1ac27a0ecSDave Kleikamp /* 2617ba13bSMingming Cao * linux/fs/ext4/fsync.c 3ac27a0ecSDave Kleikamp * 4ac27a0ecSDave Kleikamp * Copyright (C) 1993 Stephen Tweedie (sct@redhat.com) 5ac27a0ecSDave Kleikamp * from 6ac27a0ecSDave Kleikamp * Copyright (C) 1992 Remy Card (card@masi.ibp.fr) 7ac27a0ecSDave Kleikamp * Laboratoire MASI - Institut Blaise Pascal 8ac27a0ecSDave Kleikamp * Universite Pierre et Marie Curie (Paris VI) 9ac27a0ecSDave Kleikamp * from 10ac27a0ecSDave Kleikamp * linux/fs/minix/truncate.c Copyright (C) 1991, 1992 Linus Torvalds 11ac27a0ecSDave Kleikamp * 12617ba13bSMingming Cao * ext4fs fsync primitive 13ac27a0ecSDave Kleikamp * 14ac27a0ecSDave Kleikamp * Big-endian to little-endian byte-swapping/bitmaps by 15ac27a0ecSDave Kleikamp * David S. Miller (davem@caip.rutgers.edu), 1995 16ac27a0ecSDave Kleikamp * 17ac27a0ecSDave Kleikamp * Removed unnecessary code duplication for little endian machines 18ac27a0ecSDave Kleikamp * and excessive __inline__s. 19ac27a0ecSDave Kleikamp * Andi Kleen, 1997 20ac27a0ecSDave Kleikamp * 21ac27a0ecSDave Kleikamp * Major simplications and cleanup - we only need to do the metadata, because 22ac27a0ecSDave Kleikamp * we can depend on generic_block_fdatasync() to sync the data blocks. 23ac27a0ecSDave Kleikamp */ 24ac27a0ecSDave Kleikamp 25ac27a0ecSDave Kleikamp #include <linux/time.h> 26ac27a0ecSDave Kleikamp #include <linux/fs.h> 27ac27a0ecSDave Kleikamp #include <linux/sched.h> 28ac27a0ecSDave Kleikamp #include <linux/writeback.h> 29dab291afSMingming Cao #include <linux/jbd2.h> 30d755fb38SEric Sandeen #include <linux/blkdev.h> 319bffad1eSTheodore Ts'o 323dcf5451SChristoph Hellwig #include "ext4.h" 333dcf5451SChristoph Hellwig #include "ext4_jbd2.h" 34ac27a0ecSDave Kleikamp 359bffad1eSTheodore Ts'o #include <trace/events/ext4.h> 369bffad1eSTheodore Ts'o 374a873a47STheodore Ts'o static void dump_completed_IO(struct inode * inode) 384a873a47STheodore Ts'o { 394a873a47STheodore Ts'o #ifdef EXT4_DEBUG 404a873a47STheodore Ts'o struct list_head *cur, *before, *after; 414a873a47STheodore Ts'o ext4_io_end_t *io, *io0, *io1; 424a873a47STheodore Ts'o unsigned long flags; 434a873a47STheodore Ts'o 444a873a47STheodore Ts'o if (list_empty(&EXT4_I(inode)->i_completed_io_list)){ 454a873a47STheodore Ts'o ext4_debug("inode %lu completed_io list is empty\n", inode->i_ino); 464a873a47STheodore Ts'o return; 474a873a47STheodore Ts'o } 484a873a47STheodore Ts'o 494a873a47STheodore Ts'o ext4_debug("Dump inode %lu completed_io list \n", inode->i_ino); 504a873a47STheodore Ts'o spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags); 514a873a47STheodore Ts'o list_for_each_entry(io, &EXT4_I(inode)->i_completed_io_list, list){ 524a873a47STheodore Ts'o cur = &io->list; 534a873a47STheodore Ts'o before = cur->prev; 544a873a47STheodore Ts'o io0 = container_of(before, ext4_io_end_t, list); 554a873a47STheodore Ts'o after = cur->next; 564a873a47STheodore Ts'o io1 = container_of(after, ext4_io_end_t, list); 574a873a47STheodore Ts'o 584a873a47STheodore Ts'o ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n", 594a873a47STheodore Ts'o io, inode->i_ino, io0, io1); 604a873a47STheodore Ts'o } 614a873a47STheodore Ts'o spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags); 624a873a47STheodore Ts'o #endif 634a873a47STheodore Ts'o } 644a873a47STheodore Ts'o 654a873a47STheodore Ts'o /* 664a873a47STheodore Ts'o * This function is called from ext4_sync_file(). 