]>
Commit | Line | Data |
---|---|---|
7c6b6602 | 1 | /* |
7387863d DDAG |
2 | * FUSE: Filesystem in Userspace |
3 | * Copyright (C) 2001-2007 Miklos Szeredi <miklos@szeredi.hu> | |
4 | * | |
5 | * This program can be distributed under the terms of the GNU GPLv2. | |
6 | * See the file COPYING. | |
7 | */ | |
7c6b6602 | 8 | |
7387863d | 9 | /* |
7c6b6602 DDAG |
10 | * |
11 | * This file system mirrors the existing file system hierarchy of the | |
12 | * system, starting at the root file system. This is implemented by | |
13 | * just "passing through" all requests to the corresponding user-space | |
14 | * libc functions. In contrast to passthrough.c and passthrough_fh.c, | |
15 | * this implementation uses the low-level API. Its performance should | |
16 | * be the least bad among the three, but many operations are not | |
17 | * implemented. In particular, it is not possible to remove files (or | |
18 | * directories) because the code necessary to defer actual removal | |
19 | * until the file is not opened anymore would make the example much | |
20 | * more complicated. | |
21 | * | |
22 | * When writeback caching is enabled (-o writeback mount option), it | |
23 | * is only possible to write to files for which the mounting user has | |
24 | * read permissions. This is because the writeback cache requires the | |
25 | * kernel to be able to issue read requests for all files (which the | |
26 | * passthrough filesystem cannot satisfy if it can't read the file in | |
27 | * the underlying filesystem). | |
28 | * | |
29 | * Compile with: | |
30 | * | |
7387863d DDAG |
31 | * gcc -Wall passthrough_ll.c `pkg-config fuse3 --cflags --libs` -o |
32 | * passthrough_ll | |
7c6b6602 DDAG |
33 | * |
34 | * ## Source code ## | |
35 | * \include passthrough_ll.c | |
36 | */ | |
37 | ||
09863ebc | 38 | #include "qemu/osdep.h" |
50fb955a | 39 | #include "qemu/timer.h" |
f6f3573c | 40 | #include "fuse_virtio.h" |
d240314a | 41 | #include "fuse_log.h" |
09863ebc | 42 | #include "fuse_lowlevel.h" |
7c6b6602 | 43 | #include <assert.h> |
2405f3c0 | 44 | #include <cap-ng.h> |
7387863d | 45 | #include <dirent.h> |
7c6b6602 | 46 | #include <errno.h> |
36f38469 | 47 | #include <glib.h> |
7c6b6602 | 48 | #include <inttypes.h> |
7387863d | 49 | #include <limits.h> |
7c6b6602 | 50 | #include <pthread.h> |
7387863d DDAG |
51 | #include <stdbool.h> |
52 | #include <stddef.h> | |
53 | #include <stdio.h> | |
54 | #include <stdlib.h> | |
55 | #include <string.h> | |
7c6b6602 | 56 | #include <sys/file.h> |
5baa3b8e | 57 | #include <sys/mount.h> |
8e1d4ef2 | 58 | #include <sys/prctl.h> |
01a6dc95 | 59 | #include <sys/resource.h> |
929cfb7a | 60 | #include <sys/syscall.h> |
8e1d4ef2 SH |
61 | #include <sys/types.h> |
62 | #include <sys/wait.h> | |
7c6b6602 | 63 | #include <sys/xattr.h> |
f185621d | 64 | #include <syslog.h> |
7387863d | 65 | #include <unistd.h> |
7c6b6602 DDAG |
66 | |
67 | #include "passthrough_helpers.h" | |
4f8bde99 | 68 | #include "seccomp.h" |
7c6b6602 | 69 | |
25c13572 SH |
70 | struct lo_map_elem { |
71 | union { | |
92fb57b8 | 72 | struct lo_inode *inode; |
b39bce12 | 73 | struct lo_dirp *dirp; |
73b4d19d | 74 | int fd; |
25c13572 SH |
75 | ssize_t freelist; |
76 | }; | |
77 | bool in_use; | |
78 | }; | |
79 | ||
80 | /* Maps FUSE fh or ino values to internal objects */ | |
81 | struct lo_map { | |
82 | struct lo_map_elem *elems; | |
83 | size_t nelems; | |
84 | ssize_t freelist; | |
85 | }; | |
86 | ||
7c6b6602 | 87 | struct lo_inode { |
7387863d DDAG |
88 | struct lo_inode *next; /* protected by lo->mutex */ |
89 | struct lo_inode *prev; /* protected by lo->mutex */ | |
90 | int fd; | |
91 | bool is_symlink; | |
92 | ino_t ino; | |
93 | dev_t dev; | |
94 | uint64_t refcount; /* protected by lo->mutex */ | |
92fb57b8 | 95 | fuse_ino_t fuse_ino; |
7c6b6602 DDAG |
96 | }; |
97 | ||
929cfb7a VG |
98 | struct lo_cred { |
99 | uid_t euid; | |
100 | gid_t egid; | |
101 | }; | |
102 | ||
7c6b6602 | 103 | enum { |
7387863d DDAG |
104 | CACHE_NEVER, |
105 | CACHE_NORMAL, | |
106 | CACHE_ALWAYS, | |
7c6b6602 DDAG |
107 | }; |
108 | ||
109 | struct lo_data { | |
7387863d DDAG |
110 | pthread_mutex_t mutex; |
111 | int debug; | |
5fe319a7 | 112 | int norace; |
7387863d DDAG |
113 | int writeback; |
114 | int flock; | |
115 | int xattr; | |
116 | const char *source; | |
117 | double timeout; | |
118 | int cache; | |
119 | int timeout_set; | |
59aef494 MS |
120 | int readdirplus_set; |
121 | int readdirplus_clear; | |
7387863d | 122 | struct lo_inode root; /* protected by lo->mutex */ |
92fb57b8 | 123 | struct lo_map ino_map; /* protected by lo->mutex */ |
b39bce12 | 124 | struct lo_map dirp_map; /* protected by lo->mutex */ |
73b4d19d | 125 | struct lo_map fd_map; /* protected by lo->mutex */ |
9f59d175 SH |
126 | |
127 | /* An O_PATH file descriptor to /proc/self/fd/ */ | |
128 | int proc_self_fd; | |
7c6b6602 DDAG |
129 | }; |
130 | ||
131 | static const struct fuse_opt lo_opts[] = { | |
7387863d DDAG |
132 | { "writeback", offsetof(struct lo_data, writeback), 1 }, |
133 | { "no_writeback", offsetof(struct lo_data, writeback), 0 }, | |
134 | { "source=%s", offsetof(struct lo_data, source), 0 }, | |
135 | { "flock", offsetof(struct lo_data, flock), 1 }, | |
136 | { "no_flock", offsetof(struct lo_data, flock), 0 }, | |
137 | { "xattr", offsetof(struct lo_data, xattr), 1 }, | |
138 | { "no_xattr", offsetof(struct lo_data, xattr), 0 }, | |
139 | { "timeout=%lf", offsetof(struct lo_data, timeout), 0 }, | |
140 | { "timeout=", offsetof(struct lo_data, timeout_set), 1 }, | |
141 | { "cache=never", offsetof(struct lo_data, cache), CACHE_NEVER }, | |
142 | { "cache=auto", offsetof(struct lo_data, cache), CACHE_NORMAL }, | |
143 | { "cache=always", offsetof(struct lo_data, cache), CACHE_ALWAYS }, | |
5fe319a7 | 144 | { "norace", offsetof(struct lo_data, norace), 1 }, |
59aef494 MS |
145 | { "readdirplus", offsetof(struct lo_data, readdirplus_set), 1 }, |
146 | { "no_readdirplus", offsetof(struct lo_data, readdirplus_clear), 1 }, | |
7387863d | 147 | FUSE_OPT_END |
7c6b6602 | 148 | }; |
f185621d | 149 | static bool use_syslog = false; |
d240314a | 150 | static int current_log_level; |
95d27157 MS |
151 | static void unref_inode_lolocked(struct lo_data *lo, struct lo_inode *inode, |
152 | uint64_t n); | |
5fe319a7 | 153 | |
2405f3c0 DDAG |
154 | static struct { |
155 | pthread_mutex_t mutex; | |
156 | void *saved; | |
157 | } cap; | |
158 | /* That we loaded cap-ng in the current thread from the saved */ | |
159 | static __thread bool cap_loaded = 0; | |
160 | ||
5fe319a7 MS |
161 | static struct lo_inode *lo_find(struct lo_data *lo, struct stat *st); |
162 | ||
25dae28c SH |
163 | static int is_dot_or_dotdot(const char *name) |
164 | { | |
165 | return name[0] == '.' && | |
166 | (name[1] == '\0' || (name[1] == '.' && name[2] == '\0')); | |
167 | } | |
168 | ||
169 | /* Is `path` a single path component that is not "." or ".."? */ | |
170 | static int is_safe_path_component(const char *path) | |
171 | { | |
172 | if (strchr(path, '/')) { | |
173 | return 0; | |
174 | } | |
175 | ||
176 | return !is_dot_or_dotdot(path); | |
177 | } | |
5fe319a7 | 178 | |
7c6b6602 DDAG |
179 | static struct lo_data *lo_data(fuse_req_t req) |
180 | { | |
7387863d | 181 | return (struct lo_data *)fuse_req_userdata(req); |
7c6b6602 DDAG |
182 | } |
183 | ||
2405f3c0 DDAG |
184 | /* |
185 | * Load capng's state from our saved state if the current thread | |
186 | * hadn't previously been loaded. | |
187 | * returns 0 on success | |
188 | */ | |
189 | static int load_capng(void) | |
190 | { | |
191 | if (!cap_loaded) { | |
192 | pthread_mutex_lock(&cap.mutex); | |
193 | capng_restore_state(&cap.saved); | |
194 | /* | |
195 | * restore_state free's the saved copy | |
196 | * so make another. | |
197 | */ | |
198 | cap.saved = capng_save_state(); | |
199 | if (!cap.saved) { | |
200 | fuse_log(FUSE_LOG_ERR, "capng_save_state (thread)\n"); | |
201 | return -EINVAL; | |
202 | } | |
203 | pthread_mutex_unlock(&cap.mutex); | |
204 | ||
205 | /* | |
206 | * We want to use the loaded state for our pid, | |
207 | * not the original | |
208 | */ | |
209 | capng_setpid(syscall(SYS_gettid)); | |
210 | cap_loaded = true; | |
211 | } | |
212 | return 0; | |
213 | } | |
214 | ||
ee884652 VG |
215 | /* |
216 | * Helpers for dropping and regaining effective capabilities. Returns 0 | |
217 | * on success, error otherwise | |
218 | */ | |
219 | static int drop_effective_cap(const char *cap_name, bool *cap_dropped) | |
220 | { | |
221 | int cap, ret; | |
222 | ||
223 | cap = capng_name_to_capability(cap_name); | |
224 | if (cap < 0) { | |
225 | ret = errno; | |
226 | fuse_log(FUSE_LOG_ERR, "capng_name_to_capability(%s) failed:%s\n", | |
227 | cap_name, strerror(errno)); | |
228 | goto out; | |
229 | } | |
230 | ||
231 | if (load_capng()) { | |
232 | ret = errno; | |
233 | fuse_log(FUSE_LOG_ERR, "load_capng() failed\n"); | |
234 | goto out; | |
235 | } | |
236 | ||
237 | /* We dont have this capability in effective set already. */ | |
238 | if (!capng_have_capability(CAPNG_EFFECTIVE, cap)) { | |
239 | ret = 0; | |
240 | goto out; | |
241 | } | |
242 | ||
243 | if (capng_update(CAPNG_DROP, CAPNG_EFFECTIVE, cap)) { | |
244 | ret = errno; | |
245 | fuse_log(FUSE_LOG_ERR, "capng_update(DROP,) failed\n"); | |
246 | goto out; | |
247 | } | |
248 | ||
249 | if (capng_apply(CAPNG_SELECT_CAPS)) { | |
250 | ret = errno; | |
251 | fuse_log(FUSE_LOG_ERR, "drop:capng_apply() failed\n"); | |
252 | goto out; | |
253 | } | |
254 | ||
255 | ret = 0; | |
256 | if (cap_dropped) { | |
257 | *cap_dropped = true; | |
258 | } | |
259 | ||
260 | out: | |
261 | return ret; | |
262 | } | |
263 | ||
264 | static int gain_effective_cap(const char *cap_name) | |
265 | { | |
266 | int cap; | |
267 | int ret = 0; | |
268 | ||
269 | cap = capng_name_to_capability(cap_name); | |
270 | if (cap < 0) { | |
271 | ret = errno; | |
272 | fuse_log(FUSE_LOG_ERR, "capng_name_to_capability(%s) failed:%s\n", | |
273 | cap_name, strerror(errno)); | |
274 | goto out; | |
275 | } | |
276 | ||
277 | if (load_capng()) { | |
278 | ret = errno; | |
279 | fuse_log(FUSE_LOG_ERR, "load_capng() failed\n"); | |
280 | goto out; | |
281 | } | |
282 | ||
283 | if (capng_update(CAPNG_ADD, CAPNG_EFFECTIVE, cap)) { | |
284 | ret = errno; | |
285 | fuse_log(FUSE_LOG_ERR, "capng_update(ADD,) failed\n"); | |
286 | goto out; | |
287 | } | |
288 | ||
289 | if (capng_apply(CAPNG_SELECT_CAPS)) { | |
290 | ret = errno; | |
