1470decc6SDave Kleikamp /* 2f7f4bccbSMingming Cao * linux/fs/jbd2/journal.c 3470decc6SDave Kleikamp * 4470decc6SDave Kleikamp * Written by Stephen C. Tweedie <sct@redhat.com>, 1998 5470decc6SDave Kleikamp * 6470decc6SDave Kleikamp * Copyright 1998 Red Hat corp --- All Rights Reserved 7470decc6SDave Kleikamp * 8470decc6SDave Kleikamp * This file is part of the Linux kernel and is made available under 9470decc6SDave Kleikamp * the terms of the GNU General Public License, version 2, or at your 10470decc6SDave Kleikamp * option, any later version, incorporated herein by reference. 11470decc6SDave Kleikamp * 12470decc6SDave Kleikamp * Generic filesystem journal-writing code; part of the ext2fs 13470decc6SDave Kleikamp * journaling system. 14470decc6SDave Kleikamp * 15470decc6SDave Kleikamp * This file manages journals: areas of disk reserved for logging 16470decc6SDave Kleikamp * transactional updates. This includes the kernel journaling thread 17470decc6SDave Kleikamp * which is responsible for scheduling updates to the log. 18470decc6SDave Kleikamp * 19470decc6SDave Kleikamp * We do not actually manage the physical storage of the journal in this 20470decc6SDave Kleikamp * file: that is left to a per-journal policy function, which allows us 21470decc6SDave Kleikamp * to store the journal within a filesystem-specified area for ext2 22470decc6SDave Kleikamp * journaling (ext2 can use a reserved inode for storing the log). 23470decc6SDave Kleikamp */ 24470decc6SDave Kleikamp 25470decc6SDave Kleikamp #include <linux/module.h> 26470decc6SDave Kleikamp #include <linux/time.h> 27470decc6SDave Kleikamp #include <linux/fs.h> 28f7f4bccbSMingming Cao #include <linux/jbd2.h> 29470decc6SDave Kleikamp #include <linux/errno.h> 30470decc6SDave Kleikamp #include <linux/slab.h> 31470decc6SDave Kleikamp #include <linux/init.h> 32470decc6SDave Kleikamp #include <linux/mm.h> 337dfb7103SNigel Cunningham #include <linux/freezer.h> 34470decc6SDave Kleikamp #include <linux/pagemap.h> 35470decc6SDave Kleikamp #include <linux/kthread.h> 36470decc6SDave Kleikamp #include <linux/poison.h> 37470decc6SDave Kleikamp #include <linux/proc_fs.h> 388e85fb3fSJohann Lombardi #include <linux/seq_file.h> 39c225aa57SSimon Holm Thøgersen #include <linux/math64.h> 40879c5e6bSTheodore Ts'o #include <linux/hash.h> 41d2eecb03STheodore Ts'o #include <linux/log2.h> 42d2eecb03STheodore Ts'o #include <linux/vmalloc.h> 4347def826STheodore Ts'o #include <linux/backing-dev.h> 4439e3ac25SBrian King #include <linux/bitops.h> 45670be5a7STheodore Ts'o #include <linux/ratelimit.h> 46879c5e6bSTheodore Ts'o 47879c5e6bSTheodore Ts'o #define CREATE_TRACE_POINTS 48879c5e6bSTheodore Ts'o #include <trace/events/jbd2.h> 49470decc6SDave Kleikamp 50470decc6SDave Kleikamp #include <asm/uaccess.h> 51470decc6SDave Kleikamp #include <asm/page.h> 52470decc6SDave Kleikamp 53b6e96d00STheodore Ts'o #ifdef CONFIG_JBD2_DEBUG 54b6e96d00STheodore Ts'o ushort jbd2_journal_enable_debug __read_mostly; 55b6e96d00STheodore Ts'o EXPORT_SYMBOL(jbd2_journal_enable_debug); 56b6e96d00STheodore Ts'o 57b6e96d00STheodore Ts'o module_param_named(jbd2_debug, jbd2_journal_enable_debug, ushort, 0644); 58b6e96d00STheodore Ts'o MODULE_PARM_DESC(jbd2_debug, "Debugging level for jbd2"); 59b6e96d00STheodore Ts'o #endif 60b6e96d00STheodore Ts'o 61f7f4bccbSMingming Cao EXPORT_SYMBOL(jbd2_journal_extend); 62f7f4bccbSMingming Cao EXPORT_SYMBOL(jbd2_journal_stop); 63f7f4bccbSMingming Cao EXPORT_SYMBOL(jbd2_journal_lock_updates); 64f7f4bccbSMingming Cao EXPORT_SYMBOL(jbd2_journal_unlock_updates); 65f7f4bccbSMingming Cao EXPORT_SYMBOL(jbd2_journal_get_write_access); 66f7f4bccbSMingming Cao EXPORT_SYMBOL(jbd2_journal_get_create_access); 67f7f4bccbSMingming Cao EXPORT_SYMBOL(jbd2_journal_get_undo_access); 68e06c8227SJoel Becker EXPORT_SYMBOL(jbd2_journal_set_triggers); 69f7f4bccbSMingming Cao EXPORT_SYMBOL(jbd2_journal_dirty_metadata); 70f7f4bccbSMingming Cao EXPORT_SYMBOL(jbd2_journal_forget); 71470decc6SDave Kleikamp #if 0 72470decc6SDave Kleikamp EXPORT_SYMBOL(journal_sync_buffer); 73470decc6SDave Kleikamp #endif 74f7f4bccbSMingming Cao EXPORT_SYMBOL(jbd2_journal_flush); 75f7f4bccbSMingming Cao EXPORT_SYMBOL(jbd2_journal_revoke); 76470decc6SDave Kleikamp 77f7f4bccbSMingming Cao EXPORT_SYMBOL(jbd2_journal_init_dev); 78f7f4bccbSMingming Cao EXPORT_SYMBOL(jbd2_journal_init_inode); 79f7f4bccbSMingming Cao EXPORT_SYMBOL(jbd2_journal_check_used_features); 80f7f4bccbSMingming Cao EXPORT_SYMBOL(jbd2_journal_check_available_features); 81f7f4bccbSMingming Cao EXPORT_SYMBOL(jbd2_journal_set_features); 82f7f4bccbSMingming Cao EXPORT_SYMBOL(jbd2_journal_load); 83f7f4bccbSMingming Cao EXPORT_SYMBOL(jbd2_journal_destroy); 84f7f4bccbSMingming Cao EXPORT_SYMBOL(jbd2_journal_abort); 85f7f4bccbSMingming Cao EXPORT_SYMBOL(jbd2_journal_errno); 86f7f4bccbSMingming Cao EXPORT_SYMBOL(jbd2_journal_ack_err); 87f7f4bccbSMingming Cao EXPORT_SYMBOL(jbd2_journal_clear_err); 88f7f4bccbSMingming Cao EXPORT_SYMBOL(jbd2_log_wait_commit); 893b799d15STheodore Ts'o EXPORT_SYMBOL(jbd2_log_start_commit); 90f7f4bccbSMingming Cao EXPORT_SYMBOL(jbd2_journal_start_commit); 91f7f4bccbSMingming Cao EXPORT_SYMBOL(jbd2_journal_force_commit_nested); 92f7f4bccbSMingming Cao EXPORT_SYMBOL(jbd2_journal_wipe); 93f7f4bccbSMingming Cao EXPORT_SYMBOL(jbd2_journal_blocks_per_page); 94f7f4bccbSMingming Cao EXPORT_SYMBOL(jbd2_journal_invalidatepage); 95f7f4bccbSMingming Cao EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers); 96f7f4bccbSMingming Cao EXPORT_SYMBOL(jbd2_journal_force_commit); 97c851ed54SJan Kara EXPORT_SYMBOL(jbd2_journal_file_inode); 98c851ed54SJan Kara EXPORT_SYMBOL(jbd2_journal_init_jbd_inode); 99c851ed54SJan Kara EXPORT_SYMBOL(jbd2_journal_release_jbd_inode); 100c851ed54SJan Kara EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate); 1018aefcd55STheodore Ts'o EXPORT_SYMBOL(jbd2_inode_cache); 102470decc6SDave Kleikamp 103470decc6SDave Kleikamp static void __journal_abort_soft (journal_t *journal, int errno); 104d2eecb03STheodore Ts'o static int jbd2_journal_create_slab(size_t slab_size); 105470decc6SDave Kleikamp 106169f1a2aSPaul Gortmaker #ifdef CONFIG_JBD2_DEBUG 107169f1a2aSPaul Gortmaker void __jbd2_debug(int level, const char *file, const char *func, 108169f1a2aSPaul Gortmaker unsigned int line, const char *fmt, ...) 109169f1a2aSPaul Gortmaker { 110169f1a2aSPaul Gortmaker struct va_format vaf; 111169f1a2aSPaul Gortmaker va_list args; 112169f1a2aSPaul Gortmaker 113169f1a2aSPaul Gortmaker if (level > jbd2_journal_enable_debug) 114169f1a2aSPaul Gortmaker return; 115169f1a2aSPaul Gortmaker va_start(args, fmt); 116169f1a2aSPaul Gortmaker vaf.fmt = fmt; 117169f1a2aSPaul Gortmaker vaf.va = &args; 118169f1a2aSPaul Gortmaker printk(KERN_DEBUG "%s: (%s, %u): %pV\n", file, func, line, &vaf); 119169f1a2aSPaul Gortmaker va_end(args); 120169f1a2aSPaul Gortmaker } 121169f1a2aSPaul Gortmaker EXPORT_SYMBOL(__jbd2_debug); 122169f1a2aSPaul Gortmaker #endif 123169f1a2aSPaul Gortmaker 12425ed6e8aSDarrick J. Wong /* Checksumming functions */ 1257747e6d0SRashika Kheria static int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb) 12625ed6e8aSDarrick J. Wong { 127db9ee220SDarrick J. Wong if (!jbd2_journal_has_csum_v2or3(j)) 12825ed6e8aSDarrick J. Wong return 1; 12925ed6e8aSDarrick J. Wong 13025ed6e8aSDarrick J. Wong return sb->s_checksum_type == JBD2_CRC32C_CHKSUM; 13125ed6e8aSDarrick J. Wong } 13225ed6e8aSDarrick J. Wong 13318a6ea1eSDarrick J. Wong static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb) 1344fd5ea43SDarrick J. Wong { 13518a6ea1eSDarrick J. Wong __u32 csum; 13618a6ea1eSDarrick J. Wong __be32 old_csum; 1374fd5ea43SDarrick J. Wong 1384fd5ea43SDarrick J. Wong old_csum = sb->s_checksum; 1394fd5ea43SDarrick J. Wong sb->s_checksum = 0; 1404fd5ea43SDarrick J. Wong csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t)); 1414fd5ea43SDarrick J. Wong sb->s_checksum = old_csum; 1424fd5ea43SDarrick J. Wong 1434fd5ea43SDarrick J. Wong return cpu_to_be32(csum); 1444fd5ea43SDarrick J. Wong } 1454fd5ea43SDarrick J. Wong 1467747e6d0SRashika Kheria static int jbd2_superblock_csum_verify(journal_t *j, journal_superblock_t *sb) 1474fd5ea43SDarrick J. Wong { 148db9ee220SDarrick J. Wong if (!jbd2_journal_has_csum_v2or3(j)) 1494fd5ea43SDarrick J. Wong return 1; 1504fd5ea43SDarrick J. Wong 1514fd5ea43SDarrick J. Wong return sb->s_checksum == jbd2_superblock_csum(j, sb); 1524fd5ea43SDarrick J. Wong } 1534fd5ea43SDarrick J. Wong 1547747e6d0SRashika Kheria static void jbd2_superblock_csum_set(journal_t *j, journal_superblock_t *sb) 1554fd5ea43SDarrick J. Wong { 156db9ee220SDarrick J. Wong if (!jbd2_journal_has_csum_v2or3(j)) 1574fd5ea43SDarrick J. Wong return; 1584fd5ea43SDarrick J. Wong 1594fd5ea43SDarrick J. Wong sb->s_checksum = jbd2_superblock_csum(j, sb); 1604fd5ea43SDarrick J. Wong } 1614fd5ea43SDarrick J. Wong 162470decc6SDave Kleikamp /* 163470decc6SDave Kleikamp * Helper function used to manage commit timeouts 164470decc6SDave Kleikamp */ 165470decc6SDave Kleikamp 166470decc6SDave Kleikamp static void commit_timeout(unsigned long __data) 167470decc6SDave Kleikamp { 168470decc6SDave Kleikamp struct task_struct * p = (struct task_struct *) __data; 169470decc6SDave Kleikamp 170470decc6SDave Kleikamp wake_up_process(p); 171470decc6SDave Kleikamp } 172470decc6SDave Kleikamp 173470decc6SDave Kleikamp /* 174f7f4bccbSMingming Cao * kjournald2: The main thread function used to manage a logging device 175470decc6SDave Kleikamp * journal. 176470decc6SDave Kleikamp * 177470decc6SDave Kleikamp * This kernel thread is responsible for two things: 178470decc6SDave Kleikamp * 179470decc6SDave Kleikamp * 1) COMMIT: Every so often we need to commit the current state of the 180470decc6SDave Kleikamp * filesystem to disk. The journal thread is responsible for writing 181470decc6SDave Kleikamp * all of the metadata buffers to disk. 182470decc6SDave Kleikamp * 183470decc6SDave Kleikamp * 2) CHECKPOINT: We cannot reuse a used section of the log file until all 184470decc6SDave Kleikamp * of the data in that part of the log has been rewritten elsewhere on 185470decc6SDave Kleikamp * the disk. Flushing these old buffers to reclaim space in the log is 186470decc6SDave Kleikamp * known as checkpointing, and this thread is responsible for that job. 187470decc6SDave Kleikamp */ 188470decc6SDave Kleikamp 189f7f4bccbSMingming Cao static int kjournald2(void *arg) 190470decc6SDave Kleikamp { 191470decc6SDave Kleikamp journal_t *journal = arg; 192470decc6SDave Kleikamp transaction_t *transaction; 193470decc6SDave Kleikamp 194470decc6SDave Kleikamp /* 195470decc6SDave Kleikamp * Set up an interval timer which can be used to trigger a commit wakeup 196470decc6SDave Kleikamp * after the commit interval expires 197470decc6SDave Kleikamp */ 198470decc6SDave Kleikamp setup_timer(&journal->j_commit_timer, commit_timeout, 199470decc6SDave Kleikamp (unsigned long)current); 200470decc6SDave Kleikamp 20135c80422SNigel Cunningham set_freezable(); 20235c80422SNigel Cunningham 203470decc6SDave Kleikamp /* Record that the journal thread is running */ 204470decc6SDave Kleikamp journal->j_task = current; 205470decc6SDave Kleikamp wake_up(&journal->j_wait_done_commit); 206470decc6SDave Kleikamp 207470decc6SDave Kleikamp /* 208470decc6SDave Kleikamp * And now, wait forever for commit wakeup events. 209470decc6SDave Kleikamp */ 210a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 211470decc6SDave Kleikamp 212470decc6SDave Kleikamp loop: 213f7f4bccbSMingming Cao if (journal->j_flags & JBD2_UNMOUNT) 214470decc6SDave Kleikamp goto end_loop; 215470decc6SDave Kleikamp 216470decc6SDave Kleikamp jbd_debug(1, "commit_sequence=%d, commit_request=%d\n", 217470decc6SDave Kleikamp journal->j_commit_sequence, journal->j_commit_request); 218470decc6SDave Kleikamp 219470decc6SDave Kleikamp if (journal->j_commit_sequence != journal->j_commit_request) { 220470decc6SDave Kleikamp jbd_debug(1, "OK, requests differ\n"); 221a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 222470decc6SDave Kleikamp del_timer_sync(&journal->j_commit_timer); 223f7f4bccbSMingming Cao jbd2_journal_commit_transaction(journal); 224a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 225470decc6SDave Kleikamp goto loop; 226470decc6SDave Kleikamp } 227470decc6SDave Kleikamp 228470decc6SDave Kleikamp wake_up(&journal->j_wait_done_commit); 229470decc6SDave Kleikamp if (freezing(current)) { 230470decc6SDave Kleikamp /* 231470decc6SDave Kleikamp * The simpler the better. Flushing journal isn't a 232470decc6SDave Kleikamp * good idea, because that depends on threads that may 233470decc6SDave Kleikamp * be already stopped. 234470decc6SDave Kleikamp */ 235f7f4bccbSMingming Cao jbd_debug(1, "Now suspending kjournald2\n"); 236a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 237a0acae0eSTejun Heo try_to_freeze(); 238a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 239470decc6SDave Kleikamp } else { 240470decc6SDave Kleikamp /* 241470decc6SDave Kleikamp * We assume on resume that commits are already there, 242470decc6SDave Kleikamp * so we don't sleep 243470decc6SDave Kleikamp */ 244470decc6SDave Kleikamp DEFINE_WAIT(wait); 245470decc6SDave Kleikamp int should_sleep = 1; 246470decc6SDave Kleikamp 247470decc6SDave Kleikamp prepare_to_wait(&journal->j_wait_commit, &wait, 248470decc6SDave Kleikamp TASK_INTERRUPTIBLE); 249470decc6SDave Kleikamp if (journal->j_commit_sequence != journal->j_commit_request) 250470decc6SDave Kleikamp should_sleep = 0; 251470decc6SDave Kleikamp transaction = journal->j_running_transaction; 252470decc6SDave Kleikamp if (transaction && time_after_eq(jiffies, 253470decc6SDave Kleikamp transaction->t_expires)) 254470decc6SDave Kleikamp should_sleep = 0; 255f7f4bccbSMingming Cao if (journal->j_flags & JBD2_UNMOUNT) 256470decc6SDave Kleikamp should_sleep = 0; 257470decc6SDave Kleikamp if (should_sleep) { 258a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 259470decc6SDave Kleikamp schedule(); 260a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 261470decc6SDave Kleikamp } 262470decc6SDave Kleikamp finish_wait(&journal->j_wait_commit, &wait); 263470decc6SDave Kleikamp } 264470decc6SDave Kleikamp 265f7f4bccbSMingming Cao jbd_debug(1, "kjournald2 wakes\n"); 266470decc6SDave Kleikamp 267470decc6SDave Kleikamp /* 268470decc6SDave Kleikamp * Were we woken up by a commit wakeup event? 269470decc6SDave Kleikamp */ 270470decc6SDave Kleikamp transaction = journal->j_running_transaction; 271470decc6SDave Kleikamp if (transaction && time_after_eq(jiffies, transaction->t_expires)) { 272470decc6SDave Kleikamp journal->j_commit_request = transaction->t_tid; 273470decc6SDave Kleikamp jbd_debug(1, "woke because of timeout\n"); 274470decc6SDave Kleikamp } 275470decc6SDave Kleikamp goto loop; 276470decc6SDave Kleikamp 277470decc6SDave Kleikamp end_loop: 278a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 279470decc6SDave Kleikamp del_timer_sync(&journal->j_commit_timer); 280470decc6SDave Kleikamp journal->j_task = NULL; 281470decc6SDave Kleikamp wake_up(&journal->j_wait_done_commit); 282470decc6SDave Kleikamp jbd_debug(1, "Journal thread exiting.\n"); 283470decc6SDave Kleikamp return 0; 284470decc6SDave Kleikamp } 285470decc6SDave Kleikamp 28697f06784SPavel Emelianov static int jbd2_journal_start_thread(journal_t *journal) 287470decc6SDave Kleikamp { 28897f06784SPavel Emelianov struct task_struct *t; 28997f06784SPavel Emelianov 29090576c0bSTheodore Ts'o t = kthread_run(kjournald2, journal, "jbd2/%s", 29190576c0bSTheodore Ts'o journal->j_devname); 29297f06784SPavel Emelianov if (IS_ERR(t)) 29397f06784SPavel Emelianov return PTR_ERR(t); 29497f06784SPavel Emelianov 2951076d17aSAl Viro wait_event(journal->j_wait_done_commit, journal->j_task != NULL); 29697f06784SPavel Emelianov return 0; 297470decc6SDave Kleikamp } 298470decc6SDave Kleikamp 299470decc6SDave Kleikamp static void journal_kill_thread(journal_t *journal) 300470decc6SDave Kleikamp { 301a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 302f7f4bccbSMingming Cao journal->j_flags |= JBD2_UNMOUNT; 303470decc6SDave Kleikamp 304470decc6SDave Kleikamp while (journal->j_task) { 305a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 3063469a32aSTheodore Ts'o wake_up(&journal->j_wait_commit); 3071076d17aSAl Viro wait_event(journal->j_wait_done_commit, journal->j_task == NULL); 308a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 309470decc6SDave Kleikamp } 310a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 311470decc6SDave Kleikamp } 312470decc6SDave Kleikamp 313470decc6SDave Kleikamp /* 314f7f4bccbSMingming Cao * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal. 315470decc6SDave Kleikamp * 316470decc6SDave Kleikamp * Writes a metadata buffer to a given disk block. The actual IO is not 317470decc6SDave Kleikamp * performed but a new buffer_head is constructed which labels the data 318470decc6SDave Kleikamp * to be written with the correct destination disk block. 319470decc6SDave Kleikamp * 320470decc6SDave Kleikamp * Any magic-number escaping which needs to be done will cause a 321470decc6SDave Kleikamp * copy-out here. If the buffer happens to start with the 322f7f4bccbSMingming Cao * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the 323470decc6SDave Kleikamp * magic number is only written to the log for descripter blocks. In 324470decc6SDave Kleikamp * this case, we copy the data and replace the first word with 0, and we 325470decc6SDave Kleikamp * return a result code which indicates that this buffer needs to be 326470decc6SDave Kleikamp * marked as an escaped buffer in the corresponding log descriptor 327470decc6SDave Kleikamp * block. The missing word can then be restored when the block is read 328470decc6SDave Kleikamp * during recovery. 