Commit | Line | Data |
---|---|---|
7c37fbda NB |
1 | Written by: Neil Brown <neilb@suse.de> |
2 | ||
3 | Overlay Filesystem | |
4 | ================== | |
5 | ||
6 | This document describes a prototype for a new approach to providing | |
7 | overlay-filesystem functionality in Linux (sometimes referred to as | |
8 | union-filesystems). An overlay-filesystem tries to present a | |
9 | filesystem which is the result over overlaying one filesystem on top | |
10 | of the other. | |
11 | ||
12 | The result will inevitably fail to look exactly like a normal | |
13 | filesystem for various technical reasons. The expectation is that | |
14 | many use cases will be able to ignore these differences. | |
15 | ||
16 | This approach is 'hybrid' because the objects that appear in the | |
17 | filesystem do not all appear to belong to that filesystem. In many | |
18 | cases an object accessed in the union will be indistinguishable | |
19 | from accessing the corresponding object from the original filesystem. | |
20 | This is most obvious from the 'st_dev' field returned by stat(2). | |
21 | ||
22 | While directories will report an st_dev from the overlay-filesystem, | |
23 | all non-directory objects will report an st_dev from the lower or | |
24 | upper filesystem that is providing the object. Similarly st_ino will | |
25 | only be unique when combined with st_dev, and both of these can change | |
26 | over the lifetime of a non-directory object. Many applications and | |
27 | tools ignore these values and will not be affected. | |
28 | ||
29 | Upper and Lower | |
30 | --------------- | |
31 | ||
32 | An overlay filesystem combines two filesystems - an 'upper' filesystem | |
33 | and a 'lower' filesystem. When a name exists in both filesystems, the | |
34 | object in the 'upper' filesystem is visible while the object in the | |
35 | 'lower' filesystem is either hidden or, in the case of directories, | |
36 | merged with the 'upper' object. | |
37 | ||
38 | It would be more correct to refer to an upper and lower 'directory | |
39 | tree' rather than 'filesystem' as it is quite possible for both | |
40 | directory trees to be in the same filesystem and there is no | |
41 | requirement that the root of a filesystem be given for either upper or | |
42 | lower. | |
43 | ||
44 | The lower filesystem can be any filesystem supported by Linux and does | |
45 | not need to be writable. The lower filesystem can even be another | |
46 | overlayfs. The upper filesystem will normally be writable and if it | |
47 | is it must support the creation of trusted.* extended attributes, and | |
48 | must provide valid d_type in readdir responses, so NFS is not suitable. | |
49 | ||
50 | A read-only overlay of two read-only filesystems may use any | |
51 | filesystem type. | |
52 | ||
53 | Directories | |
54 | ----------- | |
55 | ||
56 | Overlaying mainly involves directories. If a given name appears in both | |
57 | upper and lower filesystems and refers to a non-directory in either, | |
58 | then the lower object is hidden - the name refers only to the upper | |
59 | object. | |
60 | ||
61 | Where both upper and lower objects are directories, a merged directory | |
62 | is formed. | |
63 | ||
64 | At mount time, the two directories given as mount options "lowerdir" and | |
65 | "upperdir" are combined into a merged directory: | |
66 | ||
ef94b186 | 67 | mount -t overlay overlay -olowerdir=/lower,upperdir=/upper,\ |
7c37fbda NB |
68 | workdir=/work /merged |
69 | ||
70 | The "workdir" needs to be an empty directory on the same filesystem | |
71 | as upperdir. | |
72 | ||
73 | Then whenever a lookup is requested in such a merged directory, the | |
74 | lookup is performed in each actual directory and the combined result | |
75 | is cached in the dentry belonging to the overlay filesystem. If both | |
76 | actual lookups find directories, both are stored and a merged | |
77 | directory is created, otherwise only one is stored: the upper if it | |
78 | exists, else the lower. | |
79 | ||
80 | Only the lists of names from directories are merged. Other content | |
81 | such as metadata and extended attributes are reported for the upper | |
82 | directory only. These attributes of the lower directory are hidden. | |
83 | ||
84 | whiteouts and opaque directories | |
85 | -------------------------------- | |
86 | ||
87 | In order to support rm and rmdir without changing the lower | |
88 | filesystem, an overlay filesystem needs to record in the upper filesystem | |
89 | that files have been removed. This is done using whiteouts and opaque | |
90 | directories (non-directories are always opaque). | |
91 | ||
92 | A whiteout is created as a character device with 0/0 device number. | |
93 | When a whiteout is found in the upper level of a merged directory, any | |
94 | matching name in the lower level is ignored, and the whiteout itself | |
95 | is also hidden. | |
96 | ||
97 | A directory is made opaque by setting the xattr "trusted.overlay.opaque" | |
98 | to "y". Where the upper filesystem contains an opaque directory, any | |
99 | directory in the lower filesystem with the same name is ignored. | |
100 | ||
101 | readdir | |
102 | ------- | |
103 | ||
104 | When a 'readdir' request is made on a merged directory, the upper and | |
105 | lower directories are each read and the name lists merged in the | |
106 | obvious way (upper is read first, then lower - entries that already | |
107 | exist are not re-added). This merged name list is cached in the | |
108 | 'struct file' and so remains as long as the file is kept open. If the | |
