fs/notify/fanotify/fanotify_user.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/fanotify.h>
   3 #include <linux/fcntl.h>
   4 #include <linux/file.h>
   5 #include <linux/fs.h>
   6 #include <linux/anon_inodes.h>
   7 #include <linux/fsnotify_backend.h>
   8 #include <linux/init.h>
   9 #include <linux/mount.h>
  10 #include <linux/namei.h>
  11 #include <linux/poll.h>
  12 #include <linux/security.h>
  13 #include <linux/syscalls.h>
  14 #include <linux/slab.h>
  15 #include <linux/types.h>
  16 #include <linux/uaccess.h>
  17 #include <linux/compat.h>
  18 #include <linux/sched/signal.h>
  19
  20 #include <asm/ioctls.h>
  21
  22 #include "../../mount.h"
  23 #include "../fdinfo.h"
  24 #include "fanotify.h"
  25
  26 #define FANOTIFY_DEFAULT_MAX_EVENTS     16384
  27 #define FANOTIFY_DEFAULT_MAX_MARKS      8192
  28 #define FANOTIFY_DEFAULT_MAX_LISTENERS  128
  29
  30 /*
  31  * All flags that may be specified in parameter event_f_flags of fanotify_init.
  32  *
  33  * Internal and external open flags are stored together in field f_flags of
  34  * struct file. Only external open flags shall be allowed in event_f_flags.
  35  * Internal flags like FMODE_NONOTIFY, FMODE_EXEC, FMODE_NOCMTIME shall be
  36  * excluded.
  37  */
  38 #define FANOTIFY_INIT_ALL_EVENT_F_BITS                          ( \
  39                 O_ACCMODE       | O_APPEND      | O_NONBLOCK    | \
  40                 __O_SYNC        | O_DSYNC       | O_CLOEXEC     | \
  41                 O_LARGEFILE     | O_NOATIME     )
  42
  43 extern const struct fsnotify_ops fanotify_fsnotify_ops;
  44
  45 struct kmem_cache *fanotify_mark_cache __read_mostly;
  46 struct kmem_cache *fanotify_event_cachep __read_mostly;
  47 struct kmem_cache *fanotify_perm_event_cachep __read_mostly;
  48
  49 /*
  50  * Get an fsnotify notification event if one exists and is small
  51  * enough to fit in "count". Return an error pointer if the count
  52  * is not large enough.
  53  *
  54  * Called with the group->notification_lock held.
  55  */
  56 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
  57                                             size_t count)
  58 {
  59         assert_spin_locked(&group->notification_lock);
  60
  61         pr_debug("%s: group=%p count=%zd\n", __func__, group, count);
  62
  63         if (fsnotify_notify_queue_is_empty(group))
  64                 return NULL;
  65
  66         if (FAN_EVENT_METADATA_LEN > count)
  67                 return ERR_PTR(-EINVAL);
  68
  69         /* held the notification_lock the whole time, so this is the
  70          * same event we peeked above */
  71         return fsnotify_remove_first_event(group);
  72 }
  73
  74 static int create_fd(struct fsnotify_group *group,
  75                      struct fanotify_event_info *event,
  76                      struct file **file)
  77 {
  78         int client_fd;
  79         struct file *new_file;
  80
  81         pr_debug("%s: group=%p event=%p\n", __func__, group, event);
  82
  83         client_fd = get_unused_fd_flags(group->fanotify_data.f_flags);
  84         if (client_fd < 0)
  85                 return client_fd;
  86
  87         /*
  88          * we need a new file handle for the userspace program so it can read even if it was
  89          * originally opened O_WRONLY.
  90          */
  91         /* it's possible this event was an overflow event.  in that case dentry and mnt
  92          * are NULL;  That's fine, just don't call dentry open */
  93         if (event->path.dentry && event->path.mnt)
  94                 new_file = dentry_open(&event->path,
  95                                        group->fanotify_data.f_flags | FMODE_NONOTIFY,
  96                                        current_cred());
  97         else
  98                 new_file = ERR_PTR(-EOVERFLOW);
  99         if (IS_ERR(new_file)) {
 100                 /*
 101                  * we still send an event even if we can't open the file.  this
 102                  * can happen when say tasks are gone and we try to open their
 103                  * /proc files or we try to open a WRONLY file like in sysfs
 104                  * we just send the errno to userspace since there isn't much
 105                  * else we can do.
