11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds ** Write ahead logging implementation copyright Chris Mason 2000 31da177e4SLinus Torvalds ** 41da177e4SLinus Torvalds ** The background commits make this code very interelated, and 51da177e4SLinus Torvalds ** overly complex. I need to rethink things a bit....The major players: 61da177e4SLinus Torvalds ** 71da177e4SLinus Torvalds ** journal_begin -- call with the number of blocks you expect to log. 81da177e4SLinus Torvalds ** If the current transaction is too 91da177e4SLinus Torvalds ** old, it will block until the current transaction is 101da177e4SLinus Torvalds ** finished, and then start a new one. 111da177e4SLinus Torvalds ** Usually, your transaction will get joined in with 121da177e4SLinus Torvalds ** previous ones for speed. 131da177e4SLinus Torvalds ** 141da177e4SLinus Torvalds ** journal_join -- same as journal_begin, but won't block on the current 151da177e4SLinus Torvalds ** transaction regardless of age. Don't ever call 161da177e4SLinus Torvalds ** this. Ever. There are only two places it should be 171da177e4SLinus Torvalds ** called from, and they are both inside this file. 181da177e4SLinus Torvalds ** 191da177e4SLinus Torvalds ** journal_mark_dirty -- adds blocks into this transaction. clears any flags 201da177e4SLinus Torvalds ** that might make them get sent to disk 211da177e4SLinus Torvalds ** and then marks them BH_JDirty. Puts the buffer head 221da177e4SLinus Torvalds ** into the current transaction hash. 231da177e4SLinus Torvalds ** 241da177e4SLinus Torvalds ** journal_end -- if the current transaction is batchable, it does nothing 251da177e4SLinus Torvalds ** otherwise, it could do an async/synchronous commit, or 261da177e4SLinus Torvalds ** a full flush of all log and real blocks in the 271da177e4SLinus Torvalds ** transaction. 281da177e4SLinus Torvalds ** 291da177e4SLinus Torvalds ** flush_old_commits -- if the current transaction is too old, it is ended and 301da177e4SLinus Torvalds ** commit blocks are sent to disk. Forces commit blocks 311da177e4SLinus Torvalds ** to disk for all backgrounded commits that have been 321da177e4SLinus Torvalds ** around too long. 331da177e4SLinus Torvalds ** -- Note, if you call this as an immediate flush from 341da177e4SLinus Torvalds ** from within kupdate, it will ignore the immediate flag 351da177e4SLinus Torvalds */ 361da177e4SLinus Torvalds 371da177e4SLinus Torvalds #include <asm/uaccess.h> 381da177e4SLinus Torvalds #include <asm/system.h> 391da177e4SLinus Torvalds 401da177e4SLinus Torvalds #include <linux/time.h> 411da177e4SLinus Torvalds #include <asm/semaphore.h> 421da177e4SLinus Torvalds 431da177e4SLinus Torvalds #include <linux/vmalloc.h> 441da177e4SLinus Torvalds #include <linux/reiserfs_fs.h> 451da177e4SLinus Torvalds 461da177e4SLinus Torvalds #include <linux/kernel.h> 471da177e4SLinus Torvalds #include <linux/errno.h> 481da177e4SLinus Torvalds #include <linux/fcntl.h> 491da177e4SLinus Torvalds #include <linux/stat.h> 501da177e4SLinus Torvalds #include <linux/string.h> 511da177e4SLinus Torvalds #include <linux/smp_lock.h> 521da177e4SLinus Torvalds #include <linux/buffer_head.h> 531da177e4SLinus Torvalds #include <linux/workqueue.h> 541da177e4SLinus Torvalds #include <linux/writeback.h> 551da177e4SLinus Torvalds #include <linux/blkdev.h> 563fcfab16SAndrew Morton #include <linux/backing-dev.h> 571da177e4SLinus Torvalds 581da177e4SLinus Torvalds /* gets a struct reiserfs_journal_list * from a list head */ 591da177e4SLinus Torvalds #define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \ 601da177e4SLinus Torvalds j_list)) 611da177e4SLinus Torvalds #define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \ 621da177e4SLinus Torvalds j_working_list)) 631da177e4SLinus Torvalds 641da177e4SLinus Torvalds /* the number of mounted filesystems. This is used to decide when to 651da177e4SLinus Torvalds ** start and kill the commit workqueue 661da177e4SLinus Torvalds */ 671da177e4SLinus Torvalds static int reiserfs_mounted_fs_count; 681da177e4SLinus Torvalds 691da177e4SLinus Torvalds static struct workqueue_struct *commit_wq; 701da177e4SLinus Torvalds 711da177e4SLinus Torvalds #define JOURNAL_TRANS_HALF 1018 /* must be correct to keep the desc and commit 721da177e4SLinus Torvalds structs at 4k */ 731da177e4SLinus Torvalds #define BUFNR 64 /*read ahead */ 741da177e4SLinus Torvalds 751da177e4SLinus Torvalds /* cnode stat bits. Move these into reiserfs_fs.h */ 761da177e4SLinus Torvalds 771da177e4SLinus Torvalds #define BLOCK_FREED 2 /* this block was freed, and can't be written. */ 781da177e4SLinus Torvalds #define BLOCK_FREED_HOLDER 3 /* this block was freed during this transaction, and can't be written */ 791da177e4SLinus Torvalds 801da177e4SLinus Torvalds #define BLOCK_NEEDS_FLUSH 4 /* used in flush_journal_list */ 811da177e4SLinus Torvalds #define BLOCK_DIRTIED 5 821da177e4SLinus Torvalds 831da177e4SLinus Torvalds /* journal list state bits */ 841da177e4SLinus Torvalds #define LIST_TOUCHED 1 851da177e4SLinus Torvalds #define LIST_DIRTY 2 861da177e4SLinus Torvalds #define LIST_COMMIT_PENDING 4 /* someone will commit this list */ 871da177e4SLinus Torvalds 881da177e4SLinus Torvalds /* flags for do_journal_end */ 891da177e4SLinus Torvalds #define FLUSH_ALL 1 /* flush commit and real blocks */ 901da177e4SLinus Torvalds #define COMMIT_NOW 2 /* end and commit this transaction */ 911da177e4SLinus Torvalds #define WAIT 4 /* wait for the log blocks to hit the disk */ 921da177e4SLinus Torvalds 93bd4c625cSLinus Torvalds static int do_journal_end(struct reiserfs_transaction_handle *, 94bd4c625cSLinus Torvalds struct super_block *, unsigned long nblocks, 95bd4c625cSLinus Torvalds int flags); 96bd4c625cSLinus Torvalds static int flush_journal_list(struct super_block *s, 97bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl, int flushall); 98bd4c625cSLinus Torvalds static int flush_commit_list(struct super_block *s, 99bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl, int flushall); 1001da177e4SLinus Torvalds static int can_dirty(struct reiserfs_journal_cnode *cn); 101bd4c625cSLinus Torvalds static int journal_join(struct reiserfs_transaction_handle *th, 102bd4c625cSLinus Torvalds struct super_block *p_s_sb, unsigned long nblocks); 1031da177e4SLinus Torvalds static int release_journal_dev(struct super_block *super, 1041da177e4SLinus Torvalds struct reiserfs_journal *journal); 1051da177e4SLinus Torvalds static int dirty_one_transaction(struct super_block *s, 1061da177e4SLinus Torvalds struct reiserfs_journal_list *jl); 107c4028958SDavid Howells static void flush_async_commits(struct work_struct *work); 1081da177e4SLinus Torvalds static void queue_log_writer(struct super_block *s); 1091da177e4SLinus Torvalds 1101da177e4SLinus Torvalds /* values for join in do_journal_begin_r */ 1111da177e4SLinus Torvalds enum { 1121da177e4SLinus Torvalds JBEGIN_REG = 0, /* regular journal begin */ 1131da177e4SLinus Torvalds JBEGIN_JOIN = 1, /* join the running transaction if at all possible */ 1141da177e4SLinus Torvalds JBEGIN_ABORT = 2, /* called from cleanup code, ignores aborted flag */ 1151da177e4SLinus Torvalds }; 1161da177e4SLinus Torvalds 1171da177e4SLinus Torvalds static int do_journal_begin_r(struct reiserfs_transaction_handle *th, 1181da177e4SLinus Torvalds struct super_block *p_s_sb, 1191da177e4SLinus Torvalds unsigned long nblocks, int join); 1201da177e4SLinus Torvalds 121bd4c625cSLinus Torvalds static void init_journal_hash(struct super_block *p_s_sb) 122bd4c625cSLinus Torvalds { 1231da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 124bd4c625cSLinus Torvalds memset(journal->j_hash_table, 0, 125bd4c625cSLinus Torvalds JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *)); 1261da177e4SLinus Torvalds } 1271da177e4SLinus Torvalds 1281da177e4SLinus Torvalds /* 1291da177e4SLinus Torvalds ** clears BH_Dirty and sticks the buffer on the clean list. Called because I can't allow refile_buffer to 1301da177e4SLinus Torvalds ** make schedule happen after I've freed a block. Look at remove_from_transaction and journal_mark_freed for 1311da177e4SLinus Torvalds ** more details. 1321da177e4SLinus Torvalds */ 133bd4c625cSLinus Torvalds static int reiserfs_clean_and_file_buffer(struct buffer_head *bh) 134bd4c625cSLinus Torvalds { 1351da177e4SLinus Torvalds if (bh) { 1361da177e4SLinus Torvalds clear_buffer_dirty(bh); 1371da177e4SLinus Torvalds clear_buffer_journal_test(bh); 1381da177e4SLinus Torvalds } 1391da177e4SLinus Torvalds return 0; 1401da177e4SLinus Torvalds } 1411da177e4SLinus Torvalds 1421da177e4SLinus Torvalds static void disable_barrier(struct super_block *s) 1431da177e4SLinus Torvalds { 1441da177e4SLinus Torvalds REISERFS_SB(s)->s_mount_opt &= ~(1 << REISERFS_BARRIER_FLUSH); 145bd4c625cSLinus Torvalds printk("reiserfs: disabling flush barriers on %s\n", 146bd4c625cSLinus Torvalds reiserfs_bdevname(s)); 1471da177e4SLinus Torvalds } 1481da177e4SLinus Torvalds 149bd4c625cSLinus Torvalds static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block 150bd4c625cSLinus Torvalds *p_s_sb) 151bd4c625cSLinus Torvalds { 1521da177e4SLinus Torvalds struct reiserfs_bitmap_node *bn; 1531da177e4SLinus Torvalds static int id; 1541da177e4SLinus Torvalds 155d739b42bSPekka Enberg bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS); 1561da177e4SLinus Torvalds if (!bn) { 1571da177e4SLinus Torvalds return NULL; 1581da177e4SLinus Torvalds } 159d739b42bSPekka Enberg bn->data = kzalloc(p_s_sb->s_blocksize, GFP_NOFS); 1601da177e4SLinus Torvalds if (!bn->data) { 161d739b42bSPekka Enberg kfree(bn); 1621da177e4SLinus Torvalds return NULL; 1631da177e4SLinus Torvalds } 1641da177e4SLinus Torvalds bn->id = id++; 1651da177e4SLinus Torvalds INIT_LIST_HEAD(&bn->list); 1661da177e4SLinus Torvalds return bn; 1671da177e4SLinus Torvalds } 1681da177e4SLinus Torvalds 169bd4c625cSLinus Torvalds static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *p_s_sb) 170bd4c625cSLinus Torvalds { 1711da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 1721da177e4SLinus Torvalds struct reiserfs_bitmap_node *bn = NULL; 1731da177e4SLinus Torvalds struct list_head *entry = journal->j_bitmap_nodes.next; 1741da177e4SLinus Torvalds 1751da177e4SLinus Torvalds journal->j_used_bitmap_nodes++; 1761da177e4SLinus Torvalds repeat: 1771da177e4SLinus Torvalds 1781da177e4SLinus Torvalds if (entry != &journal->j_bitmap_nodes) { 1791da177e4SLinus Torvalds bn = list_entry(entry, struct reiserfs_bitmap_node, list); 1801da177e4SLinus Torvalds list_del(entry); 1811da177e4SLinus Torvalds memset(bn->data, 0, p_s_sb->s_blocksize); 1821da177e4SLinus Torvalds journal->j_free_bitmap_nodes--; 1831da177e4SLinus Torvalds return bn; 1841da177e4SLinus Torvalds } 1851da177e4SLinus Torvalds bn = allocate_bitmap_node(p_s_sb); 1861da177e4SLinus Torvalds if (!bn) { 1871da177e4SLinus Torvalds yield(); 1881da177e4SLinus Torvalds goto repeat; 1891da177e4SLinus Torvalds } 1901da177e4SLinus Torvalds return bn; 1911da177e4SLinus Torvalds } 1921da177e4SLinus Torvalds static inline void free_bitmap_node(struct super_block *p_s_sb, 193bd4c625cSLinus Torvalds struct reiserfs_bitmap_node *bn) 194bd4c625cSLinus Torvalds { 1951da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 1961da177e4SLinus Torvalds journal->j_used_bitmap_nodes--; 1971da177e4SLinus Torvalds if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) { 198d739b42bSPekka Enberg kfree(bn->data); 199d739b42bSPekka Enberg kfree(bn); 2001da177e4SLinus Torvalds } else { 2011da177e4SLinus Torvalds list_add(&bn->list, &journal->j_bitmap_nodes); 2021da177e4SLinus Torvalds journal->j_free_bitmap_nodes++; 2031da177e4SLinus Torvalds } 2041da177e4SLinus Torvalds } 2051da177e4SLinus Torvalds 206bd4c625cSLinus Torvalds static void allocate_bitmap_nodes(struct super_block *p_s_sb) 207bd4c625cSLinus Torvalds { 2081da177e4SLinus Torvalds int i; 2091da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 2101da177e4SLinus Torvalds struct reiserfs_bitmap_node *bn = NULL; 2111da177e4SLinus Torvalds for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) { 2121da177e4SLinus Torvalds bn = allocate_bitmap_node(p_s_sb); 2131da177e4SLinus Torvalds if (bn) { 2141da177e4SLinus Torvalds list_add(&bn->list, &journal->j_bitmap_nodes); 2151da177e4SLinus Torvalds journal->j_free_bitmap_nodes++; 2161da177e4SLinus Torvalds } else { 2171da177e4SLinus Torvalds break; // this is ok, we'll try again when more are needed 2181da177e4SLinus Torvalds } 2191da177e4SLinus Torvalds } 2201da177e4SLinus Torvalds } 2211da177e4SLinus Torvalds 2221da177e4SLinus Torvalds static int set_bit_in_list_bitmap(struct super_block *p_s_sb, int block, 223bd4c625cSLinus Torvalds struct reiserfs_list_bitmap *jb) 224bd4c625cSLinus Torvalds { 2251da177e4SLinus Torvalds int bmap_nr = block / (p_s_sb->s_blocksize << 3); 2261da177e4SLinus Torvalds int bit_nr = block % (p_s_sb->s_blocksize << 3); 2271da177e4SLinus Torvalds 2281da177e4SLinus Torvalds if (!jb->bitmaps[bmap_nr]) { 2291da177e4SLinus Torvalds jb->bitmaps[bmap_nr] = get_bitmap_node(p_s_sb); 2301da177e4SLinus Torvalds } 2311da177e4SLinus Torvalds set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data); 2321da177e4SLinus Torvalds return 0; 2331da177e4SLinus Torvalds } 2341da177e4SLinus Torvalds 2351da177e4SLinus Torvalds static void cleanup_bitmap_list(struct super_block *p_s_sb, 236bd4c625cSLinus Torvalds struct reiserfs_list_bitmap *jb) 237bd4c625cSLinus Torvalds { 2381da177e4SLinus Torvalds int i; 2391da177e4SLinus Torvalds if (jb->bitmaps == NULL) 2401da177e4SLinus Torvalds return; 2411da177e4SLinus Torvalds 2421da177e4SLinus Torvalds for (i = 0; i < SB_BMAP_NR(p_s_sb); i++) { 2431da177e4SLinus Torvalds if (jb->bitmaps[i]) { 2441da177e4SLinus Torvalds free_bitmap_node(p_s_sb, jb->bitmaps[i]); 2451da177e4SLinus Torvalds jb->bitmaps[i] = NULL; 2461da177e4SLinus Torvalds } 2471da177e4SLinus Torvalds } 2481da177e4SLinus Torvalds } 2491da177e4SLinus Torvalds 2501da177e4SLinus Torvalds /* 2511da177e4SLinus Torvalds ** only call this on FS unmount. 2521da177e4SLinus Torvalds */ 2531da177e4SLinus Torvalds static int free_list_bitmaps(struct super_block *p_s_sb, 254bd4c625cSLinus Torvalds struct reiserfs_list_bitmap *jb_array) 255bd4c625cSLinus Torvalds { 2561da177e4SLinus Torvalds int i; 2571da177e4SLinus Torvalds struct reiserfs_list_bitmap *jb; 2581da177e4SLinus Torvalds for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) { 2591da177e4SLinus Torvalds jb = jb_array + i; 2601da177e4SLinus Torvalds jb->journal_list = NULL; 2611da177e4SLinus Torvalds cleanup_bitmap_list(p_s_sb, jb); 2621da177e4SLinus Torvalds vfree(jb->bitmaps); 2631da177e4SLinus Torvalds jb->bitmaps = NULL; 2641da177e4SLinus Torvalds } 2651da177e4SLinus Torvalds return 0; 2661da177e4SLinus Torvalds } 2671da177e4SLinus Torvalds 268bd4c625cSLinus Torvalds static int free_bitmap_nodes(struct super_block *p_s_sb) 269bd4c625cSLinus Torvalds { 2701da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 2711da177e4SLinus Torvalds struct list_head *next = journal->j_bitmap_nodes.next; 2721da177e4SLinus Torvalds struct reiserfs_bitmap_node *bn; 2731da177e4SLinus Torvalds 2741da177e4SLinus Torvalds while (next != &journal->j_bitmap_nodes) { 2751da177e4SLinus Torvalds bn = list_entry(next, struct reiserfs_bitmap_node, list); 2761da177e4SLinus Torvalds list_del(next); 277d739b42bSPekka Enberg kfree(bn->data); 278d739b42bSPekka Enberg kfree(bn); 2791da177e4SLinus Torvalds next = journal->j_bitmap_nodes.next; 2801da177e4SLinus Torvalds journal->j_free_bitmap_nodes--; 2811da177e4SLinus Torvalds } 2821da177e4SLinus Torvalds 2831da177e4SLinus Torvalds return 0; 2841da177e4SLinus Torvalds } 2851da177e4SLinus Torvalds 2861da177e4SLinus Torvalds /* 2871da177e4SLinus Torvalds ** get memory for JOURNAL_NUM_BITMAPS worth of bitmaps. 2881da177e4SLinus Torvalds ** jb_array is the array to be filled in. 2891da177e4SLinus Torvalds */ 2901da177e4SLinus Torvalds int reiserfs_allocate_list_bitmaps(struct super_block *p_s_sb, 2911da177e4SLinus Torvalds struct reiserfs_list_bitmap *jb_array, 292bd4c625cSLinus Torvalds int bmap_nr) 293bd4c625cSLinus Torvalds { 2941da177e4SLinus Torvalds int i; 2951da177e4SLinus Torvalds int failed = 0; 2961da177e4SLinus Torvalds struct reiserfs_list_bitmap *jb; 2971da177e4SLinus Torvalds int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *); 2981da177e4SLinus Torvalds 2991da177e4SLinus Torvalds for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) { 3001da177e4SLinus Torvalds jb = jb_array + i; 3011da177e4SLinus Torvalds jb->journal_list = NULL; 3021da177e4SLinus Torvalds jb->bitmaps = vmalloc(mem); 3031da177e4SLinus Torvalds if (!jb->bitmaps) { 304bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 305bd4c625cSLinus Torvalds "clm-2000, unable to allocate bitmaps for journal lists"); 3061da177e4SLinus Torvalds failed = 1; 3071da177e4SLinus Torvalds break; 3081da177e4SLinus Torvalds } 3091da177e4SLinus Torvalds memset(jb->bitmaps, 0, mem); 3101da177e4SLinus Torvalds } 3111da177e4SLinus Torvalds if (failed) { 3121da177e4SLinus Torvalds free_list_bitmaps(p_s_sb, jb_array); 3131da177e4SLinus Torvalds return -1; 3141da177e4SLinus Torvalds } 3151da177e4SLinus Torvalds return 0; 3161da177e4SLinus Torvalds } 3171da177e4SLinus Torvalds 3181da177e4SLinus Torvalds /* 3191da177e4SLinus Torvalds ** find an available list bitmap. If you can't find one, flush a commit list 3201da177e4SLinus Torvalds ** and try again 3211da177e4SLinus Torvalds */ 322bd4c625cSLinus Torvalds static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *p_s_sb, 323bd4c625cSLinus Torvalds struct reiserfs_journal_list 324bd4c625cSLinus Torvalds *jl) 325bd4c625cSLinus Torvalds { 3261da177e4SLinus Torvalds int i, j; 3271da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 3281da177e4SLinus Torvalds struct reiserfs_list_bitmap *jb = NULL; 3291da177e4SLinus Torvalds 3301da177e4SLinus Torvalds for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) { 3311da177e4SLinus Torvalds i = journal->j_list_bitmap_index; 3321da177e4SLinus Torvalds journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS; 3331da177e4SLinus Torvalds jb = journal->j_list_bitmap + i; 3341da177e4SLinus Torvalds if (journal->j_list_bitmap[i].journal_list) { 335bd4c625cSLinus Torvalds flush_commit_list(p_s_sb, 336bd4c625cSLinus Torvalds journal->j_list_bitmap[i]. 337bd4c625cSLinus Torvalds journal_list, 1); 3381da177e4SLinus Torvalds if (!journal->j_list_bitmap[i].journal_list) { 3391da177e4SLinus Torvalds break; 3401da177e4SLinus Torvalds } 3411da177e4SLinus Torvalds } else { 3421da177e4SLinus Torvalds break; 3431da177e4SLinus Torvalds } 3441da177e4SLinus Torvalds } 3451da177e4SLinus Torvalds if (jb->journal_list) { /* double check to make sure if flushed correctly */ 3461da177e4SLinus Torvalds return NULL; 3471da177e4SLinus Torvalds } 3481da177e4SLinus Torvalds jb->journal_list = jl; 3491da177e4SLinus Torvalds return jb; 3501da177e4SLinus Torvalds } 3511da177e4SLinus Torvalds 3521da177e4SLinus Torvalds /* 3531da177e4SLinus Torvalds ** allocates a new chunk of X nodes, and links them all together as a list. 3541da177e4SLinus Torvalds ** Uses the cnode->next and cnode->prev pointers 3551da177e4SLinus Torvalds ** returns NULL on failure 3561da177e4SLinus Torvalds */ 357bd4c625cSLinus Torvalds static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes) 358bd4c625cSLinus Torvalds { 3591da177e4SLinus Torvalds struct reiserfs_journal_cnode *head; 3601da177e4SLinus Torvalds int i; 3611da177e4SLinus Torvalds if (num_cnodes <= 0) { 3621da177e4SLinus Torvalds return NULL; 3631da177e4SLinus Torvalds } 3641da177e4SLinus Torvalds head = vmalloc(num_cnodes * sizeof(struct reiserfs_journal_cnode)); 3651da177e4SLinus Torvalds if (!head) { 3661da177e4SLinus Torvalds return NULL; 3671da177e4SLinus Torvalds } 3681da177e4SLinus Torvalds memset(head, 0, num_cnodes * sizeof(struct reiserfs_journal_cnode)); 3691da177e4SLinus Torvalds head[0].prev = NULL; 3701da177e4SLinus Torvalds head[0].next = head + 1; 3711da177e4SLinus Torvalds for (i = 1; i < num_cnodes; i++) { 3721da177e4SLinus Torvalds head[i].prev = head + (i - 1); 3731da177e4SLinus Torvalds head[i].next = head + (i + 1); /* if last one, overwrite it after the if */ 3741da177e4SLinus Torvalds } 3751da177e4SLinus Torvalds head[num_cnodes - 1].next = NULL; 3761da177e4SLinus Torvalds return head; 3771da177e4SLinus Torvalds } 3781da177e4SLinus Torvalds 3791da177e4SLinus Torvalds /* 3801da177e4SLinus Torvalds ** pulls a cnode off the free list, or returns NULL on failure 3811da177e4SLinus Torvalds */ 382bd4c625cSLinus Torvalds static struct reiserfs_journal_cnode *get_cnode(struct super_block *p_s_sb) 383bd4c625cSLinus Torvalds { 3841da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 3851da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 3861da177e4SLinus Torvalds 3871da177e4SLinus Torvalds reiserfs_check_lock_depth(p_s_sb, "get_cnode"); 3881da177e4SLinus Torvalds 3891da177e4SLinus Torvalds if (journal->j_cnode_free <= 0) { 3901da177e4SLinus Torvalds return NULL; 3911da177e4SLinus Torvalds } 3921da177e4SLinus Torvalds journal->j_cnode_used++; 3931da177e4SLinus Torvalds journal->j_cnode_free--; 3941da177e4SLinus Torvalds cn = journal->j_cnode_free_list; 3951da177e4SLinus Torvalds if (!cn) { 3961da177e4SLinus Torvalds return cn; 3971da177e4SLinus Torvalds } 3981da177e4SLinus Torvalds if (cn->next) { 3991da177e4SLinus Torvalds cn->next->prev = NULL; 4001da177e4SLinus Torvalds } 4011da177e4SLinus Torvalds journal->j_cnode_free_list = cn->next; 4021da177e4SLinus Torvalds memset(cn, 0, sizeof(struct reiserfs_journal_cnode)); 4031da177e4SLinus Torvalds return cn; 4041da177e4SLinus Torvalds } 4051da177e4SLinus Torvalds 4061da177e4SLinus Torvalds /* 4071da177e4SLinus Torvalds ** returns a cnode to the free list 4081da177e4SLinus Torvalds */ 409bd4c625cSLinus Torvalds static void free_cnode(struct super_block *p_s_sb, 410bd4c625cSLinus Torvalds struct reiserfs_journal_cnode *cn) 411bd4c625cSLinus Torvalds { 4121da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 4131da177e4SLinus Torvalds 4141da177e4SLinus Torvalds reiserfs_check_lock_depth(p_s_sb, "free_cnode"); 4151da177e4SLinus Torvalds 4161da177e4SLinus Torvalds journal->j_cnode_used--; 4171da177e4SLinus Torvalds journal->j_cnode_free++; 4181da177e4SLinus Torvalds /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */ 4191da177e4SLinus Torvalds cn->next = journal->j_cnode_free_list; 4201da177e4SLinus Torvalds if (journal->j_cnode_free_list) { 4211da177e4SLinus Torvalds journal->j_cnode_free_list->prev = cn; 4221da177e4SLinus Torvalds } 4231da177e4SLinus Torvalds cn->prev = NULL; /* not needed with the memset, but I might kill the memset, and forget to do this */ 4241da177e4SLinus Torvalds journal->j_cnode_free_list = cn; 4251da177e4SLinus Torvalds } 4261da177e4SLinus Torvalds 427bd4c625cSLinus Torvalds static void clear_prepared_bits(struct buffer_head *bh) 428bd4c625cSLinus Torvalds { 4291da177e4SLinus Torvalds clear_buffer_journal_prepared(bh); 4301da177e4SLinus Torvalds clear_buffer_journal_restore_dirty(bh); 4311da177e4SLinus Torvalds } 4321da177e4SLinus Torvalds 4331da177e4SLinus Torvalds /* utility function to force a BUG if it is called without the big 4341da177e4SLinus Torvalds ** kernel lock held. caller is the string printed just before calling BUG() 4351da177e4SLinus Torvalds */ 436bd4c625cSLinus Torvalds void reiserfs_check_lock_depth(struct super_block *sb, char *caller) 437bd4c625cSLinus Torvalds { 4381da177e4SLinus Torvalds #ifdef CONFIG_SMP 4391da177e4SLinus Torvalds if (current->lock_depth < 0) { 440bd4c625cSLinus Torvalds reiserfs_panic(sb, "%s called without kernel lock held", 441bd4c625cSLinus Torvalds caller); 4421da177e4SLinus Torvalds } 4431da177e4SLinus Torvalds #else 4441da177e4SLinus Torvalds ; 4451da177e4SLinus Torvalds #endif 4461da177e4SLinus Torvalds } 4471da177e4SLinus Torvalds 4481da177e4SLinus Torvalds /* return a cnode with same dev, block number and size in table, or null if not found */ 449bd4c625cSLinus Torvalds static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct 450bd4c625cSLinus Torvalds super_block 451bd4c625cSLinus Torvalds *sb, 452bd4c625cSLinus Torvalds struct 453bd4c625cSLinus Torvalds reiserfs_journal_cnode 454bd4c625cSLinus Torvalds **table, 4551da177e4SLinus Torvalds long bl) 4561da177e4SLinus Torvalds { 4571da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 4581da177e4SLinus Torvalds cn = journal_hash(table, sb, bl); 4591da177e4SLinus Torvalds while (cn) { 4601da177e4SLinus Torvalds if (cn->blocknr == bl && cn->sb == sb) 4611da177e4SLinus Torvalds return cn; 4621da177e4SLinus Torvalds cn = cn->hnext; 4631da177e4SLinus Torvalds } 4641da177e4SLinus Torvalds return (struct reiserfs_journal_cnode *)0; 4651da177e4SLinus Torvalds } 4661da177e4SLinus Torvalds 4671da177e4SLinus Torvalds /* 4681da177e4SLinus Torvalds ** this actually means 'can this block be reallocated yet?'. If you set search_all, a block can only be allocated 4691da177e4SLinus Torvalds ** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever 4701da177e4SLinus Torvalds ** being overwritten by a replay after crashing. 4711da177e4SLinus Torvalds ** 4721da177e4SLinus Torvalds ** If you don't set search_all, a block can only be allocated if it is not in the current transaction. Since deleting 4731da177e4SLinus Torvalds ** a block removes it from the current transaction, this case should never happen. If you don't set search_all, make 4741da177e4SLinus Torvalds ** sure you never write the block without logging it. 4751da177e4SLinus Torvalds ** 4761da177e4SLinus Torvalds ** next_zero_bit is a suggestion about the next block to try for find_forward. 4771da177e4SLinus Torvalds ** when bl is rejected because it is set in a journal list bitmap, we search 4781da177e4SLinus Torvalds ** for the next zero bit in the bitmap that rejected bl. Then, we return that 4791da177e4SLinus Torvalds ** through next_zero_bit for find_forward to try. 4801da177e4SLinus Torvalds ** 4811da177e4SLinus Torvalds ** Just because we return something in next_zero_bit does not mean we won't 4821da177e4SLinus Torvalds ** reject it on the next call to reiserfs_in_journal 4831da177e4SLinus Torvalds ** 4841da177e4SLinus Torvalds */ 4851da177e4SLinus Torvalds int reiserfs_in_journal(struct super_block *p_s_sb, 4861da177e4SLinus Torvalds int bmap_nr, int bit_nr, int search_all, 487bd4c625cSLinus Torvalds b_blocknr_t * next_zero_bit) 488bd4c625cSLinus Torvalds { 4891da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 4901da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 4911da177e4SLinus Torvalds struct reiserfs_list_bitmap *jb; 4921da177e4SLinus Torvalds int i; 4931da177e4SLinus Torvalds unsigned long bl; 4941da177e4SLinus Torvalds 4951da177e4SLinus Torvalds *next_zero_bit = 0; /* always start this at zero. */ 4961da177e4SLinus Torvalds 4971da177e4SLinus Torvalds PROC_INFO_INC(p_s_sb, journal.in_journal); 4981da177e4SLinus Torvalds /* If we aren't doing a search_all, this is a metablock, and it will be logged before use. 4991da177e4SLinus Torvalds ** if we crash before the transaction that freed it commits, this transaction won't 5001da177e4SLinus Torvalds ** have committed either, and the block will never be written 5011da177e4SLinus Torvalds */ 5021da177e4SLinus Torvalds if (search_all) { 5031da177e4SLinus Torvalds for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) { 5041da177e4SLinus Torvalds PROC_INFO_INC(p_s_sb, journal.in_journal_bitmap); 5051da177e4SLinus Torvalds jb = journal->j_list_bitmap + i; 5061da177e4SLinus Torvalds if (jb->journal_list && jb->bitmaps[bmap_nr] && 507bd4c625cSLinus Torvalds test_bit(bit_nr, 508bd4c625cSLinus Torvalds (unsigned long *)jb->bitmaps[bmap_nr]-> 509bd4c625cSLinus Torvalds data)) { 510bd4c625cSLinus Torvalds *next_zero_bit = 511bd4c625cSLinus Torvalds find_next_zero_bit((unsigned long *) 512bd4c625cSLinus Torvalds (jb->bitmaps[bmap_nr]-> 513bd4c625cSLinus Torvalds data), 514bd4c625cSLinus Torvalds p_s_sb->s_blocksize << 3, 515bd4c625cSLinus Torvalds bit_nr + 1); 5161da177e4SLinus Torvalds return 1; 5171da177e4SLinus Torvalds } 5181da177e4SLinus Torvalds } 5191da177e4SLinus Torvalds } 5201da177e4SLinus Torvalds 5211da177e4SLinus Torvalds bl = bmap_nr * (p_s_sb->s_blocksize << 3) + bit_nr; 5221da177e4SLinus Torvalds /* is it in any old transactions? */ 523bd4c625cSLinus Torvalds if (search_all 524bd4c625cSLinus Torvalds && (cn = 525bd4c625cSLinus Torvalds get_journal_hash_dev(p_s_sb, journal->j_list_hash_table, bl))) { 5261da177e4SLinus Torvalds return 1; 5271da177e4SLinus Torvalds } 5281da177e4SLinus Torvalds 5291da177e4SLinus Torvalds /* is it in the current transaction. This should never happen */ 5301da177e4SLinus Torvalds if ((cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, bl))) { 5311da177e4SLinus Torvalds BUG(); 5321da177e4SLinus Torvalds return 1; 5331da177e4SLinus Torvalds } 5341da177e4SLinus Torvalds 5351da177e4SLinus Torvalds PROC_INFO_INC(p_s_sb, journal.in_journal_reusable); 5361da177e4SLinus Torvalds /* safe for reuse */ 5371da177e4SLinus Torvalds return 0; 5381da177e4SLinus Torvalds } 5391da177e4SLinus Torvalds 5401da177e4SLinus Torvalds /* insert cn into table 5411da177e4SLinus Torvalds */ 542bd4c625cSLinus Torvalds static inline void insert_journal_hash(struct reiserfs_journal_cnode **table, 543bd4c625cSLinus Torvalds struct reiserfs_journal_cnode *cn) 544bd4c625cSLinus Torvalds { 5451da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn_orig; 5461da177e4SLinus Torvalds 5471da177e4SLinus Torvalds cn_orig = journal_hash(table, cn->sb, cn->blocknr); 5481da177e4SLinus Torvalds cn->hnext = cn_orig; 5491da177e4SLinus Torvalds cn->hprev = NULL; 5501da177e4SLinus Torvalds if (cn_orig) { 5511da177e4SLinus Torvalds cn_orig->hprev = cn; 5521da177e4SLinus Torvalds } 5531da177e4SLinus Torvalds journal_hash(table, cn->sb, cn->blocknr) = cn; 5541da177e4SLinus Torvalds } 5551da177e4SLinus Torvalds 5561da177e4SLinus Torvalds /* lock the current transaction */ 55777933d72SJesper Juhl static inline void lock_journal(struct super_block *p_s_sb) 558bd4c625cSLinus Torvalds { 5591da177e4SLinus Torvalds PROC_INFO_INC(p_s_sb, journal.lock_journal); 5601da177e4SLinus Torvalds down(&SB_JOURNAL(p_s_sb)->j_lock); 5611da177e4SLinus Torvalds } 5621da177e4SLinus Torvalds 5631da177e4SLinus Torvalds /* unlock the current transaction */ 56477933d72SJesper Juhl static inline void unlock_journal(struct super_block *p_s_sb) 565bd4c625cSLinus Torvalds { 5661da177e4SLinus Torvalds up(&SB_JOURNAL(p_s_sb)->j_lock); 5671da177e4SLinus Torvalds } 5681da177e4SLinus Torvalds 5691da177e4SLinus Torvalds static inline void get_journal_list(struct reiserfs_journal_list *jl) 5701da177e4SLinus Torvalds { 5711da177e4SLinus Torvalds jl->j_refcount++; 5721da177e4SLinus Torvalds } 5731da177e4SLinus Torvalds 5741da177e4SLinus Torvalds static inline void put_journal_list(struct super_block *s, 5751da177e4SLinus Torvalds struct reiserfs_journal_list *jl) 5761da177e4SLinus Torvalds { 5771da177e4SLinus Torvalds if (jl->j_refcount < 1) { 578bd4c625cSLinus Torvalds reiserfs_panic(s, "trans id %lu, refcount at %d", 579bd4c625cSLinus Torvalds jl->j_trans_id, jl->j_refcount); 5801da177e4SLinus Torvalds } 5811da177e4SLinus Torvalds if (--jl->j_refcount == 0) 582d739b42bSPekka Enberg kfree(jl); 5831da177e4SLinus Torvalds } 5841da177e4SLinus Torvalds 5851da177e4SLinus Torvalds /* 5861da177e4SLinus Torvalds ** this used to be much more involved, and I'm keeping it just in case things get ugly again. 