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); 9741617e1aSJan Kara EXPORT_SYMBOL(jbd2_journal_inode_add_write); 9841617e1aSJan Kara EXPORT_SYMBOL(jbd2_journal_inode_add_wait); 99c851ed54SJan Kara EXPORT_SYMBOL(jbd2_journal_init_jbd_inode); 100c851ed54SJan Kara EXPORT_SYMBOL(jbd2_journal_release_jbd_inode); 101c851ed54SJan Kara EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate); 1028aefcd55STheodore Ts'o EXPORT_SYMBOL(jbd2_inode_cache); 103470decc6SDave Kleikamp 104470decc6SDave Kleikamp static void __journal_abort_soft (journal_t *journal, int errno); 105d2eecb03STheodore Ts'o static int jbd2_journal_create_slab(size_t slab_size); 106470decc6SDave Kleikamp 107169f1a2aSPaul Gortmaker #ifdef CONFIG_JBD2_DEBUG 108169f1a2aSPaul Gortmaker void __jbd2_debug(int level, const char *file, const char *func, 109169f1a2aSPaul Gortmaker unsigned int line, const char *fmt, ...) 110169f1a2aSPaul Gortmaker { 111169f1a2aSPaul Gortmaker struct va_format vaf; 112169f1a2aSPaul Gortmaker va_list args; 113169f1a2aSPaul Gortmaker 114169f1a2aSPaul Gortmaker if (level > jbd2_journal_enable_debug) 115169f1a2aSPaul Gortmaker return; 116169f1a2aSPaul Gortmaker va_start(args, fmt); 117169f1a2aSPaul Gortmaker vaf.fmt = fmt; 118169f1a2aSPaul Gortmaker vaf.va = &args; 119169f1a2aSPaul Gortmaker printk(KERN_DEBUG "%s: (%s, %u): %pV\n", file, func, line, &vaf); 120169f1a2aSPaul Gortmaker va_end(args); 121169f1a2aSPaul Gortmaker } 122169f1a2aSPaul Gortmaker EXPORT_SYMBOL(__jbd2_debug); 123169f1a2aSPaul Gortmaker #endif 124169f1a2aSPaul Gortmaker 12525ed6e8aSDarrick J. Wong /* Checksumming functions */ 1267747e6d0SRashika Kheria static int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb) 12725ed6e8aSDarrick J. Wong { 1288595798cSDarrick J. Wong if (!jbd2_journal_has_csum_v2or3_feature(j)) 12925ed6e8aSDarrick J. Wong return 1; 13025ed6e8aSDarrick J. Wong 13125ed6e8aSDarrick J. Wong return sb->s_checksum_type == JBD2_CRC32C_CHKSUM; 13225ed6e8aSDarrick J. Wong } 13325ed6e8aSDarrick J. Wong 13418a6ea1eSDarrick J. Wong static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb) 1354fd5ea43SDarrick J. Wong { 13618a6ea1eSDarrick J. Wong __u32 csum; 13718a6ea1eSDarrick J. Wong __be32 old_csum; 1384fd5ea43SDarrick J. Wong 1394fd5ea43SDarrick J. Wong old_csum = sb->s_checksum; 1404fd5ea43SDarrick J. Wong sb->s_checksum = 0; 1414fd5ea43SDarrick J. Wong csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t)); 1424fd5ea43SDarrick J. Wong sb->s_checksum = old_csum; 1434fd5ea43SDarrick J. Wong 1444fd5ea43SDarrick J. Wong return cpu_to_be32(csum); 1454fd5ea43SDarrick J. Wong } 1464fd5ea43SDarrick J. Wong 1477747e6d0SRashika Kheria static int jbd2_superblock_csum_verify(journal_t *j, journal_superblock_t *sb) 1484fd5ea43SDarrick J. Wong { 149db9ee220SDarrick J. Wong if (!jbd2_journal_has_csum_v2or3(j)) 1504fd5ea43SDarrick J. Wong return 1; 1514fd5ea43SDarrick J. Wong 1524fd5ea43SDarrick J. Wong return sb->s_checksum == jbd2_superblock_csum(j, sb); 1534fd5ea43SDarrick J. Wong } 1544fd5ea43SDarrick J. Wong 1557747e6d0SRashika Kheria static void jbd2_superblock_csum_set(journal_t *j, journal_superblock_t *sb) 1564fd5ea43SDarrick J. Wong { 157db9ee220SDarrick J. Wong if (!jbd2_journal_has_csum_v2or3(j)) 1584fd5ea43SDarrick J. Wong return; 1594fd5ea43SDarrick J. Wong 1604fd5ea43SDarrick J. Wong sb->s_checksum = jbd2_superblock_csum(j, sb); 1614fd5ea43SDarrick J. Wong } 1624fd5ea43SDarrick J. Wong 163470decc6SDave Kleikamp /* 164470decc6SDave Kleikamp * Helper function used to manage commit timeouts 165470decc6SDave Kleikamp */ 166470decc6SDave Kleikamp 167470decc6SDave Kleikamp static void commit_timeout(unsigned long __data) 168470decc6SDave Kleikamp { 169470decc6SDave Kleikamp struct task_struct * p = (struct task_struct *) __data; 170470decc6SDave Kleikamp 171470decc6SDave Kleikamp wake_up_process(p); 172470decc6SDave Kleikamp } 173470decc6SDave Kleikamp 174470decc6SDave Kleikamp /* 175f7f4bccbSMingming Cao * kjournald2: The main thread function used to manage a logging device 176470decc6SDave Kleikamp * journal. 177470decc6SDave Kleikamp * 178470decc6SDave Kleikamp * This kernel thread is responsible for two things: 179470decc6SDave Kleikamp * 180470decc6SDave Kleikamp * 1) COMMIT: Every so often we need to commit the current state of the 181470decc6SDave Kleikamp * filesystem to disk. The journal thread is responsible for writing 182470decc6SDave Kleikamp * all of the metadata buffers to disk. 183470decc6SDave Kleikamp * 184470decc6SDave Kleikamp * 2) CHECKPOINT: We cannot reuse a used section of the log file until all 185470decc6SDave Kleikamp * of the data in that part of the log has been rewritten elsewhere on 186470decc6SDave Kleikamp * the disk. Flushing these old buffers to reclaim space in the log is 187470decc6SDave Kleikamp * known as checkpointing, and this thread is responsible for that job. 188470decc6SDave Kleikamp */ 189470decc6SDave Kleikamp 190f7f4bccbSMingming Cao static int kjournald2(void *arg) 191470decc6SDave Kleikamp { 192470decc6SDave Kleikamp journal_t *journal = arg; 193470decc6SDave Kleikamp transaction_t *transaction; 194470decc6SDave Kleikamp 195470decc6SDave Kleikamp /* 196470decc6SDave Kleikamp * Set up an interval timer which can be used to trigger a commit wakeup 197470decc6SDave Kleikamp * after the commit interval expires 198470decc6SDave Kleikamp */ 199470decc6SDave Kleikamp setup_timer(&journal->j_commit_timer, commit_timeout, 200470decc6SDave Kleikamp (unsigned long)current); 201470decc6SDave Kleikamp 20235c80422SNigel Cunningham set_freezable(); 20335c80422SNigel Cunningham 204470decc6SDave Kleikamp /* Record that the journal thread is running */ 205470decc6SDave Kleikamp journal->j_task = current; 206470decc6SDave Kleikamp wake_up(&journal->j_wait_done_commit); 207470decc6SDave Kleikamp 208470decc6SDave Kleikamp /* 209470decc6SDave Kleikamp * And now, wait forever for commit wakeup events. 210470decc6SDave Kleikamp */ 211a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 212470decc6SDave Kleikamp 213470decc6SDave Kleikamp loop: 214f7f4bccbSMingming Cao if (journal->j_flags & JBD2_UNMOUNT) 215470decc6SDave Kleikamp goto end_loop; 216470decc6SDave Kleikamp 217470decc6SDave Kleikamp jbd_debug(1, "commit_sequence=%d, commit_request=%d\n", 218470decc6SDave Kleikamp journal->j_commit_sequence, journal->j_commit_request); 219470decc6SDave Kleikamp 220470decc6SDave Kleikamp if (journal->j_commit_sequence != journal->j_commit_request) { 221470decc6SDave Kleikamp jbd_debug(1, "OK, requests differ\n"); 222a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 223470decc6SDave Kleikamp del_timer_sync(&journal->j_commit_timer); 224f7f4bccbSMingming Cao jbd2_journal_commit_transaction(journal); 225a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 226470decc6SDave Kleikamp goto loop; 227470decc6SDave Kleikamp } 228470decc6SDave Kleikamp 229470decc6SDave Kleikamp wake_up(&journal->j_wait_done_commit); 230470decc6SDave Kleikamp if (freezing(current)) { 231470decc6SDave Kleikamp /* 232470decc6SDave Kleikamp * The simpler the better. Flushing journal isn't a 233470decc6SDave Kleikamp * good idea, because that depends on threads that may 234470decc6SDave Kleikamp * be already stopped. 235470decc6SDave Kleikamp */ 236f7f4bccbSMingming Cao jbd_debug(1, "Now suspending kjournald2\n"); 237a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 238a0acae0eSTejun Heo try_to_freeze(); 239a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 240470decc6SDave Kleikamp } else { 241470decc6SDave Kleikamp /* 242470decc6SDave Kleikamp * We assume on resume that commits are already there, 243470decc6SDave Kleikamp * so we don't sleep 244470decc6SDave Kleikamp */ 245470decc6SDave Kleikamp DEFINE_WAIT(wait); 246470decc6SDave Kleikamp int should_sleep = 1; 247470decc6SDave Kleikamp 248470decc6SDave Kleikamp prepare_to_wait(&journal->j_wait_commit, &wait, 249470decc6SDave Kleikamp TASK_INTERRUPTIBLE); 250470decc6SDave Kleikamp if (journal->j_commit_sequence != journal->j_commit_request) 251470decc6SDave Kleikamp should_sleep = 0; 252470decc6SDave Kleikamp transaction = journal->j_running_transaction; 253470decc6SDave Kleikamp if (transaction && time_after_eq(jiffies, 254470decc6SDave Kleikamp transaction->t_expires)) 255470decc6SDave Kleikamp should_sleep = 0; 256f7f4bccbSMingming Cao if (journal->j_flags & JBD2_UNMOUNT) 257470decc6SDave Kleikamp should_sleep = 0; 258470decc6SDave Kleikamp if (should_sleep) { 259a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 260470decc6SDave Kleikamp schedule(); 261a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 262470decc6SDave Kleikamp } 263470decc6SDave Kleikamp finish_wait(&journal->j_wait_commit, &wait); 264470decc6SDave Kleikamp } 265470decc6SDave Kleikamp 266f7f4bccbSMingming Cao jbd_debug(1, "kjournald2 wakes\n"); 267470decc6SDave Kleikamp 268470decc6SDave Kleikamp /* 269470decc6SDave Kleikamp * Were we woken up by a commit wakeup event? 270470decc6SDave Kleikamp */ 271470decc6SDave Kleikamp transaction = journal->j_running_transaction; 272470decc6SDave Kleikamp if (transaction && time_after_eq(jiffies, transaction->t_expires)) { 273470decc6SDave Kleikamp journal->j_commit_request = transaction->t_tid; 274470decc6SDave Kleikamp jbd_debug(1, "woke because of timeout\n"); 275470decc6SDave Kleikamp } 276470decc6SDave Kleikamp goto loop; 277470decc6SDave Kleikamp 278470decc6SDave Kleikamp end_loop: 279a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 280470decc6SDave Kleikamp del_timer_sync(&journal->j_commit_timer); 281470decc6SDave Kleikamp journal->j_task = NULL; 282470decc6SDave Kleikamp wake_up(&journal->j_wait_done_commit); 283470decc6SDave Kleikamp jbd_debug(1, "Journal thread exiting.\n"); 284470decc6SDave Kleikamp return 0; 285470decc6SDave Kleikamp } 286470decc6SDave Kleikamp 28797f06784SPavel Emelianov static int jbd2_journal_start_thread(journal_t *journal) 288470decc6SDave Kleikamp { 28997f06784SPavel Emelianov struct task_struct *t; 29097f06784SPavel Emelianov 29190576c0bSTheodore Ts'o t = kthread_run(kjournald2, journal, "jbd2/%s", 29290576c0bSTheodore Ts'o journal->j_devname); 29397f06784SPavel Emelianov if (IS_ERR(t)) 29497f06784SPavel Emelianov return PTR_ERR(t); 29597f06784SPavel Emelianov 2961076d17aSAl Viro wait_event(journal->j_wait_done_commit, journal->j_task != NULL); 29797f06784SPavel Emelianov return 0; 298470decc6SDave Kleikamp } 299470decc6SDave Kleikamp 300470decc6SDave Kleikamp static void journal_kill_thread(journal_t *journal) 301470decc6SDave Kleikamp { 302a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 303f7f4bccbSMingming Cao journal->j_flags |= JBD2_UNMOUNT; 304470decc6SDave Kleikamp 305470decc6SDave Kleikamp while (journal->j_task) { 306a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 3073469a32aSTheodore Ts'o wake_up(&journal->j_wait_commit); 3081076d17aSAl Viro wait_event(journal->j_wait_done_commit, journal->j_task == NULL); 309a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 310470decc6SDave Kleikamp } 311a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 312470decc6SDave Kleikamp } 313470decc6SDave Kleikamp 314470decc6SDave Kleikamp /* 315f7f4bccbSMingming Cao * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal. 316470decc6SDave Kleikamp * 317470decc6SDave Kleikamp * Writes a metadata buffer to a given disk block. The actual IO is not 318470decc6SDave Kleikamp * performed but a new buffer_head is constructed which labels the data 319470decc6SDave Kleikamp * to be written with the correct destination disk block. 320470decc6SDave Kleikamp * 321470decc6SDave Kleikamp * Any magic-number escaping which needs to be done will cause a 322470decc6SDave Kleikamp * copy-out here. If the buffer happens to start with the 323f7f4bccbSMingming Cao * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the 324470decc6SDave Kleikamp * magic number is only written to the log for descripter blocks. In 325470decc6SDave Kleikamp * this case, we copy the data and replace the first word with 0, and we 326470decc6SDave Kleikamp * return a result code which indicates that this buffer needs to be 327470decc6SDave Kleikamp * marked as an escaped buffer in the corresponding log descriptor 328470decc6SDave Kleikamp * block. The missing word can then be restored when the block is read 329470decc6SDave Kleikamp * during recovery. 330470decc6SDave Kleikamp * 331470decc6SDave Kleikamp * If the source buffer has already been modified by a new transaction 332470decc6SDave Kleikamp * since we took the last commit snapshot, we use the frozen copy of 333470decc6SDave Kleikamp * that data for IO. If we end up using the existing buffer_head's data 334f5113effSJan Kara * for the write, then we have to make sure nobody modifies it while the 335f5113effSJan Kara * IO is in progress. do_get_write_access() handles this. 336470decc6SDave Kleikamp * 337f5113effSJan Kara * The function returns a pointer to the buffer_head to be used for IO. 338470decc6SDave Kleikamp * 339470decc6SDave Kleikamp * 340470decc6SDave Kleikamp * Return value: 341470decc6SDave Kleikamp * <0: Error 342470decc6SDave Kleikamp * >=0: Finished OK 343470decc6SDave Kleikamp * 344470decc6SDave Kleikamp * On success: 345470decc6SDave Kleikamp * Bit 0 set == escape performed on the data 346470decc6SDave Kleikamp * Bit 1 set == buffer copy-out performed (kfree the data after IO) 347470decc6SDave Kleikamp */ 348470decc6SDave Kleikamp 349f7f4bccbSMingming Cao int jbd2_journal_write_metadata_buffer(transaction_t *transaction, 350470decc6SDave Kleikamp struct journal_head *jh_in, 351f5113effSJan Kara struct buffer_head **bh_out, 352f5113effSJan Kara sector_t blocknr) 353470decc6SDave Kleikamp { 354470decc6SDave Kleikamp int need_copy_out = 0; 355470decc6SDave Kleikamp int done_copy_out = 0; 356470decc6SDave Kleikamp int do_escape = 0; 357470decc6SDave Kleikamp char *mapped_data; 358470decc6SDave Kleikamp struct buffer_head *new_bh; 359470decc6SDave Kleikamp struct page *new_page; 360470decc6SDave Kleikamp unsigned int new_offset; 361470decc6SDave Kleikamp struct buffer_head *bh_in = jh2bh(jh_in); 36296577c43Sdingdinghua journal_t *journal = transaction->t_journal; 363470decc6SDave Kleikamp 364470decc6SDave Kleikamp /* 365470decc6SDave Kleikamp * The buffer really shouldn't be locked: only the current committing 366470decc6SDave Kleikamp * transaction is allowed to write it, so nobody else is allowed 367470decc6SDave Kleikamp * to do any IO. 368470decc6SDave Kleikamp * 369470decc6SDave Kleikamp * akpm: except if we're journalling data, and write() output is 370470decc6SDave Kleikamp * also part of a shared mapping, and another thread has 371470decc6SDave Kleikamp * decided to launch a writepage() against this buffer. 372470decc6SDave Kleikamp */ 373470decc6SDave Kleikamp J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in)); 374470decc6SDave Kleikamp 3756ccaf3e2SMichal Hocko new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL); 37647def826STheodore Ts'o 37796577c43Sdingdinghua /* keep subsequent assertions sane */ 37896577c43Sdingdinghua atomic_set(&new_bh->b_count, 1); 379470decc6SDave Kleikamp 380f5113effSJan Kara jbd_lock_bh_state(bh_in); 381f5113effSJan Kara repeat: 382470decc6SDave Kleikamp /* 383470decc6SDave Kleikamp * If a new transaction has already done a buffer copy-out, then 384470decc6SDave Kleikamp * we use that version of the data for the commit. 385470decc6SDave Kleikamp */ 386470decc6SDave Kleikamp if (jh_in->b_frozen_data) { 387470decc6SDave Kleikamp done_copy_out = 1; 388470decc6SDave Kleikamp new_page = virt_to_page(jh_in->b_frozen_data); 389470decc6SDave Kleikamp new_offset = offset_in_page(jh_in->b_frozen_data); 390470decc6SDave Kleikamp } else { 391470decc6SDave Kleikamp new_page = jh2bh(jh_in)->b_page; 392470decc6SDave Kleikamp new_offset = offset_in_page(jh2bh(jh_in)->b_data); 393470decc6SDave Kleikamp } 394470decc6SDave Kleikamp 395303a8f2aSCong Wang mapped_data = kmap_atomic(new_page); 396470decc6SDave Kleikamp /* 39713ceef09SJan Kara * Fire data frozen trigger if data already wasn't frozen. Do this 39813ceef09SJan Kara * before checking for escaping, as the trigger may modify the magic 39913ceef09SJan Kara * offset. If a copy-out happens afterwards, it will have the correct 40013ceef09SJan Kara * data in the buffer. 