1// SPDX-License-Identifier: GPL-2.0
2#include <linux/fanotify.h>
3#include <linux/fdtable.h>
4#include <linux/fsnotify_backend.h>
5#include <linux/init.h>
6#include <linux/jiffies.h>
7#include <linux/kernel.h> /* UINT_MAX */
8#include <linux/mount.h>
9#include <linux/sched.h>
10#include <linux/sched/user.h>
11#include <linux/sched/signal.h>
12#include <linux/types.h>
13#include <linux/wait.h>
14#include <linux/audit.h>
15#include <linux/sched/mm.h>
16#include <linux/statfs.h>
17#include <linux/stringhash.h>
18
19#include "fanotify.h"
20
21static bool fanotify_path_equal(const struct path *p1, const struct path *p2)
22{
23 return p1->mnt == p2->mnt && p1->dentry == p2->dentry;
24}
25
26static unsigned int fanotify_hash_path(const struct path *path)
27{
28 return hash_ptr(ptr: path->dentry, FANOTIFY_EVENT_HASH_BITS) ^
29 hash_ptr(ptr: path->mnt, FANOTIFY_EVENT_HASH_BITS);
30}
31
32static inline bool fanotify_fsid_equal(__kernel_fsid_t *fsid1,
33 __kernel_fsid_t *fsid2)
34{
35 return fsid1->val[0] == fsid2->val[0] && fsid1->val[1] == fsid2->val[1];
36}
37
38static unsigned int fanotify_hash_fsid(__kernel_fsid_t *fsid)
39{
40 return hash_32(val: fsid->val[0], FANOTIFY_EVENT_HASH_BITS) ^
41 hash_32(val: fsid->val[1], FANOTIFY_EVENT_HASH_BITS);
42}
43
44static bool fanotify_fh_equal(struct fanotify_fh *fh1,
45 struct fanotify_fh *fh2)
46{
47 if (fh1->type != fh2->type || fh1->len != fh2->len)
48 return false;
49
50 return !fh1->len ||
51 !memcmp(p: fanotify_fh_buf(fh: fh1), q: fanotify_fh_buf(fh: fh2), size: fh1->len);
52}
53
54static unsigned int fanotify_hash_fh(struct fanotify_fh *fh)
55{
56 long salt = (long)fh->type | (long)fh->len << 8;
57
58 /*
59 * full_name_hash() works long by long, so it handles fh buf optimally.
60 */
61 return full_name_hash(salt: (void *)salt, fanotify_fh_buf(fh), fh->len);
62}
63
64static bool fanotify_fid_event_equal(struct fanotify_fid_event *ffe1,
65 struct fanotify_fid_event *ffe2)
66{
67 /* Do not merge fid events without object fh */
68 if (!ffe1->object_fh.len)
69 return false;
70
71 return fanotify_fsid_equal(fsid1: &ffe1->fsid, fsid2: &ffe2->fsid) &&
72 fanotify_fh_equal(fh1: &ffe1->object_fh, fh2: &ffe2->object_fh);
73}
74
75static bool fanotify_info_equal(struct fanotify_info *info1,
76 struct fanotify_info *info2)
77{
78 if (info1->dir_fh_totlen != info2->dir_fh_totlen ||
79 info1->dir2_fh_totlen != info2->dir2_fh_totlen ||
80 info1->file_fh_totlen != info2->file_fh_totlen ||
81 info1->name_len != info2->name_len ||
82 info1->name2_len != info2->name2_len)
83 return false;
84
85 if (info1->dir_fh_totlen &&
86 !fanotify_fh_equal(fh1: fanotify_info_dir_fh(info: info1),
87 fh2: fanotify_info_dir_fh(info: info2)))
88 return false;
89
90 if (info1->dir2_fh_totlen &&
91 !fanotify_fh_equal(fh1: fanotify_info_dir2_fh(info: info1),
92 fh2: fanotify_info_dir2_fh(info: info2)))
93 return false;
94
95 if (info1->file_fh_totlen &&
96 !fanotify_fh_equal(fh1: fanotify_info_file_fh(info: info1),
97 fh2: fanotify_info_file_fh(info: info2)))
98 return false;
99
100 if (info1->name_len &&
101 memcmp(p: fanotify_info_name(info: info1), q: fanotify_info_name(info: info2),
102 size: info1->name_len))
103 return false;
104
105 return !info1->name2_len ||
106 !memcmp(p: fanotify_info_name2(info: info1), q: fanotify_info_name2(info: info2),
107 size: info1->name2_len);
108}
109
110static bool fanotify_name_event_equal(struct fanotify_name_event *fne1,
111 struct fanotify_name_event *fne2)
112{
113 struct fanotify_info *info1 = &fne1->info;
114 struct fanotify_info *info2 = &fne2->info;
115
116 /* Do not merge name events without dir fh */
117 if (!info1->dir_fh_totlen)
118 return false;
119
120 if (!fanotify_fsid_equal(fsid1: &fne1->fsid, fsid2: &fne2->fsid))
121 return false;
122
123 return fanotify_info_equal(info1, info2);
124}
125
126static bool fanotify_error_event_equal(struct fanotify_error_event *fee1,
127 struct fanotify_error_event *fee2)
128{
129 /* Error events against the same file system are always merged. */
130 if (!fanotify_fsid_equal(fsid1: &fee1->fsid, fsid2: &fee2->fsid))
131 return false;
132
133 return true;
134}
135
136static bool fanotify_should_merge(struct fanotify_event *old,
137 struct fanotify_event *new)
138{
139 pr_debug("%s: old=%p new=%p\n", __func__, old, new);
140
141 if (old->hash != new->hash ||
142 old->type != new->type || old->pid != new->pid)
143 return false;
144
145 /*
146 * We want to merge many dirent events in the same dir (i.e.