5871da177e4SLinus Torvalds ** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a 5881da177e4SLinus Torvalds ** transaction. 5891da177e4SLinus Torvalds */ 590bd4c625cSLinus Torvalds static void cleanup_freed_for_journal_list(struct super_block *p_s_sb, 591bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl) 592bd4c625cSLinus Torvalds { 5931da177e4SLinus Torvalds 5941da177e4SLinus Torvalds struct reiserfs_list_bitmap *jb = jl->j_list_bitmap; 5951da177e4SLinus Torvalds if (jb) { 5961da177e4SLinus Torvalds cleanup_bitmap_list(p_s_sb, jb); 5971da177e4SLinus Torvalds } 5981da177e4SLinus Torvalds jl->j_list_bitmap->journal_list = NULL; 5991da177e4SLinus Torvalds jl->j_list_bitmap = NULL; 6001da177e4SLinus Torvalds } 6011da177e4SLinus Torvalds 6021da177e4SLinus Torvalds static int journal_list_still_alive(struct super_block *s, 6031da177e4SLinus Torvalds unsigned long trans_id) 6041da177e4SLinus Torvalds { 6051da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 6061da177e4SLinus Torvalds struct list_head *entry = &journal->j_journal_list; 6071da177e4SLinus Torvalds struct reiserfs_journal_list *jl; 6081da177e4SLinus Torvalds 6091da177e4SLinus Torvalds if (!list_empty(entry)) { 6101da177e4SLinus Torvalds jl = JOURNAL_LIST_ENTRY(entry->next); 6111da177e4SLinus Torvalds if (jl->j_trans_id <= trans_id) { 6121da177e4SLinus Torvalds return 1; 6131da177e4SLinus Torvalds } 6141da177e4SLinus Torvalds } 6151da177e4SLinus Torvalds return 0; 6161da177e4SLinus Torvalds } 6171da177e4SLinus Torvalds 618bd4c625cSLinus Torvalds static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate) 619bd4c625cSLinus Torvalds { 6201da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 6211da177e4SLinus Torvalds 6221da177e4SLinus Torvalds if (buffer_journaled(bh)) { 623bd4c625cSLinus Torvalds reiserfs_warning(NULL, 624bd4c625cSLinus Torvalds "clm-2084: pinned buffer %lu:%s sent to disk", 6251da177e4SLinus Torvalds bh->b_blocknr, bdevname(bh->b_bdev, b)); 6261da177e4SLinus Torvalds } 6271da177e4SLinus Torvalds if (uptodate) 6281da177e4SLinus Torvalds set_buffer_uptodate(bh); 6291da177e4SLinus Torvalds else 6301da177e4SLinus Torvalds clear_buffer_uptodate(bh); 6311da177e4SLinus Torvalds unlock_buffer(bh); 6321da177e4SLinus Torvalds put_bh(bh); 6331da177e4SLinus Torvalds } 6341da177e4SLinus Torvalds 635bd4c625cSLinus Torvalds static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate) 636bd4c625cSLinus Torvalds { 6371da177e4SLinus Torvalds if (uptodate) 6381da177e4SLinus Torvalds set_buffer_uptodate(bh); 6391da177e4SLinus Torvalds else 6401da177e4SLinus Torvalds clear_buffer_uptodate(bh); 6411da177e4SLinus Torvalds unlock_buffer(bh); 6421da177e4SLinus Torvalds put_bh(bh); 6431da177e4SLinus Torvalds } 6441da177e4SLinus Torvalds 645bd4c625cSLinus Torvalds static void submit_logged_buffer(struct buffer_head *bh) 646bd4c625cSLinus Torvalds { 6471da177e4SLinus Torvalds get_bh(bh); 6481da177e4SLinus Torvalds bh->b_end_io = reiserfs_end_buffer_io_sync; 6491da177e4SLinus Torvalds clear_buffer_journal_new(bh); 6501da177e4SLinus Torvalds clear_buffer_dirty(bh); 6511da177e4SLinus Torvalds if (!test_clear_buffer_journal_test(bh)) 6521da177e4SLinus Torvalds BUG(); 6531da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 6541da177e4SLinus Torvalds BUG(); 6551da177e4SLinus Torvalds submit_bh(WRITE, bh); 6561da177e4SLinus Torvalds } 6571da177e4SLinus Torvalds 658bd4c625cSLinus Torvalds static void submit_ordered_buffer(struct buffer_head *bh) 659bd4c625cSLinus Torvalds { 6601da177e4SLinus Torvalds get_bh(bh); 6611da177e4SLinus Torvalds bh->b_end_io = reiserfs_end_ordered_io; 6621da177e4SLinus Torvalds clear_buffer_dirty(bh); 6631da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 6641da177e4SLinus Torvalds BUG(); 6651da177e4SLinus Torvalds submit_bh(WRITE, bh); 6661da177e4SLinus Torvalds } 6671da177e4SLinus Torvalds 668bd4c625cSLinus Torvalds static int submit_barrier_buffer(struct buffer_head *bh) 669bd4c625cSLinus Torvalds { 6701da177e4SLinus Torvalds get_bh(bh); 6711da177e4SLinus Torvalds bh->b_end_io = reiserfs_end_ordered_io; 6721da177e4SLinus Torvalds clear_buffer_dirty(bh); 6731da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 6741da177e4SLinus Torvalds BUG(); 6751da177e4SLinus Torvalds return submit_bh(WRITE_BARRIER, bh); 6761da177e4SLinus Torvalds } 6771da177e4SLinus Torvalds 6781da177e4SLinus Torvalds static void check_barrier_completion(struct super_block *s, 679bd4c625cSLinus Torvalds struct buffer_head *bh) 680bd4c625cSLinus Torvalds { 6811da177e4SLinus Torvalds if (buffer_eopnotsupp(bh)) { 6821da177e4SLinus Torvalds clear_buffer_eopnotsupp(bh); 6831da177e4SLinus Torvalds disable_barrier(s); 6841da177e4SLinus Torvalds set_buffer_uptodate(bh); 6851da177e4SLinus Torvalds set_buffer_dirty(bh); 6861da177e4SLinus Torvalds sync_dirty_buffer(bh); 6871da177e4SLinus Torvalds } 6881da177e4SLinus Torvalds } 6891da177e4SLinus Torvalds 6901da177e4SLinus Torvalds #define CHUNK_SIZE 32 6911da177e4SLinus Torvalds struct buffer_chunk { 6921da177e4SLinus Torvalds struct buffer_head *bh[CHUNK_SIZE]; 6931da177e4SLinus Torvalds int nr; 6941da177e4SLinus Torvalds }; 6951da177e4SLinus Torvalds 696bd4c625cSLinus Torvalds static void write_chunk(struct buffer_chunk *chunk) 697bd4c625cSLinus Torvalds { 6981da177e4SLinus Torvalds int i; 69922e2c507SJens Axboe get_fs_excl(); 7001da177e4SLinus Torvalds for (i = 0; i < chunk->nr; i++) { 7011da177e4SLinus Torvalds submit_logged_buffer(chunk->bh[i]); 7021da177e4SLinus Torvalds } 7031da177e4SLinus Torvalds chunk->nr = 0; 70422e2c507SJens Axboe put_fs_excl(); 7051da177e4SLinus Torvalds } 7061da177e4SLinus Torvalds 707bd4c625cSLinus Torvalds static void write_ordered_chunk(struct buffer_chunk *chunk) 708bd4c625cSLinus Torvalds { 7091da177e4SLinus Torvalds int i; 71022e2c507SJens Axboe get_fs_excl(); 7111da177e4SLinus Torvalds for (i = 0; i < chunk->nr; i++) { 7121da177e4SLinus Torvalds submit_ordered_buffer(chunk->bh[i]); 7131da177e4SLinus Torvalds } 7141da177e4SLinus Torvalds chunk->nr = 0; 71522e2c507SJens Axboe put_fs_excl(); 7161da177e4SLinus Torvalds } 7171da177e4SLinus Torvalds 7181da177e4SLinus Torvalds static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh, 719bd4c625cSLinus Torvalds spinlock_t * lock, void (fn) (struct buffer_chunk *)) 7201da177e4SLinus Torvalds { 7211da177e4SLinus Torvalds int ret = 0; 72214a61442SEric Sesterhenn BUG_ON(chunk->nr >= CHUNK_SIZE); 7231da177e4SLinus Torvalds chunk->bh[chunk->nr++] = bh; 7241da177e4SLinus Torvalds if (chunk->nr >= CHUNK_SIZE) { 7251da177e4SLinus Torvalds ret = 1; 7261da177e4SLinus Torvalds if (lock) 7271da177e4SLinus Torvalds spin_unlock(lock); 7281da177e4SLinus Torvalds fn(chunk); 7291da177e4SLinus Torvalds if (lock) 7301da177e4SLinus Torvalds spin_lock(lock); 7311da177e4SLinus Torvalds } 7321da177e4SLinus Torvalds return ret; 7331da177e4SLinus Torvalds } 7341da177e4SLinus Torvalds 7351da177e4SLinus Torvalds static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0); 736bd4c625cSLinus Torvalds static struct reiserfs_jh *alloc_jh(void) 737bd4c625cSLinus Torvalds { 7381da177e4SLinus Torvalds struct reiserfs_jh *jh; 7391da177e4SLinus Torvalds while (1) { 7401da177e4SLinus Torvalds jh = kmalloc(sizeof(*jh), GFP_NOFS); 7411da177e4SLinus Torvalds if (jh) { 7421da177e4SLinus Torvalds atomic_inc(&nr_reiserfs_jh); 7431da177e4SLinus Torvalds return jh; 7441da177e4SLinus Torvalds } 7451da177e4SLinus Torvalds yield(); 7461da177e4SLinus Torvalds } 7471da177e4SLinus Torvalds } 7481da177e4SLinus Torvalds 7491da177e4SLinus Torvalds /* 7501da177e4SLinus Torvalds * we want to free the jh when the buffer has been written 7511da177e4SLinus Torvalds * and waited on 7521da177e4SLinus Torvalds */ 753bd4c625cSLinus Torvalds void reiserfs_free_jh(struct buffer_head *bh) 754bd4c625cSLinus Torvalds { 7551da177e4SLinus Torvalds struct reiserfs_jh *jh; 7561da177e4SLinus Torvalds 7571da177e4SLinus Torvalds jh = bh->b_private; 7581da177e4SLinus Torvalds if (jh) { 7591da177e4SLinus Torvalds bh->b_private = NULL; 7601da177e4SLinus Torvalds jh->bh = NULL; 7611da177e4SLinus Torvalds list_del_init(&jh->list); 7621da177e4SLinus Torvalds kfree(jh); 7631da177e4SLinus Torvalds if (atomic_read(&nr_reiserfs_jh) <= 0) 7641da177e4SLinus Torvalds BUG(); 7651da177e4SLinus Torvalds atomic_dec(&nr_reiserfs_jh); 7661da177e4SLinus Torvalds put_bh(bh); 7671da177e4SLinus Torvalds } 7681da177e4SLinus Torvalds } 7691da177e4SLinus Torvalds 7701da177e4SLinus Torvalds static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh, 7711da177e4SLinus Torvalds int tail) 7721da177e4SLinus Torvalds { 7731da177e4SLinus Torvalds struct reiserfs_jh *jh; 7741da177e4SLinus Torvalds 7751da177e4SLinus Torvalds if (bh->b_private) { 7761da177e4SLinus Torvalds spin_lock(&j->j_dirty_buffers_lock); 7771da177e4SLinus Torvalds if (!bh->b_private) { 7781da177e4SLinus Torvalds spin_unlock(&j->j_dirty_buffers_lock); 7791da177e4SLinus Torvalds goto no_jh; 7801da177e4SLinus Torvalds } 7811da177e4SLinus Torvalds jh = bh->b_private; 7821da177e4SLinus Torvalds list_del_init(&jh->list); 7831da177e4SLinus Torvalds } else { 7841da177e4SLinus Torvalds no_jh: 7851da177e4SLinus Torvalds get_bh(bh); 7861da177e4SLinus Torvalds jh = alloc_jh(); 7871da177e4SLinus Torvalds spin_lock(&j->j_dirty_buffers_lock); 7881da177e4SLinus Torvalds /* buffer must be locked for __add_jh, should be able to have 7891da177e4SLinus Torvalds * two adds at the same time 7901da177e4SLinus Torvalds */ 79114a61442SEric Sesterhenn BUG_ON(bh->b_private); 7921da177e4SLinus Torvalds jh->bh = bh; 7931da177e4SLinus Torvalds bh->b_private = jh; 7941da177e4SLinus Torvalds } 7951da177e4SLinus Torvalds jh->jl = j->j_current_jl; 7961da177e4SLinus Torvalds if (tail) 7971da177e4SLinus Torvalds list_add_tail(&jh->list, &jh->jl->j_tail_bh_list); 7981da177e4SLinus Torvalds else { 7991da177e4SLinus Torvalds list_add_tail(&jh->list, &jh->jl->j_bh_list); 8001da177e4SLinus Torvalds } 8011da177e4SLinus Torvalds spin_unlock(&j->j_dirty_buffers_lock); 8021da177e4SLinus Torvalds return 0; 8031da177e4SLinus Torvalds } 8041da177e4SLinus Torvalds 805bd4c625cSLinus Torvalds int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh) 806bd4c625cSLinus Torvalds { 8071da177e4SLinus Torvalds return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1); 8081da177e4SLinus Torvalds } 809bd4c625cSLinus Torvalds int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh) 810bd4c625cSLinus Torvalds { 8111da177e4SLinus Torvalds return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0); 8121da177e4SLinus Torvalds } 8131da177e4SLinus Torvalds 8141da177e4SLinus Torvalds #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list) 8151da177e4SLinus Torvalds static int write_ordered_buffers(spinlock_t * lock, 8161da177e4SLinus Torvalds struct reiserfs_journal *j, 8171da177e4SLinus Torvalds struct reiserfs_journal_list *jl, 8181da177e4SLinus Torvalds struct list_head *list) 8191da177e4SLinus Torvalds { 8201da177e4SLinus Torvalds struct buffer_head *bh; 8211da177e4SLinus Torvalds struct reiserfs_jh *jh; 8221da177e4SLinus Torvalds int ret = j->j_errno; 8231da177e4SLinus Torvalds struct buffer_chunk chunk; 8241da177e4SLinus Torvalds struct list_head tmp; 8251da177e4SLinus Torvalds INIT_LIST_HEAD(&tmp); 8261da177e4SLinus Torvalds 8271da177e4SLinus Torvalds chunk.nr = 0; 8281da177e4SLinus Torvalds spin_lock(lock); 8291da177e4SLinus Torvalds while (!list_empty(list)) { 8301da177e4SLinus Torvalds jh = JH_ENTRY(list->next); 8311da177e4SLinus Torvalds bh = jh->bh; 8321da177e4SLinus Torvalds get_bh(bh); 8331da177e4SLinus Torvalds if (test_set_buffer_locked(bh)) { 8341da177e4SLinus Torvalds if (!buffer_dirty(bh)) { 835f116629dSAkinobu Mita list_move(&jh->list, &tmp); 8361da177e4SLinus Torvalds goto loop_next; 8371da177e4SLinus Torvalds } 8381da177e4SLinus Torvalds spin_unlock(lock); 8391da177e4SLinus Torvalds if (chunk.nr) 8401da177e4SLinus Torvalds write_ordered_chunk(&chunk); 8411da177e4SLinus Torvalds wait_on_buffer(bh); 8421da177e4SLinus Torvalds cond_resched(); 8431da177e4SLinus Torvalds spin_lock(lock); 8441da177e4SLinus Torvalds goto loop_next; 8451da177e4SLinus Torvalds } 8463d4492f8SChris Mason /* in theory, dirty non-uptodate buffers should never get here, 8473d4492f8SChris Mason * but the upper layer io error paths still have a few quirks. 8483d4492f8SChris Mason * Handle them here as gracefully as we can 8493d4492f8SChris Mason */ 8503d4492f8SChris Mason if (!buffer_uptodate(bh) && buffer_dirty(bh)) { 8513d4492f8SChris Mason clear_buffer_dirty(bh); 8523d4492f8SChris Mason ret = -EIO; 8533d4492f8SChris Mason } 8541da177e4SLinus Torvalds if (buffer_dirty(bh)) { 855f116629dSAkinobu Mita list_move(&jh->list, &tmp); 8561da177e4SLinus Torvalds add_to_chunk(&chunk, bh, lock, write_ordered_chunk); 8571da177e4SLinus Torvalds } else { 8581da177e4SLinus Torvalds reiserfs_free_jh(bh); 8591da177e4SLinus Torvalds unlock_buffer(bh); 8601da177e4SLinus Torvalds } 8611da177e4SLinus Torvalds loop_next: 8621da177e4SLinus Torvalds put_bh(bh); 8631da177e4SLinus Torvalds cond_resched_lock(lock); 8641da177e4SLinus Torvalds } 8651da177e4SLinus Torvalds if (chunk.nr) { 8661da177e4SLinus Torvalds spin_unlock(lock); 8671da177e4SLinus Torvalds write_ordered_chunk(&chunk); 8681da177e4SLinus Torvalds spin_lock(lock); 8691da177e4SLinus Torvalds } 8701da177e4SLinus Torvalds while (!list_empty(&tmp)) { 8711da177e4SLinus Torvalds jh = JH_ENTRY(tmp.prev); 8721da177e4SLinus Torvalds bh = jh->bh; 8731da177e4SLinus Torvalds get_bh(bh); 8741da177e4SLinus Torvalds reiserfs_free_jh(bh); 8751da177e4SLinus Torvalds 8761da177e4SLinus Torvalds if (buffer_locked(bh)) { 8771da177e4SLinus Torvalds spin_unlock(lock); 8781da177e4SLinus Torvalds wait_on_buffer(bh); 8791da177e4SLinus Torvalds spin_lock(lock); 8801da177e4SLinus Torvalds } 8811da177e4SLinus Torvalds if (!buffer_uptodate(bh)) { 8821da177e4SLinus Torvalds ret = -EIO; 8831da177e4SLinus Torvalds } 884d62b1b87SChris Mason /* ugly interaction with invalidatepage here. 885d62b1b87SChris Mason * reiserfs_invalidate_page will pin any buffer that has a valid 886d62b1b87SChris Mason * journal head from an older transaction. If someone else sets 887d62b1b87SChris Mason * our buffer dirty after we write it in the first loop, and 888d62b1b87SChris Mason * then someone truncates the page away, nobody will ever write 889d62b1b87SChris Mason * the buffer. We're safe if we write the page one last time 890d62b1b87SChris Mason * after freeing the journal header. 891d62b1b87SChris Mason */ 892d62b1b87SChris Mason if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) { 893d62b1b87SChris Mason spin_unlock(lock); 894d62b1b87SChris Mason ll_rw_block(WRITE, 1, &bh); 895d62b1b87SChris Mason spin_lock(lock); 896d62b1b87SChris Mason } 8971da177e4SLinus Torvalds put_bh(bh); 8981da177e4SLinus Torvalds cond_resched_lock(lock); 8991da177e4SLinus Torvalds } 9001da177e4SLinus Torvalds spin_unlock(lock); 9011da177e4SLinus Torvalds return ret; 9021da177e4SLinus Torvalds } 9031da177e4SLinus Torvalds 904bd4c625cSLinus Torvalds static int flush_older_commits(struct super_block *s, 905bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl) 906bd4c625cSLinus Torvalds { 9071da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 9081da177e4SLinus Torvalds struct reiserfs_journal_list *other_jl; 9091da177e4SLinus Torvalds struct reiserfs_journal_list *first_jl; 9101da177e4SLinus Torvalds struct list_head *entry; 9111da177e4SLinus Torvalds unsigned long trans_id = jl->j_trans_id; 9121da177e4SLinus Torvalds unsigned long other_trans_id; 9131da177e4SLinus Torvalds unsigned long first_trans_id; 9141da177e4SLinus Torvalds 9151da177e4SLinus Torvalds find_first: 9161da177e4SLinus Torvalds /* 9171da177e4SLinus Torvalds * first we walk backwards to find the oldest uncommitted transation 9181da177e4SLinus Torvalds */ 9191da177e4SLinus Torvalds first_jl = jl; 9201da177e4SLinus Torvalds entry = jl->j_list.prev; 9211da177e4SLinus Torvalds while (1) { 9221da177e4SLinus Torvalds other_jl = JOURNAL_LIST_ENTRY(entry); 9231da177e4SLinus Torvalds if (entry == &journal->j_journal_list || 9241da177e4SLinus Torvalds atomic_read(&other_jl->j_older_commits_done)) 9251da177e4SLinus Torvalds break; 9261da177e4SLinus Torvalds 9271da177e4SLinus Torvalds first_jl = other_jl; 9281da177e4SLinus Torvalds entry = other_jl->j_list.prev; 9291da177e4SLinus Torvalds } 9301da177e4SLinus Torvalds 9311da177e4SLinus Torvalds /* if we didn't find any older uncommitted transactions, return now */ 9321da177e4SLinus Torvalds if (first_jl == jl) { 9331da177e4SLinus Torvalds return 0; 9341da177e4SLinus Torvalds } 9351da177e4SLinus Torvalds 9361da177e4SLinus Torvalds first_trans_id = first_jl->j_trans_id; 9371da177e4SLinus Torvalds 9381da177e4SLinus Torvalds entry = &first_jl->j_list; 9391da177e4SLinus Torvalds while (1) { 9401da177e4SLinus Torvalds other_jl = JOURNAL_LIST_ENTRY(entry); 9411da177e4SLinus Torvalds other_trans_id = other_jl->j_trans_id; 9421da177e4SLinus Torvalds 9431da177e4SLinus Torvalds if (other_trans_id < trans_id) { 9441da177e4SLinus Torvalds if (atomic_read(&other_jl->j_commit_left) != 0) { 9451da177e4SLinus Torvalds flush_commit_list(s, other_jl, 0); 9461da177e4SLinus Torvalds 9471da177e4SLinus Torvalds /* list we were called with is gone, return */ 9481da177e4SLinus Torvalds if (!journal_list_still_alive(s, trans_id)) 9491da177e4SLinus Torvalds return 1; 9501da177e4SLinus Torvalds 9511da177e4SLinus Torvalds /* the one we just flushed is gone, this means all 9521da177e4SLinus Torvalds * older lists are also gone, so first_jl is no longer 9531da177e4SLinus Torvalds * valid either. Go back to the beginning. 9541da177e4SLinus Torvalds */ 955bd4c625cSLinus Torvalds if (!journal_list_still_alive 956bd4c625cSLinus Torvalds (s, other_trans_id)) { 9571da177e4SLinus Torvalds goto find_first; 9581da177e4SLinus Torvalds } 9591da177e4SLinus Torvalds } 9601da177e4SLinus Torvalds entry = entry->next; 9611da177e4SLinus Torvalds if (entry == &journal->j_journal_list) 9621da177e4SLinus Torvalds return 0; 9631da177e4SLinus Torvalds } else { 9641da177e4SLinus Torvalds return 0; 9651da177e4SLinus Torvalds } 9661da177e4SLinus Torvalds } 9671da177e4SLinus Torvalds return 0; 9681da177e4SLinus Torvalds } 969bd4c625cSLinus Torvalds int reiserfs_async_progress_wait(struct super_block *s) 970bd4c625cSLinus Torvalds { 9711da177e4SLinus Torvalds DEFINE_WAIT(wait); 9721da177e4SLinus Torvalds struct reiserfs_journal *j = SB_JOURNAL(s); 9731da177e4SLinus Torvalds if (atomic_read(&j->j_async_throttle)) 9743fcfab16SAndrew Morton congestion_wait(WRITE, HZ / 10); 9751da177e4SLinus Torvalds return 0; 9761da177e4SLinus Torvalds } 9771da177e4SLinus Torvalds 9781da177e4SLinus Torvalds /* 9791da177e4SLinus Torvalds ** if this journal list still has commit blocks unflushed, send them to disk. 9801da177e4SLinus Torvalds ** 9811da177e4SLinus Torvalds ** log areas must be flushed in order (transaction 2 can't commit before transaction 1) 9821da177e4SLinus Torvalds ** Before the commit block can by written, every other log block must be safely on disk 9831da177e4SLinus Torvalds ** 9841da177e4SLinus Torvalds */ 985bd4c625cSLinus Torvalds static int flush_commit_list(struct super_block *s, 986bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl, int flushall) 987bd4c625cSLinus Torvalds { 9881da177e4SLinus Torvalds int i; 9891da177e4SLinus Torvalds int bn; 9901da177e4SLinus Torvalds struct buffer_head *tbh = NULL; 9911da177e4SLinus Torvalds unsigned long trans_id = jl->j_trans_id; 9921da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 9931da177e4SLinus Torvalds int barrier = 0; 9941da177e4SLinus Torvalds int retval = 0; 995e0e851cfSChris Mason int write_len; 9961da177e4SLinus Torvalds 9971da177e4SLinus Torvalds reiserfs_check_lock_depth(s, "flush_commit_list"); 9981da177e4SLinus Torvalds 9991da177e4SLinus Torvalds if (atomic_read(&jl->j_older_commits_done)) { 10001da177e4SLinus Torvalds return 0; 10011da177e4SLinus Torvalds } 10021da177e4SLinus Torvalds 100322e2c507SJens Axboe get_fs_excl(); 100422e2c507SJens Axboe 10051da177e4SLinus Torvalds /* before we can put our commit blocks on disk, we have to make sure everyone older than 10061da177e4SLinus Torvalds ** us is on disk too 10071da177e4SLinus Torvalds */ 10081da177e4SLinus Torvalds BUG_ON(jl->j_len <= 0); 10091da177e4SLinus Torvalds BUG_ON(trans_id == journal->j_trans_id); 10101da177e4SLinus Torvalds 10111da177e4SLinus Torvalds get_journal_list(jl); 10121da177e4SLinus Torvalds if (flushall) { 10131da177e4SLinus Torvalds if (flush_older_commits(s, jl) == 1) { 10141da177e4SLinus Torvalds /* list disappeared during flush_older_commits. return */ 10151da177e4SLinus Torvalds goto put_jl; 10161da177e4SLinus Torvalds } 10171da177e4SLinus Torvalds } 10181da177e4SLinus Torvalds 10191da177e4SLinus Torvalds /* make sure nobody is trying to flush this one at the same time */ 10201da177e4SLinus Torvalds down(&jl->j_commit_lock); 10211da177e4SLinus Torvalds if (!journal_list_still_alive(s, trans_id)) { 10221da177e4SLinus Torvalds up(&jl->j_commit_lock); 10231da177e4SLinus Torvalds goto put_jl; 10241da177e4SLinus Torvalds } 10251da177e4SLinus Torvalds BUG_ON(jl->j_trans_id == 0); 10261da177e4SLinus Torvalds 10271da177e4SLinus Torvalds /* this commit is done, exit */ 10281da177e4SLinus Torvalds if (atomic_read(&(jl->j_commit_left)) <= 0) { 10291da177e4SLinus Torvalds if (flushall) { 10301da177e4SLinus Torvalds atomic_set(&(jl->j_older_commits_done), 1); 10311da177e4SLinus Torvalds } 10321da177e4SLinus Torvalds up(&jl->j_commit_lock); 10331da177e4SLinus Torvalds goto put_jl; 10341da177e4SLinus Torvalds } 10351da177e4SLinus Torvalds 10361da177e4SLinus Torvalds if (!list_empty(&jl->j_bh_list)) { 10373d4492f8SChris Mason int ret; 10381da177e4SLinus Torvalds unlock_kernel(); 10393d4492f8SChris Mason ret = write_ordered_buffers(&journal->j_dirty_buffers_lock, 10401da177e4SLinus Torvalds journal, jl, &jl->j_bh_list); 10413d4492f8SChris Mason if (ret < 0 && retval == 0) 10423d4492f8SChris Mason retval = ret; 10431da177e4SLinus Torvalds lock_kernel(); 10441da177e4SLinus Torvalds } 10451da177e4SLinus Torvalds BUG_ON(!list_empty(&jl->j_bh_list)); 10461da177e4SLinus Torvalds /* 10471da177e4SLinus Torvalds * for the description block and all the log blocks, submit any buffers 1048e0e851cfSChris Mason * that haven't already reached the disk. Try to write at least 256 1049e0e851cfSChris Mason * log blocks. later on, we will only wait on blocks that correspond 1050e0e851cfSChris Mason * to this transaction, but while we're unplugging we might as well 1051e0e851cfSChris Mason * get a chunk of data on there. 