus Torvalds jl->j_state |= LIST_COMMIT_PENDING; 3646a228bf8fSJeff Mahoney atomic_set(&journal->j_jlock, 1); 36471da177e4SLinus Torvalds if (flush) { 36481da177e4SLinus Torvalds journal->j_next_full_flush = 1; 36491da177e4SLinus Torvalds } 3650a9dd3643SJeff Mahoney unlock_journal(sb); 36511da177e4SLinus Torvalds 3652098297b2SJeff Mahoney /* 3653098297b2SJeff Mahoney * sleep while the current transaction is 3654098297b2SJeff Mahoney * still j_jlocked 3655098297b2SJeff Mahoney */ 36561da177e4SLinus Torvalds while (journal->j_trans_id == trans_id) { 36571da177e4SLinus Torvalds if (atomic_read(&journal->j_jlock)) { 3658a9dd3643SJeff Mahoney queue_log_writer(sb); 36591da177e4SLinus Torvalds } else { 3660a9dd3643SJeff Mahoney lock_journal(sb); 36611da177e4SLinus Torvalds if (journal->j_trans_id == trans_id) { 3662a228bf8fSJeff Mahoney atomic_set(&journal->j_jlock, 3663bd4c625cSLinus Torvalds 1); 36641da177e4SLinus Torvalds } 3665a9dd3643SJeff Mahoney unlock_journal(sb); 36661da177e4SLinus Torvalds } 36671da177e4SLinus Torvalds } 366814a61442SEric Sesterhenn BUG_ON(journal->j_trans_id == trans_id); 366914a61442SEric Sesterhenn 3670bd4c625cSLinus Torvalds if (commit_now 3671a9dd3643SJeff Mahoney && journal_list_still_alive(sb, trans_id) 3672bd4c625cSLinus Torvalds && wait_on_commit) { 3673a9dd3643SJeff Mahoney flush_commit_list(sb, jl, 1); 36741da177e4SLinus Torvalds } 36751da177e4SLinus Torvalds return 0; 36761da177e4SLinus Torvalds } 3677a9dd3643SJeff Mahoney unlock_journal(sb); 36781da177e4SLinus Torvalds return 0; 36791da177e4SLinus Torvalds } 36801da177e4SLinus Torvalds 36811da177e4SLinus Torvalds /* deal with old transactions where we are the last writers */ 368234d08260SArnd Bergmann now = ktime_get_seconds(); 36831da177e4SLinus Torvalds if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) { 36841da177e4SLinus Torvalds commit_now = 1; 36851da177e4SLinus Torvalds journal->j_next_async_flush = 1; 36861da177e4SLinus Torvalds } 36871da177e4SLinus Torvalds /* don't batch when someone is waiting on j_join_wait */ 36881da177e4SLinus Torvalds /* don't batch when syncing the commit or flushing the whole trans */ 3689a228bf8fSJeff Mahoney if (!(journal->j_must_wait > 0) && !(atomic_read(&journal->j_jlock)) 3690bd4c625cSLinus Torvalds && !flush && !commit_now && (journal->j_len < journal->j_max_batch) 3691bd4c625cSLinus Torvalds && journal->j_len_alloc < journal->j_max_batch 3692bd4c625cSLinus Torvalds && journal->j_cnode_free > (journal->j_trans_max * 3)) { 36931da177e4SLinus Torvalds journal->j_bcount++; 3694a9dd3643SJeff Mahoney unlock_journal(sb); 36951da177e4SLinus Torvalds return 0; 36961da177e4SLinus Torvalds } 36971da177e4SLinus Torvalds 3698a9dd3643SJeff Mahoney if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(sb)) { 3699a9dd3643SJeff Mahoney reiserfs_panic(sb, "journal-003", 3700c3a9c210SJeff Mahoney "j_start (%ld) is too high", 3701bd4c625cSLinus Torvalds journal->j_start); 37021da177e4SLinus Torvalds } 37031da177e4SLinus Torvalds return 1; 37041da177e4SLinus Torvalds } 37051da177e4SLinus Torvalds 37061da177e4SLinus Torvalds /* 3707098297b2SJeff Mahoney * Does all the work that makes deleting blocks safe. 3708098297b2SJeff Mahoney * when deleting a block mark BH_JNew, just remove it from the current 3709098297b2SJeff Mahoney * transaction, clean it's buffer_head and move on. 3710098297b2SJeff Mahoney * 3711098297b2SJeff Mahoney * otherwise: 3712098297b2SJeff Mahoney * set a bit for the block in the journal bitmap. That will prevent it from 3713098297b2SJeff Mahoney * being allocated for unformatted nodes before this transaction has finished. 3714098297b2SJeff Mahoney * 3715098297b2SJeff Mahoney * mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers. 3716098297b2SJeff Mahoney * That will prevent any old transactions with this block from trying to flush 3717098297b2SJeff Mahoney * to the real location. Since we aren't removing the cnode from the 3718098297b2SJeff Mahoney * journal_list_hash, *the block can't be reallocated yet. 3719098297b2SJeff Mahoney * 3720098297b2SJeff Mahoney * Then remove it from the current transaction, decrementing any counters and 3721098297b2SJeff Mahoney * filing it on the clean list. 37221da177e4SLinus Torvalds */ 3723bd4c625cSLinus Torvalds int journal_mark_freed(struct reiserfs_transaction_handle *th, 3724a9dd3643SJeff Mahoney struct super_block *sb, b_blocknr_t blocknr) 3725bd4c625cSLinus Torvalds { 3726a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 37271da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn = NULL; 37281da177e4SLinus Torvalds struct buffer_head *bh = NULL; 37291da177e4SLinus Torvalds struct reiserfs_list_bitmap *jb = NULL; 37301da177e4SLinus Torvalds int cleaned = 0; 37311da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 37321da177e4SLinus Torvalds 3733a9dd3643SJeff Mahoney cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr); 37341da177e4SLinus Torvalds if (cn && cn->bh) { 37351da177e4SLinus Torvalds bh = cn->bh; 37361da177e4SLinus Torvalds get_bh(bh); 37371da177e4SLinus Torvalds } 37381da177e4SLinus Torvalds /* if it is journal new, we just remove it from this transaction */ 37391da177e4SLinus Torvalds if (bh && buffer_journal_new(bh)) { 37401da177e4SLinus Torvalds clear_buffer_journal_new(bh); 37411da177e4SLinus Torvalds clear_prepared_bits(bh); 37421da177e4SLinus Torvalds reiserfs_clean_and_file_buffer(bh); 3743a9dd3643SJeff Mahoney cleaned = remove_from_transaction(sb, blocknr, cleaned); 37441da177e4SLinus Torvalds } else { 3745098297b2SJeff Mahoney /* 3746098297b2SJeff Mahoney * set the bit for this block in the journal bitmap 3747098297b2SJeff Mahoney * for this transaction 3748098297b2SJeff Mahoney */ 37491da177e4SLinus Torvalds jb = journal->j_current_jl->j_list_bitmap; 37501da177e4SLinus Torvalds if (!jb) { 3751a9dd3643SJeff Mahoney reiserfs_panic(sb, "journal-1702", 3752c3a9c210SJeff Mahoney "journal_list_bitmap is NULL"); 37531da177e4SLinus Torvalds } 3754a9dd3643SJeff Mahoney set_bit_in_list_bitmap(sb, blocknr, jb); 37551da177e4SLinus Torvalds 37561da177e4SLinus Torvalds /* Note, the entire while loop is not allowed to schedule. */ 37571da177e4SLinus Torvalds 37581da177e4SLinus Torvalds if (bh) { 37591da177e4SLinus Torvalds clear_prepared_bits(bh); 37601da177e4SLinus Torvalds reiserfs_clean_and_file_buffer(bh); 37611da177e4SLinus Torvalds } 3762a9dd3643SJeff Mahoney cleaned = remove_from_transaction(sb, blocknr, cleaned); 37631da177e4SLinus Torvalds 3764098297b2SJeff Mahoney /* 3765098297b2SJeff Mahoney * find all older transactions with this block, 3766098297b2SJeff Mahoney * make sure they don't try to write it out 3767098297b2SJeff Mahoney */ 3768a9dd3643SJeff Mahoney cn = get_journal_hash_dev(sb, journal->j_list_hash_table, 3769bd4c625cSLinus Torvalds blocknr); 37701da177e4SLinus Torvalds while (cn) { 3771a9dd3643SJeff Mahoney if (sb == cn->sb && blocknr == cn->blocknr) { 37721da177e4SLinus Torvalds set_bit(BLOCK_FREED, &cn->state); 37731da177e4SLinus Torvalds if (cn->bh) { 3774098297b2SJeff Mahoney /* 3775098297b2SJeff Mahoney * remove_from_transaction will brelse 3776098297b2SJeff Mahoney * the buffer if it was in the current 3777098297b2SJeff Mahoney * trans 37781da177e4SLinus Torvalds */ 3779098297b2SJeff Mahoney if (!cleaned) { 3780bd4c625cSLinus Torvalds clear_buffer_journal_dirty(cn-> 3781bd4c625cSLinus Torvalds bh); 37821da177e4SLinus Torvalds clear_buffer_dirty(cn->bh); 3783bd4c625cSLinus Torvalds clear_buffer_journal_test(cn-> 3784bd4c625cSLinus Torvalds bh); 37851da177e4SLinus Torvalds cleaned = 1; 37861da177e4SLinus Torvalds put_bh(cn->bh); 3787bd4c625cSLinus Torvalds if (atomic_read 3788a228bf8fSJeff Mahoney (&cn->bh->b_count) < 0) { 3789a9dd3643SJeff Mahoney reiserfs_warning(sb, 379045b03d5eSJeff Mahoney "journal-2138", 379145b03d5eSJeff Mahoney "cn->bh->b_count < 0"); 37921da177e4SLinus Torvalds } 37931da177e4SLinus Torvalds } 3794098297b2SJeff Mahoney /* 3795098297b2SJeff Mahoney * since we are clearing the bh, 3796098297b2SJeff Mahoney * we MUST dec nonzerolen 3797098297b2SJeff Mahoney */ 3798098297b2SJeff Mahoney if (cn->jlist) { 3799a228bf8fSJeff Mahoney atomic_dec(&cn->jlist-> 3800a228bf8fSJeff Mahoney j_nonzerolen); 38011da177e4SLinus Torvalds } 38021da177e4SLinus Torvalds cn->bh = NULL; 38031da177e4SLinus Torvalds } 38041da177e4SLinus Torvalds } 38051da177e4SLinus Torvalds cn = cn->hnext; 38061da177e4SLinus Torvalds } 38071da177e4SLinus Torvalds } 38081da177e4SLinus Torvalds 3809398c95bdSChris Mason if (bh) 3810398c95bdSChris Mason release_buffer_page(bh); /* get_hash grabs the buffer */ 38111da177e4SLinus Torvalds return 0; 38121da177e4SLinus Torvalds } 38131da177e4SLinus Torvalds 3814bd4c625cSLinus Torvalds void reiserfs_update_inode_transaction(struct inode *inode) 3815bd4c625cSLinus Torvalds { 38161da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb); 38171da177e4SLinus Torvalds REISERFS_I(inode)->i_jl = journal->j_current_jl; 38181da177e4SLinus Torvalds REISERFS_I(inode)->i_trans_id = journal->j_trans_id; 38191da177e4SLinus Torvalds } 38201da177e4SLinus Torvalds 38211da177e4SLinus Torvalds /* 38221da177e4SLinus Torvalds * returns -1 on error, 0 if no commits/barriers were done and 1 38231da177e4SLinus Torvalds * if a transaction was actually committed and the barrier was done 38241da177e4SLinus Torvalds */ 38251da177e4SLinus Torvalds static int __commit_trans_jl(struct inode *inode, unsigned long id, 38261da177e4SLinus Torvalds struct reiserfs_journal_list *jl) 38271da177e4SLinus Torvalds { 38281da177e4SLinus Torvalds struct reiserfs_transaction_handle th; 38291da177e4SLinus Torvalds struct super_block *sb = inode->i_sb; 38301da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(sb); 38311da177e4SLinus Torvalds int ret = 0; 38321da177e4SLinus Torvalds 3833098297b2SJeff Mahoney /* 3834098297b2SJeff Mahoney * is it from the current transaction, 3835098297b2SJeff Mahoney * or from an unknown transaction? 