147 * creates/unlinks/renames), but we do not want to merge dirent
148 * events referring to subdirs with dirent events referring to
149 * non subdirs, otherwise, user won't be able to tell from a
150 * mask FAN_CREATE|FAN_DELETE|FAN_ONDIR if it describes mkdir+
151 * unlink pair or rmdir+create pair of events.
152 */
153 if ((old->mask & FS_ISDIR) != (new->mask & FS_ISDIR))
154 return false;
155
156 /*
157 * FAN_RENAME event is reported with special info record types,
158 * so we cannot merge it with other events.
159 */
160 if ((old->mask & FAN_RENAME) != (new->mask & FAN_RENAME))
161 return false;
162
163 switch (old->type) {
164 case FANOTIFY_EVENT_TYPE_PATH:
165 return fanotify_path_equal(p1: fanotify_event_path(event: old),
166 p2: fanotify_event_path(event: new));
167 case FANOTIFY_EVENT_TYPE_FID:
168 return fanotify_fid_event_equal(ffe1: FANOTIFY_FE(event: old),
169 ffe2: FANOTIFY_FE(event: new));
170 case FANOTIFY_EVENT_TYPE_FID_NAME:
171 return fanotify_name_event_equal(fne1: FANOTIFY_NE(event: old),
172 fne2: FANOTIFY_NE(event: new));
173 case FANOTIFY_EVENT_TYPE_FS_ERROR:
174 return fanotify_error_event_equal(fee1: FANOTIFY_EE(event: old),
175 fee2: FANOTIFY_EE(event: new));
176 default:
177 WARN_ON_ONCE(1);
178 }
179
180 return false;
181}
182
183/* Limit event merges to limit CPU overhead per event */
184#define FANOTIFY_MAX_MERGE_EVENTS 128
185
186/* and the list better be locked by something too! */
187static int fanotify_merge(struct fsnotify_group *group,
188 struct fsnotify_event *event)
189{
190 struct fanotify_event *old, *new = FANOTIFY_E(fse: event);
191 unsigned int bucket = fanotify_event_hash_bucket(group, event: new);
192 struct hlist_head *hlist = &group->fanotify_data.merge_hash[bucket];
193 int i = 0;
194
195 pr_debug("%s: group=%p event=%p bucket=%u\n", __func__,
196 group, event, bucket);
197
198 /*
199 * Don't merge a permission event with any other event so that we know
200 * the event structure we have created in fanotify_handle_event() is the
201 * one we should check for permission response.
202 */
203 if (fanotify_is_perm_event(mask: new->mask))
204 return 0;
205
206 hlist_for_each_entry(old, hlist, merge_list) {
207 if (++i > FANOTIFY_MAX_MERGE_EVENTS)
208 break;
209 if (fanotify_should_merge(old, new)) {
210 old->mask |= new->mask;
211
212 if (fanotify_is_error_event(mask: old->mask))
213 FANOTIFY_EE(event: old)->err_count++;
214
215 return 1;
216 }
217 }
218
219 return 0;
220}
221
222/*
223 * Wait for response to permission event. The function also takes care of
224 * freeing the permission event (or offloads that in case the wait is canceled
225 * by a signal). The function returns 0 in case access got allowed by userspace,
226 * -EPERM in case userspace disallowed the access, and -ERESTARTSYS in case
227 * the wait got interrupted by a signal.
228 */
229static int fanotify_get_response(struct fsnotify_group *group,
230 struct fanotify_perm_event *event,
231 struct fsnotify_iter_info *iter_info)
232{
233 int ret;
234
235 pr_debug("%s: group=%p event=%p\n", __func__, group, event);
236
237 ret = wait_event_killable(group->fanotify_data.access_waitq,
238 event->state == FAN_EVENT_ANSWERED);
239 /* Signal pending? */
240 if (ret < 0) {
241 spin_lock(lock: &group->notification_lock);
242 /* Event reported to userspace and no answer yet? */
243 if (event->state == FAN_EVENT_REPORTED) {
244 /* Event will get freed once userspace answers to it */
245 event->state = FAN_EVENT_CANCELED;
246 spin_unlock(lock: &group->notification_lock);
247 return ret;
248 }
249 /* Event not yet reported? Just remove it. */
250 if (event->state == FAN_EVENT_INIT) {
251 fsnotify_remove_queued_event(group, event: &event->fae.fse);
252 /* Permission events are not supposed to be hashed */
253 WARN_ON_ONCE(!hlist_unhashed(&event->fae.merge_list));
254 }
255 /*
256 * Event may be also answered in case signal delivery raced
257 * with wakeup. In that case we have nothing to do besides
258 * freeing the event and reporting error.
