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 499a3cc754aSTian 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 */ 504a3cc754aSTian 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 { 604ac55e78dSMatthew Wilcox (Oracle) struct folio *folio = bh->b_folio; 605be6b969cSMatthew Wilcox (Oracle) if (!folio->mapping && folio_trylock(folio)) { 606be6b969cSMatthew Wilcox (Oracle) folio_get(folio); 607398c95bdSChris Mason put_bh(bh); 608be6b969cSMatthew Wilcox (Oracle) if (!folio->mapping) 60968189fefSMatthew Wilcox (Oracle) try_to_free_buffers(folio); 610be6b969cSMatthew Wilcox (Oracle) folio_unlock(folio); 611be6b969cSMatthew Wilcox (Oracle) folio_put(folio); 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(); 6531420c4a5SBart Van Assche submit_bh(REQ_OP_WRITE, 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(); 6631420c4a5SBart Van Assche submit_bh(REQ_OP_WRITE, 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 /* 861d97dfc94SMatthew Wilcox (Oracle) * ugly interaction with invalidate_folio here. 862d97dfc94SMatthew Wilcox (Oracle) * reiserfs_invalidate_folio 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 */ 869ac55e78dSMatthew Wilcox (Oracle) if (buffer_dirty(bh) && unlikely(bh->b_folio->mapping == NULL)) { 870d62b1b87SChris Mason spin_unlock(lock); 871d554822eSZhang Yi write_dirty_buffer(bh, 0); 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); 954b831281bSNeilBrown wait_var_event_timeout(&j->j_async_throttle, 955b831281bSNeilBrown atomic_read(&j->j_async_throttle) == 0, 956b831281bSNeilBrown HZ / 10); 957278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 9588ebc4232SFrederic Weisbecker } 9598ebc4232SFrederic Weisbecker 9601da177e4SLinus Torvalds return 0; 9611da177e4SLinus Torvalds } 9621da177e4SLinus Torvalds 9631da177e4SLinus Torvalds /* 964098297b2SJeff Mahoney * if this journal list still has commit blocks unflushed, send them to disk. 965098297b2SJeff Mahoney * 966098297b2SJeff Mahoney * log areas must be flushed in order (transaction 2 can't commit before 967098297b2SJeff Mahoney * transaction 1) Before the commit block can by written, every other log 968098297b2SJeff Mahoney * block must be safely on disk 9691da177e4SLinus Torvalds */ 970bd4c625cSLinus Torvalds static int flush_commit_list(struct super_block *s, 971bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl, int flushall) 972bd4c625cSLinus Torvalds { 9731da177e4SLinus Torvalds int i; 9743ee16670SJeff Mahoney b_blocknr_t bn; 9751da177e4SLinus Torvalds struct buffer_head *tbh = NULL; 976600ed416SJeff Mahoney unsigned int trans_id = jl->j_trans_id; 9771da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 9781da177e4SLinus Torvalds int retval = 0; 979e0e851cfSChris Mason int write_len; 980278f6679SJeff Mahoney int depth; 9811da177e4SLinus Torvalds 9821da177e4SLinus Torvalds reiserfs_check_lock_depth(s, "flush_commit_list"); 9831da177e4SLinus Torvalds 9841da177e4SLinus Torvalds if (atomic_read(&jl->j_older_commits_done)) { 9851da177e4SLinus Torvalds return 0; 9861da177e4SLinus Torvalds } 9871da177e4SLinus Torvalds 988098297b2SJeff Mahoney /* 989098297b2SJeff Mahoney * before we can put our commit blocks on disk, we have to make 990098297b2SJeff Mahoney * sure everyone older than us is on disk too 9911da177e4SLinus Torvalds */ 9921da177e4SLinus Torvalds BUG_ON(jl->j_len <= 0); 9931da177e4SLinus Torvalds BUG_ON(trans_id == journal->j_trans_id); 9941da177e4SLinus Torvalds 9951da177e4SLinus Torvalds get_journal_list(jl); 9961da177e4SLinus Torvalds if (flushall) { 9971da177e4SLinus Torvalds if (flush_older_commits(s, jl) == 1) { 998098297b2SJeff Mahoney /* 999098297b2SJeff Mahoney * list disappeared during flush_older_commits. 1000098297b2SJeff Mahoney * return 1001098297b2SJeff Mahoney */ 10021da177e4SLinus Torvalds goto put_jl; 10031da177e4SLinus Torvalds } 10041da177e4SLinus Torvalds } 10051da177e4SLinus Torvalds 10061da177e4SLinus Torvalds /* make sure nobody is trying to flush this one at the same time */ 10078ebc4232SFrederic Weisbecker reiserfs_mutex_lock_safe(&jl->j_commit_mutex, s); 10088ebc4232SFrederic Weisbecker 10091da177e4SLinus Torvalds if (!journal_list_still_alive(s, trans_id)) { 101090415deaSJeff Mahoney mutex_unlock(&jl->j_commit_mutex); 10111da177e4SLinus Torvalds goto put_jl; 10121da177e4SLinus Torvalds } 10131da177e4SLinus Torvalds BUG_ON(jl->j_trans_id == 0); 10141da177e4SLinus Torvalds 10151da177e4SLinus Torvalds /* this commit is done, exit */ 1016a228bf8fSJeff Mahoney if (atomic_read(&jl->j_commit_left) <= 0) { 10171da177e4SLinus Torvalds if (flushall) { 1018a228bf8fSJeff Mahoney atomic_set(&jl->j_older_commits_done, 1); 10191da177e4SLinus Torvalds } 102090415deaSJeff Mahoney mutex_unlock(&jl->j_commit_mutex); 10211da177e4SLinus Torvalds goto put_jl; 10221da177e4SLinus Torvalds } 10231da177e4SLinus Torvalds 10241da177e4SLinus Torvalds if (!list_empty(&jl->j_bh_list)) { 10253d4492f8SChris Mason int ret; 10268ebc4232SFrederic Weisbecker 10278ebc4232SFrederic Weisbecker /* 10288ebc4232SFrederic Weisbecker * We might sleep in numerous places inside 10298ebc4232SFrederic Weisbecker * write_ordered_buffers. Relax the write lock. 10308ebc4232SFrederic Weisbecker */ 1031278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(s); 10323d4492f8SChris Mason ret = write_ordered_buffers(&journal->j_dirty_buffers_lock, 10331da177e4SLinus Torvalds journal, jl, &jl->j_bh_list); 10343d4492f8SChris Mason if (ret < 0 && retval == 0) 10353d4492f8SChris Mason retval = ret; 1036278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 10371da177e4SLinus Torvalds } 10381da177e4SLinus Torvalds BUG_ON(!list_empty(&jl->j_bh_list)); 10391da177e4SLinus Torvalds /* 10401da177e4SLinus Torvalds * for the description block and all the log blocks, submit any buffers 1041e0e851cfSChris Mason * that haven't already reached the disk. Try to write at least 256 1042e0e851cfSChris Mason * log blocks. later on, we will only wait on blocks that correspond 1043e0e851cfSChris Mason * to this transaction, but while we're unplugging we might as well 1044e0e851cfSChris Mason * get a chunk of data on there. 10451da177e4SLinus Torvalds */ 10461da177e4SLinus Torvalds atomic_inc(&journal->j_async_throttle); 1047e0e851cfSChris Mason write_len = jl->j_len + 1; 1048e0e851cfSChris Mason if (write_len < 256) 1049e0e851cfSChris Mason write_len = 256; 1050e0e851cfSChris Mason for (i = 0 ; i < write_len ; i++) { 10511da177e4SLinus Torvalds bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) % 10521da177e4SLinus Torvalds SB_ONDISK_JOURNAL_SIZE(s); 10531da177e4SLinus Torvalds tbh = journal_find_get_block(s, bn); 1054e0e851cfSChris Mason if (tbh) { 10556e3647acSFrederic Weisbecker if (buffer_dirty(tbh)) { 1056278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(s); 1057d554822eSZhang Yi write_dirty_buffer(tbh, 0); 1058278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 10596e3647acSFrederic Weisbecker } 10601da177e4SLinus Torvalds put_bh(tbh) ; 10611da177e4SLinus Torvalds } 1062e0e851cfSChris Mason } 1063b831281bSNeilBrown if (atomic_dec_and_test(&journal->j_async_throttle)) 1064b831281bSNeilBrown wake_up_var(&journal->j_async_throttle); 10651da177e4SLinus Torvalds 10661da177e4SLinus Torvalds for (i = 0; i < (jl->j_len + 1); i++) { 10671da177e4SLinus Torvalds bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + 10681da177e4SLinus Torvalds (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s); 10691da177e4SLinus Torvalds tbh = journal_find_get_block(s, bn); 10708ebc4232SFrederic Weisbecker 1071278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(s); 1072278f6679SJeff Mahoney __wait_on_buffer(tbh); 1073278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 1074098297b2SJeff Mahoney /* 1075098297b2SJeff Mahoney * since we're using ll_rw_blk above, it might have skipped 1076098297b2SJeff Mahoney * over a locked buffer. Double check here 1077098297b2SJeff Mahoney */ 10788ebc4232SFrederic Weisbecker /* redundant, sync_dirty_buffer() checks */ 10798ebc4232SFrederic Weisbecker if (buffer_dirty(tbh)) { 1080278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(s); 10811da177e4SLinus Torvalds sync_dirty_buffer(tbh); 1082278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 10838ebc4232SFrederic Weisbecker } 10841da177e4SLinus Torvalds if (unlikely(!buffer_uptodate(tbh))) { 10851da177e4SLinus Torvalds #ifdef CONFIG_REISERFS_CHECK 108645b03d5eSJeff Mahoney reiserfs_warning(s, "journal-601", 108745b03d5eSJeff Mahoney "buffer write failed"); 10881da177e4SLinus Torvalds #endif 10891da177e4SLinus Torvalds retval = -EIO; 10901da177e4SLinus Torvalds } 1091098297b2SJeff Mahoney /* once for journal_find_get_block */ 1092098297b2SJeff Mahoney put_bh(tbh); 1093098297b2SJeff Mahoney /* once due to original getblk in do_journal_end */ 1094098297b2SJeff Mahoney put_bh(tbh); 1095a228bf8fSJeff Mahoney atomic_dec(&jl->j_commit_left); 10961da177e4SLinus Torvalds } 10971da177e4SLinus Torvalds 1098a228bf8fSJeff Mahoney BUG_ON(atomic_read(&jl->j_commit_left) != 1); 10991da177e4SLinus Torvalds 1100098297b2SJeff Mahoney /* 1101098297b2SJeff Mahoney * If there was a write error in the journal - we can't commit 11025d5e8156SJeff Mahoney * this transaction - it will be invalid and, if successful, 1103beb7dd86SRobert P. J. Day * will just end up propagating the write error out to 1104098297b2SJeff Mahoney * the file system. 1105098297b2SJeff Mahoney */ 11065d5e8156SJeff Mahoney if (likely(!retval && !reiserfs_is_journal_aborted (journal))) { 11071da177e4SLinus Torvalds if (buffer_dirty(jl->j_commit_bh)) 11081da177e4SLinus Torvalds BUG(); 11091da177e4SLinus Torvalds mark_buffer_dirty(jl->j_commit_bh) ; 1110278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(s); 11117cd33ad2SChristoph Hellwig if (reiserfs_barrier_flush(s)) 111270fd7614SChristoph Hellwig __sync_dirty_buffer(jl->j_commit_bh, 1113d8747d64SJan Kara REQ_SYNC | REQ_PREFLUSH | REQ_FUA); 11147cd33ad2SChristoph Hellwig else 11151da177e4SLinus Torvalds sync_dirty_buffer(jl->j_commit_bh); 1116278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 11175d5e8156SJeff Mahoney } 11181da177e4SLinus Torvalds 1119098297b2SJeff Mahoney /* 1120098297b2SJeff Mahoney * If there was a write error in the journal - we can't commit this 11211da177e4SLinus Torvalds * transaction - it will be invalid and, if successful, will just end 1122098297b2SJeff Mahoney * up propagating the write error out to the filesystem. 1123098297b2SJeff Mahoney */ 11241da177e4SLinus Torvalds if (unlikely(!buffer_uptodate(jl->j_commit_bh))) { 11251da177e4SLinus Torvalds #ifdef CONFIG_REISERFS_CHECK 112645b03d5eSJeff Mahoney reiserfs_warning(s, "journal-615", "buffer write failed"); 11271da177e4SLinus Torvalds #endif 11281da177e4SLinus Torvalds retval = -EIO; 11291da177e4SLinus Torvalds } 11301da177e4SLinus Torvalds bforget(jl->j_commit_bh); 11311da177e4SLinus Torvalds if (journal->j_last_commit_id != 0 && 11321da177e4SLinus Torvalds (jl->j_trans_id - journal->j_last_commit_id) != 1) { 113345b03d5eSJeff Mahoney reiserfs_warning(s, "clm-2200", "last commit %lu, current %lu", 1134bd4c625cSLinus Torvalds journal->j_last_commit_id, jl->j_trans_id); 11351da177e4SLinus Torvalds } 11361da177e4SLinus Torvalds journal->j_last_commit_id = jl->j_trans_id; 11371da177e4SLinus Torvalds 1138098297b2SJeff Mahoney /* 1139098297b2SJeff Mahoney * now, every commit block is on the disk. It is safe to allow 1140098297b2SJeff Mahoney * blocks freed during this transaction to be reallocated 1141098297b2SJeff Mahoney */ 11421da177e4SLinus Torvalds cleanup_freed_for_journal_list(s, jl); 11431da177e4SLinus Torvalds 11441da177e4SLinus Torvalds retval = retval ? retval : journal->j_errno; 11451da177e4SLinus Torvalds 11461da177e4SLinus Torvalds /* mark the metadata dirty */ 11471da177e4SLinus Torvalds if (!retval) 11481da177e4SLinus Torvalds dirty_one_transaction(s, jl); 1149a228bf8fSJeff Mahoney atomic_dec(&jl->j_commit_left); 11501da177e4SLinus Torvalds 11511da177e4SLinus Torvalds if (flushall) { 1152a228bf8fSJeff Mahoney atomic_set(&jl->j_older_commits_done, 1); 11531da177e4SLinus Torvalds } 115490415deaSJeff Mahoney mutex_unlock(&jl->j_commit_mutex); 11551da177e4SLinus Torvalds put_jl: 11561da177e4SLinus Torvalds put_journal_list(s, jl); 11571da177e4SLinus Torvalds 11581da177e4SLinus Torvalds if (retval) 1159bd4c625cSLinus Torvalds reiserfs_abort(s, retval, "Journal write error in %s", 1160fbe5498bSHarvey Harrison __func__); 11611da177e4SLinus Torvalds return retval; 11621da177e4SLinus Torvalds } 11631da177e4SLinus Torvalds 11641da177e4SLinus Torvalds /* 1165098297b2SJeff Mahoney * flush_journal_list frequently needs to find a newer transaction for a 1166098297b2SJeff Mahoney * given block. This does that, or returns NULL if it can't find anything 11671da177e4SLinus Torvalds */ 1168bd4c625cSLinus Torvalds static struct reiserfs_journal_list *find_newer_jl_for_cn(struct 1169bd4c625cSLinus Torvalds reiserfs_journal_cnode 1170bd4c625cSLinus Torvalds *cn) 1171bd4c625cSLinus Torvalds { 11721da177e4SLinus Torvalds struct super_block *sb = cn->sb; 11731da177e4SLinus Torvalds b_blocknr_t blocknr = cn->blocknr; 11741da177e4SLinus Torvalds 11751da177e4SLinus Torvalds cn = cn->hprev; 11761da177e4SLinus Torvalds while (cn) { 11771da177e4SLinus Torvalds if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) { 11781da177e4SLinus Torvalds return cn->jlist; 11791da177e4SLinus Torvalds } 11801da177e4SLinus Torvalds cn = cn->hprev; 11811da177e4SLinus Torvalds } 11821da177e4SLinus Torvalds return NULL; 11831da177e4SLinus Torvalds } 11841da177e4SLinus Torvalds 1185bd4c625cSLinus Torvalds static void remove_journal_hash(struct super_block *, 1186bd4c625cSLinus Torvalds struct reiserfs_journal_cnode **, 1187bd4c625cSLinus Torvalds struct reiserfs_journal_list *, unsigned long, 1188bd4c625cSLinus Torvalds int); 11891da177e4SLinus Torvalds 11901da177e4SLinus Torvalds /* 1191098297b2SJeff Mahoney * once all the real blocks have been flushed, it is safe to remove them 1192098297b2SJeff Mahoney * from the journal list for this transaction. Aside from freeing the 1193098297b2SJeff Mahoney * cnode, this also allows the block to be reallocated for data blocks 1194098297b2SJeff Mahoney * if it had been deleted. 11951da177e4SLinus Torvalds */ 1196a9dd3643SJeff Mahoney static void remove_all_from_journal_list(struct super_block *sb, 1197bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl, 1198bd4c625cSLinus Torvalds int debug) 1199bd4c625cSLinus Torvalds { 1200a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 12011da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn, *last; 12021da177e4SLinus Torvalds cn = jl->j_realblock; 12031da177e4SLinus Torvalds 1204098297b2SJeff Mahoney /* 1205098297b2SJeff Mahoney * which is better, to lock once around the whole loop, or 1206098297b2SJeff Mahoney * to lock for each call to remove_journal_hash? 12071da177e4SLinus Torvalds */ 12081da177e4SLinus Torvalds while (cn) { 12091da177e4SLinus Torvalds if (cn->blocknr != 0) { 12101da177e4SLinus Torvalds if (debug) { 1211a9dd3643SJeff Mahoney reiserfs_warning(sb, "reiserfs-2201", 1212bd4c625cSLinus Torvalds "block %u, bh is %d, state %ld", 1213bd4c625cSLinus Torvalds cn->blocknr, cn->bh ? 1 : 0, 1214bd4c625cSLinus Torvalds cn->state); 12151da177e4SLinus Torvalds } 12161da177e4SLinus Torvalds cn->state = 0; 1217a9dd3643SJeff Mahoney remove_journal_hash(sb, journal->j_list_hash_table, 1218bd4c625cSLinus Torvalds jl, cn->blocknr, 1); 12191da177e4SLinus Torvalds } 12201da177e4SLinus Torvalds last = cn; 12211da177e4SLinus Torvalds cn = cn->next; 1222a9dd3643SJeff Mahoney free_cnode(sb, last); 12231da177e4SLinus Torvalds } 12241da177e4SLinus Torvalds jl->j_realblock = NULL; 12251da177e4SLinus Torvalds } 12261da177e4SLinus Torvalds 12271da177e4SLinus Torvalds /* 1228098297b2SJeff Mahoney * if this timestamp is greater than the timestamp we wrote last to the 1229098297b2SJeff Mahoney * header block, write it to the header block. once this is done, I can 1230098297b2SJeff Mahoney * safely say the log area for this transaction won't ever be replayed, 1231098297b2SJeff Mahoney * and I can start releasing blocks in this transaction for reuse as data 1232098297b2SJeff Mahoney * blocks. called by flush_journal_list, before it calls 1233098297b2SJeff Mahoney * remove_all_from_journal_list 12341da177e4SLinus Torvalds */ 1235a9dd3643SJeff Mahoney static int _update_journal_header_block(struct super_block *sb, 1236bd4c625cSLinus Torvalds unsigned long offset, 1237600ed416SJeff Mahoney unsigned int trans_id) 1238bd4c625cSLinus Torvalds { 12391da177e4SLinus Torvalds struct reiserfs_journal_header *jh; 1240a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 1241278f6679SJeff Mahoney int depth; 12421da177e4SLinus Torvalds 12431da177e4SLinus Torvalds if (reiserfs_is_journal_aborted(journal)) 12441da177e4SLinus Torvalds return -EIO; 12451da177e4SLinus Torvalds 12461da177e4SLinus Torvalds if (trans_id >= journal->j_last_flush_trans_id) { 12471da177e4SLinus Torvalds if (buffer_locked((journal->j_header_bh))) { 1248278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(sb); 1249278f6679SJeff Mahoney __wait_on_buffer(journal->j_header_bh); 1250278f6679SJeff Mahoney reiserfs_write_lock_nested(sb, depth); 12511da177e4SLinus Torvalds if (unlikely(!buffer_uptodate(journal->j_header_bh))) { 12521da177e4SLinus Torvalds #ifdef CONFIG_REISERFS_CHECK 1253a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-699", 125445b03d5eSJeff Mahoney "buffer write failed"); 12551da177e4SLinus Torvalds #endif 12561da177e4SLinus Torvalds return -EIO; 12571da177e4SLinus Torvalds } 12581da177e4SLinus Torvalds } 12591da177e4SLinus Torvalds journal->j_last_flush_trans_id = trans_id; 12601da177e4SLinus Torvalds journal->j_first_unflushed_offset = offset; 1261bd4c625cSLinus Torvalds jh = (struct reiserfs_journal_header *)(journal->j_header_bh-> 1262bd4c625cSLinus Torvalds b_data); 12631da177e4SLinus Torvalds jh->j_last_flush_trans_id = cpu_to_le32(trans_id); 12641da177e4SLinus Torvalds jh->j_first_unflushed_offset = cpu_to_le32(offset); 12651da177e4SLinus Torvalds jh->j_mount_id = cpu_to_le32(journal->j_mount_id); 12661da177e4SLinus Torvalds 12671da177e4SLinus Torvalds set_buffer_dirty(journal->j_header_bh); 1268278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(sb); 12697cd33ad2SChristoph Hellwig 12707cd33ad2SChristoph Hellwig if (reiserfs_barrier_flush(sb)) 127170fd7614SChristoph Hellwig __sync_dirty_buffer(journal->j_header_bh, 1272d8747d64SJan Kara REQ_SYNC | REQ_PREFLUSH | REQ_FUA); 12737cd33ad2SChristoph Hellwig else 12741da177e4SLinus Torvalds sync_dirty_buffer(journal->j_header_bh); 12757cd33ad2SChristoph Hellwig 1276278f6679SJeff Mahoney reiserfs_write_lock_nested(sb, depth); 12771da177e4SLinus Torvalds if (!buffer_uptodate(journal->j_header_bh)) { 1278a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-837", 127945b03d5eSJeff Mahoney "IO error during journal replay"); 12801da177e4SLinus Torvalds return -EIO; 12811da177e4SLinus Torvalds } 12821da177e4SLinus Torvalds } 12831da177e4SLinus Torvalds return 0; 12841da177e4SLinus Torvalds } 12851da177e4SLinus Torvalds 1286a9dd3643SJeff Mahoney static int update_journal_header_block(struct super_block *sb, 12871da177e4SLinus Torvalds unsigned long offset, 1288600ed416SJeff Mahoney unsigned int trans_id) 1289bd4c625cSLinus Torvalds { 1290a9dd3643SJeff Mahoney return _update_journal_header_block(sb, offset, trans_id); 12911da177e4SLinus Torvalds } 1292bd4c625cSLinus Torvalds 12931da177e4SLinus Torvalds /* 12941da177e4SLinus Torvalds ** flush any and all journal lists older than you are 12951da177e4SLinus Torvalds ** can only be called from flush_journal_list 12961da177e4SLinus Torvalds */ 1297a9dd3643SJeff Mahoney static int flush_older_journal_lists(struct super_block *sb, 12981da177e4SLinus Torvalds struct reiserfs_journal_list *jl) 12991da177e4SLinus Torvalds { 13001da177e4SLinus Torvalds struct list_head *entry; 13011da177e4SLinus Torvalds struct reiserfs_journal_list *other_jl; 1302a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 1303600ed416SJeff Mahoney unsigned int trans_id = jl->j_trans_id; 13041da177e4SLinus Torvalds 1305098297b2SJeff Mahoney /* 1306098297b2SJeff Mahoney * we know we are the only ones flushing things, no extra race 13071da177e4SLinus Torvalds * protection is required. 13081da177e4SLinus Torvalds */ 13091da177e4SLinus Torvalds restart: 13101da177e4SLinus Torvalds entry = journal->j_journal_list.next; 13111da177e4SLinus Torvalds /* Did we wrap? */ 13121da177e4SLinus Torvalds if (entry == &journal->j_journal_list) 13131da177e4SLinus Torvalds return 0; 13141da177e4SLinus Torvalds other_jl = JOURNAL_LIST_ENTRY(entry); 13151da177e4SLinus Torvalds if (other_jl->j_trans_id < trans_id) { 13161da177e4SLinus Torvalds BUG_ON(other_jl->j_refcount <= 0); 13171da177e4SLinus Torvalds /* do not flush all */ 1318a9dd3643SJeff Mahoney flush_journal_list(sb, other_jl, 0); 13191da177e4SLinus Torvalds 13201da177e4SLinus Torvalds /* other_jl is now deleted from the list */ 13211da177e4SLinus Torvalds goto restart; 13221da177e4SLinus Torvalds } 13231da177e4SLinus Torvalds return 0; 13241da177e4SLinus Torvalds } 13251da177e4SLinus Torvalds 13261da177e4SLinus Torvalds static void del_from_work_list(struct super_block *s, 1327bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl) 1328bd4c625cSLinus Torvalds { 13291da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 13301da177e4SLinus Torvalds if (!list_empty(&jl->j_working_list)) { 13311da177e4SLinus Torvalds list_del_init(&jl->j_working_list); 13321da177e4SLinus Torvalds journal->j_num_work_lists--; 13331da177e4SLinus Torvalds } 13341da177e4SLinus Torvalds } 13351da177e4SLinus Torvalds 1336098297b2SJeff Mahoney /* 1337098297b2SJeff Mahoney * flush a journal list, both commit and real blocks 1338098297b2SJeff Mahoney * 1339098297b2SJeff Mahoney * always set flushall to 1, unless you are calling from inside 1340098297b2SJeff Mahoney * flush_journal_list 1341098297b2SJeff Mahoney * 1342098297b2SJeff Mahoney * IMPORTANT. This can only be called while there are no journal writers, 1343098297b2SJeff Mahoney * and the journal is locked. That means it can only be called from 1344098297b2SJeff Mahoney * do_journal_end, or by journal_release 13451da177e4SLinus Torvalds */ 13461da177e4SLinus Torvalds static int flush_journal_list(struct super_block *s, 1347bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl, int flushall) 1348bd4c625cSLinus Torvalds { 13491da177e4SLinus Torvalds struct reiserfs_journal_list *pjl; 13506e9ca45fSzhengbin struct reiserfs_journal_cnode *cn; 13511da177e4SLinus Torvalds int count; 13521da177e4SLinus Torvalds int was_jwait = 0; 13531da177e4SLinus Torvalds int was_dirty = 0; 13541da177e4SLinus Torvalds struct buffer_head *saved_bh; 13551da177e4SLinus Torvalds unsigned long j_len_saved = jl->j_len; 13561da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 13571da177e4SLinus Torvalds int err = 0; 1358278f6679SJeff Mahoney int depth; 13591da177e4SLinus Torvalds 13601da177e4SLinus Torvalds BUG_ON(j_len_saved <= 0); 13611da177e4SLinus Torvalds 13621da177e4SLinus Torvalds if (atomic_read(&journal->j_wcount) != 0) { 136345b03d5eSJeff Mahoney reiserfs_warning(s, "clm-2048", "called with wcount %d", 13641da177e4SLinus Torvalds atomic_read(&journal->j_wcount)); 13651da177e4SLinus Torvalds } 13661da177e4SLinus Torvalds 13671da177e4SLinus Torvalds /* if flushall == 0, the lock is already held */ 13681da177e4SLinus Torvalds if (flushall) { 13698ebc4232SFrederic Weisbecker reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s); 1370afe70259SJeff Mahoney } else if (mutex_trylock(&journal->j_flush_mutex)) { 13711da177e4SLinus Torvalds BUG(); 13721da177e4SLinus Torvalds } 13731da177e4SLinus Torvalds 13741da177e4SLinus Torvalds count = 0; 13751da177e4SLinus Torvalds if (j_len_saved > journal->j_trans_max) { 1376c3a9c210SJeff Mahoney reiserfs_panic(s, "journal-715", "length is %lu, trans id %lu", 1377bd4c625cSLinus Torvalds j_len_saved, jl->j_trans_id); 13781da177e4SLinus Torvalds return 0; 13791da177e4SLinus Torvalds } 13801da177e4SLinus Torvalds 13811da177e4SLinus Torvalds /* if all the work is already done, get out of here */ 1382a228bf8fSJeff Mahoney if (atomic_read(&jl->j_nonzerolen) <= 0 && 1383a228bf8fSJeff Mahoney atomic_read(&jl->j_commit_left) <= 0) { 13841da177e4SLinus Torvalds goto flush_older_and_return; 13851da177e4SLinus Torvalds } 13861da177e4SLinus Torvalds 1387098297b2SJeff Mahoney /* 1388098297b2SJeff Mahoney * start by putting the commit list on disk. This will also flush 1389098297b2SJeff Mahoney * the commit lists of any olders transactions 13901da177e4SLinus Torvalds */ 13911da177e4SLinus Torvalds flush_commit_list(s, jl, 1); 13921da177e4SLinus Torvalds 1393bd4c625cSLinus Torvalds if (!