291 | fuse_log(FUSE_LOG_ERR, "gain:capng_apply() failed\n"); | |
292 | goto out; | |
293 | } | |
294 | ret = 0; | |
295 | ||
296 | out: | |
297 | return ret; | |
298 | } | |
299 | ||
92fb57b8 | 300 | static void lo_map_init(struct lo_map *map) |
25c13572 SH |
301 | { |
302 | map->elems = NULL; | |
303 | map->nelems = 0; | |
304 | map->freelist = -1; | |
305 | } | |
306 | ||
92fb57b8 | 307 | static void lo_map_destroy(struct lo_map *map) |
25c13572 SH |
308 | { |
309 | free(map->elems); | |
310 | } | |
311 | ||
312 | static int lo_map_grow(struct lo_map *map, size_t new_nelems) | |
313 | { | |
314 | struct lo_map_elem *new_elems; | |
315 | size_t i; | |
316 | ||
317 | if (new_nelems <= map->nelems) { | |
318 | return 1; | |
319 | } | |
320 | ||
321 | new_elems = realloc(map->elems, sizeof(map->elems[0]) * new_nelems); | |
322 | if (!new_elems) { | |
323 | return 0; | |
324 | } | |
325 | ||
326 | for (i = map->nelems; i < new_nelems; i++) { | |
327 | new_elems[i].freelist = i + 1; | |
328 | new_elems[i].in_use = false; | |
329 | } | |
330 | new_elems[new_nelems - 1].freelist = -1; | |
331 | ||
332 | map->elems = new_elems; | |
333 | map->freelist = map->nelems; | |
334 | map->nelems = new_nelems; | |
335 | return 1; | |
336 | } | |
337 | ||
92fb57b8 | 338 | static struct lo_map_elem *lo_map_alloc_elem(struct lo_map *map) |
25c13572 SH |
339 | { |
340 | struct lo_map_elem *elem; | |
341 | ||
342 | if (map->freelist == -1 && !lo_map_grow(map, map->nelems + 256)) { | |
343 | return NULL; | |
344 | } | |
345 | ||
346 | elem = &map->elems[map->freelist]; | |
347 | map->freelist = elem->freelist; | |
348 | ||
349 | elem->in_use = true; | |
350 | ||
351 | return elem; | |
352 | } | |
353 | ||
92fb57b8 | 354 | static struct lo_map_elem *lo_map_reserve(struct lo_map *map, size_t key) |
25c13572 SH |
355 | { |
356 | ssize_t *prev; | |
357 | ||
358 | if (!lo_map_grow(map, key + 1)) { | |
359 | return NULL; | |
360 | } | |
361 | ||
362 | for (prev = &map->freelist; *prev != -1; | |
363 | prev = &map->elems[*prev].freelist) { | |
364 | if (*prev == key) { | |
365 | struct lo_map_elem *elem = &map->elems[key]; | |
366 | ||
367 | *prev = elem->freelist; | |
368 | elem->in_use = true; | |
369 | return elem; | |
370 | } | |
371 | } | |
372 | return NULL; | |
373 | } | |
374 | ||
92fb57b8 | 375 | static struct lo_map_elem *lo_map_get(struct lo_map *map, size_t key) |
25c13572 SH |
376 | { |
377 | if (key >= map->nelems) { | |
378 | return NULL; | |
379 | } | |
380 | if (!map->elems[key].in_use) { | |
381 | return NULL; | |
382 | } | |
383 | return &map->elems[key]; | |
384 | } | |
385 | ||
92fb57b8 | 386 | static void lo_map_remove(struct lo_map *map, size_t key) |
25c13572 SH |
387 | { |
388 | struct lo_map_elem *elem; | |
389 | ||
390 | if (key >= map->nelems) { | |
391 | return; | |
392 | } | |
393 | ||
394 | elem = &map->elems[key]; | |
395 | if (!elem->in_use) { | |
396 | return; | |
397 | } | |
398 | ||
399 | elem->in_use = false; | |
400 | ||
401 | elem->freelist = map->freelist; | |
402 | map->freelist = key; | |
403 | } | |
404 | ||
73b4d19d SH |
405 | /* Assumes lo->mutex is held */ |
406 | static ssize_t lo_add_fd_mapping(fuse_req_t req, int fd) | |
407 | { | |
408 | struct lo_map_elem *elem; | |
409 | ||
410 | elem = lo_map_alloc_elem(&lo_data(req)->fd_map); | |
411 | if (!elem) { | |
412 | return -1; | |
413 | } | |
414 | ||
415 | elem->fd = fd; | |
416 | return elem - lo_data(req)->fd_map.elems; | |
417 | } | |
418 | ||
b39bce12 SH |
419 | /* Assumes lo->mutex is held */ |
420 | static ssize_t lo_add_dirp_mapping(fuse_req_t req, struct lo_dirp *dirp) | |
421 | { | |
422 | struct lo_map_elem *elem; | |
423 | ||
424 | elem = lo_map_alloc_elem(&lo_data(req)->dirp_map); | |
425 | if (!elem) { | |
426 | return -1; | |
427 | } | |
428 | ||
429 | elem->dirp = dirp; | |
430 | return elem - lo_data(req)->dirp_map.elems; | |
431 | } | |
432 | ||
92fb57b8 SH |
433 | /* Assumes lo->mutex is held */ |
434 | static ssize_t lo_add_inode_mapping(fuse_req_t req, struct lo_inode *inode) | |
435 | { | |
436 | struct lo_map_elem *elem; | |
437 | ||
438 | elem = lo_map_alloc_elem(&lo_data(req)->ino_map); | |
439 | if (!elem) { | |
440 | return -1; | |
441 | } | |
442 | ||
443 | elem->inode = inode; | |
444 | return elem - lo_data(req)->ino_map.elems; | |
445 | } | |
446 | ||
7c6b6602 DDAG |
447 | static struct lo_inode *lo_inode(fuse_req_t req, fuse_ino_t ino) |
448 | { | |
92fb57b8 SH |
449 | struct lo_data *lo = lo_data(req); |
450 | struct lo_map_elem *elem; | |
451 | ||
452 | pthread_mutex_lock(&lo->mutex); | |
453 | elem = lo_map_get(&lo->ino_map, ino); | |
454 | pthread_mutex_unlock(&lo->mutex); | |
455 | ||
456 | if (!elem) { | |
457 | return NULL; | |
7387863d | 458 | } |
92fb57b8 SH |
459 | |
460 | return elem->inode; | |
7c6b6602 DDAG |
461 | } |
462 | ||
463 | static int lo_fd(fuse_req_t req, fuse_ino_t ino) | |
464 | { | |
92fb57b8 SH |
465 | struct lo_inode *inode = lo_inode(req, ino); |
466 | return inode ? inode->fd : -1; | |
7c6b6602 DDAG |
467 | } |
468 | ||
7387863d | 469 | static void lo_init(void *userdata, struct fuse_conn_info *conn) |
7c6b6602 | 470 | { |
7387863d DDAG |
471 | struct lo_data *lo = (struct lo_data *)userdata; |
472 | ||
473 | if (conn->capable & FUSE_CAP_EXPORT_SUPPORT) { | |
474 | conn->want |= FUSE_CAP_EXPORT_SUPPORT; | |
475 | } | |
476 | ||
477 | if (lo->writeback && conn->capable & FUSE_CAP_WRITEBACK_CACHE) { | |
d240314a | 478 | fuse_log(FUSE_LOG_DEBUG, "lo_init: activating writeback\n"); |
7387863d DDAG |
479 | conn->want |= FUSE_CAP_WRITEBACK_CACHE; |
480 | } | |
481 | if (lo->flock && conn->capable & FUSE_CAP_FLOCK_LOCKS) { | |
d240314a | 482 | fuse_log(FUSE_LOG_DEBUG, "lo_init: activating flock locks\n"); |
7387863d DDAG |
483 | conn->want |= FUSE_CAP_FLOCK_LOCKS; |
484 | } | |
59aef494 MS |
485 | if ((lo->cache == CACHE_NEVER && !lo->readdirplus_set) || |
486 | lo->readdirplus_clear) { | |
ddcbabcb MS |
487 | fuse_log(FUSE_LOG_DEBUG, "lo_init: disabling readdirplus\n"); |
488 | conn->want &= ~FUSE_CAP_READDIRPLUS; | |
489 | } | |
7c6b6602 DDAG |
490 | } |
491 | ||
492 | static void lo_getattr(fuse_req_t req, fuse_ino_t ino, | |
7387863d | 493 | struct fuse_file_info *fi) |
7c6b6602 | 494 | { |
7387863d DDAG |
495 | int res; |
496 | struct stat buf; | |
497 | struct lo_data *lo = lo_data(req); | |
7c6b6602 | 498 | |
7387863d | 499 | (void)fi; |
7c6b6602 | 500 | |
7387863d DDAG |
501 | res = |
502 | fstatat(lo_fd(req, ino), "", &buf, AT_EMPTY_PATH | AT_SYMLINK_NOFOLLOW); | |
503 | if (res == -1) { | |
504 | return (void)fuse_reply_err(req, errno); | |
505 | } | |
7c6b6602 | 506 | |
7387863d | 507 | fuse_reply_attr(req, &buf, lo->timeout); |
7c6b6602 DDAG |
508 | } |
509 | ||
5fe319a7 MS |
510 | static int lo_parent_and_name(struct lo_data *lo, struct lo_inode *inode, |
511 | char path[PATH_MAX], struct lo_inode **parent) | |
7c6b6602 | 512 | { |
7387863d | 513 | char procname[64]; |
5fe319a7 MS |
514 | char *last; |
515 | struct stat stat; | |
516 | struct lo_inode *p; | |
517 | int retries = 2; | |
518 | int res; | |
519 | ||
520 | retry: | |
9f59d175 | 521 | sprintf(procname, "%i", inode->fd); |
5fe319a7 | 522 | |
9f59d175 | 523 | res = readlinkat(lo->proc_self_fd, procname, path, PATH_MAX); |
5fe319a7 MS |
524 | if (res < 0) { |
525 | fuse_log(FUSE_LOG_WARNING, "%s: readlink failed: %m\n", __func__); | |
526 | goto fail_noretry; | |
527 | } | |
528 | ||
529 | if (res >= PATH_MAX) { | |
530 | fuse_log(FUSE_LOG_WARNING, "%s: readlink overflowed\n", __func__); | |
531 | goto fail_noretry; | |
532 | } | |
533 | path[res] = '\0'; | |
534 | ||
535 | last = strrchr(path, '/'); | |
536 | if (last == NULL) { | |
537 | /* Shouldn't happen */ | |
538 | fuse_log( | |
539 | FUSE_LOG_WARNING, | |
540 | "%s: INTERNAL ERROR: bad path read from proc\n", __func__); | |
541 | goto fail_noretry; | |
542 | } | |
543 | if (last == path) { | |
544 | p = &lo->root; | |
545 | pthread_mutex_lock(&lo->mutex); | |
546 | p->refcount++; | |
547 | pthread_mutex_unlock(&lo->mutex); | |
548 | } else { | |
549 | *last = '\0'; | |
550 | res = fstatat(AT_FDCWD, last == path ? "/" : path, &stat, 0); | |
551 | if (res == -1) { | |
552 | if (!retries) { | |
553 | fuse_log(FUSE_LOG_WARNING, | |
554 | "%s: failed to stat parent: %m\n", __func__); | |
555 | } | |
556 | goto fail; | |
557 | } | |
558 | p = lo_find(lo, &stat); | |
559 | if (p == NULL) { | |
560 | if (!retries) { | |
561 | fuse_log(FUSE_LOG_WARNING, | |
562 | "%s: failed to find parent\n", __func__); | |
563 | } | |
564 | goto fail; | |
565 | } | |
566 | } | |
567 | last++; | |
568 | res = fstatat(p->fd, last, &stat, AT_SYMLINK_NOFOLLOW); | |
569 | if (res == -1) { | |
570 | if (!retries) { | |
571 | fuse_log(FUSE_LOG_WARNING, | |
572 | "%s: failed to stat last\n", __func__); | |
573 | } | |
574 | goto fail_unref; | |
575 | } | |
576 | if (stat.st_dev != inode->dev || stat.st_ino != inode->ino) { | |
577 | if (!retries) { | |
578 | fuse_log(FUSE_LOG_WARNING, | |
579 | "%s: failed to match last\n", __func__); | |
580 | } | |
581 | goto fail_unref; | |
582 | } | |
583 | *parent = p; | |
584 | memmove(path, last, strlen(last) + 1); | |
585 | ||
586 | return 0; | |
587 | ||
588 | fail_unref: | |
95d27157 | 589 | unref_inode_lolocked(lo, p, 1); |
5fe319a7 MS |
590 | fail: |
591 | if (retries) { | |
592 | retries--; | |
593 | goto retry; | |
594 | } | |
595 | fail_noretry: | |
596 | errno = EIO; | |
597 | return -1; | |
598 | } | |
599 | ||
600 | static int utimensat_empty(struct lo_data *lo, struct lo_inode *inode, | |
601 | const struct timespec *tv) | |
602 | { | |
603 | int res; | |
604 | struct lo_inode *parent; | |
605 | char path[PATH_MAX]; | |
7387863d DDAG |
606 | |
607 | if (inode->is_symlink) { | |
5fe319a7 | 608 | res = utimensat(inode->fd, "", tv, AT_EMPTY_PATH); |
7387863d DDAG |
609 | if (res == -1 && errno == EINVAL) { |
610 | /* Sorry, no race free way to set times on symlink. */ | |
5fe319a7 MS |
611 | if (lo->norace) { |
612 | errno = EPERM; | |
613 | } else { | |
614 | goto fallback; | |
615 | } | |
7387863d DDAG |
616 | } |
617 | return res; | |
618 | } | |
9f59d175 | 619 | sprintf(path, "%i", inode->fd); |
5fe319a7 | 620 | |
9f59d175 | 621 | return utimensat(lo->proc_self_fd, path, tv, 0); |
7387863d | 622 | |
5fe319a7 MS |
623 | fallback: |
624 | res = lo_parent_and_name(lo, inode, path, &parent); | |
625 | if (res != -1) { | |
626 | res = utimensat(parent->fd, path, tv, AT_SYMLINK_NOFOLLOW); | |
95d27157 | 627 | unref_inode_lolocked(lo, parent, 1); |
