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>
21 static bool fanotify_path_equal(const struct path *p1, const struct path *p2)
23 return p1->mnt == p2->mnt && p1->dentry == p2->dentry;
26 static unsigned int fanotify_hash_path(const struct path *path)
28 return hash_ptr(path->dentry, FANOTIFY_EVENT_HASH_BITS) ^
29 hash_ptr(path->mnt, FANOTIFY_EVENT_HASH_BITS);
32 static inline bool fanotify_fsid_equal(__kernel_fsid_t *fsid1,
33 __kernel_fsid_t *fsid2)
35 return fsid1->val[0] == fsid2->val[0] && fsid1->val[1] == fsid2->val[1];
38 static unsigned int fanotify_hash_fsid(__kernel_fsid_t *fsid)
40 return hash_32(fsid->val[0], FANOTIFY_EVENT_HASH_BITS) ^
41 hash_32(fsid->val[1], FANOTIFY_EVENT_HASH_BITS);
44 static bool fanotify_fh_equal(struct fanotify_fh *fh1,
45 struct fanotify_fh *fh2)
47 if (fh1->type != fh2->type || fh1->len != fh2->len)
51 !memcmp(fanotify_fh_buf(fh1), fanotify_fh_buf(fh2), fh1->len);
54 static unsigned int fanotify_hash_fh(struct fanotify_fh *fh)
56 long salt = (long)fh->type | (long)fh->len << 8;
59 * full_name_hash() works long by long, so it handles fh buf optimally.
61 return full_name_hash((void *)salt, fanotify_fh_buf(fh), fh->len);
64 static bool fanotify_fid_event_equal(struct fanotify_fid_event *ffe1,
65 struct fanotify_fid_event *ffe2)
67 /* Do not merge fid events without object fh */
68 if (!ffe1->object_fh.len)
71 return fanotify_fsid_equal(&ffe1->fsid, &ffe2->fsid) &&
72 fanotify_fh_equal(&ffe1->object_fh, &ffe2->object_fh);
75 static bool fanotify_info_equal(struct fanotify_info *info1,
76 struct fanotify_info *info2)
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)
85 if (info1->dir_fh_totlen &&
86 !fanotify_fh_equal(fanotify_info_dir_fh(info1),
87 fanotify_info_dir_fh(info2)))
90 if (info1->dir2_fh_totlen &&
91 !fanotify_fh_equal(fanotify_info_dir2_fh(info1),
92 fanotify_info_dir2_fh(info2)))
95 if (info1->file_fh_totlen &&
96 !fanotify_fh_equal(fanotify_info_file_fh(info1),
97 fanotify_info_file_fh(info2)))
100 if (info1->name_len &&
101 memcmp(fanotify_info_name(info1), fanotify_info_name(info2),
105 return !info1->name2_len ||
106 !memcmp(fanotify_info_name2(info1), fanotify_info_name2(info2),
110 static bool fanotify_name_event_equal(struct fanotify_name_event *fne1,
111 struct fanotify_name_event *fne2)
113 struct fanotify_info *info1 = &fne1->info;
114 struct fanotify_info *info2 = &fne2->info;
116 /* Do not merge name events without dir fh */
117 if (!info1->dir_fh_totlen)
120 if (!fanotify_fsid_equal(&fne1->fsid, &fne2->fsid))
123 return fanotify_info_equal(info1, info2);
126 static bool fanotify_error_event_equal(struct fanotify_error_event *fee1,
127 struct fanotify_error_event *fee2)
129 /* Error events against the same file system are always merged. */
130 if (!fanotify_fsid_equal(&fee1->fsid, &fee2->fsid))
136 static bool fanotify_should_merge(struct fanotify_event *old,
137 struct fanotify_event *new)
139 pr_debug("%s: old=%p new=%p\n", __func__, old, new);
141 if (old->hash != new->hash ||
142 old->type != new->type || old->pid != new->pid)
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.
153 if ((old->mask & FS_ISDIR) != (new->mask & FS_ISDIR))
157 * FAN_RENAME event is reported with special info record types,
158 * so we cannot merge it with other events.