329470decc6SDave Kleikamp * 330470decc6SDave Kleikamp * If the source buffer has already been modified by a new transaction 331470decc6SDave Kleikamp * since we took the last commit snapshot, we use the frozen copy of 332470decc6SDave Kleikamp * that data for IO. If we end up using the existing buffer_head's data 333f5113effSJan Kara * for the write, then we have to make sure nobody modifies it while the 334f5113effSJan Kara * IO is in progress. do_get_write_access() handles this. 335470decc6SDave Kleikamp * 336f5113effSJan Kara * The function returns a pointer to the buffer_head to be used for IO. 337470decc6SDave Kleikamp * 338470decc6SDave Kleikamp * 339470decc6SDave Kleikamp * Return value: 340470decc6SDave Kleikamp * <0: Error 341470decc6SDave Kleikamp * >=0: Finished OK 342470decc6SDave Kleikamp * 343470decc6SDave Kleikamp * On success: 344470decc6SDave Kleikamp * Bit 0 set == escape performed on the data 345470decc6SDave Kleikamp * Bit 1 set == buffer copy-out performed (kfree the data after IO) 346470decc6SDave Kleikamp */ 347470decc6SDave Kleikamp 348f7f4bccbSMingming Cao int jbd2_journal_write_metadata_buffer(transaction_t *transaction, 349470decc6SDave Kleikamp struct journal_head *jh_in, 350f5113effSJan Kara struct buffer_head **bh_out, 351f5113effSJan Kara sector_t blocknr) 352470decc6SDave Kleikamp { 353470decc6SDave Kleikamp int need_copy_out = 0; 354470decc6SDave Kleikamp int done_copy_out = 0; 355470decc6SDave Kleikamp int do_escape = 0; 356470decc6SDave Kleikamp char *mapped_data; 357470decc6SDave Kleikamp struct buffer_head *new_bh; 358470decc6SDave Kleikamp struct page *new_page; 359470decc6SDave Kleikamp unsigned int new_offset; 360470decc6SDave Kleikamp struct buffer_head *bh_in = jh2bh(jh_in); 36196577c43Sdingdinghua journal_t *journal = transaction->t_journal; 362470decc6SDave Kleikamp 363470decc6SDave Kleikamp /* 364470decc6SDave Kleikamp * The buffer really shouldn't be locked: only the current committing 365470decc6SDave Kleikamp * transaction is allowed to write it, so nobody else is allowed 366470decc6SDave Kleikamp * to do any IO. 367470decc6SDave Kleikamp * 368470decc6SDave Kleikamp * akpm: except if we're journalling data, and write() output is 369470decc6SDave Kleikamp * also part of a shared mapping, and another thread has 370470decc6SDave Kleikamp * decided to launch a writepage() against this buffer. 371470decc6SDave Kleikamp */ 372470decc6SDave Kleikamp J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in)); 373470decc6SDave Kleikamp 37447def826STheodore Ts'o retry_alloc: 37547def826STheodore Ts'o new_bh = alloc_buffer_head(GFP_NOFS); 37647def826STheodore Ts'o if (!new_bh) { 37747def826STheodore Ts'o /* 37847def826STheodore Ts'o * Failure is not an option, but __GFP_NOFAIL is going 37947def826STheodore Ts'o * away; so we retry ourselves here. 38047def826STheodore Ts'o */ 38147def826STheodore Ts'o congestion_wait(BLK_RW_ASYNC, HZ/50); 38247def826STheodore Ts'o goto retry_alloc; 38347def826STheodore Ts'o } 38447def826STheodore Ts'o 38596577c43Sdingdinghua /* keep subsequent assertions sane */ 38696577c43Sdingdinghua atomic_set(&new_bh->b_count, 1); 387470decc6SDave Kleikamp 388f5113effSJan Kara jbd_lock_bh_state(bh_in); 389f5113effSJan Kara repeat: 390470decc6SDave Kleikamp /* 391470decc6SDave Kleikamp * If a new transaction has already done a buffer copy-out, then 392470decc6SDave Kleikamp * we use that version of the data for the commit. 393470decc6SDave Kleikamp */ 394470decc6SDave Kleikamp if (jh_in->b_frozen_data) { 395470decc6SDave Kleikamp done_copy_out = 1; 396470decc6SDave Kleikamp new_page = virt_to_page(jh_in->b_frozen_data); 397470decc6SDave Kleikamp new_offset = offset_in_page(jh_in->b_frozen_data); 398470decc6SDave Kleikamp } else { 399470decc6SDave Kleikamp new_page = jh2bh(jh_in)->b_page; 400470decc6SDave Kleikamp new_offset = offset_in_page(jh2bh(jh_in)->b_data); 401470decc6SDave Kleikamp } 402470decc6SDave Kleikamp 403303a8f2aSCong Wang mapped_data = kmap_atomic(new_page); 404470decc6SDave Kleikamp /* 40513ceef09SJan Kara * Fire data frozen trigger if data already wasn't frozen. Do this 40613ceef09SJan Kara * before checking for escaping, as the trigger may modify the magic 40713ceef09SJan Kara * offset. If a copy-out happens afterwards, it will have the correct 40813ceef09SJan Kara * data in the buffer. 409e06c8227SJoel Becker */ 41013ceef09SJan Kara if (!done_copy_out) 41113ceef09SJan Kara jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset, 41213ceef09SJan Kara jh_in->b_triggers); 413e06c8227SJoel Becker 414e06c8227SJoel Becker /* 415470decc6SDave Kleikamp * Check for escaping 416470decc6SDave Kleikamp */ 417470decc6SDave Kleikamp if (*((__be32 *)(mapped_data + new_offset)) == 418f7f4bccbSMingming Cao cpu_to_be32(JBD2_MAGIC_NUMBER)) { 419470decc6SDave Kleikamp need_copy_out = 1; 420470decc6SDave Kleikamp do_escape = 1; 421470decc6SDave Kleikamp } 422303a8f2aSCong Wang kunmap_atomic(mapped_data); 423470decc6SDave Kleikamp 424470decc6SDave Kleikamp /* 425470decc6SDave Kleikamp * Do we need to do a data copy? 426470decc6SDave Kleikamp */ 427470decc6SDave Kleikamp if (need_copy_out && !done_copy_out) { 428470decc6SDave Kleikamp char *tmp; 429470decc6SDave Kleikamp 430470decc6SDave Kleikamp jbd_unlock_bh_state(bh_in); 431af1e76d6SMingming Cao tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS); 432e6ec116bSTheodore Ts'o if (!tmp) { 433f5113effSJan Kara brelse(new_bh); 434e6ec116bSTheodore Ts'o return -ENOMEM; 435e6ec116bSTheodore Ts'o } 436470decc6SDave Kleikamp jbd_lock_bh_state(bh_in); 437470decc6SDave Kleikamp if (jh_in->b_frozen_data) { 438af1e76d6SMingming Cao jbd2_free(tmp, bh_in->b_size); 439470decc6SDave Kleikamp goto repeat; 440470decc6SDave Kleikamp } 441470decc6SDave Kleikamp 442470decc6SDave Kleikamp jh_in->b_frozen_data = tmp; 443303a8f2aSCong Wang mapped_data = kmap_atomic(new_page); 444f5113effSJan Kara memcpy(tmp, mapped_data + new_offset, bh_in->b_size); 445303a8f2aSCong Wang kunmap_atomic(mapped_data); 446470decc6SDave Kleikamp 447470decc6SDave Kleikamp new_page = virt_to_page(tmp); 448470decc6SDave Kleikamp new_offset = offset_in_page(tmp); 449470decc6SDave Kleikamp done_copy_out = 1; 450e06c8227SJoel Becker 451e06c8227SJoel Becker /* 452e06c8227SJoel Becker * This isn't strictly necessary, as we're using frozen 453e06c8227SJoel Becker * data for the escaping, but it keeps consistency with 454e06c8227SJoel Becker * b_frozen_data usage. 455e06c8227SJoel Becker */ 456e06c8227SJoel Becker jh_in->b_frozen_triggers = jh_in->b_triggers; 457470decc6SDave Kleikamp } 458470decc6SDave Kleikamp 459470decc6SDave Kleikamp /* 460470decc6SDave Kleikamp * Did we need to do an escaping? Now we've done all the 461470decc6SDave Kleikamp * copying, we can finally do so. 462470decc6SDave Kleikamp */ 463470decc6SDave Kleikamp if (do_escape) { 464303a8f2aSCong Wang mapped_data = kmap_atomic(new_page); 465470decc6SDave Kleikamp *((unsigned int *)(mapped_data + new_offset)) = 0; 466303a8f2aSCong Wang kunmap_atomic(mapped_data); 467470decc6SDave Kleikamp } 468470decc6SDave Kleikamp 469470decc6SDave Kleikamp set_bh_page(new_bh, new_page, new_offset); 470f5113effSJan Kara new_bh->b_size = bh_in->b_size; 471f5113effSJan Kara new_bh->b_bdev = journal->j_dev; 472470decc6SDave Kleikamp new_bh->b_blocknr = blocknr; 473b34090e5SJan Kara new_bh->b_private = bh_in; 474470decc6SDave Kleikamp set_buffer_mapped(new_bh); 475470decc6SDave Kleikamp set_buffer_dirty(new_bh); 476470decc6SDave Kleikamp 477f5113effSJan Kara *bh_out = new_bh; 478470decc6SDave Kleikamp 479470decc6SDave Kleikamp /* 480470decc6SDave Kleikamp * The to-be-written buffer needs to get moved to the io queue, 481470decc6SDave Kleikamp * and the original buffer whose contents we are shadowing or 482470decc6SDave Kleikamp * copying is moved to the transaction's shadow queue. 483470decc6SDave Kleikamp */ 484470decc6SDave Kleikamp JBUFFER_TRACE(jh_in, "file as BJ_Shadow"); 48596577c43Sdingdinghua spin_lock(&journal->j_list_lock); 48696577c43Sdingdinghua __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow); 48796577c43Sdingdinghua spin_unlock(&journal->j_list_lock); 488b34090e5SJan Kara set_buffer_shadow(bh_in); 48996577c43Sdingdinghua jbd_unlock_bh_state(bh_in); 49096577c43Sdingdinghua 491470decc6SDave Kleikamp return do_escape | (done_copy_out << 1); 492470decc6SDave Kleikamp } 493470decc6SDave Kleikamp 494470decc6SDave Kleikamp /* 495470decc6SDave Kleikamp * Allocation code for the journal file. Manage the space left in the 496470decc6SDave Kleikamp * journal, so that we can begin checkpointing when appropriate. 497470decc6SDave Kleikamp */ 498470decc6SDave Kleikamp 499470decc6SDave Kleikamp /* 500e4471831STheodore Ts'o * Called with j_state_lock locked for writing. 501e4471831STheodore Ts'o * Returns true if a transaction commit was started. 502470decc6SDave Kleikamp */ 503f7f4bccbSMingming Cao int __jbd2_log_start_commit(journal_t *journal, tid_t target) 504470decc6SDave Kleikamp { 505e7b04ac0SEric Sandeen /* Return if the txn has already requested to be committed */ 506e7b04ac0SEric Sandeen if (journal->j_commit_request == target) 507e7b04ac0SEric Sandeen return 0; 508e7b04ac0SEric Sandeen 509470decc6SDave Kleikamp /* 510deeeaf13STheodore Ts'o * The only transaction we can possibly wait upon is the 511deeeaf13STheodore Ts'o * currently running transaction (if it exists). Otherwise, 512deeeaf13STheodore Ts'o * the target tid must be an old one. 513470decc6SDave Kleikamp */ 514deeeaf13STheodore Ts'o if (journal->j_running_transaction && 515deeeaf13STheodore Ts'o journal->j_running_transaction->t_tid == target) { 516470decc6SDave Kleikamp /* 517bcf3d0bcSAndrea Gelmini * We want a new commit: OK, mark the request and wakeup the 518470decc6SDave Kleikamp * commit thread. We do _not_ do the commit ourselves. 519470decc6SDave Kleikamp */ 520470decc6SDave Kleikamp 521470decc6SDave Kleikamp journal->j_commit_request = target; 522f2a44523SEryu Guan jbd_debug(1, "JBD2: requesting commit %d/%d\n", 523470decc6SDave Kleikamp journal->j_commit_request, 524470decc6SDave Kleikamp journal->j_commit_sequence); 5259fff24aaSTheodore Ts'o journal->j_running_transaction->t_requested = jiffies; 526470decc6SDave Kleikamp wake_up(&journal->j_wait_commit); 527470decc6SDave Kleikamp return 1; 528deeeaf13STheodore Ts'o } else if (!tid_geq(journal->j_commit_request, target)) 529deeeaf13STheodore Ts'o /* This should never happen, but if it does, preserve 530deeeaf13STheodore Ts'o the evidence before kjournald goes into a loop and 531deeeaf13STheodore Ts'o increments j_commit_sequence beyond all recognition. */ 532f2a44523SEryu Guan WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n", 5331be2add6STheodore Ts'o journal->j_commit_request, 5341be2add6STheodore Ts'o journal->j_commit_sequence, 535deeeaf13STheodore Ts'o target, journal->j_running_transaction ? 536deeeaf13STheodore Ts'o journal->j_running_transaction->t_tid : 0); 537470decc6SDave Kleikamp return 0; 538470decc6SDave Kleikamp } 539470decc6SDave Kleikamp 540f7f4bccbSMingming Cao int jbd2_log_start_commit(journal_t *journal, tid_t tid) 541470decc6SDave Kleikamp { 542470decc6SDave Kleikamp int ret; 543470decc6SDave Kleikamp 544a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 545f7f4bccbSMingming Cao ret = __jbd2_log_start_commit(journal, tid); 546a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 547470decc6SDave Kleikamp return ret; 548470decc6SDave Kleikamp } 549470decc6SDave Kleikamp 550470decc6SDave Kleikamp /* 5519ff86446SDmitry Monakhov * Force and wait any uncommitted transactions. We can only force the running 5529ff86446SDmitry Monakhov * transaction if we don't have an active handle, otherwise, we will deadlock. 5539ff86446SDmitry Monakhov * Returns: <0 in case of error, 5549ff86446SDmitry Monakhov * 0 if nothing to commit, 5559ff86446SDmitry Monakhov * 1 if transaction was successfully committed. 556470decc6SDave Kleikamp */ 5579ff86446SDmitry Monakhov static int __jbd2_journal_force_commit(journal_t *journal) 558470decc6SDave Kleikamp { 559470decc6SDave Kleikamp transaction_t *transaction = NULL; 560470decc6SDave Kleikamp tid_t tid; 5619ff86446SDmitry Monakhov int need_to_start = 0, ret = 0; 562470decc6SDave Kleikamp 563a931da6aSTheodore Ts'o read_lock(&journal->j_state_lock); 564470decc6SDave Kleikamp if (journal->j_running_transaction && !current->journal_info) { 565470decc6SDave Kleikamp transaction = journal->j_running_transaction; 566e4471831STheodore Ts'o if (!tid_geq(journal->j_commit_request, transaction->t_tid)) 567e4471831STheodore Ts'o need_to_start = 1; 568470decc6SDave Kleikamp } else if (journal->j_committing_transaction) 569470decc6SDave Kleikamp transaction = journal->j_committing_transaction; 570470decc6SDave Kleikamp 571470decc6SDave Kleikamp if (!transaction) { 5729ff86446SDmitry Monakhov /* Nothing to commit */ 573a931da6aSTheodore Ts'o read_unlock(&journal->j_state_lock); 5749ff86446SDmitry Monakhov return 0; 575470decc6SDave Kleikamp } 576470decc6SDave Kleikamp tid = transaction->t_tid; 577a931da6aSTheodore Ts'o read_unlock(&journal->j_state_lock); 578e4471831STheodore Ts'o if (need_to_start) 579e4471831STheodore Ts'o jbd2_log_start_commit(journal, tid); 5809ff86446SDmitry Monakhov ret = jbd2_log_wait_commit(journal, tid); 5819ff86446SDmitry Monakhov if (!ret) 5829ff86446SDmitry Monakhov ret = 1; 5839ff86446SDmitry Monakhov 5849ff86446SDmitry Monakhov return ret; 5859ff86446SDmitry Monakhov } 5869ff86446SDmitry Monakhov 5879ff86446SDmitry Monakhov /** 5889ff86446SDmitry Monakhov * Force and wait upon a commit if the calling process is not within 5899ff86446SDmitry Monakhov * transaction. This is used for forcing out undo-protected data which contains 5909ff86446SDmitry Monakhov * bitmaps, when the fs is running out of space. 5919ff86446SDmitry Monakhov * 5929ff86446SDmitry Monakhov * @journal: journal to force 5939ff86446SDmitry Monakhov * Returns true if progress was made. 5949ff86446SDmitry Monakhov */ 5959ff86446SDmitry Monakhov int jbd2_journal_force_commit_nested(journal_t *journal) 5969ff86446SDmitry Monakhov { 5979ff86446SDmitry Monakhov int ret; 5989ff86446SDmitry Monakhov 5999ff86446SDmitry Monakhov ret = __jbd2_journal_force_commit(journal); 6009ff86446SDmitry Monakhov return ret > 0; 6019ff86446SDmitry Monakhov } 6029ff86446SDmitry Monakhov 6039ff86446SDmitry Monakhov /** 6049ff86446SDmitry Monakhov * int journal_force_commit() - force any uncommitted transactions 6059ff86446SDmitry Monakhov * @journal: journal to force 6069ff86446SDmitry Monakhov * 6079ff86446SDmitry Monakhov * Caller want unconditional commit. We can only force the running transaction 6089ff86446SDmitry Monakhov * if we don't have an active handle, otherwise, we will deadlock. 6099ff86446SDmitry Monakhov */ 6109ff86446SDmitry Monakhov int jbd2_journal_force_commit(journal_t *journal) 6119ff86446SDmitry Monakhov { 6129ff86446SDmitry Monakhov int ret; 6139ff86446SDmitry Monakhov 6149ff86446SDmitry Monakhov J_ASSERT(!current->journal_info); 6159ff86446SDmitry Monakhov ret = __jbd2_journal_force_commit(journal); 6169ff86446SDmitry Monakhov if (ret > 0) 6179ff86446SDmitry Monakhov ret = 0; 6189ff86446SDmitry Monakhov return ret; 619470decc6SDave Kleikamp } 620470decc6SDave Kleikamp 621470decc6SDave Kleikamp /* 622470decc6SDave Kleikamp * Start a commit of the current running transaction (if any). Returns true 623c88ccea3SJan Kara * if a transaction is going to be committed (or is currently already 624c88ccea3SJan Kara * committing), and fills its tid in at *ptid 625470decc6SDave Kleikamp */ 626f7f4bccbSMingming Cao int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid) 627470decc6SDave Kleikamp { 628470decc6SDave Kleikamp int ret = 0; 629470decc6SDave Kleikamp 630a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 631470decc6SDave Kleikamp if (journal->j_running_transaction) { 632470decc6SDave Kleikamp tid_t tid = journal->j_running_transaction->t_tid; 633470decc6SDave Kleikamp 634c88ccea3SJan Kara __jbd2_log_start_commit(journal, tid); 635c88ccea3SJan Kara /* There's a running transaction and we've just made sure 636c88ccea3SJan Kara * it's commit has been scheduled. */ 637c88ccea3SJan Kara if (ptid) 638470decc6SDave Kleikamp *ptid = tid; 639c88ccea3SJan Kara ret = 1; 640c88ccea3SJan Kara } else if (journal->j_committing_transaction) { 641470decc6SDave Kleikamp /* 64212810ad7SArtem Bityutskiy * If commit has been started, then we have to wait for 64312810ad7SArtem Bityutskiy * completion of that transaction. 644470decc6SDave Kleikamp */ 645c88ccea3SJan Kara if (ptid) 646470decc6SDave Kleikamp *ptid = journal->j_committing_transaction->t_tid; 647470decc6SDave Kleikamp ret = 1; 648470decc6SDave Kleikamp } 649a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 650470decc6SDave Kleikamp return ret; 651470decc6SDave Kleikamp } 652470decc6SDave Kleikamp 653470decc6SDave Kleikamp /* 654bbd2be36SJan Kara * Return 1 if a given transaction has not yet sent barrier request 655bbd2be36SJan Kara * connected with a transaction commit. If 0 is returned, transaction 656bbd2be36SJan Kara * may or may not have sent the barrier. Used to avoid sending barrier 657bbd2be36SJan Kara * twice in common cases. 658bbd2be36SJan Kara */ 659bbd2be36SJan Kara int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid) 660bbd2be36SJan Kara { 661bbd2be36SJan Kara int ret = 0; 662bbd2be36SJan Kara transaction_t *commit_trans; 663bbd2be36SJan Kara 664bbd2be36SJan Kara if (!