(jl->j_state & LIST_DIRTY) 1394bd4c625cSLinus Torvalds && !reiserfs_is_journal_aborted(journal)) 13951da177e4SLinus Torvalds BUG(); 13961da177e4SLinus Torvalds 13971da177e4SLinus Torvalds /* are we done now? */ 1398a228bf8fSJeff Mahoney if (atomic_read(&jl->j_nonzerolen) <= 0 && 1399a228bf8fSJeff Mahoney atomic_read(&jl->j_commit_left) <= 0) { 14001da177e4SLinus Torvalds goto flush_older_and_return; 14011da177e4SLinus Torvalds } 14021da177e4SLinus Torvalds 1403098297b2SJeff Mahoney /* 1404098297b2SJeff Mahoney * loop through each cnode, see if we need to write it, 1405098297b2SJeff Mahoney * or wait on a more recent transaction, or just ignore it 14061da177e4SLinus Torvalds */ 1407a228bf8fSJeff Mahoney if (atomic_read(&journal->j_wcount) != 0) { 1408c3a9c210SJeff Mahoney reiserfs_panic(s, "journal-844", "journal list is flushing, " 1409c3a9c210SJeff Mahoney "wcount is not 0"); 14101da177e4SLinus Torvalds } 14111da177e4SLinus Torvalds cn = jl->j_realblock; 14121da177e4SLinus Torvalds while (cn) { 14131da177e4SLinus Torvalds was_jwait = 0; 14141da177e4SLinus Torvalds was_dirty = 0; 14151da177e4SLinus Torvalds saved_bh = NULL; 14161da177e4SLinus Torvalds /* blocknr of 0 is no longer in the hash, ignore it */ 14171da177e4SLinus Torvalds if (cn->blocknr == 0) { 14181da177e4SLinus Torvalds goto free_cnode; 14191da177e4SLinus Torvalds } 14201da177e4SLinus Torvalds 1421098297b2SJeff Mahoney /* 1422098297b2SJeff Mahoney * This transaction failed commit. 1423098297b2SJeff Mahoney * Don't write out to the disk 1424098297b2SJeff Mahoney */ 14251da177e4SLinus Torvalds if (!(jl->j_state & LIST_DIRTY)) 14261da177e4SLinus Torvalds goto free_cnode; 14271da177e4SLinus Torvalds 14281da177e4SLinus Torvalds pjl = find_newer_jl_for_cn(cn); 1429098297b2SJeff Mahoney /* 1430098297b2SJeff Mahoney * the order is important here. We check pjl to make sure we 1431098297b2SJeff Mahoney * don't clear BH_JDirty_wait if we aren't the one writing this 1432098297b2SJeff Mahoney * block to disk 14331da177e4SLinus Torvalds */ 14341da177e4SLinus Torvalds if (!pjl && cn->bh) { 14351da177e4SLinus Torvalds saved_bh = cn->bh; 14361da177e4SLinus Torvalds 1437098297b2SJeff Mahoney /* 1438098297b2SJeff Mahoney * we do this to make sure nobody releases the 1439098297b2SJeff Mahoney * buffer while we are working with it 14401da177e4SLinus Torvalds */ 14411da177e4SLinus Torvalds get_bh(saved_bh); 14421da177e4SLinus Torvalds 14431da177e4SLinus Torvalds if (buffer_journal_dirty(saved_bh)) { 14441da177e4SLinus Torvalds BUG_ON(!can_dirty(cn)); 14451da177e4SLinus Torvalds was_jwait = 1; 14461da177e4SLinus Torvalds was_dirty = 1; 14471da177e4SLinus Torvalds } else if (can_dirty(cn)) { 1448098297b2SJeff Mahoney /* 1449098297b2SJeff Mahoney * everything with !pjl && jwait 1450098297b2SJeff Mahoney * should be writable 1451098297b2SJeff Mahoney */ 14521da177e4SLinus Torvalds BUG(); 14531da177e4SLinus Torvalds } 14541da177e4SLinus Torvalds } 14551da177e4SLinus Torvalds 1456098297b2SJeff Mahoney /* 1457098297b2SJeff Mahoney * if someone has this block in a newer transaction, just make 1458098297b2SJeff Mahoney * sure they are committed, and don't try writing it to disk 14591da177e4SLinus Torvalds */ 14601da177e4SLinus Torvalds if (pjl) { 14611da177e4SLinus Torvalds if (atomic_read(&pjl->j_commit_left)) 14621da177e4SLinus Torvalds flush_commit_list(s, pjl, 1); 14631da177e4SLinus Torvalds goto free_cnode; 14641da177e4SLinus Torvalds } 14651da177e4SLinus Torvalds 1466098297b2SJeff Mahoney /* 1467098297b2SJeff Mahoney * bh == NULL when the block got to disk on its own, OR, 1468098297b2SJeff Mahoney * the block got freed in a future transaction 14691da177e4SLinus Torvalds */ 14701da177e4SLinus Torvalds if (saved_bh == NULL) { 14711da177e4SLinus Torvalds goto free_cnode; 14721da177e4SLinus Torvalds } 14731da177e4SLinus Torvalds 1474098297b2SJeff Mahoney /* 1475098297b2SJeff Mahoney * this should never happen. kupdate_one_transaction has 1476098297b2SJeff Mahoney * this list locked while it works, so we should never see a 1477098297b2SJeff Mahoney * buffer here that is not marked JDirty_wait 14781da177e4SLinus Torvalds */ 14791da177e4SLinus Torvalds if ((!was_jwait) && !buffer_locked(saved_bh)) { 148045b03d5eSJeff Mahoney reiserfs_warning(s, "journal-813", 148145b03d5eSJeff Mahoney "BAD! buffer %llu %cdirty %cjwait, " 148265f2b263SColin Ian King "not in a newer transaction", 1483bd4c625cSLinus Torvalds (unsigned long long)saved_bh-> 1484bd4c625cSLinus Torvalds b_blocknr, was_dirty ? ' ' : '!', 1485bd4c625cSLinus Torvalds was_jwait ? ' ' : '!'); 14861da177e4SLinus Torvalds } 14871da177e4SLinus Torvalds if (was_dirty) { 1488098297b2SJeff Mahoney /* 1489098297b2SJeff Mahoney * we inc again because saved_bh gets decremented 1490098297b2SJeff Mahoney * at free_cnode 1491098297b2SJeff Mahoney */ 14921da177e4SLinus Torvalds get_bh(saved_bh); 14931da177e4SLinus Torvalds set_bit(BLOCK_NEEDS_FLUSH, &cn->state); 14941da177e4SLinus Torvalds lock_buffer(saved_bh); 14951da177e4SLinus Torvalds BUG_ON(cn->blocknr != saved_bh->b_blocknr); 14961da177e4SLinus Torvalds if (buffer_dirty(saved_bh)) 14971da177e4SLinus Torvalds submit_logged_buffer(saved_bh); 14981da177e4SLinus Torvalds else 14991da177e4SLinus Torvalds unlock_buffer(saved_bh); 15001da177e4SLinus Torvalds count++; 15011da177e4SLinus Torvalds } else { 150245b03d5eSJeff Mahoney reiserfs_warning(s, "clm-2082", 150345b03d5eSJeff Mahoney "Unable to flush buffer %llu in %s", 1504bd4c625cSLinus Torvalds (unsigned long long)saved_bh-> 1505fbe5498bSHarvey Harrison b_blocknr, __func__); 15061da177e4SLinus Torvalds } 15071da177e4SLinus Torvalds free_cnode: 15081da177e4SLinus Torvalds cn = cn->next; 15091da177e4SLinus Torvalds if (saved_bh) { 1510098297b2SJeff Mahoney /* 1511098297b2SJeff Mahoney * we incremented this to keep others from 1512098297b2SJeff Mahoney * taking the buffer head away 1513098297b2SJeff Mahoney */ 15141da177e4SLinus Torvalds put_bh(saved_bh); 1515a228bf8fSJeff Mahoney if (atomic_read(&saved_bh->b_count) < 0) { 151645b03d5eSJeff Mahoney reiserfs_warning(s, "journal-945", 151745b03d5eSJeff Mahoney "saved_bh->b_count < 0"); 15181da177e4SLinus Torvalds } 15191da177e4SLinus Torvalds } 15201da177e4SLinus Torvalds } 15211da177e4SLinus Torvalds if (count > 0) { 15221da177e4SLinus Torvalds cn = jl->j_realblock; 15231da177e4SLinus Torvalds while (cn) { 15241da177e4SLinus Torvalds if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) { 15251da177e4SLinus Torvalds if (!cn->bh) { 1526c3a9c210SJeff Mahoney reiserfs_panic(s, "journal-1011", 1527c3a9c210SJeff Mahoney "cn->bh is NULL"); 15281da177e4SLinus Torvalds } 15298ebc4232SFrederic Weisbecker 1530278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(s); 1531278f6679SJeff Mahoney __wait_on_buffer(cn->bh); 1532278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 15338ebc4232SFrederic Weisbecker 15341da177e4SLinus Torvalds if (!cn->bh) { 1535c3a9c210SJeff Mahoney reiserfs_panic(s, "journal-1012", 1536c3a9c210SJeff Mahoney "cn->bh is NULL"); 15371da177e4SLinus Torvalds } 15381da177e4SLinus Torvalds if (unlikely(!buffer_uptodate(cn->bh))) { 15391da177e4SLinus Torvalds #ifdef CONFIG_REISERFS_CHECK 154045b03d5eSJeff Mahoney reiserfs_warning(s, "journal-949", 154145b03d5eSJeff Mahoney "buffer write failed"); 15421da177e4SLinus Torvalds #endif 15431da177e4SLinus Torvalds err = -EIO; 15441da177e4SLinus Torvalds } 1545098297b2SJeff Mahoney /* 1546098297b2SJeff Mahoney * note, we must clear the JDirty_wait bit 1547098297b2SJeff Mahoney * after the up to date check, otherwise we 1548098297b2SJeff Mahoney * race against our flushpage routine 15491da177e4SLinus Torvalds */ 1550bd4c625cSLinus Torvalds BUG_ON(!test_clear_buffer_journal_dirty 1551bd4c625cSLinus Torvalds (cn->bh)); 15521da177e4SLinus Torvalds 1553398c95bdSChris Mason /* drop one ref for us */ 15541da177e4SLinus Torvalds put_bh(cn->bh); 1555398c95bdSChris Mason /* drop one ref for journal_mark_dirty */ 1556398c95bdSChris Mason release_buffer_page(cn->bh); 15571da177e4SLinus Torvalds } 15581da177e4SLinus Torvalds cn = cn->next; 15591da177e4SLinus Torvalds } 15601da177e4SLinus Torvalds } 15611da177e4SLinus Torvalds 15621da177e4SLinus Torvalds if (err) 1563bd4c625cSLinus Torvalds reiserfs_abort(s, -EIO, 1564bd4c625cSLinus Torvalds "Write error while pushing transaction to disk in %s", 1565fbe5498bSHarvey Harrison __func__); 15661da177e4SLinus Torvalds flush_older_and_return: 15671da177e4SLinus Torvalds 1568098297b2SJeff Mahoney /* 1569098297b2SJeff Mahoney * before we can update the journal header block, we _must_ flush all 1570098297b2SJeff Mahoney * real blocks from all older transactions to disk. This is because 1571098297b2SJeff Mahoney * once the header block is updated, this transaction will not be 1572098297b2SJeff Mahoney * replayed after a crash 15731da177e4SLinus Torvalds */ 15741da177e4SLinus Torvalds if (flushall) { 15751da177e4SLinus Torvalds flush_older_journal_lists(s, jl); 15761da177e4SLinus Torvalds } 15771da177e4SLinus Torvalds 15781da177e4SLinus Torvalds err = journal->j_errno; 1579098297b2SJeff Mahoney /* 1580098297b2SJeff Mahoney * before we can remove everything from the hash tables for this 1581098297b2SJeff Mahoney * transaction, we must make sure it can never be replayed 1582098297b2SJeff Mahoney * 1583098297b2SJeff Mahoney * since we are only called from do_journal_end, we know for sure there 1584098297b2SJeff Mahoney * are no allocations going on while we are flushing journal lists. So, 1585098297b2SJeff Mahoney * we only need to update the journal header block for the last list 1586098297b2SJeff Mahoney * being flushed 15871da177e4SLinus Torvalds */ 15881da177e4SLinus Torvalds if (!err && flushall) { 1589bd4c625cSLinus Torvalds err = 1590bd4c625cSLinus Torvalds update_journal_header_block(s, 1591bd4c625cSLinus Torvalds (jl->j_start + jl->j_len + 1592bd4c625cSLinus Torvalds 2) % SB_ONDISK_JOURNAL_SIZE(s), 1593bd4c625cSLinus Torvalds jl->j_trans_id); 15941da177e4SLinus Torvalds if (err) 1595bd4c625cSLinus Torvalds reiserfs_abort(s, -EIO, 1596bd4c625cSLinus Torvalds "Write error while updating journal header in %s", 1597fbe5498bSHarvey Harrison __func__); 15981da177e4SLinus Torvalds } 15991da177e4SLinus Torvalds remove_all_from_journal_list(s, jl, 0); 16001da177e4SLinus Torvalds list_del_init(&jl->j_list); 16011da177e4SLinus Torvalds journal->j_num_lists--; 16021da177e4SLinus Torvalds del_from_work_list(s, jl); 16031da177e4SLinus Torvalds 16041da177e4SLinus Torvalds if (journal->j_last_flush_id != 0 && 16051da177e4SLinus Torvalds (jl->j_trans_id - journal->j_last_flush_id) != 1) { 160645b03d5eSJeff Mahoney reiserfs_warning(s, "clm-2201", "last flush %lu, current %lu", 1607bd4c625cSLinus Torvalds journal->j_last_flush_id, jl->j_trans_id); 16081da177e4SLinus Torvalds } 16091da177e4SLinus Torvalds journal->j_last_flush_id = jl->j_trans_id; 16101da177e4SLinus Torvalds 1611098297b2SJeff Mahoney /* 1612098297b2SJeff Mahoney * not strictly required since we are freeing the list, but it should 16131da177e4SLinus Torvalds * help find code using dead lists later on 16141da177e4SLinus Torvalds */ 16151da177e4SLinus Torvalds jl->j_len = 0; 1616a228bf8fSJeff Mahoney atomic_set(&jl->j_nonzerolen, 0); 16171da177e4SLinus Torvalds jl->j_start = 0; 16181da177e4SLinus Torvalds jl->j_realblock = NULL; 16191da177e4SLinus Torvalds jl->j_commit_bh = NULL; 16201da177e4SLinus Torvalds jl->j_trans_id = 0; 16211da177e4SLinus Torvalds jl->j_state = 0; 16221da177e4SLinus Torvalds put_journal_list(s, jl); 16231da177e4SLinus Torvalds if (flushall) 1624afe70259SJeff Mahoney mutex_unlock(&journal->j_flush_mutex); 16251da177e4SLinus Torvalds return err; 16261da177e4SLinus Torvalds } 16271da177e4SLinus Torvalds 16281da177e4SLinus Torvalds static int write_one_transaction(struct super_block *s, 16291da177e4SLinus Torvalds struct reiserfs_journal_list *jl, 16301da177e4SLinus Torvalds struct buffer_chunk *chunk) 16311da177e4SLinus Torvalds { 16321da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 16331da177e4SLinus Torvalds int ret = 0; 16341da177e4SLinus Torvalds 16351da177e4SLinus Torvalds jl->j_state |= LIST_TOUCHED; 16361da177e4SLinus Torvalds del_from_work_list(s, jl); 16371da177e4SLinus Torvalds if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) { 16381da177e4SLinus Torvalds return 0; 16391da177e4SLinus Torvalds } 16401da177e4SLinus Torvalds 16411da177e4SLinus Torvalds cn = jl->j_realblock; 16421da177e4SLinus Torvalds while (cn) { 1643098297b2SJeff Mahoney /* 1644098297b2SJeff Mahoney * if the blocknr == 0, this has been cleared from the hash, 1645098297b2SJeff Mahoney * skip it 16461da177e4SLinus Torvalds */ 16471da177e4SLinus Torvalds if (cn->blocknr == 0) { 16481da177e4SLinus Torvalds goto next; 16491da177e4SLinus Torvalds } 16501da177e4SLinus Torvalds if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) { 16511da177e4SLinus Torvalds struct buffer_head *tmp_bh; 1652098297b2SJeff Mahoney /* 1653098297b2SJeff Mahoney * we can race against journal_mark_freed when we try 16541da177e4SLinus Torvalds * to lock_buffer(cn->bh), so we have to inc the buffer 16551da177e4SLinus Torvalds * count, and recheck things after locking 16561da177e4SLinus Torvalds */ 16571da177e4SLinus Torvalds tmp_bh = cn->bh; 16581da177e4SLinus Torvalds get_bh(tmp_bh); 16591da177e4SLinus Torvalds lock_buffer(tmp_bh); 16601da177e4SLinus Torvalds if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) { 16611da177e4SLinus Torvalds if (!buffer_journal_dirty(tmp_bh) || 16621da177e4SLinus Torvalds buffer_journal_prepared(tmp_bh)) 16631da177e4SLinus Torvalds BUG(); 16641da177e4SLinus Torvalds add_to_chunk(chunk, tmp_bh, NULL, write_chunk); 16651da177e4SLinus Torvalds ret++; 16661da177e4SLinus Torvalds } else { 16671da177e4SLinus Torvalds /* note, cn->bh might be null now */ 16681da177e4SLinus Torvalds unlock_buffer(tmp_bh); 16691da177e4SLinus Torvalds } 16701da177e4SLinus Torvalds put_bh(tmp_bh); 16711da177e4SLinus Torvalds } 16721da177e4SLinus Torvalds next: 16731da177e4SLinus Torvalds cn = cn->next; 16741da177e4SLinus Torvalds cond_resched(); 16751da177e4SLinus Torvalds } 16761da177e4SLinus Torvalds return ret; 16771da177e4SLinus Torvalds } 16781da177e4SLinus Torvalds 16791da177e4SLinus Torvalds /* used by flush_commit_list */ 1680dc0dde61SHariprasad Kelam static void dirty_one_transaction(struct super_block *s, 16811da177e4SLinus Torvalds struct reiserfs_journal_list *jl) 16821da177e4SLinus Torvalds { 16831da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 16841da177e4SLinus Torvalds struct reiserfs_journal_list *pjl; 16851da177e4SLinus Torvalds 16861da177e4SLinus Torvalds jl->j_state |= LIST_DIRTY; 16871da177e4SLinus Torvalds cn = jl->j_realblock; 16881da177e4SLinus Torvalds while (cn) { 1689098297b2SJeff Mahoney /* 1690098297b2SJeff Mahoney * look for a more recent transaction that logged this 1691098297b2SJeff Mahoney * buffer. Only the most recent transaction with a buffer in 1692098297b2SJeff Mahoney * it is allowed to send that buffer to disk 16931da177e4SLinus Torvalds */ 16941da177e4SLinus Torvalds pjl = find_newer_jl_for_cn(cn); 1695bd4c625cSLinus Torvalds if (!pjl && cn->blocknr && cn->bh 1696bd4c625cSLinus Torvalds && buffer_journal_dirty(cn->bh)) { 16971da177e4SLinus Torvalds BUG_ON(!can_dirty(cn)); 1698098297b2SJeff Mahoney /* 1699098297b2SJeff Mahoney * if the buffer is prepared, it will either be logged 17001da177e4SLinus Torvalds * or restored. If restored, we need to make sure 17011da177e4SLinus Torvalds * it actually gets marked dirty 17021da177e4SLinus Torvalds */ 17031da177e4SLinus Torvalds clear_buffer_journal_new(cn->bh); 17041da177e4SLinus Torvalds if (buffer_journal_prepared(cn->bh)) { 17051da177e4SLinus Torvalds set_buffer_journal_restore_dirty(cn->bh); 17061da177e4SLinus Torvalds } else { 17071da177e4SLinus Torvalds set_buffer_journal_test(cn->bh); 17081da177e4SLinus Torvalds mark_buffer_dirty(cn->bh); 17091da177e4SLinus Torvalds } 17101da177e4SLinus Torvalds } 17111da177e4SLinus Torvalds cn = cn->next; 17121da177e4SLinus Torvalds } 17131da177e4SLinus Torvalds } 17141da177e4SLinus Torvalds 17151da177e4SLinus Torvalds static int kupdate_transactions(struct super_block *s, 17161da177e4SLinus Torvalds struct reiserfs_journal_list *jl, 17171da177e4SLinus Torvalds struct reiserfs_journal_list **next_jl, 1718600ed416SJeff Mahoney unsigned int *next_trans_id, 1719bd4c625cSLinus Torvalds int num_blocks, int num_trans) 1720bd4c625cSLinus Torvalds { 17211da177e4SLinus Torvalds int ret = 0; 17221da177e4SLinus Torvalds int written = 0; 17231da177e4SLinus Torvalds int transactions_flushed = 0; 1724600ed416SJeff Mahoney unsigned int orig_trans_id = jl->j_trans_id; 17251da177e4SLinus Torvalds struct buffer_chunk chunk; 17261da177e4SLinus Torvalds struct list_head *entry; 17271da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 17281da177e4SLinus Torvalds chunk.nr = 0; 17291da177e4SLinus Torvalds 1730a412f9efSFrederic Weisbecker reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s); 17311da177e4SLinus Torvalds if (!journal_list_still_alive(s, orig_trans_id)) { 17321da177e4SLinus Torvalds goto done; 17331da177e4SLinus Torvalds } 17341da177e4SLinus Torvalds 1735098297b2SJeff Mahoney /* 1736098297b2SJeff Mahoney * we've got j_flush_mutex held, nobody is going to delete any 17371da177e4SLinus Torvalds * of these lists out from underneath us 17381da177e4SLinus Torvalds */ 17391da177e4SLinus Torvalds while ((num_trans && transactions_flushed < num_trans) || 17401da177e4SLinus Torvalds (!num_trans && written < num_blocks)) { 17411da177e4SLinus Torvalds 17421da177e4SLinus Torvalds if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) || 1743bd4c625cSLinus Torvalds atomic_read(&jl->j_commit_left) 1744bd4c625cSLinus Torvalds || !(jl->j_state & LIST_DIRTY)) { 17451da177e4SLinus Torvalds del_from_work_list(s, jl); 17461da177e4SLinus Torvalds break; 17471da177e4SLinus Torvalds } 17481da177e4SLinus Torvalds ret = write_one_transaction(s, jl, &chunk); 17491da177e4SLinus Torvalds 17501da177e4SLinus Torvalds if (ret < 0) 17511da177e4SLinus Torvalds goto done; 17521da177e4SLinus Torvalds transactions_flushed++; 17531da177e4SLinus Torvalds written += ret; 17541da177e4SLinus Torvalds entry = jl->j_list.next; 17551da177e4SLinus Torvalds 17561da177e4SLinus Torvalds /* did we wrap? */ 17571da177e4SLinus Torvalds if (entry == &journal->j_journal_list) { 17581da177e4SLinus Torvalds break; 17591da177e4SLinus Torvalds } 17601da177e4SLinus Torvalds jl = JOURNAL_LIST_ENTRY(entry); 17611da177e4SLinus Torvalds 17621da177e4SLinus Torvalds /* don't bother with older transactions */ 17631da177e4SLinus Torvalds if (jl->j_trans_id <= orig_trans_id) 17641da177e4SLinus Torvalds break; 17651da177e4SLinus Torvalds } 17661da177e4SLinus Torvalds if (chunk.nr) { 17671da177e4SLinus Torvalds write_chunk(&chunk); 17681da177e4SLinus Torvalds } 17691da177e4SLinus Torvalds 17701da177e4SLinus Torvalds done: 1771afe70259SJeff Mahoney mutex_unlock(&journal->j_flush_mutex); 17721da177e4SLinus Torvalds return ret; 17731da177e4SLinus Torvalds } 17741da177e4SLinus Torvalds 1775098297b2SJeff Mahoney /* 1776098297b2SJeff Mahoney * for o_sync and fsync heavy applications, they tend to use 1777098297b2SJeff Mahoney * all the journa list slots with tiny transactions. These 1778098297b2SJeff Mahoney * trigger lots and lots of calls to update the header block, which 1779098297b2SJeff Mahoney * adds seeks and slows things down. 1780098297b2SJeff Mahoney * 1781098297b2SJeff Mahoney * This function tries to clear out a large chunk of the journal lists 1782098297b2SJeff Mahoney * at once, which makes everything faster since only the newest journal 1783098297b2SJeff Mahoney * list updates the header block 17841da177e4SLinus Torvalds */ 17851da177e4SLinus Torvalds static int flush_used_journal_lists(struct super_block *s, 1786bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl) 1787bd4c625cSLinus Torvalds { 17881da177e4SLinus Torvalds unsigned long len = 0; 17891da177e4SLinus Torvalds unsigned long cur_len; 17901da177e4SLinus Torvalds int i; 17911da177e4SLinus Torvalds int limit = 256; 17921da177e4SLinus Torvalds struct reiserfs_journal_list *tjl; 17931da177e4SLinus Torvalds struct reiserfs_journal_list *flush_jl; 1794600ed416SJeff Mahoney unsigned int trans_id; 17951da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 17961da177e4SLinus Torvalds 17971da177e4SLinus Torvalds flush_jl = tjl = jl; 17981da177e4SLinus Torvalds 17991da177e4SLinus Torvalds /* in data logging mode, try harder to flush a lot of blocks */ 18001da177e4SLinus Torvalds if (reiserfs_data_log(s)) 18011da177e4SLinus Torvalds limit = 1024; 18021da177e4SLinus Torvalds /* flush for 256 transactions or limit blocks, whichever comes first */ 18031da177e4SLinus Torvalds for (i = 0; i < 256 && len < limit; i++) { 18041da177e4SLinus Torvalds if (atomic_read(&tjl->j_commit_left) || 18051da177e4SLinus Torvalds tjl->j_trans_id < jl->j_trans_id) { 18061da177e4SLinus Torvalds break; 18071da177e4SLinus Torvalds } 18081da177e4SLinus Torvalds cur_len = atomic_read(&tjl->j_nonzerolen); 18091da177e4SLinus Torvalds if (cur_len > 0) { 18101da177e4SLinus Torvalds tjl->j_state &= ~LIST_TOUCHED; 18111da177e4SLinus Torvalds } 18121da177e4SLinus Torvalds len += cur_len; 18131da177e4SLinus Torvalds flush_jl = tjl; 18141da177e4SLinus Torvalds if (tjl->j_list.next == &journal->j_journal_list) 18151da177e4SLinus Torvalds break; 18161da177e4SLinus Torvalds tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next); 18171da177e4SLinus Torvalds } 1818721a769cSJeff Mahoney get_journal_list(jl); 1819721a769cSJeff Mahoney get_journal_list(flush_jl); 1820098297b2SJeff Mahoney 1821098297b2SJeff Mahoney /* 1822098297b2SJeff Mahoney * try to find a group of blocks we can flush across all the 1823098297b2SJeff Mahoney * transactions, but only bother if we've actually spanned 1824098297b2SJeff Mahoney * across multiple lists 18251da177e4SLinus Torvalds */ 18263e9fd5a4SJason Yan if (flush_jl != jl) 18273e9fd5a4SJason Yan kupdate_transactions(s, jl, &tjl, &trans_id, len, i); 18283e9fd5a4SJason Yan 18291da177e4SLinus Torvalds flush_journal_list(s, flush_jl, 1); 1830721a769cSJeff Mahoney put_journal_list(s, flush_jl); 1831721a769cSJeff Mahoney put_journal_list(s, jl); 18321da177e4SLinus Torvalds return 0; 18331da177e4SLinus Torvalds } 18341da177e4SLinus Torvalds 18351da177e4SLinus Torvalds /* 1836098297b2SJeff Mahoney * removes any nodes in table with name block and dev as bh. 1837098297b2SJeff Mahoney * only touchs the hnext and hprev pointers. 