blob: 6b5741e651a2a11aef2590fb4b100030756bad1c [file] [log] [blame]
Miklos Szeredi06e1c5d2005-04-28 13:27:13 +00001Definitions
2~~~~~~~~~~~
3
4Userspace filesystem:
5
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.
9
10Filesystem daemon:
11
12 The process(es) providing the data and metadata of the filesystem.
13
14Non-privileged mount (or user mount):
15
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.
20
21Mount owner:
22
23 The user who does the mounting.
24
25User:
26
27 The user who is performing filesystem operations.
28
29What is FUSE?
30~~~~~~~~~~~~~
31
32FUSE is a userspace filesystem framework. It consists of a kernel
33module (fuse.ko), a userspace library (libfuse.*) and a mount utility
34(fusermount).
35
36One of the most important features of FUSE is allowing secure,
37non-privileged mounts. This opens up new possibilities for the use of
38filesystems. A good example is sshfs: a secure network filesystem
39using the sftp protocol.
40
Miklos Szerediac1b49d2005-07-05 13:25:52 +000041The userspace library and utilities are available from the FUSE
42homepage:
43
44 http://fuse.sourceforge.net/
45
Miklos Szeredibb9f5172005-07-03 18:03:11 +000046Mount options
47~~~~~~~~~~~~~
48
Miklos Szeredie2e23c92005-07-05 12:47:24 +000049'fd=N'
50
51 The file descriptor to use for communication between the userspace
52 filesystem and the kernel. The file descriptor must have been
53 obtained by opening the FUSE device ('/dev/fuse').
54
55'rootmode=M'
56
57 The file mode of the filesystem's root in octal representation.
58
59'user_id=N'
60
61 The numeric user id of the mount owner.
62
63'group_id=N'
64
65 The numeric group id of the mount owner.
66
Miklos Szeredibb9f5172005-07-03 18:03:11 +000067'default_permissions'
68
69 By default FUSE doesn't check file access permissions, the
70 filesystem is free to implement it's access policy or leave it to
71 the underlying file access mechanism (e.g. in case of network
72 filesystems). This option enables permission checking, restricting
73 access based on file mode. This is option is usually useful
74 together with the 'allow_other' mount option.
75
76'allow_other'
77
78 This option overrides the security measure restricting file access
79 to the user mounting the filesystem. This option is by default only
80 allowed to root, but this restriction can be removed with a
81 (userspace) configuration option.
82
Miklos Szeredibb9f5172005-07-03 18:03:11 +000083'max_read=N'
84
85 With this option the maximum size of read operations can be set.
86 The default is infinite. Note that the size of read requests is
87 limited anyway to 32 pages (which is 128kbyte on i386).
88
Miklos Szeredi06e1c5d2005-04-28 13:27:13 +000089How do non-privileged mounts work?
90~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
91
92Since the mount() system call is a privileged operation, a helper
93program (fusermount) is needed, which is installed setuid root.
94
95The implication of providing non-privileged mounts is that the mount
96owner must not be able to use this capability to compromise the
97system. Obvious requirements arising from this are:
98
99 A) mount owner should not be able to get elevated privileges with the
100 help of the mounted filesystem
101
102 B) mount owner should not get illegitimate access to information from
103 other users' and the super user's processes
104
105 C) mount owner should not be able to induce undesired behavior in
106 other users' or the super user's processes
107
108How are requirements fulfilled?
109~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
110
111 A) The mount owner could gain elevated privileges by either:
112
113 1) creating a filesystem containing a device file, then opening
114 this device
115
116 2) creating a filesystem containing a suid or sgid application,
117 then executing this application
118
119 The solution is not to allow opening device files and ignore
120 setuid and setgid bits when executing programs. To ensure this
121 fusermount always adds "nosuid" and "nodev" to the mount options
122 for non-privileged mounts.
123
124 B) If another user is accessing files or directories in the
125 filesystem, the filesystem daemon serving requests can record the
126 exact sequence and timing of operations performed. This
127 information is otherwise inaccessible to the mount owner, so this
128 counts as an information leak.
129
130 The solution to this problem will be presented in point 2) of C).
131
132 C) There are several ways in which the mount owner can induce
133 undesired behavior in other users' processes, such as:
134
135 1) mounting a filesystem over a file or directory which the mount
136 owner could otherwise not be able to modify (or could only
137 make limited modifications).
138
139 This is solved in fusermount, by checking the access
140 permissions on the mountpoint and only allowing the mount if
141 the mount owner can do unlimited modification (has write
142 access to the mountpoint, and mountpoint is not a "sticky"
143 directory)
144
145 2) Even if 1) is solved the mount owner can change the behavior
Miklos Szeredibaab6cb2005-04-30 07:59:09 +0000146 of other users' processes.
Miklos Szeredi06e1c5d2005-04-28 13:27:13 +0000147
Miklos Szeredie2e23c92005-07-05 12:47:24 +0000148 i) It can slow down or indefinitely delay the execution of a
Miklos Szeredi06e1c5d2005-04-28 13:27:13 +0000149 filesystem operation creating a DoS against the user or the
150 whole system. For example a suid application locking a
151 system file, and then accessing a file on the mount owner's
152 filesystem could be stopped, and thus causing the system
153 file to be locked forever.
154
Miklos Szeredie2e23c92005-07-05 12:47:24 +0000155 ii) It can present files or directories of unlimited length, or
Miklos Szeredi06e1c5d2005-04-28 13:27:13 +0000156 directory structures of unlimited depth, possibly causing a
157 system process to eat up diskspace, memory or other
158 resources, again causing DoS.