(journal->j_flags & JBD2_BARRIER)) 665bbd2be36SJan Kara return 0; 666bbd2be36SJan Kara read_lock(&journal->j_state_lock); 667bbd2be36SJan Kara /* Transaction already committed? */ 668bbd2be36SJan Kara if (tid_geq(journal->j_commit_sequence, tid)) 669bbd2be36SJan Kara goto out; 670bbd2be36SJan Kara commit_trans = journal->j_committing_transaction; 671bbd2be36SJan Kara if (!commit_trans || commit_trans->t_tid != tid) { 672bbd2be36SJan Kara ret = 1; 673bbd2be36SJan Kara goto out; 674bbd2be36SJan Kara } 675bbd2be36SJan Kara /* 676bbd2be36SJan Kara * Transaction is being committed and we already proceeded to 677bbd2be36SJan Kara * submitting a flush to fs partition? 678bbd2be36SJan Kara */ 679bbd2be36SJan Kara if (journal->j_fs_dev != journal->j_dev) { 680bbd2be36SJan Kara if (!commit_trans->t_need_data_flush || 681bbd2be36SJan Kara commit_trans->t_state >= T_COMMIT_DFLUSH) 682bbd2be36SJan Kara goto out; 683bbd2be36SJan Kara } else { 684bbd2be36SJan Kara if (commit_trans->t_state >= T_COMMIT_JFLUSH) 685bbd2be36SJan Kara goto out; 686bbd2be36SJan Kara } 687bbd2be36SJan Kara ret = 1; 688bbd2be36SJan Kara out: 689bbd2be36SJan Kara read_unlock(&journal->j_state_lock); 690bbd2be36SJan Kara return ret; 691bbd2be36SJan Kara } 692bbd2be36SJan Kara EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier); 693bbd2be36SJan Kara 694bbd2be36SJan Kara /* 695470decc6SDave Kleikamp * Wait for a specified commit to complete. 696470decc6SDave Kleikamp * The caller may not hold the journal lock. 697470decc6SDave Kleikamp */ 698f7f4bccbSMingming Cao int jbd2_log_wait_commit(journal_t *journal, tid_t tid) 699470decc6SDave Kleikamp { 700470decc6SDave Kleikamp int err = 0; 701470decc6SDave Kleikamp 702a931da6aSTheodore Ts'o read_lock(&journal->j_state_lock); 703e23291b9SJose R. Santos #ifdef CONFIG_JBD2_DEBUG 704470decc6SDave Kleikamp if (!tid_geq(journal->j_commit_request, tid)) { 70575685071SJan Kara printk(KERN_ERR 706470decc6SDave Kleikamp "%s: error: j_commit_request=%d, tid=%d\n", 707329d291fSHarvey Harrison __func__, journal->j_commit_request, tid); 708470decc6SDave Kleikamp } 709470decc6SDave Kleikamp #endif 710470decc6SDave Kleikamp while (tid_gt(tid, journal->j_commit_sequence)) { 711f2a44523SEryu Guan jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n", 712470decc6SDave Kleikamp tid, journal->j_commit_sequence); 713a931da6aSTheodore Ts'o read_unlock(&journal->j_state_lock); 7143469a32aSTheodore Ts'o wake_up(&journal->j_wait_commit); 715470decc6SDave Kleikamp wait_event(journal->j_wait_done_commit, 716470decc6SDave Kleikamp !tid_gt(tid, journal->j_commit_sequence)); 717a931da6aSTheodore Ts'o read_lock(&journal->j_state_lock); 718470decc6SDave Kleikamp } 719a931da6aSTheodore Ts'o read_unlock(&journal->j_state_lock); 720470decc6SDave Kleikamp 72175685071SJan Kara if (unlikely(is_journal_aborted(journal))) 722470decc6SDave Kleikamp err = -EIO; 723470decc6SDave Kleikamp return err; 724470decc6SDave Kleikamp } 725470decc6SDave Kleikamp 726470decc6SDave Kleikamp /* 727d76a3a77STheodore Ts'o * When this function returns the transaction corresponding to tid 728d76a3a77STheodore Ts'o * will be completed. If the transaction has currently running, start 729d76a3a77STheodore Ts'o * committing that transaction before waiting for it to complete. If 730d76a3a77STheodore Ts'o * the transaction id is stale, it is by definition already completed, 731d76a3a77STheodore Ts'o * so just return SUCCESS. 732d76a3a77STheodore Ts'o */ 733d76a3a77STheodore Ts'o int jbd2_complete_transaction(journal_t *journal, tid_t tid) 734d76a3a77STheodore Ts'o { 735d76a3a77STheodore Ts'o int need_to_wait = 1; 736d76a3a77STheodore Ts'o 737d76a3a77STheodore Ts'o read_lock(&journal->j_state_lock); 738d76a3a77STheodore Ts'o if (journal->j_running_transaction && 739d76a3a77STheodore Ts'o journal->j_running_transaction->t_tid == tid) { 740d76a3a77STheodore Ts'o if (journal->j_commit_request != tid) { 741d76a3a77STheodore Ts'o /* transaction not yet started, so request it */ 742d76a3a77STheodore Ts'o read_unlock(&journal->j_state_lock); 743d76a3a77STheodore Ts'o jbd2_log_start_commit(journal, tid); 744d76a3a77STheodore Ts'o goto wait_commit; 745d76a3a77STheodore Ts'o } 746d76a3a77STheodore Ts'o } else if (!(journal->j_committing_transaction && 747d76a3a77STheodore Ts'o journal->j_committing_transaction->t_tid == tid)) 748d76a3a77STheodore Ts'o need_to_wait = 0; 749d76a3a77STheodore Ts'o read_unlock(&journal->j_state_lock); 750d76a3a77STheodore Ts'o if (!need_to_wait) 751d76a3a77STheodore Ts'o return 0; 752d76a3a77STheodore Ts'o wait_commit: 753d76a3a77STheodore Ts'o return jbd2_log_wait_commit(journal, tid); 754d76a3a77STheodore Ts'o } 755d76a3a77STheodore Ts'o EXPORT_SYMBOL(jbd2_complete_transaction); 756d76a3a77STheodore Ts'o 757d76a3a77STheodore Ts'o /* 758470decc6SDave Kleikamp * Log buffer allocation routines: 759470decc6SDave Kleikamp */ 760470decc6SDave Kleikamp 76118eba7aaSMingming Cao int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp) 762470decc6SDave Kleikamp { 763470decc6SDave Kleikamp unsigned long blocknr; 764470decc6SDave Kleikamp 765a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 766470decc6SDave Kleikamp J_ASSERT(journal->j_free > 1); 767470decc6SDave Kleikamp 768470decc6SDave Kleikamp blocknr = journal->j_head; 769470decc6SDave Kleikamp journal->j_head++; 770470decc6SDave Kleikamp journal->j_free--; 771470decc6SDave Kleikamp if (journal->j_head == journal->j_last) 772470decc6SDave Kleikamp journal->j_head = journal->j_first; 773a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 774f7f4bccbSMingming Cao return jbd2_journal_bmap(journal, blocknr, retp); 775470decc6SDave Kleikamp } 776470decc6SDave Kleikamp 777470decc6SDave Kleikamp /* 778470decc6SDave Kleikamp * Conversion of logical to physical block numbers for the journal 779470decc6SDave Kleikamp * 780470decc6SDave Kleikamp * On external journals the journal blocks are identity-mapped, so 781470decc6SDave Kleikamp * this is a no-op. If needed, we can use j_blk_offset - everything is 782470decc6SDave Kleikamp * ready. 783470decc6SDave Kleikamp */ 784f7f4bccbSMingming Cao int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr, 78518eba7aaSMingming Cao unsigned long long *retp) 786470decc6SDave Kleikamp { 787470decc6SDave Kleikamp int err = 0; 78818eba7aaSMingming Cao unsigned long long ret; 789470decc6SDave Kleikamp 790470decc6SDave Kleikamp if (journal->j_inode) { 791470decc6SDave Kleikamp ret = bmap(journal->j_inode, blocknr); 792470decc6SDave Kleikamp if (ret) 793470decc6SDave Kleikamp *retp = ret; 794470decc6SDave Kleikamp else { 795470decc6SDave Kleikamp printk(KERN_ALERT "%s: journal block not found " 796470decc6SDave Kleikamp "at offset %lu on %s\n", 79705496769STheodore Ts'o __func__, blocknr, journal->j_devname); 798470decc6SDave Kleikamp err = -EIO; 799470decc6SDave Kleikamp __journal_abort_soft(journal, err); 800470decc6SDave Kleikamp } 801470decc6SDave Kleikamp } else { 802470decc6SDave Kleikamp *retp = blocknr; /* +journal->j_blk_offset */ 803470decc6SDave Kleikamp } 804470decc6SDave Kleikamp return err; 805470decc6SDave Kleikamp } 806470decc6SDave Kleikamp 807470decc6SDave Kleikamp /* 808470decc6SDave Kleikamp * We play buffer_head aliasing tricks to write data/metadata blocks to 809470decc6SDave Kleikamp * the journal without copying their contents, but for journal 810470decc6SDave Kleikamp * descriptor blocks we do need to generate bona fide buffers. 811470decc6SDave Kleikamp * 812f7f4bccbSMingming Cao * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying 813470decc6SDave Kleikamp * the buffer's contents they really should run flush_dcache_page(bh->b_page). 814470decc6SDave Kleikamp * But we don't bother doing that, so there will be coherency problems with 815470decc6SDave Kleikamp * mmaps of blockdevs which hold live JBD-controlled filesystems. 816470decc6SDave Kleikamp */ 817e5a120aeSJan Kara struct buffer_head *jbd2_journal_get_descriptor_buffer(journal_t *journal) 818470decc6SDave Kleikamp { 819470decc6SDave Kleikamp struct buffer_head *bh; 82018eba7aaSMingming Cao unsigned long long blocknr; 821470decc6SDave Kleikamp int err; 822470decc6SDave Kleikamp 823f7f4bccbSMingming Cao err = jbd2_journal_next_log_block(journal, &blocknr); 824470decc6SDave Kleikamp 825470decc6SDave Kleikamp if (err) 826470decc6SDave Kleikamp return NULL; 827470decc6SDave Kleikamp 828470decc6SDave Kleikamp bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize); 8294b905671SJan Kara if (!bh) 8304b905671SJan Kara return NULL; 831470decc6SDave Kleikamp lock_buffer(bh); 832470decc6SDave Kleikamp memset(bh->b_data, 0, journal->j_blocksize); 833470decc6SDave Kleikamp set_buffer_uptodate(bh); 834470decc6SDave Kleikamp unlock_buffer(bh); 835470decc6SDave Kleikamp BUFFER_TRACE(bh, "return this buffer"); 836e5a120aeSJan Kara return bh; 837470decc6SDave Kleikamp } 838470decc6SDave Kleikamp 83979feb521SJan Kara /* 84079feb521SJan Kara * Return tid of the oldest transaction in the journal and block in the journal 84179feb521SJan Kara * where the transaction starts. 84279feb521SJan Kara * 84379feb521SJan Kara * If the journal is now empty, return which will be the next transaction ID 84479feb521SJan Kara * we will write and where will that transaction start. 84579feb521SJan Kara * 84679feb521SJan Kara * The return value is 0 if journal tail cannot be pushed any further, 1 if 84779feb521SJan Kara * it can. 84879feb521SJan Kara */ 84979feb521SJan Kara int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid, 85079feb521SJan Kara unsigned long *block) 85179feb521SJan Kara { 85279feb521SJan Kara transaction_t *transaction; 85379feb521SJan Kara int ret; 85479feb521SJan Kara 85579feb521SJan Kara read_lock(&journal->j_state_lock); 85679feb521SJan Kara spin_lock(&journal->j_list_lock); 85779feb521SJan Kara transaction = journal->j_checkpoint_transactions; 85879feb521SJan Kara if (transaction) { 85979feb521SJan Kara *tid = transaction->t_tid; 86079feb521SJan Kara *block = transaction->t_log_start; 86179feb521SJan Kara } else if ((transaction = journal->j_committing_transaction) != NULL) { 86279feb521SJan Kara *tid = transaction->t_tid; 86379feb521SJan Kara *block = transaction->t_log_start; 86479feb521SJan Kara } else if ((transaction = journal->j_running_transaction) != NULL) { 86579feb521SJan Kara *tid = transaction->t_tid; 86679feb521SJan Kara *block = journal->j_head; 86779feb521SJan Kara } else { 86879feb521SJan Kara *tid = journal->j_transaction_sequence; 86979feb521SJan Kara *block = journal->j_head; 87079feb521SJan Kara } 87179feb521SJan Kara ret = tid_gt(*tid, journal->j_tail_sequence); 87279feb521SJan Kara spin_unlock(&journal->j_list_lock); 87379feb521SJan Kara read_unlock(&journal->j_state_lock); 87479feb521SJan Kara 87579feb521SJan Kara return ret; 87679feb521SJan Kara } 87779feb521SJan Kara 87879feb521SJan Kara /* 87979feb521SJan Kara * Update information in journal structure and in on disk journal superblock 88079feb521SJan Kara * about log tail. This function does not check whether information passed in 88179feb521SJan Kara * really pushes log tail further. It's responsibility of the caller to make 88279feb521SJan Kara * sure provided log tail information is valid (e.g. by holding 88379feb521SJan Kara * j_checkpoint_mutex all the time between computing log tail and calling this 88479feb521SJan Kara * function as is the case with jbd2_cleanup_journal_tail()). 88579feb521SJan Kara * 88679feb521SJan Kara * Requires j_checkpoint_mutex 88779feb521SJan Kara */ 88879feb521SJan Kara void __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block) 88979feb521SJan Kara { 89079feb521SJan Kara unsigned long freed; 89179feb521SJan Kara 89279feb521SJan Kara BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); 89379feb521SJan Kara 89479feb521SJan Kara /* 89579feb521SJan Kara * We cannot afford for write to remain in drive's caches since as 89679feb521SJan Kara * soon as we update j_tail, next transaction can start reusing journal 89779feb521SJan Kara * space and if we lose sb update during power failure we'd replay 89879feb521SJan Kara * old transaction with possibly newly overwritten data. 89979feb521SJan Kara */ 90079feb521SJan Kara jbd2_journal_update_sb_log_tail(journal, tid, block, WRITE_FUA); 90179feb521SJan Kara write_lock(&journal->j_state_lock); 90279feb521SJan Kara freed = block - journal->j_tail; 90379feb521SJan Kara if (block < journal->j_tail) 90479feb521SJan Kara freed += journal->j_last - journal->j_first; 90579feb521SJan Kara 90679feb521SJan Kara trace_jbd2_update_log_tail(journal, tid, block, freed); 90779feb521SJan Kara jbd_debug(1, 90879feb521SJan Kara "Cleaning journal tail from %d to %d (offset %lu), " 90979feb521SJan Kara "freeing %lu\n", 91079feb521SJan Kara journal->j_tail_sequence, tid, block, freed); 91179feb521SJan Kara 91279feb521SJan Kara journal->j_free += freed; 91379feb521SJan Kara journal->j_tail_sequence = tid; 91479feb521SJan Kara journal->j_tail = block; 91579feb521SJan Kara write_unlock(&journal->j_state_lock); 91679feb521SJan Kara } 91779feb521SJan Kara 9183339578fSJan Kara /* 9193339578fSJan Kara * This is a variaon of __jbd2_update_log_tail which checks for validity of 9203339578fSJan Kara * provided log tail and locks j_checkpoint_mutex. So it is safe against races 9213339578fSJan Kara * with other threads updating log tail. 9223339578fSJan Kara */ 9233339578fSJan Kara void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block) 9243339578fSJan Kara { 9253339578fSJan Kara mutex_lock(&journal->j_checkpoint_mutex); 9263339578fSJan Kara if (tid_gt(tid, journal->j_tail_sequence)) 9273339578fSJan Kara __jbd2_update_log_tail(journal, tid, block); 9283339578fSJan Kara mutex_unlock(&journal->j_checkpoint_mutex); 9293339578fSJan Kara } 9303339578fSJan Kara 9318e85fb3fSJohann Lombardi struct jbd2_stats_proc_session { 9328e85fb3fSJohann Lombardi journal_t *journal; 9338e85fb3fSJohann Lombardi struct transaction_stats_s *stats; 9348e85fb3fSJohann Lombardi int start; 9358e85fb3fSJohann Lombardi int max; 9368e85fb3fSJohann Lombardi }; 9378e85fb3fSJohann Lombardi 9388e85fb3fSJohann Lombardi static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos) 9398e85fb3fSJohann Lombardi { 9408e85fb3fSJohann Lombardi return *pos ? NULL : SEQ_START_TOKEN; 9418e85fb3fSJohann Lombardi } 9428e85fb3fSJohann Lombardi 9438e85fb3fSJohann Lombardi static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos) 9448e85fb3fSJohann Lombardi { 9458e85fb3fSJohann Lombardi return NULL; 9468e85fb3fSJohann Lombardi } 9478e85fb3fSJohann Lombardi 9488e85fb3fSJohann Lombardi static int jbd2_seq_info_show(struct seq_file *seq, void *v) 9498e85fb3fSJohann Lombardi { 9508e85fb3fSJohann Lombardi struct jbd2_stats_proc_session *s = seq->private; 9518e85fb3fSJohann Lombardi 9528e85fb3fSJohann Lombardi if (v != SEQ_START_TOKEN) 9538e85fb3fSJohann Lombardi return 0; 9549fff24aaSTheodore Ts'o seq_printf(seq, "%lu transactions (%lu requested), " 9559fff24aaSTheodore Ts'o "each up to %u blocks\n", 9569fff24aaSTheodore Ts'o s->stats->ts_tid, s->stats->ts_requested, 9578e85fb3fSJohann Lombardi s->journal->j_max_transaction_buffers); 9588e85fb3fSJohann Lombardi if (s->stats->ts_tid == 0) 9598e85fb3fSJohann Lombardi return 0; 9608e85fb3fSJohann Lombardi seq_printf(seq, "average: \n %ums waiting for transaction\n", 961bf699327STheodore Ts'o jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid)); 9629fff24aaSTheodore Ts'o seq_printf(seq, " %ums request delay\n", 9639fff24aaSTheodore Ts'o (s->stats->ts_requested == 0) ? 