linux/fs/notify/fanotify/fanotify_user.c
<<
>>
Prefs
   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
  25extern const struct fsnotify_ops fanotify_fsnotify_ops;
  26
  27static struct kmem_cache *fanotify_mark_cache __read_mostly;
  28static struct kmem_cache *fanotify_response_event_cache __read_mostly;
  29
  30struct 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 */
  43static 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
  61static int create_fd(struct fsnotify_group *group, struct fsnotify_event *event)
  62{
  63        int client_fd;
  64        struct dentry *dentry;
  65        struct vfsmount *mnt;
  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        dentry = dget(event->path.dentry);
  85        mnt = mntget(event->path.mnt);
  86        /* it's possible this event was an overflow event.  in that case dentry and mnt
  87         * are NULL;  That's fine, just don't call dentry open */
  88        if (dentry && mnt)
  89                new_file = dentry_open(dentry, mnt,
  90                                       group->fanotify_data.f_flags | FMODE_NONOTIFY,
  91                                       current_cred());
  92        else
  93                new_file = ERR_PTR(-EOVERFLOW);
  94        if (IS_ERR(new_file)) {
  95                /*
  96                 * we still send an event even if we can't open the file.  this
  97                 * can happen when say tasks are gone and we try to open their
  98                 * /proc files or we try to open a WRONLY file like in sysfs
  99                 * we just send the errno to userspace since there isn't much
 100                 * else we can do.
 101                 */
 102                put_unused_fd(client_fd);
 103                client_fd = PTR_ERR(new_file);
 104        } else {
 105                fd_install(client_fd, new_file);
 106        }
 107
 108        return client_fd;
 109}
 110
 111static int fill_event_metadata(struct fsnotify_group *group,
 112                                   struct fanotify_event_metadata *metadata,
 113                                   struct fsnotify_event *event)
 114{
 115        int ret = 0;
 116
 117        pr_debug("%s: group=%p metadata=%p event=%p\n", __func__,
 118                 group, metadata, event);
 119
 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);
 129                if (metadata->fd < 0)
 130                        ret = metadata->fd;
 131        }
 132
 133        return ret;
 134}
 135
 136#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 137static 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
 158static 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
 196static 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
 227static void remove_access_response(struct fsnotify_group *group,
 228                                   struct fsnotify_event *event,
 229                                   __s32 fd)
 230{
 231        struct fanotify_response_event *re;
 232
 233        if (!(event->mask & FAN_ALL_PERM_EVENTS))
 234                return;
 235
 236        re = dequeue_re(group, fd);
 237        if (!re)
 238                return;
 239
 240        BUG_ON(re->event != event);
 241
 242        kmem_cache_free(fanotify_response_event_cache, re);
 243
 244        return;
 245}
 246#else
 247static int prepare_for_access_response(struct fsnotify_group *group,
 248                                       struct fsnotify_event *event,
 249                                       __s32 fd)
 250{
 251        return 0;
 252}
 253
 254static void remove_access_response(struct fsnotify_group *group,
 255                                   struct fsnotify_event *event,
 256                                   __s32 fd)
 257{
 258        return;
 259}
 260#endif
 261
 262static ssize_t copy_event_to_user(struct fsnotify_group *group,
 263                                  struct fsnotify_event *event,
 264                                  char __user *buf)
 265{
 266        struct fanotify_event_metadata fanotify_event_metadata;
 267        int fd, ret;
 268
 269        pr_debug("%s: group=%p event=%p\n", __func__, group, event);
 270
 271        ret = fill_event_metadata(group, &fanotify_event_metadata, event);
 272        if (ret < 0)
 273                goto out;
 274
 275        fd = fanotify_event_metadata.fd;
 276        ret = prepare_for_access_response(group, event, fd);
 277        if (ret)
 278                goto out_close_fd;
 279
 280        ret = -EFAULT;
