blob: 341aaa4ce481a5a9af1f85d17be35e83e2b8b075 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001<?xml version="1.0" encoding="UTF-8"?>
2<!DOCTYPE book PUBLIC "-//OASIS//DTD DocBook XML V4.1.2//EN"
3 "http://www.oasis-open.org/docbook/xml/4.1.2/docbookx.dtd" []>
4
5<book id="LinuxJBDAPI">
6 <bookinfo>
7 <title>The Linux Journalling API</title>
8 <authorgroup>
9 <author>
10 <firstname>Roger</firstname>
11 <surname>Gammans</surname>
12 <affiliation>
13 <address>
14 <email>rgammans@computer-surgery.co.uk</email>
15 </address>
16 </affiliation>
17 </author>
18 </authorgroup>
19
20 <authorgroup>
21 <author>
22 <firstname>Stephen</firstname>
23 <surname>Tweedie</surname>
24 <affiliation>
25 <address>
26 <email>sct@redhat.com</email>
27 </address>
28 </affiliation>
29 </author>
30 </authorgroup>
31
32 <copyright>
33 <year>2002</year>
34 <holder>Roger Gammans</holder>
35 </copyright>
36
37<legalnotice>
38 <para>
39 This documentation is free software; you can redistribute
40 it and/or modify it under the terms of the GNU General Public
41 License as published by the Free Software Foundation; either
42 version 2 of the License, or (at your option) any later
43 version.
44 </para>
45
46 <para>
47 This program is distributed in the hope that it will be
48 useful, but WITHOUT ANY WARRANTY; without even the implied
49 warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
50 See the GNU General Public License for more details.
51 </para>
52
53 <para>
54 You should have received a copy of the GNU General Public
55 License along with this program; if not, write to the Free
56 Software Foundation, Inc., 59 Temple Place, Suite 330, Boston,
57 MA 02111-1307 USA
58 </para>
59
60 <para>
61 For more details see the file COPYING in the source
62 distribution of Linux.
63 </para>
64 </legalnotice>
65 </bookinfo>
66
67<toc></toc>
68
69 <chapter id="Overview">
70 <title>Overview</title>
71 <sect1>
72 <title>Details</title>
73<para>
74The journalling layer is easy to use. You need to
75first of all create a journal_t data structure. There are
76two calls to do this dependent on how you decide to allocate the physical
77media on which the journal resides. The journal_init_inode() call
78is for journals stored in filesystem inodes, or the journal_init_dev()
79call can be use for journal stored on a raw device (in a continuous range
80of blocks). A journal_t is a typedef for a struct pointer, so when
81you are finally finished make sure you call journal_destroy() on it
82to free up any used kernel memory.
83</para>
84
85<para>
86Once you have got your journal_t object you need to 'mount' or load the journal
87file, unless of course you haven't initialised it yet - in which case you
88need to call journal_create().
89</para>
90
91<para>
92Most of the time however your journal file will already have been created, but
93before you load it you must call journal_wipe() to empty the journal file.
94Hang on, you say , what if the filesystem wasn't cleanly umount()'d . Well, it is the
95job of the client file system to detect this and skip the call to journal_wipe().
96</para>
97
98<para>
99In either case the next call should be to journal_load() which prepares the
100journal file for use. Note that journal_wipe(..,0) calls journal_skip_recovery()
101for you if it detects any outstanding transactions in the journal and similarly
102journal_load() will call journal_recover() if necessary.
103I would advise reading fs/ext3/super.c for examples on this stage.
104[RGG: Why is the journal_wipe() call necessary - doesn't this needlessly
105complicate the API. Or isn't a good idea for the journal layer to hide
106dirty mounts from the client fs]
107</para>
108
109<para>
110Now you can go ahead and start modifying the underlying
111filesystem. Almost.
112</para>
113
114
115<para>
116
117You still need to actually journal your filesystem changes, this
118is done by wrapping them into transactions. Additionally you
Tobias Klauserd533f672005-09-10 00:26:46 -0700119also need to wrap the modification of each of the buffers
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120with calls to the journal layer, so it knows what the modifications
121you are actually making are. To do this use journal_start() which
122returns a transaction handle.
123</para>
124
125<para>
126journal_start()
127and its counterpart journal_stop(), which indicates the end of a transaction
128are nestable calls, so you can reenter a transaction if necessary,
129but remember you must call journal_stop() the same number of times as
130journal_start() before the transaction is completed (or more accurately
Tobias Klauserd533f672005-09-10 00:26:46 -0700131leaves the update phase). Ext3/VFS makes use of this feature to simplify
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132quota support.
133</para>
134
135<para>
136Inside each transaction you need to wrap the modifications to the
137individual buffers (blocks). Before you start to modify a buffer you
138need to call journal_get_{create,write,undo}_access() as appropriate,
139this allows the journalling layer to copy the unmodified data if it
140needs to. After all the buffer may be part of a previously uncommitted
141transaction.
142At this point you are at last ready to modify a buffer, and once
143you are have done so you need to call journal_dirty_{meta,}data().
144Or if you've asked for access to a buffer you now know is now longer
145required to be pushed back on the device you can call journal_forget()
146in much the same way as you might have used bforget() in the past.
147</para>
148
149<para>
150A journal_flush() may be called at any time to commit and checkpoint
151all your transactions.
152</para>
153
154<para>
155Then at umount time , in your put_super() (2.4) or write_super() (2.5)
156you can then call journal_destroy() to clean up your in-core journal object.
157</para>
158
159
160<para>
161Unfortunately there a couple of ways the journal layer can cause a deadlock.
