1 #include <linux/fcntl.h>
2 #include <linux/file.h>
4 #include <linux/anon_inodes.h>
5 #include <linux/fsnotify_backend.h>
6 #include <linux/init.h>
7 #include <linux/mount.h>
8 #include <linux/namei.h>
9 #include <linux/poll.h>
10 #include <linux/security.h>
11 #include <linux/syscalls.h>
12 #include <linux/types.h>
13 #include <linux/uaccess.h>
15 #include <asm/ioctls.h>
19 static struct kmem_cache *fanotify_mark_cache __read_mostly;
22 * Get an fsnotify notification event if one exists and is small
23 * enough to fit in "count". Return an error pointer if the count
24 * is not large enough.
26 * Called with the group->notification_mutex held.
28 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
31 BUG_ON(!mutex_is_locked(&group->notification_mutex));
33 pr_debug("%s: group=%p count=%zd\n", __func__, group, count);
35 if (fsnotify_notify_queue_is_empty(group))
38 if (FAN_EVENT_METADATA_LEN > count)
39 return ERR_PTR(-EINVAL);
41 /* held the notification_mutex the whole time, so this is the
42 * same event we peeked above */
43 return fsnotify_remove_notify_event(group);
46 static int create_and_fill_fd(struct fsnotify_group *group,
47 struct fanotify_event_metadata *metadata,
48 struct fsnotify_event *event)
51 struct dentry *dentry;
53 struct file *new_file;
55 pr_debug("%s: group=%p metadata=%p event=%p\n", __func__, group,
58 client_fd = get_unused_fd();
62 if (event->data_type != FSNOTIFY_EVENT_PATH) {
64 put_unused_fd(client_fd);
69 * we need a new file handle for the userspace program so it can read even if it was
70 * originally opened O_WRONLY.
72 dentry = dget(event->path.dentry);
73 mnt = mntget(event->path.mnt);
74 /* it's possible this event was an overflow event. in that case dentry and mnt
75 * are NULL; That's fine, just don't call dentry open */
77 new_file = dentry_open(dentry, mnt,
78 O_RDONLY | O_LARGEFILE | FMODE_NONOTIFY,
81 new_file = ERR_PTR(-EOVERFLOW);
82 if (IS_ERR(new_file)) {
84 * we still send an event even if we can't open the file. this
85 * can happen when say tasks are gone and we try to open their
86 * /proc files or we try to open a WRONLY file like in sysfs
87 * we just send the errno to userspace since there isn't much
90 put_unused_fd(client_fd);
91 client_fd = PTR_ERR(new_file);
93 fd_install(client_fd, new_file);
96 metadata->fd = client_fd;
101 static ssize_t fill_event_metadata(struct fsnotify_group *group,
102 struct fanotify_event_metadata *metadata,
103 struct fsnotify_event *event)
105 pr_debug("%s: group=%p metadata=%p event=%p\n", __func__,
106 group, metadata, event);
108 metadata->event_len = FAN_EVENT_METADATA_LEN;
109 metadata->vers = FANOTIFY_METADATA_VERSION;
110 metadata->mask = fanotify_outgoing_mask(event->mask);
112 return create_and_fill_fd(group, metadata, event);
116 static ssize_t copy_event_to_user(struct fsnotify_group *group,
117 struct fsnotify_event *event,
120 struct fanotify_event_metadata fanotify_event_metadata;
123 pr_debug("%s: group=%p event=%p\n", __func__, group, event);
125 ret = fill_event_metadata(group, &fanotify_event_metadata, event);
129 if (copy_to_user(buf, &fanotify_event_metadata, FAN_EVENT_METADATA_LEN))
132 return FAN_EVENT_METADATA_LEN;
135 /* intofiy userspace file descriptor functions */
136 static unsigned int fanotify_poll(struct file *file, poll_table *wait)
138 struct fsnotify_group *group = file->private_data;
141 poll_wait(file, &group->notification_waitq, wait);
142 mutex_lock(&group->notification_mutex);
143 if (!fsnotify_notify_queue_is_empty(group))
144 ret = POLLIN | POLLRDNORM;
145 mutex_unlock(&group->notification_mutex);
