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

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