0 : 9649fff24aaSTheodore Ts'o jiffies_to_msecs(s->stats->run.rs_request_delay / 9659fff24aaSTheodore Ts'o s->stats->ts_requested)); 9668e85fb3fSJohann Lombardi seq_printf(seq, " %ums running transaction\n", 967bf699327STheodore Ts'o jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid)); 9688e85fb3fSJohann Lombardi seq_printf(seq, " %ums transaction was being locked\n", 969bf699327STheodore Ts'o jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid)); 9708e85fb3fSJohann Lombardi seq_printf(seq, " %ums flushing data (in ordered mode)\n", 971bf699327STheodore Ts'o jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid)); 9728e85fb3fSJohann Lombardi seq_printf(seq, " %ums logging transaction\n", 973bf699327STheodore Ts'o jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid)); 974c225aa57SSimon Holm Thøgersen seq_printf(seq, " %lluus average transaction commit time\n", 975c225aa57SSimon Holm Thøgersen div_u64(s->journal->j_average_commit_time, 1000)); 9768e85fb3fSJohann Lombardi seq_printf(seq, " %lu handles per transaction\n", 977bf699327STheodore Ts'o s->stats->run.rs_handle_count / s->stats->ts_tid); 9788e85fb3fSJohann Lombardi seq_printf(seq, " %lu blocks per transaction\n", 979bf699327STheodore Ts'o s->stats->run.rs_blocks / s->stats->ts_tid); 9808e85fb3fSJohann Lombardi seq_printf(seq, " %lu logged blocks per transaction\n", 981bf699327STheodore Ts'o s->stats->run.rs_blocks_logged / s->stats->ts_tid); 9828e85fb3fSJohann Lombardi return 0; 9838e85fb3fSJohann Lombardi } 9848e85fb3fSJohann Lombardi 9858e85fb3fSJohann Lombardi static void jbd2_seq_info_stop(struct seq_file *seq, void *v) 9868e85fb3fSJohann Lombardi { 9878e85fb3fSJohann Lombardi } 9888e85fb3fSJohann Lombardi 98988e9d34cSJames Morris static const struct seq_operations jbd2_seq_info_ops = { 9908e85fb3fSJohann Lombardi .start = jbd2_seq_info_start, 9918e85fb3fSJohann Lombardi .next = jbd2_seq_info_next, 9928e85fb3fSJohann Lombardi .stop = jbd2_seq_info_stop, 9938e85fb3fSJohann Lombardi .show = jbd2_seq_info_show, 9948e85fb3fSJohann Lombardi }; 9958e85fb3fSJohann Lombardi 9968e85fb3fSJohann Lombardi static int jbd2_seq_info_open(struct inode *inode, struct file *file) 9978e85fb3fSJohann Lombardi { 998d9dda78bSAl Viro journal_t *journal = PDE_DATA(inode); 9998e85fb3fSJohann Lombardi struct jbd2_stats_proc_session *s; 10008e85fb3fSJohann Lombardi int rc, size; 10018e85fb3fSJohann Lombardi 10028e85fb3fSJohann Lombardi s = kmalloc(sizeof(*s), GFP_KERNEL); 10038e85fb3fSJohann Lombardi if (s == NULL) 10048e85fb3fSJohann Lombardi return -ENOMEM; 10058e85fb3fSJohann Lombardi size = sizeof(struct transaction_stats_s); 10068e85fb3fSJohann Lombardi s->stats = kmalloc(size, GFP_KERNEL); 10078e85fb3fSJohann Lombardi if (s->stats == NULL) { 10088e85fb3fSJohann Lombardi kfree(s); 10098e85fb3fSJohann Lombardi return -ENOMEM; 10108e85fb3fSJohann Lombardi } 10118e85fb3fSJohann Lombardi spin_lock(&journal->j_history_lock); 10128e85fb3fSJohann Lombardi memcpy(s->stats, &journal->j_stats, size); 10138e85fb3fSJohann Lombardi s->journal = journal; 10148e85fb3fSJohann Lombardi spin_unlock(&journal->j_history_lock); 10158e85fb3fSJohann Lombardi 10168e85fb3fSJohann Lombardi rc = seq_open(file, &jbd2_seq_info_ops); 10178e85fb3fSJohann Lombardi if (rc == 0) { 10188e85fb3fSJohann Lombardi struct seq_file *m = file->private_data; 10198e85fb3fSJohann Lombardi m->private = s; 10208e85fb3fSJohann Lombardi } else { 10218e85fb3fSJohann Lombardi kfree(s->stats); 10228e85fb3fSJohann Lombardi kfree(s); 10238e85fb3fSJohann Lombardi } 10248e85fb3fSJohann Lombardi return rc; 10258e85fb3fSJohann Lombardi 10268e85fb3fSJohann Lombardi } 10278e85fb3fSJohann Lombardi 10288e85fb3fSJohann Lombardi static int jbd2_seq_info_release(struct inode *inode, struct file *file) 10298e85fb3fSJohann Lombardi { 10308e85fb3fSJohann Lombardi struct seq_file *seq = file->private_data; 10318e85fb3fSJohann Lombardi struct jbd2_stats_proc_session *s = seq->private; 10328e85fb3fSJohann Lombardi kfree(s->stats); 10338e85fb3fSJohann Lombardi kfree(s); 10348e85fb3fSJohann Lombardi return seq_release(inode, file); 10358e85fb3fSJohann Lombardi } 10368e85fb3fSJohann Lombardi 1037828c0950SAlexey Dobriyan static const struct file_operations jbd2_seq_info_fops = { 10388e85fb3fSJohann Lombardi .owner = THIS_MODULE, 10398e85fb3fSJohann Lombardi .open = jbd2_seq_info_open, 10408e85fb3fSJohann Lombardi .read = seq_read, 10418e85fb3fSJohann Lombardi .llseek = seq_lseek, 10428e85fb3fSJohann Lombardi .release = jbd2_seq_info_release, 10438e85fb3fSJohann Lombardi }; 10448e85fb3fSJohann Lombardi 10458e85fb3fSJohann Lombardi static struct proc_dir_entry *proc_jbd2_stats; 10468e85fb3fSJohann Lombardi 10478e85fb3fSJohann Lombardi static void jbd2_stats_proc_init(journal_t *journal) 10488e85fb3fSJohann Lombardi { 104905496769STheodore Ts'o journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats); 10508e85fb3fSJohann Lombardi if (journal->j_proc_entry) { 105179da3664SDenis V. Lunev proc_create_data("info", S_IRUGO, journal->j_proc_entry, 105279da3664SDenis V. Lunev &jbd2_seq_info_fops, journal); 10538e85fb3fSJohann Lombardi } 10548e85fb3fSJohann Lombardi } 10558e85fb3fSJohann Lombardi 10568e85fb3fSJohann Lombardi static void jbd2_stats_proc_exit(journal_t *journal) 10578e85fb3fSJohann Lombardi { 10588e85fb3fSJohann Lombardi remove_proc_entry("info", journal->j_proc_entry); 105905496769STheodore Ts'o remove_proc_entry(journal->j_devname, proc_jbd2_stats); 10608e85fb3fSJohann Lombardi } 10618e85fb3fSJohann Lombardi 1062470decc6SDave Kleikamp /* 1063470decc6SDave Kleikamp * Management for journal control blocks: functions to create and 1064470decc6SDave Kleikamp * destroy journal_t structures, and to initialise and read existing 1065470decc6SDave Kleikamp * journal blocks from disk. */ 1066470decc6SDave Kleikamp 1067470decc6SDave Kleikamp /* First: create and setup a journal_t object in memory. We initialise 1068470decc6SDave Kleikamp * very few fields yet: that has to wait until we have created the 1069470decc6SDave Kleikamp * journal structures from from scratch, or loaded them from disk. */ 1070470decc6SDave Kleikamp 1071470decc6SDave Kleikamp static journal_t * journal_init_common (void) 1072470decc6SDave Kleikamp { 1073470decc6SDave Kleikamp journal_t *journal; 1074470decc6SDave Kleikamp int err; 1075470decc6SDave Kleikamp 10763ebfdf88SAndrew Morton journal = kzalloc(sizeof(*journal), GFP_KERNEL); 1077470decc6SDave Kleikamp if (!journal) 1078b7271b0aSTheodore Ts'o return NULL; 1079470decc6SDave Kleikamp 1080470decc6SDave Kleikamp init_waitqueue_head(&journal->j_wait_transaction_locked); 1081470decc6SDave Kleikamp init_waitqueue_head(&journal->j_wait_done_commit); 1082470decc6SDave Kleikamp init_waitqueue_head(&journal->j_wait_commit); 1083470decc6SDave Kleikamp init_waitqueue_head(&journal->j_wait_updates); 10848f7d89f3SJan Kara init_waitqueue_head(&journal->j_wait_reserved); 1085470decc6SDave Kleikamp mutex_init(&journal->j_barrier); 1086470decc6SDave Kleikamp mutex_init(&journal->j_checkpoint_mutex); 1087470decc6SDave Kleikamp spin_lock_init(&journal->j_revoke_lock); 1088470decc6SDave Kleikamp spin_lock_init(&journal->j_list_lock); 1089a931da6aSTheodore Ts'o rwlock_init(&journal->j_state_lock); 1090470decc6SDave Kleikamp 1091cd02ff0bSMingming Cao journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE); 109230773840STheodore Ts'o journal->j_min_batch_time = 0; 109330773840STheodore Ts'o journal->j_max_batch_time = 15000; /* 15ms */ 10948f7d89f3SJan Kara atomic_set(&journal->j_reserved_credits, 0); 1095470decc6SDave Kleikamp 1096470decc6SDave Kleikamp /* The journal is marked for error until we succeed with recovery! */ 1097f7f4bccbSMingming Cao journal->j_flags = JBD2_ABORT; 1098470decc6SDave Kleikamp 1099470decc6SDave Kleikamp /* Set up a default-sized revoke table for the new mount. */ 1100f7f4bccbSMingming Cao err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH); 1101470decc6SDave Kleikamp if (err) { 1102470decc6SDave Kleikamp kfree(journal); 1103b7271b0aSTheodore Ts'o return NULL; 1104470decc6SDave Kleikamp } 11058e85fb3fSJohann Lombardi 1106bf699327STheodore Ts'o spin_lock_init(&journal->j_history_lock); 11078e85fb3fSJohann Lombardi 1108470decc6SDave Kleikamp return journal; 1109470decc6SDave Kleikamp } 1110470decc6SDave Kleikamp 1111f7f4bccbSMingming Cao /* jbd2_journal_init_dev and jbd2_journal_init_inode: 1112470decc6SDave Kleikamp * 1113470decc6SDave Kleikamp * Create a journal structure assigned some fixed set of disk blocks to 1114470decc6SDave Kleikamp * the journal. We don't actually touch those disk blocks yet, but we 1115470decc6SDave Kleikamp * need to set up all of the mapping information to tell the journaling 1116470decc6SDave Kleikamp * system where the journal blocks are. 1117470decc6SDave Kleikamp * 1118470decc6SDave Kleikamp */ 1119470decc6SDave Kleikamp 1120470decc6SDave Kleikamp /** 11215648ba5bSRandy Dunlap * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure 1122470decc6SDave Kleikamp * @bdev: Block device on which to create the journal 1123470decc6SDave Kleikamp * @fs_dev: Device which hold journalled filesystem for this journal. 1124470decc6SDave Kleikamp * @start: Block nr Start of journal. 1125470decc6SDave Kleikamp * @len: Length of the journal in blocks. 1126470decc6SDave Kleikamp * @blocksize: blocksize of journalling device 11275648ba5bSRandy Dunlap * 11285648ba5bSRandy Dunlap * Returns: a newly created journal_t * 1129470decc6SDave Kleikamp * 1130f7f4bccbSMingming Cao * jbd2_journal_init_dev creates a journal which maps a fixed contiguous 1131470decc6SDave Kleikamp * range of blocks on an arbitrary block device. 1132470decc6SDave Kleikamp * 1133470decc6SDave Kleikamp */ 1134f7f4bccbSMingming Cao journal_t * jbd2_journal_init_dev(struct block_device *bdev, 1135470decc6SDave Kleikamp struct block_device *fs_dev, 113618eba7aaSMingming Cao unsigned long long start, int len, int blocksize) 1137470decc6SDave Kleikamp { 1138470decc6SDave Kleikamp journal_t *journal = journal_init_common(); 1139470decc6SDave Kleikamp struct buffer_head *bh; 1140470decc6SDave Kleikamp int n; 1141470decc6SDave Kleikamp 1142470decc6SDave Kleikamp if (!journal) 1143470decc6SDave Kleikamp return NULL; 1144470decc6SDave Kleikamp 1145470decc6SDave Kleikamp /* journal descriptor can store up to n blocks -bzzz */ 1146470decc6SDave Kleikamp journal->j_blocksize = blocksize; 11470587aa3dSyangsheng journal->j_dev = bdev; 11480587aa3dSyangsheng journal->j_fs_dev = fs_dev; 11490587aa3dSyangsheng journal->j_blk_offset = start; 11500587aa3dSyangsheng journal->j_maxlen = len; 11510587aa3dSyangsheng bdevname(journal->j_dev, journal->j_devname); 1152*81ae394bSRasmus Villemoes strreplace(journal->j_devname, '/', '!'); 11534b905671SJan Kara jbd2_stats_proc_init(journal); 1154470decc6SDave Kleikamp n = journal->j_blocksize / sizeof(journal_block_tag_t); 1155470decc6SDave Kleikamp journal->j_wbufsize = n; 1156470decc6SDave Kleikamp journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL); 1157470decc6SDave Kleikamp if (!journal->j_wbuf) { 115825985edcSLucas De Marchi printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n", 1159329d291fSHarvey Harrison __func__); 11604b905671SJan Kara goto out_err; 1161470decc6SDave Kleikamp } 1162470decc6SDave Kleikamp 1163470decc6SDave Kleikamp bh = __getblk(journal->j_dev, start, journal->j_blocksize); 11644b905671SJan Kara if (!bh) { 11654b905671SJan Kara printk(KERN_ERR 11664b905671SJan Kara "%s: Cannot get buffer for journal superblock\n", 11674b905671SJan Kara __func__); 11684b905671SJan Kara goto out_err; 11694b905671SJan Kara } 1170470decc6SDave Kleikamp journal->j_sb_buffer = bh; 1171470decc6SDave Kleikamp journal->j_superblock = (journal_superblock_t *)bh->b_data; 11724b905671SJan Kara 1173470decc6SDave Kleikamp return journal; 11744b905671SJan Kara out_err: 11757b02bec0STao Ma kfree(journal->j_wbuf); 11764b905671SJan Kara jbd2_stats_proc_exit(journal); 11774b905671SJan Kara kfree(journal); 11784b905671SJan Kara return NULL; 1179470decc6SDave Kleikamp } 1180470decc6SDave Kleikamp 1181470decc6SDave Kleikamp /** 1182f7f4bccbSMingming Cao * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode. 1183470decc6SDave Kleikamp * @inode: An inode to create the journal in 1184470decc6SDave Kleikamp * 1185f7f4bccbSMingming Cao * jbd2_journal_init_inode creates a journal which maps an on-disk inode as 1186470decc6SDave Kleikamp * the journal. The inode must exist already, must support bmap() and 1187470decc6SDave Kleikamp * must have all data blocks preallocated. 1188470decc6SDave Kleikamp */ 1189f7f4bccbSMingming Cao journal_t * jbd2_journal_init_inode (struct inode *inode) 1190470decc6SDave Kleikamp { 1191470decc6SDave Kleikamp struct buffer_head *bh; 1192470decc6SDave Kleikamp journal_t *journal = journal_init_common(); 119305496769STheodore Ts'o char *p; 1194470decc6SDave Kleikamp int err; 1195470decc6SDave Kleikamp int n; 119618eba7aaSMingming Cao unsigned long long blocknr; 1197470decc6SDave Kleikamp 1198470decc6SDave Kleikamp if (!journal) 1199470decc6SDave Kleikamp return NULL; 1200470decc6SDave Kleikamp 1201470decc6SDave Kleikamp journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev; 1202470decc6SDave Kleikamp journal->j_inode = inode; 120305496769STheodore Ts'o bdevname(journal->j_dev, journal->j_devname); 1204*81ae394bSRasmus Villemoes p = strreplace(journal->j_devname, '/', '!'); 120590576c0bSTheodore Ts'o sprintf(p, "-%lu", journal->j_inode->i_ino); 1206470decc6SDave Kleikamp jbd_debug(1, 1207470decc6SDave Kleikamp "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n", 1208470decc6SDave Kleikamp journal, inode->i_sb->s_id, inode->i_ino, 1209470decc6SDave Kleikamp (long long) inode->i_size, 1210470decc6SDave Kleikamp inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize); 1211470decc6SDave Kleikamp 1212470decc6SDave Kleikamp journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits; 1213470decc6SDave Kleikamp journal->j_blocksize = inode->i_sb->s_blocksize; 12148e85fb3fSJohann Lombardi jbd2_stats_proc_init(journal); 1215470decc6SDave Kleikamp 1216470decc6SDave Kleikamp /* journal descriptor can store up to n blocks -bzzz */ 1217470decc6SDave Kleikamp n = journal->j_blocksize / sizeof(journal_block_tag_t); 1218470decc6SDave Kleikamp journal->j_wbufsize = n; 1219470decc6SDave Kleikamp journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL); 1220470decc6SDave Kleikamp if (!journal->j_wbuf) { 122125985edcSLucas De Marchi printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n", 1222329d291fSHarvey Harrison __func__); 12234b905671SJan Kara goto out_err; 1224470decc6SDave Kleikamp } 1225470decc6SDave Kleikamp 1226f7f4bccbSMingming Cao err = jbd2_journal_bmap(journal, 0, &blocknr); 1227470decc6SDave Kleikamp /* If that failed, give up */ 1228470decc6SDave Kleikamp if (err) { 12293c26bdb4SJustin P. Mattock printk(KERN_ERR "%s: Cannot locate journal superblock\n", 1230329d291fSHarvey Harrison __func__); 12314b905671SJan Kara goto out_err; 1232470decc6SDave Kleikamp } 1233470decc6SDave Kleikamp 1234a49058faSGioh Kim bh = getblk_unmovable(journal->j_dev, blocknr, journal->j_blocksize); 12354b905671SJan Kara if (!bh) { 12364b905671SJan Kara printk(KERN_ERR 12374b905671SJan Kara "%s: Cannot get buffer for journal superblock\n", 12384b905671SJan Kara __func__); 12394b905671SJan Kara goto out_err; 12404b905671SJan Kara } 1241470decc6SDave Kleikamp journal->j_sb_buffer = bh; 1242470decc6SDave Kleikamp journal->j_superblock = (journal_superblock_t *)bh->b_data; 1243470decc6SDave Kleikamp 1244470decc6SDave Kleikamp return journal; 12454b905671SJan Kara out_err: 12467b02bec0STao Ma kfree(journal->j_wbuf); 12474b905671SJan Kara jbd2_stats_proc_exit(journal); 12484b905671SJan Kara kfree(journal); 12494b905671SJan Kara return NULL; 1250470decc6SDave Kleikamp } 1251470decc6SDave Kleikamp 1252470decc6SDave Kleikamp /* 1253470decc6SDave Kleikamp * If the journal init or create aborts, we need to mark the journal 1254470decc6SDave Kleikamp * superblock as being NULL to prevent the journal destroy from writing 1255470decc6SDave Kleikamp * back a bogus superblock. 1256470decc6SDave Kleikamp */ 1257470decc6SDave Kleikamp static void journal_fail_superblock (journal_t *journal) 1258470decc6SDave Kleikamp { 1259470decc6SDave Kleikamp struct buffer_head *bh = journal->j_sb_buffer; 1260470decc6SDave Kleikamp brelse(bh); 1261470decc6SDave Kleikamp journal->j_sb_buffer = NULL; 1262470decc6SDave Kleikamp } 1263470decc6SDave Kleikamp 1264470decc6SDave Kleikamp /* 1265470decc6SDave Kleikamp * Given a journal_t structure, initialise the various fields for 1266470decc6SDave Kleikamp * startup of a new journaling session. We use this both when creating 1267470decc6SDave Kleikamp * a journal, and after recovering an old journal to reset it for 1268470decc6SDave Kleikamp * subsequent use. 1269470decc6SDave Kleikamp */ 1270470decc6SDave Kleikamp 1271470decc6SDave Kleikamp static int journal_reset(journal_t *journal) 1272470decc6SDave Kleikamp { 1273470decc6SDave Kleikamp journal_superblock_t *sb = journal->j_superblock; 127418eba7aaSMingming Cao unsigned long long first, last; 1275470decc6SDave Kleikamp 1276470decc6SDave Kleikamp first = be32_to_cpu(sb->s_first); 1277470decc6SDave Kleikamp last = be32_to_cpu(sb->s_maxlen); 1278f6f50e28SJan Kara if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) { 1279f2a44523SEryu Guan printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n", 1280f6f50e28SJan Kara first, last); 1281f6f50e28SJan Kara journal_fail_superblock(journal); 1282f6f50e28SJan Kara return -EINVAL; 1283f6f50e28SJan Kara } 1284470decc6SDave Kleikamp 1285470decc6SDave Kleikamp journal->j_first = first; 1286470decc6SDave Kleikamp journal->j_last = last; 1287470decc6SDave Kleikamp 1288470decc6SDave Kleikamp journal->j_head = first; 1289470decc6SDave Kleikamp journal->j_tail = first; 1290470decc6SDave Kleikamp journal->j_free = last - first; 1291470decc6SDave Kleikamp 1292470decc6SDave Kleikamp journal->j_tail_sequence = journal->j_transaction_sequence; 1293470decc6SDave Kleikamp journal->j_commit_sequence = journal->j_transaction_sequence - 1; 1294470decc6SDave Kleikamp journal->j_commit_request = journal->j_commit_sequence; 1295470decc6SDave Kleikamp 1296470decc6SDave Kleikamp journal->j_max_transaction_buffers = journal->j_maxlen / 4; 1297470decc6SDave Kleikamp 1298470decc6SDave Kleikamp /* 1299470decc6SDave Kleikamp * As a special case, if the on-disk copy is already marked as needing 130024bcc89cSJan Kara * no recovery (s_start == 0), then we can safely defer the superblock 130124bcc89cSJan Kara * update until the next commit by setting JBD2_FLUSHED. This avoids 1302470decc6SDave Kleikamp * attempting a write to a potential-readonly device. 1303470decc6SDave Kleikamp */ 130424bcc89cSJan Kara if (sb->s_start == 0) { 1305f2a44523SEryu Guan jbd_debug(1, "JBD2: Skipping superblock update on recovered sb " 1306470decc6SDave Kleikamp "(start %ld, seq %d, errno %d)\n", 1307470decc6SDave Kleikamp journal->j_tail, journal->j_tail_sequence, 1308470decc6SDave Kleikamp journal->j_errno); 130924bcc89cSJan Kara journal->j_flags |= JBD2_FLUSHED; 131024bcc89cSJan Kara } else { 1311a78bb11dSJan Kara /* Lock here to make assertions happy... */ 1312a78bb11dSJan Kara mutex_lock(&journal->j_checkpoint_mutex); 131379feb521SJan Kara /* 131479feb521SJan Kara * Update log tail information. We use WRITE_FUA since new 131579feb521SJan Kara * transaction will start reusing journal space and so we 131679feb521SJan Kara * must make sure information about current log tail is on 131779feb521SJan Kara * disk before that. 131879feb521SJan Kara */ 131979feb521SJan Kara jbd2_journal_update_sb_log_tail(journal, 132079feb521SJan Kara journal->j_tail_sequence, 132179feb521SJan Kara journal->j_tail, 132279feb521SJan Kara WRITE_FUA); 1323a78bb11dSJan Kara mutex_unlock(&journal->j_checkpoint_mutex); 132424bcc89cSJan Kara } 132524bcc89cSJan Kara return jbd2_journal_start_thread(journal); 1326470decc6SDave Kleikamp } 1327470decc6SDave Kleikamp 132879feb521SJan Kara static void jbd2_write_superblock(journal_t *journal, int write_op) 132924bcc89cSJan Kara { 133024bcc89cSJan Kara struct buffer_head *bh = journal->j_sb_buffer; 1331fe52d17cSTheodore Ts'o journal_superblock_t *sb = journal->j_superblock; 133279feb521SJan Kara int ret; 133324bcc89cSJan Kara 133479feb521SJan Kara trace_jbd2_write_superblock(journal, write_op); 133579feb521SJan Kara if (!