401e06c8227SJoel Becker */ 40213ceef09SJan Kara if (!done_copy_out) 40313ceef09SJan Kara jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset, 40413ceef09SJan Kara jh_in->b_triggers); 405e06c8227SJoel Becker 406e06c8227SJoel Becker /* 407470decc6SDave Kleikamp * Check for escaping 408470decc6SDave Kleikamp */ 409470decc6SDave Kleikamp if (*((__be32 *)(mapped_data + new_offset)) == 410f7f4bccbSMingming Cao cpu_to_be32(JBD2_MAGIC_NUMBER)) { 411470decc6SDave Kleikamp need_copy_out = 1; 412470decc6SDave Kleikamp do_escape = 1; 413470decc6SDave Kleikamp } 414303a8f2aSCong Wang kunmap_atomic(mapped_data); 415470decc6SDave Kleikamp 416470decc6SDave Kleikamp /* 417470decc6SDave Kleikamp * Do we need to do a data copy? 418470decc6SDave Kleikamp */ 419470decc6SDave Kleikamp if (need_copy_out && !done_copy_out) { 420470decc6SDave Kleikamp char *tmp; 421470decc6SDave Kleikamp 422470decc6SDave Kleikamp jbd_unlock_bh_state(bh_in); 423af1e76d6SMingming Cao tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS); 424e6ec116bSTheodore Ts'o if (!tmp) { 425f5113effSJan Kara brelse(new_bh); 426e6ec116bSTheodore Ts'o return -ENOMEM; 427e6ec116bSTheodore Ts'o } 428470decc6SDave Kleikamp jbd_lock_bh_state(bh_in); 429470decc6SDave Kleikamp if (jh_in->b_frozen_data) { 430af1e76d6SMingming Cao jbd2_free(tmp, bh_in->b_size); 431470decc6SDave Kleikamp goto repeat; 432470decc6SDave Kleikamp } 433470decc6SDave Kleikamp 434470decc6SDave Kleikamp jh_in->b_frozen_data = tmp; 435303a8f2aSCong Wang mapped_data = kmap_atomic(new_page); 436f5113effSJan Kara memcpy(tmp, mapped_data + new_offset, bh_in->b_size); 437303a8f2aSCong Wang kunmap_atomic(mapped_data); 438470decc6SDave Kleikamp 439470decc6SDave Kleikamp new_page = virt_to_page(tmp); 440470decc6SDave Kleikamp new_offset = offset_in_page(tmp); 441470decc6SDave Kleikamp done_copy_out = 1; 442e06c8227SJoel Becker 443e06c8227SJoel Becker /* 444e06c8227SJoel Becker * This isn't strictly necessary, as we're using frozen 445e06c8227SJoel Becker * data for the escaping, but it keeps consistency with 446e06c8227SJoel Becker * b_frozen_data usage. 447e06c8227SJoel Becker */ 448e06c8227SJoel Becker jh_in->b_frozen_triggers = jh_in->b_triggers; 449470decc6SDave Kleikamp } 450470decc6SDave Kleikamp 451470decc6SDave Kleikamp /* 452470decc6SDave Kleikamp * Did we need to do an escaping? Now we've done all the 453470decc6SDave Kleikamp * copying, we can finally do so. 454470decc6SDave Kleikamp */ 455470decc6SDave Kleikamp if (do_escape) { 456303a8f2aSCong Wang mapped_data = kmap_atomic(new_page); 457470decc6SDave Kleikamp *((unsigned int *)(mapped_data + new_offset)) = 0; 458303a8f2aSCong Wang kunmap_atomic(mapped_data); 459470decc6SDave Kleikamp } 460470decc6SDave Kleikamp 461470decc6SDave Kleikamp set_bh_page(new_bh, new_page, new_offset); 462f5113effSJan Kara new_bh->b_size = bh_in->b_size; 463f5113effSJan Kara new_bh->b_bdev = journal->j_dev; 464470decc6SDave Kleikamp new_bh->b_blocknr = blocknr; 465b34090e5SJan Kara new_bh->b_private = bh_in; 466470decc6SDave Kleikamp set_buffer_mapped(new_bh); 467470decc6SDave Kleikamp set_buffer_dirty(new_bh); 468470decc6SDave Kleikamp 469f5113effSJan Kara *bh_out = new_bh; 470470decc6SDave Kleikamp 471470decc6SDave Kleikamp /* 472470decc6SDave Kleikamp * The to-be-written buffer needs to get moved to the io queue, 473470decc6SDave Kleikamp * and the original buffer whose contents we are shadowing or 474470decc6SDave Kleikamp * copying is moved to the transaction's shadow queue. 475470decc6SDave Kleikamp */ 476470decc6SDave Kleikamp JBUFFER_TRACE(jh_in, "file as BJ_Shadow"); 47796577c43Sdingdinghua spin_lock(&journal->j_list_lock); 47896577c43Sdingdinghua __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow); 47996577c43Sdingdinghua spin_unlock(&journal->j_list_lock); 480b34090e5SJan Kara set_buffer_shadow(bh_in); 48196577c43Sdingdinghua jbd_unlock_bh_state(bh_in); 48296577c43Sdingdinghua 483470decc6SDave Kleikamp return do_escape | (done_copy_out << 1); 484470decc6SDave Kleikamp } 485470decc6SDave Kleikamp 486470decc6SDave Kleikamp /* 487470decc6SDave Kleikamp * Allocation code for the journal file. Manage the space left in the 488470decc6SDave Kleikamp * journal, so that we can begin checkpointing when appropriate. 489470decc6SDave Kleikamp */ 490470decc6SDave Kleikamp 491470decc6SDave Kleikamp /* 492e4471831STheodore Ts'o * Called with j_state_lock locked for writing. 493e4471831STheodore Ts'o * Returns true if a transaction commit was started. 494470decc6SDave Kleikamp */ 495f7f4bccbSMingming Cao int __jbd2_log_start_commit(journal_t *journal, tid_t target) 496470decc6SDave Kleikamp { 497e7b04ac0SEric Sandeen /* Return if the txn has already requested to be committed */ 498e7b04ac0SEric Sandeen if (journal->j_commit_request == target) 499e7b04ac0SEric Sandeen return 0; 500e7b04ac0SEric Sandeen 501470decc6SDave Kleikamp /* 502deeeaf13STheodore Ts'o * The only transaction we can possibly wait upon is the 503deeeaf13STheodore Ts'o * currently running transaction (if it exists). Otherwise, 504deeeaf13STheodore Ts'o * the target tid must be an old one. 505470decc6SDave Kleikamp */ 506deeeaf13STheodore Ts'o if (journal->j_running_transaction && 507deeeaf13STheodore Ts'o journal->j_running_transaction->t_tid == target) { 508470decc6SDave Kleikamp /* 509bcf3d0bcSAndrea Gelmini * We want a new commit: OK, mark the request and wakeup the 510470decc6SDave Kleikamp * commit thread. We do _not_ do the commit ourselves. 511470decc6SDave Kleikamp */ 512470decc6SDave Kleikamp 513470decc6SDave Kleikamp journal->j_commit_request = target; 514f2a44523SEryu Guan jbd_debug(1, "JBD2: requesting commit %d/%d\n", 515470decc6SDave Kleikamp journal->j_commit_request, 516470decc6SDave Kleikamp journal->j_commit_sequence); 5179fff24aaSTheodore Ts'o journal->j_running_transaction->t_requested = jiffies; 518470decc6SDave Kleikamp wake_up(&journal->j_wait_commit); 519470decc6SDave Kleikamp return 1; 520deeeaf13STheodore Ts'o } else if (!tid_geq(journal->j_commit_request, target)) 521deeeaf13STheodore Ts'o /* This should never happen, but if it does, preserve 522deeeaf13STheodore Ts'o the evidence before kjournald goes into a loop and 523deeeaf13STheodore Ts'o increments j_commit_sequence beyond all recognition. */ 524f2a44523SEryu Guan WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n", 5251be2add6STheodore Ts'o journal->j_commit_request, 5261be2add6STheodore Ts'o journal->j_commit_sequence, 527deeeaf13STheodore Ts'o target, journal->j_running_transaction ? 528deeeaf13STheodore Ts'o journal->j_running_transaction->t_tid : 0); 529470decc6SDave Kleikamp return 0; 530470decc6SDave Kleikamp } 531470decc6SDave Kleikamp 532f7f4bccbSMingming Cao int jbd2_log_start_commit(journal_t *journal, tid_t tid) 533470decc6SDave Kleikamp { 534470decc6SDave Kleikamp int ret; 535470decc6SDave Kleikamp 536a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 537f7f4bccbSMingming Cao ret = __jbd2_log_start_commit(journal, tid); 538a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 539470decc6SDave Kleikamp return ret; 540470decc6SDave Kleikamp } 541470decc6SDave Kleikamp 542470decc6SDave Kleikamp /* 5439ff86446SDmitry Monakhov * Force and wait any uncommitted transactions. We can only force the running 5449ff86446SDmitry Monakhov * transaction if we don't have an active handle, otherwise, we will deadlock. 5459ff86446SDmitry Monakhov * Returns: <0 in case of error, 5469ff86446SDmitry Monakhov * 0 if nothing to commit, 5479ff86446SDmitry Monakhov * 1 if transaction was successfully committed. 548470decc6SDave Kleikamp */ 5499ff86446SDmitry Monakhov static int __jbd2_journal_force_commit(journal_t *journal) 550470decc6SDave Kleikamp { 551470decc6SDave Kleikamp transaction_t *transaction = NULL; 552470decc6SDave Kleikamp tid_t tid; 5539ff86446SDmitry Monakhov int need_to_start = 0, ret = 0; 554470decc6SDave Kleikamp 555a931da6aSTheodore Ts'o read_lock(&journal->j_state_lock); 556470decc6SDave Kleikamp if (journal->j_running_transaction && !current->journal_info) { 557470decc6SDave Kleikamp transaction = journal->j_running_transaction; 558e4471831STheodore Ts'o if (!tid_geq(journal->j_commit_request, transaction->t_tid)) 559e4471831STheodore Ts'o need_to_start = 1; 560470decc6SDave Kleikamp } else if (journal->j_committing_transaction) 561470decc6SDave Kleikamp transaction = journal->j_committing_transaction; 562470decc6SDave Kleikamp 563470decc6SDave Kleikamp if (!transaction) { 5649ff86446SDmitry Monakhov /* Nothing to commit */ 565a931da6aSTheodore Ts'o read_unlock(&journal->j_state_lock); 5669ff86446SDmitry Monakhov return 0; 567470decc6SDave Kleikamp } 568470decc6SDave Kleikamp tid = transaction->t_tid; 569a931da6aSTheodore Ts'o read_unlock(&journal->j_state_lock); 570e4471831STheodore Ts'o if (need_to_start) 571e4471831STheodore Ts'o jbd2_log_start_commit(journal, tid); 5729ff86446SDmitry Monakhov ret = jbd2_log_wait_commit(journal, tid); 5739ff86446SDmitry Monakhov if (!ret) 5749ff86446SDmitry Monakhov ret = 1; 5759ff86446SDmitry Monakhov 5769ff86446SDmitry Monakhov return ret; 5779ff86446SDmitry Monakhov } 5789ff86446SDmitry Monakhov 5799ff86446SDmitry Monakhov /** 5809ff86446SDmitry Monakhov * Force and wait upon a commit if the calling process is not within 5819ff86446SDmitry Monakhov * transaction. This is used for forcing out undo-protected data which contains 5829ff86446SDmitry Monakhov * bitmaps, when the fs is running out of space. 5839ff86446SDmitry Monakhov * 5849ff86446SDmitry Monakhov * @journal: journal to force 5859ff86446SDmitry Monakhov * Returns true if progress was made. 5869ff86446SDmitry Monakhov */ 5879ff86446SDmitry Monakhov int jbd2_journal_force_commit_nested(journal_t *journal) 5889ff86446SDmitry Monakhov { 5899ff86446SDmitry Monakhov int ret; 5909ff86446SDmitry Monakhov 5919ff86446SDmitry Monakhov ret = __jbd2_journal_force_commit(journal); 5929ff86446SDmitry Monakhov return ret > 0; 5939ff86446SDmitry Monakhov } 5949ff86446SDmitry Monakhov 5959ff86446SDmitry Monakhov /** 5969ff86446SDmitry Monakhov * int journal_force_commit() - force any uncommitted transactions 5979ff86446SDmitry Monakhov * @journal: journal to force 5989ff86446SDmitry Monakhov * 5999ff86446SDmitry Monakhov * Caller want unconditional commit. We can only force the running transaction 6009ff86446SDmitry Monakhov * if we don't have an active handle, otherwise, we will deadlock. 6019ff86446SDmitry Monakhov */ 6029ff86446SDmitry Monakhov int jbd2_journal_force_commit(journal_t *journal) 6039ff86446SDmitry Monakhov { 6049ff86446SDmitry Monakhov int ret; 6059ff86446SDmitry Monakhov 6069ff86446SDmitry Monakhov J_ASSERT(!current->journal_info); 6079ff86446SDmitry Monakhov ret = __jbd2_journal_force_commit(journal); 6089ff86446SDmitry Monakhov if (ret > 0) 6099ff86446SDmitry Monakhov ret = 0; 6109ff86446SDmitry Monakhov return ret; 611470decc6SDave Kleikamp } 612470decc6SDave Kleikamp 613470decc6SDave Kleikamp /* 614470decc6SDave Kleikamp * Start a commit of the current running transaction (if any). Returns true 615c88ccea3SJan Kara * if a transaction is going to be committed (or is currently already 616c88ccea3SJan Kara * committing), and fills its tid in at *ptid 617470decc6SDave Kleikamp */ 618f7f4bccbSMingming Cao int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid) 619470decc6SDave Kleikamp { 620470decc6SDave Kleikamp int ret = 0; 621470decc6SDave Kleikamp 622a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 623470decc6SDave Kleikamp if (journal->j_running_transaction) { 624470decc6SDave Kleikamp tid_t tid = journal->j_running_transaction->t_tid; 625470decc6SDave Kleikamp 626c88ccea3SJan Kara __jbd2_log_start_commit(journal, tid); 627c88ccea3SJan Kara /* There's a running transaction and we've just made sure 628c88ccea3SJan Kara * it's commit has been scheduled. */ 629c88ccea3SJan Kara if (ptid) 630470decc6SDave Kleikamp *ptid = tid; 631c88ccea3SJan Kara ret = 1; 632c88ccea3SJan Kara } else if (journal->j_committing_transaction) { 633470decc6SDave Kleikamp /* 63412810ad7SArtem Bityutskiy * If commit has been started, then we have to wait for 63512810ad7SArtem Bityutskiy * completion of that transaction. 636470decc6SDave Kleikamp */ 637c88ccea3SJan Kara if (ptid) 638470decc6SDave Kleikamp *ptid = journal->j_committing_transaction->t_tid; 639470decc6SDave Kleikamp ret = 1; 640470decc6SDave Kleikamp } 641a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 642470decc6SDave Kleikamp return ret; 643470decc6SDave Kleikamp } 644470decc6SDave Kleikamp 645470decc6SDave Kleikamp /* 646bbd2be36SJan Kara * Return 1 if a given transaction has not yet sent barrier request 647bbd2be36SJan Kara * connected with a transaction commit. If 0 is returned, transaction 648bbd2be36SJan Kara * may or may not have sent the barrier. Used to avoid sending barrier 649bbd2be36SJan Kara * twice in common cases. 650bbd2be36SJan Kara */ 651bbd2be36SJan Kara int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid) 652bbd2be36SJan Kara { 653bbd2be36SJan Kara int ret = 0; 654bbd2be36SJan Kara transaction_t *commit_trans; 655bbd2be36SJan Kara 656bbd2be36SJan Kara if (!(journal->j_flags & JBD2_BARRIER)) 657bbd2be36SJan Kara return 0; 658bbd2be36SJan Kara read_lock(&journal->j_state_lock); 659bbd2be36SJan Kara /* Transaction already committed? */ 660bbd2be36SJan Kara if (tid_geq(journal->j_commit_sequence, tid)) 661bbd2be36SJan Kara goto out; 662bbd2be36SJan Kara commit_trans = journal->j_committing_transaction; 663bbd2be36SJan Kara if (!commit_trans || commit_trans->t_tid != tid) { 664bbd2be36SJan Kara ret = 1; 665bbd2be36SJan Kara goto out; 666bbd2be36SJan Kara } 667bbd2be36SJan Kara /* 668bbd2be36SJan Kara * Transaction is being committed and we already proceeded to 669bbd2be36SJan Kara * submitting a flush to fs partition? 670bbd2be36SJan Kara */ 671bbd2be36SJan Kara if (journal->j_fs_dev != journal->j_dev) { 672bbd2be36SJan Kara if (!commit_trans->t_need_data_flush || 673bbd2be36SJan Kara commit_trans->t_state >= T_COMMIT_DFLUSH) 674bbd2be36SJan Kara goto out; 675bbd2be36SJan Kara } else { 676bbd2be36SJan Kara if (commit_trans->t_state >= T_COMMIT_JFLUSH) 677bbd2be36SJan Kara goto out; 678bbd2be36SJan Kara } 679bbd2be36SJan Kara ret = 1; 680bbd2be36SJan Kara out: 681bbd2be36SJan Kara read_unlock(&journal->j_state_lock); 682bbd2be36SJan Kara return ret; 683bbd2be36SJan Kara } 684bbd2be36SJan Kara EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier); 685bbd2be36SJan Kara 686bbd2be36SJan Kara /* 687470decc6SDave Kleikamp * Wait for a specified commit to complete. 688470decc6SDave Kleikamp * The caller may not hold the journal lock. 689470decc6SDave Kleikamp */ 690f7f4bccbSMingming Cao int jbd2_log_wait_commit(journal_t *journal, tid_t tid) 691470decc6SDave Kleikamp { 692470decc6SDave Kleikamp int err = 0; 693470decc6SDave Kleikamp 694a931da6aSTheodore Ts'o read_lock(&journal->j_state_lock); 695e23291b9SJose R. Santos #ifdef CONFIG_JBD2_DEBUG 696470decc6SDave Kleikamp if (!tid_geq(journal->j_commit_request, tid)) { 69775685071SJan Kara printk(KERN_ERR 698470decc6SDave Kleikamp "%s: error: j_commit_request=%d, tid=%d\n", 699329d291fSHarvey Harrison __func__, journal->j_commit_request, tid); 700470decc6SDave Kleikamp } 701470decc6SDave Kleikamp #endif 702470decc6SDave Kleikamp while (tid_gt(tid, journal->j_commit_sequence)) { 703f2a44523SEryu Guan jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n", 704470decc6SDave Kleikamp tid, journal->j_commit_sequence); 705a931da6aSTheodore Ts'o read_unlock(&journal->j_state_lock); 7063469a32aSTheodore Ts'o wake_up(&journal->j_wait_commit); 707470decc6SDave Kleikamp wait_event(journal->j_wait_done_commit, 708470decc6SDave Kleikamp !tid_gt(tid, journal->j_commit_sequence)); 709a931da6aSTheodore Ts'o read_lock(&journal->j_state_lock); 710470decc6SDave Kleikamp } 711a931da6aSTheodore Ts'o read_unlock(&journal->j_state_lock); 712470decc6SDave Kleikamp 71375685071SJan Kara if (unlikely(is_journal_aborted(journal))) 714470decc6SDave Kleikamp err = -EIO; 715470decc6SDave Kleikamp return err; 716470decc6SDave Kleikamp } 717470decc6SDave Kleikamp 718470decc6SDave Kleikamp /* 719d76a3a77STheodore Ts'o * When this function returns the transaction corresponding to tid 720d76a3a77STheodore Ts'o * will be completed. If the transaction has currently running, start 721d76a3a77STheodore Ts'o * committing that transaction before waiting for it to complete. If 722d76a3a77STheodore Ts'o * the transaction id is stale, it is by definition already completed, 723d76a3a77STheodore Ts'o * so just return SUCCESS. 724d76a3a77STheodore Ts'o */ 725d76a3a77STheodore Ts'o int jbd2_complete_transaction(journal_t *journal, tid_t tid) 726d76a3a77STheodore Ts'o { 727d76a3a77STheodore Ts'o int need_to_wait = 1; 728d76a3a77STheodore Ts'o 729d76a3a77STheodore Ts'o read_lock(&journal->j_state_lock); 730d76a3a77STheodore Ts'o if (journal->j_running_transaction && 731d76a3a77STheodore Ts'o journal->j_running_transaction->t_tid == tid) { 732d76a3a77STheodore Ts'o if (journal->j_commit_request != tid) { 733d76a3a77STheodore Ts'o /* transaction not yet started, so request it */ 734d76a3a77STheodore Ts'o read_unlock(&journal->j_state_lock); 735d76a3a77STheodore Ts'o jbd2_log_start_commit(journal, tid); 736d76a3a77STheodore Ts'o goto wait_commit; 737d76a3a77STheodore Ts'o } 738d76a3a77STheodore Ts'o } else if (!(journal->j_committing_transaction && 739d76a3a77STheodore Ts'o journal->j_committing_transaction->t_tid == tid)) 740d76a3a77STheodore Ts'o need_to_wait = 0; 741d76a3a77STheodore Ts'o read_unlock(&journal->j_state_lock); 742d76a3a77STheodore Ts'o if (!need_to_wait) 743d76a3a77STheodore Ts'o return 0; 744d76a3a77STheodore Ts'o wait_commit: 745d76a3a77STheodore Ts'o return jbd2_log_wait_commit(journal, tid); 746d76a3a77STheodore Ts'o } 747d76a3a77STheodore Ts'o EXPORT_SYMBOL(jbd2_complete_transaction); 748d76a3a77STheodore Ts'o 749d76a3a77STheodore Ts'o /* 750470decc6SDave Kleikamp * Log buffer allocation routines: 751470decc6SDave Kleikamp */ 752470decc6SDave Kleikamp 75318eba7aaSMingming Cao int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp) 754470decc6SDave Kleikamp { 755470decc6SDave Kleikamp unsigned long blocknr; 756470decc6SDave Kleikamp 757a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 758470decc6SDave Kleikamp J_ASSERT(journal->j_free > 1); 759470decc6SDave Kleikamp 760470decc6SDave Kleikamp blocknr = journal->j_head; 761470decc6SDave Kleikamp journal->j_head++; 762470decc6SDave Kleikamp journal->j_free--; 763470decc6SDave Kleikamp if (journal->j_head == journal->j_last) 764470decc6SDave Kleikamp journal->j_head = journal->j_first; 765a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 766f7f4bccbSMingming Cao return jbd2_journal_bmap(journal, blocknr, retp); 767470decc6SDave Kleikamp } 768470decc6SDave Kleikamp 769470decc6SDave Kleikamp /* 770470decc6SDave Kleikamp * Conversion of logical to physical block numbers for the journal 771470decc6SDave Kleikamp * 772470decc6SDave Kleikamp * On external journals the journal blocks are identity-mapped, so 773470decc6SDave Kleikamp * this is a no-op. If needed, we can use j_blk_offset - everything is 774470decc6SDave Kleikamp * ready. 