3836098297b2SJeff Mahoney */ 38371da177e4SLinus Torvalds if (id == journal->j_trans_id) { 38381da177e4SLinus Torvalds jl = journal->j_current_jl; 3839098297b2SJeff Mahoney /* 3840098297b2SJeff Mahoney * try to let other writers come in and 3841098297b2SJeff Mahoney * grow this transaction 3842098297b2SJeff Mahoney */ 38431da177e4SLinus Torvalds let_transaction_grow(sb, id); 38441da177e4SLinus Torvalds if (journal->j_trans_id != id) { 38451da177e4SLinus Torvalds goto flush_commit_only; 38461da177e4SLinus Torvalds } 38471da177e4SLinus Torvalds 38481da177e4SLinus Torvalds ret = journal_begin(&th, sb, 1); 38491da177e4SLinus Torvalds if (ret) 38501da177e4SLinus Torvalds return ret; 38511da177e4SLinus Torvalds 38521da177e4SLinus Torvalds /* someone might have ended this transaction while we joined */ 38531da177e4SLinus Torvalds if (journal->j_trans_id != id) { 3854bd4c625cSLinus Torvalds reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb), 3855bd4c625cSLinus Torvalds 1); 385609f1b80bSJeff Mahoney journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb)); 385758d85426SJeff Mahoney ret = journal_end(&th); 38581da177e4SLinus Torvalds goto flush_commit_only; 38591da177e4SLinus Torvalds } 38601da177e4SLinus Torvalds 386158d85426SJeff Mahoney ret = journal_end_sync(&th); 38621da177e4SLinus Torvalds if (!ret) 38631da177e4SLinus Torvalds ret = 1; 38641da177e4SLinus Torvalds 38651da177e4SLinus Torvalds } else { 3866098297b2SJeff Mahoney /* 3867098297b2SJeff Mahoney * this gets tricky, we have to make sure the journal list in 38681da177e4SLinus Torvalds * the inode still exists. We know the list is still around 38691da177e4SLinus Torvalds * if we've got a larger transaction id than the oldest list 38701da177e4SLinus Torvalds */ 38711da177e4SLinus Torvalds flush_commit_only: 38721da177e4SLinus Torvalds if (journal_list_still_alive(inode->i_sb, id)) { 38731da177e4SLinus Torvalds /* 38741da177e4SLinus Torvalds * we only set ret to 1 when we know for sure 38751da177e4SLinus Torvalds * the barrier hasn't been started yet on the commit 38761da177e4SLinus Torvalds * block. 38771da177e4SLinus Torvalds */ 38781da177e4SLinus Torvalds if (atomic_read(&jl->j_commit_left) > 1) 38791da177e4SLinus Torvalds ret = 1; 38801da177e4SLinus Torvalds flush_commit_list(sb, jl, 1); 38811da177e4SLinus Torvalds if (journal->j_errno) 38821da177e4SLinus Torvalds ret = journal->j_errno; 38831da177e4SLinus Torvalds } 38841da177e4SLinus Torvalds } 38851da177e4SLinus Torvalds /* otherwise the list is gone, and long since committed */ 38861da177e4SLinus Torvalds return ret; 38871da177e4SLinus Torvalds } 38881da177e4SLinus Torvalds 3889bd4c625cSLinus Torvalds int reiserfs_commit_for_inode(struct inode *inode) 3890bd4c625cSLinus Torvalds { 3891600ed416SJeff Mahoney unsigned int id = REISERFS_I(inode)->i_trans_id; 38921da177e4SLinus Torvalds struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl; 38931da177e4SLinus Torvalds 3894098297b2SJeff Mahoney /* 3895098297b2SJeff Mahoney * for the whole inode, assume unset id means it was 38961da177e4SLinus Torvalds * changed in the current transaction. More conservative 38971da177e4SLinus Torvalds */ 38981da177e4SLinus Torvalds if (!id || !jl) { 38991da177e4SLinus Torvalds reiserfs_update_inode_transaction(inode); 39001da177e4SLinus Torvalds id = REISERFS_I(inode)->i_trans_id; 39011da177e4SLinus Torvalds /* jl will be updated in __commit_trans_jl */ 39021da177e4SLinus Torvalds } 39031da177e4SLinus Torvalds 39041da177e4SLinus Torvalds return __commit_trans_jl(inode, id, jl); 39051da177e4SLinus Torvalds } 39061da177e4SLinus Torvalds 3907a9dd3643SJeff Mahoney void reiserfs_restore_prepared_buffer(struct super_block *sb, 3908bd4c625cSLinus Torvalds struct buffer_head *bh) 3909bd4c625cSLinus Torvalds { 3910a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 3911a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.restore_prepared); 39121da177e4SLinus Torvalds if (!bh) { 39131da177e4SLinus Torvalds return; 39141da177e4SLinus Torvalds } 39151da177e4SLinus Torvalds if (test_clear_buffer_journal_restore_dirty(bh) && 39161da177e4SLinus Torvalds buffer_journal_dirty(bh)) { 39171da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 3918278f6679SJeff Mahoney reiserfs_write_lock(sb); 3919a9dd3643SJeff Mahoney cn = get_journal_hash_dev(sb, 39201da177e4SLinus Torvalds journal->j_list_hash_table, 39211da177e4SLinus Torvalds bh->b_blocknr); 39221da177e4SLinus Torvalds if (cn && can_dirty(cn)) { 39231da177e4SLinus Torvalds set_buffer_journal_test(bh); 39241da177e4SLinus Torvalds mark_buffer_dirty(bh); 39251da177e4SLinus Torvalds } 3926278f6679SJeff Mahoney reiserfs_write_unlock(sb); 39271da177e4SLinus Torvalds } 39281da177e4SLinus Torvalds clear_buffer_journal_prepared(bh); 39291da177e4SLinus Torvalds } 39301da177e4SLinus Torvalds 39311da177e4SLinus Torvalds extern struct tree_balance *cur_tb; 