journalling.rst (9176B)
1The Linux Journalling API 2========================= 3 4Overview 5-------- 6 7Details 8~~~~~~~ 9 10The journalling layer is easy to use. You need to first of all create a 11journal_t data structure. There are two calls to do this dependent on 12how you decide to allocate the physical media on which the journal 13resides. The jbd2_journal_init_inode() call is for journals stored in 14filesystem inodes, or the jbd2_journal_init_dev() call can be used 15for journal stored on a raw device (in a continuous range of blocks). A 16journal_t is a typedef for a struct pointer, so when you are finally 17finished make sure you call jbd2_journal_destroy() on it to free up 18any used kernel memory. 19 20Once you have got your journal_t object you need to 'mount' or load the 21journal file. The journalling layer expects the space for the journal 22was already allocated and initialized properly by the userspace tools. 23When loading the journal you must call jbd2_journal_load() to process 24journal contents. If the client file system detects the journal contents 25does not need to be processed (or even need not have valid contents), it 26may call jbd2_journal_wipe() to clear the journal contents before 27calling jbd2_journal_load(). 28 29Note that jbd2_journal_wipe(..,0) calls 30jbd2_journal_skip_recovery() for you if it detects any outstanding 31transactions in the journal and similarly jbd2_journal_load() will 32call jbd2_journal_recover() if necessary. I would advise reading 33ext4_load_journal() in fs/ext4/super.c for examples on this stage. 34 35Now you can go ahead and start modifying the underlying filesystem. 36Almost. 37 38You still need to actually journal your filesystem changes, this is done 39by wrapping them into transactions. Additionally you also need to wrap 40the modification of each of the buffers with calls to the journal layer, 41so it knows what the modifications you are actually making are. To do 42this use jbd2_journal_start() which returns a transaction handle. 43 44jbd2_journal_start() and its counterpart jbd2_journal_stop(), 45which indicates the end of a transaction are nestable calls, so you can 46reenter a transaction if necessary, but remember you must call 47jbd2_journal_stop() the same number of times as 48jbd2_journal_start() before the transaction is completed (or more 49accurately leaves the update phase). Ext4/VFS makes use of this feature to 50simplify handling of inode dirtying, quota support, etc. 51 52Inside each transaction you need to wrap the modifications to the 53individual buffers (blocks). Before you start to modify a buffer you 54need to call jbd2_journal_get_create_access() / 55jbd2_journal_get_write_access() / 56jbd2_journal_get_undo_access() as appropriate, this allows the 57journalling layer to copy the unmodified 58data if it needs to. After all the buffer may be part of a previously 59uncommitted transaction. At this point you are at last ready to modify a 60buffer, and once you are have done so you need to call 61jbd2_journal_dirty_metadata(). Or if you've asked for access to a 62buffer you now know is now longer required to be pushed back on the 63device you can call jbd2_journal_forget() in much the same way as you 64might have used bforget() in the past. 65 66A jbd2_journal_flush() may be called at any time to commit and 67checkpoint all your transactions. 68 69Then at umount time , in your put_super() you can then call 70jbd2_journal_destroy() to clean up your in-core journal object. 71 72Unfortunately there a couple of ways the journal layer can cause a 73deadlock. The first thing to note is that each task can only have a 74single outstanding transaction at any one time, remember nothing commits 75until the outermost jbd2_journal_stop(). This means you must complete 76the transaction at the end of each file/inode/address etc. operation you 77perform, so that the journalling system isn't re-entered on another 78journal. Since transactions can't be nested/batched across differing 79journals, and another filesystem other than yours (say ext4) may be 80modified in a later syscall. 81 82The second case to bear in mind is that jbd2_journal_start() can block 83if there isn't enough space in the journal for your transaction (based 84on the passed nblocks param) - when it blocks it merely(!) needs to wait 85for transactions to complete and be committed from other tasks, so 86essentially we are waiting for jbd2_journal_stop(). So to avoid 87deadlocks you must treat jbd2_journal_start() / 88jbd2_journal_stop() as if they were semaphores and include them in 89your semaphore ordering rules to prevent 90deadlocks. Note that jbd2_journal_extend() has similar blocking 91behaviour to jbd2_journal_start() so you can deadlock here just as 92easily as on jbd2_journal_start(). 