775470decc6SDave Kleikamp */ 776f7f4bccbSMingming Cao int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr, 77718eba7aaSMingming Cao unsigned long long *retp) 778470decc6SDave Kleikamp { 779470decc6SDave Kleikamp int err = 0; 78018eba7aaSMingming Cao unsigned long long ret; 781470decc6SDave Kleikamp 782470decc6SDave Kleikamp if (journal->j_inode) { 783470decc6SDave Kleikamp ret = bmap(journal->j_inode, blocknr); 784470decc6SDave Kleikamp if (ret) 785470decc6SDave Kleikamp *retp = ret; 786470decc6SDave Kleikamp else { 787470decc6SDave Kleikamp printk(KERN_ALERT "%s: journal block not found " 788470decc6SDave Kleikamp "at offset %lu on %s\n", 78905496769STheodore Ts'o __func__, blocknr, journal->j_devname); 790470decc6SDave Kleikamp err = -EIO; 791470decc6SDave Kleikamp __journal_abort_soft(journal, err); 792470decc6SDave Kleikamp } 793470decc6SDave Kleikamp } else { 794470decc6SDave Kleikamp *retp = blocknr; /* +journal->j_blk_offset */ 795470decc6SDave Kleikamp } 796470decc6SDave Kleikamp return err; 797470decc6SDave Kleikamp } 798470decc6SDave Kleikamp 799470decc6SDave Kleikamp /* 800470decc6SDave Kleikamp * We play buffer_head aliasing tricks to write data/metadata blocks to 801470decc6SDave Kleikamp * the journal without copying their contents, but for journal 802470decc6SDave Kleikamp * descriptor blocks we do need to generate bona fide buffers. 803470decc6SDave Kleikamp * 804f7f4bccbSMingming Cao * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying 805470decc6SDave Kleikamp * the buffer's contents they really should run flush_dcache_page(bh->b_page). 806470decc6SDave Kleikamp * But we don't bother doing that, so there will be coherency problems with 807470decc6SDave Kleikamp * mmaps of blockdevs which hold live JBD-controlled filesystems. 808470decc6SDave Kleikamp */ 80932ab6715SJan Kara struct buffer_head * 81032ab6715SJan Kara jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type) 811470decc6SDave Kleikamp { 81232ab6715SJan Kara journal_t *journal = transaction->t_journal; 813470decc6SDave Kleikamp struct buffer_head *bh; 81418eba7aaSMingming Cao unsigned long long blocknr; 81532ab6715SJan Kara journal_header_t *header; 816470decc6SDave Kleikamp int err; 817470decc6SDave Kleikamp 818f7f4bccbSMingming Cao err = jbd2_journal_next_log_block(journal, &blocknr); 819470decc6SDave Kleikamp 820470decc6SDave Kleikamp if (err) 821470decc6SDave Kleikamp return NULL; 822470decc6SDave Kleikamp 823470decc6SDave Kleikamp bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize); 8244b905671SJan Kara if (!bh) 8254b905671SJan Kara return NULL; 826470decc6SDave Kleikamp lock_buffer(bh); 827470decc6SDave Kleikamp memset(bh->b_data, 0, journal->j_blocksize); 82832ab6715SJan Kara header = (journal_header_t *)bh->b_data; 82932ab6715SJan Kara header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER); 83032ab6715SJan Kara header->h_blocktype = cpu_to_be32(type); 83132ab6715SJan Kara header->h_sequence = cpu_to_be32(transaction->t_tid); 832470decc6SDave Kleikamp set_buffer_uptodate(bh); 833470decc6SDave Kleikamp unlock_buffer(bh); 834470decc6SDave Kleikamp BUFFER_TRACE(bh, "return this buffer"); 835e5a120aeSJan Kara return bh; 836470decc6SDave Kleikamp } 837470decc6SDave Kleikamp 8381101cd4dSJan Kara void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh) 8391101cd4dSJan Kara { 8401101cd4dSJan Kara struct jbd2_journal_block_tail *tail; 8411101cd4dSJan Kara __u32 csum; 8421101cd4dSJan Kara 8431101cd4dSJan Kara if (!jbd2_journal_has_csum_v2or3(j)) 8441101cd4dSJan Kara return; 8451101cd4dSJan Kara 8461101cd4dSJan Kara tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize - 8471101cd4dSJan Kara sizeof(struct jbd2_journal_block_tail)); 8481101cd4dSJan Kara tail->t_checksum = 0; 8491101cd4dSJan Kara csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize); 8501101cd4dSJan Kara tail->t_checksum = cpu_to_be32(csum); 8511101cd4dSJan Kara } 8521101cd4dSJan Kara 85379feb521SJan Kara /* 85479feb521SJan Kara * Return tid of the oldest transaction in the journal and block in the journal 85579feb521SJan Kara * where the transaction starts. 85679feb521SJan Kara * 85779feb521SJan Kara * If the journal is now empty, return which will be the next transaction ID 85879feb521SJan Kara * we will write and where will that transaction start. 85979feb521SJan Kara * 86079feb521SJan Kara * The return value is 0 if journal tail cannot be pushed any further, 1 if 86179feb521SJan Kara * it can. 86279feb521SJan Kara */ 86379feb521SJan Kara int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid, 86479feb521SJan Kara unsigned long *block) 86579feb521SJan Kara { 86679feb521SJan Kara transaction_t *transaction; 86779feb521SJan Kara int ret; 86879feb521SJan Kara 86979feb521SJan Kara read_lock(&journal->j_state_lock); 87079feb521SJan Kara spin_lock(&journal->j_list_lock); 87179feb521SJan Kara transaction = journal->j_checkpoint_transactions; 87279feb521SJan Kara if (transaction) { 87379feb521SJan Kara *tid = transaction->t_tid; 87479feb521SJan Kara *block = transaction->t_log_start; 87579feb521SJan Kara } else if ((transaction = journal->j_committing_transaction) != NULL) { 87679feb521SJan Kara *tid = transaction->t_tid; 87779feb521SJan Kara *block = transaction->t_log_start; 87879feb521SJan Kara } else if ((transaction = journal->j_running_transaction) != NULL) { 87979feb521SJan Kara *tid = transaction->t_tid; 88079feb521SJan Kara *block = journal->j_head; 88179feb521SJan Kara } else { 88279feb521SJan Kara *tid = journal->j_transaction_sequence; 88379feb521SJan Kara *block = journal->j_head; 88479feb521SJan Kara } 88579feb521SJan Kara ret = tid_gt(*tid, journal->j_tail_sequence); 88679feb521SJan Kara spin_unlock(&journal->j_list_lock); 88779feb521SJan Kara read_unlock(&journal->j_state_lock); 88879feb521SJan Kara 88979feb521SJan Kara return ret; 89079feb521SJan Kara } 89179feb521SJan Kara 89279feb521SJan Kara /* 89379feb521SJan Kara * Update information in journal structure and in on disk journal superblock 89479feb521SJan Kara * about log tail. This function does not check whether information passed in 89579feb521SJan Kara * really pushes log tail further. It's responsibility of the caller to make 89679feb521SJan Kara * sure provided log tail information is valid (e.g. by holding 89779feb521SJan Kara * j_checkpoint_mutex all the time between computing log tail and calling this 89879feb521SJan Kara * function as is the case with jbd2_cleanup_journal_tail()). 89979feb521SJan Kara * 90079feb521SJan Kara * Requires j_checkpoint_mutex 90179feb521SJan Kara */ 9026f6a6fdaSJoseph Qi int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block) 90379feb521SJan Kara { 90479feb521SJan Kara unsigned long freed; 9056f6a6fdaSJoseph Qi int ret; 90679feb521SJan Kara 90779feb521SJan Kara BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); 90879feb521SJan Kara 90979feb521SJan Kara /* 91079feb521SJan Kara * We cannot afford for write to remain in drive's caches since as 91179feb521SJan Kara * soon as we update j_tail, next transaction can start reusing journal 91279feb521SJan Kara * space and if we lose sb update during power failure we'd replay 91379feb521SJan Kara * old transaction with possibly newly overwritten data. 91479feb521SJan Kara */ 9156f6a6fdaSJoseph Qi ret = jbd2_journal_update_sb_log_tail(journal, tid, block, WRITE_FUA); 9166f6a6fdaSJoseph Qi if (ret) 9176f6a6fdaSJoseph Qi goto out; 9186f6a6fdaSJoseph Qi 91979feb521SJan Kara write_lock(&journal->j_state_lock); 92079feb521SJan Kara freed = block - journal->j_tail; 92179feb521SJan Kara if (block < journal->j_tail) 92279feb521SJan Kara freed += journal->j_last - journal->j_first; 92379feb521SJan Kara 92479feb521SJan Kara trace_jbd2_update_log_tail(journal, tid, block, freed); 92579feb521SJan Kara jbd_debug(1, 92679feb521SJan Kara "Cleaning journal tail from %d to %d (offset %lu), " 92779feb521SJan Kara "freeing %lu\n", 92879feb521SJan Kara journal->j_tail_sequence, tid, block, freed); 92979feb521SJan Kara 93079feb521SJan Kara journal->j_free += freed; 93179feb521SJan Kara journal->j_tail_sequence = tid; 93279feb521SJan Kara journal->j_tail = block; 93379feb521SJan Kara write_unlock(&journal->j_state_lock); 9346f6a6fdaSJoseph Qi 9356f6a6fdaSJoseph Qi out: 9366f6a6fdaSJoseph Qi return ret; 93779feb521SJan Kara } 93879feb521SJan Kara 9393339578fSJan Kara /* 9403339578fSJan Kara * This is a variaon of __jbd2_update_log_tail which checks for validity of 9413339578fSJan Kara * provided log tail and locks j_checkpoint_mutex. So it is safe against races 9423339578fSJan Kara * with other threads updating log tail. 9433339578fSJan Kara */ 9443339578fSJan Kara void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block) 9453339578fSJan Kara { 9463339578fSJan Kara mutex_lock(&journal->j_checkpoint_mutex); 9473339578fSJan Kara if (tid_gt(tid, journal->j_tail_sequence)) 9483339578fSJan Kara __jbd2_update_log_tail(journal, tid, block); 9493339578fSJan Kara mutex_unlock(&journal->j_checkpoint_mutex); 9503339578fSJan Kara } 9513339578fSJan Kara 9528e85fb3fSJohann Lombardi struct jbd2_stats_proc_session { 9538e85fb3fSJohann Lombardi journal_t *journal; 9548e85fb3fSJohann Lombardi struct transaction_stats_s *stats; 9558e85fb3fSJohann Lombardi int start; 9568e85fb3fSJohann Lombardi int max; 9578e85fb3fSJohann Lombardi }; 9588e85fb3fSJohann Lombardi 9598e85fb3fSJohann Lombardi static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos) 9608e85fb3fSJohann Lombardi { 9618e85fb3fSJohann Lombardi return *pos ? NULL : SEQ_START_TOKEN; 9628e85fb3fSJohann Lombardi } 9638e85fb3fSJohann Lombardi 9648e85fb3fSJohann Lombardi static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos) 9658e85fb3fSJohann Lombardi { 9668e85fb3fSJohann Lombardi return NULL; 9678e85fb3fSJohann Lombardi } 9688e85fb3fSJohann Lombardi 9698e85fb3fSJohann Lombardi static int jbd2_seq_info_show(struct seq_file *seq, void *v) 9708e85fb3fSJohann Lombardi { 9718e85fb3fSJohann Lombardi struct jbd2_stats_proc_session *s = seq->private; 9728e85fb3fSJohann Lombardi 9738e85fb3fSJohann Lombardi if (v != SEQ_START_TOKEN) 9748e85fb3fSJohann Lombardi return 0; 9759fff24aaSTheodore Ts'o seq_printf(seq, "%lu transactions (%lu requested), " 9769fff24aaSTheodore Ts'o "each up to %u blocks\n", 9779fff24aaSTheodore Ts'o s->stats->ts_tid, s->stats->ts_requested, 9788e85fb3fSJohann Lombardi s->journal->j_max_transaction_buffers); 9798e85fb3fSJohann Lombardi if (s->stats->ts_tid == 0) 9808e85fb3fSJohann Lombardi return 0; 9818e85fb3fSJohann Lombardi seq_printf(seq, "average: \n %ums waiting for transaction\n", 982bf699327STheodore Ts'o jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid)); 9839fff24aaSTheodore Ts'o seq_printf(seq, " %ums request delay\n", 9849fff24aaSTheodore Ts'o (s->stats->ts_requested == 0) ? 0 : 9859fff24aaSTheodore Ts'o jiffies_to_msecs(s->stats->run.rs_request_delay / 9869fff24aaSTheodore Ts'o s->stats->ts_requested)); 9878e85fb3fSJohann Lombardi seq_printf(seq, " %ums running transaction\n", 988bf699327STheodore Ts'o jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid)); 9898e85fb3fSJohann Lombardi seq_printf(seq, " %ums transaction was being locked\n", 990bf699327STheodore Ts'o jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid)); 9918e85fb3fSJohann Lombardi seq_printf(seq, " %ums flushing data (in ordered mode)\n", 992bf699327STheodore Ts'o jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid)); 9938e85fb3fSJohann Lombardi seq_printf(seq, " %ums logging transaction\n", 994bf699327STheodore Ts'o jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid)); 995c225aa57SSimon Holm Thøgersen seq_printf(seq, " %lluus average transaction commit time\n", 996c225aa57SSimon Holm Thøgersen div_u64(s->journal->j_average_commit_time, 1000)); 9978e85fb3fSJohann Lombardi seq_printf(seq, " %lu handles per transaction\n", 998bf699327STheodore Ts'o s->stats->run.rs_handle_count / s->stats->ts_tid); 9998e85fb3fSJohann Lombardi seq_printf(seq, " %lu blocks per transaction\n", 1000bf699327STheodore Ts'o s->stats->run.rs_blocks / s->stats->ts_tid); 10018e85fb3fSJohann Lombardi seq_printf(seq, " %lu logged blocks per transaction\n", 1002bf699327STheodore Ts'o s->stats->run.rs_blocks_logged / s->stats->ts_tid); 10038e85fb3fSJohann Lombardi return 0; 10048e85fb3fSJohann Lombardi } 10058e85fb3fSJohann Lombardi 10068e85fb3fSJohann Lombardi static void jbd2_seq_info_stop(struct seq_file *seq, void *v) 10078e85fb3fSJohann Lombardi { 10088e85fb3fSJohann Lombardi } 10098e85fb3fSJohann Lombardi 101088e9d34cSJames Morris static const struct seq_operations jbd2_seq_info_ops = { 10118e85fb3fSJohann Lombardi .start = jbd2_seq_info_start, 10128e85fb3fSJohann Lombardi .next = jbd2_seq_info_next, 10138e85fb3fSJohann Lombardi .stop = jbd2_seq_info_stop, 10148e85fb3fSJohann Lombardi .show = jbd2_seq_info_show, 10158e85fb3fSJohann Lombardi }; 10168e85fb3fSJohann Lombardi 10178e85fb3fSJohann Lombardi static int jbd2_seq_info_open(struct inode *inode, struct file *file) 10188e85fb3fSJohann Lombardi { 1019d9dda78bSAl Viro journal_t *journal = PDE_DATA(inode); 10208e85fb3fSJohann Lombardi struct jbd2_stats_proc_session *s; 10218e85fb3fSJohann Lombardi int rc, size; 10228e85fb3fSJohann Lombardi 10238e85fb3fSJohann Lombardi s = kmalloc(sizeof(*s), GFP_KERNEL); 10248e85fb3fSJohann Lombardi if (s == NULL) 10258e85fb3fSJohann Lombardi return -ENOMEM; 10268e85fb3fSJohann Lombardi size = sizeof(struct transaction_stats_s); 10278e85fb3fSJohann Lombardi s->stats = kmalloc(size, GFP_KERNEL); 10288e85fb3fSJohann Lombardi if (s->stats == NULL) { 10298e85fb3fSJohann Lombardi kfree(s); 10308e85fb3fSJohann Lombardi return -ENOMEM; 10318e85fb3fSJohann Lombardi } 10328e85fb3fSJohann Lombardi spin_lock(&journal->j_history_lock); 10338e85fb3fSJohann Lombardi memcpy(s->stats, &journal->j_stats, size); 10348e85fb3fSJohann Lombardi s->journal = journal; 10358e85fb3fSJohann Lombardi spin_unlock(&journal->j_history_lock); 10368e85fb3fSJohann Lombardi 10378e85fb3fSJohann Lombardi rc = seq_open(file, &jbd2_seq_info_ops); 10388e85fb3fSJohann Lombardi if (rc == 0) { 10398e85fb3fSJohann Lombardi struct seq_file *m = file->private_data; 10408e85fb3fSJohann Lombardi m->private = s; 10418e85fb3fSJohann Lombardi } else { 10428e85fb3fSJohann Lombardi kfree(s->stats); 10438e85fb3fSJohann Lombardi kfree(s); 10448e85fb3fSJohann Lombardi } 10458e85fb3fSJohann Lombardi return rc; 10468e85fb3fSJohann Lombardi 10478e85fb3fSJohann Lombardi } 10488e85fb3fSJohann Lombardi 10498e85fb3fSJohann Lombardi static int jbd2_seq_info_release(struct inode *inode, struct file *file) 10508e85fb3fSJohann Lombardi { 10518e85fb3fSJohann Lombardi struct seq_file *seq = file->private_data; 10528e85fb3fSJohann Lombardi struct jbd2_stats_proc_session *s = seq->private; 10538e85fb3fSJohann Lombardi kfree(s->stats); 10548e85fb3fSJohann Lombardi kfree(s); 10558e85fb3fSJohann Lombardi return seq_release(inode, file); 10568e85fb3fSJohann Lombardi } 10578e85fb3fSJohann Lombardi 1058828c0950SAlexey Dobriyan static const struct file_operations jbd2_seq_info_fops = { 10598e85fb3fSJohann Lombardi .owner = THIS_MODULE, 10608e85fb3fSJohann Lombardi .open = jbd2_seq_info_open, 10618e85fb3fSJohann Lombardi .read = seq_read, 10628e85fb3fSJohann Lombardi .llseek = seq_lseek, 10638e85fb3fSJohann Lombardi .release = jbd2_seq_info_release, 10648e85fb3fSJohann Lombardi }; 10658e85fb3fSJohann Lombardi 10668e85fb3fSJohann Lombardi static struct proc_dir_entry *proc_jbd2_stats; 10678e85fb3fSJohann Lombardi 10688e85fb3fSJohann Lombardi static void jbd2_stats_proc_init(journal_t *journal) 10698e85fb3fSJohann Lombardi { 107005496769STheodore Ts'o journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats); 10718e85fb3fSJohann Lombardi if (journal->j_proc_entry) { 107279da3664SDenis V. Lunev proc_create_data("info", S_IRUGO, journal->j_proc_entry, 107379da3664SDenis V. Lunev &jbd2_seq_info_fops, journal); 10748e85fb3fSJohann Lombardi } 10758e85fb3fSJohann Lombardi } 10768e85fb3fSJohann Lombardi 10778e85fb3fSJohann Lombardi static void jbd2_stats_proc_exit(journal_t *journal) 10788e85fb3fSJohann Lombardi { 10798e85fb3fSJohann Lombardi remove_proc_entry("info", journal->j_proc_entry); 108005496769STheodore Ts'o remove_proc_entry(journal->j_devname, proc_jbd2_stats); 10818e85fb3fSJohann Lombardi } 10828e85fb3fSJohann Lombardi 1083470decc6SDave Kleikamp /* 1084470decc6SDave Kleikamp * Management for journal control blocks: functions to create