 106                  */
 107                 put_unused_fd(client_fd);
 108                 client_fd = PTR_ERR(new_file);
 109         } else {
 110                 *file = new_file;
 111         }
 112
 113         return client_fd;
 114 }
 115
 116 static int fill_event_metadata(struct fsnotify_group *group,
 117                                struct fanotify_event_metadata *metadata,
 118                                struct fsnotify_event *fsn_event,
 119                                struct file **file)
 120 {
 121         int ret = 0;
 122         struct fanotify_event_info *event;
 123
 124         pr_debug("%s: group=%p metadata=%p event=%p\n", __func__,
 125                  group, metadata, fsn_event);
 126
 127         *file = NULL;
 128         event = container_of(fsn_event, struct fanotify_event_info, fse);
 129         metadata->event_len = FAN_EVENT_METADATA_LEN;
 130         metadata->metadata_len = FAN_EVENT_METADATA_LEN;
 131         metadata->vers = FANOTIFY_METADATA_VERSION;
 132         metadata->reserved = 0;
 133         metadata->mask = fsn_event->mask & FAN_ALL_OUTGOING_EVENTS;
 134         metadata->pid = pid_vnr(event->tgid);
 135         if (unlikely(fsn_event->mask & FAN_Q_OVERFLOW))
 136                 metadata->fd = FAN_NOFD;
 137         else {
 138                 metadata->fd = create_fd(group, event, file);
 139                 if (metadata->fd < 0)
 140                         ret = metadata->fd;
 141         }
 142
 143         return ret;
 144 }
 145
 146 static struct fanotify_perm_event_info *dequeue_event(
 147                                 struct fsnotify_group *group, int fd)
 148 {
 149         struct fanotify_perm_event_info *event, *return_e = NULL;
 150
 151         spin_lock(&group->notification_lock);
 152         list_for_each_entry(event, &group->fanotify_data.access_list,
 153                             fae.fse.list) {
 154                 if (event->fd != fd)
 155                         continue;
 156
 157                 list_del_init(&event->fae.fse.list);
 158                 return_e = event;
 159                 break;
 160         }
 161         spin_unlock(&group->notification_lock);
 162
 163         pr_debug("%s: found return_re=%p\n", __func__, return_e);
 164
 165         return return_e;
 166 }
 167
 168 static int process_access_response(struct fsnotify_group *group,
 169                                    struct fanotify_response *response_struct)
 170 {
 171         struct fanotify_perm_event_info *event;
 172         int fd = response_struct->fd;
 173         int response = response_struct->response;
 174
 175         pr_debug("%s: group=%p fd=%d response=%d\n", __func__, group,
 176                  fd, response);
 177         /*
 178          * make sure the response is valid, if invalid we do nothing and either
 179          * userspace can send a valid response or we will clean it up after the
 180          * timeout
 181          */
 182         switch (response) {
 183         case FAN_ALLOW:
 184         case FAN_DENY:
 185                 break;
 186         default:
 187                 return -EINVAL;
 188         }
 189
 190         if (fd < 0)
 191                 return -EINVAL;
 192
 193         event = dequeue_event(group, fd);
 194         if (!event)
 195                 return -ENOENT;
 196
 197         event->response = response;
 198         wake_up(&group->fanotify_data.access_waitq);
 199
 200         return 0;
 201 }
 202
 203 static ssize_t copy_event_to_user(struct fsnotify_group *group,
 204                                   struct fsnotify_event *event,
 205                                   char __user *buf)
 206 {
 207         struct fanotify_event_metadata fanotify_event_metadata;
 208         struct file *f;
 209         int fd, ret;
 210
 211         pr_debug("%s: group=%p event=%p\n", __func__, group, event);
 212
 213         ret = fill_event_metadata(group, &fanotify_event_metadata, event, &f);
 214         if (ret < 0)
 215                 return ret;
 216
 217         fd = fanotify_event_metadata.fd;
 218         ret = -EFAULT;
 219         if (copy_to_user(buf, &fanotify_event_metadata,
 220                          fanotify_event_metadata.event_len))
 221                 goto out_close_fd;
 222
 223         if (fanotify_is_perm_event(event->mask))
 224                 FANOTIFY_PE(event)->fd = fd;
 225
 226         if (fd != FAN_NOFD)
 227                 fd_install(fd, f);
 228         return fanotify_event_metadata.event_len;
 229
 230 out_close_fd:
 231         if (fd != FAN_NOFD) {
 232                 put_unused_fd(fd);
 233                 fput(f);
 234         }
 235         return ret;
 236 }
 237
 238 /* intofiy userspace file descriptor functions */
 239 static unsigned int fanotify_poll(struct file *file, poll_table *wait)
 240 {
 241         struct fsnotify_group *group = file->private_data;
 242         int ret = 0;
 243