5fe319a7 MS |
628 | } |
629 | ||
630 | return res; | |
7c6b6602 DDAG |
631 | } |
632 | ||
73b4d19d SH |
633 | static int lo_fi_fd(fuse_req_t req, struct fuse_file_info *fi) |
634 | { | |
635 | struct lo_data *lo = lo_data(req); | |
636 | struct lo_map_elem *elem; | |
637 | ||
638 | pthread_mutex_lock(&lo->mutex); | |
639 | elem = lo_map_get(&lo->fd_map, fi->fh); | |
640 | pthread_mutex_unlock(&lo->mutex); | |
641 | ||
642 | if (!elem) { | |
643 | return -1; | |
644 | } | |
645 | ||
646 | return elem->fd; | |
647 | } | |
648 | ||
7c6b6602 | 649 | static void lo_setattr(fuse_req_t req, fuse_ino_t ino, struct stat *attr, |
7387863d | 650 | int valid, struct fuse_file_info *fi) |
7c6b6602 | 651 | { |
7387863d DDAG |
652 | int saverr; |
653 | char procname[64]; | |
5fe319a7 | 654 | struct lo_data *lo = lo_data(req); |
92fb57b8 SH |
655 | struct lo_inode *inode; |
656 | int ifd; | |
7387863d | 657 | int res; |
73b4d19d | 658 | int fd; |
7387863d | 659 | |
92fb57b8 SH |
660 | inode = lo_inode(req, ino); |
661 | if (!inode) { | |
662 | fuse_reply_err(req, EBADF); | |
663 | return; | |
664 | } | |
665 | ||
666 | ifd = inode->fd; | |
667 | ||
73b4d19d SH |
668 | /* If fi->fh is invalid we'll report EBADF later */ |
669 | if (fi) { | |
670 | fd = lo_fi_fd(req, fi); | |
671 | } | |
672 | ||
7387863d DDAG |
673 | if (valid & FUSE_SET_ATTR_MODE) { |
674 | if (fi) { | |
73b4d19d | 675 | res = fchmod(fd, attr->st_mode); |
7387863d | 676 | } else { |
9f59d175 SH |
677 | sprintf(procname, "%i", ifd); |
678 | res = fchmodat(lo->proc_self_fd, procname, attr->st_mode, 0); | |
7387863d DDAG |
679 | } |
680 | if (res == -1) { | |
681 | goto out_err; | |
682 | } | |
683 | } | |
684 | if (valid & (FUSE_SET_ATTR_UID | FUSE_SET_ATTR_GID)) { | |
685 | uid_t uid = (valid & FUSE_SET_ATTR_UID) ? attr->st_uid : (uid_t)-1; | |
686 | gid_t gid = (valid & FUSE_SET_ATTR_GID) ? attr->st_gid : (gid_t)-1; | |
687 | ||
688 | res = fchownat(ifd, "", uid, gid, AT_EMPTY_PATH | AT_SYMLINK_NOFOLLOW); | |
689 | if (res == -1) { | |
690 | goto out_err; | |
691 | } | |
692 | } | |
693 | if (valid & FUSE_SET_ATTR_SIZE) { | |
9f59d175 SH |
694 | int truncfd; |
695 | ||
7387863d | 696 | if (fi) { |
9f59d175 | 697 | truncfd = fd; |
7387863d | 698 | } else { |
9f59d175 SH |
699 | sprintf(procname, "%i", ifd); |
700 | truncfd = openat(lo->proc_self_fd, procname, O_RDWR); | |
701 | if (truncfd < 0) { | |
702 | goto out_err; | |
703 | } | |
704 | } | |
705 | ||
706 | res = ftruncate(truncfd, attr->st_size); | |
707 | if (!fi) { | |
708 | saverr = errno; | |
709 | close(truncfd); | |
710 | errno = saverr; | |
7387863d DDAG |
711 | } |
712 | if (res == -1) { | |
713 | goto out_err; | |
714 | } | |
715 | } | |
716 | if (valid & (FUSE_SET_ATTR_ATIME | FUSE_SET_ATTR_MTIME)) { | |
717 | struct timespec tv[2]; | |
718 | ||
719 | tv[0].tv_sec = 0; | |
720 | tv[1].tv_sec = 0; | |
721 | tv[0].tv_nsec = UTIME_OMIT; | |
722 | tv[1].tv_nsec = UTIME_OMIT; | |
723 | ||
724 | if (valid & FUSE_SET_ATTR_ATIME_NOW) { | |
725 | tv[0].tv_nsec = UTIME_NOW; | |
726 | } else if (valid & FUSE_SET_ATTR_ATIME) { | |
727 | tv[0] = attr->st_atim; | |
728 | } | |
729 | ||
730 | if (valid & FUSE_SET_ATTR_MTIME_NOW) { | |
731 | tv[1].tv_nsec = UTIME_NOW; | |
732 | } else if (valid & FUSE_SET_ATTR_MTIME) { | |
733 | tv[1] = attr->st_mtim; | |
734 | } | |
735 | ||
736 | if (fi) { | |
73b4d19d | 737 | res = futimens(fd, tv); |
7387863d | 738 | } else { |
5fe319a7 | 739 | res = utimensat_empty(lo, inode, tv); |
7387863d DDAG |
740 | } |
741 | if (res == -1) { | |
742 | goto out_err; | |
743 | } | |
744 | } | |
745 | ||
746 | return lo_getattr(req, ino, fi); | |
7c6b6602 DDAG |
747 | |
748 | out_err: | |
7387863d DDAG |
749 | saverr = errno; |
750 | fuse_reply_err(req, saverr); | |
7c6b6602 DDAG |
751 | } |
752 | ||
753 | static struct lo_inode *lo_find(struct lo_data *lo, struct stat *st) | |
754 | { | |
7387863d DDAG |
755 | struct lo_inode *p; |
756 | struct lo_inode *ret = NULL; | |
757 | ||
758 | pthread_mutex_lock(&lo->mutex); | |
759 | for (p = lo->root.next; p != &lo->root; p = p->next) { | |
760 | if (p->ino == st->st_ino && p->dev == st->st_dev) { | |
761 | assert(p->refcount > 0); | |
762 | ret = p; | |
763 | ret->refcount++; | |
764 | break; | |
765 | } | |
766 | } | |
767 | pthread_mutex_unlock(&lo->mutex); | |
768 | return ret; | |
7c6b6602 DDAG |
769 | } |
770 | ||
771 | static int lo_do_lookup(fuse_req_t req, fuse_ino_t parent, const char *name, | |
7387863d | 772 | struct fuse_entry_param *e) |
7c6b6602 | 773 | { |
7387863d DDAG |
774 | int newfd; |
775 | int res; | |
776 | int saverr; | |
777 | struct lo_data *lo = lo_data(req); | |
854684bc | 778 | struct lo_inode *inode, *dir = lo_inode(req, parent); |
7387863d | 779 | |
9de4fab5 MS |
780 | /* |
781 | * name_to_handle_at() and open_by_handle_at() can reach here with fuse | |
782 | * mount point in guest, but we don't have its inode info in the | |
783 | * ino_map. | |
784 | */ | |
785 | if (!dir) { | |
786 | return ENOENT; | |
787 | } | |
788 | ||
7387863d DDAG |
789 | memset(e, 0, sizeof(*e)); |
790 | e->attr_timeout = lo->timeout; | |
791 | e->entry_timeout = lo->timeout; | |
792 | ||
854684bc SH |
793 | /* Do not allow escaping root directory */ |
794 | if (dir == &lo->root && strcmp(name, "..") == 0) { | |
795 | name = "."; | |
796 | } | |
797 | ||
9de4fab5 | 798 | newfd = openat(dir->fd, name, O_PATH | O_NOFOLLOW); |
7387863d DDAG |
799 | if (newfd == -1) { |
800 | goto out_err; | |
801 | } | |
802 | ||
803 | res = fstatat(newfd, "", &e->attr, AT_EMPTY_PATH | AT_SYMLINK_NOFOLLOW); | |
804 | if (res == -1) { | |
805 | goto out_err; | |
806 | } | |
807 | ||
9de4fab5 | 808 | inode = lo_find(lo, &e->attr); |
7387863d DDAG |
809 | if (inode) { |
810 | close(newfd); | |
811 | newfd = -1; | |
812 | } else { | |
813 | struct lo_inode *prev, *next; | |
814 | ||
815 | saverr = ENOMEM; | |
816 | inode = calloc(1, sizeof(struct lo_inode)); | |
817 | if (!inode) { | |
818 | goto out_err; | |
819 | } | |
820 | ||
821 | inode->is_symlink = S_ISLNK(e->attr.st_mode); | |
822 | inode->refcount = 1; | |
823 | inode->fd = newfd; | |
9de4fab5 | 824 | newfd = -1; |
7387863d DDAG |
825 | inode->ino = e->attr.st_ino; |
826 | inode->dev = e->attr.st_dev; | |
827 | ||
828 | pthread_mutex_lock(&lo->mutex); | |
92fb57b8 | 829 | inode->fuse_ino = lo_add_inode_mapping(req, inode); |
7387863d DDAG |
830 | prev = &lo->root; |
831 | next = prev->next; | |
832 | next->prev = inode; | |
833 | inode->next = next; | |
834 | inode->prev = prev; | |
835 | prev->next = inode; | |
836 | pthread_mutex_unlock(&lo->mutex); | |
837 | } | |
92fb57b8 | 838 | e->ino = inode->fuse_ino; |
7387863d | 839 | |
d240314a EG |
840 | fuse_log(FUSE_LOG_DEBUG, " %lli/%s -> %lli\n", (unsigned long long)parent, |
841 | name, (unsigned long long)e->ino); | |
7387863d DDAG |
842 | |
843 | return 0; | |
7c6b6602 DDAG |
844 | |
845 | out_err: | |
7387863d DDAG |
846 | saverr = errno; |
847 | if (newfd != -1) { | |
848 | close(newfd); | |
849 | } | |
850 | return saverr; | |
7c6b6602 DDAG |
851 | } |
852 | ||
853 | static void lo_lookup(fuse_req_t req, fuse_ino_t parent, const char *name) | |
854 | { | |
7387863d DDAG |
855 | struct fuse_entry_param e; |
856 | int err; | |
857 | ||
d240314a EG |
858 | fuse_log(FUSE_LOG_DEBUG, "lo_lookup(parent=%" PRIu64 ", name=%s)\n", parent, |
859 | name); | |
7387863d | 860 | |
25dae28c SH |
861 | /* |
862 | * Don't use is_safe_path_component(), allow "." and ".." for NFS export | |
863 | * support. | |
864 | */ | |
865 | if (strchr(name, '/')) { | |
866 | fuse_reply_err(req, EINVAL); | |
867 | return; | |
868 | } | |
869 | ||
7387863d DDAG |
870 | err = lo_do_lookup(req, parent, name, &e); |
871 | if (err) { | |
872 | fuse_reply_err(req, err); | |
873 | } else { | |
874 | fuse_reply_entry(req, &e); | |
875 | } | |
7c6b6602 DDAG |
876 | } |
877 | ||
929cfb7a VG |
878 | /* |
879 | * On some archs, setres*id is limited to 2^16 but they | |
880 | * provide setres*id32 variants that allow 2^32. | |
881 | * Others just let setres*id do 2^32 anyway. | |
882 | */ | |
883 | #ifdef SYS_setresgid32 | |
884 | #define OURSYS_setresgid SYS_setresgid32 | |
885 | #else | |
886 | #define OURSYS_setresgid SYS_setresgid | |
887 | #endif | |
888 | ||
889 | #ifdef SYS_setresuid32 | |
890 | #define OURSYS_setresuid SYS_setresuid32 | |
891 | #else | |
892 | #define OURSYS_setresuid SYS_setresuid | |
893 | #endif | |
894 | ||
895 | /* | |
896 | * Change to uid/gid of caller so that file is created with | |
897 | * ownership of caller. | |
898 | * TODO: What about selinux context? | |
899 | */ | |
900 | static int lo_change_cred(fuse_req_t req, struct lo_cred *old) | |
901 | { | |
902 | int res; | |
903 | ||
904 | old->euid = geteuid(); | |
905 | old->egid = getegid(); | |
906 | ||
907 | res = syscall(OURSYS_setresgid, -1, fuse_req_ctx(req)->gid, -1); | |
908 | if (res == -1) { | |
909 | return errno; | |
910 | } | |
911 | ||
912 | res = syscall(OURSYS_setresuid, -1, fuse_req_ctx(req)->uid, -1); | |
913 | if (res == -1) { | |
914 | int errno_save = errno; | |
915 | ||
916 | syscall(OURSYS_setresgid, -1, old->egid, -1); | |
917 | return errno_save; | |
918 | } | |
919 | ||
920 | return 0; | |
921 | } | |
922 | ||
923 | /* Regain Privileges */ | |
924 | static void lo_restore_cred(struct lo_cred *old) | |
925 | { | |
926 | int res; | |
927 | ||
928 | res = syscall(OURSYS_setresuid, -1, old->euid, -1); | |
929 | if (res == -1) { | |
930 | fuse_log(FUSE_LOG_ERR, "seteuid(%u): %m\n", old->euid); | |
931 | exit(1); | |
932 | } | |
933 | ||
934 | res = syscall(OURSYS_setresgid, -1, old->egid, -1); | |
935 | if (res == -1) { | |
936 | fuse_log(FUSE_LOG_ERR, "setegid(%u): %m\n", old->egid); | |
937 | exit(1); | |
938 | } | |
939 | } | |
940 | ||
7c6b6602 | 941 | static void lo_mknod_symlink(fuse_req_t req, fuse_ino_t parent, |
7387863d DDAG |
942 | const char *name, mode_t mode, dev_t rdev, |
943 | const char *link) | |
7c6b6602 | 944 | { |
7387863d DDAG |
945 | int res; |
946 | int saverr; | |
92fb57b8 | 947 | struct lo_inode *dir; |
7387863d | 948 | struct fuse_entry_param e; |
929cfb7a | 949 | struct lo_cred old = {}; |
7c6b6602 | 950 | |
25dae28c SH |
951 | if (!is_safe_path_component(name)) { |
952 | fuse_reply_err(req, EINVAL); | |
953 | return; | |
954 | } | |
955 | ||
92fb57b8 SH |
956 | dir = lo_inode(req, parent); |
957 | if (!dir) { | |
958 | fuse_reply_err(req, EBADF); | |
959 | return; | |
960 | } | |
961 | ||
7387863d | 962 | saverr = ENOMEM; |