10521da177e4SLinus Torvalds */ 10531da177e4SLinus Torvalds atomic_inc(&journal->j_async_throttle); 1054e0e851cfSChris Mason write_len = jl->j_len + 1; 1055e0e851cfSChris Mason if (write_len < 256) 1056e0e851cfSChris Mason write_len = 256; 1057e0e851cfSChris Mason for (i = 0 ; i < write_len ; i++) { 10581da177e4SLinus Torvalds bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) % 10591da177e4SLinus Torvalds SB_ONDISK_JOURNAL_SIZE(s); 10601da177e4SLinus Torvalds tbh = journal_find_get_block(s, bn); 1061e0e851cfSChris Mason if (tbh) { 1062e0e851cfSChris Mason if (buffer_dirty(tbh)) 1063e0e851cfSChris Mason ll_rw_block(WRITE, 1, &tbh) ; 10641da177e4SLinus Torvalds put_bh(tbh) ; 10651da177e4SLinus Torvalds } 1066e0e851cfSChris Mason } 10671da177e4SLinus Torvalds atomic_dec(&journal->j_async_throttle); 10681da177e4SLinus Torvalds 10695d5e8156SJeff Mahoney /* We're skipping the commit if there's an error */ 10705d5e8156SJeff Mahoney if (retval || reiserfs_is_journal_aborted(journal)) 10715d5e8156SJeff Mahoney barrier = 0; 10725d5e8156SJeff Mahoney 10731da177e4SLinus Torvalds /* wait on everything written so far before writing the commit 10741da177e4SLinus Torvalds * if we are in barrier mode, send the commit down now 10751da177e4SLinus Torvalds */ 10761da177e4SLinus Torvalds barrier = reiserfs_barrier_flush(s); 10771da177e4SLinus Torvalds if (barrier) { 10781da177e4SLinus Torvalds int ret; 10791da177e4SLinus Torvalds lock_buffer(jl->j_commit_bh); 10801da177e4SLinus Torvalds ret = submit_barrier_buffer(jl->j_commit_bh); 10811da177e4SLinus Torvalds if (ret == -EOPNOTSUPP) { 10821da177e4SLinus Torvalds set_buffer_uptodate(jl->j_commit_bh); 10831da177e4SLinus Torvalds disable_barrier(s); 10841da177e4SLinus Torvalds barrier = 0; 10851da177e4SLinus Torvalds } 10861da177e4SLinus Torvalds } 10871da177e4SLinus Torvalds for (i = 0; i < (jl->j_len + 1); i++) { 10881da177e4SLinus Torvalds bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + 10891da177e4SLinus Torvalds (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s); 10901da177e4SLinus Torvalds tbh = journal_find_get_block(s, bn); 10911da177e4SLinus Torvalds wait_on_buffer(tbh); 10921da177e4SLinus Torvalds // since we're using ll_rw_blk above, it might have skipped over 10931da177e4SLinus Torvalds // a locked buffer. Double check here 10941da177e4SLinus Torvalds // 10951da177e4SLinus Torvalds if (buffer_dirty(tbh)) /* redundant, sync_dirty_buffer() checks */ 10961da177e4SLinus Torvalds sync_dirty_buffer(tbh); 10971da177e4SLinus Torvalds if (unlikely(!buffer_uptodate(tbh))) { 10981da177e4SLinus Torvalds #ifdef CONFIG_REISERFS_CHECK 10991da177e4SLinus Torvalds reiserfs_warning(s, "journal-601, buffer write failed"); 11001da177e4SLinus Torvalds #endif 11011da177e4SLinus Torvalds retval = -EIO; 11021da177e4SLinus Torvalds } 11031da177e4SLinus Torvalds put_bh(tbh); /* once for journal_find_get_block */ 11041da177e4SLinus Torvalds put_bh(tbh); /* once due to original getblk in do_journal_end */ 11051da177e4SLinus Torvalds atomic_dec(&(jl->j_commit_left)); 11061da177e4SLinus Torvalds } 11071da177e4SLinus Torvalds 11081da177e4SLinus Torvalds BUG_ON(atomic_read(&(jl->j_commit_left)) != 1); 11091da177e4SLinus Torvalds 11101da177e4SLinus Torvalds if (!barrier) { 11115d5e8156SJeff Mahoney /* If there was a write error in the journal - we can't commit 11125d5e8156SJeff Mahoney * this transaction - it will be invalid and, if successful, 11135d5e8156SJeff Mahoney * will just end up propogating the write error out to 11145d5e8156SJeff Mahoney * the file system. */ 11155d5e8156SJeff Mahoney if (likely(!retval && !reiserfs_is_journal_aborted (journal))) { 11161da177e4SLinus Torvalds if (buffer_dirty(jl->j_commit_bh)) 11171da177e4SLinus Torvalds BUG(); 11181da177e4SLinus Torvalds mark_buffer_dirty(jl->j_commit_bh) ; 11191da177e4SLinus Torvalds sync_dirty_buffer(jl->j_commit_bh) ; 11205d5e8156SJeff Mahoney } 11211da177e4SLinus Torvalds } else 11221da177e4SLinus Torvalds wait_on_buffer(jl->j_commit_bh); 11231da177e4SLinus Torvalds 11241da177e4SLinus Torvalds check_barrier_completion(s, jl->j_commit_bh); 11251da177e4SLinus Torvalds 11261da177e4SLinus Torvalds /* If there was a write error in the journal - we can't commit this 11271da177e4SLinus Torvalds * transaction - it will be invalid and, if successful, will just end 11281da177e4SLinus Torvalds * up propogating the write error out to the filesystem. */ 11291da177e4SLinus Torvalds if (unlikely(!buffer_uptodate(jl->j_commit_bh))) { 11301da177e4SLinus Torvalds #ifdef CONFIG_REISERFS_CHECK 11311da177e4SLinus Torvalds reiserfs_warning(s, "journal-615: buffer write failed"); 11321da177e4SLinus Torvalds #endif 11331da177e4SLinus Torvalds retval = -EIO; 11341da177e4SLinus Torvalds } 11351da177e4SLinus Torvalds bforget(jl->j_commit_bh); 11361da177e4SLinus Torvalds if (journal->j_last_commit_id != 0 && 11371da177e4SLinus Torvalds (jl->j_trans_id - journal->j_last_commit_id) != 1) { 11381da177e4SLinus Torvalds reiserfs_warning(s, "clm-2200: last commit %lu, current %lu", 1139bd4c625cSLinus Torvalds journal->j_last_commit_id, jl->j_trans_id); 11401da177e4SLinus Torvalds } 11411da177e4SLinus Torvalds journal->j_last_commit_id = jl->j_trans_id; 11421da177e4SLinus Torvalds 11431da177e4SLinus Torvalds /* now, every commit block is on the disk. It is safe to allow blocks freed during this transaction to be reallocated */ 11441da177e4SLinus Torvalds cleanup_freed_for_journal_list(s, jl); 11451da177e4SLinus Torvalds 11461da177e4SLinus Torvalds retval = retval ? retval : journal->j_errno; 11471da177e4SLinus Torvalds 11481da177e4SLinus Torvalds /* mark the metadata dirty */ 11491da177e4SLinus Torvalds if (!retval) 11501da177e4SLinus Torvalds dirty_one_transaction(s, jl); 11511da177e4SLinus Torvalds atomic_dec(&(jl->j_commit_left)); 11521da177e4SLinus Torvalds 11531da177e4SLinus Torvalds if (flushall) { 11541da177e4SLinus Torvalds atomic_set(&(jl->j_older_commits_done), 1); 11551da177e4SLinus Torvalds } 11561da177e4SLinus Torvalds up(&jl->j_commit_lock); 11571da177e4SLinus Torvalds put_jl: 11581da177e4SLinus Torvalds put_journal_list(s, jl); 11591da177e4SLinus Torvalds 11601da177e4SLinus Torvalds if (retval) 1161bd4c625cSLinus Torvalds reiserfs_abort(s, retval, "Journal write error in %s", 1162bd4c625cSLinus Torvalds __FUNCTION__); 116322e2c507SJens Axboe put_fs_excl(); 11641da177e4SLinus Torvalds return retval; 11651da177e4SLinus Torvalds } 11661da177e4SLinus Torvalds 11671da177e4SLinus Torvalds /* 11681da177e4SLinus Torvalds ** flush_journal_list frequently needs to find a newer transaction for a given block. This does that, or 11691da177e4SLinus Torvalds ** returns NULL if it can't find anything 11701da177e4SLinus Torvalds */ 1171bd4c625cSLinus Torvalds static struct reiserfs_journal_list *find_newer_jl_for_cn(struct 1172bd4c625cSLinus Torvalds reiserfs_journal_cnode 1173bd4c625cSLinus Torvalds *cn) 1174bd4c625cSLinus Torvalds { 11751da177e4SLinus Torvalds struct super_block *sb = cn->sb; 11761da177e4SLinus Torvalds b_blocknr_t blocknr = cn->blocknr; 11771da177e4SLinus Torvalds 11781da177e4SLinus Torvalds cn = cn->hprev; 11791da177e4SLinus Torvalds while (cn) { 11801da177e4SLinus Torvalds if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) { 11811da177e4SLinus Torvalds return cn->jlist; 11821da177e4SLinus Torvalds } 11831da177e4SLinus Torvalds cn = cn->hprev; 11841da177e4SLinus Torvalds } 11851da177e4SLinus Torvalds return NULL; 11861da177e4SLinus Torvalds } 11871da177e4SLinus Torvalds 1188a3172027SChris Mason static int newer_jl_done(struct reiserfs_journal_cnode *cn) 1189a3172027SChris Mason { 1190a3172027SChris Mason struct super_block *sb = cn->sb; 1191a3172027SChris Mason b_blocknr_t blocknr = cn->blocknr; 1192a3172027SChris Mason 1193a3172027SChris Mason cn = cn->hprev; 1194a3172027SChris Mason while (cn) { 1195a3172027SChris Mason if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist && 1196a3172027SChris Mason atomic_read(&cn->jlist->j_commit_left) != 0) 1197a3172027SChris Mason return 0; 1198a3172027SChris Mason cn = cn->hprev; 1199a3172027SChris Mason } 1200a3172027SChris Mason return 1; 1201a3172027SChris Mason } 1202a3172027SChris Mason 1203bd4c625cSLinus Torvalds static void remove_journal_hash(struct super_block *, 1204bd4c625cSLinus Torvalds struct reiserfs_journal_cnode **, 1205bd4c625cSLinus Torvalds struct reiserfs_journal_list *, unsigned long, 1206bd4c625cSLinus Torvalds int); 12071da177e4SLinus Torvalds 12081da177e4SLinus Torvalds /* 12091da177e4SLinus Torvalds ** once all the real blocks have been flushed, it is safe to remove them from the 12101da177e4SLinus Torvalds ** journal list for this transaction. Aside from freeing the cnode, this also allows the 12111da177e4SLinus Torvalds ** block to be reallocated for data blocks if it had been deleted. 12121da177e4SLinus Torvalds */ 1213bd4c625cSLinus Torvalds static void remove_all_from_journal_list(struct super_block *p_s_sb, 1214bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl, 1215bd4c625cSLinus Torvalds int debug) 1216bd4c625cSLinus Torvalds { 12171da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 12181da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn, *last; 12191da177e4SLinus Torvalds cn = jl->j_realblock; 12201da177e4SLinus Torvalds 12211da177e4SLinus Torvalds /* which is better, to lock once around the whole loop, or 12221da177e4SLinus Torvalds ** to lock for each call to remove_journal_hash? 12231da177e4SLinus Torvalds */ 12241da177e4SLinus Torvalds while (cn) { 12251da177e4SLinus Torvalds if (cn->blocknr != 0) { 12261da177e4SLinus Torvalds if (debug) { 1227bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 1228bd4c625cSLinus Torvalds "block %u, bh is %d, state %ld", 1229bd4c625cSLinus Torvalds cn->blocknr, cn->bh ? 1 : 0, 1230bd4c625cSLinus Torvalds cn->state); 12311da177e4SLinus Torvalds } 12321da177e4SLinus Torvalds cn->state = 0; 1233bd4c625cSLinus Torvalds remove_journal_hash(p_s_sb, journal->j_list_hash_table, 1234bd4c625cSLinus Torvalds jl, cn->blocknr, 1); 12351da177e4SLinus Torvalds } 12361da177e4SLinus Torvalds last = cn; 12371da177e4SLinus Torvalds cn = cn->next; 12381da177e4SLinus Torvalds free_cnode(p_s_sb, last); 12391da177e4SLinus Torvalds } 12401da177e4SLinus Torvalds jl->j_realblock = NULL; 12411da177e4SLinus Torvalds } 12421da177e4SLinus Torvalds 12431da177e4SLinus Torvalds /* 12441da177e4SLinus Torvalds ** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block. 12451da177e4SLinus Torvalds ** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start 12461da177e4SLinus Torvalds ** releasing blocks in this transaction for reuse as data blocks. 12471da177e4SLinus Torvalds ** called by flush_journal_list, before it calls remove_all_from_journal_list 12481da177e4SLinus Torvalds ** 12491da177e4SLinus Torvalds */ 1250bd4c625cSLinus Torvalds static int _update_journal_header_block(struct super_block *p_s_sb, 1251bd4c625cSLinus Torvalds unsigned long offset, 1252bd4c625cSLinus Torvalds unsigned long trans_id) 1253bd4c625cSLinus Torvalds { 12541da177e4SLinus Torvalds struct reiserfs_journal_header *jh; 12551da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 12561da177e4SLinus Torvalds 12571da177e4SLinus Torvalds if (reiserfs_is_journal_aborted(journal)) 12581da177e4SLinus Torvalds return -EIO; 12591da177e4SLinus Torvalds 12601da177e4SLinus Torvalds if (trans_id >= journal->j_last_flush_trans_id) { 12611da177e4SLinus Torvalds if (buffer_locked((journal->j_header_bh))) { 12621da177e4SLinus Torvalds wait_on_buffer((journal->j_header_bh)); 12631da177e4SLinus Torvalds if (unlikely(!buffer_uptodate(journal->j_header_bh))) { 12641da177e4SLinus Torvalds #ifdef CONFIG_REISERFS_CHECK 1265bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 1266bd4c625cSLinus Torvalds "journal-699: buffer write failed"); 12671da177e4SLinus Torvalds #endif 12681da177e4SLinus Torvalds return -EIO; 12691da177e4SLinus Torvalds } 12701da177e4SLinus Torvalds } 12711da177e4SLinus Torvalds journal->j_last_flush_trans_id = trans_id; 12721da177e4SLinus Torvalds journal->j_first_unflushed_offset = offset; 1273bd4c625cSLinus Torvalds jh = (struct reiserfs_journal_header *)(journal->j_header_bh-> 1274bd4c625cSLinus Torvalds b_data); 12751da177e4SLinus Torvalds jh->j_last_flush_trans_id = cpu_to_le32(trans_id); 12761da177e4SLinus Torvalds jh->j_first_unflushed_offset = cpu_to_le32(offset); 12771da177e4SLinus Torvalds jh->j_mount_id = cpu_to_le32(journal->j_mount_id); 12781da177e4SLinus Torvalds 12791da177e4SLinus Torvalds if (reiserfs_barrier_flush(p_s_sb)) { 12801da177e4SLinus Torvalds int ret; 12811da177e4SLinus Torvalds lock_buffer(journal->j_header_bh); 12821da177e4SLinus Torvalds ret = submit_barrier_buffer(journal->j_header_bh); 12831da177e4SLinus Torvalds if (ret == -EOPNOTSUPP) { 12841da177e4SLinus Torvalds set_buffer_uptodate(journal->j_header_bh); 12851da177e4SLinus Torvalds disable_barrier(p_s_sb); 12861da177e4SLinus Torvalds goto sync; 12871da177e4SLinus Torvalds } 12881da177e4SLinus Torvalds wait_on_buffer(journal->j_header_bh); 12891da177e4SLinus Torvalds check_barrier_completion(p_s_sb, journal->j_header_bh); 12901da177e4SLinus Torvalds } else { 12911da177e4SLinus Torvalds sync: 12921da177e4SLinus Torvalds set_buffer_dirty(journal->j_header_bh); 12931da177e4SLinus Torvalds sync_dirty_buffer(journal->j_header_bh); 12941da177e4SLinus Torvalds } 12951da177e4SLinus Torvalds if (!buffer_uptodate(journal->j_header_bh)) { 1296bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 1297bd4c625cSLinus Torvalds "journal-837: IO error during journal replay"); 12981da177e4SLinus Torvalds return -EIO; 12991da177e4SLinus Torvalds } 13001da177e4SLinus Torvalds } 13011da177e4SLinus Torvalds return 0; 13021da177e4SLinus Torvalds } 13031da177e4SLinus Torvalds 13041da177e4SLinus Torvalds static int update_journal_header_block(struct super_block *p_s_sb, 13051da177e4SLinus Torvalds unsigned long offset, 1306bd4c625cSLinus Torvalds unsigned long trans_id) 1307bd4c625cSLinus Torvalds { 13081da177e4SLinus Torvalds return _update_journal_header_block(p_s_sb, offset, trans_id); 13091da177e4SLinus Torvalds } 1310bd4c625cSLinus Torvalds 13111da177e4SLinus Torvalds /* 13121da177e4SLinus Torvalds ** flush any and all journal lists older than you are 13131da177e4SLinus Torvalds ** can only be called from flush_journal_list 13141da177e4SLinus Torvalds */ 13151da177e4SLinus Torvalds static int flush_older_journal_lists(struct super_block *p_s_sb, 13161da177e4SLinus Torvalds struct reiserfs_journal_list *jl) 13171da177e4SLinus Torvalds { 13181da177e4SLinus Torvalds struct list_head *entry; 13191da177e4SLinus Torvalds struct reiserfs_journal_list *other_jl; 13201da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 13211da177e4SLinus Torvalds unsigned long trans_id = jl->j_trans_id; 13221da177e4SLinus Torvalds 13231da177e4SLinus Torvalds /* we know we are the only ones flushing things, no extra race 13241da177e4SLinus Torvalds * protection is required. 13251da177e4SLinus Torvalds */ 13261da177e4SLinus Torvalds restart: 13271da177e4SLinus Torvalds entry = journal->j_journal_list.next; 13281da177e4SLinus Torvalds /* Did we wrap? */ 13291da177e4SLinus Torvalds if (entry == &journal->j_journal_list) 13301da177e4SLinus Torvalds return 0; 13311da177e4SLinus Torvalds other_jl = JOURNAL_LIST_ENTRY(entry); 13321da177e4SLinus Torvalds if (other_jl->j_trans_id < trans_id) { 13331da177e4SLinus Torvalds BUG_ON(other_jl->j_refcount <= 0); 13341da177e4SLinus Torvalds /* do not flush all */ 13351da177e4SLinus Torvalds flush_journal_list(p_s_sb, other_jl, 0); 13361da177e4SLinus Torvalds 13371da177e4SLinus Torvalds /* other_jl is now deleted from the list */ 13381da177e4SLinus Torvalds goto restart; 13391da177e4SLinus Torvalds } 13401da177e4SLinus Torvalds return 0; 13411da177e4SLinus Torvalds } 13421da177e4SLinus Torvalds 13431da177e4SLinus Torvalds static void del_from_work_list(struct super_block *s, 1344bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl) 1345bd4c625cSLinus Torvalds { 13461da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 13471da177e4SLinus Torvalds if (!list_empty(&jl->j_working_list)) { 13481da177e4SLinus Torvalds list_del_init(&jl->j_working_list); 13491da177e4SLinus Torvalds journal->j_num_work_lists--; 13501da177e4SLinus Torvalds } 13511da177e4SLinus Torvalds } 13521da177e4SLinus Torvalds 13531da177e4SLinus Torvalds /* flush a journal list, both commit and real blocks 13541da177e4SLinus Torvalds ** 13551da177e4SLinus Torvalds ** always set flushall to 1, unless you are calling from inside 13561da177e4SLinus Torvalds ** flush_journal_list 13571da177e4SLinus Torvalds ** 13581da177e4SLinus Torvalds ** IMPORTANT. This can only be called while there are no journal writers, 13591da177e4SLinus Torvalds ** and the journal is locked. That means it can only be called from 13601da177e4SLinus Torvalds ** do_journal_end, or by journal_release 13611da177e4SLinus Torvalds */ 13621da177e4SLinus Torvalds static int flush_journal_list(struct super_block *s, 1363bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl, int flushall) 1364bd4c625cSLinus Torvalds { 13651da177e4SLinus Torvalds struct reiserfs_journal_list *pjl; 13661da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn, *last; 13671da177e4SLinus Torvalds int count; 13681da177e4SLinus Torvalds int was_jwait = 0; 13691da177e4SLinus Torvalds int was_dirty = 0; 13701da177e4SLinus Torvalds struct buffer_head *saved_bh; 13711da177e4SLinus Torvalds unsigned long j_len_saved = jl->j_len; 13721da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 13731da177e4SLinus Torvalds int err = 0; 13741da177e4SLinus Torvalds 13751da177e4SLinus Torvalds BUG_ON(j_len_saved <= 0); 13761da177e4SLinus Torvalds 13771da177e4SLinus Torvalds if (atomic_read(&journal->j_wcount) != 0) { 1378bd4c625cSLinus Torvalds reiserfs_warning(s, 1379bd4c625cSLinus Torvalds "clm-2048: flush_journal_list called with wcount %d", 13801da177e4SLinus Torvalds atomic_read(&journal->j_wcount)); 13811da177e4SLinus Torvalds } 13821da177e4SLinus Torvalds BUG_ON(jl->j_trans_id == 0); 13831da177e4SLinus Torvalds 13841da177e4SLinus Torvalds /* if flushall == 0, the lock is already held */ 13851da177e4SLinus Torvalds if (flushall) { 13861da177e4SLinus Torvalds down(&journal->j_flush_sem); 13871da177e4SLinus Torvalds } else if (!down_trylock(&journal->j_flush_sem)) { 13881da177e4SLinus Torvalds BUG(); 13891da177e4SLinus Torvalds } 13901da177e4SLinus Torvalds 13911da177e4SLinus Torvalds count = 0; 13921da177e4SLinus Torvalds if (j_len_saved > journal->j_trans_max) { 1393bd4c625cSLinus Torvalds reiserfs_panic(s, 1394bd4c625cSLinus Torvalds "journal-715: flush_journal_list, length is %lu, trans id %lu\n", 1395bd4c625cSLinus Torvalds j_len_saved, jl->j_trans_id); 13961da177e4SLinus Torvalds return 0; 13971da177e4SLinus Torvalds } 13981da177e4SLinus Torvalds 139922e2c507SJens Axboe get_fs_excl(); 140022e2c507SJens Axboe 14011da177e4SLinus Torvalds /* if all the work is already done, get out of here */ 14021da177e4SLinus Torvalds if (atomic_read(&(jl->j_nonzerolen)) <= 0 && 14031da177e4SLinus Torvalds atomic_read(&(jl->j_commit_left)) <= 0) { 14041da177e4SLinus Torvalds goto flush_older_and_return; 14051da177e4SLinus Torvalds } 14061da177e4SLinus Torvalds 14071da177e4SLinus Torvalds /* start by putting the commit list on disk. This will also flush 14081da177e4SLinus Torvalds ** the commit lists of any olders transactions 14091da177e4SLinus Torvalds */ 14101da177e4SLinus Torvalds flush_commit_list(s, jl, 1); 14111da177e4SLinus Torvalds 1412bd4c625cSLinus Torvalds if (!(jl->j_state & LIST_DIRTY) 1413bd4c625cSLinus Torvalds && !reiserfs_is_journal_aborted(journal)) 14141da177e4SLinus Torvalds BUG(); 14151da177e4SLinus Torvalds 14161da177e4SLinus Torvalds /* are we done now? */ 14171da177e4SLinus Torvalds if (atomic_read(&(jl->j_nonzerolen)) <= 0 && 14181da177e4SLinus Torvalds atomic_read(&(jl->j_commit_left)) <= 0) { 14191da177e4SLinus Torvalds goto flush_older_and_return; 14201da177e4SLinus Torvalds } 14211da177e4SLinus Torvalds 14221da177e4SLinus Torvalds /* loop through each cnode, see if we need to write it, 14231da177e4SLinus Torvalds ** or wait on a more recent transaction, or just ignore it 14241da177e4SLinus Torvalds */ 14251da177e4SLinus Torvalds if (atomic_read(&(journal->j_wcount)) != 0) { 1426bd4c625cSLinus Torvalds reiserfs_panic(s, 1427bd4c625cSLinus Torvalds "journal-844: panic journal list is flushing, wcount is not 0\n"); 14281da177e4SLinus Torvalds } 14291da177e4SLinus Torvalds cn = jl->j_realblock; 14301da177e4SLinus Torvalds while (cn) { 14311da177e4SLinus Torvalds was_jwait = 0; 14321da177e4SLinus Torvalds was_dirty = 0; 14331da177e4SLinus Torvalds saved_bh = NULL; 14341da177e4SLinus Torvalds /* blocknr of 0 is no longer in the hash, ignore it */ 14351da177e4SLinus Torvalds if (cn->blocknr == 0) { 14361da177e4SLinus Torvalds goto free_cnode; 14371da177e4SLinus Torvalds } 14381da177e4SLinus Torvalds 14391da177e4SLinus Torvalds /* This transaction failed commit. Don't write out to the disk */ 14401da177e4SLinus Torvalds if (!(jl->j_state & LIST_DIRTY)) 14411da177e4SLinus Torvalds goto free_cnode; 14421da177e4SLinus Torvalds 14431da177e4SLinus Torvalds pjl = find_newer_jl_for_cn(cn); 14441da177e4SLinus Torvalds /* the order is important here. We check pjl to make sure we 14451da177e4SLinus Torvalds ** don't clear BH_JDirty_wait if we aren't the one writing this 14461da177e4SLinus Torvalds ** block to disk 14471da177e4SLinus Torvalds */ 14481da177e4SLinus Torvalds if (!pjl && cn->bh) { 14491da177e4SLinus Torvalds saved_bh = cn->bh; 14501da177e4SLinus Torvalds 14511da177e4SLinus Torvalds /* we do this to make sure nobody releases the buffer while 14521da177e4SLinus Torvalds ** we are working with it 14531da177e4SLinus Torvalds */ 14541da177e4SLinus Torvalds get_bh(saved_bh); 14551da177e4SLinus Torvalds 14561da177e4SLinus Torvalds if (buffer_journal_dirty(saved_bh)) { 14571da177e4SLinus Torvalds BUG_ON(!can_dirty(cn)); 14581da177e4SLinus Torvalds was_jwait = 1; 14591da177e4SLinus Torvalds was_dirty = 1; 14601da177e4SLinus Torvalds } else if (can_dirty(cn)) { 14611da177e4SLinus Torvalds /* everything with !pjl && jwait should be writable */ 14621da177e4SLinus Torvalds BUG(); 14631da177e4SLinus Torvalds } 14641da177e4SLinus Torvalds } 14651da177e4SLinus Torvalds 14661da177e4SLinus Torvalds /* if someone has this block in a newer transaction, just make 14670779bf2dSMatt LaPlante ** sure they are committed, and don't try writing it to disk 14681da177e4SLinus Torvalds */ 14691da177e4SLinus Torvalds if (pjl) { 14701da177e4SLinus Torvalds if (atomic_read(&pjl->j_commit_left)) 14711da177e4SLinus Torvalds flush_commit_list(s, pjl, 1); 14721da177e4SLinus Torvalds goto free_cnode; 14731da177e4SLinus Torvalds } 14741da177e4SLinus Torvalds 14751da177e4SLinus Torvalds /* bh == NULL when the block got to disk on its own, OR, 14761da177e4SLinus Torvalds ** the block got freed in a future transaction 14771da177e4SLinus Torvalds */ 14781da177e4SLinus Torvalds if (saved_bh == NULL) { 14791da177e4SLinus Torvalds goto free_cnode; 14801da177e4SLinus Torvalds } 14811da177e4SLinus Torvalds 14821da177e4SLinus Torvalds /* this should never happen. kupdate_one_transaction has this list 14831da177e4SLinus Torvalds ** locked while it works, so we should never see a buffer here that 14841da177e4SLinus Torvalds ** is not marked JDirty_wait 14851da177e4SLinus Torvalds */ 14861da177e4SLinus Torvalds if ((!was_jwait) && !buffer_locked(saved_bh)) { 1487bd4c625cSLinus Torvalds reiserfs_warning(s, 1488bd4c625cSLinus Torvalds "journal-813: BAD! buffer %llu %cdirty %cjwait, " 14891da177e4SLinus Torvalds "not in a newer tranasction", 1490bd4c625cSLinus Torvalds (unsigned long long)saved_bh-> 1491bd4c625cSLinus Torvalds b_blocknr, was_dirty ? ' ' : '!', 1492bd4c625cSLinus Torvalds was_jwait ? ' ' : '!'); 14931da177e4SLinus Torvalds } 14941da177e4SLinus Torvalds if (was_dirty) { 14951da177e4SLinus Torvalds /* we inc again because saved_bh gets decremented at free_cnode */ 14961da177e4SLinus Torvalds get_bh(saved_bh); 14971da177e4SLinus Torvalds set_bit(BLOCK_NEEDS_FLUSH, &cn->state); 14981da177e4SLinus Torvalds lock_buffer(saved_bh); 14991da177e4SLinus Torvalds BUG_ON(cn->blocknr != saved_bh->b_blocknr); 15001da177e4SLinus Torvalds if (buffer_dirty(saved_bh)) 15011da177e4SLinus Torvalds submit_logged_buffer(saved_bh); 15021da177e4SLinus Torvalds else 15031da177e4SLinus Torvalds unlock_buffer(saved_bh); 15041da177e4SLinus Torvalds count++; 15051da177e4SLinus Torvalds } else { 1506bd4c625cSLinus Torvalds reiserfs_warning(s, 1507bd4c625cSLinus Torvalds "clm-2082: Unable to flush buffer %llu in %s", 1508bd4c625cSLinus Torvalds (unsigned long long)saved_bh-> 1509bd4c625cSLinus Torvalds b_blocknr, __FUNCTION__); 15101da177e4SLinus Torvalds } 15111da177e4SLinus Torvalds free_cnode: 15121da177e4SLinus Torvalds last = cn; 15131da177e4SLinus Torvalds cn = cn->next; 15141da177e4SLinus Torvalds if (saved_bh) { 15151da177e4SLinus Torvalds /* we incremented this to keep others from taking the buffer head away */ 15161da177e4SLinus Torvalds put_bh(saved_bh); 15171da177e4SLinus Torvalds if (atomic_read(&(saved_bh->b_count)) < 0) { 1518bd4c625cSLinus Torvalds reiserfs_warning(s, 1519bd4c625cSLinus Torvalds "journal-945: saved_bh->b_count < 0"); 15201da177e4SLinus Torvalds } 15211da177e4SLinus Torvalds } 15221da177e4SLinus Torvalds } 15231da177e4SLinus Torvalds if (count > 0) { 15241da177e4SLinus Torvalds cn = jl->j_realblock; 15251da177e4SLinus Torvalds while (cn) { 15261da177e4SLinus Torvalds if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) { 15271da177e4SLinus Torvalds if (!cn->bh) { 1528bd4c625cSLinus Torvalds reiserfs_panic(s, 1529bd4c625cSLinus Torvalds "journal-1011: cn->bh is NULL\n"); 15301da177e4SLinus Torvalds } 15311da177e4SLinus Torvalds wait_on_buffer(cn->bh); 15321da177e4SLinus Torvalds if (!cn->bh) { 1533bd4c625cSLinus Torvalds reiserfs_panic(s, 1534bd4c625cSLinus Torvalds "journal-1012: cn->bh is NULL\n"); 15351da177e4SLinus Torvalds } 15361da177e4SLinus Torvalds if (unlikely(!buffer_uptodate(cn->bh))) { 15371da177e4SLinus Torvalds #ifdef CONFIG_REISERFS_CHECK 1538bd4c625cSLinus Torvalds reiserfs_warning(s, 1539bd4c625cSLinus Torvalds "journal-949: buffer write failed\n"); 15401da177e4SLinus Torvalds #endif 15411da177e4SLinus Torvalds err = -EIO; 15421da177e4SLinus Torvalds } 15431da177e4SLinus Torvalds /* note, we must clear the JDirty_wait bit after the up to date 15441da177e4SLinus Torvalds ** check, otherwise we race against our flushpage routine 15451da177e4SLinus Torvalds */ 1546bd4c625cSLinus Torvalds BUG_ON(!test_clear_buffer_journal_dirty 1547bd4c625cSLinus Torvalds (cn->bh)); 15481da177e4SLinus Torvalds 15491da177e4SLinus Torvalds /* undo the inc from journal_mark_dirty */ 15501da177e4SLinus Torvalds put_bh(cn->bh); 15511da177e4SLinus Torvalds brelse(cn->bh); 15521da177e4SLinus Torvalds } 15531da177e4SLinus Torvalds cn = cn->next; 15541da177e4SLinus Torvalds } 15551da177e4SLinus Torvalds } 15561da177e4SLinus Torvalds 15571da177e4SLinus Torvalds if (err) 1558bd4c625cSLinus Torvalds reiserfs_abort(s, -EIO, 1559bd4c625cSLinus Torvalds "Write error while pushing transaction to disk in %s", 1560bd4c625cSLinus Torvalds __FUNCTION__); 15611da177e4SLinus Torvalds flush_older_and_return: 15621da177e4SLinus Torvalds 15631da177e4SLinus Torvalds /* before we can update the journal header block, we _must_ flush all 15641da177e4SLinus Torvalds ** real blocks from all older transactions to disk. This is because 15651da177e4SLinus Torvalds ** once the header block is updated, this transaction will not be 15661da177e4SLinus Torvalds ** replayed after a crash 15671da177e4SLinus Torvalds */ 15681da177e4SLinus Torvalds if (flushall) { 15691da177e4SLinus Torvalds flush_older_journal_lists(s, jl); 15701da177e4SLinus Torvalds } 15711da177e4SLinus Torvalds 15721da177e4SLinus Torvalds err = journal->j_errno; 15731da177e4SLinus Torvalds /* before we can remove everything from the hash tables for this 15741da177e4SLinus Torvalds ** transaction, we must make sure it can never be replayed 15751da177e4SLinus Torvalds ** 15761da177e4SLinus Torvalds ** since we are only called from do_journal_end, we know for sure there 15771da177e4SLinus Torvalds ** are no allocations going on while we are flushing journal lists. So, 15781da177e4SLinus Torvalds ** we only need to update the journal header block for the last list 15791da177e4SLinus Torvalds ** being flushed 15801da177e4SLinus Torvalds */ 15811da177e4SLinus Torvalds if (!err && flushall) { 1582bd4c625cSLinus Torvalds err = 1583bd4c625cSLinus Torvalds update_journal_header_block(s, 1584bd4c625cSLinus Torvalds (jl->j_start + jl->j_len + 1585bd4c625cSLinus Torvalds 2) % SB_ONDISK_JOURNAL_SIZE(s), 1586bd4c625cSLinus Torvalds jl->j_trans_id); 15871da177e4SLinus Torvalds if (err) 1588bd4c625cSLinus Torvalds reiserfs_abort(s, -EIO, 1589bd4c625cSLinus Torvalds "Write error while updating journal header in %s", 1590bd4c625cSLinus Torvalds __FUNCTION__); 15911da177e4SLinus Torvalds } 15921da177e4SLinus Torvalds remove_all_from_journal_list(s, jl, 0); 15931da177e4SLinus Torvalds list_del_init(&jl->j_list); 15941da177e4SLinus Torvalds journal->j_num_lists--; 15951da177e4SLinus Torvalds del_from_work_list(s, jl); 15961da177e4SLinus Torvalds 15971da177e4SLinus Torvalds if (journal->j_last_flush_id != 0 && 15981da177e4SLinus Torvalds (jl->j_trans_id - journal->j_last_flush_id) != 1) { 15991da177e4SLinus Torvalds reiserfs_warning(s, "clm-2201: last flush %lu, current %lu", 1600bd4c625cSLinus Torvalds journal->j_last_flush_id, jl->j_trans_id); 16011da177e4SLinus Torvalds } 16021da177e4SLinus Torvalds journal->j_last_flush_id = jl->j_trans_id; 16031da177e4SLinus Torvalds 16041da177e4SLinus Torvalds /* not strictly required since we are freeing the list, but it should 16051da177e4SLinus Torvalds * help find code using dead lists later on 16061da177e4SLinus Torvalds */ 16071da177e4SLinus Torvalds jl->j_len = 0; 16081da177e4SLinus Torvalds atomic_set(&(jl->j_nonzerolen), 0); 16091da177e4SLinus Torvalds jl->j_start = 0; 16101da177e4SLinus Torvalds jl->j_realblock = NULL; 16111da177e4SLinus Torvalds jl->j_commit_bh = NULL; 16121da177e4SLinus Torvalds jl->j_trans_id = 0; 16131da177e4SLinus Torvalds jl->j_state = 0; 16141da177e4SLinus Torvalds put_journal_list(s, jl); 16151da177e4SLinus Torvalds if (flushall) 16161da177e4SLinus Torvalds up(&journal->j_flush_sem); 161722e2c507SJens Axboe put_fs_excl(); 16181da177e4SLinus Torvalds return err; 16191da177e4SLinus Torvalds } 16201da177e4SLinus Torvalds 1621a3172027SChris Mason static int test_transaction(struct super_block *s, 1622a3172027SChris Mason struct reiserfs_journal_list *jl) 1623a3172027SChris Mason { 1624a3172027SChris Mason struct reiserfs_journal_cnode *cn; 1625a3172027SChris Mason 1626a3172027SChris Mason if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) 1627a3172027SChris Mason return 1; 1628a3172027SChris Mason 1629a3172027SChris Mason cn = jl->j_realblock; 1630a3172027SChris Mason while (cn) { 1631a3172027SChris Mason /* if the blocknr == 0, this has been cleared from the hash, 1632a3172027SChris Mason ** skip it 1633a3172027SChris Mason */ 1634a3172027SChris Mason if (cn->blocknr == 0) { 1635a3172027SChris Mason goto next; 1636a3172027SChris Mason } 1637a3172027SChris Mason if (cn->bh && !newer_jl_done(cn)) 1638a3172027SChris Mason return 0; 1639a3172027SChris Mason next: 1640a3172027SChris Mason cn = cn->next; 1641a3172027SChris Mason cond_resched(); 1642a3172027SChris Mason } 1643a3172027SChris Mason return 0; 1644a3172027SChris Mason } 1645a3172027SChris Mason 16461da177e4SLinus Torvalds static int write_one_transaction(struct super_block *s, 16471da177e4SLinus Torvalds struct reiserfs_journal_list *jl, 16481da177e4SLinus Torvalds struct buffer_chunk *chunk) 16491da177e4SLinus Torvalds { 16501da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 16511da177e4SLinus Torvalds int ret = 0; 16521da177e4SLinus Torvalds 16531da177e4SLinus Torvalds jl->j_state |= LIST_TOUCHED; 16541da177e4SLinus Torvalds del_from_work_list(s, jl); 16551da177e4SLinus Torvalds if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) { 16561da177e4SLinus Torvalds return 0; 16571da177e4SLinus Torvalds } 16581da177e4SLinus Torvalds 16591da177e4SLinus Torvalds cn = jl->j_realblock; 16601da177e4SLinus Torvalds while (cn) { 16611da177e4SLinus Torvalds /* if the blocknr == 0, this has been cleared from the hash, 16621da177e4SLinus Torvalds ** skip it 16631da177e4SLinus Torvalds */ 16641da177e4SLinus Torvalds if (cn->blocknr == 0) { 16651da177e4SLinus Torvalds goto next; 16661da177e4SLinus Torvalds } 16671da177e4SLinus Torvalds if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) { 16681da177e4SLinus Torvalds struct buffer_head *tmp_bh; 16691da177e4SLinus Torvalds /* we can race against journal_mark_freed when we try 16701da177e4SLinus Torvalds * to lock_buffer(cn->bh), so we have to inc the buffer 16711da177e4SLinus Torvalds * count, and recheck things after locking 16721da177e4SLinus Torvalds */ 16731da177e4SLinus Torvalds tmp_bh = cn->bh; 16741da177e4SLinus Torvalds get_bh(tmp_bh); 16751da177e4SLinus Torvalds lock_buffer(tmp_bh); 16761da177e4SLinus Torvalds if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) { 16771da177e4SLinus Torvalds if (!buffer_journal_dirty(tmp_bh) || 16781da177e4SLinus Torvalds buffer_journal_prepared(tmp_bh)) 16791da177e4SLinus Torvalds BUG(); 16801da177e4SLinus Torvalds add_to_chunk(chunk, tmp_bh, NULL, write_chunk); 16811da177e4SLinus Torvalds ret++; 16821da177e4SLinus Torvalds } else { 16831da177e4SLinus Torvalds /* note, cn->bh might be null now */ 16841da177e4SLinus Torvalds unlock_buffer(tmp_bh); 16851da177e4SLinus Torvalds } 16861da177e4SLinus Torvalds put_bh(tmp_bh); 16871da177e4SLinus Torvalds } 16881da177e4SLinus Torvalds next: 16891da177e4SLinus Torvalds cn = cn->next; 16901da177e4SLinus Torvalds cond_resched(); 16911da177e4SLinus Torvalds } 16921da177e4SLinus Torvalds return ret; 16931da177e4SLinus Torvalds } 16941da177e4SLinus Torvalds 16951da177e4SLinus Torvalds /* used by flush_commit_list */ 16961da177e4SLinus Torvalds static int dirty_one_transaction(struct super_block *s, 16971da177e4SLinus Torvalds struct reiserfs_journal_list *jl) 16981da177e4SLinus Torvalds { 16991da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 17001da177e4SLinus Torvalds struct reiserfs_journal_list *pjl; 17011da177e4SLinus Torvalds int ret = 0; 17021da177e4SLinus Torvalds 17031da177e4SLinus Torvalds jl->j_state |= LIST_DIRTY; 17041da177e4SLinus Torvalds cn = jl->j_realblock; 17051da177e4SLinus Torvalds while (cn) { 17061da177e4SLinus Torvalds /* look for a more recent transaction that logged this 17071da177e4SLinus Torvalds ** buffer. Only the most recent transaction with a buffer in 17081da177e4SLinus Torvalds ** it is allowed to send that buffer to disk 17091da177e4SLinus Torvalds */ 17101da177e4SLinus Torvalds pjl = find_newer_jl_for_cn(cn); 1711bd4c625cSLinus Torvalds if (!pjl && cn->blocknr && cn->bh 1712bd4c625cSLinus Torvalds && buffer_journal_dirty(cn->bh)) { 17131da177e4SLinus Torvalds BUG_ON(!can_dirty(cn)); 17141da177e4SLinus Torvalds /* if the buffer is prepared, it will either be logged 17151da177e4SLinus Torvalds * or restored. If restored, we need to make sure 17161da177e4SLinus Torvalds * it actually gets marked dirty 17171da177e4SLinus Torvalds */ 17181da177e4SLinus Torvalds clear_buffer_journal_new(cn->bh); 17191da177e4SLinus Torvalds if (buffer_journal_prepared(cn->bh)) { 17201da177e4SLinus Torvalds set_buffer_journal_restore_dirty(cn->bh); 17211da177e4SLinus Torvalds } else { 17221da177e4SLinus Torvalds set_buffer_journal_test(cn->bh); 17231da177e4SLinus Torvalds mark_buffer_dirty(cn->bh); 17241da177e4SLinus Torvalds } 17251da177e4SLinus Torvalds } 17261da177e4SLinus Torvalds cn = cn->next; 17271da177e4SLinus Torvalds } 17281da177e4SLinus Torvalds return ret; 17291da177e4SLinus Torvalds } 17301da177e4SLinus Torvalds 17311da177e4SLinus Torvalds static int kupdate_transactions(struct super_block *s, 17321da177e4SLinus Torvalds struct reiserfs_journal_list *jl, 17331da177e4SLinus Torvalds struct reiserfs_journal_list **next_jl, 17341da177e4SLinus Torvalds unsigned long *next_trans_id, 1735bd4c625cSLinus Torvalds int num_blocks, int num_trans) 1736bd4c625cSLinus Torvalds { 17371da177e4SLinus Torvalds int ret = 0; 17381da177e4SLinus Torvalds int written = 0; 17391da177e4SLinus Torvalds int transactions_flushed = 0; 17401da177e4SLinus Torvalds unsigned long orig_trans_id = jl->j_trans_id; 17411da177e4SLinus Torvalds struct buffer_chunk chunk; 17421da177e4SLinus Torvalds struct list_head *entry; 17431da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 17441da177e4SLinus Torvalds chunk.nr = 0; 17451da177e4SLinus Torvalds 17461da177e4SLinus Torvalds down(&journal->j_flush_sem); 17471da177e4SLinus Torvalds if (!journal_list_still_alive(s, orig_trans_id)) { 17481da177e4SLinus Torvalds goto done; 17491da177e4SLinus Torvalds } 17501da177e4SLinus Torvalds 17511da177e4SLinus Torvalds /* we've got j_flush_sem held, nobody is going to delete any 17521da177e4SLinus Torvalds * of these lists out from underneath us 17531da177e4SLinus Torvalds */ 17541da177e4SLinus Torvalds while ((num_trans && transactions_flushed < num_trans) || 17551da177e4SLinus Torvalds (!num_trans && written < num_blocks)) { 17561da177e4SLinus Torvalds 17571da177e4SLinus Torvalds if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) || 1758bd4c625cSLinus Torvalds atomic_read(&jl->j_commit_left) 1759bd4c625cSLinus Torvalds || !