 244         poll_wait(file, &group->notification_waitq, wait);
 245         spin_lock(&group->notification_lock);
 246         if (!fsnotify_notify_queue_is_empty(group))
 247                 ret = POLLIN | POLLRDNORM;
 248         spin_unlock(&group->notification_lock);
 249
 250         return ret;
 251 }
 252
 253 static ssize_t fanotify_read(struct file *file, char __user *buf,
 254                              size_t count, loff_t *pos)
 255 {
 256         struct fsnotify_group *group;
 257         struct fsnotify_event *kevent;
 258         char __user *start;
 259         int ret;
 260         DEFINE_WAIT_FUNC(wait, woken_wake_function);
 261
 262         start = buf;
 263         group = file->private_data;
 264
 265         pr_debug("%s: group=%p\n", __func__, group);
 266
 267         add_wait_queue(&group->notification_waitq, &wait);
 268         while (1) {
 269                 spin_lock(&group->notification_lock);
 270                 kevent = get_one_event(group, count);
 271                 spin_unlock(&group->notification_lock);
 272
 273                 if (IS_ERR(kevent)) {
 274                         ret = PTR_ERR(kevent);
 275                         break;
 276                 }
 277
 278                 if (!kevent) {
 279                         ret = -EAGAIN;
 280                         if (file->f_flags & O_NONBLOCK)
 281                                 break;
 282
 283                         ret = -ERESTARTSYS;
 284                         if (signal_pending(current))
 285                                 break;
 286
 287                         if (start != buf)
 288                                 break;
 289
 290                         wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
 291                         continue;
 292                 }
 293
 294                 ret = copy_event_to_user(group, kevent, buf);
 295                 if (unlikely(ret == -EOPENSTALE)) {
 296                         /*
 297                          * We cannot report events with stale fd so drop it.
 298                          * Setting ret to 0 will continue the event loop and
 299                          * do the right thing if there are no more events to
 300                          * read (i.e. return bytes read, -EAGAIN or wait).
 301                          */
 302                         ret = 0;
 303                 }
 304
 305                 /*
 306                  * Permission events get queued to wait for response.  Other
 307                  * events can be destroyed now.
 308                  */
 309                 if (!fanotify_is_perm_event(kevent->mask)) {
 310                         fsnotify_destroy_event(group, kevent);
 311                 } else {
 312                         if (ret <= 0) {
 313                                 FANOTIFY_PE(kevent)->response = FAN_DENY;
 314                                 wake_up(&group->fanotify_data.access_waitq);
 315                         } else {
 316                                 spin_lock(&group->notification_lock);
 317                                 list_add_tail(&kevent->list,
 318                                         &group->fanotify_data.access_list);
 319                                 spin_unlock(&group->notification_lock);
 320                         }
 321                 }
 322                 if (ret < 0)
 323                         break;
 324                 buf += ret;
 325                 count -= ret;
 326         }
 327         remove_wait_queue(&group->notification_waitq, &wait);
 328
 329         if (start != buf && ret != -EFAULT)
 330                 ret = buf - start;
 331         return ret;
 332 }
 333
 334 static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
 335 {
 336         struct fanotify_response response = { .fd = -1, .response = -1 };
 337         struct fsnotify_group *group;
 338         int ret;
 339
 340         if (!IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS))
 341                 return -EINVAL;
 342
 343         group = file->private_data;
 344
 345         if (count > sizeof(response))
 346                 count = sizeof(response);
 347
 348         pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
 349
 350         if (copy_from_user(&response, buf, count))
 351                 return -EFAULT;
 352
 353         ret = process_access_response(group, &response);
 354         if (ret < 0)
 355                 count = ret;
 356
 357         return count;
 358 }
 359
 360 static int fanotify_release(struct inode *ignored, struct file *file)
 361 {
 362         struct fsnotify_group *group = file->private_data;
 363         struct fanotify_perm_event_info *event, *next;
 364         struct fsnotify_event *fsn_event;
 365
 366         /*
 367          * Stop new events from arriving in the notification queue. since
 368          * userspace cannot use fanotify fd anymore, no event can enter or
 369          * leave access_list by now either.
