fs/ext4/fsync.c

   1 /*
   2  *  linux/fs/ext4/fsync.c
   3  *
   4  *  Copyright (C) 1993  Stephen Tweedie (sct@redhat.com)
   5  *  from
   6  *  Copyright (C) 1992  Remy Card (card@masi.ibp.fr)
   7  *                      Laboratoire MASI - Institut Blaise Pascal
   8  *                      Universite Pierre et Marie Curie (Paris VI)
   9  *  from
  10  *  linux/fs/minix/truncate.c   Copyright (C) 1991, 1992  Linus Torvalds
  11  *
  12  *  ext4fs fsync primitive
  13  *
  14  *  Big-endian to little-endian byte-swapping/bitmaps by
  15  *        David S. Miller (davem@caip.rutgers.edu), 1995
  16  *
  17  *  Removed unnecessary code duplication for little endian machines
  18  *  and excessive __inline__s.
  19  *        Andi Kleen, 1997
  20  *
  21  * Major simplications and cleanup - we only need to do the metadata, because
  22  * we can depend on generic_block_fdatasync() to sync the data blocks.
  23  */
  24
  25 #include <linux/time.h>
  26 #include <linux/fs.h>
  27 #include <linux/sched.h>
  28 #include <linux/writeback.h>
  29 #include <linux/blkdev.h>
  30
  31 #include "ext4.h"
  32 #include "ext4_jbd2.h"
  33
  34 #include <trace/events/ext4.h>
  35
  36 /*
  37  * If we're not journaling and this is a just-created file, we have to
  38  * sync our parent directory (if it was freshly created) since
  39  * otherwise it will only be written by writeback, leaving a huge
  40  * window during which a crash may lose the file.  This may apply for
  41  * the parent directory's parent as well, and so on recursively, if
  42  * they are also freshly created.
  43  */
  44 static int ext4_sync_parent(struct inode *inode)
  45 {
  46         struct dentry *dentry = NULL;
  47         struct inode *next;
  48         int ret = 0;
  49
  50         if (!ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY))
  51                 return 0;
  52         inode = igrab(inode);
  53         while (ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY)) {
  54                 ext4_clear_inode_state(inode, EXT4_STATE_NEWENTRY);
  55                 dentry = d_find_any_alias(inode);
  56                 if (!dentry)
  57                         break;
  58                 next = igrab(d_inode(dentry->d_parent));
  59                 dput(dentry);
  60                 if (!next)
  61                         break;
  62                 iput(inode);
  63                 inode = next;
  64                 ret = sync_mapping_buffers(inode->i_mapping);
  65                 if (ret)
  66                         break;
  67                 ret = sync_inode_metadata(inode, 1);
  68                 if (ret)
  69                         break;
  70         }
  71         iput(inode);
  72         return ret;
  73 }
  74
  75 /*
  76  * akpm: A new design for ext4_sync_file().
  77  *
  78  * This is only called from sys_fsync(), sys_fdatasync() and sys_msync().
  79  * There cannot be a transaction open by this task.
  80  * Another task could have dirtied this inode.  Its data can be in any
  81  * state in the journalling system.
  82  *
  83  * What we do is just kick off a commit and wait on it.  This will snapshot the
  84  * inode to disk.
  85  */
  86
  87 int ext4_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
  88 {
  89         struct inode *inode = file->f_mapping->host;
  90         struct ext4_inode_info *ei = EXT4_I(inode);
  91         journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
  92         int ret = 0, err;
  93         tid_t commit_tid;
  94         bool needs_barrier = false;
  95
  96         J_ASSERT(ext4_journal_current_handle() == NULL);
  97
  98         trace_ext4_sync_file_enter(file, datasync);
  99
 100         if (inode->i_sb->s_flags & MS_RDONLY) {
 101                 /* Make sure that we read updated s_mount_flags value */
 102                 smp_rmb();
 103                 if (EXT4_SB(inode->i_sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
 104                         ret = -EROFS;
 105                 goto out;
 106         }
 107
 108         if (!journal) {
 109                 ret = generic_file_fsync(file, start, end, datasync);
 110                 if (!ret && !hlist_empty(&inode->i_dentry))
 111                         ret = ext4_sync_parent(inode);
 112                 goto out;
 113         }
 114
 115         ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
 116         if (ret)
 117                 return ret;
 118         /*
 119          * data=writeback,ordered:
 120          *  The caller's filemap_fdatawrite()/wait will sync the data.
 121          *  Metadata is in the journal, we wait for proper transaction to
 122          *  commit here.
 123          *
 124          * data=journal:
 125          *  filemap_fdatawrite won't do anything (the buffers are clean).
 126          *  ext4_force_commit will write the file data into the journal and
 127          *  will wait on that.
 128          *  filemap_fdatawait() will encounter a ton of newly-dirtied pages
 129          *  (they were dirtied by commit).  But that's OK - the blocks are
 130          *  safe in-journal, which is all fsync() needs to ensure.
 131          */
 132         if (ext4_should_journal_data(inode)) {
 133                 ret = ext4_force_commit(inode->i_sb);
 134                 goto out;
 135         }
 136
 137         commit_tid = datasync ? ei->i_datasync_tid : ei->i_sync_tid;
 138         if (journal->j_flags & JBD2_BARRIER &&
 139             !jbd2_trans_will_send_data_barrier(journal, commit_tid))
 140                 needs_barrier = true;
 141         ret = jbd2_complete_transaction(journal, commit_tid);
 142         if (needs_barrier) {
 143                 err = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
 144                 if (!ret)
 145                         ret = err;
 146         }
 147 out:
 148         trace_ext4_sync_file_exit(inode, ret);
 149         return ret;
 150 }