(jl->j_state & LIST_DIRTY)) { 17601da177e4SLinus Torvalds del_from_work_list(s, jl); 17611da177e4SLinus Torvalds break; 17621da177e4SLinus Torvalds } 17631da177e4SLinus Torvalds ret = write_one_transaction(s, jl, &chunk); 17641da177e4SLinus Torvalds 17651da177e4SLinus Torvalds if (ret < 0) 17661da177e4SLinus Torvalds goto done; 17671da177e4SLinus Torvalds transactions_flushed++; 17681da177e4SLinus Torvalds written += ret; 17691da177e4SLinus Torvalds entry = jl->j_list.next; 17701da177e4SLinus Torvalds 17711da177e4SLinus Torvalds /* did we wrap? */ 17721da177e4SLinus Torvalds if (entry == &journal->j_journal_list) { 17731da177e4SLinus Torvalds break; 17741da177e4SLinus Torvalds } 17751da177e4SLinus Torvalds jl = JOURNAL_LIST_ENTRY(entry); 17761da177e4SLinus Torvalds 17771da177e4SLinus Torvalds /* don't bother with older transactions */ 17781da177e4SLinus Torvalds if (jl->j_trans_id <= orig_trans_id) 17791da177e4SLinus Torvalds break; 17801da177e4SLinus Torvalds } 17811da177e4SLinus Torvalds if (chunk.nr) { 17821da177e4SLinus Torvalds write_chunk(&chunk); 17831da177e4SLinus Torvalds } 17841da177e4SLinus Torvalds 17851da177e4SLinus Torvalds done: 17861da177e4SLinus Torvalds up(&journal->j_flush_sem); 17871da177e4SLinus Torvalds return ret; 17881da177e4SLinus Torvalds } 17891da177e4SLinus Torvalds 17901da177e4SLinus Torvalds /* for o_sync and fsync heavy applications, they tend to use 17911da177e4SLinus Torvalds ** all the journa list slots with tiny transactions. These 17921da177e4SLinus Torvalds ** trigger lots and lots of calls to update the header block, which 17931da177e4SLinus Torvalds ** adds seeks and slows things down. 17941da177e4SLinus Torvalds ** 17951da177e4SLinus Torvalds ** This function tries to clear out a large chunk of the journal lists 17961da177e4SLinus Torvalds ** at once, which makes everything faster since only the newest journal 17971da177e4SLinus Torvalds ** list updates the header block 17981da177e4SLinus Torvalds */ 17991da177e4SLinus Torvalds static int flush_used_journal_lists(struct super_block *s, 1800bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl) 1801bd4c625cSLinus Torvalds { 18021da177e4SLinus Torvalds unsigned long len = 0; 18031da177e4SLinus Torvalds unsigned long cur_len; 18041da177e4SLinus Torvalds int ret; 18051da177e4SLinus Torvalds int i; 18061da177e4SLinus Torvalds int limit = 256; 18071da177e4SLinus Torvalds struct reiserfs_journal_list *tjl; 18081da177e4SLinus Torvalds struct reiserfs_journal_list *flush_jl; 18091da177e4SLinus Torvalds unsigned long trans_id; 18101da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 18111da177e4SLinus Torvalds 18121da177e4SLinus Torvalds flush_jl = tjl = jl; 18131da177e4SLinus Torvalds 18141da177e4SLinus Torvalds /* in data logging mode, try harder to flush a lot of blocks */ 18151da177e4SLinus Torvalds if (reiserfs_data_log(s)) 18161da177e4SLinus Torvalds limit = 1024; 18171da177e4SLinus Torvalds /* flush for 256 transactions or limit blocks, whichever comes first */ 18181da177e4SLinus Torvalds for (i = 0; i < 256 && len < limit; i++) { 18191da177e4SLinus Torvalds if (atomic_read(&tjl->j_commit_left) || 18201da177e4SLinus Torvalds tjl->j_trans_id < jl->j_trans_id) { 18211da177e4SLinus Torvalds break; 18221da177e4SLinus Torvalds } 18231da177e4SLinus Torvalds cur_len = atomic_read(&tjl->j_nonzerolen); 18241da177e4SLinus Torvalds if (cur_len > 0) { 18251da177e4SLinus Torvalds tjl->j_state &= ~LIST_TOUCHED; 18261da177e4SLinus Torvalds } 18271da177e4SLinus Torvalds len += cur_len; 18281da177e4SLinus Torvalds flush_jl = tjl; 18291da177e4SLinus Torvalds if (tjl->j_list.next == &journal->j_journal_list) 18301da177e4SLinus Torvalds break; 18311da177e4SLinus Torvalds tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next); 18321da177e4SLinus Torvalds } 18331da177e4SLinus Torvalds /* try to find a group of blocks we can flush across all the 18341da177e4SLinus Torvalds ** transactions, but only bother if we've actually spanned 18351da177e4SLinus Torvalds ** across multiple lists 18361da177e4SLinus Torvalds */ 18371da177e4SLinus Torvalds if (flush_jl != jl) { 18381da177e4SLinus Torvalds ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i); 18391da177e4SLinus Torvalds } 18401da177e4SLinus Torvalds flush_journal_list(s, flush_jl, 1); 18411da177e4SLinus Torvalds return 0; 18421da177e4SLinus Torvalds } 18431da177e4SLinus Torvalds 18441da177e4SLinus Torvalds /* 18451da177e4SLinus Torvalds ** removes any nodes in table with name block and dev as bh. 18461da177e4SLinus Torvalds ** only touchs the hnext and hprev pointers. 18471da177e4SLinus Torvalds */ 18481da177e4SLinus Torvalds void remove_journal_hash(struct super_block *sb, 18491da177e4SLinus Torvalds struct reiserfs_journal_cnode **table, 18501da177e4SLinus Torvalds struct reiserfs_journal_list *jl, 18511da177e4SLinus Torvalds unsigned long block, int remove_freed) 18521da177e4SLinus Torvalds { 18531da177e4SLinus Torvalds struct reiserfs_journal_cnode *cur; 18541da177e4SLinus Torvalds struct reiserfs_journal_cnode **head; 18551da177e4SLinus Torvalds 18561da177e4SLinus Torvalds head = &(journal_hash(table, sb, block)); 18571da177e4SLinus Torvalds if (!head) { 18581da177e4SLinus Torvalds return; 18591da177e4SLinus Torvalds } 18601da177e4SLinus Torvalds cur = *head; 18611da177e4SLinus Torvalds while (cur) { 1862bd4c625cSLinus Torvalds if (cur->blocknr == block && cur->sb == sb 1863bd4c625cSLinus Torvalds && (jl == NULL || jl == cur->jlist) 1864bd4c625cSLinus Torvalds && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) { 18651da177e4SLinus Torvalds if (cur->hnext) { 18661da177e4SLinus Torvalds cur->hnext->hprev = cur->hprev; 18671da177e4SLinus Torvalds } 18681da177e4SLinus Torvalds if (cur->hprev) { 18691da177e4SLinus Torvalds cur->hprev->hnext = cur->hnext; 18701da177e4SLinus Torvalds } else { 18711da177e4SLinus Torvalds *head = cur->hnext; 18721da177e4SLinus Torvalds } 18731da177e4SLinus Torvalds cur->blocknr = 0; 18741da177e4SLinus Torvalds cur->sb = NULL; 18751da177e4SLinus Torvalds cur->state = 0; 18761da177e4SLinus Torvalds if (cur->bh && cur->jlist) /* anybody who clears the cur->bh will also dec the nonzerolen */ 18771da177e4SLinus Torvalds atomic_dec(&(cur->jlist->j_nonzerolen)); 18781da177e4SLinus Torvalds cur->bh = NULL; 18791da177e4SLinus Torvalds cur->jlist = NULL; 18801da177e4SLinus Torvalds } 18811da177e4SLinus Torvalds cur = cur->hnext; 18821da177e4SLinus Torvalds } 18831da177e4SLinus Torvalds } 18841da177e4SLinus Torvalds 1885bd4c625cSLinus Torvalds static void free_journal_ram(struct super_block *p_s_sb) 1886bd4c625cSLinus Torvalds { 18871da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 1888d739b42bSPekka Enberg kfree(journal->j_current_jl); 18891da177e4SLinus Torvalds journal->j_num_lists--; 18901da177e4SLinus Torvalds 18911da177e4SLinus Torvalds vfree(journal->j_cnode_free_orig); 18921da177e4SLinus Torvalds free_list_bitmaps(p_s_sb, journal->j_list_bitmap); 18931da177e4SLinus Torvalds free_bitmap_nodes(p_s_sb); /* must be after free_list_bitmaps */ 18941da177e4SLinus Torvalds if (journal->j_header_bh) { 18951da177e4SLinus Torvalds brelse(journal->j_header_bh); 18961da177e4SLinus Torvalds } 18971da177e4SLinus Torvalds /* j_header_bh is on the journal dev, make sure not to release the journal 18981da177e4SLinus Torvalds * dev until we brelse j_header_bh 18991da177e4SLinus Torvalds */ 19001da177e4SLinus Torvalds release_journal_dev(p_s_sb, journal); 19011da177e4SLinus Torvalds vfree(journal); 19021da177e4SLinus Torvalds } 19031da177e4SLinus Torvalds 19041da177e4SLinus Torvalds /* 19051da177e4SLinus Torvalds ** call on unmount. Only set error to 1 if you haven't made your way out 19061da177e4SLinus Torvalds ** of read_super() yet. Any other caller must keep error at 0. 19071da177e4SLinus Torvalds */ 1908bd4c625cSLinus Torvalds static int do_journal_release(struct reiserfs_transaction_handle *th, 1909bd4c625cSLinus Torvalds struct super_block *p_s_sb, int error) 1910bd4c625cSLinus Torvalds { 19111da177e4SLinus Torvalds struct reiserfs_transaction_handle myth; 19121da177e4SLinus Torvalds int flushed = 0; 19131da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 19141da177e4SLinus Torvalds 19151da177e4SLinus Torvalds /* we only want to flush out transactions if we were called with error == 0 19161da177e4SLinus Torvalds */ 19171da177e4SLinus Torvalds if (!error && !(p_s_sb->s_flags & MS_RDONLY)) { 19181da177e4SLinus Torvalds /* end the current trans */ 19191da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 19201da177e4SLinus Torvalds do_journal_end(th, p_s_sb, 10, FLUSH_ALL); 19211da177e4SLinus Torvalds 19221da177e4SLinus Torvalds /* make sure something gets logged to force our way into the flush code */ 19231da177e4SLinus Torvalds if (!journal_join(&myth, p_s_sb, 1)) { 1924bd4c625cSLinus Torvalds reiserfs_prepare_for_journal(p_s_sb, 1925bd4c625cSLinus Torvalds SB_BUFFER_WITH_SB(p_s_sb), 1926bd4c625cSLinus Torvalds 1); 1927bd4c625cSLinus Torvalds journal_mark_dirty(&myth, p_s_sb, 1928bd4c625cSLinus Torvalds SB_BUFFER_WITH_SB(p_s_sb)); 19291da177e4SLinus Torvalds do_journal_end(&myth, p_s_sb, 1, FLUSH_ALL); 19301da177e4SLinus Torvalds flushed = 1; 19311da177e4SLinus Torvalds } 19321da177e4SLinus Torvalds } 19331da177e4SLinus Torvalds 19341da177e4SLinus Torvalds /* this also catches errors during the do_journal_end above */ 19351da177e4SLinus Torvalds if (!error && reiserfs_is_journal_aborted(journal)) { 19361da177e4SLinus Torvalds memset(&myth, 0, sizeof(myth)); 19371da177e4SLinus Torvalds if (!journal_join_abort(&myth, p_s_sb, 1)) { 1938bd4c625cSLinus Torvalds reiserfs_prepare_for_journal(p_s_sb, 1939bd4c625cSLinus Torvalds SB_BUFFER_WITH_SB(p_s_sb), 1940bd4c625cSLinus Torvalds 1); 1941bd4c625cSLinus Torvalds journal_mark_dirty(&myth, p_s_sb, 1942bd4c625cSLinus Torvalds SB_BUFFER_WITH_SB(p_s_sb)); 19431da177e4SLinus Torvalds do_journal_end(&myth, p_s_sb, 1, FLUSH_ALL); 19441da177e4SLinus Torvalds } 19451da177e4SLinus Torvalds } 19461da177e4SLinus Torvalds 19471da177e4SLinus Torvalds reiserfs_mounted_fs_count--; 19481da177e4SLinus Torvalds /* wait for all commits to finish */ 19491da177e4SLinus Torvalds cancel_delayed_work(&SB_JOURNAL(p_s_sb)->j_work); 19501da177e4SLinus Torvalds flush_workqueue(commit_wq); 19511da177e4SLinus Torvalds if (!reiserfs_mounted_fs_count) { 19521da177e4SLinus Torvalds destroy_workqueue(commit_wq); 19531da177e4SLinus Torvalds commit_wq = NULL; 19541da177e4SLinus Torvalds } 19551da177e4SLinus Torvalds 19561da177e4SLinus Torvalds free_journal_ram(p_s_sb); 19571da177e4SLinus Torvalds 19581da177e4SLinus Torvalds return 0; 19591da177e4SLinus Torvalds } 19601da177e4SLinus Torvalds 19611da177e4SLinus Torvalds /* 19621da177e4SLinus Torvalds ** call on unmount. flush all journal trans, release all alloc'd ram 19631da177e4SLinus Torvalds */ 1964bd4c625cSLinus Torvalds int journal_release(struct reiserfs_transaction_handle *th, 1965bd4c625cSLinus Torvalds struct super_block *p_s_sb) 1966bd4c625cSLinus Torvalds { 19671da177e4SLinus Torvalds return do_journal_release(th, p_s_sb, 0); 19681da177e4SLinus Torvalds } 1969bd4c625cSLinus Torvalds 19701da177e4SLinus Torvalds /* 19711da177e4SLinus Torvalds ** only call from an error condition inside reiserfs_read_super! 19721da177e4SLinus Torvalds */ 1973bd4c625cSLinus Torvalds int journal_release_error(struct reiserfs_transaction_handle *th, 1974bd4c625cSLinus Torvalds struct super_block *p_s_sb) 1975bd4c625cSLinus Torvalds { 19761da177e4SLinus Torvalds return do_journal_release(th, p_s_sb, 1); 19771da177e4SLinus Torvalds } 19781da177e4SLinus Torvalds 19791da177e4SLinus Torvalds /* compares description block with commit block. returns 1 if they differ, 0 if they are the same */ 1980bd4c625cSLinus Torvalds static int journal_compare_desc_commit(struct super_block *p_s_sb, 1981bd4c625cSLinus Torvalds struct reiserfs_journal_desc *desc, 1982bd4c625cSLinus Torvalds struct reiserfs_journal_commit *commit) 1983bd4c625cSLinus Torvalds { 19841da177e4SLinus Torvalds if (get_commit_trans_id(commit) != get_desc_trans_id(desc) || 19851da177e4SLinus Torvalds get_commit_trans_len(commit) != get_desc_trans_len(desc) || 19861da177e4SLinus Torvalds get_commit_trans_len(commit) > SB_JOURNAL(p_s_sb)->j_trans_max || 1987bd4c625cSLinus Torvalds get_commit_trans_len(commit) <= 0) { 19881da177e4SLinus Torvalds return 1; 19891da177e4SLinus Torvalds } 19901da177e4SLinus Torvalds return 0; 19911da177e4SLinus Torvalds } 1992bd4c625cSLinus Torvalds 19931da177e4SLinus Torvalds /* returns 0 if it did not find a description block 19941da177e4SLinus Torvalds ** returns -1 if it found a corrupt commit block 19951da177e4SLinus Torvalds ** returns 1 if both desc and commit were valid 19961da177e4SLinus Torvalds */ 1997bd4c625cSLinus Torvalds static int journal_transaction_is_valid(struct super_block *p_s_sb, 1998bd4c625cSLinus Torvalds struct buffer_head *d_bh, 1999bd4c625cSLinus Torvalds unsigned long *oldest_invalid_trans_id, 2000bd4c625cSLinus Torvalds unsigned long *newest_mount_id) 2001bd4c625cSLinus Torvalds { 20021da177e4SLinus Torvalds struct reiserfs_journal_desc *desc; 20031da177e4SLinus Torvalds struct reiserfs_journal_commit *commit; 20041da177e4SLinus Torvalds struct buffer_head *c_bh; 20051da177e4SLinus Torvalds unsigned long offset; 20061da177e4SLinus Torvalds 20071da177e4SLinus Torvalds if (!d_bh) 20081da177e4SLinus Torvalds return 0; 20091da177e4SLinus Torvalds 20101da177e4SLinus Torvalds desc = (struct reiserfs_journal_desc *)d_bh->b_data; 2011bd4c625cSLinus Torvalds if (get_desc_trans_len(desc) > 0 2012bd4c625cSLinus Torvalds && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) { 2013bd4c625cSLinus Torvalds if (oldest_invalid_trans_id && *oldest_invalid_trans_id 2014bd4c625cSLinus Torvalds && get_desc_trans_id(desc) > *oldest_invalid_trans_id) { 2015bd4c625cSLinus Torvalds reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, 2016bd4c625cSLinus Torvalds "journal-986: transaction " 20171da177e4SLinus Torvalds "is valid returning because trans_id %d is greater than " 2018bd4c625cSLinus Torvalds "oldest_invalid %lu", 2019bd4c625cSLinus Torvalds get_desc_trans_id(desc), 20201da177e4SLinus Torvalds *oldest_invalid_trans_id); 20211da177e4SLinus Torvalds return 0; 20221da177e4SLinus Torvalds } 2023bd4c625cSLinus Torvalds if (newest_mount_id 2024bd4c625cSLinus Torvalds && *newest_mount_id > get_desc_mount_id(desc)) { 2025bd4c625cSLinus Torvalds reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, 2026bd4c625cSLinus Torvalds "journal-1087: transaction " 20271da177e4SLinus Torvalds "is valid returning because mount_id %d is less than " 2028bd4c625cSLinus Torvalds "newest_mount_id %lu", 2029bd4c625cSLinus Torvalds get_desc_mount_id(desc), 20301da177e4SLinus Torvalds *newest_mount_id); 20311da177e4SLinus Torvalds return -1; 20321da177e4SLinus Torvalds } 20331da177e4SLinus Torvalds if (get_desc_trans_len(desc) > SB_JOURNAL(p_s_sb)->j_trans_max) { 2034bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 2035bd4c625cSLinus Torvalds "journal-2018: Bad transaction length %d encountered, ignoring transaction", 2036bd4c625cSLinus Torvalds get_desc_trans_len(desc)); 20371da177e4SLinus Torvalds return -1; 20381da177e4SLinus Torvalds } 20391da177e4SLinus Torvalds offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb); 20401da177e4SLinus Torvalds 20411da177e4SLinus Torvalds /* ok, we have a journal description block, lets see if the transaction was valid */ 2042bd4c625cSLinus Torvalds c_bh = 2043bd4c625cSLinus Torvalds journal_bread(p_s_sb, 2044bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + 2045bd4c625cSLinus Torvalds ((offset + get_desc_trans_len(desc) + 2046bd4c625cSLinus Torvalds 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb))); 20471da177e4SLinus Torvalds if (!c_bh) 20481da177e4SLinus Torvalds return 0; 20491da177e4SLinus Torvalds commit = (struct reiserfs_journal_commit *)c_bh->b_data; 20501da177e4SLinus Torvalds if (journal_compare_desc_commit(p_s_sb, desc, commit)) { 20511da177e4SLinus Torvalds reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, 20521da177e4SLinus Torvalds "journal_transaction_is_valid, commit offset %ld had bad " 20531da177e4SLinus Torvalds "time %d or length %d", 2054bd4c625cSLinus Torvalds c_bh->b_blocknr - 2055bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb), 20561da177e4SLinus Torvalds get_commit_trans_id(commit), 20571da177e4SLinus Torvalds get_commit_trans_len(commit)); 20581da177e4SLinus Torvalds brelse(c_bh); 20591da177e4SLinus Torvalds if (oldest_invalid_trans_id) { 2060bd4c625cSLinus Torvalds *oldest_invalid_trans_id = 2061bd4c625cSLinus Torvalds get_desc_trans_id(desc); 2062bd4c625cSLinus Torvalds reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, 2063bd4c625cSLinus Torvalds "journal-1004: " 20641da177e4SLinus Torvalds "transaction_is_valid setting oldest invalid trans_id " 2065bd4c625cSLinus Torvalds "to %d", 2066bd4c625cSLinus Torvalds get_desc_trans_id(desc)); 20671da177e4SLinus Torvalds } 20681da177e4SLinus Torvalds return -1; 20691da177e4SLinus Torvalds } 20701da177e4SLinus Torvalds brelse(c_bh); 2071bd4c625cSLinus Torvalds reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, 2072bd4c625cSLinus Torvalds "journal-1006: found valid " 20731da177e4SLinus Torvalds "transaction start offset %llu, len %d id %d", 2074bd4c625cSLinus Torvalds d_bh->b_blocknr - 2075bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb), 2076bd4c625cSLinus Torvalds get_desc_trans_len(desc), 2077bd4c625cSLinus Torvalds get_desc_trans_id(desc)); 20781da177e4SLinus Torvalds return 1; 20791da177e4SLinus Torvalds } else { 20801da177e4SLinus Torvalds return 0; 20811da177e4SLinus Torvalds } 20821da177e4SLinus Torvalds } 20831da177e4SLinus Torvalds 2084bd4c625cSLinus Torvalds static void brelse_array(struct buffer_head **heads, int num) 2085bd4c625cSLinus Torvalds { 20861da177e4SLinus Torvalds int i; 20871da177e4SLinus Torvalds for (i = 0; i < num; i++) { 20881da177e4SLinus Torvalds brelse(heads[i]); 20891da177e4SLinus Torvalds } 20901da177e4SLinus Torvalds } 20911da177e4SLinus Torvalds 20921da177e4SLinus Torvalds /* 20931da177e4SLinus Torvalds ** given the start, and values for the oldest acceptable transactions, 20941da177e4SLinus Torvalds ** this either reads in a replays a transaction, or returns because the transaction 20951da177e4SLinus Torvalds ** is invalid, or too old. 20961da177e4SLinus Torvalds */ 2097bd4c625cSLinus Torvalds static int journal_read_transaction(struct super_block *p_s_sb, 2098bd4c625cSLinus Torvalds unsigned long cur_dblock, 2099bd4c625cSLinus Torvalds unsigned long oldest_start, 2100bd4c625cSLinus Torvalds unsigned long oldest_trans_id, 2101bd4c625cSLinus Torvalds unsigned long newest_mount_id) 2102bd4c625cSLinus Torvalds { 21031da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 21041da177e4SLinus Torvalds struct reiserfs_journal_desc *desc; 21051da177e4SLinus Torvalds struct reiserfs_journal_commit *commit; 21061da177e4SLinus Torvalds unsigned long trans_id = 0; 21071da177e4SLinus Torvalds struct buffer_head *c_bh; 21081da177e4SLinus Torvalds struct buffer_head *d_bh; 21091da177e4SLinus Torvalds struct buffer_head **log_blocks = NULL; 21101da177e4SLinus Torvalds struct buffer_head **real_blocks = NULL; 21111da177e4SLinus Torvalds unsigned long trans_offset; 21121da177e4SLinus Torvalds int i; 21131da177e4SLinus Torvalds int trans_half; 21141da177e4SLinus Torvalds 21151da177e4SLinus Torvalds d_bh = journal_bread(p_s_sb, cur_dblock); 21161da177e4SLinus Torvalds if (!d_bh) 21171da177e4SLinus Torvalds return 1; 21181da177e4SLinus Torvalds desc = (struct reiserfs_journal_desc *)d_bh->b_data; 21191da177e4SLinus Torvalds trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb); 21201da177e4SLinus Torvalds reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1037: " 21211da177e4SLinus Torvalds "journal_read_transaction, offset %llu, len %d mount_id %d", 21221da177e4SLinus Torvalds d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb), 21231da177e4SLinus Torvalds get_desc_trans_len(desc), get_desc_mount_id(desc)); 21241da177e4SLinus Torvalds if (get_desc_trans_id(desc) < oldest_trans_id) { 21251da177e4SLinus Torvalds reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1039: " 21261da177e4SLinus Torvalds "journal_read_trans skipping because %lu is too old", 2127bd4c625cSLinus Torvalds cur_dblock - 2128bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb)); 21291da177e4SLinus Torvalds brelse(d_bh); 21301da177e4SLinus Torvalds return 1; 21311da177e4SLinus Torvalds } 21321da177e4SLinus Torvalds if (get_desc_mount_id(desc) != newest_mount_id) { 21331da177e4SLinus Torvalds reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1146: " 21341da177e4SLinus Torvalds "journal_read_trans skipping because %d is != " 21351da177e4SLinus Torvalds "newest_mount_id %lu", get_desc_mount_id(desc), 21361da177e4SLinus Torvalds newest_mount_id); 21371da177e4SLinus Torvalds brelse(d_bh); 21381da177e4SLinus Torvalds return 1; 21391da177e4SLinus Torvalds } 21401da177e4SLinus Torvalds c_bh = journal_bread(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + 21411da177e4SLinus Torvalds ((trans_offset + get_desc_trans_len(desc) + 1) % 21421da177e4SLinus Torvalds SB_ONDISK_JOURNAL_SIZE(p_s_sb))); 21431da177e4SLinus Torvalds if (!c_bh) { 21441da177e4SLinus Torvalds brelse(d_bh); 21451da177e4SLinus Torvalds return 1; 21461da177e4SLinus Torvalds } 21471da177e4SLinus Torvalds commit = (struct reiserfs_journal_commit *)c_bh->b_data; 21481da177e4SLinus Torvalds if (journal_compare_desc_commit(p_s_sb, desc, commit)) { 2149bd4c625cSLinus Torvalds reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, 2150bd4c625cSLinus Torvalds "journal_read_transaction, " 21511da177e4SLinus Torvalds "commit offset %llu had bad time %d or length %d", 2152bd4c625cSLinus Torvalds c_bh->b_blocknr - 2153bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb), 2154bd4c625cSLinus Torvalds get_commit_trans_id(commit), 2155bd4c625cSLinus Torvalds get_commit_trans_len(commit)); 21561da177e4SLinus Torvalds brelse(c_bh); 21571da177e4SLinus Torvalds brelse(d_bh); 21581da177e4SLinus Torvalds return 1; 21591da177e4SLinus Torvalds } 21601da177e4SLinus Torvalds trans_id = get_desc_trans_id(desc); 21611da177e4SLinus Torvalds /* now we know we've got a good transaction, and it was inside the valid time ranges */ 2162d739b42bSPekka Enberg log_blocks = kmalloc(get_desc_trans_len(desc) * 2163d739b42bSPekka Enberg sizeof(struct buffer_head *), GFP_NOFS); 2164d739b42bSPekka Enberg real_blocks = kmalloc(get_desc_trans_len(desc) * 2165d739b42bSPekka Enberg sizeof(struct buffer_head *), GFP_NOFS); 21661da177e4SLinus Torvalds if (!log_blocks || !real_blocks) { 21671da177e4SLinus Torvalds brelse(c_bh); 21681da177e4SLinus Torvalds brelse(d_bh); 2169d739b42bSPekka Enberg kfree(log_blocks); 2170d739b42bSPekka Enberg kfree(real_blocks); 2171bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 2172bd4c625cSLinus Torvalds "journal-1169: kmalloc failed, unable to mount FS"); 21731da177e4SLinus Torvalds return -1; 21741da177e4SLinus Torvalds } 21751da177e4SLinus Torvalds /* get all the buffer heads */ 21761da177e4SLinus Torvalds trans_half = journal_trans_half(p_s_sb->s_blocksize); 21771da177e4SLinus Torvalds for (i = 0; i < get_desc_trans_len(desc); i++) { 2178bd4c625cSLinus Torvalds log_blocks[i] = 2179bd4c625cSLinus Torvalds journal_getblk(p_s_sb, 2180bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + 2181bd4c625cSLinus Torvalds (trans_offset + 1 + 2182bd4c625cSLinus Torvalds i) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)); 21831da177e4SLinus Torvalds if (i < trans_half) { 2184bd4c625cSLinus Torvalds real_blocks[i] = 2185bd4c625cSLinus Torvalds sb_getblk(p_s_sb, 2186bd4c625cSLinus Torvalds le32_to_cpu(desc->j_realblock[i])); 21871da177e4SLinus Torvalds } else { 2188bd4c625cSLinus Torvalds real_blocks[i] = 2189bd4c625cSLinus Torvalds sb_getblk(p_s_sb, 2190bd4c625cSLinus Torvalds le32_to_cpu(commit-> 2191bd4c625cSLinus Torvalds j_realblock[i - trans_half])); 21921da177e4SLinus Torvalds } 21931da177e4SLinus Torvalds if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(p_s_sb)) { 2194bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 2195bd4c625cSLinus Torvalds "journal-1207: REPLAY FAILURE fsck required! Block to replay is outside of filesystem"); 21961da177e4SLinus Torvalds goto abort_replay; 21971da177e4SLinus Torvalds } 21981da177e4SLinus Torvalds /* make sure we don't try to replay onto log or reserved area */ 2199bd4c625cSLinus Torvalds if (is_block_in_log_or_reserved_area 2200bd4c625cSLinus Torvalds (p_s_sb, real_blocks[i]->b_blocknr)) { 2201bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 2202bd4c625cSLinus Torvalds "journal-1204: REPLAY FAILURE fsck required! Trying to replay onto a log block"); 22031da177e4SLinus Torvalds abort_replay: 22041da177e4SLinus Torvalds brelse_array(log_blocks, i); 22051da177e4SLinus Torvalds brelse_array(real_blocks, i); 22061da177e4SLinus Torvalds brelse(c_bh); 22071da177e4SLinus Torvalds brelse(d_bh); 2208d739b42bSPekka Enberg kfree(log_blocks); 2209d739b42bSPekka Enberg kfree(real_blocks); 22101da177e4SLinus Torvalds return -1; 22111da177e4SLinus Torvalds } 22121da177e4SLinus Torvalds } 22131da177e4SLinus Torvalds /* read in the log blocks, memcpy to the corresponding real block */ 22141da177e4SLinus Torvalds ll_rw_block(READ, get_desc_trans_len(desc), log_blocks); 22151da177e4SLinus Torvalds for (i = 0; i < get_desc_trans_len(desc); i++) { 22161da177e4SLinus Torvalds wait_on_buffer(log_blocks[i]); 22171da177e4SLinus Torvalds if (!buffer_uptodate(log_blocks[i])) { 2218bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 2219bd4c625cSLinus Torvalds "journal-1212: REPLAY FAILURE fsck required! buffer write failed"); 2220bd4c625cSLinus Torvalds brelse_array(log_blocks + i, 2221bd4c625cSLinus Torvalds get_desc_trans_len(desc) - i); 22221da177e4SLinus Torvalds brelse_array(real_blocks, get_desc_trans_len(desc)); 22231da177e4SLinus Torvalds brelse(c_bh); 22241da177e4SLinus Torvalds brelse(d_bh); 2225d739b42bSPekka Enberg kfree(log_blocks); 2226d739b42bSPekka Enberg kfree(real_blocks); 22271da177e4SLinus Torvalds return -1; 22281da177e4SLinus Torvalds } 2229bd4c625cSLinus Torvalds memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data, 2230bd4c625cSLinus Torvalds real_blocks[i]->b_size); 22311da177e4SLinus Torvalds set_buffer_uptodate(real_blocks[i]); 22321da177e4SLinus Torvalds brelse(log_blocks[i]); 22331da177e4SLinus Torvalds } 22341da177e4SLinus Torvalds /* flush out the real blocks */ 22351da177e4SLinus Torvalds for (i = 0; i < get_desc_trans_len(desc); i++) { 22361da177e4SLinus Torvalds set_buffer_dirty(real_blocks[i]); 223753778ffdSJan Kara ll_rw_block(SWRITE, 1, real_blocks + i); 22381da177e4SLinus Torvalds } 22391da177e4SLinus Torvalds for (i = 0; i < get_desc_trans_len(desc); i++) { 22401da177e4SLinus Torvalds wait_on_buffer(real_blocks[i]); 22411da177e4SLinus Torvalds if (!buffer_uptodate(real_blocks[i])) { 2242bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 2243bd4c625cSLinus Torvalds "journal-1226: REPLAY FAILURE, fsck required! buffer write failed"); 2244bd4c625cSLinus Torvalds brelse_array(real_blocks + i, 2245bd4c625cSLinus Torvalds get_desc_trans_len(desc) - i); 22461da177e4SLinus Torvalds brelse(c_bh); 22471da177e4SLinus Torvalds brelse(d_bh); 2248d739b42bSPekka Enberg kfree(log_blocks); 2249d739b42bSPekka Enberg kfree(real_blocks); 22501da177e4SLinus Torvalds return -1; 22511da177e4SLinus Torvalds } 22521da177e4SLinus Torvalds brelse(real_blocks[i]); 22531da177e4SLinus Torvalds } 2254bd4c625cSLinus Torvalds cur_dblock = 2255bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + 2256bd4c625cSLinus Torvalds ((trans_offset + get_desc_trans_len(desc) + 2257bd4c625cSLinus Torvalds 2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)); 2258bd4c625cSLinus Torvalds reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, 2259bd4c625cSLinus Torvalds "journal-1095: setting journal " "start to offset %ld", 22601da177e4SLinus Torvalds cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb)); 22611da177e4SLinus Torvalds 22621da177e4SLinus Torvalds /* init starting values for the first transaction, in case this is the last transaction to be replayed. */ 22631da177e4SLinus Torvalds journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb); 22641da177e4SLinus Torvalds journal->j_last_flush_trans_id = trans_id; 22651da177e4SLinus Torvalds journal->j_trans_id = trans_id + 1; 2266a44c94a7SAlexander Zarochentsev /* check for trans_id overflow */ 2267a44c94a7SAlexander Zarochentsev if (journal->j_trans_id == 0) 2268a44c94a7SAlexander Zarochentsev journal->j_trans_id = 10; 22691da177e4SLinus Torvalds brelse(c_bh); 22701da177e4SLinus Torvalds brelse(d_bh); 2271d739b42bSPekka Enberg kfree(log_blocks); 2272d739b42bSPekka Enberg kfree(real_blocks); 22731da177e4SLinus Torvalds return 0; 22741da177e4SLinus Torvalds } 22751da177e4SLinus Torvalds 22761da177e4SLinus Torvalds /* This function reads blocks starting from block and to max_block of bufsize 22771da177e4SLinus Torvalds size (but no more than BUFNR blocks at a time). This proved to improve 22781da177e4SLinus Torvalds mounting speed on self-rebuilding raid5 arrays at least. 