and 1085470decc6SDave Kleikamp * destroy journal_t structures, and to initialise and read existing 1086470decc6SDave Kleikamp * journal blocks from disk. */ 1087470decc6SDave Kleikamp 1088470decc6SDave Kleikamp /* First: create and setup a journal_t object in memory. We initialise 1089470decc6SDave Kleikamp * very few fields yet: that has to wait until we have created the 1090470decc6SDave Kleikamp * journal structures from from scratch, or loaded them from disk. */ 1091470decc6SDave Kleikamp 1092470decc6SDave Kleikamp static journal_t * journal_init_common (void) 1093470decc6SDave Kleikamp { 1094*ab714affSJan Kara static struct lock_class_key jbd2_trans_commit_key; 1095470decc6SDave Kleikamp journal_t *journal; 1096470decc6SDave Kleikamp int err; 1097470decc6SDave Kleikamp 10983ebfdf88SAndrew Morton journal = kzalloc(sizeof(*journal), GFP_KERNEL); 1099470decc6SDave Kleikamp if (!journal) 1100b7271b0aSTheodore Ts'o return NULL; 1101470decc6SDave Kleikamp 1102470decc6SDave Kleikamp init_waitqueue_head(&journal->j_wait_transaction_locked); 1103470decc6SDave Kleikamp init_waitqueue_head(&journal->j_wait_done_commit); 1104470decc6SDave Kleikamp init_waitqueue_head(&journal->j_wait_commit); 1105470decc6SDave Kleikamp init_waitqueue_head(&journal->j_wait_updates); 11068f7d89f3SJan Kara init_waitqueue_head(&journal->j_wait_reserved); 1107470decc6SDave Kleikamp mutex_init(&journal->j_barrier); 1108470decc6SDave Kleikamp mutex_init(&journal->j_checkpoint_mutex); 1109470decc6SDave Kleikamp spin_lock_init(&journal->j_revoke_lock); 1110470decc6SDave Kleikamp spin_lock_init(&journal->j_list_lock); 1111a931da6aSTheodore Ts'o rwlock_init(&journal->j_state_lock); 1112470decc6SDave Kleikamp 1113cd02ff0bSMingming Cao journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE); 111430773840STheodore Ts'o journal->j_min_batch_time = 0; 111530773840STheodore Ts'o journal->j_max_batch_time = 15000; /* 15ms */ 11168f7d89f3SJan Kara atomic_set(&journal->j_reserved_credits, 0); 1117470decc6SDave Kleikamp 1118470decc6SDave Kleikamp /* The journal is marked for error until we succeed with recovery! */ 1119f7f4bccbSMingming Cao journal->j_flags = JBD2_ABORT; 1120470decc6SDave Kleikamp 1121470decc6SDave Kleikamp /* Set up a default-sized revoke table for the new mount. */ 1122f7f4bccbSMingming Cao err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH); 1123470decc6SDave Kleikamp if (err) { 1124470decc6SDave Kleikamp kfree(journal); 1125b7271b0aSTheodore Ts'o return NULL; 1126470decc6SDave Kleikamp } 11278e85fb3fSJohann Lombardi 1128bf699327STheodore Ts'o spin_lock_init(&journal->j_history_lock); 11298e85fb3fSJohann Lombardi 1130*ab714affSJan Kara lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle", 1131*ab714affSJan Kara &jbd2_trans_commit_key, 0); 1132*ab714affSJan Kara 1133470decc6SDave Kleikamp return journal; 1134470decc6SDave Kleikamp } 1135470decc6SDave Kleikamp 1136f7f4bccbSMingming Cao /* jbd2_journal_init_dev and jbd2_journal_init_inode: 1137470decc6SDave Kleikamp * 1138470decc6SDave Kleikamp * Create a journal structure assigned some fixed set of disk blocks to 1139470decc6SDave Kleikamp * the journal. We don't actually touch those disk blocks yet, but we 1140470decc6SDave Kleikamp * need to set up all of the mapping information to tell the journaling 1141470decc6SDave Kleikamp * system where the journal blocks are. 1142470decc6SDave Kleikamp * 1143470decc6SDave Kleikamp */ 1144470decc6SDave Kleikamp 1145470decc6SDave Kleikamp /** 11465648ba5bSRandy Dunlap * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure 1147470decc6SDave Kleikamp * @bdev: Block device on which to create the journal 1148470decc6SDave Kleikamp * @fs_dev: Device which hold journalled filesystem for this journal. 1149470decc6SDave Kleikamp * @start: Block nr Start of journal. 1150470decc6SDave Kleikamp * @len: Length of the journal in blocks. 1151470decc6SDave Kleikamp * @blocksize: blocksize of journalling device 11525648ba5bSRandy Dunlap * 11535648ba5bSRandy Dunlap * Returns: a newly created journal_t * 1154470decc6SDave Kleikamp * 1155f7f4bccbSMingming Cao * jbd2_journal_init_dev creates a journal which maps a fixed contiguous 1156470decc6SDave Kleikamp * range of blocks on an arbitrary block device. 1157470decc6SDave Kleikamp * 1158470decc6SDave Kleikamp */ 1159f7f4bccbSMingming Cao journal_t * jbd2_journal_init_dev(struct block_device *bdev, 1160470decc6SDave Kleikamp struct block_device *fs_dev, 116118eba7aaSMingming Cao unsigned long long start, int len, int blocksize) 1162470decc6SDave Kleikamp { 1163470decc6SDave Kleikamp journal_t *journal = journal_init_common(); 1164470decc6SDave Kleikamp struct buffer_head *bh; 1165470decc6SDave Kleikamp int n; 1166470decc6SDave Kleikamp 1167470decc6SDave Kleikamp if (!journal) 1168470decc6SDave Kleikamp return NULL; 1169470decc6SDave Kleikamp 1170470decc6SDave Kleikamp /* journal descriptor can store up to n blocks -bzzz */ 1171470decc6SDave Kleikamp journal->j_blocksize = blocksize; 11720587aa3dSyangsheng journal->j_dev = bdev; 11730587aa3dSyangsheng journal->j_fs_dev = fs_dev; 11740587aa3dSyangsheng journal->j_blk_offset = start; 11750587aa3dSyangsheng journal->j_maxlen = len; 11760587aa3dSyangsheng bdevname(journal->j_dev, journal->j_devname); 117781ae394bSRasmus Villemoes strreplace(journal->j_devname, '/', '!'); 11784b905671SJan Kara jbd2_stats_proc_init(journal); 1179470decc6SDave Kleikamp n = journal->j_blocksize / sizeof(journal_block_tag_t); 1180470decc6SDave Kleikamp journal->j_wbufsize = n; 1181470decc6SDave Kleikamp journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL); 1182470decc6SDave Kleikamp if (!journal->j_wbuf) { 118325985edcSLucas De Marchi printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n", 1184329d291fSHarvey Harrison __func__); 11854b905671SJan Kara goto out_err; 1186470decc6SDave Kleikamp } 1187470decc6SDave Kleikamp 1188470decc6SDave Kleikamp bh = __getblk(journal->j_dev, start, journal->j_blocksize); 11894b905671SJan Kara if (!bh) { 11904b905671SJan Kara printk(KERN_ERR 11914b905671SJan Kara "%s: Cannot get buffer for journal superblock\n", 11924b905671SJan Kara __func__); 11934b905671SJan Kara goto out_err; 11944b905671SJan Kara } 1195470decc6SDave Kleikamp journal->j_sb_buffer = bh; 1196470decc6SDave Kleikamp journal->j_superblock = (journal_superblock_t *)bh->b_data; 11974b905671SJan Kara 1198470decc6SDave Kleikamp return journal; 11994b905671SJan Kara out_err: 12007b02bec0STao Ma kfree(journal->j_wbuf); 12014b905671SJan Kara jbd2_stats_proc_exit(journal); 12024b905671SJan Kara kfree(journal); 12034b905671SJan Kara return NULL; 1204470decc6SDave Kleikamp } 1205470decc6SDave Kleikamp 1206470decc6SDave Kleikamp /** 1207f7f4bccbSMingming Cao * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode. 1208470decc6SDave Kleikamp * @inode: An inode to create the journal in 1209470decc6SDave Kleikamp * 1210f7f4bccbSMingming Cao * jbd2_journal_init_inode creates a journal which maps an on-disk inode as 1211470decc6SDave Kleikamp * the journal. The inode must exist already, must support bmap() and 1212470decc6SDave Kleikamp * must have all data blocks preallocated. 1213470decc6SDave Kleikamp */ 1214f7f4bccbSMingming Cao journal_t * jbd2_journal_init_inode (struct inode *inode) 1215470decc6SDave Kleikamp { 1216470decc6SDave Kleikamp struct buffer_head *bh; 1217470decc6SDave Kleikamp journal_t *journal = journal_init_common(); 121805496769STheodore Ts'o char *p; 1219470decc6SDave Kleikamp int err; 1220470decc6SDave Kleikamp int n; 122118eba7aaSMingming Cao unsigned long long blocknr; 1222470decc6SDave Kleikamp 1223470decc6SDave Kleikamp if (!journal) 1224470decc6SDave Kleikamp return NULL; 1225470decc6SDave Kleikamp 1226470decc6SDave Kleikamp journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev; 1227470decc6SDave Kleikamp journal->j_inode = inode; 122805496769STheodore Ts'o bdevname(journal->j_dev, journal->j_devname); 122981ae394bSRasmus Villemoes p = strreplace(journal->j_devname, '/', '!'); 123090576c0bSTheodore Ts'o sprintf(p, "-%lu", journal->j_inode->i_ino); 1231470decc6SDave Kleikamp jbd_debug(1, 1232470decc6SDave Kleikamp "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n", 1233470decc6SDave Kleikamp journal, inode->i_sb->s_id, inode->i_ino, 1234470decc6SDave Kleikamp (long long) inode->i_size, 1235470decc6SDave Kleikamp inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize); 1236470decc6SDave Kleikamp 1237470decc6SDave Kleikamp journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits; 1238470decc6SDave Kleikamp journal->j_blocksize = inode->i_sb->s_blocksize; 12398e85fb3fSJohann Lombardi jbd2_stats_proc_init(journal); 1240470decc6SDave Kleikamp 1241470decc6SDave Kleikamp /* journal descriptor can store up to n blocks -bzzz */ 1242470decc6SDave Kleikamp n = journal->j_blocksize / sizeof(journal_block_tag_t); 1243470decc6SDave Kleikamp journal->j_wbufsize = n; 1244470decc6SDave Kleikamp journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL); 1245470decc6SDave Kleikamp if (!journal->j_wbuf) { 124625985edcSLucas De Marchi printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n", 1247329d291fSHarvey Harrison __func__); 12484b905671SJan Kara goto out_err; 1249470decc6SDave Kleikamp } 1250470decc6SDave Kleikamp 1251f7f4bccbSMingming Cao err = jbd2_journal_bmap(journal, 0, &blocknr); 1252470decc6SDave Kleikamp /* If that failed, give up */ 1253470decc6SDave Kleikamp if (err) { 12543c26bdb4SJustin P. Mattock printk(KERN_ERR "%s: Cannot locate journal superblock\n", 1255329d291fSHarvey Harrison __func__); 12564b905671SJan Kara goto out_err; 1257470decc6SDave Kleikamp } 1258470decc6SDave Kleikamp 1259a49058faSGioh Kim bh = getblk_unmovable(journal->j_dev, blocknr, journal->j_blocksize); 12604b905671SJan Kara if (!bh) { 12614b905671SJan Kara printk(KERN_ERR 12624b905671SJan Kara "%s: Cannot get buffer for journal superblock\n", 12634b905671SJan Kara __func__); 12644b905671SJan Kara goto out_err; 12654b905671SJan Kara } 1266470decc6SDave Kleikamp journal->j_sb_buffer = bh; 1267470decc6SDave Kleikamp journal->j_superblock = (journal_superblock_t *)bh->b_data; 1268470decc6SDave Kleikamp 1269470decc6SDave Kleikamp return journal; 12704b905671SJan Kara out_err: 12717b02bec0STao Ma kfree(journal->j_wbuf); 12724b905671SJan Kara jbd2_stats_proc_exit(journal); 12734b905671SJan Kara kfree(journal); 12744b905671SJan Kara return NULL; 1275470decc6SDave Kleikamp } 1276470decc6SDave Kleikamp 1277470decc6SDave Kleikamp /* 1278470decc6SDave Kleikamp * If the journal init or create aborts, we need to mark the journal 1279470decc6SDave Kleikamp * superblock as being NULL to prevent the journal destroy from writing 1280470decc6SDave Kleikamp * back a bogus superblock. 1281470decc6SDave Kleikamp */ 1282470decc6SDave Kleikamp static void journal_fail_superblock (journal_t *journal) 1283470decc6SDave Kleikamp { 1284470decc6SDave Kleikamp struct buffer_head *bh = journal->j_sb_buffer; 1285470decc6SDave Kleikamp brelse(bh); 1286470decc6SDave Kleikamp journal->j_sb_buffer = NULL; 1287470decc6SDave Kleikamp } 1288470decc6SDave Kleikamp 1289470decc6SDave Kleikamp /* 1290470decc6SDave Kleikamp * Given a journal_t structure, initialise the various fields for 1291470decc6SDave Kleikamp * startup of a new journaling session. We use this both when creating 1292470decc6SDave Kleikamp * a journal, and after recovering an old journal to reset it for 1293470decc6SDave Kleikamp * subsequent use. 1294470decc6SDave Kleikamp */ 1295470decc6SDave Kleikamp 1296470decc6SDave Kleikamp static int journal_reset(journal_t *journal) 1297470decc6SDave Kleikamp { 1298470decc6SDave Kleikamp journal_superblock_t *sb = journal->j_superblock; 129918eba7aaSMingming Cao unsigned long long first, last; 1300470decc6SDave Kleikamp 1301470decc6SDave Kleikamp first = be32_to_cpu(sb->s_first); 1302470decc6SDave Kleikamp last = be32_to_cpu(sb->s_maxlen); 1303f6f50e28SJan Kara if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) { 1304f2a44523SEryu Guan printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n", 1305f6f50e28SJan Kara first, last); 1306f6f50e28SJan Kara journal_fail_superblock(journal); 1307f6f50e28SJan Kara return -EINVAL; 1308f6f50e28SJan Kara } 1309470decc6SDave Kleikamp 1310470decc6SDave Kleikamp journal->j_first = first; 1311470decc6SDave Kleikamp journal->j_last = last; 1312470decc6SDave Kleikamp 1313470decc6SDave Kleikamp journal->j_head = first; 1314470decc6SDave Kleikamp journal->j_tail = first; 1315470decc6SDave Kleikamp journal->j_free = last - first; 1316470decc6SDave Kleikamp 1317470decc6SDave Kleikamp journal->j_tail_sequence = journal->j_transaction_sequence; 1318470decc6SDave Kleikamp journal->j_commit_sequence = journal->j_transaction_sequence - 1; 1319470decc6SDave Kleikamp journal->j_commit_request = journal->j_commit_sequence; 1320470decc6SDave Kleikamp 1321470decc6SDave Kleikamp journal->j_max_transaction_buffers = journal->j_maxlen / 4; 1322470decc6SDave Kleikamp 1323470decc6SDave Kleikamp /* 1324470decc6SDave Kleikamp * As a special case, if the on-disk copy is already marked as needing 132524bcc89cSJan Kara * no recovery (s_start == 0), then we can safely defer the superblock 132624bcc89cSJan Kara * update until the next commit by setting JBD2_FLUSHED. This avoids 1327470decc6SDave Kleikamp * attempting a write to a potential-readonly device. 1328470decc6SDave Kleikamp */ 132924bcc89cSJan Kara if (sb->s_start == 0) { 1330f2a44523SEryu Guan jbd_debug(1, "JBD2: Skipping superblock update on recovered sb " 1331470decc6SDave Kleikamp "(start %ld, seq %d, errno %d)\n", 1332470decc6SDave Kleikamp journal->j_tail, journal->j_tail_sequence, 1333470decc6SDave Kleikamp journal->j_errno); 133424bcc89cSJan Kara journal->j_flags |= JBD2_FLUSHED; 133524bcc89cSJan Kara } else { 1336a78bb11dSJan Kara /* Lock here to make assertions happy... */ 1337a78bb11dSJan Kara mutex_lock(&journal->j_checkpoint_mutex); 133879feb521SJan Kara /* 133979feb521SJan Kara * Update log tail information. We use WRITE_FUA since new 134079feb521SJan Kara * transaction will start reusing journal space and so we 134179feb521SJan Kara * must make sure information about current log tail is on 134279feb521SJan Kara * disk before that. 134379feb521SJan Kara */ 134479feb521SJan Kara jbd2_journal_update_sb_log_tail(journal, 134579feb521SJan Kara journal->j_tail_sequence, 134679feb521SJan Kara journal->j_tail, 134779feb521SJan Kara WRITE_FUA); 1348a78bb11dSJan Kara mutex_unlock(&journal->j_checkpoint_mutex); 134924bcc89cSJan Kara } 135024bcc89cSJan Kara return jbd2_journal_start_thread(journal); 1351470decc6SDave Kleikamp } 1352470decc6SDave Kleikamp 13536f6a6fdaSJoseph Qi static int jbd2_write_superblock(journal_t *journal, int write_op) 135424bcc89cSJan Kara { 135524bcc89cSJan Kara struct buffer_head *bh = journal->j_sb_buffer; 1356fe52d17cSTheodore Ts'o journal_superblock_t *sb = journal->j_superblock; 135779feb521SJan Kara int ret; 135824bcc89cSJan Kara 135979feb521SJan Kara trace_jbd2_write_superblock(journal, write_op); 136079feb521SJan Kara if (!(journal->j_flags & JBD2_BARRIER)) 136179feb521SJan Kara write_op &= ~(REQ_FUA | REQ_FLUSH); 136279feb521SJan Kara lock_buffer(bh); 1363914258bfSTheodore Ts'o if (buffer_write_io_error(bh)) { 1364914258bfSTheodore Ts'o /* 1365914258bfSTheodore Ts'o * Oh, dear. A previous attempt to write the journal 1366914258bfSTheodore Ts'o * superblock failed. This could happen because the 1367914258bfSTheodore Ts'o * USB device was yanked out. Or it could happen to 1368914258bfSTheodore Ts'o * be a transient write error and maybe the block will 1369914258bfSTheodore Ts'o * be remapped. Nothing we can do but to retry the 1370914258bfSTheodore Ts'o * write and hope for the best. 1371914258bfSTheodore Ts'o */ 1372914258bfSTheodore Ts'o printk(KERN_ERR "JBD2: previous I/O error detected " 1373914258bfSTheodore Ts'o "for journal superblock update for %s.\n", 1374914258bfSTheodore Ts'o journal->j_devname); 1375914258bfSTheodore Ts'o clear_buffer_write_io_error(bh); 1376914258bfSTheodore Ts'o set_buffer_uptodate(bh); 1377914258bfSTheodore Ts'o } 1378fe52d17cSTheodore Ts'o jbd2_superblock_csum_set(journal, sb); 137979feb521SJan Kara get_bh(bh); 138079feb521SJan Kara bh->b_end_io = end_buffer_write_sync; 138179feb521SJan Kara ret = submit_bh(write_op, bh); 138279feb521SJan Kara wait_on_buffer(bh); 1383914258bfSTheodore Ts'o if (buffer_write_io_error(bh)) { 1384914258bfSTheodore Ts'o clear_buffer_write_io_error(bh); 1385914258bfSTheodore Ts'o set_buffer_uptodate(bh); 138679feb521SJan Kara ret = -EIO; 138779feb521SJan Kara } 138879feb521SJan Kara if (ret) { 138979feb521SJan Kara printk(KERN_ERR "JBD2: Error %d detected when updating " 139079feb521SJan Kara "journal superblock for %s.\n", ret, 139179feb521SJan Kara journal->j_devname); 13926f6a6fdaSJoseph Qi jbd2_journal_abort(journal, ret); 1393914258bfSTheodore Ts'o } 13946f6a6fdaSJoseph Qi 13956f6a6fdaSJoseph Qi return ret; 139624bcc89cSJan Kara } 1397470decc6SDave Kleikamp 139824bcc89cSJan Kara /** 139924bcc89cSJan Kara * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk. 140024bcc89cSJan Kara * @journal: The journal to update. 