7c6b6602 | 963 | |
929cfb7a VG |
964 | saverr = lo_change_cred(req, &old); |
965 | if (saverr) { | |
966 | goto out; | |
967 | } | |
968 | ||
7387863d | 969 | res = mknod_wrapper(dir->fd, name, link, mode, rdev); |
7c6b6602 | 970 | |
7387863d | 971 | saverr = errno; |
929cfb7a VG |
972 | |
973 | lo_restore_cred(&old); | |
974 | ||
7387863d DDAG |
975 | if (res == -1) { |
976 | goto out; | |
977 | } | |
7c6b6602 | 978 | |
7387863d DDAG |
979 | saverr = lo_do_lookup(req, parent, name, &e); |
980 | if (saverr) { | |
981 | goto out; | |
982 | } | |
7c6b6602 | 983 | |
d240314a EG |
984 | fuse_log(FUSE_LOG_DEBUG, " %lli/%s -> %lli\n", (unsigned long long)parent, |
985 | name, (unsigned long long)e.ino); | |
7c6b6602 | 986 | |
7387863d DDAG |
987 | fuse_reply_entry(req, &e); |
988 | return; | |
7c6b6602 DDAG |
989 | |
990 | out: | |
7387863d | 991 | fuse_reply_err(req, saverr); |
7c6b6602 DDAG |
992 | } |
993 | ||
7387863d DDAG |
994 | static void lo_mknod(fuse_req_t req, fuse_ino_t parent, const char *name, |
995 | mode_t mode, dev_t rdev) | |
7c6b6602 | 996 | { |
7387863d | 997 | lo_mknod_symlink(req, parent, name, mode, rdev, NULL); |
7c6b6602 DDAG |
998 | } |
999 | ||
1000 | static void lo_mkdir(fuse_req_t req, fuse_ino_t parent, const char *name, | |
7387863d | 1001 | mode_t mode) |
7c6b6602 | 1002 | { |
7387863d | 1003 | lo_mknod_symlink(req, parent, name, S_IFDIR | mode, 0, NULL); |
7c6b6602 DDAG |
1004 | } |
1005 | ||
7387863d DDAG |
1006 | static void lo_symlink(fuse_req_t req, const char *link, fuse_ino_t parent, |
1007 | const char *name) | |
7c6b6602 | 1008 | { |
7387863d | 1009 | lo_mknod_symlink(req, parent, name, S_IFLNK, 0, link); |
7c6b6602 DDAG |
1010 | } |
1011 | ||
5fe319a7 MS |
1012 | static int linkat_empty_nofollow(struct lo_data *lo, struct lo_inode *inode, |
1013 | int dfd, const char *name) | |
7c6b6602 | 1014 | { |
7387863d | 1015 | int res; |
5fe319a7 MS |
1016 | struct lo_inode *parent; |
1017 | char path[PATH_MAX]; | |
7c6b6602 | 1018 | |
7387863d DDAG |
1019 | if (inode->is_symlink) { |
1020 | res = linkat(inode->fd, "", dfd, name, AT_EMPTY_PATH); | |
1021 | if (res == -1 && (errno == ENOENT || errno == EINVAL)) { | |
1022 | /* Sorry, no race free way to hard-link a symlink. */ | |
5fe319a7 MS |
1023 | if (lo->norace) { |
1024 | errno = EPERM; | |
1025 | } else { | |
1026 | goto fallback; | |
1027 | } | |
7387863d DDAG |
1028 | } |
1029 | return res; | |
1030 | } | |
7c6b6602 | 1031 | |
9f59d175 | 1032 | sprintf(path, "%i", inode->fd); |
5fe319a7 | 1033 | |
9f59d175 | 1034 | return linkat(lo->proc_self_fd, path, dfd, name, AT_SYMLINK_FOLLOW); |
5fe319a7 MS |
1035 | |
1036 | fallback: | |
1037 | res = lo_parent_and_name(lo, inode, path, &parent); | |
1038 | if (res != -1) { | |
1039 | res = linkat(parent->fd, path, dfd, name, 0); | |
95d27157 | 1040 | unref_inode_lolocked(lo, parent, 1); |
5fe319a7 | 1041 | } |
7c6b6602 | 1042 | |
5fe319a7 | 1043 | return res; |
7c6b6602 DDAG |
1044 | } |
1045 | ||
1046 | static void lo_link(fuse_req_t req, fuse_ino_t ino, fuse_ino_t parent, | |
7387863d | 1047 | const char *name) |
7c6b6602 | 1048 | { |
7387863d DDAG |
1049 | int res; |
1050 | struct lo_data *lo = lo_data(req); | |
92fb57b8 | 1051 | struct lo_inode *inode; |
7387863d DDAG |
1052 | struct fuse_entry_param e; |
1053 | int saverr; | |
1054 | ||
25dae28c SH |
1055 | if (!is_safe_path_component(name)) { |
1056 | fuse_reply_err(req, EINVAL); | |
1057 | return; | |
1058 | } | |
1059 | ||
92fb57b8 SH |
1060 | inode = lo_inode(req, ino); |
1061 | if (!inode) { | |
1062 | fuse_reply_err(req, EBADF); | |
1063 | return; | |
1064 | } | |
1065 | ||
7387863d DDAG |
1066 | memset(&e, 0, sizeof(struct fuse_entry_param)); |
1067 | e.attr_timeout = lo->timeout; | |
1068 | e.entry_timeout = lo->timeout; | |
1069 | ||
5fe319a7 | 1070 | res = linkat_empty_nofollow(lo, inode, lo_fd(req, parent), name); |
7387863d DDAG |
1071 | if (res == -1) { |
1072 | goto out_err; | |
1073 | } | |
1074 | ||
1075 | res = fstatat(inode->fd, "", &e.attr, AT_EMPTY_PATH | AT_SYMLINK_NOFOLLOW); | |
1076 | if (res == -1) { | |
1077 | goto out_err; | |
1078 | } | |
1079 | ||
1080 | pthread_mutex_lock(&lo->mutex); | |
1081 | inode->refcount++; | |
1082 | pthread_mutex_unlock(&lo->mutex); | |
92fb57b8 | 1083 | e.ino = inode->fuse_ino; |
7387863d | 1084 | |
d240314a EG |
1085 | fuse_log(FUSE_LOG_DEBUG, " %lli/%s -> %lli\n", (unsigned long long)parent, |
1086 | name, (unsigned long long)e.ino); | |
7387863d DDAG |
1087 | |
1088 | fuse_reply_entry(req, &e); | |
1089 | return; | |
7c6b6602 DDAG |
1090 | |
1091 | out_err: | |
7387863d DDAG |
1092 | saverr = errno; |
1093 | fuse_reply_err(req, saverr); | |
7c6b6602 DDAG |
1094 | } |
1095 | ||
1096 | static void lo_rmdir(fuse_req_t req, fuse_ino_t parent, const char *name) | |
1097 | { | |
7387863d | 1098 | int res; |
25dae28c SH |
1099 | if (!is_safe_path_component(name)) { |
1100 | fuse_reply_err(req, EINVAL); | |
1101 | return; | |
1102 | } | |
7c6b6602 | 1103 | |
7387863d | 1104 | res = unlinkat(lo_fd(req, parent), name, AT_REMOVEDIR); |
7c6b6602 | 1105 | |
7387863d | 1106 | fuse_reply_err(req, res == -1 ? errno : 0); |
7c6b6602 DDAG |
1107 | } |
1108 | ||
1109 | static void lo_rename(fuse_req_t req, fuse_ino_t parent, const char *name, | |
7387863d DDAG |
1110 | fuse_ino_t newparent, const char *newname, |
1111 | unsigned int flags) | |
7c6b6602 | 1112 | { |
7387863d | 1113 | int res; |
7c6b6602 | 1114 | |
25dae28c SH |
1115 | if (!is_safe_path_component(name) || !is_safe_path_component(newname)) { |
1116 | fuse_reply_err(req, EINVAL); | |
1117 | return; | |
1118 | } | |
1119 | ||
7387863d | 1120 | if (flags) { |
f0ab7d6f | 1121 | #ifndef SYS_renameat2 |
7387863d | 1122 | fuse_reply_err(req, EINVAL); |
f0ab7d6f MS |
1123 | #else |
1124 | res = syscall(SYS_renameat2, lo_fd(req, parent), name, | |
1125 | lo_fd(req, newparent), newname, flags); | |
1126 | if (res == -1 && errno == ENOSYS) { | |
1127 | fuse_reply_err(req, EINVAL); | |
1128 | } else { | |
1129 | fuse_reply_err(req, res == -1 ? errno : 0); | |
1130 | } | |
1131 | #endif | |
7387863d DDAG |
1132 | return; |
1133 | } | |
7c6b6602 | 1134 | |
7387863d | 1135 | res = renameat(lo_fd(req, parent), name, lo_fd(req, newparent), newname); |
7c6b6602 | 1136 | |
7387863d | 1137 | fuse_reply_err(req, res == -1 ? errno : 0); |
7c6b6602 DDAG |
1138 | } |
1139 | ||
1140 | static void lo_unlink(fuse_req_t req, fuse_ino_t parent, const char *name) | |
1141 | { | |
7387863d | 1142 | int res; |
7c6b6602 | 1143 | |
25dae28c SH |
1144 | if (!is_safe_path_component(name)) { |
1145 | fuse_reply_err(req, EINVAL); | |
1146 | return; | |
1147 | } | |
1148 | ||
7387863d | 1149 | res = unlinkat(lo_fd(req, parent), name, 0); |
7c6b6602 | 1150 | |
7387863d | 1151 | fuse_reply_err(req, res == -1 ? errno : 0); |
7c6b6602 DDAG |
1152 | } |
1153 | ||
95d27157 MS |
1154 | static void unref_inode_lolocked(struct lo_data *lo, struct lo_inode *inode, |
1155 | uint64_t n) | |
7c6b6602 | 1156 | { |
7387863d DDAG |
1157 | if (!inode) { |
1158 | return; | |
1159 | } | |
1160 | ||
1161 | pthread_mutex_lock(&lo->mutex); | |
1162 | assert(inode->refcount >= n); | |
1163 | inode->refcount -= n; | |
1164 | if (!inode->refcount) { | |
1165 | struct lo_inode *prev, *next; | |
1166 | ||
1167 | prev = inode->prev; | |
1168 | next = inode->next; | |
1169 | next->prev = prev; | |
1170 | prev->next = next; | |
1171 | ||
92fb57b8 | 1172 | lo_map_remove(&lo->ino_map, inode->fuse_ino); |
7387863d DDAG |
1173 | pthread_mutex_unlock(&lo->mutex); |
1174 | close(inode->fd); | |
1175 | free(inode); | |
7387863d DDAG |
1176 | } else { |
1177 | pthread_mutex_unlock(&lo->mutex); | |
1178 | } | |
7c6b6602 DDAG |
1179 | } |
1180 | ||
1181 | static void lo_forget_one(fuse_req_t req, fuse_ino_t ino, uint64_t nlookup) | |
1182 | { | |
7387863d | 1183 | struct lo_data *lo = lo_data(req); |
92fb57b8 SH |
1184 | struct lo_inode *inode; |
1185 | ||
1186 | inode = lo_inode(req, ino); | |
1187 | if (!inode) { | |
1188 | return; | |
1189 | } | |
7c6b6602 | 1190 | |
d240314a EG |
1191 | fuse_log(FUSE_LOG_DEBUG, " forget %lli %lli -%lli\n", |
1192 | (unsigned long long)ino, (unsigned long long)inode->refcount, | |
1193 | (unsigned long long)nlookup); | |
7c6b6602 | 1194 | |
95d27157 | 1195 | unref_inode_lolocked(lo, inode, nlookup); |
7c6b6602 DDAG |
1196 | } |
1197 | ||
1198 | static void lo_forget(fuse_req_t req, fuse_ino_t ino, uint64_t nlookup) | |
1199 | { | |
7387863d DDAG |
1200 | lo_forget_one(req, ino, nlookup); |
1201 | fuse_reply_none(req); | |
7c6b6602 DDAG |
1202 | } |
1203 | ||
1204 | static void lo_forget_multi(fuse_req_t req, size_t count, | |
7387863d | 1205 | struct fuse_forget_data *forgets) |
7c6b6602 | 1206 | { |
7387863d | 1207 | int i; |
7c6b6602 | 1208 | |
7387863d DDAG |
1209 | for (i = 0; i < count; i++) { |
1210 | lo_forget_one(req, forgets[i].ino, forgets[i].nlookup); | |
1211 | } | |
1212 | fuse_reply_none(req); | |
7c6b6602 DDAG |
1213 | } |
1214 | ||
1215 | static void lo_readlink(fuse_req_t req, fuse_ino_t ino) | |
1216 | { | |
7387863d DDAG |
1217 | char buf[PATH_MAX + 1]; |
1218 | int res; | |
7c6b6602 | 1219 | |
7387863d DDAG |
1220 | res = readlinkat(lo_fd(req, ino), "", buf, sizeof(buf)); |
1221 | if (res == -1) { | |
1222 | return (void)fuse_reply_err(req, errno); | |
1223 | } | |
7c6b6602 | 1224 | |
7387863d DDAG |
1225 | if (res == sizeof(buf)) { |
1226 | return (void)fuse_reply_err(req, ENAMETOOLONG); | |
1227 | } | |
7c6b6602 | 1228 | |
7387863d | 1229 | buf[res] = '\0'; |
7c6b6602 | 1230 | |
7387863d | 1231 | fuse_reply_readlink(req, buf); |
7c6b6602 DDAG |
1232 | } |
1233 | ||
1234 | struct lo_dirp { | |
7387863d DDAG |
1235 | DIR *dp; |
1236 | struct dirent *entry; | |
1237 | off_t offset; | |
7c6b6602 DDAG |
1238 | }; |
1239 | ||
b39bce12 | 1240 | static struct lo_dirp *lo_dirp(fuse_req_t req, struct fuse_file_info *fi) |
7c6b6602 | 1241 | { |
b39bce12 SH |
1242 | struct lo_data *lo = lo_data(req); |
1243 | struct lo_map_elem *elem; | |
1244 | ||
1245 | pthread_mutex_lock(&lo->mutex); | |
1246 | elem = lo_map_get(&lo->dirp_map, fi->fh); | |
1247 | pthread_mutex_unlock(&lo->mutex); | |
1248 | if (!elem) { | |
1249 | return NULL; | |
1250 | } | |
1251 | ||
1252 | return elem->dirp; | |
7c6b6602 DDAG |
1253 | } |
1254 | ||
7387863d DDAG |
1255 | static void lo_opendir(fuse_req_t req, fuse_ino_t ino, |
1256 | struct fuse_file_info *fi) | |
7c6b6602 | 1257 | { |
7387863d DDAG |
1258 | int error = ENOMEM; |
1259 | struct lo_data *lo = lo_data(req); | |
