ia64/linux-2.6.18-xen.hg

view Documentation/filesystems/fuse.txt @ 524:7f8b544237bf

netfront: Allow netfront in domain 0.

This is useful if your physical network device is in a utility domain.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
author Keir Fraser <keir.fraser@citrix.com>
date Tue Apr 15 15:18:58 2008 +0100 (2008-04-15)
parents 831230e53067
children
line source
1 Definitions
2 ~~~~~~~~~~~
4 Userspace filesystem:
6 A filesystem in which data and metadata are provided by an ordinary
7 userspace process. The filesystem can be accessed normally through
8 the kernel interface.
10 Filesystem daemon:
12 The process(es) providing the data and metadata of the filesystem.
14 Non-privileged mount (or user mount):
16 A userspace filesystem mounted by a non-privileged (non-root) user.
17 The filesystem daemon is running with the privileges of the mounting
18 user. NOTE: this is not the same as mounts allowed with the "user"
19 option in /etc/fstab, which is not discussed here.
21 Filesystem connection:
23 A connection between the filesystem daemon and the kernel. The
24 connection exists until either the daemon dies, or the filesystem is
25 umounted. Note that detaching (or lazy umounting) the filesystem
26 does _not_ break the connection, in this case it will exist until
27 the last reference to the filesystem is released.
29 Mount owner:
31 The user who does the mounting.
33 User:
35 The user who is performing filesystem operations.
37 What is FUSE?
38 ~~~~~~~~~~~~~
40 FUSE is a userspace filesystem framework. It consists of a kernel
41 module (fuse.ko), a userspace library (libfuse.*) and a mount utility
42 (fusermount).
44 One of the most important features of FUSE is allowing secure,
45 non-privileged mounts. This opens up new possibilities for the use of
46 filesystems. A good example is sshfs: a secure network filesystem
47 using the sftp protocol.
49 The userspace library and utilities are available from the FUSE
50 homepage:
52 http://fuse.sourceforge.net/
54 Mount options
55 ~~~~~~~~~~~~~
57 'fd=N'
59 The file descriptor to use for communication between the userspace
60 filesystem and the kernel. The file descriptor must have been
61 obtained by opening the FUSE device ('/dev/fuse').
63 'rootmode=M'
65 The file mode of the filesystem's root in octal representation.
67 'user_id=N'
69 The numeric user id of the mount owner.
71 'group_id=N'
73 The numeric group id of the mount owner.
75 'default_permissions'
77 By default FUSE doesn't check file access permissions, the
78 filesystem is free to implement it's access policy or leave it to
79 the underlying file access mechanism (e.g. in case of network
80 filesystems). This option enables permission checking, restricting
81 access based on file mode. This is option is usually useful
82 together with the 'allow_other' mount option.
84 'allow_other'
86 This option overrides the security measure restricting file access
87 to the user mounting the filesystem. This option is by default only
88 allowed to root, but this restriction can be removed with a
89 (userspace) configuration option.
91 'max_read=N'
93 With this option the maximum size of read operations can be set.
94 The default is infinite. Note that the size of read requests is
95 limited anyway to 32 pages (which is 128kbyte on i386).
97 Control filesystem
98 ~~~~~~~~~~~~~~~~~~
100 There's a control filesystem for FUSE, which can be mounted by:
102 mount -t fusectl none /sys/fs/fuse/connections
104 Mounting it under the '/sys/fs/fuse/connections' directory makes it
105 backwards compatible with earlier versions.
107 Under the fuse control filesystem each connection has a directory
108 named by a unique number.
110 For each connection the following files exist within this directory:
112 'waiting'
114 The number of requests which are waiting to be transfered to
115 userspace or being processed by the filesystem daemon. If there is
116 no filesystem activity and 'waiting' is non-zero, then the
117 filesystem is hung or deadlocked.
119 'abort'
121 Writing anything into this file will abort the filesystem
122 connection. This means that all waiting requests will be aborted an
123 error returned for all aborted and new requests.
125 Only the owner of the mount may read or write these files.
127 Interrupting filesystem operations
128 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
130 If a process issuing a FUSE filesystem request is interrupted, the
131 following will happen:
133 1) If the request is not yet sent to userspace AND the signal is
134 fatal (SIGKILL or unhandled fatal signal), then the request is
135 dequeued and returns immediately.
137 2) If the request is not yet sent to userspace AND the signal is not
138 fatal, then an 'interrupted' flag is set for the request. When
139 the request has been successfully transfered to userspace and
140 this flag is set, an INTERRUPT request is queued.