160 if ((old->mask & FAN_RENAME) != (new->mask & FAN_RENAME))
164 case FANOTIFY_EVENT_TYPE_PATH:
165 return fanotify_path_equal(fanotify_event_path(old),
166 fanotify_event_path(new));
167 case FANOTIFY_EVENT_TYPE_FID:
168 return fanotify_fid_event_equal(FANOTIFY_FE(old),
170 case FANOTIFY_EVENT_TYPE_FID_NAME:
171 return fanotify_name_event_equal(FANOTIFY_NE(old),
173 case FANOTIFY_EVENT_TYPE_FS_ERROR:
174 return fanotify_error_event_equal(FANOTIFY_EE(old),
183 /* Limit event merges to limit CPU overhead per event */
184 #define FANOTIFY_MAX_MERGE_EVENTS 128
186 /* and the list better be locked by something too! */
187 static int fanotify_merge(struct fsnotify_group *group,
188 struct fsnotify_event *event)
190 struct fanotify_event *old, *new = FANOTIFY_E(event);
191 unsigned int bucket = fanotify_event_hash_bucket(group, new);
192 struct hlist_head *hlist = &group->fanotify_data.merge_hash[bucket];
195 pr_debug("%s: group=%p event=%p bucket=%u\n", __func__,
196 group, event, bucket);
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.
203 if (fanotify_is_perm_event(new->mask))
206 hlist_for_each_entry(old, hlist, merge_list) {
207 if (++i > FANOTIFY_MAX_MERGE_EVENTS)
209 if (fanotify_should_merge(old, new)) {
210 old->mask |= new->mask;
212 if (fanotify_is_error_event(old->mask))
213 FANOTIFY_EE(old)->err_count++;
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.
229 static int fanotify_get_response(struct fsnotify_group *group,
230 struct fanotify_perm_event *event,
231 struct fsnotify_iter_info *iter_info)
235 pr_debug("%s: group=%p event=%p\n", __func__, group, event);
237 ret = wait_event_killable(group->fanotify_data.access_waitq,
238 event->state == FAN_EVENT_ANSWERED);
239 /* Signal pending? */
241 spin_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(&group->notification_lock);
249 /* Event not yet reported? Just remove it. */
250 if (event->state == FAN_EVENT_INIT) {
251 fsnotify_remove_queued_event(group, &event->fae.fse);
252 /* Permission events are not supposed to be hashed */
253 WARN_ON_ONCE(!hlist_unhashed(&event->fae.merge_list));
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.
260 spin_unlock(&group->notification_lock);
264 /* userspace responded, convert to something usable */
265 switch (event->response & FANOTIFY_RESPONSE_ACCESS) {
274 /* Check if the response should be audited */
275 if (event->response & FAN_AUDIT)
276 audit_fanotify(event->response & ~FAN_AUDIT,
279 pr_debug("%s: group=%p event=%p about to return ret=%d\n", __func__,
282 fsnotify_destroy_event(group, &event->fae.fse);
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.
293 static 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,
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;
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);
312 /* Do we have path to open a file descriptor? */
315 /* Path type events are only relevant for files and dirs */
316 if (!d_is_reg(path->dentry) && !d_can_lookup(path->dentry))
318 } else if (!(fid_mode & FAN_REPORT_FID)) {
319 /* Do we have a directory inode to report? */
324 fsnotify_foreach_iter_mark_type(iter_info, mark, type) {
326 * Apply ignore mask depending on event flags in ignore mask.
329 fsnotify_effective_ignore_mask(mark, ondir, type);
332 * Send the event depending on event flags in mark mask.
334 if (!fsnotify_mask_applicable(mark->mask, ondir, type))
337 marks_mask |= mark->mask;
339 /* Record the mark types of this group that matched the event */
340 *match_mask |= 1U << type;
343 test_mask = event_mask & marks_mask & ~marks_ignore_mask;
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.
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.
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.
359 /* Do not report event flags without any event */
360 if (!(test_mask & ~FANOTIFY_EVENT_FLAGS))
363 user_mask &= ~FANOTIFY_EVENT_FLAGS;
366 return test_mask & user_mask;
370 * Check size needed to encode fanotify_fh.
372 * Return size of encoded fh without fanotify_fh header.