259 */
260 spin_unlock(lock: &group->notification_lock);
261 goto out;
262 }
263
264 /* userspace responded, convert to something usable */
265 switch (event->response & FANOTIFY_RESPONSE_ACCESS) {
266 case FAN_ALLOW:
267 ret = 0;
268 break;
269 case FAN_DENY:
270 default:
271 ret = -EPERM;
272 }
273
274 /* Check if the response should be audited */
275 if (event->response & FAN_AUDIT)
276 audit_fanotify(response: event->response & ~FAN_AUDIT,
277 friar: &event->audit_rule);
278
279 pr_debug("%s: group=%p event=%p about to return ret=%d\n", __func__,
280 group, event, ret);
281out:
282 fsnotify_destroy_event(group, event: &event->fae.fse);
283
284 return ret;
285}
286
287/*
288 * This function returns a mask for an event that only contains the flags
289 * that have been specifically requested by the user. Flags that may have
290 * been included within the event mask, but have not been explicitly
291 * requested by the user, will not be present in the returned mask.
292 */
293static u32 fanotify_group_event_mask(struct fsnotify_group *group,
294 struct fsnotify_iter_info *iter_info,
295 u32 *match_mask, u32 event_mask,
296 const void *data, int data_type,
297 struct inode *dir)
298{
299 __u32 marks_mask = 0, marks_ignore_mask = 0;
300 __u32 test_mask, user_mask = FANOTIFY_OUTGOING_EVENTS |
301 FANOTIFY_EVENT_FLAGS;
302 const struct path *path = fsnotify_data_path(data, data_type);
303 unsigned int fid_mode = FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS);
304 struct fsnotify_mark *mark;
305 bool ondir = event_mask & FAN_ONDIR;
306 int type;
307
308 pr_debug("%s: report_mask=%x mask=%x data=%p data_type=%d\n",
309 __func__, iter_info->report_mask, event_mask, data, data_type);
310
311 if (!fid_mode) {
312 /* Do we have path to open a file descriptor? */
313 if (!path)
314 return 0;
315 /* Path type events are only relevant for files and dirs */
316 if (!d_is_reg(dentry: path->dentry) && !d_can_lookup(dentry: path->dentry))
317 return 0;
318 } else if (!(fid_mode & FAN_REPORT_FID)) {
319 /* Do we have a directory inode to report? */
320 if (!dir && !ondir)
321 return 0;
322 }
323
324 fsnotify_foreach_iter_mark_type(iter_info, mark, type) {
325 /*
326 * Apply ignore mask depending on event flags in ignore mask.
327 */
328 marks_ignore_mask |=
329 fsnotify_effective_ignore_mask(mark, is_dir: ondir, iter_type: type);
330
331 /*
332 * Send the event depending on event flags in mark mask.
333 */
334 if (!fsnotify_mask_applicable(mask: mark->mask, is_dir: ondir, iter_type: type))
335 continue;
336
337 marks_mask |= mark->mask;
338
339 /* Record the mark types of this group that matched the event */
340 *match_mask |= 1U << type;
341 }
342
343 test_mask = event_mask & marks_mask & ~marks_ignore_mask;
344
345 /*
346 * For dirent modification events (create/delete/move) that do not carry
347 * the child entry name information, we report FAN_ONDIR for mkdir/rmdir
348 * so user can differentiate them from creat/unlink.
349 *
350 * For backward compatibility and consistency, do not report FAN_ONDIR
351 * to user in legacy fanotify mode (reporting fd) and report FAN_ONDIR
352 * to user in fid mode for all event types.
353 *
354 * We never report FAN_EVENT_ON_CHILD to user, but we do pass it in to
355 * fanotify_alloc_event() when group is reporting fid as indication
356 * that event happened on child.
357 */
358 if (fid_mode) {
359 /* Do not report event flags without any event */
360 if (!(test_mask & ~FANOTIFY_EVENT_FLAGS))
361 return 0;
362 } else {
363 user_mask &= ~FANOTIFY_EVENT_FLAGS;
364 }
365
366 return test_mask & user_mask;
367}
368
369/*
370 * Check size needed to encode fanotify_fh.
371 *
372 * Return size of encoded fh without fanotify_fh header.
373 * Return 0 on failure to encode.
374 */
375static int fanotify_encode_fh_len(struct inode *inode)
376{
377 int dwords = 0;
378 int fh_len;
379
380 if (!inode)
381 return 0;
382
383 exportfs_encode_fid(inode, NULL, max_len: &dwords);
384 fh_len = dwords << 2;
385
386 /*
387 * struct fanotify_error_event might be preallocated and is
388 * limited to MAX_HANDLE_SZ. This should never happen, but
389 * safeguard by forcing an invalid file handle.
390 */
391 if (WARN_ON_ONCE(fh_len > MAX_HANDLE_SZ))
392 return 0;
393
394 return fh_len;
395}
396
397/*
398 * Encode fanotify_fh.
399 *
400 * Return total size of encoded fh including fanotify_fh header.
401 * Return 0 on failure to encode.
402 */
403static int fanotify_encode_fh(struct fanotify_fh *fh, struct inode *inode,
404 unsigned int fh_len, unsigned int *hash,
405 gfp_t gfp)
406{
407 int dwords, type = 0;
408 char *ext_buf = NULL;
409 void *buf = fh->buf;
410 int err;
411
412 fh->type = FILEID_ROOT;
413 fh->len = 0;
414 fh->flags = 0;
415
416 /*
417 * Invalid FHs are used by FAN_FS_ERROR for errors not
418 * linked to any inode. The f_handle won't be reported
419 * back to userspace.
420 */
421 if (!inode)
422 goto out;
423
424 /*
425 * !gpf means preallocated variable size fh, but fh_len could
426 * be zero in that case if encoding fh len failed.
