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 <linux/config.h> 381da177e4SLinus Torvalds #include <asm/uaccess.h> 391da177e4SLinus Torvalds #include <asm/system.h> 401da177e4SLinus Torvalds 411da177e4SLinus Torvalds #include <linux/time.h> 421da177e4SLinus Torvalds #include <asm/semaphore.h> 431da177e4SLinus Torvalds 441da177e4SLinus Torvalds #include <linux/vmalloc.h> 451da177e4SLinus Torvalds #include <linux/reiserfs_fs.h> 461da177e4SLinus Torvalds 471da177e4SLinus Torvalds #include <linux/kernel.h> 481da177e4SLinus Torvalds #include <linux/errno.h> 491da177e4SLinus Torvalds #include <linux/fcntl.h> 501da177e4SLinus Torvalds #include <linux/stat.h> 511da177e4SLinus Torvalds #include <linux/string.h> 521da177e4SLinus Torvalds #include <linux/smp_lock.h> 531da177e4SLinus Torvalds #include <linux/buffer_head.h> 541da177e4SLinus Torvalds #include <linux/workqueue.h> 551da177e4SLinus Torvalds #include <linux/writeback.h> 561da177e4SLinus Torvalds #include <linux/blkdev.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); 1071da177e4SLinus Torvalds static void flush_async_commits(void *p); 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; 7221da177e4SLinus Torvalds if (chunk->nr >= CHUNK_SIZE) 7231da177e4SLinus Torvalds BUG(); 7241da177e4SLinus Torvalds chunk->bh[chunk->nr++] = bh; 7251da177e4SLinus Torvalds if (chunk->nr >= CHUNK_SIZE) { 7261da177e4SLinus Torvalds ret = 1; 7271da177e4SLinus Torvalds if (lock) 7281da177e4SLinus Torvalds spin_unlock(lock); 7291da177e4SLinus Torvalds fn(chunk); 7301da177e4SLinus Torvalds if (lock) 7311da177e4SLinus Torvalds spin_lock(lock); 7321da177e4SLinus Torvalds } 7331da177e4SLinus Torvalds return ret; 7341da177e4SLinus Torvalds } 7351da177e4SLinus Torvalds 7361da177e4SLinus Torvalds static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0); 737bd4c625cSLinus Torvalds static struct reiserfs_jh *alloc_jh(void) 738bd4c625cSLinus Torvalds { 7391da177e4SLinus Torvalds struct reiserfs_jh *jh; 7401da177e4SLinus Torvalds while (1) { 7411da177e4SLinus Torvalds jh = kmalloc(sizeof(*jh), GFP_NOFS); 7421da177e4SLinus Torvalds if (jh) { 7431da177e4SLinus Torvalds atomic_inc(&nr_reiserfs_jh); 7441da177e4SLinus Torvalds return jh; 7451da177e4SLinus Torvalds } 7461da177e4SLinus Torvalds yield(); 7471da177e4SLinus Torvalds } 7481da177e4SLinus Torvalds } 7491da177e4SLinus Torvalds 7501da177e4SLinus Torvalds /* 7511da177e4SLinus Torvalds * we want to free the jh when the buffer has been written 7521da177e4SLinus Torvalds * and waited on 7531da177e4SLinus Torvalds */ 754bd4c625cSLinus Torvalds void reiserfs_free_jh(struct buffer_head *bh) 755bd4c625cSLinus Torvalds { 7561da177e4SLinus Torvalds struct reiserfs_jh *jh; 7571da177e4SLinus Torvalds 7581da177e4SLinus Torvalds jh = bh->b_private; 7591da177e4SLinus Torvalds if (jh) { 7601da177e4SLinus Torvalds bh->b_private = NULL; 7611da177e4SLinus Torvalds jh->bh = NULL; 7621da177e4SLinus Torvalds list_del_init(&jh->list); 7631da177e4SLinus Torvalds kfree(jh); 7641da177e4SLinus Torvalds if (atomic_read(&nr_reiserfs_jh) <= 0) 7651da177e4SLinus Torvalds BUG(); 7661da177e4SLinus Torvalds atomic_dec(&nr_reiserfs_jh); 7671da177e4SLinus Torvalds put_bh(bh); 7681da177e4SLinus Torvalds } 7691da177e4SLinus Torvalds } 7701da177e4SLinus Torvalds 7711da177e4SLinus Torvalds static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh, 7721da177e4SLinus Torvalds int tail) 7731da177e4SLinus Torvalds { 7741da177e4SLinus Torvalds struct reiserfs_jh *jh; 7751da177e4SLinus Torvalds 7761da177e4SLinus Torvalds if (bh->b_private) { 7771da177e4SLinus Torvalds spin_lock(&j->j_dirty_buffers_lock); 7781da177e4SLinus Torvalds if (!bh->b_private) { 7791da177e4SLinus Torvalds spin_unlock(&j->j_dirty_buffers_lock); 7801da177e4SLinus Torvalds goto no_jh; 7811da177e4SLinus Torvalds } 7821da177e4SLinus Torvalds jh = bh->b_private; 7831da177e4SLinus Torvalds list_del_init(&jh->list); 7841da177e4SLinus Torvalds } else { 7851da177e4SLinus Torvalds no_jh: 7861da177e4SLinus Torvalds get_bh(bh); 7871da177e4SLinus Torvalds jh = alloc_jh(); 7881da177e4SLinus Torvalds spin_lock(&j->j_dirty_buffers_lock); 7891da177e4SLinus Torvalds /* buffer must be locked for __add_jh, should be able to have 7901da177e4SLinus Torvalds * two adds at the same time 7911da177e4SLinus Torvalds */ 7921da177e4SLinus Torvalds if (bh->b_private) 7931da177e4SLinus Torvalds BUG(); 7941da177e4SLinus Torvalds jh->bh = bh; 7951da177e4SLinus Torvalds bh->b_private = jh; 7961da177e4SLinus Torvalds } 7971da177e4SLinus Torvalds jh->jl = j->j_current_jl; 7981da177e4SLinus Torvalds if (tail) 7991da177e4SLinus Torvalds list_add_tail(&jh->list, &jh->jl->j_tail_bh_list); 8001da177e4SLinus Torvalds else { 8011da177e4SLinus Torvalds list_add_tail(&jh->list, &jh->jl->j_bh_list); 8021da177e4SLinus Torvalds } 8031da177e4SLinus Torvalds spin_unlock(&j->j_dirty_buffers_lock); 8041da177e4SLinus Torvalds return 0; 8051da177e4SLinus Torvalds } 8061da177e4SLinus Torvalds 807bd4c625cSLinus Torvalds int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh) 808bd4c625cSLinus Torvalds { 8091da177e4SLinus Torvalds return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1); 8101da177e4SLinus Torvalds } 811bd4c625cSLinus Torvalds int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh) 812bd4c625cSLinus Torvalds { 8131da177e4SLinus Torvalds return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0); 8141da177e4SLinus Torvalds } 8151da177e4SLinus Torvalds 8161da177e4SLinus Torvalds #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list) 8171da177e4SLinus Torvalds static int write_ordered_buffers(spinlock_t * lock, 8181da177e4SLinus Torvalds struct reiserfs_journal *j, 8191da177e4SLinus Torvalds struct reiserfs_journal_list *jl, 8201da177e4SLinus Torvalds struct list_head *list) 8211da177e4SLinus Torvalds { 8221da177e4SLinus Torvalds struct buffer_head *bh; 8231da177e4SLinus Torvalds struct reiserfs_jh *jh; 8241da177e4SLinus Torvalds int ret = j->j_errno; 8251da177e4SLinus Torvalds struct buffer_chunk chunk; 8261da177e4SLinus Torvalds struct list_head tmp; 8271da177e4SLinus Torvalds INIT_LIST_HEAD(&tmp); 8281da177e4SLinus Torvalds 8291da177e4SLinus Torvalds chunk.nr = 0; 8301da177e4SLinus Torvalds spin_lock(lock); 8311da177e4SLinus Torvalds while (!list_empty(list)) { 8321da177e4SLinus Torvalds jh = JH_ENTRY(list->next); 8331da177e4SLinus Torvalds bh = jh->bh; 8341da177e4SLinus Torvalds get_bh(bh); 8351da177e4SLinus Torvalds if (test_set_buffer_locked(bh)) { 8361da177e4SLinus Torvalds if (!buffer_dirty(bh)) { 8371da177e4SLinus Torvalds list_del_init(&jh->list); 8381da177e4SLinus Torvalds list_add(&jh->list, &tmp); 8391da177e4SLinus Torvalds goto loop_next; 8401da177e4SLinus Torvalds } 8411da177e4SLinus Torvalds spin_unlock(lock); 8421da177e4SLinus Torvalds if (chunk.nr) 8431da177e4SLinus Torvalds write_ordered_chunk(&chunk); 8441da177e4SLinus Torvalds wait_on_buffer(bh); 8451da177e4SLinus Torvalds cond_resched(); 8461da177e4SLinus Torvalds spin_lock(lock); 8471da177e4SLinus Torvalds goto loop_next; 8481da177e4SLinus Torvalds } 8491da177e4SLinus Torvalds if (buffer_dirty(bh)) { 8501da177e4SLinus Torvalds list_del_init(&jh->list); 8511da177e4SLinus Torvalds list_add(&jh->list, &tmp); 8521da177e4SLinus Torvalds add_to_chunk(&chunk, bh, lock, write_ordered_chunk); 8531da177e4SLinus Torvalds } else { 8541da177e4SLinus Torvalds reiserfs_free_jh(bh); 8551da177e4SLinus Torvalds unlock_buffer(bh); 8561da177e4SLinus Torvalds } 8571da177e4SLinus Torvalds loop_next: 8581da177e4SLinus Torvalds put_bh(bh); 8591da177e4SLinus Torvalds cond_resched_lock(lock); 8601da177e4SLinus Torvalds } 8611da177e4SLinus Torvalds if (chunk.nr) { 8621da177e4SLinus Torvalds spin_unlock(lock); 8631da177e4SLinus Torvalds write_ordered_chunk(&chunk); 8641da177e4SLinus Torvalds spin_lock(lock); 8651da177e4SLinus Torvalds } 8661da177e4SLinus Torvalds while (!list_empty(&tmp)) { 8671da177e4SLinus Torvalds jh = JH_ENTRY(tmp.prev); 8681da177e4SLinus Torvalds bh = jh->bh; 8691da177e4SLinus Torvalds get_bh(bh); 8701da177e4SLinus Torvalds reiserfs_free_jh(bh); 8711da177e4SLinus Torvalds 8721da177e4SLinus Torvalds if (buffer_locked(bh)) { 8731da177e4SLinus Torvalds spin_unlock(lock); 8741da177e4SLinus Torvalds wait_on_buffer(bh); 8751da177e4SLinus Torvalds spin_lock(lock); 8761da177e4SLinus Torvalds } 8771da177e4SLinus Torvalds if (!buffer_uptodate(bh)) { 8781da177e4SLinus Torvalds ret = -EIO; 8791da177e4SLinus Torvalds } 8801da177e4SLinus Torvalds put_bh(bh); 8811da177e4SLinus Torvalds cond_resched_lock(lock); 8821da177e4SLinus Torvalds } 8831da177e4SLinus Torvalds spin_unlock(lock); 8841da177e4SLinus Torvalds return ret; 8851da177e4SLinus Torvalds } 8861da177e4SLinus Torvalds 887bd4c625cSLinus Torvalds static int flush_older_commits(struct super_block *s, 888bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl) 889bd4c625cSLinus Torvalds { 8901da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 8911da177e4SLinus Torvalds struct reiserfs_journal_list *other_jl; 8921da177e4SLinus Torvalds struct reiserfs_journal_list *first_jl; 8931da177e4SLinus Torvalds struct list_head *entry; 8941da177e4SLinus Torvalds unsigned long trans_id = jl->j_trans_id; 8951da177e4SLinus Torvalds unsigned long other_trans_id; 8961da177e4SLinus Torvalds unsigned long first_trans_id; 8971da177e4SLinus Torvalds 8981da177e4SLinus Torvalds find_first: 8991da177e4SLinus Torvalds /* 9001da177e4SLinus Torvalds * first we walk backwards to find the oldest uncommitted transation 9011da177e4SLinus Torvalds */ 9021da177e4SLinus Torvalds first_jl = jl; 9031da177e4SLinus Torvalds entry = jl->j_list.prev; 9041da177e4SLinus Torvalds while (1) { 9051da177e4SLinus Torvalds other_jl = JOURNAL_LIST_ENTRY(entry); 9061da177e4SLinus Torvalds if (entry == &journal->j_journal_list || 9071da177e4SLinus Torvalds atomic_read(&other_jl->j_older_commits_done)) 9081da177e4SLinus Torvalds break; 9091da177e4SLinus Torvalds 9101da177e4SLinus Torvalds first_jl = other_jl; 9111da177e4SLinus Torvalds entry = other_jl->j_list.prev; 9121da177e4SLinus Torvalds } 9131da177e4SLinus Torvalds 9141da177e4SLinus Torvalds /* if we didn't find any older uncommitted transactions, return now */ 9151da177e4SLinus Torvalds if (first_jl == jl) { 9161da177e4SLinus Torvalds return 0; 9171da177e4SLinus Torvalds } 9181da177e4SLinus Torvalds 9191da177e4SLinus Torvalds first_trans_id = first_jl->j_trans_id; 9201da177e4SLinus Torvalds 9211da177e4SLinus Torvalds entry = &first_jl->j_list; 9221da177e4SLinus Torvalds while (1) { 9231da177e4SLinus Torvalds other_jl = JOURNAL_LIST_ENTRY(entry); 9241da177e4SLinus Torvalds other_trans_id = other_jl->j_trans_id; 9251da177e4SLinus Torvalds 9261da177e4SLinus Torvalds if (other_trans_id < trans_id) { 9271da177e4SLinus Torvalds if (atomic_read(&other_jl->j_commit_left) != 0) { 9281da177e4SLinus Torvalds flush_commit_list(s, other_jl, 0); 9291da177e4SLinus Torvalds 9301da177e4SLinus Torvalds /* list we were called with is gone, return */ 9311da177e4SLinus Torvalds if (!journal_list_still_alive(s, trans_id)) 9321da177e4SLinus Torvalds return 1; 9331da177e4SLinus Torvalds 9341da177e4SLinus Torvalds /* the one we just flushed is gone, this means all 9351da177e4SLinus Torvalds * older lists are also gone, so first_jl is no longer 9361da177e4SLinus Torvalds * valid either. Go back to the beginning. 9371da177e4SLinus Torvalds */ 938bd4c625cSLinus Torvalds if (!journal_list_still_alive 939bd4c625cSLinus Torvalds (s, other_trans_id)) { 9401da177e4SLinus Torvalds goto find_first; 9411da177e4SLinus Torvalds } 9421da177e4SLinus Torvalds } 9431da177e4SLinus Torvalds entry = entry->next; 9441da177e4SLinus Torvalds if (entry == &journal->j_journal_list) 9451da177e4SLinus Torvalds return 0; 9461da177e4SLinus Torvalds } else { 9471da177e4SLinus Torvalds return 0; 9481da177e4SLinus Torvalds } 9491da177e4SLinus Torvalds } 9501da177e4SLinus Torvalds return 0; 9511da177e4SLinus Torvalds } 952bd4c625cSLinus Torvalds int reiserfs_async_progress_wait(struct super_block *s) 953bd4c625cSLinus Torvalds { 9541da177e4SLinus Torvalds DEFINE_WAIT(wait); 9551da177e4SLinus Torvalds struct reiserfs_journal *j = SB_JOURNAL(s); 9561da177e4SLinus Torvalds if (atomic_read(&j->j_async_throttle)) 9571da177e4SLinus Torvalds blk_congestion_wait(WRITE, HZ / 10); 9581da177e4SLinus Torvalds return 0; 9591da177e4SLinus Torvalds } 9601da177e4SLinus Torvalds 9611da177e4SLinus Torvalds /* 9621da177e4SLinus Torvalds ** if this journal list still has commit blocks unflushed, send them to disk. 9631da177e4SLinus Torvalds ** 9641da177e4SLinus Torvalds ** log areas must be flushed in order (transaction 2 can't commit before transaction 1) 9651da177e4SLinus Torvalds ** Before the commit block can by written, every other log block must be safely on disk 9661da177e4SLinus Torvalds ** 9671da177e4SLinus Torvalds */ 968bd4c625cSLinus Torvalds static int flush_commit_list(struct super_block *s, 969bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl, int flushall) 970bd4c625cSLinus Torvalds { 9711da177e4SLinus Torvalds int i; 9721da177e4SLinus Torvalds int bn; 9731da177e4SLinus Torvalds struct buffer_head *tbh = NULL; 9741da177e4SLinus Torvalds unsigned long trans_id = jl->j_trans_id; 9751da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 9761da177e4SLinus Torvalds int barrier = 0; 9771da177e4SLinus Torvalds int retval = 0; 9781da177e4SLinus Torvalds 9791da177e4SLinus Torvalds reiserfs_check_lock_depth(s, "flush_commit_list"); 9801da177e4SLinus Torvalds 9811da177e4SLinus Torvalds if (atomic_read(&jl->j_older_commits_done)) { 9821da177e4SLinus Torvalds return 0; 9831da177e4SLinus Torvalds } 9841da177e4SLinus Torvalds 98522e2c507SJens Axboe get_fs_excl(); 98622e2c507SJens Axboe 9871da177e4SLinus Torvalds /* before we can put our commit blocks on disk, we have to make sure everyone older than 9881da177e4SLinus Torvalds ** us is on disk too 9891da177e4SLinus Torvalds */ 9901da177e4SLinus Torvalds BUG_ON(jl->j_len <= 0); 9911da177e4SLinus Torvalds BUG_ON(trans_id == journal->j_trans_id); 9921da177e4SLinus Torvalds 9931da177e4SLinus Torvalds get_journal_list(jl); 9941da177e4SLinus Torvalds if (flushall) { 9951da177e4SLinus Torvalds if (flush_older_commits(s, jl) == 1) { 9961da177e4SLinus Torvalds /* list disappeared during flush_older_commits. return */ 9971da177e4SLinus Torvalds goto put_jl; 9981da177e4SLinus Torvalds } 9991da177e4SLinus Torvalds } 10001da177e4SLinus Torvalds 10011da177e4SLinus Torvalds /* make sure nobody is trying to flush this one at the same time */ 10021da177e4SLinus Torvalds down(&jl->j_commit_lock); 10031da177e4SLinus Torvalds if (!journal_list_still_alive(s, trans_id)) { 10041da177e4SLinus Torvalds up(&jl->j_commit_lock); 10051da177e4SLinus Torvalds goto put_jl; 10061da177e4SLinus Torvalds } 10071da177e4SLinus Torvalds BUG_ON(jl->j_trans_id == 0); 10081da177e4SLinus Torvalds 10091da177e4SLinus Torvalds /* this commit is done, exit */ 10101da177e4SLinus Torvalds if (atomic_read(&(jl->j_commit_left)) <= 0) { 10111da177e4SLinus Torvalds if (flushall) { 10121da177e4SLinus Torvalds atomic_set(&(jl->j_older_commits_done), 1); 10131da177e4SLinus Torvalds } 10141da177e4SLinus Torvalds up(&jl->j_commit_lock); 10151da177e4SLinus Torvalds goto put_jl; 10161da177e4SLinus Torvalds } 10171da177e4SLinus Torvalds 10181da177e4SLinus Torvalds if (!list_empty(&jl->j_bh_list)) { 10191da177e4SLinus Torvalds unlock_kernel(); 10201da177e4SLinus Torvalds write_ordered_buffers(&journal->j_dirty_buffers_lock, 10211da177e4SLinus Torvalds journal, jl, &jl->j_bh_list); 10221da177e4SLinus Torvalds lock_kernel(); 10231da177e4SLinus Torvalds } 10241da177e4SLinus Torvalds BUG_ON(!list_empty(&jl->j_bh_list)); 10251da177e4SLinus Torvalds /* 10261da177e4SLinus Torvalds * for the description block and all the log blocks, submit any buffers 10271da177e4SLinus Torvalds * that haven't already reached the disk 10281da177e4SLinus Torvalds */ 10291da177e4SLinus Torvalds atomic_inc(&journal->j_async_throttle); 10301da177e4SLinus Torvalds for (i = 0; i < (jl->j_len + 1); i++) { 10311da177e4SLinus Torvalds bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) % 10321da177e4SLinus Torvalds SB_ONDISK_JOURNAL_SIZE(s); 10331da177e4SLinus Torvalds tbh = journal_find_get_block(s, bn); 10341da177e4SLinus Torvalds if (buffer_dirty(tbh)) /* redundant, ll_rw_block() checks */ 103553778ffdSJan Kara ll_rw_block(SWRITE, 1, &tbh); 10361da177e4SLinus Torvalds put_bh(tbh); 10371da177e4SLinus Torvalds } 10381da177e4SLinus Torvalds atomic_dec(&journal->j_async_throttle); 10391da177e4SLinus Torvalds 10405d5e8156SJeff Mahoney /* We're skipping the commit if there's an error */ 10415d5e8156SJeff Mahoney if (retval || reiserfs_is_journal_aborted(journal)) 10425d5e8156SJeff Mahoney barrier = 0; 10435d5e8156SJeff Mahoney 10441da177e4SLinus Torvalds /* wait on everything written so far before writing the commit 10451da177e4SLinus Torvalds * if we are in barrier mode, send the commit down now 10461da177e4SLinus Torvalds */ 10471da177e4SLinus Torvalds barrier = reiserfs_barrier_flush(s); 10481da177e4SLinus Torvalds if (barrier) { 10491da177e4SLinus Torvalds int ret; 10501da177e4SLinus Torvalds lock_buffer(jl->j_commit_bh); 10511da177e4SLinus Torvalds ret = submit_barrier_buffer(jl->j_commit_bh); 10521da177e4SLinus Torvalds if (ret == -EOPNOTSUPP) { 10531da177e4SLinus Torvalds set_buffer_uptodate(jl->j_commit_bh); 10541da177e4SLinus Torvalds disable_barrier(s); 10551da177e4SLinus Torvalds barrier = 0; 10561da177e4SLinus Torvalds } 10571da177e4SLinus Torvalds } 10581da177e4SLinus Torvalds for (i = 0; i < (jl->j_len + 1); i++) { 10591da177e4SLinus Torvalds bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + 10601da177e4SLinus Torvalds (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s); 10611da177e4SLinus Torvalds tbh = journal_find_get_block(s, bn); 10621da177e4SLinus Torvalds wait_on_buffer(tbh); 10631da177e4SLinus Torvalds // since we're using ll_rw_blk above, it might have skipped over 10641da177e4SLinus Torvalds // a locked buffer. Double check here 10651da177e4SLinus Torvalds // 10661da177e4SLinus Torvalds if (buffer_dirty(tbh)) /* redundant, sync_dirty_buffer() checks */ 10671da177e4SLinus Torvalds sync_dirty_buffer(tbh); 10681da177e4SLinus Torvalds if (unlikely(!buffer_uptodate(tbh))) { 10691da177e4SLinus Torvalds #ifdef CONFIG_REISERFS_CHECK 10701da177e4SLinus Torvalds reiserfs_warning(s, "journal-601, buffer write failed"); 10711da177e4SLinus Torvalds #endif 10721da177e4SLinus Torvalds retval = -EIO; 10731da177e4SLinus Torvalds } 10741da177e4SLinus Torvalds put_bh(tbh); /* once for journal_find_get_block */ 10751da177e4SLinus Torvalds put_bh(tbh); /* once due to original getblk in do_journal_end */ 10761da177e4SLinus Torvalds atomic_dec(&(jl->j_commit_left)); 10771da177e4SLinus Torvalds } 10781da177e4SLinus Torvalds 10791da177e4SLinus Torvalds BUG_ON(atomic_read(&(jl->j_commit_left)) != 1); 10801da177e4SLinus Torvalds 10811da177e4SLinus Torvalds if (!barrier) { 10825d5e8156SJeff Mahoney /* If there was a write error in the journal - we can't commit 10835d5e8156SJeff Mahoney * this transaction - it will be invalid and, if successful, 10845d5e8156SJeff Mahoney * will just end up propogating the write error out to 10855d5e8156SJeff Mahoney * the file system. */ 10865d5e8156SJeff Mahoney if (likely(!retval && !reiserfs_is_journal_aborted (journal))) { 10871da177e4SLinus Torvalds if (buffer_dirty(jl->j_commit_bh)) 10881da177e4SLinus Torvalds BUG(); 10891da177e4SLinus Torvalds mark_buffer_dirty(jl->j_commit_bh) ; 10901da177e4SLinus Torvalds sync_dirty_buffer(jl->j_commit_bh) ; 10915d5e8156SJeff Mahoney } 10921da177e4SLinus Torvalds } else 10931da177e4SLinus Torvalds wait_on_buffer(jl->j_commit_bh); 10941da177e4SLinus Torvalds 10951da177e4SLinus Torvalds check_barrier_completion(s, jl->j_commit_bh); 10961da177e4SLinus Torvalds 10971da177e4SLinus Torvalds /* If there was a write error in the journal - we can't commit this 10981da177e4SLinus Torvalds * transaction - it will be invalid and, if successful, will just end 10991da177e4SLinus Torvalds * up propogating the write error out to the filesystem. */ 11001da177e4SLinus Torvalds if (unlikely(!buffer_uptodate(jl->j_commit_bh))) { 11011da177e4SLinus Torvalds #ifdef CONFIG_REISERFS_CHECK 11021da177e4SLinus Torvalds reiserfs_warning(s, "journal-615: buffer write failed"); 11031da177e4SLinus Torvalds #endif 11041da177e4SLinus Torvalds retval = -EIO; 11051da177e4SLinus Torvalds } 11061da177e4SLinus Torvalds bforget(jl->j_commit_bh); 11071da177e4SLinus Torvalds if (journal->j_last_commit_id != 0 && 11081da177e4SLinus Torvalds (jl->j_trans_id - journal->j_last_commit_id) != 1) { 11091da177e4SLinus Torvalds reiserfs_warning(s, "clm-2200: last commit %lu, current %lu", 1110bd4c625cSLinus Torvalds journal->j_last_commit_id, jl->j_trans_id); 11111da177e4SLinus Torvalds } 11121da177e4SLinus Torvalds journal->j_last_commit_id = jl->j_trans_id; 11131da177e4SLinus Torvalds 11141da177e4SLinus Torvalds /* now, every commit block is on the disk. It is safe to allow blocks freed during this transaction to be reallocated */ 11151da177e4SLinus Torvalds cleanup_freed_for_journal_list(s, jl); 11161da177e4SLinus Torvalds 11171da177e4SLinus Torvalds retval = retval ? retval : journal->j_errno; 11181da177e4SLinus Torvalds 11191da177e4SLinus Torvalds /* mark the metadata dirty */ 11201da177e4SLinus Torvalds if (!retval) 11211da177e4SLinus Torvalds dirty_one_transaction(s, jl); 11221da177e4SLinus Torvalds atomic_dec(&(jl->j_commit_left)); 11231da177e4SLinus Torvalds 11241da177e4SLinus Torvalds if (flushall) { 11251da177e4SLinus Torvalds atomic_set(&(jl->j_older_commits_done), 1); 11261da177e4SLinus Torvalds } 11271da177e4SLinus Torvalds up(&jl->j_commit_lock); 11281da177e4SLinus Torvalds put_jl: 11291da177e4SLinus Torvalds put_journal_list(s, jl); 11301da177e4SLinus Torvalds 11311da177e4SLinus Torvalds if (retval) 1132bd4c625cSLinus Torvalds reiserfs_abort(s, retval, "Journal write error in %s", 1133bd4c625cSLinus Torvalds __FUNCTION__); 113422e2c507SJens Axboe put_fs_excl(); 11351da177e4SLinus Torvalds return retval; 11361da177e4SLinus Torvalds } 11371da177e4SLinus Torvalds 11381da177e4SLinus Torvalds /* 11391da177e4SLinus Torvalds ** flush_journal_list frequently needs to find a newer transaction for a given block. This does that, or 11401da177e4SLinus Torvalds ** returns NULL if it can't find anything 11411da177e4SLinus Torvalds */ 1142bd4c625cSLinus Torvalds static struct reiserfs_journal_list *find_newer_jl_for_cn(struct 1143bd4c625cSLinus Torvalds reiserfs_journal_cnode 1144bd4c625cSLinus Torvalds *cn) 1145bd4c625cSLinus Torvalds { 11461da177e4SLinus Torvalds struct super_block *sb = cn->sb; 11471da177e4SLinus Torvalds b_blocknr_t blocknr = cn->blocknr; 11481da177e4SLinus Torvalds 11491da177e4SLinus Torvalds cn = cn->hprev; 11501da177e4SLinus Torvalds while (cn) { 11511da177e4SLinus Torvalds if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) { 11521da177e4SLinus Torvalds return cn->jlist; 11531da177e4SLinus Torvalds } 11541da177e4SLinus Torvalds cn = cn->hprev; 11551da177e4SLinus Torvalds } 11561da177e4SLinus Torvalds return NULL; 11571da177e4SLinus Torvalds } 11581da177e4SLinus Torvalds 1159bd4c625cSLinus Torvalds static void remove_journal_hash(struct super_block *, 1160bd4c625cSLinus Torvalds struct reiserfs_journal_cnode **, 1161bd4c625cSLinus Torvalds struct reiserfs_journal_list *, unsigned long, 1162bd4c625cSLinus Torvalds int); 11631da177e4SLinus Torvalds 11641da177e4SLinus Torvalds /* 11651da177e4SLinus Torvalds ** once all the real blocks have been flushed, it is safe to remove them from the 11661da177e4SLinus Torvalds ** journal list for this transaction. Aside from freeing the cnode, this also allows the 11671da177e4SLinus Torvalds ** block to be reallocated for data blocks if it had been deleted. 