 370          */
 371         fsnotify_group_stop_queueing(group);
 372
 373         /*
 374          * Process all permission events on access_list and notification queue
 375          * and simulate reply from userspace.
 376          */
 377         spin_lock(&group->notification_lock);
 378         list_for_each_entry_safe(event, next, &group->fanotify_data.access_list,
 379                                  fae.fse.list) {
 380                 pr_debug("%s: found group=%p event=%p\n", __func__, group,
 381                          event);
 382
 383                 list_del_init(&event->fae.fse.list);
 384                 event->response = FAN_ALLOW;
 385         }
 386
 387         /*
 388          * Destroy all non-permission events. For permission events just
 389          * dequeue them and set the response. They will be freed once the
 390          * response is consumed and fanotify_get_response() returns.
 391          */
 392         while (!fsnotify_notify_queue_is_empty(group)) {
 393                 fsn_event = fsnotify_remove_first_event(group);
 394                 if (!(fsn_event->mask & FAN_ALL_PERM_EVENTS)) {
 395                         spin_unlock(&group->notification_lock);
 396                         fsnotify_destroy_event(group, fsn_event);
 397                         spin_lock(&group->notification_lock);
 398                 } else {
 399                         FANOTIFY_PE(fsn_event)->response = FAN_ALLOW;
 400                 }
 401         }
 402         spin_unlock(&group->notification_lock);
 403
 404         /* Response for all permission events it set, wakeup waiters */
 405         wake_up(&group->fanotify_data.access_waitq);
 406
 407         /* matches the fanotify_init->fsnotify_alloc_group */
 408         fsnotify_destroy_group(group);
 409
 410         return 0;
 411 }
 412
 413 static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 414 {
 415         struct fsnotify_group *group;
 416         struct fsnotify_event *fsn_event;
 417         void __user *p;
 418         int ret = -ENOTTY;
 419         size_t send_len = 0;
 420
 421         group = file->private_data;
 422
 423         p = (void __user *) arg;
 424
 425         switch (cmd) {
 426         case FIONREAD:
 427                 spin_lock(&group->notification_lock);
 428                 list_for_each_entry(fsn_event, &group->notification_list, list)
 429                         send_len += FAN_EVENT_METADATA_LEN;
 430                 spin_unlock(&group->notification_lock);
 431                 ret = put_user(send_len, (int __user *) p);
 432                 break;
 433         }
 434
 435         return ret;
 436 }
 437
 438 static const struct file_operations fanotify_fops = {
 439         .show_fdinfo    = fanotify_show_fdinfo,
 440         .poll           = fanotify_poll,
 441         .read           = fanotify_read,
 442         .write          = fanotify_write,
 443         .fasync         = NULL,
 444         .release        = fanotify_release,
 445         .unlocked_ioctl = fanotify_ioctl,
 446         .compat_ioctl   = fanotify_ioctl,
 447         .llseek         = noop_llseek,
 448 };
 449
 450 static int fanotify_find_path(int dfd, const char __user *filename,
 451                               struct path *path, unsigned int flags)
 452 {
 453         int ret;
 454
 455         pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
 456                  dfd, filename, flags);
 457
 458         if (filename == NULL) {
 459                 struct fd f = fdget(dfd);
 460
 461                 ret = -EBADF;
 462                 if (!f.file)
 463                         goto out;
 464
 465                 ret = -ENOTDIR;
 466                 if ((flags & FAN_MARK_ONLYDIR) &&
 467                     !(S_ISDIR(file_inode(f.file)->i_mode))) {
 468                         fdput(f);
 469                         goto out;
 470                 }
 471
 472                 *path = f.file->f_path;
 473                 path_get(path);
 474                 fdput(f);
 475         } else {
 476                 unsigned int lookup_flags = 0;
 477
 478                 if (!(flags & FAN_MARK_DONT_FOLLOW))
 479                         lookup_flags |= LOOKUP_FOLLOW;
 480                 if (flags & FAN_MARK_ONLYDIR)
 481                         lookup_flags |= LOOKUP_DIRECTORY;
 482
 483                 ret = user_path_at(dfd, filename, lookup_flags, path);
 484                 if (ret)
 485                         goto out;
 486         }
 487
 488         /* you can only watch an inode if you have read permissions on it */
 489         ret = inode_permission(path->dentry->d_inode, MAY_READ);
 490         if (ret)
 491                 path_put(path);
 492 out:
 493         return ret;
 494 }
 495
 496 static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
 497                                             __u32 mask,
 498                                             unsigned int flags,
 499                                             int *destroy)
 500 {
 501         __u32 oldmask = 0;
 502
 503         spin_lock(&fsn_mark->lock);
 504         if (!(flags & FAN_MARK_IGNORED_MASK)) {
 505                 __u32 tmask = fsn_mark->mask & ~mask;
 506
 507                 if (flags & FAN_MARK_ONDIR)
 508                         tmask &= ~FAN_ONDIR;
 509
 510                 oldmask = fsn_mark->mask;
 511                 fsn_mark->mask = tmask;
 512         } else {
 513                 __u32 tmask = fsn_mark->ignored_mask & ~mask;
 514                 if (flags & FAN_MARK_ONDIR)
 515                         tmask &= ~FAN_ONDIR;
 516                 fsn_mark->ignored_mask = tmask;
 517         }
 518         *destroy = !(fsn_mark->mask | fsn_mark->ignored_mask);
 519         spin_unlock(&fsn_mark->lock);
 520
 521         return mask & oldmask;
 522 }
 523
 524 static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
 525                                          struct vfsmount *mnt, __u32 mask,
 526                                          unsigned int flags)
 527 {
 528         struct fsnotify_mark *fsn_mark = NULL;
 529         __u32 removed;
 530         int destroy_mark;
 531
 532         mutex_lock(&group->mark_mutex);
 533         fsn_mark = fsnotify_find_mark(&real_mount(mnt)->mnt_fsnotify_marks,
 534                                       group);
 535         if (!fsn_mark) {
 536                 mutex_unlock(&group->mark_mutex);
 537                 return -ENOENT;
 538         }
 539
 540         removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
 541                                                  &destroy_mark);
 542         if (removed & real_mount(mnt)->mnt_fsnotify_mask)
 543                 fsnotify_recalc_mask(real_mount(mnt)->mnt_fsnotify_marks);
 544         if (destroy_mark)
 545                 fsnotify_detach_mark(fsn_mark);
 546         mutex_unlock(&group->mark_mutex);
 547         if (destroy_mark)
 548                 fsnotify_free_mark(fsn_mark);
 549
 550         fsnotify_put_mark(fsn_mark);
 551         return 0;
 552 }
 553
 554 static int fanotify_remove_inode_mark(struct fsnotify_group *group,
 555                                       struct inode *inode, __u32 mask,
 556                                       unsigned int flags)
 557 {
 558         struct fsnotify_mark *fsn_mark = NULL;
 559         __u32 removed;
 560         int destroy_mark;
 561
 562         mutex_lock(&group->mark_mutex);
 563         fsn_mark = fsnotify_find_mark(&inode->i_fsnotify_marks, group);
 564         if (!fsn_mark) {
 565                 mutex_unlock(&group->mark_mutex);
 566                 return -ENOENT;
 567         }
 568
 569         removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
 570                                                  &destroy_mark);
 571         if (removed & inode->i_fsnotify_mask)
 572                 fsnotify_recalc_mask(inode->i_fsnotify_marks);
 573         if (destroy_mark)
 574                 fsnotify_detach_mark(fsn_mark);
 575         mutex_unlock(&group->mark_mutex);
 576         if (destroy_mark)
 577                 fsnotify_free_mark(fsn_mark);
 578
 579         /* matches the fsnotify_find_mark() */
 580         fsnotify_put_mark(fsn_mark);
 581
 582         return 0;
 583 }
 584
 585 static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
 586                                        __u32 mask,
 587                                        unsigned int flags)
 588 {
 589         __u32 oldmask = -1;
 590
 591         spin_lock(&fsn_mark->lock);
 592         if (!(flags & FAN_MARK_IGNORED_MASK)) {
 593                 __u32 tmask = fsn_mark->mask | mask;
 594
 595                 if (flags & FAN_MARK_ONDIR)
 596                         tmask |= FAN_ONDIR;
 597
 598                 oldmask = fsn_mark->mask;
 599                 fsn_mark->mask = tmask;
 600         } else {
 601                 __u32 tmask = fsn_mark->ignored_mask | mask;
 602                 if (flags & FAN_MARK_ONDIR)
 603                         tmask |= FAN_ONDIR;
 604
 605                 fsn_mark->ignored_mask = tmask;
 606                 if (flags & FAN_MARK_IGNORED_SURV_MODIFY)
 607                         fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
 608         }
 609         spin_unlock(&fsn_mark->lock);
 610
 611         return mask & ~oldmask;
 612 }
 613
 614 static struct fsnotify_mark *fanotify_add_new_mark(struct fsnotify_group *group,
