|  | <?xml version="1.0" encoding="UTF-8"?> | 
|  | <!DOCTYPE book PUBLIC "-//OASIS//DTD DocBook XML V4.1.2//EN" | 
|  | "http://www.oasis-open.org/docbook/xml/4.1.2/docbookx.dtd" []> | 
|  |  | 
|  | <book id="Linux-filesystems-API"> | 
|  | <bookinfo> | 
|  | <title>Linux Filesystems API</title> | 
|  |  | 
|  | <legalnotice> | 
|  | <para> | 
|  | This documentation is free software; you can redistribute | 
|  | it and/or modify it under the terms of the GNU General Public | 
|  | License as published by the Free Software Foundation; either | 
|  | version 2 of the License, or (at your option) any later | 
|  | version. | 
|  | </para> | 
|  |  | 
|  | <para> | 
|  | This program is distributed in the hope that it will be | 
|  | useful, but WITHOUT ANY WARRANTY; without even the implied | 
|  | warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. | 
|  | See the GNU General Public License for more details. | 
|  | </para> | 
|  |  | 
|  | <para> | 
|  | You should have received a copy of the GNU General Public | 
|  | License along with this program; if not, write to the Free | 
|  | Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, | 
|  | MA 02111-1307 USA | 
|  | </para> | 
|  |  | 
|  | <para> | 
|  | For more details see the file COPYING in the source | 
|  | distribution of Linux. | 
|  | </para> | 
|  | </legalnotice> | 
|  | </bookinfo> | 
|  |  | 
|  | <toc></toc> | 
|  |  | 
|  | <chapter id="vfs"> | 
|  | <title>The Linux VFS</title> | 
|  | <sect1 id="the_filesystem_types"><title>The Filesystem types</title> | 
|  | !Iinclude/linux/fs.h | 
|  | </sect1> | 
|  | <sect1 id="the_directory_cache"><title>The Directory Cache</title> | 
|  | !Efs/dcache.c | 
|  | !Iinclude/linux/dcache.h | 
|  | </sect1> | 
|  | <sect1 id="inode_handling"><title>Inode Handling</title> | 
|  | !Efs/inode.c | 
|  | !Efs/bad_inode.c | 
|  | </sect1> | 
|  | <sect1 id="registration_and_superblocks"><title>Registration and Superblocks</title> | 
|  | !Efs/super.c | 
|  | </sect1> | 
|  | <sect1 id="file_locks"><title>File Locks</title> | 
|  | !Efs/locks.c | 
|  | !Ifs/locks.c | 
|  | </sect1> | 
|  | <sect1 id="other_functions"><title>Other Functions</title> | 
|  | !Efs/mpage.c | 
|  | !Efs/namei.c | 
|  | !Efs/buffer.c | 
|  | !Efs/bio.c | 
|  | !Efs/seq_file.c | 
|  | !Efs/filesystems.c | 
|  | !Efs/fs-writeback.c | 
|  | !Efs/block_dev.c | 
|  | </sect1> | 
|  | </chapter> | 
|  |  | 
|  | <chapter id="proc"> | 
|  | <title>The proc filesystem</title> | 
|  |  | 
|  | <sect1 id="sysctl_interface"><title>sysctl interface</title> | 
|  | !Ekernel/sysctl.c | 
|  | </sect1> | 
|  |  | 
|  | <sect1 id="proc_filesystem_interface"><title>proc filesystem interface</title> | 
|  | !Ifs/proc/base.c | 
|  | </sect1> | 
|  | </chapter> | 
|  |  | 
|  | <chapter id="fs_events"> | 
|  | <title>Events based on file descriptors</title> | 
|  | !Efs/eventfd.c | 
|  | </chapter> | 
|  |  | 
|  | <chapter id="sysfs"> | 
|  | <title>The Filesystem for Exporting Kernel Objects</title> | 
|  | !Efs/sysfs/file.c | 
|  | !Efs/sysfs/symlink.c | 
|  | </chapter> | 
|  |  | 
|  | <chapter id="debugfs"> | 
|  | <title>The debugfs filesystem</title> | 
|  |  | 
|  | <sect1 id="debugfs_interface"><title>debugfs interface</title> | 
|  | !Efs/debugfs/inode.c | 
|  | !Efs/debugfs/file.c | 
|  | </sect1> | 
|  | </chapter> | 
|  |  | 
|  | <chapter id="LinuxJDBAPI"> | 
|  | <chapterinfo> | 
|  | <title>The Linux Journalling API</title> | 
|  |  | 
|  | <authorgroup> | 
|  | <author> | 
|  | <firstname>Roger</firstname> | 
|  | <surname>Gammans</surname> | 
|  | <affiliation> | 
|  | <address> | 
|  | <email>rgammans@computer-surgery.co.uk</email> | 
|  | </address> | 
|  | </affiliation> | 
|  | </author> | 
|  | </authorgroup> | 
|  |  | 
|  | <authorgroup> | 
|  | <author> | 
|  | <firstname>Stephen</firstname> | 
|  | <surname>Tweedie</surname> | 
|  | <affiliation> | 
|  | <address> | 
|  | <email>sct@redhat.com</email> | 
|  | </address> | 