11681da177e4SLinus Torvalds */ 1169bd4c625cSLinus Torvalds static void remove_all_from_journal_list(struct super_block *p_s_sb, 1170bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl, 1171bd4c625cSLinus Torvalds int debug) 1172bd4c625cSLinus Torvalds { 11731da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 11741da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn, *last; 11751da177e4SLinus Torvalds cn = jl->j_realblock; 11761da177e4SLinus Torvalds 11771da177e4SLinus Torvalds /* which is better, to lock once around the whole loop, or 11781da177e4SLinus Torvalds ** to lock for each call to remove_journal_hash? 11791da177e4SLinus Torvalds */ 11801da177e4SLinus Torvalds while (cn) { 11811da177e4SLinus Torvalds if (cn->blocknr != 0) { 11821da177e4SLinus Torvalds if (debug) { 1183bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 1184bd4c625cSLinus Torvalds "block %u, bh is %d, state %ld", 1185bd4c625cSLinus Torvalds cn->blocknr, cn->bh ? 1 : 0, 1186bd4c625cSLinus Torvalds cn->state); 11871da177e4SLinus Torvalds } 11881da177e4SLinus Torvalds cn->state = 0; 1189bd4c625cSLinus Torvalds remove_journal_hash(p_s_sb, journal->j_list_hash_table, 1190bd4c625cSLinus Torvalds jl, cn->blocknr, 1); 11911da177e4SLinus Torvalds } 11921da177e4SLinus Torvalds last = cn; 11931da177e4SLinus Torvalds cn = cn->next; 11941da177e4SLinus Torvalds free_cnode(p_s_sb, last); 11951da177e4SLinus Torvalds } 11961da177e4SLinus Torvalds jl->j_realblock = NULL; 11971da177e4SLinus Torvalds } 11981da177e4SLinus Torvalds 11991da177e4SLinus Torvalds /* 12001da177e4SLinus Torvalds ** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block. 12011da177e4SLinus Torvalds ** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start 12021da177e4SLinus Torvalds ** releasing blocks in this transaction for reuse as data blocks. 12031da177e4SLinus Torvalds ** called by flush_journal_list, before it calls remove_all_from_journal_list 12041da177e4SLinus Torvalds ** 12051da177e4SLinus Torvalds */ 1206bd4c625cSLinus Torvalds static int _update_journal_header_block(struct super_block *p_s_sb, 1207bd4c625cSLinus Torvalds unsigned long offset, 1208bd4c625cSLinus Torvalds unsigned long trans_id) 1209bd4c625cSLinus Torvalds { 12101da177e4SLinus Torvalds struct reiserfs_journal_header *jh; 12111da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 12121da177e4SLinus Torvalds 12131da177e4SLinus Torvalds if (reiserfs_is_journal_aborted(journal)) 12141da177e4SLinus Torvalds return -EIO; 12151da177e4SLinus Torvalds 12161da177e4SLinus Torvalds if (trans_id >= journal->j_last_flush_trans_id) { 12171da177e4SLinus Torvalds if (buffer_locked((journal->j_header_bh))) { 12181da177e4SLinus Torvalds wait_on_buffer((journal->j_header_bh)); 12191da177e4SLinus Torvalds if (unlikely(!buffer_uptodate(journal->j_header_bh))) { 12201da177e4SLinus Torvalds #ifdef CONFIG_REISERFS_CHECK 1221bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 1222bd4c625cSLinus Torvalds "journal-699: buffer write failed"); 12231da177e4SLinus Torvalds #endif 12241da177e4SLinus Torvalds return -EIO; 12251da177e4SLinus Torvalds } 12261da177e4SLinus Torvalds } 12271da177e4SLinus Torvalds journal->j_last_flush_trans_id = trans_id; 12281da177e4SLinus Torvalds journal->j_first_unflushed_offset = offset; 1229bd4c625cSLinus Torvalds jh = (struct reiserfs_journal_header *)(journal->j_header_bh-> 1230bd4c625cSLinus Torvalds b_data); 12311da177e4SLinus Torvalds jh->j_last_flush_trans_id = cpu_to_le32(trans_id); 12321da177e4SLinus Torvalds jh->j_first_unflushed_offset = cpu_to_le32(offset); 12331da177e4SLinus Torvalds jh->j_mount_id = cpu_to_le32(journal->j_mount_id); 12341da177e4SLinus Torvalds 12351da177e4SLinus Torvalds if (reiserfs_barrier_flush(p_s_sb)) { 12361da177e4SLinus Torvalds int ret; 12371da177e4SLinus Torvalds lock_buffer(journal->j_header_bh); 12381da177e4SLinus Torvalds ret = submit_barrier_buffer(journal->j_header_bh); 12391da177e4SLinus Torvalds if (ret == -EOPNOTSUPP) { 12401da177e4SLinus Torvalds set_buffer_uptodate(journal->j_header_bh); 12411da177e4SLinus Torvalds disable_barrier(p_s_sb); 12421da177e4SLinus Torvalds goto sync; 12431da177e4SLinus Torvalds } 12441da177e4SLinus Torvalds wait_on_buffer(journal->j_header_bh); 12451da177e4SLinus Torvalds check_barrier_completion(p_s_sb, journal->j_header_bh); 12461da177e4SLinus Torvalds } else { 12471da177e4SLinus Torvalds sync: 12481da177e4SLinus Torvalds set_buffer_dirty(journal->j_header_bh); 12491da177e4SLinus Torvalds sync_dirty_buffer(journal->j_header_bh); 12501da177e4SLinus Torvalds } 12511da177e4SLinus Torvalds if (!buffer_uptodate(journal->j_header_bh)) { 1252bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 1253bd4c625cSLinus Torvalds "journal-837: IO error during journal replay"); 12541da177e4SLinus Torvalds return -EIO; 12551da177e4SLinus Torvalds } 12561da177e4SLinus Torvalds } 12571da177e4SLinus Torvalds return 0; 12581da177e4SLinus Torvalds } 12591da177e4SLinus Torvalds 12601da177e4SLinus Torvalds static int update_journal_header_block(struct super_block *p_s_sb, 12611da177e4SLinus Torvalds unsigned long offset, 1262bd4c625cSLinus Torvalds unsigned long trans_id) 1263bd4c625cSLinus Torvalds { 12641da177e4SLinus Torvalds return _update_journal_header_block(p_s_sb, offset, trans_id); 12651da177e4SLinus Torvalds } 1266bd4c625cSLinus Torvalds 12671da177e4SLinus Torvalds /* 12681da177e4SLinus Torvalds ** flush any and all journal lists older than you are 12691da177e4SLinus Torvalds ** can only be called from flush_journal_list 12701da177e4SLinus Torvalds */ 12711da177e4SLinus Torvalds static int flush_older_journal_lists(struct super_block *p_s_sb, 12721da177e4SLinus Torvalds struct reiserfs_journal_list *jl) 12731da177e4SLinus Torvalds { 12741da177e4SLinus Torvalds struct list_head *entry; 12751da177e4SLinus Torvalds struct reiserfs_journal_list *other_jl; 12761da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 12771da177e4SLinus Torvalds unsigned long trans_id = jl->j_trans_id; 12781da177e4SLinus Torvalds 12791da177e4SLinus Torvalds /* we know we are the only ones flushing things, no extra race 12801da177e4SLinus Torvalds * protection is required. 12811da177e4SLinus Torvalds */ 12821da177e4SLinus Torvalds restart: 12831da177e4SLinus Torvalds entry = journal->j_journal_list.next; 12841da177e4SLinus Torvalds /* Did we wrap? */ 12851da177e4SLinus Torvalds if (entry == &journal->j_journal_list) 12861da177e4SLinus Torvalds return 0; 12871da177e4SLinus Torvalds other_jl = JOURNAL_LIST_ENTRY(entry); 12881da177e4SLinus Torvalds if (other_jl->j_trans_id < trans_id) { 12891da177e4SLinus Torvalds BUG_ON(other_jl->j_refcount <= 0); 12901da177e4SLinus Torvalds /* do not flush all */ 12911da177e4SLinus Torvalds flush_journal_list(p_s_sb, other_jl, 0); 12921da177e4SLinus Torvalds 12931da177e4SLinus Torvalds /* other_jl is now deleted from the list */ 12941da177e4SLinus Torvalds goto restart; 12951da177e4SLinus Torvalds } 12961da177e4SLinus Torvalds return 0; 12971da177e4SLinus Torvalds } 12981da177e4SLinus Torvalds 12991da177e4SLinus Torvalds static void del_from_work_list(struct super_block *s, 1300bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl) 1301bd4c625cSLinus Torvalds { 13021da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 13031da177e4SLinus Torvalds if (!list_empty(&jl->j_working_list)) { 13041da177e4SLinus Torvalds list_del_init(&jl->j_working_list); 13051da177e4SLinus Torvalds journal->j_num_work_lists--; 13061da177e4SLinus Torvalds } 13071da177e4SLinus Torvalds } 13081da177e4SLinus Torvalds 13091da177e4SLinus Torvalds /* flush a journal list, both commit and real blocks 13101da177e4SLinus Torvalds ** 13111da177e4SLinus Torvalds ** always set flushall to 1, unless you are calling from inside 13121da177e4SLinus Torvalds ** flush_journal_list 13131da177e4SLinus Torvalds ** 13141da177e4SLinus Torvalds ** IMPORTANT. This can only be called while there are no journal writers, 13151da177e4SLinus Torvalds ** and the journal is locked. That means it can only be called from 13161da177e4SLinus Torvalds ** do_journal_end, or by journal_release 13171da177e4SLinus Torvalds */ 13181da177e4SLinus Torvalds static int flush_journal_list(struct super_block *s, 1319bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl, int flushall) 1320bd4c625cSLinus Torvalds { 13211da177e4SLinus Torvalds struct reiserfs_journal_list *pjl; 13221da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn, *last; 13231da177e4SLinus Torvalds int count; 13241da177e4SLinus Torvalds int was_jwait = 0; 13251da177e4SLinus Torvalds int was_dirty = 0; 13261da177e4SLinus Torvalds struct buffer_head *saved_bh; 13271da177e4SLinus Torvalds unsigned long j_len_saved = jl->j_len; 13281da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 13291da177e4SLinus Torvalds int err = 0; 13301da177e4SLinus Torvalds 13311da177e4SLinus Torvalds BUG_ON(j_len_saved <= 0); 13321da177e4SLinus Torvalds 13331da177e4SLinus Torvalds if (atomic_read(&journal->j_wcount) != 0) { 1334bd4c625cSLinus Torvalds reiserfs_warning(s, 1335bd4c625cSLinus Torvalds "clm-2048: flush_journal_list called with wcount %d", 13361da177e4SLinus Torvalds atomic_read(&journal->j_wcount)); 13371da177e4SLinus Torvalds } 13381da177e4SLinus Torvalds BUG_ON(jl->j_trans_id == 0); 13391da177e4SLinus Torvalds 13401da177e4SLinus Torvalds /* if flushall == 0, the lock is already held */ 13411da177e4SLinus Torvalds if (flushall) { 13421da177e4SLinus Torvalds down(&journal->j_flush_sem); 13431da177e4SLinus Torvalds } else if (!down_trylock(&journal->j_flush_sem)) { 13441da177e4SLinus Torvalds BUG(); 13451da177e4SLinus Torvalds } 13461da177e4SLinus Torvalds 13471da177e4SLinus Torvalds count = 0; 13481da177e4SLinus Torvalds if (j_len_saved > journal->j_trans_max) { 1349bd4c625cSLinus Torvalds reiserfs_panic(s, 1350bd4c625cSLinus Torvalds "journal-715: flush_journal_list, length is %lu, trans id %lu\n", 1351bd4c625cSLinus Torvalds j_len_saved, jl->j_trans_id); 13521da177e4SLinus Torvalds return 0; 13531da177e4SLinus Torvalds } 13541da177e4SLinus Torvalds 135522e2c507SJens Axboe get_fs_excl(); 135622e2c507SJens Axboe 13571da177e4SLinus Torvalds /* if all the work is already done, get out of here */ 13581da177e4SLinus Torvalds if (atomic_read(&(jl->j_nonzerolen)) <= 0 && 13591da177e4SLinus Torvalds atomic_read(&(jl->j_commit_left)) <= 0) { 13601da177e4SLinus Torvalds goto flush_older_and_return; 13611da177e4SLinus Torvalds } 13621da177e4SLinus Torvalds 13631da177e4SLinus Torvalds /* start by putting the commit list on disk. This will also flush 13641da177e4SLinus Torvalds ** the commit lists of any olders transactions 13651da177e4SLinus Torvalds */ 13661da177e4SLinus Torvalds flush_commit_list(s, jl, 1); 13671da177e4SLinus Torvalds 1368bd4c625cSLinus Torvalds if (!(jl->j_state & LIST_DIRTY) 1369bd4c625cSLinus Torvalds && !reiserfs_is_journal_aborted(journal)) 13701da177e4SLinus Torvalds BUG(); 13711da177e4SLinus Torvalds 13721da177e4SLinus Torvalds /* are we done now? */ 13731da177e4SLinus Torvalds if (atomic_read(&(jl->j_nonzerolen)) <= 0 && 13741da177e4SLinus Torvalds atomic_read(&(jl->j_commit_left)) <= 0) { 13751da177e4SLinus Torvalds goto flush_older_and_return; 13761da177e4SLinus Torvalds } 13771da177e4SLinus Torvalds 13781da177e4SLinus Torvalds /* loop through each cnode, see if we need to write it, 13791da177e4SLinus Torvalds ** or wait on a more recent transaction, or just ignore it 13801da177e4SLinus Torvalds */ 13811da177e4SLinus Torvalds if (atomic_read(&(journal->j_wcount)) != 0) { 1382bd4c625cSLinus Torvalds reiserfs_panic(s, 1383bd4c625cSLinus Torvalds "journal-844: panic journal list is flushing, wcount is not 0\n"); 13841da177e4SLinus Torvalds } 13851da177e4SLinus Torvalds cn = jl->j_realblock; 13861da177e4SLinus Torvalds while (cn) { 13871da177e4SLinus Torvalds was_jwait = 0; 13881da177e4SLinus Torvalds was_dirty = 0; 13891da177e4SLinus Torvalds saved_bh = NULL; 13901da177e4SLinus Torvalds /* blocknr of 0 is no longer in the hash, ignore it */ 13911da177e4SLinus Torvalds if (cn->blocknr == 0) { 13921da177e4SLinus Torvalds goto free_cnode; 13931da177e4SLinus Torvalds } 13941da177e4SLinus Torvalds 13951da177e4SLinus Torvalds /* This transaction failed commit. Don't write out to the disk */ 13961da177e4SLinus Torvalds if (!(jl->j_state & LIST_DIRTY)) 13971da177e4SLinus Torvalds goto free_cnode; 13981da177e4SLinus Torvalds 13991da177e4SLinus Torvalds pjl = find_newer_jl_for_cn(cn); 14001da177e4SLinus Torvalds /* the order is important here. We check pjl to make sure we 14011da177e4SLinus Torvalds ** don't clear BH_JDirty_wait if we aren't the one writing this 14021da177e4SLinus Torvalds ** block to disk 14031da177e4SLinus Torvalds */ 14041da177e4SLinus Torvalds if (!pjl && cn->bh) { 14051da177e4SLinus Torvalds saved_bh = cn->bh; 14061da177e4SLinus Torvalds 14071da177e4SLinus Torvalds /* we do this to make sure nobody releases the buffer while 14081da177e4SLinus Torvalds ** we are working with it 14091da177e4SLinus Torvalds */ 14101da177e4SLinus Torvalds get_bh(saved_bh); 14111da177e4SLinus Torvalds 14121da177e4SLinus Torvalds if (buffer_journal_dirty(saved_bh)) { 14131da177e4SLinus Torvalds BUG_ON(!can_dirty(cn)); 14141da177e4SLinus Torvalds was_jwait = 1; 14151da177e4SLinus Torvalds was_dirty = 1; 14161da177e4SLinus Torvalds } else if (can_dirty(cn)) { 14171da177e4SLinus Torvalds /* everything with !pjl && jwait should be writable */ 14181da177e4SLinus Torvalds BUG(); 14191da177e4SLinus Torvalds } 14201da177e4SLinus Torvalds } 14211da177e4SLinus Torvalds 14221da177e4SLinus Torvalds /* if someone has this block in a newer transaction, just make 14231da177e4SLinus Torvalds ** sure they are commited, and don't try writing it to disk 14241da177e4SLinus Torvalds */ 14251da177e4SLinus Torvalds if (pjl) { 14261da177e4SLinus Torvalds if (atomic_read(&pjl->j_commit_left)) 14271da177e4SLinus Torvalds flush_commit_list(s, pjl, 1); 14281da177e4SLinus Torvalds goto free_cnode; 14291da177e4SLinus Torvalds } 14301da177e4SLinus Torvalds 14311da177e4SLinus Torvalds /* bh == NULL when the block got to disk on its own, OR, 14321da177e4SLinus Torvalds ** the block got freed in a future transaction 14331da177e4SLinus Torvalds */ 14341da177e4SLinus Torvalds if (saved_bh == NULL) { 14351da177e4SLinus Torvalds goto free_cnode; 14361da177e4SLinus Torvalds } 14371da177e4SLinus Torvalds 14381da177e4SLinus Torvalds /* this should never happen. kupdate_one_transaction has this list 14391da177e4SLinus Torvalds ** locked while it works, so we should never see a buffer here that 14401da177e4SLinus Torvalds ** is not marked JDirty_wait 14411da177e4SLinus Torvalds */ 14421da177e4SLinus Torvalds if ((!was_jwait) && !buffer_locked(saved_bh)) { 1443bd4c625cSLinus Torvalds reiserfs_warning(s, 1444bd4c625cSLinus Torvalds "journal-813: BAD! buffer %llu %cdirty %cjwait, " 14451da177e4SLinus Torvalds "not in a newer tranasction", 1446bd4c625cSLinus Torvalds (unsigned long long)saved_bh-> 1447bd4c625cSLinus Torvalds b_blocknr, was_dirty ? ' ' : '!', 1448bd4c625cSLinus Torvalds was_jwait ? ' ' : '!'); 14491da177e4SLinus Torvalds } 14501da177e4SLinus Torvalds if (was_dirty) { 14511da177e4SLinus Torvalds /* we inc again because saved_bh gets decremented at free_cnode */ 14521da177e4SLinus Torvalds get_bh(saved_bh); 14531da177e4SLinus Torvalds set_bit(BLOCK_NEEDS_FLUSH, &cn->state); 14541da177e4SLinus Torvalds lock_buffer(saved_bh); 14551da177e4SLinus Torvalds BUG_ON(cn->blocknr != saved_bh->b_blocknr); 14561da177e4SLinus Torvalds if (buffer_dirty(saved_bh)) 14571da177e4SLinus Torvalds submit_logged_buffer(saved_bh); 14581da177e4SLinus Torvalds else 14591da177e4SLinus Torvalds unlock_buffer(saved_bh); 14601da177e4SLinus Torvalds count++; 14611da177e4SLinus Torvalds } else { 1462bd4c625cSLinus Torvalds reiserfs_warning(s, 1463bd4c625cSLinus Torvalds "clm-2082: Unable to flush buffer %llu in %s", 1464bd4c625cSLinus Torvalds (unsigned long long)saved_bh-> 1465bd4c625cSLinus Torvalds b_blocknr, __FUNCTION__); 14661da177e4SLinus Torvalds } 14671da177e4SLinus Torvalds free_cnode: 14681da177e4SLinus Torvalds last = cn; 14691da177e4SLinus Torvalds cn = cn->next; 14701da177e4SLinus Torvalds if (saved_bh) { 14711da177e4SLinus Torvalds /* we incremented this to keep others from taking the buffer head away */ 14721da177e4SLinus Torvalds put_bh(saved_bh); 14731da177e4SLinus Torvalds if (atomic_read(&(saved_bh->b_count)) < 0) { 1474bd4c625cSLinus Torvalds reiserfs_warning(s, 1475bd4c625cSLinus Torvalds "journal-945: saved_bh->b_count < 0"); 14761da177e4SLinus Torvalds } 14771da177e4SLinus Torvalds } 14781da177e4SLinus Torvalds } 14791da177e4SLinus Torvalds if (count > 0) { 14801da177e4SLinus Torvalds cn = jl->j_realblock; 14811da177e4SLinus Torvalds while (cn) { 14821da177e4SLinus Torvalds if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) { 14831da177e4SLinus Torvalds if (!cn->bh) { 1484bd4c625cSLinus Torvalds reiserfs_panic(s, 1485bd4c625cSLinus Torvalds "journal-1011: cn->bh is NULL\n"); 14861da177e4SLinus Torvalds } 14871da177e4SLinus Torvalds wait_on_buffer(cn->bh); 14881da177e4SLinus Torvalds if (!cn->bh) { 1489bd4c625cSLinus Torvalds reiserfs_panic(s, 1490bd4c625cSLinus Torvalds "journal-1012: cn->bh is NULL\n"); 14911da177e4SLinus Torvalds } 14921da177e4SLinus Torvalds if (unlikely(!buffer_uptodate(cn->bh))) { 14931da177e4SLinus Torvalds #ifdef CONFIG_REISERFS_CHECK 1494bd4c625cSLinus Torvalds reiserfs_warning(s, 1495bd4c625cSLinus Torvalds "journal-949: buffer write failed\n"); 14961da177e4SLinus Torvalds #endif 14971da177e4SLinus Torvalds err = -EIO; 14981da177e4SLinus Torvalds } 14991da177e4SLinus Torvalds /* note, we must clear the JDirty_wait bit after the up to date 15001da177e4SLinus Torvalds ** check, otherwise we race against our flushpage routine 15011da177e4SLinus Torvalds */ 1502bd4c625cSLinus Torvalds BUG_ON(!test_clear_buffer_journal_dirty 1503bd4c625cSLinus Torvalds (cn->bh)); 15041da177e4SLinus Torvalds 15051da177e4SLinus Torvalds /* undo the inc from journal_mark_dirty */ 15061da177e4SLinus Torvalds put_bh(cn->bh); 15071da177e4SLinus Torvalds brelse(cn->bh); 15081da177e4SLinus Torvalds } 15091da177e4SLinus Torvalds cn = cn->next; 15101da177e4SLinus Torvalds } 15111da177e4SLinus Torvalds } 15121da177e4SLinus Torvalds 15131da177e4SLinus Torvalds if (err) 1514bd4c625cSLinus Torvalds reiserfs_abort(s, -EIO, 1515bd4c625cSLinus Torvalds "Write error while pushing transaction to disk in %s", 1516bd4c625cSLinus Torvalds __FUNCTION__); 15171da177e4SLinus Torvalds flush_older_and_return: 15181da177e4SLinus Torvalds 15191da177e4SLinus Torvalds /* before we can update the journal header block, we _must_ flush all 15201da177e4SLinus Torvalds ** real blocks from all older transactions to disk. This is because 15211da177e4SLinus Torvalds ** once the header block is updated, this transaction will not be 15221da177e4SLinus Torvalds ** replayed after a crash 15231da177e4SLinus Torvalds */ 15241da177e4SLinus Torvalds if (flushall) { 15251da177e4SLinus Torvalds flush_older_journal_lists(s, jl); 15261da177e4SLinus Torvalds } 15271da177e4SLinus Torvalds 15281da177e4SLinus Torvalds err = journal->j_errno; 15291da177e4SLinus Torvalds /* before we can remove everything from the hash tables for this 15301da177e4SLinus Torvalds ** transaction, we must make sure it can never be replayed 15311da177e4SLinus Torvalds ** 15321da177e4SLinus Torvalds ** since we are only called from do_journal_end, we know for sure there 15331da177e4SLinus Torvalds ** are no allocations going on while we are flushing journal lists. So, 15341da177e4SLinus Torvalds ** we only need to update the journal header block for the last list 15351da177e4SLinus Torvalds ** being flushed 15361da177e4SLinus Torvalds */ 15371da177e4SLinus Torvalds if (!err && flushall) { 1538bd4c625cSLinus Torvalds err = 1539bd4c625cSLinus Torvalds update_journal_header_block(s, 1540bd4c625cSLinus Torvalds (jl->j_start + jl->j_len + 1541bd4c625cSLinus Torvalds 2) % SB_ONDISK_JOURNAL_SIZE(s), 1542bd4c625cSLinus Torvalds jl->j_trans_id); 15431da177e4SLinus Torvalds if (err) 1544bd4c625cSLinus Torvalds reiserfs_abort(s, -EIO, 1545bd4c625cSLinus Torvalds "Write error while updating journal header in %s", 1546bd4c625cSLinus Torvalds __FUNCTION__); 15471da177e4SLinus Torvalds } 15481da177e4SLinus Torvalds remove_all_from_journal_list(s, jl, 0); 15491da177e4SLinus Torvalds list_del_init(&jl->j_list); 15501da177e4SLinus Torvalds journal->j_num_lists--; 15511da177e4SLinus Torvalds del_from_work_list(s, jl); 15521da177e4SLinus Torvalds 15531da177e4SLinus Torvalds if (journal->j_last_flush_id != 0 && 15541da177e4SLinus Torvalds (jl->j_trans_id - journal->j_last_flush_id) != 1) { 15551da177e4SLinus Torvalds reiserfs_warning(s, "clm-2201: last flush %lu, current %lu", 1556bd4c625cSLinus Torvalds journal->j_last_flush_id, jl->j_trans_id); 15571da177e4SLinus Torvalds } 15581da177e4SLinus Torvalds journal->j_last_flush_id = jl->j_trans_id; 15591da177e4SLinus Torvalds 15601da177e4SLinus Torvalds /* not strictly required since we are freeing the list, but it should 15611da177e4SLinus Torvalds * help find code using dead lists later on 15621da177e4SLinus Torvalds */ 15631da177e4SLinus Torvalds jl->j_len = 0; 15641da177e4SLinus Torvalds atomic_set(&(jl->j_nonzerolen), 0); 15651da177e4SLinus Torvalds jl->j_start = 0; 15661da177e4SLinus Torvalds jl->j_realblock = NULL; 15671da177e4SLinus Torvalds jl->j_commit_bh = NULL; 15681da177e4SLinus Torvalds jl->j_trans_id = 0; 15691da177e4SLinus Torvalds jl->j_state = 0; 15701da177e4SLinus Torvalds put_journal_list(s, jl); 15711da177e4SLinus Torvalds if (flushall) 15721da177e4SLinus Torvalds up(&journal->j_flush_sem); 157322e2c507SJens Axboe put_fs_excl(); 15741da177e4SLinus Torvalds return err; 15751da177e4SLinus Torvalds } 15761da177e4SLinus Torvalds 15771da177e4SLinus Torvalds static int write_one_transaction(struct super_block *s, 15781da177e4SLinus Torvalds struct reiserfs_journal_list *jl, 15791da177e4SLinus Torvalds struct buffer_chunk *chunk) 15801da177e4SLinus Torvalds { 15811da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 15821da177e4SLinus Torvalds int ret = 0; 15831da177e4SLinus Torvalds 15841da177e4SLinus Torvalds jl->j_state |= LIST_TOUCHED; 15851da177e4SLinus Torvalds del_from_work_list(s, jl); 15861da177e4SLinus Torvalds if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) { 15871da177e4SLinus Torvalds return 0; 15881da177e4SLinus Torvalds } 15891da177e4SLinus Torvalds 15901da177e4SLinus Torvalds cn = jl->j_realblock; 15911da177e4SLinus Torvalds while (cn) { 15921da177e4SLinus Torvalds /* if the blocknr == 0, this has been cleared from the hash, 15931da177e4SLinus Torvalds ** skip it 15941da177e4SLinus Torvalds */ 15951da177e4SLinus Torvalds if (cn->blocknr == 0) { 15961da177e4SLinus Torvalds goto next; 15971da177e4SLinus Torvalds } 15981da177e4SLinus Torvalds if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) { 15991da177e4SLinus Torvalds struct buffer_head *tmp_bh; 16001da177e4SLinus Torvalds /* we can race against journal_mark_freed when we try 16011da177e4SLinus Torvalds * to lock_buffer(cn->bh), so we have to inc the buffer 16021da177e4SLinus Torvalds * count, and recheck things after locking 16031da177e4SLinus Torvalds */ 16041da177e4SLinus Torvalds tmp_bh = cn->bh; 16051da177e4SLinus Torvalds get_bh(tmp_bh); 16061da177e4SLinus Torvalds lock_buffer(tmp_bh); 16071da177e4SLinus Torvalds if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) { 16081da177e4SLinus Torvalds if (!buffer_journal_dirty(tmp_bh) || 16091da177e4SLinus Torvalds buffer_journal_prepared(tmp_bh)) 16101da177e4SLinus Torvalds BUG(); 16111da177e4SLinus Torvalds add_to_chunk(chunk, tmp_bh, NULL, write_chunk); 16121da177e4SLinus Torvalds ret++; 16131da177e4SLinus Torvalds } else { 16141da177e4SLinus Torvalds /* note, cn->bh might be null now */ 16151da177e4SLinus Torvalds unlock_buffer(tmp_bh); 16161da177e4SLinus Torvalds } 16171da177e4SLinus Torvalds put_bh(tmp_bh); 16181da177e4SLinus Torvalds } 16191da177e4SLinus Torvalds next: 16201da177e4SLinus Torvalds cn = cn->next; 16211da177e4SLinus Torvalds cond_resched(); 16221da177e4SLinus Torvalds } 16231da177e4SLinus Torvalds return ret; 16241da177e4SLinus Torvalds } 16251da177e4SLinus Torvalds 16261da177e4SLinus Torvalds /* used by flush_commit_list */ 16271da177e4SLinus Torvalds static int dirty_one_transaction(struct super_block *s, 16281da177e4SLinus Torvalds struct reiserfs_journal_list *jl) 16291da177e4SLinus Torvalds { 16301da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 16311da177e4SLinus Torvalds struct reiserfs_journal_list *pjl; 16321da177e4SLinus Torvalds int ret = 0; 16331da177e4SLinus Torvalds 16341da177e4SLinus Torvalds jl->j_state |= LIST_DIRTY; 16351da177e4SLinus Torvalds cn = jl->j_realblock; 16361da177e4SLinus Torvalds while (cn) { 16371da177e4SLinus Torvalds /* look for a more recent transaction that logged this 16381da177e4SLinus Torvalds ** buffer. Only the most recent transaction with a buffer in 16391da177e4SLinus Torvalds ** it is allowed to send that buffer to disk 16401da177e4SLinus Torvalds */ 16411da177e4SLinus Torvalds pjl = find_newer_jl_for_cn(cn); 1642bd4c625cSLinus Torvalds if (!pjl && cn->blocknr && cn->bh 1643bd4c625cSLinus Torvalds && buffer_journal_dirty(cn->bh)) { 16441da177e4SLinus Torvalds BUG_ON(!can_dirty(cn)); 16451da177e4SLinus Torvalds /* if the buffer is prepared, it will either be logged 16461da177e4SLinus Torvalds * or restored. If restored, we need to make sure 16471da177e4SLinus Torvalds * it actually gets marked dirty 16481da177e4SLinus Torvalds */ 16491da177e4SLinus Torvalds clear_buffer_journal_new(cn->bh); 16501da177e4SLinus Torvalds if (buffer_journal_prepared(cn->bh)) { 16511da177e4SLinus Torvalds set_buffer_journal_restore_dirty(cn->bh); 16521da177e4SLinus Torvalds } else { 16531da177e4SLinus Torvalds set_buffer_journal_test(cn->bh); 16541da177e4SLinus Torvalds mark_buffer_dirty(cn->bh); 16551da177e4SLinus Torvalds } 16561da177e4SLinus Torvalds } 16571da177e4SLinus Torvalds cn = cn->next; 16581da177e4SLinus Torvalds } 16591da177e4SLinus Torvalds return ret; 16601da177e4SLinus Torvalds } 16611da177e4SLinus Torvalds 16621da177e4SLinus Torvalds static int kupdate_transactions(struct super_block *s, 16631da177e4SLinus Torvalds struct reiserfs_journal_list *jl, 16641da177e4SLinus Torvalds struct reiserfs_journal_list **next_jl, 16651da177e4SLinus Torvalds unsigned long *next_trans_id, 1666bd4c625cSLinus Torvalds int num_blocks, int num_trans) 1667bd4c625cSLinus Torvalds { 16681da177e4SLinus Torvalds int ret = 0; 16691da177e4SLinus Torvalds int written = 0; 16701da177e4SLinus Torvalds int transactions_flushed = 0; 16711da177e4SLinus Torvalds unsigned long orig_trans_id = jl->j_trans_id; 16721da177e4SLinus Torvalds struct buffer_chunk chunk; 16731da177e4SLinus Torvalds struct list_head *entry; 16741da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 16751da177e4SLinus Torvalds chunk.nr = 0; 16761da177e4SLinus Torvalds 16771da177e4SLinus Torvalds down(&journal->j_flush_sem); 16781da177e4SLinus Torvalds if (!journal_list_still_alive(s, orig_trans_id)) { 16791da177e4SLinus Torvalds goto done; 16801da177e4SLinus Torvalds } 16811da177e4SLinus Torvalds 16821da177e4SLinus Torvalds /* we've got j_flush_sem held, nobody is going to delete any 16831da177e4SLinus Torvalds * of these lists out from underneath us 16841da177e4SLinus Torvalds */ 16851da177e4SLinus Torvalds while ((num_trans && transactions_flushed < num_trans) || 16861da177e4SLinus Torvalds (!num_trans && written < num_blocks)) { 16871da177e4SLinus Torvalds 16881da177e4SLinus Torvalds if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) || 1689bd4c625cSLinus Torvalds atomic_read(&jl->j_commit_left) 1690bd4c625cSLinus Torvalds || !