140179feb521SJan Kara * @tail_tid: TID of the new transaction at the tail of the log 140279feb521SJan Kara * @tail_block: The first block of the transaction at the tail of the log 140379feb521SJan Kara * @write_op: With which operation should we write the journal sb 140424bcc89cSJan Kara * 140524bcc89cSJan Kara * Update a journal's superblock information about log tail and write it to 140624bcc89cSJan Kara * disk, waiting for the IO to complete. 140724bcc89cSJan Kara */ 14086f6a6fdaSJoseph Qi int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid, 140979feb521SJan Kara unsigned long tail_block, int write_op) 141024bcc89cSJan Kara { 141124bcc89cSJan Kara journal_superblock_t *sb = journal->j_superblock; 14126f6a6fdaSJoseph Qi int ret; 14132201c590SSeiji Aguchi 1414a78bb11dSJan Kara BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); 141579feb521SJan Kara jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n", 141679feb521SJan Kara tail_block, tail_tid); 1417470decc6SDave Kleikamp 141879feb521SJan Kara sb->s_sequence = cpu_to_be32(tail_tid); 141979feb521SJan Kara sb->s_start = cpu_to_be32(tail_block); 142024bcc89cSJan Kara 14216f6a6fdaSJoseph Qi ret = jbd2_write_superblock(journal, write_op); 14226f6a6fdaSJoseph Qi if (ret) 14236f6a6fdaSJoseph Qi goto out; 142424bcc89cSJan Kara 142524bcc89cSJan Kara /* Log is no longer empty */ 1426a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 142724bcc89cSJan Kara WARN_ON(!sb->s_sequence); 1428f7f4bccbSMingming Cao journal->j_flags &= ~JBD2_FLUSHED; 142924bcc89cSJan Kara write_unlock(&journal->j_state_lock); 14306f6a6fdaSJoseph Qi 14316f6a6fdaSJoseph Qi out: 14326f6a6fdaSJoseph Qi return ret; 143324bcc89cSJan Kara } 143424bcc89cSJan Kara 143524bcc89cSJan Kara /** 143624bcc89cSJan Kara * jbd2_mark_journal_empty() - Mark on disk journal as empty. 143724bcc89cSJan Kara * @journal: The journal to update. 1438c0a2ad9bSOGAWA Hirofumi * @write_op: With which operation should we write the journal sb 143924bcc89cSJan Kara * 144024bcc89cSJan Kara * Update a journal's dynamic superblock fields to show that journal is empty. 144124bcc89cSJan Kara * Write updated superblock to disk waiting for IO to complete. 144224bcc89cSJan Kara */ 1443c0a2ad9bSOGAWA Hirofumi static void jbd2_mark_journal_empty(journal_t *journal, int write_op) 144424bcc89cSJan Kara { 144524bcc89cSJan Kara journal_superblock_t *sb = journal->j_superblock; 144624bcc89cSJan Kara 1447a78bb11dSJan Kara BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); 1448470decc6SDave Kleikamp read_lock(&journal->j_state_lock); 1449eeecef0aSEric Sandeen /* Is it already empty? */ 1450eeecef0aSEric Sandeen if (sb->s_start == 0) { 1451eeecef0aSEric Sandeen read_unlock(&journal->j_state_lock); 1452eeecef0aSEric Sandeen return; 1453eeecef0aSEric Sandeen } 145424bcc89cSJan Kara jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n", 145524bcc89cSJan Kara journal->j_tail_sequence); 1456470decc6SDave Kleikamp 1457470decc6SDave Kleikamp sb->s_sequence = cpu_to_be32(journal->j_tail_sequence); 145824bcc89cSJan Kara sb->s_start = cpu_to_be32(0); 145924bcc89cSJan Kara read_unlock(&journal->j_state_lock); 146024bcc89cSJan Kara 1461c0a2ad9bSOGAWA Hirofumi jbd2_write_superblock(journal, write_op); 146224bcc89cSJan Kara 146324bcc89cSJan Kara /* Log is no longer empty */ 146424bcc89cSJan Kara write_lock(&journal->j_state_lock); 1465f7f4bccbSMingming Cao journal->j_flags |= JBD2_FLUSHED; 1466a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 1467470decc6SDave Kleikamp } 1468470decc6SDave Kleikamp 146924bcc89cSJan Kara 147024bcc89cSJan Kara /** 147124bcc89cSJan Kara * jbd2_journal_update_sb_errno() - Update error in the journal. 147224bcc89cSJan Kara * @journal: The journal to update. 147324bcc89cSJan Kara * 147424bcc89cSJan Kara * Update a journal's errno. Write updated superblock to disk waiting for IO 147524bcc89cSJan Kara * to complete. 147624bcc89cSJan Kara */ 1477d796c52eSTheodore Ts'o void jbd2_journal_update_sb_errno(journal_t *journal) 147824bcc89cSJan Kara { 147924bcc89cSJan Kara journal_superblock_t *sb = journal->j_superblock; 148024bcc89cSJan Kara 148124bcc89cSJan Kara read_lock(&journal->j_state_lock); 148224bcc89cSJan Kara jbd_debug(1, "JBD2: updating superblock error (errno %d)\n", 148324bcc89cSJan Kara journal->j_errno); 1484470decc6SDave Kleikamp sb->s_errno = cpu_to_be32(journal->j_errno); 1485470decc6SDave Kleikamp read_unlock(&journal->j_state_lock); 1486470decc6SDave Kleikamp 1487564bc402SDaeho Jeong jbd2_write_superblock(journal, WRITE_FUA); 1488470decc6SDave Kleikamp } 1489d796c52eSTheodore Ts'o EXPORT_SYMBOL(jbd2_journal_update_sb_errno); 1490470decc6SDave Kleikamp 1491470decc6SDave Kleikamp /* 1492470decc6SDave Kleikamp * Read the superblock for a given journal, performing initial 1493470decc6SDave Kleikamp * validation of the format. 1494470decc6SDave Kleikamp */ 1495470decc6SDave Kleikamp static int journal_get_superblock(journal_t *journal) 1496470decc6SDave Kleikamp { 1497470decc6SDave Kleikamp struct buffer_head *bh; 1498470decc6SDave Kleikamp journal_superblock_t *sb; 1499470decc6SDave Kleikamp int err = -EIO; 1500470decc6SDave Kleikamp 1501470decc6SDave Kleikamp bh = journal->j_sb_buffer; 1502470decc6SDave Kleikamp 1503470decc6SDave Kleikamp J_ASSERT(bh != NULL); 1504470decc6SDave Kleikamp if (!buffer_uptodate(bh)) { 1505470decc6SDave Kleikamp ll_rw_block(READ, 1, &bh); 1506470decc6SDave Kleikamp wait_on_buffer(bh); 1507470decc6SDave Kleikamp if (!buffer_uptodate(bh)) { 1508470decc6SDave Kleikamp printk(KERN_ERR 1509f2a44523SEryu Guan "JBD2: IO error reading journal superblock\n"); 1510470decc6SDave Kleikamp goto out; 1511470decc6SDave Kleikamp } 1512470decc6SDave Kleikamp } 1513470decc6SDave Kleikamp 151425ed6e8aSDarrick J. Wong if (buffer_verified(bh)) 151525ed6e8aSDarrick J. Wong return 0; 151625ed6e8aSDarrick J. Wong 1517470decc6SDave Kleikamp sb = journal->j_superblock; 1518470decc6SDave Kleikamp 1519470decc6SDave Kleikamp err = -EINVAL; 1520470decc6SDave Kleikamp 1521f7f4bccbSMingming Cao if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) || 1522470decc6SDave Kleikamp sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) { 1523f2a44523SEryu Guan printk(KERN_WARNING "JBD2: no valid journal superblock found\n"); 1524470decc6SDave Kleikamp goto out; 1525470decc6SDave Kleikamp } 1526470decc6SDave Kleikamp 1527470decc6SDave Kleikamp switch(be32_to_cpu(sb->s_header.h_blocktype)) { 1528f7f4bccbSMingming Cao case JBD2_SUPERBLOCK_V1: 1529470decc6SDave Kleikamp journal->j_format_version = 1; 1530470decc6SDave Kleikamp break; 1531f7f4bccbSMingming Cao case JBD2_SUPERBLOCK_V2: 1532470decc6SDave Kleikamp journal->j_format_version = 2; 1533470decc6SDave Kleikamp break; 1534470decc6SDave Kleikamp default: 1535f2a44523SEryu Guan printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n"); 1536470decc6SDave Kleikamp goto out; 1537470decc6SDave Kleikamp } 1538470decc6SDave Kleikamp 1539470decc6SDave Kleikamp if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen) 1540470decc6SDave Kleikamp journal->j_maxlen = be32_to_cpu(sb->s_maxlen); 1541470decc6SDave Kleikamp else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) { 1542f2a44523SEryu Guan printk(KERN_WARNING "JBD2: journal file too short\n"); 1543470decc6SDave Kleikamp goto out; 1544470decc6SDave Kleikamp } 1545470decc6SDave Kleikamp 15468762202dSEryu Guan if (be32_to_cpu(sb->s_first) == 0 || 15478762202dSEryu Guan be32_to_cpu(sb->s_first) >= journal->j_maxlen) { 15488762202dSEryu Guan printk(KERN_WARNING 15498762202dSEryu Guan "JBD2: Invalid start block of journal: %u\n", 15508762202dSEryu Guan be32_to_cpu(sb->s_first)); 15518762202dSEryu Guan goto out; 15528762202dSEryu Guan } 15538762202dSEryu Guan 155456316a0dSDarrick J. Wong if (jbd2_has_feature_csum2(journal) && 155556316a0dSDarrick J. Wong jbd2_has_feature_csum3(journal)) { 1556db9ee220SDarrick J. Wong /* Can't have checksum v2 and v3 at the same time! */ 1557db9ee220SDarrick J. Wong printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 " 1558db9ee220SDarrick J. Wong "at the same time!\n"); 1559db9ee220SDarrick J. Wong goto out; 1560db9ee220SDarrick J. Wong } 1561db9ee220SDarrick J. Wong 15628595798cSDarrick J. Wong if (jbd2_journal_has_csum_v2or3_feature(journal) && 156356316a0dSDarrick J. Wong jbd2_has_feature_checksum(journal)) { 1564feb8c6d3SDarrick J. Wong /* Can't have checksum v1 and v2 on at the same time! */ 1565feb8c6d3SDarrick J. Wong printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 " 1566feb8c6d3SDarrick J. Wong "at the same time!\n"); 1567feb8c6d3SDarrick J. Wong goto out; 1568feb8c6d3SDarrick J. Wong } 1569feb8c6d3SDarrick J. Wong 157025ed6e8aSDarrick J. Wong if (!jbd2_verify_csum_type(journal, sb)) { 1571a67c848aSDmitry Monakhov printk(KERN_ERR "JBD2: Unknown checksum type\n"); 157225ed6e8aSDarrick J. Wong goto out; 157325ed6e8aSDarrick J. Wong } 157425ed6e8aSDarrick J. Wong 157501b5adceSDarrick J. Wong /* Load the checksum driver */ 15768595798cSDarrick J. Wong if (jbd2_journal_has_csum_v2or3_feature(journal)) { 157701b5adceSDarrick J. Wong journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0); 157801b5adceSDarrick J. Wong if (IS_ERR(journal->j_chksum_driver)) { 1579a67c848aSDmitry Monakhov printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n"); 158001b5adceSDarrick J. Wong err = PTR_ERR(journal->j_chksum_driver); 158101b5adceSDarrick J. Wong journal->j_chksum_driver = NULL; 158201b5adceSDarrick J. Wong goto out; 158301b5adceSDarrick J. Wong } 158401b5adceSDarrick J. Wong } 158501b5adceSDarrick J. Wong 15864fd5ea43SDarrick J. Wong /* Check superblock checksum */ 15874fd5ea43SDarrick J. Wong if (!jbd2_superblock_csum_verify(journal, sb)) { 1588a67c848aSDmitry Monakhov printk(KERN_ERR "JBD2: journal checksum error\n"); 15896a797d27SDarrick J. Wong err = -EFSBADCRC; 15904fd5ea43SDarrick J. Wong goto out; 15914fd5ea43SDarrick J. Wong } 15924fd5ea43SDarrick J. Wong 15934fd5ea43SDarrick J. Wong /* Precompute checksum seed for all metadata */ 1594db9ee220SDarrick J. Wong if (jbd2_journal_has_csum_v2or3(journal)) 15954fd5ea43SDarrick J. Wong journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid, 15964fd5ea43SDarrick J. Wong sizeof(sb->s_uuid)); 15974fd5ea43SDarrick J. Wong 159825ed6e8aSDarrick J. Wong set_buffer_verified(bh); 159925ed6e8aSDarrick J. Wong 1600470decc6SDave Kleikamp return 0; 1601470decc6SDave Kleikamp 1602470decc6SDave Kleikamp out: 1603470decc6SDave Kleikamp journal_fail_superblock(journal); 1604470decc6SDave Kleikamp return err; 1605470decc6SDave Kleikamp } 1606470decc6SDave Kleikamp 1607470decc6SDave Kleikamp /* 1608470decc6SDave Kleikamp * Load the on-disk journal superblock and read the key fields into the 1609470decc6SDave Kleikamp * journal_t. 1610470decc6SDave Kleikamp */ 1611470decc6SDave Kleikamp 1612470decc6SDave Kleikamp static int load_superblock(journal_t *journal) 1613470decc6SDave Kleikamp { 1614470decc6SDave Kleikamp int err; 1615470decc6SDave Kleikamp journal_superblock_t *sb; 1616470decc6SDave Kleikamp 1617470decc6SDave Kleikamp err = journal_get_superblock(journal); 1618470decc6SDave Kleikamp if (err) 1619470decc6SDave Kleikamp return err; 1620470decc6SDave Kleikamp 1621470decc6SDave Kleikamp sb = journal->j_superblock; 1622470decc6SDave Kleikamp 1623470decc6SDave Kleikamp journal->j_tail_sequence = be32_to_cpu(sb->s_sequence); 1624470decc6SDave Kleikamp journal->j_tail = be32_to_cpu(sb->s_start); 1625470decc6SDave Kleikamp journal->j_first = be32_to_cpu(sb->s_first); 1626470decc6SDave Kleikamp journal->j_last = be32_to_cpu(sb->s_maxlen); 1627470decc6SDave Kleikamp journal->j_errno = be32_to_cpu(sb->s_errno); 1628470decc6SDave Kleikamp 1629470decc6SDave Kleikamp return 0; 1630470decc6SDave Kleikamp } 1631470decc6SDave Kleikamp 1632470decc6SDave Kleikamp 1633470decc6SDave Kleikamp /** 1634f7f4bccbSMingming Cao * int jbd2_journal_load() - Read journal from disk. 1635470decc6SDave Kleikamp * @journal: Journal to act on. 1636470decc6SDave Kleikamp * 1637470decc6SDave Kleikamp * Given a journal_t structure which tells us which disk blocks contain 1638470decc6SDave Kleikamp * a journal, read the journal from disk to initialise the in-memory 1639470decc6SDave Kleikamp * structures. 1640470decc6SDave Kleikamp */ 1641f7f4bccbSMingming Cao int jbd2_journal_load(journal_t *journal) 1642470decc6SDave Kleikamp { 1643470decc6SDave Kleikamp int err; 1644470decc6SDave Kleikamp journal_superblock_t *sb; 1645470decc6SDave Kleikamp 1646470decc6SDave Kleikamp err = load_superblock(journal); 1647470decc6SDave Kleikamp if (err) 1648470decc6SDave Kleikamp return err; 1649470decc6SDave Kleikamp 1650470decc6SDave Kleikamp sb = journal->j_superblock; 1651470decc6SDave Kleikamp /* If this is a V2 superblock, then we have to check the 1652470decc6SDave Kleikamp * features flags on it. */ 1653470decc6SDave Kleikamp 1654470decc6SDave Kleikamp if (journal->j_format_version >= 2) { 1655470decc6SDave Kleikamp if ((sb->s_feature_ro_compat & 1656f7f4bccbSMingming Cao ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) || 1657470decc6SDave Kleikamp (sb->s_feature_incompat & 1658f7f4bccbSMingming Cao ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) { 1659470decc6SDave Kleikamp printk(KERN_WARNING 1660f2a44523SEryu Guan "JBD2: Unrecognised features on journal\n"); 1661470decc6SDave Kleikamp return -EINVAL; 1662470decc6SDave Kleikamp } 1663470decc6SDave Kleikamp } 1664470decc6SDave Kleikamp 1665d2eecb03STheodore Ts'o /* 1666d2eecb03STheodore Ts'o * Create a slab for this blocksize 1667d2eecb03STheodore Ts'o */ 1668d2eecb03STheodore Ts'o err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize)); 1669d2eecb03STheodore Ts'o if (err) 1670d2eecb03STheodore Ts'o return err; 1671d2eecb03STheodore Ts'o 1672470decc6SDave Kleikamp /* Let the recovery code check whether it needs to recover any 1673470decc6SDave Kleikamp * data from the journal. */ 1674f7f4bccbSMingming Cao if (jbd2_journal_recover(journal)) 1675470decc6SDave Kleikamp goto recovery_error; 1676470decc6SDave Kleikamp 1677e6a47428STheodore Ts'o if (journal->j_failed_commit) { 1678e6a47428STheodore Ts'o printk(KERN_ERR "JBD2: journal transaction %u on %s " 1679e6a47428STheodore Ts'o "is corrupt.\n", journal->j_failed_commit, 1680e6a47428STheodore Ts'o journal->j_devname); 16816a797d27SDarrick J. Wong return -EFSCORRUPTED; 1682e6a47428STheodore Ts'o } 1683e6a47428STheodore Ts'o 1684470decc6SDave Kleikamp /* OK, we've finished with the dynamic journal bits: 1685470decc6SDave Kleikamp * reinitialise the dynamic contents of the superblock in memory 1686470decc6SDave Kleikamp * and reset them on disk. */ 1687470decc6SDave Kleikamp if (journal_reset(journal)) 1688470decc6SDave Kleikamp goto recovery_error; 1689470decc6SDave Kleikamp 1690f7f4bccbSMingming Cao journal->j_flags &= ~JBD2_ABORT; 1691f7f4bccbSMingming Cao journal->j_flags |= JBD2_LOADED; 1692470decc6SDave Kleikamp return 0; 1693470decc6SDave Kleikamp 1694470decc6SDave Kleikamp recovery_error: 1695f2a44523SEryu Guan printk(KERN_WARNING "JBD2: recovery failed\n"); 1696470decc6SDave Kleikamp return -EIO; 1697470decc6SDave Kleikamp } 1698470decc6SDave Kleikamp 1699470decc6SDave Kleikamp /** 1700f7f4bccbSMingming Cao * void jbd2_journal_destroy() - Release a journal_t structure. 1701470decc6SDave Kleikamp * @journal: Journal to act on. 1702470decc6SDave Kleikamp * 1703470decc6SDave Kleikamp * Release a journal_t structure once it is no longer in use by the 1704470decc6SDave Kleikamp * journaled object. 170544519fafSHidehiro Kawai * Return <0 if we couldn't clean up the journal. 1706470decc6SDave Kleikamp */ 170744519fafSHidehiro Kawai int jbd2_journal_destroy(journal_t *journal) 1708470decc6SDave Kleikamp { 170944519fafSHidehiro Kawai int err = 0; 171044519fafSHidehiro Kawai 1711470decc6SDave Kleikamp /* Wait for the commit thread to wake up and die. */ 1712470decc6SDave Kleikamp journal_kill_thread(journal); 1713470decc6SDave Kleikamp 1714470decc6SDave Kleikamp /* Force a final log commit */ 1715470decc6SDave Kleikamp if (journal->j_running_transaction) 1716f7f4bccbSMingming Cao jbd2_journal_commit_transaction(journal); 1717470decc6SDave Kleikamp 1718470decc6SDave Kleikamp /* Force any old transactions to disk */ 1719470decc6SDave Kleikamp 1720470decc6SDave Kleikamp /* Totally anal locking here... */ 1721470decc6SDave Kleikamp spin_lock(&journal->j_list_lock); 1722470decc6SDave Kleikamp while (journal->j_checkpoint_transactions != NULL) { 1723470decc6SDave Kleikamp spin_unlock(&journal->j_list_lock); 17241a0d3786STheodore Ts'o mutex_lock(&journal->j_checkpoint_mutex); 1725841df7dfSJan Kara err = jbd2_log_do_checkpoint(journal); 17261a0d3786STheodore Ts'o mutex_unlock(&journal->j_checkpoint_mutex); 1727841df7dfSJan Kara /* 1728841df7dfSJan Kara * If checkpointing failed, just free the buffers to avoid 1729841df7dfSJan Kara * looping forever 1730841df7dfSJan Kara */ 1731841df7dfSJan Kara if (err) { 1732841df7dfSJan Kara jbd2_journal_destroy_checkpoint(journal); 1733841df7dfSJan Kara spin_lock(&journal->j_list_lock); 1734841df7dfSJan Kara break; 1735841df7dfSJan Kara } 1736470decc6SDave Kleikamp spin_lock(&journal->j_list_lock); 1737470decc6SDave Kleikamp } 1738470decc6SDave Kleikamp 1739470decc6SDave Kleikamp J_ASSERT(journal->j_running_transaction == NULL); 1740470decc6SDave Kleikamp J_ASSERT(journal->j_committing_transaction == NULL); 1741470decc6SDave Kleikamp J_ASSERT(journal->j_checkpoint_transactions == NULL); 1742470decc6SDave Kleikamp spin_unlock(&journal->j_list_lock); 1743470decc6SDave Kleikamp 174444519fafSHidehiro Kawai if (journal->j_sb_buffer) { 174544519fafSHidehiro Kawai if (!is_journal_aborted(journal)) { 1746a78bb11dSJan Kara mutex_lock(&journal->j_checkpoint_mutex); 1747c0a2ad9bSOGAWA Hirofumi 1748c0a2ad9bSOGAWA Hirofumi write_lock(&journal->j_state_lock); 1749c0a2ad9bSOGAWA Hirofumi journal->j_tail_sequence = 1750c0a2ad9bSOGAWA Hirofumi ++journal->j_transaction_sequence; 1751c0a2ad9bSOGAWA Hirofumi write_unlock(&journal->j_state_lock); 1752c0a2ad9bSOGAWA Hirofumi 1753c0a2ad9bSOGAWA Hirofumi jbd2_mark_journal_empty(journal, WRITE_FLUSH_FUA); 1754a78bb11dSJan Kara mutex_unlock(&journal->j_checkpoint_mutex); 1755a78bb11dSJan Kara } else 175644519fafSHidehiro Kawai err = -EIO; 1757470decc6SDave Kleikamp brelse(journal->j_sb_buffer); 1758470decc6SDave Kleikamp } 1759470decc6SDave Kleikamp 17608e85fb3fSJohann Lombardi if (journal->j_proc_entry) 17618e85fb3fSJohann Lombardi jbd2_stats_proc_exit(journal); 1762470decc6SDave Kleikamp iput(journal->j_inode); 1763470decc6SDave Kleikamp if (journal->j_revoke) 1764f7f4bccbSMingming Cao jbd2_journal_destroy_revoke(journal); 176501b5adceSDarrick J. Wong if (journal->j_chksum_driver) 176601b5adceSDarrick J. Wong crypto_free_shash(journal->j_chksum_driver); 1767470decc6SDave Kleikamp kfree(journal->j_wbuf); 1768470decc6SDave Kleikamp kfree(journal); 176944519fafSHidehiro Kawai 177044519fafSHidehiro Kawai return err; 1771470decc6SDave Kleikamp } 1772470decc6SDave Kleikamp 1773470decc6SDave Kleikamp 1774470decc6SDave Kleikamp /** 1775f7f4bccbSMingming Cao *int jbd2_journal_check_used_features () - Check if features specified are used. 1776470decc6SDave Kleikamp * @journal: Journal to check. 