exporting.rst 7.1 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165
  1. :orphan:
  2. Making Filesystems Exportable
  3. =============================
  4. Overview
  5. --------
  6. All filesystem operations require a dentry (or two) as a starting
  7. point. Local applications have a reference-counted hold on suitable
  8. dentries via open file descriptors or cwd/root. However remote
  9. applications that access a filesystem via a remote filesystem protocol
  10. such as NFS may not be able to hold such a reference, and so need a
  11. different way to refer to a particular dentry. As the alternative
  12. form of reference needs to be stable across renames, truncates, and
  13. server-reboot (among other things, though these tend to be the most
  14. problematic), there is no simple answer like 'filename'.
  15. The mechanism discussed here allows each filesystem implementation to
  16. specify how to generate an opaque (outside of the filesystem) byte
  17. string for any dentry, and how to find an appropriate dentry for any
  18. given opaque byte string.
  19. This byte string will be called a "filehandle fragment" as it
  20. corresponds to part of an NFS filehandle.
  21. A filesystem which supports the mapping between filehandle fragments
  22. and dentries will be termed "exportable".
  23. Dcache Issues
  24. -------------
  25. The dcache normally contains a proper prefix of any given filesystem
  26. tree. This means that if any filesystem object is in the dcache, then
  27. all of the ancestors of that filesystem object are also in the dcache.
  28. As normal access is by filename this prefix is created naturally and
  29. maintained easily (by each object maintaining a reference count on
  30. its parent).
  31. However when objects are included into the dcache by interpreting a
  32. filehandle fragment, there is no automatic creation of a path prefix
  33. for the object. This leads to two related but distinct features of
  34. the dcache that are not needed for normal filesystem access.
  35. 1. The dcache must sometimes contain objects that are not part of the
  36. proper prefix. i.e that are not connected to the root.
  37. 2. The dcache must be prepared for a newly found (via ->lookup) directory
  38. to already have a (non-connected) dentry, and must be able to move
  39. that dentry into place (based on the parent and name in the
  40. ->lookup). This is particularly needed for directories as
  41. it is a dcache invariant that directories only have one dentry.
  42. To implement these features, the dcache has:
  43. a. A dentry flag DCACHE_DISCONNECTED which is set on
  44. any dentry that might not be part of the proper prefix.
  45. This is set when anonymous dentries are created, and cleared when a
  46. dentry is noticed to be a child of a dentry which is in the proper
  47. prefix. If the refcount on a dentry with this flag set
  48. becomes zero, the dentry is immediately discarded, rather than being
  49. kept in the dcache. If a dentry that is not already in the dcache
  50. is repeatedly accessed by filehandle (as NFSD might do), an new dentry
  51. will be a allocated for each access, and discarded at the end of
  52. the access.
  53. Note that such a dentry can acquire children, name, ancestors, etc.
  54. without losing DCACHE_DISCONNECTED - that flag is only cleared when
  55. subtree is successfully reconnected to root. Until then dentries
  56. in such subtree are retained only as long as there are references;
  57. refcount reaching zero means immediate eviction, same as for unhashed
  58. dentries. That guarantees that we won't need to hunt them down upon
  59. umount.
  60. b. A primitive for creation of secondary roots - d_obtain_root(inode).
  61. Those do _not_ bear DCACHE_DISCONNECTED. They are placed on the
  62. per-superblock list (->s_roots), so they can be located at umount
  63. time for eviction purposes.
  64. c. Helper routines to allocate anonymous dentries, and to help attach
  65. loose directory dentries at lookup time. They are:
  66. d_obtain_alias(inode) will return a dentry for the given inode.
  67. If the inode already has a dentry, one of those is returned.
  68. If it doesn't, a new anonymous (IS_ROOT and
  69. DCACHE_DISCONNECTED) dentry is allocated and attached.
  70. In the case of a directory, care is taken that only one dentry
  71. can ever be attached.
  72. d_splice_alias(inode, dentry) will introduce a new dentry into the tree;
  73. either the passed-in dentry or a preexisting alias for the given inode
  74. (such as an anonymous one created by d_obtain_alias), if appropriate.
  75. It returns NULL when the passed-in dentry is used, following the calling
  76. convention of ->lookup.
  77. Filesystem Issues
  78. -----------------
  79. For a filesystem to be exportable it must:
  80. 1. provide the filehandle fragment routines described below.
  81. 2. make sure that d_splice_alias is used rather than d_add
  82. when ->lookup finds an inode for a given parent and name.
  83. If inode is NULL, d_splice_alias(inode, dentry) is equivalent to::
  84. d_add(dentry, inode), NULL
  85. Similarly, d_splice_alias(ERR_PTR(err), dentry) = ERR_PTR(err)
  86. Typically the ->lookup routine will simply end with a::
  87. return d_splice_alias(inode, dentry);
  88. }
  89. A file system implementation declares that instances of the filesystem
  90. are exportable by setting the s_export_op field in the struct
  91. super_block. This field must point to a "struct export_operations"
  92. struct which has the following members:
  93. encode_fh (optional)
  94. Takes a dentry and creates a filehandle fragment which can later be used
  95. to find or create a dentry for the same object. The default
  96. implementation creates a filehandle fragment that encodes a 32bit inode
  97. and generation number for the inode encoded, and if necessary the
  98. same information for the parent.
  99. fh_to_dentry (mandatory)
  100. Given a filehandle fragment, this should find the implied object and
  101. create a dentry for it (possibly with d_obtain_alias).
  102. fh_to_parent (optional but strongly recommended)
  103. Given a filehandle fragment, this should find the parent of the
  104. implied object and create a dentry for it (possibly with
  105. d_obtain_alias). May fail if the filehandle fragment is too small.
  106. get_parent (optional but strongly recommended)
  107. When given a dentry for a directory, this should return a dentry for
  108. the parent. Quite possibly the parent dentry will have been allocated
  109. by d_alloc_anon. The default get_parent function just returns an error
  110. so any filehandle lookup that requires finding a parent will fail.
  111. ->lookup("..") is *not* used as a default as it can leave ".." entries
  112. in the dcache which are too messy to work with.
  113. get_name (optional)
  114. When given a parent dentry and a child dentry, this should find a name
  115. in the directory identified by the parent dentry, which leads to the
  116. object identified by the child dentry. If no get_name function is
  117. supplied, a default implementation is provided which uses vfs_readdir
  118. to find potential names, and matches inode numbers to find the correct
  119. match.
  120. A filehandle fragment consists of an array of 1 or more 4byte words,
  121. together with a one byte "type".
  122. The decode_fh routine should not depend on the stated size that is
  123. passed to it. This size may be larger than the original filehandle
  124. generated by encode_fh, in which case it will have been padded with
  125. nuls. Rather, the encode_fh routine should choose a "type" which
  126. indicates the decode_fh how much of the filehandle is valid, and how
  127. it should be interpreted.