109 | directory is opened and read by two processes at the same time, they | |
110 | will each have separate caches. A seekdir to the start of the | |
111 | directory (offset 0) followed by a readdir will cause the cache to be | |
112 | discarded and rebuilt. | |
113 | ||
114 | This means that changes to the merged directory do not appear while a | |
115 | directory is being read. This is unlikely to be noticed by many | |
116 | programs. | |
117 | ||
118 | seek offsets are assigned sequentially when the directories are read. | |
119 | Thus if | |
120 | - read part of a directory | |
121 | - remember an offset, and close the directory | |
122 | - re-open the directory some time later | |
123 | - seek to the remembered offset | |
124 | ||
125 | there may be little correlation between the old and new locations in | |
126 | the list of filenames, particularly if anything has changed in the | |
127 | directory. | |
128 | ||
129 | Readdir on directories that are not merged is simply handled by the | |
130 | underlying directory (upper or lower). | |
131 | ||
132 | ||
133 | Non-directories | |
134 | --------------- | |
135 | ||
136 | Objects that are not directories (files, symlinks, device-special | |
137 | files etc.) are presented either from the upper or lower filesystem as | |
138 | appropriate. When a file in the lower filesystem is accessed in a way | |
139 | the requires write-access, such as opening for write access, changing | |
140 | some metadata etc., the file is first copied from the lower filesystem | |
141 | to the upper filesystem (copy_up). Note that creating a hard-link | |
142 | also requires copy_up, though of course creation of a symlink does | |
143 | not. | |
144 | ||
145 | The copy_up may turn out to be unnecessary, for example if the file is | |
146 | opened for read-write but the data is not modified. | |
147 | ||
148 | The copy_up process first makes sure that the containing directory | |
149 | exists in the upper filesystem - creating it and any parents as | |
150 | necessary. It then creates the object with the same metadata (owner, | |
151 | mode, mtime, symlink-target etc.) and then if the object is a file, the | |
152 | data is copied from the lower to the upper filesystem. Finally any | |
153 | extended attributes are copied up. | |
154 | ||
155 | Once the copy_up is complete, the overlay filesystem simply | |
156 | provides direct access to the newly created file in the upper | |
157 | filesystem - future operations on the file are barely noticed by the | |
158 | overlay filesystem (though an operation on the name of the file such as | |
159 | rename or unlink will of course be noticed and handled). | |
160 | ||
161 | ||
a78d9f0d MS |
162 | Multiple lower layers |
163 | --------------------- | |
164 | ||
165 | Multiple lower layers can now be given using the the colon (":") as a | |
166 | separator character between the directory names. For example: | |
167 | ||
168 | mount -t overlay overlay -olowerdir=/lower1:/lower2:/lower3 /merged | |
169 | ||
6d900f5a MS |
170 | As the example shows, "upperdir=" and "workdir=" may be omitted. In |
171 | that case the overlay will be read-only. | |
172 | ||
173 | The specified lower directories will be stacked beginning from the | |
174 | rightmost one and going left. In the above example lower1 will be the | |
175 | top, lower2 the middle and lower3 the bottom layer. | |
a78d9f0d MS |
176 | |
177 | ||
7c37fbda NB |
178 | Non-standard behavior |
179 | --------------------- | |
180 | ||
181 | The copy_up operation essentially creates a new, identical file and | |
182 | moves it over to the old name. The new file may be on a different | |
183 | filesystem, so both st_dev and st_ino of the file may change. | |
184 | ||
185 | Any open files referring to this inode will access the old data and | |
186 | metadata. Similarly any file locks obtained before copy_up will not | |
187 | apply to the copied up file. | |
188 | ||
189 | On a file opened with O_RDONLY fchmod(2), fchown(2), futimesat(2) and | |
190 | fsetxattr(2) will fail with EROFS. | |
191 | ||
192 | If a file with multiple hard links is copied up, then this will | |
193 | "break" the link. Changes will not be propagated to other names | |
194 | referring to the same inode. | |
195 | ||
196 | Symlinks in /proc/PID/ and /proc/PID/fd which point to a non-directory | |
197 | object in overlayfs will not contain valid absolute paths, only | |
198 | relative paths leading up to the filesystem's root. This will be | |
199 | fixed in the future. | |
200 | ||
201 | Some operations are not atomic, for example a crash during copy_up or | |
202 | rename will leave the filesystem in an inconsistent state. This will | |
203 | be addressed in the future. | |
204 | ||
205 | Changes to underlying filesystems | |
206 | --------------------------------- | |
207 | ||
208 | Offline changes, when the overlay is not mounted, are allowed to either | |
209 | the upper or the lower trees. | |
210 | ||
211 | Changes to the underlying filesystems while part of a mounted overlay | |
212 | filesystem are not allowed. If the underlying filesystem is changed, | |
213 | the behavior of the overlay is undefined, though it will not result in | |
214 | a crash or deadlock. | |
2b7a8f36 MS |
215 | |
216 | Testsuite | |
217 | --------- | |
218 | ||
219 | There's testsuite developed by David Howells at: | |
220 | ||
221 | git://git.infradead.org/users/dhowells/unionmount-testsuite.git | |
222 | ||
223 | Run as root: | |
224 | ||
225 | # cd unionmount-testsuite | |
226 | # ./run --ov |