1 .. SPDX-License-Identifier: GPL-2.0
2
3 ====
4 FUSE
5 ====
6
7 Definitions
8 ===========
9
10 Userspace filesystem:
11 A filesystem in which data and metadata are
12 userspace process. The filesystem can be ac
13 the kernel interface.
14
15 Filesystem daemon:
16 The process(es) providing the data and metad
17
18 Non-privileged mount (or user mount):
19 A userspace filesystem mounted by a non-priv
20 The filesystem daemon is running with the pr
21 user. NOTE: this is not the same as mounts
22 option in /etc/fstab, which is not discussed
23
24 Filesystem connection:
25 A connection between the filesystem daemon a
26 connection exists until either the daemon di
27 umounted. Note that detaching (or lazy umou
28 does *not* break the connection, in this cas
29 the last reference to the filesystem is rele
30
31 Mount owner:
32 The user who does the mounting.
33
34 User:
35 The user who is performing filesystem operat
36
37 What is FUSE?
38 =============
39
40 FUSE is a userspace filesystem framework. It
41 module (fuse.ko), a userspace library (libfuse
42 (fusermount).
43
44 One of the most important features of FUSE is
45 non-privileged mounts. This opens up new poss
46 filesystems. A good example is sshfs: a secur
47 using the sftp protocol.
48
49 The userspace library and utilities are availa
50 `FUSE homepage: <https://github.com/libfuse/>`
51
52 Filesystem type
53 ===============
54
55 The filesystem type given to mount(2) can be o
56
57 fuse
58 This is the usual way to mount a FUSE fi
59 argument of the mount system call may co
60 which is not interpreted by the kernel.
61
62 fuseblk
63 The filesystem is block device based. T
64 mount system call is interpreted as the
65
66 Mount options
67 =============
68
69 fd=N
70 The file descriptor to use for communication
71 filesystem and the kernel. The file descrip
72 obtained by opening the FUSE device ('/dev/f
73
74 rootmode=M
75 The file mode of the filesystem's root in oc
76
77 user_id=N
78 The numeric user id of the mount owner.
79
80 group_id=N
81 The numeric group id of the mount owner.
82
83 default_permissions
84 By default FUSE doesn't check file access pe
85 filesystem is free to implement its access p
86 the underlying file access mechanism (e.g. i
87 filesystems). This option enables permissio
88 access based on file mode. It is usually us
89 'allow_other' mount option.
90
91 allow_other
92 This option overrides the security measure r
93 to the user mounting the filesystem. This o
94 allowed to root, but this restriction can be
95 (userspace) configuration option.
96
97 max_read=N
98 With this option the maximum size of read op
99 The default is infinite. Note that the size
100 limited anyway to 32 pages (which is 128kbyt
101
102 blksize=N
103 Set the block size for the filesystem. The
104 option is only valid for 'fuseblk' type moun
105
106 Control filesystem
107 ==================
108
109 There's a control filesystem for FUSE, which c
110
111 mount -t fusectl none /sys/fs/fuse/connectio
112
113 Mounting it under the '/sys/fs/fuse/connection
114 backwards compatible with earlier versions.
115
116 Under the fuse control filesystem each connect
117 named by a unique number.
118
119 For each connection the following files exist
120
121 waiting
122 The number of requests which are wai
123 userspace or being processed by the
124 no filesystem activity and 'waiting'
125 filesystem is hung or deadlocked.
126
127 abort
128 Writing anything into this file will
129 connection. This means that all wai
130 error returned for all aborted and n
131
132 Only the owner of the mount may read or write
133
134 Interrupting filesystem operations
135 ##################################
136
137 If a process issuing a FUSE filesystem request
138 following will happen:
139
140 - If the request is not yet sent to userspa
141 fatal (SIGKILL or unhandled fatal signal)
142 dequeued and returns immediately.