(journal->j_flags & JBD2_BARRIER)) 133679feb521SJan Kara write_op &= ~(REQ_FUA | REQ_FLUSH); 133779feb521SJan Kara lock_buffer(bh); 1338914258bfSTheodore Ts'o if (buffer_write_io_error(bh)) { 1339914258bfSTheodore Ts'o /* 1340914258bfSTheodore Ts'o * Oh, dear. A previous attempt to write the journal 1341914258bfSTheodore Ts'o * superblock failed. This could happen because the 1342914258bfSTheodore Ts'o * USB device was yanked out. Or it could happen to 1343914258bfSTheodore Ts'o * be a transient write error and maybe the block will 1344914258bfSTheodore Ts'o * be remapped. Nothing we can do but to retry the 1345914258bfSTheodore Ts'o * write and hope for the best. 1346914258bfSTheodore Ts'o */ 1347914258bfSTheodore Ts'o printk(KERN_ERR "JBD2: previous I/O error detected " 1348914258bfSTheodore Ts'o "for journal superblock update for %s.\n", 1349914258bfSTheodore Ts'o journal->j_devname); 1350914258bfSTheodore Ts'o clear_buffer_write_io_error(bh); 1351914258bfSTheodore Ts'o set_buffer_uptodate(bh); 1352914258bfSTheodore Ts'o } 1353fe52d17cSTheodore Ts'o jbd2_superblock_csum_set(journal, sb); 135479feb521SJan Kara get_bh(bh); 135579feb521SJan Kara bh->b_end_io = end_buffer_write_sync; 135679feb521SJan Kara ret = submit_bh(write_op, bh); 135779feb521SJan Kara wait_on_buffer(bh); 1358914258bfSTheodore Ts'o if (buffer_write_io_error(bh)) { 1359914258bfSTheodore Ts'o clear_buffer_write_io_error(bh); 1360914258bfSTheodore Ts'o set_buffer_uptodate(bh); 136179feb521SJan Kara ret = -EIO; 136279feb521SJan Kara } 136379feb521SJan Kara if (ret) { 136479feb521SJan Kara printk(KERN_ERR "JBD2: Error %d detected when updating " 136579feb521SJan Kara "journal superblock for %s.\n", ret, 136679feb521SJan Kara journal->j_devname); 1367914258bfSTheodore Ts'o } 136824bcc89cSJan Kara } 1369470decc6SDave Kleikamp 137024bcc89cSJan Kara /** 137124bcc89cSJan Kara * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk. 137224bcc89cSJan Kara * @journal: The journal to update. 137379feb521SJan Kara * @tail_tid: TID of the new transaction at the tail of the log 137479feb521SJan Kara * @tail_block: The first block of the transaction at the tail of the log 137579feb521SJan Kara * @write_op: With which operation should we write the journal sb 137624bcc89cSJan Kara * 137724bcc89cSJan Kara * Update a journal's superblock information about log tail and write it to 137824bcc89cSJan Kara * disk, waiting for the IO to complete. 137924bcc89cSJan Kara */ 138079feb521SJan Kara void jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid, 138179feb521SJan Kara unsigned long tail_block, int write_op) 138224bcc89cSJan Kara { 138324bcc89cSJan Kara journal_superblock_t *sb = journal->j_superblock; 13842201c590SSeiji Aguchi 1385a78bb11dSJan Kara BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); 138679feb521SJan Kara jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n", 138779feb521SJan Kara tail_block, tail_tid); 1388470decc6SDave Kleikamp 138979feb521SJan Kara sb->s_sequence = cpu_to_be32(tail_tid); 139079feb521SJan Kara sb->s_start = cpu_to_be32(tail_block); 139124bcc89cSJan Kara 139279feb521SJan Kara jbd2_write_superblock(journal, write_op); 139324bcc89cSJan Kara 139424bcc89cSJan Kara /* Log is no longer empty */ 1395a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 139624bcc89cSJan Kara WARN_ON(!sb->s_sequence); 1397f7f4bccbSMingming Cao journal->j_flags &= ~JBD2_FLUSHED; 139824bcc89cSJan Kara write_unlock(&journal->j_state_lock); 139924bcc89cSJan Kara } 140024bcc89cSJan Kara 140124bcc89cSJan Kara /** 140224bcc89cSJan Kara * jbd2_mark_journal_empty() - Mark on disk journal as empty. 140324bcc89cSJan Kara * @journal: The journal to update. 140424bcc89cSJan Kara * 140524bcc89cSJan Kara * Update a journal's dynamic superblock fields to show that journal is empty. 140624bcc89cSJan Kara * Write updated superblock to disk waiting for IO to complete. 140724bcc89cSJan Kara */ 140824bcc89cSJan Kara static void jbd2_mark_journal_empty(journal_t *journal) 140924bcc89cSJan Kara { 141024bcc89cSJan Kara journal_superblock_t *sb = journal->j_superblock; 141124bcc89cSJan Kara 1412a78bb11dSJan Kara BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); 1413470decc6SDave Kleikamp read_lock(&journal->j_state_lock); 1414eeecef0aSEric Sandeen /* Is it already empty? */ 1415eeecef0aSEric Sandeen if (sb->s_start == 0) { 1416eeecef0aSEric Sandeen read_unlock(&journal->j_state_lock); 1417eeecef0aSEric Sandeen return; 1418eeecef0aSEric Sandeen } 141924bcc89cSJan Kara jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n", 142024bcc89cSJan Kara journal->j_tail_sequence); 1421470decc6SDave Kleikamp 1422470decc6SDave Kleikamp sb->s_sequence = cpu_to_be32(journal->j_tail_sequence); 142324bcc89cSJan Kara sb->s_start = cpu_to_be32(0); 142424bcc89cSJan Kara read_unlock(&journal->j_state_lock); 142524bcc89cSJan Kara 142679feb521SJan Kara jbd2_write_superblock(journal, WRITE_FUA); 142724bcc89cSJan Kara 142824bcc89cSJan Kara /* Log is no longer empty */ 142924bcc89cSJan Kara write_lock(&journal->j_state_lock); 1430f7f4bccbSMingming Cao journal->j_flags |= JBD2_FLUSHED; 1431a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 1432470decc6SDave Kleikamp } 1433470decc6SDave Kleikamp 143424bcc89cSJan Kara 143524bcc89cSJan Kara /** 143624bcc89cSJan Kara * jbd2_journal_update_sb_errno() - Update error in the journal. 143724bcc89cSJan Kara * @journal: The journal to update. 143824bcc89cSJan Kara * 143924bcc89cSJan Kara * Update a journal's errno. Write updated superblock to disk waiting for IO 144024bcc89cSJan Kara * to complete. 144124bcc89cSJan Kara */ 1442d796c52eSTheodore Ts'o void jbd2_journal_update_sb_errno(journal_t *journal) 144324bcc89cSJan Kara { 144424bcc89cSJan Kara journal_superblock_t *sb = journal->j_superblock; 144524bcc89cSJan Kara 144624bcc89cSJan Kara read_lock(&journal->j_state_lock); 144724bcc89cSJan Kara jbd_debug(1, "JBD2: updating superblock error (errno %d)\n", 144824bcc89cSJan Kara journal->j_errno); 1449470decc6SDave Kleikamp sb->s_errno = cpu_to_be32(journal->j_errno); 1450470decc6SDave Kleikamp read_unlock(&journal->j_state_lock); 1451470decc6SDave Kleikamp 145279feb521SJan Kara jbd2_write_superblock(journal, WRITE_SYNC); 1453470decc6SDave Kleikamp } 1454d796c52eSTheodore Ts'o EXPORT_SYMBOL(jbd2_journal_update_sb_errno); 1455470decc6SDave Kleikamp 1456470decc6SDave Kleikamp /* 1457470decc6SDave Kleikamp * Read the superblock for a given journal, performing initial 1458470decc6SDave Kleikamp * validation of the format. 1459470decc6SDave Kleikamp */ 1460470decc6SDave Kleikamp static int journal_get_superblock(journal_t *journal) 1461470decc6SDave Kleikamp { 1462470decc6SDave Kleikamp struct buffer_head *bh; 1463470decc6SDave Kleikamp journal_superblock_t *sb; 1464470decc6SDave Kleikamp int err = -EIO; 1465470decc6SDave Kleikamp 1466470decc6SDave Kleikamp bh = journal->j_sb_buffer; 1467470decc6SDave Kleikamp 1468470decc6SDave Kleikamp J_ASSERT(bh != NULL); 1469470decc6SDave Kleikamp if (!buffer_uptodate(bh)) { 1470470decc6SDave Kleikamp ll_rw_block(READ, 1, &bh); 1471470decc6SDave Kleikamp wait_on_buffer(bh); 1472470decc6SDave Kleikamp if (!buffer_uptodate(bh)) { 1473470decc6SDave Kleikamp printk(KERN_ERR 1474f2a44523SEryu Guan "JBD2: IO error reading journal superblock\n"); 1475470decc6SDave Kleikamp goto out; 1476470decc6SDave Kleikamp } 1477470decc6SDave Kleikamp } 1478470decc6SDave Kleikamp 147925ed6e8aSDarrick J. Wong if (buffer_verified(bh)) 148025ed6e8aSDarrick J. Wong return 0; 148125ed6e8aSDarrick J. Wong 1482470decc6SDave Kleikamp sb = journal->j_superblock; 1483470decc6SDave Kleikamp 1484470decc6SDave Kleikamp err = -EINVAL; 1485470decc6SDave Kleikamp 1486f7f4bccbSMingming Cao if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) || 1487470decc6SDave Kleikamp sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) { 1488f2a44523SEryu Guan printk(KERN_WARNING "JBD2: no valid journal superblock found\n"); 1489470decc6SDave Kleikamp goto out; 1490470decc6SDave Kleikamp } 1491470decc6SDave Kleikamp 1492470decc6SDave Kleikamp switch(be32_to_cpu(sb->s_header.h_blocktype)) { 1493f7f4bccbSMingming Cao case JBD2_SUPERBLOCK_V1: 1494470decc6SDave Kleikamp journal->j_format_version = 1; 1495470decc6SDave Kleikamp break; 1496f7f4bccbSMingming Cao case JBD2_SUPERBLOCK_V2: 1497470decc6SDave Kleikamp journal->j_format_version = 2; 1498470decc6SDave Kleikamp break; 1499470decc6SDave Kleikamp default: 1500f2a44523SEryu Guan printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n"); 1501470decc6SDave Kleikamp goto out; 1502470decc6SDave Kleikamp } 1503470decc6SDave Kleikamp 1504470decc6SDave Kleikamp if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen) 1505470decc6SDave Kleikamp journal->j_maxlen = be32_to_cpu(sb->s_maxlen); 1506470decc6SDave Kleikamp else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) { 1507f2a44523SEryu Guan printk(KERN_WARNING "JBD2: journal file too short\n"); 1508470decc6SDave Kleikamp goto out; 1509470decc6SDave Kleikamp } 1510470decc6SDave Kleikamp 15118762202dSEryu Guan if (be32_to_cpu(sb->s_first) == 0 || 15128762202dSEryu Guan be32_to_cpu(sb->s_first) >= journal->j_maxlen) { 15138762202dSEryu Guan printk(KERN_WARNING 15148762202dSEryu Guan "JBD2: Invalid start block of journal: %u\n", 15158762202dSEryu Guan be32_to_cpu(sb->s_first)); 15168762202dSEryu Guan goto out; 15178762202dSEryu Guan } 15188762202dSEryu Guan 1519db9ee220SDarrick J. Wong if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2) && 1520db9ee220SDarrick J. Wong JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V3)) { 1521db9ee220SDarrick J. Wong /* Can't have checksum v2 and v3 at the same time! */ 1522db9ee220SDarrick J. Wong printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 " 1523db9ee220SDarrick J. Wong "at the same time!\n"); 1524db9ee220SDarrick J. Wong goto out; 1525db9ee220SDarrick J. Wong } 1526db9ee220SDarrick J. Wong 1527feb8c6d3SDarrick J. Wong if (jbd2_journal_has_csum_v2or3(journal) && 1528feb8c6d3SDarrick J. Wong JBD2_HAS_COMPAT_FEATURE(journal, JBD2_FEATURE_COMPAT_CHECKSUM)) { 1529feb8c6d3SDarrick J. Wong /* Can't have checksum v1 and v2 on at the same time! */ 1530feb8c6d3SDarrick J. Wong printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 " 1531feb8c6d3SDarrick J. Wong "at the same time!\n"); 1532feb8c6d3SDarrick J. Wong goto out; 1533feb8c6d3SDarrick J. Wong } 1534feb8c6d3SDarrick J. Wong 153525ed6e8aSDarrick J. Wong if (!jbd2_verify_csum_type(journal, sb)) { 1536a67c848aSDmitry Monakhov printk(KERN_ERR "JBD2: Unknown checksum type\n"); 153725ed6e8aSDarrick J. Wong goto out; 153825ed6e8aSDarrick J. Wong } 153925ed6e8aSDarrick J. Wong 154001b5adceSDarrick J. Wong /* Load the checksum driver */ 1541db9ee220SDarrick J. Wong if (jbd2_journal_has_csum_v2or3(journal)) { 154201b5adceSDarrick J. Wong journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0); 154301b5adceSDarrick J. Wong if (IS_ERR(journal->j_chksum_driver)) { 1544a67c848aSDmitry Monakhov printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n"); 154501b5adceSDarrick J. Wong err = PTR_ERR(journal->j_chksum_driver); 154601b5adceSDarrick J. Wong journal->j_chksum_driver = NULL; 154701b5adceSDarrick J. Wong goto out; 154801b5adceSDarrick J. Wong } 154901b5adceSDarrick J. Wong } 155001b5adceSDarrick J. Wong 15514fd5ea43SDarrick J. Wong /* Check superblock checksum */ 15524fd5ea43SDarrick J. Wong if (!jbd2_superblock_csum_verify(journal, sb)) { 1553a67c848aSDmitry Monakhov printk(KERN_ERR "JBD2: journal checksum error\n"); 15544fd5ea43SDarrick J. Wong goto out; 15554fd5ea43SDarrick J. Wong } 15564fd5ea43SDarrick J. Wong 15574fd5ea43SDarrick J. Wong /* Precompute checksum seed for all metadata */ 1558db9ee220SDarrick J. Wong if (jbd2_journal_has_csum_v2or3(journal)) 15594fd5ea43SDarrick J. Wong journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid, 15604fd5ea43SDarrick J. Wong sizeof(sb->s_uuid)); 15614fd5ea43SDarrick J. Wong 156225ed6e8aSDarrick J. Wong set_buffer_verified(bh); 156325ed6e8aSDarrick J. Wong 1564470decc6SDave Kleikamp return 0; 1565470decc6SDave Kleikamp 1566470decc6SDave Kleikamp out: 1567470decc6SDave Kleikamp journal_fail_superblock(journal); 1568470decc6SDave Kleikamp return err; 1569470decc6SDave Kleikamp } 1570470decc6SDave Kleikamp 1571470decc6SDave Kleikamp /* 1572470decc6SDave Kleikamp * Load the on-disk journal superblock and read the key fields into the 1573470decc6SDave Kleikamp * journal_t. 1574470decc6SDave Kleikamp */ 1575470decc6SDave Kleikamp 1576470decc6SDave Kleikamp static int load_superblock(journal_t *journal) 1577470decc6SDave Kleikamp { 1578470decc6SDave Kleikamp int err; 1579470decc6SDave Kleikamp journal_superblock_t *sb; 1580470decc6SDave Kleikamp 1581470decc6SDave Kleikamp err = journal_get_superblock(journal); 1582470decc6SDave Kleikamp if (err) 1583470decc6SDave Kleikamp return err; 1584470decc6SDave Kleikamp 1585470decc6SDave Kleikamp sb = journal->j_superblock; 1586470decc6SDave Kleikamp 1587470decc6SDave Kleikamp journal->j_tail_sequence = be32_to_cpu(sb->s_sequence); 1588470decc6SDave Kleikamp journal->j_tail = be32_to_cpu(sb->s_start); 1589470decc6SDave Kleikamp journal->j_first = be32_to_cpu(sb->s_first); 1590470decc6SDave Kleikamp journal->j_last = be32_to_cpu(sb->s_maxlen); 1591470decc6SDave Kleikamp journal->j_errno = be32_to_cpu(sb->s_errno); 1592470decc6SDave Kleikamp 1593470decc6SDave Kleikamp return 0; 1594470decc6SDave Kleikamp } 1595470decc6SDave Kleikamp 1596470decc6SDave Kleikamp 1597470decc6SDave Kleikamp /** 1598f7f4bccbSMingming Cao * int jbd2_journal_load() - Read journal from disk. 1599470decc6SDave Kleikamp * @journal: Journal to act on. 1600470decc6SDave Kleikamp * 1601470decc6SDave Kleikamp * Given a journal_t structure which tells us which disk blocks contain 1602470decc6SDave Kleikamp * a journal, read the journal from disk to initialise the in-memory 1603470decc6SDave Kleikamp * structures. 1604470decc6SDave Kleikamp */ 1605f7f4bccbSMingming Cao int jbd2_journal_load(journal_t *journal) 1606470decc6SDave Kleikamp { 1607470decc6SDave Kleikamp int err; 1608470decc6SDave Kleikamp journal_superblock_t *sb; 1609470decc6SDave Kleikamp 1610470decc6SDave Kleikamp err = load_superblock(journal); 1611470decc6SDave Kleikamp if (err) 1612470decc6SDave Kleikamp return err; 1613470decc6SDave Kleikamp 1614470decc6SDave Kleikamp sb = journal->j_superblock; 1615470decc6SDave Kleikamp /* If this is a V2 superblock, then we have to check the 1616470decc6SDave Kleikamp * features flags on it. */ 1617470decc6SDave Kleikamp 1618470decc6SDave Kleikamp if (journal->j_format_version >= 2) { 1619470decc6SDave Kleikamp if ((sb->s_feature_ro_compat & 1620f7f4bccbSMingming Cao ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) || 1621470decc6SDave Kleikamp (sb->s_feature_incompat & 1622f7f4bccbSMingming Cao ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) { 1623470decc6SDave Kleikamp printk(KERN_WARNING 1624f2a44523SEryu Guan "JBD2: Unrecognised features on journal\n"); 1625470decc6SDave Kleikamp return -EINVAL; 1626470decc6SDave Kleikamp } 1627470decc6SDave Kleikamp } 1628470decc6SDave Kleikamp 1629d2eecb03STheodore Ts'o /* 1630d2eecb03STheodore Ts'o * Create a slab for this blocksize 1631d2eecb03STheodore Ts'o */ 1632d2eecb03STheodore Ts'o err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize)); 1633d2eecb03STheodore Ts'o if (err) 1634d2eecb03STheodore Ts'o return err; 1635d2eecb03STheodore Ts'o 1636470decc6SDave Kleikamp /* Let the recovery code check whether it needs to recover any 1637470decc6SDave Kleikamp * data from the journal. */ 1638f7f4bccbSMingming Cao if (jbd2_journal_recover(journal)) 1639470decc6SDave Kleikamp goto recovery_error; 1640470decc6SDave Kleikamp 1641e6a47428STheodore Ts'o if (journal->j_failed_commit) { 1642e6a47428STheodore Ts'o printk(KERN_ERR "JBD2: journal transaction %u on %s " 1643e6a47428STheodore Ts'o "is corrupt.\n", journal->j_failed_commit, 1644e6a47428STheodore Ts'o journal->j_devname); 1645e6a47428STheodore Ts'o return -EIO; 1646e6a47428STheodore Ts'o } 1647e6a47428STheodore Ts'o 1648470decc6SDave Kleikamp /* OK, we've finished with the dynamic journal bits: 1649470decc6SDave Kleikamp * reinitialise the dynamic contents of the superblock in memory 1650470decc6SDave Kleikamp * and reset them on disk. */ 1651470decc6SDave Kleikamp if (journal_reset(journal)) 1652470decc6SDave Kleikamp goto recovery_error; 1653470decc6SDave Kleikamp 1654f7f4bccbSMingming Cao journal->j_flags &= ~JBD2_ABORT; 1655f7f4bccbSMingming Cao journal->j_flags |= JBD2_LOADED; 1656470decc6SDave Kleikamp return 0; 1657470decc6SDave Kleikamp 1658470decc6SDave Kleikamp recovery_error: 1659f2a44523SEryu Guan printk(KERN_WARNING "JBD2: recovery failed\n"); 1660470decc6SDave Kleikamp return -EIO; 1661470decc6SDave Kleikamp } 1662470decc6SDave Kleikamp 1663470decc6SDave Kleikamp /** 1664f7f4bccbSMingming Cao * void jbd2_journal_destroy() - Release a journal_t structure. 