(jl->j_state & LIST_DIRTY)) { 16911da177e4SLinus Torvalds del_from_work_list(s, jl); 16921da177e4SLinus Torvalds break; 16931da177e4SLinus Torvalds } 16941da177e4SLinus Torvalds ret = write_one_transaction(s, jl, &chunk); 16951da177e4SLinus Torvalds 16961da177e4SLinus Torvalds if (ret < 0) 16971da177e4SLinus Torvalds goto done; 16981da177e4SLinus Torvalds transactions_flushed++; 16991da177e4SLinus Torvalds written += ret; 17001da177e4SLinus Torvalds entry = jl->j_list.next; 17011da177e4SLinus Torvalds 17021da177e4SLinus Torvalds /* did we wrap? */ 17031da177e4SLinus Torvalds if (entry == &journal->j_journal_list) { 17041da177e4SLinus Torvalds break; 17051da177e4SLinus Torvalds } 17061da177e4SLinus Torvalds jl = JOURNAL_LIST_ENTRY(entry); 17071da177e4SLinus Torvalds 17081da177e4SLinus Torvalds /* don't bother with older transactions */ 17091da177e4SLinus Torvalds if (jl->j_trans_id <= orig_trans_id) 17101da177e4SLinus Torvalds break; 17111da177e4SLinus Torvalds } 17121da177e4SLinus Torvalds if (chunk.nr) { 17131da177e4SLinus Torvalds write_chunk(&chunk); 17141da177e4SLinus Torvalds } 17151da177e4SLinus Torvalds 17161da177e4SLinus Torvalds done: 17171da177e4SLinus Torvalds up(&journal->j_flush_sem); 17181da177e4SLinus Torvalds return ret; 17191da177e4SLinus Torvalds } 17201da177e4SLinus Torvalds 17211da177e4SLinus Torvalds /* for o_sync and fsync heavy applications, they tend to use 17221da177e4SLinus Torvalds ** all the journa list slots with tiny transactions. These 17231da177e4SLinus Torvalds ** trigger lots and lots of calls to update the header block, which 17241da177e4SLinus Torvalds ** adds seeks and slows things down. 17251da177e4SLinus Torvalds ** 17261da177e4SLinus Torvalds ** This function tries to clear out a large chunk of the journal lists 17271da177e4SLinus Torvalds ** at once, which makes everything faster since only the newest journal 17281da177e4SLinus Torvalds ** list updates the header block 17291da177e4SLinus Torvalds */ 17301da177e4SLinus Torvalds static int flush_used_journal_lists(struct super_block *s, 1731bd4c625cSLinus Torvalds struct reiserfs_journal_list *jl) 1732bd4c625cSLinus Torvalds { 17331da177e4SLinus Torvalds unsigned long len = 0; 17341da177e4SLinus Torvalds unsigned long cur_len; 17351da177e4SLinus Torvalds int ret; 17361da177e4SLinus Torvalds int i; 17371da177e4SLinus Torvalds int limit = 256; 17381da177e4SLinus Torvalds struct reiserfs_journal_list *tjl; 17391da177e4SLinus Torvalds struct reiserfs_journal_list *flush_jl; 17401da177e4SLinus Torvalds unsigned long trans_id; 17411da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 17421da177e4SLinus Torvalds 17431da177e4SLinus Torvalds flush_jl = tjl = jl; 17441da177e4SLinus Torvalds 17451da177e4SLinus Torvalds /* in data logging mode, try harder to flush a lot of blocks */ 17461da177e4SLinus Torvalds if (reiserfs_data_log(s)) 17471da177e4SLinus Torvalds limit = 1024; 17481da177e4SLinus Torvalds /* flush for 256 transactions or limit blocks, whichever comes first */ 17491da177e4SLinus Torvalds for (i = 0; i < 256 && len < limit; i++) { 17501da177e4SLinus Torvalds if (atomic_read(&tjl->j_commit_left) || 17511da177e4SLinus Torvalds tjl->j_trans_id < jl->j_trans_id) { 17521da177e4SLinus Torvalds break; 17531da177e4SLinus Torvalds } 17541da177e4SLinus Torvalds cur_len = atomic_read(&tjl->j_nonzerolen); 17551da177e4SLinus Torvalds if (cur_len > 0) { 17561da177e4SLinus Torvalds tjl->j_state &= ~LIST_TOUCHED; 17571da177e4SLinus Torvalds } 17581da177e4SLinus Torvalds len += cur_len; 17591da177e4SLinus Torvalds flush_jl = tjl; 17601da177e4SLinus Torvalds if (tjl->j_list.next == &journal->j_journal_list) 17611da177e4SLinus Torvalds break; 17621da177e4SLinus Torvalds tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next); 17631da177e4SLinus Torvalds } 17641da177e4SLinus Torvalds /* try to find a group of blocks we can flush across all the 17651da177e4SLinus Torvalds ** transactions, but only bother if we've actually spanned 17661da177e4SLinus Torvalds ** across multiple lists 17671da177e4SLinus Torvalds */ 17681da177e4SLinus Torvalds if (flush_jl != jl) { 17691da177e4SLinus Torvalds ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i); 17701da177e4SLinus Torvalds } 17711da177e4SLinus Torvalds flush_journal_list(s, flush_jl, 1); 17721da177e4SLinus Torvalds return 0; 17731da177e4SLinus Torvalds } 17741da177e4SLinus Torvalds 17751da177e4SLinus Torvalds /* 17761da177e4SLinus Torvalds ** removes any nodes in table with name block and dev as bh. 17771da177e4SLinus Torvalds ** only touchs the hnext and hprev pointers. 17781da177e4SLinus Torvalds */ 17791da177e4SLinus Torvalds void remove_journal_hash(struct super_block *sb, 17801da177e4SLinus Torvalds struct reiserfs_journal_cnode **table, 17811da177e4SLinus Torvalds struct reiserfs_journal_list *jl, 17821da177e4SLinus Torvalds unsigned long block, int remove_freed) 17831da177e4SLinus Torvalds { 17841da177e4SLinus Torvalds struct reiserfs_journal_cnode *cur; 17851da177e4SLinus Torvalds struct reiserfs_journal_cnode **head; 17861da177e4SLinus Torvalds 17871da177e4SLinus Torvalds head = &(journal_hash(table, sb, block)); 17881da177e4SLinus Torvalds if (!head) { 17891da177e4SLinus Torvalds return; 17901da177e4SLinus Torvalds } 17911da177e4SLinus Torvalds cur = *head; 17921da177e4SLinus Torvalds while (cur) { 1793bd4c625cSLinus Torvalds if (cur->blocknr == block && cur->sb == sb 1794bd4c625cSLinus Torvalds && (jl == NULL || jl == cur->jlist) 1795bd4c625cSLinus Torvalds && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) { 17961da177e4SLinus Torvalds if (cur->hnext) { 17971da177e4SLinus Torvalds cur->hnext->hprev = cur->hprev; 17981da177e4SLinus Torvalds } 17991da177e4SLinus Torvalds if (cur->hprev) { 18001da177e4SLinus Torvalds cur->hprev->hnext = cur->hnext; 18011da177e4SLinus Torvalds } else { 18021da177e4SLinus Torvalds *head = cur->hnext; 18031da177e4SLinus Torvalds } 18041da177e4SLinus Torvalds cur->blocknr = 0; 18051da177e4SLinus Torvalds cur->sb = NULL; 18061da177e4SLinus Torvalds cur->state = 0; 18071da177e4SLinus Torvalds if (cur->bh && cur->jlist) /* anybody who clears the cur->bh will also dec the nonzerolen */ 18081da177e4SLinus Torvalds atomic_dec(&(cur->jlist->j_nonzerolen)); 18091da177e4SLinus Torvalds cur->bh = NULL; 18101da177e4SLinus Torvalds cur->jlist = NULL; 18111da177e4SLinus Torvalds } 18121da177e4SLinus Torvalds cur = cur->hnext; 18131da177e4SLinus Torvalds } 18141da177e4SLinus Torvalds } 18151da177e4SLinus Torvalds 1816bd4c625cSLinus Torvalds static void free_journal_ram(struct super_block *p_s_sb) 1817bd4c625cSLinus Torvalds { 18181da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 1819d739b42bSPekka Enberg kfree(journal->j_current_jl); 18201da177e4SLinus Torvalds journal->j_num_lists--; 18211da177e4SLinus Torvalds 18221da177e4SLinus Torvalds vfree(journal->j_cnode_free_orig); 18231da177e4SLinus Torvalds free_list_bitmaps(p_s_sb, journal->j_list_bitmap); 18241da177e4SLinus Torvalds free_bitmap_nodes(p_s_sb); /* must be after free_list_bitmaps */ 18251da177e4SLinus Torvalds if (journal->j_header_bh) { 18261da177e4SLinus Torvalds brelse(journal->j_header_bh); 18271da177e4SLinus Torvalds } 18281da177e4SLinus Torvalds /* j_header_bh is on the journal dev, make sure not to release the journal 18291da177e4SLinus Torvalds * dev until we brelse j_header_bh 18301da177e4SLinus Torvalds */ 18311da177e4SLinus Torvalds release_journal_dev(p_s_sb, journal); 18321da177e4SLinus Torvalds vfree(journal); 18331da177e4SLinus Torvalds } 18341da177e4SLinus Torvalds 18351da177e4SLinus Torvalds /* 18361da177e4SLinus Torvalds ** call on unmount. Only set error to 1 if you haven't made your way out 18371da177e4SLinus Torvalds ** of read_super() yet. Any other caller must keep error at 0. 18381da177e4SLinus Torvalds */ 1839bd4c625cSLinus Torvalds static int do_journal_release(struct reiserfs_transaction_handle *th, 1840bd4c625cSLinus Torvalds struct super_block *p_s_sb, int error) 1841bd4c625cSLinus Torvalds { 18421da177e4SLinus Torvalds struct reiserfs_transaction_handle myth; 18431da177e4SLinus Torvalds int flushed = 0; 18441da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 18451da177e4SLinus Torvalds 18461da177e4SLinus Torvalds /* we only want to flush out transactions if we were called with error == 0 18471da177e4SLinus Torvalds */ 18481da177e4SLinus Torvalds if (!error && !(p_s_sb->s_flags & MS_RDONLY)) { 18491da177e4SLinus Torvalds /* end the current trans */ 18501da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 18511da177e4SLinus Torvalds do_journal_end(th, p_s_sb, 10, FLUSH_ALL); 18521da177e4SLinus Torvalds 18531da177e4SLinus Torvalds /* make sure something gets logged to force our way into the flush code */ 18541da177e4SLinus Torvalds if (!journal_join(&myth, p_s_sb, 1)) { 1855bd4c625cSLinus Torvalds reiserfs_prepare_for_journal(p_s_sb, 1856bd4c625cSLinus Torvalds SB_BUFFER_WITH_SB(p_s_sb), 1857bd4c625cSLinus Torvalds 1); 1858bd4c625cSLinus Torvalds journal_mark_dirty(&myth, p_s_sb, 1859bd4c625cSLinus Torvalds SB_BUFFER_WITH_SB(p_s_sb)); 18601da177e4SLinus Torvalds do_journal_end(&myth, p_s_sb, 1, FLUSH_ALL); 18611da177e4SLinus Torvalds flushed = 1; 18621da177e4SLinus Torvalds } 18631da177e4SLinus Torvalds } 18641da177e4SLinus Torvalds 18651da177e4SLinus Torvalds /* this also catches errors during the do_journal_end above */ 18661da177e4SLinus Torvalds if (!error && reiserfs_is_journal_aborted(journal)) { 18671da177e4SLinus Torvalds memset(&myth, 0, sizeof(myth)); 18681da177e4SLinus Torvalds if (!journal_join_abort(&myth, p_s_sb, 1)) { 1869bd4c625cSLinus Torvalds reiserfs_prepare_for_journal(p_s_sb, 1870bd4c625cSLinus Torvalds SB_BUFFER_WITH_SB(p_s_sb), 1871bd4c625cSLinus Torvalds 1); 1872bd4c625cSLinus Torvalds journal_mark_dirty(&myth, p_s_sb, 1873bd4c625cSLinus Torvalds SB_BUFFER_WITH_SB(p_s_sb)); 18741da177e4SLinus Torvalds do_journal_end(&myth, p_s_sb, 1, FLUSH_ALL); 18751da177e4SLinus Torvalds } 18761da177e4SLinus Torvalds } 18771da177e4SLinus Torvalds 18781da177e4SLinus Torvalds reiserfs_mounted_fs_count--; 18791da177e4SLinus Torvalds /* wait for all commits to finish */ 18801da177e4SLinus Torvalds cancel_delayed_work(&SB_JOURNAL(p_s_sb)->j_work); 18811da177e4SLinus Torvalds flush_workqueue(commit_wq); 18821da177e4SLinus Torvalds if (!reiserfs_mounted_fs_count) { 18831da177e4SLinus Torvalds destroy_workqueue(commit_wq); 18841da177e4SLinus Torvalds commit_wq = NULL; 18851da177e4SLinus Torvalds } 18861da177e4SLinus Torvalds 18871da177e4SLinus Torvalds free_journal_ram(p_s_sb); 18881da177e4SLinus Torvalds 18891da177e4SLinus Torvalds return 0; 18901da177e4SLinus Torvalds } 18911da177e4SLinus Torvalds 18921da177e4SLinus Torvalds /* 18931da177e4SLinus Torvalds ** call on unmount. flush all journal trans, release all alloc'd ram 18941da177e4SLinus Torvalds */ 1895bd4c625cSLinus Torvalds int journal_release(struct reiserfs_transaction_handle *th, 1896bd4c625cSLinus Torvalds struct super_block *p_s_sb) 1897bd4c625cSLinus Torvalds { 18981da177e4SLinus Torvalds return do_journal_release(th, p_s_sb, 0); 18991da177e4SLinus Torvalds } 1900bd4c625cSLinus Torvalds 19011da177e4SLinus Torvalds /* 19021da177e4SLinus Torvalds ** only call from an error condition inside reiserfs_read_super! 19031da177e4SLinus Torvalds */ 1904bd4c625cSLinus Torvalds int journal_release_error(struct reiserfs_transaction_handle *th, 1905bd4c625cSLinus Torvalds struct super_block *p_s_sb) 1906bd4c625cSLinus Torvalds { 19071da177e4SLinus Torvalds return do_journal_release(th, p_s_sb, 1); 19081da177e4SLinus Torvalds } 19091da177e4SLinus Torvalds 19101da177e4SLinus Torvalds /* compares description block with commit block. returns 1 if they differ, 0 if they are the same */ 1911bd4c625cSLinus Torvalds static int journal_compare_desc_commit(struct super_block *p_s_sb, 1912bd4c625cSLinus Torvalds struct reiserfs_journal_desc *desc, 1913bd4c625cSLinus Torvalds struct reiserfs_journal_commit *commit) 1914bd4c625cSLinus Torvalds { 19151da177e4SLinus Torvalds if (get_commit_trans_id(commit) != get_desc_trans_id(desc) || 19161da177e4SLinus Torvalds get_commit_trans_len(commit) != get_desc_trans_len(desc) || 19171da177e4SLinus Torvalds get_commit_trans_len(commit) > SB_JOURNAL(p_s_sb)->j_trans_max || 1918bd4c625cSLinus Torvalds get_commit_trans_len(commit) <= 0) { 19191da177e4SLinus Torvalds return 1; 19201da177e4SLinus Torvalds } 19211da177e4SLinus Torvalds return 0; 19221da177e4SLinus Torvalds } 1923bd4c625cSLinus Torvalds 19241da177e4SLinus Torvalds /* returns 0 if it did not find a description block 19251da177e4SLinus Torvalds ** returns -1 if it found a corrupt commit block 19261da177e4SLinus Torvalds ** returns 1 if both desc and commit were valid 19271da177e4SLinus Torvalds */ 1928bd4c625cSLinus Torvalds static int journal_transaction_is_valid(struct super_block *p_s_sb, 1929bd4c625cSLinus Torvalds struct buffer_head *d_bh, 1930bd4c625cSLinus Torvalds unsigned long *oldest_invalid_trans_id, 1931bd4c625cSLinus Torvalds unsigned long *newest_mount_id) 1932bd4c625cSLinus Torvalds { 19331da177e4SLinus Torvalds struct reiserfs_journal_desc *desc; 19341da177e4SLinus Torvalds struct reiserfs_journal_commit *commit; 19351da177e4SLinus Torvalds struct buffer_head *c_bh; 19361da177e4SLinus Torvalds unsigned long offset; 19371da177e4SLinus Torvalds 19381da177e4SLinus Torvalds if (!d_bh) 19391da177e4SLinus Torvalds return 0; 19401da177e4SLinus Torvalds 19411da177e4SLinus Torvalds desc = (struct reiserfs_journal_desc *)d_bh->b_data; 1942bd4c625cSLinus Torvalds if (get_desc_trans_len(desc) > 0 1943bd4c625cSLinus Torvalds && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) { 1944bd4c625cSLinus Torvalds if (oldest_invalid_trans_id && *oldest_invalid_trans_id 1945bd4c625cSLinus Torvalds && get_desc_trans_id(desc) > *oldest_invalid_trans_id) { 1946bd4c625cSLinus Torvalds reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, 1947bd4c625cSLinus Torvalds "journal-986: transaction " 19481da177e4SLinus Torvalds "is valid returning because trans_id %d is greater than " 1949bd4c625cSLinus Torvalds "oldest_invalid %lu", 1950bd4c625cSLinus Torvalds get_desc_trans_id(desc), 19511da177e4SLinus Torvalds *oldest_invalid_trans_id); 19521da177e4SLinus Torvalds return 0; 19531da177e4SLinus Torvalds } 1954bd4c625cSLinus Torvalds if (newest_mount_id 1955bd4c625cSLinus Torvalds && *newest_mount_id > get_desc_mount_id(desc)) { 1956bd4c625cSLinus Torvalds reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, 1957bd4c625cSLinus Torvalds "journal-1087: transaction " 19581da177e4SLinus Torvalds "is valid returning because mount_id %d is less than " 1959bd4c625cSLinus Torvalds "newest_mount_id %lu", 1960bd4c625cSLinus Torvalds get_desc_mount_id(desc), 19611da177e4SLinus Torvalds *newest_mount_id); 19621da177e4SLinus Torvalds return -1; 19631da177e4SLinus Torvalds } 19641da177e4SLinus Torvalds if (get_desc_trans_len(desc) > SB_JOURNAL(p_s_sb)->j_trans_max) { 1965bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 1966bd4c625cSLinus Torvalds "journal-2018: Bad transaction length %d encountered, ignoring transaction", 1967bd4c625cSLinus Torvalds get_desc_trans_len(desc)); 19681da177e4SLinus Torvalds return -1; 19691da177e4SLinus Torvalds } 19701da177e4SLinus Torvalds offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb); 19711da177e4SLinus Torvalds 19721da177e4SLinus Torvalds /* ok, we have a journal description block, lets see if the transaction was valid */ 1973bd4c625cSLinus Torvalds c_bh = 1974bd4c625cSLinus Torvalds journal_bread(p_s_sb, 1975bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + 1976bd4c625cSLinus Torvalds ((offset + get_desc_trans_len(desc) + 1977bd4c625cSLinus Torvalds 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb))); 19781da177e4SLinus Torvalds if (!c_bh) 19791da177e4SLinus Torvalds return 0; 19801da177e4SLinus Torvalds commit = (struct reiserfs_journal_commit *)c_bh->b_data; 19811da177e4SLinus Torvalds if (journal_compare_desc_commit(p_s_sb, desc, commit)) { 19821da177e4SLinus Torvalds reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, 19831da177e4SLinus Torvalds "journal_transaction_is_valid, commit offset %ld had bad " 19841da177e4SLinus Torvalds "time %d or length %d", 1985bd4c625cSLinus Torvalds c_bh->b_blocknr - 1986bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb), 19871da177e4SLinus Torvalds get_commit_trans_id(commit), 19881da177e4SLinus Torvalds get_commit_trans_len(commit)); 19891da177e4SLinus Torvalds brelse(c_bh); 19901da177e4SLinus Torvalds if (oldest_invalid_trans_id) { 1991bd4c625cSLinus Torvalds *oldest_invalid_trans_id = 1992bd4c625cSLinus Torvalds get_desc_trans_id(desc); 1993bd4c625cSLinus Torvalds reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, 1994bd4c625cSLinus Torvalds "journal-1004: " 19951da177e4SLinus Torvalds "transaction_is_valid setting oldest invalid trans_id " 1996bd4c625cSLinus Torvalds "to %d", 1997bd4c625cSLinus Torvalds get_desc_trans_id(desc)); 19981da177e4SLinus Torvalds } 19991da177e4SLinus Torvalds return -1; 20001da177e4SLinus Torvalds } 20011da177e4SLinus Torvalds brelse(c_bh); 2002bd4c625cSLinus Torvalds reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, 2003bd4c625cSLinus Torvalds "journal-1006: found valid " 20041da177e4SLinus Torvalds "transaction start offset %llu, len %d id %d", 2005bd4c625cSLinus Torvalds d_bh->b_blocknr - 2006bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb), 2007bd4c625cSLinus Torvalds get_desc_trans_len(desc), 2008bd4c625cSLinus Torvalds get_desc_trans_id(desc)); 20091da177e4SLinus Torvalds return 1; 20101da177e4SLinus Torvalds } else { 20111da177e4SLinus Torvalds return 0; 20121da177e4SLinus Torvalds } 20131da177e4SLinus Torvalds } 20141da177e4SLinus Torvalds 2015bd4c625cSLinus Torvalds static void brelse_array(struct buffer_head **heads, int num) 2016bd4c625cSLinus Torvalds { 20171da177e4SLinus Torvalds int i; 20181da177e4SLinus Torvalds for (i = 0; i < num; i++) { 20191da177e4SLinus Torvalds brelse(heads[i]); 20201da177e4SLinus Torvalds } 20211da177e4SLinus Torvalds } 20221da177e4SLinus Torvalds 20231da177e4SLinus Torvalds /* 20241da177e4SLinus Torvalds ** given the start, and values for the oldest acceptable transactions, 20251da177e4SLinus Torvalds ** this either reads in a replays a transaction, or returns because the transaction 20261da177e4SLinus Torvalds ** is invalid, or too old. 20271da177e4SLinus Torvalds */ 2028bd4c625cSLinus Torvalds static int journal_read_transaction(struct super_block *p_s_sb, 2029bd4c625cSLinus Torvalds unsigned long cur_dblock, 2030bd4c625cSLinus Torvalds unsigned long oldest_start, 2031bd4c625cSLinus Torvalds unsigned long oldest_trans_id, 2032bd4c625cSLinus Torvalds unsigned long newest_mount_id) 2033bd4c625cSLinus Torvalds { 20341da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 20351da177e4SLinus Torvalds struct reiserfs_journal_desc *desc; 20361da177e4SLinus Torvalds struct reiserfs_journal_commit *commit; 20371da177e4SLinus Torvalds unsigned long trans_id = 0; 20381da177e4SLinus Torvalds struct buffer_head *c_bh; 20391da177e4SLinus Torvalds struct buffer_head *d_bh; 20401da177e4SLinus Torvalds struct buffer_head **log_blocks = NULL; 20411da177e4SLinus Torvalds struct buffer_head **real_blocks = NULL; 20421da177e4SLinus Torvalds unsigned long trans_offset; 20431da177e4SLinus Torvalds int i; 20441da177e4SLinus Torvalds int trans_half; 20451da177e4SLinus Torvalds 20461da177e4SLinus Torvalds d_bh = journal_bread(p_s_sb, cur_dblock); 20471da177e4SLinus Torvalds if (!d_bh) 20481da177e4SLinus Torvalds return 1; 20491da177e4SLinus Torvalds desc = (struct reiserfs_journal_desc *)d_bh->b_data; 20501da177e4SLinus Torvalds trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb); 20511da177e4SLinus Torvalds reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1037: " 20521da177e4SLinus Torvalds "journal_read_transaction, offset %llu, len %d mount_id %d", 20531da177e4SLinus Torvalds d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb), 20541da177e4SLinus Torvalds get_desc_trans_len(desc), get_desc_mount_id(desc)); 20551da177e4SLinus Torvalds if (get_desc_trans_id(desc) < oldest_trans_id) { 20561da177e4SLinus Torvalds reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1039: " 20571da177e4SLinus Torvalds "journal_read_trans skipping because %lu is too old", 2058bd4c625cSLinus Torvalds cur_dblock - 2059bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb)); 20601da177e4SLinus Torvalds brelse(d_bh); 20611da177e4SLinus Torvalds return 1; 20621da177e4SLinus Torvalds } 20631da177e4SLinus Torvalds if (get_desc_mount_id(desc) != newest_mount_id) { 20641da177e4SLinus Torvalds reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1146: " 20651da177e4SLinus Torvalds "journal_read_trans skipping because %d is != " 20661da177e4SLinus Torvalds "newest_mount_id %lu", get_desc_mount_id(desc), 20671da177e4SLinus Torvalds newest_mount_id); 20681da177e4SLinus Torvalds