373 * Return 0 on failure to encode.
375 static int fanotify_encode_fh_len(struct inode *inode)
383 exportfs_encode_fid(inode, NULL, &dwords);
384 fh_len = dwords << 2;
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.
391 if (WARN_ON_ONCE(fh_len > MAX_HANDLE_SZ))
398 * Encode fanotify_fh.
400 * Return total size of encoded fh including fanotify_fh header.
401 * Return 0 on failure to encode.
403 static int fanotify_encode_fh(struct fanotify_fh *fh, struct inode *inode,
404 unsigned int fh_len, unsigned int *hash,
407 int dwords, type = 0;
408 char *ext_buf = NULL;
412 fh->type = FILEID_ROOT;
417 * Invalid FHs are used by FAN_FS_ERROR for errors not
418 * linked to any inode. The f_handle won't be reported
425 * !gpf means preallocated variable size fh, but fh_len could
426 * be zero in that case if encoding fh len failed.
429 if (fh_len < 4 || WARN_ON_ONCE(fh_len % 4) || fh_len > MAX_HANDLE_SZ)
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 */
436 ext_buf = kmalloc(fh_len, gfp);
440 *fanotify_fh_ext_buf_ptr(fh) = ext_buf;
442 fh->flags |= FANOTIFY_FH_FLAG_EXT_BUF;
445 dwords = fh_len >> 2;
446 type = exportfs_encode_fid(inode, buf, &dwords);
448 if (type <= 0 || type == FILEID_INVALID || fh_len != dwords << 2)
456 * Mix fh into event merge key. Hash might be NULL in case of
457 * unhashed FID events (i.e. FAN_FS_ERROR).
460 *hash ^= fanotify_hash_fh(fh);
462 return FANOTIFY_FH_HDR_LEN + fh_len;
465 pr_warn_ratelimited("fanotify: failed to encode fid (type=%d, len=%d, err=%i)\n",
468 *fanotify_fh_ext_buf_ptr(fh) = NULL;
469 /* Report the event without a file identifier on encode error */
470 fh->type = FILEID_INVALID;
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.
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.
483 static inline bool fanotify_report_child_fid(unsigned int fid_mode, u32 mask)
485 if (mask & ALL_FSNOTIFY_DIRENT_EVENTS)
486 return (fid_mode & FAN_REPORT_TARGET_FID);
488 return (fid_mode & FAN_REPORT_FID) && !(mask & FAN_ONDIR);
492 * The inode to use as identifier when reporting fid depends on the event
493 * and the group flags.
495 * With the group flag FAN_REPORT_TARGET_FID, always report the child fid.
497 * Without the group flag FAN_REPORT_TARGET_FID, report the modified directory
498 * fid on dirent events and the child fid otherwise.
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.
505 static struct inode *fanotify_fid_inode(u32 event_mask, const void *data,
506 int data_type, struct inode *dir,
507 unsigned int fid_mode)
509 if ((event_mask & ALL_FSNOTIFY_DIRENT_EVENTS) &&
510 !(fid_mode & FAN_REPORT_TARGET_FID))
513 return fsnotify_data_inode(data, data_type);
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.