150 static ssize_t fanotify_read(struct file *file, char __user *buf,
151 size_t count, loff_t *pos)
153 struct fsnotify_group *group;
154 struct fsnotify_event *kevent;
160 group = file->private_data;
162 pr_debug("%s: group=%p\n", __func__, group);
165 prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
167 mutex_lock(&group->notification_mutex);
168 kevent = get_one_event(group, count);
169 mutex_unlock(&group->notification_mutex);
172 ret = PTR_ERR(kevent);
175 ret = copy_event_to_user(group, kevent, buf);
176 fsnotify_put_event(kevent);
185 if (file->f_flags & O_NONBLOCK)
188 if (signal_pending(current))
197 finish_wait(&group->notification_waitq, &wait);
198 if (start != buf && ret != -EFAULT)
203 static int fanotify_release(struct inode *ignored, struct file *file)
205 struct fsnotify_group *group = file->private_data;
207 pr_debug("%s: file=%p group=%p\n", __func__, file, group);
209 /* matches the fanotify_init->fsnotify_alloc_group */
210 fsnotify_put_group(group);
215 static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
217 struct fsnotify_group *group;
218 struct fsnotify_event_holder *holder;
223 group = file->private_data;
225 p = (void __user *) arg;
229 mutex_lock(&group->notification_mutex);
230 list_for_each_entry(holder, &group->notification_list, event_list)
231 send_len += FAN_EVENT_METADATA_LEN;
232 mutex_unlock(&group->notification_mutex);
233 ret = put_user(send_len, (int __user *) p);
240 static const struct file_operations fanotify_fops = {
241 .poll = fanotify_poll,
242 .read = fanotify_read,
244 .release = fanotify_release,
245 .unlocked_ioctl = fanotify_ioctl,
246 .compat_ioctl = fanotify_ioctl,
249 static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
251 kmem_cache_free(fanotify_mark_cache, fsn_mark);
254 static int fanotify_find_path(int dfd, const char __user *filename,
255 struct path *path, unsigned int flags)
259 pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
260 dfd, filename, flags);
262 if (filename == NULL) {
267 file = fget_light(dfd, &fput_needed);
272 if ((flags & FAN_MARK_ONLYDIR) &&
273 !(S_ISDIR(file->f_path.dentry->d_inode->i_mode))) {
274 fput_light(file, fput_needed);
278 *path = file->f_path;
280 fput_light(file, fput_needed);
282 unsigned int lookup_flags = 0;
284 if (!(flags & FAN_MARK_DONT_FOLLOW))
285 lookup_flags |= LOOKUP_FOLLOW;
286 if (flags & FAN_MARK_ONLYDIR)
287 lookup_flags |= LOOKUP_DIRECTORY;
289 ret = user_path_at(dfd, filename, lookup_flags, path);
294 /* you can only watch an inode if you have read permissions on it */
295 ret = inode_permission(path->dentry->d_inode, MAY_READ);
302 static int fanotify_remove_mark(struct fsnotify_group *group,
306 struct fsnotify_mark *fsn_mark;
309 pr_debug("%s: group=%p inode=%p mask=%x\n", __func__,
312 fsn_mark = fsnotify_find_mark(group, inode);
316 spin_lock(&fsn_mark->lock);
317 fsn_mark->mask &= ~mask;
318 new_mask = fsn_mark->mask;
319 spin_unlock(&fsn_mark->lock);
322 fsnotify_destroy_mark(fsn_mark);
324 fsnotify_recalc_inode_mask(inode);
326 fsnotify_recalc_group_mask(group);
328 /* matches the fsnotify_find_mark() */
329 fsnotify_put_mark(fsn_mark);
334 static int fanotify_add_mark(struct fsnotify_group *group,
338 struct fsnotify_mark *fsn_mark;
339 __u32 old_mask, new_mask;
342 pr_debug("%s: group=%p inode=%p mask=%x\n", __func__,
345 fsn_mark = fsnotify_find_mark(group, inode);
347 struct fsnotify_mark *new_fsn_mark;
350 new_fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
354 fsnotify_init_mark(new_fsn_mark, fanotify_free_mark);
355 ret = fsnotify_add_mark(new_fsn_mark, group, inode, 0);
357 fanotify_free_mark(new_fsn_mark);