18381da177e4SLinus Torvalds */ 18395d552ad7SBharath Vedartham static void remove_journal_hash(struct super_block *sb, 18401da177e4SLinus Torvalds struct reiserfs_journal_cnode **table, 18411da177e4SLinus Torvalds struct reiserfs_journal_list *jl, 18421da177e4SLinus Torvalds unsigned long block, int remove_freed) 18431da177e4SLinus Torvalds { 18441da177e4SLinus Torvalds struct reiserfs_journal_cnode *cur; 18451da177e4SLinus Torvalds struct reiserfs_journal_cnode **head; 18461da177e4SLinus Torvalds 18471da177e4SLinus Torvalds head = &(journal_hash(table, sb, block)); 18481da177e4SLinus Torvalds if (!head) { 18491da177e4SLinus Torvalds return; 18501da177e4SLinus Torvalds } 18511da177e4SLinus Torvalds cur = *head; 18521da177e4SLinus Torvalds while (cur) { 1853bd4c625cSLinus Torvalds if (cur->blocknr == block && cur->sb == sb 1854bd4c625cSLinus Torvalds && (jl == NULL || jl == cur->jlist) 1855bd4c625cSLinus Torvalds && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) { 18561da177e4SLinus Torvalds if (cur->hnext) { 18571da177e4SLinus Torvalds cur->hnext->hprev = cur->hprev; 18581da177e4SLinus Torvalds } 18591da177e4SLinus Torvalds if (cur->hprev) { 18601da177e4SLinus Torvalds cur->hprev->hnext = cur->hnext; 18611da177e4SLinus Torvalds } else { 18621da177e4SLinus Torvalds *head = cur->hnext; 18631da177e4SLinus Torvalds } 18641da177e4SLinus Torvalds cur->blocknr = 0; 18651da177e4SLinus Torvalds cur->sb = NULL; 18661da177e4SLinus Torvalds cur->state = 0; 1867098297b2SJeff Mahoney /* 1868098297b2SJeff Mahoney * anybody who clears the cur->bh will also 1869098297b2SJeff Mahoney * dec the nonzerolen 1870098297b2SJeff Mahoney */ 1871098297b2SJeff Mahoney if (cur->bh && cur->jlist) 1872a228bf8fSJeff Mahoney atomic_dec(&cur->jlist->j_nonzerolen); 18731da177e4SLinus Torvalds cur->bh = NULL; 18741da177e4SLinus Torvalds cur->jlist = NULL; 18751da177e4SLinus Torvalds } 18761da177e4SLinus Torvalds cur = cur->hnext; 18771da177e4SLinus Torvalds } 18781da177e4SLinus Torvalds } 18791da177e4SLinus Torvalds 1880a9dd3643SJeff Mahoney static void free_journal_ram(struct super_block *sb) 1881bd4c625cSLinus Torvalds { 1882a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 1883d739b42bSPekka Enberg kfree(journal->j_current_jl); 18841da177e4SLinus Torvalds journal->j_num_lists--; 18851da177e4SLinus Torvalds 18861da177e4SLinus Torvalds vfree(journal->j_cnode_free_orig); 1887a9dd3643SJeff Mahoney free_list_bitmaps(sb, journal->j_list_bitmap); 1888a9dd3643SJeff Mahoney free_bitmap_nodes(sb); /* must be after free_list_bitmaps */ 18891da177e4SLinus Torvalds if (journal->j_header_bh) { 18901da177e4SLinus Torvalds brelse(journal->j_header_bh); 18911da177e4SLinus Torvalds } 1892098297b2SJeff Mahoney /* 1893098297b2SJeff Mahoney * j_header_bh is on the journal dev, make sure 1894098297b2SJeff Mahoney * not to release the journal dev until we brelse j_header_bh 18951da177e4SLinus Torvalds */ 1896a9dd3643SJeff Mahoney release_journal_dev(sb, journal); 18971da177e4SLinus Torvalds vfree(journal); 18981da177e4SLinus Torvalds } 18991da177e4SLinus Torvalds 19001da177e4SLinus Torvalds /* 1901098297b2SJeff Mahoney * call on unmount. Only set error to 1 if you haven't made your way out 1902098297b2SJeff Mahoney * of read_super() yet. Any other caller must keep error at 0. 19031da177e4SLinus Torvalds */ 1904bd4c625cSLinus Torvalds static int do_journal_release(struct reiserfs_transaction_handle *th, 1905a9dd3643SJeff Mahoney struct super_block *sb, int error) 1906bd4c625cSLinus Torvalds { 19071da177e4SLinus Torvalds struct reiserfs_transaction_handle myth; 1908a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 19091da177e4SLinus Torvalds 1910098297b2SJeff Mahoney /* 1911098297b2SJeff Mahoney * we only want to flush out transactions if we were 1912098297b2SJeff Mahoney * called with error == 0 19131da177e4SLinus Torvalds */ 1914bc98a42cSDavid Howells if (!error && !sb_rdonly(sb)) { 19151da177e4SLinus Torvalds /* end the current trans */ 19161da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 191758d85426SJeff Mahoney do_journal_end(th, FLUSH_ALL); 19181da177e4SLinus Torvalds 1919098297b2SJeff Mahoney /* 1920098297b2SJeff Mahoney * make sure something gets logged to force 1921098297b2SJeff Mahoney * our way into the flush code 1922098297b2SJeff Mahoney */ 1923b491dd17SJeff Mahoney if (!journal_join(&myth, sb)) { 1924a9dd3643SJeff Mahoney reiserfs_prepare_for_journal(sb, 1925a9dd3643SJeff Mahoney SB_BUFFER_WITH_SB(sb), 1926bd4c625cSLinus Torvalds 1); 192709f1b80bSJeff Mahoney journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb)); 192858d85426SJeff Mahoney do_journal_end(&myth, FLUSH_ALL); 19291da177e4SLinus Torvalds } 19301da177e4SLinus Torvalds } 19311da177e4SLinus Torvalds 19321da177e4SLinus Torvalds /* this also catches errors during the do_journal_end above */ 19331da177e4SLinus Torvalds if (!error && reiserfs_is_journal_aborted(journal)) { 19341da177e4SLinus Torvalds memset(&myth, 0, sizeof(myth)); 1935b491dd17SJeff Mahoney if (!journal_join_abort(&myth, sb)) { 1936a9dd3643SJeff Mahoney reiserfs_prepare_for_journal(sb, 1937a9dd3643SJeff Mahoney SB_BUFFER_WITH_SB(sb), 1938bd4c625cSLinus Torvalds 1); 193909f1b80bSJeff Mahoney journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb)); 194058d85426SJeff Mahoney do_journal_end(&myth, FLUSH_ALL); 19411da177e4SLinus Torvalds } 19421da177e4SLinus Torvalds } 19431da177e4SLinus Torvalds 19448ebc4232SFrederic Weisbecker 19458ebc4232SFrederic Weisbecker /* 19468ebc4232SFrederic Weisbecker * We must release the write lock here because 19478ebc4232SFrederic Weisbecker * the workqueue job (flush_async_commit) needs this lock 19488ebc4232SFrederic Weisbecker */ 19498ebc4232SFrederic Weisbecker reiserfs_write_unlock(sb); 1950033369d1SArtem Bityutskiy 195101777836SJan Kara /* 195201777836SJan Kara * Cancel flushing of old commits. Note that neither of these works 195301777836SJan Kara * will be requeued because superblock is being shutdown and doesn't 19541751e8a6SLinus Torvalds * have SB_ACTIVE set. 195501777836SJan Kara */ 195671b0576bSJan Kara reiserfs_cancel_old_flush(sb); 195701777836SJan Kara /* wait for all commits to finish */ 195801777836SJan Kara cancel_delayed_work_sync(&SB_JOURNAL(sb)->j_work); 19591da177e4SLinus Torvalds 1960a9dd3643SJeff Mahoney free_journal_ram(sb); 19611da177e4SLinus Torvalds 19620523676dSFrederic Weisbecker reiserfs_write_lock(sb); 19630523676dSFrederic Weisbecker 19641da177e4SLinus Torvalds return 0; 19651da177e4SLinus Torvalds } 19661da177e4SLinus Torvalds 1967098297b2SJeff Mahoney /* * call on unmount. flush all journal trans, release all alloc'd ram */ 1968bd4c625cSLinus Torvalds int journal_release(struct reiserfs_transaction_handle *th, 1969a9dd3643SJeff Mahoney struct super_block *sb) 1970bd4c625cSLinus Torvalds { 1971a9dd3643SJeff Mahoney return do_journal_release(th, sb, 0); 19721da177e4SLinus Torvalds } 1973bd4c625cSLinus Torvalds 1974098297b2SJeff Mahoney /* only call from an error condition inside reiserfs_read_super! */ 1975bd4c625cSLinus Torvalds int journal_release_error(struct reiserfs_transaction_handle *th, 1976a9dd3643SJeff Mahoney struct super_block *sb) 1977bd4c625cSLinus Torvalds { 1978a9dd3643SJeff Mahoney return do_journal_release(th, sb, 1); 19791da177e4SLinus Torvalds } 19801da177e4SLinus Torvalds 1981098297b2SJeff Mahoney /* 1982098297b2SJeff Mahoney * compares description block with commit block. 1983098297b2SJeff Mahoney * returns 1 if they differ, 0 if they are the same 1984098297b2SJeff Mahoney */ 1985a9dd3643SJeff Mahoney static int journal_compare_desc_commit(struct super_block *sb, 1986bd4c625cSLinus Torvalds struct reiserfs_journal_desc *desc, 1987bd4c625cSLinus Torvalds struct reiserfs_journal_commit *commit) 1988bd4c625cSLinus Torvalds { 19891da177e4SLinus Torvalds if (get_commit_trans_id(commit) != get_desc_trans_id(desc) || 19901da177e4SLinus Torvalds get_commit_trans_len(commit) != get_desc_trans_len(desc) || 1991a9dd3643SJeff Mahoney get_commit_trans_len(commit) > SB_JOURNAL(sb)->j_trans_max || 1992bd4c625cSLinus Torvalds get_commit_trans_len(commit) <= 0) { 19931da177e4SLinus Torvalds return 1; 19941da177e4SLinus Torvalds } 19951da177e4SLinus Torvalds return 0; 19961da177e4SLinus Torvalds } 1997bd4c625cSLinus Torvalds 1998098297b2SJeff Mahoney /* 1999098297b2SJeff Mahoney * returns 0 if it did not find a description block 2000098297b2SJeff Mahoney * returns -1 if it found a corrupt commit block 2001098297b2SJeff Mahoney * returns 1 if both desc and commit were valid 2002098297b2SJeff Mahoney * NOTE: only called during fs mount 20031da177e4SLinus Torvalds */ 2004a9dd3643SJeff Mahoney static int journal_transaction_is_valid(struct super_block *sb, 2005bd4c625cSLinus Torvalds struct buffer_head *d_bh, 2006600ed416SJeff Mahoney unsigned int *oldest_invalid_trans_id, 2007bd4c625cSLinus Torvalds unsigned long *newest_mount_id) 2008bd4c625cSLinus Torvalds { 20091da177e4SLinus Torvalds struct reiserfs_journal_desc *desc; 20101da177e4SLinus Torvalds struct reiserfs_journal_commit *commit; 20111da177e4SLinus Torvalds struct buffer_head *c_bh; 20121da177e4SLinus Torvalds unsigned long offset; 20131da177e4SLinus Torvalds 20141da177e4SLinus Torvalds if (!d_bh) 20151da177e4SLinus Torvalds return 0; 20161da177e4SLinus Torvalds 20171da177e4SLinus Torvalds desc = (struct reiserfs_journal_desc *)d_bh->b_data; 2018bd4c625cSLinus Torvalds if (get_desc_trans_len(desc) > 0 2019bd4c625cSLinus Torvalds && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) { 2020bd4c625cSLinus Torvalds if (oldest_invalid_trans_id && *oldest_invalid_trans_id 2021bd4c625cSLinus Torvalds && get_desc_trans_id(desc) > *oldest_invalid_trans_id) { 2022a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2023bd4c625cSLinus Torvalds "journal-986: transaction " 20241da177e4SLinus Torvalds "is valid returning because trans_id %d is greater than " 2025bd4c625cSLinus Torvalds "oldest_invalid %lu", 2026bd4c625cSLinus Torvalds get_desc_trans_id(desc), 20271da177e4SLinus Torvalds *oldest_invalid_trans_id); 20281da177e4SLinus Torvalds return 0; 20291da177e4SLinus Torvalds } 2030bd4c625cSLinus Torvalds if (newest_mount_id 2031bd4c625cSLinus Torvalds && *newest_mount_id > get_desc_mount_id(desc)) { 2032a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2033bd4c625cSLinus Torvalds "journal-1087: transaction " 20341da177e4SLinus Torvalds "is valid returning because mount_id %d is less than " 2035bd4c625cSLinus Torvalds "newest_mount_id %lu", 2036bd4c625cSLinus Torvalds get_desc_mount_id(desc), 20371da177e4SLinus Torvalds *newest_mount_id); 20381da177e4SLinus Torvalds return -1; 20391da177e4SLinus Torvalds } 2040a9dd3643SJeff Mahoney if (get_desc_trans_len(desc) > SB_JOURNAL(sb)->j_trans_max) { 2041a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-2018", 204245b03d5eSJeff Mahoney "Bad transaction length %d " 204345b03d5eSJeff Mahoney "encountered, ignoring transaction", 2044bd4c625cSLinus Torvalds get_desc_trans_len(desc)); 20451da177e4SLinus Torvalds return -1; 20461da177e4SLinus Torvalds } 2047a9dd3643SJeff Mahoney offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb); 20481da177e4SLinus Torvalds 2049098297b2SJeff Mahoney /* 2050098297b2SJeff Mahoney * ok, we have a journal description block, 2051098297b2SJeff Mahoney * let's see if the transaction was valid 2052098297b2SJeff Mahoney */ 2053bd4c625cSLinus Torvalds c_bh = 2054a9dd3643SJeff Mahoney journal_bread(sb, 2055a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2056bd4c625cSLinus Torvalds ((offset + get_desc_trans_len(desc) + 2057a9dd3643SJeff Mahoney 1) % SB_ONDISK_JOURNAL_SIZE(sb))); 20581da177e4SLinus Torvalds if (!c_bh) 20591da177e4SLinus Torvalds return 0; 20601da177e4SLinus Torvalds commit = (struct reiserfs_journal_commit *)c_bh->b_data; 2061a9dd3643SJeff Mahoney if (journal_compare_desc_commit(sb, desc, commit)) { 2062a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 20631da177e4SLinus Torvalds "journal_transaction_is_valid, commit offset %ld had bad " 20641da177e4SLinus Torvalds "time %d or length %d", 2065bd4c625cSLinus Torvalds c_bh->b_blocknr - 2066a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb), 20671da177e4SLinus Torvalds get_commit_trans_id(commit), 20681da177e4SLinus Torvalds get_commit_trans_len(commit)); 20691da177e4SLinus Torvalds brelse(c_bh); 20701da177e4SLinus Torvalds if (oldest_invalid_trans_id) { 2071bd4c625cSLinus Torvalds *oldest_invalid_trans_id = 2072bd4c625cSLinus Torvalds get_desc_trans_id(desc); 2073a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2074bd4c625cSLinus Torvalds "journal-1004: " 20751da177e4SLinus Torvalds "transaction_is_valid setting oldest invalid trans_id " 2076bd4c625cSLinus Torvalds "to %d", 2077bd4c625cSLinus Torvalds get_desc_trans_id(desc)); 20781da177e4SLinus Torvalds } 20791da177e4SLinus Torvalds return -1; 20801da177e4SLinus Torvalds } 20811da177e4SLinus Torvalds brelse(c_bh); 2082a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2083bd4c625cSLinus Torvalds "journal-1006: found valid " 20841da177e4SLinus Torvalds "transaction start offset %llu, len %d id %d", 2085bd4c625cSLinus Torvalds d_bh->b_blocknr - 2086a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb), 2087bd4c625cSLinus Torvalds get_desc_trans_len(desc), 2088bd4c625cSLinus Torvalds get_desc_trans_id(desc)); 20891da177e4SLinus Torvalds return 1; 20901da177e4SLinus Torvalds } else { 20911da177e4SLinus Torvalds return 0; 20921da177e4SLinus Torvalds } 20931da177e4SLinus Torvalds } 20941da177e4SLinus Torvalds 2095bd4c625cSLinus Torvalds static void brelse_array(struct buffer_head **heads, int num) 2096bd4c625cSLinus Torvalds { 20971da177e4SLinus Torvalds int i; 20981da177e4SLinus Torvalds for (i = 0; i < num; i++) { 20991da177e4SLinus Torvalds brelse(heads[i]); 21001da177e4SLinus Torvalds } 21011da177e4SLinus Torvalds } 21021da177e4SLinus Torvalds 21031da177e4SLinus Torvalds /* 2104098297b2SJeff Mahoney * given the start, and values for the oldest acceptable transactions, 2105098297b2SJeff Mahoney * this either reads in a replays a transaction, or returns because the 2106098297b2SJeff Mahoney * transaction is invalid, or too old. 2107098297b2SJeff Mahoney * NOTE: only called during fs mount 21081da177e4SLinus Torvalds */ 2109a9dd3643SJeff Mahoney static int journal_read_transaction(struct super_block *sb, 2110bd4c625cSLinus Torvalds unsigned long cur_dblock, 2111bd4c625cSLinus Torvalds unsigned long oldest_start, 2112600ed416SJeff Mahoney unsigned int oldest_trans_id, 2113bd4c625cSLinus Torvalds unsigned long newest_mount_id) 2114bd4c625cSLinus Torvalds { 2115a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 21161da177e4SLinus Torvalds struct reiserfs_journal_desc *desc; 21171da177e4SLinus Torvalds struct reiserfs_journal_commit *commit; 2118600ed416SJeff Mahoney unsigned int trans_id = 0; 21191da177e4SLinus Torvalds struct buffer_head *c_bh; 21201da177e4SLinus Torvalds struct buffer_head *d_bh; 21211da177e4SLinus Torvalds struct buffer_head **log_blocks = NULL; 21221da177e4SLinus Torvalds struct buffer_head **real_blocks = NULL; 2123600ed416SJeff Mahoney unsigned int trans_offset; 21241da177e4SLinus Torvalds int i; 21251da177e4SLinus Torvalds int trans_half; 21261da177e4SLinus Torvalds 2127a9dd3643SJeff Mahoney d_bh = journal_bread(sb, cur_dblock); 21281da177e4SLinus Torvalds if (!d_bh) 21291da177e4SLinus Torvalds return 1; 21301da177e4SLinus Torvalds desc = (struct reiserfs_journal_desc *)d_bh->b_data; 2131a9dd3643SJeff Mahoney trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb); 2132a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1037: " 21331da177e4SLinus Torvalds "journal_read_transaction, offset %llu, len %d mount_id %d", 2134a9dd3643SJeff Mahoney d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb), 21351da177e4SLinus Torvalds get_desc_trans_len(desc), get_desc_mount_id(desc)); 21361da177e4SLinus Torvalds if (get_desc_trans_id(desc) < oldest_trans_id) { 2137a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1039: " 21381da177e4SLinus Torvalds "journal_read_trans skipping because %lu is too old", 2139bd4c625cSLinus Torvalds cur_dblock - 2140a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb)); 21411da177e4SLinus Torvalds brelse(d_bh); 21421da177e4SLinus Torvalds return 1; 21431da177e4SLinus Torvalds } 21441da177e4SLinus Torvalds if (get_desc_mount_id(desc) != newest_mount_id) { 2145a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1146: " 21461da177e4SLinus Torvalds "journal_read_trans skipping because %d is != " 21471da177e4SLinus Torvalds "newest_mount_id %lu", get_desc_mount_id(desc), 21481da177e4SLinus Torvalds newest_mount_id); 21491da177e4SLinus Torvalds brelse(d_bh); 21501da177e4SLinus Torvalds return 1; 21511da177e4SLinus Torvalds } 2152a9dd3643SJeff Mahoney c_bh = journal_bread(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 21531da177e4SLinus Torvalds ((trans_offset + get_desc_trans_len(desc) + 1) % 2154a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb))); 21551da177e4SLinus Torvalds if (!c_bh) { 21561da177e4SLinus Torvalds brelse(d_bh); 21571da177e4SLinus Torvalds return 1; 21581da177e4SLinus Torvalds } 21591da177e4SLinus Torvalds commit = (struct reiserfs_journal_commit *)c_bh->b_data; 2160a9dd3643SJeff Mahoney if (journal_compare_desc_commit(sb, desc, commit)) { 2161a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2162bd4c625cSLinus Torvalds "journal_read_transaction, " 21631da177e4SLinus Torvalds "commit offset %llu had bad time %d or length %d", 2164bd4c625cSLinus Torvalds c_bh->b_blocknr - 2165a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb), 2166bd4c625cSLinus Torvalds get_commit_trans_id(commit), 2167bd4c625cSLinus Torvalds get_commit_trans_len(commit)); 21681da177e4SLinus Torvalds brelse(c_bh); 21691da177e4SLinus Torvalds brelse(d_bh); 21701da177e4SLinus Torvalds return 1; 21711da177e4SLinus Torvalds } 21723f8b5ee3SJeff Mahoney 21733f8b5ee3SJeff Mahoney if (bdev_read_only(sb->s_bdev)) { 21743f8b5ee3SJeff Mahoney reiserfs_warning(sb, "clm-2076", 21753f8b5ee3SJeff Mahoney "device is readonly, unable to replay log"); 21763f8b5ee3SJeff Mahoney brelse(c_bh); 21773f8b5ee3SJeff Mahoney brelse(d_bh); 21783f8b5ee3SJeff Mahoney return -EROFS; 21793f8b5ee3SJeff Mahoney } 21803f8b5ee3SJeff Mahoney 21811da177e4SLinus Torvalds trans_id = get_desc_trans_id(desc); 2182098297b2SJeff Mahoney /* 2183098297b2SJeff Mahoney * now we know we've got a good transaction, and it was 2184098297b2SJeff Mahoney * inside the valid time ranges 2185098297b2SJeff Mahoney */ 21866da2ec56SKees Cook log_blocks = kmalloc_array(get_desc_trans_len(desc), 21876da2ec56SKees Cook sizeof(struct buffer_head *), 21886da2ec56SKees Cook GFP_NOFS); 21896da2ec56SKees Cook real_blocks = kmalloc_array(get_desc_trans_len(desc), 21906da2ec56SKees Cook sizeof(struct buffer_head *), 21916da2ec56SKees Cook GFP_NOFS); 21921da177e4SLinus Torvalds if (!log_blocks || !real_blocks) { 21931da177e4SLinus Torvalds brelse(c_bh); 21941da177e4SLinus Torvalds brelse(d_bh); 2195d739b42bSPekka Enberg kfree(log_blocks); 2196d739b42bSPekka Enberg kfree(real_blocks); 2197a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1169", 219845b03d5eSJeff Mahoney "kmalloc failed, unable to mount FS"); 21991da177e4SLinus Torvalds return -1; 22001da177e4SLinus Torvalds } 22011da177e4SLinus Torvalds /* get all the buffer heads */ 2202a9dd3643SJeff Mahoney trans_half = journal_trans_half(sb->s_blocksize); 22031da177e4SLinus Torvalds for (i = 0; i < get_desc_trans_len(desc); i++) { 2204bd4c625cSLinus Torvalds log_blocks[i] = 2205a9dd3643SJeff Mahoney journal_getblk(sb, 2206a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2207bd4c625cSLinus Torvalds (trans_offset + 1 + 2208a9dd3643SJeff Mahoney i) % SB_ONDISK_JOURNAL_SIZE(sb)); 22091da177e4SLinus Torvalds if (i < trans_half) { 2210bd4c625cSLinus Torvalds real_blocks[i] = 2211a9dd3643SJeff Mahoney sb_getblk(sb, 2212bd4c625cSLinus Torvalds le32_to_cpu(desc->j_realblock[i])); 22131da177e4SLinus Torvalds } else { 2214bd4c625cSLinus Torvalds real_blocks[i] = 2215a9dd3643SJeff Mahoney sb_getblk(sb, 2216bd4c625cSLinus Torvalds le32_to_cpu(commit-> 2217bd4c625cSLinus Torvalds j_realblock[i - trans_half])); 22181da177e4SLinus Torvalds } 2219a9dd3643SJeff Mahoney if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(sb)) { 2220a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1207", 222145b03d5eSJeff Mahoney "REPLAY FAILURE fsck required! " 222245b03d5eSJeff Mahoney "Block to replay is outside of " 222345b03d5eSJeff Mahoney "filesystem"); 22241da177e4SLinus Torvalds goto abort_replay; 22251da177e4SLinus Torvalds } 22261da177e4SLinus Torvalds /* make sure we don't try to replay onto log or reserved area */ 2227bd4c625cSLinus Torvalds if (is_block_in_log_or_reserved_area 2228a9dd3643SJeff Mahoney (sb, real_blocks[i]->b_blocknr)) { 2229a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1204", 223045b03d5eSJeff Mahoney "REPLAY FAILURE fsck required! " 223145b03d5eSJeff Mahoney "Trying to replay onto a log block"); 22321da177e4SLinus Torvalds abort_replay: 22331da177e4SLinus Torvalds brelse_array(log_blocks, i); 22341da177e4SLinus Torvalds brelse_array(real_blocks, i); 22351da177e4SLinus Torvalds brelse(c_bh); 22361da177e4SLinus Torvalds brelse(d_bh); 2237d739b42bSPekka Enberg kfree(log_blocks); 2238d739b42bSPekka Enberg kfree(real_blocks); 22391da177e4SLinus Torvalds return -1; 22401da177e4SLinus Torvalds } 22411da177e4SLinus Torvalds } 22421da177e4SLinus Torvalds /* read in the log blocks, memcpy to the corresponding real block */ 2243d554822eSZhang Yi bh_read_batch(get_desc_trans_len(desc), log_blocks); 22441da177e4SLinus Torvalds for (i = 0; i < get_desc_trans_len(desc); i++) { 22458ebc4232SFrederic Weisbecker 22461da177e4SLinus Torvalds wait_on_buffer(log_blocks[i]); 22471da177e4SLinus Torvalds if (!buffer_uptodate(log_blocks[i])) { 2248a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1212", 224945b03d5eSJeff Mahoney "REPLAY FAILURE fsck required! " 225045b03d5eSJeff Mahoney "buffer write failed"); 2251bd4c625cSLinus Torvalds brelse_array(log_blocks + i, 2252bd4c625cSLinus Torvalds get_desc_trans_len(desc) - i); 22531da177e4SLinus Torvalds brelse_array(real_blocks, get_desc_trans_len(desc)); 22541da177e4SLinus Torvalds brelse(c_bh); 22551da177e4SLinus Torvalds brelse(d_bh); 2256d739b42bSPekka Enberg kfree(log_blocks); 2257d739b42bSPekka Enberg kfree(real_blocks); 22581da177e4SLinus Torvalds return -1; 22591da177e4SLinus Torvalds } 2260bd4c625cSLinus Torvalds memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data, 2261bd4c625cSLinus Torvalds real_blocks[i]->b_size); 22621da177e4SLinus Torvalds set_buffer_uptodate(real_blocks[i]); 22631da177e4SLinus Torvalds brelse(log_blocks[i]); 22641da177e4SLinus Torvalds } 22651da177e4SLinus Torvalds /* flush out the real blocks */ 22661da177e4SLinus Torvalds for (i = 0; i < get_desc_trans_len(desc); i++) { 22671da177e4SLinus Torvalds set_buffer_dirty(real_blocks[i]); 22682a222ca9SMike Christie write_dirty_buffer(real_blocks[i], 0); 22691da177e4SLinus Torvalds } 22701da177e4SLinus Torvalds for (i = 0; i < get_desc_trans_len(desc); i++) { 22711da177e4SLinus Torvalds wait_on_buffer(real_blocks[i]); 22721da177e4SLinus Torvalds if (!buffer_uptodate(real_blocks[i])) { 2273a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1226", 227445b03d5eSJeff Mahoney "REPLAY FAILURE, fsck required! " 227545b03d5eSJeff Mahoney "buffer write failed"); 2276bd4c625cSLinus Torvalds brelse_array(real_blocks + i, 2277bd4c625cSLinus Torvalds get_desc_trans_len(desc) - i); 22781da177e4SLinus Torvalds brelse(c_bh); 22791da177e4SLinus Torvalds brelse(d_bh); 2280d739b42bSPekka Enberg kfree(log_blocks); 2281d739b42bSPekka Enberg kfree(real_blocks); 22821da177e4SLinus Torvalds return -1; 22831da177e4SLinus Torvalds } 22841da177e4SLinus Torvalds brelse(real_blocks[i]); 22851da177e4SLinus Torvalds } 2286bd4c625cSLinus Torvalds cur_dblock = 2287a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2288bd4c625cSLinus Torvalds ((trans_offset + get_desc_trans_len(desc) + 2289a9dd3643SJeff Mahoney 2) % SB_ONDISK_JOURNAL_SIZE(sb)); 2290a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2291bd4c625cSLinus Torvalds "journal-1095: setting journal " "start to offset %ld", 2292a9dd3643SJeff Mahoney cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb)); 22931da177e4SLinus Torvalds 2294098297b2SJeff Mahoney /* 2295098297b2SJeff Mahoney * init starting values for the first transaction, in case 2296098297b2SJeff Mahoney * this is the last transaction to be replayed. 