674a873a47STheodore Ts'o * 684a873a47STheodore Ts'o * When IO is completed, the work to convert unwritten extents to 694a873a47STheodore Ts'o * written is queued on workqueue but may not get immediately 704a873a47STheodore Ts'o * scheduled. When fsync is called, we need to ensure the 714a873a47STheodore Ts'o * conversion is complete before fsync returns. 724a873a47STheodore Ts'o * The inode keeps track of a list of pending/completed IO that 734a873a47STheodore Ts'o * might needs to do the conversion. This function walks through 744a873a47STheodore Ts'o * the list and convert the related unwritten extents for completed IO 754a873a47STheodore Ts'o * to written. 764a873a47STheodore Ts'o * The function return the number of pending IOs on success. 774a873a47STheodore Ts'o */ 783889fd57SJiaying Zhang extern int ext4_flush_completed_IO(struct inode *inode) 794a873a47STheodore Ts'o { 804a873a47STheodore Ts'o ext4_io_end_t *io; 814a873a47STheodore Ts'o struct ext4_inode_info *ei = EXT4_I(inode); 824a873a47STheodore Ts'o unsigned long flags; 834a873a47STheodore Ts'o int ret = 0; 844a873a47STheodore Ts'o int ret2 = 0; 854a873a47STheodore Ts'o 864a873a47STheodore Ts'o if (list_empty(&ei->i_completed_io_list)) 874a873a47STheodore Ts'o return ret; 884a873a47STheodore Ts'o 894a873a47STheodore Ts'o dump_completed_IO(inode); 904a873a47STheodore Ts'o spin_lock_irqsave(&ei->i_completed_io_lock, flags); 914a873a47STheodore Ts'o while (!list_empty(&ei->i_completed_io_list)){ 924a873a47STheodore Ts'o io = list_entry(ei->i_completed_io_list.next, 934a873a47STheodore Ts'o ext4_io_end_t, list); 944a873a47STheodore Ts'o /* 954a873a47STheodore Ts'o * Calling ext4_end_io_nolock() to convert completed 964a873a47STheodore Ts'o * IO to written. 974a873a47STheodore Ts'o * 984a873a47STheodore Ts'o * When ext4_sync_file() is called, run_queue() may already 994a873a47STheodore Ts'o * about to flush the work corresponding to this io structure. 1004a873a47STheodore Ts'o * It will be upset if it founds the io structure related 1014a873a47STheodore Ts'o * to the work-to-be schedule is freed. 1024a873a47STheodore Ts'o * 1034a873a47STheodore Ts'o * Thus we need to keep the io structure still valid here after 1044a873a47STheodore Ts'o * convertion finished. The io structure has a flag to 1054a873a47STheodore Ts'o * avoid double converting from both fsync and background work 1064a873a47STheodore Ts'o * queue work. 1074a873a47STheodore Ts'o */ 1084a873a47STheodore Ts'o spin_unlock_irqrestore(&ei->i_completed_io_lock, flags); 1094a873a47STheodore Ts'o ret = ext4_end_io_nolock(io); 1104a873a47STheodore Ts'o spin_lock_irqsave(&ei->i_completed_io_lock, flags); 1114a873a47STheodore Ts'o if (ret < 0) 1124a873a47STheodore Ts'o ret2 = ret; 1134a873a47STheodore Ts'o else 1144a873a47STheodore Ts'o list_del_init(&io->list); 1154a873a47STheodore Ts'o } 1164a873a47STheodore Ts'o spin_unlock_irqrestore(&ei->i_completed_io_lock, flags); 1174a873a47STheodore Ts'o return (ret2 < 0) ? ret2 : 0; 1184a873a47STheodore Ts'o } 1194a873a47STheodore Ts'o 120ac27a0ecSDave Kleikamp /* 12114ece102SFrank Mayhar * If we're not journaling and this is a just-created file, we have to 12214ece102SFrank Mayhar * sync our parent directory (if it was freshly created) since 12314ece102SFrank Mayhar * otherwise it will only be written by writeback, leaving a huge 12414ece102SFrank Mayhar * window during which a crash may lose the file. This may apply for 12514ece102SFrank Mayhar * the parent directory's parent as well, and so on recursively, if 12614ece102SFrank Mayhar * they are also freshly created. 12714ece102SFrank Mayhar */ 12814ece102SFrank Mayhar static void ext4_sync_parent(struct inode *inode) 12914ece102SFrank Mayhar { 13014ece102SFrank Mayhar struct dentry *dentry = NULL; 13114ece102SFrank Mayhar 13214ece102SFrank Mayhar while (inode && ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY)) { 13314ece102SFrank Mayhar ext4_clear_inode_state(inode, EXT4_STATE_NEWENTRY); 13414ece102SFrank Mayhar dentry = list_entry(inode->i_dentry.next, 13514ece102SFrank Mayhar struct dentry, d_alias); 13614ece102SFrank Mayhar if (!dentry || !dentry->d_parent || !dentry->d_parent->d_inode) 13714ece102SFrank Mayhar break; 13814ece102SFrank Mayhar inode = dentry->d_parent->d_inode; 13914ece102SFrank Mayhar sync_mapping_buffers(inode->i_mapping); 14014ece102SFrank Mayhar } 14114ece102SFrank Mayhar } 14214ece102SFrank Mayhar 14314ece102SFrank Mayhar /* 144617ba13bSMingming Cao * akpm: A new design for ext4_sync_file(). 145ac27a0ecSDave Kleikamp * 146ac27a0ecSDave Kleikamp * This is only called from sys_fsync(), sys_fdatasync() and sys_msync(). 147ac27a0ecSDave Kleikamp * There cannot be a transaction open by this task. 148ac27a0ecSDave Kleikamp * Another task could have dirtied this inode. Its data can be in any 149ac27a0ecSDave Kleikamp * state in the journalling system. 150ac27a0ecSDave Kleikamp * 151ac27a0ecSDave Kleikamp * What we do is just kick off a commit and wait on it. This will snapshot the 152ac27a0ecSDave Kleikamp * inode to disk. 1538d5d02e6SMingming Cao * 1548d5d02e6SMingming Cao * i_mutex lock is held when entering and exiting this function 155ac27a0ecSDave Kleikamp */ 156ac27a0ecSDave Kleikamp 1577ea80859SChristoph Hellwig int ext4_sync_file(struct file *file, int datasync) 158ac27a0ecSDave Kleikamp { 1597ea80859SChristoph Hellwig struct inode *inode = file->f_mapping->host; 160b436b9beSJan Kara struct ext4_inode_info *ei = EXT4_I(inode); 161d755fb38SEric Sandeen journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; 162b436b9beSJan Kara int ret; 163b436b9beSJan Kara tid_t commit_tid; 164ac27a0ecSDave Kleikamp 165ac39849dSAneesh Kumar K.V J_ASSERT(ext4_journal_current_handle() == NULL); 166ac27a0ecSDave Kleikamp 1670562e0baSJiaying Zhang trace_ext4_sync_file_enter(file, datasync); 