1 <!DOCTYPE book PUBLIC "-//OASIS//DTD DocBook V3.1//EN"[]>
2 <book id="LinuxJBDAPI">
4 <title>The Linux Journalling API</title>
7 <firstname>Roger</firstname>
8 <surname>Gammans</surname>
11 <email>rgammans@computer-surgery.co.uk</email>
19 <firstname>Stephen</firstname>
20 <surname>Tweedie</surname>
23 <email>sct@redhat.com</email>
31 <holder>Roger Gammans</holder>
36 This documentation is free software; you can redistribute
37 it and/or modify it under the terms of the GNU General Public
38 License as published by the Free Software Foundation; either
39 version 2 of the License, or (at your option) any later
44 This program is distributed in the hope that it will be
45 useful, but WITHOUT ANY WARRANTY; without even the implied
46 warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
47 See the GNU General Public License for more details.
51 You should have received a copy of the GNU General Public
52 License along with this program; if not, write to the Free
53 Software Foundation, Inc., 59 Temple Place, Suite 330, Boston,
58 For more details see the file COPYING in the source
59 distribution of Linux.
66 <chapter id="Overview">
67 <title>Overview</title>
69 <title>Details</title>
71 The journalling layer is easy to use. You need to
72 first of all create a journal_t data structure. There are
73 two calls to do this dependent on how you decide to allocate the physical
74 media on which the journal resides. The journal_init_inode() call
75 is for journals stored in filesystem inodes, or the journal_init_dev()
76 call can be use for journal stored on a raw device (in a continuous range
77 of blocks). A journal_t is a typedef for a struct pointer, so when
78 you are finally finished make sure you call journal_destroy() on it
79 to free up any used kernel memory.
83 Once you have got your journal_t object you need to 'mount' or load the journal
84 file, unless of course you haven't initialised it yet - in which case you
85 need to call journal_create().
89 Most of the time however your journal file will already have been created, but
90 before you load it you must call journal_wipe() to empty the journal file.
91 Hang on, you say , what if the filesystem wasn't cleanly umount()'d . Well, it is the
92 job of the client file system to detect this and skip the call to journal_wipe().
96 In either case the next call should be to journal_load() which prepares the
97 journal file for use. Note that journal_wipe(..,0) calls journal_skip_recovery()
98 for you if it detects any outstanding transactions in the journal and similarly
99 journal_load() will call journal_recover() if necessary.
100 I would advise reading fs/ext3/super.c for examples on this stage.
101 [RGG: Why is the journal_wipe() call necessary - doesn't this needlessly
102 complicate the API. Or isn't a good idea for the journal layer to hide
103 dirty mounts from the client fs]
107 Now you can go ahead and start modifying the underlying
114 You still need to actually journal your filesystem changes, this
115 is done by wrapping them into transactions. Additionally you
116 also need to wrap the modification of each of the the buffers
117 with calls to the journal layer, so it knows what the modifications
118 you are actually making are. To do this use journal_start() which
119 returns a transaction handle.
124 and its counterpart journal_stop(), which indicates the end of a transaction
125 are nestable calls, so you can reenter a transaction if necessary,
126 but remember you must call journal_stop() the same number of times as
127 journal_start() before the transaction is completed (or more accurately
128 leaves the the update phase). Ext3/VFS makes use of this feature to simplify
133 Inside each transaction you need to wrap the modifications to the
134 individual buffers (blocks). Before you start to modify a buffer you
135 need to call journal_get_{create,write,undo}_access() as appropriate,
136 this allows the journalling layer to copy the unmodified data if it
137 needs to. After all the buffer may be part of a previously uncommitted
139 At this point you are at last ready to modify a buffer, and once
140 you are have done so you need to call journal_dirty_{meta,}data().
141 Or if you've asked for access to a buffer you now know is now longer
142 required to be pushed back on the device you can call journal_forget()
143 in much the same way as you might have used bforget() in the past.