2297098297b2SJeff Mahoney */ 2298a9dd3643SJeff Mahoney journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb); 22991da177e4SLinus Torvalds journal->j_last_flush_trans_id = trans_id; 23001da177e4SLinus Torvalds journal->j_trans_id = trans_id + 1; 2301a44c94a7SAlexander Zarochentsev /* check for trans_id overflow */ 2302a44c94a7SAlexander Zarochentsev if (journal->j_trans_id == 0) 2303a44c94a7SAlexander Zarochentsev journal->j_trans_id = 10; 23041da177e4SLinus Torvalds brelse(c_bh); 23051da177e4SLinus Torvalds brelse(d_bh); 2306d739b42bSPekka Enberg kfree(log_blocks); 2307d739b42bSPekka Enberg kfree(real_blocks); 23081da177e4SLinus Torvalds return 0; 23091da177e4SLinus Torvalds } 23101da177e4SLinus Torvalds 2311098297b2SJeff Mahoney /* 2312098297b2SJeff Mahoney * This function reads blocks starting from block and to max_block of bufsize 2313098297b2SJeff Mahoney * size (but no more than BUFNR blocks at a time). This proved to improve 2314098297b2SJeff Mahoney * mounting speed on self-rebuilding raid5 arrays at least. 2315098297b2SJeff Mahoney * Right now it is only used from journal code. But later we might use it 2316098297b2SJeff Mahoney * from other places. 2317098297b2SJeff Mahoney * Note: Do not use journal_getblk/sb_getblk functions here! 2318098297b2SJeff Mahoney */ 23193ee16670SJeff Mahoney static struct buffer_head *reiserfs_breada(struct block_device *dev, 23203ee16670SJeff Mahoney b_blocknr_t block, int bufsize, 23213ee16670SJeff Mahoney b_blocknr_t max_block) 23221da177e4SLinus Torvalds { 23231da177e4SLinus Torvalds struct buffer_head *bhlist[BUFNR]; 23241da177e4SLinus Torvalds unsigned int blocks = BUFNR; 23251da177e4SLinus Torvalds struct buffer_head *bh; 23261da177e4SLinus Torvalds int i, j; 23271da177e4SLinus Torvalds 23281da177e4SLinus Torvalds bh = __getblk(dev, block, bufsize); 23291da177e4SLinus Torvalds if (buffer_uptodate(bh)) 23301da177e4SLinus Torvalds return (bh); 23311da177e4SLinus Torvalds 23321da177e4SLinus Torvalds if (block + BUFNR > max_block) { 23331da177e4SLinus Torvalds blocks = max_block - block; 23341da177e4SLinus Torvalds } 23351da177e4SLinus Torvalds bhlist[0] = bh; 23361da177e4SLinus Torvalds j = 1; 23371da177e4SLinus Torvalds for (i = 1; i < blocks; i++) { 23381da177e4SLinus Torvalds bh = __getblk(dev, block + i, bufsize); 23391da177e4SLinus Torvalds if (buffer_uptodate(bh)) { 23401da177e4SLinus Torvalds brelse(bh); 23411da177e4SLinus Torvalds break; 2342bd4c625cSLinus Torvalds } else 2343bd4c625cSLinus Torvalds bhlist[j++] = bh; 23441da177e4SLinus Torvalds } 2345d554822eSZhang Yi bh = bhlist[0]; 2346d554822eSZhang Yi bh_read_nowait(bh, 0); 2347d554822eSZhang Yi bh_readahead_batch(j - 1, &bhlist[1], 0); 23481da177e4SLinus Torvalds for (i = 1; i < j; i++) 23491da177e4SLinus Torvalds brelse(bhlist[i]); 23501da177e4SLinus Torvalds wait_on_buffer(bh); 23511da177e4SLinus Torvalds if (buffer_uptodate(bh)) 23521da177e4SLinus Torvalds return bh; 23531da177e4SLinus Torvalds brelse(bh); 23541da177e4SLinus Torvalds return NULL; 23551da177e4SLinus Torvalds } 23561da177e4SLinus Torvalds 23571da177e4SLinus Torvalds /* 2358098297b2SJeff Mahoney * read and replay the log 2359098297b2SJeff Mahoney * on a clean unmount, the journal header's next unflushed pointer will be 2360098297b2SJeff Mahoney * to an invalid transaction. This tests that before finding all the 2361098297b2SJeff Mahoney * transactions in the log, which makes normal mount times fast. 2362098297b2SJeff Mahoney * 2363098297b2SJeff Mahoney * After a crash, this starts with the next unflushed transaction, and 2364098297b2SJeff Mahoney * replays until it finds one too old, or invalid. 2365098297b2SJeff Mahoney * 2366098297b2SJeff Mahoney * On exit, it sets things up so the first transaction will work correctly. 2367098297b2SJeff Mahoney * NOTE: only called during fs mount 23681da177e4SLinus Torvalds */ 2369a9dd3643SJeff Mahoney static int journal_read(struct super_block *sb) 2370bd4c625cSLinus Torvalds { 2371a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 23721da177e4SLinus Torvalds struct reiserfs_journal_desc *desc; 2373600ed416SJeff Mahoney unsigned int oldest_trans_id = 0; 2374600ed416SJeff Mahoney unsigned int oldest_invalid_trans_id = 0; 237534d08260SArnd Bergmann time64_t start; 23761da177e4SLinus Torvalds unsigned long oldest_start = 0; 23771da177e4SLinus Torvalds unsigned long cur_dblock = 0; 23781da177e4SLinus Torvalds unsigned long newest_mount_id = 9; 23791da177e4SLinus Torvalds struct buffer_head *d_bh; 23801da177e4SLinus Torvalds struct reiserfs_journal_header *jh; 23811da177e4SLinus Torvalds int valid_journal_header = 0; 23821da177e4SLinus Torvalds int replay_count = 0; 23831da177e4SLinus Torvalds int continue_replay = 1; 23841da177e4SLinus Torvalds int ret; 23851da177e4SLinus Torvalds 2386a9dd3643SJeff Mahoney cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(sb); 2387a1c6f057SDmitry Monakhov reiserfs_info(sb, "checking transaction log (%pg)\n", 2388a1c6f057SDmitry Monakhov journal->j_dev_bd); 238934d08260SArnd Bergmann start = ktime_get_seconds(); 23901da177e4SLinus Torvalds 2391098297b2SJeff Mahoney /* 2392098297b2SJeff Mahoney * step 1, read in the journal header block. Check the transaction 2393098297b2SJeff Mahoney * it says is the first unflushed, and if that transaction is not 2394098297b2SJeff Mahoney * valid, replay is done 23951da177e4SLinus Torvalds */ 2396a9dd3643SJeff Mahoney journal->j_header_bh = journal_bread(sb, 2397a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) 2398a9dd3643SJeff Mahoney + SB_ONDISK_JOURNAL_SIZE(sb)); 23991da177e4SLinus Torvalds if (!journal->j_header_bh) { 24001da177e4SLinus Torvalds return 1; 24011da177e4SLinus Torvalds } 24021da177e4SLinus Torvalds jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data); 2403c499ec24SVladimir V. Saveliev if (le32_to_cpu(jh->j_first_unflushed_offset) < 2404a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb) 2405bd4c625cSLinus Torvalds && le32_to_cpu(jh->j_last_flush_trans_id) > 0) { 2406bd4c625cSLinus Torvalds oldest_start = 2407a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 24081da177e4SLinus Torvalds le32_to_cpu(jh->j_first_unflushed_offset); 24091da177e4SLinus Torvalds oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1; 24101da177e4SLinus Torvalds newest_mount_id = le32_to_cpu(jh->j_mount_id); 2411a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2412bd4c625cSLinus Torvalds "journal-1153: found in " 24131da177e4SLinus Torvalds "header: first_unflushed_offset %d, last_flushed_trans_id " 24141da177e4SLinus Torvalds "%lu", le32_to_cpu(jh->j_first_unflushed_offset), 24151da177e4SLinus Torvalds le32_to_cpu(jh->j_last_flush_trans_id)); 24161da177e4SLinus Torvalds valid_journal_header = 1; 24171da177e4SLinus Torvalds 2418098297b2SJeff Mahoney /* 2419098297b2SJeff Mahoney * now, we try to read the first unflushed offset. If it 2420098297b2SJeff Mahoney * is not valid, there is nothing more we can do, and it 2421098297b2SJeff Mahoney * makes no sense to read through the whole log. 24221da177e4SLinus Torvalds */ 2423bd4c625cSLinus Torvalds d_bh = 2424a9dd3643SJeff Mahoney journal_bread(sb, 2425a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2426bd4c625cSLinus Torvalds le32_to_cpu(jh->j_first_unflushed_offset)); 2427a9dd3643SJeff Mahoney ret = journal_transaction_is_valid(sb, d_bh, NULL, NULL); 24281da177e4SLinus Torvalds if (!ret) { 24291da177e4SLinus Torvalds continue_replay = 0; 24301da177e4SLinus Torvalds } 24311da177e4SLinus Torvalds brelse(d_bh); 24321da177e4SLinus Torvalds goto start_log_replay; 24331da177e4SLinus Torvalds } 24341da177e4SLinus Torvalds 2435098297b2SJeff Mahoney /* 2436098297b2SJeff Mahoney * ok, there are transactions that need to be replayed. start 2437098297b2SJeff Mahoney * with the first log block, find all the valid transactions, and 2438098297b2SJeff Mahoney * pick out the oldest. 24391da177e4SLinus Torvalds */ 2440bd4c625cSLinus Torvalds while (continue_replay 2441bd4c625cSLinus Torvalds && cur_dblock < 2442a9dd3643SJeff Mahoney (SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2443a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb))) { 2444098297b2SJeff Mahoney /* 2445098297b2SJeff Mahoney * Note that it is required for blocksize of primary fs 2446098297b2SJeff Mahoney * device and journal device to be the same 2447098297b2SJeff Mahoney */ 2448bd4c625cSLinus Torvalds d_bh = 2449bd4c625cSLinus Torvalds reiserfs_breada(journal->j_dev_bd, cur_dblock, 2450a9dd3643SJeff Mahoney sb->s_blocksize, 2451a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2452a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb)); 2453bd4c625cSLinus Torvalds ret = 2454a9dd3643SJeff Mahoney journal_transaction_is_valid(sb, d_bh, 2455bd4c625cSLinus Torvalds &oldest_invalid_trans_id, 2456bd4c625cSLinus Torvalds &newest_mount_id); 24571da177e4SLinus Torvalds if (ret == 1) { 24581da177e4SLinus Torvalds desc = (struct reiserfs_journal_desc *)d_bh->b_data; 24591da177e4SLinus Torvalds if (oldest_start == 0) { /* init all oldest_ values */ 24601da177e4SLinus Torvalds oldest_trans_id = get_desc_trans_id(desc); 24611da177e4SLinus Torvalds oldest_start = d_bh->b_blocknr; 24621da177e4SLinus Torvalds newest_mount_id = get_desc_mount_id(desc); 2463a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2464bd4c625cSLinus Torvalds "journal-1179: Setting " 24651da177e4SLinus Torvalds "oldest_start to offset %llu, trans_id %lu", 2466bd4c625cSLinus Torvalds oldest_start - 2467bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK 2468a9dd3643SJeff Mahoney (sb), oldest_trans_id); 24691da177e4SLinus Torvalds } else if (oldest_trans_id > get_desc_trans_id(desc)) { 24701da177e4SLinus Torvalds /* one we just read was older */ 24711da177e4SLinus Torvalds oldest_trans_id = get_desc_trans_id(desc); 24721da177e4SLinus Torvalds oldest_start = d_bh->b_blocknr; 2473a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2474bd4c625cSLinus Torvalds "journal-1180: Resetting " 24751da177e4SLinus Torvalds "oldest_start to offset %lu, trans_id %lu", 2476bd4c625cSLinus Torvalds oldest_start - 2477bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK 2478a9dd3643SJeff Mahoney (sb), oldest_trans_id); 24791da177e4SLinus Torvalds } 24801da177e4SLinus Torvalds if (newest_mount_id < get_desc_mount_id(desc)) { 24811da177e4SLinus Torvalds newest_mount_id = get_desc_mount_id(desc); 2482a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2483bd4c625cSLinus Torvalds "journal-1299: Setting " 2484bd4c625cSLinus Torvalds "newest_mount_id to %d", 2485bd4c625cSLinus Torvalds get_desc_mount_id(desc)); 24861da177e4SLinus Torvalds } 24871da177e4SLinus Torvalds cur_dblock += get_desc_trans_len(desc) + 2; 24881da177e4SLinus Torvalds } else { 24891da177e4SLinus Torvalds cur_dblock++; 24901da177e4SLinus Torvalds } 24911da177e4SLinus Torvalds brelse(d_bh); 24921da177e4SLinus Torvalds } 24931da177e4SLinus Torvalds 24941da177e4SLinus Torvalds start_log_replay: 24951da177e4SLinus Torvalds cur_dblock = oldest_start; 24961da177e4SLinus Torvalds if (oldest_trans_id) { 2497a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2498bd4c625cSLinus Torvalds "journal-1206: Starting replay " 24991da177e4SLinus Torvalds "from offset %llu, trans_id %lu", 2500a9dd3643SJeff Mahoney cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb), 25011da177e4SLinus Torvalds oldest_trans_id); 25021da177e4SLinus Torvalds 25031da177e4SLinus Torvalds } 25041da177e4SLinus Torvalds replay_count = 0; 25051da177e4SLinus Torvalds while (continue_replay && oldest_trans_id > 0) { 2506bd4c625cSLinus Torvalds ret = 2507a9dd3643SJeff Mahoney journal_read_transaction(sb, cur_dblock, oldest_start, 2508bd4c625cSLinus Torvalds oldest_trans_id, newest_mount_id); 25091da177e4SLinus Torvalds if (ret < 0) { 25101da177e4SLinus Torvalds return ret; 25111da177e4SLinus Torvalds } else if (ret != 0) { 25121da177e4SLinus Torvalds break; 25131da177e4SLinus Torvalds } 2514bd4c625cSLinus Torvalds cur_dblock = 2515a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + journal->j_start; 25161da177e4SLinus Torvalds replay_count++; 25171da177e4SLinus Torvalds if (cur_dblock == oldest_start) 25181da177e4SLinus Torvalds break; 25191da177e4SLinus Torvalds } 25201da177e4SLinus Torvalds 25211da177e4SLinus Torvalds if (oldest_trans_id == 0) { 2522a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, 2523bd4c625cSLinus Torvalds "journal-1225: No valid " "transactions found"); 25241da177e4SLinus Torvalds } 2525098297b2SJeff Mahoney /* 2526098297b2SJeff Mahoney * j_start does not get set correctly if we don't replay any 2527098297b2SJeff Mahoney * transactions. if we had a valid journal_header, set j_start 2528098297b2SJeff Mahoney * to the first unflushed transaction value, copy the trans_id 2529098297b2SJeff Mahoney * from the header 25301da177e4SLinus Torvalds */ 25311da177e4SLinus Torvalds if (valid_journal_header && replay_count == 0) { 25321da177e4SLinus Torvalds journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset); 2533bd4c625cSLinus Torvalds journal->j_trans_id = 2534bd4c625cSLinus Torvalds le32_to_cpu(jh->j_last_flush_trans_id) + 1; 2535a44c94a7SAlexander Zarochentsev /* check for trans_id overflow */ 2536a44c94a7SAlexander Zarochentsev if (journal->j_trans_id == 0) 2537a44c94a7SAlexander Zarochentsev journal->j_trans_id = 10; 2538bd4c625cSLinus Torvalds journal->j_last_flush_trans_id = 2539bd4c625cSLinus Torvalds le32_to_cpu(jh->j_last_flush_trans_id); 25401da177e4SLinus Torvalds journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1; 25411da177e4SLinus Torvalds } else { 25421da177e4SLinus Torvalds journal->j_mount_id = newest_mount_id + 1; 25431da177e4SLinus Torvalds } 2544a9dd3643SJeff Mahoney reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1299: Setting " 25451da177e4SLinus Torvalds "newest_mount_id to %lu", journal->j_mount_id); 25461da177e4SLinus Torvalds journal->j_first_unflushed_offset = journal->j_start; 25471da177e4SLinus Torvalds if (replay_count > 0) { 2548a9dd3643SJeff Mahoney reiserfs_info(sb, 2549bd4c625cSLinus Torvalds "replayed %d transactions in %lu seconds\n", 255034d08260SArnd Bergmann replay_count, ktime_get_seconds() - start); 25511da177e4SLinus Torvalds } 2552278f6679SJeff Mahoney /* needed to satisfy the locking in _update_journal_header_block */ 2553278f6679SJeff Mahoney reiserfs_write_lock(sb); 2554a9dd3643SJeff Mahoney if (!bdev_read_only(sb->s_bdev) && 2555a9dd3643SJeff Mahoney _update_journal_header_block(sb, journal->j_start, 2556bd4c625cSLinus Torvalds journal->j_last_flush_trans_id)) { 2557278f6679SJeff Mahoney reiserfs_write_unlock(sb); 2558098297b2SJeff Mahoney /* 2559098297b2SJeff Mahoney * replay failed, caller must call free_journal_ram and abort 2560098297b2SJeff Mahoney * the mount 25611da177e4SLinus Torvalds */ 25621da177e4SLinus Torvalds return -1; 25631da177e4SLinus Torvalds } 2564278f6679SJeff Mahoney reiserfs_write_unlock(sb); 25651da177e4SLinus Torvalds return 0; 25661da177e4SLinus Torvalds } 25671da177e4SLinus Torvalds 25681da177e4SLinus Torvalds static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s) 25691da177e4SLinus Torvalds { 25701da177e4SLinus Torvalds struct reiserfs_journal_list *jl; 25718c777cc4SPekka Enberg jl = kzalloc(sizeof(struct reiserfs_journal_list), 25728c777cc4SPekka Enberg GFP_NOFS | __GFP_NOFAIL); 25731da177e4SLinus Torvalds INIT_LIST_HEAD(&jl->j_list); 25741da177e4SLinus Torvalds INIT_LIST_HEAD(&jl->j_working_list); 25751da177e4SLinus Torvalds INIT_LIST_HEAD(&jl->j_tail_bh_list); 25761da177e4SLinus Torvalds INIT_LIST_HEAD(&jl->j_bh_list); 257790415deaSJeff Mahoney mutex_init(&jl->j_commit_mutex); 25781da177e4SLinus Torvalds SB_JOURNAL(s)->j_num_lists++; 25791da177e4SLinus Torvalds get_journal_list(jl); 25801da177e4SLinus Torvalds return jl; 25811da177e4SLinus Torvalds } 25821da177e4SLinus Torvalds 2583a9dd3643SJeff Mahoney static void journal_list_init(struct super_block *sb) 2584bd4c625cSLinus Torvalds { 2585a9dd3643SJeff Mahoney SB_JOURNAL(sb)->j_current_jl = alloc_journal_list(sb); 25861da177e4SLinus Torvalds } 25871da177e4SLinus Torvalds 25884385bab1SAl Viro static void release_journal_dev(struct super_block *super, 25891da177e4SLinus Torvalds struct reiserfs_journal *journal) 25901da177e4SLinus Torvalds { 259186098fa0SChristoph Hellwig if (journal->j_dev_bd != NULL) { 25924385bab1SAl Viro blkdev_put(journal->j_dev_bd, journal->j_dev_mode); 25931da177e4SLinus Torvalds journal->j_dev_bd = NULL; 25941da177e4SLinus Torvalds } 25951da177e4SLinus Torvalds } 25961da177e4SLinus Torvalds 25971da177e4SLinus Torvalds static int journal_init_dev(struct super_block *super, 25981da177e4SLinus Torvalds struct reiserfs_journal *journal, 25991da177e4SLinus Torvalds const char *jdev_name) 26001da177e4SLinus Torvalds { 26011da177e4SLinus Torvalds int result; 26021da177e4SLinus Torvalds dev_t jdev; 2603e525fd89STejun Heo fmode_t blkdev_mode = FMODE_READ | FMODE_WRITE | FMODE_EXCL; 26041da177e4SLinus Torvalds 26051da177e4SLinus Torvalds result = 0; 26061da177e4SLinus Torvalds 26071da177e4SLinus Torvalds journal->j_dev_bd = NULL; 26081da177e4SLinus Torvalds jdev = SB_ONDISK_JOURNAL_DEVICE(super) ? 26091da177e4SLinus Torvalds new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev; 26101da177e4SLinus Torvalds 26111da177e4SLinus Torvalds if (bdev_read_only(super->s_bdev)) 26121da177e4SLinus Torvalds blkdev_mode = FMODE_READ; 26131da177e4SLinus Torvalds 26141da177e4SLinus Torvalds /* there is no "jdev" option and journal is on separate device */ 26151da177e4SLinus Torvalds if ((!jdev_name || !jdev_name[0])) { 2616e525fd89STejun Heo if (jdev == super->s_dev) 2617e525fd89STejun Heo blkdev_mode &= ~FMODE_EXCL; 2618d4d77629STejun Heo journal->j_dev_bd = blkdev_get_by_dev(jdev, blkdev_mode, 2619*0718afd4SChristoph Hellwig journal, NULL); 2620e5eb8caaSAl Viro journal->j_dev_mode = blkdev_mode; 26211da177e4SLinus Torvalds if (IS_ERR(journal->j_dev_bd)) { 26221da177e4SLinus Torvalds result = PTR_ERR(journal->j_dev_bd); 26231da177e4SLinus Torvalds journal->j_dev_bd = NULL; 262445b03d5eSJeff Mahoney reiserfs_warning(super, "sh-458", 2625ea3edd4dSChristoph Hellwig "cannot init journal device unknown-block(%u,%u): %i", 2626ea3edd4dSChristoph Hellwig MAJOR(jdev), MINOR(jdev), result); 26271da177e4SLinus Torvalds return result; 2628e525fd89STejun Heo } else if (jdev != super->s_dev) 26291da177e4SLinus Torvalds set_blocksize(journal->j_dev_bd, super->s_blocksize); 263086098fa0SChristoph Hellwig 26311da177e4SLinus Torvalds return 0; 26321da177e4SLinus Torvalds } 26331da177e4SLinus Torvalds 2634e5eb8caaSAl Viro journal->j_dev_mode = blkdev_mode; 2635*0718afd4SChristoph Hellwig journal->j_dev_bd = blkdev_get_by_path(jdev_name, blkdev_mode, journal, 2636*0718afd4SChristoph Hellwig NULL); 263786098fa0SChristoph Hellwig if (IS_ERR(journal->j_dev_bd)) { 263886098fa0SChristoph Hellwig result = PTR_ERR(journal->j_dev_bd); 263986098fa0SChristoph Hellwig journal->j_dev_bd = NULL; 26409ad553abSAndrew Morton reiserfs_warning(super, "sh-457", 264186098fa0SChristoph Hellwig "journal_init_dev: Cannot open '%s': %i", 264286098fa0SChristoph Hellwig jdev_name, result); 264386098fa0SChristoph Hellwig return result; 264486098fa0SChristoph Hellwig } 264586098fa0SChristoph Hellwig 26461da177e4SLinus Torvalds set_blocksize(journal->j_dev_bd, super->s_blocksize); 2647bd4c625cSLinus Torvalds reiserfs_info(super, 2648a1c6f057SDmitry Monakhov "journal_init_dev: journal device: %pg\n", 2649a1c6f057SDmitry Monakhov journal->j_dev_bd); 265086098fa0SChristoph Hellwig return 0; 26511da177e4SLinus Torvalds } 26521da177e4SLinus Torvalds 2653098297b2SJeff Mahoney /* 2654cf3d0b81SEdward Shishkin * When creating/tuning a file system user can assign some 2655cf3d0b81SEdward Shishkin * journal params within boundaries which depend on the ratio 2656cf3d0b81SEdward Shishkin * blocksize/standard_blocksize. 2657cf3d0b81SEdward Shishkin * 2658cf3d0b81SEdward Shishkin * For blocks >= standard_blocksize transaction size should 2659cf3d0b81SEdward Shishkin * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more 2660cf3d0b81SEdward Shishkin * then JOURNAL_TRANS_MAX_DEFAULT. 2661cf3d0b81SEdward Shishkin * 2662cf3d0b81SEdward Shishkin * For blocks < standard_blocksize these boundaries should be 2663cf3d0b81SEdward Shishkin * decreased proportionally. 