journalling.rst 9.0 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215
  1. The Linux Journalling API
  2. =========================
  3. Overview
  4. --------
  5. Details
  6. ~~~~~~~
  7. The journalling layer is easy to use. You need to first of all create a
  8. journal_t data structure. There are two calls to do this dependent on
  9. how you decide to allocate the physical media on which the journal
  10. resides. The jbd2_journal_init_inode() call is for journals stored in
  11. filesystem inodes, or the jbd2_journal_init_dev() call can be used
  12. for journal stored on a raw device (in a continuous range of blocks). A
  13. journal_t is a typedef for a struct pointer, so when you are finally
  14. finished make sure you call jbd2_journal_destroy() on it to free up
  15. any used kernel memory.
  16. Once you have got your journal_t object you need to 'mount' or load the
  17. journal file. The journalling layer expects the space for the journal
  18. was already allocated and initialized properly by the userspace tools.
  19. When loading the journal you must call jbd2_journal_load() to process
  20. journal contents. If the client file system detects the journal contents
  21. does not need to be processed (or even need not have valid contents), it
  22. may call jbd2_journal_wipe() to clear the journal contents before
  23. calling jbd2_journal_load().
  24. Note that jbd2_journal_wipe(..,0) calls
  25. jbd2_journal_skip_recovery() for you if it detects any outstanding
  26. transactions in the journal and similarly jbd2_journal_load() will
  27. call jbd2_journal_recover() if necessary. I would advise reading
  28. ext4_load_journal() in fs/ext4/super.c for examples on this stage.
  29. Now you can go ahead and start modifying the underlying filesystem.
  30. Almost.
  31. You still need to actually journal your filesystem changes, this is done
  32. by wrapping them into transactions. Additionally you also need to wrap
  33. the modification of each of the buffers with calls to the journal layer,
  34. so it knows what the modifications you are actually making are. To do
  35. this use jbd2_journal_start() which returns a transaction handle.
  36. jbd2_journal_start() and its counterpart jbd2_journal_stop(),
  37. which indicates the end of a transaction are nestable calls, so you can
  38. reenter a transaction if necessary, but remember you must call
  39. jbd2_journal_stop() the same number of times as
  40. jbd2_journal_start() before the transaction is completed (or more
  41. accurately leaves the update phase). Ext4/VFS makes use of this feature to
  42. simplify handling of inode dirtying, quota support, etc.
  43. Inside each transaction you need to wrap the modifications to the
  44. individual buffers (blocks). Before you start to modify a buffer you
  45. need to call jbd2_journal_get_create_access() /
  46. jbd2_journal_get_write_access() /
  47. jbd2_journal_get_undo_access() as appropriate, this allows the
  48. journalling layer to copy the unmodified
  49. data if it needs to. After all the buffer may be part of a previously
  50. uncommitted transaction. At this point you are at last ready to modify a
  51. buffer, and once you are have done so you need to call
  52. jbd2_journal_dirty_metadata(). Or if you've asked for access to a
  53. buffer you now know is now longer required to be pushed back on the
  54. device you can call jbd2_journal_forget() in much the same way as you
  55. might have used bforget() in the past.
  56. A jbd2_journal_flush() may be called at any time to commit and
  57. checkpoint all your transactions.
  58. Then at umount time , in your put_super() you can then call
  59. jbd2_journal_destroy() to clean up your in-core journal object.
  60. Unfortunately there a couple of ways the journal layer can cause a
  61. deadlock. The first thing to note is that each task can only have a
  62. single outstanding transaction at any one time, remember nothing commits
  63. until the outermost jbd2_journal_stop(). This means you must complete
  64. the transaction at the end of each file/inode/address etc. operation you
  65. perform, so that the journalling system isn't re-entered on another
  66. journal. Since transactions can't be nested/batched across differing
  67. journals, and another filesystem other than yours (say ext4) may be
  68. modified in a later syscall.
  69. The second case to bear in mind is that jbd2_journal_start() can block
  70. if there isn't enough space in the journal for your transaction (based
  71. on the passed nblocks param) - when it blocks it merely(!) needs to wait
  72. for transactions to complete and be committed from other tasks, so
  73. essentially we are waiting for jbd2_journal_stop(). So to avoid
  74. deadlocks you must treat jbd2_journal_start() /
  75. jbd2_journal_stop() as if they were semaphores and include them in
  76. your semaphore ordering rules to prevent
  77. deadlocks. Note that jbd2_journal_extend() has similar blocking
  78. behaviour to jbd2_journal_start() so you can deadlock here just as
  79. easily as on jbd2_journal_start().
  80. Try to reserve the right number of blocks the first time. ;-). This will
  81. be the maximum number of blocks you are going to touch in this
  82. transaction. I advise having a look at at least ext4_jbd.h to see the
  83. basis on which ext4 uses to make these decisions.