150 A journal_flush() may be called at any time to commit and checkpoint
151 all your transactions.
155 Then at umount time , in your put_super() (2.4) or write_super() (2.5)
156 you can then call journal_destroy() to clean up your in-core journal object.
161 Unfortunately there a couple of ways the journal layer can cause a deadlock.
162 The first thing to note is that each task can only have
163 a single outstanding transaction at any one time, remember nothing
164 commits until the outermost journal_stop(). This means
165 you must complete the transaction at the end of each file/inode/address
166 etc. operation you perform, so that the journalling system isn't re-entered
167 on another journal. Since transactions can't be nested/batched
168 across differing journals, and another filesystem other than
169 yours (say ext3) may be modified in a later syscall.
173 The second case to bear in mind is that journal_start() can
174 block if there isn't enough space in the journal for your transaction
175 (based on the passed nblocks param) - when it blocks it merely(!) needs to
176 wait for transactions to complete and be committed from other tasks,
177 so essentially we are waiting for journal_stop(). So to avoid
178 deadlocks you must treat journal_start/stop() as if they
179 were semaphores and include them in your semaphore ordering rules to prevent
180 deadlocks. Note that journal_extend() has similar blocking behaviour to
181 journal_start() so you can deadlock here just as easily as on journal_start().
185 Try to reserve the right number of blocks the first time. ;-).
188 Another wriggle to watch out for is your on-disk block allocation strategy.
189 why? Because, if you undo a delete, you need to ensure you haven't reused any
190 of the freed blocks in a later transaction. One simple way of doing this
191 is make sure any blocks you allocate only have checkpointed transactions
192 listed against them. Ext3 does this in ext3_test_allocatable().
196 Lock is also providing through journal_{un,}lock_updates(),
197 ext3 uses this when it wants a window with a clean and stable fs for a moment.
203 journal_lock_updates() //stop new stuff happening..
204 journal_flush() // checkpoint everything.
205 ..do stuff on stable fs
206 journal_unlock_updates() // carry on with filesystem use.
210 The opportunities for abuse and DOS attacks with this should be obvious,
211 if you allow unprivileged userspace to trigger codepaths containing these
217 <title>Summary</title>
219 Using the journal is a matter of wrapping the different context changes,
220 being each mount, each modification (transaction) and each changed buffer
221 to tell the journalling layer about them.
225 Here is a some pseudo code to give you an idea of how it works, as
230 journal_t* my_jnrl = journal_create();
231 journal_init_{dev,inode}(jnrl,...)
232 if (clean) journal_wipe();
235 foreach(transaction) { /*transactions must be
240 handle_t * xct=journal_start(my_jnrl);
242 journal_get_{create,write,undo}_access(xact,bh);
243 if ( myfs_modify(bh) ) { /* returns true
245 journal_dirty_{meta,}data(xact,bh);
252 journal_destroy(my_jrnl);
259 <title>Data Types</title>
261 The journalling layer uses typedefs to 'hide' the concrete definitions
262 of the structures used. As a client of the JBD layer you can
263 just rely on the using the pointer as a magic cookie of some sort.
265 Obviously the hiding is not enforced as this is 'C'.
267 <sect1><title>Structures</title>
268 !Iinclude/linux/jbd.h
273 <title>Functions</title>
275 The functions here are split into two groups those that
276 affect a journal as a whole, and those which are used to
279 <sect1><title>Journal Level</title>
283 <sect1><title>Transasction Level</title>
284 !Efs/jbd/transaction.c
288 <title>See also</title>
291 <ulink url="ftp://ftp.uk.linux.org/pub/linux/sct/fs/jfs/journal-design.ps.gz">
292 Journaling the Linux ext2fs Filesystem,LinuxExpo 98, Stephen Tweedie
298 <ulink url="http://olstrans.sourceforge.net/release/OLS2000-ext3/OLS2000-ext3.html">
299 Ext3 Journalling FileSystem , OLS 2000, Dr. Stephen Tweedie