1777470decc6SDave Kleikamp * @compat: bitmask of compatible features 1778470decc6SDave Kleikamp * @ro: bitmask of features that force read-only mount 1779470decc6SDave Kleikamp * @incompat: bitmask of incompatible features 1780470decc6SDave Kleikamp * 1781470decc6SDave Kleikamp * Check whether the journal uses all of a given set of 1782470decc6SDave Kleikamp * features. Return true (non-zero) if it does. 1783470decc6SDave Kleikamp **/ 1784470decc6SDave Kleikamp 1785f7f4bccbSMingming Cao int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat, 1786470decc6SDave Kleikamp unsigned long ro, unsigned long incompat) 1787470decc6SDave Kleikamp { 1788470decc6SDave Kleikamp journal_superblock_t *sb; 1789470decc6SDave Kleikamp 1790470decc6SDave Kleikamp if (!compat && !ro && !incompat) 1791470decc6SDave Kleikamp return 1; 17921113e1b5SPatrick J. LoPresti /* Load journal superblock if it is not loaded yet. */ 17931113e1b5SPatrick J. LoPresti if (journal->j_format_version == 0 && 17941113e1b5SPatrick J. LoPresti journal_get_superblock(journal) != 0) 17951113e1b5SPatrick J. LoPresti return 0; 1796470decc6SDave Kleikamp if (journal->j_format_version == 1) 1797470decc6SDave Kleikamp return 0; 1798470decc6SDave Kleikamp 1799470decc6SDave Kleikamp sb = journal->j_superblock; 1800470decc6SDave Kleikamp 1801470decc6SDave Kleikamp if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) && 1802470decc6SDave Kleikamp ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) && 1803470decc6SDave Kleikamp ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat)) 1804470decc6SDave Kleikamp return 1; 1805470decc6SDave Kleikamp 1806470decc6SDave Kleikamp return 0; 1807470decc6SDave Kleikamp } 1808470decc6SDave Kleikamp 1809470decc6SDave Kleikamp /** 1810f7f4bccbSMingming Cao * int jbd2_journal_check_available_features() - Check feature set in journalling layer 1811470decc6SDave Kleikamp * @journal: Journal to check. 1812470decc6SDave Kleikamp * @compat: bitmask of compatible features 1813470decc6SDave Kleikamp * @ro: bitmask of features that force read-only mount 1814470decc6SDave Kleikamp * @incompat: bitmask of incompatible features 1815470decc6SDave Kleikamp * 1816470decc6SDave Kleikamp * Check whether the journaling code supports the use of 1817470decc6SDave Kleikamp * all of a given set of features on this journal. Return true 1818470decc6SDave Kleikamp * (non-zero) if it can. */ 1819470decc6SDave Kleikamp 1820f7f4bccbSMingming Cao int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat, 1821470decc6SDave Kleikamp unsigned long ro, unsigned long incompat) 1822470decc6SDave Kleikamp { 1823470decc6SDave Kleikamp if (!compat && !ro && !incompat) 1824470decc6SDave Kleikamp return 1; 1825470decc6SDave Kleikamp 1826470decc6SDave Kleikamp /* We can support any known requested features iff the 1827470decc6SDave Kleikamp * superblock is in version 2. Otherwise we fail to support any 1828470decc6SDave Kleikamp * extended sb features. */ 1829470decc6SDave Kleikamp 1830470decc6SDave Kleikamp if (journal->j_format_version != 2) 1831470decc6SDave Kleikamp return 0; 1832470decc6SDave Kleikamp 1833f7f4bccbSMingming Cao if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat && 1834f7f4bccbSMingming Cao (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro && 1835f7f4bccbSMingming Cao (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat) 1836470decc6SDave Kleikamp return 1; 1837470decc6SDave Kleikamp 1838470decc6SDave Kleikamp return 0; 1839470decc6SDave Kleikamp } 1840470decc6SDave Kleikamp 1841470decc6SDave Kleikamp /** 1842f7f4bccbSMingming Cao * int jbd2_journal_set_features () - Mark a given journal feature in the superblock 1843470decc6SDave Kleikamp * @journal: Journal to act on. 1844470decc6SDave Kleikamp * @compat: bitmask of compatible features 1845470decc6SDave Kleikamp * @ro: bitmask of features that force read-only mount 1846470decc6SDave Kleikamp * @incompat: bitmask of incompatible features 1847470decc6SDave Kleikamp * 1848470decc6SDave Kleikamp * Mark a given journal feature as present on the 1849470decc6SDave Kleikamp * superblock. Returns true if the requested features could be set. 1850470decc6SDave Kleikamp * 1851470decc6SDave Kleikamp */ 1852470decc6SDave Kleikamp 1853f7f4bccbSMingming Cao int jbd2_journal_set_features (journal_t *journal, unsigned long compat, 1854470decc6SDave Kleikamp unsigned long ro, unsigned long incompat) 1855470decc6SDave Kleikamp { 185625ed6e8aSDarrick J. Wong #define INCOMPAT_FEATURE_ON(f) \ 185725ed6e8aSDarrick J. Wong ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f))) 185825ed6e8aSDarrick J. Wong #define COMPAT_FEATURE_ON(f) \ 185925ed6e8aSDarrick J. Wong ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f))) 1860470decc6SDave Kleikamp journal_superblock_t *sb; 1861470decc6SDave Kleikamp 1862f7f4bccbSMingming Cao if (jbd2_journal_check_used_features(journal, compat, ro, incompat)) 1863470decc6SDave Kleikamp return 1; 1864470decc6SDave Kleikamp 1865f7f4bccbSMingming Cao if (!jbd2_journal_check_available_features(journal, compat, ro, incompat)) 1866470decc6SDave Kleikamp return 0; 1867470decc6SDave Kleikamp 1868db9ee220SDarrick J. Wong /* If enabling v2 checksums, turn on v3 instead */ 1869db9ee220SDarrick J. Wong if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) { 1870db9ee220SDarrick J. Wong incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2; 1871db9ee220SDarrick J. Wong incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3; 1872db9ee220SDarrick J. Wong } 1873db9ee220SDarrick J. Wong 1874db9ee220SDarrick J. Wong /* Asking for checksumming v3 and v1? Only give them v3. */ 1875db9ee220SDarrick J. Wong if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 && 187625ed6e8aSDarrick J. Wong compat & JBD2_FEATURE_COMPAT_CHECKSUM) 187725ed6e8aSDarrick J. Wong compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM; 187825ed6e8aSDarrick J. Wong 1879470decc6SDave Kleikamp jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n", 1880470decc6SDave Kleikamp compat, ro, incompat); 1881470decc6SDave Kleikamp 1882470decc6SDave Kleikamp sb = journal->j_superblock; 1883470decc6SDave Kleikamp 1884db9ee220SDarrick J. Wong /* If enabling v3 checksums, update superblock */ 1885db9ee220SDarrick J. Wong if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) { 188625ed6e8aSDarrick J. Wong sb->s_checksum_type = JBD2_CRC32C_CHKSUM; 188725ed6e8aSDarrick J. Wong sb->s_feature_compat &= 188825ed6e8aSDarrick J. Wong ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM); 188901b5adceSDarrick J. Wong 189001b5adceSDarrick J. Wong /* Load the checksum driver */ 189101b5adceSDarrick J. Wong if (journal->j_chksum_driver == NULL) { 189201b5adceSDarrick J. Wong journal->j_chksum_driver = crypto_alloc_shash("crc32c", 189301b5adceSDarrick J. Wong 0, 0); 189401b5adceSDarrick J. Wong if (IS_ERR(journal->j_chksum_driver)) { 1895a67c848aSDmitry Monakhov printk(KERN_ERR "JBD2: Cannot load crc32c " 189601b5adceSDarrick J. Wong "driver.\n"); 189701b5adceSDarrick J. Wong journal->j_chksum_driver = NULL; 189801b5adceSDarrick J. Wong return 0; 189901b5adceSDarrick J. Wong } 19004fd5ea43SDarrick J. Wong 19014fd5ea43SDarrick J. Wong /* Precompute checksum seed for all metadata */ 19024fd5ea43SDarrick J. Wong journal->j_csum_seed = jbd2_chksum(journal, ~0, 19034fd5ea43SDarrick J. Wong sb->s_uuid, 19044fd5ea43SDarrick J. Wong sizeof(sb->s_uuid)); 190525ed6e8aSDarrick J. Wong } 190632f38691SDarrick J. Wong } 190725ed6e8aSDarrick J. Wong 190825ed6e8aSDarrick J. Wong /* If enabling v1 checksums, downgrade superblock */ 190925ed6e8aSDarrick J. Wong if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM)) 191025ed6e8aSDarrick J. Wong sb->s_feature_incompat &= 1911db9ee220SDarrick J. Wong ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 | 1912db9ee220SDarrick J. Wong JBD2_FEATURE_INCOMPAT_CSUM_V3); 191325ed6e8aSDarrick J. Wong 1914470decc6SDave Kleikamp sb->s_feature_compat |= cpu_to_be32(compat); 1915470decc6SDave Kleikamp sb->s_feature_ro_compat |= cpu_to_be32(ro); 1916470decc6SDave Kleikamp sb->s_feature_incompat |= cpu_to_be32(incompat); 1917470decc6SDave Kleikamp 1918470decc6SDave Kleikamp return 1; 191925ed6e8aSDarrick J. Wong #undef COMPAT_FEATURE_ON 192025ed6e8aSDarrick J. Wong #undef INCOMPAT_FEATURE_ON 1921470decc6SDave Kleikamp } 1922470decc6SDave Kleikamp 1923818d276cSGirish Shilamkar /* 1924818d276cSGirish Shilamkar * jbd2_journal_clear_features () - Clear a given journal feature in the 1925818d276cSGirish Shilamkar * superblock 1926818d276cSGirish Shilamkar * @journal: Journal to act on. 1927818d276cSGirish Shilamkar * @compat: bitmask of compatible features 1928818d276cSGirish Shilamkar * @ro: bitmask of features that force read-only mount 1929818d276cSGirish Shilamkar * @incompat: bitmask of incompatible features 1930818d276cSGirish Shilamkar * 1931818d276cSGirish Shilamkar * Clear a given journal feature as present on the 1932818d276cSGirish Shilamkar * superblock. 1933818d276cSGirish Shilamkar */ 1934818d276cSGirish Shilamkar void jbd2_journal_clear_features(journal_t *journal, unsigned long compat, 1935818d276cSGirish Shilamkar unsigned long ro, unsigned long incompat) 1936818d276cSGirish Shilamkar { 1937818d276cSGirish Shilamkar journal_superblock_t *sb; 1938818d276cSGirish Shilamkar 1939818d276cSGirish Shilamkar jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n", 1940818d276cSGirish Shilamkar compat, ro, incompat); 1941818d276cSGirish Shilamkar 1942818d276cSGirish Shilamkar sb = journal->j_superblock; 1943818d276cSGirish Shilamkar 1944818d276cSGirish Shilamkar sb->s_feature_compat &= ~cpu_to_be32(compat); 1945818d276cSGirish Shilamkar sb->s_feature_ro_compat &= ~cpu_to_be32(ro); 1946818d276cSGirish Shilamkar sb->s_feature_incompat &= ~cpu_to_be32(incompat); 1947818d276cSGirish Shilamkar } 1948818d276cSGirish Shilamkar EXPORT_SYMBOL(jbd2_journal_clear_features); 1949470decc6SDave Kleikamp 1950470decc6SDave Kleikamp /** 1951f7f4bccbSMingming Cao * int jbd2_journal_flush () - Flush journal 1952470decc6SDave Kleikamp * @journal: Journal to act on. 1953470decc6SDave Kleikamp * 1954470decc6SDave Kleikamp * Flush all data for a given journal to disk and empty the journal. 1955470decc6SDave Kleikamp * Filesystems can use this when remounting readonly to ensure that 1956470decc6SDave Kleikamp * recovery does not need to happen on remount. 1957470decc6SDave Kleikamp */ 1958470decc6SDave Kleikamp 1959f7f4bccbSMingming Cao int jbd2_journal_flush(journal_t *journal) 1960470decc6SDave Kleikamp { 1961470decc6SDave Kleikamp int err = 0; 1962470decc6SDave Kleikamp transaction_t *transaction = NULL; 1963470decc6SDave Kleikamp 1964a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 1965470decc6SDave Kleikamp 1966470decc6SDave Kleikamp /* Force everything buffered to the log... */ 1967470decc6SDave Kleikamp if (journal->j_running_transaction) { 1968470decc6SDave Kleikamp transaction = journal->j_running_transaction; 1969f7f4bccbSMingming Cao __jbd2_log_start_commit(journal, transaction->t_tid); 1970470decc6SDave Kleikamp } else if (journal->j_committing_transaction) 1971470decc6SDave Kleikamp transaction = journal->j_committing_transaction; 1972470decc6SDave Kleikamp 1973470decc6SDave Kleikamp /* Wait for the log commit to complete... */ 1974470decc6SDave Kleikamp if (transaction) { 1975470decc6SDave Kleikamp tid_t tid = transaction->t_tid; 1976470decc6SDave Kleikamp 1977a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 1978f7f4bccbSMingming Cao jbd2_log_wait_commit(journal, tid); 1979470decc6SDave Kleikamp } else { 1980a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 1981470decc6SDave Kleikamp } 1982470decc6SDave Kleikamp 1983470decc6SDave Kleikamp /* ...and flush everything in the log out to disk. */ 1984470decc6SDave Kleikamp spin_lock(&journal->j_list_lock); 1985470decc6SDave Kleikamp while (!err && journal->j_checkpoint_transactions != NULL) { 1986470decc6SDave Kleikamp spin_unlock(&journal->j_list_lock); 198744519fafSHidehiro Kawai mutex_lock(&journal->j_checkpoint_mutex); 1988f7f4bccbSMingming Cao err = jbd2_log_do_checkpoint(journal); 198944519fafSHidehiro Kawai mutex_unlock(&journal->j_checkpoint_mutex); 1990470decc6SDave Kleikamp spin_lock(&journal->j_list_lock); 1991470decc6SDave Kleikamp } 1992470decc6SDave Kleikamp spin_unlock(&journal->j_list_lock); 199344519fafSHidehiro Kawai 199444519fafSHidehiro Kawai if (is_journal_aborted(journal)) 199544519fafSHidehiro Kawai return -EIO; 199644519fafSHidehiro Kawai 1997a78bb11dSJan Kara mutex_lock(&journal->j_checkpoint_mutex); 19986f6a6fdaSJoseph Qi if (!err) { 19996f6a6fdaSJoseph Qi err = jbd2_cleanup_journal_tail(journal); 20006f6a6fdaSJoseph Qi if (err < 0) { 20016f6a6fdaSJoseph Qi mutex_unlock(&journal->j_checkpoint_mutex); 20026f6a6fdaSJoseph Qi goto out; 20036f6a6fdaSJoseph Qi } 20046f6a6fdaSJoseph Qi err = 0; 20056f6a6fdaSJoseph Qi } 2006470decc6SDave Kleikamp 2007470decc6SDave Kleikamp /* Finally, mark the journal as really needing no recovery. 2008470decc6SDave Kleikamp * This sets s_start==0 in the underlying superblock, which is 2009470decc6SDave Kleikamp * the magic code for a fully-recovered superblock. Any future 2010470decc6SDave Kleikamp * commits of data to the journal will restore the current 2011470decc6SDave Kleikamp * s_start value. */ 2012c0a2ad9bSOGAWA Hirofumi jbd2_mark_journal_empty(journal, WRITE_FUA); 2013a78bb11dSJan Kara mutex_unlock(&journal->j_checkpoint_mutex); 2014a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 2015470decc6SDave Kleikamp J_ASSERT(!journal->j_running_transaction); 2016470decc6SDave Kleikamp J_ASSERT(!journal->j_committing_transaction); 2017470decc6SDave Kleikamp J_ASSERT(!journal->j_checkpoint_transactions); 2018470decc6SDave Kleikamp J_ASSERT(journal->j_head == journal->j_tail); 2019470decc6SDave Kleikamp J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence); 2020a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 20216f6a6fdaSJoseph Qi out: 20226f6a6fdaSJoseph Qi return err; 2023470decc6SDave Kleikamp } 2024470decc6SDave Kleikamp 2025470decc6SDave Kleikamp /** 2026f7f4bccbSMingming Cao * int jbd2_journal_wipe() - Wipe journal contents 2027470decc6SDave Kleikamp * @journal: Journal to act on. 2028470decc6SDave Kleikamp * @write: flag (see below) 2029470decc6SDave Kleikamp * 2030470decc6SDave Kleikamp * Wipe out all of the contents of a journal, safely. This will produce 2031470decc6SDave Kleikamp * a warning if the journal contains any valid recovery information. 2032f7f4bccbSMingming Cao * Must be called between journal_init_*() and jbd2_journal_load(). 2033470decc6SDave Kleikamp * 2034470decc6SDave Kleikamp * If 'write' is non-zero, then we wipe out the journal on disk; otherwise 2035470decc6SDave Kleikamp * we merely suppress recovery. 2036470decc6SDave Kleikamp */ 2037470decc6SDave Kleikamp 2038f7f4bccbSMingming Cao int jbd2_journal_wipe(journal_t *journal, int write) 2039470decc6SDave Kleikamp { 2040470decc6SDave Kleikamp int err = 0; 2041470decc6SDave Kleikamp 2042f7f4bccbSMingming Cao J_ASSERT (!