1260 | struct lo_dirp *d; | |
1261 | int fd; | |
b39bce12 | 1262 | ssize_t fh; |
7387863d DDAG |
1263 | |
1264 | d = calloc(1, sizeof(struct lo_dirp)); | |
1265 | if (d == NULL) { | |
1266 | goto out_err; | |
1267 | } | |
1268 | ||
1269 | fd = openat(lo_fd(req, ino), ".", O_RDONLY); | |
1270 | if (fd == -1) { | |
1271 | goto out_errno; | |
1272 | } | |
1273 | ||
1274 | d->dp = fdopendir(fd); | |
1275 | if (d->dp == NULL) { | |
1276 | goto out_errno; | |
1277 | } | |
1278 | ||
1279 | d->offset = 0; | |
1280 | d->entry = NULL; | |
1281 | ||
b39bce12 SH |
1282 | pthread_mutex_lock(&lo->mutex); |
1283 | fh = lo_add_dirp_mapping(req, d); | |
1284 | pthread_mutex_unlock(&lo->mutex); | |
1285 | if (fh == -1) { | |
1286 | goto out_err; | |
1287 | } | |
1288 | ||
1289 | fi->fh = fh; | |
7387863d DDAG |
1290 | if (lo->cache == CACHE_ALWAYS) { |
1291 | fi->keep_cache = 1; | |
1292 | } | |
1293 | fuse_reply_open(req, fi); | |
1294 | return; | |
7c6b6602 DDAG |
1295 | |
1296 | out_errno: | |
7387863d | 1297 | error = errno; |
7c6b6602 | 1298 | out_err: |
7387863d | 1299 | if (d) { |
b39bce12 SH |
1300 | if (d->dp) { |
1301 | closedir(d->dp); | |
1302 | } | |
7387863d DDAG |
1303 | if (fd != -1) { |
1304 | close(fd); | |
1305 | } | |
1306 | free(d); | |
1307 | } | |
1308 | fuse_reply_err(req, error); | |
7c6b6602 DDAG |
1309 | } |
1310 | ||
7c6b6602 | 1311 | static void lo_do_readdir(fuse_req_t req, fuse_ino_t ino, size_t size, |
7387863d | 1312 | off_t offset, struct fuse_file_info *fi, int plus) |
7c6b6602 | 1313 | { |
752272da | 1314 | struct lo_data *lo = lo_data(req); |
b39bce12 | 1315 | struct lo_dirp *d; |
752272da | 1316 | struct lo_inode *dinode; |
b39bce12 | 1317 | char *buf = NULL; |
7387863d DDAG |
1318 | char *p; |
1319 | size_t rem = size; | |
752272da | 1320 | int err = EBADF; |
7387863d | 1321 | |
752272da SH |
1322 | dinode = lo_inode(req, ino); |
1323 | if (!dinode) { | |
1324 | goto error; | |
1325 | } | |
7387863d | 1326 | |
b39bce12 SH |
1327 | d = lo_dirp(req, fi); |
1328 | if (!d) { | |
1329 | goto error; | |
1330 | } | |
1331 | ||
752272da | 1332 | err = ENOMEM; |
7387863d DDAG |
1333 | buf = calloc(1, size); |
1334 | if (!buf) { | |
7387863d DDAG |
1335 | goto error; |
1336 | } | |
1337 | p = buf; | |
1338 | ||
1339 | if (offset != d->offset) { | |
1340 | seekdir(d->dp, offset); | |
1341 | d->entry = NULL; | |
1342 | d->offset = offset; | |
1343 | } | |
1344 | while (1) { | |
1345 | size_t entsize; | |
1346 | off_t nextoff; | |
1347 | const char *name; | |
1348 | ||
1349 | if (!d->entry) { | |
1350 | errno = 0; | |
1351 | d->entry = readdir(d->dp); | |
1352 | if (!d->entry) { | |
1353 | if (errno) { /* Error */ | |
1354 | err = errno; | |
1355 | goto error; | |
1356 | } else { /* End of stream */ | |
1357 | break; | |
1358 | } | |
1359 | } | |
1360 | } | |
1361 | nextoff = d->entry->d_off; | |
1362 | name = d->entry->d_name; | |
752272da | 1363 | |
7387863d | 1364 | fuse_ino_t entry_ino = 0; |
752272da SH |
1365 | struct fuse_entry_param e = (struct fuse_entry_param){ |
1366 | .attr.st_ino = d->entry->d_ino, | |
1367 | .attr.st_mode = d->entry->d_type << 12, | |
1368 | }; | |
1369 | ||
1370 | /* Hide root's parent directory */ | |
1371 | if (dinode == &lo->root && strcmp(name, "..") == 0) { | |
1372 | e.attr.st_ino = lo->root.ino; | |
1373 | e.attr.st_mode = DT_DIR << 12; | |
1374 | } | |
1375 | ||
7387863d | 1376 | if (plus) { |
752272da | 1377 | if (!is_dot_or_dotdot(name)) { |
7387863d DDAG |
1378 | err = lo_do_lookup(req, ino, name, &e); |
1379 | if (err) { | |
1380 | goto error; | |
1381 | } | |
1382 | entry_ino = e.ino; | |
1383 | } | |
1384 | ||
1385 | entsize = fuse_add_direntry_plus(req, p, rem, name, &e, nextoff); | |
1386 | } else { | |
752272da | 1387 | entsize = fuse_add_direntry(req, p, rem, name, &e.attr, nextoff); |
7387863d DDAG |
1388 | } |
1389 | if (entsize > rem) { | |
1390 | if (entry_ino != 0) { | |
1391 | lo_forget_one(req, entry_ino, 1); | |
1392 | } | |
1393 | break; | |
1394 | } | |
1395 | ||
1396 | p += entsize; | |
1397 | rem -= entsize; | |
1398 | ||
1399 | d->entry = NULL; | |
1400 | d->offset = nextoff; | |
1401 | } | |
7c6b6602 DDAG |
1402 | |
1403 | err = 0; | |
1404 | error: | |
7387863d DDAG |
1405 | /* |
1406 | * If there's an error, we can only signal it if we haven't stored | |
1407 | * any entries yet - otherwise we'd end up with wrong lookup | |
1408 | * counts for the entries that are already in the buffer. So we | |
1409 | * return what we've collected until that point. | |
1410 | */ | |
1411 | if (err && rem == size) { | |
1412 | fuse_reply_err(req, err); | |
1413 | } else { | |
1414 | fuse_reply_buf(req, buf, size - rem); | |
1415 | } | |
7c6b6602 DDAG |
1416 | free(buf); |
1417 | } | |
1418 | ||
1419 | static void lo_readdir(fuse_req_t req, fuse_ino_t ino, size_t size, | |
7387863d | 1420 | off_t offset, struct fuse_file_info *fi) |
7c6b6602 | 1421 | { |
7387863d | 1422 | lo_do_readdir(req, ino, size, offset, fi, 0); |
7c6b6602 DDAG |
1423 | } |
1424 | ||
1425 | static void lo_readdirplus(fuse_req_t req, fuse_ino_t ino, size_t size, | |
7387863d | 1426 | off_t offset, struct fuse_file_info *fi) |
7c6b6602 | 1427 | { |
7387863d | 1428 | lo_do_readdir(req, ino, size, offset, fi, 1); |
7c6b6602 DDAG |
1429 | } |
1430 | ||
7387863d DDAG |
1431 | static void lo_releasedir(fuse_req_t req, fuse_ino_t ino, |
1432 | struct fuse_file_info *fi) | |
7c6b6602 | 1433 | { |
b39bce12 SH |
1434 | struct lo_data *lo = lo_data(req); |
1435 | struct lo_dirp *d; | |
1436 | ||
7387863d | 1437 | (void)ino; |
b39bce12 SH |
1438 | |
1439 | d = lo_dirp(req, fi); | |
1440 | if (!d) { | |
1441 | fuse_reply_err(req, EBADF); | |
1442 | return; | |
1443 | } | |
1444 | ||
1445 | pthread_mutex_lock(&lo->mutex); | |
1446 | lo_map_remove(&lo->dirp_map, fi->fh); | |
1447 | pthread_mutex_unlock(&lo->mutex); | |
1448 | ||
7387863d DDAG |
1449 | closedir(d->dp); |
1450 | free(d); | |
1451 | fuse_reply_err(req, 0); | |
7c6b6602 DDAG |
1452 | } |
1453 | ||
1454 | static void lo_create(fuse_req_t req, fuse_ino_t parent, const char *name, | |
7387863d | 1455 | mode_t mode, struct fuse_file_info *fi) |
7c6b6602 | 1456 | { |
7387863d DDAG |
1457 | int fd; |
1458 | struct lo_data *lo = lo_data(req); | |
1459 | struct fuse_entry_param e; | |
1460 | int err; | |
929cfb7a | 1461 | struct lo_cred old = {}; |
7387863d | 1462 | |
d240314a EG |
1463 | fuse_log(FUSE_LOG_DEBUG, "lo_create(parent=%" PRIu64 ", name=%s)\n", parent, |
1464 | name); | |
7387863d | 1465 | |
25dae28c SH |
1466 | if (!is_safe_path_component(name)) { |
1467 | fuse_reply_err(req, EINVAL); | |
1468 | return; | |
1469 | } | |
1470 | ||
929cfb7a VG |
1471 | err = lo_change_cred(req, &old); |
1472 | if (err) { | |
1473 | goto out; | |
1474 | } | |
1475 | ||
7387863d DDAG |
1476 | fd = openat(lo_fd(req, parent), name, (fi->flags | O_CREAT) & ~O_NOFOLLOW, |
1477 | mode); | |
929cfb7a VG |
1478 | err = fd == -1 ? errno : 0; |
1479 | lo_restore_cred(&old); | |
7387863d | 1480 | |
929cfb7a | 1481 | if (!err) { |
73b4d19d SH |
1482 | ssize_t fh; |
1483 | ||
1484 | pthread_mutex_lock(&lo->mutex); | |
1485 | fh = lo_add_fd_mapping(req, fd); | |
1486 | pthread_mutex_unlock(&lo->mutex); | |
1487 | if (fh == -1) { | |
1488 | close(fd); | |
1489 | fuse_reply_err(req, ENOMEM); | |
1490 | return; | |
1491 | } | |
1492 | ||
1493 | fi->fh = fh; | |
929cfb7a VG |
1494 | err = lo_do_lookup(req, parent, name, &e); |
1495 | } | |
7387863d DDAG |
1496 | if (lo->cache == CACHE_NEVER) { |
1497 | fi->direct_io = 1; | |
1498 | } else if (lo->cache == CACHE_ALWAYS) { | |
1499 | fi->keep_cache = 1; | |
1500 | } | |
1501 | ||
929cfb7a | 1502 | out: |
7387863d DDAG |
1503 | if (err) { |
1504 | fuse_reply_err(req, err); | |
1505 | } else { | |
1506 | fuse_reply_create(req, &e, fi); | |
1507 | } | |
7c6b6602 DDAG |
1508 | } |
1509 | ||
1510 | static void lo_fsyncdir(fuse_req_t req, fuse_ino_t ino, int datasync, | |
7387863d | 1511 | struct fuse_file_info *fi) |
7c6b6602 | 1512 | { |
7387863d | 1513 | int res; |
b39bce12 SH |
1514 | struct lo_dirp *d; |
1515 | int fd; | |
1516 | ||
7387863d | 1517 | (void)ino; |
b39bce12 SH |
1518 | |
1519 | d = lo_dirp(req, fi); | |
1520 | if (!d) { | |
1521 | fuse_reply_err(req, EBADF); | |
1522 | return; | |
1523 | } | |
1524 | ||
1525 | fd = dirfd(d->dp); | |
7387863d DDAG |
1526 | if (datasync) { |
1527 | res = fdatasync(fd); | |
1528 | } else { | |
1529 | res = fsync(fd); | |
1530 | } | |
1531 | fuse_reply_err(req, res == -1 ? errno : 0); | |
7c6b6602 DDAG |
1532 | } |
1533 | ||
1534 | static void lo_open(fuse_req_t req, fuse_ino_t ino, struct fuse_file_info *fi) | |
1535 | { | |
7387863d | 1536 | int fd; |
73b4d19d | 1537 | ssize_t fh; |
7387863d DDAG |
1538 | char buf[64]; |
1539 | struct lo_data *lo = lo_data(req); | |
1540 | ||
d240314a EG |
1541 | fuse_log(FUSE_LOG_DEBUG, "lo_open(ino=%" PRIu64 ", flags=%d)\n", ino, |
1542 | fi->flags); | |
7387863d DDAG |
1543 | |
1544 | /* | |
1545 | * With writeback cache, kernel may send read requests even | |
1546 | * when userspace opened write-only | |
1547 | */ | |
1548 | if (lo->writeback && (fi->flags & O_ACCMODE) == O_WRONLY) { | |
1549 | fi->flags &= ~O_ACCMODE; | |
1550 | fi->flags |= O_RDWR; | |
1551 | } | |
1552 | ||
1553 | /* | |
1554 | * With writeback cache, O_APPEND is handled by the kernel. | |
1555 | * This breaks atomicity (since the file may change in the | |
1556 | * underlying filesystem, so that the kernel's idea of the | |
1557 | * end of the file isn't accurate anymore). In this example, | |
1558 | * we just accept that. A more rigorous filesystem may want | |
1559 | * to return an error here | |
1560 | */ | |
1561 | if (lo->writeback && (fi->flags & O_APPEND)) { | |
1562 | fi->flags &= ~O_APPEND; | |
1563 | } | |
1564 | ||
9f59d175 SH |
1565 | sprintf(buf, "%i", lo_fd(req, ino)); |
1566 | fd = openat(lo->proc_self_fd, buf, fi->flags & ~O_NOFOLLOW); | |
7387863d DDAG |
1567 | if (fd == -1) { |
1568 | return (void)fuse_reply_err(req, errno); | |
1569 | } | |
1570 | ||
73b4d19d SH |
1571 | pthread_mutex_lock(&lo->mutex); |
1572 | fh = lo_add_fd_mapping(req, fd); | |
1573 | pthread_mutex_unlock(&lo->mutex); | |
1574 | if (fh == -1) { | |
1575 | close(fd); | |
1576 | fuse_reply_err(req, ENOMEM); | |
1577 | return; | |
1578 | } | |
1579 | ||
1580 | fi->fh = fh; | |
7387863d DDAG |
1581 | if (lo->cache == CACHE_NEVER) { |
1582 | fi->direct_io = 1; | |
1583 | } else if (lo->cache == CACHE_ALWAYS) { | |
1584 | fi->keep_cache = 1; | |
1585 | } | |