|  | </affiliation> | 
|  | </author> | 
|  | </authorgroup> | 
|  |  | 
|  | <copyright> | 
|  | <year>2002</year> | 
|  | <holder>Roger Gammans</holder> | 
|  | </copyright> | 
|  | </chapterinfo> | 
|  |  | 
|  | <title>The Linux Journalling API</title> | 
|  |  | 
|  | <sect1 id="journaling_overview"> | 
|  | <title>Overview</title> | 
|  | <sect2 id="journaling_details"> | 
|  | <title>Details</title> | 
|  | <para> | 
|  | The journalling layer is  easy to use. You need to | 
|  | first of all create a journal_t data structure. There are | 
|  | two calls to do this dependent on how you decide to allocate the physical | 
|  | media on which the journal resides. The journal_init_inode() call | 
|  | is for journals stored in filesystem inodes, or the journal_init_dev() | 
|  | call can be use for journal stored on a raw device (in a continuous range | 
|  | of blocks). A journal_t is a typedef for a struct pointer, so when | 
|  | you are finally finished make sure you call journal_destroy() on it | 
|  | to free up any used kernel memory. | 
|  | </para> | 
|  |  | 
|  | <para> | 
|  | Once you have got your journal_t object you need to 'mount' or load the journal | 
|  | file, unless of course you haven't initialised it yet - in which case you | 
|  | need to call journal_create(). | 
|  | </para> | 
|  |  | 
|  | <para> | 
|  | Most of the time however your journal file will already have been created, but | 
|  | before you load it you must call journal_wipe() to empty the journal file. | 
|  | Hang on, you say , what if the filesystem wasn't cleanly umount()'d . Well, it is the | 
|  | job of the client file system to detect this and skip the call to journal_wipe(). | 
|  | </para> | 
|  |  | 
|  | <para> | 
|  | In either case the next call should be to journal_load() which prepares the | 
|  | journal file for use. Note that journal_wipe(..,0) calls journal_skip_recovery() | 
|  | for you if it detects any outstanding transactions in the journal and similarly | 
|  | journal_load() will call journal_recover() if necessary. | 
|  | I would advise reading fs/ext3/super.c for examples on this stage. | 
|  | [RGG: Why is the journal_wipe() call necessary - doesn't this needlessly | 
|  | complicate the API. Or isn't a good idea for the journal layer to hide | 
|  | dirty mounts from the client fs] | 
|  | </para> | 
|  |  | 
|  | <para> | 
|  | Now you can go ahead and start modifying the underlying | 
|  | filesystem. Almost. | 
|  | </para> | 
|  |  | 
|  | <para> | 
|  |  | 
|  | You still need to actually journal your filesystem changes, this | 
|  | is done by wrapping them into transactions. Additionally you | 
|  | also need to wrap the modification of each of the buffers | 
|  | with calls to the journal layer, so it knows what the modifications | 
|  | you are actually making are. To do this use  journal_start() which | 
|  | returns a transaction handle. | 
|  | </para> | 
|  |  | 
|  | <para> | 
|  | journal_start() | 
|  | and its counterpart journal_stop(), which indicates the end of a transaction | 
|  | are nestable calls, so you can reenter a transaction if necessary, | 
|  | but remember you must call journal_stop() the same number of times as | 
|  | journal_start() before the transaction is completed (or more accurately | 
|  | leaves the update phase). Ext3/VFS makes use of this feature to simplify | 
|  | quota support. | 
|  | </para> | 
|  |  | 
|  | <para> | 
|  | Inside each transaction you need to wrap the modifications to the | 
|  | individual buffers (blocks). Before you start to modify a buffer you | 
|  | need to call journal_get_{create,write,undo}_access() as appropriate, | 
|  | this allows the journalling layer to copy the unmodified data if it | 
|  | needs to. After all the buffer may be part of a previously uncommitted | 
|  | transaction. | 
|  | At this point you are at last ready to modify a buffer, and once | 
|  | you are have done so you need to call journal_dirty_{meta,}data(). | 
