fs/notify/fanotify/fanotify_user.c

   1 #include <linux/fanotify.h>
   2 #include <linux/fcntl.h>
   3 #include <linux/file.h>
   4 #include <linux/fs.h>
   5 #include <linux/anon_inodes.h>
   6 #include <linux/fsnotify_backend.h>
   7 #include <linux/init.h>
   8 #include <linux/mount.h>
   9 #include <linux/namei.h>
  10 #include <linux/poll.h>
  11 #include <linux/security.h>
  12 #include <linux/syscalls.h>
  13 #include <linux/slab.h>
  14 #include <linux/types.h>
  15 #include <linux/uaccess.h>
  16
  17 #include <asm/ioctls.h>
  18
  19 #include "../../mount.h"
  20
  21 #define FANOTIFY_DEFAULT_MAX_EVENTS     16384
  22 #define FANOTIFY_DEFAULT_MAX_MARKS      8192
  23 #define FANOTIFY_DEFAULT_MAX_LISTENERS  128
  24
  25 extern const struct fsnotify_ops fanotify_fsnotify_ops;
  26
  27 static struct kmem_cache *fanotify_mark_cache __read_mostly;
  28 static struct kmem_cache *fanotify_response_event_cache __read_mostly;
  29
  30 struct fanotify_response_event {
  31         struct list_head list;
  32         __s32 fd;
  33         struct fsnotify_event *event;
  34 };
  35
  36 /*
  37  * Get an fsnotify notification event if one exists and is small
  38  * enough to fit in "count". Return an error pointer if the count
  39  * is not large enough.
  40  *
  41  * Called with the group->notification_mutex held.
  42  */
  43 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
  44                                             size_t count)
  45 {
  46         BUG_ON(!mutex_is_locked(&group->notification_mutex));
  47
  48         pr_debug("%s: group=%p count=%zd\n", __func__, group, count);
  49
  50         if (fsnotify_notify_queue_is_empty(group))
  51                 return NULL;
  52
  53         if (FAN_EVENT_METADATA_LEN > count)
  54                 return ERR_PTR(-EINVAL);
  55
  56         /* held the notification_mutex the whole time, so this is the
  57          * same event we peeked above */
  58         return fsnotify_remove_notify_event(group);
  59 }
  60
  61 static int create_fd(struct fsnotify_group *group,
  62                         struct fsnotify_event *event,
  63                         struct file **file)
  64 {
  65         int client_fd;
  66         struct file *new_file;
  67
  68         pr_debug("%s: group=%p event=%p\n", __func__, group, event);
  69
  70         client_fd = get_unused_fd();
  71         if (client_fd < 0)
  72                 return client_fd;
  73
  74         if (event->data_type != FSNOTIFY_EVENT_PATH) {
  75                 WARN_ON(1);
  76                 put_unused_fd(client_fd);
  77                 return -EINVAL;
  78         }
  79
  80         /*
  81          * we need a new file handle for the userspace program so it can read even if it was
  82          * originally opened O_WRONLY.
  83          */
  84         /* it's possible this event was an overflow event.  in that case dentry and mnt
  85          * are NULL;  That's fine, just don't call dentry open */
  86         if (event->path.dentry && event->path.mnt)
  87                 new_file = dentry_open(&event->path,
  88                                        group->fanotify_data.f_flags | FMODE_NONOTIFY,
  89                                        current_cred());
  90         else
  91                 new_file = ERR_PTR(-EOVERFLOW);
  92         if (IS_ERR(new_file)) {
  93                 /*
  94                  * we still send an event even if we can't open the file.  this
  95                  * can happen when say tasks are gone and we try to open their
  96                  * /proc files or we try to open a WRONLY file like in sysfs
  97                  * we just send the errno to userspace since there isn't much
  98                  * else we can do.