39321da177e4SLinus Torvalds /* 3933098297b2SJeff Mahoney * before we can change a metadata block, we have to make sure it won't 3934098297b2SJeff Mahoney * be written to disk while we are altering it. So, we must: 3935098297b2SJeff Mahoney * clean it 3936098297b2SJeff Mahoney * wait on it. 39371da177e4SLinus Torvalds */ 3938a9dd3643SJeff Mahoney int reiserfs_prepare_for_journal(struct super_block *sb, 3939bd4c625cSLinus Torvalds struct buffer_head *bh, int wait) 3940bd4c625cSLinus Torvalds { 3941a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.prepare); 39421da177e4SLinus Torvalds 3943ca5de404SNick Piggin if (!trylock_buffer(bh)) { 39441da177e4SLinus Torvalds if (!wait) 39451da177e4SLinus Torvalds return 0; 39461da177e4SLinus Torvalds lock_buffer(bh); 39471da177e4SLinus Torvalds } 39481da177e4SLinus Torvalds set_buffer_journal_prepared(bh); 39491da177e4SLinus Torvalds if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) { 39501da177e4SLinus Torvalds clear_buffer_journal_test(bh); 39511da177e4SLinus Torvalds set_buffer_journal_restore_dirty(bh); 39521da177e4SLinus Torvalds } 39531da177e4SLinus Torvalds unlock_buffer(bh); 39541da177e4SLinus Torvalds return 1; 39551da177e4SLinus Torvalds } 39561da177e4SLinus Torvalds 39571da177e4SLinus Torvalds /* 3958098297b2SJeff Mahoney * long and ugly. If flush, will not return until all commit 3959098297b2SJeff Mahoney * blocks and all real buffers in the trans are on disk. 3960098297b2SJeff Mahoney * If no_async, won't return until all commit blocks are on disk. 3961098297b2SJeff Mahoney * 3962098297b2SJeff Mahoney * keep reading, there are comments as you go along 3963098297b2SJeff Mahoney * 3964098297b2SJeff Mahoney * If the journal is aborted, we just clean up. Things like flushing 3965098297b2SJeff Mahoney * journal lists, etc just won't happen. 39661da177e4SLinus Torvalds */ 396758d85426SJeff Mahoney static int do_journal_end(struct reiserfs_transaction_handle *th, int flags) 3968bd4c625cSLinus Torvalds { 396958d85426SJeff Mahoney struct super_block *sb = th->t_super; 3970a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 39711da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn, *next, *jl_cn; 39721da177e4SLinus Torvalds struct reiserfs_journal_cnode *last_cn = NULL; 39731da177e4SLinus Torvalds struct reiserfs_journal_desc *desc; 39741da177e4SLinus Torvalds struct reiserfs_journal_commit *commit; 39751da177e4SLinus Torvalds struct buffer_head *c_bh; /* commit bh */ 39761da177e4SLinus Torvalds struct buffer_head *d_bh; /* desc bh */ 39771da177e4SLinus Torvalds int cur_write_start = 0; /* start index of current log write */ 39781da177e4SLinus Torvalds int i; 3979a44c94a7SAlexander Zarochentsev int flush; 3980a44c94a7SAlexander Zarochentsev int wait_on_commit; 39811da177e4SLinus Torvalds struct reiserfs_journal_list *jl, *temp_jl; 39821da177e4SLinus Torvalds struct list_head *entry, *safe; 39831da177e4SLinus Torvalds unsigned long jindex; 3984600ed416SJeff Mahoney unsigned int commit_trans_id; 39851da177e4SLinus Torvalds int trans_half; 3986278f6679SJeff Mahoney int depth; 39871da177e4SLinus Torvalds 39881da177e4SLinus Torvalds BUG_ON(th->t_refcount > 1); 39891da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 399058d85426SJeff Mahoney BUG_ON(!th->t_super); 39911da177e4SLinus Torvalds 3992098297b2SJeff Mahoney /* 3993098297b2SJeff Mahoney * protect flush_older_commits from doing mistakes if the 3994098297b2SJeff Mahoney * transaction ID counter gets overflowed. 3995098297b2SJeff Mahoney */ 3996600ed416SJeff Mahoney if (th->t_trans_id == ~0U) 3997a44c94a7SAlexander Zarochentsev flags |= FLUSH_ALL | COMMIT_NOW | WAIT; 3998a44c94a7SAlexander Zarochentsev flush = flags & FLUSH_ALL; 3999a44c94a7SAlexander Zarochentsev wait_on_commit = flags & WAIT; 4000a44c94a7SAlexander Zarochentsev 40011da177e4SLinus Torvalds current->journal_info = th->t_handle_save; 4002a9dd3643SJeff Mahoney reiserfs_check_lock_depth(sb, "journal end"); 40031da177e4SLinus Torvalds if (journal->j_len == 0) { 4004a9dd3643SJeff Mahoney reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb), 4005bd4c625cSLinus Torvalds 1); 400609f1b80bSJeff Mahoney journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb)); 40071da177e4SLinus Torvalds } 40081da177e4SLinus Torvalds 4009a9dd3643SJeff Mahoney lock_journal(sb); 40101da177e4SLinus Torvalds if (journal->j_next_full_flush) { 40111da177e4SLinus Torvalds flags |= FLUSH_ALL; 40121da177e4SLinus Torvalds flush = 1; 40131da177e4SLinus Torvalds } 40141da177e4SLinus Torvalds if (journal->j_next_async_flush) { 40151da177e4SLinus Torvalds flags |= COMMIT_NOW | WAIT; 40161da177e4SLinus Torvalds wait_on_commit = 1; 40171da177e4SLinus Torvalds } 40181da177e4SLinus Torvalds 4019098297b2SJeff Mahoney /* 4020098297b2SJeff Mahoney * check_journal_end locks the journal, and unlocks if it does 4021098297b2SJeff Mahoney * not return 1 it tells us if we should continue with the 