427 */
428 err = -ENOENT;
429 if (fh_len < 4 || WARN_ON_ONCE(fh_len % 4) || fh_len > MAX_HANDLE_SZ)
430 goto out_err;
431
432 /* No external buffer in a variable size allocated fh */
433 if (gfp && fh_len > FANOTIFY_INLINE_FH_LEN) {
434 /* Treat failure to allocate fh as failure to encode fh */
435 err = -ENOMEM;
436 ext_buf = kmalloc(size: fh_len, flags: gfp);
437 if (!ext_buf)
438 goto out_err;
439
440 *fanotify_fh_ext_buf_ptr(fh) = ext_buf;
441 buf = ext_buf;
442 fh->flags |= FANOTIFY_FH_FLAG_EXT_BUF;
443 }
444
445 dwords = fh_len >> 2;
446 type = exportfs_encode_fid(inode, fid: buf, max_len: &dwords);
447 err = -EINVAL;
448 if (type <= 0 || type == FILEID_INVALID || fh_len != dwords << 2)
449 goto out_err;
450
451 fh->type = type;
452 fh->len = fh_len;
453
454out:
455 /*
456 * Mix fh into event merge key. Hash might be NULL in case of
457 * unhashed FID events (i.e. FAN_FS_ERROR).
458 */
459 if (hash)
460 *hash ^= fanotify_hash_fh(fh);
461
462 return FANOTIFY_FH_HDR_LEN + fh_len;
463
464out_err:
465 pr_warn_ratelimited("fanotify: failed to encode fid (type=%d, len=%d, err=%i)\n",
466 type, fh_len, err);
467 kfree(objp: ext_buf);
468 *fanotify_fh_ext_buf_ptr(fh) = NULL;
469 /* Report the event without a file identifier on encode error */
470 fh->type = FILEID_INVALID;
471 fh->len = 0;
472 return 0;
473}
474
475/*
476 * FAN_REPORT_FID is ambiguous in that it reports the fid of the child for
477 * some events and the fid of the parent for create/delete/move events.
478 *
479 * With the FAN_REPORT_TARGET_FID flag, the fid of the child is reported
480 * also in create/delete/move events in addition to the fid of the parent
481 * and the name of the child.
482 */
483static inline bool fanotify_report_child_fid(unsigned int fid_mode, u32 mask)
484{
485 if (mask & ALL_FSNOTIFY_DIRENT_EVENTS)
486 return (fid_mode & FAN_REPORT_TARGET_FID);
487
488 return (fid_mode & FAN_REPORT_FID) && !(mask & FAN_ONDIR);
489}
490
491/*
492 * The inode to use as identifier when reporting fid depends on the event
493 * and the group flags.
494 *
495 * With the group flag FAN_REPORT_TARGET_FID, always report the child fid.
496 *
497 * Without the group flag FAN_REPORT_TARGET_FID, report the modified directory
498 * fid on dirent events and the child fid otherwise.
499 *
500 * For example:
501 * FS_ATTRIB reports the child fid even if reported on a watched parent.
502 * FS_CREATE reports the modified dir fid without FAN_REPORT_TARGET_FID.
503 * and reports the created child fid with FAN_REPORT_TARGET_FID.
504 */
505static struct inode *fanotify_fid_inode(u32 event_mask, const void *data,
506 int data_type, struct inode *dir,
507 unsigned int fid_mode)
508{
509 if ((event_mask & ALL_FSNOTIFY_DIRENT_EVENTS) &&
510 !(fid_mode & FAN_REPORT_TARGET_FID))
511 return dir;
512
513 return fsnotify_data_inode(data, data_type);
514}
515
516/*
517 * The inode to use as identifier when reporting dir fid depends on the event.
518 * Report the modified directory inode on dirent modification events.
519 * Report the "victim" inode if "victim" is a directory.
520 * Report the parent inode if "victim" is not a directory and event is
521 * reported to parent.
522 * Otherwise, do not report dir fid.
