6 A filesystem in which data and metadata are provided by an ordinary
7 userspace process. The filesystem can be accessed normally through
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.
31 The user who does the mounting.
35 The user who is performing filesystem operations.
40 FUSE is a userspace filesystem framework. It consists of a kernel
41 module (fuse.ko), a userspace library (libfuse.*) and a mount utility
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
52 http://fuse.sourceforge.net/
57 The filesystem type given to mount(2) can be one of the following:
61 This is the usual way to mount a FUSE filesystem. The first
62 argument of the mount system call may contain an arbitrary string,
63 which is not interpreted by the kernel.
67 The filesystem is block device based. The first argument of the
68 mount system call is interpreted as the name of the device.
75 The file descriptor to use for communication between the userspace
76 filesystem and the kernel. The file descriptor must have been
77 obtained by opening the FUSE device ('/dev/fuse').
81 The file mode of the filesystem's root in octal representation.
85 The numeric user id of the mount owner.
89 The numeric group id of the mount owner.
93 By default FUSE doesn't check file access permissions, the
94 filesystem is free to implement it's access policy or leave it to
95 the underlying file access mechanism (e.g. in case of network
96 filesystems). This option enables permission checking, restricting
97 access based on file mode. This is option is usually useful
98 together with the 'allow_other' mount option.
102 This option overrides the security measure restricting file access
103 to the user mounting the filesystem. This option is by default only
104 allowed to root, but this restriction can be removed with a
105 (userspace) configuration option.
109 With this option the maximum size of read operations can be set.
110 The default is infinite. Note that the size of read requests is
111 limited anyway to 32 pages (which is 128kbyte on i386).
116 There's a control filesystem for FUSE, which can be mounted by:
118 mount -t fusectl none /sys/fs/fuse/connections
120 Mounting it under the '/sys/fs/fuse/connections' directory makes it
121 backwards compatible with earlier versions.
123 Under the fuse control filesystem each connection has a directory
124 named by a unique number.
126 For each connection the following files exist within this directory:
130 The number of requests which are waiting to be transferred to
131 userspace or being processed by the filesystem daemon. If there is
132 no filesystem activity and 'waiting' is non-zero, then the
133 filesystem is hung or deadlocked.
137 Writing anything into this file will abort the filesystem
138 connection. This means that all waiting requests will be aborted an
139 error returned for all aborted and new requests.
141 Only the owner of the mount may read or write these files.
143 Interrupting filesystem operations
144 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
146 If a process issuing a FUSE filesystem request is interrupted, the
147 following will happen:
149 1) If the request is not yet sent to userspace AND the signal is
150 fatal (SIGKILL or unhandled fatal signal), then the request is
151 dequeued and returns immediately.
153 2) If the request is not yet sent to userspace AND the signal is not
154 fatal, then an 'interrupted' flag is set for the request. When
155 the request has been successfully transferred to userspace and
156 this flag is set, an INTERRUPT request is queued.
158 3) If the request is already sent to userspace, then an INTERRUPT
161 INTERRUPT requests take precedence over other requests, so the
162 userspace filesystem will receive queued INTERRUPTs before any others.
164 The userspace filesystem may ignore the INTERRUPT requests entirely,
165 or may honor them by sending a reply to the _original_ request, with
166 the error set to EINTR.
168 It is also possible that there's a race between processing the
169 original request and it's INTERRUPT request. There are two possibilities:
171 1) The INTERRUPT request is processed before the original request is
174 2) The INTERRUPT request is processed after the original request has
177 If the filesystem cannot find the original request, it should wait for
178 some timeout and/or a number of new requests to arrive, after which it
179 should reply to the INTERRUPT request with an EAGAIN error. In case
180 1) the INTERRUPT request will be requeued. In case 2) the INTERRUPT
181 reply will be ignored.
183 Aborting a filesystem connection
184 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
186 It is possible to get into certain situations where the filesystem is
187 not responding. Reasons for this may be:
189 a) Broken userspace filesystem implementation
191 b) Network connection down
193 c) Accidental deadlock
195 d) Malicious deadlock
197 (For more on c) and d) see later sections)
199 In either of these cases it may be useful to abort the connection to
200 the filesystem. There are several ways to do this:
202 - Kill the filesystem daemon. Works in case of a) and b)
204 - Kill the filesystem daemon and all users of the filesystem. Works
205 in all cases except some malicious deadlocks
207 - Use forced umount (umount -f). Works in all cases but only if
208 filesystem is still attached (it hasn't been lazy unmounted)
210 - Abort filesystem through the FUSE control filesystem. Most
211 powerful method, always works.
213 How do non-privileged mounts work?
214 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
216 Since the mount() system call is a privileged operation, a helper
217 program (fusermount) is needed, which is installed setuid root.
219 The implication of providing non-privileged mounts is that the mount
220 owner must not be able to use this capability to compromise the
221 system. Obvious requirements arising from this are:
223 A) mount owner should not be able to get elevated privileges with the
224 help of the mounted filesystem
226 B) mount owner should not get illegitimate access to information from
227 other users' and the super user's processes
229 C) mount owner should not be able to induce undesired behavior in
230 other users' or the super user's processes
232 How are requirements fulfilled?
233 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
235 A) The mount owner could gain elevated privileges by either:
237 1) creating a filesystem containing a device file, then opening
240 2) creating a filesystem containing a suid or sgid application,
241 then executing this application
243 The solution is not to allow opening device files and ignore
244 setuid and setgid bits when executing programs. To ensure this
245 fusermount always adds "nosuid" and "nodev" to the mount options
246 for non-privileged mounts.