4022098297b2SJeff Mahoney * journal_end, or just return 40231da177e4SLinus Torvalds */ 402458d85426SJeff Mahoney if (!check_journal_end(th, flags)) { 4025033369d1SArtem Bityutskiy reiserfs_schedule_old_flush(sb); 4026a9dd3643SJeff Mahoney wake_queued_writers(sb); 4027a9dd3643SJeff Mahoney reiserfs_async_progress_wait(sb); 40281da177e4SLinus Torvalds goto out; 40291da177e4SLinus Torvalds } 40301da177e4SLinus Torvalds 40311da177e4SLinus Torvalds /* check_journal_end might set these, check again */ 40321da177e4SLinus Torvalds if (journal->j_next_full_flush) { 40331da177e4SLinus Torvalds flush = 1; 40341da177e4SLinus Torvalds } 40351da177e4SLinus Torvalds 40361da177e4SLinus Torvalds /* 4037098297b2SJeff Mahoney * j must wait means we have to flush the log blocks, and the 4038098297b2SJeff Mahoney * real blocks for this transaction 40391da177e4SLinus Torvalds */ 40401da177e4SLinus Torvalds if (journal->j_must_wait > 0) { 40411da177e4SLinus Torvalds flush = 1; 40421da177e4SLinus Torvalds } 40431da177e4SLinus Torvalds #ifdef REISERFS_PREALLOCATE 4044098297b2SJeff Mahoney /* 4045098297b2SJeff Mahoney * quota ops might need to nest, setup the journal_info pointer 4046098297b2SJeff Mahoney * for them and raise the refcount so that it is > 0. 4047098297b2SJeff Mahoney */ 40481da177e4SLinus Torvalds current->journal_info = th; 4049ef43bc4fSJan Kara th->t_refcount++; 4050098297b2SJeff Mahoney 4051098297b2SJeff Mahoney /* it should not involve new blocks into the transaction */ 4052098297b2SJeff Mahoney reiserfs_discard_all_prealloc(th); 4053098297b2SJeff Mahoney 4054ef43bc4fSJan Kara th->t_refcount--; 40551da177e4SLinus Torvalds current->journal_info = th->t_handle_save; 40561da177e4SLinus Torvalds #endif 40571da177e4SLinus Torvalds 40581da177e4SLinus Torvalds /* setup description block */ 4059bd4c625cSLinus Torvalds d_bh = 4060a9dd3643SJeff Mahoney journal_getblk(sb, 4061a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 4062bd4c625cSLinus Torvalds journal->j_start); 40631da177e4SLinus Torvalds set_buffer_uptodate(d_bh); 40641da177e4SLinus Torvalds desc = (struct reiserfs_journal_desc *)(d_bh)->b_data; 40651da177e4SLinus Torvalds memset(d_bh->b_data, 0, d_bh->b_size); 40661da177e4SLinus Torvalds memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8); 40671da177e4SLinus Torvalds set_desc_trans_id(desc, journal->j_trans_id); 40681da177e4SLinus Torvalds 4069098297b2SJeff Mahoney /* 4070098297b2SJeff Mahoney * setup commit block. Don't write (keep it clean too) this one 4071098297b2SJeff Mahoney * until after everyone else is written 4072098297b2SJeff Mahoney */ 4073a9dd3643SJeff Mahoney c_bh = journal_getblk(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 4074bd4c625cSLinus Torvalds ((journal->j_start + journal->j_len + 4075a9dd3643SJeff Mahoney 1) % SB_ONDISK_JOURNAL_SIZE(sb))); 40761da177e4SLinus Torvalds commit = (struct reiserfs_journal_commit *)c_bh->b_data; 40771da177e4SLinus Torvalds memset(c_bh->b_data, 0, c_bh->b_size); 40781da177e4SLinus Torvalds set_commit_trans_id(commit, journal->j_trans_id); 40791da177e4SLinus Torvalds set_buffer_uptodate(c_bh); 40801da177e4SLinus Torvalds 40811da177e4SLinus Torvalds /* init this journal list */ 40821da177e4SLinus Torvalds jl = journal->j_current_jl; 40831da177e4SLinus Torvalds 4084098297b2SJeff Mahoney /* 4085098297b2SJeff Mahoney * we lock the commit before doing anything because 40861da177e4SLinus Torvalds * we want to make sure nobody tries to run flush_commit_list until 40871da177e4SLinus Torvalds * the new transaction is fully setup, and we've already flushed the 40881da177e4SLinus Torvalds * ordered bh list 40891da177e4SLinus Torvalds */ 40908ebc4232SFrederic Weisbecker reiserfs_mutex_lock_safe(&jl->j_commit_mutex, sb); 40911da177e4SLinus Torvalds 40921da177e4SLinus Torvalds /* save the transaction id in case we need to commit it later */ 40931da177e4SLinus Torvalds commit_trans_id = jl->j_trans_id; 40941da177e4SLinus Torvalds 40951da177e4SLinus Torvalds atomic_set(&jl->j_older_commits_done, 0); 40961da177e4SLinus Torvalds jl->j_trans_id = journal->j_trans_id; 40971da177e4SLinus Torvalds jl->j_timestamp = journal->j_trans_start_time; 40981da177e4SLinus Torvalds jl->j_commit_bh = c_bh; 40991da177e4SLinus Torvalds jl->j_start = journal->j_start; 41001da177e4SLinus Torvalds jl->j_len = journal->j_len; 41011da177e4SLinus Torvalds atomic_set(&jl->j_nonzerolen, journal->j_len); 41021da177e4SLinus Torvalds atomic_set(&jl->j_commit_left, journal->j_len + 2); 41031da177e4SLinus Torvalds jl->j_realblock = NULL; 41041da177e4SLinus Torvalds 4105098297b2SJeff Mahoney /* 4106098297b2SJeff Mahoney * The ENTIRE FOR LOOP MUST not cause schedule to occur. 