523 */
524static struct inode *fanotify_dfid_inode(u32 event_mask, const void *data,
525 int data_type, struct inode *dir)
526{
527 struct inode *inode = fsnotify_data_inode(data, data_type);
528
529 if (event_mask & ALL_FSNOTIFY_DIRENT_EVENTS)
530 return dir;
531
532 if (inode && S_ISDIR(inode->i_mode))
533 return inode;
534
535 return dir;
536}
537
538static struct fanotify_event *fanotify_alloc_path_event(const struct path *path,
539 unsigned int *hash,
540 gfp_t gfp)
541{
542 struct fanotify_path_event *pevent;
543
544 pevent = kmem_cache_alloc(cachep: fanotify_path_event_cachep, flags: gfp);
545 if (!pevent)
546 return NULL;
547
548 pevent->fae.type = FANOTIFY_EVENT_TYPE_PATH;
549 pevent->path = *path;
550 *hash ^= fanotify_hash_path(path);
551 path_get(path);
552
553 return &pevent->fae;
554}
555
556static struct fanotify_event *fanotify_alloc_perm_event(const struct path *path,
557 gfp_t gfp)
558{
559 struct fanotify_perm_event *pevent;
560
561 pevent = kmem_cache_alloc(cachep: fanotify_perm_event_cachep, flags: gfp);
562 if (!pevent)
563 return NULL;
564
565 pevent->fae.type = FANOTIFY_EVENT_TYPE_PATH_PERM;
566 pevent->response = 0;
567 pevent->hdr.type = FAN_RESPONSE_INFO_NONE;
568 pevent->hdr.pad = 0;
569 pevent->hdr.len = 0;
570 pevent->state = FAN_EVENT_INIT;
571 pevent->path = *path;
572 path_get(path);
573
574 return &pevent->fae;
575}
576
577static struct fanotify_event *fanotify_alloc_fid_event(struct inode *id,
578 __kernel_fsid_t *fsid,
579 unsigned int *hash,
580 gfp_t gfp)
581{
582 struct fanotify_fid_event *ffe;
583
584 ffe = kmem_cache_alloc(cachep: fanotify_fid_event_cachep, flags: gfp);
585 if (!ffe)
586 return NULL;
587
588 ffe->fae.type = FANOTIFY_EVENT_TYPE_FID;
589 ffe->fsid = *fsid;
590 *hash ^= fanotify_hash_fsid(fsid);
591 fanotify_encode_fh(fh: &ffe->object_fh, inode: id, fh_len: fanotify_encode_fh_len(inode: id),
592 hash, gfp);
593
594 return &ffe->fae;
595}
596
597static struct fanotify_event *fanotify_alloc_name_event(struct inode *dir,
598 __kernel_fsid_t *fsid,
599 const struct qstr *name,
600 struct inode *child,
601 struct dentry *moved,
602 unsigned int *hash,
603 gfp_t gfp)
604{
605 struct fanotify_name_event *fne;
606 struct fanotify_info *info;
607 struct fanotify_fh *dfh, *ffh;
608 struct inode *dir2 = moved ? d_inode(dentry: moved->d_parent) : NULL;
609 const struct qstr *name2 = moved ? &moved->d_name : NULL;
610 unsigned int dir_fh_len = fanotify_encode_fh_len(inode: dir);
611 unsigned int dir2_fh_len = fanotify_encode_fh_len(inode: dir2);
612 unsigned int child_fh_len = fanotify_encode_fh_len(inode: child);
613 unsigned long name_len = name ? name->len : 0;
614 unsigned long name2_len = name2 ? name2->len : 0;
615 unsigned int len, size;
616
617 /* Reserve terminating null byte even for empty name */
618 size = sizeof(*fne) + name_len + name2_len + 2;
619 if (dir_fh_len)
620 size += FANOTIFY_FH_HDR_LEN + dir_fh_len;
621 if (dir2_fh_len)
622 size += FANOTIFY_FH_HDR_LEN + dir2_fh_len;
623 if (child_fh_len)
624 size += FANOTIFY_FH_HDR_LEN + child_fh_len;
625 fne = kmalloc(size, flags: gfp);
626 if (!fne)
627 return NULL;
628
629 fne->fae.type = FANOTIFY_EVENT_TYPE_FID_NAME;
630 fne->fsid = *fsid;
631 *hash ^= fanotify_hash_fsid(fsid);
632 info = &fne->info;
633 fanotify_info_init(info);
634 if (dir_fh_len) {
635 dfh = fanotify_info_dir_fh(info);
636 len = fanotify_encode_fh(fh: dfh, inode: dir, fh_len: dir_fh_len, hash, gfp: 0);
637 fanotify_info_set_dir_fh(info, totlen: len);
638 }
639 if (dir2_fh_len) {
640 dfh = fanotify_info_dir2_fh(info);
641 len = fanotify_encode_fh(fh: dfh, inode: dir2, fh_len: dir2_fh_len, hash, gfp: 0);
642 fanotify_info_set_dir2_fh(info, totlen: len);
643 }
644 if (child_fh_len) {
645 ffh = fanotify_info_file_fh(info);
646 len = fanotify_encode_fh(fh: ffh, inode: child, fh_len: child_fh_len, hash, gfp: 0);
647 fanotify_info_set_file_fh(info, totlen: len);
648 }
649 if (name_len) {
650 fanotify_info_copy_name(info, name);
651 *hash ^= full_name_hash(salt: (void *)name_len, name->name, name_len);
652 }
653 if (name2_len) {
654 fanotify_info_copy_name2(info, name: name2);
655 *hash ^= full_name_hash(salt: (void *)name2_len, name2->name,
656 name2_len);
657 }
658
659 pr_debug("%s: size=%u dir_fh_len=%u child_fh_len=%u name_len=%u name='%.*s'\n",
660 __func__, size, dir_fh_len, child_fh_len,
661 info->name_len, info->name_len, fanotify_info_name(info));
662
663 if (dir2_fh_len) {
664 pr_debug("%s: dir2_fh_len=%u name2_len=%u name2='%.*s'\n",
665 __func__, dir2_fh_len, info->name2_len,
666 info->name2_len, fanotify_info_name2(info));
667 }
668
669 return &fne->fae;
670}
671
672static struct fanotify_event *fanotify_alloc_error_event(
673 struct fsnotify_group *group,
674 __kernel_fsid_t *fsid,
675 const void *data, int data_type,
676 unsigned int *hash)
677{
678 struct fs_error_report *report =
679 fsnotify_data_error_report(data, data_type);
680 struct inode *inode;
681 struct fanotify_error_event *fee;
682 int fh_len;
683
684 if (WARN_ON_ONCE(!report))
685 return NULL;
686