22791da177e4SLinus Torvalds Right now it is only used from journal code. But later we might use it 22801da177e4SLinus Torvalds from other places. 22811da177e4SLinus Torvalds Note: Do not use journal_getblk/sb_getblk functions here! */ 2282bd4c625cSLinus Torvalds static struct buffer_head *reiserfs_breada(struct block_device *dev, int block, 2283bd4c625cSLinus Torvalds int bufsize, unsigned int max_block) 22841da177e4SLinus Torvalds { 22851da177e4SLinus Torvalds struct buffer_head *bhlist[BUFNR]; 22861da177e4SLinus Torvalds unsigned int blocks = BUFNR; 22871da177e4SLinus Torvalds struct buffer_head *bh; 22881da177e4SLinus Torvalds int i, j; 22891da177e4SLinus Torvalds 22901da177e4SLinus Torvalds bh = __getblk(dev, block, bufsize); 22911da177e4SLinus Torvalds if (buffer_uptodate(bh)) 22921da177e4SLinus Torvalds return (bh); 22931da177e4SLinus Torvalds 22941da177e4SLinus Torvalds if (block + BUFNR > max_block) { 22951da177e4SLinus Torvalds blocks = max_block - block; 22961da177e4SLinus Torvalds } 22971da177e4SLinus Torvalds bhlist[0] = bh; 22981da177e4SLinus Torvalds j = 1; 22991da177e4SLinus Torvalds for (i = 1; i < blocks; i++) { 23001da177e4SLinus Torvalds bh = __getblk(dev, block + i, bufsize); 23011da177e4SLinus Torvalds if (buffer_uptodate(bh)) { 23021da177e4SLinus Torvalds brelse(bh); 23031da177e4SLinus Torvalds break; 2304bd4c625cSLinus Torvalds } else 2305bd4c625cSLinus Torvalds bhlist[j++] = bh; 23061da177e4SLinus Torvalds } 23071da177e4SLinus Torvalds ll_rw_block(READ, j, bhlist); 23081da177e4SLinus Torvalds for (i = 1; i < j; i++) 23091da177e4SLinus Torvalds brelse(bhlist[i]); 23101da177e4SLinus Torvalds bh = bhlist[0]; 23111da177e4SLinus Torvalds wait_on_buffer(bh); 23121da177e4SLinus Torvalds if (buffer_uptodate(bh)) 23131da177e4SLinus Torvalds return bh; 23141da177e4SLinus Torvalds brelse(bh); 23151da177e4SLinus Torvalds return NULL; 23161da177e4SLinus Torvalds } 23171da177e4SLinus Torvalds 23181da177e4SLinus Torvalds /* 23191da177e4SLinus Torvalds ** read and replay the log 23201da177e4SLinus Torvalds ** on a clean unmount, the journal header's next unflushed pointer will be to an invalid 23211da177e4SLinus Torvalds ** transaction. This tests that before finding all the transactions in the log, which makes normal mount times fast. 23221da177e4SLinus Torvalds ** 23231da177e4SLinus Torvalds ** After a crash, this starts with the next unflushed transaction, and replays until it finds one too old, or invalid. 23241da177e4SLinus Torvalds ** 23251da177e4SLinus Torvalds ** On exit, it sets things up so the first transaction will work correctly. 23261da177e4SLinus Torvalds */ 2327bd4c625cSLinus Torvalds static int journal_read(struct super_block *p_s_sb) 2328bd4c625cSLinus Torvalds { 23291da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 23301da177e4SLinus Torvalds struct reiserfs_journal_desc *desc; 23311da177e4SLinus Torvalds unsigned long oldest_trans_id = 0; 23321da177e4SLinus Torvalds unsigned long oldest_invalid_trans_id = 0; 23331da177e4SLinus Torvalds time_t start; 23341da177e4SLinus Torvalds unsigned long oldest_start = 0; 23351da177e4SLinus Torvalds unsigned long cur_dblock = 0; 23361da177e4SLinus Torvalds unsigned long newest_mount_id = 9; 23371da177e4SLinus Torvalds struct buffer_head *d_bh; 23381da177e4SLinus Torvalds struct reiserfs_journal_header *jh; 23391da177e4SLinus Torvalds int valid_journal_header = 0; 23401da177e4SLinus Torvalds int replay_count = 0; 23411da177e4SLinus Torvalds int continue_replay = 1; 23421da177e4SLinus Torvalds int ret; 23431da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 23441da177e4SLinus Torvalds 23451da177e4SLinus Torvalds cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb); 23461da177e4SLinus Torvalds reiserfs_info(p_s_sb, "checking transaction log (%s)\n", 23471da177e4SLinus Torvalds bdevname(journal->j_dev_bd, b)); 23481da177e4SLinus Torvalds start = get_seconds(); 23491da177e4SLinus Torvalds 23501da177e4SLinus Torvalds /* step 1, read in the journal header block. Check the transaction it says 23511da177e4SLinus Torvalds ** is the first unflushed, and if that transaction is not valid, 23521da177e4SLinus Torvalds ** replay is done 23531da177e4SLinus Torvalds */ 23541da177e4SLinus Torvalds journal->j_header_bh = journal_bread(p_s_sb, 2355bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) 2356bd4c625cSLinus Torvalds + SB_ONDISK_JOURNAL_SIZE(p_s_sb)); 23571da177e4SLinus Torvalds if (!journal->j_header_bh) { 23581da177e4SLinus Torvalds return 1; 23591da177e4SLinus Torvalds } 23601da177e4SLinus Torvalds jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data); 2361c499ec24SVladimir V. Saveliev if (le32_to_cpu(jh->j_first_unflushed_offset) < 2362bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_SIZE(p_s_sb) 2363bd4c625cSLinus Torvalds && le32_to_cpu(jh->j_last_flush_trans_id) > 0) { 2364bd4c625cSLinus Torvalds oldest_start = 2365bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + 23661da177e4SLinus Torvalds le32_to_cpu(jh->j_first_unflushed_offset); 23671da177e4SLinus Torvalds oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1; 23681da177e4SLinus Torvalds newest_mount_id = le32_to_cpu(jh->j_mount_id); 2369bd4c625cSLinus Torvalds reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, 2370bd4c625cSLinus Torvalds "journal-1153: found in " 23711da177e4SLinus Torvalds "header: first_unflushed_offset %d, last_flushed_trans_id " 23721da177e4SLinus Torvalds "%lu", le32_to_cpu(jh->j_first_unflushed_offset), 23731da177e4SLinus Torvalds le32_to_cpu(jh->j_last_flush_trans_id)); 23741da177e4SLinus Torvalds valid_journal_header = 1; 23751da177e4SLinus Torvalds 23761da177e4SLinus Torvalds /* now, we try to read the first unflushed offset. If it is not valid, 23771da177e4SLinus Torvalds ** there is nothing more we can do, and it makes no sense to read 23781da177e4SLinus Torvalds ** through the whole log. 23791da177e4SLinus Torvalds */ 2380bd4c625cSLinus Torvalds d_bh = 2381bd4c625cSLinus Torvalds journal_bread(p_s_sb, 2382bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + 2383bd4c625cSLinus Torvalds le32_to_cpu(jh->j_first_unflushed_offset)); 23841da177e4SLinus Torvalds ret = journal_transaction_is_valid(p_s_sb, d_bh, NULL, NULL); 23851da177e4SLinus Torvalds if (!ret) { 23861da177e4SLinus Torvalds continue_replay = 0; 23871da177e4SLinus Torvalds } 23881da177e4SLinus Torvalds brelse(d_bh); 23891da177e4SLinus Torvalds goto start_log_replay; 23901da177e4SLinus Torvalds } 23911da177e4SLinus Torvalds 23921da177e4SLinus Torvalds if (continue_replay && bdev_read_only(p_s_sb->s_bdev)) { 23931da177e4SLinus Torvalds reiserfs_warning(p_s_sb, 23941da177e4SLinus Torvalds "clm-2076: device is readonly, unable to replay log"); 23951da177e4SLinus Torvalds return -1; 23961da177e4SLinus Torvalds } 23971da177e4SLinus Torvalds 23981da177e4SLinus Torvalds /* ok, there are transactions that need to be replayed. start with the first log block, find 23991da177e4SLinus Torvalds ** all the valid transactions, and pick out the oldest. 24001da177e4SLinus Torvalds */ 2401bd4c625cSLinus Torvalds while (continue_replay 2402bd4c625cSLinus Torvalds && cur_dblock < 2403bd4c625cSLinus Torvalds (SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + 2404bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_SIZE(p_s_sb))) { 24051da177e4SLinus Torvalds /* Note that it is required for blocksize of primary fs device and journal 24061da177e4SLinus Torvalds device to be the same */ 2407bd4c625cSLinus Torvalds d_bh = 2408bd4c625cSLinus Torvalds reiserfs_breada(journal->j_dev_bd, cur_dblock, 2409bd4c625cSLinus Torvalds p_s_sb->s_blocksize, 2410bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + 2411bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_SIZE(p_s_sb)); 2412bd4c625cSLinus Torvalds ret = 2413bd4c625cSLinus Torvalds journal_transaction_is_valid(p_s_sb, d_bh, 2414bd4c625cSLinus Torvalds &oldest_invalid_trans_id, 2415bd4c625cSLinus Torvalds &newest_mount_id); 24161da177e4SLinus Torvalds if (ret == 1) { 24171da177e4SLinus Torvalds desc = (struct reiserfs_journal_desc *)d_bh->b_data; 24181da177e4SLinus Torvalds if (oldest_start == 0) { /* init all oldest_ values */ 24191da177e4SLinus Torvalds oldest_trans_id = get_desc_trans_id(desc); 24201da177e4SLinus Torvalds oldest_start = d_bh->b_blocknr; 24211da177e4SLinus Torvalds newest_mount_id = get_desc_mount_id(desc); 2422bd4c625cSLinus Torvalds reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, 2423bd4c625cSLinus Torvalds "journal-1179: Setting " 24241da177e4SLinus Torvalds "oldest_start to offset %llu, trans_id %lu", 2425bd4c625cSLinus Torvalds oldest_start - 2426bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK 2427bd4c625cSLinus Torvalds (p_s_sb), oldest_trans_id); 24281da177e4SLinus Torvalds } else if (oldest_trans_id > get_desc_trans_id(desc)) { 24291da177e4SLinus Torvalds /* one we just read was older */ 24301da177e4SLinus Torvalds oldest_trans_id = get_desc_trans_id(desc); 24311da177e4SLinus Torvalds oldest_start = d_bh->b_blocknr; 2432bd4c625cSLinus Torvalds reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, 2433bd4c625cSLinus Torvalds "journal-1180: Resetting " 24341da177e4SLinus Torvalds "oldest_start to offset %lu, trans_id %lu", 2435bd4c625cSLinus Torvalds oldest_start - 2436bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK 2437bd4c625cSLinus Torvalds (p_s_sb), oldest_trans_id); 24381da177e4SLinus Torvalds } 24391da177e4SLinus Torvalds if (newest_mount_id < get_desc_mount_id(desc)) { 24401da177e4SLinus Torvalds newest_mount_id = get_desc_mount_id(desc); 2441bd4c625cSLinus Torvalds reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, 2442bd4c625cSLinus Torvalds "journal-1299: Setting " 2443bd4c625cSLinus Torvalds "newest_mount_id to %d", 2444bd4c625cSLinus Torvalds get_desc_mount_id(desc)); 24451da177e4SLinus Torvalds } 24461da177e4SLinus Torvalds cur_dblock += get_desc_trans_len(desc) + 2; 24471da177e4SLinus Torvalds } else { 24481da177e4SLinus Torvalds cur_dblock++; 24491da177e4SLinus Torvalds } 24501da177e4SLinus Torvalds brelse(d_bh); 24511da177e4SLinus Torvalds } 24521da177e4SLinus Torvalds 24531da177e4SLinus Torvalds start_log_replay: 24541da177e4SLinus Torvalds cur_dblock = oldest_start; 24551da177e4SLinus Torvalds if (oldest_trans_id) { 2456bd4c625cSLinus Torvalds reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, 2457bd4c625cSLinus Torvalds "journal-1206: Starting replay " 24581da177e4SLinus Torvalds "from offset %llu, trans_id %lu", 24591da177e4SLinus Torvalds cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb), 24601da177e4SLinus Torvalds oldest_trans_id); 24611da177e4SLinus Torvalds 24621da177e4SLinus Torvalds } 24631da177e4SLinus Torvalds replay_count = 0; 24641da177e4SLinus Torvalds while (continue_replay && oldest_trans_id > 0) { 2465bd4c625cSLinus Torvalds ret = 2466bd4c625cSLinus Torvalds journal_read_transaction(p_s_sb, cur_dblock, oldest_start, 2467bd4c625cSLinus Torvalds oldest_trans_id, newest_mount_id); 24681da177e4SLinus Torvalds if (ret < 0) { 24691da177e4SLinus Torvalds return ret; 24701da177e4SLinus Torvalds } else if (ret != 0) { 24711da177e4SLinus Torvalds break; 24721da177e4SLinus Torvalds } 2473bd4c625cSLinus Torvalds cur_dblock = 2474bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + journal->j_start; 24751da177e4SLinus Torvalds replay_count++; 24761da177e4SLinus Torvalds if (cur_dblock == oldest_start) 24771da177e4SLinus Torvalds break; 24781da177e4SLinus Torvalds } 24791da177e4SLinus Torvalds 24801da177e4SLinus Torvalds if (oldest_trans_id == 0) { 2481bd4c625cSLinus Torvalds reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, 2482bd4c625cSLinus Torvalds "journal-1225: No valid " "transactions found"); 24831da177e4SLinus Torvalds } 24841da177e4SLinus Torvalds /* j_start does not get set correctly if we don't replay any transactions. 24851da177e4SLinus Torvalds ** if we had a valid journal_header, set j_start to the first unflushed transaction value, 24861da177e4SLinus Torvalds ** copy the trans_id from the header 24871da177e4SLinus Torvalds */ 24881da177e4SLinus Torvalds if (valid_journal_header && replay_count == 0) { 24891da177e4SLinus Torvalds journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset); 2490bd4c625cSLinus Torvalds journal->j_trans_id = 2491bd4c625cSLinus Torvalds le32_to_cpu(jh->j_last_flush_trans_id) + 1; 2492a44c94a7SAlexander Zarochentsev /* check for trans_id overflow */ 2493a44c94a7SAlexander Zarochentsev if (journal->j_trans_id == 0) 2494a44c94a7SAlexander Zarochentsev journal->j_trans_id = 10; 2495bd4c625cSLinus Torvalds journal->j_last_flush_trans_id = 2496bd4c625cSLinus Torvalds le32_to_cpu(jh->j_last_flush_trans_id); 24971da177e4SLinus Torvalds journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1; 24981da177e4SLinus Torvalds } else { 24991da177e4SLinus Torvalds journal->j_mount_id = newest_mount_id + 1; 25001da177e4SLinus Torvalds } 25011da177e4SLinus Torvalds reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1299: Setting " 25021da177e4SLinus Torvalds "newest_mount_id to %lu", journal->j_mount_id); 25031da177e4SLinus Torvalds journal->j_first_unflushed_offset = journal->j_start; 25041da177e4SLinus Torvalds if (replay_count > 0) { 2505bd4c625cSLinus Torvalds reiserfs_info(p_s_sb, 2506bd4c625cSLinus Torvalds "replayed %d transactions in %lu seconds\n", 25071da177e4SLinus Torvalds replay_count, get_seconds() - start); 25081da177e4SLinus Torvalds } 25091da177e4SLinus Torvalds if (!bdev_read_only(p_s_sb->s_bdev) && 25101da177e4SLinus Torvalds _update_journal_header_block(p_s_sb, journal->j_start, 2511bd4c625cSLinus Torvalds journal->j_last_flush_trans_id)) { 25121da177e4SLinus Torvalds /* replay failed, caller must call free_journal_ram and abort 25131da177e4SLinus Torvalds ** the mount 25141da177e4SLinus Torvalds */ 25151da177e4SLinus Torvalds return -1; 25161da177e4SLinus Torvalds } 25171da177e4SLinus Torvalds return 0; 25181da177e4SLinus Torvalds } 25191da177e4SLinus Torvalds 25201da177e4SLinus Torvalds static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s) 25211da177e4SLinus Torvalds { 25221da177e4SLinus Torvalds struct reiserfs_journal_list *jl; 25238c777cc4SPekka Enberg jl = kzalloc(sizeof(struct reiserfs_journal_list), 25248c777cc4SPekka Enberg GFP_NOFS | __GFP_NOFAIL); 25251da177e4SLinus Torvalds INIT_LIST_HEAD(&jl->j_list); 25261da177e4SLinus Torvalds INIT_LIST_HEAD(&jl->j_working_list); 25271da177e4SLinus Torvalds INIT_LIST_HEAD(&jl->j_tail_bh_list); 25281da177e4SLinus Torvalds INIT_LIST_HEAD(&jl->j_bh_list); 25291da177e4SLinus Torvalds sema_init(&jl->j_commit_lock, 1); 25301da177e4SLinus Torvalds SB_JOURNAL(s)->j_num_lists++; 25311da177e4SLinus Torvalds get_journal_list(jl); 25321da177e4SLinus Torvalds return jl; 25331da177e4SLinus Torvalds } 25341da177e4SLinus Torvalds 2535bd4c625cSLinus Torvalds static void journal_list_init(struct super_block *p_s_sb) 2536bd4c625cSLinus Torvalds { 25371da177e4SLinus Torvalds SB_JOURNAL(p_s_sb)->j_current_jl = alloc_journal_list(p_s_sb); 25381da177e4SLinus Torvalds } 25391da177e4SLinus Torvalds 25401da177e4SLinus Torvalds static int release_journal_dev(struct super_block *super, 25411da177e4SLinus Torvalds struct reiserfs_journal *journal) 25421da177e4SLinus Torvalds { 25431da177e4SLinus Torvalds int result; 25441da177e4SLinus Torvalds 25451da177e4SLinus Torvalds result = 0; 25461da177e4SLinus Torvalds 25471da177e4SLinus Torvalds if (journal->j_dev_file != NULL) { 25481da177e4SLinus Torvalds result = filp_close(journal->j_dev_file, NULL); 25491da177e4SLinus Torvalds journal->j_dev_file = NULL; 25501da177e4SLinus Torvalds journal->j_dev_bd = NULL; 25511da177e4SLinus Torvalds } else if (journal->j_dev_bd != NULL) { 25521da177e4SLinus Torvalds result = blkdev_put(journal->j_dev_bd); 25531da177e4SLinus Torvalds journal->j_dev_bd = NULL; 25541da177e4SLinus Torvalds } 25551da177e4SLinus Torvalds 25561da177e4SLinus Torvalds if (result != 0) { 2557bd4c625cSLinus Torvalds reiserfs_warning(super, 2558bd4c625cSLinus Torvalds "sh-457: release_journal_dev: Cannot release journal device: %i", 2559bd4c625cSLinus Torvalds result); 25601da177e4SLinus Torvalds } 25611da177e4SLinus Torvalds return result; 25621da177e4SLinus Torvalds } 25631da177e4SLinus Torvalds 25641da177e4SLinus Torvalds static int journal_init_dev(struct super_block *super, 25651da177e4SLinus Torvalds struct reiserfs_journal *journal, 25661da177e4SLinus Torvalds const char *jdev_name) 25671da177e4SLinus Torvalds { 25681da177e4SLinus Torvalds int result; 25691da177e4SLinus Torvalds dev_t jdev; 25701da177e4SLinus Torvalds int blkdev_mode = FMODE_READ | FMODE_WRITE; 25711da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 25721da177e4SLinus Torvalds 25731da177e4SLinus Torvalds result = 0; 25741da177e4SLinus Torvalds 25751da177e4SLinus Torvalds journal->j_dev_bd = NULL; 25761da177e4SLinus Torvalds journal->j_dev_file = NULL; 25771da177e4SLinus Torvalds jdev = SB_ONDISK_JOURNAL_DEVICE(super) ? 25781da177e4SLinus Torvalds new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev; 25791da177e4SLinus Torvalds 25801da177e4SLinus Torvalds if (bdev_read_only(super->s_bdev)) 25811da177e4SLinus Torvalds blkdev_mode = FMODE_READ; 25821da177e4SLinus Torvalds 25831da177e4SLinus Torvalds /* there is no "jdev" option and journal is on separate device */ 25841da177e4SLinus Torvalds if ((!jdev_name || !jdev_name[0])) { 25851da177e4SLinus Torvalds journal->j_dev_bd = open_by_devnum(jdev, blkdev_mode); 25861da177e4SLinus Torvalds if (IS_ERR(journal->j_dev_bd)) { 25871da177e4SLinus Torvalds result = PTR_ERR(journal->j_dev_bd); 25881da177e4SLinus Torvalds journal->j_dev_bd = NULL; 25891da177e4SLinus Torvalds reiserfs_warning(super, "sh-458: journal_init_dev: " 25901da177e4SLinus Torvalds "cannot init journal device '%s': %i", 25911da177e4SLinus Torvalds __bdevname(jdev, b), result); 25921da177e4SLinus Torvalds return result; 25931da177e4SLinus Torvalds } else if (jdev != super->s_dev) 25941da177e4SLinus Torvalds set_blocksize(journal->j_dev_bd, super->s_blocksize); 25951da177e4SLinus Torvalds return 0; 25961da177e4SLinus Torvalds } 25971da177e4SLinus Torvalds 25981da177e4SLinus Torvalds journal->j_dev_file = filp_open(jdev_name, 0, 0); 25991da177e4SLinus Torvalds if (!IS_ERR(journal->j_dev_file)) { 26001da177e4SLinus Torvalds struct inode *jdev_inode = journal->j_dev_file->f_mapping->host; 26011da177e4SLinus Torvalds if (!S_ISBLK(jdev_inode->i_mode)) { 26021da177e4SLinus Torvalds reiserfs_warning(super, "journal_init_dev: '%s' is " 26031da177e4SLinus Torvalds "not a block device", jdev_name); 26041da177e4SLinus Torvalds result = -ENOTBLK; 260574f9f974SEdward Shishkin release_journal_dev(super, journal); 26061da177e4SLinus Torvalds } else { 26071da177e4SLinus Torvalds /* ok */ 26081da177e4SLinus Torvalds journal->j_dev_bd = I_BDEV(jdev_inode); 26091da177e4SLinus Torvalds set_blocksize(journal->j_dev_bd, super->s_blocksize); 2610bd4c625cSLinus Torvalds reiserfs_info(super, 2611bd4c625cSLinus Torvalds "journal_init_dev: journal device: %s\n", 261274f9f974SEdward Shishkin bdevname(journal->j_dev_bd, b)); 26131da177e4SLinus Torvalds } 26141da177e4SLinus Torvalds } else { 26151da177e4SLinus Torvalds result = PTR_ERR(journal->j_dev_file); 26161da177e4SLinus Torvalds journal->j_dev_file = NULL; 26171da177e4SLinus Torvalds reiserfs_warning(super, 26181da177e4SLinus Torvalds "journal_init_dev: Cannot open '%s': %i", 26191da177e4SLinus Torvalds jdev_name, result); 26201da177e4SLinus Torvalds } 26211da177e4SLinus Torvalds return result; 26221da177e4SLinus Torvalds } 26231da177e4SLinus Torvalds 26241da177e4SLinus Torvalds /* 26251da177e4SLinus Torvalds ** must be called once on fs mount. calls journal_read for you 26261da177e4SLinus Torvalds */ 2627bd4c625cSLinus Torvalds int journal_init(struct super_block *p_s_sb, const char *j_dev_name, 2628bd4c625cSLinus Torvalds int old_format, unsigned int commit_max_age) 2629bd4c625cSLinus Torvalds { 26301da177e4SLinus Torvalds int num_cnodes = SB_ONDISK_JOURNAL_SIZE(p_s_sb) * 2; 26311da177e4SLinus Torvalds struct buffer_head *bhjh; 26321da177e4SLinus Torvalds struct reiserfs_super_block *rs; 26331da177e4SLinus Torvalds struct reiserfs_journal_header *jh; 26341da177e4SLinus Torvalds struct reiserfs_journal *journal; 26351da177e4SLinus Torvalds struct reiserfs_journal_list *jl; 26361da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 26371da177e4SLinus Torvalds 26381da177e4SLinus Torvalds journal = SB_JOURNAL(p_s_sb) = vmalloc(sizeof(struct reiserfs_journal)); 26391da177e4SLinus Torvalds if (!journal) { 2640bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 2641bd4c625cSLinus Torvalds "journal-1256: unable to get memory for journal structure"); 26421da177e4SLinus Torvalds return 1; 26431da177e4SLinus Torvalds } 26441da177e4SLinus Torvalds memset(journal, 0, sizeof(struct reiserfs_journal)); 26451da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_bitmap_nodes); 26461da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_prealloc_list); 26471da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_working_list); 26481da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_journal_list); 26491da177e4SLinus Torvalds journal->j_persistent_trans = 0; 26501da177e4SLinus Torvalds if (reiserfs_allocate_list_bitmaps(p_s_sb, 26511da177e4SLinus Torvalds journal->j_list_bitmap, 26521da177e4SLinus Torvalds SB_BMAP_NR(p_s_sb))) 26531da177e4SLinus Torvalds goto free_and_return; 26541da177e4SLinus Torvalds allocate_bitmap_nodes(p_s_sb); 26551da177e4SLinus Torvalds 26561da177e4SLinus Torvalds /* reserved for journal area support */ 26571da177e4SLinus Torvalds SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) = (old_format ? 