168ede86cc4STheodore Ts'o 169b436b9beSJan Kara if (inode->i_sb->s_flags & MS_RDONLY) 170b436b9beSJan Kara return 0; 171b436b9beSJan Kara 1723889fd57SJiaying Zhang ret = ext4_flush_completed_IO(inode); 1738d5d02e6SMingming Cao if (ret < 0) 1740562e0baSJiaying Zhang goto out; 175b436b9beSJan Kara 17614ece102SFrank Mayhar if (!journal) { 1771b061d92SChristoph Hellwig ret = generic_file_fsync(file, datasync); 17814ece102SFrank Mayhar if (!ret && !list_empty(&inode->i_dentry)) 17914ece102SFrank Mayhar ext4_sync_parent(inode); 1800562e0baSJiaying Zhang goto out; 18114ece102SFrank Mayhar } 182b436b9beSJan Kara 183ac27a0ecSDave Kleikamp /* 184b436b9beSJan Kara * data=writeback,ordered: 185ac27a0ecSDave Kleikamp * The caller's filemap_fdatawrite()/wait will sync the data. 186b436b9beSJan Kara * Metadata is in the journal, we wait for proper transaction to 187b436b9beSJan Kara * commit here. 188ac27a0ecSDave Kleikamp * 189ac27a0ecSDave Kleikamp * data=journal: 190ac27a0ecSDave Kleikamp * filemap_fdatawrite won't do anything (the buffers are clean). 191617ba13bSMingming Cao * ext4_force_commit will write the file data into the journal and 192ac27a0ecSDave Kleikamp * will wait on that. 193ac27a0ecSDave Kleikamp * filemap_fdatawait() will encounter a ton of newly-dirtied pages 194ac27a0ecSDave Kleikamp * (they were dirtied by commit). But that's OK - the blocks are 195ac27a0ecSDave Kleikamp * safe in-journal, which is all fsync() needs to ensure. 196ac27a0ecSDave Kleikamp */ 1970562e0baSJiaying Zhang if (ext4_should_journal_data(inode)) { 1980562e0baSJiaying Zhang ret = ext4_force_commit(inode->i_sb); 1990562e0baSJiaying Zhang goto out; 2000562e0baSJiaying Zhang } 201ac27a0ecSDave Kleikamp 202b436b9beSJan Kara commit_tid = datasync ? ei->i_datasync_tid : ei->i_sync_tid; 203cc3e1beaSTheodore Ts'o if (jbd2_log_start_commit(journal, commit_tid)) { 204cc3e1beaSTheodore Ts'o /* 205cc3e1beaSTheodore Ts'o * When the journal is on a different device than the 206cc3e1beaSTheodore Ts'o * fs data disk, we need to issue the barrier in 207cc3e1beaSTheodore Ts'o * writeback mode. (In ordered mode, the jbd2 layer 208cc3e1beaSTheodore Ts'o * will take care of issuing the barrier. In 209cc3e1beaSTheodore Ts'o * data=journal, all of the data blocks are written to 210cc3e1beaSTheodore Ts'o * the journal device.) 211cc3e1beaSTheodore Ts'o */ 212cc3e1beaSTheodore Ts'o if (ext4_should_writeback_data(inode) && 213cc3e1beaSTheodore Ts'o (journal->j_fs_dev != journal->j_dev) && 214cc3e1beaSTheodore Ts'o (journal->j_flags & JBD2_BARRIER)) 215fbd9b09aSDmitry Monakhov blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, 216dd3932edSChristoph Hellwig NULL); 2170671e704SDmitry Monakhov ret = jbd2_log_wait_commit(journal, commit_tid); 218cc3e1beaSTheodore Ts'o } else if (journal->j_flags & JBD2_BARRIER) 219dd3932edSChristoph Hellwig blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL); 2200562e0baSJiaying Zhang out: 2210562e0baSJiaying Zhang trace_ext4_sync_file_exit(inode, ret); 222ac27a0ecSDave Kleikamp return ret; 223ac27a0ecSDave Kleikamp } 224