(journal->j_flags & JBD2_LOADED)); 2043470decc6SDave Kleikamp 2044470decc6SDave Kleikamp err = load_superblock(journal); 2045470decc6SDave Kleikamp if (err) 2046470decc6SDave Kleikamp return err; 2047470decc6SDave Kleikamp 2048470decc6SDave Kleikamp if (!journal->j_tail) 2049470decc6SDave Kleikamp goto no_recovery; 2050470decc6SDave Kleikamp 2051f2a44523SEryu Guan printk(KERN_WARNING "JBD2: %s recovery information on journal\n", 2052470decc6SDave Kleikamp write ? "Clearing" : "Ignoring"); 2053470decc6SDave Kleikamp 2054f7f4bccbSMingming Cao err = jbd2_journal_skip_recovery(journal); 2055a78bb11dSJan Kara if (write) { 2056a78bb11dSJan Kara /* Lock to make assertions happy... */ 2057a78bb11dSJan Kara mutex_lock(&journal->j_checkpoint_mutex); 2058c0a2ad9bSOGAWA Hirofumi jbd2_mark_journal_empty(journal, WRITE_FUA); 2059a78bb11dSJan Kara mutex_unlock(&journal->j_checkpoint_mutex); 2060a78bb11dSJan Kara } 2061470decc6SDave Kleikamp 2062470decc6SDave Kleikamp no_recovery: 2063470decc6SDave Kleikamp return err; 2064470decc6SDave Kleikamp } 2065470decc6SDave Kleikamp 2066470decc6SDave Kleikamp /* 2067470decc6SDave Kleikamp * Journal abort has very specific semantics, which we describe 2068470decc6SDave Kleikamp * for journal abort. 2069470decc6SDave Kleikamp * 2070bfcd3555SAlberto Bertogli * Two internal functions, which provide abort to the jbd layer 2071470decc6SDave Kleikamp * itself are here. 2072470decc6SDave Kleikamp */ 2073470decc6SDave Kleikamp 2074470decc6SDave Kleikamp /* 2075470decc6SDave Kleikamp * Quick version for internal journal use (doesn't lock the journal). 2076470decc6SDave Kleikamp * Aborts hard --- we mark the abort as occurred, but do _nothing_ else, 2077470decc6SDave Kleikamp * and don't attempt to make any other journal updates. 2078470decc6SDave Kleikamp */ 2079f7f4bccbSMingming Cao void __jbd2_journal_abort_hard(journal_t *journal) 2080470decc6SDave Kleikamp { 2081470decc6SDave Kleikamp transaction_t *transaction; 2082470decc6SDave Kleikamp 2083f7f4bccbSMingming Cao if (journal->j_flags & JBD2_ABORT) 2084470decc6SDave Kleikamp return; 2085470decc6SDave Kleikamp 2086470decc6SDave Kleikamp printk(KERN_ERR "Aborting journal on device %s.\n", 208705496769STheodore Ts'o journal->j_devname); 2088470decc6SDave Kleikamp 2089a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 2090f7f4bccbSMingming Cao journal->j_flags |= JBD2_ABORT; 2091470decc6SDave Kleikamp transaction = journal->j_running_transaction; 2092470decc6SDave Kleikamp if (transaction) 2093f7f4bccbSMingming Cao __jbd2_log_start_commit(journal, transaction->t_tid); 2094a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 2095470decc6SDave Kleikamp } 2096470decc6SDave Kleikamp 2097470decc6SDave Kleikamp /* Soft abort: record the abort error status in the journal superblock, 2098470decc6SDave Kleikamp * but don't do any other IO. */ 2099470decc6SDave Kleikamp static void __journal_abort_soft (journal_t *journal, int errno) 2100470decc6SDave Kleikamp { 2101f7f4bccbSMingming Cao if (journal->j_flags & JBD2_ABORT) 2102470decc6SDave Kleikamp return; 2103470decc6SDave Kleikamp 2104470decc6SDave Kleikamp if (!journal->j_errno) 2105470decc6SDave Kleikamp journal->j_errno = errno; 2106470decc6SDave Kleikamp 2107f7f4bccbSMingming Cao __jbd2_journal_abort_hard(journal); 2108470decc6SDave Kleikamp 21094327ba52SDaeho Jeong if (errno) { 211024bcc89cSJan Kara jbd2_journal_update_sb_errno(journal); 21114327ba52SDaeho Jeong write_lock(&journal->j_state_lock); 21124327ba52SDaeho Jeong journal->j_flags |= JBD2_REC_ERR; 21134327ba52SDaeho Jeong write_unlock(&journal->j_state_lock); 21144327ba52SDaeho Jeong } 2115470decc6SDave Kleikamp } 2116470decc6SDave Kleikamp 2117470decc6SDave Kleikamp /** 2118f7f4bccbSMingming Cao * void jbd2_journal_abort () - Shutdown the journal immediately. 2119470decc6SDave Kleikamp * @journal: the journal to shutdown. 2120470decc6SDave Kleikamp * @errno: an error number to record in the journal indicating 2121470decc6SDave Kleikamp * the reason for the shutdown. 2122470decc6SDave Kleikamp * 2123470decc6SDave Kleikamp * Perform a complete, immediate shutdown of the ENTIRE 2124470decc6SDave Kleikamp * journal (not of a single transaction). This operation cannot be 2125470decc6SDave Kleikamp * undone without closing and reopening the journal. 2126470decc6SDave Kleikamp * 2127f7f4bccbSMingming Cao * The jbd2_journal_abort function is intended to support higher level error 2128470decc6SDave Kleikamp * recovery mechanisms such as the ext2/ext3 remount-readonly error 2129470decc6SDave Kleikamp * mode. 2130470decc6SDave Kleikamp * 2131470decc6SDave Kleikamp * Journal abort has very specific semantics. Any existing dirty, 2132470decc6SDave Kleikamp * unjournaled buffers in the main filesystem will still be written to 2133470decc6SDave Kleikamp * disk by bdflush, but the journaling mechanism will be suspended 2134470decc6SDave Kleikamp * immediately and no further transaction commits will be honoured. 2135470decc6SDave Kleikamp * 2136470decc6SDave Kleikamp * Any dirty, journaled buffers will be written back to disk without 2137470decc6SDave Kleikamp * hitting the journal. Atomicity cannot be guaranteed on an aborted 2138470decc6SDave Kleikamp * filesystem, but we _do_ attempt to leave as much data as possible 2139470decc6SDave Kleikamp * behind for fsck to use for cleanup. 2140470decc6SDave Kleikamp * 2141470decc6SDave Kleikamp * Any attempt to get a new transaction handle on a journal which is in 2142470decc6SDave Kleikamp * ABORT state will just result in an -EROFS error return. A 2143f7f4bccbSMingming Cao * jbd2_journal_stop on an existing handle will return -EIO if we have 2144470decc6SDave Kleikamp * entered abort state during the update. 2145470decc6SDave Kleikamp * 2146470decc6SDave Kleikamp * Recursive transactions are not disturbed by journal abort until the 2147f7f4bccbSMingming Cao * final jbd2_journal_stop, which will receive the -EIO error. 2148470decc6SDave Kleikamp * 2149f7f4bccbSMingming Cao * Finally, the jbd2_journal_abort call allows the caller to supply an errno 2150470decc6SDave Kleikamp * which will be recorded (if possible) in the journal superblock. This 2151470decc6SDave Kleikamp * allows a client to record failure conditions in the middle of a 2152470decc6SDave Kleikamp * transaction without having to complete the transaction to record the 2153470decc6SDave Kleikamp * failure to disk. ext3_error, for example, now uses this 2154470decc6SDave Kleikamp * functionality. 2155470decc6SDave Kleikamp * 2156470decc6SDave Kleikamp * Errors which originate from within the journaling layer will NOT 2157470decc6SDave Kleikamp * supply an errno; a null errno implies that absolutely no further 2158470decc6SDave Kleikamp * writes are done to the journal (unless there are any already in 2159470decc6SDave Kleikamp * progress). 2160470decc6SDave Kleikamp * 2161470decc6SDave Kleikamp */ 2162470decc6SDave Kleikamp 2163f7f4bccbSMingming Cao void jbd2_journal_abort(journal_t *journal, int errno) 2164470decc6SDave Kleikamp { 2165470decc6SDave Kleikamp __journal_abort_soft(journal, errno); 2166470decc6SDave Kleikamp } 2167470decc6SDave Kleikamp 2168470decc6SDave Kleikamp /** 2169f7f4bccbSMingming Cao * int jbd2_journal_errno () - returns the journal's error state. 2170470decc6SDave Kleikamp * @journal: journal to examine. 2171470decc6SDave Kleikamp * 2172bfcd3555SAlberto Bertogli * This is the errno number set with jbd2_journal_abort(), the last 2173470decc6SDave Kleikamp * time the journal was mounted - if the journal was stopped 2174470decc6SDave Kleikamp * without calling abort this will be 0. 2175470decc6SDave Kleikamp * 2176470decc6SDave Kleikamp * If the journal has been aborted on this mount time -EROFS will 2177470decc6SDave Kleikamp * be returned. 2178470decc6SDave Kleikamp */ 2179f7f4bccbSMingming Cao int jbd2_journal_errno(journal_t *journal) 2180470decc6SDave Kleikamp { 2181470decc6SDave Kleikamp int err; 2182470decc6SDave Kleikamp 2183a931da6aSTheodore Ts'o read_lock(&journal->j_state_lock); 2184f7f4bccbSMingming Cao if (journal->j_flags & JBD2_ABORT) 2185470decc6SDave Kleikamp err = -EROFS; 2186470decc6SDave Kleikamp else 2187470decc6SDave Kleikamp err = journal->j_errno; 2188a931da6aSTheodore Ts'o read_unlock(&journal->j_state_lock); 2189470decc6SDave Kleikamp return err; 2190470decc6SDave Kleikamp } 2191470decc6SDave Kleikamp 2192470decc6SDave Kleikamp /** 2193f7f4bccbSMingming Cao * int jbd2_journal_clear_err () - clears the journal's error state 2194470decc6SDave Kleikamp * @journal: journal to act on. 2195470decc6SDave Kleikamp * 2196bfcd3555SAlberto Bertogli * An error must be cleared or acked to take a FS out of readonly 2197470decc6SDave Kleikamp * mode. 2198470decc6SDave Kleikamp */ 2199f7f4bccbSMingming Cao int jbd2_journal_clear_err(journal_t *journal) 2200470decc6SDave Kleikamp { 2201470decc6SDave Kleikamp int err = 0; 2202470decc6SDave Kleikamp 2203a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 2204f7f4bccbSMingming Cao if (journal->j_flags & JBD2_ABORT) 2205470decc6SDave Kleikamp err = -EROFS; 2206470decc6SDave Kleikamp else 2207470decc6SDave Kleikamp journal->j_errno = 0; 2208a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 2209470decc6SDave Kleikamp return err; 2210470decc6SDave Kleikamp } 2211470decc6SDave Kleikamp 2212470decc6SDave Kleikamp /** 2213f7f4bccbSMingming Cao * void jbd2_journal_ack_err() - Ack journal err. 2214470decc6SDave Kleikamp * @journal: journal to act on. 2215470decc6SDave Kleikamp * 2216bfcd3555SAlberto Bertogli * An error must be cleared or acked to take a FS out of readonly 2217470decc6SDave Kleikamp * mode. 2218470decc6SDave Kleikamp */ 2219f7f4bccbSMingming Cao void jbd2_journal_ack_err(journal_t *journal) 2220470decc6SDave Kleikamp { 2221a931da6aSTheodore Ts'o write_lock(&journal->j_state_lock); 2222470decc6SDave Kleikamp if (journal->j_errno) 2223f7f4bccbSMingming Cao journal->j_flags |= JBD2_ACK_ERR; 2224a931da6aSTheodore Ts'o write_unlock(&journal->j_state_lock); 2225470decc6SDave Kleikamp } 2226470decc6SDave Kleikamp 2227f7f4bccbSMingming Cao int jbd2_journal_blocks_per_page(struct inode *inode) 2228470decc6SDave Kleikamp { 222909cbfeafSKirill A. Shutemov return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits); 2230470decc6SDave Kleikamp } 2231470decc6SDave Kleikamp 2232470decc6SDave Kleikamp /* 2233b517bea1SZach Brown * helper functions to deal with 32 or 64bit block numbers. 2234b517bea1SZach Brown */ 2235b517bea1SZach Brown size_t journal_tag_bytes(journal_t *journal) 2236b517bea1SZach Brown { 2237db9ee220SDarrick J. Wong size_t sz; 2238db9ee220SDarrick J. Wong 223956316a0dSDarrick J. Wong if (jbd2_has_feature_csum3(journal)) 2240db9ee220SDarrick J. Wong return sizeof(journal_block_tag3_t); 2241db9ee220SDarrick J. Wong 2242db9ee220SDarrick J. Wong sz = sizeof(journal_block_tag_t); 2243c3900875SDarrick J. Wong 224456316a0dSDarrick J. Wong if (jbd2_has_feature_csum2(journal)) 2245db9ee220SDarrick J. Wong sz += sizeof(__u16); 2246c3900875SDarrick J. Wong 224756316a0dSDarrick J. Wong if (jbd2_has_feature_64bit(journal)) 2248db9ee220SDarrick J. Wong return sz; 2249b517bea1SZach Brown else 2250db9ee220SDarrick J. Wong return sz - sizeof(__u32); 2251b517bea1SZach Brown } 2252b517bea1SZach Brown 2253b517bea1SZach Brown /* 2254d2eecb03STheodore Ts'o * JBD memory management 2255d2eecb03STheodore Ts'o * 2256d2eecb03STheodore Ts'o * These functions are used to allocate block-sized chunks of memory 2257d2eecb03STheodore Ts'o * used for making copies of buffer_head data. Very often it will be 2258d2eecb03STheodore Ts'o * page-sized chunks of data, but sometimes it will be in 2259d2eecb03STheodore Ts'o * sub-page-size chunks. (For example, 16k pages on Power systems 2260d2eecb03STheodore Ts'o * with a 4k block file system.) For blocks smaller than a page, we 2261d2eecb03STheodore Ts'o * use a SLAB allocator. There are slab caches for each block size, 2262d2eecb03STheodore Ts'o * which are allocated at mount time, if necessary, and we only free 2263d2eecb03STheodore Ts'o * (all of) the slab caches when/if the jbd2 module is unloaded. For 2264d2eecb03STheodore Ts'o * this reason we don't need to a mutex to protect access to 2265d2eecb03STheodore Ts'o * jbd2_slab[] allocating or releasing memory; only in 2266d2eecb03STheodore Ts'o * jbd2_journal_create_slab(). 2267d2eecb03STheodore Ts'o */ 2268d2eecb03STheodore Ts'o #define JBD2_MAX_SLABS 8 2269d2eecb03STheodore Ts'o static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS]; 2270d2eecb03STheodore Ts'o 2271d2eecb03STheodore Ts'o static const char *jbd2_slab_names[JBD2_MAX_SLABS] = { 2272d2eecb03STheodore Ts'o "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k", 2273d2eecb03STheodore Ts'o "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k" 2274d2eecb03STheodore Ts'o }; 2275d2eecb03STheodore Ts'o 2276d2eecb03STheodore Ts'o 2277d2eecb03STheodore Ts'o static void jbd2_journal_destroy_slabs(void) 2278d2eecb03STheodore Ts'o { 2279d2eecb03STheodore Ts'o int i; 2280d2eecb03STheodore Ts'o 2281d2eecb03STheodore Ts'o for (i = 0; i < JBD2_MAX_SLABS; i++) { 2282d2eecb03STheodore Ts'o if (jbd2_slab[i]) 2283d2eecb03STheodore Ts'o kmem_cache_destroy(jbd2_slab[i]); 2284d2eecb03STheodore Ts'o jbd2_slab[i] = NULL; 2285d2eecb03STheodore Ts'o } 2286d2eecb03STheodore Ts'o } 2287d2eecb03STheodore Ts'o 2288d2eecb03STheodore Ts'o static int jbd2_journal_create_slab(size_t size) 2289d2eecb03STheodore Ts'o { 229051dfacdeSThomas Gleixner static DEFINE_MUTEX(jbd2_slab_create_mutex); 2291d2eecb03STheodore Ts'o int i = order_base_2(size) - 10; 2292d2eecb03STheodore Ts'o size_t slab_size; 2293d2eecb03STheodore Ts'o 2294d2eecb03STheodore Ts'o if (size == PAGE_SIZE) 2295d2eecb03STheodore Ts'o return 0; 2296d2eecb03STheodore Ts'o 2297d2eecb03STheodore Ts'o if (i >= JBD2_MAX_SLABS) 2298d2eecb03STheodore Ts'o return -EINVAL; 2299d2eecb03STheodore Ts'o 2300d2eecb03STheodore Ts'o if (unlikely(i < 0)) 2301d2eecb03STheodore Ts'o i = 0; 230251dfacdeSThomas Gleixner mutex_lock(&jbd2_slab_create_mutex); 2303d2eecb03STheodore Ts'o if (jbd2_slab[i]) { 230451dfacdeSThomas Gleixner mutex_unlock(&jbd2_slab_create_mutex); 2305d2eecb03STheodore Ts'o return 0; /* Already created */ 2306d2eecb03STheodore Ts'o } 2307d2eecb03STheodore Ts'o 2308d2eecb03STheodore Ts'o slab_size = 1 << (i+10); 2309d2eecb03STheodore Ts'o jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size, 2310d2eecb03STheodore Ts'o slab_size, 0, NULL); 231151dfacdeSThomas Gleixner mutex_unlock(&jbd2_slab_create_mutex); 2312d2eecb03STheodore Ts'o if (!jbd2_slab[i]) { 2313d2eecb03STheodore Ts'o printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n"); 2314d2eecb03STheodore Ts'o return -ENOMEM; 2315d2eecb03STheodore Ts'o } 2316d2eecb03STheodore Ts'o return 0; 2317d2eecb03STheodore Ts'o } 2318d2eecb03STheodore Ts'o 2319d2eecb03STheodore Ts'o static struct kmem_cache *get_slab(size_t size) 2320d2eecb03STheodore Ts'o { 2321d2eecb03STheodore Ts'o int i = order_base_2(size) - 10; 2322d2eecb03STheodore Ts'o 2323d2eecb03STheodore Ts'o BUG_ON(i >= JBD2_MAX_SLABS); 2324d2eecb03STheodore Ts'o if (unlikely(i < 0)) 2325d2eecb03STheodore Ts'o i = 0; 23268ac97b74SBill Pemberton BUG_ON(jbd2_slab[i] == NULL); 2327d2eecb03STheodore Ts'o return jbd2_slab[i]; 2328d2eecb03STheodore Ts'o } 2329d2eecb03STheodore Ts'o 2330d2eecb03STheodore Ts'o void *jbd2_alloc(size_t size, gfp_t flags) 2331d2eecb03STheodore Ts'o { 2332d2eecb03STheodore Ts'o void *ptr; 2333d2eecb03STheodore Ts'o 2334d2eecb03STheodore Ts'o BUG_ON(size & (size-1)); /* Must be a power of 2 */ 2335d2eecb03STheodore Ts'o 2336d2eecb03STheodore Ts'o flags |= __GFP_REPEAT; 2337d2eecb03STheodore Ts'o if (size == PAGE_SIZE) 2338d2eecb03STheodore Ts'o ptr = (void *)__get_free_pages(flags, 0); 2339d2eecb03STheodore Ts'o else if (size > PAGE_SIZE) { 2340d2eecb03STheodore Ts'o int order = get_order(size); 2341d2eecb03STheodore Ts'o 2342d2eecb03STheodore Ts'o if (order < 3) 2343d2eecb03STheodore Ts'o ptr = (void *)__get_free_pages(flags, order); 2344d2eecb03STheodore Ts'o else 2345d2eecb03STheodore Ts'o ptr = vmalloc(size); 2346d2eecb03STheodore Ts'o } else 2347d2eecb03STheodore Ts'o ptr = kmem_cache_alloc(get_slab(size), flags); 2348d2eecb03STheodore Ts'o 2349d2eecb03STheodore Ts'o /* Check alignment; SLUB has gotten this wrong in the past, 2350d2eecb03STheodore Ts'o * and this can lead to user data corruption! */ 2351d2eecb03STheodore Ts'o BUG_ON(((unsigned long) ptr) & (size-1)); 2352d2eecb03STheodore Ts'o 2353d2eecb03STheodore Ts'o return ptr; 2354d2eecb03STheodore Ts'o } 2355d2eecb03STheodore Ts'o 2356d2eecb03STheodore Ts'o void jbd2_free(void *ptr, size_t size) 2357d2eecb03STheodore Ts'o { 2358d2eecb03STheodore Ts'o if (size == PAGE_SIZE) { 2359d2eecb03STheodore Ts'o free_pages((unsigned long)ptr, 0); 2360d2eecb03STheodore Ts'o return; 2361d2eecb03STheodore Ts'o } 2362d2eecb03STheodore Ts'o if (size > PAGE_SIZE) { 2363d2eecb03STheodore Ts'o int order = get_order(size); 2364d2eecb03STheodore Ts'o 2365d2eecb03STheodore Ts'o if (order < 3) 2366d2eecb03STheodore Ts'o free_pages((unsigned long)ptr, order); 2367d2eecb03STheodore Ts'o else 2368d2eecb03STheodore Ts'o vfree(ptr); 2369d2eecb03STheodore Ts'o return; 2370d2eecb03STheodore Ts'o } 2371d2eecb03STheodore Ts'o kmem_cache_free(get_slab(size), ptr); 2372d2eecb03STheodore Ts'o }; 2373d2eecb03STheodore Ts'o 2374d2eecb03STheodore Ts'o /* 2375470decc6SDave Kleikamp * Journal_head storage management 2376470decc6SDave Kleikamp */ 2377e18b890bSChristoph Lameter static struct kmem_cache *jbd2_journal_head_cache; 2378e23291b9SJose R. Santos #ifdef CONFIG_JBD2_DEBUG 2379470decc6SDave Kleikamp static atomic_t nr_journal_heads = ATOMIC_INIT(0); 2380470decc6SDave Kleikamp #endif 2381470decc6SDave Kleikamp 23824185a2acSYongqiang Yang static int jbd2_journal_init_journal_head_cache(void) 