2658bd4c625cSLinus Torvalds REISERFS_OLD_DISK_OFFSET_IN_BYTES 2659bd4c625cSLinus Torvalds / p_s_sb->s_blocksize + 2660bd4c625cSLinus Torvalds SB_BMAP_NR(p_s_sb) + 2661bd4c625cSLinus Torvalds 1 : 2662bd4c625cSLinus Torvalds REISERFS_DISK_OFFSET_IN_BYTES / 2663bd4c625cSLinus Torvalds p_s_sb->s_blocksize + 2); 26641da177e4SLinus Torvalds 26651da177e4SLinus Torvalds /* Sanity check to see is the standard journal fitting withing first bitmap 26661da177e4SLinus Torvalds (actual for small blocksizes) */ 26671da177e4SLinus Torvalds if (!SB_ONDISK_JOURNAL_DEVICE(p_s_sb) && 2668bd4c625cSLinus Torvalds (SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) + 2669bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_SIZE(p_s_sb) > p_s_sb->s_blocksize * 8)) { 2670bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 2671bd4c625cSLinus Torvalds "journal-1393: journal does not fit for area " 26721da177e4SLinus Torvalds "addressed by first of bitmap blocks. It starts at " 26731da177e4SLinus Torvalds "%u and its size is %u. Block size %ld", 26741da177e4SLinus Torvalds SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb), 2675bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_SIZE(p_s_sb), 2676bd4c625cSLinus Torvalds p_s_sb->s_blocksize); 26771da177e4SLinus Torvalds goto free_and_return; 26781da177e4SLinus Torvalds } 26791da177e4SLinus Torvalds 26801da177e4SLinus Torvalds if (journal_init_dev(p_s_sb, journal, j_dev_name) != 0) { 2681bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 2682bd4c625cSLinus Torvalds "sh-462: unable to initialize jornal device"); 26831da177e4SLinus Torvalds goto free_and_return; 26841da177e4SLinus Torvalds } 26851da177e4SLinus Torvalds 26861da177e4SLinus Torvalds rs = SB_DISK_SUPER_BLOCK(p_s_sb); 26871da177e4SLinus Torvalds 26881da177e4SLinus Torvalds /* read journal header */ 26891da177e4SLinus Torvalds bhjh = journal_bread(p_s_sb, 2690bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + 2691bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_SIZE(p_s_sb)); 26921da177e4SLinus Torvalds if (!bhjh) { 2693bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 2694bd4c625cSLinus Torvalds "sh-459: unable to read journal header"); 26951da177e4SLinus Torvalds goto free_and_return; 26961da177e4SLinus Torvalds } 26971da177e4SLinus Torvalds jh = (struct reiserfs_journal_header *)(bhjh->b_data); 26981da177e4SLinus Torvalds 26991da177e4SLinus Torvalds /* make sure that journal matches to the super block */ 2700bd4c625cSLinus Torvalds if (is_reiserfs_jr(rs) 2701bd4c625cSLinus Torvalds && (le32_to_cpu(jh->jh_journal.jp_journal_magic) != 2702bd4c625cSLinus Torvalds sb_jp_journal_magic(rs))) { 2703bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 2704bd4c625cSLinus Torvalds "sh-460: journal header magic %x " 27051da177e4SLinus Torvalds "(device %s) does not match to magic found in super " 2706bd4c625cSLinus Torvalds "block %x", jh->jh_journal.jp_journal_magic, 27071da177e4SLinus Torvalds bdevname(journal->j_dev_bd, b), 27081da177e4SLinus Torvalds sb_jp_journal_magic(rs)); 27091da177e4SLinus Torvalds brelse(bhjh); 27101da177e4SLinus Torvalds goto free_and_return; 27111da177e4SLinus Torvalds } 27121da177e4SLinus Torvalds 27131da177e4SLinus Torvalds journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max); 27141da177e4SLinus Torvalds journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch); 2715bd4c625cSLinus Torvalds journal->j_max_commit_age = 2716bd4c625cSLinus Torvalds le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age); 27171da177e4SLinus Torvalds journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE; 27181da177e4SLinus Torvalds 27191da177e4SLinus Torvalds if (journal->j_trans_max) { 27201da177e4SLinus Torvalds /* make sure these parameters are available, assign it if they are not */ 27211da177e4SLinus Torvalds __u32 initial = journal->j_trans_max; 27221da177e4SLinus Torvalds __u32 ratio = 1; 27231da177e4SLinus Torvalds 27241da177e4SLinus Torvalds if (p_s_sb->s_blocksize < 4096) 27251da177e4SLinus Torvalds ratio = 4096 / p_s_sb->s_blocksize; 27261da177e4SLinus Torvalds 2727bd4c625cSLinus Torvalds if (SB_ONDISK_JOURNAL_SIZE(p_s_sb) / journal->j_trans_max < 2728bd4c625cSLinus Torvalds JOURNAL_MIN_RATIO) 2729bd4c625cSLinus Torvalds journal->j_trans_max = 2730bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_SIZE(p_s_sb) / JOURNAL_MIN_RATIO; 27311da177e4SLinus Torvalds if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio) 2732bd4c625cSLinus Torvalds journal->j_trans_max = 2733bd4c625cSLinus Torvalds JOURNAL_TRANS_MAX_DEFAULT / ratio; 27341da177e4SLinus Torvalds if (journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio) 2735bd4c625cSLinus Torvalds journal->j_trans_max = 2736bd4c625cSLinus Torvalds JOURNAL_TRANS_MIN_DEFAULT / ratio; 27371da177e4SLinus Torvalds 27381da177e4SLinus Torvalds if (journal->j_trans_max != initial) 2739bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 2740bd4c625cSLinus Torvalds "sh-461: journal_init: wrong transaction max size (%u). Changed to %u", 27411da177e4SLinus Torvalds initial, journal->j_trans_max); 27421da177e4SLinus Torvalds 27431da177e4SLinus Torvalds journal->j_max_batch = journal->j_trans_max * 27441da177e4SLinus Torvalds JOURNAL_MAX_BATCH_DEFAULT / JOURNAL_TRANS_MAX_DEFAULT; 27451da177e4SLinus Torvalds } 27461da177e4SLinus Torvalds 27471da177e4SLinus Torvalds if (!journal->j_trans_max) { 27481da177e4SLinus Torvalds /*we have the file system was created by old version of mkreiserfs 27491da177e4SLinus Torvalds so this field contains zero value */ 27501da177e4SLinus Torvalds journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT; 27511da177e4SLinus Torvalds journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT; 27521da177e4SLinus Torvalds journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE; 27531da177e4SLinus Torvalds 27541da177e4SLinus Torvalds /* for blocksize >= 4096 - max transaction size is 1024. For block size < 4096 27551da177e4SLinus Torvalds trans max size is decreased proportionally */ 27561da177e4SLinus Torvalds if (p_s_sb->s_blocksize < 4096) { 27571da177e4SLinus Torvalds journal->j_trans_max /= (4096 / p_s_sb->s_blocksize); 27581da177e4SLinus Torvalds journal->j_max_batch = (journal->j_trans_max) * 9 / 10; 27591da177e4SLinus Torvalds } 27601da177e4SLinus Torvalds } 27611da177e4SLinus Torvalds 27621da177e4SLinus Torvalds journal->j_default_max_commit_age = journal->j_max_commit_age; 27631da177e4SLinus Torvalds 27641da177e4SLinus Torvalds if (commit_max_age != 0) { 27651da177e4SLinus Torvalds journal->j_max_commit_age = commit_max_age; 27661da177e4SLinus Torvalds journal->j_max_trans_age = commit_max_age; 27671da177e4SLinus Torvalds } 27681da177e4SLinus Torvalds 27691da177e4SLinus Torvalds reiserfs_info(p_s_sb, "journal params: device %s, size %u, " 27701da177e4SLinus Torvalds "journal first block %u, max trans len %u, max batch %u, " 27711da177e4SLinus Torvalds "max commit age %u, max trans age %u\n", 27721da177e4SLinus Torvalds bdevname(journal->j_dev_bd, b), 27731da177e4SLinus Torvalds SB_ONDISK_JOURNAL_SIZE(p_s_sb), 27741da177e4SLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb), 27751da177e4SLinus Torvalds journal->j_trans_max, 27761da177e4SLinus Torvalds journal->j_max_batch, 2777bd4c625cSLinus Torvalds journal->j_max_commit_age, journal->j_max_trans_age); 27781da177e4SLinus Torvalds 27791da177e4SLinus Torvalds brelse(bhjh); 27801da177e4SLinus Torvalds 27811da177e4SLinus Torvalds journal->j_list_bitmap_index = 0; 27821da177e4SLinus Torvalds journal_list_init(p_s_sb); 27831da177e4SLinus Torvalds 2784bd4c625cSLinus Torvalds memset(journal->j_list_hash_table, 0, 2785bd4c625cSLinus Torvalds JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *)); 27861da177e4SLinus Torvalds 27871da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_dirty_buffers); 27881da177e4SLinus Torvalds spin_lock_init(&journal->j_dirty_buffers_lock); 27891da177e4SLinus Torvalds 27901da177e4SLinus Torvalds journal->j_start = 0; 27911da177e4SLinus Torvalds journal->j_len = 0; 27921da177e4SLinus Torvalds journal->j_len_alloc = 0; 27931da177e4SLinus Torvalds atomic_set(&(journal->j_wcount), 0); 27941da177e4SLinus Torvalds atomic_set(&(journal->j_async_throttle), 0); 27951da177e4SLinus Torvalds journal->j_bcount = 0; 27961da177e4SLinus Torvalds journal->j_trans_start_time = 0; 27971da177e4SLinus Torvalds journal->j_last = NULL; 27981da177e4SLinus Torvalds journal->j_first = NULL; 27991da177e4SLinus Torvalds init_waitqueue_head(&(journal->j_join_wait)); 28001da177e4SLinus Torvalds sema_init(&journal->j_lock, 1); 28011da177e4SLinus Torvalds sema_init(&journal->j_flush_sem, 1); 28021da177e4SLinus Torvalds 28031da177e4SLinus Torvalds journal->j_trans_id = 10; 28041da177e4SLinus Torvalds journal->j_mount_id = 10; 28051da177e4SLinus Torvalds journal->j_state = 0; 28061da177e4SLinus Torvalds atomic_set(&(journal->j_jlock), 0); 28071da177e4SLinus Torvalds journal->j_cnode_free_list = allocate_cnodes(num_cnodes); 28081da177e4SLinus Torvalds journal->j_cnode_free_orig = journal->j_cnode_free_list; 28091da177e4SLinus Torvalds journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0; 28101da177e4SLinus Torvalds journal->j_cnode_used = 0; 28111da177e4SLinus Torvalds journal->j_must_wait = 0; 28121da177e4SLinus Torvalds 2813576f6d79SJeff Mahoney if (journal->j_cnode_free == 0) { 2814576f6d79SJeff Mahoney reiserfs_warning(p_s_sb, "journal-2004: Journal cnode memory " 2815576f6d79SJeff Mahoney "allocation failed (%ld bytes). Journal is " 2816576f6d79SJeff Mahoney "too large for available memory. Usually " 2817576f6d79SJeff Mahoney "this is due to a journal that is too large.", 2818576f6d79SJeff Mahoney sizeof (struct reiserfs_journal_cnode) * num_cnodes); 2819576f6d79SJeff Mahoney goto free_and_return; 2820576f6d79SJeff Mahoney } 2821576f6d79SJeff Mahoney 28221da177e4SLinus Torvalds init_journal_hash(p_s_sb); 28231da177e4SLinus Torvalds jl = journal->j_current_jl; 28241da177e4SLinus Torvalds jl->j_list_bitmap = get_list_bitmap(p_s_sb, jl); 28251da177e4SLinus Torvalds if (!jl->j_list_bitmap) { 2826bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 2827bd4c625cSLinus Torvalds "journal-2005, get_list_bitmap failed for journal list 0"); 28281da177e4SLinus Torvalds goto free_and_return; 28291da177e4SLinus Torvalds } 28301da177e4SLinus Torvalds if (journal_read(p_s_sb) < 0) { 28311da177e4SLinus Torvalds reiserfs_warning(p_s_sb, "Replay Failure, unable to mount"); 28321da177e4SLinus Torvalds goto free_and_return; 28331da177e4SLinus Torvalds } 28341da177e4SLinus Torvalds 28351da177e4SLinus Torvalds reiserfs_mounted_fs_count++; 28361da177e4SLinus Torvalds if (reiserfs_mounted_fs_count <= 1) 28371da177e4SLinus Torvalds commit_wq = create_workqueue("reiserfs"); 28381da177e4SLinus Torvalds 2839c4028958SDavid Howells INIT_DELAYED_WORK(&journal->j_work, flush_async_commits); 2840c4028958SDavid Howells journal->j_work_sb = p_s_sb; 28411da177e4SLinus Torvalds return 0; 28421da177e4SLinus Torvalds free_and_return: 28431da177e4SLinus Torvalds free_journal_ram(p_s_sb); 28441da177e4SLinus Torvalds return 1; 28451da177e4SLinus Torvalds } 28461da177e4SLinus Torvalds 28471da177e4SLinus Torvalds /* 28481da177e4SLinus Torvalds ** test for a polite end of the current transaction. Used by file_write, and should 28491da177e4SLinus Torvalds ** be used by delete to make sure they don't write more than can fit inside a single 28501da177e4SLinus Torvalds ** transaction 28511da177e4SLinus Torvalds */ 2852bd4c625cSLinus Torvalds int journal_transaction_should_end(struct reiserfs_transaction_handle *th, 2853bd4c625cSLinus Torvalds int new_alloc) 2854bd4c625cSLinus Torvalds { 28551da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(th->t_super); 28561da177e4SLinus Torvalds time_t now = get_seconds(); 28571da177e4SLinus Torvalds /* cannot restart while nested */ 28581da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 28591da177e4SLinus Torvalds if (th->t_refcount > 1) 28601da177e4SLinus Torvalds return 0; 28611da177e4SLinus Torvalds if (journal->j_must_wait > 0 || 28621da177e4SLinus Torvalds (journal->j_len_alloc + new_alloc) >= journal->j_max_batch || 28631da177e4SLinus Torvalds atomic_read(&(journal->j_jlock)) || 28641da177e4SLinus Torvalds (now - journal->j_trans_start_time) > journal->j_max_trans_age || 28651da177e4SLinus Torvalds journal->j_cnode_free < (journal->j_trans_max * 3)) { 28661da177e4SLinus Torvalds return 1; 28671da177e4SLinus Torvalds } 28686ae1ea44SChris Mason /* protected by the BKL here */ 28696ae1ea44SChris Mason journal->j_len_alloc += new_alloc; 28706ae1ea44SChris Mason th->t_blocks_allocated += new_alloc ; 28711da177e4SLinus Torvalds return 0; 28721da177e4SLinus Torvalds } 28731da177e4SLinus Torvalds 28741da177e4SLinus Torvalds /* this must be called inside a transaction, and requires the 28751da177e4SLinus Torvalds ** kernel_lock to be held 28761da177e4SLinus Torvalds */ 2877bd4c625cSLinus Torvalds void reiserfs_block_writes(struct reiserfs_transaction_handle *th) 2878bd4c625cSLinus Torvalds { 28791da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(th->t_super); 28801da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 28811da177e4SLinus Torvalds journal->j_must_wait = 1; 28821da177e4SLinus Torvalds set_bit(J_WRITERS_BLOCKED, &journal->j_state); 28831da177e4SLinus Torvalds return; 28841da177e4SLinus Torvalds } 28851da177e4SLinus Torvalds 28861da177e4SLinus Torvalds /* this must be called without a transaction started, and does not 28871da177e4SLinus Torvalds ** require BKL 28881da177e4SLinus Torvalds */ 2889bd4c625cSLinus Torvalds void reiserfs_allow_writes(struct super_block *s) 2890bd4c625cSLinus Torvalds { 28911da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 28921da177e4SLinus Torvalds clear_bit(J_WRITERS_BLOCKED, &journal->j_state); 28931da177e4SLinus Torvalds wake_up(&journal->j_join_wait); 28941da177e4SLinus Torvalds } 28951da177e4SLinus Torvalds 28961da177e4SLinus Torvalds /* this must be called without a transaction started, and does not 28971da177e4SLinus Torvalds ** require BKL 28981da177e4SLinus Torvalds */ 2899bd4c625cSLinus Torvalds void reiserfs_wait_on_write_block(struct super_block *s) 2900bd4c625cSLinus Torvalds { 29011da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 29021da177e4SLinus Torvalds wait_event(journal->j_join_wait, 29031da177e4SLinus Torvalds !test_bit(J_WRITERS_BLOCKED, &journal->j_state)); 29041da177e4SLinus Torvalds } 29051da177e4SLinus Torvalds 2906bd4c625cSLinus Torvalds static void queue_log_writer(struct super_block *s) 2907bd4c625cSLinus Torvalds { 29081da177e4SLinus Torvalds wait_queue_t wait; 29091da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 29101da177e4SLinus Torvalds set_bit(J_WRITERS_QUEUED, &journal->j_state); 29111da177e4SLinus Torvalds 29121da177e4SLinus Torvalds /* 29131da177e4SLinus Torvalds * we don't want to use wait_event here because 29141da177e4SLinus Torvalds * we only want to wait once. 29151da177e4SLinus Torvalds */ 29161da177e4SLinus Torvalds init_waitqueue_entry(&wait, current); 29171da177e4SLinus Torvalds add_wait_queue(&journal->j_join_wait, &wait); 29181da177e4SLinus Torvalds set_current_state(TASK_UNINTERRUPTIBLE); 29191da177e4SLinus Torvalds if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) 29201da177e4SLinus Torvalds schedule(); 29215ab2f7e0SMilind Arun Choudhary __set_current_state(TASK_RUNNING); 29221da177e4SLinus Torvalds remove_wait_queue(&journal->j_join_wait, &wait); 29231da177e4SLinus Torvalds } 29241da177e4SLinus Torvalds 2925bd4c625cSLinus Torvalds static void wake_queued_writers(struct super_block *s) 2926bd4c625cSLinus Torvalds { 29271da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 29281da177e4SLinus Torvalds if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state)) 29291da177e4SLinus Torvalds wake_up(&journal->j_join_wait); 29301da177e4SLinus Torvalds } 29311da177e4SLinus Torvalds 2932bd4c625cSLinus Torvalds static void let_transaction_grow(struct super_block *sb, unsigned long trans_id) 29331da177e4SLinus Torvalds { 29341da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(sb); 29351da177e4SLinus Torvalds unsigned long bcount = journal->j_bcount; 29361da177e4SLinus Torvalds while (1) { 2937041e0e3bSNishanth Aravamudan schedule_timeout_uninterruptible(1); 29381da177e4SLinus Torvalds journal->j_current_jl->j_state |= LIST_COMMIT_PENDING; 29391da177e4SLinus Torvalds while ((atomic_read(&journal->j_wcount) > 0 || 29401da177e4SLinus Torvalds atomic_read(&journal->j_jlock)) && 29411da177e4SLinus Torvalds journal->j_trans_id == trans_id) { 29421da177e4SLinus Torvalds queue_log_writer(sb); 29431da177e4SLinus Torvalds } 29441da177e4SLinus Torvalds if (journal->j_trans_id != trans_id) 29451da177e4SLinus Torvalds break; 29461da177e4SLinus Torvalds if (bcount == journal->j_bcount) 29471da177e4SLinus Torvalds break; 29481da177e4SLinus Torvalds bcount = journal->j_bcount; 29491da177e4SLinus Torvalds } 29501da177e4SLinus Torvalds } 29511da177e4SLinus Torvalds 29521da177e4SLinus Torvalds /* join == true if you must join an existing transaction. 29531da177e4SLinus Torvalds ** join == false if you can deal with waiting for others to finish 29541da177e4SLinus Torvalds ** 29551da177e4SLinus Torvalds ** this will block until the transaction is joinable. send the number of blocks you 29561da177e4SLinus Torvalds ** expect to use in nblocks. 29571da177e4SLinus Torvalds */ 2958bd4c625cSLinus Torvalds static int do_journal_begin_r(struct reiserfs_transaction_handle *th, 2959bd4c625cSLinus Torvalds struct super_block *p_s_sb, unsigned long nblocks, 2960bd4c625cSLinus Torvalds int join) 2961bd4c625cSLinus Torvalds { 29621da177e4SLinus Torvalds time_t now = get_seconds(); 29631da177e4SLinus Torvalds int old_trans_id; 29641da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 29651da177e4SLinus Torvalds struct reiserfs_transaction_handle myth; 29661da177e4SLinus Torvalds int sched_count = 0; 29671da177e4SLinus Torvalds int retval; 29681da177e4SLinus Torvalds 29691da177e4SLinus Torvalds reiserfs_check_lock_depth(p_s_sb, "journal_begin"); 297014a61442SEric Sesterhenn BUG_ON(nblocks > journal->j_trans_max); 29711da177e4SLinus Torvalds 29721da177e4SLinus Torvalds PROC_INFO_INC(p_s_sb, journal.journal_being); 29731da177e4SLinus Torvalds /* set here for journal_join */ 29741da177e4SLinus Torvalds th->t_refcount = 1; 29751da177e4SLinus Torvalds th->t_super = p_s_sb; 29761da177e4SLinus Torvalds 29771da177e4SLinus Torvalds relock: 29781da177e4SLinus Torvalds lock_journal(p_s_sb); 29791da177e4SLinus Torvalds if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) { 29801da177e4SLinus Torvalds unlock_journal(p_s_sb); 29811da177e4SLinus Torvalds retval = journal->j_errno; 29821da177e4SLinus Torvalds goto out_fail; 29831da177e4SLinus Torvalds } 29841da177e4SLinus Torvalds journal->j_bcount++; 29851da177e4SLinus Torvalds 29861da177e4SLinus Torvalds if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) { 29871da177e4SLinus Torvalds unlock_journal(p_s_sb); 29881da177e4SLinus Torvalds reiserfs_wait_on_write_block(p_s_sb); 29891da177e4SLinus Torvalds PROC_INFO_INC(p_s_sb, journal.journal_relock_writers); 29901da177e4SLinus Torvalds goto relock; 29911da177e4SLinus Torvalds } 29921da177e4SLinus Torvalds now = get_seconds(); 29931da177e4SLinus Torvalds 29941da177e4SLinus Torvalds /* if there is no room in the journal OR 29951da177e4SLinus Torvalds ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning 29961da177e4SLinus Torvalds ** we don't sleep if there aren't other writers 29971da177e4SLinus Torvalds */ 29981da177e4SLinus Torvalds 29991da177e4SLinus Torvalds if ((!join && journal->j_must_wait > 0) || 3000bd4c625cSLinus Torvalds (!join 3001bd4c625cSLinus Torvalds && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch) 3002bd4c625cSLinus Torvalds || (!join && atomic_read(&journal->j_wcount) > 0 3003bd4c625cSLinus Torvalds && journal->j_trans_start_time > 0 3004bd4c625cSLinus Torvalds && (now - journal->j_trans_start_time) > 3005bd4c625cSLinus Torvalds journal->j_max_trans_age) || (!join 3006bd4c625cSLinus Torvalds && atomic_read(&journal->j_jlock)) 3007bd4c625cSLinus Torvalds || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) { 30081da177e4SLinus Torvalds 30091da177e4SLinus Torvalds old_trans_id = journal->j_trans_id; 30101da177e4SLinus Torvalds unlock_journal(p_s_sb); /* allow others to finish this transaction */ 30111da177e4SLinus Torvalds 30121da177e4SLinus Torvalds if (!join && (journal->j_len_alloc + nblocks + 2) >= 30131da177e4SLinus Torvalds journal->j_max_batch && 3014bd4c625cSLinus Torvalds ((journal->j_len + nblocks + 2) * 100) < 3015bd4c625cSLinus Torvalds (journal->j_len_alloc * 75)) { 30161da177e4SLinus Torvalds if (atomic_read(&journal->j_wcount) > 10) { 30171da177e4SLinus Torvalds sched_count++; 30181da177e4SLinus Torvalds queue_log_writer(p_s_sb); 30191da177e4SLinus Torvalds goto relock; 30201da177e4SLinus Torvalds } 30211da177e4SLinus Torvalds } 30221da177e4SLinus Torvalds /* don't mess with joining the transaction if all we have to do is 30231da177e4SLinus Torvalds * wait for someone else to do a commit 30241da177e4SLinus Torvalds */ 30251da177e4SLinus Torvalds if (atomic_read(&journal->j_jlock)) { 30261da177e4SLinus Torvalds while (journal->j_trans_id == old_trans_id && 30271da177e4SLinus Torvalds atomic_read(&journal->j_jlock)) { 30281da177e4SLinus Torvalds queue_log_writer(p_s_sb); 30291da177e4SLinus Torvalds } 30301da177e4SLinus Torvalds goto relock; 30311da177e4SLinus Torvalds } 30321da177e4SLinus Torvalds retval = journal_join(&myth, p_s_sb, 1); 30331da177e4SLinus Torvalds if (retval) 30341da177e4SLinus Torvalds goto out_fail; 30351da177e4SLinus Torvalds 30361da177e4SLinus Torvalds /* someone might have ended the transaction while we joined */ 30371da177e4SLinus Torvalds if (old_trans_id != journal->j_trans_id) { 30381da177e4SLinus Torvalds retval = do_journal_end(&myth, p_s_sb, 1, 0); 30391da177e4SLinus Torvalds } else { 30401da177e4SLinus Torvalds retval = do_journal_end(&myth, p_s_sb, 1, COMMIT_NOW); 30411da177e4SLinus Torvalds } 30421da177e4SLinus Torvalds 30431da177e4SLinus Torvalds if (retval) 30441da177e4SLinus Torvalds goto out_fail; 30451da177e4SLinus Torvalds 30461da177e4SLinus Torvalds PROC_INFO_INC(p_s_sb, journal.journal_relock_wcount); 30471da177e4SLinus Torvalds goto relock; 30481da177e4SLinus Torvalds } 30491da177e4SLinus Torvalds /* we are the first writer, set trans_id */ 30501da177e4SLinus Torvalds if (journal->j_trans_start_time == 0) { 30511da177e4SLinus Torvalds journal->j_trans_start_time = get_seconds(); 30521da177e4SLinus Torvalds } 30531da177e4SLinus Torvalds atomic_inc(&(journal->j_wcount)); 30541da177e4SLinus Torvalds journal->j_len_alloc += nblocks; 30551da177e4SLinus Torvalds th->t_blocks_logged = 0; 30561da177e4SLinus Torvalds th->t_blocks_allocated = nblocks; 30571da177e4SLinus Torvalds th->t_trans_id = journal->j_trans_id; 30581da177e4SLinus Torvalds unlock_journal(p_s_sb); 30591da177e4SLinus Torvalds INIT_LIST_HEAD(&th->t_list); 306022e2c507SJens Axboe get_fs_excl(); 30611da177e4SLinus Torvalds return 0; 30621da177e4SLinus Torvalds 30631da177e4SLinus Torvalds out_fail: 30641da177e4SLinus Torvalds memset(th, 0, sizeof(*th)); 30651da177e4SLinus Torvalds /* Re-set th->t_super, so we can properly keep track of how many 30661da177e4SLinus Torvalds * persistent transactions there are. We need to do this so if this 30671da177e4SLinus Torvalds * call is part of a failed restart_transaction, we can free it later */ 30681da177e4SLinus Torvalds th->t_super = p_s_sb; 30691da177e4SLinus Torvalds return retval; 30701da177e4SLinus Torvalds } 30711da177e4SLinus Torvalds 3072bd4c625cSLinus Torvalds struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct 3073bd4c625cSLinus Torvalds super_block 3074bd4c625cSLinus Torvalds *s, 3075bd4c625cSLinus Torvalds int nblocks) 3076bd4c625cSLinus Torvalds { 30771da177e4SLinus Torvalds int ret; 30781da177e4SLinus Torvalds struct reiserfs_transaction_handle *th; 30791da177e4SLinus Torvalds 30801da177e4SLinus Torvalds /* if we're nesting into an existing transaction. It will be 30811da177e4SLinus Torvalds ** persistent on its own 30821da177e4SLinus Torvalds */ 30831da177e4SLinus Torvalds if (reiserfs_transaction_running(s)) { 30841da177e4SLinus Torvalds th = current->journal_info; 30851da177e4SLinus Torvalds th->t_refcount++; 308614a61442SEric Sesterhenn BUG_ON(th->t_refcount < 2); 308714a61442SEric Sesterhenn 30881da177e4SLinus Torvalds return th; 30891da177e4SLinus Torvalds } 3090d739b42bSPekka Enberg th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS); 30911da177e4SLinus Torvalds if (!th) 30921da177e4SLinus Torvalds return NULL; 30931da177e4SLinus Torvalds ret = journal_begin(th, s, nblocks); 30941da177e4SLinus Torvalds if (ret) { 3095d739b42bSPekka Enberg kfree(th); 30961da177e4SLinus Torvalds return NULL; 30971da177e4SLinus Torvalds } 30981da177e4SLinus Torvalds 30991da177e4SLinus Torvalds SB_JOURNAL(s)->j_persistent_trans++; 31001da177e4SLinus Torvalds return th; 31011da177e4SLinus Torvalds } 31021da177e4SLinus Torvalds 3103bd4c625cSLinus Torvalds int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th) 3104bd4c625cSLinus Torvalds { 31051da177e4SLinus Torvalds struct super_block *s = th->t_super; 31061da177e4SLinus Torvalds int ret = 0; 31071da177e4SLinus Torvalds if (th->t_trans_id) 31081da177e4SLinus Torvalds ret = journal_end(th, th->t_super, th->t_blocks_allocated); 31091da177e4SLinus Torvalds else 31101da177e4SLinus Torvalds ret = -EIO; 31111da177e4SLinus Torvalds if (th->t_refcount == 0) { 31121da177e4SLinus Torvalds SB_JOURNAL(s)->j_persistent_trans--; 3113d739b42bSPekka Enberg kfree(th); 31141da177e4SLinus Torvalds } 31151da177e4SLinus Torvalds return ret; 31161da177e4SLinus Torvalds } 31171da177e4SLinus Torvalds 3118bd4c625cSLinus Torvalds static int journal_join(struct reiserfs_transaction_handle *th, 3119bd4c625cSLinus Torvalds struct super_block *p_s_sb, unsigned long nblocks) 3120bd4c625cSLinus Torvalds { 31211da177e4SLinus Torvalds struct reiserfs_transaction_handle *cur_th = current->journal_info; 31221da177e4SLinus Torvalds 31231da177e4SLinus Torvalds /* this keeps do_journal_end from NULLing out the current->journal_info 31241da177e4SLinus Torvalds ** pointer 31251da177e4SLinus Torvalds */ 31261da177e4SLinus Torvalds th->t_handle_save = cur_th; 312714a61442SEric Sesterhenn BUG_ON(cur_th && cur_th->t_refcount > 1); 31281da177e4SLinus Torvalds return do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_JOIN); 31291da177e4SLinus Torvalds } 31301da177e4SLinus Torvalds 3131bd4c625cSLinus Torvalds int journal_join_abort(struct reiserfs_transaction_handle *th, 3132bd4c625cSLinus Torvalds struct super_block *p_s_sb, unsigned long nblocks) 3133bd4c625cSLinus Torvalds { 31341da177e4SLinus Torvalds struct reiserfs_transaction_handle *cur_th = current->journal_info; 31351da177e4SLinus Torvalds 31361da177e4SLinus Torvalds /* this keeps do_journal_end from NULLing out the current->journal_info 31371da177e4SLinus Torvalds ** pointer 31381da177e4SLinus Torvalds */ 31391da177e4SLinus Torvalds th->t_handle_save = cur_th; 314014a61442SEric Sesterhenn BUG_ON(cur_th && cur_th->t_refcount > 1); 31411da177e4SLinus Torvalds return do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_ABORT); 31421da177e4SLinus Torvalds } 31431da177e4SLinus Torvalds 3144bd4c625cSLinus Torvalds int journal_begin(struct reiserfs_transaction_handle *th, 3145bd4c625cSLinus Torvalds struct super_block *p_s_sb, unsigned long nblocks) 3146bd4c625cSLinus Torvalds { 31471da177e4SLinus Torvalds struct reiserfs_transaction_handle *cur_th = current->journal_info; 31481da177e4SLinus Torvalds int ret; 31491da177e4SLinus Torvalds 31501da177e4SLinus Torvalds th->t_handle_save = NULL; 31511da177e4SLinus Torvalds if (cur_th) { 31521da177e4SLinus Torvalds /* we are nesting into the current transaction */ 31531da177e4SLinus Torvalds if (cur_th->t_super == p_s_sb) { 31541da177e4SLinus Torvalds BUG_ON(!cur_th->t_refcount); 31551da177e4SLinus Torvalds cur_th->t_refcount++; 31561da177e4SLinus Torvalds memcpy(th, cur_th, sizeof(*th)); 31571da177e4SLinus Torvalds if (th->t_refcount <= 1) 3158bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 3159bd4c625cSLinus Torvalds "BAD: refcount <= 1, but journal_info != 0"); 31601da177e4SLinus Torvalds return 0; 31611da177e4SLinus Torvalds } else { 31621da177e4SLinus Torvalds /* we've ended up with a handle from a different filesystem. 31631da177e4SLinus Torvalds ** save it and restore on journal_end. This should never 31641da177e4SLinus Torvalds ** really happen... 31651da177e4SLinus Torvalds */ 3166bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 3167bd4c625cSLinus Torvalds "clm-2100: nesting info a different FS"); 31681da177e4SLinus Torvalds th->t_handle_save = current->journal_info; 31691da177e4SLinus Torvalds current->journal_info = th; 31701da177e4SLinus Torvalds } 31711da177e4SLinus Torvalds } else { 31721da177e4SLinus Torvalds current->journal_info = th; 31731da177e4SLinus Torvalds } 31741da177e4SLinus Torvalds ret = do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_REG); 317514a61442SEric Sesterhenn BUG_ON(current->journal_info != th); 31761da177e4SLinus Torvalds 31771da177e4SLinus Torvalds /* I guess this boils down to being the reciprocal of clm-2100 above. 31781da177e4SLinus Torvalds * If do_journal_begin_r fails, we need to put it back, since journal_end 31791da177e4SLinus Torvalds * won't be called to do it. */ 31801da177e4SLinus Torvalds if (ret) 31811da177e4SLinus Torvalds current->journal_info = th->t_handle_save; 31821da177e4SLinus Torvalds else 31831da177e4SLinus Torvalds BUG_ON(!th->t_refcount); 31841da177e4SLinus Torvalds 31851da177e4SLinus Torvalds return ret; 31861da177e4SLinus Torvalds } 31871da177e4SLinus Torvalds 31881da177e4SLinus Torvalds /* 31891da177e4SLinus Torvalds ** puts bh into the current transaction. If it was already there, reorders removes the 31901da177e4SLinus Torvalds ** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order). 31911da177e4SLinus Torvalds ** 31921da177e4SLinus Torvalds ** if it was dirty, cleans and files onto the clean list. I can't let it be dirty again until the 31931da177e4SLinus Torvalds ** transaction is committed. 31941da177e4SLinus Torvalds ** 31951da177e4SLinus Torvalds ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len. 31961da177e4SLinus Torvalds */ 3197bd4c625cSLinus Torvalds int journal_mark_dirty(struct reiserfs_transaction_handle *th, 3198bd4c625cSLinus Torvalds struct super_block *p_s_sb, struct buffer_head *bh) 3199bd4c625cSLinus Torvalds { 32001da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 32011da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn = NULL; 32021da177e4SLinus Torvalds int count_already_incd = 0; 32031da177e4SLinus Torvalds int prepared = 0; 32041da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 32051da177e4SLinus Torvalds 32061da177e4SLinus Torvalds PROC_INFO_INC(p_s_sb, journal.mark_dirty); 32071da177e4SLinus Torvalds if (th->t_trans_id != journal->j_trans_id) { 3208bd4c625cSLinus Torvalds reiserfs_panic(th->t_super, 3209bd4c625cSLinus Torvalds "journal-1577: handle trans id %ld != current trans id %ld\n", 32101da177e4SLinus Torvalds th->t_trans_id, journal->j_trans_id); 32111da177e4SLinus Torvalds } 32121da177e4SLinus Torvalds 32131da177e4SLinus Torvalds p_s_sb->s_dirt = 1; 32141da177e4SLinus Torvalds 32151da177e4SLinus Torvalds prepared = test_clear_buffer_journal_prepared(bh); 32161da177e4SLinus Torvalds clear_buffer_journal_restore_dirty(bh); 32171da177e4SLinus Torvalds /* already in this transaction, we are done */ 32181da177e4SLinus Torvalds if (buffer_journaled(bh)) { 32191da177e4SLinus Torvalds PROC_INFO_INC(p_s_sb, journal.mark_dirty_already); 32201da177e4SLinus Torvalds return 0; 32211da177e4SLinus Torvalds } 32221da177e4SLinus Torvalds 32231da177e4SLinus Torvalds /* this must be turned into a panic instead of a warning. We can't allow 32241da177e4SLinus Torvalds ** a dirty or journal_dirty or locked buffer to be logged, as some changes 32251da177e4SLinus Torvalds ** could get to disk too early. NOT GOOD. 32261da177e4SLinus Torvalds */ 32271da177e4SLinus Torvalds if (!prepared || buffer_dirty(bh)) { 32281da177e4SLinus Torvalds reiserfs_warning(p_s_sb, "journal-1777: buffer %llu bad state " 32291da177e4SLinus Torvalds "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT", 3230bd4c625cSLinus Torvalds (unsigned long long)bh->b_blocknr, 3231bd4c625cSLinus Torvalds prepared ? ' ' : '!', 32321da177e4SLinus Torvalds buffer_locked(bh) ? ' ' : '!', 32331da177e4SLinus Torvalds buffer_dirty(bh) ? ' ' : '!', 32341da177e4SLinus Torvalds buffer_journal_dirty(bh) ? ' ' : '!'); 32351da177e4SLinus Torvalds } 32361da177e4SLinus Torvalds 32371da177e4SLinus Torvalds if (atomic_read(&(journal->j_wcount)) <= 0) { 3238bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 3239bd4c625cSLinus Torvalds "journal-1409: journal_mark_dirty returning because j_wcount was %d", 3240bd4c625cSLinus Torvalds atomic_read(&(journal->j_wcount))); 32411da177e4SLinus Torvalds return 1; 32421da177e4SLinus Torvalds } 32431da177e4SLinus Torvalds /* this error means I've screwed up, and we've overflowed the transaction. 