2664cf3d0b81SEdward Shishkin */ 2665cf3d0b81SEdward Shishkin #define REISERFS_STANDARD_BLKSIZE (4096) 2666cf3d0b81SEdward Shishkin 2667a9dd3643SJeff Mahoney static int check_advise_trans_params(struct super_block *sb, 2668cf3d0b81SEdward Shishkin struct reiserfs_journal *journal) 2669cf3d0b81SEdward Shishkin { 2670cf3d0b81SEdward Shishkin if (journal->j_trans_max) { 2671098297b2SJeff Mahoney /* Non-default journal params. Do sanity check for them. */ 2672cf3d0b81SEdward Shishkin int ratio = 1; 2673a9dd3643SJeff Mahoney if (sb->s_blocksize < REISERFS_STANDARD_BLKSIZE) 2674a9dd3643SJeff Mahoney ratio = REISERFS_STANDARD_BLKSIZE / sb->s_blocksize; 2675cf3d0b81SEdward Shishkin 2676cf3d0b81SEdward Shishkin if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio || 2677cf3d0b81SEdward Shishkin journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio || 2678a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb) / journal->j_trans_max < 2679cf3d0b81SEdward Shishkin JOURNAL_MIN_RATIO) { 2680a9dd3643SJeff Mahoney reiserfs_warning(sb, "sh-462", 268145b03d5eSJeff Mahoney "bad transaction max size (%u). " 268245b03d5eSJeff Mahoney "FSCK?", journal->j_trans_max); 2683cf3d0b81SEdward Shishkin return 1; 2684cf3d0b81SEdward Shishkin } 2685cf3d0b81SEdward Shishkin if (journal->j_max_batch != (journal->j_trans_max) * 2686cf3d0b81SEdward Shishkin JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) { 2687a9dd3643SJeff Mahoney reiserfs_warning(sb, "sh-463", 268845b03d5eSJeff Mahoney "bad transaction max batch (%u). " 268945b03d5eSJeff Mahoney "FSCK?", journal->j_max_batch); 2690cf3d0b81SEdward Shishkin return 1; 2691cf3d0b81SEdward Shishkin } 2692cf3d0b81SEdward Shishkin } else { 2693098297b2SJeff Mahoney /* 2694098297b2SJeff Mahoney * Default journal params. 2695098297b2SJeff Mahoney * The file system was created by old version 2696098297b2SJeff Mahoney * of mkreiserfs, so some fields contain zeros, 2697098297b2SJeff Mahoney * and we need to advise proper values for them 2698098297b2SJeff Mahoney */ 2699a9dd3643SJeff Mahoney if (sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) { 2700a9dd3643SJeff Mahoney reiserfs_warning(sb, "sh-464", "bad blocksize (%u)", 2701a9dd3643SJeff Mahoney sb->s_blocksize); 270245b03d5eSJeff Mahoney return 1; 270345b03d5eSJeff Mahoney } 2704cf3d0b81SEdward Shishkin journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT; 2705cf3d0b81SEdward Shishkin journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT; 2706cf3d0b81SEdward Shishkin journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE; 2707cf3d0b81SEdward Shishkin } 2708cf3d0b81SEdward Shishkin return 0; 2709cf3d0b81SEdward Shishkin } 2710cf3d0b81SEdward Shishkin 2711098297b2SJeff Mahoney /* must be called once on fs mount. calls journal_read for you */ 2712a9dd3643SJeff Mahoney int journal_init(struct super_block *sb, const char *j_dev_name, 2713bd4c625cSLinus Torvalds int old_format, unsigned int commit_max_age) 2714bd4c625cSLinus Torvalds { 2715a9dd3643SJeff Mahoney int num_cnodes = SB_ONDISK_JOURNAL_SIZE(sb) * 2; 27161da177e4SLinus Torvalds struct buffer_head *bhjh; 27171da177e4SLinus Torvalds struct reiserfs_super_block *rs; 27181da177e4SLinus Torvalds struct reiserfs_journal_header *jh; 27191da177e4SLinus Torvalds struct reiserfs_journal *journal; 27201da177e4SLinus Torvalds struct reiserfs_journal_list *jl; 272198ea3f50SFrederic Weisbecker int ret; 27221da177e4SLinus Torvalds 2723558feb08SJoe Perches journal = SB_JOURNAL(sb) = vzalloc(sizeof(struct reiserfs_journal)); 27241da177e4SLinus Torvalds if (!journal) { 2725a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1256", 272645b03d5eSJeff Mahoney "unable to get memory for journal structure"); 27271da177e4SLinus Torvalds return 1; 27281da177e4SLinus Torvalds } 27291da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_bitmap_nodes); 27301da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_prealloc_list); 27311da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_working_list); 27321da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_journal_list); 27331da177e4SLinus Torvalds journal->j_persistent_trans = 0; 273437c69b98SFrederic Weisbecker if (reiserfs_allocate_list_bitmaps(sb, journal->j_list_bitmap, 273537c69b98SFrederic Weisbecker reiserfs_bmap_count(sb))) 27361da177e4SLinus Torvalds goto free_and_return; 273798ea3f50SFrederic Weisbecker 2738a9dd3643SJeff Mahoney allocate_bitmap_nodes(sb); 27391da177e4SLinus Torvalds 27401da177e4SLinus Torvalds /* reserved for journal area support */ 2741a9dd3643SJeff Mahoney SB_JOURNAL_1st_RESERVED_BLOCK(sb) = (old_format ? 2742bd4c625cSLinus Torvalds REISERFS_OLD_DISK_OFFSET_IN_BYTES 2743a9dd3643SJeff Mahoney / sb->s_blocksize + 2744a9dd3643SJeff Mahoney reiserfs_bmap_count(sb) + 2745bd4c625cSLinus Torvalds 1 : 2746bd4c625cSLinus Torvalds REISERFS_DISK_OFFSET_IN_BYTES / 2747a9dd3643SJeff Mahoney sb->s_blocksize + 2); 27481da177e4SLinus Torvalds 2749098297b2SJeff Mahoney /* 2750098297b2SJeff Mahoney * Sanity check to see is the standard journal fitting 2751098297b2SJeff Mahoney * within first bitmap (actual for small blocksizes) 2752098297b2SJeff Mahoney */ 2753a9dd3643SJeff Mahoney if (!SB_ONDISK_JOURNAL_DEVICE(sb) && 2754a9dd3643SJeff Mahoney (SB_JOURNAL_1st_RESERVED_BLOCK(sb) + 2755a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb) > sb->s_blocksize * 8)) { 2756a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1393", 275745b03d5eSJeff Mahoney "journal does not fit for area addressed " 275845b03d5eSJeff Mahoney "by first of bitmap blocks. It starts at " 27591da177e4SLinus Torvalds "%u and its size is %u. Block size %ld", 2760a9dd3643SJeff Mahoney SB_JOURNAL_1st_RESERVED_BLOCK(sb), 2761a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb), 2762a9dd3643SJeff Mahoney sb->s_blocksize); 27631da177e4SLinus Torvalds goto free_and_return; 27641da177e4SLinus Torvalds } 27651da177e4SLinus Torvalds 2766a149127bSPavel Skripkin /* 2767a149127bSPavel Skripkin * Sanity check to see if journal first block is correct. 2768a149127bSPavel Skripkin * If journal first block is invalid it can cause 2769a149127bSPavel Skripkin * zeroing important superblock members. 2770a149127bSPavel Skripkin */ 2771a149127bSPavel Skripkin if (!SB_ONDISK_JOURNAL_DEVICE(sb) && 2772a149127bSPavel Skripkin SB_ONDISK_JOURNAL_1st_BLOCK(sb) < SB_JOURNAL_1st_RESERVED_BLOCK(sb)) { 2773a149127bSPavel Skripkin reiserfs_warning(sb, "journal-1393", 2774a149127bSPavel Skripkin "journal 1st super block is invalid: 1st reserved block %d, but actual 1st block is %d", 2775a149127bSPavel Skripkin SB_JOURNAL_1st_RESERVED_BLOCK(sb), 2776a149127bSPavel Skripkin SB_ONDISK_JOURNAL_1st_BLOCK(sb)); 2777a149127bSPavel Skripkin goto free_and_return; 2778a149127bSPavel Skripkin } 2779a149127bSPavel Skripkin 2780a9dd3643SJeff Mahoney if (journal_init_dev(sb, journal, j_dev_name) != 0) { 2781a9dd3643SJeff Mahoney reiserfs_warning(sb, "sh-462", 27826774def6SMasanari Iida "unable to initialize journal device"); 27831da177e4SLinus Torvalds goto free_and_return; 27841da177e4SLinus Torvalds } 27851da177e4SLinus Torvalds 2786a9dd3643SJeff Mahoney rs = SB_DISK_SUPER_BLOCK(sb); 27871da177e4SLinus Torvalds 27881da177e4SLinus Torvalds /* read journal header */ 2789a9dd3643SJeff Mahoney bhjh = journal_bread(sb, 2790a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 2791a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb)); 27921da177e4SLinus Torvalds if (!bhjh) { 2793a9dd3643SJeff Mahoney reiserfs_warning(sb, "sh-459", 279445b03d5eSJeff Mahoney "unable to read journal header"); 27951da177e4SLinus Torvalds goto free_and_return; 27961da177e4SLinus Torvalds } 27971da177e4SLinus Torvalds jh = (struct reiserfs_journal_header *)(bhjh->b_data); 27981da177e4SLinus Torvalds 27991da177e4SLinus Torvalds /* make sure that journal matches to the super block */ 2800bd4c625cSLinus Torvalds if (is_reiserfs_jr(rs) 2801bd4c625cSLinus Torvalds && (le32_to_cpu(jh->jh_journal.jp_journal_magic) != 2802bd4c625cSLinus Torvalds sb_jp_journal_magic(rs))) { 2803a9dd3643SJeff Mahoney reiserfs_warning(sb, "sh-460", 2804a1c6f057SDmitry Monakhov "journal header magic %x (device %pg) does " 280545b03d5eSJeff Mahoney "not match to magic found in super block %x", 280645b03d5eSJeff Mahoney jh->jh_journal.jp_journal_magic, 2807a1c6f057SDmitry Monakhov journal->j_dev_bd, 28081da177e4SLinus Torvalds sb_jp_journal_magic(rs)); 28091da177e4SLinus Torvalds brelse(bhjh); 28101da177e4SLinus Torvalds goto free_and_return; 28111da177e4SLinus Torvalds } 28121da177e4SLinus Torvalds 28131da177e4SLinus Torvalds journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max); 28141da177e4SLinus Torvalds journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch); 2815bd4c625cSLinus Torvalds journal->j_max_commit_age = 2816bd4c625cSLinus Torvalds le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age); 28171da177e4SLinus Torvalds journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE; 28181da177e4SLinus Torvalds 2819a9dd3643SJeff Mahoney if (check_advise_trans_params(sb, journal) != 0) 2820cf3d0b81SEdward Shishkin goto free_and_return; 28211da177e4SLinus Torvalds journal->j_default_max_commit_age = journal->j_max_commit_age; 28221da177e4SLinus Torvalds 28231da177e4SLinus Torvalds if (commit_max_age != 0) { 28241da177e4SLinus Torvalds journal->j_max_commit_age = commit_max_age; 28251da177e4SLinus Torvalds journal->j_max_trans_age = commit_max_age; 28261da177e4SLinus Torvalds } 28271da177e4SLinus Torvalds 2828a1c6f057SDmitry Monakhov reiserfs_info(sb, "journal params: device %pg, size %u, " 28291da177e4SLinus Torvalds "journal first block %u, max trans len %u, max batch %u, " 28301da177e4SLinus Torvalds "max commit age %u, max trans age %u\n", 2831a1c6f057SDmitry Monakhov journal->j_dev_bd, 2832a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb), 2833a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb), 28341da177e4SLinus Torvalds journal->j_trans_max, 28351da177e4SLinus Torvalds journal->j_max_batch, 2836bd4c625cSLinus Torvalds journal->j_max_commit_age, journal->j_max_trans_age); 28371da177e4SLinus Torvalds 28381da177e4SLinus Torvalds brelse(bhjh); 28391da177e4SLinus Torvalds 28401da177e4SLinus Torvalds journal->j_list_bitmap_index = 0; 2841a9dd3643SJeff Mahoney journal_list_init(sb); 28421da177e4SLinus Torvalds 2843bd4c625cSLinus Torvalds memset(journal->j_list_hash_table, 0, 2844bd4c625cSLinus Torvalds JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *)); 28451da177e4SLinus Torvalds 28461da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_dirty_buffers); 28471da177e4SLinus Torvalds spin_lock_init(&journal->j_dirty_buffers_lock); 28481da177e4SLinus Torvalds 28491da177e4SLinus Torvalds journal->j_start = 0; 28501da177e4SLinus Torvalds journal->j_len = 0; 28511da177e4SLinus Torvalds journal->j_len_alloc = 0; 2852a228bf8fSJeff Mahoney atomic_set(&journal->j_wcount, 0); 2853a228bf8fSJeff Mahoney atomic_set(&journal->j_async_throttle, 0); 28541da177e4SLinus Torvalds journal->j_bcount = 0; 28551da177e4SLinus Torvalds journal->j_trans_start_time = 0; 28561da177e4SLinus Torvalds journal->j_last = NULL; 28571da177e4SLinus Torvalds journal->j_first = NULL; 2858a228bf8fSJeff Mahoney init_waitqueue_head(&journal->j_join_wait); 2859f68215c4SJeff Mahoney mutex_init(&journal->j_mutex); 2860afe70259SJeff Mahoney mutex_init(&journal->j_flush_mutex); 28611da177e4SLinus Torvalds 28621da177e4SLinus Torvalds journal->j_trans_id = 10; 28631da177e4SLinus Torvalds journal->j_mount_id = 10; 28641da177e4SLinus Torvalds journal->j_state = 0; 2865a228bf8fSJeff Mahoney atomic_set(&journal->j_jlock, 0); 28661da177e4SLinus Torvalds journal->j_cnode_free_list = allocate_cnodes(num_cnodes); 28671da177e4SLinus Torvalds journal->j_cnode_free_orig = journal->j_cnode_free_list; 28681da177e4SLinus Torvalds journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0; 28691da177e4SLinus Torvalds journal->j_cnode_used = 0; 28701da177e4SLinus Torvalds journal->j_must_wait = 0; 28711da177e4SLinus Torvalds 2872576f6d79SJeff Mahoney if (journal->j_cnode_free == 0) { 2873a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-2004", "Journal cnode memory " 2874576f6d79SJeff Mahoney "allocation failed (%ld bytes). Journal is " 2875576f6d79SJeff Mahoney "too large for available memory. Usually " 2876576f6d79SJeff Mahoney "this is due to a journal that is too large.", 2877576f6d79SJeff Mahoney sizeof (struct reiserfs_journal_cnode) * num_cnodes); 2878576f6d79SJeff Mahoney goto free_and_return; 2879576f6d79SJeff Mahoney } 2880576f6d79SJeff Mahoney 2881a9dd3643SJeff Mahoney init_journal_hash(sb); 28821da177e4SLinus Torvalds jl = journal->j_current_jl; 288337c69b98SFrederic Weisbecker 288437c69b98SFrederic Weisbecker /* 288537c69b98SFrederic Weisbecker * get_list_bitmap() may call flush_commit_list() which 288637c69b98SFrederic Weisbecker * requires the lock. Calling flush_commit_list() shouldn't happen 288737c69b98SFrederic Weisbecker * this early but I like to be paranoid. 288837c69b98SFrederic Weisbecker */ 288937c69b98SFrederic Weisbecker reiserfs_write_lock(sb); 2890a9dd3643SJeff Mahoney jl->j_list_bitmap = get_list_bitmap(sb, jl); 289137c69b98SFrederic Weisbecker reiserfs_write_unlock(sb); 28921da177e4SLinus Torvalds if (!jl->j_list_bitmap) { 2893a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-2005", 289445b03d5eSJeff Mahoney "get_list_bitmap failed for journal list 0"); 28951da177e4SLinus Torvalds goto free_and_return; 28961da177e4SLinus Torvalds } 289737c69b98SFrederic Weisbecker 289837c69b98SFrederic Weisbecker ret = journal_read(sb); 289937c69b98SFrederic Weisbecker if (ret < 0) { 2900a9dd3643SJeff Mahoney reiserfs_warning(sb, "reiserfs-2006", 290145b03d5eSJeff Mahoney "Replay Failure, unable to mount"); 29021da177e4SLinus Torvalds goto free_and_return; 29031da177e4SLinus Torvalds } 29041da177e4SLinus Torvalds 2905c4028958SDavid Howells INIT_DELAYED_WORK(&journal->j_work, flush_async_commits); 2906a9dd3643SJeff Mahoney journal->j_work_sb = sb; 29071da177e4SLinus Torvalds return 0; 29081da177e4SLinus Torvalds free_and_return: 2909a9dd3643SJeff Mahoney free_journal_ram(sb); 29101da177e4SLinus Torvalds return 1; 29111da177e4SLinus Torvalds } 29121da177e4SLinus Torvalds 29131da177e4SLinus Torvalds /* 2914098297b2SJeff Mahoney * test for a polite end of the current transaction. Used by file_write, 2915098297b2SJeff Mahoney * and should be used by delete to make sure they don't write more than 2916098297b2SJeff Mahoney * can fit inside a single transaction 29171da177e4SLinus Torvalds */ 2918bd4c625cSLinus Torvalds int journal_transaction_should_end(struct reiserfs_transaction_handle *th, 2919bd4c625cSLinus Torvalds int new_alloc) 2920bd4c625cSLinus Torvalds { 29211da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(th->t_super); 292234d08260SArnd Bergmann time64_t now = ktime_get_seconds(); 29231da177e4SLinus Torvalds /* cannot restart while nested */ 29241da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 29251da177e4SLinus Torvalds if (th->t_refcount > 1) 29261da177e4SLinus Torvalds return 0; 29271da177e4SLinus Torvalds if (journal->j_must_wait > 0 || 29281da177e4SLinus Torvalds (journal->j_len_alloc + new_alloc) >= journal->j_max_batch || 2929a228bf8fSJeff Mahoney atomic_read(&journal->j_jlock) || 29301da177e4SLinus Torvalds (now - journal->j_trans_start_time) > journal->j_max_trans_age || 29311da177e4SLinus Torvalds journal->j_cnode_free < (journal->j_trans_max * 3)) { 29321da177e4SLinus Torvalds return 1; 29331da177e4SLinus Torvalds } 2934b18c1c6eSDavidlohr Bueso 29356ae1ea44SChris Mason journal->j_len_alloc += new_alloc; 29366ae1ea44SChris Mason th->t_blocks_allocated += new_alloc ; 29371da177e4SLinus Torvalds return 0; 29381da177e4SLinus Torvalds } 29391da177e4SLinus Torvalds 2940098297b2SJeff Mahoney /* this must be called inside a transaction */ 2941bd4c625cSLinus Torvalds void reiserfs_block_writes(struct reiserfs_transaction_handle *th) 2942bd4c625cSLinus Torvalds { 29431da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(th->t_super); 29441da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 29451da177e4SLinus Torvalds journal->j_must_wait = 1; 29461da177e4SLinus Torvalds set_bit(J_WRITERS_BLOCKED, &journal->j_state); 29471da177e4SLinus Torvalds return; 29481da177e4SLinus Torvalds } 29491da177e4SLinus Torvalds 2950098297b2SJeff Mahoney /* this must be called without a transaction started */ 2951bd4c625cSLinus Torvalds void reiserfs_allow_writes(struct super_block *s) 2952bd4c625cSLinus Torvalds { 29531da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 29541da177e4SLinus Torvalds clear_bit(J_WRITERS_BLOCKED, &journal->j_state); 29551da177e4SLinus Torvalds wake_up(&journal->j_join_wait); 29561da177e4SLinus Torvalds } 29571da177e4SLinus Torvalds 2958098297b2SJeff Mahoney /* this must be called without a transaction started */ 2959bd4c625cSLinus Torvalds void reiserfs_wait_on_write_block(struct super_block *s) 2960bd4c625cSLinus Torvalds { 29611da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 29621da177e4SLinus Torvalds wait_event(journal->j_join_wait, 29631da177e4SLinus Torvalds !test_bit(J_WRITERS_BLOCKED, &journal->j_state)); 29641da177e4SLinus Torvalds } 29651da177e4SLinus Torvalds 2966bd4c625cSLinus Torvalds static void queue_log_writer(struct super_block *s) 2967bd4c625cSLinus Torvalds { 2968ac6424b9SIngo Molnar wait_queue_entry_t wait; 29691da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 29701da177e4SLinus Torvalds set_bit(J_WRITERS_QUEUED, &journal->j_state); 29711da177e4SLinus Torvalds 29721da177e4SLinus Torvalds /* 29731da177e4SLinus Torvalds * we don't want to use wait_event here because 29741da177e4SLinus Torvalds * we only want to wait once. 29751da177e4SLinus Torvalds */ 29761da177e4SLinus Torvalds init_waitqueue_entry(&wait, current); 29771da177e4SLinus Torvalds add_wait_queue(&journal->j_join_wait, &wait); 29781da177e4SLinus Torvalds set_current_state(TASK_UNINTERRUPTIBLE); 29798ebc4232SFrederic Weisbecker if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) { 2980278f6679SJeff Mahoney int depth = reiserfs_write_unlock_nested(s); 29811da177e4SLinus Torvalds schedule(); 2982278f6679SJeff Mahoney reiserfs_write_lock_nested(s, depth); 29838ebc4232SFrederic Weisbecker } 29845ab2f7e0SMilind Arun Choudhary __set_current_state(TASK_RUNNING); 29851da177e4SLinus Torvalds remove_wait_queue(&journal->j_join_wait, &wait); 29861da177e4SLinus Torvalds } 29871da177e4SLinus Torvalds 2988bd4c625cSLinus Torvalds static void wake_queued_writers(struct super_block *s) 2989bd4c625cSLinus Torvalds { 29901da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 29911da177e4SLinus Torvalds if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state)) 29921da177e4SLinus Torvalds wake_up(&journal->j_join_wait); 29931da177e4SLinus Torvalds } 29941da177e4SLinus Torvalds 2995600ed416SJeff Mahoney static void let_transaction_grow(struct super_block *sb, unsigned int trans_id) 29961da177e4SLinus Torvalds { 29971da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(sb); 29981da177e4SLinus Torvalds unsigned long bcount = journal->j_bcount; 29991da177e4SLinus Torvalds while (1) { 3000278f6679SJeff Mahoney int depth; 3001278f6679SJeff Mahoney 3002278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(sb); 3003041e0e3bSNishanth Aravamudan schedule_timeout_uninterruptible(1); 3004278f6679SJeff Mahoney reiserfs_write_lock_nested(sb, depth); 3005278f6679SJeff Mahoney 30061da177e4SLinus Torvalds journal->j_current_jl->j_state |= LIST_COMMIT_PENDING; 30071da177e4SLinus Torvalds while ((atomic_read(&journal->j_wcount) > 0 || 30081da177e4SLinus Torvalds atomic_read(&journal->j_jlock)) && 30091da177e4SLinus Torvalds journal->j_trans_id == trans_id) { 30101da177e4SLinus Torvalds queue_log_writer(sb); 30111da177e4SLinus Torvalds } 30121da177e4SLinus Torvalds if (journal->j_trans_id != trans_id) 30131da177e4SLinus Torvalds break; 30141da177e4SLinus Torvalds if (bcount == journal->j_bcount) 30151da177e4SLinus Torvalds break; 30161da177e4SLinus Torvalds bcount = journal->j_bcount; 30171da177e4SLinus Torvalds } 30181da177e4SLinus Torvalds } 30191da177e4SLinus Torvalds 3020098297b2SJeff Mahoney /* 3021098297b2SJeff Mahoney * join == true if you must join an existing transaction. 3022098297b2SJeff Mahoney * join == false if you can deal with waiting for others to finish 3023098297b2SJeff Mahoney * 3024098297b2SJeff Mahoney * this will block until the transaction is joinable. send the number of 3025098297b2SJeff Mahoney * blocks you expect to use in nblocks. 30261da177e4SLinus Torvalds */ 3027bd4c625cSLinus Torvalds static int do_journal_begin_r(struct reiserfs_transaction_handle *th, 3028a9dd3643SJeff Mahoney struct super_block *sb, unsigned long nblocks, 3029bd4c625cSLinus Torvalds int join) 3030bd4c625cSLinus Torvalds { 303134d08260SArnd Bergmann time64_t now = ktime_get_seconds(); 3032600ed416SJeff Mahoney unsigned int old_trans_id; 3033a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 30341da177e4SLinus Torvalds struct reiserfs_transaction_handle myth; 30351da177e4SLinus Torvalds int retval; 3036278f6679SJeff Mahoney int depth; 30371da177e4SLinus Torvalds 3038a9dd3643SJeff Mahoney reiserfs_check_lock_depth(sb, "journal_begin"); 303914a61442SEric Sesterhenn BUG_ON(nblocks > journal->j_trans_max); 30401da177e4SLinus Torvalds 3041a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.journal_being); 30421da177e4SLinus Torvalds /* set here for journal_join */ 30431da177e4SLinus Torvalds th->t_refcount = 1; 3044a9dd3643SJeff Mahoney th->t_super = sb; 30451da177e4SLinus Torvalds 30461da177e4SLinus Torvalds relock: 3047a9dd3643SJeff Mahoney lock_journal(sb); 30481da177e4SLinus Torvalds if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) { 3049a9dd3643SJeff Mahoney unlock_journal(sb); 30501da177e4SLinus Torvalds retval = journal->j_errno; 30511da177e4SLinus Torvalds goto out_fail; 30521da177e4SLinus Torvalds } 30531da177e4SLinus Torvalds journal->j_bcount++; 30541da177e4SLinus Torvalds 30551da177e4SLinus Torvalds if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) { 3056a9dd3643SJeff Mahoney unlock_journal(sb); 3057278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(sb); 3058a9dd3643SJeff Mahoney reiserfs_wait_on_write_block(sb); 3059278f6679SJeff Mahoney reiserfs_write_lock_nested(sb, depth); 3060a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.journal_relock_writers); 30611da177e4SLinus Torvalds goto relock; 30621da177e4SLinus Torvalds } 306334d08260SArnd Bergmann now = ktime_get_seconds(); 30641da177e4SLinus Torvalds 3065098297b2SJeff Mahoney /* 3066098297b2SJeff Mahoney * if there is no room in the journal OR 3067098297b2SJeff Mahoney * if this transaction is too old, and we weren't called joinable, 3068098297b2SJeff Mahoney * wait for it to finish before beginning we don't sleep if there 3069098297b2SJeff Mahoney * aren't other writers 30701da177e4SLinus Torvalds */ 30711da177e4SLinus Torvalds 30721da177e4SLinus Torvalds if ((!join && journal->j_must_wait > 0) || 3073bd4c625cSLinus Torvalds (!join 3074bd4c625cSLinus Torvalds && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch) 3075bd4c625cSLinus Torvalds || (!join && atomic_read(&journal->j_wcount) > 0 3076bd4c625cSLinus Torvalds && journal->j_trans_start_time > 0 3077bd4c625cSLinus Torvalds && (now - journal->j_trans_start_time) > 3078bd4c625cSLinus Torvalds journal->j_max_trans_age) || (!join 3079bd4c625cSLinus Torvalds && atomic_read(&journal->j_jlock)) 3080bd4c625cSLinus Torvalds || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) { 30811da177e4SLinus Torvalds 30821da177e4SLinus Torvalds old_trans_id = journal->j_trans_id; 3083098297b2SJeff Mahoney /* allow others to finish this transaction */ 3084098297b2SJeff Mahoney unlock_journal(sb); 30851da177e4SLinus Torvalds 30861da177e4SLinus Torvalds if (!join && (journal->j_len_alloc + nblocks + 2) >= 30871da177e4SLinus Torvalds journal->j_max_batch && 3088bd4c625cSLinus Torvalds ((journal->j_len + nblocks + 2) * 100) < 3089bd4c625cSLinus Torvalds (journal->j_len_alloc * 75)) { 30901da177e4SLinus Torvalds if (atomic_read(&journal->j_wcount) > 10) { 3091a9dd3643SJeff Mahoney queue_log_writer(sb); 30921da177e4SLinus Torvalds goto relock; 30931da177e4SLinus Torvalds } 30941da177e4SLinus Torvalds } 3095098297b2SJeff Mahoney /* 3096098297b2SJeff Mahoney * don't mess with joining the transaction if all we 3097098297b2SJeff Mahoney * have to do is wait for someone else to do a commit 30981da177e4SLinus Torvalds */ 30991da177e4SLinus Torvalds if (atomic_read(&journal->j_jlock)) { 31001da177e4SLinus Torvalds while (journal->j_trans_id == old_trans_id && 31011da177e4SLinus Torvalds atomic_read(&journal->j_jlock)) { 3102a9dd3643SJeff Mahoney queue_log_writer(sb); 31031da177e4SLinus Torvalds } 31041da177e4SLinus Torvalds goto relock; 31051da177e4SLinus Torvalds } 3106b491dd17SJeff Mahoney retval = journal_join(&myth, sb); 31071da177e4SLinus Torvalds if (retval) 31081da177e4SLinus Torvalds goto out_fail; 31091da177e4SLinus Torvalds 31101da177e4SLinus Torvalds /* someone might have ended the transaction while we joined */ 31111da177e4SLinus Torvalds if (old_trans_id != journal->j_trans_id) { 311258d85426SJeff Mahoney retval = do_journal_end(&myth, 0); 31131da177e4SLinus Torvalds } else { 311458d85426SJeff Mahoney retval = do_journal_end(&myth, COMMIT_NOW); 31151da177e4SLinus Torvalds } 31161da177e4SLinus Torvalds 31171da177e4SLinus Torvalds if (retval) 31181da177e4SLinus Torvalds goto out_fail; 31191da177e4SLinus Torvalds 3120a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.journal_relock_wcount); 31211da177e4SLinus Torvalds goto relock; 31221da177e4SLinus Torvalds } 31231da177e4SLinus Torvalds /* we are the first writer, set trans_id */ 31241da177e4SLinus Torvalds if (journal->j_trans_start_time == 0) { 312534d08260SArnd Bergmann journal->j_trans_start_time = ktime_get_seconds(); 31261da177e4SLinus Torvalds } 3127a228bf8fSJeff Mahoney atomic_inc(&journal->j_wcount); 31281da177e4SLinus Torvalds journal->j_len_alloc += nblocks; 31291da177e4SLinus Torvalds th->t_blocks_logged = 0; 31301da177e4SLinus Torvalds th->t_blocks_allocated = nblocks; 31311da177e4SLinus Torvalds th->t_trans_id = journal->j_trans_id; 3132a9dd3643SJeff Mahoney unlock_journal(sb); 31331da177e4SLinus Torvalds INIT_LIST_HEAD(&th->t_list); 31341da177e4SLinus Torvalds return 0; 31351da177e4SLinus Torvalds 31361da177e4SLinus Torvalds out_fail: 31371da177e4SLinus Torvalds memset(th, 0, sizeof(*th)); 3138098297b2SJeff Mahoney /* 3139098297b2SJeff Mahoney * Re-set th->t_super, so we can properly keep track of how many 31401da177e4SLinus Torvalds * persistent transactions there are. We need to do this so if this 3141098297b2SJeff Mahoney * call is part of a failed restart_transaction, we can free it later 3142098297b2SJeff Mahoney */ 3143a9dd3643SJeff Mahoney th->t_super = sb; 31441da177e4SLinus Torvalds return retval; 31451da177e4SLinus Torvalds } 31461da177e4SLinus Torvalds 3147bd4c625cSLinus Torvalds struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct 3148bd4c625cSLinus Torvalds super_block 3149bd4c625cSLinus Torvalds *s, 3150bd4c625cSLinus Torvalds int nblocks) 3151bd4c625cSLinus Torvalds { 31521da177e4SLinus Torvalds int ret; 31531da177e4SLinus Torvalds struct reiserfs_transaction_handle *th; 31541da177e4SLinus Torvalds 3155098297b2SJeff Mahoney /* 3156098297b2SJeff Mahoney * if we're nesting into an existing transaction. It will be 3157098297b2SJeff Mahoney * persistent on its own 31581da177e4SLinus Torvalds */ 31591da177e4SLinus Torvalds if (reiserfs_transaction_running(s)) { 31601da177e4SLinus Torvalds th = current->journal_info; 31611da177e4SLinus Torvalds th->t_refcount++; 316214a61442SEric Sesterhenn BUG_ON(th->t_refcount < 2); 316314a61442SEric Sesterhenn 31641da177e4SLinus Torvalds return th; 31651da177e4SLinus Torvalds } 3166d739b42bSPekka Enberg th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS); 31671da177e4SLinus Torvalds if (!th) 31681da177e4SLinus Torvalds return NULL; 31691da177e4SLinus Torvalds ret = journal_begin(th, s, nblocks); 31701da177e4SLinus Torvalds if (ret) { 3171d739b42bSPekka Enberg kfree(th); 31721da177e4SLinus Torvalds return NULL; 31731da177e4SLinus Torvalds } 31741da177e4SLinus Torvalds 31751da177e4SLinus Torvalds SB_JOURNAL(s)->j_persistent_trans++; 31761da177e4SLinus Torvalds return th; 31771da177e4SLinus Torvalds } 31781da177e4SLinus Torvalds 3179bd4c625cSLinus Torvalds int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th) 3180bd4c625cSLinus Torvalds { 31811da177e4SLinus Torvalds struct super_block *s = th->t_super; 31821da177e4SLinus Torvalds int ret = 0; 31831da177e4SLinus Torvalds if (th->t_trans_id) 318458d85426SJeff Mahoney ret = journal_end(th); 31851da177e4SLinus Torvalds else 31861da177e4SLinus Torvalds ret = -EIO; 31871da177e4SLinus Torvalds if (th->t_refcount == 0) { 31881da177e4SLinus Torvalds SB_JOURNAL(s)->j_persistent_trans--; 3189d739b42bSPekka Enberg kfree(th); 31901da177e4SLinus Torvalds } 31911da177e4SLinus Torvalds return ret; 31921da177e4SLinus Torvalds } 31931da177e4SLinus Torvalds 3194bd4c625cSLinus Torvalds static int journal_join(struct reiserfs_transaction_handle *th, 3195b491dd17SJeff Mahoney struct super_block *sb) 3196bd4c625cSLinus Torvalds { 31971da177e4SLinus Torvalds struct reiserfs_transaction_handle *cur_th = current->journal_info; 31981da177e4SLinus Torvalds 3199098297b2SJeff Mahoney /* 3200098297b2SJeff Mahoney * this keeps do_journal_end from NULLing out the 3201098297b2SJeff Mahoney * current->journal_info pointer 32021da177e4SLinus Torvalds */ 32031da177e4SLinus Torvalds th->t_handle_save = cur_th; 320414a61442SEric Sesterhenn BUG_ON(cur_th && cur_th->t_refcount > 1); 3205b491dd17SJeff Mahoney return do_journal_begin_r(th, sb, 1, JBEGIN_JOIN); 32061da177e4SLinus Torvalds } 32071da177e4SLinus Torvalds 3208bd4c625cSLinus Torvalds int journal_join_abort(struct reiserfs_transaction_handle *th, 3209b491dd17SJeff Mahoney struct super_block *sb) 3210bd4c625cSLinus Torvalds { 32111da177e4SLinus Torvalds struct reiserfs_transaction_handle *cur_th = current->journal_info; 32121da177e4SLinus Torvalds 3213098297b2SJeff Mahoney /* 3214098297b2SJeff Mahoney * this keeps do_journal_end from NULLing out the 3215098297b2SJeff Mahoney * current->journal_info pointer 32161da177e4SLinus Torvalds */ 32171da177e4SLinus Torvalds th->t_handle_save = cur_th; 321814a61442SEric Sesterhenn BUG_ON(cur_th && cur_th->t_refcount > 1); 3219b491dd17SJeff Mahoney return do_journal_begin_r(th, sb, 1, JBEGIN_ABORT); 32201da177e4SLinus Torvalds } 32211da177e4SLinus Torvalds 3222bd4c625cSLinus Torvalds int journal_begin(struct reiserfs_transaction_handle *th, 3223a9dd3643SJeff Mahoney struct super_block *sb, unsigned long nblocks) 3224bd4c625cSLinus Torvalds { 32251da177e4SLinus Torvalds struct reiserfs_transaction_handle *cur_th = current->journal_info; 32261da177e4SLinus Torvalds int ret; 32271da177e4SLinus Torvalds 32281da177e4SLinus Torvalds th->t_handle_save = NULL; 32291da177e4SLinus Torvalds if (cur_th) { 32301da177e4SLinus Torvalds /* we are nesting into the current transaction */ 3231a9dd3643SJeff Mahoney if (cur_th->t_super == sb) { 32321da177e4SLinus Torvalds BUG_ON(!cur_th->t_refcount); 32331da177e4SLinus Torvalds cur_th->t_refcount++; 32341da177e4SLinus Torvalds memcpy(th, cur_th, sizeof(*th)); 32351da177e4SLinus Torvalds if (th->t_refcount <= 1) 3236a9dd3643SJeff Mahoney reiserfs_warning(sb, "reiserfs-2005", 323745b03d5eSJeff Mahoney "BAD: refcount <= 1, but " 323845b03d5eSJeff Mahoney "journal_info != 0"); 32391da177e4SLinus Torvalds return 0; 32401da177e4SLinus Torvalds } else { 3241098297b2SJeff Mahoney /* 3242098297b2SJeff Mahoney * we've ended up with a handle from a different 3243098297b2SJeff Mahoney * filesystem. save it and restore on journal_end. 3244098297b2SJeff Mahoney * This should never really happen... 32451da177e4SLinus Torvalds */ 3246a9dd3643SJeff Mahoney reiserfs_warning(sb, "clm-2100", 324745b03d5eSJeff Mahoney "nesting info a different FS"); 32481da177e4SLinus Torvalds th->t_handle_save = current->journal_info; 32491da177e4SLinus Torvalds current->journal_info = th; 32501da177e4SLinus Torvalds } 32511da177e4SLinus Torvalds } else { 32521da177e4SLinus Torvalds current->journal_info = th; 32531da177e4SLinus Torvalds } 3254a9dd3643SJeff Mahoney ret = do_journal_begin_r(th, sb, nblocks, JBEGIN_REG); 325514a61442SEric Sesterhenn BUG_ON(current->journal_info != th); 32561da177e4SLinus Torvalds 3257098297b2SJeff Mahoney /* 3258098297b2SJeff Mahoney * I guess this boils down to being the reciprocal of clm-2100 above. 3259098297b2SJeff Mahoney * If do_journal_begin_r fails, we need to put it back, since 3260098297b2SJeff Mahoney * journal_end won't be called to do it. */ 32611da177e4SLinus Torvalds if (ret) 32621da177e4SLinus Torvalds current->journal_info = th->t_handle_save; 32631da177e4SLinus Torvalds else 32641da177e4SLinus Torvalds BUG_ON(!th->t_refcount); 32651da177e4SLinus Torvalds 32661da177e4SLinus Torvalds return ret; 32671da177e4SLinus Torvalds } 32681da177e4SLinus Torvalds 32691da177e4SLinus Torvalds /* 3270098297b2SJeff Mahoney * puts bh into the current transaction. If it was already there, reorders 3271098297b2SJeff Mahoney * removes the old pointers from the hash, and puts new ones in (to make 3272098297b2SJeff Mahoney * sure replay happen in the right order). 3273098297b2SJeff Mahoney * 3274098297b2SJeff Mahoney * if it was dirty, cleans and files onto the clean list. I can't let it 3275098297b2SJeff Mahoney * be dirty again until the transaction is committed. 3276098297b2SJeff Mahoney * 3277098297b2SJeff Mahoney * if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len. 32781da177e4SLinus Torvalds */ 3279bd4c625cSLinus Torvalds int journal_mark_dirty(struct reiserfs_transaction_handle *th, 328009f1b80bSJeff Mahoney struct buffer_head *bh) 3281bd4c625cSLinus Torvalds { 328209f1b80bSJeff Mahoney struct super_block *sb = th->t_super; 3283a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 32841da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn = NULL; 32851da177e4SLinus Torvalds int count_already_incd = 0; 32861da177e4SLinus Torvalds int prepared = 0; 32871da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 32881da177e4SLinus Torvalds 3289a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.mark_dirty); 32901da177e4SLinus Torvalds if (th->t_trans_id != journal->j_trans_id) { 3291c3a9c210SJeff Mahoney reiserfs_panic(th->t_super, "journal-1577", 3292c3a9c210SJeff Mahoney "handle trans id %ld != current trans id %ld", 32931da177e4SLinus Torvalds th->t_trans_id, journal->j_trans_id); 32941da177e4SLinus Torvalds } 32951da177e4SLinus Torvalds 32961da177e4SLinus Torvalds prepared = test_clear_buffer_journal_prepared(bh); 32971da177e4SLinus Torvalds clear_buffer_journal_restore_dirty(bh); 32981da177e4SLinus Torvalds /* already in this transaction, we are done */ 32991da177e4SLinus Torvalds if (buffer_journaled(bh)) { 3300a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.mark_dirty_already); 33011da177e4SLinus Torvalds return 0; 33021da177e4SLinus Torvalds } 33031da177e4SLinus Torvalds 3304098297b2SJeff Mahoney /* 3305098297b2SJeff Mahoney * this must be turned into a panic instead of a warning. We can't 3306098297b2SJeff Mahoney * allow a dirty or journal_dirty or locked buffer to be logged, as 3307098297b2SJeff Mahoney * some changes could get to disk too early. NOT GOOD. 33081da177e4SLinus Torvalds */ 33091da177e4SLinus Torvalds if (!prepared || buffer_dirty(bh)) { 3310a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1777", 331145b03d5eSJeff Mahoney "buffer %llu bad state " 33121da177e4SLinus Torvalds "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT", 3313bd4c625cSLinus Torvalds (unsigned long long)bh->b_blocknr, 3314bd4c625cSLinus Torvalds prepared ? ' ' : '!', 33151da177e4SLinus Torvalds buffer_locked(bh) ? ' ' : '!', 33161da177e4SLinus Torvalds buffer_dirty(bh) ? ' ' : '!', 33171da177e4SLinus Torvalds buffer_journal_dirty(bh) ? ' ' : '!'); 33181da177e4SLinus Torvalds } 33191da177e4SLinus Torvalds 3320a228bf8fSJeff Mahoney if (atomic_read(&journal->j_wcount) <= 0) { 3321a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1409", 332245b03d5eSJeff Mahoney "returning because j_wcount was %d", 3323a228bf8fSJeff Mahoney atomic_read(&journal->j_wcount)); 33241da177e4SLinus Torvalds return 1; 33251da177e4SLinus Torvalds } 3326098297b2SJeff Mahoney /* 3327098297b2SJeff Mahoney * this error means I've screwed up, and we've overflowed 3328098297b2SJeff Mahoney * the transaction. Nothing can be done here, except make the 3329098297b2SJeff Mahoney * FS readonly or panic. 33301da177e4SLinus Torvalds */ 33311da177e4SLinus Torvalds if (journal->j_len >= journal->j_trans_max) { 3332c3a9c210SJeff Mahoney reiserfs_panic(th->t_super, "journal-1413", 3333c3a9c210SJeff Mahoney "j_len (%lu) is too big", 3334bd4c625cSLinus Torvalds journal->j_len); 33351da177e4SLinus Torvalds } 33361da177e4SLinus Torvalds 33371da177e4SLinus Torvalds if (buffer_journal_dirty(bh)) { 33381da177e4SLinus Torvalds count_already_incd = 1; 3339a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.mark_dirty_notjournal); 33401da177e4SLinus Torvalds clear_buffer_journal_dirty(bh); 33411da177e4SLinus Torvalds } 33421da177e4SLinus Torvalds 33431da177e4SLinus Torvalds if (journal->j_len > journal->j_len_alloc) { 33441da177e4SLinus Torvalds journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT; 33451da177e4SLinus Torvalds } 33461da177e4SLinus Torvalds 33471da177e4SLinus Torvalds set_buffer_journaled(bh); 33481da177e4SLinus Torvalds 33491da177e4SLinus Torvalds /* now put this guy on the end */ 33501da177e4SLinus Torvalds if (!cn) { 3351a9dd3643SJeff Mahoney cn = get_cnode(sb); 33521da177e4SLinus Torvalds if (!cn) { 3353a9dd3643SJeff Mahoney reiserfs_panic(sb, "journal-4", "get_cnode failed!"); 33541da177e4SLinus Torvalds } 33551da177e4SLinus Torvalds 33561da177e4SLinus Torvalds if (th->t_blocks_logged == th->t_blocks_allocated) { 33571da177e4SLinus Torvalds th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT; 33581da177e4SLinus Torvalds journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT; 33591da177e4SLinus Torvalds } 33601da177e4SLinus Torvalds th->t_blocks_logged++; 33611da177e4SLinus Torvalds journal->j_len++; 33621da177e4SLinus Torvalds 33631da177e4SLinus Torvalds cn->bh = bh; 33641da177e4SLinus Torvalds cn->blocknr = bh->b_blocknr; 3365a9dd3643SJeff Mahoney cn->sb = sb; 33661da177e4SLinus Torvalds cn->jlist = NULL; 33671da177e4SLinus Torvalds insert_journal_hash(journal->j_hash_table, cn); 33681da177e4SLinus Torvalds if (!count_already_incd) { 33691da177e4SLinus Torvalds get_bh(bh); 33701da177e4SLinus Torvalds } 33711da177e4SLinus Torvalds } 33721da177e4SLinus Torvalds cn->next = NULL; 33731da177e4SLinus Torvalds cn->prev = journal->j_last; 33741da177e4SLinus Torvalds cn->bh = bh; 33751da177e4SLinus Torvalds if (journal->j_last) { 33761da177e4SLinus Torvalds journal->j_last->next = cn; 33771da177e4SLinus Torvalds journal->j_last = cn; 33781da177e4SLinus Torvalds } else { 33791da177e4SLinus Torvalds journal->j_first = cn; 33801da177e4SLinus Torvalds journal->j_last = cn; 33811da177e4SLinus Torvalds } 3382033369d1SArtem Bityutskiy reiserfs_schedule_old_flush(sb); 33831da177e4SLinus Torvalds return 0; 33841da177e4SLinus Torvalds } 33851da177e4SLinus Torvalds 338658d85426SJeff Mahoney int journal_end(struct reiserfs_transaction_handle *th) 3387bd4c625cSLinus Torvalds { 338858d85426SJeff Mahoney struct super_block *sb = th->t_super; 33891da177e4SLinus Torvalds if (!current->journal_info && th->t_refcount > 1) 3390a9dd3643SJeff Mahoney reiserfs_warning(sb, "REISER-NESTING", 339145b03d5eSJeff Mahoney "th NULL, refcount %d", th->t_refcount); 33921da177e4SLinus Torvalds 33931da177e4SLinus Torvalds if (!th->t_trans_id) { 33941da177e4SLinus Torvalds WARN_ON(1); 33951da177e4SLinus Torvalds return -EIO; 33961da177e4SLinus Torvalds } 33971da177e4SLinus Torvalds 33981da177e4SLinus Torvalds th->t_refcount--; 33991da177e4SLinus Torvalds if (th->t_refcount > 0) { 3400bd4c625cSLinus Torvalds struct reiserfs_transaction_handle *cur_th = 3401bd4c625cSLinus Torvalds current->journal_info; 34021da177e4SLinus Torvalds 3403098297b2SJeff Mahoney /* 3404098297b2SJeff Mahoney * we aren't allowed to close a nested transaction on a 3405098297b2SJeff Mahoney * different filesystem from the one in the task struct 34061da177e4SLinus Torvalds */ 340714a61442SEric Sesterhenn BUG_ON(cur_th->t_super != th->t_super); 34081da177e4SLinus Torvalds 34091da177e4SLinus Torvalds if (th != cur_th) { 34101da177e4SLinus Torvalds memcpy(current->journal_info, th, sizeof(*th)); 34111da177e4SLinus Torvalds th->t_trans_id = 0; 34121da177e4SLinus Torvalds } 34131da177e4SLinus Torvalds return 0; 34141da177e4SLinus Torvalds } else { 341558d85426SJeff Mahoney return do_journal_end(th, 0); 34161da177e4SLinus Torvalds } 34171da177e4SLinus Torvalds } 34181da177e4SLinus Torvalds 3419098297b2SJeff Mahoney /* 3420098297b2SJeff Mahoney * removes from the current transaction, relsing and descrementing any counters. 3421098297b2SJeff Mahoney * also files the removed buffer directly onto the clean list 3422098297b2SJeff Mahoney * 3423098297b2SJeff Mahoney * called by journal_mark_freed when a block has been deleted 3424098297b2SJeff Mahoney * 3425098297b2SJeff Mahoney * returns 1 if it cleaned and relsed the buffer. 0 otherwise 34261da177e4SLinus Torvalds */ 3427a9dd3643SJeff Mahoney static int remove_from_transaction(struct super_block *sb, 3428bd4c625cSLinus Torvalds b_blocknr_t blocknr, int already_cleaned) 3429bd4c625cSLinus Torvalds { 34301da177e4SLinus Torvalds struct buffer_head *bh; 34311da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 3432a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 34331da177e4SLinus Torvalds int ret = 0; 34341da177e4SLinus Torvalds 3435a9dd3643SJeff Mahoney cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr); 34361da177e4SLinus Torvalds if (!cn || !cn->bh) { 34371da177e4SLinus Torvalds return ret; 34381da177e4SLinus Torvalds } 34391da177e4SLinus Torvalds bh = cn->bh; 34401da177e4SLinus Torvalds if (cn->prev) { 34411da177e4SLinus Torvalds cn->prev->next = cn->next; 34421da177e4SLinus Torvalds } 34431da177e4SLinus Torvalds if (cn->next) { 34441da177e4SLinus Torvalds cn->next->prev = cn->prev; 34451da177e4SLinus Torvalds } 34461da177e4SLinus Torvalds if (cn == journal->j_first) { 34471da177e4SLinus Torvalds journal->j_first = cn->next; 34481da177e4SLinus Torvalds } 34491da177e4SLinus Torvalds if (cn == journal->j_last) { 34501da177e4SLinus Torvalds journal->j_last = cn->prev; 34511da177e4SLinus Torvalds } 3452a9dd3643SJeff Mahoney remove_journal_hash(sb, journal->j_hash_table, NULL, 3453bd4c625cSLinus Torvalds bh->b_blocknr, 0); 34541da177e4SLinus Torvalds clear_buffer_journaled(bh); /* don't log this one */ 34551da177e4SLinus Torvalds 34561da177e4SLinus Torvalds if (!already_cleaned) { 34571da177e4SLinus Torvalds clear_buffer_journal_dirty(bh); 34581da177e4SLinus Torvalds clear_buffer_dirty(bh); 34591da177e4SLinus Torvalds clear_buffer_journal_test(bh); 34601da177e4SLinus Torvalds put_bh(bh); 3461a228bf8fSJeff Mahoney if (atomic_read(&bh->b_count) < 0) { 3462a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-1752", 346345b03d5eSJeff Mahoney "b_count < 0"); 34641da177e4SLinus Torvalds } 34651da177e4SLinus Torvalds ret = 1; 34661da177e4SLinus Torvalds } 34671da177e4SLinus Torvalds journal->j_len--; 34681da177e4SLinus Torvalds journal->j_len_alloc--; 3469a9dd3643SJeff Mahoney free_cnode(sb, cn); 34701da177e4SLinus Torvalds return ret; 34711da177e4SLinus Torvalds } 34721da177e4SLinus Torvalds 34731da177e4SLinus Torvalds /* 3474098297b2SJeff Mahoney * for any cnode in a journal list, it can only be dirtied of all the 3475098297b2SJeff Mahoney * transactions that include it are committed to disk. 3476098297b2SJeff Mahoney * this checks through each transaction, and returns 1 if you are allowed 3477098297b2SJeff Mahoney * to dirty, and 0 if you aren't 3478098297b2SJeff Mahoney * 3479098297b2SJeff Mahoney * it is called by dirty_journal_list, which is called after 3480098297b2SJeff Mahoney * flush_commit_list has gotten all the log blocks for a given 3481098297b2SJeff Mahoney * transaction on disk 3482098297b2SJeff Mahoney * 34831da177e4SLinus Torvalds */ 3484bd4c625cSLinus Torvalds static int can_dirty(struct reiserfs_journal_cnode *cn) 3485bd4c625cSLinus Torvalds { 34861da177e4SLinus Torvalds struct super_block *sb = cn->sb; 34871da177e4SLinus Torvalds b_blocknr_t blocknr = cn->blocknr; 34881da177e4SLinus Torvalds struct reiserfs_journal_cnode *cur = cn->hprev; 34891da177e4SLinus Torvalds int can_dirty = 1; 34901da177e4SLinus Torvalds 3491098297b2SJeff Mahoney /* 3492098297b2SJeff Mahoney * first test hprev. These are all newer than cn, so any node here 3493098297b2SJeff Mahoney * with the same block number and dev means this node can't be sent 3494098297b2SJeff Mahoney * to disk right now. 34951da177e4SLinus Torvalds */ 34961da177e4SLinus Torvalds while (cur && can_dirty) { 34971da177e4SLinus Torvalds if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb && 34981da177e4SLinus Torvalds cur->blocknr == blocknr) { 34991da177e4SLinus Torvalds can_dirty = 0; 35001da177e4SLinus Torvalds } 35011da177e4SLinus Torvalds cur = cur->hprev; 35021da177e4SLinus Torvalds } 3503098297b2SJeff Mahoney /* 3504098297b2SJeff Mahoney * then test hnext. These are all older than cn. As long as they 3505098297b2SJeff Mahoney * are committed to the log, it is safe to write cn to disk 35061da177e4SLinus Torvalds */ 35071da177e4SLinus Torvalds cur = cn->hnext; 35081da177e4SLinus Torvalds while (cur && can_dirty) { 35091da177e4SLinus Torvalds if (cur->jlist && cur->jlist->j_len > 0 && 3510a228bf8fSJeff Mahoney atomic_read(&cur->jlist->j_commit_left) > 0 && cur->bh && 35111da177e4SLinus Torvalds cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) { 35121da177e4SLinus Torvalds can_dirty = 0; 35131da177e4SLinus Torvalds } 35141da177e4SLinus Torvalds cur = cur->hnext; 35151da177e4SLinus Torvalds } 35161da177e4SLinus Torvalds return can_dirty; 35171da177e4SLinus Torvalds } 35181da177e4SLinus Torvalds 3519098297b2SJeff Mahoney /* 3520098297b2SJeff Mahoney * syncs the commit blocks, but does not force the real buffers to disk 3521098297b2SJeff Mahoney * will wait until the current transaction is done/committed before returning 35221da177e4SLinus Torvalds */ 352358d85426SJeff Mahoney int journal_end_sync(struct reiserfs_transaction_handle *th) 3524bd4c625cSLinus Torvalds { 352558d85426SJeff Mahoney struct super_block *sb = th->t_super; 3526a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 35271da177e4SLinus Torvalds 35281da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 35291da177e4SLinus Torvalds /* you can sync while nested, very, very bad */ 353014a61442SEric Sesterhenn BUG_ON(th->t_refcount > 1); 35311da177e4SLinus Torvalds if (journal->j_len == 0) { 3532a9dd3643SJeff Mahoney reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb), 3533bd4c625cSLinus Torvalds 1); 353409f1b80bSJeff Mahoney journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb)); 35351da177e4SLinus Torvalds } 353658d85426SJeff Mahoney return do_journal_end(th, COMMIT_NOW | WAIT); 35371da177e4SLinus Torvalds } 