fs/ext4/fsync.c

0001 // SPDX-License-Identifier: GPL-2.0
0002 /*
0003  *  linux/fs/ext4/fsync.c
0004  *
0005  *  Copyright (C) 1993  Stephen Tweedie (sct@redhat.com)
0006  *  from
0007  *  Copyright (C) 1992  Remy Card (card@masi.ibp.fr)
0008  *                      Laboratoire MASI - Institut Blaise Pascal
0009  *                      Universite Pierre et Marie Curie (Paris VI)
0010  *  from
0011  *  linux/fs/minix/truncate.c   Copyright (C) 1991, 1992  Linus Torvalds
0012  *
0013  *  ext4fs fsync primitive
0014  *
0015  *  Big-endian to little-endian byte-swapping/bitmaps by
0016  *        David S. Miller (davem@caip.rutgers.edu), 1995
0017  *
0018  *  Removed unnecessary code duplication for little endian machines
0019  *  and excessive __inline__s.
0020  *        Andi Kleen, 1997
0021  *
0022  * Major simplications and cleanup - we only need to do the metadata, because
0023  * we can depend on generic_block_fdatasync() to sync the data blocks.
0024  */
0025
0026 #include <linux/time.h>
0027 #include <linux/fs.h>
0028 #include <linux/sched.h>
0029 #include <linux/writeback.h>
0030 #include <linux/blkdev.h>
0031
0032 #include "ext4.h"
0033 #include "ext4_jbd2.h"
0034
0035 #include <trace/events/ext4.h>
0036
0037 /*
0038  * If we're not journaling and this is a just-created file, we have to
0039  * sync our parent directory (if it was freshly created) since
0040  * otherwise it will only be written by writeback, leaving a huge
0041  * window during which a crash may lose the file.  This may apply for
0042  * the parent directory's parent as well, and so on recursively, if
0043  * they are also freshly created.
0044  */
0045 static int ext4_sync_parent(struct inode *inode)
0046 {
0047     struct dentry *dentry, *next;
0048     int ret = 0;
0049
0050     if (!ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY))
0051         return 0;
0052     dentry = d_find_any_alias(inode);
0053     if (!dentry)
0054         return 0;
0055     while (ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY)) {
0056         ext4_clear_inode_state(inode, EXT4_STATE_NEWENTRY);
0057
0058         next = dget_parent(dentry);
0059         dput(dentry);
0060         dentry = next;
0061         inode = dentry->d_inode;
0062
0063         /*
0064          * The directory inode may have gone through rmdir by now. But
0065          * the inode itself and its blocks are still allocated (we hold
0066          * a reference to the inode via its dentry), so it didn't go
0067          * through ext4_evict_inode()) and so we are safe to flush
0068          * metadata blocks and the inode.
0069          */
0070         ret = sync_mapping_buffers(inode->i_mapping);
0071         if (ret)
0072             break;
0073         ret = sync_inode_metadata(inode, 1);
0074         if (ret)
0075             break;
0076     }
0077     dput(dentry);
0078     return ret;
0079 }
0080
0081 static int ext4_fsync_nojournal(struct inode *inode, bool datasync,
0082                 bool *needs_barrier)
0083 {
0084     int ret, err;
0085
0086     ret = sync_mapping_buffers(inode->i_mapping);
0087     if (!(inode->i_state & I_DIRTY_ALL))
0088         return ret;
0089     if (datasync && !(inode->i_state & I_DIRTY_DATASYNC))
0090         return ret;
0091
0092     err = sync_inode_metadata(inode, 1);
0093     if (!ret)
0094         ret = err;
0095
0096     if (!ret)
0097         ret = ext4_sync_parent(inode);
0098     if (test_opt(inode->i_sb, BARRIER))
0099         *needs_barrier = true;
0100
0101     return ret;
0102 }
0103
0104 static int ext4_fsync_journal(struct inode *inode, bool datasync,
0105                  bool *needs_barrier)
0106 {
0107     struct ext4_inode_info *ei = EXT4_I(inode);
0108     journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
0109     tid_t commit_tid = datasync ? ei->i_datasync_tid : ei->i_sync_tid;
0110
0111     if (journal->j_flags & JBD2_BARRIER &&
0112         !jbd2_trans_will_send_data_barrier(journal, commit_tid))
0113         *needs_barrier = true;
0114
0115     return ext4_fc_commit(journal, commit_tid);
0116 }
0117
0118 /*
0119  * akpm: A new design for ext4_sync_file().
0120  *
0121  * This is only called from sys_fsync(), sys_fdatasync() and sys_msync().
0122  * There cannot be a transaction open by this task.
0123  * Another task could have dirtied this inode.  Its data can be in any
0124  * state in the journalling system.
0125  *
0126  * What we do is just kick off a commit and wait on it.  This will snapshot the
0127  * inode to disk.
0128  */
0129 int ext4_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
0130 {
0131     int ret = 0, err;
0132     bool needs_barrier = false;
0133     struct inode *inode = file->f_mapping->host;
0134     struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
0135
0136     if (unlikely(ext4_forced_shutdown(sbi)))
0137         return -EIO;
0138
0139     ASSERT(ext4_journal_current_handle() == NULL);
0140
0141     trace_ext4_sync_file_enter(file, datasync);
0142
0143     if (sb_rdonly(inode->i_sb)) {
0144         /* Make sure that we read updated s_mount_flags value */
0145         smp_rmb();
0146         if (ext4_test_mount_flag(inode->i_sb, EXT4_MF_FS_ABORTED))
0147             ret = -EROFS;
0148         goto out;
0149     }
0150
0151     ret = file_write_and_wait_range(file, start, end);
0152     if (ret)
0153         goto out;
0154
0155     /*
0156      * data=writeback,ordered:
0157      *  The caller's filemap_fdatawrite()/wait will sync the data.
0158      *  Metadata is in the journal, we wait for proper transaction to
0159      *  commit here.
0160      *
0161      * data=journal:
0162      *  filemap_fdatawrite won't do anything (the buffers are clean).
0163      *  ext4_force_commit will write the file data into the journal and
0164      *  will wait on that.
0165      *  filemap_fdatawait() will encounter a ton of newly-dirtied pages
0166      *  (they were dirtied by commit).  But that's OK - the blocks are
0167      *  safe in-journal, which is all fsync() needs to ensure.
0168      */
0169     if (!sbi->s_journal)
0170         ret = ext4_fsync_nojournal(inode, datasync, &needs_barrier);
0171     else if (ext4_should_journal_data(inode))
0172         ret = ext4_force_commit(inode->i_sb);
0173     else
0174         ret = ext4_fsync_journal(inode, datasync, &needs_barrier);
0175
0176     if (needs_barrier) {
0177         err = blkdev_issue_flush(inode->i_sb->s_bdev);
0178         if (!ret)
0179             ret = err;
0180     }
0181 out:
0182     err = file_check_and_advance_wb_err(file);
0183     if (ret == 0)
0184         ret = err;
0185     trace_ext4_sync_file_exit(inode, ret);
0186     return ret;
0187 }