1586 | fuse_reply_open(req, fi); | |
7c6b6602 DDAG |
1587 | } |
1588 | ||
7387863d DDAG |
1589 | static void lo_release(fuse_req_t req, fuse_ino_t ino, |
1590 | struct fuse_file_info *fi) | |
7c6b6602 | 1591 | { |
73b4d19d SH |
1592 | struct lo_data *lo = lo_data(req); |
1593 | int fd; | |
1594 | ||
7387863d | 1595 | (void)ino; |
7c6b6602 | 1596 | |
73b4d19d SH |
1597 | fd = lo_fi_fd(req, fi); |
1598 | ||
1599 | pthread_mutex_lock(&lo->mutex); | |
1600 | lo_map_remove(&lo->fd_map, fi->fh); | |
1601 | pthread_mutex_unlock(&lo->mutex); | |
1602 | ||
1603 | close(fd); | |
7387863d | 1604 | fuse_reply_err(req, 0); |
7c6b6602 DDAG |
1605 | } |
1606 | ||
1607 | static void lo_flush(fuse_req_t req, fuse_ino_t ino, struct fuse_file_info *fi) | |
1608 | { | |
7387863d DDAG |
1609 | int res; |
1610 | (void)ino; | |
73b4d19d | 1611 | res = close(dup(lo_fi_fd(req, fi))); |
7387863d | 1612 | fuse_reply_err(req, res == -1 ? errno : 0); |
7c6b6602 DDAG |
1613 | } |
1614 | ||
1615 | static void lo_fsync(fuse_req_t req, fuse_ino_t ino, int datasync, | |
7387863d | 1616 | struct fuse_file_info *fi) |
7c6b6602 | 1617 | { |
7387863d | 1618 | int res; |
1b209805 VG |
1619 | int fd; |
1620 | char *buf; | |
1621 | ||
1622 | fuse_log(FUSE_LOG_DEBUG, "lo_fsync(ino=%" PRIu64 ", fi=0x%p)\n", ino, | |
1623 | (void *)fi); | |
1624 | ||
1625 | if (!fi) { | |
9f59d175 SH |
1626 | struct lo_data *lo = lo_data(req); |
1627 | ||
1628 | res = asprintf(&buf, "%i", lo_fd(req, ino)); | |
1b209805 VG |
1629 | if (res == -1) { |
1630 | return (void)fuse_reply_err(req, errno); | |
1631 | } | |
1632 | ||
9f59d175 | 1633 | fd = openat(lo->proc_self_fd, buf, O_RDWR); |
1b209805 VG |
1634 | free(buf); |
1635 | if (fd == -1) { | |
1636 | return (void)fuse_reply_err(req, errno); | |
1637 | } | |
1638 | } else { | |
73b4d19d | 1639 | fd = lo_fi_fd(req, fi); |
1b209805 VG |
1640 | } |
1641 | ||
7387863d | 1642 | if (datasync) { |
1b209805 | 1643 | res = fdatasync(fd); |
7387863d | 1644 | } else { |
1b209805 VG |
1645 | res = fsync(fd); |
1646 | } | |
1647 | if (!fi) { | |
1648 | close(fd); | |
7387863d DDAG |
1649 | } |
1650 | fuse_reply_err(req, res == -1 ? errno : 0); | |
7c6b6602 DDAG |
1651 | } |
1652 | ||
7387863d DDAG |
1653 | static void lo_read(fuse_req_t req, fuse_ino_t ino, size_t size, off_t offset, |
1654 | struct fuse_file_info *fi) | |
7c6b6602 | 1655 | { |
7387863d | 1656 | struct fuse_bufvec buf = FUSE_BUFVEC_INIT(size); |
7c6b6602 | 1657 | |
d240314a EG |
1658 | fuse_log(FUSE_LOG_DEBUG, |
1659 | "lo_read(ino=%" PRIu64 ", size=%zd, " | |
1660 | "off=%lu)\n", | |
1661 | ino, size, (unsigned long)offset); | |
7c6b6602 | 1662 | |
7387863d | 1663 | buf.buf[0].flags = FUSE_BUF_IS_FD | FUSE_BUF_FD_SEEK; |
73b4d19d | 1664 | buf.buf[0].fd = lo_fi_fd(req, fi); |
7387863d | 1665 | buf.buf[0].pos = offset; |
7c6b6602 | 1666 | |
8c3fe75e | 1667 | fuse_reply_data(req, &buf); |
7c6b6602 DDAG |
1668 | } |
1669 | ||
1670 | static void lo_write_buf(fuse_req_t req, fuse_ino_t ino, | |
7387863d DDAG |
1671 | struct fuse_bufvec *in_buf, off_t off, |
1672 | struct fuse_file_info *fi) | |
7c6b6602 | 1673 | { |
7387863d DDAG |
1674 | (void)ino; |
1675 | ssize_t res; | |
1676 | struct fuse_bufvec out_buf = FUSE_BUFVEC_INIT(fuse_buf_size(in_buf)); | |
ee884652 | 1677 | bool cap_fsetid_dropped = false; |
7387863d DDAG |
1678 | |
1679 | out_buf.buf[0].flags = FUSE_BUF_IS_FD | FUSE_BUF_FD_SEEK; | |
73b4d19d | 1680 | out_buf.buf[0].fd = lo_fi_fd(req, fi); |
7387863d DDAG |
1681 | out_buf.buf[0].pos = off; |
1682 | ||
d240314a EG |
1683 | fuse_log(FUSE_LOG_DEBUG, |
1684 | "lo_write_buf(ino=%" PRIu64 ", size=%zd, off=%lu)\n", ino, | |
1685 | out_buf.buf[0].size, (unsigned long)off); | |
7387863d | 1686 | |
ee884652 VG |
1687 | /* |
1688 | * If kill_priv is set, drop CAP_FSETID which should lead to kernel | |
1689 | * clearing setuid/setgid on file. | |
1690 | */ | |
1691 | if (fi->kill_priv) { | |
1692 | res = drop_effective_cap("FSETID", &cap_fsetid_dropped); | |
1693 | if (res != 0) { | |
1694 | fuse_reply_err(req, res); | |
1695 | return; | |
1696 | } | |
1697 | } | |
1698 | ||
8c3fe75e | 1699 | res = fuse_buf_copy(&out_buf, in_buf); |
7387863d DDAG |
1700 | if (res < 0) { |
1701 | fuse_reply_err(req, -res); | |
1702 | } else { | |
1703 | fuse_reply_write(req, (size_t)res); | |
1704 | } | |
ee884652 VG |
1705 | |
1706 | if (cap_fsetid_dropped) { | |
1707 | res = gain_effective_cap("FSETID"); | |
1708 | if (res) { | |
1709 | fuse_log(FUSE_LOG_ERR, "Failed to gain CAP_FSETID\n"); | |
1710 | } | |
1711 | } | |
7c6b6602 DDAG |
1712 | } |
1713 | ||
1714 | static void lo_statfs(fuse_req_t req, fuse_ino_t ino) | |
1715 | { | |
7387863d DDAG |
1716 | int res; |
1717 | struct statvfs stbuf; | |
1718 | ||
1719 | res = fstatvfs(lo_fd(req, ino), &stbuf); | |
1720 | if (res == -1) { | |
1721 | fuse_reply_err(req, errno); | |
1722 | } else { | |
1723 | fuse_reply_statfs(req, &stbuf); | |
1724 | } | |
7c6b6602 DDAG |
1725 | } |
1726 | ||
7387863d DDAG |
1727 | static void lo_fallocate(fuse_req_t req, fuse_ino_t ino, int mode, off_t offset, |
1728 | off_t length, struct fuse_file_info *fi) | |
7c6b6602 | 1729 | { |
7387863d DDAG |
1730 | int err = EOPNOTSUPP; |
1731 | (void)ino; | |
7c6b6602 | 1732 | |
9776457c | 1733 | #ifdef CONFIG_FALLOCATE |
73b4d19d | 1734 | err = fallocate(lo_fi_fd(req, fi), mode, offset, length); |
7387863d DDAG |
1735 | if (err < 0) { |
1736 | err = errno; | |
1737 | } | |
7c6b6602 | 1738 | |
9776457c | 1739 | #elif defined(CONFIG_POSIX_FALLOCATE) |
7387863d DDAG |
1740 | if (mode) { |
1741 | fuse_reply_err(req, EOPNOTSUPP); | |
1742 | return; | |
1743 | } | |
7c6b6602 | 1744 | |
73b4d19d | 1745 | err = posix_fallocate(lo_fi_fd(req, fi), offset, length); |
7c6b6602 DDAG |
1746 | #endif |
1747 | ||
7387863d | 1748 | fuse_reply_err(req, err); |
7c6b6602 DDAG |
1749 | } |
1750 | ||
1751 | static void lo_flock(fuse_req_t req, fuse_ino_t ino, struct fuse_file_info *fi, | |
7387863d | 1752 | int op) |
7c6b6602 | 1753 | { |
7387863d DDAG |
1754 | int res; |
1755 | (void)ino; | |
7c6b6602 | 1756 | |
73b4d19d | 1757 | res = flock(lo_fi_fd(req, fi), op); |
7c6b6602 | 1758 | |
7387863d | 1759 | fuse_reply_err(req, res == -1 ? errno : 0); |
7c6b6602 DDAG |
1760 | } |
1761 | ||
1762 | static void lo_getxattr(fuse_req_t req, fuse_ino_t ino, const char *name, | |
7387863d | 1763 | size_t size) |
7c6b6602 | 1764 | { |
9f59d175 | 1765 | struct lo_data *lo = lo_data(req); |
7387863d DDAG |
1766 | char *value = NULL; |
1767 | char procname[64]; | |
92fb57b8 | 1768 | struct lo_inode *inode; |
7387863d DDAG |
1769 | ssize_t ret; |
1770 | int saverr; | |
9f59d175 | 1771 | int fd = -1; |
7387863d | 1772 | |
92fb57b8 SH |
1773 | inode = lo_inode(req, ino); |
1774 | if (!inode) { | |
1775 | fuse_reply_err(req, EBADF); | |
1776 | return; | |
1777 | } | |
1778 | ||
7387863d DDAG |
1779 | saverr = ENOSYS; |
1780 | if (!lo_data(req)->xattr) { | |
1781 | goto out; | |
1782 | } | |
1783 | ||
d240314a EG |
1784 | fuse_log(FUSE_LOG_DEBUG, "lo_getxattr(ino=%" PRIu64 ", name=%s size=%zd)\n", |
1785 | ino, name, size); | |
7387863d DDAG |
1786 | |
1787 | if (inode->is_symlink) { | |
1788 | /* Sorry, no race free way to getxattr on symlink. */ | |
1789 | saverr = EPERM; | |
1790 | goto out; | |
1791 | } | |
1792 | ||
9f59d175 SH |
1793 | sprintf(procname, "%i", inode->fd); |
1794 | fd = openat(lo->proc_self_fd, procname, O_RDONLY); | |
1795 | if (fd < 0) { | |
1796 | goto out_err; | |
1797 | } | |
7387863d DDAG |
1798 | |
1799 | if (size) { | |
1800 | value = malloc(size); | |
1801 | if (!value) { | |
1802 | goto out_err; | |
1803 | } | |
1804 | ||
9f59d175 | 1805 | ret = fgetxattr(fd, name, value, size); |
7387863d DDAG |
1806 | if (ret == -1) { |
1807 | goto out_err; | |
1808 | } | |
1809 | saverr = 0; | |
1810 | if (ret == 0) { | |
1811 | goto out; | |
1812 | } | |
1813 | ||
1814 | fuse_reply_buf(req, value, ret); | |
1815 | } else { | |
9f59d175 | 1816 | ret = fgetxattr(fd, name, NULL, 0); |
7387863d DDAG |
1817 | if (ret == -1) { |
1818 | goto out_err; | |
1819 | } | |
1820 | ||
1821 | fuse_reply_xattr(req, ret); | |
1822 | } | |
7c6b6602 | 1823 | out_free: |
7387863d | 1824 | free(value); |
9f59d175 SH |
1825 | |
1826 | if (fd >= 0) { | |
1827 | close(fd); | |
1828 | } | |
7387863d | 1829 | return; |
7c6b6602 DDAG |
1830 | |
1831 | out_err: | |
7387863d | 1832 | saverr = errno; |
7c6b6602 | 1833 | out: |
7387863d DDAG |
1834 | fuse_reply_err(req, saverr); |
1835 | goto out_free; | |
7c6b6602 DDAG |
1836 | } |
1837 | ||
1838 | static void lo_listxattr(fuse_req_t req, fuse_ino_t ino, size_t size) | |
1839 | { | |
9f59d175 | 1840 | struct lo_data *lo = lo_data(req); |
7387863d DDAG |
1841 | char *value = NULL; |
1842 | char procname[64]; | |
92fb57b8 | 1843 | struct lo_inode *inode; |
7387863d DDAG |
1844 | ssize_t ret; |
1845 | int saverr; | |
9f59d175 | 1846 | int fd = -1; |
7387863d | 1847 | |
92fb57b8 SH |
1848 | inode = lo_inode(req, ino); |
1849 | if (!inode) { | |
1850 | fuse_reply_err(req, EBADF); | |
1851 | return; | |
1852 | } | |
1853 | ||
7387863d DDAG |
1854 | saverr = ENOSYS; |
1855 | if (!lo_data(req)->xattr) { | |
1856 | goto out; | |
1857 | } | |
1858 | ||
d240314a EG |
1859 | fuse_log(FUSE_LOG_DEBUG, "lo_listxattr(ino=%" PRIu64 ", size=%zd)\n", ino, |
1860 | size); | |
7387863d DDAG |
1861 | |
1862 | if (inode->is_symlink) { | |
1863 | /* Sorry, no race free way to listxattr on symlink. */ | |
1864 | saverr = EPERM; | |
1865 | goto out; | |
1866 | } | |
1867 | ||
9f59d175 SH |
1868 | sprintf(procname, "%i", inode->fd); |
1869 | fd = openat(lo->proc_self_fd, procname, O_RDONLY); | |
1870 | if (fd < 0) { | |
1871 | goto out_err; | |
1872 | } | |
7387863d DDAG |
1873 | |
1874 | if (size) { | |
1875 | value = malloc(size); | |
1876 | if (!value) { | |
1877 | goto out_err; | |
1878 | } | |
1879 | ||
9f59d175 | 1880 | ret = flistxattr(fd, value, size); |
7387863d DDAG |
1881 | if (ret == -1) { |
1882 | goto out_err; | |
1883 | } | |
1884 | saverr = 0; | |
1885 | if (ret == 0) { | |
1886 | goto out; | |
1887 | } | |
1888 | ||
1889 | fuse_reply_buf(req, value, ret); | |
1890 | } else { | |
9f59d175 | 1891 | ret = flistxattr(fd, NULL, 0); |
7387863d DDAG |
1892 | if (ret == -1) { |
1893 | goto out_err; | |
1894 | } | |
1895 | ||
1896 | fuse_reply_xattr(req, ret); | |
1897 | } | |
7c6b6602 | 1898 | out_free: |
7387863d | 1899 | free(value); |
9f59d175 SH |
1900 | |
1901 | if (fd >= 0) { | |
1902 | close(fd); | |
1903 | } | |
7387863d | 1904 | return; |
7c6b6602 DDAG |