  99                  */
 100                 put_unused_fd(client_fd);
 101                 client_fd = PTR_ERR(new_file);
 102         } else {
 103                 *file = new_file;
 104         }
 105
 106         return client_fd;
 107 }
 108
 109 static int fill_event_metadata(struct fsnotify_group *group,
 110                                    struct fanotify_event_metadata *metadata,
 111                                    struct fsnotify_event *event,
 112                                    struct file **file)
 113 {
 114         int ret = 0;
 115
 116         pr_debug("%s: group=%p metadata=%p event=%p\n", __func__,
 117                  group, metadata, event);
 118
 119         *file = NULL;
 120         metadata->event_len = FAN_EVENT_METADATA_LEN;
 121         metadata->metadata_len = FAN_EVENT_METADATA_LEN;
 122         metadata->vers = FANOTIFY_METADATA_VERSION;
 123         metadata->mask = event->mask & FAN_ALL_OUTGOING_EVENTS;
 124         metadata->pid = pid_vnr(event->tgid);
 125         if (unlikely(event->mask & FAN_Q_OVERFLOW))
 126                 metadata->fd = FAN_NOFD;
 127         else {
 128                 metadata->fd = create_fd(group, event, file);
 129                 if (metadata->fd < 0)
 130                         ret = metadata->fd;
 131         }
 132
 133         return ret;
 134 }
 135
 136 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 137 static struct fanotify_response_event *dequeue_re(struct fsnotify_group *group,
 138                                                   __s32 fd)
 139 {
 140         struct fanotify_response_event *re, *return_re = NULL;
 141
 142         mutex_lock(&group->fanotify_data.access_mutex);
 143         list_for_each_entry(re, &group->fanotify_data.access_list, list) {
 144                 if (re->fd != fd)
 145                         continue;
 146
 147                 list_del_init(&re->list);
 148                 return_re = re;
 149                 break;
 150         }
 151         mutex_unlock(&group->fanotify_data.access_mutex);
 152
 153         pr_debug("%s: found return_re=%p\n", __func__, return_re);
 154
 155         return return_re;
 156 }
 157
 158 static int process_access_response(struct fsnotify_group *group,
 159                                    struct fanotify_response *response_struct)
 160 {
 161         struct fanotify_response_event *re;
 162         __s32 fd = response_struct->fd;
 163         __u32 response = response_struct->response;
 164
 165         pr_debug("%s: group=%p fd=%d response=%d\n", __func__, group,
 166                  fd, response);
 167         /*
 168          * make sure the response is valid, if invalid we do nothing and either
 169          * userspace can send a valid response or we will clean it up after the
 170          * timeout
 171          */
 172         switch (response) {
 173         case FAN_ALLOW:
 174         case FAN_DENY:
 175                 break;
 176         default:
 177                 return -EINVAL;
 178         }
 179
 180         if (fd < 0)
 181                 return -EINVAL;
 182
 183         re = dequeue_re(group, fd);
 184         if (!re)
 185                 return -ENOENT;
 186
 187         re->event->response = response;
 188
 189         wake_up(&group->fanotify_data.access_waitq);
 190
 191         kmem_cache_free(fanotify_response_event_cache, re);
 192
 193         return 0;
 194 }
 195
 196 static int prepare_for_access_response(struct fsnotify_group *group,
 197                                        struct fsnotify_event *event,
 198                                        __s32 fd)
 199 {
 200         struct fanotify_response_event *re;
 201
 202         if (!(event->mask & FAN_ALL_PERM_EVENTS))
 203                 return 0;
 204
 205         re = kmem_cache_alloc(fanotify_response_event_cache, GFP_KERNEL);
 206         if (!re)
 207                 return -ENOMEM;
 208
 209         re->event = event;
 210         re->fd = fd;
 211
 212         mutex_lock(&group->fanotify_data.access_mutex);
 213
 214         if (atomic_read(&group->fanotify_data.bypass_perm)) {
 215                 mutex_unlock(&group->fanotify_data.access_mutex);
 216                 kmem_cache_free(fanotify_response_event_cache, re);
 217                 event->response = FAN_ALLOW;
 218                 return 0;
 219         }
 220
 221         list_add_tail(&re->list, &group->fanotify_data.access_list);
 222         mutex_unlock(&group->fanotify_data.access_mutex);
 223
 224         return 0;
 225 }
 226
 227 #else
 228 static int prepare_for_access_response(struct fsnotify_group *group,
 229                                        struct fsnotify_event *event,
 230                                        __s32 fd)
 231 {