524 static struct inode *fanotify_dfid_inode(u32 event_mask, const void *data,
525 int data_type, struct inode *dir)
527 struct inode *inode = fsnotify_data_inode(data, data_type);
529 if (event_mask & ALL_FSNOTIFY_DIRENT_EVENTS)
532 if (inode && S_ISDIR(inode->i_mode))
538 static struct fanotify_event *fanotify_alloc_path_event(const struct path *path,
542 struct fanotify_path_event *pevent;
544 pevent = kmem_cache_alloc(fanotify_path_event_cachep, gfp);
548 pevent->fae.type = FANOTIFY_EVENT_TYPE_PATH;
549 pevent->path = *path;
550 *hash ^= fanotify_hash_path(path);
556 static struct fanotify_event *fanotify_alloc_perm_event(const struct path *path,
559 struct fanotify_perm_event *pevent;
561 pevent = kmem_cache_alloc(fanotify_perm_event_cachep, gfp);
565 pevent->fae.type = FANOTIFY_EVENT_TYPE_PATH_PERM;
566 pevent->response = 0;
567 pevent->hdr.type = FAN_RESPONSE_INFO_NONE;
570 pevent->state = FAN_EVENT_INIT;
571 pevent->path = *path;
577 static struct fanotify_event *fanotify_alloc_fid_event(struct inode *id,
578 __kernel_fsid_t *fsid,
582 struct fanotify_fid_event *ffe;
584 ffe = kmem_cache_alloc(fanotify_fid_event_cachep, gfp);
588 ffe->fae.type = FANOTIFY_EVENT_TYPE_FID;
590 *hash ^= fanotify_hash_fsid(fsid);
591 fanotify_encode_fh(&ffe->object_fh, id, fanotify_encode_fh_len(id),
597 static struct fanotify_event *fanotify_alloc_name_event(struct inode *dir,
598 __kernel_fsid_t *fsid,
599 const struct qstr *name,
601 struct dentry *moved,
605 struct fanotify_name_event *fne;
606 struct fanotify_info *info;
607 struct fanotify_fh *dfh, *ffh;
608 struct inode *dir2 = moved ? d_inode(moved->d_parent) : NULL;
609 const struct qstr *name2 = moved ? &moved->d_name : NULL;
610 unsigned int dir_fh_len = fanotify_encode_fh_len(dir);
611 unsigned int dir2_fh_len = fanotify_encode_fh_len(dir2);
612 unsigned int child_fh_len = fanotify_encode_fh_len(child);
613 unsigned long name_len = name ? name->len : 0;
614 unsigned long name2_len = name2 ? name2->len : 0;
615 unsigned int len, size;
617 /* Reserve terminating null byte even for empty name */
618 size = sizeof(*fne) + name_len + name2_len + 2;
620 size += FANOTIFY_FH_HDR_LEN + dir_fh_len;
622 size += FANOTIFY_FH_HDR_LEN + dir2_fh_len;
624 size += FANOTIFY_FH_HDR_LEN + child_fh_len;
625 fne = kmalloc(size, gfp);
629 fne->fae.type = FANOTIFY_EVENT_TYPE_FID_NAME;
631 *hash ^= fanotify_hash_fsid(fsid);
633 fanotify_info_init(info);
635 dfh = fanotify_info_dir_fh(info);
636 len = fanotify_encode_fh(dfh, dir, dir_fh_len, hash, 0);
637 fanotify_info_set_dir_fh(info, len);
640 dfh = fanotify_info_dir2_fh(info);
641 len = fanotify_encode_fh(dfh, dir2, dir2_fh_len, hash, 0);
642 fanotify_info_set_dir2_fh(info, len);
645 ffh = fanotify_info_file_fh(info);
646 len = fanotify_encode_fh(ffh, child, child_fh_len, hash, 0);
647 fanotify_info_set_file_fh(info, len);
650 fanotify_info_copy_name(info, name);
651 *hash ^= full_name_hash((void *)name_len, name->name, name_len);
654 fanotify_info_copy_name2(info, name2);
655 *hash ^= full_name_hash((void *)name2_len, name2->name,
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));
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));
672 static struct fanotify_event *fanotify_alloc_error_event(
673 struct fsnotify_group *group,
674 __kernel_fsid_t *fsid,
675 const void *data, int data_type,
678 struct fs_error_report *report =
679 fsnotify_data_error_report(data, data_type);
681 struct fanotify_error_event *fee;
684 if (WARN_ON_ONCE(!report))
687 fee = mempool_alloc(&group->fanotify_data.error_events_pool, GFP_NOFS);
691 fee->fae.type = FANOTIFY_EVENT_TYPE_FS_ERROR;
692 fee->error = report->error;
696 inode = report->inode;
697 fh_len = fanotify_encode_fh_len(inode);
699 /* Bad fh_len. Fallback to using an invalid fh. Should never happen. */
700 if (!fh_len && inode)
703 fanotify_encode_fh(&fee->object_fh, inode, fh_len, NULL, 0);
705 *hash ^= fanotify_hash_fsid(fsid);
710 static 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)
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(mask, data, data_type, dir,
721 struct inode *dirid = fanotify_dfid_inode(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;
731 if ((fid_mode & FAN_REPORT_DIR_FID) && dirid) {
733 * For certain events and group flags, report the child fid
734 * in addition to reporting the parent fid and maybe child name.