143
144 - If the request is not yet sent to userspa
145 fatal, then an interrupted flag is set fo
146 the request has been successfully transfe
147 this flag is set, an INTERRUPT request is
148
149 - If the request is already sent to userspa
150 request is queued.
151
152 INTERRUPT requests take precedence over other
153 userspace filesystem will receive queued INTER
154
155 The userspace filesystem may ignore the INTERR
156 or may honor them by sending a reply to the *o
157 the error set to EINTR.
158
159 It is also possible that there's a race betwee
160 original request and its INTERRUPT request. T
161
162 1. The INTERRUPT request is processed before
163 processed
164
165 2. The INTERRUPT request is processed after
166 been answered
167
168 If the filesystem cannot find the original req
169 some timeout and/or a number of new requests t
170 should reply to the INTERRUPT request with an
171 1) the INTERRUPT request will be requeued. In
172 reply will be ignored.
173
174 Aborting a filesystem connection
175 ================================
176
177 It is possible to get into certain situations
178 not responding. Reasons for this may be:
179
180 a) Broken userspace filesystem implementatio
181
182 b) Network connection down
183
184 c) Accidental deadlock
185
186 d) Malicious deadlock
187
188 (For more on c) and d) see later sections)
189
190 In either of these cases it may be useful to a
191 the filesystem. There are several ways to do
192
193 - Kill the filesystem daemon. Works in case
194
195 - Kill the filesystem daemon and all users o
196 in all cases except some malicious deadloc
197
198 - Use forced umount (umount -f). Works in a
199 filesystem is still attached (it hasn't be
200
201 - Abort filesystem through the FUSE control
202 powerful method, always works.
203
204 How do non-privileged mounts work?
205 ==================================
206
207 Since the mount() system call is a privileged
208 program (fusermount) is needed, which is insta
209
210 The implication of providing non-privileged mo
211 owner must not be able to use this capability
212 system. Obvious requirements arising from thi
213
214 A) mount owner should not be able to get elev
215 help of the mounted filesystem
216
217 B) mount owner should not get illegitimate ac
218 other users' and the super user's processe
219
220 C) mount owner should not be able to induce u
221 other users' or the super user's processes
222
223 How are requirements fulfilled?
224 ===============================
225
226 A) The mount owner could gain elevated privil
227
228 1. creating a filesystem containing a devi
229
230 2. creating a filesystem containing a suid
231
232 The solution is not to allow opening devic
233 setuid and setgid bits when executing prog
234 fusermount always adds "nosuid" and "nodev
235 for non-privileged mounts.
236
237 B) If another user is accessing files or dire
238 filesystem, the filesystem daemon serving
239 exact sequence and timing of operations pe
240 information is otherwise inaccessible to t
241 counts as an information leak.
242
243 The solution to this problem will be prese
244
245 C) There are several ways in which the mount
246 undesired behavior in other users' process
247
248 1) mounting a filesystem over a file or d
249 owner could otherwise not be able to m
250 make limited modifications).
251
252 This is solved in fusermount, by check
253 permissions on the mountpoint and only
254 the mount owner can do unlimited modif
255 access to the mountpoint, and mountpoi
256 directory)
257
258 2) Even if 1) is solved the mount owner c
259 of other users' processes.
260
261 i) It can slow down or indefinitely d
262 filesystem operation creating a Do
263 whole system. For example a suid
264 system file, and then accessing a
265 filesystem could be stopped, and t
266 file to be locked forever.
267
268 ii) It can present files or directori
269 directory structures of unlimited
270 system process to eat up diskspac
271 resources, again causing *DoS*.
272
273 The solution to this as well as B) is
274 to access the filesystem, which could
275 monitored or manipulated by the mount
276 mount owner can ptrace a process, it c
277 without using a FUSE mount, the same c
278 ptrace can be used to check if a proce
279 the filesystem or not.