brelse(d_bh); 20691da177e4SLinus Torvalds return 1; 20701da177e4SLinus Torvalds } 20711da177e4SLinus Torvalds c_bh = journal_bread(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + 20721da177e4SLinus Torvalds ((trans_offset + get_desc_trans_len(desc) + 1) % 20731da177e4SLinus Torvalds SB_ONDISK_JOURNAL_SIZE(p_s_sb))); 20741da177e4SLinus Torvalds if (!c_bh) { 20751da177e4SLinus Torvalds brelse(d_bh); 20761da177e4SLinus Torvalds return 1; 20771da177e4SLinus Torvalds } 20781da177e4SLinus Torvalds commit = (struct reiserfs_journal_commit *)c_bh->b_data; 20791da177e4SLinus Torvalds if (journal_compare_desc_commit(p_s_sb, desc, commit)) { 2080bd4c625cSLinus Torvalds reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, 2081bd4c625cSLinus Torvalds "journal_read_transaction, " 20821da177e4SLinus Torvalds "commit offset %llu had bad time %d or length %d", 2083bd4c625cSLinus Torvalds c_bh->b_blocknr - 2084bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb), 2085bd4c625cSLinus Torvalds get_commit_trans_id(commit), 2086bd4c625cSLinus Torvalds get_commit_trans_len(commit)); 20871da177e4SLinus Torvalds brelse(c_bh); 20881da177e4SLinus Torvalds brelse(d_bh); 20891da177e4SLinus Torvalds return 1; 20901da177e4SLinus Torvalds } 20911da177e4SLinus Torvalds trans_id = get_desc_trans_id(desc); 20921da177e4SLinus Torvalds /* now we know we've got a good transaction, and it was inside the valid time ranges */ 2093d739b42bSPekka Enberg log_blocks = kmalloc(get_desc_trans_len(desc) * 2094d739b42bSPekka Enberg sizeof(struct buffer_head *), GFP_NOFS); 2095d739b42bSPekka Enberg real_blocks = kmalloc(get_desc_trans_len(desc) * 2096d739b42bSPekka Enberg sizeof(struct buffer_head *), GFP_NOFS); 20971da177e4SLinus Torvalds if (!log_blocks || !real_blocks) { 20981da177e4SLinus Torvalds brelse(c_bh); 20991da177e4SLinus Torvalds brelse(d_bh); 2100d739b42bSPekka Enberg kfree(log_blocks); 2101d739b42bSPekka Enberg kfree(real_blocks); 2102bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 2103bd4c625cSLinus Torvalds "journal-1169: kmalloc failed, unable to mount FS"); 21041da177e4SLinus Torvalds return -1; 21051da177e4SLinus Torvalds } 21061da177e4SLinus Torvalds /* get all the buffer heads */ 21071da177e4SLinus Torvalds trans_half = journal_trans_half(p_s_sb->s_blocksize); 21081da177e4SLinus Torvalds for (i = 0; i < get_desc_trans_len(desc); i++) { 2109bd4c625cSLinus Torvalds log_blocks[i] = 2110bd4c625cSLinus Torvalds journal_getblk(p_s_sb, 2111bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + 2112bd4c625cSLinus Torvalds (trans_offset + 1 + 2113bd4c625cSLinus Torvalds i) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)); 21141da177e4SLinus Torvalds if (i < trans_half) { 2115bd4c625cSLinus Torvalds real_blocks[i] = 2116bd4c625cSLinus Torvalds sb_getblk(p_s_sb, 2117bd4c625cSLinus Torvalds le32_to_cpu(desc->j_realblock[i])); 21181da177e4SLinus Torvalds } else { 2119bd4c625cSLinus Torvalds real_blocks[i] = 2120bd4c625cSLinus Torvalds sb_getblk(p_s_sb, 2121bd4c625cSLinus Torvalds le32_to_cpu(commit-> 2122bd4c625cSLinus Torvalds j_realblock[i - trans_half])); 21231da177e4SLinus Torvalds } 21241da177e4SLinus Torvalds if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(p_s_sb)) { 2125bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 2126bd4c625cSLinus Torvalds "journal-1207: REPLAY FAILURE fsck required! Block to replay is outside of filesystem"); 21271da177e4SLinus Torvalds goto abort_replay; 21281da177e4SLinus Torvalds } 21291da177e4SLinus Torvalds /* make sure we don't try to replay onto log or reserved area */ 2130bd4c625cSLinus Torvalds if (is_block_in_log_or_reserved_area 2131bd4c625cSLinus Torvalds (p_s_sb, real_blocks[i]->b_blocknr)) { 2132bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 2133bd4c625cSLinus Torvalds "journal-1204: REPLAY FAILURE fsck required! Trying to replay onto a log block"); 21341da177e4SLinus Torvalds abort_replay: 21351da177e4SLinus Torvalds brelse_array(log_blocks, i); 21361da177e4SLinus Torvalds brelse_array(real_blocks, i); 21371da177e4SLinus Torvalds brelse(c_bh); 21381da177e4SLinus Torvalds brelse(d_bh); 2139d739b42bSPekka Enberg kfree(log_blocks); 2140d739b42bSPekka Enberg kfree(real_blocks); 21411da177e4SLinus Torvalds return -1; 21421da177e4SLinus Torvalds } 21431da177e4SLinus Torvalds } 21441da177e4SLinus Torvalds /* read in the log blocks, memcpy to the corresponding real block */ 21451da177e4SLinus Torvalds ll_rw_block(READ, get_desc_trans_len(desc), log_blocks); 21461da177e4SLinus Torvalds for (i = 0; i < get_desc_trans_len(desc); i++) { 21471da177e4SLinus Torvalds wait_on_buffer(log_blocks[i]); 21481da177e4SLinus Torvalds if (!buffer_uptodate(log_blocks[i])) { 2149bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 2150bd4c625cSLinus Torvalds "journal-1212: REPLAY FAILURE fsck required! buffer write failed"); 2151bd4c625cSLinus Torvalds brelse_array(log_blocks + i, 2152bd4c625cSLinus Torvalds get_desc_trans_len(desc) - i); 21531da177e4SLinus Torvalds brelse_array(real_blocks, get_desc_trans_len(desc)); 21541da177e4SLinus Torvalds brelse(c_bh); 21551da177e4SLinus Torvalds brelse(d_bh); 2156d739b42bSPekka Enberg kfree(log_blocks); 2157d739b42bSPekka Enberg kfree(real_blocks); 21581da177e4SLinus Torvalds return -1; 21591da177e4SLinus Torvalds } 2160bd4c625cSLinus Torvalds memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data, 2161bd4c625cSLinus Torvalds real_blocks[i]->b_size); 21621da177e4SLinus Torvalds set_buffer_uptodate(real_blocks[i]); 21631da177e4SLinus Torvalds brelse(log_blocks[i]); 21641da177e4SLinus Torvalds } 21651da177e4SLinus Torvalds /* flush out the real blocks */ 21661da177e4SLinus Torvalds for (i = 0; i < get_desc_trans_len(desc); i++) { 21671da177e4SLinus Torvalds set_buffer_dirty(real_blocks[i]); 216853778ffdSJan Kara ll_rw_block(SWRITE, 1, real_blocks + i); 21691da177e4SLinus Torvalds } 21701da177e4SLinus Torvalds for (i = 0; i < get_desc_trans_len(desc); i++) { 21711da177e4SLinus Torvalds wait_on_buffer(real_blocks[i]); 21721da177e4SLinus Torvalds if (!buffer_uptodate(real_blocks[i])) { 2173bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 2174bd4c625cSLinus Torvalds "journal-1226: REPLAY FAILURE, fsck required! buffer write failed"); 2175bd4c625cSLinus Torvalds brelse_array(real_blocks + i, 2176bd4c625cSLinus Torvalds get_desc_trans_len(desc) - i); 21771da177e4SLinus Torvalds brelse(c_bh); 21781da177e4SLinus Torvalds brelse(d_bh); 2179d739b42bSPekka Enberg kfree(log_blocks); 2180d739b42bSPekka Enberg kfree(real_blocks); 21811da177e4SLinus Torvalds return -1; 21821da177e4SLinus Torvalds } 21831da177e4SLinus Torvalds brelse(real_blocks[i]); 21841da177e4SLinus Torvalds } 2185bd4c625cSLinus Torvalds cur_dblock = 2186bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + 2187bd4c625cSLinus Torvalds ((trans_offset + get_desc_trans_len(desc) + 2188bd4c625cSLinus Torvalds 2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)); 2189bd4c625cSLinus Torvalds reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, 2190bd4c625cSLinus Torvalds "journal-1095: setting journal " "start to offset %ld", 21911da177e4SLinus Torvalds cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb)); 21921da177e4SLinus Torvalds 21931da177e4SLinus Torvalds /* init starting values for the first transaction, in case this is the last transaction to be replayed. */ 21941da177e4SLinus Torvalds journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb); 21951da177e4SLinus Torvalds journal->j_last_flush_trans_id = trans_id; 21961da177e4SLinus Torvalds journal->j_trans_id = trans_id + 1; 21971da177e4SLinus Torvalds brelse(c_bh); 21981da177e4SLinus Torvalds brelse(d_bh); 2199d739b42bSPekka Enberg kfree(log_blocks); 2200d739b42bSPekka Enberg kfree(real_blocks); 22011da177e4SLinus Torvalds return 0; 22021da177e4SLinus Torvalds } 22031da177e4SLinus Torvalds 22041da177e4SLinus Torvalds /* This function reads blocks starting from block and to max_block of bufsize 22051da177e4SLinus Torvalds size (but no more than BUFNR blocks at a time). This proved to improve 22061da177e4SLinus Torvalds mounting speed on self-rebuilding raid5 arrays at least. 22071da177e4SLinus Torvalds Right now it is only used from journal code. But later we might use it 22081da177e4SLinus Torvalds from other places. 22091da177e4SLinus Torvalds Note: Do not use journal_getblk/sb_getblk functions here! */ 2210bd4c625cSLinus Torvalds static struct buffer_head *reiserfs_breada(struct block_device *dev, int block, 2211bd4c625cSLinus Torvalds int bufsize, unsigned int max_block) 22121da177e4SLinus Torvalds { 22131da177e4SLinus Torvalds struct buffer_head *bhlist[BUFNR]; 22141da177e4SLinus Torvalds unsigned int blocks = BUFNR; 22151da177e4SLinus Torvalds struct buffer_head *bh; 22161da177e4SLinus Torvalds int i, j; 22171da177e4SLinus Torvalds 22181da177e4SLinus Torvalds bh = __getblk(dev, block, bufsize); 22191da177e4SLinus Torvalds if (buffer_uptodate(bh)) 22201da177e4SLinus Torvalds return (bh); 22211da177e4SLinus Torvalds 22221da177e4SLinus Torvalds if (block + BUFNR > max_block) { 22231da177e4SLinus Torvalds blocks = max_block - block; 22241da177e4SLinus Torvalds } 22251da177e4SLinus Torvalds bhlist[0] = bh; 22261da177e4SLinus Torvalds j = 1; 22271da177e4SLinus Torvalds for (i = 1; i < blocks; i++) { 22281da177e4SLinus Torvalds bh = __getblk(dev, block + i, bufsize); 22291da177e4SLinus Torvalds if (buffer_uptodate(bh)) { 22301da177e4SLinus Torvalds brelse(bh); 22311da177e4SLinus Torvalds break; 2232bd4c625cSLinus Torvalds } else 2233bd4c625cSLinus Torvalds bhlist[j++] = bh; 22341da177e4SLinus Torvalds } 22351da177e4SLinus Torvalds ll_rw_block(READ, j, bhlist); 22361da177e4SLinus Torvalds for (i = 1; i < j; i++) 22371da177e4SLinus Torvalds brelse(bhlist[i]); 22381da177e4SLinus Torvalds bh = bhlist[0]; 22391da177e4SLinus Torvalds wait_on_buffer(bh); 22401da177e4SLinus Torvalds if (buffer_uptodate(bh)) 22411da177e4SLinus Torvalds return bh; 22421da177e4SLinus Torvalds brelse(bh); 22431da177e4SLinus Torvalds return NULL; 22441da177e4SLinus Torvalds } 22451da177e4SLinus Torvalds 22461da177e4SLinus Torvalds /* 22471da177e4SLinus Torvalds ** read and replay the log 22481da177e4SLinus Torvalds ** on a clean unmount, the journal header's next unflushed pointer will be to an invalid 22491da177e4SLinus Torvalds ** transaction. This tests that before finding all the transactions in the log, which makes normal mount times fast. 22501da177e4SLinus Torvalds ** 22511da177e4SLinus Torvalds ** After a crash, this starts with the next unflushed transaction, and replays until it finds one too old, or invalid. 22521da177e4SLinus Torvalds ** 22531da177e4SLinus Torvalds ** On exit, it sets things up so the first transaction will work correctly. 22541da177e4SLinus Torvalds */ 2255bd4c625cSLinus Torvalds static int journal_read(struct super_block *p_s_sb) 2256bd4c625cSLinus Torvalds { 22571da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 22581da177e4SLinus Torvalds struct reiserfs_journal_desc *desc; 22591da177e4SLinus Torvalds unsigned long oldest_trans_id = 0; 22601da177e4SLinus Torvalds unsigned long oldest_invalid_trans_id = 0; 22611da177e4SLinus Torvalds time_t start; 22621da177e4SLinus Torvalds unsigned long oldest_start = 0; 22631da177e4SLinus Torvalds unsigned long cur_dblock = 0; 22641da177e4SLinus Torvalds unsigned long newest_mount_id = 9; 22651da177e4SLinus Torvalds struct buffer_head *d_bh; 22661da177e4SLinus Torvalds struct reiserfs_journal_header *jh; 22671da177e4SLinus Torvalds int valid_journal_header = 0; 22681da177e4SLinus Torvalds int replay_count = 0; 22691da177e4SLinus Torvalds int continue_replay = 1; 22701da177e4SLinus Torvalds int ret; 22711da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 22721da177e4SLinus Torvalds 22731da177e4SLinus Torvalds cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb); 22741da177e4SLinus Torvalds reiserfs_info(p_s_sb, "checking transaction log (%s)\n", 22751da177e4SLinus Torvalds bdevname(journal->j_dev_bd, b)); 22761da177e4SLinus Torvalds start = get_seconds(); 22771da177e4SLinus Torvalds 22781da177e4SLinus Torvalds /* step 1, read in the journal header block. Check the transaction it says 22791da177e4SLinus Torvalds ** is the first unflushed, and if that transaction is not valid, 22801da177e4SLinus Torvalds ** replay is done 22811da177e4SLinus Torvalds */ 22821da177e4SLinus Torvalds journal->j_header_bh = journal_bread(p_s_sb, 2283bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) 2284bd4c625cSLinus Torvalds + SB_ONDISK_JOURNAL_SIZE(p_s_sb)); 22851da177e4SLinus Torvalds if (!journal->j_header_bh) { 22861da177e4SLinus Torvalds return 1; 22871da177e4SLinus Torvalds } 22881da177e4SLinus Torvalds jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data); 22891da177e4SLinus Torvalds if (le32_to_cpu(jh->j_first_unflushed_offset) >= 0 && 2290bd4c625cSLinus Torvalds le32_to_cpu(jh->j_first_unflushed_offset) < 2291bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_SIZE(p_s_sb) 2292bd4c625cSLinus Torvalds && le32_to_cpu(jh->j_last_flush_trans_id) > 0) { 2293bd4c625cSLinus Torvalds oldest_start = 2294bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + 22951da177e4SLinus Torvalds le32_to_cpu(jh->j_first_unflushed_offset); 22961da177e4SLinus Torvalds oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1; 22971da177e4SLinus Torvalds newest_mount_id = le32_to_cpu(jh->j_mount_id); 2298bd4c625cSLinus Torvalds reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, 2299bd4c625cSLinus Torvalds "journal-1153: found in " 23001da177e4SLinus Torvalds "header: first_unflushed_offset %d, last_flushed_trans_id " 23011da177e4SLinus Torvalds "%lu", le32_to_cpu(jh->j_first_unflushed_offset), 23021da177e4SLinus Torvalds le32_to_cpu(jh->j_last_flush_trans_id)); 23031da177e4SLinus Torvalds valid_journal_header = 1; 23041da177e4SLinus Torvalds 23051da177e4SLinus Torvalds /* now, we try to read the first unflushed offset. If it is not valid, 23061da177e4SLinus Torvalds ** there is nothing more we can do, and it makes no sense to read 23071da177e4SLinus Torvalds ** through the whole log. 23081da177e4SLinus Torvalds */ 2309bd4c625cSLinus Torvalds d_bh = 2310bd4c625cSLinus Torvalds journal_bread(p_s_sb, 2311bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + 2312bd4c625cSLinus Torvalds le32_to_cpu(jh->j_first_unflushed_offset)); 23131da177e4SLinus Torvalds ret = journal_transaction_is_valid(p_s_sb, d_bh, NULL, NULL); 23141da177e4SLinus Torvalds if (!ret) { 23151da177e4SLinus Torvalds continue_replay = 0; 23161da177e4SLinus Torvalds } 23171da177e4SLinus Torvalds brelse(d_bh); 23181da177e4SLinus Torvalds goto start_log_replay; 23191da177e4SLinus Torvalds } 23201da177e4SLinus Torvalds 23211da177e4SLinus Torvalds if (continue_replay && bdev_read_only(p_s_sb->s_bdev)) { 23221da177e4SLinus Torvalds reiserfs_warning(p_s_sb, 23231da177e4SLinus Torvalds "clm-2076: device is readonly, unable to replay log"); 23241da177e4SLinus Torvalds return -1; 23251da177e4SLinus Torvalds } 23261da177e4SLinus Torvalds 23271da177e4SLinus Torvalds /* ok, there are transactions that need to be replayed. start with the first log block, find 23281da177e4SLinus Torvalds ** all the valid transactions, and pick out the oldest. 23291da177e4SLinus Torvalds */ 2330bd4c625cSLinus Torvalds while (continue_replay 2331bd4c625cSLinus Torvalds && cur_dblock < 2332bd4c625cSLinus Torvalds (SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + 2333bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_SIZE(p_s_sb))) { 23341da177e4SLinus Torvalds /* Note that it is required for blocksize of primary fs device and journal 23351da177e4SLinus Torvalds device to be the same */ 2336bd4c625cSLinus Torvalds d_bh = 2337bd4c625cSLinus Torvalds reiserfs_breada(journal->j_dev_bd, cur_dblock, 2338bd4c625cSLinus Torvalds p_s_sb->s_blocksize, 2339bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + 2340bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_SIZE(p_s_sb)); 2341bd4c625cSLinus Torvalds ret = 2342bd4c625cSLinus Torvalds journal_transaction_is_valid(p_s_sb, d_bh, 2343bd4c625cSLinus Torvalds &oldest_invalid_trans_id, 2344bd4c625cSLinus Torvalds &newest_mount_id); 23451da177e4SLinus Torvalds if (ret == 1) { 23461da177e4SLinus Torvalds desc = (struct reiserfs_journal_desc *)d_bh->b_data; 23471da177e4SLinus Torvalds if (oldest_start == 0) { /* init all oldest_ values */ 23481da177e4SLinus Torvalds oldest_trans_id = get_desc_trans_id(desc); 23491da177e4SLinus Torvalds oldest_start = d_bh->b_blocknr; 23501da177e4SLinus Torvalds newest_mount_id = get_desc_mount_id(desc); 2351bd4c625cSLinus Torvalds reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, 2352bd4c625cSLinus Torvalds "journal-1179: Setting " 23531da177e4SLinus Torvalds "oldest_start to offset %llu, trans_id %lu", 2354bd4c625cSLinus Torvalds oldest_start - 2355bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK 2356bd4c625cSLinus Torvalds (p_s_sb), oldest_trans_id); 23571da177e4SLinus Torvalds } else if (oldest_trans_id > get_desc_trans_id(desc)) { 23581da177e4SLinus Torvalds /* one we just read was older */ 23591da177e4SLinus Torvalds oldest_trans_id = get_desc_trans_id(desc); 23601da177e4SLinus Torvalds oldest_start = d_bh->b_blocknr; 2361bd4c625cSLinus Torvalds reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, 2362bd4c625cSLinus Torvalds "journal-1180: Resetting " 23631da177e4SLinus Torvalds "oldest_start to offset %lu, trans_id %lu", 2364bd4c625cSLinus Torvalds oldest_start - 2365bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK 2366bd4c625cSLinus Torvalds (p_s_sb), oldest_trans_id); 23671da177e4SLinus Torvalds } 23681da177e4SLinus Torvalds if (newest_mount_id < get_desc_mount_id(desc)) { 23691da177e4SLinus Torvalds newest_mount_id = get_desc_mount_id(desc); 2370bd4c625cSLinus Torvalds reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, 2371bd4c625cSLinus Torvalds "journal-1299: Setting " 2372bd4c625cSLinus Torvalds "newest_mount_id to %d", 2373bd4c625cSLinus Torvalds get_desc_mount_id(desc)); 23741da177e4SLinus Torvalds } 23751da177e4SLinus Torvalds cur_dblock += get_desc_trans_len(desc) + 2; 23761da177e4SLinus Torvalds } else { 23771da177e4SLinus Torvalds cur_dblock++; 23781da177e4SLinus Torvalds } 23791da177e4SLinus Torvalds brelse(d_bh); 23801da177e4SLinus Torvalds } 23811da177e4SLinus Torvalds 23821da177e4SLinus Torvalds start_log_replay: 23831da177e4SLinus Torvalds cur_dblock = oldest_start; 23841da177e4SLinus Torvalds if (oldest_trans_id) { 2385bd4c625cSLinus Torvalds reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, 2386bd4c625cSLinus Torvalds "journal-1206: Starting replay " 23871da177e4SLinus Torvalds "from offset %llu, trans_id %lu", 23881da177e4SLinus Torvalds cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb), 23891da177e4SLinus Torvalds oldest_trans_id); 23901da177e4SLinus Torvalds 23911da177e4SLinus Torvalds } 23921da177e4SLinus Torvalds replay_count = 0; 23931da177e4SLinus Torvalds while (continue_replay && oldest_trans_id > 0) { 2394bd4c625cSLinus Torvalds ret = 2395bd4c625cSLinus Torvalds journal_read_transaction(p_s_sb, cur_dblock, oldest_start, 2396bd4c625cSLinus Torvalds oldest_trans_id, newest_mount_id); 23971da177e4SLinus Torvalds if (ret < 0) { 23981da177e4SLinus Torvalds return ret; 23991da177e4SLinus Torvalds } else if (ret != 0) { 24001da177e4SLinus Torvalds break; 24011da177e4SLinus Torvalds } 2402bd4c625cSLinus Torvalds cur_dblock = 2403bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + journal->j_start; 24041da177e4SLinus Torvalds replay_count++; 24051da177e4SLinus Torvalds if (cur_dblock == oldest_start) 24061da177e4SLinus Torvalds break; 24071da177e4SLinus Torvalds } 24081da177e4SLinus Torvalds 24091da177e4SLinus Torvalds if (oldest_trans_id == 0) { 2410bd4c625cSLinus Torvalds reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, 2411bd4c625cSLinus Torvalds "journal-1225: No valid " "transactions found"); 24121da177e4SLinus Torvalds } 24131da177e4SLinus Torvalds /* j_start does not get set correctly if we don't replay any transactions. 24141da177e4SLinus Torvalds ** if we had a valid journal_header, set j_start to the first unflushed transaction value, 24151da177e4SLinus Torvalds ** copy the trans_id from the header 24161da177e4SLinus Torvalds */ 24171da177e4SLinus Torvalds if (valid_journal_header && replay_count == 0) { 24181da177e4SLinus Torvalds journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset); 2419bd4c625cSLinus Torvalds journal->j_trans_id = 2420bd4c625cSLinus Torvalds le32_to_cpu(jh->j_last_flush_trans_id) + 1; 2421bd4c625cSLinus Torvalds journal->j_last_flush_trans_id = 2422bd4c625cSLinus Torvalds le32_to_cpu(jh->j_last_flush_trans_id); 24231da177e4SLinus Torvalds journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1; 24241da177e4SLinus Torvalds } else { 24251da177e4SLinus Torvalds journal->j_mount_id = newest_mount_id + 1; 24261da177e4SLinus Torvalds } 24271da177e4SLinus Torvalds reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1299: Setting " 24281da177e4SLinus Torvalds "newest_mount_id to %lu", journal->j_mount_id); 24291da177e4SLinus Torvalds journal->j_first_unflushed_offset = journal->j_start; 24301da177e4SLinus Torvalds if (replay_count > 0) { 2431bd4c625cSLinus Torvalds reiserfs_info(p_s_sb, 2432bd4c625cSLinus Torvalds "replayed %d transactions in %lu seconds\n", 24331da177e4SLinus Torvalds replay_count, get_seconds() - start); 24341da177e4SLinus Torvalds } 24351da177e4SLinus Torvalds if (!bdev_read_only(p_s_sb->s_bdev) && 24361da177e4SLinus Torvalds _update_journal_header_block(p_s_sb, journal->j_start, 2437bd4c625cSLinus Torvalds journal->j_last_flush_trans_id)) { 24381da177e4SLinus Torvalds /* replay failed, caller must call free_journal_ram and abort 24391da177e4SLinus Torvalds ** the mount 24401da177e4SLinus Torvalds */ 24411da177e4SLinus Torvalds return -1; 24421da177e4SLinus Torvalds } 24431da177e4SLinus Torvalds return 0; 24441da177e4SLinus Torvalds } 24451da177e4SLinus Torvalds 24461da177e4SLinus Torvalds static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s) 24471da177e4SLinus Torvalds { 24481da177e4SLinus Torvalds struct reiserfs_journal_list *jl; 24491da177e4SLinus Torvalds retry: 2450d739b42bSPekka Enberg jl = kzalloc(sizeof(struct reiserfs_journal_list), GFP_NOFS); 24511da177e4SLinus Torvalds if (!jl) { 24521da177e4SLinus Torvalds yield(); 24531da177e4SLinus Torvalds goto retry; 24541da177e4SLinus Torvalds } 24551da177e4SLinus Torvalds INIT_LIST_HEAD(&jl->j_list); 24561da177e4SLinus Torvalds INIT_LIST_HEAD(&jl->j_working_list); 24571da177e4SLinus Torvalds INIT_LIST_HEAD(&jl->j_tail_bh_list); 24581da177e4SLinus Torvalds INIT_LIST_HEAD(&jl->j_bh_list); 24591da177e4SLinus Torvalds sema_init(&jl->j_commit_lock, 1); 24601da177e4SLinus Torvalds SB_JOURNAL(s)->j_num_lists++; 24611da177e4SLinus Torvalds get_journal_list(jl); 24621da177e4SLinus Torvalds return jl; 24631da177e4SLinus Torvalds } 24641da177e4SLinus Torvalds 2465bd4c625cSLinus Torvalds static void journal_list_init(struct super_block *p_s_sb) 2466bd4c625cSLinus Torvalds { 24671da177e4SLinus Torvalds SB_JOURNAL(p_s_sb)->j_current_jl = alloc_journal_list(p_s_sb); 24681da177e4SLinus Torvalds } 24691da177e4SLinus Torvalds 24701da177e4SLinus Torvalds static int release_journal_dev(struct super_block *super, 24711da177e4SLinus Torvalds struct reiserfs_journal *journal) 24721da177e4SLinus Torvalds { 24731da177e4SLinus Torvalds int result; 24741da177e4SLinus Torvalds 24751da177e4SLinus Torvalds result = 0; 24761da177e4SLinus Torvalds 24771da177e4SLinus Torvalds if (journal->j_dev_file != NULL) { 24781da177e4SLinus Torvalds result = filp_close(journal->j_dev_file, NULL); 24791da177e4SLinus Torvalds journal->j_dev_file = NULL; 24801da177e4SLinus Torvalds journal->j_dev_bd = NULL; 24811da177e4SLinus Torvalds } else if (journal->j_dev_bd != NULL) { 24821da177e4SLinus Torvalds result = blkdev_put(journal->j_dev_bd); 24831da177e4SLinus Torvalds journal->j_dev_bd = NULL; 24841da177e4SLinus Torvalds } 24851da177e4SLinus Torvalds 24861da177e4SLinus Torvalds if (result != 0) { 2487bd4c625cSLinus Torvalds reiserfs_warning(super, 2488bd4c625cSLinus Torvalds "sh-457: release_journal_dev: Cannot release journal device: %i", 2489bd4c625cSLinus Torvalds result); 24901da177e4SLinus Torvalds } 24911da177e4SLinus Torvalds return result; 24921da177e4SLinus Torvalds } 24931da177e4SLinus Torvalds 24941da177e4SLinus Torvalds static int journal_init_dev(struct super_block *super, 24951da177e4SLinus Torvalds struct reiserfs_journal *journal, 24961da177e4SLinus Torvalds const char *jdev_name) 24971da177e4SLinus Torvalds { 24981da177e4SLinus Torvalds int result; 24991da177e4SLinus Torvalds dev_t jdev; 25001da177e4SLinus Torvalds int blkdev_mode = FMODE_READ | FMODE_WRITE; 25011da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 25021da177e4SLinus Torvalds 25031da177e4SLinus Torvalds result = 0; 25041da177e4SLinus Torvalds 25051da177e4SLinus Torvalds journal->j_dev_bd = NULL; 25061da177e4SLinus Torvalds journal->j_dev_file = NULL; 25071da177e4SLinus Torvalds jdev = SB_ONDISK_JOURNAL_DEVICE(super) ? 