35381da177e4SLinus Torvalds 3539098297b2SJeff Mahoney /* writeback the pending async commits to disk */ 3540c4028958SDavid Howells static void flush_async_commits(struct work_struct *work) 3541bd4c625cSLinus Torvalds { 3542c4028958SDavid Howells struct reiserfs_journal *journal = 3543c4028958SDavid Howells container_of(work, struct reiserfs_journal, j_work.work); 3544a9dd3643SJeff Mahoney struct super_block *sb = journal->j_work_sb; 35451da177e4SLinus Torvalds struct reiserfs_journal_list *jl; 35461da177e4SLinus Torvalds struct list_head *entry; 35471da177e4SLinus Torvalds 35488ebc4232SFrederic Weisbecker reiserfs_write_lock(sb); 35491da177e4SLinus Torvalds if (!list_empty(&journal->j_journal_list)) { 35501da177e4SLinus Torvalds /* last entry is the youngest, commit it and you get everything */ 35511da177e4SLinus Torvalds entry = journal->j_journal_list.prev; 35521da177e4SLinus Torvalds jl = JOURNAL_LIST_ENTRY(entry); 3553a9dd3643SJeff Mahoney flush_commit_list(sb, jl, 1); 35541da177e4SLinus Torvalds } 35558ebc4232SFrederic Weisbecker reiserfs_write_unlock(sb); 35561da177e4SLinus Torvalds } 35571da177e4SLinus Torvalds 35581da177e4SLinus Torvalds /* 3559098297b2SJeff Mahoney * flushes any old transactions to disk 3560098297b2SJeff Mahoney * ends the current transaction if it is too old 35611da177e4SLinus Torvalds */ 356225729b0eSArtem Bityutskiy void reiserfs_flush_old_commits(struct super_block *sb) 3563bd4c625cSLinus Torvalds { 356434d08260SArnd Bergmann time64_t now; 35651da177e4SLinus Torvalds struct reiserfs_transaction_handle th; 3566a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 35671da177e4SLinus Torvalds 356834d08260SArnd Bergmann now = ktime_get_seconds(); 3569098297b2SJeff Mahoney /* 3570098297b2SJeff Mahoney * safety check so we don't flush while we are replaying the log during 35711da177e4SLinus Torvalds * mount 35721da177e4SLinus Torvalds */ 357325729b0eSArtem Bityutskiy if (list_empty(&journal->j_journal_list)) 357425729b0eSArtem Bityutskiy return; 35751da177e4SLinus Torvalds 3576098297b2SJeff Mahoney /* 3577098297b2SJeff Mahoney * check the current transaction. If there are no writers, and it is 35781da177e4SLinus Torvalds * too old, finish it, and force the commit blocks to disk 35791da177e4SLinus Torvalds */ 35801da177e4SLinus Torvalds if (atomic_read(&journal->j_wcount) <= 0 && 35811da177e4SLinus Torvalds journal->j_trans_start_time > 0 && 35821da177e4SLinus Torvalds journal->j_len > 0 && 3583bd4c625cSLinus Torvalds (now - journal->j_trans_start_time) > journal->j_max_trans_age) { 3584b491dd17SJeff Mahoney if (!journal_join(&th, sb)) { 3585a9dd3643SJeff Mahoney reiserfs_prepare_for_journal(sb, 3586a9dd3643SJeff Mahoney SB_BUFFER_WITH_SB(sb), 3587bd4c625cSLinus Torvalds 1); 358809f1b80bSJeff Mahoney journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb)); 35891da177e4SLinus Torvalds 3590098297b2SJeff Mahoney /* 3591098297b2SJeff Mahoney * we're only being called from kreiserfsd, it makes 3592098297b2SJeff Mahoney * no sense to do an async commit so that kreiserfsd 3593098297b2SJeff Mahoney * can do it later 35941da177e4SLinus Torvalds */ 359558d85426SJeff Mahoney do_journal_end(&th, COMMIT_NOW | WAIT); 35961da177e4SLinus Torvalds } 35971da177e4SLinus Torvalds } 35981da177e4SLinus Torvalds } 35991da177e4SLinus Torvalds 36001da177e4SLinus Torvalds /* 3601098297b2SJeff Mahoney * returns 0 if do_journal_end should return right away, returns 1 if 3602098297b2SJeff Mahoney * do_journal_end should finish the commit 3603098297b2SJeff Mahoney * 3604098297b2SJeff Mahoney * if the current transaction is too old, but still has writers, this will 3605098297b2SJeff Mahoney * wait on j_join_wait until all the writers are done. By the time it 3606098297b2SJeff Mahoney * wakes up, the transaction it was called has already ended, so it just 3607098297b2SJeff Mahoney * flushes the commit list and returns 0. 3608098297b2SJeff Mahoney * 3609098297b2SJeff Mahoney * Won't batch when flush or commit_now is set. Also won't batch when 3610098297b2SJeff Mahoney * others are waiting on j_join_wait. 3611098297b2SJeff Mahoney * 3612098297b2SJeff Mahoney * Note, we can't allow the journal_end to proceed while there are still 3613098297b2SJeff Mahoney * writers in the log. 36141da177e4SLinus Torvalds */ 361558d85426SJeff Mahoney static int check_journal_end(struct reiserfs_transaction_handle *th, int flags) 3616bd4c625cSLinus Torvalds { 36171da177e4SLinus Torvalds 361834d08260SArnd Bergmann time64_t now; 36191da177e4SLinus Torvalds int flush = flags & FLUSH_ALL; 36201da177e4SLinus Torvalds int commit_now = flags & COMMIT_NOW; 36211da177e4SLinus Torvalds int wait_on_commit = flags & WAIT; 36221da177e4SLinus Torvalds struct reiserfs_journal_list *jl; 362358d85426SJeff Mahoney struct super_block *sb = th->t_super; 3624a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 36251da177e4SLinus Torvalds 36261da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 36271da177e4SLinus Torvalds 36281da177e4SLinus Torvalds if (th->t_trans_id != journal->j_trans_id) { 3629c3a9c210SJeff Mahoney reiserfs_panic(th->t_super, "journal-1577", 3630c3a9c210SJeff Mahoney "handle trans id %ld != current trans id %ld", 36311da177e4SLinus Torvalds th->t_trans_id, journal->j_trans_id); 36321da177e4SLinus Torvalds } 36331da177e4SLinus Torvalds 36341da177e4SLinus Torvalds journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged); 3635098297b2SJeff Mahoney /* <= 0 is allowed. unmounting might not call begin */ 3636a228bf8fSJeff Mahoney if (atomic_read(&journal->j_wcount) > 0) 3637a228bf8fSJeff Mahoney atomic_dec(&journal->j_wcount); 36381da177e4SLinus Torvalds 3639098297b2SJeff Mahoney /* 3640098297b2SJeff Mahoney * BUG, deal with case where j_len is 0, but people previously 3641098297b2SJeff Mahoney * freed blocks need to be released will be dealt with by next 3642098297b2SJeff Mahoney * transaction that actually writes something, but should be taken 3643098297b2SJeff Mahoney * care of in this trans 36441da177e4SLinus Torvalds */ 364514a61442SEric Sesterhenn BUG_ON(journal->j_len == 0); 364614a61442SEric Sesterhenn 3647098297b2SJeff Mahoney /* 3648098297b2SJeff Mahoney * if wcount > 0, and we are called to with flush or commit_now, 3649098297b2SJeff Mahoney * we wait on j_join_wait. We will wake up when the last writer has 3650098297b2SJeff Mahoney * finished the transaction, and started it on its way to the disk. 3651098297b2SJeff Mahoney * Then, we flush the commit or journal list, and just return 0 3652098297b2SJeff Mahoney * because the rest of journal end was already done for this 3653098297b2SJeff Mahoney * transaction. 36541da177e4SLinus Torvalds */ 3655a228bf8fSJeff Mahoney if (atomic_read(&journal->j_wcount) > 0) { 36561da177e4SLinus Torvalds if (flush || commit_now) { 36571da177e4SLinus Torvalds unsigned trans_id; 36581da177e4SLinus Torvalds 36591da177e4SLinus Torvalds jl = journal->j_current_jl; 36601da177e4SLinus Torvalds trans_id = jl->j_trans_id; 36611da177e4SLinus Torvalds if (wait_on_commit) 36621da177e4SLinus Torvalds jl->j_state |= LIST_COMMIT_PENDING; 3663a228bf8fSJeff Mahoney atomic_set(&journal->j_jlock, 1); 36641da177e4SLinus Torvalds if (flush) { 36651da177e4SLinus Torvalds journal->j_next_full_flush = 1; 36661da177e4SLinus Torvalds } 3667a9dd3643SJeff Mahoney unlock_journal(sb); 36681da177e4SLinus Torvalds 3669098297b2SJeff Mahoney /* 3670098297b2SJeff Mahoney * sleep while the current transaction is 3671098297b2SJeff Mahoney * still j_jlocked 3672098297b2SJeff Mahoney */ 36731da177e4SLinus Torvalds while (journal->j_trans_id == trans_id) { 36741da177e4SLinus Torvalds if (atomic_read(&journal->j_jlock)) { 3675a9dd3643SJeff Mahoney queue_log_writer(sb); 36761da177e4SLinus Torvalds } else { 3677a9dd3643SJeff Mahoney lock_journal(sb); 36781da177e4SLinus Torvalds if (journal->j_trans_id == trans_id) { 3679a228bf8fSJeff Mahoney atomic_set(&journal->j_jlock, 3680bd4c625cSLinus Torvalds 1); 36811da177e4SLinus Torvalds } 3682a9dd3643SJeff Mahoney unlock_journal(sb); 36831da177e4SLinus Torvalds } 36841da177e4SLinus Torvalds } 368514a61442SEric Sesterhenn BUG_ON(journal->j_trans_id == trans_id); 368614a61442SEric Sesterhenn 3687bd4c625cSLinus Torvalds if (commit_now 3688a9dd3643SJeff Mahoney && journal_list_still_alive(sb, trans_id) 3689bd4c625cSLinus Torvalds && wait_on_commit) { 3690a9dd3643SJeff Mahoney flush_commit_list(sb, jl, 1); 36911da177e4SLinus Torvalds } 36921da177e4SLinus Torvalds return 0; 36931da177e4SLinus Torvalds } 3694a9dd3643SJeff Mahoney unlock_journal(sb); 36951da177e4SLinus Torvalds return 0; 36961da177e4SLinus Torvalds } 36971da177e4SLinus Torvalds 36981da177e4SLinus Torvalds /* deal with old transactions where we are the last writers */ 369934d08260SArnd Bergmann now = ktime_get_seconds(); 37001da177e4SLinus Torvalds if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) { 37011da177e4SLinus Torvalds commit_now = 1; 37021da177e4SLinus Torvalds journal->j_next_async_flush = 1; 37031da177e4SLinus Torvalds } 37041da177e4SLinus Torvalds /* don't batch when someone is waiting on j_join_wait */ 37051da177e4SLinus Torvalds /* don't batch when syncing the commit or flushing the whole trans */ 3706a228bf8fSJeff Mahoney if (!(journal->j_must_wait > 0) && !(atomic_read(&journal->j_jlock)) 3707bd4c625cSLinus Torvalds && !flush && !commit_now && (journal->j_len < journal->j_max_batch) 3708bd4c625cSLinus Torvalds && journal->j_len_alloc < journal->j_max_batch 3709bd4c625cSLinus Torvalds && journal->j_cnode_free > (journal->j_trans_max * 3)) { 37101da177e4SLinus Torvalds journal->j_bcount++; 3711a9dd3643SJeff Mahoney unlock_journal(sb); 37121da177e4SLinus Torvalds return 0; 37131da177e4SLinus Torvalds } 37141da177e4SLinus Torvalds 3715a9dd3643SJeff Mahoney if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(sb)) { 3716a9dd3643SJeff Mahoney reiserfs_panic(sb, "journal-003", 3717c3a9c210SJeff Mahoney "j_start (%ld) is too high", 3718bd4c625cSLinus Torvalds journal->j_start); 37191da177e4SLinus Torvalds } 37201da177e4SLinus Torvalds return 1; 37211da177e4SLinus Torvalds } 37221da177e4SLinus Torvalds 37231da177e4SLinus Torvalds /* 3724098297b2SJeff Mahoney * Does all the work that makes deleting blocks safe. 3725098297b2SJeff Mahoney * when deleting a block mark BH_JNew, just remove it from the current 3726098297b2SJeff Mahoney * transaction, clean it's buffer_head and move on. 3727098297b2SJeff Mahoney * 3728098297b2SJeff Mahoney * otherwise: 3729098297b2SJeff Mahoney * set a bit for the block in the journal bitmap. That will prevent it from 3730098297b2SJeff Mahoney * being allocated for unformatted nodes before this transaction has finished. 3731098297b2SJeff Mahoney * 3732098297b2SJeff Mahoney * mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers. 3733098297b2SJeff Mahoney * That will prevent any old transactions with this block from trying to flush 3734098297b2SJeff Mahoney * to the real location. Since we aren't removing the cnode from the 3735098297b2SJeff Mahoney * journal_list_hash, *the block can't be reallocated yet. 3736098297b2SJeff Mahoney * 3737098297b2SJeff Mahoney * Then remove it from the current transaction, decrementing any counters and 3738098297b2SJeff Mahoney * filing it on the clean list. 37391da177e4SLinus Torvalds */ 3740bd4c625cSLinus Torvalds int journal_mark_freed(struct reiserfs_transaction_handle *th, 3741a9dd3643SJeff Mahoney struct super_block *sb, b_blocknr_t blocknr) 3742bd4c625cSLinus Torvalds { 3743a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 37441da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn = NULL; 37451da177e4SLinus Torvalds struct buffer_head *bh = NULL; 37461da177e4SLinus Torvalds struct reiserfs_list_bitmap *jb = NULL; 37471da177e4SLinus Torvalds int cleaned = 0; 37481da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 37491da177e4SLinus Torvalds 3750a9dd3643SJeff Mahoney cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr); 37511da177e4SLinus Torvalds if (cn && cn->bh) { 37521da177e4SLinus Torvalds bh = cn->bh; 37531da177e4SLinus Torvalds get_bh(bh); 37541da177e4SLinus Torvalds } 37551da177e4SLinus Torvalds /* if it is journal new, we just remove it from this transaction */ 37561da177e4SLinus Torvalds if (bh && buffer_journal_new(bh)) { 37571da177e4SLinus Torvalds clear_buffer_journal_new(bh); 37581da177e4SLinus Torvalds clear_prepared_bits(bh); 37591da177e4SLinus Torvalds reiserfs_clean_and_file_buffer(bh); 3760a9dd3643SJeff Mahoney cleaned = remove_from_transaction(sb, blocknr, cleaned); 37611da177e4SLinus Torvalds } else { 3762098297b2SJeff Mahoney /* 3763098297b2SJeff Mahoney * set the bit for this block in the journal bitmap 3764098297b2SJeff Mahoney * for this transaction 3765098297b2SJeff Mahoney */ 37661da177e4SLinus Torvalds jb = journal->j_current_jl->j_list_bitmap; 37671da177e4SLinus Torvalds if (!jb) { 3768a9dd3643SJeff Mahoney reiserfs_panic(sb, "journal-1702", 3769c3a9c210SJeff Mahoney "journal_list_bitmap is NULL"); 37701da177e4SLinus Torvalds } 3771a9dd3643SJeff Mahoney set_bit_in_list_bitmap(sb, blocknr, jb); 37721da177e4SLinus Torvalds 37731da177e4SLinus Torvalds /* Note, the entire while loop is not allowed to schedule. */ 37741da177e4SLinus Torvalds 37751da177e4SLinus Torvalds if (bh) { 37761da177e4SLinus Torvalds clear_prepared_bits(bh); 37771da177e4SLinus Torvalds reiserfs_clean_and_file_buffer(bh); 37781da177e4SLinus Torvalds } 3779a9dd3643SJeff Mahoney cleaned = remove_from_transaction(sb, blocknr, cleaned); 37801da177e4SLinus Torvalds 3781098297b2SJeff Mahoney /* 3782098297b2SJeff Mahoney * find all older transactions with this block, 3783098297b2SJeff Mahoney * make sure they don't try to write it out 3784098297b2SJeff Mahoney */ 3785a9dd3643SJeff Mahoney cn = get_journal_hash_dev(sb, journal->j_list_hash_table, 3786bd4c625cSLinus Torvalds blocknr); 37871da177e4SLinus Torvalds while (cn) { 3788a9dd3643SJeff Mahoney if (sb == cn->sb && blocknr == cn->blocknr) { 37891da177e4SLinus Torvalds set_bit(BLOCK_FREED, &cn->state); 37901da177e4SLinus Torvalds if (cn->bh) { 3791098297b2SJeff Mahoney /* 3792098297b2SJeff Mahoney * remove_from_transaction will brelse 3793098297b2SJeff Mahoney * the buffer if it was in the current 3794098297b2SJeff Mahoney * trans 37951da177e4SLinus Torvalds */ 3796098297b2SJeff Mahoney if (!cleaned) { 3797bd4c625cSLinus Torvalds clear_buffer_journal_dirty(cn-> 3798bd4c625cSLinus Torvalds bh); 37991da177e4SLinus Torvalds clear_buffer_dirty(cn->bh); 3800bd4c625cSLinus Torvalds clear_buffer_journal_test(cn-> 3801bd4c625cSLinus Torvalds bh); 38021da177e4SLinus Torvalds cleaned = 1; 38031da177e4SLinus Torvalds put_bh(cn->bh); 3804bd4c625cSLinus Torvalds if (atomic_read 3805a228bf8fSJeff Mahoney (&cn->bh->b_count) < 0) { 3806a9dd3643SJeff Mahoney reiserfs_warning(sb, 380745b03d5eSJeff Mahoney "journal-2138", 380845b03d5eSJeff Mahoney "cn->bh->b_count < 0"); 38091da177e4SLinus Torvalds } 38101da177e4SLinus Torvalds } 3811098297b2SJeff Mahoney /* 3812098297b2SJeff Mahoney * since we are clearing the bh, 3813098297b2SJeff Mahoney * we MUST dec nonzerolen 3814098297b2SJeff Mahoney */ 3815098297b2SJeff Mahoney if (cn->jlist) { 3816a228bf8fSJeff Mahoney atomic_dec(&cn->jlist-> 3817a228bf8fSJeff Mahoney j_nonzerolen); 38181da177e4SLinus Torvalds } 38191da177e4SLinus Torvalds cn->bh = NULL; 38201da177e4SLinus Torvalds } 38211da177e4SLinus Torvalds } 38221da177e4SLinus Torvalds cn = cn->hnext; 38231da177e4SLinus Torvalds } 38241da177e4SLinus Torvalds } 38251da177e4SLinus Torvalds 3826398c95bdSChris Mason if (bh) 3827398c95bdSChris Mason release_buffer_page(bh); /* get_hash grabs the buffer */ 38281da177e4SLinus Torvalds return 0; 38291da177e4SLinus Torvalds } 38301da177e4SLinus Torvalds 3831bd4c625cSLinus Torvalds void reiserfs_update_inode_transaction(struct inode *inode) 3832bd4c625cSLinus Torvalds { 38331da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb); 38341da177e4SLinus Torvalds REISERFS_I(inode)->i_jl = journal->j_current_jl; 38351da177e4SLinus Torvalds REISERFS_I(inode)->i_trans_id = journal->j_trans_id; 38361da177e4SLinus Torvalds } 38371da177e4SLinus Torvalds 38381da177e4SLinus Torvalds /* 38391da177e4SLinus Torvalds * returns -1 on error, 0 if no commits/barriers were done and 1 38401da177e4SLinus Torvalds * if a transaction was actually committed and the barrier was done 38411da177e4SLinus Torvalds */ 38421da177e4SLinus Torvalds static int __commit_trans_jl(struct inode *inode, unsigned long id, 38431da177e4SLinus Torvalds struct reiserfs_journal_list *jl) 38441da177e4SLinus Torvalds { 38451da177e4SLinus Torvalds struct reiserfs_transaction_handle th; 38461da177e4SLinus Torvalds struct super_block *sb = inode->i_sb; 38471da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(sb); 38481da177e4SLinus Torvalds int ret = 0; 38491da177e4SLinus Torvalds 3850098297b2SJeff Mahoney /* 3851098297b2SJeff Mahoney * is it from the current transaction, 3852098297b2SJeff Mahoney * or from an unknown transaction? 3853098297b2SJeff Mahoney */ 38541da177e4SLinus Torvalds if (id == journal->j_trans_id) { 38551da177e4SLinus Torvalds jl = journal->j_current_jl; 3856098297b2SJeff Mahoney /* 3857098297b2SJeff Mahoney * try to let other writers come in and 3858098297b2SJeff Mahoney * grow this transaction 3859098297b2SJeff Mahoney */ 38601da177e4SLinus Torvalds let_transaction_grow(sb, id); 38611da177e4SLinus Torvalds if (journal->j_trans_id != id) { 38621da177e4SLinus Torvalds goto flush_commit_only; 38631da177e4SLinus Torvalds } 38641da177e4SLinus Torvalds 38651da177e4SLinus Torvalds ret = journal_begin(&th, sb, 1); 38661da177e4SLinus Torvalds if (ret) 38671da177e4SLinus Torvalds return ret; 38681da177e4SLinus Torvalds 38691da177e4SLinus Torvalds /* someone might have ended this transaction while we joined */ 38701da177e4SLinus Torvalds if (journal->j_trans_id != id) { 3871bd4c625cSLinus Torvalds reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb), 3872bd4c625cSLinus Torvalds 1); 387309f1b80bSJeff Mahoney journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb)); 387458d85426SJeff Mahoney ret = journal_end(&th); 38751da177e4SLinus Torvalds goto flush_commit_only; 38761da177e4SLinus Torvalds } 38771da177e4SLinus Torvalds 387858d85426SJeff Mahoney ret = journal_end_sync(&th); 38791da177e4SLinus Torvalds if (!ret) 38801da177e4SLinus Torvalds ret = 1; 38811da177e4SLinus Torvalds 38821da177e4SLinus Torvalds } else { 3883098297b2SJeff Mahoney /* 3884098297b2SJeff Mahoney * this gets tricky, we have to make sure the journal list in 38851da177e4SLinus Torvalds * the inode still exists. We know the list is still around 38861da177e4SLinus Torvalds * if we've got a larger transaction id than the oldest list 38871da177e4SLinus Torvalds */ 38881da177e4SLinus Torvalds flush_commit_only: 38891da177e4SLinus Torvalds if (journal_list_still_alive(inode->i_sb, id)) { 38901da177e4SLinus Torvalds /* 38911da177e4SLinus Torvalds * we only set ret to 1 when we know for sure 38921da177e4SLinus Torvalds * the barrier hasn't been started yet on the commit 38931da177e4SLinus Torvalds * block. 38941da177e4SLinus Torvalds */ 38951da177e4SLinus Torvalds if (atomic_read(&jl->j_commit_left) > 1) 38961da177e4SLinus Torvalds ret = 1; 38971da177e4SLinus Torvalds flush_commit_list(sb, jl, 1); 38981da177e4SLinus Torvalds if (journal->j_errno) 38991da177e4SLinus Torvalds ret = journal->j_errno; 39001da177e4SLinus Torvalds } 39011da177e4SLinus Torvalds } 39021da177e4SLinus Torvalds /* otherwise the list is gone, and long since committed */ 39031da177e4SLinus Torvalds return ret; 39041da177e4SLinus Torvalds } 39051da177e4SLinus Torvalds 3906bd4c625cSLinus Torvalds int reiserfs_commit_for_inode(struct inode *inode) 3907bd4c625cSLinus Torvalds { 3908600ed416SJeff Mahoney unsigned int id = REISERFS_I(inode)->i_trans_id; 39091da177e4SLinus Torvalds struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl; 39101da177e4SLinus Torvalds 3911098297b2SJeff Mahoney /* 3912098297b2SJeff Mahoney * for the whole inode, assume unset id means it was 39131da177e4SLinus Torvalds * changed in the current transaction. More conservative 39141da177e4SLinus Torvalds */ 39151da177e4SLinus Torvalds if (!id || !jl) { 39161da177e4SLinus Torvalds reiserfs_update_inode_transaction(inode); 39171da177e4SLinus Torvalds id = REISERFS_I(inode)->i_trans_id; 39181da177e4SLinus Torvalds /* jl will be updated in __commit_trans_jl */ 39191da177e4SLinus Torvalds } 39201da177e4SLinus Torvalds 39211da177e4SLinus Torvalds return __commit_trans_jl(inode, id, jl); 39221da177e4SLinus Torvalds } 39231da177e4SLinus Torvalds 3924a9dd3643SJeff Mahoney void reiserfs_restore_prepared_buffer(struct super_block *sb, 3925bd4c625cSLinus Torvalds struct buffer_head *bh) 3926bd4c625cSLinus Torvalds { 3927a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 3928a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.restore_prepared); 39291da177e4SLinus Torvalds if (!bh) { 39301da177e4SLinus Torvalds return; 39311da177e4SLinus Torvalds } 39321da177e4SLinus Torvalds if (test_clear_buffer_journal_restore_dirty(bh) && 39331da177e4SLinus Torvalds buffer_journal_dirty(bh)) { 39341da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 3935278f6679SJeff Mahoney reiserfs_write_lock(sb); 3936a9dd3643SJeff Mahoney cn = get_journal_hash_dev(sb, 39371da177e4SLinus Torvalds journal->j_list_hash_table, 39381da177e4SLinus Torvalds bh->b_blocknr); 39391da177e4SLinus Torvalds if (cn && can_dirty(cn)) { 39401da177e4SLinus Torvalds set_buffer_journal_test(bh); 39411da177e4SLinus Torvalds mark_buffer_dirty(bh); 39421da177e4SLinus Torvalds } 3943278f6679SJeff Mahoney reiserfs_write_unlock(sb); 39441da177e4SLinus Torvalds } 39451da177e4SLinus Torvalds clear_buffer_journal_prepared(bh); 39461da177e4SLinus Torvalds } 39471da177e4SLinus Torvalds 39481da177e4SLinus Torvalds extern struct tree_balance *cur_tb; 39491da177e4SLinus Torvalds /* 3950098297b2SJeff Mahoney * before we can change a metadata block, we have to make sure it won't 3951098297b2SJeff Mahoney * be written to disk while we are altering it. So, we must: 3952098297b2SJeff Mahoney * clean it 3953098297b2SJeff Mahoney * wait on it. 39541da177e4SLinus Torvalds */ 3955a9dd3643SJeff Mahoney int reiserfs_prepare_for_journal(struct super_block *sb, 3956bd4c625cSLinus Torvalds struct buffer_head *bh, int wait) 3957bd4c625cSLinus Torvalds { 3958a9dd3643SJeff Mahoney PROC_INFO_INC(sb, journal.prepare); 39591da177e4SLinus Torvalds 3960ca5de404SNick Piggin if (!trylock_buffer(bh)) { 39611da177e4SLinus Torvalds if (!wait) 39621da177e4SLinus Torvalds return 0; 39631da177e4SLinus Torvalds lock_buffer(bh); 39641da177e4SLinus Torvalds } 39651da177e4SLinus Torvalds set_buffer_journal_prepared(bh); 39661da177e4SLinus Torvalds if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) { 39671da177e4SLinus Torvalds clear_buffer_journal_test(bh); 39681da177e4SLinus Torvalds set_buffer_journal_restore_dirty(bh); 39691da177e4SLinus Torvalds } 39701da177e4SLinus Torvalds unlock_buffer(bh); 39711da177e4SLinus Torvalds return 1; 39721da177e4SLinus Torvalds } 39731da177e4SLinus Torvalds 39741da177e4SLinus Torvalds /* 3975098297b2SJeff Mahoney * long and ugly. If flush, will not return until all commit 3976098297b2SJeff Mahoney * blocks and all real buffers in the trans are on disk. 3977098297b2SJeff Mahoney * If no_async, won't return until all commit blocks are on disk. 3978098297b2SJeff Mahoney * 3979098297b2SJeff Mahoney * keep reading, there are comments as you go along 3980098297b2SJeff Mahoney * 3981098297b2SJeff Mahoney * If the journal is aborted, we just clean up. Things like flushing 3982098297b2SJeff Mahoney * journal lists, etc just won't happen. 