687 fee = mempool_alloc(pool: &group->fanotify_data.error_events_pool, GFP_NOFS);
688 if (!fee)
689 return NULL;
690
691 fee->fae.type = FANOTIFY_EVENT_TYPE_FS_ERROR;
692 fee->error = report->error;
693 fee->err_count = 1;
694 fee->fsid = *fsid;
695
696 inode = report->inode;
697 fh_len = fanotify_encode_fh_len(inode);
698
699 /* Bad fh_len. Fallback to using an invalid fh. Should never happen. */
700 if (!fh_len && inode)
701 inode = NULL;
702
703 fanotify_encode_fh(fh: &fee->object_fh, inode, fh_len, NULL, gfp: 0);
704
705 *hash ^= fanotify_hash_fsid(fsid);
706
707 return &fee->fae;
708}
709
710static struct fanotify_event *fanotify_alloc_event(
711 struct fsnotify_group *group,
712 u32 mask, const void *data, int data_type,
713 struct inode *dir, const struct qstr *file_name,
714 __kernel_fsid_t *fsid, u32 match_mask)
715{
716 struct fanotify_event *event = NULL;
717 gfp_t gfp = GFP_KERNEL_ACCOUNT;
718 unsigned int fid_mode = FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS);
719 struct inode *id = fanotify_fid_inode(event_mask: mask, data, data_type, dir,
720 fid_mode);
721 struct inode *dirid = fanotify_dfid_inode(event_mask: mask, data, data_type, dir);
722 const struct path *path = fsnotify_data_path(data, data_type);
723 struct mem_cgroup *old_memcg;
724 struct dentry *moved = NULL;
725 struct inode *child = NULL;
726 bool name_event = false;
727 unsigned int hash = 0;
728 bool ondir = mask & FAN_ONDIR;
729 struct pid *pid;
730
731 if ((fid_mode & FAN_REPORT_DIR_FID) && dirid) {
732 /*
733 * For certain events and group flags, report the child fid
734 * in addition to reporting the parent fid and maybe child name.
735 */
736 if (fanotify_report_child_fid(fid_mode, mask) && id != dirid)
737 child = id;
738
739 id = dirid;
740
741 /*
742 * We record file name only in a group with FAN_REPORT_NAME
743 * and when we have a directory inode to report.
744 *
745 * For directory entry modification event, we record the fid of
746 * the directory and the name of the modified entry.
747 *
748 * For event on non-directory that is reported to parent, we
749 * record the fid of the parent and the name of the child.
750 *
751 * Even if not reporting name, we need a variable length
752 * fanotify_name_event if reporting both parent and child fids.
753 */
754 if (!(fid_mode & FAN_REPORT_NAME)) {
755 name_event = !!child;
756 file_name = NULL;
757 } else if ((mask & ALL_FSNOTIFY_DIRENT_EVENTS) || !ondir) {
758 name_event = true;
759 }
760
761 /*
762 * In the special case of FAN_RENAME event, use the match_mask
763 * to determine if we need to report only the old parent+name,
764 * only the new parent+name or both.
765 * 'dirid' and 'file_name' are the old parent+name and
766 * 'moved' has the new parent+name.
767 */
768 if (mask & FAN_RENAME) {
769 bool report_old, report_new;
770
771 if (WARN_ON_ONCE(!match_mask))
772 return NULL;
773
774 /* Report both old and new parent+name if sb watching */
775 report_old = report_new =
776 match_mask & (1U << FSNOTIFY_ITER_TYPE_SB);
777 report_old |=
778 match_mask & (1U << FSNOTIFY_ITER_TYPE_INODE);
779 report_new |=
780 match_mask & (1U << FSNOTIFY_ITER_TYPE_INODE2);
781
782 if (!report_old) {
783 /* Do not report old parent+name */
784 dirid = NULL;
785 file_name = NULL;
786 }
787 if (report_new) {
788 /* Report new parent+name */
789 moved = fsnotify_data_dentry(data, data_type);
790 }
791 }
792 }
793
794 /*
795 * For queues with unlimited length lost events are not expected and
796 * can possibly have security implications. Avoid losing events when
797 * memory is short. For the limited size queues, avoid OOM killer in the
798 * target monitoring memcg as it may have security repercussion.
799 */
800 if (group->max_events == UINT_MAX)
801 gfp |= __GFP_NOFAIL;
802 else
803 gfp |= __GFP_RETRY_MAYFAIL;
804
805 /* Whoever is interested in the event, pays for the allocation. */
806 old_memcg = set_active_memcg(group->memcg);
807
808 if (fanotify_is_perm_event(mask)) {
809 event = fanotify_alloc_perm_event(path, gfp);
810 } else if (fanotify_is_error_event(mask)) {
811 event = fanotify_alloc_error_event(group, fsid, data,
812 data_type, hash: &hash);
813 } else if (name_event && (file_name || moved || child)) {
814 event = fanotify_alloc_name_event(dir: dirid, fsid, name: file_name, child,
815 moved, hash: &hash, gfp);
816 } else if (fid_mode) {
817 event = fanotify_alloc_fid_event(id, fsid, hash: &hash, gfp);
818 } else {
819 event = fanotify_alloc_path_event(path, hash: &hash, gfp);
820 }
821
822 if (!event)
823 goto out;
824
825 if (FAN_GROUP_FLAG(group, FAN_REPORT_TID))
826 pid = get_pid(pid: task_pid(current));
827 else
828 pid = get_pid(pid: task_tgid(current));
829
830 /* Mix event info, FAN_ONDIR flag and pid into event merge key */
831 hash ^= hash_long((unsigned long)pid | ondir, FANOTIFY_EVENT_HASH_BITS);
832 fanotify_init_event(event, hash, mask);
833 event->pid = pid;
834
835out:
836 set_active_memcg(old_memcg);
837 return event;
838}
839
840/*
841 * Get cached fsid of the filesystem containing the object from any connector.