|  | Or if you've asked for access to a buffer you now know is now longer | 
|  | required to be pushed back on the device you can call journal_forget() | 
|  | in much the same way as you might have used bforget() in the past. | 
|  | </para> | 
|  |  | 
|  | <para> | 
|  | A journal_flush() may be called at any time to commit and checkpoint | 
|  | all your transactions. | 
|  | </para> | 
|  |  | 
|  | <para> | 
|  | Then at umount time , in your put_super() you can then call journal_destroy() | 
|  | to clean up your in-core journal object. | 
|  | </para> | 
|  |  | 
|  | <para> | 
|  | Unfortunately there a couple of ways the journal layer can cause a deadlock. | 
|  | The first thing to note is that each task can only have | 
|  | a single outstanding transaction at any one time, remember nothing | 
|  | commits until the outermost journal_stop(). This means | 
|  | you must complete the transaction at the end of each file/inode/address | 
|  | etc. operation you perform, so that the journalling system isn't re-entered | 
|  | on another journal. Since transactions can't be nested/batched | 
|  | across differing journals, and another filesystem other than | 
|  | yours (say ext3) may be modified in a later syscall. | 
|  | </para> | 
|  |  | 
|  | <para> | 
|  | The second case to bear in mind is that journal_start() can | 
|  | block if there isn't enough space in the journal for your transaction | 
|  | (based on the passed nblocks param) - when it blocks it merely(!) needs to | 
|  | wait for transactions to complete and be committed from other tasks, | 
|  | so essentially we are waiting for journal_stop(). So to avoid | 
|  | deadlocks you must treat journal_start/stop() as if they | 
|  | were semaphores and include them in your semaphore ordering rules to prevent | 
|  | deadlocks. Note that journal_extend() has similar blocking behaviour to | 
|  | journal_start() so you can deadlock here just as easily as on journal_start(). | 
|  | </para> | 
|  |  | 
|  | <para> | 
|  | Try to reserve the right number of blocks the first time. ;-). This will | 
|  | be the maximum number of blocks you are going to touch in this transaction. | 
|  | I advise having a look at at least ext3_jbd.h to see the basis on which | 
|  | ext3 uses to make these decisions. | 
|  | </para> | 
|  |  | 
|  | <para> | 
|  | Another wriggle to watch out for is your on-disk block allocation strategy. | 
|  | why? Because, if you undo a delete, you need to ensure you haven't reused any | 
|  | of the freed blocks in a later transaction. One simple way of doing this | 
|  | is make sure any blocks you allocate only have checkpointed transactions | 
|  | listed against them. Ext3 does this in ext3_test_allocatable(). | 
|  | </para> | 
|  |  | 
|  | <para> | 
|  | Lock is also providing through journal_{un,}lock_updates(), | 
|  | ext3 uses this when it wants a window with a clean and stable fs for a moment. | 
|  | eg. | 
|  | </para> | 
|  |  | 
|  | <programlisting> | 
|  |  | 
|  | journal_lock_updates() //stop new stuff happening.. | 
|  | journal_flush()        // checkpoint everything. | 
|  | ..do stuff on stable fs | 
|  | journal_unlock_updates() // carry on with filesystem use. | 
|  | </programlisting> | 
|  |  | 
|  | <para> | 
|  | The opportunities for abuse and DOS attacks with this should be obvious, | 
|  | if you allow unprivileged userspace to trigger codepaths containing these | 
|  | calls. | 
|  | </para> | 
|  |  | 
|  | <para> | 
|  | A new feature of jbd since 2.5.25 is commit callbacks with the new | 
|  | journal_callback_set() function you can now ask the journalling layer | 
|  | to call you back when the transaction is finally committed to disk, so that | 
|  | you can do some of your own management. The key to this is the journal_callback | 
|  | struct, this maintains the internal callback information but you can | 
|  | extend it like this:- | 
|  | </para> | 