162The first thing to note is that each task can only have
163a single outstanding transaction at any one time, remember nothing
164commits until the outermost journal_stop(). This means
165you must complete the transaction at the end of each file/inode/address
166etc. operation you perform, so that the journalling system isn't re-entered
167on another journal. Since transactions can't be nested/batched
168across differing journals, and another filesystem other than
169yours (say ext3) may be modified in a later syscall.
170</para>
171
172<para>
173The second case to bear in mind is that journal_start() can
174block 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
176wait for transactions to complete and be committed from other tasks,
177so essentially we are waiting for journal_stop(). So to avoid
178deadlocks you must treat journal_start/stop() as if they
179were semaphores and include them in your semaphore ordering rules to prevent
180deadlocks. Note that journal_extend() has similar blocking behaviour to
181journal_start() so you can deadlock here just as easily as on journal_start().
182</para>
183
184<para>
185Try to reserve the right number of blocks the first time. ;-). This will
186be the maximum number of blocks you are going to touch in this transaction.
187I advise having a look at at least ext3_jbd.h to see the basis on which
188ext3 uses to make these decisions.
189</para>
190
191<para>
192Another wriggle to watch out for is your on-disk block allocation strategy.
193why? Because, if you undo a delete, you need to ensure you haven't reused any
194of the freed blocks in a later transaction. One simple way of doing this
195is make sure any blocks you allocate only have checkpointed transactions
196listed against them. Ext3 does this in ext3_test_allocatable().
197</para>
198
199<para>
200Lock is also providing through journal_{un,}lock_updates(),
201ext3 uses this when it wants a window with a clean and stable fs for a moment.
202eg.
203</para>
204
205<programlisting>
206
207 journal_lock_updates() //stop new stuff happening..
208 journal_flush() // checkpoint everything.
209 ..do stuff on stable fs
210 journal_unlock_updates() // carry on with filesystem use.
211</programlisting>
212
213<para>
214The opportunities for abuse and DOS attacks with this should be obvious,
215if you allow unprivileged userspace to trigger codepaths containing these
216calls.
217</para>
218
219<para>
220A new feature of jbd since 2.5.25 is commit callbacks with the new
221journal_callback_set() function you can now ask the journalling layer
222to call you back when the transaction is finally committed to disk, so that
223you can do some of your own management. The key to this is the journal_callback
224struct, this maintains the internal callback information but you can
225extend it like this:-
226</para>
227<programlisting>
228 struct myfs_callback_s {
229 //Data structure element required by jbd..
230 struct journal_callback for_jbd;
231 // Stuff for myfs allocated together.
232 myfs_inode* i_commited;
233
234 }
235</programlisting>
236
237<para>
238this would be useful if you needed to know when data was committed to a
239particular inode.
240</para>
241
242</sect1>
243
244<sect1>
245<title>Summary</title>
246<para>
247Using the journal is a matter of wrapping the different context changes,
248being each mount, each modification (transaction) and each changed buffer
249to tell the journalling layer about them.
250</para>
251
252<para>
253Here is a some pseudo code to give you an idea of how it works, as
254an example.
255</para>
256
257<programlisting>
258 journal_t* my_jnrl = journal_create();
259 journal_init_{dev,inode}(jnrl,...)
260 if (clean) journal_wipe();
261 journal_load();
262
263 foreach(transaction) { /*transactions must be
264 completed before
265 a syscall returns to
266 userspace*/
267
268 handle_t * xct=journal_start(my_jnrl);
269 foreach(bh) {
270 journal_get_{create,write,undo}_access(xact,bh);
271 if ( myfs_modify(bh) ) { /* returns true
272 if makes changes */
273 journal_dirty_{meta,}data(xact,bh);
274 } else {
275 journal_forget(bh);
276 }
277 }
278 journal_stop(xct);
279 }
280 journal_destroy(my_jrnl);
281</programlisting>
282</sect1>
283
284</chapter>
285
286 <chapter id="adt">
287 <title>Data Types</title>
288 <para>
289 The journalling layer uses typedefs to 'hide' the concrete definitions
290 of the structures used. As a client of the JBD layer you can
291 just rely on the using the pointer as a magic cookie of some sort.
292
293 Obviously the hiding is not enforced as this is 'C'.
294 </para>
295 <sect1><title>Structures</title>
296!Iinclude/linux/jbd.h
297 </sect1>
298</chapter>
299
300 <chapter id="calls">
301 <title>Functions</title>
302 <para>
303 The functions here are split into two groups those that
304 affect a journal as a whole, and those which are used to
305 manage transactions
306</para>
307 <sect1><title>Journal Level</title>
308!Efs/jbd/journal.c
309!Efs/jbd/recovery.c
310 </sect1>
311 <sect1><title>Transasction Level</title>
312!Efs/jbd/transaction.c
313 </sect1>
314</chapter>
315<chapter>
316 <title>See also</title>
317 <para>
318 <citation>
319 <ulink url="ftp://ftp.uk.linux.org/pub/linux/sct/fs/jfs/journal-design.ps.gz">
320 Journaling the Linux ext2fs Filesystem,LinuxExpo 98, Stephen Tweedie
321 </ulink>
322 </citation>
323 </para>
324 <para>
325 <citation>
326 <ulink url="http://olstrans.sourceforge.net/release/OLS2000-ext3/OLS2000-ext3.html">
327 Ext3 Journalling FileSystem , OLS 2000, Dr. Stephen Tweedie
328 </ulink>
329 </citation>
330 </para>
331</chapter>
332
333</book>