4107098297b2SJeff Mahoney * for each real block, add it to the journal list hash, 4108098297b2SJeff Mahoney * copy into real block index array in the commit or desc block 41091da177e4SLinus Torvalds */ 4110a9dd3643SJeff Mahoney trans_half = journal_trans_half(sb->s_blocksize); 41111da177e4SLinus Torvalds for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) { 41121da177e4SLinus Torvalds if (buffer_journaled(cn->bh)) { 4113a9dd3643SJeff Mahoney jl_cn = get_cnode(sb); 41141da177e4SLinus Torvalds if (!jl_cn) { 4115a9dd3643SJeff Mahoney reiserfs_panic(sb, "journal-1676", 4116c3a9c210SJeff Mahoney "get_cnode returned NULL"); 41171da177e4SLinus Torvalds } 41181da177e4SLinus Torvalds if (i == 0) { 41191da177e4SLinus Torvalds jl->j_realblock = jl_cn; 41201da177e4SLinus Torvalds } 41211da177e4SLinus Torvalds jl_cn->prev = last_cn; 41221da177e4SLinus Torvalds jl_cn->next = NULL; 41231da177e4SLinus Torvalds if (last_cn) { 41241da177e4SLinus Torvalds last_cn->next = jl_cn; 41251da177e4SLinus Torvalds } 41261da177e4SLinus Torvalds last_cn = jl_cn; 4127098297b2SJeff Mahoney /* 4128098297b2SJeff Mahoney * make sure the block we are trying to log 4129098297b2SJeff Mahoney * is not a block of journal or reserved area 4130098297b2SJeff Mahoney */ 4131bd4c625cSLinus Torvalds if (is_block_in_log_or_reserved_area 4132a9dd3643SJeff Mahoney (sb, cn->bh->b_blocknr)) { 4133a9dd3643SJeff Mahoney reiserfs_panic(sb, "journal-2332", 4134c3a9c210SJeff Mahoney "Trying to log block %lu, " 4135c3a9c210SJeff Mahoney "which is a log block", 4136bd4c625cSLinus Torvalds cn->bh->b_blocknr); 41371da177e4SLinus Torvalds } 41381da177e4SLinus Torvalds jl_cn->blocknr = cn->bh->b_blocknr; 41391da177e4SLinus Torvalds jl_cn->state = 0; 4140a9dd3643SJeff Mahoney jl_cn->sb = sb; 41411da177e4SLinus Torvalds jl_cn->bh = cn->bh; 41421da177e4SLinus Torvalds jl_cn->jlist = jl; 41431da177e4SLinus Torvalds insert_journal_hash(journal->j_list_hash_table, jl_cn); 41441da177e4SLinus Torvalds if (i < trans_half) { 4145bd4c625cSLinus Torvalds desc->j_realblock[i] = 4146bd4c625cSLinus Torvalds cpu_to_le32(cn->bh->b_blocknr); 41471da177e4SLinus Torvalds } else { 4148bd4c625cSLinus Torvalds commit->j_realblock[i - trans_half] = 4149bd4c625cSLinus Torvalds cpu_to_le32(cn->bh->b_blocknr); 41501da177e4SLinus Torvalds } 41511da177e4SLinus Torvalds } else { 41521da177e4SLinus Torvalds i--; 41531da177e4SLinus Torvalds } 41541da177e4SLinus Torvalds } 41551da177e4SLinus Torvalds set_desc_trans_len(desc, journal->j_len); 41561da177e4SLinus Torvalds set_desc_mount_id(desc, journal->j_mount_id); 41571da177e4SLinus Torvalds set_desc_trans_id(desc, journal->j_trans_id); 41581da177e4SLinus Torvalds set_commit_trans_len(commit, journal->j_len); 41591da177e4SLinus Torvalds 4160098297b2SJeff Mahoney /* 4161098297b2SJeff Mahoney * special check in case all buffers in the journal 4162098297b2SJeff Mahoney * were marked for not logging 4163098297b2SJeff Mahoney */ 416414a61442SEric Sesterhenn BUG_ON(journal->j_len == 0); 41651da177e4SLinus Torvalds 4166098297b2SJeff Mahoney /* 4167098297b2SJeff Mahoney * we're about to dirty all the log blocks, mark the description block 41681da177e4SLinus Torvalds * dirty now too. Don't mark the commit block dirty until all the 41691da177e4SLinus Torvalds * others are on disk 41701da177e4SLinus Torvalds */ 41711da177e4SLinus Torvalds mark_buffer_dirty(d_bh); 41721da177e4SLinus Torvalds 4173098297b2SJeff Mahoney /* 4174098297b2SJeff Mahoney * first data block is j_start + 1, so add one to 4175098297b2SJeff Mahoney * cur_write_start wherever you use it 4176098297b2SJeff Mahoney */ 41771da177e4SLinus Torvalds cur_write_start = journal->j_start; 41781da177e4SLinus Torvalds cn = journal->j_first; 41791da177e4SLinus Torvalds jindex = 1; /* start at one so we don't get the desc again */ 41801da177e4SLinus Torvalds while (cn) { 41811da177e4SLinus Torvalds clear_buffer_journal_new(cn->bh); 41821da177e4SLinus Torvalds /* copy all the real blocks into log area. dirty log blocks */ 41831da177e4SLinus Torvalds if (buffer_journaled(cn->bh)) { 41841da177e4SLinus Torvalds struct buffer_head *tmp_bh; 41851da177e4SLinus Torvalds char *addr; 41861da177e4SLinus Torvalds struct page *page; 4187bd4c625cSLinus Torvalds tmp_bh = 4188a9dd3643SJeff Mahoney journal_getblk(sb, 4189a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 4190bd4c625cSLinus Torvalds ((cur_write_start + 4191bd4c625cSLinus Torvalds jindex) % 4192a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb))); 41931da177e4SLinus Torvalds set_buffer_uptodate(tmp_bh); 41941da177e4SLinus Torvalds page = cn->bh->b_page; 41951da177e4SLinus Torvalds addr = kmap(page); 4196bd4c625cSLinus Torvalds memcpy(tmp_bh->b_data, 4197bd4c625cSLinus Torvalds addr + offset_in_page(cn->bh->b_data), 41981da177e4SLinus Torvalds cn->bh->b_size); 41991da177e4SLinus Torvalds kunmap(page); 42001da177e4SLinus Torvalds mark_buffer_dirty(tmp_bh); 42011da177e4SLinus Torvalds jindex++; 42021da177e4SLinus Torvalds set_buffer_journal_dirty(cn->bh); 42031da177e4SLinus Torvalds clear_buffer_journaled(cn->bh); 42041da177e4SLinus Torvalds } else { 4205098297b2SJeff Mahoney /* 4206098297b2SJeff Mahoney * JDirty cleared sometime during transaction. 