 232         return 0;
 233 }
 234
 235 #endif
 236
 237 static ssize_t copy_event_to_user(struct fsnotify_group *group,
 238                                   struct fsnotify_event *event,
 239                                   char __user *buf)
 240 {
 241         struct fanotify_event_metadata fanotify_event_metadata;
 242         struct file *f;
 243         int fd, ret;
 244
 245         pr_debug("%s: group=%p event=%p\n", __func__, group, event);
 246
 247         ret = fill_event_metadata(group, &fanotify_event_metadata, event, &f);
 248         if (ret < 0)
 249                 goto out;
 250
 251         fd = fanotify_event_metadata.fd;
 252         ret = -EFAULT;
 253         if (copy_to_user(buf, &fanotify_event_metadata,
 254                          fanotify_event_metadata.event_len))
 255                 goto out_close_fd;
 256
 257         ret = prepare_for_access_response(group, event, fd);
 258         if (ret)
 259                 goto out_close_fd;
 260
 261         fd_install(fd, f);
 262         return fanotify_event_metadata.event_len;
 263
 264 out_close_fd:
 265         if (fd != FAN_NOFD) {
 266                 put_unused_fd(fd);
 267                 fput(f);
 268         }
 269 out:
 270 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 271         if (event->mask & FAN_ALL_PERM_EVENTS) {
 272                 event->response = FAN_DENY;
 273                 wake_up(&group->fanotify_data.access_waitq);
 274         }
 275 #endif
 276         return ret;
 277 }
 278
 279 /* intofiy userspace file descriptor functions */
 280 static unsigned int fanotify_poll(struct file *file, poll_table *wait)
 281 {
 282         struct fsnotify_group *group = file->private_data;
 283         int ret = 0;
 284
 285         poll_wait(file, &group->notification_waitq, wait);
 286         mutex_lock(&group->notification_mutex);
 287         if (!fsnotify_notify_queue_is_empty(group))
 288                 ret = POLLIN | POLLRDNORM;
 289         mutex_unlock(&group->notification_mutex);
 290
 291         return ret;
 292 }
 293
 294 static ssize_t fanotify_read(struct file *file, char __user *buf,
 295                              size_t count, loff_t *pos)
 296 {
 297         struct fsnotify_group *group;
 298         struct fsnotify_event *kevent;
 299         char __user *start;
 300         int ret;
 301         DEFINE_WAIT(wait);
 302
 303         start = buf;
 304         group = file->private_data;
 305
 306         pr_debug("%s: group=%p\n", __func__, group);
 307
 308         while (1) {
 309                 prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
 310
 311                 mutex_lock(&group->notification_mutex);
 312                 kevent = get_one_event(group, count);
 313                 mutex_unlock(&group->notification_mutex);
 314
 315                 if (kevent) {
 316                         ret = PTR_ERR(kevent);
 317                         if (IS_ERR(kevent))
 318                                 break;
 319                         ret = copy_event_to_user(group, kevent, buf);
 320                         fsnotify_put_event(kevent);
 321                         if (ret < 0)
 322                                 break;
 323                         buf += ret;
 324                         count -= ret;
 325                         continue;
 326                 }
 327
 328                 ret = -EAGAIN;
 329                 if (file->f_flags & O_NONBLOCK)
 330                         break;
 331                 ret = -ERESTARTSYS;
 332                 if (signal_pending(current))
 333                         break;
 334
 335                 if (start != buf)
 336                         break;
 337
 338                 schedule();
 339         }
 340
 341         finish_wait(&group->notification_waitq, &wait);
 342         if (start != buf && ret != -EFAULT)
 343                 ret = buf - start;
 344         return ret;
 345 }
 346
 347 static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
 348 {
 349 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 350         struct fanotify_response response = { .fd = -1, .response = -1 };
 351         struct fsnotify_group *group;
 352         int ret;
 353
 354         group = file->private_data;
 355
 356         if (count > sizeof(response))
 357                 count = sizeof(response);
 358
 359         pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
 360
 361         if (copy_from_user(&response, buf, count))
 362                 return -EFAULT;
 363
 364         ret = process_access_response(group, &response);
 365         if (ret < 0)
 366                 count = ret;
 367
 368         return count;
 369 #else
 370         return -EINVAL;
 371 #endif
 372 }
 373