25081da177e4SLinus Torvalds new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev; 25091da177e4SLinus Torvalds 25101da177e4SLinus Torvalds if (bdev_read_only(super->s_bdev)) 25111da177e4SLinus Torvalds blkdev_mode = FMODE_READ; 25121da177e4SLinus Torvalds 25131da177e4SLinus Torvalds /* there is no "jdev" option and journal is on separate device */ 25141da177e4SLinus Torvalds if ((!jdev_name || !jdev_name[0])) { 25151da177e4SLinus Torvalds journal->j_dev_bd = open_by_devnum(jdev, blkdev_mode); 25161da177e4SLinus Torvalds if (IS_ERR(journal->j_dev_bd)) { 25171da177e4SLinus Torvalds result = PTR_ERR(journal->j_dev_bd); 25181da177e4SLinus Torvalds journal->j_dev_bd = NULL; 25191da177e4SLinus Torvalds reiserfs_warning(super, "sh-458: journal_init_dev: " 25201da177e4SLinus Torvalds "cannot init journal device '%s': %i", 25211da177e4SLinus Torvalds __bdevname(jdev, b), result); 25221da177e4SLinus Torvalds return result; 25231da177e4SLinus Torvalds } else if (jdev != super->s_dev) 25241da177e4SLinus Torvalds set_blocksize(journal->j_dev_bd, super->s_blocksize); 25251da177e4SLinus Torvalds return 0; 25261da177e4SLinus Torvalds } 25271da177e4SLinus Torvalds 25281da177e4SLinus Torvalds journal->j_dev_file = filp_open(jdev_name, 0, 0); 25291da177e4SLinus Torvalds if (!IS_ERR(journal->j_dev_file)) { 25301da177e4SLinus Torvalds struct inode *jdev_inode = journal->j_dev_file->f_mapping->host; 25311da177e4SLinus Torvalds if (!S_ISBLK(jdev_inode->i_mode)) { 25321da177e4SLinus Torvalds reiserfs_warning(super, "journal_init_dev: '%s' is " 25331da177e4SLinus Torvalds "not a block device", jdev_name); 25341da177e4SLinus Torvalds result = -ENOTBLK; 253574f9f974SEdward Shishkin release_journal_dev(super, journal); 25361da177e4SLinus Torvalds } else { 25371da177e4SLinus Torvalds /* ok */ 25381da177e4SLinus Torvalds journal->j_dev_bd = I_BDEV(jdev_inode); 25391da177e4SLinus Torvalds set_blocksize(journal->j_dev_bd, super->s_blocksize); 2540bd4c625cSLinus Torvalds reiserfs_info(super, 2541bd4c625cSLinus Torvalds "journal_init_dev: journal device: %s\n", 254274f9f974SEdward Shishkin bdevname(journal->j_dev_bd, b)); 25431da177e4SLinus Torvalds } 25441da177e4SLinus Torvalds } else { 25451da177e4SLinus Torvalds result = PTR_ERR(journal->j_dev_file); 25461da177e4SLinus Torvalds journal->j_dev_file = NULL; 25471da177e4SLinus Torvalds reiserfs_warning(super, 25481da177e4SLinus Torvalds "journal_init_dev: Cannot open '%s': %i", 25491da177e4SLinus Torvalds jdev_name, result); 25501da177e4SLinus Torvalds } 25511da177e4SLinus Torvalds return result; 25521da177e4SLinus Torvalds } 25531da177e4SLinus Torvalds 25541da177e4SLinus Torvalds /* 25551da177e4SLinus Torvalds ** must be called once on fs mount. calls journal_read for you 25561da177e4SLinus Torvalds */ 2557bd4c625cSLinus Torvalds int journal_init(struct super_block *p_s_sb, const char *j_dev_name, 2558bd4c625cSLinus Torvalds int old_format, unsigned int commit_max_age) 2559bd4c625cSLinus Torvalds { 25601da177e4SLinus Torvalds int num_cnodes = SB_ONDISK_JOURNAL_SIZE(p_s_sb) * 2; 25611da177e4SLinus Torvalds struct buffer_head *bhjh; 25621da177e4SLinus Torvalds struct reiserfs_super_block *rs; 25631da177e4SLinus Torvalds struct reiserfs_journal_header *jh; 25641da177e4SLinus Torvalds struct reiserfs_journal *journal; 25651da177e4SLinus Torvalds struct reiserfs_journal_list *jl; 25661da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 25671da177e4SLinus Torvalds 25681da177e4SLinus Torvalds journal = SB_JOURNAL(p_s_sb) = vmalloc(sizeof(struct reiserfs_journal)); 25691da177e4SLinus Torvalds if (!journal) { 2570bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 2571bd4c625cSLinus Torvalds "journal-1256: unable to get memory for journal structure"); 25721da177e4SLinus Torvalds return 1; 25731da177e4SLinus Torvalds } 25741da177e4SLinus Torvalds memset(journal, 0, sizeof(struct reiserfs_journal)); 25751da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_bitmap_nodes); 25761da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_prealloc_list); 25771da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_working_list); 25781da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_journal_list); 25791da177e4SLinus Torvalds journal->j_persistent_trans = 0; 25801da177e4SLinus Torvalds if (reiserfs_allocate_list_bitmaps(p_s_sb, 25811da177e4SLinus Torvalds journal->j_list_bitmap, 25821da177e4SLinus Torvalds SB_BMAP_NR(p_s_sb))) 25831da177e4SLinus Torvalds goto free_and_return; 25841da177e4SLinus Torvalds allocate_bitmap_nodes(p_s_sb); 25851da177e4SLinus Torvalds 25861da177e4SLinus Torvalds /* reserved for journal area support */ 25871da177e4SLinus Torvalds SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) = (old_format ? 2588bd4c625cSLinus Torvalds REISERFS_OLD_DISK_OFFSET_IN_BYTES 2589bd4c625cSLinus Torvalds / p_s_sb->s_blocksize + 2590bd4c625cSLinus Torvalds SB_BMAP_NR(p_s_sb) + 2591bd4c625cSLinus Torvalds 1 : 2592bd4c625cSLinus Torvalds REISERFS_DISK_OFFSET_IN_BYTES / 2593bd4c625cSLinus Torvalds p_s_sb->s_blocksize + 2); 25941da177e4SLinus Torvalds 25951da177e4SLinus Torvalds /* Sanity check to see is the standard journal fitting withing first bitmap 25961da177e4SLinus Torvalds (actual for small blocksizes) */ 25971da177e4SLinus Torvalds if (!SB_ONDISK_JOURNAL_DEVICE(p_s_sb) && 2598bd4c625cSLinus Torvalds (SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) + 2599bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_SIZE(p_s_sb) > p_s_sb->s_blocksize * 8)) { 2600bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 2601bd4c625cSLinus Torvalds "journal-1393: journal does not fit for area " 26021da177e4SLinus Torvalds "addressed by first of bitmap blocks. It starts at " 26031da177e4SLinus Torvalds "%u and its size is %u. Block size %ld", 26041da177e4SLinus Torvalds SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb), 2605bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_SIZE(p_s_sb), 2606bd4c625cSLinus Torvalds p_s_sb->s_blocksize); 26071da177e4SLinus Torvalds goto free_and_return; 26081da177e4SLinus Torvalds } 26091da177e4SLinus Torvalds 26101da177e4SLinus Torvalds if (journal_init_dev(p_s_sb, journal, j_dev_name) != 0) { 2611bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 2612bd4c625cSLinus Torvalds "sh-462: unable to initialize jornal device"); 26131da177e4SLinus Torvalds goto free_and_return; 26141da177e4SLinus Torvalds } 26151da177e4SLinus Torvalds 26161da177e4SLinus Torvalds rs = SB_DISK_SUPER_BLOCK(p_s_sb); 26171da177e4SLinus Torvalds 26181da177e4SLinus Torvalds /* read journal header */ 26191da177e4SLinus Torvalds bhjh = journal_bread(p_s_sb, 2620bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + 2621bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_SIZE(p_s_sb)); 26221da177e4SLinus Torvalds if (!bhjh) { 2623bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 2624bd4c625cSLinus Torvalds "sh-459: unable to read journal header"); 26251da177e4SLinus Torvalds goto free_and_return; 26261da177e4SLinus Torvalds } 26271da177e4SLinus Torvalds jh = (struct reiserfs_journal_header *)(bhjh->b_data); 26281da177e4SLinus Torvalds 26291da177e4SLinus Torvalds /* make sure that journal matches to the super block */ 2630bd4c625cSLinus Torvalds if (is_reiserfs_jr(rs) 2631bd4c625cSLinus Torvalds && (le32_to_cpu(jh->jh_journal.jp_journal_magic) != 2632bd4c625cSLinus Torvalds sb_jp_journal_magic(rs))) { 2633bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 2634bd4c625cSLinus Torvalds "sh-460: journal header magic %x " 26351da177e4SLinus Torvalds "(device %s) does not match to magic found in super " 2636bd4c625cSLinus Torvalds "block %x", jh->jh_journal.jp_journal_magic, 26371da177e4SLinus Torvalds bdevname(journal->j_dev_bd, b), 26381da177e4SLinus Torvalds sb_jp_journal_magic(rs)); 26391da177e4SLinus Torvalds brelse(bhjh); 26401da177e4SLinus Torvalds goto free_and_return; 26411da177e4SLinus Torvalds } 26421da177e4SLinus Torvalds 26431da177e4SLinus Torvalds journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max); 26441da177e4SLinus Torvalds journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch); 2645bd4c625cSLinus Torvalds journal->j_max_commit_age = 2646bd4c625cSLinus Torvalds le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age); 26471da177e4SLinus Torvalds journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE; 26481da177e4SLinus Torvalds 26491da177e4SLinus Torvalds if (journal->j_trans_max) { 26501da177e4SLinus Torvalds /* make sure these parameters are available, assign it if they are not */ 26511da177e4SLinus Torvalds __u32 initial = journal->j_trans_max; 26521da177e4SLinus Torvalds __u32 ratio = 1; 26531da177e4SLinus Torvalds 26541da177e4SLinus Torvalds if (p_s_sb->s_blocksize < 4096) 26551da177e4SLinus Torvalds ratio = 4096 / p_s_sb->s_blocksize; 26561da177e4SLinus Torvalds 2657bd4c625cSLinus Torvalds if (SB_ONDISK_JOURNAL_SIZE(p_s_sb) / journal->j_trans_max < 2658bd4c625cSLinus Torvalds JOURNAL_MIN_RATIO) 2659bd4c625cSLinus Torvalds journal->j_trans_max = 2660bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_SIZE(p_s_sb) / JOURNAL_MIN_RATIO; 26611da177e4SLinus Torvalds if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio) 2662bd4c625cSLinus Torvalds journal->j_trans_max = 2663bd4c625cSLinus Torvalds JOURNAL_TRANS_MAX_DEFAULT / ratio; 26641da177e4SLinus Torvalds if (journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio) 2665bd4c625cSLinus Torvalds journal->j_trans_max = 2666bd4c625cSLinus Torvalds JOURNAL_TRANS_MIN_DEFAULT / ratio; 26671da177e4SLinus Torvalds 26681da177e4SLinus Torvalds if (journal->j_trans_max != initial) 2669bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 2670bd4c625cSLinus Torvalds "sh-461: journal_init: wrong transaction max size (%u). Changed to %u", 26711da177e4SLinus Torvalds initial, journal->j_trans_max); 26721da177e4SLinus Torvalds 26731da177e4SLinus Torvalds journal->j_max_batch = journal->j_trans_max * 26741da177e4SLinus Torvalds JOURNAL_MAX_BATCH_DEFAULT / JOURNAL_TRANS_MAX_DEFAULT; 26751da177e4SLinus Torvalds } 26761da177e4SLinus Torvalds 26771da177e4SLinus Torvalds if (!journal->j_trans_max) { 26781da177e4SLinus Torvalds /*we have the file system was created by old version of mkreiserfs 26791da177e4SLinus Torvalds so this field contains zero value */ 26801da177e4SLinus Torvalds journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT; 26811da177e4SLinus Torvalds journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT; 26821da177e4SLinus Torvalds journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE; 26831da177e4SLinus Torvalds 26841da177e4SLinus Torvalds /* for blocksize >= 4096 - max transaction size is 1024. For block size < 4096 26851da177e4SLinus Torvalds trans max size is decreased proportionally */ 26861da177e4SLinus Torvalds if (p_s_sb->s_blocksize < 4096) { 26871da177e4SLinus Torvalds journal->j_trans_max /= (4096 / p_s_sb->s_blocksize); 26881da177e4SLinus Torvalds journal->j_max_batch = (journal->j_trans_max) * 9 / 10; 26891da177e4SLinus Torvalds } 26901da177e4SLinus Torvalds } 26911da177e4SLinus Torvalds 26921da177e4SLinus Torvalds journal->j_default_max_commit_age = journal->j_max_commit_age; 26931da177e4SLinus Torvalds 26941da177e4SLinus Torvalds if (commit_max_age != 0) { 26951da177e4SLinus Torvalds journal->j_max_commit_age = commit_max_age; 26961da177e4SLinus Torvalds journal->j_max_trans_age = commit_max_age; 26971da177e4SLinus Torvalds } 26981da177e4SLinus Torvalds 26991da177e4SLinus Torvalds reiserfs_info(p_s_sb, "journal params: device %s, size %u, " 27001da177e4SLinus Torvalds "journal first block %u, max trans len %u, max batch %u, " 27011da177e4SLinus Torvalds "max commit age %u, max trans age %u\n", 27021da177e4SLinus Torvalds bdevname(journal->j_dev_bd, b), 27031da177e4SLinus Torvalds SB_ONDISK_JOURNAL_SIZE(p_s_sb), 27041da177e4SLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb), 27051da177e4SLinus Torvalds journal->j_trans_max, 27061da177e4SLinus Torvalds journal->j_max_batch, 2707bd4c625cSLinus Torvalds journal->j_max_commit_age, journal->j_max_trans_age); 27081da177e4SLinus Torvalds 27091da177e4SLinus Torvalds brelse(bhjh); 27101da177e4SLinus Torvalds 27111da177e4SLinus Torvalds journal->j_list_bitmap_index = 0; 27121da177e4SLinus Torvalds journal_list_init(p_s_sb); 27131da177e4SLinus Torvalds 2714bd4c625cSLinus Torvalds memset(journal->j_list_hash_table, 0, 2715bd4c625cSLinus Torvalds JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *)); 27161da177e4SLinus Torvalds 27171da177e4SLinus Torvalds INIT_LIST_HEAD(&journal->j_dirty_buffers); 27181da177e4SLinus Torvalds spin_lock_init(&journal->j_dirty_buffers_lock); 27191da177e4SLinus Torvalds 27201da177e4SLinus Torvalds journal->j_start = 0; 27211da177e4SLinus Torvalds journal->j_len = 0; 27221da177e4SLinus Torvalds journal->j_len_alloc = 0; 27231da177e4SLinus Torvalds atomic_set(&(journal->j_wcount), 0); 27241da177e4SLinus Torvalds atomic_set(&(journal->j_async_throttle), 0); 27251da177e4SLinus Torvalds journal->j_bcount = 0; 27261da177e4SLinus Torvalds journal->j_trans_start_time = 0; 27271da177e4SLinus Torvalds journal->j_last = NULL; 27281da177e4SLinus Torvalds journal->j_first = NULL; 27291da177e4SLinus Torvalds init_waitqueue_head(&(journal->j_join_wait)); 27301da177e4SLinus Torvalds sema_init(&journal->j_lock, 1); 27311da177e4SLinus Torvalds sema_init(&journal->j_flush_sem, 1); 27321da177e4SLinus Torvalds 27331da177e4SLinus Torvalds journal->j_trans_id = 10; 27341da177e4SLinus Torvalds journal->j_mount_id = 10; 27351da177e4SLinus Torvalds journal->j_state = 0; 27361da177e4SLinus Torvalds atomic_set(&(journal->j_jlock), 0); 27371da177e4SLinus Torvalds journal->j_cnode_free_list = allocate_cnodes(num_cnodes); 27381da177e4SLinus Torvalds journal->j_cnode_free_orig = journal->j_cnode_free_list; 27391da177e4SLinus Torvalds journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0; 27401da177e4SLinus Torvalds journal->j_cnode_used = 0; 27411da177e4SLinus Torvalds journal->j_must_wait = 0; 27421da177e4SLinus Torvalds 2743576f6d79SJeff Mahoney if (journal->j_cnode_free == 0) { 2744576f6d79SJeff Mahoney reiserfs_warning(p_s_sb, "journal-2004: Journal cnode memory " 2745576f6d79SJeff Mahoney "allocation failed (%ld bytes). Journal is " 2746576f6d79SJeff Mahoney "too large for available memory. Usually " 2747576f6d79SJeff Mahoney "this is due to a journal that is too large.", 2748576f6d79SJeff Mahoney sizeof (struct reiserfs_journal_cnode) * num_cnodes); 2749576f6d79SJeff Mahoney goto free_and_return; 2750576f6d79SJeff Mahoney } 2751576f6d79SJeff Mahoney 27521da177e4SLinus Torvalds init_journal_hash(p_s_sb); 27531da177e4SLinus Torvalds jl = journal->j_current_jl; 27541da177e4SLinus Torvalds jl->j_list_bitmap = get_list_bitmap(p_s_sb, jl); 27551da177e4SLinus Torvalds if (!jl->j_list_bitmap) { 2756bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 2757bd4c625cSLinus Torvalds "journal-2005, get_list_bitmap failed for journal list 0"); 27581da177e4SLinus Torvalds goto free_and_return; 27591da177e4SLinus Torvalds } 27601da177e4SLinus Torvalds if (journal_read(p_s_sb) < 0) { 27611da177e4SLinus Torvalds reiserfs_warning(p_s_sb, "Replay Failure, unable to mount"); 27621da177e4SLinus Torvalds goto free_and_return; 27631da177e4SLinus Torvalds } 27641da177e4SLinus Torvalds 27651da177e4SLinus Torvalds reiserfs_mounted_fs_count++; 27661da177e4SLinus Torvalds if (reiserfs_mounted_fs_count <= 1) 27671da177e4SLinus Torvalds commit_wq = create_workqueue("reiserfs"); 27681da177e4SLinus Torvalds 27691da177e4SLinus Torvalds INIT_WORK(&journal->j_work, flush_async_commits, p_s_sb); 27701da177e4SLinus Torvalds return 0; 27711da177e4SLinus Torvalds free_and_return: 27721da177e4SLinus Torvalds free_journal_ram(p_s_sb); 27731da177e4SLinus Torvalds return 1; 27741da177e4SLinus Torvalds } 27751da177e4SLinus Torvalds 27761da177e4SLinus Torvalds /* 27771da177e4SLinus Torvalds ** test for a polite end of the current transaction. Used by file_write, and should 27781da177e4SLinus Torvalds ** be used by delete to make sure they don't write more than can fit inside a single 27791da177e4SLinus Torvalds ** transaction 27801da177e4SLinus Torvalds */ 2781bd4c625cSLinus Torvalds int journal_transaction_should_end(struct reiserfs_transaction_handle *th, 2782bd4c625cSLinus Torvalds int new_alloc) 2783bd4c625cSLinus Torvalds { 27841da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(th->t_super); 27851da177e4SLinus Torvalds time_t now = get_seconds(); 27861da177e4SLinus Torvalds /* cannot restart while nested */ 27871da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 27881da177e4SLinus Torvalds if (th->t_refcount > 1) 27891da177e4SLinus Torvalds return 0; 27901da177e4SLinus Torvalds if (journal->j_must_wait > 0 || 27911da177e4SLinus Torvalds (journal->j_len_alloc + new_alloc) >= journal->j_max_batch || 27921da177e4SLinus Torvalds atomic_read(&(journal->j_jlock)) || 27931da177e4SLinus Torvalds (now - journal->j_trans_start_time) > journal->j_max_trans_age || 27941da177e4SLinus Torvalds journal->j_cnode_free < (journal->j_trans_max * 3)) { 27951da177e4SLinus Torvalds return 1; 27961da177e4SLinus Torvalds } 27971da177e4SLinus Torvalds return 0; 27981da177e4SLinus Torvalds } 27991da177e4SLinus Torvalds 28001da177e4SLinus Torvalds /* this must be called inside a transaction, and requires the 28011da177e4SLinus Torvalds ** kernel_lock to be held 28021da177e4SLinus Torvalds */ 2803bd4c625cSLinus Torvalds void reiserfs_block_writes(struct reiserfs_transaction_handle *th) 2804bd4c625cSLinus Torvalds { 28051da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(th->t_super); 28061da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 28071da177e4SLinus Torvalds journal->j_must_wait = 1; 28081da177e4SLinus Torvalds set_bit(J_WRITERS_BLOCKED, &journal->j_state); 28091da177e4SLinus Torvalds return; 28101da177e4SLinus Torvalds } 28111da177e4SLinus Torvalds 28121da177e4SLinus Torvalds /* this must be called without a transaction started, and does not 28131da177e4SLinus Torvalds ** require BKL 28141da177e4SLinus Torvalds */ 2815bd4c625cSLinus Torvalds void reiserfs_allow_writes(struct super_block *s) 2816bd4c625cSLinus Torvalds { 28171da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 28181da177e4SLinus Torvalds clear_bit(J_WRITERS_BLOCKED, &journal->j_state); 28191da177e4SLinus Torvalds wake_up(&journal->j_join_wait); 28201da177e4SLinus Torvalds } 28211da177e4SLinus Torvalds 28221da177e4SLinus Torvalds /* this must be called without a transaction started, and does not 28231da177e4SLinus Torvalds ** require BKL 28241da177e4SLinus Torvalds */ 2825bd4c625cSLinus Torvalds void reiserfs_wait_on_write_block(struct super_block *s) 2826bd4c625cSLinus Torvalds { 28271da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 28281da177e4SLinus Torvalds wait_event(journal->j_join_wait, 28291da177e4SLinus Torvalds !test_bit(J_WRITERS_BLOCKED, &journal->j_state)); 28301da177e4SLinus Torvalds } 28311da177e4SLinus Torvalds 2832bd4c625cSLinus Torvalds static void queue_log_writer(struct super_block *s) 2833bd4c625cSLinus Torvalds { 28341da177e4SLinus Torvalds wait_queue_t wait; 28351da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 28361da177e4SLinus Torvalds set_bit(J_WRITERS_QUEUED, &journal->j_state); 28371da177e4SLinus Torvalds 28381da177e4SLinus Torvalds /* 28391da177e4SLinus Torvalds * we don't want to use wait_event here because 28401da177e4SLinus Torvalds * we only want to wait once. 28411da177e4SLinus Torvalds */ 28421da177e4SLinus Torvalds init_waitqueue_entry(&wait, current); 28431da177e4SLinus Torvalds add_wait_queue(&journal->j_join_wait, &wait); 28441da177e4SLinus Torvalds set_current_state(TASK_UNINTERRUPTIBLE); 28451da177e4SLinus Torvalds if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) 28461da177e4SLinus Torvalds schedule(); 28471da177e4SLinus Torvalds current->state = TASK_RUNNING; 28481da177e4SLinus Torvalds remove_wait_queue(&journal->j_join_wait, &wait); 28491da177e4SLinus Torvalds } 28501da177e4SLinus Torvalds 2851bd4c625cSLinus Torvalds static void wake_queued_writers(struct super_block *s) 2852bd4c625cSLinus Torvalds { 28531da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 28541da177e4SLinus Torvalds if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state)) 28551da177e4SLinus Torvalds wake_up(&journal->j_join_wait); 28561da177e4SLinus Torvalds } 28571da177e4SLinus Torvalds 2858bd4c625cSLinus Torvalds static void let_transaction_grow(struct super_block *sb, unsigned long trans_id) 28591da177e4SLinus Torvalds { 28601da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(sb); 28611da177e4SLinus Torvalds unsigned long bcount = journal->j_bcount; 28621da177e4SLinus Torvalds while (1) { 2863041e0e3bSNishanth Aravamudan schedule_timeout_uninterruptible(1); 28641da177e4SLinus Torvalds journal->j_current_jl->j_state |= LIST_COMMIT_PENDING; 28651da177e4SLinus Torvalds while ((atomic_read(&journal->j_wcount) > 0 || 28661da177e4SLinus Torvalds atomic_read(&journal->j_jlock)) && 28671da177e4SLinus Torvalds journal->j_trans_id == trans_id) { 28681da177e4SLinus Torvalds queue_log_writer(sb); 28691da177e4SLinus Torvalds } 28701da177e4SLinus Torvalds if (journal->j_trans_id != trans_id) 28711da177e4SLinus Torvalds break; 28721da177e4SLinus Torvalds if (bcount == journal->j_bcount) 28731da177e4SLinus Torvalds break; 28741da177e4SLinus Torvalds bcount = journal->j_bcount; 28751da177e4SLinus Torvalds } 28761da177e4SLinus Torvalds } 28771da177e4SLinus Torvalds 28781da177e4SLinus Torvalds /* join == true if you must join an existing transaction. 28791da177e4SLinus Torvalds ** join == false if you can deal with waiting for others to finish 28801da177e4SLinus Torvalds ** 28811da177e4SLinus Torvalds ** this will block until the transaction is joinable. send the number of blocks you 28821da177e4SLinus Torvalds ** expect to use in nblocks. 