1905 | |
1906 | out_err: | |
7387863d | 1907 | saverr = errno; |
7c6b6602 | 1908 | out: |
7387863d DDAG |
1909 | fuse_reply_err(req, saverr); |
1910 | goto out_free; | |
7c6b6602 DDAG |
1911 | } |
1912 | ||
1913 | static void lo_setxattr(fuse_req_t req, fuse_ino_t ino, const char *name, | |
7387863d | 1914 | const char *value, size_t size, int flags) |
7c6b6602 | 1915 | { |
7387863d | 1916 | char procname[64]; |
9f59d175 | 1917 | struct lo_data *lo = lo_data(req); |
92fb57b8 | 1918 | struct lo_inode *inode; |
7387863d DDAG |
1919 | ssize_t ret; |
1920 | int saverr; | |
9f59d175 | 1921 | int fd = -1; |
7c6b6602 | 1922 | |
92fb57b8 SH |
1923 | inode = lo_inode(req, ino); |
1924 | if (!inode) { | |
1925 | fuse_reply_err(req, EBADF); | |
1926 | return; | |
1927 | } | |
1928 | ||
7387863d DDAG |
1929 | saverr = ENOSYS; |
1930 | if (!lo_data(req)->xattr) { | |
1931 | goto out; | |
1932 | } | |
7c6b6602 | 1933 | |
d240314a EG |
1934 | fuse_log(FUSE_LOG_DEBUG, "lo_setxattr(ino=%" PRIu64 |
1935 | ", name=%s value=%s size=%zd)\n", ino, name, value, size); | |
7c6b6602 | 1936 | |
7387863d DDAG |
1937 | if (inode->is_symlink) { |
1938 | /* Sorry, no race free way to setxattr on symlink. */ | |
1939 | saverr = EPERM; | |
1940 | goto out; | |
1941 | } | |
7c6b6602 | 1942 | |
9f59d175 SH |
1943 | sprintf(procname, "%i", inode->fd); |
1944 | fd = openat(lo->proc_self_fd, procname, O_RDWR); | |
1945 | if (fd < 0) { | |
1946 | saverr = errno; | |
1947 | goto out; | |
1948 | } | |
7c6b6602 | 1949 | |
9f59d175 | 1950 | ret = fsetxattr(fd, name, value, size, flags); |
7387863d | 1951 | saverr = ret == -1 ? errno : 0; |
7c6b6602 DDAG |
1952 | |
1953 | out: | |
9f59d175 SH |
1954 | if (fd >= 0) { |
1955 | close(fd); | |
1956 | } | |
7387863d | 1957 | fuse_reply_err(req, saverr); |
7c6b6602 DDAG |
1958 | } |
1959 | ||
1960 | static void lo_removexattr(fuse_req_t req, fuse_ino_t ino, const char *name) | |
1961 | { | |
7387863d | 1962 | char procname[64]; |
9f59d175 | 1963 | struct lo_data *lo = lo_data(req); |
92fb57b8 | 1964 | struct lo_inode *inode; |
7387863d DDAG |
1965 | ssize_t ret; |
1966 | int saverr; | |
9f59d175 | 1967 | int fd = -1; |
7c6b6602 | 1968 | |
92fb57b8 SH |
1969 | inode = lo_inode(req, ino); |
1970 | if (!inode) { | |
1971 | fuse_reply_err(req, EBADF); | |
1972 | return; | |
1973 | } | |
1974 | ||
7387863d DDAG |
1975 | saverr = ENOSYS; |
1976 | if (!lo_data(req)->xattr) { | |
1977 | goto out; | |
1978 | } | |
7c6b6602 | 1979 | |
d240314a EG |
1980 | fuse_log(FUSE_LOG_DEBUG, "lo_removexattr(ino=%" PRIu64 ", name=%s)\n", ino, |
1981 | name); | |
7c6b6602 | 1982 | |
7387863d DDAG |
1983 | if (inode->is_symlink) { |
1984 | /* Sorry, no race free way to setxattr on symlink. */ | |
1985 | saverr = EPERM; | |
1986 | goto out; | |
1987 | } | |
7c6b6602 | 1988 | |
9f59d175 SH |
1989 | sprintf(procname, "%i", inode->fd); |
1990 | fd = openat(lo->proc_self_fd, procname, O_RDWR); | |
1991 | if (fd < 0) { | |
1992 | saverr = errno; | |
1993 | goto out; | |
1994 | } | |
7c6b6602 | 1995 | |
9f59d175 | 1996 | ret = fremovexattr(fd, name); |
7387863d | 1997 | saverr = ret == -1 ? errno : 0; |
7c6b6602 DDAG |
1998 | |
1999 | out: | |
9f59d175 SH |
2000 | if (fd >= 0) { |
2001 | close(fd); | |
2002 | } | |
7387863d | 2003 | fuse_reply_err(req, saverr); |
7c6b6602 DDAG |
2004 | } |
2005 | ||
2006 | #ifdef HAVE_COPY_FILE_RANGE | |
2007 | static void lo_copy_file_range(fuse_req_t req, fuse_ino_t ino_in, off_t off_in, | |
7387863d DDAG |
2008 | struct fuse_file_info *fi_in, fuse_ino_t ino_out, |
2009 | off_t off_out, struct fuse_file_info *fi_out, | |
2010 | size_t len, int flags) | |
7c6b6602 | 2011 | { |
73b4d19d | 2012 | int in_fd, out_fd; |
7387863d DDAG |
2013 | ssize_t res; |
2014 | ||
73b4d19d SH |
2015 | in_fd = lo_fi_fd(req, fi_in); |
2016 | out_fd = lo_fi_fd(req, fi_out); | |
2017 | ||
2018 | fuse_log(FUSE_LOG_DEBUG, | |
2019 | "lo_copy_file_range(ino=%" PRIu64 "/fd=%d, " | |
2020 | "off=%lu, ino=%" PRIu64 "/fd=%d, " | |
2021 | "off=%lu, size=%zd, flags=0x%x)\n", | |
2022 | ino_in, in_fd, off_in, ino_out, out_fd, off_out, len, flags); | |
7387863d | 2023 | |
73b4d19d | 2024 | res = copy_file_range(in_fd, &off_in, out_fd, &off_out, len, flags); |
7387863d DDAG |
2025 | if (res < 0) { |
2026 | fuse_reply_err(req, -errno); | |
2027 | } else { | |
2028 | fuse_reply_write(req, res); | |
2029 | } | |
7c6b6602 DDAG |
2030 | } |
2031 | #endif | |
2032 | ||
2033 | static void lo_lseek(fuse_req_t req, fuse_ino_t ino, off_t off, int whence, | |
7387863d | 2034 | struct fuse_file_info *fi) |
7c6b6602 | 2035 | { |
7387863d DDAG |
2036 | off_t res; |
2037 | ||
2038 | (void)ino; | |
73b4d19d | 2039 | res = lseek(lo_fi_fd(req, fi), off, whence); |
7387863d DDAG |
2040 | if (res != -1) { |
2041 | fuse_reply_lseek(req, res); | |
2042 | } else { | |
2043 | fuse_reply_err(req, errno); | |
2044 | } | |
7c6b6602 DDAG |
2045 | } |
2046 | ||
2047 | static struct fuse_lowlevel_ops lo_oper = { | |
7387863d DDAG |
2048 | .init = lo_init, |
2049 | .lookup = lo_lookup, | |
2050 | .mkdir = lo_mkdir, | |
2051 | .mknod = lo_mknod, | |
2052 | .symlink = lo_symlink, | |
2053 | .link = lo_link, | |
2054 | .unlink = lo_unlink, | |
2055 | .rmdir = lo_rmdir, | |
2056 | .rename = lo_rename, | |
2057 | .forget = lo_forget, | |
2058 | .forget_multi = lo_forget_multi, | |
2059 | .getattr = lo_getattr, | |
2060 | .setattr = lo_setattr, | |
2061 | .readlink = lo_readlink, | |
2062 | .opendir = lo_opendir, | |
2063 | .readdir = lo_readdir, | |
2064 | .readdirplus = lo_readdirplus, | |
2065 | .releasedir = lo_releasedir, | |
2066 | .fsyncdir = lo_fsyncdir, | |
2067 | .create = lo_create, | |
2068 | .open = lo_open, | |
2069 | .release = lo_release, | |
2070 | .flush = lo_flush, | |
2071 | .fsync = lo_fsync, | |
2072 | .read = lo_read, | |
2073 | .write_buf = lo_write_buf, | |
2074 | .statfs = lo_statfs, | |
2075 | .fallocate = lo_fallocate, | |
2076 | .flock = lo_flock, | |
2077 | .getxattr = lo_getxattr, | |
2078 | .listxattr = lo_listxattr, | |
2079 | .setxattr = lo_setxattr, | |
2080 | .removexattr = lo_removexattr, | |
7c6b6602 | 2081 | #ifdef HAVE_COPY_FILE_RANGE |
7387863d | 2082 | .copy_file_range = lo_copy_file_range, |
7c6b6602 | 2083 | #endif |
7387863d | 2084 | .lseek = lo_lseek, |
7c6b6602 DDAG |
2085 | }; |
2086 | ||
45018fbb SH |
2087 | /* Print vhost-user.json backend program capabilities */ |
2088 | static void print_capabilities(void) | |
2089 | { | |
2090 | printf("{\n"); | |
2091 | printf(" \"type\": \"fs\"\n"); | |
2092 | printf("}\n"); | |
2093 | } | |
2094 | ||
d74830d1 | 2095 | /* |
8e1d4ef2 | 2096 | * Move to a new mount, net, and pid namespaces to isolate this process. |
d74830d1 | 2097 | */ |
8e1d4ef2 | 2098 | static void setup_namespaces(struct lo_data *lo, struct fuse_session *se) |
d74830d1 | 2099 | { |
8e1d4ef2 SH |
2100 | pid_t child; |
2101 | ||
2102 | /* | |
2103 | * Create a new pid namespace for *child* processes. We'll have to | |
2104 | * fork in order to enter the new pid namespace. A new mount namespace | |
2105 | * is also needed so that we can remount /proc for the new pid | |
2106 | * namespace. | |
2107 | * | |
2108 | * Our UNIX domain sockets have been created. Now we can move to | |
2109 | * an empty network namespace to prevent TCP/IP and other network | |
2110 | * activity in case this process is compromised. | |
2111 | */ | |
2112 | if (unshare(CLONE_NEWPID | CLONE_NEWNS | CLONE_NEWNET) != 0) { | |
2113 | fuse_log(FUSE_LOG_ERR, "unshare(CLONE_NEWPID | CLONE_NEWNS): %m\n"); | |
2114 | exit(1); | |
2115 | } | |
2116 | ||
2117 | child = fork(); | |
2118 | if (child < 0) { | |
2119 | fuse_log(FUSE_LOG_ERR, "fork() failed: %m\n"); | |
2120 | exit(1); | |
2121 | } | |
2122 | if (child > 0) { | |
2123 | pid_t waited; | |
2124 | int wstatus; | |
2125 | ||
2126 | /* The parent waits for the child */ | |
2127 | do { | |
2128 | waited = waitpid(child, &wstatus, 0); | |
2129 | } while (waited < 0 && errno == EINTR && !se->exited); | |
2130 | ||
2131 | /* We were terminated by a signal, see fuse_signals.c */ | |
2132 | if (se->exited) { | |
2133 | exit(0); | |
2134 | } | |
2135 | ||
2136 | if (WIFEXITED(wstatus)) { | |
2137 | exit(WEXITSTATUS(wstatus)); | |
2138 | } | |
2139 | ||
2140 | exit(1); | |
2141 | } | |
2142 | ||
2143 | /* Send us SIGTERM when the parent thread terminates, see prctl(2) */ | |
2144 | prctl(PR_SET_PDEATHSIG, SIGTERM); | |
2145 | ||
2146 | /* | |
2147 | * If the mounts have shared propagation then we want to opt out so our | |
2148 | * mount changes don't affect the parent mount namespace. | |
2149 | */ | |
2150 | if (mount(NULL, "/", NULL, MS_REC | MS_SLAVE, NULL) < 0) { | |
2151 | fuse_log(FUSE_LOG_ERR, "mount(/, MS_REC|MS_SLAVE): %m\n"); | |
2152 | exit(1); | |
2153 | } | |
2154 | ||
2155 | /* The child must remount /proc to use the new pid namespace */ | |
2156 | if (mount("proc", "/proc", "proc", | |
2157 | MS_NODEV | MS_NOEXEC | MS_NOSUID | MS_RELATIME, NULL) < 0) { | |
2158 | fuse_log(FUSE_LOG_ERR, "mount(/proc): %m\n"); | |
2159 | exit(1); | |
2160 | } | |
2161 | ||
2162 | /* Now we can get our /proc/self/fd directory file descriptor */ | |
2163 | lo->proc_self_fd = open("/proc/self/fd", O_PATH); | |
2164 | if (lo->proc_self_fd == -1) { | |
2165 | fuse_log(FUSE_LOG_ERR, "open(/proc/self/fd, O_PATH): %m\n"); | |
d74830d1 SH |
2166 | exit(1); |
2167 | } | |
2168 | } | |
2169 | ||
2405f3c0 DDAG |
2170 | /* |
2171 | * Capture the capability state, we'll need to restore this for individual | |
2172 | * threads later; see load_capng. | |
2173 | */ | |
2174 | static void setup_capng(void) | |
2175 | { | |
2176 | /* Note this accesses /proc so has to happen before the sandbox */ | |
2177 | if (capng_get_caps_process()) { | |
2178 | fuse_log(FUSE_LOG_ERR, "capng_get_caps_process\n"); | |
2179 | exit(1); | |
2180 | } | |
2181 | pthread_mutex_init(&cap.mutex, NULL); | |
2182 | pthread_mutex_lock(&cap.mutex); | |
2183 | cap.saved = capng_save_state(); | |
2184 | if (!cap.saved) { | |
2185 | fuse_log(FUSE_LOG_ERR, "capng_save_state\n"); | |
2186 | exit(1); | |
2187 | } | |
2188 | pthread_mutex_unlock(&cap.mutex); | |
2189 | } | |
2190 | ||
2191 | static void cleanup_capng(void) | |
2192 | { | |
2193 | free(cap.saved); | |
2194 | cap.saved = NULL; | |
2195 | pthread_mutex_destroy(&cap.mutex); | |
2196 | } | |
2197 | ||
2198 | ||
8e1d4ef2 SH |
2199 | /* |
2200 | * Make the source directory our root so symlinks cannot escape and no other | |
2201 | * files are accessible. Assumes unshare(CLONE_NEWNS) was already called. | |