1665470decc6SDave Kleikamp * @journal: Journal to act on. 1666470decc6SDave Kleikamp * 1667470decc6SDave Kleikamp * Release a journal_t structure once it is no longer in use by the 1668470decc6SDave Kleikamp * journaled object. 166944519fafSHidehiro Kawai * Return <0 if we couldn't clean up the journal. 1670470decc6SDave Kleikamp */ 167144519fafSHidehiro Kawai int jbd2_journal_destroy(journal_t *journal) 1672470decc6SDave Kleikamp { 167344519fafSHidehiro Kawai int err = 0; 167444519fafSHidehiro Kawai 1675470decc6SDave Kleikamp /* Wait for the commit thread to wake up and die. */ 1676470decc6SDave Kleikamp journal_kill_thread(journal); 1677470decc6SDave Kleikamp 1678470decc6SDave Kleikamp /* Force a final log commit */ 1679470decc6SDave Kleikamp if (journal->j_running_transaction) 1680f7f4bccbSMingming Cao jbd2_journal_commit_transaction(journal); 1681470decc6SDave Kleikamp 1682470decc6SDave Kleikamp /* Force any old transactions to disk */ 1683470decc6SDave Kleikamp 1684470decc6SDave Kleikamp /* Totally anal locking here... */ 1685470decc6SDave Kleikamp spin_lock(&journal->j_list_lock); 1686470decc6SDave Kleikamp while (journal->j_checkpoint_transactions != NULL) { 1687470decc6SDave Kleikamp spin_unlock(&journal->j_list_lock); 16881a0d3786STheodore Ts'o mutex_lock(&journal->j_checkpoint_mutex); 1689f7f4bccbSMingming Cao jbd2_log_do_checkpoint(journal); 16901a0d3786STheodore Ts'o mutex_unlock(&journal->j_checkpoint_mutex); 1691470decc6SDave Kleikamp spin_lock(&journal->j_list_lock); 1692470decc6SDave Kleikamp } 1693470decc6SDave Kleikamp 1694470decc6SDave Kleikamp J_ASSERT(journal->j_running_transaction == NULL); 1695470decc6SDave Kleikamp J_ASSERT(journal->j_committing_transaction == NULL); 1696470decc6SDave Kleikamp J_ASSERT(journal->j_checkpoint_transactions == NULL); 1697470decc6SDave Kleikamp spin_unlock(&journal->j_list_lock); 1698470decc6SDave Kleikamp 169944519fafSHidehiro Kawai if (journal->j_sb_buffer) { 170044519fafSHidehiro Kawai if (!is_journal_aborted(journal)) { 1701a78bb11dSJan Kara mutex_lock(&journal->j_checkpoint_mutex); 170224bcc89cSJan Kara jbd2_mark_journal_empty(journal); 1703a78bb11dSJan Kara mutex_unlock(&journal->j_checkpoint_mutex); 1704a78bb11dSJan Kara } else 170544519fafSHidehiro Kawai err = -EIO; 1706470decc6SDave Kleikamp brelse(journal->j_sb_buffer); 1707470decc6SDave Kleikamp } 1708470decc6SDave Kleikamp 17098e85fb3fSJohann Lombardi if (journal->j_proc_entry) 17108e85fb3fSJohann Lombardi jbd2_stats_proc_exit(journal); 1711470decc6SDave Kleikamp iput(journal->j_inode); 1712470decc6SDave Kleikamp if (journal->j_revoke) 1713f7f4bccbSMingming Cao jbd2_journal_destroy_revoke(journal); 171401b5adceSDarrick J. Wong if (journal->j_chksum_driver) 171501b5adceSDarrick J. Wong crypto_free_shash(journal->j_chksum_driver); 1716470decc6SDave Kleikamp kfree(journal->j_wbuf); 1717470decc6SDave Kleikamp kfree(journal); 171844519fafSHidehiro Kawai 171944519fafSHidehiro Kawai return err; 1720470decc6SDave Kleikamp } 1721470decc6SDave Kleikamp 1722470decc6SDave Kleikamp 1723470decc6SDave Kleikamp /** 1724f7f4bccbSMingming Cao *int jbd2_journal_check_used_features () - Check if features specified are used. 1725470decc6SDave Kleikamp * @journal: Journal to check. 1726470decc6SDave Kleikamp * @compat: bitmask of compatible features 1727470decc6SDave Kleikamp * @ro: bitmask of features that force read-only mount 1728470decc6SDave Kleikamp * @incompat: bitmask of incompatible features 1729470decc6SDave Kleikamp * 1730470decc6SDave Kleikamp * Check whether the journal uses all of a given set of 1731470decc6SDave Kleikamp * features. Return true (non-zero) if it does. 1732470decc6SDave Kleikamp **/ 1733470decc6SDave Kleikamp 1734f7f4bccbSMingming Cao int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat, 1735470decc6SDave Kleikamp unsigned long ro, unsigned long incompat) 1736470decc6SDave Kleikamp { 1737470decc6SDave Kleikamp journal_superblock_t *sb; 1738470decc6SDave Kleikamp 1739470decc6SDave Kleikamp if (!compat && !ro && !incompat) 1740470decc6SDave Kleikamp return 1; 17411113e1b5SPatrick J. LoPresti /* Load journal superblock if it is not loaded yet. */ 17421113e1b5SPatrick J. LoPresti if (journal->j_format_version == 0 && 17431113e1b5SPatrick J. LoPresti journal_get_superblock(journal) != 0) 17441113e1b5SPatrick J. LoPresti return 0; 1745470decc6SDave Kleikamp if (journal->j_format_version == 1) 1746470decc6SDave Kleikamp return 0; 1747470decc6SDave Kleikamp 1748470decc6SDave Kleikamp sb = journal->j_superblock; 1749470decc6SDave Kleikamp 1750470decc6SDave Kleikamp if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) && 1751470decc6SDave Kleikamp ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) && 1752470decc6SDave Kleikamp ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat)) 1753470decc6SDave Kleikamp return 1; 1754470decc6SDave Kleikamp 1755470decc6SDave Kleikamp return 0; 1756470decc6SDave Kleikamp } 1757470decc6SDave Kleikamp 1758470decc6SDave Kleikamp /** 1759f7f4bccbSMingming Cao * int jbd2_journal_check_available_features() - Check feature set in journalling layer 1760470decc6SDave Kleikamp * @journal: Journal to check. 1761470decc6SDave Kleikamp * @compat: bitmask of compatible features 1762470decc6SDave Kleikamp * @ro: bitmask of features that force read-only mount 1763470decc6SDave Kleikamp * @incompat: bitmask of incompatible features 1764470decc6SDave Kleikamp * 1765470decc6SDave Kleikamp * Check whether the journaling code supports the use of 1766470decc6SDave Kleikamp * all of a given set of features on this journal. Return true 1767470decc6SDave Kleikamp * (non-zero) if it can. */ 1768470decc6SDave Kleikamp 1769f7f4bccbSMingming Cao int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat, 1770470decc6SDave Kleikamp unsigned long ro, unsigned long incompat) 1771470decc6SDave Kleikamp { 1772470decc6SDave Kleikamp if (!compat && !ro && !incompat) 1773470decc6SDave Kleikamp return 1; 1774470decc6SDave Kleikamp 1775470decc6SDave Kleikamp /* We can support any known requested features iff the 1776470decc6SDave Kleikamp * superblock is in version 2. Otherwise we fail to support any 1777470decc6SDave Kleikamp * extended sb features. */ 1778470decc6SDave Kleikamp 1779470decc6SDave Kleikamp if (journal->j_format_version != 2) 1780470decc6SDave Kleikamp return 0; 1781470decc6SDave Kleikamp 1782f7f4bccbSMingming Cao if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat && 1783f7f4bccbSMingming Cao (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro && 1784f7f4bccbSMingming Cao (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat) 1785470decc6SDave Kleikamp return 1; 1786470decc6SDave Kleikamp 1787470decc6SDave Kleikamp return 0; 1788470decc6SDave Kleikamp } 1789470decc6SDave Kleikamp 1790470decc6SDave Kleikamp /** 1791f7f4bccbSMingming Cao * int jbd2_journal_set_features () - Mark a given journal feature in the superblock 1792470decc6SDave Kleikamp * @journal: Journal to act on. 1793470decc6SDave Kleikamp * @compat: bitmask of compatible features 1794470decc6SDave Kleikamp * @ro: bitmask of features that force read-only mount 1795470decc6SDave Kleikamp * @incompat: bitmask of incompatible features 1796470decc6SDave Kleikamp * 1797470decc6SDave Kleikamp * Mark a given journal feature as present on the 1798470decc6SDave Kleikamp * superblock. Returns true if the requested features could be set. 1799470decc6SDave Kleikamp * 1800470decc6SDave Kleikamp */ 1801470decc6SDave Kleikamp 1802f7f4bccbSMingming Cao int jbd2_journal_set_features (journal_t *journal, unsigned long compat, 1803470decc6SDave Kleikamp unsigned long ro, unsigned long incompat) 1804470decc6SDave Kleikamp { 180525ed6e8aSDarrick J. Wong #define INCOMPAT_FEATURE_ON(f) \ 180625ed6e8aSDarrick J. Wong ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f))) 180725ed6e8aSDarrick J. Wong #define COMPAT_FEATURE_ON(f) \ 180825ed6e8aSDarrick J. Wong ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f))) 1809470decc6SDave Kleikamp journal_superblock_t *sb; 1810470decc6SDave Kleikamp 1811f7f4bccbSMingming Cao if (jbd2_journal_check_used_features(journal, compat, ro, incompat)) 1812470decc6SDave Kleikamp return 1; 1813470decc6SDave Kleikamp 1814f7f4bccbSMingming Cao if (!jbd2_journal_check_available_features(journal, compat, ro, incompat)) 1815470decc6SDave Kleikamp return 0; 1816470decc6SDave Kleikamp 1817db9ee220SDarrick J. Wong /* If enabling v2 checksums, turn on v3 instead */ 1818db9ee220SDarrick J. Wong if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) { 1819db9ee220SDarrick J. Wong incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2; 1820db9ee220SDarrick J. Wong incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3; 1821db9ee220SDarrick J. Wong } 1822db9ee220SDarrick J. Wong 1823db9ee220SDarrick J. Wong /* Asking for checksumming v3 and v1? Only give them v3. */ 1824db9ee220SDarrick J. Wong if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 && 182525ed6e8aSDarrick J. Wong compat & JBD2_FEATURE_COMPAT_CHECKSUM) 182625ed6e8aSDarrick J. Wong compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM; 182725ed6e8aSDarrick J. Wong 1828470decc6SDave Kleikamp jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n", 1829470decc6SDave Kleikamp compat, ro, incompat); 1830470decc6SDave Kleikamp 1831470decc6SDave Kleikamp sb = journal->j_superblock; 1832470decc6SDave Kleikamp 1833db9ee220SDarrick J. Wong /* If enabling v3 checksums, update superblock */ 1834db9ee220SDarrick J. Wong if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) { 183525ed6e8aSDarrick J. Wong sb->s_checksum_type = JBD2_CRC32C_CHKSUM; 183625ed6e8aSDarrick J. Wong sb->s_feature_compat &= 183725ed6e8aSDarrick J. Wong ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM); 183801b5adceSDarrick J. Wong 183901b5adceSDarrick J. Wong /* Load the checksum driver */ 184001b5adceSDarrick J. Wong if (journal->j_chksum_driver == NULL) { 184101b5adceSDarrick J. Wong journal->j_chksum_driver = crypto_alloc_shash("crc32c", 184201b5adceSDarrick J. Wong 0, 0); 184301b5adceSDarrick J. Wong if (IS_ERR(journal->j_chksum_driver)) { 1844a67c848aSDmitry Monakhov printk(KERN_ERR "JBD2: Cannot load crc32c " 184501b5adceSDarrick J. Wong "driver.\n"); 184601b5adceSDarrick J. Wong journal->j_chksum_driver = NULL; 184701b5adceSDarrick J. Wong return 0; 184801b5adceSDarrick J. Wong } 18494fd5ea43SDarrick J. Wong 18504fd5ea43SDarrick J. Wong /* Precompute checksum seed for all metadata */ 18514fd5ea43SDarrick J. Wong journal->j_csum_seed = jbd2_chksum(journal, ~0, 18524fd5ea43SDarrick J. Wong sb->s_uuid, 18534fd5ea43SDarrick J. Wong sizeof(sb->s_uuid)); 185425ed6e8aSDarrick J. Wong } 185532f38691SDarrick J. Wong } 185625ed6e8aSDarrick J. Wong 185725ed6e8aSDarrick J. Wong /* If enabling v1 checksums, downgrade superblock */ 185825ed6e8aSDarrick J. Wong if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM)) 185925ed6e8aSDarrick J. Wong sb->s_feature_incompat &= 1860db9ee220SDarrick J. Wong ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 | 1861db9ee220SDarrick J. Wong JBD2_FEATURE_INCOMPAT_CSUM_V3); 186225ed6e8aSDarrick J. Wong 1863470decc6SDave Kleikamp sb->s_feature_compat |= cpu_to_be32(compat); 1864470decc6SDave Kleikamp sb->s_feature_ro_compat |= cpu_to_be32(ro); 1865470decc6SDave Kleikamp sb->s_feature_incompat |= cpu_to_be32(incompat); 1866470decc6SDave Kleikamp 1867470decc6SDave Kleikamp return 1; 186825ed6e8aSDarrick J. Wong #undef COMPAT_FEATURE_ON 186925ed6e8aSDarrick J. Wong #undef INCOMPAT_FEATURE_ON 1870470decc6SDave Kleikamp } 1871470decc6SDave Kleikamp 1872818d276cSGirish Shilamkar /* 1873818d276cSGirish Shilamkar * jbd2_journal_clear_features () - Clear a given journal feature in the 1874818d276cSGirish Shilamkar * superblock 1875818d276cSGirish Shilamkar * @journal: Journal to act on. 1876818d276cSGirish Shilamkar * @compat: bitmask of compatible features 1877818d276cSGirish Shilamkar * @ro: bitmask of features that force read-only mount 1878818d276cSGirish Shilamkar * @incompat: bitmask of incompatible features 1879818d276cSGirish Shilamkar * 1880818d276cSGirish Shilamkar * Clear a given journal feature as present on the 1881818d276cSGirish Shilamkar * superblock. 1882818d276cSGirish Shilamkar */ 1883818d276cSGirish Shilamkar void jbd2_journal_clear_features(journal_t *journal, unsigned long compat, 1884818d276cSGirish Shilamkar unsigned long ro, unsigned long incompat) 1885818d276cSGirish Shilamkar { 1886818d276cSGirish Shilamkar journal_superblock_t *sb; 1887818d276cSGirish Shilamkar 1888818d276cSGirish Shilamkar jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n", 1889818d276cSGirish Shilamkar compat, ro, incompat); 1890818d276cSGirish Shilamkar 1891818d276cSGirish Shilamkar sb = journal->j_superblock; 1892818d276cSGirish Shilamkar 1893818d276cSGirish Shilamkar sb->s_feature_compat &= ~cpu_to_be32(compat); 1894818d276cSGirish Shilamkar sb->s_feature_ro_compat &= ~cpu_to_be32(ro); 1895818d276cSGirish Shilamkar sb->s_feature_incompat &= ~cpu_to_be32(incompat); 1896818d276cSGirish Shilamkar } 1897818d276cSGirish Shilamkar EXPORT_SYMBOL(jbd2_journal_clear_features); 1898470decc6SDave Kleikamp 1899470decc6SDave Kleikamp /** 1900f7f4bccbSMingming Cao * int jbd2_journal_flush () - Flush journal 1901470decc6SDave Kleikamp * @journal: Journal to act on. 1902470decc6SDave Kleikamp * 1903470decc6SDave Kleikamp * Flush all data for a given journal to disk and empty the journal. 1904470decc6SDave Kleikamp * Filesystems can use this when remounting readonly to ensure that 1905470decc6SDave Kleikamp * recovery does not need to happen on remount. 1906470decc6SDave Kleikamp */ 1907470decc6SDave Kleikamp 1908f7f4bccbSMingming Cao int jbd2_journal_flush(journal_t *journal) 1909470decc6SDave Kleikamp { 1910470decc6SDave Kleikamp int err = 0; 1911470decc6SDave Kleikamp transaction_t *transaction = NULL; 1912470decc6SDave Kleikamp 1913a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 1914470decc6SDave Kleikamp 1915470decc6SDave Kleikamp /* Force everything buffered to the log... */ 1916470decc6SDave Kleikamp if (journal->j_running_transaction) { 1917470decc6SDave Kleikamp transaction = journal->j_running_transaction; 1918f7f4bccbSMingming Cao __jbd2_log_start_commit(journal, transaction->t_tid); 1919470decc6SDave Kleikamp } else if (journal->j_committing_transaction) 1920470decc6SDave Kleikamp transaction = journal->j_committing_transaction; 1921470decc6SDave Kleikamp 1922470decc6SDave Kleikamp /* Wait for the log commit to complete... */ 1923470decc6SDave Kleikamp if (transaction) { 1924470decc6SDave Kleikamp tid_t tid = transaction->t_tid; 1925470decc6SDave Kleikamp 1926a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 1927f7f4bccbSMingming Cao jbd2_log_wait_commit(journal, tid); 1928470decc6SDave Kleikamp } else { 1929a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 1930470decc6SDave Kleikamp } 1931470decc6SDave Kleikamp 1932470decc6SDave Kleikamp /* ...and flush everything in the log out to disk. */ 1933470decc6SDave Kleikamp spin_lock(&journal->j_list_lock); 1934470decc6SDave Kleikamp while (!err && journal->j_checkpoint_transactions != NULL) { 1935470decc6SDave Kleikamp spin_unlock(&journal->j_list_lock); 193644519fafSHidehiro Kawai mutex_lock(&journal->j_checkpoint_mutex); 1937f7f4bccbSMingming Cao err = jbd2_log_do_checkpoint(journal); 193844519fafSHidehiro Kawai mutex_unlock(&journal->j_checkpoint_mutex); 1939470decc6SDave Kleikamp spin_lock(&journal->j_list_lock); 1940470decc6SDave Kleikamp } 1941470decc6SDave Kleikamp spin_unlock(&journal->j_list_lock); 194244519fafSHidehiro Kawai 194344519fafSHidehiro Kawai if (is_journal_aborted(journal)) 194444519fafSHidehiro Kawai return -EIO; 194544519fafSHidehiro Kawai 1946a78bb11dSJan Kara mutex_lock(&journal->j_checkpoint_mutex); 1947f7f4bccbSMingming Cao jbd2_cleanup_journal_tail(journal); 1948470decc6SDave Kleikamp 1949470decc6SDave Kleikamp /* Finally, mark the journal as really needing no recovery. 1950470decc6SDave Kleikamp * This sets s_start==0 in the underlying superblock, which is 1951470decc6SDave Kleikamp * the magic code for a fully-recovered superblock. Any future 1952470decc6SDave Kleikamp * commits of data to the journal will restore the current 1953470decc6SDave Kleikamp * s_start value. */ 195424bcc89cSJan Kara jbd2_mark_journal_empty(journal); 1955a78bb11dSJan Kara mutex_unlock(&journal->j_checkpoint_mutex); 1956a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 1957470decc6SDave Kleikamp J_ASSERT(!journal->j_running_transaction); 1958470decc6SDave Kleikamp J_ASSERT(!journal->j_committing_transaction); 1959470decc6SDave Kleikamp J_ASSERT(!journal->j_checkpoint_transactions); 1960470decc6SDave Kleikamp J_ASSERT(journal->j_head == journal->j_tail); 1961470decc6SDave Kleikamp J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence); 1962a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 196344519fafSHidehiro Kawai return 0; 1964470decc6SDave Kleikamp } 1965470decc6SDave Kleikamp 1966470decc6SDave Kleikamp /** 1967f7f4bccbSMingming Cao * int jbd2_journal_wipe() - Wipe journal contents 1968470decc6SDave Kleikamp * @journal: Journal to act on. 1969470decc6SDave Kleikamp * @write: flag (see below) 1970470decc6SDave Kleikamp * 1971470decc6SDave Kleikamp * Wipe out all of the contents of a journal, safely. This will produce 1972470decc6SDave Kleikamp * a warning if the journal contains any valid recovery information. 1973f7f4bccbSMingming Cao * Must be called between journal_init_*() and jbd2_journal_load(). 1974470decc6SDave Kleikamp * 1975470decc6SDave Kleikamp * If 'write' is non-zero, then we wipe out the journal on disk; otherwise 1976470decc6SDave Kleikamp * we merely suppress recovery. 