280
281 Note that the *ptrace* check is not st
282 prevent C/2/i, it is enough to check i
283 privilege to send signal to the proces
284 filesystem, since *SIGSTOP* can be use
285
286 I think these limitations are unacceptable?
287 ===========================================
288
289 If a sysadmin trusts the users enough, or can
290 measures, that system processes will never ent
291 mounts, it can relax the last limitation in se
292
293 - With the 'user_allow_other' config option.
294 set, the mounting user can add the 'allow_
295 disables the check for other users' proces
296
297 User namespaces have an unintuitive intera
298 an unprivileged user - normally restricted
299 'allow_other' - could do so in a user name
300 privileged. If any process could access su
301 this would give the mounting user the abil
302 processes in user namespaces where they're
303 reason 'allow_other' restricts access to u
304 or a descendant.
305
306 - With the 'allow_sys_admin_access' module o
307 set, super user's processes have unrestric
308 irrespective of allow_other setting or use
309 mounting user.
310
311 Note that both of these relaxations expose the
312 information leak or *DoS* as described in poin
313 preceding section.
314
315 Kernel - userspace interface
316 ============================
317
318 The following diagram shows how a filesystem o
319 example unlink) is performed in FUSE. ::
320
321
322 | "rm /mnt/fuse/file" | FUSE
323 | |
324 | | >sys_
325 | | >fu
326 | | >
327 | |
328 | |
329 | >sys_unlink() |
330 | >fuse_unlink() |
331 | [get request from |
332 | fc->unused_list] |
333 | >request_send() |
334 | [queue req on fc->pending] |
335 | [wake up fc->waitq] |
336 | >request_wait_answer() |
337 | [sleep on req->waitq] |
338 | | <
339 | | [
340 | | [
341 | | [
342 | | <fu
343 | | <sys_
344 | |
345 | | [perf
346 | |
347 | | >sys_
348 | | >fu
349 | | [
350 | | [
351 | | [
352 | [woken up] | [
353 | | <fu
354 | | <sys_
355 | <request_wait_answer() |
356 | <request_send() |
357 | [add request to |
358 | fc->unused_list] |
359 | <fuse_unlink() |
360 | <sys_unlink() |
361
362 .. note:: Everything in the description above
363
364 There are a couple of ways in which to deadloc
365 Since we are talking about unprivileged usersp
366 something must be done about these.
367
368 **Scenario 1 - Simple deadlock**::
369
370 | "rm /mnt/fuse/file" | FUSE
371 | |
372 | >sys_unlink("/mnt/fuse/file") |
373 | [acquire inode semaphore |
374 | for "file"] |
375 | >fuse_unlink() |
376 | [sleep on req->waitq] |
377 | | <sys_
378 | | >sys_
379 | | [ac
380 | | fo
381 | | *DE
382
383 The solution for this is to allow the filesyst
384
385 **Scenario 2 - Tricky deadlock**
386
387
388 This one needs a carefully crafted filesystem.
389 the above, only the call back to the filesyste
390 but is caused by a pagefault. ::
391
392 | Kamikaze filesystem thread 1 | Kamik
393 | |
394 | [fd = open("/mnt/fuse/file")] | [requ
395 | [mmap fd to 'addr'] |
396 | [close fd] | [FLUS
397 | [read a byte from addr] |
398 | >do_page_fault() |
399 | [find or create page] |
400 | [lock page] |
401 | >fuse_readpage() |
402 | [queue READ request] |
403 | [sleep on req->waitq] |
404 | | [read
405 | | [crea
406 | | >sys_
407 | | >fu
408 | | [
409 | | [
410 | | [
411 | |
412 | |
413 | |
414 | |
415
416 The solution is basically the same as above.
417
418 An additional problem is that while the write
419 to the request, the request must not be interr
420 because the destination address of the copy ma
421 request has returned.
422
423 This is solved with doing the copy atomically,
424 while the page(s) belonging to the write buffe
425 get_user_pages(). The 'req->locked' flag indi
426 taking place, and abort is delayed until this
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.