2202 | */ | |
2203 | static void setup_mounts(const char *source) | |
5baa3b8e SH |
2204 | { |
2205 | int oldroot; | |
2206 | int newroot; | |
2207 | ||
8e1d4ef2 SH |
2208 | if (mount(source, source, NULL, MS_BIND, NULL) < 0) { |
2209 | fuse_log(FUSE_LOG_ERR, "mount(%s, %s, MS_BIND): %m\n", source, source); | |
2210 | exit(1); | |
2211 | } | |
2212 | ||
2213 | /* This magic is based on lxc's lxc_pivot_root() */ | |
5baa3b8e SH |
2214 | oldroot = open("/", O_DIRECTORY | O_RDONLY | O_CLOEXEC); |
2215 | if (oldroot < 0) { | |
2216 | fuse_log(FUSE_LOG_ERR, "open(/): %m\n"); | |
2217 | exit(1); | |
2218 | } | |
2219 | ||
2220 | newroot = open(source, O_DIRECTORY | O_RDONLY | O_CLOEXEC); | |
2221 | if (newroot < 0) { | |
2222 | fuse_log(FUSE_LOG_ERR, "open(%s): %m\n", source); | |
2223 | exit(1); | |
2224 | } | |
2225 | ||
2226 | if (fchdir(newroot) < 0) { | |
2227 | fuse_log(FUSE_LOG_ERR, "fchdir(newroot): %m\n"); | |
2228 | exit(1); | |
2229 | } | |
2230 | ||
2231 | if (syscall(__NR_pivot_root, ".", ".") < 0) { | |
2232 | fuse_log(FUSE_LOG_ERR, "pivot_root(., .): %m\n"); | |
2233 | exit(1); | |
2234 | } | |
2235 | ||
2236 | if (fchdir(oldroot) < 0) { | |
2237 | fuse_log(FUSE_LOG_ERR, "fchdir(oldroot): %m\n"); | |
2238 | exit(1); | |
2239 | } | |
2240 | ||
2241 | if (mount("", ".", "", MS_SLAVE | MS_REC, NULL) < 0) { | |
2242 | fuse_log(FUSE_LOG_ERR, "mount(., MS_SLAVE | MS_REC): %m\n"); | |
2243 | exit(1); | |
2244 | } | |
2245 | ||
2246 | if (umount2(".", MNT_DETACH) < 0) { | |
2247 | fuse_log(FUSE_LOG_ERR, "umount2(., MNT_DETACH): %m\n"); | |
2248 | exit(1); | |
2249 | } | |
2250 | ||
2251 | if (fchdir(newroot) < 0) { | |
2252 | fuse_log(FUSE_LOG_ERR, "fchdir(newroot): %m\n"); | |
2253 | exit(1); | |
2254 | } | |
2255 | ||
2256 | close(newroot); | |
2257 | close(oldroot); | |
2258 | } | |
2259 | ||
5baa3b8e SH |
2260 | /* |
2261 | * Lock down this process to prevent access to other processes or files outside | |
2262 | * source directory. This reduces the impact of arbitrary code execution bugs. | |
2263 | */ | |
f185621d SH |
2264 | static void setup_sandbox(struct lo_data *lo, struct fuse_session *se, |
2265 | bool enable_syslog) | |
5baa3b8e | 2266 | { |
8e1d4ef2 SH |
2267 | setup_namespaces(lo, se); |
2268 | setup_mounts(lo->source); | |
f185621d | 2269 | setup_seccomp(enable_syslog); |
5baa3b8e SH |
2270 | } |
2271 | ||
01a6dc95 SH |
2272 | /* Raise the maximum number of open file descriptors */ |
2273 | static void setup_nofile_rlimit(void) | |
2274 | { | |
2275 | const rlim_t max_fds = 1000000; | |
2276 | struct rlimit rlim; | |
2277 | ||
2278 | if (getrlimit(RLIMIT_NOFILE, &rlim) < 0) { | |
2279 | fuse_log(FUSE_LOG_ERR, "getrlimit(RLIMIT_NOFILE): %m\n"); | |
2280 | exit(1); | |
2281 | } | |
2282 | ||
2283 | if (rlim.rlim_cur >= max_fds) { | |
2284 | return; /* nothing to do */ | |
2285 | } | |
2286 | ||
2287 | rlim.rlim_cur = max_fds; | |
2288 | rlim.rlim_max = max_fds; | |
2289 | ||
2290 | if (setrlimit(RLIMIT_NOFILE, &rlim) < 0) { | |
2291 | /* Ignore SELinux denials */ | |
2292 | if (errno == EPERM) { | |
2293 | return; | |
2294 | } | |
2295 | ||
2296 | fuse_log(FUSE_LOG_ERR, "setrlimit(RLIMIT_NOFILE): %m\n"); | |
2297 | exit(1); | |
2298 | } | |
2299 | } | |
2300 | ||
f185621d SH |
2301 | static void log_func(enum fuse_log_level level, const char *fmt, va_list ap) |
2302 | { | |
36f38469 MM |
2303 | g_autofree char *localfmt = NULL; |
2304 | ||
d240314a EG |
2305 | if (current_log_level < level) { |
2306 | return; | |
2307 | } | |
2308 | ||
36f38469 | 2309 | if (current_log_level == FUSE_LOG_DEBUG) { |
50fb955a MM |
2310 | if (!use_syslog) { |
2311 | localfmt = g_strdup_printf("[%" PRId64 "] [ID: %08ld] %s", | |
2312 | get_clock(), syscall(__NR_gettid), fmt); | |
2313 | } else { | |
2314 | localfmt = g_strdup_printf("[ID: %08ld] %s", syscall(__NR_gettid), | |
2315 | fmt); | |
2316 | } | |
36f38469 MM |
2317 | fmt = localfmt; |
2318 | } | |
2319 | ||
f185621d SH |
2320 | if (use_syslog) { |
2321 | int priority = LOG_ERR; | |
2322 | switch (level) { | |
2323 | case FUSE_LOG_EMERG: | |
2324 | priority = LOG_EMERG; | |
2325 | break; | |
2326 | case FUSE_LOG_ALERT: | |
2327 | priority = LOG_ALERT; | |
2328 | break; | |
2329 | case FUSE_LOG_CRIT: | |
2330 | priority = LOG_CRIT; | |
2331 | break; | |
2332 | case FUSE_LOG_ERR: | |
2333 | priority = LOG_ERR; | |
2334 | break; | |
2335 | case FUSE_LOG_WARNING: | |
2336 | priority = LOG_WARNING; | |
2337 | break; | |
2338 | case FUSE_LOG_NOTICE: | |
2339 | priority = LOG_NOTICE; | |
2340 | break; | |
2341 | case FUSE_LOG_INFO: | |
2342 | priority = LOG_INFO; | |
2343 | break; | |
2344 | case FUSE_LOG_DEBUG: | |
2345 | priority = LOG_DEBUG; | |
2346 | break; | |
2347 | } | |
2348 | vsyslog(priority, fmt, ap); | |
2349 | } else { | |
2350 | vfprintf(stderr, fmt, ap); | |
2351 | } | |
2352 | } | |
2353 | ||
7c6b6602 DDAG |
2354 | int main(int argc, char *argv[]) |
2355 | { | |
7387863d DDAG |
2356 | struct fuse_args args = FUSE_ARGS_INIT(argc, argv); |
2357 | struct fuse_session *se; | |
2358 | struct fuse_cmdline_opts opts; | |
9f59d175 SH |
2359 | struct lo_data lo = { |
2360 | .debug = 0, | |
2361 | .writeback = 0, | |
2362 | .proc_self_fd = -1, | |
2363 | }; | |
92fb57b8 | 2364 | struct lo_map_elem *root_elem; |
7387863d DDAG |
2365 | int ret = -1; |
2366 | ||
2367 | /* Don't mask creation mode, kernel already did that */ | |
2368 | umask(0); | |
2369 | ||
2370 | pthread_mutex_init(&lo.mutex, NULL); | |
2371 | lo.root.next = lo.root.prev = &lo.root; | |
2372 | lo.root.fd = -1; | |
92fb57b8 | 2373 | lo.root.fuse_ino = FUSE_ROOT_ID; |
7387863d DDAG |
2374 | lo.cache = CACHE_NORMAL; |
2375 | ||
92fb57b8 SH |
2376 | /* |
2377 | * Set up the ino map like this: | |
2378 | * [0] Reserved (will not be used) | |
2379 | * [1] Root inode | |
2380 | */ | |
2381 | lo_map_init(&lo.ino_map); | |
2382 | lo_map_reserve(&lo.ino_map, 0)->in_use = false; | |
2383 | root_elem = lo_map_reserve(&lo.ino_map, lo.root.fuse_ino); | |
2384 | root_elem->inode = &lo.root; | |
2385 | ||
b39bce12 | 2386 | lo_map_init(&lo.dirp_map); |
73b4d19d | 2387 | lo_map_init(&lo.fd_map); |
b39bce12 | 2388 | |
7387863d DDAG |
2389 | if (fuse_parse_cmdline(&args, &opts) != 0) { |
2390 | return 1; | |
2391 | } | |
f185621d SH |
2392 | fuse_set_log_func(log_func); |
2393 | use_syslog = opts.syslog; | |
2394 | if (use_syslog) { | |
2395 | openlog("virtiofsd", LOG_PID, LOG_DAEMON); | |
2396 | } | |
7387863d | 2397 | if (opts.show_help) { |
67aab022 | 2398 | printf("usage: %s [options]\n\n", argv[0]); |
7387863d | 2399 | fuse_cmdline_help(); |
4ff075f7 | 2400 | printf(" -o source=PATH shared directory tree\n"); |
7387863d DDAG |
2401 | fuse_lowlevel_help(); |
2402 | ret = 0; | |
2403 | goto err_out1; | |
2404 | } else if (opts.show_version) { | |
2405 | fuse_lowlevel_version(); | |
2406 | ret = 0; | |
2407 | goto err_out1; | |
45018fbb SH |
2408 | } else if (opts.print_capabilities) { |
2409 | print_capabilities(); | |
2410 | ret = 0; | |
2411 | goto err_out1; | |
7387863d DDAG |
2412 | } |
2413 | ||
7387863d DDAG |
2414 | if (fuse_opt_parse(&args, &lo, lo_opts, NULL) == -1) { |
2415 | return 1; | |
2416 | } | |
2417 | ||
d240314a EG |
2418 | /* |
2419 | * log_level is 0 if not configured via cmd options (0 is LOG_EMERG, | |
2420 | * and we don't use this log level). | |
2421 | */ | |
2422 | if (opts.log_level != 0) { | |
2423 | current_log_level = opts.log_level; | |
2424 | } | |
7387863d | 2425 | lo.debug = opts.debug; |
d240314a EG |
2426 | if (lo.debug) { |
2427 | current_log_level = FUSE_LOG_DEBUG; | |
2428 | } | |
7387863d | 2429 | lo.root.refcount = 2; |
d240314a | 2430 | |
7387863d DDAG |
2431 | if (lo.source) { |
2432 | struct stat stat; | |
2433 | int res; | |
2434 | ||
2435 | res = lstat(lo.source, &stat); | |
2436 | if (res == -1) { | |
2437 | fuse_log(FUSE_LOG_ERR, "failed to stat source (\"%s\"): %m\n", | |
2438 | lo.source); | |
2439 | exit(1); | |
2440 | } | |
2441 | if (!S_ISDIR(stat.st_mode)) { | |
2442 | fuse_log(FUSE_LOG_ERR, "source is not a directory\n"); | |
2443 | exit(1); | |
2444 | } | |
2445 | ||
2446 | } else { | |
2447 | lo.source = "/"; | |
2448 | } | |
2449 | lo.root.is_symlink = false; | |
2450 | if (!lo.timeout_set) { | |
2451 | switch (lo.cache) { | |
2452 | case CACHE_NEVER: | |
2453 | lo.timeout = 0.0; | |
2454 | break; | |
2455 | ||
2456 | case CACHE_NORMAL: | |
2457 | lo.timeout = 1.0; | |
2458 | break; | |
2459 | ||
2460 | case CACHE_ALWAYS: | |
2461 | lo.timeout = 86400.0; | |
2462 | break; | |
2463 | } | |
2464 | } else if (lo.timeout < 0) { | |
2465 | fuse_log(FUSE_LOG_ERR, "timeout is negative (%lf)\n", lo.timeout); | |
2466 | exit(1); | |
2467 | } | |
2468 | ||
2469 | lo.root.fd = open(lo.source, O_PATH); | |
5baa3b8e | 2470 | |
7387863d DDAG |
2471 | if (lo.root.fd == -1) { |
2472 | fuse_log(FUSE_LOG_ERR, "open(\"%s\", O_PATH): %m\n", lo.source); | |
2473 | exit(1); | |
2474 | } | |
2475 | ||
2476 | se = fuse_session_new(&args, &lo_oper, sizeof(lo_oper), &lo); | |
2477 | if (se == NULL) { | |
2478 | goto err_out1; | |
2479 | } | |
2480 | ||
2481 | if (fuse_set_signal_handlers(se) != 0) { | |
2482 | goto err_out2; | |
2483 | } | |
2484 | ||
67aab022 | 2485 | if (fuse_session_mount(se) != 0) { |
7387863d DDAG |
2486 | goto err_out3; |
2487 | } | |
2488 | ||
2489 | fuse_daemonize(opts.foreground); | |
2490 | ||
01a6dc95 SH |
2491 | setup_nofile_rlimit(); |
2492 | ||
2405f3c0 DDAG |
2493 | /* Must be before sandbox since it wants /proc */ |
2494 | setup_capng(); | |
2495 | ||
f185621d | 2496 | setup_sandbox(&lo, se, opts.syslog); |
5baa3b8e | 2497 | |
7387863d | 2498 | /* Block until ctrl+c or fusermount -u */ |
f6f3573c | 2499 | ret = virtio_loop(se); |
7387863d DDAG |
2500 | |
2501 | fuse_session_unmount(se); | |
2405f3c0 | 2502 | cleanup_capng(); |
7c6b6602 | 2503 | err_out3: |
7387863d | 2504 | fuse_remove_signal_handlers(se); |
7c6b6602 | 2505 | err_out2: |
7387863d | 2506 | fuse_session_destroy(se); |
7c6b6602 | 2507 | err_out1: |
7387863d | 2508 | fuse_opt_free_args(&args); |
7c6b6602 | 2509 | |
73b4d19d | 2510 | lo_map_destroy(&lo.fd_map); |
b39bce12 | 2511 | lo_map_destroy(&lo.dirp_map); |
92fb57b8 SH |
2512 | lo_map_destroy(&lo.ino_map); |
2513 | ||
9f59d175 SH |
2514 | if (lo.proc_self_fd >= 0) { |
2515 | close(lo.proc_self_fd); | |
2516 | } | |
2517 | ||
7387863d DDAG |
2518 | if (lo.root.fd >= 0) { |
2519 | close(lo.root.fd); | |
2520 | } | |
7c6b6602 | 2521 | |
7387863d | 2522 | return ret ? 1 : 0; |
7c6b6602 | 2523 | } |