 615                                                    struct inode *inode,
 616                                                    struct vfsmount *mnt)
 617 {
 618         struct fsnotify_mark *mark;
 619         int ret;
 620
 621         if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
 622                 return ERR_PTR(-ENOSPC);
 623
 624         mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
 625         if (!mark)
 626                 return ERR_PTR(-ENOMEM);
 627
 628         fsnotify_init_mark(mark, group);
 629         ret = fsnotify_add_mark_locked(mark, inode, mnt, 0);
 630         if (ret) {
 631                 fsnotify_put_mark(mark);
 632                 return ERR_PTR(ret);
 633         }
 634
 635         return mark;
 636 }
 637
 638
 639 static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
 640                                       struct vfsmount *mnt, __u32 mask,
 641                                       unsigned int flags)
 642 {
 643         struct fsnotify_mark *fsn_mark;
 644         __u32 added;
 645
 646         mutex_lock(&group->mark_mutex);
 647         fsn_mark = fsnotify_find_mark(&real_mount(mnt)->mnt_fsnotify_marks,
 648                                       group);
 649         if (!fsn_mark) {
 650                 fsn_mark = fanotify_add_new_mark(group, NULL, mnt);
 651                 if (IS_ERR(fsn_mark)) {
 652                         mutex_unlock(&group->mark_mutex);
 653                         return PTR_ERR(fsn_mark);
 654                 }
 655         }
 656         added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
 657         if (added & ~real_mount(mnt)->mnt_fsnotify_mask)
 658                 fsnotify_recalc_mask(real_mount(mnt)->mnt_fsnotify_marks);
 659         mutex_unlock(&group->mark_mutex);
 660
 661         fsnotify_put_mark(fsn_mark);
 662         return 0;
 663 }
 664
 665 static int fanotify_add_inode_mark(struct fsnotify_group *group,
 666                                    struct inode *inode, __u32 mask,
 667                                    unsigned int flags)
 668 {
 669         struct fsnotify_mark *fsn_mark;
 670         __u32 added;
 671
 672         pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
 673
 674         /*
 675          * If some other task has this inode open for write we should not add
 676          * an ignored mark, unless that ignored mark is supposed to survive
 677          * modification changes anyway.
 678          */
 679         if ((flags & FAN_MARK_IGNORED_MASK) &&
 680             !(flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
 681             (atomic_read(&inode->i_writecount) > 0))
 682                 return 0;
 683
 684         mutex_lock(&group->mark_mutex);
 685         fsn_mark = fsnotify_find_mark(&inode->i_fsnotify_marks, group);
 686         if (!fsn_mark) {
 687                 fsn_mark = fanotify_add_new_mark(group, inode, NULL);
 688                 if (IS_ERR(fsn_mark)) {
 689                         mutex_unlock(&group->mark_mutex);
 690                         return PTR_ERR(fsn_mark);
 691                 }
 692         }
 693         added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
 694         if (added & ~inode->i_fsnotify_mask)
 695                 fsnotify_recalc_mask(inode->i_fsnotify_marks);
 696         mutex_unlock(&group->mark_mutex);
 697
 698         fsnotify_put_mark(fsn_mark);
 699         return 0;
 700 }
 701
 702 /* fanotify syscalls */
 703 SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
 704 {
 705         struct fsnotify_group *group;
 706         int f_flags, fd;
 707         struct user_struct *user;
 708         struct fanotify_event_info *oevent;
 709
 710         pr_debug("%s: flags=%d event_f_flags=%d\n",
 711                 __func__, flags, event_f_flags);
 712
 713         if (!capable(CAP_SYS_ADMIN))
 714                 return -EPERM;
 715
 716         if (flags & ~FAN_ALL_INIT_FLAGS)
 717                 return -EINVAL;
 718
 719         if (event_f_flags & ~FANOTIFY_INIT_ALL_EVENT_F_BITS)
 720                 return -EINVAL;
 721
 722         switch (event_f_flags & O_ACCMODE) {
 723         case O_RDONLY:
 724         case O_RDWR:
 725         case O_WRONLY:
 726                 break;
 727         default:
 728                 return -EINVAL;
 729         }
 730
 731         user = get_current_user();
 732         if (atomic_read(&user->fanotify_listeners) > FANOTIFY_DEFAULT_MAX_LISTENERS) {
 733                 free_uid(user);
 734                 return -EMFILE;
 735         }
 736
 737         f_flags = O_RDWR | FMODE_NONOTIFY;
 738         if (flags & FAN_CLOEXEC)
 739                 f_flags |= O_CLOEXEC;
 740         if (flags & FAN_NONBLOCK)