361 fsn_mark = new_fsn_mark;
366 spin_lock(&fsn_mark->lock);
367 old_mask = fsn_mark->mask;
368 fsn_mark->mask |= mask;
369 new_mask = fsn_mark->mask;
370 spin_unlock(&fsn_mark->lock);
372 /* we made changes to a mask, update the group mask and the inode mask
373 * so things happen quickly. */
374 if (old_mask != new_mask) {
375 /* more bits in old than in new? */
376 int dropped = (old_mask & ~new_mask);
377 /* more bits in this mark than the inode's mask? */
378 int do_inode = (new_mask & ~inode->i_fsnotify_mask);
379 /* more bits in this mark than the group? */
380 int do_group = (new_mask & ~group->mask);
382 /* update the inode with this new mark */
383 if (dropped || do_inode)
384 fsnotify_recalc_inode_mask(inode);
386 /* update the group mask with the new mask */
387 if (dropped || do_group)
388 fsnotify_recalc_group_mask(group);
391 /* match the init or the find.... */
392 fsnotify_put_mark(fsn_mark);
397 static int fanotify_update_mark(struct fsnotify_group *group,
398 struct inode *inode, int flags,
401 pr_debug("%s: group=%p inode=%p flags=%x mask=%x\n", __func__,
402 group, inode, flags, mask);
404 if (flags & FAN_MARK_ADD)
405 fanotify_add_mark(group, inode, mask);
406 else if (flags & FAN_MARK_REMOVE)
407 fanotify_remove_mark(group, inode, mask);
414 static bool fanotify_mark_validate_input(int flags,
417 pr_debug("%s: flags=%x mask=%x\n", __func__, flags, mask);
419 /* are flags valid of this operation? */
420 if (!fanotify_mark_flags_valid(flags))
422 /* is the mask valid? */
423 if (!fanotify_mask_valid(mask))
428 /* fanotify syscalls */
429 SYSCALL_DEFINE3(fanotify_init, unsigned int, flags, unsigned int, event_f_flags,
430 unsigned int, priority)
432 struct fsnotify_group *group;
435 pr_debug("%s: flags=%d event_f_flags=%d priority=%d\n",
436 __func__, flags, event_f_flags, priority);
443 if (!capable(CAP_SYS_ADMIN))
446 if (flags & ~FAN_ALL_INIT_FLAGS)
449 f_flags = (O_RDONLY | FMODE_NONOTIFY);
450 if (flags & FAN_CLOEXEC)
451 f_flags |= O_CLOEXEC;
452 if (flags & FAN_NONBLOCK)
453 f_flags |= O_NONBLOCK;
455 /* fsnotify_alloc_group takes a ref. Dropped in fanotify_release */
456 group = fsnotify_alloc_group(&fanotify_fsnotify_ops);
458 return PTR_ERR(group);
460 fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
467 fsnotify_put_group(group);
471 SYSCALL_DEFINE5(fanotify_mark, int, fanotify_fd, unsigned int, flags,
472 __u64, mask, int, dfd, const char __user *, pathname)
475 struct fsnotify_group *group;
478 int ret, fput_needed;
480 pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
481 __func__, fanotify_fd, flags, dfd, pathname, mask);
483 /* we only use the lower 32 bits as of right now. */
484 if (mask & ((__u64)0xffffffff << 32))
487 if (!fanotify_mark_validate_input(flags, mask))
490 filp = fget_light(fanotify_fd, &fput_needed);
494 /* verify that this is indeed an fanotify instance */
496 if (unlikely(filp->f_op != &fanotify_fops))
499 ret = fanotify_find_path(dfd, pathname, &path, flags);
503 /* inode held in place by reference to path; group by fget on fd */
504 inode = path.dentry->d_inode;
505 group = filp->private_data;
507 /* create/update an inode mark */
508 ret = fanotify_update_mark(group, inode, flags, mask);
512 fput_light(filp, fput_needed);
517 * fanotify_user_setup - Our initialization function. Note that we cannnot return
518 * error because we have compiled-in VFS hooks. So an (unlikely) failure here
519 * must result in panic().
521 static int __init fanotify_user_setup(void)
523 fanotify_mark_cache = KMEM_CACHE(fsnotify_mark, SLAB_PANIC);
527 device_initcall(fanotify_user_setup);