1977470decc6SDave Kleikamp */ 1978470decc6SDave Kleikamp 1979f7f4bccbSMingming Cao int jbd2_journal_wipe(journal_t *journal, int write) 1980470decc6SDave Kleikamp { 1981470decc6SDave Kleikamp int err = 0; 1982470decc6SDave Kleikamp 1983f7f4bccbSMingming Cao J_ASSERT (!(journal->j_flags & JBD2_LOADED)); 1984470decc6SDave Kleikamp 1985470decc6SDave Kleikamp err = load_superblock(journal); 1986470decc6SDave Kleikamp if (err) 1987470decc6SDave Kleikamp return err; 1988470decc6SDave Kleikamp 1989470decc6SDave Kleikamp if (!journal->j_tail) 1990470decc6SDave Kleikamp goto no_recovery; 1991470decc6SDave Kleikamp 1992f2a44523SEryu Guan printk(KERN_WARNING "JBD2: %s recovery information on journal\n", 1993470decc6SDave Kleikamp write ? "Clearing" : "Ignoring"); 1994470decc6SDave Kleikamp 1995f7f4bccbSMingming Cao err = jbd2_journal_skip_recovery(journal); 1996a78bb11dSJan Kara if (write) { 1997a78bb11dSJan Kara /* Lock to make assertions happy... */ 1998a78bb11dSJan Kara mutex_lock(&journal->j_checkpoint_mutex); 199924bcc89cSJan Kara jbd2_mark_journal_empty(journal); 2000a78bb11dSJan Kara mutex_unlock(&journal->j_checkpoint_mutex); 2001a78bb11dSJan Kara } 2002470decc6SDave Kleikamp 2003470decc6SDave Kleikamp no_recovery: 2004470decc6SDave Kleikamp return err; 2005470decc6SDave Kleikamp } 2006470decc6SDave Kleikamp 2007470decc6SDave Kleikamp /* 2008470decc6SDave Kleikamp * Journal abort has very specific semantics, which we describe 2009470decc6SDave Kleikamp * for journal abort. 2010470decc6SDave Kleikamp * 2011bfcd3555SAlberto Bertogli * Two internal functions, which provide abort to the jbd layer 2012470decc6SDave Kleikamp * itself are here. 2013470decc6SDave Kleikamp */ 2014470decc6SDave Kleikamp 2015470decc6SDave Kleikamp /* 2016470decc6SDave Kleikamp * Quick version for internal journal use (doesn't lock the journal). 2017470decc6SDave Kleikamp * Aborts hard --- we mark the abort as occurred, but do _nothing_ else, 2018470decc6SDave Kleikamp * and don't attempt to make any other journal updates. 2019470decc6SDave Kleikamp */ 2020f7f4bccbSMingming Cao void __jbd2_journal_abort_hard(journal_t *journal) 2021470decc6SDave Kleikamp { 2022470decc6SDave Kleikamp transaction_t *transaction; 2023470decc6SDave Kleikamp 2024f7f4bccbSMingming Cao if (journal->j_flags & JBD2_ABORT) 2025470decc6SDave Kleikamp return; 2026470decc6SDave Kleikamp 2027470decc6SDave Kleikamp printk(KERN_ERR "Aborting journal on device %s.\n", 202805496769STheodore Ts'o journal->j_devname); 2029470decc6SDave Kleikamp 2030a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 2031f7f4bccbSMingming Cao journal->j_flags |= JBD2_ABORT; 2032470decc6SDave Kleikamp transaction = journal->j_running_transaction; 2033470decc6SDave Kleikamp if (transaction) 2034f7f4bccbSMingming Cao __jbd2_log_start_commit(journal, transaction->t_tid); 2035a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 2036470decc6SDave Kleikamp } 2037470decc6SDave Kleikamp 2038470decc6SDave Kleikamp /* Soft abort: record the abort error status in the journal superblock, 2039470decc6SDave Kleikamp * but don't do any other IO. */ 2040470decc6SDave Kleikamp static void __journal_abort_soft (journal_t *journal, int errno) 2041470decc6SDave Kleikamp { 2042f7f4bccbSMingming Cao if (journal->j_flags & JBD2_ABORT) 2043470decc6SDave Kleikamp return; 2044470decc6SDave Kleikamp 2045470decc6SDave Kleikamp if (!journal->j_errno) 2046470decc6SDave Kleikamp journal->j_errno = errno; 2047470decc6SDave Kleikamp 2048f7f4bccbSMingming Cao __jbd2_journal_abort_hard(journal); 2049470decc6SDave Kleikamp 2050470decc6SDave Kleikamp if (errno) 205124bcc89cSJan Kara jbd2_journal_update_sb_errno(journal); 2052470decc6SDave Kleikamp } 2053470decc6SDave Kleikamp 2054470decc6SDave Kleikamp /** 2055f7f4bccbSMingming Cao * void jbd2_journal_abort () - Shutdown the journal immediately. 2056470decc6SDave Kleikamp * @journal: the journal to shutdown. 2057470decc6SDave Kleikamp * @errno: an error number to record in the journal indicating 2058470decc6SDave Kleikamp * the reason for the shutdown. 2059470decc6SDave Kleikamp * 2060470decc6SDave Kleikamp * Perform a complete, immediate shutdown of the ENTIRE 2061470decc6SDave Kleikamp * journal (not of a single transaction). This operation cannot be 2062470decc6SDave Kleikamp * undone without closing and reopening the journal. 2063470decc6SDave Kleikamp * 2064f7f4bccbSMingming Cao * The jbd2_journal_abort function is intended to support higher level error 2065470decc6SDave Kleikamp * recovery mechanisms such as the ext2/ext3 remount-readonly error 2066470decc6SDave Kleikamp * mode. 2067470decc6SDave Kleikamp * 2068470decc6SDave Kleikamp * Journal abort has very specific semantics. Any existing dirty, 2069470decc6SDave Kleikamp * unjournaled buffers in the main filesystem will still be written to 2070470decc6SDave Kleikamp * disk by bdflush, but the journaling mechanism will be suspended 2071470decc6SDave Kleikamp * immediately and no further transaction commits will be honoured. 2072470decc6SDave Kleikamp * 2073470decc6SDave Kleikamp * Any dirty, journaled buffers will be written back to disk without 2074470decc6SDave Kleikamp * hitting the journal. Atomicity cannot be guaranteed on an aborted 2075470decc6SDave Kleikamp * filesystem, but we _do_ attempt to leave as much data as possible 2076470decc6SDave Kleikamp * behind for fsck to use for cleanup. 2077470decc6SDave Kleikamp * 2078470decc6SDave Kleikamp * Any attempt to get a new transaction handle on a journal which is in 2079470decc6SDave Kleikamp * ABORT state will just result in an -EROFS error return. A 2080f7f4bccbSMingming Cao * jbd2_journal_stop on an existing handle will return -EIO if we have 2081470decc6SDave Kleikamp * entered abort state during the update. 2082470decc6SDave Kleikamp * 2083470decc6SDave Kleikamp * Recursive transactions are not disturbed by journal abort until the 2084f7f4bccbSMingming Cao * final jbd2_journal_stop, which will receive the -EIO error. 2085470decc6SDave Kleikamp * 2086f7f4bccbSMingming Cao * Finally, the jbd2_journal_abort call allows the caller to supply an errno 2087470decc6SDave Kleikamp * which will be recorded (if possible) in the journal superblock. This 2088470decc6SDave Kleikamp * allows a client to record failure conditions in the middle of a 2089470decc6SDave Kleikamp * transaction without having to complete the transaction to record the 2090470decc6SDave Kleikamp * failure to disk. ext3_error, for example, now uses this 2091470decc6SDave Kleikamp * functionality. 2092470decc6SDave Kleikamp * 2093470decc6SDave Kleikamp * Errors which originate from within the journaling layer will NOT 2094470decc6SDave Kleikamp * supply an errno; a null errno implies that absolutely no further 2095470decc6SDave Kleikamp * writes are done to the journal (unless there are any already in 2096470decc6SDave Kleikamp * progress). 2097470decc6SDave Kleikamp * 2098470decc6SDave Kleikamp */ 2099470decc6SDave Kleikamp 2100f7f4bccbSMingming Cao void jbd2_journal_abort(journal_t *journal, int errno) 2101470decc6SDave Kleikamp { 2102470decc6SDave Kleikamp __journal_abort_soft(journal, errno); 2103470decc6SDave Kleikamp } 2104470decc6SDave Kleikamp 2105470decc6SDave Kleikamp /** 2106f7f4bccbSMingming Cao * int jbd2_journal_errno () - returns the journal's error state. 2107470decc6SDave Kleikamp * @journal: journal to examine. 2108470decc6SDave Kleikamp * 2109bfcd3555SAlberto Bertogli * This is the errno number set with jbd2_journal_abort(), the last 2110470decc6SDave Kleikamp * time the journal was mounted - if the journal was stopped 2111470decc6SDave Kleikamp * without calling abort this will be 0. 2112470decc6SDave Kleikamp * 2113470decc6SDave Kleikamp * If the journal has been aborted on this mount time -EROFS will 2114470decc6SDave Kleikamp * be returned. 2115470decc6SDave Kleikamp */ 2116f7f4bccbSMingming Cao int jbd2_journal_errno(journal_t *journal) 2117470decc6SDave Kleikamp { 2118470decc6SDave Kleikamp int err; 2119470decc6SDave Kleikamp 2120a931da6aSTheodore Ts'o read_lock(&journal->j_state_lock); 2121f7f4bccbSMingming Cao if (journal->j_flags & JBD2_ABORT) 2122470decc6SDave Kleikamp err = -EROFS; 2123470decc6SDave Kleikamp else 2124470decc6SDave Kleikamp err = journal->j_errno; 2125a931da6aSTheodore Ts'o read_unlock(&journal->j_state_lock); 2126470decc6SDave Kleikamp return err; 2127470decc6SDave Kleikamp } 2128470decc6SDave Kleikamp 2129470decc6SDave Kleikamp /** 2130f7f4bccbSMingming Cao * int jbd2_journal_clear_err () - clears the journal's error state 2131470decc6SDave Kleikamp * @journal: journal to act on. 2132470decc6SDave Kleikamp * 2133bfcd3555SAlberto Bertogli * An error must be cleared or acked to take a FS out of readonly 2134470decc6SDave Kleikamp * mode. 2135470decc6SDave Kleikamp */ 2136f7f4bccbSMingming Cao int jbd2_journal_clear_err(journal_t *journal) 2137470decc6SDave Kleikamp { 2138470decc6SDave Kleikamp int err = 0; 2139470decc6SDave Kleikamp 2140a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 2141f7f4bccbSMingming Cao if (journal->j_flags & JBD2_ABORT) 2142470decc6SDave Kleikamp err = -EROFS; 2143470decc6SDave Kleikamp else 2144470decc6SDave Kleikamp journal->j_errno = 0; 2145a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 2146470decc6SDave Kleikamp return err; 2147470decc6SDave Kleikamp } 2148470decc6SDave Kleikamp 2149470decc6SDave Kleikamp /** 2150f7f4bccbSMingming Cao * void jbd2_journal_ack_err() - Ack journal err. 2151470decc6SDave Kleikamp * @journal: journal to act on. 2152470decc6SDave Kleikamp * 2153bfcd3555SAlberto Bertogli * An error must be cleared or acked to take a FS out of readonly 2154470decc6SDave Kleikamp * mode. 2155470decc6SDave Kleikamp */ 2156f7f4bccbSMingming Cao void jbd2_journal_ack_err(journal_t *journal) 2157470decc6SDave Kleikamp { 2158a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 2159470decc6SDave Kleikamp if (journal->j_errno) 2160f7f4bccbSMingming Cao journal->j_flags |= JBD2_ACK_ERR; 2161a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 2162470decc6SDave Kleikamp } 2163470decc6SDave Kleikamp 2164f7f4bccbSMingming Cao int jbd2_journal_blocks_per_page(struct inode *inode) 2165470decc6SDave Kleikamp { 2166470decc6SDave Kleikamp return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); 2167470decc6SDave Kleikamp } 2168470decc6SDave Kleikamp 2169470decc6SDave Kleikamp /* 2170b517bea1SZach Brown * helper functions to deal with 32 or 64bit block numbers. 2171b517bea1SZach Brown */ 2172b517bea1SZach Brown size_t journal_tag_bytes(journal_t *journal) 2173b517bea1SZach Brown { 2174db9ee220SDarrick J. Wong size_t sz; 2175db9ee220SDarrick J. Wong 2176db9ee220SDarrick J. Wong if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V3)) 2177db9ee220SDarrick J. Wong return sizeof(journal_block_tag3_t); 2178db9ee220SDarrick J. Wong 2179db9ee220SDarrick J. Wong sz = sizeof(journal_block_tag_t); 2180c3900875SDarrick J. Wong 2181c3900875SDarrick J. Wong if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2)) 2182db9ee220SDarrick J. Wong sz += sizeof(__u16); 2183c3900875SDarrick J. Wong 2184b517bea1SZach Brown if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT)) 2185db9ee220SDarrick J. Wong return sz; 2186b517bea1SZach Brown else 2187db9ee220SDarrick J. Wong return sz - sizeof(__u32); 2188b517bea1SZach Brown } 2189b517bea1SZach Brown 2190b517bea1SZach Brown /* 2191d2eecb03STheodore Ts'o * JBD memory management 2192d2eecb03STheodore Ts'o * 2193d2eecb03STheodore Ts'o * These functions are used to allocate block-sized chunks of memory 2194d2eecb03STheodore Ts'o * used for making copies of buffer_head data. Very often it will be 2195d2eecb03STheodore Ts'o * page-sized chunks of data, but sometimes it will be in 2196d2eecb03STheodore Ts'o * sub-page-size chunks. (For example, 16k pages on Power systems 2197d2eecb03STheodore Ts'o * with a 4k block file system.) For blocks smaller than a page, we 2198d2eecb03STheodore Ts'o * use a SLAB allocator. There are slab caches for each block size, 2199d2eecb03STheodore Ts'o * which are allocated at mount time, if necessary, and we only free 2200d2eecb03STheodore Ts'o * (all of) the slab caches when/if the jbd2 module is unloaded. For 2201d2eecb03STheodore Ts'o * this reason we don't need to a mutex to protect access to 2202d2eecb03STheodore Ts'o * jbd2_slab[] allocating or releasing memory; only in 2203d2eecb03STheodore Ts'o * jbd2_journal_create_slab(). 2204d2eecb03STheodore Ts'o */ 2205d2eecb03STheodore Ts'o #define JBD2_MAX_SLABS 8 2206d2eecb03STheodore Ts'o static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS]; 2207d2eecb03STheodore Ts'o 2208d2eecb03STheodore Ts'o static const char *jbd2_slab_names[JBD2_MAX_SLABS] = { 2209d2eecb03STheodore Ts'o "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k", 2210d2eecb03STheodore Ts'o "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k" 2211d2eecb03STheodore Ts'o }; 2212d2eecb03STheodore Ts'o 2213d2eecb03STheodore Ts'o 2214d2eecb03STheodore Ts'o static void jbd2_journal_destroy_slabs(void) 2215d2eecb03STheodore Ts'o { 2216d2eecb03STheodore Ts'o int i; 2217d2eecb03STheodore Ts'o 2218d2eecb03STheodore Ts'o for (i = 0; i < JBD2_MAX_SLABS; i++) { 2219d2eecb03STheodore Ts'o if (jbd2_slab[i]) 2220d2eecb03STheodore Ts'o kmem_cache_destroy(jbd2_slab[i]); 2221d2eecb03STheodore Ts'o jbd2_slab[i] = NULL; 2222d2eecb03STheodore Ts'o } 2223d2eecb03STheodore Ts'o } 2224d2eecb03STheodore Ts'o 2225d2eecb03STheodore Ts'o static int jbd2_journal_create_slab(size_t size) 2226d2eecb03STheodore Ts'o { 222751dfacdeSThomas Gleixner static DEFINE_MUTEX(jbd2_slab_create_mutex); 2228d2eecb03STheodore Ts'o int i = order_base_2(size) - 10; 2229d2eecb03STheodore Ts'o size_t slab_size; 2230d2eecb03STheodore Ts'o 2231d2eecb03STheodore Ts'o if (size == PAGE_SIZE) 2232d2eecb03STheodore Ts'o return 0; 2233d2eecb03STheodore Ts'o 2234d2eecb03STheodore Ts'o if (i >= JBD2_MAX_SLABS) 2235d2eecb03STheodore Ts'o return -EINVAL; 2236d2eecb03STheodore Ts'o 2237d2eecb03STheodore Ts'o if (unlikely(i < 0)) 2238d2eecb03STheodore Ts'o i = 0; 223951dfacdeSThomas Gleixner mutex_lock(&jbd2_slab_create_mutex); 2240d2eecb03STheodore Ts'o if (jbd2_slab[i]) { 224151dfacdeSThomas Gleixner mutex_unlock(&jbd2_slab_create_mutex); 2242d2eecb03STheodore Ts'o return 0; /* Already created */ 2243d2eecb03STheodore Ts'o } 2244d2eecb03STheodore Ts'o 2245d2eecb03STheodore Ts'o slab_size = 1 << (i+10); 2246d2eecb03STheodore Ts'o jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size, 2247d2eecb03STheodore Ts'o slab_size, 0, NULL); 224851dfacdeSThomas Gleixner mutex_unlock(&jbd2_slab_create_mutex); 2249d2eecb03STheodore Ts'o if (!jbd2_slab[i]) { 2250d2eecb03STheodore Ts'o printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n"); 2251d2eecb03STheodore Ts'o return -ENOMEM; 2252d2eecb03STheodore Ts'o } 2253d2eecb03STheodore Ts'o return 0; 2254d2eecb03STheodore Ts'o } 2255d2eecb03STheodore Ts'o 2256d2eecb03STheodore Ts'o static struct kmem_cache *get_slab(size_t size) 2257d2eecb03STheodore Ts'o { 2258d2eecb03STheodore Ts'o int i = order_base_2(size) - 10; 2259d2eecb03STheodore Ts'o 2260d2eecb03STheodore Ts'o BUG_ON(i >= JBD2_MAX_SLABS); 2261d2eecb03STheodore Ts'o if (unlikely(i < 0)) 2262d2eecb03STheodore Ts'o i = 0; 22638ac97b74SBill Pemberton BUG_ON(jbd2_slab[i] == NULL); 2264d2eecb03STheodore Ts'o return jbd2_slab[i]; 2265d2eecb03STheodore Ts'o } 2266d2eecb03STheodore Ts'o 2267d2eecb03STheodore Ts'o void *jbd2_alloc(size_t size, gfp_t flags) 2268d2eecb03STheodore Ts'o { 2269d2eecb03STheodore Ts'o void *ptr; 2270d2eecb03STheodore Ts'o 2271d2eecb03STheodore Ts'o BUG_ON(size & (size-1)); /* Must be a power of 2 */ 2272d2eecb03STheodore Ts'o 2273d2eecb03STheodore Ts'o flags |= __GFP_REPEAT; 2274d2eecb03STheodore Ts'o if (size == PAGE_SIZE) 2275d2eecb03STheodore Ts'o ptr = (void *)__get_free_pages(flags, 0); 2276d2eecb03STheodore Ts'o else if (size > PAGE_SIZE) { 2277d2eecb03STheodore Ts'o int order = get_order(size); 2278d2eecb03STheodore Ts'o 2279d2eecb03STheodore Ts'o if (order < 3) 2280d2eecb03STheodore Ts'o ptr = (void *)__get_free_pages(flags, order); 2281d2eecb03STheodore Ts'o else 2282d2eecb03STheodore Ts'o ptr = vmalloc(size); 2283d2eecb03STheodore Ts'o } else 2284d2eecb03STheodore Ts'o ptr = kmem_cache_alloc(get_slab(size), flags); 2285d2eecb03STheodore Ts'o 2286d2eecb03STheodore Ts'o /* Check alignment; SLUB has gotten this wrong in the past, 2287d2eecb03STheodore Ts'o * and this can lead to user data corruption! */ 2288d2eecb03STheodore Ts'o BUG_ON(((unsigned long) ptr) & (size-1)); 2289d2eecb03STheodore Ts'o 2290d2eecb03STheodore Ts'o return ptr; 2291d2eecb03STheodore Ts'o } 2292d2eecb03STheodore Ts'o 2293d2eecb03STheodore Ts'o void jbd2_free(void *ptr, size_t size) 2294d2eecb03STheodore Ts'o { 2295d2eecb03STheodore Ts'o if (size == PAGE_SIZE) { 2296d2eecb03STheodore Ts'o free_pages((unsigned long)ptr, 0); 2297d2eecb03STheodore Ts'o return; 2298d2eecb03STheodore Ts'o } 2299d2eecb03STheodore Ts'o if (size > PAGE_SIZE) { 2300d2eecb03STheodore Ts'o int order = get_order(size); 2301d2eecb03STheodore Ts'o 2302d2eecb03STheodore Ts'o if (order < 3) 2303d2eecb03STheodore Ts'o free_pages((unsigned long)ptr, order); 2304d2eecb03STheodore Ts'o else 2305d2eecb03STheodore Ts'o vfree(ptr); 2306d2eecb03STheodore Ts'o return; 2307d2eecb03STheodore Ts'o } 2308d2eecb03STheodore Ts'o kmem_cache_free(get_slab(size), ptr); 2309d2eecb03STheodore Ts'o }; 2310d2eecb03STheodore Ts'o 2311d2eecb03STheodore Ts'o /* 2312470decc6SDave Kleikamp * Journal_head storage management 2313470decc6SDave Kleikamp */ 2314e18b890bSChristoph Lameter static struct kmem_cache *jbd2_journal_head_cache; 