4207098297b2SJeff Mahoney * don't log this one 4208098297b2SJeff Mahoney */ 4209a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-2048", 421045b03d5eSJeff Mahoney "BAD, buffer in journal hash, " 421145b03d5eSJeff Mahoney "but not JDirty!"); 42121da177e4SLinus Torvalds brelse(cn->bh); 42131da177e4SLinus Torvalds } 42141da177e4SLinus Torvalds next = cn->next; 4215a9dd3643SJeff Mahoney free_cnode(sb, cn); 42161da177e4SLinus Torvalds cn = next; 4217278f6679SJeff Mahoney reiserfs_cond_resched(sb); 42181da177e4SLinus Torvalds } 42191da177e4SLinus Torvalds 4220098297b2SJeff Mahoney /* 4221098297b2SJeff Mahoney * we are done with both the c_bh and d_bh, but 4222098297b2SJeff Mahoney * c_bh must be written after all other commit blocks, 4223098297b2SJeff Mahoney * so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1. 42241da177e4SLinus Torvalds */ 42251da177e4SLinus Torvalds 4226a9dd3643SJeff Mahoney journal->j_current_jl = alloc_journal_list(sb); 42271da177e4SLinus Torvalds 42281da177e4SLinus Torvalds /* now it is safe to insert this transaction on the main list */ 42291da177e4SLinus Torvalds list_add_tail(&jl->j_list, &journal->j_journal_list); 42301da177e4SLinus Torvalds list_add_tail(&jl->j_working_list, &journal->j_working_list); 42311da177e4SLinus Torvalds journal->j_num_work_lists++; 42321da177e4SLinus Torvalds 42331da177e4SLinus Torvalds /* reset journal values for the next transaction */ 4234bd4c625cSLinus Torvalds journal->j_start = 4235bd4c625cSLinus Torvalds (journal->j_start + journal->j_len + 4236a9dd3643SJeff Mahoney 2) % SB_ONDISK_JOURNAL_SIZE(sb); 4237a228bf8fSJeff Mahoney atomic_set(&journal->j_wcount, 0); 42381da177e4SLinus Torvalds journal->j_bcount = 0; 42391da177e4SLinus Torvalds journal->j_last = NULL; 42401da177e4SLinus Torvalds journal->j_first = NULL; 42411da177e4SLinus Torvalds journal->j_len = 0; 42421da177e4SLinus Torvalds journal->j_trans_start_time = 0; 4243a44c94a7SAlexander Zarochentsev /* check for trans_id overflow */ 4244a44c94a7SAlexander Zarochentsev if (++journal->j_trans_id == 0) 4245a44c94a7SAlexander Zarochentsev journal->j_trans_id = 10; 42461da177e4SLinus Torvalds journal->j_current_jl->j_trans_id = journal->j_trans_id; 42471da177e4SLinus Torvalds journal->j_must_wait = 0; 42481da177e4SLinus Torvalds journal->j_len_alloc = 0; 42491da177e4SLinus Torvalds journal->j_next_full_flush = 0; 42501da177e4SLinus Torvalds journal->j_next_async_flush = 0; 4251a9dd3643SJeff Mahoney init_journal_hash(sb); 42521da177e4SLinus Torvalds 4253098297b2SJeff Mahoney /* 4254098297b2SJeff Mahoney * make sure reiserfs_add_jh sees the new current_jl before we 4255098297b2SJeff Mahoney * write out the tails 4256098297b2SJeff Mahoney */ 42571da177e4SLinus Torvalds smp_mb(); 42581da177e4SLinus Torvalds 4259098297b2SJeff Mahoney /* 4260098297b2SJeff Mahoney * tail conversion targets have to hit the disk before we end the 42611da177e4SLinus Torvalds * transaction. Otherwise a later transaction might repack the tail 4262098297b2SJeff Mahoney * before this transaction commits, leaving the data block unflushed 4263098297b2SJeff Mahoney * and clean, if we crash before the later transaction commits, the 4264098297b2SJeff Mahoney * data block is lost. 42651da177e4SLinus Torvalds */ 42661da177e4SLinus Torvalds if (!list_empty(&jl->j_tail_bh_list)) { 4267278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(sb); 42681da177e4SLinus Torvalds write_ordered_buffers(&journal->j_dirty_buffers_lock, 42691da177e4SLinus Torvalds journal, jl, &jl->j_tail_bh_list); 4270278f6679SJeff Mahoney reiserfs_write_lock_nested(sb, depth); 42711da177e4SLinus Torvalds } 427214a61442SEric Sesterhenn BUG_ON(!list_empty(&jl->j_tail_bh_list)); 427390415deaSJeff Mahoney mutex_unlock(&jl->j_commit_mutex); 42741da177e4SLinus Torvalds 4275098297b2SJeff Mahoney /* 4276098297b2SJeff Mahoney * honor the flush wishes from the caller, simple commits can 4277098297b2SJeff Mahoney * be done outside the journal lock, they are done below 4278098297b2SJeff Mahoney * 4279098297b2SJeff Mahoney * if we don't flush the commit list right now, we put it into 4280098297b2SJeff Mahoney * the work queue so the people waiting on the async progress work 4281098297b2SJeff Mahoney * queue don't wait for this proc to flush journal lists and such. 42821da177e4SLinus Torvalds */ 42831da177e4SLinus Torvalds if (flush) { 4284a9dd3643SJeff Mahoney flush_commit_list(sb, jl, 1); 4285a9dd3643SJeff Mahoney flush_journal_list(sb, jl, 1); 428601777836SJan Kara } else if (!(jl->j_state & LIST_COMMIT_PENDING)) { 428701777836SJan Kara /* 428801777836SJan Kara * Avoid queueing work when sb is being shut down. Transaction 428901777836SJan Kara * will be flushed on journal shutdown. 