28831da177e4SLinus Torvalds */ 2884bd4c625cSLinus Torvalds static int do_journal_begin_r(struct reiserfs_transaction_handle *th, 2885bd4c625cSLinus Torvalds struct super_block *p_s_sb, unsigned long nblocks, 2886bd4c625cSLinus Torvalds int join) 2887bd4c625cSLinus Torvalds { 28881da177e4SLinus Torvalds time_t now = get_seconds(); 28891da177e4SLinus Torvalds int old_trans_id; 28901da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 28911da177e4SLinus Torvalds struct reiserfs_transaction_handle myth; 28921da177e4SLinus Torvalds int sched_count = 0; 28931da177e4SLinus Torvalds int retval; 28941da177e4SLinus Torvalds 28951da177e4SLinus Torvalds reiserfs_check_lock_depth(p_s_sb, "journal_begin"); 2896bdd5b29cSJan Kara if (nblocks > journal->j_trans_max) 2897bdd5b29cSJan Kara BUG(); 28981da177e4SLinus Torvalds 28991da177e4SLinus Torvalds PROC_INFO_INC(p_s_sb, journal.journal_being); 29001da177e4SLinus Torvalds /* set here for journal_join */ 29011da177e4SLinus Torvalds th->t_refcount = 1; 29021da177e4SLinus Torvalds th->t_super = p_s_sb; 29031da177e4SLinus Torvalds 29041da177e4SLinus Torvalds relock: 29051da177e4SLinus Torvalds lock_journal(p_s_sb); 29061da177e4SLinus Torvalds if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) { 29071da177e4SLinus Torvalds unlock_journal(p_s_sb); 29081da177e4SLinus Torvalds retval = journal->j_errno; 29091da177e4SLinus Torvalds goto out_fail; 29101da177e4SLinus Torvalds } 29111da177e4SLinus Torvalds journal->j_bcount++; 29121da177e4SLinus Torvalds 29131da177e4SLinus Torvalds if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) { 29141da177e4SLinus Torvalds unlock_journal(p_s_sb); 29151da177e4SLinus Torvalds reiserfs_wait_on_write_block(p_s_sb); 29161da177e4SLinus Torvalds PROC_INFO_INC(p_s_sb, journal.journal_relock_writers); 29171da177e4SLinus Torvalds goto relock; 29181da177e4SLinus Torvalds } 29191da177e4SLinus Torvalds now = get_seconds(); 29201da177e4SLinus Torvalds 29211da177e4SLinus Torvalds /* if there is no room in the journal OR 29221da177e4SLinus Torvalds ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning 29231da177e4SLinus Torvalds ** we don't sleep if there aren't other writers 29241da177e4SLinus Torvalds */ 29251da177e4SLinus Torvalds 29261da177e4SLinus Torvalds if ((!join && journal->j_must_wait > 0) || 2927bd4c625cSLinus Torvalds (!join 2928bd4c625cSLinus Torvalds && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch) 2929bd4c625cSLinus Torvalds || (!join && atomic_read(&journal->j_wcount) > 0 2930bd4c625cSLinus Torvalds && journal->j_trans_start_time > 0 2931bd4c625cSLinus Torvalds && (now - journal->j_trans_start_time) > 2932bd4c625cSLinus Torvalds journal->j_max_trans_age) || (!join 2933bd4c625cSLinus Torvalds && atomic_read(&journal->j_jlock)) 2934bd4c625cSLinus Torvalds || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) { 29351da177e4SLinus Torvalds 29361da177e4SLinus Torvalds old_trans_id = journal->j_trans_id; 29371da177e4SLinus Torvalds unlock_journal(p_s_sb); /* allow others to finish this transaction */ 29381da177e4SLinus Torvalds 29391da177e4SLinus Torvalds if (!join && (journal->j_len_alloc + nblocks + 2) >= 29401da177e4SLinus Torvalds journal->j_max_batch && 2941bd4c625cSLinus Torvalds ((journal->j_len + nblocks + 2) * 100) < 2942bd4c625cSLinus Torvalds (journal->j_len_alloc * 75)) { 29431da177e4SLinus Torvalds if (atomic_read(&journal->j_wcount) > 10) { 29441da177e4SLinus Torvalds sched_count++; 29451da177e4SLinus Torvalds queue_log_writer(p_s_sb); 29461da177e4SLinus Torvalds goto relock; 29471da177e4SLinus Torvalds } 29481da177e4SLinus Torvalds } 29491da177e4SLinus Torvalds /* don't mess with joining the transaction if all we have to do is 29501da177e4SLinus Torvalds * wait for someone else to do a commit 29511da177e4SLinus Torvalds */ 29521da177e4SLinus Torvalds if (atomic_read(&journal->j_jlock)) { 29531da177e4SLinus Torvalds while (journal->j_trans_id == old_trans_id && 29541da177e4SLinus Torvalds atomic_read(&journal->j_jlock)) { 29551da177e4SLinus Torvalds queue_log_writer(p_s_sb); 29561da177e4SLinus Torvalds } 29571da177e4SLinus Torvalds goto relock; 29581da177e4SLinus Torvalds } 29591da177e4SLinus Torvalds retval = journal_join(&myth, p_s_sb, 1); 29601da177e4SLinus Torvalds if (retval) 29611da177e4SLinus Torvalds goto out_fail; 29621da177e4SLinus Torvalds 29631da177e4SLinus Torvalds /* someone might have ended the transaction while we joined */ 29641da177e4SLinus Torvalds if (old_trans_id != journal->j_trans_id) { 29651da177e4SLinus Torvalds retval = do_journal_end(&myth, p_s_sb, 1, 0); 29661da177e4SLinus Torvalds } else { 29671da177e4SLinus Torvalds retval = do_journal_end(&myth, p_s_sb, 1, COMMIT_NOW); 29681da177e4SLinus Torvalds } 29691da177e4SLinus Torvalds 29701da177e4SLinus Torvalds if (retval) 29711da177e4SLinus Torvalds goto out_fail; 29721da177e4SLinus Torvalds 29731da177e4SLinus Torvalds PROC_INFO_INC(p_s_sb, journal.journal_relock_wcount); 29741da177e4SLinus Torvalds goto relock; 29751da177e4SLinus Torvalds } 29761da177e4SLinus Torvalds /* we are the first writer, set trans_id */ 29771da177e4SLinus Torvalds if (journal->j_trans_start_time == 0) { 29781da177e4SLinus Torvalds journal->j_trans_start_time = get_seconds(); 29791da177e4SLinus Torvalds } 29801da177e4SLinus Torvalds atomic_inc(&(journal->j_wcount)); 29811da177e4SLinus Torvalds journal->j_len_alloc += nblocks; 29821da177e4SLinus Torvalds th->t_blocks_logged = 0; 29831da177e4SLinus Torvalds th->t_blocks_allocated = nblocks; 29841da177e4SLinus Torvalds th->t_trans_id = journal->j_trans_id; 29851da177e4SLinus Torvalds unlock_journal(p_s_sb); 29861da177e4SLinus Torvalds INIT_LIST_HEAD(&th->t_list); 298722e2c507SJens Axboe get_fs_excl(); 29881da177e4SLinus Torvalds return 0; 29891da177e4SLinus Torvalds 29901da177e4SLinus Torvalds out_fail: 29911da177e4SLinus Torvalds memset(th, 0, sizeof(*th)); 29921da177e4SLinus Torvalds /* Re-set th->t_super, so we can properly keep track of how many 29931da177e4SLinus Torvalds * persistent transactions there are. We need to do this so if this 29941da177e4SLinus Torvalds * call is part of a failed restart_transaction, we can free it later */ 29951da177e4SLinus Torvalds th->t_super = p_s_sb; 29961da177e4SLinus Torvalds return retval; 29971da177e4SLinus Torvalds } 29981da177e4SLinus Torvalds 2999bd4c625cSLinus Torvalds struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct 3000bd4c625cSLinus Torvalds super_block 3001bd4c625cSLinus Torvalds *s, 3002bd4c625cSLinus Torvalds int nblocks) 3003bd4c625cSLinus Torvalds { 30041da177e4SLinus Torvalds int ret; 30051da177e4SLinus Torvalds struct reiserfs_transaction_handle *th; 30061da177e4SLinus Torvalds 30071da177e4SLinus Torvalds /* if we're nesting into an existing transaction. It will be 30081da177e4SLinus Torvalds ** persistent on its own 30091da177e4SLinus Torvalds */ 30101da177e4SLinus Torvalds if (reiserfs_transaction_running(s)) { 30111da177e4SLinus Torvalds th = current->journal_info; 30121da177e4SLinus Torvalds th->t_refcount++; 30131da177e4SLinus Torvalds if (th->t_refcount < 2) { 30141da177e4SLinus Torvalds BUG(); 30151da177e4SLinus Torvalds } 30161da177e4SLinus Torvalds return th; 30171da177e4SLinus Torvalds } 3018d739b42bSPekka Enberg th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS); 30191da177e4SLinus Torvalds if (!th) 30201da177e4SLinus Torvalds return NULL; 30211da177e4SLinus Torvalds ret = journal_begin(th, s, nblocks); 30221da177e4SLinus Torvalds if (ret) { 3023d739b42bSPekka Enberg kfree(th); 30241da177e4SLinus Torvalds return NULL; 30251da177e4SLinus Torvalds } 30261da177e4SLinus Torvalds 30271da177e4SLinus Torvalds SB_JOURNAL(s)->j_persistent_trans++; 30281da177e4SLinus Torvalds return th; 30291da177e4SLinus Torvalds } 30301da177e4SLinus Torvalds 3031bd4c625cSLinus Torvalds int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th) 3032bd4c625cSLinus Torvalds { 30331da177e4SLinus Torvalds struct super_block *s = th->t_super; 30341da177e4SLinus Torvalds int ret = 0; 30351da177e4SLinus Torvalds if (th->t_trans_id) 30361da177e4SLinus Torvalds ret = journal_end(th, th->t_super, th->t_blocks_allocated); 30371da177e4SLinus Torvalds else 30381da177e4SLinus Torvalds ret = -EIO; 30391da177e4SLinus Torvalds if (th->t_refcount == 0) { 30401da177e4SLinus Torvalds SB_JOURNAL(s)->j_persistent_trans--; 3041d739b42bSPekka Enberg kfree(th); 30421da177e4SLinus Torvalds } 30431da177e4SLinus Torvalds return ret; 30441da177e4SLinus Torvalds } 30451da177e4SLinus Torvalds 3046bd4c625cSLinus Torvalds static int journal_join(struct reiserfs_transaction_handle *th, 3047bd4c625cSLinus Torvalds struct super_block *p_s_sb, unsigned long nblocks) 3048bd4c625cSLinus Torvalds { 30491da177e4SLinus Torvalds struct reiserfs_transaction_handle *cur_th = current->journal_info; 30501da177e4SLinus Torvalds 30511da177e4SLinus Torvalds /* this keeps do_journal_end from NULLing out the current->journal_info 30521da177e4SLinus Torvalds ** pointer 30531da177e4SLinus Torvalds */ 30541da177e4SLinus Torvalds th->t_handle_save = cur_th; 30551da177e4SLinus Torvalds if (cur_th && cur_th->t_refcount > 1) { 30561da177e4SLinus Torvalds BUG(); 30571da177e4SLinus Torvalds } 30581da177e4SLinus Torvalds return do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_JOIN); 30591da177e4SLinus Torvalds } 30601da177e4SLinus Torvalds 3061bd4c625cSLinus Torvalds int journal_join_abort(struct reiserfs_transaction_handle *th, 3062bd4c625cSLinus Torvalds struct super_block *p_s_sb, unsigned long nblocks) 3063bd4c625cSLinus Torvalds { 30641da177e4SLinus Torvalds struct reiserfs_transaction_handle *cur_th = current->journal_info; 30651da177e4SLinus Torvalds 30661da177e4SLinus Torvalds /* this keeps do_journal_end from NULLing out the current->journal_info 30671da177e4SLinus Torvalds ** pointer 30681da177e4SLinus Torvalds */ 30691da177e4SLinus Torvalds th->t_handle_save = cur_th; 30701da177e4SLinus Torvalds if (cur_th && cur_th->t_refcount > 1) { 30711da177e4SLinus Torvalds BUG(); 30721da177e4SLinus Torvalds } 30731da177e4SLinus Torvalds return do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_ABORT); 30741da177e4SLinus Torvalds } 30751da177e4SLinus Torvalds 3076bd4c625cSLinus Torvalds int journal_begin(struct reiserfs_transaction_handle *th, 3077bd4c625cSLinus Torvalds struct super_block *p_s_sb, unsigned long nblocks) 3078bd4c625cSLinus Torvalds { 30791da177e4SLinus Torvalds struct reiserfs_transaction_handle *cur_th = current->journal_info; 30801da177e4SLinus Torvalds int ret; 30811da177e4SLinus Torvalds 30821da177e4SLinus Torvalds th->t_handle_save = NULL; 30831da177e4SLinus Torvalds if (cur_th) { 30841da177e4SLinus Torvalds /* we are nesting into the current transaction */ 30851da177e4SLinus Torvalds if (cur_th->t_super == p_s_sb) { 30861da177e4SLinus Torvalds BUG_ON(!cur_th->t_refcount); 30871da177e4SLinus Torvalds cur_th->t_refcount++; 30881da177e4SLinus Torvalds memcpy(th, cur_th, sizeof(*th)); 30891da177e4SLinus Torvalds if (th->t_refcount <= 1) 3090bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 3091bd4c625cSLinus Torvalds "BAD: refcount <= 1, but journal_info != 0"); 30921da177e4SLinus Torvalds return 0; 30931da177e4SLinus Torvalds } else { 30941da177e4SLinus Torvalds /* we've ended up with a handle from a different filesystem. 30951da177e4SLinus Torvalds ** save it and restore on journal_end. This should never 30961da177e4SLinus Torvalds ** really happen... 30971da177e4SLinus Torvalds */ 3098bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 3099bd4c625cSLinus Torvalds "clm-2100: nesting info a different FS"); 31001da177e4SLinus Torvalds th->t_handle_save = current->journal_info; 31011da177e4SLinus Torvalds current->journal_info = th; 31021da177e4SLinus Torvalds } 31031da177e4SLinus Torvalds } else { 31041da177e4SLinus Torvalds current->journal_info = th; 31051da177e4SLinus Torvalds } 31061da177e4SLinus Torvalds ret = do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_REG); 31071da177e4SLinus Torvalds if (current->journal_info != th) 31081da177e4SLinus Torvalds BUG(); 31091da177e4SLinus Torvalds 31101da177e4SLinus Torvalds /* I guess this boils down to being the reciprocal of clm-2100 above. 31111da177e4SLinus Torvalds * If do_journal_begin_r fails, we need to put it back, since journal_end 31121da177e4SLinus Torvalds * won't be called to do it. */ 31131da177e4SLinus Torvalds if (ret) 31141da177e4SLinus Torvalds current->journal_info = th->t_handle_save; 31151da177e4SLinus Torvalds else 31161da177e4SLinus Torvalds BUG_ON(!th->t_refcount); 31171da177e4SLinus Torvalds 31181da177e4SLinus Torvalds return ret; 31191da177e4SLinus Torvalds } 31201da177e4SLinus Torvalds 31211da177e4SLinus Torvalds /* 31221da177e4SLinus Torvalds ** puts bh into the current transaction. If it was already there, reorders removes the 31231da177e4SLinus Torvalds ** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order). 31241da177e4SLinus Torvalds ** 31251da177e4SLinus Torvalds ** if it was dirty, cleans and files onto the clean list. I can't let it be dirty again until the 31261da177e4SLinus Torvalds ** transaction is committed. 31271da177e4SLinus Torvalds ** 31281da177e4SLinus Torvalds ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len. 31291da177e4SLinus Torvalds */ 3130bd4c625cSLinus Torvalds int journal_mark_dirty(struct reiserfs_transaction_handle *th, 3131bd4c625cSLinus Torvalds struct super_block *p_s_sb, struct buffer_head *bh) 3132bd4c625cSLinus Torvalds { 31331da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 31341da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn = NULL; 31351da177e4SLinus Torvalds int count_already_incd = 0; 31361da177e4SLinus Torvalds int prepared = 0; 31371da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 31381da177e4SLinus Torvalds 31391da177e4SLinus Torvalds PROC_INFO_INC(p_s_sb, journal.mark_dirty); 31401da177e4SLinus Torvalds if (th->t_trans_id != journal->j_trans_id) { 3141bd4c625cSLinus Torvalds reiserfs_panic(th->t_super, 3142bd4c625cSLinus Torvalds "journal-1577: handle trans id %ld != current trans id %ld\n", 31431da177e4SLinus Torvalds th->t_trans_id, journal->j_trans_id); 31441da177e4SLinus Torvalds } 31451da177e4SLinus Torvalds 31461da177e4SLinus Torvalds p_s_sb->s_dirt = 1; 31471da177e4SLinus Torvalds 31481da177e4SLinus Torvalds prepared = test_clear_buffer_journal_prepared(bh); 31491da177e4SLinus Torvalds clear_buffer_journal_restore_dirty(bh); 31501da177e4SLinus Torvalds /* already in this transaction, we are done */ 31511da177e4SLinus Torvalds if (buffer_journaled(bh)) { 31521da177e4SLinus Torvalds PROC_INFO_INC(p_s_sb, journal.mark_dirty_already); 31531da177e4SLinus Torvalds return 0; 31541da177e4SLinus Torvalds } 31551da177e4SLinus Torvalds 31561da177e4SLinus Torvalds /* this must be turned into a panic instead of a warning. We can't allow 31571da177e4SLinus Torvalds ** a dirty or journal_dirty or locked buffer to be logged, as some changes 31581da177e4SLinus Torvalds ** could get to disk too early. NOT GOOD. 31591da177e4SLinus Torvalds */ 31601da177e4SLinus Torvalds if (!prepared || buffer_dirty(bh)) { 31611da177e4SLinus Torvalds reiserfs_warning(p_s_sb, "journal-1777: buffer %llu bad state " 31621da177e4SLinus Torvalds "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT", 3163bd4c625cSLinus Torvalds (unsigned long long)bh->b_blocknr, 3164bd4c625cSLinus Torvalds prepared ? ' ' : '!', 31651da177e4SLinus Torvalds buffer_locked(bh) ? ' ' : '!', 31661da177e4SLinus Torvalds buffer_dirty(bh) ? ' ' : '!', 31671da177e4SLinus Torvalds buffer_journal_dirty(bh) ? ' ' : '!'); 31681da177e4SLinus Torvalds } 31691da177e4SLinus Torvalds 31701da177e4SLinus Torvalds if (atomic_read(&(journal->j_wcount)) <= 0) { 3171bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 3172bd4c625cSLinus Torvalds "journal-1409: journal_mark_dirty returning because j_wcount was %d", 3173bd4c625cSLinus Torvalds atomic_read(&(journal->j_wcount))); 31741da177e4SLinus Torvalds return 1; 31751da177e4SLinus Torvalds } 31761da177e4SLinus Torvalds /* this error means I've screwed up, and we've overflowed the transaction. 31771da177e4SLinus Torvalds ** Nothing can be done here, except make the FS readonly or panic. 31781da177e4SLinus Torvalds */ 31791da177e4SLinus Torvalds if (journal->j_len >= journal->j_trans_max) { 3180bd4c625cSLinus Torvalds reiserfs_panic(th->t_super, 3181bd4c625cSLinus Torvalds "journal-1413: journal_mark_dirty: j_len (%lu) is too big\n", 3182bd4c625cSLinus Torvalds journal->j_len); 31831da177e4SLinus Torvalds } 31841da177e4SLinus Torvalds 31851da177e4SLinus Torvalds if (buffer_journal_dirty(bh)) { 31861da177e4SLinus Torvalds count_already_incd = 1; 31871da177e4SLinus Torvalds PROC_INFO_INC(p_s_sb, journal.mark_dirty_notjournal); 31881da177e4SLinus Torvalds clear_buffer_journal_dirty(bh); 31891da177e4SLinus Torvalds } 31901da177e4SLinus Torvalds 31911da177e4SLinus Torvalds if (journal->j_len > journal->j_len_alloc) { 31921da177e4SLinus Torvalds journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT; 31931da177e4SLinus Torvalds } 31941da177e4SLinus Torvalds 31951da177e4SLinus Torvalds set_buffer_journaled(bh); 31961da177e4SLinus Torvalds 31971da177e4SLinus Torvalds /* now put this guy on the end */ 31981da177e4SLinus Torvalds if (!cn) { 31991da177e4SLinus Torvalds cn = get_cnode(p_s_sb); 32001da177e4SLinus Torvalds if (!cn) { 32011da177e4SLinus Torvalds reiserfs_panic(p_s_sb, "get_cnode failed!\n"); 32021da177e4SLinus Torvalds } 32031da177e4SLinus Torvalds 32041da177e4SLinus Torvalds if (th->t_blocks_logged == th->t_blocks_allocated) { 32051da177e4SLinus Torvalds th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT; 32061da177e4SLinus Torvalds journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT; 32071da177e4SLinus Torvalds } 32081da177e4SLinus Torvalds th->t_blocks_logged++; 32091da177e4SLinus Torvalds journal->j_len++; 32101da177e4SLinus Torvalds 32111da177e4SLinus Torvalds cn->bh = bh; 32121da177e4SLinus Torvalds cn->blocknr = bh->b_blocknr; 32131da177e4SLinus Torvalds cn->sb = p_s_sb; 32141da177e4SLinus Torvalds cn->jlist = NULL; 32151da177e4SLinus Torvalds insert_journal_hash(journal->j_hash_table, cn); 32161da177e4SLinus Torvalds if (!count_already_incd) { 32171da177e4SLinus Torvalds get_bh(bh); 32181da177e4SLinus Torvalds } 32191da177e4SLinus Torvalds } 32201da177e4SLinus Torvalds cn->next = NULL; 32211da177e4SLinus Torvalds cn->prev = journal->j_last; 32221da177e4SLinus Torvalds cn->bh = bh; 32231da177e4SLinus Torvalds if (journal->j_last) { 32241da177e4SLinus Torvalds journal->j_last->next = cn; 32251da177e4SLinus Torvalds journal->j_last = cn; 32261da177e4SLinus Torvalds } else { 32271da177e4SLinus Torvalds journal->j_first = cn; 32281da177e4SLinus Torvalds journal->j_last = cn; 32291da177e4SLinus Torvalds } 32301da177e4SLinus Torvalds return 0; 32311da177e4SLinus Torvalds } 32321da177e4SLinus Torvalds 3233bd4c625cSLinus Torvalds int journal_end(struct reiserfs_transaction_handle *th, 3234bd4c625cSLinus Torvalds struct super_block *p_s_sb, unsigned long nblocks) 3235bd4c625cSLinus Torvalds { 32361da177e4SLinus Torvalds if (!current->journal_info && th->t_refcount > 1) 32371da177e4SLinus Torvalds reiserfs_warning(p_s_sb, "REISER-NESTING: th NULL, refcount %d", 32381da177e4SLinus Torvalds th->t_refcount); 32391da177e4SLinus Torvalds 32401da177e4SLinus Torvalds if (!th->t_trans_id) { 32411da177e4SLinus Torvalds WARN_ON(1); 32421da177e4SLinus Torvalds return -EIO; 32431da177e4SLinus Torvalds } 32441da177e4SLinus Torvalds 32451da177e4SLinus Torvalds th->t_refcount--; 32461da177e4SLinus Torvalds if (th->t_refcount > 0) { 3247bd4c625cSLinus Torvalds struct reiserfs_transaction_handle *cur_th = 3248bd4c625cSLinus Torvalds current->journal_info; 32491da177e4SLinus Torvalds 32501da177e4SLinus Torvalds /* we aren't allowed to close a nested transaction on a different 32511da177e4SLinus Torvalds ** filesystem from the one in the task struct 32521da177e4SLinus Torvalds */ 32531da177e4SLinus Torvalds if (cur_th->t_super != th->t_super) 32541da177e4SLinus Torvalds BUG(); 32551da177e4SLinus Torvalds 32561da177e4SLinus Torvalds if (th != cur_th) { 32571da177e4SLinus Torvalds memcpy(current->journal_info, th, sizeof(*th)); 32581da177e4SLinus Torvalds th->t_trans_id = 0; 32591da177e4SLinus Torvalds } 32601da177e4SLinus Torvalds return 0; 32611da177e4SLinus Torvalds } else { 32621da177e4SLinus Torvalds return do_journal_end(th, p_s_sb, nblocks, 0); 32631da177e4SLinus Torvalds } 32641da177e4SLinus Torvalds } 32651da177e4SLinus Torvalds 32661da177e4SLinus Torvalds /* removes from the current transaction, relsing and descrementing any counters. 32671da177e4SLinus Torvalds ** also files the removed buffer directly onto the clean list 32681da177e4SLinus Torvalds ** 32691da177e4SLinus Torvalds ** called by journal_mark_freed when a block has been deleted 32701da177e4SLinus Torvalds ** 32711da177e4SLinus Torvalds ** returns 1 if it cleaned and relsed the buffer. 0 otherwise 32721da177e4SLinus Torvalds */ 3273bd4c625cSLinus Torvalds static int remove_from_transaction(struct super_block *p_s_sb, 3274bd4c625cSLinus Torvalds b_blocknr_t blocknr, int already_cleaned) 3275bd4c625cSLinus Torvalds { 32761da177e4SLinus Torvalds struct buffer_head *bh; 32771da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 32781da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 32791da177e4SLinus Torvalds int ret = 0; 32801da177e4SLinus Torvalds 32811da177e4SLinus Torvalds cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, blocknr); 32821da177e4SLinus Torvalds if (!cn || !cn->bh) { 32831da177e4SLinus Torvalds return ret; 32841da177e4SLinus Torvalds } 32851da177e4SLinus Torvalds bh = cn->bh; 32861da177e4SLinus Torvalds if (cn->prev) { 32871da177e4SLinus Torvalds cn->prev->next = cn->next; 32881da177e4SLinus Torvalds } 32891da177e4SLinus Torvalds if (cn->next) { 32901da177e4SLinus Torvalds cn->next->prev = cn->prev; 32911da177e4SLinus Torvalds } 32921da177e4SLinus Torvalds if (cn == journal->j_first) { 32931da177e4SLinus Torvalds journal->j_first = cn->next; 32941da177e4SLinus Torvalds } 32951da177e4SLinus Torvalds if (cn == journal->j_last) { 32961da177e4SLinus Torvalds journal->j_last = cn->prev; 32971da177e4SLinus Torvalds } 32981da177e4SLinus Torvalds if (bh) 3299bd4c625cSLinus Torvalds remove_journal_hash(p_s_sb, journal->j_hash_table, NULL, 3300bd4c625cSLinus Torvalds bh->b_blocknr, 0); 33011da177e4SLinus Torvalds clear_buffer_journaled(bh); /* don't log this one */ 33021da177e4SLinus Torvalds 33031da177e4SLinus Torvalds if (!already_cleaned) { 33041da177e4SLinus Torvalds clear_buffer_journal_dirty(bh); 33051da177e4SLinus Torvalds clear_buffer_dirty(bh); 33061da177e4SLinus Torvalds clear_buffer_journal_test(bh); 33071da177e4SLinus Torvalds put_bh(bh); 33081da177e4SLinus Torvalds if (atomic_read(&(bh->b_count)) < 0) { 3309bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 3310bd4c625cSLinus Torvalds "journal-1752: remove from trans, b_count < 0"); 33111da177e4SLinus Torvalds } 33121da177e4SLinus Torvalds ret = 1; 33131da177e4SLinus Torvalds } 33141da177e4SLinus Torvalds journal->j_len--; 33151da177e4SLinus Torvalds journal->j_len_alloc--; 33161da177e4SLinus Torvalds free_cnode(p_s_sb, cn); 33171da177e4SLinus Torvalds return ret; 33181da177e4SLinus Torvalds } 33191da177e4SLinus Torvalds 33201da177e4SLinus Torvalds /* 33211da177e4SLinus Torvalds ** for any cnode in a journal list, it can only be dirtied of all the 33221da177e4SLinus Torvalds ** transactions that include it are commited to disk. 33231da177e4SLinus Torvalds ** this checks through each transaction, and returns 1 if you are allowed to dirty, 33241da177e4SLinus Torvalds ** and 0 if you aren't 33251da177e4SLinus Torvalds ** 33261da177e4SLinus Torvalds ** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log 33271da177e4SLinus Torvalds ** blocks for a given transaction on disk 33281da177e4SLinus Torvalds ** 33291da177e4SLinus Torvalds */ 3330bd4c625cSLinus Torvalds static int can_dirty(struct reiserfs_journal_cnode *cn) 3331bd4c625cSLinus Torvalds { 33321da177e4SLinus Torvalds struct super_block *sb = cn->sb; 33331da177e4SLinus Torvalds b_blocknr_t blocknr = cn->blocknr; 33341da177e4SLinus Torvalds struct reiserfs_journal_cnode *cur = cn->hprev; 33351da177e4SLinus Torvalds int can_dirty = 1; 33361da177e4SLinus Torvalds 33371da177e4SLinus Torvalds /* first test hprev. These are all newer than cn, so any node here 33381da177e4SLinus Torvalds ** with the same block number and dev means this node can't be sent 33391da177e4SLinus Torvalds ** to disk right now. 33401da177e4SLinus Torvalds */ 33411da177e4SLinus Torvalds while (cur && can_dirty) { 33421da177e4SLinus Torvalds if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb && 33431da177e4SLinus Torvalds cur->blocknr == blocknr) { 33441da177e4SLinus Torvalds can_dirty = 0; 33451da177e4SLinus Torvalds } 33461da177e4SLinus Torvalds cur = cur->hprev; 33471da177e4SLinus Torvalds } 33481da177e4SLinus Torvalds /* then test hnext. These are all older than cn. As long as they 33491da177e4SLinus Torvalds ** are committed to the log, it is safe to write cn to disk 33501da177e4SLinus Torvalds */ 33511da177e4SLinus Torvalds cur = cn->hnext; 33521da177e4SLinus Torvalds while (cur && can_dirty) { 33531da177e4SLinus Torvalds if (cur->jlist && cur->jlist->j_len > 0 && 33541da177e4SLinus Torvalds atomic_read(&(cur->jlist->j_commit_left)) > 0 && cur->bh && 33551da177e4SLinus Torvalds cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) { 33561da177e4SLinus Torvalds can_dirty = 0; 33571da177e4SLinus Torvalds } 33581da177e4SLinus Torvalds cur = cur->hnext; 33591da177e4SLinus Torvalds } 33601da177e4SLinus Torvalds return can_dirty; 33611da177e4SLinus Torvalds } 33621da177e4SLinus Torvalds 33631da177e4SLinus Torvalds /* syncs the commit blocks, but does not force the real buffers to disk 33641da177e4SLinus Torvalds ** will wait until the current transaction is done/commited before returning 33651da177e4SLinus Torvalds */ 3366bd4c625cSLinus Torvalds int journal_end_sync(struct reiserfs_transaction_handle *th, 3367bd4c625cSLinus Torvalds struct super_block *p_s_sb, unsigned long nblocks) 3368bd4c625cSLinus Torvalds { 33691da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 33701da177e4SLinus Torvalds 33711da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 33721da177e4SLinus Torvalds /* you can sync while nested, very, very bad */ 33731da177e4SLinus Torvalds if (th->t_refcount > 1) { 33741da177e4SLinus Torvalds BUG(); 33751da177e4SLinus Torvalds } 33761da177e4SLinus Torvalds if (journal->j_len == 0) { 3377bd4c625cSLinus Torvalds reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb), 3378bd4c625cSLinus Torvalds 1); 33791da177e4SLinus Torvalds journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb)); 33801da177e4SLinus Torvalds } 33811da177e4SLinus Torvalds return do_journal_end(th, p_s_sb, nblocks, COMMIT_NOW | WAIT); 33821da177e4SLinus Torvalds } 33831da177e4SLinus Torvalds 33841da177e4SLinus Torvalds /* 33851da177e4SLinus Torvalds ** writeback the pending async commits to disk 33861da177e4SLinus Torvalds */ 3387bd4c625cSLinus Torvalds static void flush_async_commits(void *p) 3388bd4c625cSLinus Torvalds { 33891da177e4SLinus Torvalds struct super_block *p_s_sb = p; 33901da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 33911da177e4SLinus Torvalds struct reiserfs_journal_list *jl; 33921da177e4SLinus Torvalds struct list_head *entry; 33931da177e4SLinus Torvalds 33941da177e4SLinus Torvalds lock_kernel(); 33951da177e4SLinus Torvalds if (!list_empty(&journal->j_journal_list)) { 33961da177e4SLinus Torvalds /* last entry is the youngest, commit it and you get everything */ 33971da177e4SLinus Torvalds entry = journal->j_journal_list.prev; 33981da177e4SLinus Torvalds jl = JOURNAL_LIST_ENTRY(entry); 33991da177e4SLinus Torvalds flush_commit_list(p_s_sb, jl, 1); 34001da177e4SLinus Torvalds } 34011da177e4SLinus Torvalds unlock_kernel(); 34021da177e4SLinus Torvalds /* 34031da177e4SLinus Torvalds * this is a little racey, but there's no harm in missing 34041da177e4SLinus Torvalds * the filemap_fdata_write 34051da177e4SLinus Torvalds */ 3406bd4c625cSLinus Torvalds if (!atomic_read(&journal->j_async_throttle) 3407bd4c625cSLinus Torvalds && !reiserfs_is_journal_aborted(journal)) { 34081da177e4SLinus Torvalds atomic_inc(&journal->j_async_throttle); 34091da177e4SLinus Torvalds filemap_fdatawrite(p_s_sb->s_bdev->bd_inode->i_mapping); 34101da177e4SLinus Torvalds atomic_dec(&journal->j_async_throttle); 34111da177e4SLinus Torvalds } 34121da177e4SLinus Torvalds } 34131da177e4SLinus Torvalds 34141da177e4SLinus Torvalds /* 34151da177e4SLinus Torvalds ** flushes any old transactions to disk 34161da177e4SLinus Torvalds ** ends the current transaction if it is too old 34171da177e4SLinus Torvalds */ 3418bd4c625cSLinus Torvalds int reiserfs_flush_old_commits(struct super_block *p_s_sb) 3419bd4c625cSLinus Torvalds { 34201da177e4SLinus Torvalds time_t now; 34211da177e4SLinus Torvalds struct reiserfs_transaction_handle th; 34221da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 34231da177e4SLinus Torvalds 34241da177e4SLinus Torvalds now = get_seconds(); 34251da177e4SLinus Torvalds /* safety check so we don't flush while we are replaying the log during 34261da177e4SLinus Torvalds * mount 34271da177e4SLinus Torvalds */ 34281da177e4SLinus Torvalds if (list_empty(&journal->j_journal_list)) { 34291da177e4SLinus Torvalds return 0; 34301da177e4SLinus Torvalds } 34311da177e4SLinus Torvalds 34321da177e4SLinus Torvalds /* check the current transaction. If there are no writers, and it is 34331da177e4SLinus Torvalds * too old, finish it, and force the commit blocks to disk 34341da177e4SLinus Torvalds */ 34351da177e4SLinus Torvalds if (atomic_read(&journal->j_wcount) <= 0 && 34361da177e4SLinus Torvalds journal->j_trans_start_time > 0 && 34371da177e4SLinus Torvalds journal->j_len > 0 && 3438bd4c625cSLinus Torvalds (now - journal->j_trans_start_time) > journal->j_max_trans_age) { 34391da177e4SLinus Torvalds if (!journal_join(&th, p_s_sb, 1)) { 3440bd4c625cSLinus Torvalds reiserfs_prepare_for_journal(p_s_sb, 3441bd4c625cSLinus Torvalds SB_BUFFER_WITH_SB(p_s_sb), 3442bd4c625cSLinus Torvalds 1); 3443bd4c625cSLinus Torvalds journal_mark_dirty(&th, p_s_sb, 3444bd4c625cSLinus Torvalds SB_BUFFER_WITH_SB(p_s_sb)); 34451da177e4SLinus Torvalds 34461da177e4SLinus Torvalds /* we're only being called from kreiserfsd, it makes no sense to do 34471da177e4SLinus Torvalds ** an async commit so that kreiserfsd can do it later 34481da177e4SLinus Torvalds */ 34491da177e4SLinus Torvalds do_journal_end(&th, p_s_sb, 1, COMMIT_NOW | WAIT); 34501da177e4SLinus Torvalds } 34511da177e4SLinus Torvalds } 34521da177e4SLinus Torvalds return p_s_sb->s_dirt; 34531da177e4SLinus Torvalds } 34541da177e4SLinus Torvalds 34551da177e4SLinus Torvalds /* 34561da177e4SLinus Torvalds ** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit 34571da177e4SLinus Torvalds ** 34581da177e4SLinus Torvalds ** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all 34591da177e4SLinus Torvalds ** the writers are done. By the time it wakes up, the transaction it was called has already ended, so it just 34601da177e4SLinus Torvalds ** flushes the commit list and returns 0. 34611da177e4SLinus Torvalds ** 34621da177e4SLinus Torvalds ** Won't batch when flush or commit_now is set. Also won't batch when others are waiting on j_join_wait. 34631da177e4SLinus Torvalds ** 34641da177e4SLinus Torvalds ** Note, we can't allow the journal_end to proceed while there are still writers in the log. 34651da177e4SLinus Torvalds */ 3466bd4c625cSLinus Torvalds static int check_journal_end(struct reiserfs_transaction_handle *th, 3467bd4c625cSLinus Torvalds struct super_block *p_s_sb, unsigned long nblocks, 3468bd4c625cSLinus Torvalds int flags) 3469bd4c625cSLinus Torvalds { 34701da177e4SLinus Torvalds 34711da177e4SLinus Torvalds time_t now; 34721da177e4SLinus Torvalds int flush = flags & FLUSH_ALL; 34731da177e4SLinus Torvalds int commit_now = flags & COMMIT_NOW; 34741da177e4SLinus Torvalds int wait_on_commit = flags & WAIT; 34751da177e4SLinus Torvalds struct reiserfs_journal_list *jl; 34761da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 34771da177e4SLinus Torvalds 34781da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 34791da177e4SLinus Torvalds 34801da177e4SLinus Torvalds if (th->t_trans_id != journal->j_trans_id) { 3481bd4c625cSLinus Torvalds reiserfs_panic(th->t_super, 3482bd4c625cSLinus Torvalds "journal-1577: handle trans id %ld != current trans id %ld\n", 34831da177e4SLinus Torvalds th->t_trans_id, journal->j_trans_id); 34841da177e4SLinus Torvalds } 34851da177e4SLinus Torvalds 34861da177e4SLinus Torvalds journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged); 34871da177e4SLinus Torvalds if (atomic_read(&(journal->j_wcount)) > 0) { /* <= 0 is allowed. unmounting might not call begin */ 34881da177e4SLinus Torvalds atomic_dec(&(journal->j_wcount)); 34891da177e4SLinus Torvalds } 34901da177e4SLinus Torvalds 34911da177e4SLinus Torvalds /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released 34921da177e4SLinus Torvalds ** will be dealt with by next transaction that actually writes something, but should be taken 34931da177e4SLinus Torvalds ** care of in this trans 34941da177e4SLinus Torvalds */ 34951da177e4SLinus Torvalds if (journal->j_len == 0) { 34961da177e4SLinus Torvalds BUG(); 34971da177e4SLinus Torvalds } 34981da177e4SLinus Torvalds /* if wcount > 0, and we are called to with flush or commit_now, 34991da177e4SLinus Torvalds ** we wait on j_join_wait. We will wake up when the last writer has 35001da177e4SLinus Torvalds ** finished the transaction, and started it on its way to the disk. 35011da177e4SLinus Torvalds ** Then, we flush the commit or journal list, and just return 0 35021da177e4SLinus Torvalds ** because the rest of journal end was already done for this transaction. 35031da177e4SLinus Torvalds */ 35041da177e4SLinus Torvalds if (atomic_read(&(journal->j_wcount)) > 0) { 35051da177e4SLinus Torvalds if (flush || commit_now) { 35061da177e4SLinus Torvalds unsigned trans_id; 35071da177e4SLinus Torvalds 35081da177e4SLinus Torvalds jl = journal->j_current_jl; 35091da177e4SLinus Torvalds trans_id = jl->j_trans_id; 35101da177e4SLinus Torvalds if (wait_on_commit) 35111da177e4SLinus Torvalds jl->j_state |= LIST_COMMIT_PENDING; 35121da177e4SLinus Torvalds atomic_set(&(journal->j_jlock), 1); 35131da177e4SLinus Torvalds if (flush) { 35141da177e4SLinus Torvalds journal->j_next_full_flush = 1; 35151da177e4SLinus Torvalds } 35161da177e4SLinus Torvalds unlock_journal(p_s_sb); 35171da177e4SLinus Torvalds 35181da177e4SLinus Torvalds /* sleep while the current transaction is still j_jlocked */ 35191da177e4SLinus Torvalds while (journal->j_trans_id == trans_id) { 35201da177e4SLinus Torvalds if (atomic_read(&journal->j_jlock)) { 35211da177e4SLinus Torvalds queue_log_writer(p_s_sb); 35221da177e4SLinus Torvalds } else { 35231da177e4SLinus Torvalds lock_journal(p_s_sb); 35241da177e4SLinus Torvalds if (journal->j_trans_id == trans_id) { 3525bd4c625cSLinus Torvalds atomic_set(&(journal->j_jlock), 3526bd4c625cSLinus Torvalds 1); 35271da177e4SLinus Torvalds } 35281da177e4SLinus Torvalds unlock_journal(p_s_sb); 35291da177e4SLinus Torvalds } 35301da177e4SLinus Torvalds } 35311da177e4SLinus Torvalds if (journal->j_trans_id == trans_id) { 35321da177e4SLinus Torvalds BUG(); 35331da177e4SLinus Torvalds } 3534bd4c625cSLinus Torvalds if (commit_now 3535bd4c625cSLinus Torvalds && journal_list_still_alive(p_s_sb, trans_id) 3536bd4c625cSLinus Torvalds && wait_on_commit) { 35371da177e4SLinus Torvalds flush_commit_list(p_s_sb, jl, 1); 35381da177e4SLinus Torvalds } 35391da177e4SLinus Torvalds return 0; 35401da177e4SLinus Torvalds } 35411da177e4SLinus Torvalds unlock_journal(p_s_sb); 35421da177e4SLinus Torvalds return 0; 35431da177e4SLinus Torvalds } 35441da177e4SLinus Torvalds 35451da177e4SLinus Torvalds /* deal with old transactions where we are the last writers */ 35461da177e4SLinus Torvalds now = get_seconds(); 35471da177e4SLinus Torvalds if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) { 35481da177e4SLinus Torvalds commit_now = 1; 35491da177e4SLinus Torvalds journal->j_next_async_flush = 1; 35501da177e4SLinus Torvalds } 35511da177e4SLinus Torvalds /* don't batch when someone is waiting on j_join_wait */ 35521da177e4SLinus Torvalds /* don't batch when syncing the commit or flushing the whole trans */ 3553bd4c625cSLinus Torvalds if (!(journal->j_must_wait > 0) && !(atomic_read(&(journal->j_jlock))) 3554bd4c625cSLinus Torvalds && !flush && !commit_now && (journal->j_len < journal->j_max_batch) 3555bd4c625cSLinus Torvalds && journal->j_len_alloc < journal->j_max_batch 3556bd4c625cSLinus Torvalds && journal->j_cnode_free > (journal->j_trans_max * 3)) { 35571da177e4SLinus Torvalds journal->j_bcount++; 35581da177e4SLinus Torvalds unlock_journal(p_s_sb); 35591da177e4SLinus Torvalds return 0; 35601da177e4SLinus Torvalds } 35611da177e4SLinus Torvalds 35621da177e4SLinus Torvalds if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(p_s_sb)) { 3563bd4c625cSLinus Torvalds reiserfs_panic(p_s_sb, 3564bd4c625cSLinus Torvalds "journal-003: journal_end: j_start (%ld) is too high\n", 3565bd4c625cSLinus Torvalds journal->j_start); 35661da177e4SLinus Torvalds } 35671da177e4SLinus Torvalds return 1; 35681da177e4SLinus Torvalds } 35691da177e4SLinus Torvalds 35701da177e4SLinus Torvalds /* 35711da177e4SLinus Torvalds ** Does all the work that makes deleting blocks safe. 35721da177e4SLinus Torvalds ** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on. 35731da177e4SLinus Torvalds ** 35741da177e4SLinus Torvalds ** otherwise: 35751da177e4SLinus Torvalds ** set a bit for the block in the journal bitmap. That will prevent it from being allocated for unformatted nodes 35761da177e4SLinus Torvalds ** before this transaction has finished. 35771da177e4SLinus Torvalds ** 35781da177e4SLinus Torvalds ** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers. That will prevent any old transactions with 35791da177e4SLinus Torvalds ** this block from trying to flush to the real location. Since we aren't removing the cnode from the journal_list_hash, 35801da177e4SLinus Torvalds ** the block can't be reallocated yet. 35811da177e4SLinus Torvalds ** 35821da177e4SLinus Torvalds ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list. 35831da177e4SLinus Torvalds */ 3584bd4c625cSLinus Torvalds int journal_mark_freed(struct reiserfs_transaction_handle *th, 3585bd4c625cSLinus Torvalds struct super_block *p_s_sb, b_blocknr_t blocknr) 3586bd4c625cSLinus Torvalds { 35871da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 35881da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn = NULL; 35891da177e4SLinus Torvalds struct buffer_head *bh = NULL; 35901da177e4SLinus Torvalds struct reiserfs_list_bitmap *jb = NULL; 35911da177e4SLinus Torvalds int cleaned = 0; 35921da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 35931da177e4SLinus Torvalds 35941da177e4SLinus Torvalds cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, blocknr); 35951da177e4SLinus Torvalds if (cn && cn->bh) { 35961da177e4SLinus Torvalds bh = cn->bh; 35971da177e4SLinus Torvalds get_bh(bh); 35981da177e4SLinus Torvalds } 35991da177e4SLinus Torvalds /* if it is journal new, we just remove it from this transaction */ 36001da177e4SLinus Torvalds if (bh && buffer_journal_new(bh)) { 36011da177e4SLinus Torvalds clear_buffer_journal_new(bh); 36021da177e4SLinus Torvalds clear_prepared_bits(bh); 36031da177e4SLinus Torvalds reiserfs_clean_and_file_buffer(bh); 36041da177e4SLinus Torvalds cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned); 36051da177e4SLinus Torvalds } else { 36061da177e4SLinus Torvalds /* set the bit for this block in the journal bitmap for this transaction */ 36071da177e4SLinus Torvalds jb = journal->j_current_jl->j_list_bitmap; 36081da177e4SLinus Torvalds if (!jb) { 3609bd4c625cSLinus Torvalds reiserfs_panic(p_s_sb, 3610bd4c625cSLinus Torvalds "journal-1702: journal_mark_freed, journal_list_bitmap is NULL\n"); 36111da177e4SLinus Torvalds } 36121da177e4SLinus Torvalds set_bit_in_list_bitmap(p_s_sb, blocknr, jb); 36131da177e4SLinus Torvalds 36141da177e4SLinus Torvalds /* Note, the entire while loop is not allowed to schedule. */ 36151da177e4SLinus Torvalds 36161da177e4SLinus Torvalds if (bh) { 36171da177e4SLinus Torvalds clear_prepared_bits(bh); 36181da177e4SLinus Torvalds reiserfs_clean_and_file_buffer(bh); 36191da177e4SLinus Torvalds } 36201da177e4SLinus Torvalds cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned); 36211da177e4SLinus Torvalds 36221da177e4SLinus Torvalds /* find all older transactions with this block, make sure they don't try to write it out */ 3623bd4c625cSLinus Torvalds cn = get_journal_hash_dev(p_s_sb, journal->j_list_hash_table, 3624bd4c625cSLinus Torvalds blocknr); 36251da177e4SLinus Torvalds while (cn) { 36261da177e4SLinus Torvalds if (p_s_sb == cn->sb && blocknr == cn->blocknr) { 36271da177e4SLinus Torvalds set_bit(BLOCK_FREED, &cn->state); 36281da177e4SLinus Torvalds if (cn->bh) { 36291da177e4SLinus Torvalds if (!cleaned) { 36301da177e4SLinus Torvalds /* remove_from_transaction will brelse the buffer if it was 36311da177e4SLinus Torvalds ** in the current trans 36321da177e4SLinus Torvalds */ 3633bd4c625cSLinus Torvalds clear_buffer_journal_dirty(cn-> 3634bd4c625cSLinus Torvalds bh); 36351da177e4SLinus Torvalds clear_buffer_dirty(cn->bh); 3636bd4c625cSLinus Torvalds clear_buffer_journal_test(cn-> 3637bd4c625cSLinus Torvalds bh); 36381da177e4SLinus Torvalds cleaned = 1; 36391da177e4SLinus Torvalds put_bh(cn->bh); 3640bd4c625cSLinus Torvalds if (atomic_read 3641bd4c625cSLinus Torvalds (&(cn->bh->b_count)) < 0) { 3642bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 3643bd4c625cSLinus Torvalds "journal-2138: cn->bh->b_count < 0"); 36441da177e4SLinus Torvalds } 36451da177e4SLinus Torvalds } 36461da177e4SLinus Torvalds if (cn->jlist) { /* since we are clearing the bh, we MUST dec nonzerolen */ 3647bd4c625cSLinus Torvalds atomic_dec(& 3648bd4c625cSLinus Torvalds (cn->jlist-> 3649bd4c625cSLinus Torvalds j_nonzerolen)); 36501da177e4SLinus Torvalds } 36511da177e4SLinus Torvalds cn->bh = NULL; 36521da177e4SLinus Torvalds } 36531da177e4SLinus Torvalds } 36541da177e4SLinus Torvalds cn = cn->hnext; 36551da177e4SLinus Torvalds } 36561da177e4SLinus Torvalds } 36571da177e4SLinus Torvalds 36581da177e4SLinus Torvalds if (bh) { 36591da177e4SLinus Torvalds put_bh(bh); /* get_hash grabs the buffer */ 36601da177e4SLinus Torvalds if (atomic_read(&(bh->b_count)) < 0) { 3661bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 3662bd4c625cSLinus Torvalds "journal-2165: bh->b_count < 0"); 36631da177e4SLinus Torvalds } 36641da177e4SLinus Torvalds } 36651da177e4SLinus Torvalds return 0; 36661da177e4SLinus Torvalds } 36671da177e4SLinus Torvalds 3668bd4c625cSLinus Torvalds void reiserfs_update_inode_transaction(struct inode *inode) 3669bd4c625cSLinus Torvalds { 36701da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb); 36711da177e4SLinus Torvalds REISERFS_I(inode)->i_jl = journal->j_current_jl; 36721da177e4SLinus Torvalds REISERFS_I(inode)->i_trans_id = journal->j_trans_id; 36731da177e4SLinus Torvalds } 36741da177e4SLinus Torvalds 36751da177e4SLinus Torvalds /* 36761da177e4SLinus Torvalds * returns -1 on error, 0 if no commits/barriers were done and 1 36771da177e4SLinus Torvalds * if a transaction was actually committed and the barrier was done 36781da177e4SLinus Torvalds */ 36791da177e4SLinus Torvalds static int __commit_trans_jl(struct inode *inode, unsigned long id, 36801da177e4SLinus Torvalds struct reiserfs_journal_list *jl) 36811da177e4SLinus Torvalds { 36821da177e4SLinus Torvalds struct reiserfs_transaction_handle th; 36831da177e4SLinus Torvalds struct super_block *sb = inode->i_sb; 36841da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(sb); 36851da177e4SLinus Torvalds int ret = 0; 36861da177e4SLinus Torvalds 36871da177e4SLinus Torvalds /* is it from the current transaction, or from an unknown transaction? */ 36881da177e4SLinus Torvalds if (id == journal->j_trans_id) { 36891da177e4SLinus Torvalds jl = journal->j_current_jl; 36901da177e4SLinus Torvalds /* try to let other writers come in and grow this transaction */ 36911da177e4SLinus Torvalds let_transaction_grow(sb, id); 36921da177e4SLinus Torvalds if (journal->j_trans_id != id) { 36931da177e4SLinus Torvalds goto flush_commit_only; 36941da177e4SLinus Torvalds } 36951da177e4SLinus Torvalds 36961da177e4SLinus Torvalds ret = journal_begin(&th, sb, 1); 36971da177e4SLinus Torvalds if (ret) 36981da177e4SLinus Torvalds return ret; 36991da177e4SLinus Torvalds 37001da177e4SLinus Torvalds /* someone might have ended this transaction while we joined */ 37011da177e4SLinus Torvalds if (journal->j_trans_id != id) { 3702bd4c625cSLinus Torvalds reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb), 3703bd4c625cSLinus Torvalds 1); 37041da177e4SLinus Torvalds journal_mark_dirty(&th, sb, SB_BUFFER_WITH_SB(sb)); 37051da177e4SLinus Torvalds ret = journal_end(&th, sb, 1); 37061da177e4SLinus Torvalds goto flush_commit_only; 37071da177e4SLinus Torvalds } 37081da177e4SLinus Torvalds 37091da177e4SLinus Torvalds ret = journal_end_sync(&th, sb, 1); 37101da177e4SLinus Torvalds if (!ret) 37111da177e4SLinus Torvalds ret = 1; 37121da177e4SLinus Torvalds 37131da177e4SLinus Torvalds } else { 37141da177e4SLinus Torvalds /* this gets tricky, we have to make sure the journal list in 37151da177e4SLinus Torvalds * the inode still exists. We know the list is still around 37161da177e4SLinus Torvalds * if we've got a larger transaction id than the oldest list 37171da177e4SLinus Torvalds */ 37181da177e4SLinus Torvalds flush_commit_only: 37191da177e4SLinus Torvalds if (journal_list_still_alive(inode->i_sb, id)) { 37201da177e4SLinus Torvalds /* 37211da177e4SLinus Torvalds * we only set ret to 1 when we know for sure 37221da177e4SLinus Torvalds * the barrier hasn't been started yet on the commit 37231da177e4SLinus Torvalds * block. 37241da177e4SLinus Torvalds */ 37251da177e4SLinus Torvalds if (atomic_read(&jl->j_commit_left) > 1) 37261da177e4SLinus Torvalds ret = 1; 37271da177e4SLinus Torvalds flush_commit_list(sb, jl, 1); 37281da177e4SLinus Torvalds if (journal->j_errno) 37291da177e4SLinus Torvalds ret = journal->j_errno; 37301da177e4SLinus Torvalds } 37311da177e4SLinus Torvalds } 37321da177e4SLinus Torvalds /* otherwise the list is gone, and long since committed */ 37331da177e4SLinus Torvalds return ret; 37341da177e4SLinus Torvalds } 37351da177e4SLinus Torvalds 3736bd4c625cSLinus Torvalds int reiserfs_commit_for_inode(struct inode *inode) 3737bd4c625cSLinus Torvalds { 37381da177e4SLinus Torvalds unsigned long id = REISERFS_I(inode)->i_trans_id; 37391da177e4SLinus Torvalds struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl; 37401da177e4SLinus Torvalds 37411da177e4SLinus Torvalds /* for the whole inode, assume unset id means it was 37421da177e4SLinus Torvalds * changed in the current transaction. More conservative 37431da177e4SLinus Torvalds */ 37441da177e4SLinus Torvalds if (!id || !jl) { 37451da177e4SLinus Torvalds reiserfs_update_inode_transaction(inode); 37461da177e4SLinus Torvalds id = REISERFS_I(inode)->i_trans_id; 37471da177e4SLinus Torvalds /* jl will be updated in __commit_trans_jl */ 37481da177e4SLinus Torvalds } 37491da177e4SLinus Torvalds 37501da177e4SLinus Torvalds return __commit_trans_jl(inode, id, jl); 37511da177e4SLinus Torvalds } 37521da177e4SLinus Torvalds 37531da177e4SLinus Torvalds void reiserfs_restore_prepared_buffer(struct super_block *p_s_sb, 3754bd4c625cSLinus Torvalds struct buffer_head *bh) 3755bd4c625cSLinus Torvalds { 37561da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 37571da177e4SLinus Torvalds PROC_INFO_INC(p_s_sb, journal.restore_prepared); 37581da177e4SLinus Torvalds if (!bh) { 37591da177e4SLinus Torvalds return; 37601da177e4SLinus Torvalds } 37611da177e4SLinus Torvalds if (test_clear_buffer_journal_restore_dirty(bh) && 37621da177e4SLinus Torvalds buffer_journal_dirty(bh)) { 37631da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn; 37641da177e4SLinus Torvalds cn = get_journal_hash_dev(p_s_sb, 37651da177e4SLinus Torvalds journal->j_list_hash_table, 37661da177e4SLinus Torvalds bh->b_blocknr); 37671da177e4SLinus Torvalds if (cn && can_dirty(cn)) { 37681da177e4SLinus Torvalds set_buffer_journal_test(bh); 37691da177e4SLinus Torvalds mark_buffer_dirty(bh); 37701da177e4SLinus Torvalds } 37711da177e4SLinus Torvalds } 37721da177e4SLinus Torvalds clear_buffer_journal_prepared(bh); 37731da177e4SLinus Torvalds } 37741da177e4SLinus Torvalds 37751da177e4SLinus Torvalds extern struct tree_balance *cur_tb; 37761da177e4SLinus Torvalds /* 37771da177e4SLinus Torvalds ** before we can change a metadata block, we have to make sure it won't 37781da177e4SLinus Torvalds ** be written to disk while we are altering it. So, we must: 37791da177e4SLinus Torvalds ** clean it 37801da177e4SLinus Torvalds ** wait on it. 37811da177e4SLinus Torvalds ** 37821da177e4SLinus Torvalds */ 37831da177e4SLinus Torvalds int reiserfs_prepare_for_journal(struct super_block *p_s_sb, 3784bd4c625cSLinus Torvalds struct buffer_head *bh, int wait) 3785bd4c625cSLinus Torvalds { 37861da177e4SLinus Torvalds PROC_INFO_INC(p_s_sb, journal.prepare); 37871da177e4SLinus Torvalds 37881da177e4SLinus Torvalds if (test_set_buffer_locked(bh)) { 37891da177e4SLinus Torvalds if (!wait) 37901da177e4SLinus Torvalds return 0; 37911da177e4SLinus Torvalds lock_buffer(bh); 37921da177e4SLinus Torvalds } 37931da177e4SLinus Torvalds set_buffer_journal_prepared(bh); 37941da177e4SLinus Torvalds if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) { 37951da177e4SLinus Torvalds clear_buffer_journal_test(bh); 37961da177e4SLinus Torvalds set_buffer_journal_restore_dirty(bh); 37971da177e4SLinus Torvalds } 37981da177e4SLinus Torvalds unlock_buffer(bh); 37991da177e4SLinus Torvalds return 1; 38001da177e4SLinus Torvalds } 38011da177e4SLinus Torvalds 3802bd4c625cSLinus Torvalds static void flush_old_journal_lists(struct super_block *s) 3803bd4c625cSLinus Torvalds { 38041da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(s); 38051da177e4SLinus Torvalds struct reiserfs_journal_list *jl; 38061da177e4SLinus Torvalds struct list_head *entry; 38071da177e4SLinus Torvalds time_t now = get_seconds(); 38081da177e4SLinus Torvalds 38091da177e4SLinus Torvalds while (!list_empty(&journal->j_journal_list)) { 38101da177e4SLinus Torvalds entry = journal->j_journal_list.next; 38111da177e4SLinus Torvalds jl = JOURNAL_LIST_ENTRY(entry); 38121da177e4SLinus Torvalds /* this check should always be run, to send old lists to disk */ 38131da177e4SLinus Torvalds if (jl->j_timestamp < (now - (JOURNAL_MAX_TRANS_AGE * 4))) { 38141da177e4SLinus Torvalds flush_used_journal_lists(s, jl); 38151da177e4SLinus Torvalds } else { 38161da177e4SLinus Torvalds break; 38171da177e4SLinus Torvalds } 38181da177e4SLinus Torvalds } 38191da177e4SLinus Torvalds } 38201da177e4SLinus Torvalds 38211da177e4SLinus Torvalds /* 38221da177e4SLinus Torvalds ** long and ugly. If flush, will not return until all commit 38231da177e4SLinus Torvalds ** blocks and all real buffers in the trans are on disk. 38241da177e4SLinus Torvalds ** If no_async, won't return until all commit blocks are on disk. 38251da177e4SLinus Torvalds ** 38261da177e4SLinus Torvalds ** keep reading, there are comments as you go along 38271da177e4SLinus Torvalds ** 38281da177e4SLinus Torvalds ** If the journal is aborted, we just clean up. Things like flushing 38291da177e4SLinus Torvalds ** journal lists, etc just won't happen. 