248 B) If another user is accessing files or directories in the
249 filesystem, the filesystem daemon serving requests can record the
250 exact sequence and timing of operations performed. This
251 information is otherwise inaccessible to the mount owner, so this
252 counts as an information leak.
254 The solution to this problem will be presented in point 2) of C).
256 C) There are several ways in which the mount owner can induce
257 undesired behavior in other users' processes, such as:
259 1) mounting a filesystem over a file or directory which the mount
260 owner could otherwise not be able to modify (or could only
261 make limited modifications).
263 This is solved in fusermount, by checking the access
264 permissions on the mountpoint and only allowing the mount if
265 the mount owner can do unlimited modification (has write
266 access to the mountpoint, and mountpoint is not a "sticky"
269 2) Even if 1) is solved the mount owner can change the behavior
270 of other users' processes.
272 i) It can slow down or indefinitely delay the execution of a
273 filesystem operation creating a DoS against the user or the
274 whole system. For example a suid application locking a
275 system file, and then accessing a file on the mount owner's
276 filesystem could be stopped, and thus causing the system
277 file to be locked forever.
279 ii) It can present files or directories of unlimited length, or
280 directory structures of unlimited depth, possibly causing a
281 system process to eat up diskspace, memory or other
282 resources, again causing DoS.
284 The solution to this as well as B) is not to allow processes
285 to access the filesystem, which could otherwise not be
286 monitored or manipulated by the mount owner. Since if the
287 mount owner can ptrace a process, it can do all of the above
288 without using a FUSE mount, the same criteria as used in
289 ptrace can be used to check if a process is allowed to access
290 the filesystem or not.
292 Note that the ptrace check is not strictly necessary to
293 prevent B/2/i, it is enough to check if mount owner has enough
294 privilege to send signal to the process accessing the
295 filesystem, since SIGSTOP can be used to get a similar effect.
297 I think these limitations are unacceptable?
298 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
300 If a sysadmin trusts the users enough, or can ensure through other
301 measures, that system processes will never enter non-privileged
302 mounts, it can relax the last limitation with a "user_allow_other"
303 config option. If this config option is set, the mounting user can
304 add the "allow_other" mount option which disables the check for other
307 Kernel - userspace interface
308 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
310 The following diagram shows how a filesystem operation (in this
311 example unlink) is performed in FUSE.
313 NOTE: everything in this description is greatly simplified
315 | "rm /mnt/fuse/file" | FUSE filesystem daemon
320 | | [sleep on fc->waitq]
324 | [get request from |
327 | [queue req on fc->pending] |
328 | [wake up fc->waitq] | [woken up]
329 | >request_wait_answer() |
330 | [sleep on req->waitq] |
332 | | [remove req from fc->pending]
333 | | [copy req to read buffer]
334 | | [add req to fc->processing]
341 | | >fuse_dev_write()
342 | | [look up req in fc->processing]
343 | | [remove from fc->processing]
344 | | [copy write buffer to req]
345 | [woken up] | [wake up req->waitq]
346 | | <fuse_dev_write()
348 | <request_wait_answer() |
355 There are a couple of ways in which to deadlock a FUSE filesystem.
356 Since we are talking about unprivileged userspace programs,
357 something must be done about these.
359 Scenario 1 - Simple deadlock
360 -----------------------------
362 | "rm /mnt/fuse/file" | FUSE filesystem daemon
364 | >sys_unlink("/mnt/fuse/file") |
365 | [acquire inode semaphore |
368 | [sleep on req->waitq] |
370 | | >sys_unlink("/mnt/fuse/file")
371 | | [acquire inode semaphore
375 The solution for this is to allow the filesystem to be aborted.
377 Scenario 2 - Tricky deadlock
378 ----------------------------
380 This one needs a carefully crafted filesystem. It's a variation on
381 the above, only the call back to the filesystem is not explicit,
382 but is caused by a pagefault.
384 | Kamikaze filesystem thread 1 | Kamikaze filesystem thread 2
386 | [fd = open("/mnt/fuse/file")] | [request served normally]
387 | [mmap fd to 'addr'] |
388 | [close fd] | [FLUSH triggers 'magic' flag]
389 | [read a byte from addr] |
391 | [find or create page] |
394 | [queue READ request] |
395 | [sleep on req->waitq] |
396 | | [read request to buffer]
397 | | [create reply header before addr]
398 | | >sys_write(addr - headerlength)
399 | | >fuse_dev_write()
400 | | [look up req in fc->processing]
401 | | [remove from fc->processing]
402 | | [copy write buffer to req]
404 | | [find or create page]
408 Solution is basically the same as above.
410 An additional problem is that while the write buffer is being copied
411 to the request, the request must not be interrupted/aborted. This is
412 because the destination address of the copy may not be valid after the
413 request has returned.
415 This is solved with doing the copy atomically, and allowing abort
416 while the page(s) belonging to the write buffer are faulted with
417 get_user_pages(). The 'req->locked' flag indicates when the copy is
418 taking place, and abort is delayed until this flag is unset.