142 3) If the request is already sent to userspace, then an INTERRUPT
143 request is queued.
145 INTERRUPT requests take precedence over other requests, so the
146 userspace filesystem will receive queued INTERRUPTs before any others.
148 The userspace filesystem may ignore the INTERRUPT requests entirely,
149 or may honor them by sending a reply to the _original_ request, with
150 the error set to EINTR.
152 It is also possible that there's a race between processing the
153 original request and it's INTERRUPT request. There are two possibilities:
155 1) The INTERRUPT request is processed before the original request is
156 processed
158 2) The INTERRUPT request is processed after the original request has
159 been answered
161 If the filesystem cannot find the original request, it should wait for
162 some timeout and/or a number of new requests to arrive, after which it
163 should reply to the INTERRUPT request with an EAGAIN error. In case
164 1) the INTERRUPT request will be requeued. In case 2) the INTERRUPT
165 reply will be ignored.
167 Aborting a filesystem connection
168 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
170 It is possible to get into certain situations where the filesystem is
171 not responding. Reasons for this may be:
173 a) Broken userspace filesystem implementation
175 b) Network connection down
177 c) Accidental deadlock
179 d) Malicious deadlock
181 (For more on c) and d) see later sections)
183 In either of these cases it may be useful to abort the connection to
184 the filesystem. There are several ways to do this:
186 - Kill the filesystem daemon. Works in case of a) and b)
188 - Kill the filesystem daemon and all users of the filesystem. Works
189 in all cases except some malicious deadlocks
191 - Use forced umount (umount -f). Works in all cases but only if
192 filesystem is still attached (it hasn't been lazy unmounted)
194 - Abort filesystem through the FUSE control filesystem. Most
195 powerful method, always works.
197 How do non-privileged mounts work?
198 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
200 Since the mount() system call is a privileged operation, a helper
201 program (fusermount) is needed, which is installed setuid root.
203 The implication of providing non-privileged mounts is that the mount
204 owner must not be able to use this capability to compromise the
205 system. Obvious requirements arising from this are:
207 A) mount owner should not be able to get elevated privileges with the
208 help of the mounted filesystem
210 B) mount owner should not get illegitimate access to information from
211 other users' and the super user's processes
213 C) mount owner should not be able to induce undesired behavior in
214 other users' or the super user's processes
216 How are requirements fulfilled?
217 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
219 A) The mount owner could gain elevated privileges by either:
221 1) creating a filesystem containing a device file, then opening
222 this device
224 2) creating a filesystem containing a suid or sgid application,
225 then executing this application
227 The solution is not to allow opening device files and ignore
228 setuid and setgid bits when executing programs. To ensure this
229 fusermount always adds "nosuid" and "nodev" to the mount options
230 for non-privileged mounts.
232 B) If another user is accessing files or directories in the
233 filesystem, the filesystem daemon serving requests can record the
234 exact sequence and timing of operations performed. This
235 information is otherwise inaccessible to the mount owner, so this
236 counts as an information leak.
238 The solution to this problem will be presented in point 2) of C).
240 C) There are several ways in which the mount owner can induce
241 undesired behavior in other users' processes, such as:
243 1) mounting a filesystem over a file or directory which the mount
244 owner could otherwise not be able to modify (or could only
245 make limited modifications).
247 This is solved in fusermount, by checking the access
248 permissions on the mountpoint and only allowing the mount if
249 the mount owner can do unlimited modification (has write
250 access to the mountpoint, and mountpoint is not a "sticky"
251 directory)
253 2) Even if 1) is solved the mount owner can change the behavior
254 of other users' processes.
256 i) It can slow down or indefinitely delay the execution of a
257 filesystem operation creating a DoS against the user or the
258 whole system. For example a suid application locking a
259 system file, and then accessing a file on the mount owner's
260 filesystem could be stopped, and thus causing the system
261 file to be locked forever.
263 ii) It can present files or directories of unlimited length, or
264 directory structures of unlimited depth, possibly causing a
265 system process to eat up diskspace, memory or other
266 resources, again causing DoS.
268 The solution to this as well as B) is not to allow processes
269 to access the filesystem, which could otherwise not be
270 monitored or manipulated by the mount owner. Since if the
271 mount owner can ptrace a process, it can do all of the above
272 without using a FUSE mount, the same criteria as used in
273 ptrace can be used to check if a process is allowed to access
274 the filesystem or not.
