fs/jffs2/README.Locking

   1
   2         JFFS2 LOCKING DOCUMENTATION
   3         ---------------------------
   4
   5 This document attempts to describe the existing locking rules for
   6 JFFS2. It is not expected to remain perfectly up to date, but ought to
   7 be fairly close.
   8
   9
  10         alloc_sem
  11         ---------
  12
  13 The alloc_sem is a per-filesystem mutex, used primarily to ensure
  14 contiguous allocation of space on the medium. It is automatically
  15 obtained during space allocations (jffs2_reserve_space()) and freed
  16 upon write completion (jffs2_complete_reservation()). Note that
  17 the garbage collector will obtain this right at the beginning of
  18 jffs2_garbage_collect_pass() and release it at the end, thereby
  19 preventing any other write activity on the file system during a
  20 garbage collect pass.
  21
  22 When writing new nodes, the alloc_sem must be held until the new nodes
  23 have been properly linked into the data structures for the inode to
  24 which they belong. This is for the benefit of NAND flash - adding new
  25 nodes to an inode may obsolete old ones, and by holding the alloc_sem
  26 until this happens we ensure that any data in the write-buffer at the
  27 time this happens are part of the new node, not just something that
  28 was written afterwards. Hence, we can ensure the newly-obsoleted nodes
  29 don't actually get erased until the write-buffer has been flushed to
  30 the medium.
  31
  32 With the introduction of NAND flash support and the write-buffer,
  33 the alloc_sem is also used to protect the wbuf-related members of the
  34 jffs2_sb_info structure. Atomically reading the wbuf_len member to see
  35 if the wbuf is currently holding any data is permitted, though.
  36
  37 Ordering constraints: See f->sem.
  38
  39
  40         File Mutex f->sem
  41         ---------------------
  42
  43 This is the JFFS2-internal equivalent of the inode mutex i->i_sem.
  44 It protects the contents of the jffs2_inode_info private inode data,
  45 including the linked list of node fragments (but see the notes below on
  46 erase_completion_lock), etc.
  47
  48 The reason that the i_sem itself isn't used for this purpose is to
  49 avoid deadlocks with garbage collection -- the VFS will lock the i_sem
  50 before calling a function which may need to allocate space. The
  51 allocation may trigger garbage-collection, which may need to move a
  52 node belonging to the inode which was locked in the first place by the
  53 VFS. If the garbage collection code were to attempt to lock the i_sem
  54 of the inode from which it's garbage-collecting a physical node, this
  55 lead to deadlock, unless we played games with unlocking the i_sem
  56 before calling the space allocation functions.
  57
  58 Instead of playing such games, we just have an extra internal
  59 mutex, which is obtained by the garbage collection code and also
  60 by the normal file system code _after_ allocation of space.
  61
  62 Ordering constraints:
  63
  64         1. Never attempt to allocate space or lock alloc_sem with
  65            any f->sem held.
  66         2. Never attempt to lock two file mutexes in one thread.
  67            No ordering rules have been made for doing so.
  68         3. Never lock a page cache page with f->sem held.
  69
  70
  71         erase_completion_lock spinlock
  72         ------------------------------
  73
  74 This is used to serialise access to the eraseblock lists, to the
  75 per-eraseblock lists of physical jffs2_raw_node_ref structures, and
  76 (NB) the per-inode list of physical nodes. The latter is a special
  77 case - see below.
  78
  79 As the MTD API no longer permits erase-completion callback functions
  80 to be called from bottom-half (timer) context (on the basis that nobody
  81 ever actually implemented such a thing), it's now sufficient to use
  82 a simple spin_lock() rather than spin_lock_bh().
  83
  84 Note that the per-inode list of physical nodes (f->nodes) is a special
  85 case. Any changes to _valid_ nodes (i.e. ->flash_offset & 1 == 0) in
  86 the list are protected by the file mutex f->sem. But the erase code
  87 may remove _obsolete_ nodes from the list while holding only the
  88 erase_completion_lock. So you can walk the list only while holding the
  89 erase_completion_lock, and can drop the lock temporarily mid-walk as
  90 long as the pointer you're holding is to a _valid_ node, not an
  91 obsolete one.
  92
  93 The erase_completion_lock is also used to protect the c->gc_task
  94 pointer when the garbage collection thread exits. The code to kill the
  95 GC thread locks it, sends the signal, then unlocks it - while the GC
  96 thread itself locks it, zeroes c->gc_task, then unlocks on the exit path.
  97
  98
  99         inocache_lock spinlock
 100         ----------------------
 101
 102 This spinlock protects the hashed list (c->inocache_list) of the
 103 in-core jffs2_inode_cache objects (each inode in JFFS2 has the
 104 correspondent jffs2_inode_cache object). So, the inocache_lock
 105 has to be locked while walking the c->inocache_list hash buckets.
 106
 107 This spinlock also covers allocation of new inode numbers, which is
 108 currently just '++->highest_ino++', but might one day get more complicated
 109 if we need to deal with wrapping after 4 milliard inode numbers are used.
 110
 111 Note, the f->sem guarantees that the correspondent jffs2_inode_cache
 112 will not be removed. So, it is allowed to access it without locking
 113 the inocache_lock spinlock.
 114
 115 Ordering constraints:
 116
 117         If both erase_completion_lock and inocache_lock are needed, the
 118         c->erase_completion has to be acquired first.
 119
 120
 121         erase_free_sem
 122         --------------
 123
 124 This mutex is only used by the erase code which frees obsolete node
 125 references and the jffs2_garbage_collect_deletion_dirent() function.
 126 The latter function on NAND flash must read _obsolete_ nodes to
 127 determine whether the 'deletion dirent' under consideration can be
 128 discarded or whether it is still required to show that an inode has
 129 been unlinked. Because reading from the flash may sleep, the
 130 erase_completion_lock cannot be held, so an alternative, more
 131 heavyweight lock was required to prevent the erase code from freeing
 132 the jffs2_raw_node_ref structures in question while the garbage
 133 collection code is looking at them.
 134
 135 Suggestions for alternative solutions to this problem would be welcomed.
 136
 137
 138         wbuf_sem
 139         --------
 140
 141 This read/write semaphore protects against concurrent access to the
 142 write-behind buffer ('wbuf') used for flash chips where we must write
 143 in blocks. It protects both the contents of the wbuf and the metadata
 144 which indicates which flash region (if any) is currently covered by
 145 the buffer.
 146
 147 Ordering constraints:
 148         Lock wbuf_sem last, after the alloc_sem or and f->sem.
 149
 150
 151         c->xattr_sem
 152         ------------
 153
 154 This read/write semaphore protects against concurrent access to the
 155 xattr related objects which include stuff in superblock and ic->xref.
 156 In read-only path, write-semaphore is too much exclusion. It's enough
 157 by read-semaphore. But you must hold write-semaphore when updating,
 158 creating or deleting any xattr related object.
 159
 160 Once xattr_sem released, there would be no assurance for the existence
 161 of those objects. Thus, a series of processes is often required to retry,
 162 when updating such a object is necessary under holding read semaphore.
 163 For example, do_jffs2_getxattr() holds read-semaphore to scan xref and
 164 xdatum at first. But it retries this process with holding write-semaphore
 165 after release read-semaphore, if it's necessary to load name/value pair
 166 from medium.
 167
 168 Ordering constraints:
 169         Lock xattr_sem last, after the alloc_sem.