93 94Try to reserve the right number of blocks the first time. ;-). This will 95be the maximum number of blocks you are going to touch in this 96transaction. I advise having a look at at least ext4_jbd.h to see the 97basis on which ext4 uses to make these decisions. 98 99Another wriggle to watch out for is your on-disk block allocation 100strategy. Why? Because, if you do a delete, you need to ensure you 101haven't reused any of the freed blocks until the transaction freeing 102these blocks commits. If you reused these blocks and crash happens, 103there is no way to restore the contents of the reallocated blocks at the 104end of the last fully committed transaction. One simple way of doing 105this is to mark blocks as free in internal in-memory block allocation 106structures only after the transaction freeing them commits. Ext4 uses 107journal commit callback for this purpose. 108 109With journal commit callbacks you can ask the journalling layer to call 110a callback function when the transaction is finally committed to disk, 111so that you can do some of your own management. You ask the journalling 112layer for calling the callback by simply setting 113``journal->j_commit_callback`` function pointer and that function is 114called after each transaction commit. You can also use 115``transaction->t_private_list`` for attaching entries to a transaction 116that need processing when the transaction commits. 117 118JBD2 also provides a way to block all transaction updates via 119jbd2_journal_lock_updates() / 120jbd2_journal_unlock_updates(). Ext4 uses this when it wants a 121window with a clean and stable fs for a moment. E.g. 122 123:: 124 125 126 jbd2_journal_lock_updates() //stop new stuff happening.. 127 jbd2_journal_flush() // checkpoint everything. 128 ..do stuff on stable fs 129 jbd2_journal_unlock_updates() // carry on with filesystem use. 130 131The opportunities for abuse and DOS attacks with this should be obvious, 132if you allow unprivileged userspace to trigger codepaths containing 133these calls. 134 135Fast commits 136~~~~~~~~~~~~ 137 138JBD2 to also allows you to perform file-system specific delta commits known as 139fast commits. In order to use fast commits, you will need to set following 140callbacks that perform correspodning work: 141 142`journal->j_fc_cleanup_cb`: Cleanup function called after every full commit and 143fast commit. 144 145`journal->j_fc_replay_cb`: Replay function called for replay of fast commit 146blocks. 147 148File system is free to perform fast commits as and when it wants as long as it 149gets permission from JBD2 to do so by calling the function 150:c:func:`jbd2_fc_begin_commit()`. Once a fast commit is done, the client 151file system should tell JBD2 about it by calling 152:c:func:`jbd2_fc_end_commit()`. If file system wants JBD2 to perform a full 153commit immediately after stopping the fast commit it can do so by calling 154:c:func:`jbd2_fc_end_commit_fallback()`. This is useful if fast commit operation 155fails for some reason and the only way to guarantee consistency is for JBD2 to 156perform the full traditional commit. 157 158JBD2 helper functions to manage fast commit buffers. File system can use 159:c:func:`jbd2_fc_get_buf()` and :c:func:`jbd2_fc_wait_bufs()` to allocate 160and wait on IO completion of fast commit buffers. 161 162Currently, only Ext4 implements fast commits. For details of its implementation 163of fast commits, please refer to the top level comments in 164fs/ext4/fast_commit.c. 165 166Summary 167~~~~~~~ 168 169Using the journal is a matter of wrapping the different context changes, 170being each mount, each modification (transaction) and each changed 171buffer to tell the journalling layer about them. 172 173Data Types 174---------- 175 176The journalling layer uses typedefs to 'hide' the concrete definitions 177of the structures used. As a client of the JBD2 layer you can just rely 178on the using the pointer as a magic cookie of some sort. Obviously the 179hiding is not enforced as this is 'C'. 180 181Structures 182~~~~~~~~~~ 183 184.. kernel-doc:: include/linux/jbd2.h 185 :internal: 186 187Functions 188--------- 189 190The functions here are split into two groups those that affect a journal 191as a whole, and those which are used to manage transactions 192 193Journal Level 194~~~~~~~~~~~~~ 195 196.. kernel-doc:: fs/jbd2/journal.c 197 :export: 198 199.. kernel-doc:: fs/jbd2/recovery.c 200 :internal: 201 202Transasction Level 203~~~~~~~~~~~~~~~~~~ 204 205.. kernel-doc:: fs/jbd2/transaction.c 206 207See also 208-------- 209 210`Journaling the Linux ext2fs Filesystem, LinuxExpo 98, Stephen 211Tweedie <http://kernel.org/pub/linux/kernel/people/sct/ext3/journal-design.ps.gz>`__ 212 213`Ext3 Journalling FileSystem, OLS 2000, Dr. Stephen 214Tweedie <http://olstrans.sourceforge.net/release/OLS2000-ext3/OLS2000-ext3.html>`__ 215