159
160 The solution to this as well as B) is not to allow processes
161 to access the filesystem, which could otherwise not be
162 monitored or manipulated by the mount owner. Since if the
163 mount owner can ptrace a process, it can do all of the above
164 without using a FUSE mount, the same criteria as used in
165 ptrace can be used to check if a process is allowed to access
166 the filesystem or not.
167
Miklos Szeredie2e23c92005-07-05 12:47:24 +0000168 Note that the ptrace check is not strictly necessary to
169 prevent B/2/i, it is enough to check if mount owner has enough
170 privilege to send signal to the process accessing the
171 filesystem, since SIGSTOP can be used to get a similar effect.
172
Miklos Szeredi06e1c5d2005-04-28 13:27:13 +0000173I think these limitations are unacceptable?
174~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
175
176If a sysadmin trusts the users enough, or can ensure through other
177measures, that system processes will never enter non-privileged
178mounts, it can relax the last limitation with a "user_allow_other"
179config option. If this config option is set, the mounting user can
180add the "allow_other" mount option which disables the check for other
181users' processes.
182
Miklos Szeredibaab6cb2005-04-30 07:59:09 +0000183Kernel - userspace interface
Miklos Szeredi06e1c5d2005-04-28 13:27:13 +0000184~~~~~~~~~~~~~~~~~~~~~~~~~~~~
185
Miklos Szeredi407e6a72005-03-25 12:19:43 +0000186The following diagram shows how a filesystem operation (in this
187example unlink) is performed in FUSE.
188
189NOTE: everything in this description is greatly simplified
190
191 | "rm /mnt/fuse/file" | FUSE filesystem daemon
192 | |
193 | | >sys_read()
194 | | >fuse_dev_read()
195 | | >request_wait()
196 | | [sleep on fc->waitq]
197 | |
198 | >sys_unlink() |
199 | >fuse_unlink() |
200 | [get request from |
201 | fc->unused_list] |
202 | >request_send() |
203 | [queue req on fc->pending] |
204 | [wake up fc->waitq] | [woken up]
205 | >request_wait_answer() |
206 | [sleep on req->waitq] |
207 | | <request_wait()
208 | | [remove req from fc->pending]
209 | | [copy req to read buffer]
210 | | [add req to fc->processing]
211 | | <fuse_dev_read()
212 | | <sys_read()
213 | |
214 | | [perform unlink]
215 | |
216 | | >sys_write()
217 | | >fuse_dev_write()
218 | | [look up req in fc->processing]
219 | | [remove from fc->processing]
220 | | [copy write buffer to req]
221 | [woken up] | [wake up req->waitq]
222 | | <fuse_dev_write()
223 | | <sys_write()
224 | <request_wait_answer() |
225 | <request_send() |
226 | [add request to |
227 | fc->unused_list] |
228 | <fuse_unlink() |
229 | <sys_unlink() |
230
231There are a couple of ways in which to deadlock a FUSE filesystem.
232Since we are talking about unprivileged userspace programs,
233something must be done about these.
234
235Scenario 1 - Simple deadlock
236-----------------------------
237
238 | "rm /mnt/fuse/file" | FUSE filesystem daemon
239 | |
240 | >sys_unlink("/mnt/fuse/file") |
241 | [acquire inode semaphore |
242 | for "file"] |
243 | >fuse_unlink() |
244 | [sleep on req->waitq] |
245 | | <sys_read()
246 | | >sys_unlink("/mnt/fuse/file")
247 | | [acquire inode semaphore
248 | | for "file"]
249 | | *DEADLOCK*
250
251The solution for this is to allow requests to be interrupted while
252they are in userspace:
253
254 | [interrupted by signal] |
255 | <fuse_unlink() |
256 | [release semaphore] | [semaphore acquired]
257 | <sys_unlink() |
258 | | >fuse_unlink()
259 | | [queue req on fc->pending]
260 | | [wake up fc->waitq]
261 | | [sleep on req->waitq]
262
263If the filesystem daemon was single threaded, this will stop here,
264since there's no other thread to dequeue and execute the request.
265In this case the solution is to kill the FUSE daemon as well. If
266there are multiple serving threads, you just have to kill them as
267long as any remain.
268
269Moral: a filesystem which deadlocks, can soon find itself dead.
270
271Scenario 2 - Tricky deadlock
272----------------------------
273
274This one needs a carefully crafted filesystem. It's a variation on
275the above, only the call back to the filesystem is not explicit,
276but is caused by a pagefault.
277
278 | Kamikaze filesystem thread 1 | Kamikaze filesystem thread 2
279 | |
280 | [fd = open("/mnt/fuse/file")] | [request served normally]
281 | [mmap fd to 'addr'] |
282 | [close fd] | [FLUSH triggers 'magic' flag]
283 | [read a byte from addr] |
284 | >do_page_fault() |
285 | [find or create page] |
286 | [lock page] |
287 | >fuse_readpage() |
288 | [queue READ request] |
289 | [sleep on req->waitq] |
290 | | [read request to buffer]
291 | | [create reply header before addr]
292 | | >sys_write(addr - headerlength)
293 | | >fuse_dev_write()
294 | | [look up req in fc->processing]
295 | | [remove from fc->processing]
296 | | [copy write buffer to req]
297 | | >do_page_fault()
298 | | [find or create page]
299 | | [lock page]
300 | | * DEADLOCK *
301
302Solution is again to let the the request be interrupted (not
303elaborated further).
304
305An additional problem is that while the write buffer is being
306copied to the request, the request must not be interrupted. This
307is because the destination address of the copy may not be valid
308after the request is interrupted.
309
310This is solved with doing the copy atomically, and allowing
311interruption while the page(s) belonging to the write buffer are
312faulted with get_user_pages(). The 'req->locked' flag indicates
313when the copy is taking place, and interruption is delayed until
314this flag is unset.
315