|  | <programlisting> | 
|  | struct  myfs_callback_s { | 
|  | //Data structure element required by jbd.. | 
|  | struct journal_callback for_jbd; | 
|  | // Stuff for myfs allocated together. | 
|  | myfs_inode*    i_commited; | 
|  |  | 
|  | } | 
|  | </programlisting> | 
|  |  | 
|  | <para> | 
|  | this would be useful if you needed to know when data was committed to a | 
|  | particular inode. | 
|  | </para> | 
|  |  | 
|  | </sect2> | 
|  |  | 
|  | <sect2 id="jbd_summary"> | 
|  | <title>Summary</title> | 
|  | <para> | 
|  | Using the journal is a matter of wrapping the different context changes, | 
|  | being each mount, each modification (transaction) and each changed buffer | 
|  | to tell the journalling layer about them. | 
|  | </para> | 
|  |  | 
|  | <para> | 
|  | Here is a some pseudo code to give you an idea of how it works, as | 
|  | an example. | 
|  | </para> | 
|  |  | 
|  | <programlisting> | 
|  | journal_t* my_jnrl = journal_create(); | 
|  | journal_init_{dev,inode}(jnrl,...) | 
|  | if (clean) journal_wipe(); | 
|  | journal_load(); | 
|  |  | 
|  | foreach(transaction) { /*transactions must be | 
|  | completed before | 
|  | a syscall returns to | 
|  | userspace*/ | 
|  |  | 
|  | handle_t * xct=journal_start(my_jnrl); | 
|  | foreach(bh) { | 
|  | journal_get_{create,write,undo}_access(xact,bh); | 
|  | if ( myfs_modify(bh) ) { /* returns true | 
|  | if makes changes */ | 
|  | journal_dirty_{meta,}data(xact,bh); | 
|  | } else { | 
|  | journal_forget(bh); | 
|  | } | 
|  | } | 
|  | journal_stop(xct); | 
|  | } | 
|  | journal_destroy(my_jrnl); | 
|  | </programlisting> | 
|  | </sect2> | 
|  |  | 
|  | </sect1> | 
|  |  | 
|  | <sect1 id="data_types"> | 
|  | <title>Data Types</title> | 
|  | <para> | 
|  | The journalling layer uses typedefs to 'hide' the concrete definitions | 
|  | of the structures used. As a client of the JBD layer you can | 
|  | just rely on the using the pointer as a magic cookie  of some sort. | 
|  |  | 
|  | Obviously the hiding is not enforced as this is 'C'. | 
|  | </para> | 
|  | <sect2 id="structures"><title>Structures</title> | 
|  | !Iinclude/linux/jbd.h | 
|  | </sect2> | 
|  | </sect1> | 
|  |  | 
|  | <sect1 id="functions"> | 
|  | <title>Functions</title> | 
|  | <para> | 
|  | The functions here are split into two groups those that | 
|  | affect a journal as a whole, and those which are used to | 
|  | manage transactions | 
|  | </para> | 
|  | <sect2 id="journal_level"><title>Journal Level</title> | 
|  | !Efs/jbd/journal.c | 
|  | !Ifs/jbd/recovery.c | 
|  | </sect2> | 
|  | <sect2 id="transaction_level"><title>Transasction Level</title> | 
|  | !Efs/jbd/transaction.c | 
|  | </sect2> | 
|  | </sect1> | 
|  | <sect1 id="see_also"> | 
|  | <title>See also</title> | 
|  | <para> | 
|  | <citation> | 
|  | <ulink url="http://kernel.org/pub/linux/kernel/people/sct/ext3/journal-design.ps.gz"> | 
|  | Journaling the Linux ext2fs Filesystem, LinuxExpo 98, Stephen Tweedie | 
|  | </ulink> | 
|  | </citation> | 
|  | </para> | 
|  | <para> | 
|  | <citation> | 
|  | <ulink url="http://olstrans.sourceforge.net/release/OLS2000-ext3/OLS2000-ext3.html"> | 
|  | Ext3 Journalling FileSystem, OLS 2000, Dr. Stephen Tweedie | 
|  | </ulink> | 
|  | </citation> | 
|  | </para> | 
|  | </sect1> | 
|  |  | 
|  | </chapter> | 
|  |  | 
|  | <chapter id="splice"> | 
|  | <title>splice API</title> | 
|  | <para> | 
|  | splice is a method for moving blocks of data around inside the | 
|  | kernel, without continually transferring them between the kernel | 
|  | and user space. | 
|  | </para> | 
|  | !Ffs/splice.c | 
|  | </chapter> | 
|  |  | 
|  | <chapter id="pipes"> | 
|  | <title>pipes API</title> | 
|  | <para> | 
|  | Pipe interfaces are all for in-kernel (builtin image) use. | 
|  | They are not exported for use by modules. | 
|  | </para> | 
|  | !Iinclude/linux/pipe_fs_i.h | 
|  | !Ffs/pipe.c | 
|  | </chapter> | 
|  |  | 
|  | </book> |