32441da177e4SLinus Torvalds ** Nothing can be done here, except make the FS readonly or panic. 32451da177e4SLinus Torvalds */ 32461da177e4SLinus Torvalds if (journal->j_len >= journal->j_trans_max) { 3247bd4c625cSLinus Torvalds reiserfs_panic(th->t_super, 3248bd4c625cSLinus Torvalds "journal-1413: journal_mark_dirty: j_len (%lu) is too big\n", 3249bd4c625cSLinus Torvalds journal->j_len); 32501da177e4SLinus Torvalds } 32511da177e4SLinus Torvalds 32521da177e4SLinus Torvalds if (buffer_journal_dirty(bh)) { 32531da177e4SLinus Torvalds count_already_incd = 1; 32541da177e4SLinus Torvalds PROC_INFO_INC(p_s_sb, journal.mark_dirty_notjournal); 32551da177e4SLinus Torvalds clear_buffer_journal_dirty(bh); 32561da177e4SLinus Torvalds } 32571da177e4SLinus Torvalds 32581da177e4SLinus Torvalds if (journal->j_len > journal->j_len_alloc) { 32591da177e4SLinus Torvalds journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT; 32601da177e4SLinus Torvalds } 32611da177e4SLinus Torvalds 32621da177e4SLinus Torvalds set_buffer_journaled(bh); 32631da177e4SLinus Torvalds 32641da177e4SLinus Torvalds /* now put this guy on the end */ 32651da177e4SLinus Torvalds if (!cn) { 32661da177e4SLinus Torvalds cn = get_cnode(p_s_sb); 32671da177e4SLinus Torvalds if (!cn) { 32681da177e4SLinus Torvalds reiserfs_panic(p_s_sb, "get_cnode failed!\n"); 32691da177e4SLinus Torvalds } 32701da177e4SLinus Torvalds 32711da177e4SLinus Torvalds if (th->t_blocks_logged == th->t_blocks_allocated) { 32721da177e4SLinus Torvalds th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT; 32731da177e4SLinus Torvalds journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT; 32741da177e4SLinus Torvalds } 32751da177e4SLinus Torvalds th->t_blocks_logged++; 32761da177e4SLinus Torvalds journal->j_len++; 32771da177e4SLinus Torvalds 32781da177e4SLinus Torvalds cn->bh = bh; 32791da177e4SLinus Torvalds cn->blocknr = bh->b_blocknr; 32801da177e4SLinus Torvalds cn->sb = p_s_sb; 32811da177e4SLinus Torvalds cn->jlist = NULL; 32821da177e4SLinus Torvalds insert_journal_hash(journal->j_hash_table, cn); 32831da177e4SLinus Torvalds if (!count_already_incd) { 32841da177e4SLinus Torvalds get_bh(bh); 32851da177e4SLinus Torvalds } 32861da177e4SLinus Torvalds } 32871da177e4SLinus Torvalds cn->next = NULL; 32881da177e4SLinus Torvalds cn->prev = journal->j_last; 32891da177e4SLinus Torvalds cn->bh = bh; 32901da177e4SLinus Torvalds if (journal->j_last) { 32911da177e4SLinus Torvalds journal->j_last->next = cn; 32921da177e4SLinus Torvalds journal->j_last = cn; 32931da177e4SLinus Torvalds } else { 32941da177e4SLinus Torvalds journal->j_first = cn; 32951da177e4SLinus Torvalds journal->j_last = cn; 32961da177e4SLinus Torvalds } 32971da177e4SLinus Torvalds return 0; 32981da177e4SLinus Torvalds } 32991da177e4SLinus Torvalds 3300bd4c625cSLinus Torvalds int journal_end(struct reiserfs_transaction_handle *th, 3301bd4c625cSLinus Torvalds struct super_block *p_s_sb, unsigned long nblocks) 3302bd4c625cSLinus Torvalds { 33031da177e4SLinus Torvalds if (!current->journal_info && th->t_refcount > 1) 33041da177e4SLinus Torvalds reiserfs_warning(p_s_sb, "REISER-NESTING: th NULL, refcount %d", 33051da177e4SLinus Torvalds th->t_refcount); 33061da177e4SLinus Torvalds 33071da177e4SLinus Torvalds if (!th->t_trans_id) { 33081da177e4SLinus Torvalds WARN_ON(1); 33091da177e4SLinus Torvalds return -EIO; 33101da177e4SLinus Torvalds } 33111da177e4SLinus Torvalds 33121da177e4SLinus Torvalds th->t_refcount--; 33131da177e4SLinus Torvalds if (th->t_refcount > 0) { 3314bd4c625cSLinus Torvalds struct reiserfs_transaction_handle *cur_th = 3315bd4c625cSLinus Torvalds current->journal_info; 33161da177e4SLinus Torvalds 33171da177e4SLinus Torvalds /* we aren't allowed to close a nested transaction on a different 33181da177e4SLinus Torvalds ** filesystem from the one in the task struct 33191da177e4SLinus Torvalds */ 332014a61442SEric Sesterhenn BUG_ON(cur_th->t_super != th->t_super); 33211da177e4SLinus Torvalds 33221da177e4SLinus Torvalds if (th != cur_th) { 33231da177e4SLinus Torvalds memcpy(current->journal_info, th, sizeof(*th)); 33241da177e4SLinus Torvalds th->t_trans_id = 0; 33251da177e4SLinus Torvalds } 33261da177e4SLinus Torvalds return 0; 33271da177e4SLinus Torvalds } else { 33281da177e4SLinus Torvalds return do_journal_end(th, p_s_sb, nblocks, 0); 33291da177e4SLinus Torvalds } 33301da177e4SLinus Torvalds } 33311da177e4SLinus Torvalds 33321da177e4SLinus Torvalds /* removes from the current transaction, relsing and descrementing any counters. 33331da177e4SLinus Torvalds ** also files the removed buffer directly onto the clean list 33341da177e4SLinus Torvalds ** 33351da177e4SLinus Torvalds ** called by journal_mark_freed when a block has been deleted 33361da177e4SLinus Torvalds ** 33371da177e4SLinus Torvalds ** returns 1 if it cleaned and relsed the buffer. 0 otherwise 33381da177e4SLinus Torvalds */ 3339bd4c625cSLinus Torvalds static int remove_from_transaction(struct super_block *p_s_sb, 3340bd4c625cSLinus Torvalds b_blocknr_t blocknr, int already_cleaned) 3341bd4c625cSLinus Torvalds { 33421da177e4SLinus Torvalds struct buffer_head *bh; 33431da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 33441da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 33451da177e4SLinus Torvalds int ret = 0; 33461da177e4SLinus Torvalds 33471da177e4SLinus Torvalds cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, blocknr); 33481da177e4SLinus Torvalds if (!cn || !cn->bh) { 33491da177e4SLinus Torvalds return ret; 33501da177e4SLinus Torvalds } 33511da177e4SLinus Torvalds bh = cn->bh; 33521da177e4SLinus Torvalds if (cn->prev) { 33531da177e4SLinus Torvalds cn->prev->next = cn->next; 33541da177e4SLinus Torvalds } 33551da177e4SLinus Torvalds if (cn->next) { 33561da177e4SLinus Torvalds cn->next->prev = cn->prev; 33571da177e4SLinus Torvalds } 33581da177e4SLinus Torvalds if (cn == journal->j_first) { 33591da177e4SLinus Torvalds journal->j_first = cn->next; 33601da177e4SLinus Torvalds } 33611da177e4SLinus Torvalds if (cn == journal->j_last) { 33621da177e4SLinus Torvalds journal->j_last = cn->prev; 33631da177e4SLinus Torvalds } 33641da177e4SLinus Torvalds if (bh) 3365bd4c625cSLinus Torvalds remove_journal_hash(p_s_sb, journal->j_hash_table, NULL, 3366bd4c625cSLinus Torvalds bh->b_blocknr, 0); 33671da177e4SLinus Torvalds clear_buffer_journaled(bh); /* don't log this one */ 33681da177e4SLinus Torvalds 33691da177e4SLinus Torvalds if (!already_cleaned) { 33701da177e4SLinus Torvalds clear_buffer_journal_dirty(bh); 33711da177e4SLinus Torvalds clear_buffer_dirty(bh); 33721da177e4SLinus Torvalds clear_buffer_journal_test(bh); 33731da177e4SLinus Torvalds put_bh(bh); 33741da177e4SLinus Torvalds if (atomic_read(&(bh->b_count)) < 0) { 3375bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 3376bd4c625cSLinus Torvalds "journal-1752: remove from trans, b_count < 0"); 33771da177e4SLinus Torvalds } 33781da177e4SLinus Torvalds ret = 1; 33791da177e4SLinus Torvalds } 33801da177e4SLinus Torvalds journal->j_len--; 33811da177e4SLinus Torvalds journal->j_len_alloc--; 33821da177e4SLinus Torvalds free_cnode(p_s_sb, cn); 33831da177e4SLinus Torvalds return ret; 33841da177e4SLinus Torvalds } 33851da177e4SLinus Torvalds 33861da177e4SLinus Torvalds /* 33871da177e4SLinus Torvalds ** for any cnode in a journal list, it can only be dirtied of all the 33880779bf2dSMatt LaPlante ** transactions that include it are committed to disk. 33891da177e4SLinus Torvalds ** this checks through each transaction, and returns 1 if you are allowed to dirty, 33901da177e4SLinus Torvalds ** and 0 if you aren't 33911da177e4SLinus Torvalds ** 33921da177e4SLinus Torvalds ** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log 33931da177e4SLinus Torvalds ** blocks for a given transaction on disk 33941da177e4SLinus Torvalds ** 33951da177e4SLinus Torvalds */ 3396bd4c625cSLinus Torvalds static int can_dirty(struct reiserfs_journal_cnode *cn) 3397bd4c625cSLinus Torvalds { 33981da177e4SLinus Torvalds struct super_block *sb = cn->sb; 33991da177e4SLinus Torvalds b_blocknr_t blocknr = cn->blocknr; 34001da177e4SLinus Torvalds struct reiserfs_journal_cnode *cur = cn->hprev; 34011da177e4SLinus Torvalds int can_dirty = 1; 34021da177e4SLinus Torvalds 34031da177e4SLinus Torvalds /* first test hprev. These are all newer than cn, so any node here 34041da177e4SLinus Torvalds ** with the same block number and dev means this node can't be sent 34051da177e4SLinus Torvalds ** to disk right now. 34061da177e4SLinus Torvalds */ 34071da177e4SLinus Torvalds while (cur && can_dirty) { 34081da177e4SLinus Torvalds if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb && 34091da177e4SLinus Torvalds cur->blocknr == blocknr) { 34101da177e4SLinus Torvalds can_dirty = 0; 34111da177e4SLinus Torvalds } 34121da177e4SLinus Torvalds cur = cur->hprev; 34131da177e4SLinus Torvalds } 34141da177e4SLinus Torvalds /* then test hnext. These are all older than cn. As long as they 34151da177e4SLinus Torvalds ** are committed to the log, it is safe to write cn to disk 34161da177e4SLinus Torvalds */ 34171da177e4SLinus Torvalds cur = cn->hnext; 34181da177e4SLinus Torvalds while (cur && can_dirty) { 34191da177e4SLinus Torvalds if (cur->jlist && cur->jlist->j_len > 0 && 34201da177e4SLinus Torvalds atomic_read(&(cur->jlist->j_commit_left)) > 0 && cur->bh && 34211da177e4SLinus Torvalds cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) { 34221da177e4SLinus Torvalds can_dirty = 0; 34231da177e4SLinus Torvalds } 34241da177e4SLinus Torvalds cur = cur->hnext; 34251da177e4SLinus Torvalds } 34261da177e4SLinus Torvalds return can_dirty; 34271da177e4SLinus Torvalds } 34281da177e4SLinus Torvalds 34291da177e4SLinus Torvalds /* syncs the commit blocks, but does not force the real buffers to disk 34300779bf2dSMatt LaPlante ** will wait until the current transaction is done/committed before returning 34311da177e4SLinus Torvalds */ 3432bd4c625cSLinus Torvalds int journal_end_sync(struct reiserfs_transaction_handle *th, 3433bd4c625cSLinus Torvalds struct super_block *p_s_sb, unsigned long nblocks) 3434bd4c625cSLinus Torvalds { 34351da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 34361da177e4SLinus Torvalds 34371da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 34381da177e4SLinus Torvalds /* you can sync while nested, very, very bad */ 343914a61442SEric Sesterhenn BUG_ON(th->t_refcount > 1); 34401da177e4SLinus Torvalds if (journal->j_len == 0) { 3441bd4c625cSLinus Torvalds reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb), 3442bd4c625cSLinus Torvalds 1); 34431da177e4SLinus Torvalds journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb)); 34441da177e4SLinus Torvalds } 34451da177e4SLinus Torvalds return do_journal_end(th, p_s_sb, nblocks, COMMIT_NOW | WAIT); 34461da177e4SLinus Torvalds } 34471da177e4SLinus Torvalds 34481da177e4SLinus Torvalds /* 34491da177e4SLinus Torvalds ** writeback the pending async commits to disk 34501da177e4SLinus Torvalds */ 3451c4028958SDavid Howells static void flush_async_commits(struct work_struct *work) 3452bd4c625cSLinus Torvalds { 3453c4028958SDavid Howells struct reiserfs_journal *journal = 3454c4028958SDavid Howells container_of(work, struct reiserfs_journal, j_work.work); 3455c4028958SDavid Howells struct super_block *p_s_sb = journal->j_work_sb; 34561da177e4SLinus Torvalds struct reiserfs_journal_list *jl; 34571da177e4SLinus Torvalds struct list_head *entry; 34581da177e4SLinus Torvalds 34591da177e4SLinus Torvalds lock_kernel(); 34601da177e4SLinus Torvalds if (!list_empty(&journal->j_journal_list)) { 34611da177e4SLinus Torvalds /* last entry is the youngest, commit it and you get everything */ 34621da177e4SLinus Torvalds entry = journal->j_journal_list.prev; 34631da177e4SLinus Torvalds jl = JOURNAL_LIST_ENTRY(entry); 34641da177e4SLinus Torvalds flush_commit_list(p_s_sb, jl, 1); 34651da177e4SLinus Torvalds } 34661da177e4SLinus Torvalds unlock_kernel(); 34671da177e4SLinus Torvalds } 34681da177e4SLinus Torvalds 34691da177e4SLinus Torvalds /* 34701da177e4SLinus Torvalds ** flushes any old transactions to disk 34711da177e4SLinus Torvalds ** ends the current transaction if it is too old 34721da177e4SLinus Torvalds */ 3473bd4c625cSLinus Torvalds int reiserfs_flush_old_commits(struct super_block *p_s_sb) 3474bd4c625cSLinus Torvalds { 34751da177e4SLinus Torvalds time_t now; 34761da177e4SLinus Torvalds struct reiserfs_transaction_handle th; 34771da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 34781da177e4SLinus Torvalds 34791da177e4SLinus Torvalds now = get_seconds(); 34801da177e4SLinus Torvalds /* safety check so we don't flush while we are replaying the log during 34811da177e4SLinus Torvalds * mount 34821da177e4SLinus Torvalds */ 34831da177e4SLinus Torvalds if (list_empty(&journal->j_journal_list)) { 34841da177e4SLinus Torvalds return 0; 34851da177e4SLinus Torvalds } 34861da177e4SLinus Torvalds 34871da177e4SLinus Torvalds /* check the current transaction. If there are no writers, and it is 34881da177e4SLinus Torvalds * too old, finish it, and force the commit blocks to disk 34891da177e4SLinus Torvalds */ 34901da177e4SLinus Torvalds if (atomic_read(&journal->j_wcount) <= 0 && 34911da177e4SLinus Torvalds journal->j_trans_start_time > 0 && 34921da177e4SLinus Torvalds journal->j_len > 0 && 3493bd4c625cSLinus Torvalds (now - journal->j_trans_start_time) > journal->j_max_trans_age) { 34941da177e4SLinus Torvalds if (!journal_join(&th, p_s_sb, 1)) { 3495bd4c625cSLinus Torvalds reiserfs_prepare_for_journal(p_s_sb, 3496bd4c625cSLinus Torvalds SB_BUFFER_WITH_SB(p_s_sb), 3497bd4c625cSLinus Torvalds 1); 3498bd4c625cSLinus Torvalds journal_mark_dirty(&th, p_s_sb, 3499bd4c625cSLinus Torvalds SB_BUFFER_WITH_SB(p_s_sb)); 35001da177e4SLinus Torvalds 35011da177e4SLinus Torvalds /* we're only being called from kreiserfsd, it makes no sense to do 35021da177e4SLinus Torvalds ** an async commit so that kreiserfsd can do it later 35031da177e4SLinus Torvalds */ 35041da177e4SLinus Torvalds do_journal_end(&th, p_s_sb, 1, COMMIT_NOW | WAIT); 35051da177e4SLinus Torvalds } 35061da177e4SLinus Torvalds } 35071da177e4SLinus Torvalds return p_s_sb->s_dirt; 35081da177e4SLinus Torvalds } 35091da177e4SLinus Torvalds 35101da177e4SLinus Torvalds /* 35111da177e4SLinus Torvalds ** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit 35121da177e4SLinus Torvalds ** 35131da177e4SLinus Torvalds ** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all 35141da177e4SLinus Torvalds ** the writers are done. By the time it wakes up, the transaction it was called has already ended, so it just 35151da177e4SLinus Torvalds ** flushes the commit list and returns 0. 35161da177e4SLinus Torvalds ** 35171da177e4SLinus Torvalds ** Won't batch when flush or commit_now is set. Also won't batch when others are waiting on j_join_wait. 35181da177e4SLinus Torvalds ** 35191da177e4SLinus Torvalds ** Note, we can't allow the journal_end to proceed while there are still writers in the log. 35201da177e4SLinus Torvalds */ 3521bd4c625cSLinus Torvalds static int check_journal_end(struct reiserfs_transaction_handle *th, 3522bd4c625cSLinus Torvalds struct super_block *p_s_sb, unsigned long nblocks, 3523bd4c625cSLinus Torvalds int flags) 3524bd4c625cSLinus Torvalds { 35251da177e4SLinus Torvalds 35261da177e4SLinus Torvalds time_t now; 35271da177e4SLinus Torvalds int flush = flags & FLUSH_ALL; 35281da177e4SLinus Torvalds int commit_now = flags & COMMIT_NOW; 35291da177e4SLinus Torvalds int wait_on_commit = flags & WAIT; 35301da177e4SLinus Torvalds struct reiserfs_journal_list *jl; 35311da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 35321da177e4SLinus Torvalds 35331da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 35341da177e4SLinus Torvalds 35351da177e4SLinus Torvalds if (th->t_trans_id != journal->j_trans_id) { 3536bd4c625cSLinus Torvalds reiserfs_panic(th->t_super, 3537bd4c625cSLinus Torvalds "journal-1577: handle trans id %ld != current trans id %ld\n", 35381da177e4SLinus Torvalds th->t_trans_id, journal->j_trans_id); 35391da177e4SLinus Torvalds } 35401da177e4SLinus Torvalds 35411da177e4SLinus Torvalds journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged); 35421da177e4SLinus Torvalds if (atomic_read(&(journal->j_wcount)) > 0) { /* <= 0 is allowed. unmounting might not call begin */ 35431da177e4SLinus Torvalds atomic_dec(&(journal->j_wcount)); 35441da177e4SLinus Torvalds } 35451da177e4SLinus Torvalds 35461da177e4SLinus Torvalds /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released 35471da177e4SLinus Torvalds ** will be dealt with by next transaction that actually writes something, but should be taken 35481da177e4SLinus Torvalds ** care of in this trans 35491da177e4SLinus Torvalds */ 355014a61442SEric Sesterhenn BUG_ON(journal->j_len == 0); 355114a61442SEric Sesterhenn 35521da177e4SLinus Torvalds /* if wcount > 0, and we are called to with flush or commit_now, 35531da177e4SLinus Torvalds ** we wait on j_join_wait. We will wake up when the last writer has 35541da177e4SLinus Torvalds ** finished the transaction, and started it on its way to the disk. 35551da177e4SLinus Torvalds ** Then, we flush the commit or journal list, and just return 0 35561da177e4SLinus Torvalds ** because the rest of journal end was already done for this transaction. 35571da177e4SLinus Torvalds */ 35581da177e4SLinus Torvalds if (atomic_read(&(journal->j_wcount)) > 0) { 35591da177e4SLinus Torvalds if (flush || commit_now) { 35601da177e4SLinus Torvalds unsigned trans_id; 35611da177e4SLinus Torvalds 35621da177e4SLinus Torvalds jl = journal->j_current_jl; 35631da177e4SLinus Torvalds trans_id = jl->j_trans_id; 35641da177e4SLinus Torvalds if (wait_on_commit) 35651da177e4SLinus Torvalds jl->j_state |= LIST_COMMIT_PENDING; 35661da177e4SLinus Torvalds atomic_set(&(journal->j_jlock), 1); 35671da177e4SLinus Torvalds if (flush) { 35681da177e4SLinus Torvalds journal->j_next_full_flush = 1; 35691da177e4SLinus Torvalds } 35701da177e4SLinus Torvalds unlock_journal(p_s_sb); 35711da177e4SLinus Torvalds 35721da177e4SLinus Torvalds /* sleep while the current transaction is still j_jlocked */ 35731da177e4SLinus Torvalds while (journal->j_trans_id == trans_id) { 35741da177e4SLinus Torvalds if (atomic_read(&journal->j_jlock)) { 35751da177e4SLinus Torvalds queue_log_writer(p_s_sb); 35761da177e4SLinus Torvalds } else { 35771da177e4SLinus Torvalds lock_journal(p_s_sb); 35781da177e4SLinus Torvalds if (journal->j_trans_id == trans_id) { 3579bd4c625cSLinus Torvalds atomic_set(&(journal->j_jlock), 3580bd4c625cSLinus Torvalds 1); 35811da177e4SLinus Torvalds } 35821da177e4SLinus Torvalds unlock_journal(p_s_sb); 35831da177e4SLinus Torvalds } 35841da177e4SLinus Torvalds } 358514a61442SEric Sesterhenn BUG_ON(journal->j_trans_id == trans_id); 358614a61442SEric Sesterhenn 3587bd4c625cSLinus Torvalds if (commit_now 3588bd4c625cSLinus Torvalds && journal_list_still_alive(p_s_sb, trans_id) 3589bd4c625cSLinus Torvalds && wait_on_commit) { 35901da177e4SLinus Torvalds flush_commit_list(p_s_sb, jl, 1); 35911da177e4SLinus Torvalds } 35921da177e4SLinus Torvalds return 0; 35931da177e4SLinus Torvalds } 35941da177e4SLinus Torvalds unlock_journal(p_s_sb); 35951da177e4SLinus Torvalds return 0; 35961da177e4SLinus Torvalds } 35971da177e4SLinus Torvalds 35981da177e4SLinus Torvalds /* deal with old transactions where we are the last writers */ 35991da177e4SLinus Torvalds now = get_seconds(); 36001da177e4SLinus Torvalds if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) { 36011da177e4SLinus Torvalds commit_now = 1; 36021da177e4SLinus Torvalds journal->j_next_async_flush = 1; 36031da177e4SLinus Torvalds } 36041da177e4SLinus Torvalds /* don't batch when someone is waiting on j_join_wait */ 36051da177e4SLinus Torvalds /* don't batch when syncing the commit or flushing the whole trans */ 3606bd4c625cSLinus Torvalds if (!(journal->j_must_wait > 0) && !(atomic_read(&(journal->j_jlock))) 3607bd4c625cSLinus Torvalds && !flush && !commit_now && (journal->j_len < journal->j_max_batch) 3608bd4c625cSLinus Torvalds && journal->j_len_alloc < journal->j_max_batch 3609bd4c625cSLinus Torvalds && journal->j_cnode_free > (journal->j_trans_max * 3)) { 36101da177e4SLinus Torvalds journal->j_bcount++; 36111da177e4SLinus Torvalds unlock_journal(p_s_sb); 36121da177e4SLinus Torvalds return 0; 36131da177e4SLinus Torvalds } 36141da177e4SLinus Torvalds 36151da177e4SLinus Torvalds if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(p_s_sb)) { 3616bd4c625cSLinus Torvalds reiserfs_panic(p_s_sb, 3617bd4c625cSLinus Torvalds "journal-003: journal_end: j_start (%ld) is too high\n", 3618bd4c625cSLinus Torvalds journal->j_start); 36191da177e4SLinus Torvalds } 36201da177e4SLinus Torvalds return 1; 36211da177e4SLinus Torvalds } 36221da177e4SLinus Torvalds 36231da177e4SLinus Torvalds /* 36241da177e4SLinus Torvalds ** Does all the work that makes deleting blocks safe. 36251da177e4SLinus Torvalds ** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on. 36261da177e4SLinus Torvalds ** 36271da177e4SLinus Torvalds ** otherwise: 36281da177e4SLinus Torvalds ** set a bit for the block in the journal bitmap. That will prevent it from being allocated for unformatted nodes 36291da177e4SLinus Torvalds ** before this transaction has finished. 36301da177e4SLinus Torvalds ** 36311da177e4SLinus Torvalds ** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers. That will prevent any old transactions with 36321da177e4SLinus Torvalds ** this block from trying to flush to the real location. Since we aren't removing the cnode from the journal_list_hash, 36331da177e4SLinus Torvalds ** the block can't be reallocated yet. 36341da177e4SLinus Torvalds ** 36351da177e4SLinus Torvalds ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list. 36361da177e4SLinus Torvalds */ 3637bd4c625cSLinus Torvalds int journal_mark_freed(struct reiserfs_transaction_handle *th, 3638bd4c625cSLinus Torvalds struct super_block *p_s_sb, b_blocknr_t blocknr) 3639bd4c625cSLinus Torvalds { 36401da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 36411da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn = NULL; 36421da177e4SLinus Torvalds struct buffer_head *bh = NULL; 36431da177e4SLinus Torvalds struct reiserfs_list_bitmap *jb = NULL; 36441da177e4SLinus Torvalds int cleaned = 0; 36451da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 36461da177e4SLinus Torvalds 36471da177e4SLinus Torvalds cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, blocknr); 36481da177e4SLinus Torvalds if (cn && cn->bh) { 36491da177e4SLinus Torvalds bh = cn->bh; 36501da177e4SLinus Torvalds get_bh(bh); 36511da177e4SLinus Torvalds } 36521da177e4SLinus Torvalds /* if it is journal new, we just remove it from this transaction */ 36531da177e4SLinus Torvalds if (bh && buffer_journal_new(bh)) { 36541da177e4SLinus Torvalds clear_buffer_journal_new(bh); 36551da177e4SLinus Torvalds clear_prepared_bits(bh); 36561da177e4SLinus Torvalds reiserfs_clean_and_file_buffer(bh); 36571da177e4SLinus Torvalds cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned); 36581da177e4SLinus Torvalds } else { 36591da177e4SLinus Torvalds /* set the bit for this block in the journal bitmap for this transaction */ 36601da177e4SLinus Torvalds jb = journal->j_current_jl->j_list_bitmap; 36611da177e4SLinus Torvalds if (!jb) { 3662bd4c625cSLinus Torvalds reiserfs_panic(p_s_sb, 3663bd4c625cSLinus Torvalds "journal-1702: journal_mark_freed, journal_list_bitmap is NULL\n"); 36641da177e4SLinus Torvalds } 36651da177e4SLinus Torvalds set_bit_in_list_bitmap(p_s_sb, blocknr, jb); 36661da177e4SLinus Torvalds 36671da177e4SLinus Torvalds /* Note, the entire while loop is not allowed to schedule. */ 36681da177e4SLinus Torvalds 36691da177e4SLinus Torvalds if (bh) { 36701da177e4SLinus Torvalds clear_prepared_bits(bh); 36711da177e4SLinus Torvalds reiserfs_clean_and_file_buffer(bh); 36721da177e4SLinus Torvalds } 36731da177e4SLinus Torvalds cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned); 36741da177e4SLinus Torvalds 36751da177e4SLinus Torvalds /* find all older transactions with this block, make sure they don't try to write it out */ 3676bd4c625cSLinus Torvalds cn = get_journal_hash_dev(p_s_sb, journal->j_list_hash_table, 3677bd4c625cSLinus Torvalds blocknr); 36781da177e4SLinus Torvalds while (cn) { 36791da177e4SLinus Torvalds if (p_s_sb == cn->sb && blocknr == cn->blocknr) { 36801da177e4SLinus Torvalds set_bit(BLOCK_FREED, &cn->state); 36811da177e4SLinus Torvalds if (cn->bh) { 36821da177e4SLinus Torvalds if (!cleaned) { 36831da177e4SLinus Torvalds /* remove_from_transaction will brelse the buffer if it was 36841da177e4SLinus Torvalds ** in the current trans 36851da177e4SLinus Torvalds */ 3686bd4c625cSLinus Torvalds clear_buffer_journal_dirty(cn-> 3687bd4c625cSLinus Torvalds bh); 36881da177e4SLinus Torvalds clear_buffer_dirty(cn->bh); 3689bd4c625cSLinus Torvalds clear_buffer_journal_test(cn-> 3690bd4c625cSLinus Torvalds bh); 36911da177e4SLinus Torvalds cleaned = 1; 36921da177e4SLinus Torvalds put_bh(cn->bh); 3693bd4c625cSLinus Torvalds if (atomic_read 3694bd4c625cSLinus Torvalds (&(cn->bh->b_count)) < 0) { 3695bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 3696bd4c625cSLinus Torvalds "journal-2138: cn->bh->b_count < 0"); 36971da177e4SLinus Torvalds } 36981da177e4SLinus Torvalds } 36991da177e4SLinus Torvalds if (cn->jlist) { /* since we are clearing the bh, we MUST dec nonzerolen */ 3700bd4c625cSLinus Torvalds atomic_dec(& 3701bd4c625cSLinus Torvalds (cn->jlist-> 3702bd4c625cSLinus Torvalds j_nonzerolen)); 37031da177e4SLinus Torvalds } 37041da177e4SLinus Torvalds cn->bh = NULL; 37051da177e4SLinus Torvalds } 37061da177e4SLinus Torvalds } 37071da177e4SLinus Torvalds cn = cn->hnext; 37081da177e4SLinus Torvalds } 37091da177e4SLinus Torvalds } 37101da177e4SLinus Torvalds 37111da177e4SLinus Torvalds if (bh) { 37121da177e4SLinus Torvalds put_bh(bh); /* get_hash grabs the buffer */ 37131da177e4SLinus Torvalds if (atomic_read(&(bh->b_count)) < 0) { 3714bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 3715bd4c625cSLinus Torvalds "journal-2165: bh->b_count < 0"); 37161da177e4SLinus Torvalds } 37171da177e4SLinus Torvalds } 37181da177e4SLinus Torvalds return 0; 37191da177e4SLinus Torvalds } 37201da177e4SLinus Torvalds 3721bd4c625cSLinus Torvalds void reiserfs_update_inode_transaction(struct inode *inode) 3722bd4c625cSLinus Torvalds { 37231da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb); 37241da177e4SLinus Torvalds REISERFS_I(inode)->i_jl = journal->j_current_jl; 37251da177e4SLinus Torvalds REISERFS_I(inode)->i_trans_id = journal->j_trans_id; 37261da177e4SLinus Torvalds } 37271da177e4SLinus Torvalds 37281da177e4SLinus Torvalds /* 37291da177e4SLinus Torvalds * returns -1 on error, 0 if no commits/barriers were done and 1 37301da177e4SLinus Torvalds * if a transaction was actually committed and the barrier was done 37311da177e4SLinus Torvalds */ 37321da177e4SLinus Torvalds static int __commit_trans_jl(struct inode *inode, unsigned long id, 37331da177e4SLinus Torvalds struct reiserfs_journal_list *jl) 37341da177e4SLinus Torvalds { 37351da177e4SLinus Torvalds struct reiserfs_transaction_handle th; 37361da177e4SLinus Torvalds struct super_block *sb = inode->i_sb; 37371da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(sb); 37381da177e4SLinus Torvalds int ret = 0; 37391da177e4SLinus Torvalds 37401da177e4SLinus Torvalds /* is it from the current transaction, or from an unknown transaction? */ 37411da177e4SLinus Torvalds if (id == journal->j_trans_id) { 37421da177e4SLinus Torvalds jl = journal->j_current_jl; 37431da177e4SLinus Torvalds /* try to let other writers come in and grow this transaction */ 37441da177e4SLinus Torvalds let_transaction_grow(sb, id); 37451da177e4SLinus Torvalds if (journal->j_trans_id != id) { 37461da177e4SLinus Torvalds goto flush_commit_only; 37471da177e4SLinus Torvalds } 37481da177e4SLinus Torvalds 37491da177e4SLinus Torvalds ret = journal_begin(&th, sb, 1); 37501da177e4SLinus Torvalds if (ret) 37511da177e4SLinus Torvalds return ret; 37521da177e4SLinus Torvalds 37531da177e4SLinus Torvalds /* someone might have ended this transaction while we joined */ 37541da177e4SLinus Torvalds if (journal->j_trans_id != id) { 3755bd4c625cSLinus Torvalds reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb), 3756bd4c625cSLinus Torvalds 1); 37571da177e4SLinus Torvalds journal_mark_dirty(&th, sb, SB_BUFFER_WITH_SB(sb)); 37581da177e4SLinus Torvalds ret = journal_end(&th, sb, 1); 37591da177e4SLinus Torvalds goto flush_commit_only; 37601da177e4SLinus Torvalds } 37611da177e4SLinus Torvalds 37621da177e4SLinus Torvalds ret = journal_end_sync(&th, sb, 1); 37631da177e4SLinus Torvalds if (!ret) 37641da177e4SLinus Torvalds ret = 1; 37651da177e4SLinus Torvalds 37661da177e4SLinus Torvalds } else { 37671da177e4SLinus Torvalds /* this gets tricky, we have to make sure the journal list in 37681da177e4SLinus Torvalds * the inode still exists. We know the list is still around 37691da177e4SLinus Torvalds * if we've got a larger transaction id than the oldest list 37701da177e4SLinus Torvalds */ 37711da177e4SLinus Torvalds flush_commit_only: 37721da177e4SLinus Torvalds if (journal_list_still_alive(inode->i_sb, id)) { 37731da177e4SLinus Torvalds /* 37741da177e4SLinus Torvalds * we only set ret to 1 when we know for sure 37751da177e4SLinus Torvalds * the barrier hasn't been started yet on the commit 37761da177e4SLinus Torvalds * block. 