 374 static int fanotify_release(struct inode *ignored, struct file *file)
 375 {
 376         struct fsnotify_group *group = file->private_data;
 377
 378 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 379         struct fanotify_response_event *re, *lre;
 380
 381         mutex_lock(&group->fanotify_data.access_mutex);
 382
 383         atomic_inc(&group->fanotify_data.bypass_perm);
 384
 385         list_for_each_entry_safe(re, lre, &group->fanotify_data.access_list, list) {
 386                 pr_debug("%s: found group=%p re=%p event=%p\n", __func__, group,
 387                          re, re->event);
 388
 389                 list_del_init(&re->list);
 390                 re->event->response = FAN_ALLOW;
 391
 392                 kmem_cache_free(fanotify_response_event_cache, re);
 393         }
 394         mutex_unlock(&group->fanotify_data.access_mutex);
 395
 396         wake_up(&group->fanotify_data.access_waitq);
 397 #endif
 398         /* matches the fanotify_init->fsnotify_alloc_group */
 399         fsnotify_put_group(group);
 400
 401         return 0;
 402 }
 403
 404 static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 405 {
 406         struct fsnotify_group *group;
 407         struct fsnotify_event_holder *holder;
 408         void __user *p;
 409         int ret = -ENOTTY;
 410         size_t send_len = 0;
 411
 412         group = file->private_data;
 413
 414         p = (void __user *) arg;
 415
 416         switch (cmd) {
 417         case FIONREAD:
 418                 mutex_lock(&group->notification_mutex);
 419                 list_for_each_entry(holder, &group->notification_list, event_list)
 420                         send_len += FAN_EVENT_METADATA_LEN;
 421                 mutex_unlock(&group->notification_mutex);
 422                 ret = put_user(send_len, (int __user *) p);
 423                 break;
 424         }
 425
 426         return ret;
 427 }
 428
 429 static const struct file_operations fanotify_fops = {
 430         .poll           = fanotify_poll,
 431         .read           = fanotify_read,
 432         .write          = fanotify_write,
 433         .fasync         = NULL,
 434         .release        = fanotify_release,
 435         .unlocked_ioctl = fanotify_ioctl,
 436         .compat_ioctl   = fanotify_ioctl,
 437         .llseek         = noop_llseek,
 438 };
 439
 440 static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
 441 {
 442         kmem_cache_free(fanotify_mark_cache, fsn_mark);
 443 }
 444
 445 static int fanotify_find_path(int dfd, const char __user *filename,
 446                               struct path *path, unsigned int flags)
 447 {
 448         int ret;
 449
 450         pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
 451                  dfd, filename, flags);
 452
 453         if (filename == NULL) {
 454                 struct file *file;
 455                 int fput_needed;
 456
 457                 ret = -EBADF;
 458                 file = fget_light(dfd, &fput_needed);
 459                 if (!file)
 460                         goto out;
 461
 462                 ret = -ENOTDIR;
 463                 if ((flags & FAN_MARK_ONLYDIR) &&
 464                     !(S_ISDIR(file->f_path.dentry->d_inode->i_mode))) {
 465                         fput_light(file, fput_needed);
 466                         goto out;
 467                 }
 468
 469                 *path = file->f_path;
 470                 path_get(path);
 471                 fput_light(file, fput_needed);
 472         } else {
 473                 unsigned int lookup_flags = 0;
 474
 475                 if (!(flags & FAN_MARK_DONT_FOLLOW))
 476                         lookup_flags |= LOOKUP_FOLLOW;
 477                 if (flags & FAN_MARK_ONLYDIR)
 478                         lookup_flags |= LOOKUP_DIRECTORY;
 479
 480                 ret = user_path_at(dfd, filename, lookup_flags, path);
 481                 if (ret)
 482                         goto out;
 483         }
 484
 485         /* you can only watch an inode if you have read permissions on it */
 486         ret = inode_permission(path->dentry->d_inode, MAY_READ);
 487         if (ret)
 488                 path_put(path);
 489 out:
 490         return ret;
 491 }
 492
 493 static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
 494                                             __u32 mask,
 495                                             unsigned int flags)
 496 {
 497         __u32 oldmask;
 498
 499         spin_lock(&fsn_mark->lock);
 500         if (!(flags & FAN_MARK_IGNORED_MASK)) {
 501                 oldmask = fsn_mark->mask;
 502                 fsnotify_set_mark_mask_locked(fsn_mark, (oldmask & ~mask));
 503         } else {