736 if (fanotify_report_child_fid(fid_mode, mask) && id != dirid)
742 * We record file name only in a group with FAN_REPORT_NAME
743 * and when we have a directory inode to report.
745 * For directory entry modification event, we record the fid of
746 * the directory and the name of the modified entry.
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.
751 * Even if not reporting name, we need a variable length
752 * fanotify_name_event if reporting both parent and child fids.
754 if (!(fid_mode & FAN_REPORT_NAME)) {
755 name_event = !!child;
757 } else if ((mask & ALL_FSNOTIFY_DIRENT_EVENTS) || !ondir) {
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.
768 if (mask & FAN_RENAME) {
769 bool report_old, report_new;
771 if (WARN_ON_ONCE(!match_mask))
774 /* Report both old and new parent+name if sb watching */
775 report_old = report_new =
776 match_mask & (1U << FSNOTIFY_ITER_TYPE_SB);
778 match_mask & (1U << FSNOTIFY_ITER_TYPE_INODE);
780 match_mask & (1U << FSNOTIFY_ITER_TYPE_INODE2);
783 /* Do not report old parent+name */
788 /* Report new parent+name */
789 moved = fsnotify_data_dentry(data, data_type);
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.
800 if (group->max_events == UINT_MAX)
803 gfp |= __GFP_RETRY_MAYFAIL;
805 /* Whoever is interested in the event, pays for the allocation. */
806 old_memcg = set_active_memcg(group->memcg);
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,
813 } else if (name_event && (file_name || moved || child)) {
814 event = fanotify_alloc_name_event(dirid, fsid, file_name, child,
816 } else if (fid_mode) {
817 event = fanotify_alloc_fid_event(id, fsid, &hash, gfp);
819 event = fanotify_alloc_path_event(path, &hash, gfp);
825 if (FAN_GROUP_FLAG(group, FAN_REPORT_TID))
826 pid = get_pid(task_pid(current));
828 pid = get_pid(task_tgid(current));
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);
836 set_active_memcg(old_memcg);
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
845 static __kernel_fsid_t fanotify_get_fsid(struct fsnotify_iter_info *iter_info)
847 struct fsnotify_mark *mark;
849 __kernel_fsid_t fsid = {};
851 fsnotify_foreach_iter_mark_type(iter_info, mark, type) {
852 struct fsnotify_mark_connector *conn;
854 conn = READ_ONCE(mark->connector);
855 /* Mark is just getting destroyed or created? */
858 if (!(conn->flags & FSNOTIFY_CONN_FLAG_HAS_FSID))
860 /* Pairs with smp_wmb() in fsnotify_add_mark_list() */
863 if (WARN_ON_ONCE(!fsid.val[0] && !fsid.val[1]))
872 * Add an event to hash table for faster merge.
874 static void fanotify_insert_event(struct fsnotify_group *group,
875 struct fsnotify_event *fsn_event)
877 struct fanotify_event *event = FANOTIFY_E(fsn_event);
878 unsigned int bucket = fanotify_event_hash_bucket(group, event);
879 struct hlist_head *hlist = &group->fanotify_data.merge_hash[bucket];
881 assert_spin_locked(&group->notification_lock);
883 if (!fanotify_is_hashed_event(event->mask))
886 pr_debug("%s: group=%p event=%p bucket=%u\n", __func__,
887 group, event, bucket);
889 hlist_add_head(&event->merge_list, hlist);
892 static int fanotify_handle_event(struct fsnotify_group *group, u32 mask,
893 const void *data, int data_type,
895 const struct qstr *file_name, u32 cookie,
896 struct fsnotify_iter_info *iter_info)
899 struct fanotify_event *event;
900 struct fsnotify_event *fsn_event;
901 __kernel_fsid_t fsid = {};
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);
926 BUILD_BUG_ON(HWEIGHT32(ALL_FANOTIFY_EVENT_BITS) != 21);
928 mask = fanotify_group_event_mask(group, iter_info, &match_mask,
929 mask, data, data_type, dir);
933 pr_debug("%s: group=%p mask=%x report_mask=%x\n", __func__,
934 group, mask, match_mask);
936 if (fanotify_is_perm_event(mask)) {
938 * fsnotify_prepare_user_wait() fails if we race with mark
939 * deletion. Just let the operation pass in that case.