37771da177e4SLinus Torvalds */ 37781da177e4SLinus Torvalds if (atomic_read(&jl->j_commit_left) > 1) 37791da177e4SLinus Torvalds ret = 1; 37801da177e4SLinus Torvalds flush_commit_list(sb, jl, 1); 37811da177e4SLinus Torvalds if (journal->j_errno) 37821da177e4SLinus Torvalds ret = journal->j_errno; 37831da177e4SLinus Torvalds } 37841da177e4SLinus Torvalds } 37851da177e4SLinus Torvalds /* otherwise the list is gone, and long since committed */ 37861da177e4SLinus Torvalds return ret; 37871da177e4SLinus Torvalds } 37881da177e4SLinus Torvalds 3789bd4c625cSLinus Torvalds int reiserfs_commit_for_inode(struct inode *inode) 3790bd4c625cSLinus Torvalds { 37911da177e4SLinus Torvalds unsigned long id = REISERFS_I(inode)->i_trans_id; 37921da177e4SLinus Torvalds struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl; 37931da177e4SLinus Torvalds 37941da177e4SLinus Torvalds /* for the whole inode, assume unset id means it was 37951da177e4SLinus Torvalds * changed in the current transaction. More conservative 37961da177e4SLinus Torvalds */ 37971da177e4SLinus Torvalds if (!id || !jl) { 37981da177e4SLinus Torvalds reiserfs_update_inode_transaction(inode); 37991da177e4SLinus Torvalds id = REISERFS_I(inode)->i_trans_id; 38001da177e4SLinus Torvalds /* jl will be updated in __commit_trans_jl */ 38011da177e4SLinus Torvalds } 38021da177e4SLinus Torvalds 38031da177e4SLinus Torvalds return __commit_trans_jl(inode, id, jl); 38041da177e4SLinus Torvalds } 38051da177e4SLinus Torvalds 38061da177e4SLinus Torvalds void reiserfs_restore_prepared_buffer(struct super_block *p_s_sb, 3807bd4c625cSLinus Torvalds struct buffer_head *bh) 3808bd4c625cSLinus Torvalds { 38091da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 38101da177e4SLinus Torvalds PROC_INFO_INC(p_s_sb, journal.restore_prepared); 38111da177e4SLinus Torvalds if (!bh) { 38121da177e4SLinus Torvalds return; 38131da177e4SLinus Torvalds } 38141da177e4SLinus Torvalds if (test_clear_buffer_journal_restore_dirty(bh) && 38151da177e4SLinus Torvalds buffer_journal_dirty(bh)) { 38161da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 38171da177e4SLinus Torvalds cn = get_journal_hash_dev(p_s_sb, 38181da177e4SLinus Torvalds journal->j_list_hash_table, 38191da177e4SLinus Torvalds bh->b_blocknr); 38201da177e4SLinus Torvalds if (cn && can_dirty(cn)) { 38211da177e4SLinus Torvalds set_buffer_journal_test(bh); 38221da177e4SLinus Torvalds mark_buffer_dirty(bh); 38231da177e4SLinus Torvalds } 38241da177e4SLinus Torvalds } 38251da177e4SLinus Torvalds clear_buffer_journal_prepared(bh); 38261da177e4SLinus Torvalds } 38271da177e4SLinus Torvalds 38281da177e4SLinus Torvalds extern struct tree_balance *cur_tb; 38291da177e4SLinus Torvalds /* 38301da177e4SLinus Torvalds ** before we can change a metadata block, we have to make sure it won't 38311da177e4SLinus Torvalds ** be written to disk while we are altering it. So, we must: 38321da177e4SLinus Torvalds ** clean it 38331da177e4SLinus Torvalds ** wait on it. 38341da177e4SLinus Torvalds ** 38351da177e4SLinus Torvalds */ 38361da177e4SLinus Torvalds int reiserfs_prepare_for_journal(struct super_block *p_s_sb, 3837bd4c625cSLinus Torvalds struct buffer_head *bh, int wait) 3838bd4c625cSLinus Torvalds { 38391da177e4SLinus Torvalds PROC_INFO_INC(p_s_sb, journal.prepare); 38401da177e4SLinus Torvalds 38411da177e4SLinus Torvalds if (test_set_buffer_locked(bh)) { 38421da177e4SLinus Torvalds if (!wait) 38431da177e4SLinus Torvalds return 0; 38441da177e4SLinus Torvalds lock_buffer(bh); 38451da177e4SLinus Torvalds } 38461da177e4SLinus Torvalds set_buffer_journal_prepared(bh); 38471da177e4SLinus Torvalds if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) { 38481da177e4SLinus Torvalds clear_buffer_journal_test(bh); 38491da177e4SLinus Torvalds set_buffer_journal_restore_dirty(bh); 38501da177e4SLinus Torvalds } 38511da177e4SLinus Torvalds unlock_buffer(bh); 38521da177e4SLinus Torvalds return 1; 38531da177e4SLinus Torvalds } 38541da177e4SLinus Torvalds 3855bd4c625cSLinus Torvalds static void flush_old_journal_lists(struct super_block *s) 3856bd4c625cSLinus Torvalds { 38571da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 38581da177e4SLinus Torvalds struct reiserfs_journal_list *jl; 38591da177e4SLinus Torvalds struct list_head *entry; 38601da177e4SLinus Torvalds time_t now = get_seconds(); 38611da177e4SLinus Torvalds 38621da177e4SLinus Torvalds while (!list_empty(&journal->j_journal_list)) { 38631da177e4SLinus Torvalds entry = journal->j_journal_list.next; 38641da177e4SLinus Torvalds jl = JOURNAL_LIST_ENTRY(entry); 38651da177e4SLinus Torvalds /* this check should always be run, to send old lists to disk */ 3866a3172027SChris Mason if (jl->j_timestamp < (now - (JOURNAL_MAX_TRANS_AGE * 4)) && 3867a3172027SChris Mason atomic_read(&jl->j_commit_left) == 0 && 3868a3172027SChris Mason test_transaction(s, jl)) { 38691da177e4SLinus Torvalds flush_used_journal_lists(s, jl); 38701da177e4SLinus Torvalds } else { 38711da177e4SLinus Torvalds break; 38721da177e4SLinus Torvalds } 38731da177e4SLinus Torvalds } 38741da177e4SLinus Torvalds } 38751da177e4SLinus Torvalds 38761da177e4SLinus Torvalds /* 38771da177e4SLinus Torvalds ** long and ugly. If flush, will not return until all commit 38781da177e4SLinus Torvalds ** blocks and all real buffers in the trans are on disk. 38791da177e4SLinus Torvalds ** If no_async, won't return until all commit blocks are on disk. 38801da177e4SLinus Torvalds ** 38811da177e4SLinus Torvalds ** keep reading, there are comments as you go along 38821da177e4SLinus Torvalds ** 38831da177e4SLinus Torvalds ** If the journal is aborted, we just clean up. Things like flushing 38841da177e4SLinus Torvalds ** journal lists, etc just won't happen. 38851da177e4SLinus Torvalds */ 3886bd4c625cSLinus Torvalds static int do_journal_end(struct reiserfs_transaction_handle *th, 3887bd4c625cSLinus Torvalds struct super_block *p_s_sb, unsigned long nblocks, 3888bd4c625cSLinus Torvalds int flags) 3889bd4c625cSLinus Torvalds { 38901da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 38911da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn, *next, *jl_cn; 38921da177e4SLinus Torvalds struct reiserfs_journal_cnode *last_cn = NULL; 38931da177e4SLinus Torvalds struct reiserfs_journal_desc *desc; 38941da177e4SLinus Torvalds struct reiserfs_journal_commit *commit; 38951da177e4SLinus Torvalds struct buffer_head *c_bh; /* commit bh */ 38961da177e4SLinus Torvalds struct buffer_head *d_bh; /* desc bh */ 38971da177e4SLinus Torvalds int cur_write_start = 0; /* start index of current log write */ 38981da177e4SLinus Torvalds int old_start; 38991da177e4SLinus Torvalds int i; 3900a44c94a7SAlexander Zarochentsev int flush; 3901a44c94a7SAlexander Zarochentsev int wait_on_commit; 39021da177e4SLinus Torvalds struct reiserfs_journal_list *jl, *temp_jl; 39031da177e4SLinus Torvalds struct list_head *entry, *safe; 39041da177e4SLinus Torvalds unsigned long jindex; 39051da177e4SLinus Torvalds unsigned long commit_trans_id; 39061da177e4SLinus Torvalds int trans_half; 39071da177e4SLinus Torvalds 39081da177e4SLinus Torvalds BUG_ON(th->t_refcount > 1); 39091da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 39101da177e4SLinus Torvalds 3911a44c94a7SAlexander Zarochentsev /* protect flush_older_commits from doing mistakes if the 3912a44c94a7SAlexander Zarochentsev transaction ID counter gets overflowed. */ 3913a44c94a7SAlexander Zarochentsev if (th->t_trans_id == ~0UL) 3914a44c94a7SAlexander Zarochentsev flags |= FLUSH_ALL | COMMIT_NOW | WAIT; 3915a44c94a7SAlexander Zarochentsev flush = flags & FLUSH_ALL; 3916a44c94a7SAlexander Zarochentsev wait_on_commit = flags & WAIT; 3917a44c94a7SAlexander Zarochentsev 391822e2c507SJens Axboe put_fs_excl(); 39191da177e4SLinus Torvalds current->journal_info = th->t_handle_save; 39201da177e4SLinus Torvalds reiserfs_check_lock_depth(p_s_sb, "journal end"); 39211da177e4SLinus Torvalds if (journal->j_len == 0) { 3922bd4c625cSLinus Torvalds reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb), 3923bd4c625cSLinus Torvalds 1); 39241da177e4SLinus Torvalds journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb)); 39251da177e4SLinus Torvalds } 39261da177e4SLinus Torvalds 39271da177e4SLinus Torvalds lock_journal(p_s_sb); 39281da177e4SLinus Torvalds if (journal->j_next_full_flush) { 39291da177e4SLinus Torvalds flags |= FLUSH_ALL; 39301da177e4SLinus Torvalds flush = 1; 39311da177e4SLinus Torvalds } 39321da177e4SLinus Torvalds if (journal->j_next_async_flush) { 39331da177e4SLinus Torvalds flags |= COMMIT_NOW | WAIT; 39341da177e4SLinus Torvalds wait_on_commit = 1; 39351da177e4SLinus Torvalds } 39361da177e4SLinus Torvalds 39371da177e4SLinus Torvalds /* check_journal_end locks the journal, and unlocks if it does not return 1 39381da177e4SLinus Torvalds ** it tells us if we should continue with the journal_end, or just return 39391da177e4SLinus Torvalds */ 39401da177e4SLinus Torvalds if (!check_journal_end(th, p_s_sb, nblocks, flags)) { 39411da177e4SLinus Torvalds p_s_sb->s_dirt = 1; 39421da177e4SLinus Torvalds wake_queued_writers(p_s_sb); 39431da177e4SLinus Torvalds reiserfs_async_progress_wait(p_s_sb); 39441da177e4SLinus Torvalds goto out; 39451da177e4SLinus Torvalds } 39461da177e4SLinus Torvalds 39471da177e4SLinus Torvalds /* check_journal_end might set these, check again */ 39481da177e4SLinus Torvalds if (journal->j_next_full_flush) { 39491da177e4SLinus Torvalds flush = 1; 39501da177e4SLinus Torvalds } 39511da177e4SLinus Torvalds 39521da177e4SLinus Torvalds /* 39531da177e4SLinus Torvalds ** j must wait means we have to flush the log blocks, and the real blocks for 39541da177e4SLinus Torvalds ** this transaction 39551da177e4SLinus Torvalds */ 39561da177e4SLinus Torvalds if (journal->j_must_wait > 0) { 39571da177e4SLinus Torvalds flush = 1; 39581da177e4SLinus Torvalds } 39591da177e4SLinus Torvalds #ifdef REISERFS_PREALLOCATE 3960ef43bc4fSJan Kara /* quota ops might need to nest, setup the journal_info pointer for them 3961ef43bc4fSJan Kara * and raise the refcount so that it is > 0. */ 39621da177e4SLinus Torvalds current->journal_info = th; 3963ef43bc4fSJan Kara th->t_refcount++; 39641da177e4SLinus Torvalds reiserfs_discard_all_prealloc(th); /* it should not involve new blocks into 39651da177e4SLinus Torvalds * the transaction */ 3966ef43bc4fSJan Kara th->t_refcount--; 39671da177e4SLinus Torvalds current->journal_info = th->t_handle_save; 39681da177e4SLinus Torvalds #endif 39691da177e4SLinus Torvalds 39701da177e4SLinus Torvalds /* setup description block */ 3971bd4c625cSLinus Torvalds d_bh = 3972bd4c625cSLinus Torvalds journal_getblk(p_s_sb, 3973bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + 3974bd4c625cSLinus Torvalds journal->j_start); 39751da177e4SLinus Torvalds set_buffer_uptodate(d_bh); 39761da177e4SLinus Torvalds desc = (struct reiserfs_journal_desc *)(d_bh)->b_data; 39771da177e4SLinus Torvalds memset(d_bh->b_data, 0, d_bh->b_size); 39781da177e4SLinus Torvalds memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8); 39791da177e4SLinus Torvalds set_desc_trans_id(desc, journal->j_trans_id); 39801da177e4SLinus Torvalds 39811da177e4SLinus Torvalds /* setup commit block. Don't write (keep it clean too) this one until after everyone else is written */ 39821da177e4SLinus Torvalds c_bh = journal_getblk(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + 3983bd4c625cSLinus Torvalds ((journal->j_start + journal->j_len + 3984bd4c625cSLinus Torvalds 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb))); 39851da177e4SLinus Torvalds commit = (struct reiserfs_journal_commit *)c_bh->b_data; 39861da177e4SLinus Torvalds memset(c_bh->b_data, 0, c_bh->b_size); 39871da177e4SLinus Torvalds set_commit_trans_id(commit, journal->j_trans_id); 39881da177e4SLinus Torvalds set_buffer_uptodate(c_bh); 39891da177e4SLinus Torvalds 39901da177e4SLinus Torvalds /* init this journal list */ 39911da177e4SLinus Torvalds jl = journal->j_current_jl; 39921da177e4SLinus Torvalds 39931da177e4SLinus Torvalds /* we lock the commit before doing anything because 39941da177e4SLinus Torvalds * we want to make sure nobody tries to run flush_commit_list until 39951da177e4SLinus Torvalds * the new transaction is fully setup, and we've already flushed the 39961da177e4SLinus Torvalds * ordered bh list 39971da177e4SLinus Torvalds */ 39981da177e4SLinus Torvalds down(&jl->j_commit_lock); 39991da177e4SLinus Torvalds 40001da177e4SLinus Torvalds /* save the transaction id in case we need to commit it later */ 40011da177e4SLinus Torvalds commit_trans_id = jl->j_trans_id; 40021da177e4SLinus Torvalds 40031da177e4SLinus Torvalds atomic_set(&jl->j_older_commits_done, 0); 40041da177e4SLinus Torvalds jl->j_trans_id = journal->j_trans_id; 40051da177e4SLinus Torvalds jl->j_timestamp = journal->j_trans_start_time; 40061da177e4SLinus Torvalds jl->j_commit_bh = c_bh; 40071da177e4SLinus Torvalds jl->j_start = journal->j_start; 40081da177e4SLinus Torvalds jl->j_len = journal->j_len; 40091da177e4SLinus Torvalds atomic_set(&jl->j_nonzerolen, journal->j_len); 40101da177e4SLinus Torvalds atomic_set(&jl->j_commit_left, journal->j_len + 2); 40111da177e4SLinus Torvalds jl->j_realblock = NULL; 40121da177e4SLinus Torvalds 40131da177e4SLinus Torvalds /* The ENTIRE FOR LOOP MUST not cause schedule to occur. 40141da177e4SLinus Torvalds ** for each real block, add it to the journal list hash, 40151da177e4SLinus Torvalds ** copy into real block index array in the commit or desc block 40161da177e4SLinus Torvalds */ 40171da177e4SLinus Torvalds trans_half = journal_trans_half(p_s_sb->s_blocksize); 40181da177e4SLinus Torvalds for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) { 40191da177e4SLinus Torvalds if (buffer_journaled(cn->bh)) { 40201da177e4SLinus Torvalds jl_cn = get_cnode(p_s_sb); 40211da177e4SLinus Torvalds if (!jl_cn) { 4022bd4c625cSLinus Torvalds reiserfs_panic(p_s_sb, 4023bd4c625cSLinus Torvalds "journal-1676, get_cnode returned NULL\n"); 40241da177e4SLinus Torvalds } 40251da177e4SLinus Torvalds if (i == 0) { 40261da177e4SLinus Torvalds jl->j_realblock = jl_cn; 40271da177e4SLinus Torvalds } 40281da177e4SLinus Torvalds jl_cn->prev = last_cn; 40291da177e4SLinus Torvalds jl_cn->next = NULL; 40301da177e4SLinus Torvalds if (last_cn) { 40311da177e4SLinus Torvalds last_cn->next = jl_cn; 40321da177e4SLinus Torvalds } 40331da177e4SLinus Torvalds last_cn = jl_cn; 40341da177e4SLinus Torvalds /* make sure the block we are trying to log is not a block 40351da177e4SLinus Torvalds of journal or reserved area */ 40361da177e4SLinus Torvalds 4037bd4c625cSLinus Torvalds if (is_block_in_log_or_reserved_area 4038bd4c625cSLinus Torvalds (p_s_sb, cn->bh->b_blocknr)) { 4039bd4c625cSLinus Torvalds reiserfs_panic(p_s_sb, 4040bd4c625cSLinus Torvalds "journal-2332: Trying to log block %lu, which is a log block\n", 4041bd4c625cSLinus Torvalds cn->bh->b_blocknr); 40421da177e4SLinus Torvalds } 40431da177e4SLinus Torvalds jl_cn->blocknr = cn->bh->b_blocknr; 40441da177e4SLinus Torvalds jl_cn->state = 0; 40451da177e4SLinus Torvalds jl_cn->sb = p_s_sb; 40461da177e4SLinus Torvalds jl_cn->bh = cn->bh; 40471da177e4SLinus Torvalds jl_cn->jlist = jl; 40481da177e4SLinus Torvalds insert_journal_hash(journal->j_list_hash_table, jl_cn); 40491da177e4SLinus Torvalds if (i < trans_half) { 4050bd4c625cSLinus Torvalds desc->j_realblock[i] = 4051bd4c625cSLinus Torvalds cpu_to_le32(cn->bh->b_blocknr); 40521da177e4SLinus Torvalds } else { 4053bd4c625cSLinus Torvalds commit->j_realblock[i - trans_half] = 4054bd4c625cSLinus Torvalds cpu_to_le32(cn->bh->b_blocknr); 40551da177e4SLinus Torvalds } 40561da177e4SLinus Torvalds } else { 40571da177e4SLinus Torvalds i--; 40581da177e4SLinus Torvalds } 40591da177e4SLinus Torvalds } 40601da177e4SLinus Torvalds set_desc_trans_len(desc, journal->j_len); 40611da177e4SLinus Torvalds set_desc_mount_id(desc, journal->j_mount_id); 40621da177e4SLinus Torvalds set_desc_trans_id(desc, journal->j_trans_id); 40631da177e4SLinus Torvalds set_commit_trans_len(commit, journal->j_len); 40641da177e4SLinus Torvalds 40651da177e4SLinus Torvalds /* special check in case all buffers in the journal were marked for not logging */ 406614a61442SEric Sesterhenn BUG_ON(journal->j_len == 0); 40671da177e4SLinus Torvalds 40681da177e4SLinus Torvalds /* we're about to dirty all the log blocks, mark the description block 40691da177e4SLinus Torvalds * dirty now too. Don't mark the commit block dirty until all the 40701da177e4SLinus Torvalds * others are on disk 40711da177e4SLinus Torvalds */ 40721da177e4SLinus Torvalds mark_buffer_dirty(d_bh); 40731da177e4SLinus Torvalds 40741da177e4SLinus Torvalds /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */ 40751da177e4SLinus Torvalds cur_write_start = journal->j_start; 40761da177e4SLinus Torvalds cn = journal->j_first; 40771da177e4SLinus Torvalds jindex = 1; /* start at one so we don't get the desc again */ 40781da177e4SLinus Torvalds while (cn) { 40791da177e4SLinus Torvalds clear_buffer_journal_new(cn->bh); 40801da177e4SLinus Torvalds /* copy all the real blocks into log area. dirty log blocks */ 40811da177e4SLinus Torvalds if (buffer_journaled(cn->bh)) { 40821da177e4SLinus Torvalds struct buffer_head *tmp_bh; 40831da177e4SLinus Torvalds char *addr; 40841da177e4SLinus Torvalds struct page *page; 4085bd4c625cSLinus Torvalds tmp_bh = 4086bd4c625cSLinus Torvalds journal_getblk(p_s_sb, 4087bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + 4088bd4c625cSLinus Torvalds ((cur_write_start + 4089bd4c625cSLinus Torvalds jindex) % 4090bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_SIZE(p_s_sb))); 40911da177e4SLinus Torvalds set_buffer_uptodate(tmp_bh); 40921da177e4SLinus Torvalds page = cn->bh->b_page; 40931da177e4SLinus Torvalds addr = kmap(page); 4094bd4c625cSLinus Torvalds memcpy(tmp_bh->b_data, 4095bd4c625cSLinus Torvalds addr + offset_in_page(cn->bh->b_data), 40961da177e4SLinus Torvalds cn->bh->b_size); 40971da177e4SLinus Torvalds kunmap(page); 40981da177e4SLinus Torvalds mark_buffer_dirty(tmp_bh); 40991da177e4SLinus Torvalds jindex++; 41001da177e4SLinus Torvalds set_buffer_journal_dirty(cn->bh); 41011da177e4SLinus Torvalds clear_buffer_journaled(cn->bh); 41021da177e4SLinus Torvalds } else { 41031da177e4SLinus Torvalds /* JDirty cleared sometime during transaction. don't log this one */ 4104bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 4105bd4c625cSLinus Torvalds "journal-2048: do_journal_end: BAD, buffer in journal hash, but not JDirty!"); 41061da177e4SLinus Torvalds brelse(cn->bh); 41071da177e4SLinus Torvalds } 41081da177e4SLinus Torvalds next = cn->next; 41091da177e4SLinus Torvalds free_cnode(p_s_sb, cn); 41101da177e4SLinus Torvalds cn = next; 41111da177e4SLinus Torvalds cond_resched(); 41121da177e4SLinus Torvalds } 41131da177e4SLinus Torvalds 41141da177e4SLinus Torvalds /* we are done with both the c_bh and d_bh, but 41151da177e4SLinus Torvalds ** c_bh must be written after all other commit blocks, 41161da177e4SLinus Torvalds ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1. 41171da177e4SLinus Torvalds */ 41181da177e4SLinus Torvalds 41191da177e4SLinus Torvalds journal->j_current_jl = alloc_journal_list(p_s_sb); 41201da177e4SLinus Torvalds 41211da177e4SLinus Torvalds /* now it is safe to insert this transaction on the main list */ 41221da177e4SLinus Torvalds list_add_tail(&jl->j_list, &journal->j_journal_list); 41231da177e4SLinus Torvalds list_add_tail(&jl->j_working_list, &journal->j_working_list); 41241da177e4SLinus Torvalds journal->j_num_work_lists++; 41251da177e4SLinus Torvalds 41261da177e4SLinus Torvalds /* reset journal values for the next transaction */ 41271da177e4SLinus Torvalds old_start = journal->j_start; 4128bd4c625cSLinus Torvalds journal->j_start = 4129bd4c625cSLinus Torvalds (journal->j_start + journal->j_len + 4130bd4c625cSLinus Torvalds 2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb); 41311da177e4SLinus Torvalds atomic_set(&(journal->j_wcount), 0); 41321da177e4SLinus Torvalds journal->j_bcount = 0; 41331da177e4SLinus Torvalds journal->j_last = NULL; 41341da177e4SLinus Torvalds journal->j_first = NULL; 41351da177e4SLinus Torvalds journal->j_len = 0; 41361da177e4SLinus Torvalds journal->j_trans_start_time = 0; 4137a44c94a7SAlexander Zarochentsev /* check for trans_id overflow */ 4138a44c94a7SAlexander Zarochentsev if (++journal->j_trans_id == 0) 4139a44c94a7SAlexander Zarochentsev journal->j_trans_id = 10; 41401da177e4SLinus Torvalds journal->j_current_jl->j_trans_id = journal->j_trans_id; 41411da177e4SLinus Torvalds journal->j_must_wait = 0; 41421da177e4SLinus Torvalds journal->j_len_alloc = 0; 41431da177e4SLinus Torvalds journal->j_next_full_flush = 0; 41441da177e4SLinus Torvalds journal->j_next_async_flush = 0; 41451da177e4SLinus Torvalds init_journal_hash(p_s_sb); 41461da177e4SLinus Torvalds 41471da177e4SLinus Torvalds // make sure reiserfs_add_jh sees the new current_jl before we 41481da177e4SLinus Torvalds // write out the tails 41491da177e4SLinus Torvalds smp_mb(); 41501da177e4SLinus Torvalds 41511da177e4SLinus Torvalds /* tail conversion targets have to hit the disk before we end the 41521da177e4SLinus Torvalds * transaction. Otherwise a later transaction might repack the tail 41531da177e4SLinus Torvalds * before this transaction commits, leaving the data block unflushed and 41541da177e4SLinus Torvalds * clean, if we crash before the later transaction commits, the data block 41551da177e4SLinus Torvalds * is lost. 41561da177e4SLinus Torvalds */ 41571da177e4SLinus Torvalds if (!list_empty(&jl->j_tail_bh_list)) { 41581da177e4SLinus Torvalds unlock_kernel(); 41591da177e4SLinus Torvalds write_ordered_buffers(&journal->j_dirty_buffers_lock, 41601da177e4SLinus Torvalds journal, jl, &jl->j_tail_bh_list); 41611da177e4SLinus Torvalds lock_kernel(); 41621da177e4SLinus Torvalds } 416314a61442SEric Sesterhenn BUG_ON(!list_empty(&jl->j_tail_bh_list)); 41641da177e4SLinus Torvalds up(&jl->j_commit_lock); 41651da177e4SLinus Torvalds 41661da177e4SLinus Torvalds /* honor the flush wishes from the caller, simple commits can 41671da177e4SLinus Torvalds ** be done outside the journal lock, they are done below 41681da177e4SLinus Torvalds ** 41691da177e4SLinus Torvalds ** if we don't flush the commit list right now, we put it into 41701da177e4SLinus Torvalds ** the work queue so the people waiting on the async progress work 41711da177e4SLinus Torvalds ** queue don't wait for this proc to flush journal lists and such. 41721da177e4SLinus Torvalds */ 41731da177e4SLinus Torvalds if (flush) { 41741da177e4SLinus Torvalds flush_commit_list(p_s_sb, jl, 1); 41751da177e4SLinus Torvalds flush_journal_list(p_s_sb, jl, 1); 41761da177e4SLinus Torvalds } else if (!(jl->j_state & LIST_COMMIT_PENDING)) 41771da177e4SLinus Torvalds queue_delayed_work(commit_wq, &journal->j_work, HZ / 10); 41781da177e4SLinus Torvalds 41791da177e4SLinus Torvalds /* if the next transaction has any chance of wrapping, flush 41801da177e4SLinus Torvalds ** transactions that might get overwritten. If any journal lists are very 41811da177e4SLinus Torvalds ** old flush them as well. 41821da177e4SLinus Torvalds */ 41831da177e4SLinus Torvalds first_jl: 41841da177e4SLinus Torvalds list_for_each_safe(entry, safe, &journal->j_journal_list) { 41851da177e4SLinus Torvalds temp_jl = JOURNAL_LIST_ENTRY(entry); 41861da177e4SLinus Torvalds if (journal->j_start <= temp_jl->j_start) { 41871da177e4SLinus Torvalds if ((journal->j_start + journal->j_trans_max + 1) >= 4188bd4c625cSLinus Torvalds temp_jl->j_start) { 41891da177e4SLinus Torvalds flush_used_journal_lists(p_s_sb, temp_jl); 41901da177e4SLinus Torvalds goto first_jl; 41911da177e4SLinus Torvalds } else if ((journal->j_start + 41921da177e4SLinus Torvalds journal->j_trans_max + 1) < 4193bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_SIZE(p_s_sb)) { 41941da177e4SLinus Torvalds /* if we don't cross into the next transaction and we don't 41951da177e4SLinus Torvalds * wrap, there is no way we can overlap any later transactions 41961da177e4SLinus Torvalds * break now 41971da177e4SLinus Torvalds */ 41981da177e4SLinus Torvalds break; 41991da177e4SLinus Torvalds } 42001da177e4SLinus Torvalds } else if ((journal->j_start + 42011da177e4SLinus Torvalds journal->j_trans_max + 1) > 4202bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_SIZE(p_s_sb)) { 42031da177e4SLinus Torvalds if (((journal->j_start + journal->j_trans_max + 1) % 4204bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_SIZE(p_s_sb)) >= 4205bd4c625cSLinus Torvalds temp_jl->j_start) { 42061da177e4SLinus Torvalds flush_used_journal_lists(p_s_sb, temp_jl); 42071da177e4SLinus Torvalds goto first_jl; 42081da177e4SLinus Torvalds } else { 42091da177e4SLinus Torvalds /* we don't overlap anything from out start to the end of the 42101da177e4SLinus Torvalds * log, and our wrapped portion doesn't overlap anything at 42111da177e4SLinus Torvalds * the start of the log. We can break 42121da177e4SLinus Torvalds */ 42131da177e4SLinus Torvalds break; 42141da177e4SLinus Torvalds } 42151da177e4SLinus Torvalds } 42161da177e4SLinus Torvalds } 42171da177e4SLinus Torvalds flush_old_journal_lists(p_s_sb); 42181da177e4SLinus Torvalds 4219bd4c625cSLinus Torvalds journal->j_current_jl->j_list_bitmap = 4220bd4c625cSLinus Torvalds get_list_bitmap(p_s_sb, journal->j_current_jl); 42211da177e4SLinus Torvalds 42221da177e4SLinus Torvalds if (!(journal->j_current_jl->j_list_bitmap)) { 4223bd4c625cSLinus Torvalds reiserfs_panic(p_s_sb, 4224bd4c625cSLinus Torvalds "journal-1996: do_journal_end, could not get a list bitmap\n"); 42251da177e4SLinus Torvalds } 42261da177e4SLinus Torvalds 42271da177e4SLinus Torvalds atomic_set(&(journal->j_jlock), 0); 42281da177e4SLinus Torvalds unlock_journal(p_s_sb); 42291da177e4SLinus Torvalds /* wake up any body waiting to join. */ 42301da177e4SLinus Torvalds clear_bit(J_WRITERS_QUEUED, &journal->j_state); 42311da177e4SLinus Torvalds wake_up(&(journal->j_join_wait)); 42321da177e4SLinus Torvalds 42331da177e4SLinus Torvalds if (!flush && wait_on_commit && 42341da177e4SLinus Torvalds journal_list_still_alive(p_s_sb, commit_trans_id)) { 42351da177e4SLinus Torvalds flush_commit_list(p_s_sb, jl, 1); 42361da177e4SLinus Torvalds } 42371da177e4SLinus Torvalds out: 42381da177e4SLinus Torvalds reiserfs_check_lock_depth(p_s_sb, "journal end2"); 42391da177e4SLinus Torvalds 42401da177e4SLinus Torvalds memset(th, 0, sizeof(*th)); 42411da177e4SLinus Torvalds /* Re-set th->t_super, so we can properly keep track of how many 42421da177e4SLinus Torvalds * persistent transactions there are. We need to do this so if this 42431da177e4SLinus Torvalds * call is part of a failed restart_transaction, we can free it later */ 42441da177e4SLinus Torvalds th->t_super = p_s_sb; 42451da177e4SLinus Torvalds 42461da177e4SLinus Torvalds return journal->j_errno; 42471da177e4SLinus Torvalds } 42481da177e4SLinus Torvalds 4249bd4c625cSLinus Torvalds static void __reiserfs_journal_abort_hard(struct super_block *sb) 42501da177e4SLinus Torvalds { 42511da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(sb); 42521da177e4SLinus Torvalds if (test_bit(J_ABORTED, &journal->j_state)) 42531da177e4SLinus Torvalds return; 42541da177e4SLinus Torvalds 42551da177e4SLinus Torvalds printk(KERN_CRIT "REISERFS: Aborting journal for filesystem on %s\n", 42561da177e4SLinus Torvalds reiserfs_bdevname(sb)); 42571da177e4SLinus Torvalds 42581da177e4SLinus Torvalds sb->s_flags |= MS_RDONLY; 42591da177e4SLinus Torvalds set_bit(J_ABORTED, &journal->j_state); 42601da177e4SLinus Torvalds 42611da177e4SLinus Torvalds #ifdef CONFIG_REISERFS_CHECK 42621da177e4SLinus Torvalds dump_stack(); 42631da177e4SLinus Torvalds #endif 42641da177e4SLinus Torvalds } 42651da177e4SLinus Torvalds 4266bd4c625cSLinus Torvalds static void __reiserfs_journal_abort_soft(struct super_block *sb, int errno) 42671da177e4SLinus Torvalds { 42681da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(sb); 42691da177e4SLinus Torvalds if (test_bit(J_ABORTED, &journal->j_state)) 42701da177e4SLinus Torvalds return; 42711da177e4SLinus Torvalds 42721da177e4SLinus Torvalds if (!journal->j_errno) 42731da177e4SLinus Torvalds journal->j_errno = errno; 42741da177e4SLinus Torvalds 42751da177e4SLinus Torvalds __reiserfs_journal_abort_hard(sb); 42761da177e4SLinus Torvalds } 42771da177e4SLinus Torvalds 4278bd4c625cSLinus Torvalds void reiserfs_journal_abort(struct super_block *sb, int errno) 42791da177e4SLinus Torvalds { 42801da177e4SLinus Torvalds return __reiserfs_journal_abort_soft(sb, errno); 42811da177e4SLinus Torvalds } 4282