38301da177e4SLinus Torvalds */ 3831bd4c625cSLinus Torvalds static int do_journal_end(struct reiserfs_transaction_handle *th, 3832bd4c625cSLinus Torvalds struct super_block *p_s_sb, unsigned long nblocks, 3833bd4c625cSLinus Torvalds int flags) 3834bd4c625cSLinus Torvalds { 38351da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); 38361da177e4SLinus Torvalds struct reiserfs_journal_cnode *cn, *next, *jl_cn; 38371da177e4SLinus Torvalds struct reiserfs_journal_cnode *last_cn = NULL; 38381da177e4SLinus Torvalds struct reiserfs_journal_desc *desc; 38391da177e4SLinus Torvalds struct reiserfs_journal_commit *commit; 38401da177e4SLinus Torvalds struct buffer_head *c_bh; /* commit bh */ 38411da177e4SLinus Torvalds struct buffer_head *d_bh; /* desc bh */ 38421da177e4SLinus Torvalds int cur_write_start = 0; /* start index of current log write */ 38431da177e4SLinus Torvalds int old_start; 38441da177e4SLinus Torvalds int i; 38451da177e4SLinus Torvalds int flush = flags & FLUSH_ALL; 38461da177e4SLinus Torvalds int wait_on_commit = flags & WAIT; 38471da177e4SLinus Torvalds struct reiserfs_journal_list *jl, *temp_jl; 38481da177e4SLinus Torvalds struct list_head *entry, *safe; 38491da177e4SLinus Torvalds unsigned long jindex; 38501da177e4SLinus Torvalds unsigned long commit_trans_id; 38511da177e4SLinus Torvalds int trans_half; 38521da177e4SLinus Torvalds 38531da177e4SLinus Torvalds BUG_ON(th->t_refcount > 1); 38541da177e4SLinus Torvalds BUG_ON(!th->t_trans_id); 38551da177e4SLinus Torvalds 385622e2c507SJens Axboe put_fs_excl(); 38571da177e4SLinus Torvalds current->journal_info = th->t_handle_save; 38581da177e4SLinus Torvalds reiserfs_check_lock_depth(p_s_sb, "journal end"); 38591da177e4SLinus Torvalds if (journal->j_len == 0) { 3860bd4c625cSLinus Torvalds reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb), 3861bd4c625cSLinus Torvalds 1); 38621da177e4SLinus Torvalds journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb)); 38631da177e4SLinus Torvalds } 38641da177e4SLinus Torvalds 38651da177e4SLinus Torvalds lock_journal(p_s_sb); 38661da177e4SLinus Torvalds if (journal->j_next_full_flush) { 38671da177e4SLinus Torvalds flags |= FLUSH_ALL; 38681da177e4SLinus Torvalds flush = 1; 38691da177e4SLinus Torvalds } 38701da177e4SLinus Torvalds if (journal->j_next_async_flush) { 38711da177e4SLinus Torvalds flags |= COMMIT_NOW | WAIT; 38721da177e4SLinus Torvalds wait_on_commit = 1; 38731da177e4SLinus Torvalds } 38741da177e4SLinus Torvalds 38751da177e4SLinus Torvalds /* check_journal_end locks the journal, and unlocks if it does not return 1 38761da177e4SLinus Torvalds ** it tells us if we should continue with the journal_end, or just return 38771da177e4SLinus Torvalds */ 38781da177e4SLinus Torvalds if (!check_journal_end(th, p_s_sb, nblocks, flags)) { 38791da177e4SLinus Torvalds p_s_sb->s_dirt = 1; 38801da177e4SLinus Torvalds wake_queued_writers(p_s_sb); 38811da177e4SLinus Torvalds reiserfs_async_progress_wait(p_s_sb); 38821da177e4SLinus Torvalds goto out; 38831da177e4SLinus Torvalds } 38841da177e4SLinus Torvalds 38851da177e4SLinus Torvalds /* check_journal_end might set these, check again */ 38861da177e4SLinus Torvalds if (journal->j_next_full_flush) { 38871da177e4SLinus Torvalds flush = 1; 38881da177e4SLinus Torvalds } 38891da177e4SLinus Torvalds 38901da177e4SLinus Torvalds /* 38911da177e4SLinus Torvalds ** j must wait means we have to flush the log blocks, and the real blocks for 38921da177e4SLinus Torvalds ** this transaction 38931da177e4SLinus Torvalds */ 38941da177e4SLinus Torvalds if (journal->j_must_wait > 0) { 38951da177e4SLinus Torvalds flush = 1; 38961da177e4SLinus Torvalds } 38971da177e4SLinus Torvalds #ifdef REISERFS_PREALLOCATE 3898ef43bc4fSJan Kara /* quota ops might need to nest, setup the journal_info pointer for them 3899ef43bc4fSJan Kara * and raise the refcount so that it is > 0. */ 39001da177e4SLinus Torvalds current->journal_info = th; 3901ef43bc4fSJan Kara th->t_refcount++; 39021da177e4SLinus Torvalds reiserfs_discard_all_prealloc(th); /* it should not involve new blocks into 39031da177e4SLinus Torvalds * the transaction */ 3904ef43bc4fSJan Kara th->t_refcount--; 39051da177e4SLinus Torvalds current->journal_info = th->t_handle_save; 39061da177e4SLinus Torvalds #endif 39071da177e4SLinus Torvalds 39081da177e4SLinus Torvalds /* setup description block */ 3909bd4c625cSLinus Torvalds d_bh = 3910bd4c625cSLinus Torvalds journal_getblk(p_s_sb, 3911bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + 3912bd4c625cSLinus Torvalds journal->j_start); 39131da177e4SLinus Torvalds set_buffer_uptodate(d_bh); 39141da177e4SLinus Torvalds desc = (struct reiserfs_journal_desc *)(d_bh)->b_data; 39151da177e4SLinus Torvalds memset(d_bh->b_data, 0, d_bh->b_size); 39161da177e4SLinus Torvalds memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8); 39171da177e4SLinus Torvalds set_desc_trans_id(desc, journal->j_trans_id); 39181da177e4SLinus Torvalds 39191da177e4SLinus Torvalds /* setup commit block. Don't write (keep it clean too) this one until after everyone else is written */ 39201da177e4SLinus Torvalds c_bh = journal_getblk(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + 3921bd4c625cSLinus Torvalds ((journal->j_start + journal->j_len + 3922bd4c625cSLinus Torvalds 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb))); 39231da177e4SLinus Torvalds commit = (struct reiserfs_journal_commit *)c_bh->b_data; 39241da177e4SLinus Torvalds memset(c_bh->b_data, 0, c_bh->b_size); 39251da177e4SLinus Torvalds set_commit_trans_id(commit, journal->j_trans_id); 39261da177e4SLinus Torvalds set_buffer_uptodate(c_bh); 39271da177e4SLinus Torvalds 39281da177e4SLinus Torvalds /* init this journal list */ 39291da177e4SLinus Torvalds jl = journal->j_current_jl; 39301da177e4SLinus Torvalds 39311da177e4SLinus Torvalds /* we lock the commit before doing anything because 39321da177e4SLinus Torvalds * we want to make sure nobody tries to run flush_commit_list until 39331da177e4SLinus Torvalds * the new transaction is fully setup, and we've already flushed the 39341da177e4SLinus Torvalds * ordered bh list 39351da177e4SLinus Torvalds */ 39361da177e4SLinus Torvalds down(&jl->j_commit_lock); 39371da177e4SLinus Torvalds 39381da177e4SLinus Torvalds /* save the transaction id in case we need to commit it later */ 39391da177e4SLinus Torvalds commit_trans_id = jl->j_trans_id; 39401da177e4SLinus Torvalds 39411da177e4SLinus Torvalds atomic_set(&jl->j_older_commits_done, 0); 39421da177e4SLinus Torvalds jl->j_trans_id = journal->j_trans_id; 39431da177e4SLinus Torvalds jl->j_timestamp = journal->j_trans_start_time; 39441da177e4SLinus Torvalds jl->j_commit_bh = c_bh; 39451da177e4SLinus Torvalds jl->j_start = journal->j_start; 39461da177e4SLinus Torvalds jl->j_len = journal->j_len; 39471da177e4SLinus Torvalds atomic_set(&jl->j_nonzerolen, journal->j_len); 39481da177e4SLinus Torvalds atomic_set(&jl->j_commit_left, journal->j_len + 2); 39491da177e4SLinus Torvalds jl->j_realblock = NULL; 39501da177e4SLinus Torvalds 39511da177e4SLinus Torvalds /* The ENTIRE FOR LOOP MUST not cause schedule to occur. 39521da177e4SLinus Torvalds ** for each real block, add it to the journal list hash, 39531da177e4SLinus Torvalds ** copy into real block index array in the commit or desc block 39541da177e4SLinus Torvalds */ 39551da177e4SLinus Torvalds trans_half = journal_trans_half(p_s_sb->s_blocksize); 39561da177e4SLinus Torvalds for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) { 39571da177e4SLinus Torvalds if (buffer_journaled(cn->bh)) { 39581da177e4SLinus Torvalds jl_cn = get_cnode(p_s_sb); 39591da177e4SLinus Torvalds if (!jl_cn) { 3960bd4c625cSLinus Torvalds reiserfs_panic(p_s_sb, 3961bd4c625cSLinus Torvalds "journal-1676, get_cnode returned NULL\n"); 39621da177e4SLinus Torvalds } 39631da177e4SLinus Torvalds if (i == 0) { 39641da177e4SLinus Torvalds jl->j_realblock = jl_cn; 39651da177e4SLinus Torvalds } 39661da177e4SLinus Torvalds jl_cn->prev = last_cn; 39671da177e4SLinus Torvalds jl_cn->next = NULL; 39681da177e4SLinus Torvalds if (last_cn) { 39691da177e4SLinus Torvalds last_cn->next = jl_cn; 39701da177e4SLinus Torvalds } 39711da177e4SLinus Torvalds last_cn = jl_cn; 39721da177e4SLinus Torvalds /* make sure the block we are trying to log is not a block 39731da177e4SLinus Torvalds of journal or reserved area */ 39741da177e4SLinus Torvalds 3975bd4c625cSLinus Torvalds if (is_block_in_log_or_reserved_area 3976bd4c625cSLinus Torvalds (p_s_sb, cn->bh->b_blocknr)) { 3977bd4c625cSLinus Torvalds reiserfs_panic(p_s_sb, 3978bd4c625cSLinus Torvalds "journal-2332: Trying to log block %lu, which is a log block\n", 3979bd4c625cSLinus Torvalds cn->bh->b_blocknr); 39801da177e4SLinus Torvalds } 39811da177e4SLinus Torvalds jl_cn->blocknr = cn->bh->b_blocknr; 39821da177e4SLinus Torvalds jl_cn->state = 0; 39831da177e4SLinus Torvalds jl_cn->sb = p_s_sb; 39841da177e4SLinus Torvalds jl_cn->bh = cn->bh; 39851da177e4SLinus Torvalds jl_cn->jlist = jl; 39861da177e4SLinus Torvalds insert_journal_hash(journal->j_list_hash_table, jl_cn); 39871da177e4SLinus Torvalds if (i < trans_half) { 3988bd4c625cSLinus Torvalds desc->j_realblock[i] = 3989bd4c625cSLinus Torvalds cpu_to_le32(cn->bh->b_blocknr); 39901da177e4SLinus Torvalds } else { 3991bd4c625cSLinus Torvalds commit->j_realblock[i - trans_half] = 3992bd4c625cSLinus Torvalds cpu_to_le32(cn->bh->b_blocknr); 39931da177e4SLinus Torvalds } 39941da177e4SLinus Torvalds } else { 39951da177e4SLinus Torvalds i--; 39961da177e4SLinus Torvalds } 39971da177e4SLinus Torvalds } 39981da177e4SLinus Torvalds set_desc_trans_len(desc, journal->j_len); 39991da177e4SLinus Torvalds set_desc_mount_id(desc, journal->j_mount_id); 40001da177e4SLinus Torvalds set_desc_trans_id(desc, journal->j_trans_id); 40011da177e4SLinus Torvalds set_commit_trans_len(commit, journal->j_len); 40021da177e4SLinus Torvalds 40031da177e4SLinus Torvalds /* special check in case all buffers in the journal were marked for not logging */ 40041da177e4SLinus Torvalds if (journal->j_len == 0) { 40051da177e4SLinus Torvalds BUG(); 40061da177e4SLinus Torvalds } 40071da177e4SLinus Torvalds 40081da177e4SLinus Torvalds /* we're about to dirty all the log blocks, mark the description block 40091da177e4SLinus Torvalds * dirty now too. Don't mark the commit block dirty until all the 40101da177e4SLinus Torvalds * others are on disk 40111da177e4SLinus Torvalds */ 40121da177e4SLinus Torvalds mark_buffer_dirty(d_bh); 40131da177e4SLinus Torvalds 40141da177e4SLinus Torvalds /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */ 40151da177e4SLinus Torvalds cur_write_start = journal->j_start; 40161da177e4SLinus Torvalds cn = journal->j_first; 40171da177e4SLinus Torvalds jindex = 1; /* start at one so we don't get the desc again */ 40181da177e4SLinus Torvalds while (cn) { 40191da177e4SLinus Torvalds clear_buffer_journal_new(cn->bh); 40201da177e4SLinus Torvalds /* copy all the real blocks into log area. dirty log blocks */ 40211da177e4SLinus Torvalds if (buffer_journaled(cn->bh)) { 40221da177e4SLinus Torvalds struct buffer_head *tmp_bh; 40231da177e4SLinus Torvalds char *addr; 40241da177e4SLinus Torvalds struct page *page; 4025bd4c625cSLinus Torvalds tmp_bh = 4026bd4c625cSLinus Torvalds journal_getblk(p_s_sb, 4027bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + 4028bd4c625cSLinus Torvalds ((cur_write_start + 4029bd4c625cSLinus Torvalds jindex) % 4030bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_SIZE(p_s_sb))); 40311da177e4SLinus Torvalds set_buffer_uptodate(tmp_bh); 40321da177e4SLinus Torvalds page = cn->bh->b_page; 40331da177e4SLinus Torvalds addr = kmap(page); 4034bd4c625cSLinus Torvalds memcpy(tmp_bh->b_data, 4035bd4c625cSLinus Torvalds addr + offset_in_page(cn->bh->b_data), 40361da177e4SLinus Torvalds cn->bh->b_size); 40371da177e4SLinus Torvalds kunmap(page); 40381da177e4SLinus Torvalds mark_buffer_dirty(tmp_bh); 40391da177e4SLinus Torvalds jindex++; 40401da177e4SLinus Torvalds set_buffer_journal_dirty(cn->bh); 40411da177e4SLinus Torvalds clear_buffer_journaled(cn->bh); 40421da177e4SLinus Torvalds } else { 40431da177e4SLinus Torvalds /* JDirty cleared sometime during transaction. don't log this one */ 4044bd4c625cSLinus Torvalds reiserfs_warning(p_s_sb, 4045bd4c625cSLinus Torvalds "journal-2048: do_journal_end: BAD, buffer in journal hash, but not JDirty!"); 40461da177e4SLinus Torvalds brelse(cn->bh); 40471da177e4SLinus Torvalds } 40481da177e4SLinus Torvalds next = cn->next; 40491da177e4SLinus Torvalds free_cnode(p_s_sb, cn); 40501da177e4SLinus Torvalds cn = next; 40511da177e4SLinus Torvalds cond_resched(); 40521da177e4SLinus Torvalds } 40531da177e4SLinus Torvalds 40541da177e4SLinus Torvalds /* we are done with both the c_bh and d_bh, but 40551da177e4SLinus Torvalds ** c_bh must be written after all other commit blocks, 40561da177e4SLinus Torvalds ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1. 40571da177e4SLinus Torvalds */ 40581da177e4SLinus Torvalds 40591da177e4SLinus Torvalds journal->j_current_jl = alloc_journal_list(p_s_sb); 40601da177e4SLinus Torvalds 40611da177e4SLinus Torvalds /* now it is safe to insert this transaction on the main list */ 40621da177e4SLinus Torvalds list_add_tail(&jl->j_list, &journal->j_journal_list); 40631da177e4SLinus Torvalds list_add_tail(&jl->j_working_list, &journal->j_working_list); 40641da177e4SLinus Torvalds journal->j_num_work_lists++; 40651da177e4SLinus Torvalds 40661da177e4SLinus Torvalds /* reset journal values for the next transaction */ 40671da177e4SLinus Torvalds old_start = journal->j_start; 4068bd4c625cSLinus Torvalds journal->j_start = 4069bd4c625cSLinus Torvalds (journal->j_start + journal->j_len + 4070bd4c625cSLinus Torvalds 2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb); 40711da177e4SLinus Torvalds atomic_set(&(journal->j_wcount), 0); 40721da177e4SLinus Torvalds journal->j_bcount = 0; 40731da177e4SLinus Torvalds journal->j_last = NULL; 40741da177e4SLinus Torvalds journal->j_first = NULL; 40751da177e4SLinus Torvalds journal->j_len = 0; 40761da177e4SLinus Torvalds journal->j_trans_start_time = 0; 40771da177e4SLinus Torvalds journal->j_trans_id++; 40781da177e4SLinus Torvalds journal->j_current_jl->j_trans_id = journal->j_trans_id; 40791da177e4SLinus Torvalds journal->j_must_wait = 0; 40801da177e4SLinus Torvalds journal->j_len_alloc = 0; 40811da177e4SLinus Torvalds journal->j_next_full_flush = 0; 40821da177e4SLinus Torvalds journal->j_next_async_flush = 0; 40831da177e4SLinus Torvalds init_journal_hash(p_s_sb); 40841da177e4SLinus Torvalds 40851da177e4SLinus Torvalds // make sure reiserfs_add_jh sees the new current_jl before we 40861da177e4SLinus Torvalds // write out the tails 40871da177e4SLinus Torvalds smp_mb(); 40881da177e4SLinus Torvalds 40891da177e4SLinus Torvalds /* tail conversion targets have to hit the disk before we end the 40901da177e4SLinus Torvalds * transaction. Otherwise a later transaction might repack the tail 40911da177e4SLinus Torvalds * before this transaction commits, leaving the data block unflushed and 40921da177e4SLinus Torvalds * clean, if we crash before the later transaction commits, the data block 40931da177e4SLinus Torvalds * is lost. 40941da177e4SLinus Torvalds */ 40951da177e4SLinus Torvalds if (!list_empty(&jl->j_tail_bh_list)) { 40961da177e4SLinus Torvalds unlock_kernel(); 40971da177e4SLinus Torvalds write_ordered_buffers(&journal->j_dirty_buffers_lock, 40981da177e4SLinus Torvalds journal, jl, &jl->j_tail_bh_list); 40991da177e4SLinus Torvalds lock_kernel(); 41001da177e4SLinus Torvalds } 41011da177e4SLinus Torvalds if (!list_empty(&jl->j_tail_bh_list)) 41021da177e4SLinus Torvalds BUG(); 41031da177e4SLinus Torvalds up(&jl->j_commit_lock); 41041da177e4SLinus Torvalds 41051da177e4SLinus Torvalds /* honor the flush wishes from the caller, simple commits can 41061da177e4SLinus Torvalds ** be done outside the journal lock, they are done below 41071da177e4SLinus Torvalds ** 41081da177e4SLinus Torvalds ** if we don't flush the commit list right now, we put it into 41091da177e4SLinus Torvalds ** the work queue so the people waiting on the async progress work 41101da177e4SLinus Torvalds ** queue don't wait for this proc to flush journal lists and such. 41111da177e4SLinus Torvalds */ 41121da177e4SLinus Torvalds if (flush) { 41131da177e4SLinus Torvalds flush_commit_list(p_s_sb, jl, 1); 41141da177e4SLinus Torvalds flush_journal_list(p_s_sb, jl, 1); 41151da177e4SLinus Torvalds } else if (!(jl->j_state & LIST_COMMIT_PENDING)) 41161da177e4SLinus Torvalds queue_delayed_work(commit_wq, &journal->j_work, HZ / 10); 41171da177e4SLinus Torvalds 41181da177e4SLinus Torvalds /* if the next transaction has any chance of wrapping, flush 41191da177e4SLinus Torvalds ** transactions that might get overwritten. If any journal lists are very 41201da177e4SLinus Torvalds ** old flush them as well. 41211da177e4SLinus Torvalds */ 41221da177e4SLinus Torvalds first_jl: 41231da177e4SLinus Torvalds list_for_each_safe(entry, safe, &journal->j_journal_list) { 41241da177e4SLinus Torvalds temp_jl = JOURNAL_LIST_ENTRY(entry); 41251da177e4SLinus Torvalds if (journal->j_start <= temp_jl->j_start) { 41261da177e4SLinus Torvalds if ((journal->j_start + journal->j_trans_max + 1) >= 4127bd4c625cSLinus Torvalds temp_jl->j_start) { 41281da177e4SLinus Torvalds flush_used_journal_lists(p_s_sb, temp_jl); 41291da177e4SLinus Torvalds goto first_jl; 41301da177e4SLinus Torvalds } else if ((journal->j_start + 41311da177e4SLinus Torvalds journal->j_trans_max + 1) < 4132bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_SIZE(p_s_sb)) { 41331da177e4SLinus Torvalds /* if we don't cross into the next transaction and we don't 41341da177e4SLinus Torvalds * wrap, there is no way we can overlap any later transactions 41351da177e4SLinus Torvalds * break now 41361da177e4SLinus Torvalds */ 41371da177e4SLinus Torvalds break; 41381da177e4SLinus Torvalds } 41391da177e4SLinus Torvalds } else if ((journal->j_start + 41401da177e4SLinus Torvalds journal->j_trans_max + 1) > 4141bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_SIZE(p_s_sb)) { 41421da177e4SLinus Torvalds if (((journal->j_start + journal->j_trans_max + 1) % 4143bd4c625cSLinus Torvalds SB_ONDISK_JOURNAL_SIZE(p_s_sb)) >= 4144bd4c625cSLinus Torvalds temp_jl->j_start) { 41451da177e4SLinus Torvalds flush_used_journal_lists(p_s_sb, temp_jl); 41461da177e4SLinus Torvalds goto first_jl; 41471da177e4SLinus Torvalds } else { 41481da177e4SLinus Torvalds /* we don't overlap anything from out start to the end of the 41491da177e4SLinus Torvalds * log, and our wrapped portion doesn't overlap anything at 41501da177e4SLinus Torvalds * the start of the log. We can break 41511da177e4SLinus Torvalds */ 41521da177e4SLinus Torvalds break; 41531da177e4SLinus Torvalds } 41541da177e4SLinus Torvalds } 41551da177e4SLinus Torvalds } 41561da177e4SLinus Torvalds flush_old_journal_lists(p_s_sb); 41571da177e4SLinus Torvalds 4158bd4c625cSLinus Torvalds journal->j_current_jl->j_list_bitmap = 4159bd4c625cSLinus Torvalds get_list_bitmap(p_s_sb, journal->j_current_jl); 41601da177e4SLinus Torvalds 41611da177e4SLinus Torvalds if (!(journal->j_current_jl->j_list_bitmap)) { 4162bd4c625cSLinus Torvalds reiserfs_panic(p_s_sb, 4163bd4c625cSLinus Torvalds "journal-1996: do_journal_end, could not get a list bitmap\n"); 41641da177e4SLinus Torvalds } 41651da177e4SLinus Torvalds 41661da177e4SLinus Torvalds atomic_set(&(journal->j_jlock), 0); 41671da177e4SLinus Torvalds unlock_journal(p_s_sb); 41681da177e4SLinus Torvalds /* wake up any body waiting to join. */ 41691da177e4SLinus Torvalds clear_bit(J_WRITERS_QUEUED, &journal->j_state); 41701da177e4SLinus Torvalds wake_up(&(journal->j_join_wait)); 41711da177e4SLinus Torvalds 41721da177e4SLinus Torvalds if (!flush && wait_on_commit && 41731da177e4SLinus Torvalds journal_list_still_alive(p_s_sb, commit_trans_id)) { 41741da177e4SLinus Torvalds flush_commit_list(p_s_sb, jl, 1); 41751da177e4SLinus Torvalds } 41761da177e4SLinus Torvalds out: 41771da177e4SLinus Torvalds reiserfs_check_lock_depth(p_s_sb, "journal end2"); 41781da177e4SLinus Torvalds 41791da177e4SLinus Torvalds memset(th, 0, sizeof(*th)); 41801da177e4SLinus Torvalds /* Re-set th->t_super, so we can properly keep track of how many 41811da177e4SLinus Torvalds * persistent transactions there are. We need to do this so if this 41821da177e4SLinus Torvalds * call is part of a failed restart_transaction, we can free it later */ 41831da177e4SLinus Torvalds th->t_super = p_s_sb; 41841da177e4SLinus Torvalds 41851da177e4SLinus Torvalds return journal->j_errno; 41861da177e4SLinus Torvalds } 41871da177e4SLinus Torvalds 4188bd4c625cSLinus Torvalds static void __reiserfs_journal_abort_hard(struct super_block *sb) 41891da177e4SLinus Torvalds { 41901da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(sb); 41911da177e4SLinus Torvalds if (test_bit(J_ABORTED, &journal->j_state)) 41921da177e4SLinus Torvalds return; 41931da177e4SLinus Torvalds 41941da177e4SLinus Torvalds printk(KERN_CRIT "REISERFS: Aborting journal for filesystem on %s\n", 41951da177e4SLinus Torvalds reiserfs_bdevname(sb)); 41961da177e4SLinus Torvalds 41971da177e4SLinus Torvalds sb->s_flags |= MS_RDONLY; 41981da177e4SLinus Torvalds set_bit(J_ABORTED, &journal->j_state); 41991da177e4SLinus Torvalds 42001da177e4SLinus Torvalds #ifdef CONFIG_REISERFS_CHECK 42011da177e4SLinus Torvalds dump_stack(); 42021da177e4SLinus Torvalds #endif 42031da177e4SLinus Torvalds } 42041da177e4SLinus Torvalds 4205bd4c625cSLinus Torvalds static void __reiserfs_journal_abort_soft(struct super_block *sb, int errno) 42061da177e4SLinus Torvalds { 42071da177e4SLinus Torvalds struct reiserfs_journal *journal = SB_JOURNAL(sb); 42081da177e4SLinus Torvalds if (test_bit(J_ABORTED, &journal->j_state)) 42091da177e4SLinus Torvalds return; 42101da177e4SLinus Torvalds 42111da177e4SLinus Torvalds if (!journal->j_errno) 42121da177e4SLinus Torvalds journal->j_errno = errno; 42131da177e4SLinus Torvalds 42141da177e4SLinus Torvalds __reiserfs_journal_abort_hard(sb); 42151da177e4SLinus Torvalds } 42161da177e4SLinus Torvalds 4217bd4c625cSLinus Torvalds void reiserfs_journal_abort(struct super_block *sb, int errno) 42181da177e4SLinus Torvalds { 42191da177e4SLinus Torvalds return __reiserfs_journal_abort_soft(sb, errno); 42201da177e4SLinus Torvalds } 4221