 281        if (copy_to_user(buf, &fanotify_event_metadata,
 282                         fanotify_event_metadata.event_len))
 283                goto out_kill_access_response;
 284
 285        return fanotify_event_metadata.event_len;
 286
 287out_kill_access_response:
 288        remove_access_response(group, event, fd);
 289out_close_fd:
 290        if (fd != FAN_NOFD)
 291                sys_close(fd);
 292out:
 293#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 294        if (event->mask & FAN_ALL_PERM_EVENTS) {
 295                event->response = FAN_DENY;
 296                wake_up(&group->fanotify_data.access_waitq);
 297        }
 298#endif
 299        return ret;
 300}
 301
 302/* intofiy userspace file descriptor functions */
 303static unsigned int fanotify_poll(struct file *file, poll_table *wait)
 304{
 305        struct fsnotify_group *group = file->private_data;
 306        int ret = 0;
 307
 308        poll_wait(file, &group->notification_waitq, wait);
 309        mutex_lock(&group->notification_mutex);
 310        if (!fsnotify_notify_queue_is_empty(group))
 311                ret = POLLIN | POLLRDNORM;
 312        mutex_unlock(&group->notification_mutex);
 313
 314        return ret;
 315}
 316
 317static ssize_t fanotify_read(struct file *file, char __user *buf,
 318                             size_t count, loff_t *pos)
 319{
 320        struct fsnotify_group *group;
 321        struct fsnotify_event *kevent;
 322        char __user *start;
 323        int ret;
 324        DEFINE_WAIT(wait);
 325
 326        start = buf;
 327        group = file->private_data;
 328
 329        pr_debug("%s: group=%p\n", __func__, group);
 330
 331        while (1) {
 332                prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
 333
 334                mutex_lock(&group->notification_mutex);
 335                kevent = get_one_event(group, count);
 336                mutex_unlock(&group->notification_mutex);
 337
 338                if (kevent) {
 339                        ret = PTR_ERR(kevent);
 340                        if (IS_ERR(kevent))
 341                                break;
 342                        ret = copy_event_to_user(group, kevent, buf);
 343                        fsnotify_put_event(kevent);
 344                        if (ret < 0)
 345                                break;
 346                        buf += ret;
 347                        count -= ret;
 348                        continue;
 349                }
 350
 351                ret = -EAGAIN;
 352                if (file->f_flags & O_NONBLOCK)
 353                        break;
 354                ret = -ERESTARTSYS;
 355                if (signal_pending(current))
 356                        break;
 357
 358                if (start != buf)
 359                        break;
 360
 361                schedule();
 362        }
 363
 364        finish_wait(&group->notification_waitq, &wait);
 365        if (start != buf && ret != -EFAULT)
 366                ret = buf - start;
 367        return ret;
 368}
 369
 370static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
 371{
 372#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 373        struct fanotify_response response = { .fd = -1, .response = -1 };
 374        struct fsnotify_group *group;
 375        int ret;
 376
 377        group = file->private_data;
 378
 379        if (count > sizeof(response))
 380                count = sizeof(response);
 381
 382        pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
 383
 384        if (copy_from_user(&response, buf, count))
 385                return -EFAULT;
 386
 387        ret = process_access_response(group, &response);
 388        if (ret < 0)
 389                count = ret;
 390
 391        return count;
 392#else
 393        return -EINVAL;
 394#endif
 395}
 396
 397static int fanotify_release(struct inode *ignored, struct file *file)
 398{
 399        struct fsnotify_group *group = file->private_data;
 400
 401#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 402        struct fanotify_response_event *re, *lre;
 403
 404        mutex_lock(&group->fanotify_data.access_mutex);
 405
 406        atomic_inc(&group->fanotify_data.bypass_perm);
 407
 408        list_for_each_entry_safe(re, lre, &group->fanotify_data.access_list, list) {
 409                pr_debug("%s: found group=%p re=%p event=%p\n", __func__, group,
 410                         re, re->event);
 411
 412                list_del_init(&re->list);
 413                re->event->response = FAN_ALLOW;
