Linux Kernel Architecture

Chapter 17: Data Synchronization

17.10 Inode Synchronization

writeback_inodeswrites back installed mappings by walking through the system inodes (for the sake

of simplicity, this is calledinode writeback, but in fact not the inode but the dirty data associated with it

are written back). The function shoulders the mainburden of synchronizationbecause most system data

are provided in the form of address space mappings that make use of inodes. Figure 17-6 illustrates the

code flow diagram forwriteback_inodes. The function is slightly more complicated in reality because

some more details and corner cases need to be handled properly. We consider a simplified variant that

nevertheless contains everything that isessential when inodes are written back.

Writeback limit reached?

Iterate oversb->s_ioFailed to write pages? Move to s_dirty

Iterate over allsuperblocks

Fill I/O list

Perform checks

Writeback limit reached?

writeback_inodes

sync_sb_inodes

__writeback_single_inode

break

break

Figure 17-6: Code flow diagram forwriteback_inodes.

The function uses the data structures discussed in Chapter 8 to establish a link among superblocks,

inodes, and associated data.

17.10.1 Walking the Superblocks

When mappings are written back inode-by-inode, the initial path taken is via all system superblock

instances that represent the mounted filesystems.sync_sb_inodesis invoked for each instance in order

to write back the superblock inode data, as shown in the code flow diagram in Figure 17-6. Walking the

superblock list can be terminated by two different conditions:

1. All superblock instances have been scanned sequentially. The kernel has reached the end of
   the list, and its work is therefore done.


2. The maximum number of writeback pages specified by thewriteback_controlinstance
   has been reached. Since writeback requires obtaining various important locks, the system
   should not be disturbed for too long to make the inodes available for other parts of the
   kernel again.

17.10.2 Examining Superblock Inodes

Once it has been established with the help of the superblock structure that the filesystem contains inodes

with dirty data, the kernel hands over tosync_sb_inodes, which synchronizes the dirty superblock

inodes. The code flow diagram is in Figure 17-6.