 741                 f_flags |= O_NONBLOCK;
 742
 743         /* fsnotify_alloc_group takes a ref.  Dropped in fanotify_release */
 744         group = fsnotify_alloc_group(&fanotify_fsnotify_ops);
 745         if (IS_ERR(group)) {
 746                 free_uid(user);
 747                 return PTR_ERR(group);
 748         }
 749
 750         group->fanotify_data.user = user;
 751         atomic_inc(&user->fanotify_listeners);
 752
 753         oevent = fanotify_alloc_event(NULL, FS_Q_OVERFLOW, NULL);
 754         if (unlikely(!oevent)) {
 755                 fd = -ENOMEM;
 756                 goto out_destroy_group;
 757         }
 758         group->overflow_event = &oevent->fse;
 759
 760         if (force_o_largefile())
 761                 event_f_flags |= O_LARGEFILE;
 762         group->fanotify_data.f_flags = event_f_flags;
 763         init_waitqueue_head(&group->fanotify_data.access_waitq);
 764         INIT_LIST_HEAD(&group->fanotify_data.access_list);
 765         switch (flags & FAN_ALL_CLASS_BITS) {
 766         case FAN_CLASS_NOTIF:
 767                 group->priority = FS_PRIO_0;
 768                 break;
 769         case FAN_CLASS_CONTENT:
 770                 group->priority = FS_PRIO_1;
 771                 break;
 772         case FAN_CLASS_PRE_CONTENT:
 773                 group->priority = FS_PRIO_2;
 774                 break;
 775         default:
 776                 fd = -EINVAL;
 777                 goto out_destroy_group;
 778         }
 779
 780         if (flags & FAN_UNLIMITED_QUEUE) {
 781                 fd = -EPERM;
 782                 if (!capable(CAP_SYS_ADMIN))
 783                         goto out_destroy_group;
 784                 group->max_events = UINT_MAX;
 785         } else {
 786                 group->max_events = FANOTIFY_DEFAULT_MAX_EVENTS;
 787         }
 788
 789         if (flags & FAN_UNLIMITED_MARKS) {
 790                 fd = -EPERM;
 791                 if (!capable(CAP_SYS_ADMIN))
 792                         goto out_destroy_group;
 793                 group->fanotify_data.max_marks = UINT_MAX;
 794         } else {
 795                 group->fanotify_data.max_marks = FANOTIFY_DEFAULT_MAX_MARKS;
 796         }
 797
 798         fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
 799         if (fd < 0)
 800                 goto out_destroy_group;
 801
 802         return fd;
 803
 804 out_destroy_group:
 805         fsnotify_destroy_group(group);
 806         return fd;
 807 }
 808
 809 SYSCALL_DEFINE5(fanotify_mark, int, fanotify_fd, unsigned int, flags,
 810                               __u64, mask, int, dfd,
 811                               const char  __user *, pathname)
 812 {
 813         struct inode *inode = NULL;
 814         struct vfsmount *mnt = NULL;
 815         struct fsnotify_group *group;
 816         struct fd f;
 817         struct path path;
 818         u32 valid_mask = FAN_ALL_EVENTS | FAN_EVENT_ON_CHILD;
 819         int ret;
 820
 821         pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
 822                  __func__, fanotify_fd, flags, dfd, pathname, mask);
 823
 824         /* we only use the lower 32 bits as of right now. */
 825         if (mask & ((__u64)0xffffffff << 32))
 826                 return -EINVAL;
 827
 828         if (flags & ~FAN_ALL_MARK_FLAGS)
 829                 return -EINVAL;
 830         switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
 831         case FAN_MARK_ADD:              /* fallthrough */
 832         case FAN_MARK_REMOVE:
 833                 if (!mask)
 834                         return -EINVAL;
 835                 break;
 836         case FAN_MARK_FLUSH:
 837                 if (flags & ~(FAN_MARK_MOUNT | FAN_MARK_FLUSH))
 838                         return -EINVAL;
 839                 break;
 840         default:
 841                 return -EINVAL;
 842         }
 843
 844         if (mask & FAN_ONDIR) {
 845                 flags |= FAN_MARK_ONDIR;
 846                 mask &= ~FAN_ONDIR;
 847         }
 848
 849         if (IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS))
 850                 valid_mask |= FAN_ALL_PERM_EVENTS;
 851
 852         if (mask & ~valid_mask)
 853                 return -EINVAL;
 854
 855         f = fdget(fanotify_fd);
 856         if (unlikely(!f.file))
 857                 return -EBADF;
 858
 859         /* verify that this is indeed an fanotify instance */
 860         ret = -EINVAL;
 861         if (unlikely(f.file->f_op != &fanotify_fops))
 862                 goto fput_and_out;
 863         group = f.file->private_data;
 864
 865         /*
 866          * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF.  These are not
 867          * allowed to set permissions events.