2315e23291b9SJose R. Santos #ifdef CONFIG_JBD2_DEBUG 2316470decc6SDave Kleikamp static atomic_t nr_journal_heads = ATOMIC_INIT(0); 2317470decc6SDave Kleikamp #endif 2318470decc6SDave Kleikamp 23194185a2acSYongqiang Yang static int jbd2_journal_init_journal_head_cache(void) 2320470decc6SDave Kleikamp { 2321470decc6SDave Kleikamp int retval; 2322470decc6SDave Kleikamp 23231076d17aSAl Viro J_ASSERT(jbd2_journal_head_cache == NULL); 2324a920e941SJohann Lombardi jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head", 2325470decc6SDave Kleikamp sizeof(struct journal_head), 2326470decc6SDave Kleikamp 0, /* offset */ 232777160957SMingming Cao SLAB_TEMPORARY, /* flags */ 232820c2df83SPaul Mundt NULL); /* ctor */ 2329470decc6SDave Kleikamp retval = 0; 23301076d17aSAl Viro if (!jbd2_journal_head_cache) { 2331470decc6SDave Kleikamp retval = -ENOMEM; 2332f2a44523SEryu Guan printk(KERN_EMERG "JBD2: no memory for journal_head cache\n"); 2333470decc6SDave Kleikamp } 2334470decc6SDave Kleikamp return retval; 2335470decc6SDave Kleikamp } 2336470decc6SDave Kleikamp 23374185a2acSYongqiang Yang static void jbd2_journal_destroy_journal_head_cache(void) 2338470decc6SDave Kleikamp { 23398a9362ebSDuane Griffin if (jbd2_journal_head_cache) { 2340f7f4bccbSMingming Cao kmem_cache_destroy(jbd2_journal_head_cache); 2341f7f4bccbSMingming Cao jbd2_journal_head_cache = NULL; 2342470decc6SDave Kleikamp } 23438a9362ebSDuane Griffin } 2344470decc6SDave Kleikamp 2345470decc6SDave Kleikamp /* 2346470decc6SDave Kleikamp * journal_head splicing and dicing 2347470decc6SDave Kleikamp */ 2348470decc6SDave Kleikamp static struct journal_head *journal_alloc_journal_head(void) 2349470decc6SDave Kleikamp { 2350470decc6SDave Kleikamp struct journal_head *ret; 2351470decc6SDave Kleikamp 2352e23291b9SJose R. Santos #ifdef CONFIG_JBD2_DEBUG 2353470decc6SDave Kleikamp atomic_inc(&nr_journal_heads); 2354470decc6SDave Kleikamp #endif 23555d9cf9c6SZheng Liu ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS); 23561076d17aSAl Viro if (!ret) { 2357470decc6SDave Kleikamp jbd_debug(1, "out of memory for journal_head\n"); 2358670be5a7STheodore Ts'o pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__); 23591076d17aSAl Viro while (!ret) { 2360470decc6SDave Kleikamp yield(); 23615d9cf9c6SZheng Liu ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS); 2362470decc6SDave Kleikamp } 2363470decc6SDave Kleikamp } 2364470decc6SDave Kleikamp return ret; 2365470decc6SDave Kleikamp } 2366470decc6SDave Kleikamp 2367470decc6SDave Kleikamp static void journal_free_journal_head(struct journal_head *jh) 2368470decc6SDave Kleikamp { 2369e23291b9SJose R. Santos #ifdef CONFIG_JBD2_DEBUG 2370470decc6SDave Kleikamp atomic_dec(&nr_journal_heads); 2371cd02ff0bSMingming Cao memset(jh, JBD2_POISON_FREE, sizeof(*jh)); 2372470decc6SDave Kleikamp #endif 2373f7f4bccbSMingming Cao kmem_cache_free(jbd2_journal_head_cache, jh); 2374470decc6SDave Kleikamp } 2375470decc6SDave Kleikamp 2376470decc6SDave Kleikamp /* 2377470decc6SDave Kleikamp * A journal_head is attached to a buffer_head whenever JBD has an 2378470decc6SDave Kleikamp * interest in the buffer. 2379470decc6SDave Kleikamp * 2380470decc6SDave Kleikamp * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit 2381470decc6SDave Kleikamp * is set. This bit is tested in core kernel code where we need to take 2382470decc6SDave Kleikamp * JBD-specific actions. Testing the zeroness of ->b_private is not reliable 2383470decc6SDave Kleikamp * there. 2384470decc6SDave Kleikamp * 2385470decc6SDave Kleikamp * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one. 2386470decc6SDave Kleikamp * 2387470decc6SDave Kleikamp * When a buffer has its BH_JBD bit set it is immune from being released by 2388470decc6SDave Kleikamp * core kernel code, mainly via ->b_count. 2389470decc6SDave Kleikamp * 2390de1b7941SJan Kara * A journal_head is detached from its buffer_head when the journal_head's 2391de1b7941SJan Kara * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint 2392de1b7941SJan Kara * transaction (b_cp_transaction) hold their references to b_jcount. 2393470decc6SDave Kleikamp * 2394470decc6SDave Kleikamp * Various places in the kernel want to attach a journal_head to a buffer_head 2395470decc6SDave Kleikamp * _before_ attaching the journal_head to a transaction. To protect the 2396f7f4bccbSMingming Cao * journal_head in this situation, jbd2_journal_add_journal_head elevates the 2397470decc6SDave Kleikamp * journal_head's b_jcount refcount by one. The caller must call 2398f7f4bccbSMingming Cao * jbd2_journal_put_journal_head() to undo this. 2399470decc6SDave Kleikamp * 2400470decc6SDave Kleikamp * So the typical usage would be: 2401470decc6SDave Kleikamp * 2402470decc6SDave Kleikamp * (Attach a journal_head if needed. Increments b_jcount) 2403f7f4bccbSMingming Cao * struct journal_head *jh = jbd2_journal_add_journal_head(bh); 2404470decc6SDave Kleikamp * ... 2405de1b7941SJan Kara * (Get another reference for transaction) 2406de1b7941SJan Kara * jbd2_journal_grab_journal_head(bh); 2407470decc6SDave Kleikamp * jh->b_transaction = xxx; 2408de1b7941SJan Kara * (Put original reference) 2409f7f4bccbSMingming Cao * jbd2_journal_put_journal_head(jh); 2410470decc6SDave Kleikamp */ 2411470decc6SDave Kleikamp 2412470decc6SDave Kleikamp /* 2413470decc6SDave Kleikamp * Give a buffer_head a journal_head. 2414470decc6SDave Kleikamp * 2415470decc6SDave Kleikamp * May sleep. 2416470decc6SDave Kleikamp */ 2417f7f4bccbSMingming Cao struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh) 2418470decc6SDave Kleikamp { 2419470decc6SDave Kleikamp struct journal_head *jh; 2420470decc6SDave Kleikamp struct journal_head *new_jh = NULL; 2421470decc6SDave Kleikamp 2422470decc6SDave Kleikamp repeat: 24235d9cf9c6SZheng Liu if (!buffer_jbd(bh)) 2424470decc6SDave Kleikamp new_jh = journal_alloc_journal_head(); 2425470decc6SDave Kleikamp 2426470decc6SDave Kleikamp jbd_lock_bh_journal_head(bh); 2427470decc6SDave Kleikamp if (buffer_jbd(bh)) { 2428470decc6SDave Kleikamp jh = bh2jh(bh); 2429470decc6SDave Kleikamp } else { 2430470decc6SDave Kleikamp J_ASSERT_BH(bh, 2431470decc6SDave Kleikamp (atomic_read(&bh->b_count) > 0) || 2432470decc6SDave Kleikamp (bh->b_page && bh->b_page->mapping)); 2433470decc6SDave Kleikamp 2434470decc6SDave Kleikamp if (!new_jh) { 2435470decc6SDave Kleikamp jbd_unlock_bh_journal_head(bh); 2436470decc6SDave Kleikamp goto repeat; 2437470decc6SDave Kleikamp } 2438470decc6SDave Kleikamp 2439470decc6SDave Kleikamp jh = new_jh; 2440470decc6SDave Kleikamp new_jh = NULL; /* We consumed it */ 2441470decc6SDave Kleikamp set_buffer_jbd(bh); 2442470decc6SDave Kleikamp bh->b_private = jh; 2443470decc6SDave Kleikamp jh->b_bh = bh; 2444470decc6SDave Kleikamp get_bh(bh); 2445470decc6SDave Kleikamp BUFFER_TRACE(bh, "added journal_head"); 2446470decc6SDave Kleikamp } 2447470decc6SDave Kleikamp jh->b_jcount++; 2448470decc6SDave Kleikamp jbd_unlock_bh_journal_head(bh); 2449470decc6SDave Kleikamp if (new_jh) 2450470decc6SDave Kleikamp journal_free_journal_head(new_jh); 2451470decc6SDave Kleikamp return bh->b_private; 2452470decc6SDave Kleikamp } 2453470decc6SDave Kleikamp 2454470decc6SDave Kleikamp /* 2455470decc6SDave Kleikamp * Grab a ref against this buffer_head's journal_head. If it ended up not 2456470decc6SDave Kleikamp * having a journal_head, return NULL 2457470decc6SDave Kleikamp */ 2458f7f4bccbSMingming Cao struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh) 2459470decc6SDave Kleikamp { 2460470decc6SDave Kleikamp struct journal_head *jh = NULL; 2461470decc6SDave Kleikamp 2462470decc6SDave Kleikamp jbd_lock_bh_journal_head(bh); 2463470decc6SDave Kleikamp if (buffer_jbd(bh)) { 2464470decc6SDave Kleikamp jh = bh2jh(bh); 2465470decc6SDave Kleikamp jh->b_jcount++; 2466470decc6SDave Kleikamp } 2467470decc6SDave Kleikamp jbd_unlock_bh_journal_head(bh); 2468470decc6SDave Kleikamp return jh; 2469470decc6SDave Kleikamp } 2470470decc6SDave Kleikamp 2471470decc6SDave Kleikamp static void __journal_remove_journal_head(struct buffer_head *bh) 2472470decc6SDave Kleikamp { 2473470decc6SDave Kleikamp struct journal_head *jh = bh2jh(bh); 2474470decc6SDave Kleikamp 2475470decc6SDave Kleikamp J_ASSERT_JH(jh, jh->b_jcount >= 0); 2476de1b7941SJan Kara J_ASSERT_JH(jh, jh->b_transaction == NULL); 2477de1b7941SJan Kara J_ASSERT_JH(jh, jh->b_next_transaction == NULL); 2478de1b7941SJan Kara J_ASSERT_JH(jh, jh->b_cp_transaction == NULL); 2479470decc6SDave Kleikamp J_ASSERT_JH(jh, jh->b_jlist == BJ_None); 2480470decc6SDave Kleikamp J_ASSERT_BH(bh, buffer_jbd(bh)); 2481470decc6SDave Kleikamp J_ASSERT_BH(bh, jh2bh(jh) == bh); 2482470decc6SDave Kleikamp BUFFER_TRACE(bh, "remove journal_head"); 2483470decc6SDave Kleikamp if (jh->b_frozen_data) { 2484de1b7941SJan Kara printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__); 2485af1e76d6SMingming Cao jbd2_free(jh->b_frozen_data, bh->b_size); 2486470decc6SDave Kleikamp } 2487470decc6SDave Kleikamp if (jh->b_committed_data) { 2488de1b7941SJan Kara printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__); 2489af1e76d6SMingming Cao jbd2_free(jh->b_committed_data, bh->b_size); 2490470decc6SDave Kleikamp } 2491470decc6SDave Kleikamp bh->b_private = NULL; 2492470decc6SDave Kleikamp jh->b_bh = NULL; /* debug, really */ 2493470decc6SDave Kleikamp clear_buffer_jbd(bh); 2494470decc6SDave Kleikamp journal_free_journal_head(jh); 2495470decc6SDave Kleikamp } 2496470decc6SDave Kleikamp 2497470decc6SDave Kleikamp /* 2498de1b7941SJan Kara * Drop a reference on the passed journal_head. If it fell to zero then 2499470decc6SDave Kleikamp * release the journal_head from the buffer_head. 2500470decc6SDave Kleikamp */ 2501f7f4bccbSMingming Cao void jbd2_journal_put_journal_head(struct journal_head *jh) 2502470decc6SDave Kleikamp { 2503470decc6SDave Kleikamp struct buffer_head *bh = jh2bh(jh); 2504470decc6SDave Kleikamp 2505470decc6SDave Kleikamp jbd_lock_bh_journal_head(bh); 2506470decc6SDave Kleikamp J_ASSERT_JH(jh, jh->b_jcount > 0); 2507470decc6SDave Kleikamp --jh->b_jcount; 2508de1b7941SJan Kara if (!jh->b_jcount) { 2509470decc6SDave Kleikamp __journal_remove_journal_head(bh); 2510de1b7941SJan Kara jbd_unlock_bh_journal_head(bh); 2511470decc6SDave Kleikamp __brelse(bh); 2512de1b7941SJan Kara } else 2513470decc6SDave Kleikamp jbd_unlock_bh_journal_head(bh); 2514470decc6SDave Kleikamp } 2515470decc6SDave Kleikamp 2516470decc6SDave Kleikamp /* 2517c851ed54SJan Kara * Initialize jbd inode head 2518c851ed54SJan Kara */ 2519c851ed54SJan Kara void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode) 2520c851ed54SJan Kara { 2521c851ed54SJan Kara jinode->i_transaction = NULL; 2522c851ed54SJan Kara jinode->i_next_transaction = NULL; 2523c851ed54SJan Kara jinode->i_vfs_inode = inode; 2524c851ed54SJan Kara jinode->i_flags = 0; 2525c851ed54SJan Kara INIT_LIST_HEAD(&jinode->i_list); 2526c851ed54SJan Kara } 2527c851ed54SJan Kara 2528c851ed54SJan Kara /* 2529c851ed54SJan Kara * Function to be called before we start removing inode from memory (i.e., 2530c851ed54SJan Kara * clear_inode() is a fine place to be called from). It removes inode from 2531c851ed54SJan Kara * transaction's lists. 2532c851ed54SJan Kara */ 2533c851ed54SJan Kara void jbd2_journal_release_jbd_inode(journal_t *journal, 2534c851ed54SJan Kara struct jbd2_inode *jinode) 2535c851ed54SJan Kara { 2536c851ed54SJan Kara if (!journal) 2537c851ed54SJan Kara return; 2538c851ed54SJan Kara restart: 2539c851ed54SJan Kara spin_lock(&journal->j_list_lock); 2540c851ed54SJan Kara /* Is commit writing out inode - we have to wait */ 254139e3ac25SBrian King if (test_bit(__JI_COMMIT_RUNNING, &jinode->i_flags)) { 2542c851ed54SJan Kara wait_queue_head_t *wq; 2543c851ed54SJan Kara DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING); 2544c851ed54SJan Kara wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING); 2545c851ed54SJan Kara prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); 2546c851ed54SJan Kara spin_unlock(&journal->j_list_lock); 2547c851ed54SJan Kara schedule(); 2548c851ed54SJan Kara finish_wait(wq, &wait.wait); 2549c851ed54SJan Kara goto restart; 2550c851ed54SJan Kara } 2551c851ed54SJan Kara 2552c851ed54SJan Kara if (jinode->i_transaction) { 2553c851ed54SJan Kara list_del(&jinode->i_list); 2554c851ed54SJan Kara jinode->i_transaction = NULL; 2555c851ed54SJan Kara } 2556c851ed54SJan Kara spin_unlock(&journal->j_list_lock); 2557c851ed54SJan Kara } 2558c851ed54SJan Kara 2559470decc6SDave Kleikamp 25608e85fb3fSJohann Lombardi #ifdef CONFIG_PROC_FS 25618e85fb3fSJohann Lombardi 25628e85fb3fSJohann Lombardi #define JBD2_STATS_PROC_NAME "fs/jbd2" 25638e85fb3fSJohann Lombardi 25648e85fb3fSJohann Lombardi static void __init jbd2_create_jbd_stats_proc_entry(void) 25658e85fb3fSJohann Lombardi { 25668e85fb3fSJohann Lombardi proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL); 25678e85fb3fSJohann Lombardi } 25688e85fb3fSJohann Lombardi 25698e85fb3fSJohann Lombardi static void __exit jbd2_remove_jbd_stats_proc_entry(void) 25708e85fb3fSJohann Lombardi { 25718e85fb3fSJohann Lombardi if (proc_jbd2_stats) 25728e85fb3fSJohann Lombardi remove_proc_entry(JBD2_STATS_PROC_NAME, NULL); 25738e85fb3fSJohann Lombardi } 25748e85fb3fSJohann Lombardi 25758e85fb3fSJohann Lombardi #else 25768e85fb3fSJohann Lombardi 25778e85fb3fSJohann Lombardi #define jbd2_create_jbd_stats_proc_entry() do {} while (0) 25788e85fb3fSJohann Lombardi #define jbd2_remove_jbd_stats_proc_entry() do {} while (0) 25798e85fb3fSJohann Lombardi 25808e85fb3fSJohann Lombardi #endif 25818e85fb3fSJohann Lombardi 25828aefcd55STheodore Ts'o struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache; 2583470decc6SDave Kleikamp 25844185a2acSYongqiang Yang static int __init jbd2_journal_init_handle_cache(void) 2585470decc6SDave Kleikamp { 25868aefcd55STheodore Ts'o jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY); 2587f7f4bccbSMingming Cao if (jbd2_handle_cache == NULL) { 25888aefcd55STheodore Ts'o printk(KERN_EMERG "JBD2: failed to create handle cache\n"); 25898aefcd55STheodore Ts'o return -ENOMEM; 25908aefcd55STheodore Ts'o } 25918aefcd55STheodore Ts'o jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0); 25928aefcd55STheodore Ts'o if (jbd2_inode_cache == NULL) { 25938aefcd55STheodore Ts'o printk(KERN_EMERG "JBD2: failed to create inode cache\n"); 25948aefcd55STheodore Ts'o kmem_cache_destroy(jbd2_handle_cache); 2595470decc6SDave Kleikamp return -ENOMEM; 2596470decc6SDave Kleikamp } 2597470decc6SDave Kleikamp return 0; 2598470decc6SDave Kleikamp } 2599470decc6SDave Kleikamp 2600f7f4bccbSMingming Cao static void jbd2_journal_destroy_handle_cache(void) 2601470decc6SDave Kleikamp { 2602f7f4bccbSMingming Cao if (jbd2_handle_cache) 2603f7f4bccbSMingming Cao kmem_cache_destroy(jbd2_handle_cache); 26048aefcd55STheodore Ts'o if (jbd2_inode_cache) 26058aefcd55STheodore Ts'o kmem_cache_destroy(jbd2_inode_cache); 26068aefcd55STheodore Ts'o 2607470decc6SDave Kleikamp } 2608470decc6SDave Kleikamp 2609470decc6SDave Kleikamp /* 2610470decc6SDave Kleikamp * Module startup and shutdown 2611470decc6SDave Kleikamp */ 2612470decc6SDave Kleikamp 2613470decc6SDave Kleikamp static int __init journal_init_caches(void) 2614470decc6SDave Kleikamp { 2615470decc6SDave Kleikamp int ret; 2616470decc6SDave Kleikamp 2617f7f4bccbSMingming Cao ret = jbd2_journal_init_revoke_caches(); 2618470decc6SDave Kleikamp if (ret == 0) 26194185a2acSYongqiang Yang ret = jbd2_journal_init_journal_head_cache(); 2620470decc6SDave Kleikamp if (ret == 0) 26214185a2acSYongqiang Yang ret = jbd2_journal_init_handle_cache(); 26220c2022ecSYongqiang Yang if (ret == 0) 26230c2022ecSYongqiang Yang ret = jbd2_journal_init_transaction_cache(); 2624470decc6SDave Kleikamp return ret; 2625470decc6SDave Kleikamp } 2626470decc6SDave Kleikamp 2627f7f4bccbSMingming Cao static void jbd2_journal_destroy_caches(void) 2628470decc6SDave Kleikamp { 2629f7f4bccbSMingming Cao jbd2_journal_destroy_revoke_caches(); 26304185a2acSYongqiang Yang jbd2_journal_destroy_journal_head_cache(); 2631f7f4bccbSMingming Cao jbd2_journal_destroy_handle_cache(); 26320c2022ecSYongqiang Yang jbd2_journal_destroy_transaction_cache(); 2633d2eecb03STheodore Ts'o jbd2_journal_destroy_slabs(); 2634470decc6SDave Kleikamp } 2635470decc6SDave Kleikamp 2636470decc6SDave Kleikamp static int __init journal_init(void) 2637470decc6SDave Kleikamp { 2638470decc6SDave Kleikamp int ret; 2639470decc6SDave Kleikamp 2640470decc6SDave Kleikamp BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024); 2641470decc6SDave Kleikamp 2642470decc6SDave Kleikamp ret = journal_init_caches(); 2643620de4e1SDuane Griffin if (ret == 0) { 26448e85fb3fSJohann Lombardi jbd2_create_jbd_stats_proc_entry(); 2645620de4e1SDuane Griffin } else { 2646620de4e1SDuane Griffin jbd2_journal_destroy_caches(); 2647620de4e1SDuane Griffin } 2648470decc6SDave Kleikamp return ret; 2649470decc6SDave Kleikamp } 2650470decc6SDave Kleikamp 2651470decc6SDave Kleikamp static void __exit journal_exit(void) 2652470decc6SDave Kleikamp { 2653e23291b9SJose R. Santos #ifdef CONFIG_JBD2_DEBUG 2654470decc6SDave Kleikamp int n = atomic_read(&nr_journal_heads); 2655470decc6SDave Kleikamp if (n) 265675685071SJan Kara printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n); 2657470decc6SDave Kleikamp #endif 26588e85fb3fSJohann Lombardi jbd2_remove_jbd_stats_proc_entry(); 2659f7f4bccbSMingming Cao jbd2_journal_destroy_caches(); 2660470decc6SDave Kleikamp } 2661470decc6SDave Kleikamp 2662470decc6SDave Kleikamp MODULE_LICENSE("GPL"); 2663470decc6SDave Kleikamp module_init(journal_init); 2664470decc6SDave Kleikamp module_exit(journal_exit); 2665470decc6SDave Kleikamp 2666