 504                 oldmask = fsn_mark->ignored_mask;
 505                 fsnotify_set_mark_ignored_mask_locked(fsn_mark, (oldmask & ~mask));
 506         }
 507         spin_unlock(&fsn_mark->lock);
 508
 509         if (!(oldmask & ~mask))
 510                 fsnotify_destroy_mark(fsn_mark);
 511
 512         return mask & oldmask;
 513 }
 514
 515 static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
 516                                          struct vfsmount *mnt, __u32 mask,
 517                                          unsigned int flags)
 518 {
 519         struct fsnotify_mark *fsn_mark = NULL;
 520         __u32 removed;
 521
 522         fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
 523         if (!fsn_mark)
 524                 return -ENOENT;
 525
 526         removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags);
 527         fsnotify_put_mark(fsn_mark);
 528         if (removed & real_mount(mnt)->mnt_fsnotify_mask)
 529                 fsnotify_recalc_vfsmount_mask(mnt);
 530
 531         return 0;
 532 }
 533
 534 static int fanotify_remove_inode_mark(struct fsnotify_group *group,
 535                                       struct inode *inode, __u32 mask,
 536                                       unsigned int flags)
 537 {
 538         struct fsnotify_mark *fsn_mark = NULL;
 539         __u32 removed;
 540
 541         fsn_mark = fsnotify_find_inode_mark(group, inode);
 542         if (!fsn_mark)
 543                 return -ENOENT;
 544
 545         removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags);
 546         /* matches the fsnotify_find_inode_mark() */
 547         fsnotify_put_mark(fsn_mark);
 548         if (removed & inode->i_fsnotify_mask)
 549                 fsnotify_recalc_inode_mask(inode);
 550
 551         return 0;
 552 }
 553
 554 static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
 555                                        __u32 mask,
 556                                        unsigned int flags)
 557 {
 558         __u32 oldmask = -1;
 559
 560         spin_lock(&fsn_mark->lock);
 561         if (!(flags & FAN_MARK_IGNORED_MASK)) {
 562                 oldmask = fsn_mark->mask;
 563                 fsnotify_set_mark_mask_locked(fsn_mark, (oldmask | mask));
 564         } else {
 565                 __u32 tmask = fsn_mark->ignored_mask | mask;
 566                 fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
 567                 if (flags & FAN_MARK_IGNORED_SURV_MODIFY)
 568                         fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
 569         }
 570
 571         if (!(flags & FAN_MARK_ONDIR)) {
 572                 __u32 tmask = fsn_mark->ignored_mask | FAN_ONDIR;
 573                 fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
 574         }
 575
 576         spin_unlock(&fsn_mark->lock);
 577
 578         return mask & ~oldmask;
 579 }
 580
 581 static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
 582                                       struct vfsmount *mnt, __u32 mask,
 583                                       unsigned int flags)
 584 {
 585         struct fsnotify_mark *fsn_mark;
 586         __u32 added;
 587         int ret = 0;
 588
 589         fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
 590         if (!fsn_mark) {
 591                 if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
 592                         return -ENOSPC;
 593
 594                 fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
 595                 if (!fsn_mark)
 596                         return -ENOMEM;
 597
 598                 fsnotify_init_mark(fsn_mark, fanotify_free_mark);
 599                 ret = fsnotify_add_mark(fsn_mark, group, NULL, mnt, 0);
 600                 if (ret)
 601                         goto err;
 602         }
 603         added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
 604
 605         if (added & ~real_mount(mnt)->mnt_fsnotify_mask)
 606                 fsnotify_recalc_vfsmount_mask(mnt);
 607 err:
 608         fsnotify_put_mark(fsn_mark);
 609         return ret;
 610 }
 611
 612 static int fanotify_add_inode_mark(struct fsnotify_group *group,
 613                                    struct inode *inode, __u32 mask,
 614                                    unsigned int flags)
 615 {
 616         struct fsnotify_mark *fsn_mark;
 617         __u32 added;
 618         int ret = 0;
 619
 620         pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
 621
 622         /*
 623          * If some other task has this inode open for write we should not add
 624          * an ignored mark, unless that ignored mark is supposed to survive
 625          * modification changes anyway.