  84. Another wriggle to watch out for is your on-disk block allocation
  85. strategy. Why? Because, if you do a delete, you need to ensure you
  86. haven't reused any of the freed blocks until the transaction freeing
  87. these blocks commits. If you reused these blocks and crash happens,
  88. there is no way to restore the contents of the reallocated blocks at the
  89. end of the last fully committed transaction. One simple way of doing
  90. this is to mark blocks as free in internal in-memory block allocation
  91. structures only after the transaction freeing them commits. Ext4 uses
  92. journal commit callback for this purpose.
  93. With journal commit callbacks you can ask the journalling layer to call
  94. a callback function when the transaction is finally committed to disk,
  95. so that you can do some of your own management. You ask the journalling
  96. layer for calling the callback by simply setting
  97. ``journal->j_commit_callback`` function pointer and that function is
  98. called after each transaction commit. You can also use
  99. ``transaction->t_private_list`` for attaching entries to a transaction
  100. that need processing when the transaction commits.
  101. JBD2 also provides a way to block all transaction updates via
  102. jbd2_journal_lock_updates() /
  103. jbd2_journal_unlock_updates(). Ext4 uses this when it wants a
  104. window with a clean and stable fs for a moment. E.g.
  105. ::
  106. jbd2_journal_lock_updates() //stop new stuff happening..
  107. jbd2_journal_flush() // checkpoint everything.
  108. ..do stuff on stable fs
  109. jbd2_journal_unlock_updates() // carry on with filesystem use.
  110. The opportunities for abuse and DOS attacks with this should be obvious,
  111. if you allow unprivileged userspace to trigger codepaths containing
  112. these calls.
  113. Fast commits
  114. ~~~~~~~~~~~~
  115. JBD2 to also allows you to perform file-system specific delta commits known as
  116. fast commits. In order to use fast commits, you will need to set following
  117. callbacks that perform correspodning work:
  118. `journal->j_fc_cleanup_cb`: Cleanup function called after every full commit and
  119. fast commit.
  120. `journal->j_fc_replay_cb`: Replay function called for replay of fast commit
  121. blocks.
  122. File system is free to perform fast commits as and when it wants as long as it
  123. gets permission from JBD2 to do so by calling the function
  124. :c:func:`jbd2_fc_begin_commit()`. Once a fast commit is done, the client
  125. file system should tell JBD2 about it by calling
  126. :c:func:`jbd2_fc_end_commit()`. If file system wants JBD2 to perform a full
  127. commit immediately after stopping the fast commit it can do so by calling
  128. :c:func:`jbd2_fc_end_commit_fallback()`. This is useful if fast commit operation
  129. fails for some reason and the only way to guarantee consistency is for JBD2 to
  130. perform the full traditional commit.
  131. JBD2 helper functions to manage fast commit buffers. File system can use
  132. :c:func:`jbd2_fc_get_buf()` and :c:func:`jbd2_fc_wait_bufs()` to allocate
  133. and wait on IO completion of fast commit buffers.
  134. Currently, only Ext4 implements fast commits. For details of its implementation
  135. of fast commits, please refer to the top level comments in
  136. fs/ext4/fast_commit.c.
  137. Summary
  138. ~~~~~~~
  139. Using the journal is a matter of wrapping the different context changes,
  140. being each mount, each modification (transaction) and each changed
  141. buffer to tell the journalling layer about them.
  142. Data Types
  143. ----------
  144. The journalling layer uses typedefs to 'hide' the concrete definitions
  145. of the structures used. As a client of the JBD2 layer you can just rely
  146. on the using the pointer as a magic cookie of some sort. Obviously the
  147. hiding is not enforced as this is 'C'.
  148. Structures
  149. ~~~~~~~~~~
  150. .. kernel-doc:: include/linux/jbd2.h
  151. :internal:
  152. Functions
  153. ---------
  154. The functions here are split into two groups those that affect a journal
  155. as a whole, and those which are used to manage transactions
  156. Journal Level
  157. ~~~~~~~~~~~~~
  158. .. kernel-doc:: fs/jbd2/journal.c
  159. :export:
  160. .. kernel-doc:: fs/jbd2/recovery.c
  161. :internal:
  162. Transasction Level
  163. ~~~~~~~~~~~~~~~~~~
  164. .. kernel-doc:: fs/jbd2/transaction.c
  165. See also
  166. --------
  167. `Journaling the Linux ext2fs Filesystem, LinuxExpo 98, Stephen
  168. Tweedie <http://kernel.org/pub/linux/kernel/people/sct/ext3/journal-design.ps.gz>`__
  169. `Ext3 Journalling FileSystem, OLS 2000, Dr. Stephen
  170. Tweedie <http://olstrans.sourceforge.net/release/OLS2000-ext3/OLS2000-ext3.html>`__