39831da177e4SLinus Torvalds */ 398458d85426SJeff Mahoney static int do_journal_end(struct reiserfs_transaction_handle *th, int flags) 3985bd4c625cSLinus Torvalds { 398658d85426SJeff Mahoney struct super_block *sb = th->t_super; 3987a9dd3643SJeff Mahoney struct reiserfs_journal *journal = SB_JOURNAL(sb); 39881da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn, *next, *jl_cn; 39891da177e4SLinus Torvalds struct reiserfs_journal_cnode *last_cn = NULL; 39901da177e4SLinus Torvalds struct reiserfs_journal_desc *desc; 39911da177e4SLinus Torvalds struct reiserfs_journal_commit *commit; 39921da177e4SLinus Torvalds struct buffer_head *c_bh; /* commit bh */ 39931da177e4SLinus Torvalds struct buffer_head *d_bh; /* desc bh */ 39941da177e4SLinus Torvalds int cur_write_start = 0; /* start index of current log write */ 39951da177e4SLinus Torvalds int i; 3996a44c94a7SAlexander Zarochentsev int flush; 3997a44c94a7SAlexander Zarochentsev int wait_on_commit; 39981da177e4SLinus Torvalds struct reiserfs_journal_list *jl, *temp_jl; 39991da177e4SLinus Torvalds struct list_head *entry, *safe; 40001da177e4SLinus Torvalds unsigned long jindex; 4001600ed416SJeff Mahoney unsigned int commit_trans_id; 40021da177e4SLinus Torvalds int trans_half; 4003278f6679SJeff Mahoney int depth; 40041da177e4SLinus Torvalds 40051da177e4SLinus Torvalds BUG_ON(th->t_refcount > 1); 40061da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 400758d85426SJeff Mahoney BUG_ON(!th->t_super); 40081da177e4SLinus Torvalds 4009098297b2SJeff Mahoney /* 4010098297b2SJeff Mahoney * protect flush_older_commits from doing mistakes if the 4011098297b2SJeff Mahoney * transaction ID counter gets overflowed. 4012098297b2SJeff Mahoney */ 4013600ed416SJeff Mahoney if (th->t_trans_id == ~0U) 4014a44c94a7SAlexander Zarochentsev flags |= FLUSH_ALL | COMMIT_NOW | WAIT; 4015a44c94a7SAlexander Zarochentsev flush = flags & FLUSH_ALL; 4016a44c94a7SAlexander Zarochentsev wait_on_commit = flags & WAIT; 4017a44c94a7SAlexander Zarochentsev 40181da177e4SLinus Torvalds current->journal_info = th->t_handle_save; 4019a9dd3643SJeff Mahoney reiserfs_check_lock_depth(sb, "journal end"); 40201da177e4SLinus Torvalds if (journal->j_len == 0) { 4021a9dd3643SJeff Mahoney reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb), 4022bd4c625cSLinus Torvalds 1); 402309f1b80bSJeff Mahoney journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb)); 40241da177e4SLinus Torvalds } 40251da177e4SLinus Torvalds 4026a9dd3643SJeff Mahoney lock_journal(sb); 40271da177e4SLinus Torvalds if (journal->j_next_full_flush) { 40281da177e4SLinus Torvalds flags |= FLUSH_ALL; 40291da177e4SLinus Torvalds flush = 1; 40301da177e4SLinus Torvalds } 40311da177e4SLinus Torvalds if (journal->j_next_async_flush) { 40321da177e4SLinus Torvalds flags |= COMMIT_NOW | WAIT; 40331da177e4SLinus Torvalds wait_on_commit = 1; 40341da177e4SLinus Torvalds } 40351da177e4SLinus Torvalds 4036098297b2SJeff Mahoney /* 4037098297b2SJeff Mahoney * check_journal_end locks the journal, and unlocks if it does 4038098297b2SJeff Mahoney * not return 1 it tells us if we should continue with the 4039098297b2SJeff Mahoney * journal_end, or just return 40401da177e4SLinus Torvalds */ 404158d85426SJeff Mahoney if (!check_journal_end(th, flags)) { 4042033369d1SArtem Bityutskiy reiserfs_schedule_old_flush(sb); 4043a9dd3643SJeff Mahoney wake_queued_writers(sb); 4044a9dd3643SJeff Mahoney reiserfs_async_progress_wait(sb); 40451da177e4SLinus Torvalds goto out; 40461da177e4SLinus Torvalds } 40471da177e4SLinus Torvalds 40481da177e4SLinus Torvalds /* check_journal_end might set these, check again */ 40491da177e4SLinus Torvalds if (journal->j_next_full_flush) { 40501da177e4SLinus Torvalds flush = 1; 40511da177e4SLinus Torvalds } 40521da177e4SLinus Torvalds 40531da177e4SLinus Torvalds /* 4054098297b2SJeff Mahoney * j must wait means we have to flush the log blocks, and the 4055098297b2SJeff Mahoney * real blocks for this transaction 40561da177e4SLinus Torvalds */ 40571da177e4SLinus Torvalds if (journal->j_must_wait > 0) { 40581da177e4SLinus Torvalds flush = 1; 40591da177e4SLinus Torvalds } 40601da177e4SLinus Torvalds #ifdef REISERFS_PREALLOCATE 4061098297b2SJeff Mahoney /* 4062098297b2SJeff Mahoney * quota ops might need to nest, setup the journal_info pointer 4063098297b2SJeff Mahoney * for them and raise the refcount so that it is > 0. 4064098297b2SJeff Mahoney */ 40651da177e4SLinus Torvalds current->journal_info = th; 4066ef43bc4fSJan Kara th->t_refcount++; 4067098297b2SJeff Mahoney 4068098297b2SJeff Mahoney /* it should not involve new blocks into the transaction */ 4069098297b2SJeff Mahoney reiserfs_discard_all_prealloc(th); 4070098297b2SJeff Mahoney 4071ef43bc4fSJan Kara th->t_refcount--; 40721da177e4SLinus Torvalds current->journal_info = th->t_handle_save; 40731da177e4SLinus Torvalds #endif 40741da177e4SLinus Torvalds 40751da177e4SLinus Torvalds /* setup description block */ 4076bd4c625cSLinus Torvalds d_bh = 4077a9dd3643SJeff Mahoney journal_getblk(sb, 4078a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 4079bd4c625cSLinus Torvalds journal->j_start); 40801da177e4SLinus Torvalds set_buffer_uptodate(d_bh); 40811da177e4SLinus Torvalds desc = (struct reiserfs_journal_desc *)(d_bh)->b_data; 40821da177e4SLinus Torvalds memset(d_bh->b_data, 0, d_bh->b_size); 40831da177e4SLinus Torvalds memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8); 40841da177e4SLinus Torvalds set_desc_trans_id(desc, journal->j_trans_id); 40851da177e4SLinus Torvalds 4086098297b2SJeff Mahoney /* 4087098297b2SJeff Mahoney * setup commit block. Don't write (keep it clean too) this one 4088098297b2SJeff Mahoney * until after everyone else is written 4089098297b2SJeff Mahoney */ 4090a9dd3643SJeff Mahoney c_bh = journal_getblk(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 4091bd4c625cSLinus Torvalds ((journal->j_start + journal->j_len + 4092a9dd3643SJeff Mahoney 1) % SB_ONDISK_JOURNAL_SIZE(sb))); 40931da177e4SLinus Torvalds commit = (struct reiserfs_journal_commit *)c_bh->b_data; 40941da177e4SLinus Torvalds memset(c_bh->b_data, 0, c_bh->b_size); 40951da177e4SLinus Torvalds set_commit_trans_id(commit, journal->j_trans_id); 40961da177e4SLinus Torvalds set_buffer_uptodate(c_bh); 40971da177e4SLinus Torvalds 40981da177e4SLinus Torvalds /* init this journal list */ 40991da177e4SLinus Torvalds jl = journal->j_current_jl; 41001da177e4SLinus Torvalds 4101098297b2SJeff Mahoney /* 4102098297b2SJeff Mahoney * we lock the commit before doing anything because 41031da177e4SLinus Torvalds * we want to make sure nobody tries to run flush_commit_list until 41041da177e4SLinus Torvalds * the new transaction is fully setup, and we've already flushed the 41051da177e4SLinus Torvalds * ordered bh list 41061da177e4SLinus Torvalds */ 41078ebc4232SFrederic Weisbecker reiserfs_mutex_lock_safe(&jl->j_commit_mutex, sb); 41081da177e4SLinus Torvalds 41091da177e4SLinus Torvalds /* save the transaction id in case we need to commit it later */ 41101da177e4SLinus Torvalds commit_trans_id = jl->j_trans_id; 41111da177e4SLinus Torvalds 41121da177e4SLinus Torvalds atomic_set(&jl->j_older_commits_done, 0); 41131da177e4SLinus Torvalds jl->j_trans_id = journal->j_trans_id; 41141da177e4SLinus Torvalds jl->j_timestamp = journal->j_trans_start_time; 41151da177e4SLinus Torvalds jl->j_commit_bh = c_bh; 41161da177e4SLinus Torvalds jl->j_start = journal->j_start; 41171da177e4SLinus Torvalds jl->j_len = journal->j_len; 41181da177e4SLinus Torvalds atomic_set(&jl->j_nonzerolen, journal->j_len); 41191da177e4SLinus Torvalds atomic_set(&jl->j_commit_left, journal->j_len + 2); 41201da177e4SLinus Torvalds jl->j_realblock = NULL; 41211da177e4SLinus Torvalds 4122098297b2SJeff Mahoney /* 4123098297b2SJeff Mahoney * The ENTIRE FOR LOOP MUST not cause schedule to occur. 4124098297b2SJeff Mahoney * for each real block, add it to the journal list hash, 4125098297b2SJeff Mahoney * copy into real block index array in the commit or desc block 41261da177e4SLinus Torvalds */ 4127a9dd3643SJeff Mahoney trans_half = journal_trans_half(sb->s_blocksize); 41281da177e4SLinus Torvalds for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) { 41291da177e4SLinus Torvalds if (buffer_journaled(cn->bh)) { 4130a9dd3643SJeff Mahoney jl_cn = get_cnode(sb); 41311da177e4SLinus Torvalds if (!jl_cn) { 4132a9dd3643SJeff Mahoney reiserfs_panic(sb, "journal-1676", 4133c3a9c210SJeff Mahoney "get_cnode returned NULL"); 41341da177e4SLinus Torvalds } 41351da177e4SLinus Torvalds if (i == 0) { 41361da177e4SLinus Torvalds jl->j_realblock = jl_cn; 41371da177e4SLinus Torvalds } 41381da177e4SLinus Torvalds jl_cn->prev = last_cn; 41391da177e4SLinus Torvalds jl_cn->next = NULL; 41401da177e4SLinus Torvalds if (last_cn) { 41411da177e4SLinus Torvalds last_cn->next = jl_cn; 41421da177e4SLinus Torvalds } 41431da177e4SLinus Torvalds last_cn = jl_cn; 4144098297b2SJeff Mahoney /* 4145098297b2SJeff Mahoney * make sure the block we are trying to log 4146098297b2SJeff Mahoney * is not a block of journal or reserved area 4147098297b2SJeff Mahoney */ 4148bd4c625cSLinus Torvalds if (is_block_in_log_or_reserved_area 4149a9dd3643SJeff Mahoney (sb, cn->bh->b_blocknr)) { 4150a9dd3643SJeff Mahoney reiserfs_panic(sb, "journal-2332", 4151c3a9c210SJeff Mahoney "Trying to log block %lu, " 4152c3a9c210SJeff Mahoney "which is a log block", 4153bd4c625cSLinus Torvalds cn->bh->b_blocknr); 41541da177e4SLinus Torvalds } 41551da177e4SLinus Torvalds jl_cn->blocknr = cn->bh->b_blocknr; 41561da177e4SLinus Torvalds jl_cn->state = 0; 4157a9dd3643SJeff Mahoney jl_cn->sb = sb; 41581da177e4SLinus Torvalds jl_cn->bh = cn->bh; 41591da177e4SLinus Torvalds jl_cn->jlist = jl; 41601da177e4SLinus Torvalds insert_journal_hash(journal->j_list_hash_table, jl_cn); 41611da177e4SLinus Torvalds if (i < trans_half) { 4162bd4c625cSLinus Torvalds desc->j_realblock[i] = 4163bd4c625cSLinus Torvalds cpu_to_le32(cn->bh->b_blocknr); 41641da177e4SLinus Torvalds } else { 4165bd4c625cSLinus Torvalds commit->j_realblock[i - trans_half] = 4166bd4c625cSLinus Torvalds cpu_to_le32(cn->bh->b_blocknr); 41671da177e4SLinus Torvalds } 41681da177e4SLinus Torvalds } else { 41691da177e4SLinus Torvalds i--; 41701da177e4SLinus Torvalds } 41711da177e4SLinus Torvalds } 41721da177e4SLinus Torvalds set_desc_trans_len(desc, journal->j_len); 41731da177e4SLinus Torvalds set_desc_mount_id(desc, journal->j_mount_id); 41741da177e4SLinus Torvalds set_desc_trans_id(desc, journal->j_trans_id); 41751da177e4SLinus Torvalds set_commit_trans_len(commit, journal->j_len); 41761da177e4SLinus Torvalds 4177098297b2SJeff Mahoney /* 4178098297b2SJeff Mahoney * special check in case all buffers in the journal 4179098297b2SJeff Mahoney * were marked for not logging 4180098297b2SJeff Mahoney */ 418114a61442SEric Sesterhenn BUG_ON(journal->j_len == 0); 41821da177e4SLinus Torvalds 4183098297b2SJeff Mahoney /* 4184098297b2SJeff Mahoney * we're about to dirty all the log blocks, mark the description block 41851da177e4SLinus Torvalds * dirty now too. Don't mark the commit block dirty until all the 41861da177e4SLinus Torvalds * others are on disk 41871da177e4SLinus Torvalds */ 41881da177e4SLinus Torvalds mark_buffer_dirty(d_bh); 41891da177e4SLinus Torvalds 4190098297b2SJeff Mahoney /* 4191098297b2SJeff Mahoney * first data block is j_start + 1, so add one to 4192098297b2SJeff Mahoney * cur_write_start wherever you use it 4193098297b2SJeff Mahoney */ 41941da177e4SLinus Torvalds cur_write_start = journal->j_start; 41951da177e4SLinus Torvalds cn = journal->j_first; 41961da177e4SLinus Torvalds jindex = 1; /* start at one so we don't get the desc again */ 41971da177e4SLinus Torvalds while (cn) { 41981da177e4SLinus Torvalds clear_buffer_journal_new(cn->bh); 41991da177e4SLinus Torvalds /* copy all the real blocks into log area. dirty log blocks */ 42001da177e4SLinus Torvalds if (buffer_journaled(cn->bh)) { 42011da177e4SLinus Torvalds struct buffer_head *tmp_bh; 42021da177e4SLinus Torvalds char *addr; 42031da177e4SLinus Torvalds struct page *page; 4204bd4c625cSLinus Torvalds tmp_bh = 4205a9dd3643SJeff Mahoney journal_getblk(sb, 4206a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_1st_BLOCK(sb) + 4207bd4c625cSLinus Torvalds ((cur_write_start + 4208bd4c625cSLinus Torvalds jindex) % 4209a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb))); 42101da177e4SLinus Torvalds set_buffer_uptodate(tmp_bh); 42111da177e4SLinus Torvalds page = cn->bh->b_page; 42121da177e4SLinus Torvalds addr = kmap(page); 4213bd4c625cSLinus Torvalds memcpy(tmp_bh->b_data, 4214bd4c625cSLinus Torvalds addr + offset_in_page(cn->bh->b_data), 42151da177e4SLinus Torvalds cn->bh->b_size); 42161da177e4SLinus Torvalds kunmap(page); 42171da177e4SLinus Torvalds mark_buffer_dirty(tmp_bh); 42181da177e4SLinus Torvalds jindex++; 42191da177e4SLinus Torvalds set_buffer_journal_dirty(cn->bh); 42201da177e4SLinus Torvalds clear_buffer_journaled(cn->bh); 42211da177e4SLinus Torvalds } else { 4222098297b2SJeff Mahoney /* 4223098297b2SJeff Mahoney * JDirty cleared sometime during transaction. 4224098297b2SJeff Mahoney * don't log this one 4225098297b2SJeff Mahoney */ 4226a9dd3643SJeff Mahoney reiserfs_warning(sb, "journal-2048", 422745b03d5eSJeff Mahoney "BAD, buffer in journal hash, " 422845b03d5eSJeff Mahoney "but not JDirty!"); 42291da177e4SLinus Torvalds brelse(cn->bh); 42301da177e4SLinus Torvalds } 42311da177e4SLinus Torvalds next = cn->next; 4232a9dd3643SJeff Mahoney free_cnode(sb, cn); 42331da177e4SLinus Torvalds cn = next; 4234278f6679SJeff Mahoney reiserfs_cond_resched(sb); 42351da177e4SLinus Torvalds } 42361da177e4SLinus Torvalds 4237098297b2SJeff Mahoney /* 4238098297b2SJeff Mahoney * we are done with both the c_bh and d_bh, but 4239098297b2SJeff Mahoney * c_bh must be written after all other commit blocks, 4240098297b2SJeff Mahoney * so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1. 42411da177e4SLinus Torvalds */ 42421da177e4SLinus Torvalds 4243a9dd3643SJeff Mahoney journal->j_current_jl = alloc_journal_list(sb); 42441da177e4SLinus Torvalds 42451da177e4SLinus Torvalds /* now it is safe to insert this transaction on the main list */ 42461da177e4SLinus Torvalds list_add_tail(&jl->j_list, &journal->j_journal_list); 42471da177e4SLinus Torvalds list_add_tail(&jl->j_working_list, &journal->j_working_list); 42481da177e4SLinus Torvalds journal->j_num_work_lists++; 42491da177e4SLinus Torvalds 42501da177e4SLinus Torvalds /* reset journal values for the next transaction */ 4251bd4c625cSLinus Torvalds journal->j_start = 4252bd4c625cSLinus Torvalds (journal->j_start + journal->j_len + 4253a9dd3643SJeff Mahoney 2) % SB_ONDISK_JOURNAL_SIZE(sb); 4254a228bf8fSJeff Mahoney atomic_set(&journal->j_wcount, 0); 42551da177e4SLinus Torvalds journal->j_bcount = 0; 42561da177e4SLinus Torvalds journal->j_last = NULL; 42571da177e4SLinus Torvalds journal->j_first = NULL; 42581da177e4SLinus Torvalds journal->j_len = 0; 42591da177e4SLinus Torvalds journal->j_trans_start_time = 0; 4260a44c94a7SAlexander Zarochentsev /* check for trans_id overflow */ 4261a44c94a7SAlexander Zarochentsev if (++journal->j_trans_id == 0) 4262a44c94a7SAlexander Zarochentsev journal->j_trans_id = 10; 42631da177e4SLinus Torvalds journal->j_current_jl->j_trans_id = journal->j_trans_id; 42641da177e4SLinus Torvalds journal->j_must_wait = 0; 42651da177e4SLinus Torvalds journal->j_len_alloc = 0; 42661da177e4SLinus Torvalds journal->j_next_full_flush = 0; 42671da177e4SLinus Torvalds journal->j_next_async_flush = 0; 4268a9dd3643SJeff Mahoney init_journal_hash(sb); 42691da177e4SLinus Torvalds 4270098297b2SJeff Mahoney /* 4271098297b2SJeff Mahoney * make sure reiserfs_add_jh sees the new current_jl before we 4272098297b2SJeff Mahoney * write out the tails 4273098297b2SJeff Mahoney */ 42741da177e4SLinus Torvalds smp_mb(); 42751da177e4SLinus Torvalds 4276098297b2SJeff Mahoney /* 4277098297b2SJeff Mahoney * tail conversion targets have to hit the disk before we end the 42781da177e4SLinus Torvalds * transaction. Otherwise a later transaction might repack the tail 4279098297b2SJeff Mahoney * before this transaction commits, leaving the data block unflushed 4280098297b2SJeff Mahoney * and clean, if we crash before the later transaction commits, the 4281098297b2SJeff Mahoney * data block is lost. 42821da177e4SLinus Torvalds */ 42831da177e4SLinus Torvalds if (!list_empty(&jl->j_tail_bh_list)) { 4284278f6679SJeff Mahoney depth = reiserfs_write_unlock_nested(sb); 42851da177e4SLinus Torvalds write_ordered_buffers(&journal->j_dirty_buffers_lock, 42861da177e4SLinus Torvalds journal, jl, &jl->j_tail_bh_list); 4287278f6679SJeff Mahoney reiserfs_write_lock_nested(sb, depth); 42881da177e4SLinus Torvalds } 428914a61442SEric Sesterhenn BUG_ON(!list_empty(&jl->j_tail_bh_list)); 429090415deaSJeff Mahoney mutex_unlock(&jl->j_commit_mutex); 42911da177e4SLinus Torvalds 4292098297b2SJeff Mahoney /* 4293098297b2SJeff Mahoney * honor the flush wishes from the caller, simple commits can 4294098297b2SJeff Mahoney * be done outside the journal lock, they are done below 4295098297b2SJeff Mahoney * 4296098297b2SJeff Mahoney * if we don't flush the commit list right now, we put it into 4297098297b2SJeff Mahoney * the work queue so the people waiting on the async progress work 4298098297b2SJeff Mahoney * queue don't wait for this proc to flush journal lists and such. 42991da177e4SLinus Torvalds */ 43001da177e4SLinus Torvalds if (flush) { 4301a9dd3643SJeff Mahoney flush_commit_list(sb, jl, 1); 4302a9dd3643SJeff Mahoney flush_journal_list(sb, jl, 1); 430301777836SJan Kara } else if (!(jl->j_state & LIST_COMMIT_PENDING)) { 430401777836SJan Kara /* 430501777836SJan Kara * Avoid queueing work when sb is being shut down. Transaction 430601777836SJan Kara * will be flushed on journal shutdown. 430701777836SJan Kara */ 43081751e8a6SLinus Torvalds if (sb->s_flags & SB_ACTIVE) 4309797d9016SJeff Mahoney queue_delayed_work(REISERFS_SB(sb)->commit_wq, 4310797d9016SJeff Mahoney &journal->j_work, HZ / 10); 431101777836SJan Kara } 43121da177e4SLinus Torvalds 4313098297b2SJeff Mahoney /* 4314098297b2SJeff Mahoney * if the next transaction has any chance of wrapping, flush 4315098297b2SJeff Mahoney * transactions that might get overwritten. If any journal lists 4316098297b2SJeff Mahoney * are very old flush them as well. 43171da177e4SLinus Torvalds */ 43181da177e4SLinus Torvalds first_jl: 43191da177e4SLinus Torvalds list_for_each_safe(entry, safe, &journal->j_journal_list) { 43201da177e4SLinus Torvalds temp_jl = JOURNAL_LIST_ENTRY(entry); 43211da177e4SLinus Torvalds if (journal->j_start <= temp_jl->j_start) { 43221da177e4SLinus Torvalds if ((journal->j_start + journal->j_trans_max + 1) >= 4323bd4c625cSLinus Torvalds temp_jl->j_start) { 4324a9dd3643SJeff Mahoney flush_used_journal_lists(sb, temp_jl); 43251da177e4SLinus Torvalds goto first_jl; 43261da177e4SLinus Torvalds } else if ((journal->j_start + 43271da177e4SLinus Torvalds journal->j_trans_max + 1) < 4328a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb)) { 4329098297b2SJeff Mahoney /* 4330098297b2SJeff Mahoney * if we don't cross into the next 4331098297b2SJeff Mahoney * transaction and we don't wrap, there is 4332098297b2SJeff Mahoney * no way we can overlap any later transactions 43331da177e4SLinus Torvalds * break now 43341da177e4SLinus Torvalds */ 43351da177e4SLinus Torvalds break; 43361da177e4SLinus Torvalds } 43371da177e4SLinus Torvalds } else if ((journal->j_start + 43381da177e4SLinus Torvalds journal->j_trans_max + 1) > 4339a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb)) { 43401da177e4SLinus Torvalds if (((journal->j_start + journal->j_trans_max + 1) % 4341a9dd3643SJeff Mahoney SB_ONDISK_JOURNAL_SIZE(sb)) >= 4342bd4c625cSLinus Torvalds temp_jl->j_start) { 4343a9dd3643SJeff Mahoney flush_used_journal_lists(sb, temp_jl); 43441da177e4SLinus Torvalds goto first_jl; 43451da177e4SLinus Torvalds } else { 4346098297b2SJeff Mahoney /* 4347098297b2SJeff Mahoney * we don't overlap anything from out start 4348098297b2SJeff Mahoney * to the end of the log, and our wrapped 4349098297b2SJeff Mahoney * portion doesn't overlap anything at 43501da177e4SLinus Torvalds * the start of the log. We can break 43511da177e4SLinus Torvalds */ 43521da177e4SLinus Torvalds break; 43531da177e4SLinus Torvalds } 43541da177e4SLinus Torvalds } 43551da177e4SLinus Torvalds } 43561da177e4SLinus Torvalds 4357bd4c625cSLinus Torvalds journal->j_current_jl->j_list_bitmap = 4358a9dd3643SJeff Mahoney get_list_bitmap(sb, journal->j_current_jl); 43591da177e4SLinus Torvalds 43601da177e4SLinus Torvalds if (!(journal->j_current_jl->j_list_bitmap)) { 4361a9dd3643SJeff Mahoney reiserfs_panic(sb, "journal-1996", 4362c3a9c210SJeff Mahoney "could not get a list bitmap"); 43631da177e4SLinus Torvalds } 43641da177e4SLinus Torvalds 4365a228bf8fSJeff Mahoney atomic_set(&journal->j_jlock, 0); 4366a9dd3643SJeff Mahoney unlock_journal(sb); 43671da177e4SLinus Torvalds /* wake up any body waiting to join. */ 43681da177e4SLinus Torvalds clear_bit(J_WRITERS_QUEUED, &journal->j_state); 4369a228bf8fSJeff Mahoney wake_up(&journal->j_join_wait); 43701da177e4SLinus Torvalds 43711da177e4SLinus Torvalds if (!flush && wait_on_commit && 4372a9dd3643SJeff Mahoney journal_list_still_alive(sb, commit_trans_id)) { 4373a9dd3643SJeff Mahoney flush_commit_list(sb, jl, 1); 43741da177e4SLinus Torvalds } 43751da177e4SLinus Torvalds out: 4376a9dd3643SJeff Mahoney reiserfs_check_lock_depth(sb, "journal end2"); 43771da177e4SLinus Torvalds 43781da177e4SLinus Torvalds memset(th, 0, sizeof(*th)); 4379098297b2SJeff Mahoney /* 4380098297b2SJeff Mahoney * Re-set th->t_super, so we can properly keep track of how many 43811da177e4SLinus Torvalds * persistent transactions there are. We need to do this so if this 4382098297b2SJeff Mahoney * call is part of a failed restart_transaction, we can free it later 4383098297b2SJeff Mahoney */ 4384a9dd3643SJeff Mahoney th->t_super = sb; 43851da177e4SLinus Torvalds 43861da177e4SLinus Torvalds return journal->j_errno; 43871da177e4SLinus Torvalds } 43881da177e4SLinus Torvalds 438932e8b106SJeff Mahoney /* Send the file system read only and refuse new transactions */ 439032e8b106SJeff Mahoney void reiserfs_abort_journal(struct super_block *sb, int errno) 43911da177e4SLinus Torvalds { 43921da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(sb); 43931da177e4SLinus Torvalds if (test_bit(J_ABORTED, &journal->j_state)) 43941da177e4SLinus Torvalds return; 43951da177e4SLinus Torvalds 439632e8b106SJeff Mahoney if (!journal->j_errno) 439732e8b106SJeff Mahoney journal->j_errno = errno; 43981da177e4SLinus Torvalds 43991751e8a6SLinus Torvalds sb->s_flags |= SB_RDONLY; 44001da177e4SLinus Torvalds set_bit(J_ABORTED, &journal->j_state); 44011da177e4SLinus Torvalds 44021da177e4SLinus Torvalds #ifdef CONFIG_REISERFS_CHECK 44031da177e4SLinus Torvalds dump_stack(); 44041da177e4SLinus Torvalds #endif 44051da177e4SLinus Torvalds } 4406