 626          */
 627         if ((flags & FAN_MARK_IGNORED_MASK) &&
 628             !(flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
 629             (atomic_read(&inode->i_writecount) > 0))
 630                 return 0;
 631
 632         fsn_mark = fsnotify_find_inode_mark(group, inode);
 633         if (!fsn_mark) {
 634                 if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
 635                         return -ENOSPC;
 636
 637                 fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
 638                 if (!fsn_mark)
 639                         return -ENOMEM;
 640
 641                 fsnotify_init_mark(fsn_mark, fanotify_free_mark);
 642                 ret = fsnotify_add_mark(fsn_mark, group, inode, NULL, 0);
 643                 if (ret)
 644                         goto err;
 645         }
 646         added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
 647
 648         if (added & ~inode->i_fsnotify_mask)
 649                 fsnotify_recalc_inode_mask(inode);
 650 err:
 651         fsnotify_put_mark(fsn_mark);
 652         return ret;
 653 }
 654
 655 /* fanotify syscalls */
 656 SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
 657 {
 658         struct fsnotify_group *group;
 659         int f_flags, fd;
 660         struct user_struct *user;
 661
 662         pr_debug("%s: flags=%d event_f_flags=%d\n",
 663                 __func__, flags, event_f_flags);
 664
 665         if (!capable(CAP_SYS_ADMIN))
 666                 return -EPERM;
 667
 668         if (flags & ~FAN_ALL_INIT_FLAGS)
 669                 return -EINVAL;
 670
 671         user = get_current_user();
 672         if (atomic_read(&user->fanotify_listeners) > FANOTIFY_DEFAULT_MAX_LISTENERS) {
 673                 free_uid(user);
 674                 return -EMFILE;
 675         }
 676
 677         f_flags = O_RDWR | FMODE_NONOTIFY;
 678         if (flags & FAN_CLOEXEC)
 679                 f_flags |= O_CLOEXEC;
 680         if (flags & FAN_NONBLOCK)
 681                 f_flags |= O_NONBLOCK;
 682
 683         /* fsnotify_alloc_group takes a ref.  Dropped in fanotify_release */
 684         group = fsnotify_alloc_group(&fanotify_fsnotify_ops);
 685         if (IS_ERR(group)) {
 686                 free_uid(user);
 687                 return PTR_ERR(group);
 688         }
 689
 690         group->fanotify_data.user = user;
 691         atomic_inc(&user->fanotify_listeners);
 692
 693         group->fanotify_data.f_flags = event_f_flags;
 694 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 695         mutex_init(&group->fanotify_data.access_mutex);
 696         init_waitqueue_head(&group->fanotify_data.access_waitq);
 697         INIT_LIST_HEAD(&group->fanotify_data.access_list);
 698         atomic_set(&group->fanotify_data.bypass_perm, 0);
 699 #endif
 700         switch (flags & FAN_ALL_CLASS_BITS) {
 701         case FAN_CLASS_NOTIF:
 702                 group->priority = FS_PRIO_0;
 703                 break;
 704         case FAN_CLASS_CONTENT:
 705                 group->priority = FS_PRIO_1;
 706                 break;
 707         case FAN_CLASS_PRE_CONTENT:
 708                 group->priority = FS_PRIO_2;
 709                 break;
 710         default:
 711                 fd = -EINVAL;
 712                 goto out_put_group;
 713         }
 714
 715         if (flags & FAN_UNLIMITED_QUEUE) {
 716                 fd = -EPERM;
 717                 if (!capable(CAP_SYS_ADMIN))
 718                         goto out_put_group;
 719                 group->max_events = UINT_MAX;
 720         } else {
 721                 group->max_events = FANOTIFY_DEFAULT_MAX_EVENTS;
 722         }
 723
 724         if (flags & FAN_UNLIMITED_MARKS) {
 725                 fd = -EPERM;
 726                 if (!capable(CAP_SYS_ADMIN))
 727                         goto out_put_group;
 728                 group->fanotify_data.max_marks = UINT_MAX;
 729         } else {
 730                 group->fanotify_data.max_marks = FANOTIFY_DEFAULT_MAX_MARKS;
 731         }
 732
 733         fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
 734         if (fd < 0)
 735                 goto out_put_group;
 736
 737         return fd;
 738
 739 out_put_group:
 740         fsnotify_put_group(group);
 741         return fd;
 742 }
 743
 744 SYSCALL_DEFINE(fanotify_mark)(int fanotify_fd, unsigned int flags,
 745                               __u64 mask, int dfd,
 746                               const char  __user * pathname)
 747 {
 748         struct inode *inode = NULL;
 749         struct vfsmount *mnt = NULL;
 750         struct fsnotify_group *group;