842 * All connectors are supposed to have the same fsid, but we do not verify that
843 * here.
844 */
845static __kernel_fsid_t fanotify_get_fsid(struct fsnotify_iter_info *iter_info)
846{
847 struct fsnotify_mark *mark;
848 int type;
849 __kernel_fsid_t fsid = {};
850
851 fsnotify_foreach_iter_mark_type(iter_info, mark, type) {
852 struct fsnotify_mark_connector *conn;
853
854 conn = READ_ONCE(mark->connector);
855 /* Mark is just getting destroyed or created? */
856 if (!conn)
857 continue;
858 if (!(conn->flags & FSNOTIFY_CONN_FLAG_HAS_FSID))
859 continue;
860 /* Pairs with smp_wmb() in fsnotify_add_mark_list() */
861 smp_rmb();
862 fsid = conn->fsid;
863 if (WARN_ON_ONCE(!fsid.val[0] && !fsid.val[1]))
864 continue;
865 return fsid;
866 }
867
868 return fsid;
869}
870
871/*
872 * Add an event to hash table for faster merge.
873 */
874static void fanotify_insert_event(struct fsnotify_group *group,
875 struct fsnotify_event *fsn_event)
876{
877 struct fanotify_event *event = FANOTIFY_E(fse: fsn_event);
878 unsigned int bucket = fanotify_event_hash_bucket(group, event);
879 struct hlist_head *hlist = &group->fanotify_data.merge_hash[bucket];
880
881 assert_spin_locked(&group->notification_lock);
882
883 if (!fanotify_is_hashed_event(mask: event->mask))
884 return;
885
886 pr_debug("%s: group=%p event=%p bucket=%u\n", __func__,
887 group, event, bucket);
888
889 hlist_add_head(n: &event->merge_list, h: hlist);
890}
891
892static int fanotify_handle_event(struct fsnotify_group *group, u32 mask,
893 const void *data, int data_type,
894 struct inode *dir,
895 const struct qstr *file_name, u32 cookie,
896 struct fsnotify_iter_info *iter_info)
897{
898 int ret = 0;
899 struct fanotify_event *event;
900 struct fsnotify_event *fsn_event;
901 __kernel_fsid_t fsid = {};
902 u32 match_mask = 0;
903
904 BUILD_BUG_ON(FAN_ACCESS != FS_ACCESS);
905 BUILD_BUG_ON(FAN_MODIFY != FS_MODIFY);
906 BUILD_BUG_ON(FAN_ATTRIB != FS_ATTRIB);
907 BUILD_BUG_ON(FAN_CLOSE_NOWRITE != FS_CLOSE_NOWRITE);
908 BUILD_BUG_ON(FAN_CLOSE_WRITE != FS_CLOSE_WRITE);
909 BUILD_BUG_ON(FAN_OPEN != FS_OPEN);
910 BUILD_BUG_ON(FAN_MOVED_TO != FS_MOVED_TO);
911 BUILD_BUG_ON(FAN_MOVED_FROM != FS_MOVED_FROM);
912 BUILD_BUG_ON(FAN_CREATE != FS_CREATE);
913 BUILD_BUG_ON(FAN_DELETE != FS_DELETE);
914 BUILD_BUG_ON(FAN_DELETE_SELF != FS_DELETE_SELF);
915 BUILD_BUG_ON(FAN_MOVE_SELF != FS_MOVE_SELF);
916 BUILD_BUG_ON(FAN_EVENT_ON_CHILD != FS_EVENT_ON_CHILD);
917 BUILD_BUG_ON(FAN_Q_OVERFLOW != FS_Q_OVERFLOW);
918 BUILD_BUG_ON(FAN_OPEN_PERM != FS_OPEN_PERM);
919 BUILD_BUG_ON(FAN_ACCESS_PERM != FS_ACCESS_PERM);
920 BUILD_BUG_ON(FAN_ONDIR != FS_ISDIR);
921 BUILD_BUG_ON(FAN_OPEN_EXEC != FS_OPEN_EXEC);
922 BUILD_BUG_ON(FAN_OPEN_EXEC_PERM != FS_OPEN_EXEC_PERM);
923 BUILD_BUG_ON(FAN_FS_ERROR != FS_ERROR);
924 BUILD_BUG_ON(FAN_RENAME != FS_RENAME);
925
926 BUILD_BUG_ON(HWEIGHT32(ALL_FANOTIFY_EVENT_BITS) != 21);
927
928 mask = fanotify_group_event_mask(group, iter_info, match_mask: &match_mask,
929 event_mask: mask, data, data_type, dir);
930 if (!mask)
931 return 0;
932
933 pr_debug("%s: group=%p mask=%x report_mask=%x\n", __func__,
934 group, mask, match_mask);
935
936 if (fanotify_is_perm_event(mask)) {
937 /*
938 * fsnotify_prepare_user_wait() fails if we race with mark
939 * deletion. Just let the operation pass in that case.