941 if (!fsnotify_prepare_user_wait(iter_info))
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])
952 event = fanotify_alloc_event(group, mask, data, data_type, dir,
953 file_name, &fsid, match_mask);
955 if (unlikely(!event)) {
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.
960 if (!fanotify_is_perm_event(mask))
961 fsnotify_queue_overflow(group);
965 fsn_event = &event->fse;
966 ret = fsnotify_insert_event(group, fsn_event, fanotify_merge,
967 fanotify_insert_event);
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, fsn_event);
975 } else if (fanotify_is_perm_event(mask)) {
976 ret = fanotify_get_response(group, FANOTIFY_PERM(event),
980 if (fanotify_is_perm_event(mask))
981 fsnotify_finish_user_wait(iter_info);
986 static void fanotify_free_group_priv(struct fsnotify_group *group)
988 kfree(group->fanotify_data.merge_hash);
989 if (group->fanotify_data.ucounts)
990 dec_ucount(group->fanotify_data.ucounts,
991 UCOUNT_FANOTIFY_GROUPS);
993 if (mempool_initialized(&group->fanotify_data.error_events_pool))
994 mempool_exit(&group->fanotify_data.error_events_pool);
997 static void fanotify_free_path_event(struct fanotify_event *event)
999 path_put(fanotify_event_path(event));
1000 kmem_cache_free(fanotify_path_event_cachep, FANOTIFY_PE(event));
1003 static void fanotify_free_perm_event(struct fanotify_event *event)
1005 path_put(fanotify_event_path(event));
1006 kmem_cache_free(fanotify_perm_event_cachep, FANOTIFY_PERM(event));
1009 static void fanotify_free_fid_event(struct fanotify_event *event)
1011 struct fanotify_fid_event *ffe = FANOTIFY_FE(event);
1013 if (fanotify_fh_has_ext_buf(&ffe->object_fh))
1014 kfree(fanotify_fh_ext_buf(&ffe->object_fh));
1015 kmem_cache_free(fanotify_fid_event_cachep, ffe);
1018 static void fanotify_free_name_event(struct fanotify_event *event)
1020 kfree(FANOTIFY_NE(event));
1023 static void fanotify_free_error_event(struct fsnotify_group *group,
1024 struct fanotify_event *event)
1026 struct fanotify_error_event *fee = FANOTIFY_EE(event);
1028 mempool_free(fee, &group->fanotify_data.error_events_pool);
1031 static void fanotify_free_event(struct fsnotify_group *group,
1032 struct fsnotify_event *fsn_event)
1034 struct fanotify_event *event;
1036 event = FANOTIFY_E(fsn_event);
1037 put_pid(event->pid);
1038 switch (event->type) {
1039 case FANOTIFY_EVENT_TYPE_PATH:
1040 fanotify_free_path_event(event);
1042 case FANOTIFY_EVENT_TYPE_PATH_PERM:
1043 fanotify_free_perm_event(event);
1045 case FANOTIFY_EVENT_TYPE_FID:
1046 fanotify_free_fid_event(event);
1048 case FANOTIFY_EVENT_TYPE_FID_NAME:
1049 fanotify_free_name_event(event);
1051 case FANOTIFY_EVENT_TYPE_OVERFLOW:
1054 case FANOTIFY_EVENT_TYPE_FS_ERROR:
1055 fanotify_free_error_event(group, event);
1062 static void fanotify_freeing_mark(struct fsnotify_mark *mark,
1063 struct fsnotify_group *group)
1065 if (!FAN_GROUP_FLAG(group, FAN_UNLIMITED_MARKS))
1066 dec_ucount(group->fanotify_data.ucounts, UCOUNT_FANOTIFY_MARKS);
1069 static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
1071 kmem_cache_free(fanotify_mark_cache, fsn_mark);
1074 const 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,