 751         struct file *filp;
 752         struct path path;
 753         int ret, fput_needed;
 754
 755         pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
 756                  __func__, fanotify_fd, flags, dfd, pathname, mask);
 757
 758         /* we only use the lower 32 bits as of right now. */
 759         if (mask & ((__u64)0xffffffff << 32))
 760                 return -EINVAL;
 761
 762         if (flags & ~FAN_ALL_MARK_FLAGS)
 763                 return -EINVAL;
 764         switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
 765         case FAN_MARK_ADD:              /* fallthrough */
 766         case FAN_MARK_REMOVE:
 767                 if (!mask)
 768                         return -EINVAL;
 769         case FAN_MARK_FLUSH:
 770                 break;
 771         default:
 772                 return -EINVAL;
 773         }
 774
 775         if (mask & FAN_ONDIR) {
 776                 flags |= FAN_MARK_ONDIR;
 777                 mask &= ~FAN_ONDIR;
 778         }
 779
 780 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 781         if (mask & ~(FAN_ALL_EVENTS | FAN_ALL_PERM_EVENTS | FAN_EVENT_ON_CHILD))
 782 #else
 783         if (mask & ~(FAN_ALL_EVENTS | FAN_EVENT_ON_CHILD))
 784 #endif
 785                 return -EINVAL;
 786
 787         filp = fget_light(fanotify_fd, &fput_needed);
 788         if (unlikely(!filp))
 789                 return -EBADF;
 790
 791         /* verify that this is indeed an fanotify instance */
 792         ret = -EINVAL;
 793         if (unlikely(filp->f_op != &fanotify_fops))
 794                 goto fput_and_out;
 795         group = filp->private_data;
 796
 797         /*
 798          * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF.  These are not
 799          * allowed to set permissions events.
 800          */
 801         ret = -EINVAL;
 802         if (mask & FAN_ALL_PERM_EVENTS &&
 803             group->priority == FS_PRIO_0)
 804                 goto fput_and_out;
 805
 806         ret = fanotify_find_path(dfd, pathname, &path, flags);
 807         if (ret)
 808                 goto fput_and_out;
 809
 810         /* inode held in place by reference to path; group by fget on fd */
 811         if (!(flags & FAN_MARK_MOUNT))
 812                 inode = path.dentry->d_inode;
 813         else
 814                 mnt = path.mnt;
 815
 816         /* create/update an inode mark */
 817         switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
 818         case FAN_MARK_ADD:
 819                 if (flags & FAN_MARK_MOUNT)
 820                         ret = fanotify_add_vfsmount_mark(group, mnt, mask, flags);
 821                 else
 822                         ret = fanotify_add_inode_mark(group, inode, mask, flags);
 823                 break;
 824         case FAN_MARK_REMOVE:
 825                 if (flags & FAN_MARK_MOUNT)
 826                         ret = fanotify_remove_vfsmount_mark(group, mnt, mask, flags);
 827                 else
 828                         ret = fanotify_remove_inode_mark(group, inode, mask, flags);
 829                 break;
 830         case FAN_MARK_FLUSH:
 831                 if (flags & FAN_MARK_MOUNT)
 832                         fsnotify_clear_vfsmount_marks_by_group(group);
 833                 else
 834                         fsnotify_clear_inode_marks_by_group(group);
 835                 break;
 836         default:
 837                 ret = -EINVAL;
 838         }
 839
 840         path_put(&path);
 841 fput_and_out:
 842         fput_light(filp, fput_needed);
 843         return ret;
 844 }
 845
 846 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 847 asmlinkage long SyS_fanotify_mark(long fanotify_fd, long flags, __u64 mask,
 848                                   long dfd, long pathname)
 849 {
 850         return SYSC_fanotify_mark((int) fanotify_fd, (unsigned int) flags,
 851                                   mask, (int) dfd,
 852                                   (const char  __user *) pathname);
 853 }
 854 SYSCALL_ALIAS(sys_fanotify_mark, SyS_fanotify_mark);
 855 #endif
 856
 857 /*
 858  * fanotify_user_setup - Our initialization function.  Note that we cannot return
 859  * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
 860  * must result in panic().
 861  */
 862 static int __init fanotify_user_setup(void)
 863 {
 864         fanotify_mark_cache = KMEM_CACHE(fsnotify_mark, SLAB_PANIC);
 865         fanotify_response_event_cache = KMEM_CACHE(fanotify_response_event,
 866                                                    SLAB_PANIC);
 867
 868         return 0;
 869 }
 870 device_initcall(fanotify_user_setup);