276 Note that the ptrace check is not strictly necessary to
277 prevent B/2/i, it is enough to check if mount owner has enough
278 privilege to send signal to the process accessing the
279 filesystem, since SIGSTOP can be used to get a similar effect.
281 I think these limitations are unacceptable?
282 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
284 If a sysadmin trusts the users enough, or can ensure through other
285 measures, that system processes will never enter non-privileged
286 mounts, it can relax the last limitation with a "user_allow_other"
287 config option. If this config option is set, the mounting user can
288 add the "allow_other" mount option which disables the check for other
289 users' processes.
291 Kernel - userspace interface
292 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
294 The following diagram shows how a filesystem operation (in this
295 example unlink) is performed in FUSE.
297 NOTE: everything in this description is greatly simplified
299 | "rm /mnt/fuse/file" | FUSE filesystem daemon
300 | |
301 | | >sys_read()
302 | | >fuse_dev_read()
303 | | >request_wait()
304 | | [sleep on fc->waitq]
305 | |
306 | >sys_unlink() |
307 | >fuse_unlink() |
308 | [get request from |
309 | fc->unused_list] |
310 | >request_send() |
311 | [queue req on fc->pending] |
312 | [wake up fc->waitq] | [woken up]
313 | >request_wait_answer() |
314 | [sleep on req->waitq] |
315 | | <request_wait()
316 | | [remove req from fc->pending]
317 | | [copy req to read buffer]
318 | | [add req to fc->processing]
319 | | <fuse_dev_read()
320 | | <sys_read()
321 | |
322 | | [perform unlink]
323 | |
324 | | >sys_write()
325 | | >fuse_dev_write()
326 | | [look up req in fc->processing]
327 | | [remove from fc->processing]
328 | | [copy write buffer to req]
329 | [woken up] | [wake up req->waitq]
330 | | <fuse_dev_write()
331 | | <sys_write()
332 | <request_wait_answer() |
333 | <request_send() |
334 | [add request to |
335 | fc->unused_list] |
336 | <fuse_unlink() |
337 | <sys_unlink() |
339 There are a couple of ways in which to deadlock a FUSE filesystem.
340 Since we are talking about unprivileged userspace programs,
341 something must be done about these.
343 Scenario 1 - Simple deadlock
344 -----------------------------
346 | "rm /mnt/fuse/file" | FUSE filesystem daemon
347 | |
348 | >sys_unlink("/mnt/fuse/file") |
349 | [acquire inode semaphore |
350 | for "file"] |
351 | >fuse_unlink() |
352 | [sleep on req->waitq] |
353 | | <sys_read()
354 | | >sys_unlink("/mnt/fuse/file")
355 | | [acquire inode semaphore
356 | | for "file"]
357 | | *DEADLOCK*
359 The solution for this is to allow the filesystem to be aborted.
361 Scenario 2 - Tricky deadlock
362 ----------------------------
364 This one needs a carefully crafted filesystem. It's a variation on
365 the above, only the call back to the filesystem is not explicit,
366 but is caused by a pagefault.
368 | Kamikaze filesystem thread 1 | Kamikaze filesystem thread 2
369 | |
370 | [fd = open("/mnt/fuse/file")] | [request served normally]
371 | [mmap fd to 'addr'] |
372 | [close fd] | [FLUSH triggers 'magic' flag]
373 | [read a byte from addr] |
374 | >do_page_fault() |
375 | [find or create page] |
376 | [lock page] |
377 | >fuse_readpage() |
378 | [queue READ request] |
379 | [sleep on req->waitq] |
380 | | [read request to buffer]
381 | | [create reply header before addr]
382 | | >sys_write(addr - headerlength)
383 | | >fuse_dev_write()
384 | | [look up req in fc->processing]
385 | | [remove from fc->processing]
386 | | [copy write buffer to req]
387 | | >do_page_fault()
388 | | [find or create page]
389 | | [lock page]
390 | | * DEADLOCK *
392 Solution is basically the same as above.
394 An additional problem is that while the write buffer is being copied
395 to the request, the request must not be interrupted/aborted. This is
396 because the destination address of the copy may not be valid after the
397 request has returned.
399 This is solved with doing the copy atomically, and allowing abort
400 while the page(s) belonging to the write buffer are faulted with
401 get_user_pages(). The 'req->locked' flag indicates when the copy is
402 taking place, and abort is delayed until this flag is unset.