429001777836SJan Kara */ 42911751e8a6SLinus Torvalds if (sb->s_flags & SB_ACTIVE) 4292797d9016SJeff Mahoney queue_delayed_work(REISERFS_SB(sb)->commit_wq, 4293797d9016SJeff Mahoney &journal->j_work, HZ / 10); 429401777836SJan Kara } 42951da177e4SLinus Torvalds 4296098297b2SJeff Mahoney /* 4297098297b2SJeff Mahoney * if the next transaction has any chance of wrapping, flush 4298098297b2SJeff Mahoney * transactions that might get overwritten. If any journal lists 4299098297b2SJeff Mahoney * are very old flush them as well. 43001da177e4SLinus Torvalds */ 43011da177e4SLinus Torvalds first_jl: 43021da177e4SLinus Torvalds list_for_each_safe(entry, safe, &journal->j_journal_list) { 43031da177e4SLinus Torvalds temp_jl = JOURNAL_LIST_ENTRY(entry); 43041da177e4SLinus Torvalds if (journal->j_start <= temp_jl->j_start) { 43051da177e4SLinus Torvalds if ((journal->j_start + journal->j_trans_max + 1) >= 4306bd4c625cSLinus Torvalds temp_jl->j_start) { 4307a9dd3643SJeff Mahoney flush_used_journal_lists(sb, temp_jl); 43081da177e4SLinus Torvalds goto first_jl; 43091da177e4SLinus Torvalds } else if ((journal->j_start + 43101da177e4SLinus Torvalds journal->j_trans_max + 1) < 4311a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb)) { 4312098297b2SJeff Mahoney /* 4313098297b2SJeff Mahoney * if we don't cross into the next 4314098297b2SJeff Mahoney * transaction and we don't wrap, there is 4315098297b2SJeff Mahoney * no way we can overlap any later transactions 43161da177e4SLinus Torvalds * break now 43171da177e4SLinus Torvalds */ 43181da177e4SLinus Torvalds break; 43191da177e4SLinus Torvalds } 43201da177e4SLinus Torvalds } else if ((journal->j_start + 43211da177e4SLinus Torvalds journal->j_trans_max + 1) > 4322a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb)) { 43231da177e4SLinus Torvalds if (((journal->j_start + journal->j_trans_max + 1) % 4324a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb)) >= 4325bd4c625cSLinus Torvalds temp_jl->j_start) { 4326a9dd3643SJeff Mahoney flush_used_journal_lists(sb, temp_jl); 43271da177e4SLinus Torvalds goto first_jl; 43281da177e4SLinus Torvalds } else { 4329098297b2SJeff Mahoney /* 4330098297b2SJeff Mahoney * we don't overlap anything from out start 4331098297b2SJeff Mahoney * to the end of the log, and our wrapped 4332098297b2SJeff Mahoney * portion doesn't overlap anything at 43331da177e4SLinus Torvalds * the start of the log. We can break 43341da177e4SLinus Torvalds */ 43351da177e4SLinus Torvalds break; 43361da177e4SLinus Torvalds } 43371da177e4SLinus Torvalds } 43381da177e4SLinus Torvalds } 43391da177e4SLinus Torvalds 4340bd4c625cSLinus Torvalds journal->j_current_jl->j_list_bitmap = 4341a9dd3643SJeff Mahoney get_list_bitmap(sb, journal->j_current_jl); 43421da177e4SLinus Torvalds 43431da177e4SLinus Torvalds if (!(journal->j_current_jl->j_list_bitmap)) { 4344a9dd3643SJeff Mahoney reiserfs_panic(sb, "journal-1996", 4345c3a9c210SJeff Mahoney "could not get a list bitmap"); 43461da177e4SLinus Torvalds } 43471da177e4SLinus Torvalds 4348a228bf8fSJeff Mahoney atomic_set(&journal->j_jlock, 0); 4349a9dd3643SJeff Mahoney unlock_journal(sb); 43501da177e4SLinus Torvalds /* wake up any body waiting to join. */ 43511da177e4SLinus Torvalds clear_bit(J_WRITERS_QUEUED, &journal->j_state); 4352a228bf8fSJeff Mahoney wake_up(&journal->j_join_wait); 43531da177e4SLinus Torvalds 43541da177e4SLinus Torvalds if (!flush && wait_on_commit && 4355a9dd3643SJeff Mahoney journal_list_still_alive(sb, commit_trans_id)) { 4356a9dd3643SJeff Mahoney flush_commit_list(sb, jl, 1); 43571da177e4SLinus Torvalds } 43581da177e4SLinus Torvalds out: 4359a9dd3643SJeff Mahoney reiserfs_check_lock_depth(sb, "journal end2"); 43601da177e4SLinus Torvalds 43611da177e4SLinus Torvalds memset(th, 0, sizeof(*th)); 4362098297b2SJeff Mahoney /* 4363098297b2SJeff Mahoney * Re-set th->t_super, so we can properly keep track of how many 43641da177e4SLinus Torvalds * persistent transactions there are. We need to do this so if this 4365098297b2SJeff Mahoney * call is part of a failed restart_transaction, we can free it later 4366098297b2SJeff Mahoney */ 4367a9dd3643SJeff Mahoney th->t_super = sb; 43681da177e4SLinus Torvalds 43691da177e4SLinus Torvalds return journal->j_errno; 43701da177e4SLinus Torvalds } 43711da177e4SLinus Torvalds 437232e8b106SJeff Mahoney /* Send the file system read only and refuse new transactions */ 437332e8b106SJeff Mahoney void reiserfs_abort_journal(struct super_block *sb, int errno) 43741da177e4SLinus Torvalds { 43751da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(sb); 43761da177e4SLinus Torvalds if (test_bit(J_ABORTED, &journal->j_state)) 43771da177e4SLinus Torvalds return; 43781da177e4SLinus Torvalds 437932e8b106SJeff Mahoney if (!journal->j_errno) 438032e8b106SJeff Mahoney journal->j_errno = errno; 43811da177e4SLinus Torvalds 43821751e8a6SLinus Torvalds sb->s_flags |= SB_RDONLY; 43831da177e4SLinus Torvalds set_bit(J_ABORTED, &journal->j_state); 43841da177e4SLinus Torvalds 43851da177e4SLinus Torvalds #ifdef CONFIG_REISERFS_CHECK 43861da177e4SLinus Torvalds dump_stack(); 43871da177e4SLinus Torvalds #endif 43881da177e4SLinus Torvalds } 4389