2383470decc6SDave Kleikamp { 2384470decc6SDave Kleikamp int retval; 2385470decc6SDave Kleikamp 23861076d17aSAl Viro J_ASSERT(jbd2_journal_head_cache == NULL); 2387a920e941SJohann Lombardi jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head", 2388470decc6SDave Kleikamp sizeof(struct journal_head), 2389470decc6SDave Kleikamp 0, /* offset */ 2390de92c8caSJan Kara SLAB_TEMPORARY | SLAB_DESTROY_BY_RCU, 239120c2df83SPaul Mundt NULL); /* ctor */ 2392470decc6SDave Kleikamp retval = 0; 23931076d17aSAl Viro if (!jbd2_journal_head_cache) { 2394470decc6SDave Kleikamp retval = -ENOMEM; 2395f2a44523SEryu Guan printk(KERN_EMERG "JBD2: no memory for journal_head cache\n"); 2396470decc6SDave Kleikamp } 2397470decc6SDave Kleikamp return retval; 2398470decc6SDave Kleikamp } 2399470decc6SDave Kleikamp 24004185a2acSYongqiang Yang static void jbd2_journal_destroy_journal_head_cache(void) 2401470decc6SDave Kleikamp { 24028a9362ebSDuane Griffin if (jbd2_journal_head_cache) { 2403f7f4bccbSMingming Cao kmem_cache_destroy(jbd2_journal_head_cache); 2404f7f4bccbSMingming Cao jbd2_journal_head_cache = NULL; 2405470decc6SDave Kleikamp } 24068a9362ebSDuane Griffin } 2407470decc6SDave Kleikamp 2408470decc6SDave Kleikamp /* 2409470decc6SDave Kleikamp * journal_head splicing and dicing 2410470decc6SDave Kleikamp */ 2411470decc6SDave Kleikamp static struct journal_head *journal_alloc_journal_head(void) 2412470decc6SDave Kleikamp { 2413470decc6SDave Kleikamp struct journal_head *ret; 2414470decc6SDave Kleikamp 2415e23291b9SJose R. Santos #ifdef CONFIG_JBD2_DEBUG 2416470decc6SDave Kleikamp atomic_inc(&nr_journal_heads); 2417470decc6SDave Kleikamp #endif 24185d9cf9c6SZheng Liu ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS); 24191076d17aSAl Viro if (!ret) { 2420470decc6SDave Kleikamp jbd_debug(1, "out of memory for journal_head\n"); 2421670be5a7STheodore Ts'o pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__); 24227b506b10SMichal Hocko ret = kmem_cache_zalloc(jbd2_journal_head_cache, 24237b506b10SMichal Hocko GFP_NOFS | __GFP_NOFAIL); 2424470decc6SDave Kleikamp } 2425470decc6SDave Kleikamp return ret; 2426470decc6SDave Kleikamp } 2427470decc6SDave Kleikamp 2428470decc6SDave Kleikamp static void journal_free_journal_head(struct journal_head *jh) 2429470decc6SDave Kleikamp { 2430e23291b9SJose R. Santos #ifdef CONFIG_JBD2_DEBUG 2431470decc6SDave Kleikamp atomic_dec(&nr_journal_heads); 2432cd02ff0bSMingming Cao memset(jh, JBD2_POISON_FREE, sizeof(*jh)); 2433470decc6SDave Kleikamp #endif 2434f7f4bccbSMingming Cao kmem_cache_free(jbd2_journal_head_cache, jh); 2435470decc6SDave Kleikamp } 2436470decc6SDave Kleikamp 2437470decc6SDave Kleikamp /* 2438470decc6SDave Kleikamp * A journal_head is attached to a buffer_head whenever JBD has an 2439470decc6SDave Kleikamp * interest in the buffer. 2440470decc6SDave Kleikamp * 2441470decc6SDave Kleikamp * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit 2442470decc6SDave Kleikamp * is set. This bit is tested in core kernel code where we need to take 2443470decc6SDave Kleikamp * JBD-specific actions. Testing the zeroness of ->b_private is not reliable 2444470decc6SDave Kleikamp * there. 2445470decc6SDave Kleikamp * 2446470decc6SDave Kleikamp * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one. 2447470decc6SDave Kleikamp * 2448470decc6SDave Kleikamp * When a buffer has its BH_JBD bit set it is immune from being released by 2449470decc6SDave Kleikamp * core kernel code, mainly via ->b_count. 2450470decc6SDave Kleikamp * 2451de1b7941SJan Kara * A journal_head is detached from its buffer_head when the journal_head's 2452de1b7941SJan Kara * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint 2453de1b7941SJan Kara * transaction (b_cp_transaction) hold their references to b_jcount. 2454470decc6SDave Kleikamp * 2455470decc6SDave Kleikamp * Various places in the kernel want to attach a journal_head to a buffer_head 2456470decc6SDave Kleikamp * _before_ attaching the journal_head to a transaction. To protect the 2457f7f4bccbSMingming Cao * journal_head in this situation, jbd2_journal_add_journal_head elevates the 2458470decc6SDave Kleikamp * journal_head's b_jcount refcount by one. The caller must call 2459f7f4bccbSMingming Cao * jbd2_journal_put_journal_head() to undo this. 2460470decc6SDave Kleikamp * 2461470decc6SDave Kleikamp * So the typical usage would be: 2462470decc6SDave Kleikamp * 2463470decc6SDave Kleikamp * (Attach a journal_head if needed. Increments b_jcount) 2464f7f4bccbSMingming Cao * struct journal_head *jh = jbd2_journal_add_journal_head(bh); 2465470decc6SDave Kleikamp * ... 2466de1b7941SJan Kara * (Get another reference for transaction) 2467de1b7941SJan Kara * jbd2_journal_grab_journal_head(bh); 2468470decc6SDave Kleikamp * jh->b_transaction = xxx; 2469de1b7941SJan Kara * (Put original reference) 2470f7f4bccbSMingming Cao * jbd2_journal_put_journal_head(jh); 2471470decc6SDave Kleikamp */ 2472470decc6SDave Kleikamp 2473470decc6SDave Kleikamp /* 2474470decc6SDave Kleikamp * Give a buffer_head a journal_head. 2475470decc6SDave Kleikamp * 2476470decc6SDave Kleikamp * May sleep. 2477470decc6SDave Kleikamp */ 2478f7f4bccbSMingming Cao struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh) 2479470decc6SDave Kleikamp { 2480470decc6SDave Kleikamp struct journal_head *jh; 2481470decc6SDave Kleikamp struct journal_head *new_jh = NULL; 2482470decc6SDave Kleikamp 2483470decc6SDave Kleikamp repeat: 24845d9cf9c6SZheng Liu if (!buffer_jbd(bh)) 2485470decc6SDave Kleikamp new_jh = journal_alloc_journal_head(); 2486470decc6SDave Kleikamp 2487470decc6SDave Kleikamp jbd_lock_bh_journal_head(bh); 2488470decc6SDave Kleikamp if (buffer_jbd(bh)) { 2489470decc6SDave Kleikamp jh = bh2jh(bh); 2490470decc6SDave Kleikamp } else { 2491470decc6SDave Kleikamp J_ASSERT_BH(bh, 2492470decc6SDave Kleikamp (atomic_read(&bh->b_count) > 0) || 2493470decc6SDave Kleikamp (bh->b_page && bh->b_page->mapping)); 2494470decc6SDave Kleikamp 2495470decc6SDave Kleikamp if (!new_jh) { 2496470decc6SDave Kleikamp jbd_unlock_bh_journal_head(bh); 2497470decc6SDave Kleikamp goto repeat; 2498470decc6SDave Kleikamp } 2499470decc6SDave Kleikamp 2500470decc6SDave Kleikamp jh = new_jh; 2501470decc6SDave Kleikamp new_jh = NULL; /* We consumed it */ 2502470decc6SDave Kleikamp set_buffer_jbd(bh); 2503470decc6SDave Kleikamp bh->b_private = jh; 2504470decc6SDave Kleikamp jh->b_bh = bh; 2505470decc6SDave Kleikamp get_bh(bh); 2506470decc6SDave Kleikamp BUFFER_TRACE(bh, "added journal_head"); 2507470decc6SDave Kleikamp } 2508470decc6SDave Kleikamp jh->b_jcount++; 2509470decc6SDave Kleikamp jbd_unlock_bh_journal_head(bh); 2510470decc6SDave Kleikamp if (new_jh) 2511470decc6SDave Kleikamp journal_free_journal_head(new_jh); 2512470decc6SDave Kleikamp return bh->b_private; 2513470decc6SDave Kleikamp } 2514470decc6SDave Kleikamp 2515470decc6SDave Kleikamp /* 2516470decc6SDave Kleikamp * Grab a ref against this buffer_head's journal_head. If it ended up not 2517470decc6SDave Kleikamp * having a journal_head, return NULL 2518470decc6SDave Kleikamp */ 2519f7f4bccbSMingming Cao struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh) 2520470decc6SDave Kleikamp { 2521470decc6SDave Kleikamp struct journal_head *jh = NULL; 2522470decc6SDave Kleikamp 2523470decc6SDave Kleikamp jbd_lock_bh_journal_head(bh); 2524470decc6SDave Kleikamp if (buffer_jbd(bh)) { 2525470decc6SDave Kleikamp jh = bh2jh(bh); 2526470decc6SDave Kleikamp jh->b_jcount++; 2527470decc6SDave Kleikamp } 2528470decc6SDave Kleikamp jbd_unlock_bh_journal_head(bh); 2529470decc6SDave Kleikamp return jh; 2530470decc6SDave Kleikamp } 2531470decc6SDave Kleikamp 2532470decc6SDave Kleikamp static void __journal_remove_journal_head(struct buffer_head *bh) 2533470decc6SDave Kleikamp { 2534470decc6SDave Kleikamp struct journal_head *jh = bh2jh(bh); 2535470decc6SDave Kleikamp 2536470decc6SDave Kleikamp J_ASSERT_JH(jh, jh->b_jcount >= 0); 2537de1b7941SJan Kara J_ASSERT_JH(jh, jh->b_transaction == NULL); 2538de1b7941SJan Kara J_ASSERT_JH(jh, jh->b_next_transaction == NULL); 2539de1b7941SJan Kara J_ASSERT_JH(jh, jh->b_cp_transaction == NULL); 2540470decc6SDave Kleikamp J_ASSERT_JH(jh, jh->b_jlist == BJ_None); 2541470decc6SDave Kleikamp J_ASSERT_BH(bh, buffer_jbd(bh)); 2542470decc6SDave Kleikamp J_ASSERT_BH(bh, jh2bh(jh) == bh); 2543470decc6SDave Kleikamp BUFFER_TRACE(bh, "remove journal_head"); 2544470decc6SDave Kleikamp if (jh->b_frozen_data) { 2545de1b7941SJan Kara printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__); 2546af1e76d6SMingming Cao jbd2_free(jh->b_frozen_data, bh->b_size); 2547470decc6SDave Kleikamp } 2548470decc6SDave Kleikamp if (jh->b_committed_data) { 2549de1b7941SJan Kara printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__); 2550af1e76d6SMingming Cao jbd2_free(jh->b_committed_data, bh->b_size); 2551470decc6SDave Kleikamp } 2552470decc6SDave Kleikamp bh->b_private = NULL; 2553470decc6SDave Kleikamp jh->b_bh = NULL; /* debug, really */ 2554470decc6SDave Kleikamp clear_buffer_jbd(bh); 2555470decc6SDave Kleikamp journal_free_journal_head(jh); 2556470decc6SDave Kleikamp } 2557470decc6SDave Kleikamp 2558470decc6SDave Kleikamp /* 2559de1b7941SJan Kara * Drop a reference on the passed journal_head. If it fell to zero then 2560470decc6SDave Kleikamp * release the journal_head from the buffer_head. 2561470decc6SDave Kleikamp */ 2562f7f4bccbSMingming Cao void jbd2_journal_put_journal_head(struct journal_head *jh) 2563470decc6SDave Kleikamp { 2564470decc6SDave Kleikamp struct buffer_head *bh = jh2bh(jh); 2565470decc6SDave Kleikamp 2566470decc6SDave Kleikamp jbd_lock_bh_journal_head(bh); 2567470decc6SDave Kleikamp J_ASSERT_JH(jh, jh->b_jcount > 0); 2568470decc6SDave Kleikamp --jh->b_jcount; 2569de1b7941SJan Kara if (!jh->b_jcount) { 2570470decc6SDave Kleikamp __journal_remove_journal_head(bh); 2571de1b7941SJan Kara jbd_unlock_bh_journal_head(bh); 2572470decc6SDave Kleikamp __brelse(bh); 2573de1b7941SJan Kara } else 2574470decc6SDave Kleikamp jbd_unlock_bh_journal_head(bh); 2575470decc6SDave Kleikamp } 2576470decc6SDave Kleikamp 2577470decc6SDave Kleikamp /* 2578c851ed54SJan Kara * Initialize jbd inode head 2579c851ed54SJan Kara */ 2580c851ed54SJan Kara void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode) 2581c851ed54SJan Kara { 2582c851ed54SJan Kara jinode->i_transaction = NULL; 2583c851ed54SJan Kara jinode->i_next_transaction = NULL; 2584c851ed54SJan Kara jinode->i_vfs_inode = inode; 2585c851ed54SJan Kara jinode->i_flags = 0; 2586c851ed54SJan Kara INIT_LIST_HEAD(&jinode->i_list); 2587c851ed54SJan Kara } 2588c851ed54SJan Kara 2589c851ed54SJan Kara /* 2590c851ed54SJan Kara * Function to be called before we start removing inode from memory (i.e., 2591c851ed54SJan Kara * clear_inode() is a fine place to be called from). It removes inode from 2592c851ed54SJan Kara * transaction's lists. 2593c851ed54SJan Kara */ 2594c851ed54SJan Kara void jbd2_journal_release_jbd_inode(journal_t *journal, 2595c851ed54SJan Kara struct jbd2_inode *jinode) 2596c851ed54SJan Kara { 2597c851ed54SJan Kara if (!journal) 2598c851ed54SJan Kara return; 2599c851ed54SJan Kara restart: 2600c851ed54SJan Kara spin_lock(&journal->j_list_lock); 2601c851ed54SJan Kara /* Is commit writing out inode - we have to wait */ 2602cb0d9d47SJan Kara if (jinode->i_flags & JI_COMMIT_RUNNING) { 2603c851ed54SJan Kara wait_queue_head_t *wq; 2604c851ed54SJan Kara DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING); 2605c851ed54SJan Kara wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING); 2606c851ed54SJan Kara prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); 2607c851ed54SJan Kara spin_unlock(&journal->j_list_lock); 2608c851ed54SJan Kara schedule(); 2609c851ed54SJan Kara finish_wait(wq, &wait.wait); 2610c851ed54SJan Kara goto restart; 2611c851ed54SJan Kara } 2612c851ed54SJan Kara 2613c851ed54SJan Kara if (jinode->i_transaction) { 2614c851ed54SJan Kara list_del(&jinode->i_list); 2615c851ed54SJan Kara jinode->i_transaction = NULL; 2616c851ed54SJan Kara } 2617c851ed54SJan Kara spin_unlock(&journal->j_list_lock); 2618c851ed54SJan Kara } 2619c851ed54SJan Kara 2620470decc6SDave Kleikamp 26218e85fb3fSJohann Lombardi #ifdef CONFIG_PROC_FS 26228e85fb3fSJohann Lombardi 26238e85fb3fSJohann Lombardi #define JBD2_STATS_PROC_NAME "fs/jbd2" 26248e85fb3fSJohann Lombardi 26258e85fb3fSJohann Lombardi static void __init jbd2_create_jbd_stats_proc_entry(void) 26268e85fb3fSJohann Lombardi { 26278e85fb3fSJohann Lombardi proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL); 26288e85fb3fSJohann Lombardi } 26298e85fb3fSJohann Lombardi 26308e85fb3fSJohann Lombardi static void __exit jbd2_remove_jbd_stats_proc_entry(void) 26318e85fb3fSJohann Lombardi { 26328e85fb3fSJohann Lombardi if (proc_jbd2_stats) 26338e85fb3fSJohann Lombardi remove_proc_entry(JBD2_STATS_PROC_NAME, NULL); 26348e85fb3fSJohann Lombardi } 26358e85fb3fSJohann Lombardi 26368e85fb3fSJohann Lombardi #else 26378e85fb3fSJohann Lombardi 26388e85fb3fSJohann Lombardi #define jbd2_create_jbd_stats_proc_entry() do {} while (0) 26398e85fb3fSJohann Lombardi #define jbd2_remove_jbd_stats_proc_entry() do {} while (0) 26408e85fb3fSJohann Lombardi 26418e85fb3fSJohann Lombardi #endif 26428e85fb3fSJohann Lombardi 26438aefcd55STheodore Ts'o struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache; 2644470decc6SDave Kleikamp 26454185a2acSYongqiang Yang static int __init jbd2_journal_init_handle_cache(void) 2646470decc6SDave Kleikamp { 26478aefcd55STheodore Ts'o jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY); 2648f7f4bccbSMingming Cao if (jbd2_handle_cache == NULL) { 26498aefcd55STheodore Ts'o printk(KERN_EMERG "JBD2: failed to create handle cache\n"); 26508aefcd55STheodore Ts'o return -ENOMEM; 26518aefcd55STheodore Ts'o } 26528aefcd55STheodore Ts'o jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0); 26538aefcd55STheodore Ts'o if (jbd2_inode_cache == NULL) { 26548aefcd55STheodore Ts'o printk(KERN_EMERG "JBD2: failed to create inode cache\n"); 26558aefcd55STheodore Ts'o kmem_cache_destroy(jbd2_handle_cache); 2656470decc6SDave Kleikamp return -ENOMEM; 2657470decc6SDave Kleikamp } 2658470decc6SDave Kleikamp return 0; 2659470decc6SDave Kleikamp } 2660470decc6SDave Kleikamp 2661f7f4bccbSMingming Cao static void jbd2_journal_destroy_handle_cache(void) 2662470decc6SDave Kleikamp { 2663f7f4bccbSMingming Cao if (jbd2_handle_cache) 2664f7f4bccbSMingming Cao kmem_cache_destroy(jbd2_handle_cache); 26658aefcd55STheodore Ts'o if (jbd2_inode_cache) 26668aefcd55STheodore Ts'o kmem_cache_destroy(jbd2_inode_cache); 26678aefcd55STheodore Ts'o 2668470decc6SDave Kleikamp } 2669470decc6SDave Kleikamp 2670470decc6SDave Kleikamp /* 2671470decc6SDave Kleikamp * Module startup and shutdown 2672470decc6SDave Kleikamp */ 2673470decc6SDave Kleikamp 2674470decc6SDave Kleikamp static int __init journal_init_caches(void) 2675470decc6SDave Kleikamp { 2676470decc6SDave Kleikamp int ret; 2677470decc6SDave Kleikamp 2678f7f4bccbSMingming Cao ret = jbd2_journal_init_revoke_caches(); 2679470decc6SDave Kleikamp if (ret == 0) 26804185a2acSYongqiang Yang ret = jbd2_journal_init_journal_head_cache(); 2681470decc6SDave Kleikamp if (ret == 0) 26824185a2acSYongqiang Yang ret = jbd2_journal_init_handle_cache(); 26830c2022ecSYongqiang Yang if (ret == 0) 26840c2022ecSYongqiang Yang ret = jbd2_journal_init_transaction_cache(); 2685470decc6SDave Kleikamp return ret; 2686470decc6SDave Kleikamp } 2687470decc6SDave Kleikamp 2688f7f4bccbSMingming Cao static void jbd2_journal_destroy_caches(void) 2689470decc6SDave Kleikamp { 2690f7f4bccbSMingming Cao jbd2_journal_destroy_revoke_caches(); 26914185a2acSYongqiang Yang jbd2_journal_destroy_journal_head_cache(); 2692f7f4bccbSMingming Cao jbd2_journal_destroy_handle_cache(); 26930c2022ecSYongqiang Yang jbd2_journal_destroy_transaction_cache(); 2694d2eecb03STheodore Ts'o jbd2_journal_destroy_slabs(); 2695470decc6SDave Kleikamp } 2696470decc6SDave Kleikamp 2697470decc6SDave Kleikamp static int __init journal_init(void) 2698470decc6SDave Kleikamp { 2699470decc6SDave Kleikamp int ret; 2700470decc6SDave Kleikamp 2701470decc6SDave Kleikamp BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024); 2702470decc6SDave Kleikamp 2703470decc6SDave Kleikamp ret = journal_init_caches(); 2704620de4e1SDuane Griffin if (ret == 0) { 27058e85fb3fSJohann Lombardi jbd2_create_jbd_stats_proc_entry(); 2706620de4e1SDuane Griffin } else { 2707620de4e1SDuane Griffin jbd2_journal_destroy_caches(); 2708620de4e1SDuane Griffin } 2709470decc6SDave Kleikamp return ret; 2710470decc6SDave Kleikamp } 2711470decc6SDave Kleikamp 2712470decc6SDave Kleikamp static void __exit journal_exit(void) 2713470decc6SDave Kleikamp { 2714e23291b9SJose R. Santos #ifdef CONFIG_JBD2_DEBUG 2715470decc6SDave Kleikamp int n = atomic_read(&nr_journal_heads); 2716470decc6SDave Kleikamp if (n) 271775685071SJan Kara printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n); 2718470decc6SDave Kleikamp #endif 27198e85fb3fSJohann Lombardi jbd2_remove_jbd_stats_proc_entry(); 2720f7f4bccbSMingming Cao jbd2_journal_destroy_caches(); 2721470decc6SDave Kleikamp } 2722470decc6SDave Kleikamp 2723470decc6SDave Kleikamp MODULE_LICENSE("GPL"); 2724470decc6SDave Kleikamp module_init(journal_init); 2725470decc6SDave Kleikamp module_exit(journal_exit); 2726470decc6SDave Kleikamp 2727