 868          */
 869         ret = -EINVAL;
 870         if (mask & FAN_ALL_PERM_EVENTS &&
 871             group->priority == FS_PRIO_0)
 872                 goto fput_and_out;
 873
 874         if (flags & FAN_MARK_FLUSH) {
 875                 ret = 0;
 876                 if (flags & FAN_MARK_MOUNT)
 877                         fsnotify_clear_vfsmount_marks_by_group(group);
 878                 else
 879                         fsnotify_clear_inode_marks_by_group(group);
 880                 goto fput_and_out;
 881         }
 882
 883         ret = fanotify_find_path(dfd, pathname, &path, flags);
 884         if (ret)
 885                 goto fput_and_out;
 886
 887         /* inode held in place by reference to path; group by fget on fd */
 888         if (!(flags & FAN_MARK_MOUNT))
 889                 inode = path.dentry->d_inode;
 890         else
 891                 mnt = path.mnt;
 892
 893         /* create/update an inode mark */
 894         switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE)) {
 895         case FAN_MARK_ADD:
 896                 if (flags & FAN_MARK_MOUNT)
 897                         ret = fanotify_add_vfsmount_mark(group, mnt, mask, flags);
 898                 else
 899                         ret = fanotify_add_inode_mark(group, inode, mask, flags);
 900                 break;
 901         case FAN_MARK_REMOVE:
 902                 if (flags & FAN_MARK_MOUNT)
 903                         ret = fanotify_remove_vfsmount_mark(group, mnt, mask, flags);
 904                 else
 905                         ret = fanotify_remove_inode_mark(group, inode, mask, flags);
 906                 break;
 907         default:
 908                 ret = -EINVAL;
 909         }
 910
 911         path_put(&path);
 912 fput_and_out:
 913         fdput(f);
 914         return ret;
 915 }
 916
 917 #ifdef CONFIG_COMPAT
 918 COMPAT_SYSCALL_DEFINE6(fanotify_mark,
 919                                 int, fanotify_fd, unsigned int, flags,
 920                                 __u32, mask0, __u32, mask1, int, dfd,
 921                                 const char  __user *, pathname)
 922 {
 923         return sys_fanotify_mark(fanotify_fd, flags,
 924 #ifdef __BIG_ENDIAN
 925                                 ((__u64)mask0 << 32) | mask1,
 926 #else
 927                                 ((__u64)mask1 << 32) | mask0,
 928 #endif
 929                                  dfd, pathname);
 930 }
 931 #endif
 932
 933 /*
 934  * fanotify_user_setup - Our initialization function.  Note that we cannot return
 935  * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
 936  * must result in panic().
 937  */
 938 static int __init fanotify_user_setup(void)
 939 {
 940         fanotify_mark_cache = KMEM_CACHE(fsnotify_mark, SLAB_PANIC);
 941         fanotify_event_cachep = KMEM_CACHE(fanotify_event_info, SLAB_PANIC);
 942         if (IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS)) {
 943                 fanotify_perm_event_cachep =
 944                         KMEM_CACHE(fanotify_perm_event_info, SLAB_PANIC);
 945         }
 946
 947         return 0;
 948 }
 949 device_initcall(fanotify_user_setup);