940 */
941 if (!fsnotify_prepare_user_wait(iter_info))
942 return 0;
943 }
944
945 if (FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS)) {
946 fsid = fanotify_get_fsid(iter_info);
947 /* Racing with mark destruction or creation? */
948 if (!fsid.val[0] && !fsid.val[1])
949 return 0;
950 }
951
952 event = fanotify_alloc_event(group, mask, data, data_type, dir,
953 file_name, fsid: &fsid, match_mask);
954 ret = -ENOMEM;
955 if (unlikely(!event)) {
956 /*
957 * We don't queue overflow events for permission events as
958 * there the access is denied and so no event is in fact lost.
959 */
960 if (!fanotify_is_perm_event(mask))
961 fsnotify_queue_overflow(group);
962 goto finish;
963 }
964
965 fsn_event = &event->fse;
966 ret = fsnotify_insert_event(group, event: fsn_event, merge: fanotify_merge,
967 insert: fanotify_insert_event);
968 if (ret) {
969 /* Permission events shouldn't be merged */
970 BUG_ON(ret == 1 && mask & FANOTIFY_PERM_EVENTS);
971 /* Our event wasn't used in the end. Free it. */
972 fsnotify_destroy_event(group, event: fsn_event);
973
974 ret = 0;
975 } else if (fanotify_is_perm_event(mask)) {
976 ret = fanotify_get_response(group, event: FANOTIFY_PERM(event),
977 iter_info);
978 }
979finish:
980 if (fanotify_is_perm_event(mask))
981 fsnotify_finish_user_wait(iter_info);
982
983 return ret;
984}
985
986static void fanotify_free_group_priv(struct fsnotify_group *group)
987{
988 kfree(objp: group->fanotify_data.merge_hash);
989 if (group->fanotify_data.ucounts)
990 dec_ucount(ucounts: group->fanotify_data.ucounts,
991 type: UCOUNT_FANOTIFY_GROUPS);
992
993 if (mempool_initialized(pool: &group->fanotify_data.error_events_pool))
994 mempool_exit(pool: &group->fanotify_data.error_events_pool);
995}
996
997static void fanotify_free_path_event(struct fanotify_event *event)
998{
999 path_put(fanotify_event_path(event));
1000 kmem_cache_free(s: fanotify_path_event_cachep, objp: FANOTIFY_PE(event));
1001}
1002
1003static void fanotify_free_perm_event(struct fanotify_event *event)
1004{
1005 path_put(fanotify_event_path(event));
1006 kmem_cache_free(s: fanotify_perm_event_cachep, objp: FANOTIFY_PERM(event));
1007}
1008
1009static void fanotify_free_fid_event(struct fanotify_event *event)
1010{
1011 struct fanotify_fid_event *ffe = FANOTIFY_FE(event);
1012
1013 if (fanotify_fh_has_ext_buf(fh: &ffe->object_fh))
1014 kfree(objp: fanotify_fh_ext_buf(fh: &ffe->object_fh));
1015 kmem_cache_free(s: fanotify_fid_event_cachep, objp: ffe);
1016}
1017
1018static void fanotify_free_name_event(struct fanotify_event *event)
1019{
1020 kfree(objp: FANOTIFY_NE(event));
1021}
1022
1023static void fanotify_free_error_event(struct fsnotify_group *group,
1024 struct fanotify_event *event)
1025{
1026 struct fanotify_error_event *fee = FANOTIFY_EE(event);
1027
1028 mempool_free(element: fee, pool: &group->fanotify_data.error_events_pool);
1029}
1030
1031static void fanotify_free_event(struct fsnotify_group *group,
1032 struct fsnotify_event *fsn_event)
1033{
1034 struct fanotify_event *event;
1035
1036 event = FANOTIFY_E(fse: fsn_event);
1037 put_pid(pid: event->pid);
1038 switch (event->type) {
1039 case FANOTIFY_EVENT_TYPE_PATH:
1040 fanotify_free_path_event(event);
1041 break;
1042 case FANOTIFY_EVENT_TYPE_PATH_PERM:
1043 fanotify_free_perm_event(event);
1044 break;
1045 case FANOTIFY_EVENT_TYPE_FID:
1046 fanotify_free_fid_event(event);
1047 break;
1048 case FANOTIFY_EVENT_TYPE_FID_NAME:
1049 fanotify_free_name_event(event);
1050 break;
1051 case FANOTIFY_EVENT_TYPE_OVERFLOW:
1052 kfree(objp: event);
1053 break;
1054 case FANOTIFY_EVENT_TYPE_FS_ERROR:
1055 fanotify_free_error_event(group, event);
1056 break;
1057 default:
1058 WARN_ON_ONCE(1);
1059 }
1060}
1061
1062static void fanotify_freeing_mark(struct fsnotify_mark *mark,
1063 struct fsnotify_group *group)
1064{
1065 if (!FAN_GROUP_FLAG(group, FAN_UNLIMITED_MARKS))
1066 dec_ucount(ucounts: group->fanotify_data.ucounts, type: UCOUNT_FANOTIFY_MARKS);
1067}
1068
1069static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
1070{
1071 kmem_cache_free(s: fanotify_mark_cache, objp: fsn_mark);
1072}
1073
1074const struct fsnotify_ops fanotify_fsnotify_ops = {
1075 .handle_event = fanotify_handle_event,
1076 .free_group_priv = fanotify_free_group_priv,
1077 .free_event = fanotify_free_event,
1078 .freeing_mark = fanotify_freeing_mark,
1079 .free_mark = fanotify_free_mark,
1080};
1081

source code of linux/fs/notify/fanotify/fanotify.c