 414
 415                kmem_cache_free(fanotify_response_event_cache, re);
 416        }
 417        mutex_unlock(&group->fanotify_data.access_mutex);
 418
 419        wake_up(&group->fanotify_data.access_waitq);
 420#endif
 421        /* matches the fanotify_init->fsnotify_alloc_group */
 422        fsnotify_put_group(group);
 423
 424        return 0;
 425}
 426
 427static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 428{
 429        struct fsnotify_group *group;
 430        struct fsnotify_event_holder *holder;
 431        void __user *p;
 432        int ret = -ENOTTY;
 433        size_t send_len = 0;
 434
 435        group = file->private_data;
 436
 437        p = (void __user *) arg;
 438
 439        switch (cmd) {
 440        case FIONREAD:
 441                mutex_lock(&group->notification_mutex);
 442                list_for_each_entry(holder, &group->notification_list, event_list)
 443                        send_len += FAN_EVENT_METADATA_LEN;
 444                mutex_unlock(&group->notification_mutex);
 445                ret = put_user(send_len, (int __user *) p);
 446                break;
 447        }
 448
 449        return ret;
 450}
 451
 452static const struct file_operations fanotify_fops = {
 453        .poll           = fanotify_poll,
 454        .read           = fanotify_read,
 455        .write          = fanotify_write,
 456        .fasync         = NULL,
 457        .release        = fanotify_release,
 458        .unlocked_ioctl = fanotify_ioctl,
 459        .compat_ioctl   = fanotify_ioctl,
 460        .llseek         = noop_llseek,
 461};
 462
 463static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
 464{
 465        kmem_cache_free(fanotify_mark_cache, fsn_mark);
 466}
 467
 468static int fanotify_find_path(int dfd, const char __user *filename,
 469                              struct path *path, unsigned int flags)
 470{
 471        int ret;
 472
 473        pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
 474                 dfd, filename, flags);
 475
 476        if (filename == NULL) {
 477                struct file *file;
 478                int fput_needed;
 479
 480                ret = -EBADF;
 481                file = fget_light(dfd, &fput_needed);
 482                if (!file)
 483                        goto out;
 484
 485                ret = -ENOTDIR;
 486                if ((flags & FAN_MARK_ONLYDIR) &&
 487                    !(S_ISDIR(file->f_path.dentry->d_inode->i_mode))) {
 488                        fput_light(file, fput_needed);
 489                        goto out;
 490                }
 491
 492                *path = file->f_path;
 493                path_get(path);
 494                fput_light(file, fput_needed);
 495        } else {
 496                unsigned int lookup_flags = 0;
 497
 498                if (!(flags & FAN_MARK_DONT_FOLLOW))
 499                        lookup_flags |= LOOKUP_FOLLOW;
 500                if (flags & FAN_MARK_ONLYDIR)
 501                        lookup_flags |= LOOKUP_DIRECTORY;
 502
 503                ret = user_path_at(dfd, filename, lookup_flags, path);
 504                if (ret)
 505                        goto out;
 506        }
 507
 508        /* you can only watch an inode if you have read permissions on it */
 509        ret = inode_permission(path->dentry->d_inode, MAY_READ);
 510        if (ret)
 511                path_put(path);
 512out:
 513        return ret;
 514}
 515
 516static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
 517                                            __u32 mask,
 518                                            unsigned int flags)
 519{
 520        __u32 oldmask;
 521
 522        spin_lock(&fsn_mark->lock);
 523        if (!(flags & FAN_MARK_IGNORED_MASK)) {
 524                oldmask = fsn_mark->mask;
 525                fsnotify_set_mark_mask_locked(fsn_mark, (oldmask & ~mask));
 526        } else {
 527                oldmask = fsn_mark->ignored_mask;
 528                fsnotify_set_mark_ignored_mask_locked(fsn_mark, (oldmask & ~mask));
 529        }
 530        spin_unlock(&fsn_mark->lock);
 531
 532        if (!(oldmask & ~mask))
 533                fsnotify_destroy_mark(fsn_mark);
 534
 535        return mask & oldmask;
 536}
 537
 538static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
 539                                         struct vfsmount *mnt, __u32 mask,
 540                                         unsigned int flags)
 541{
 542        struct fsnotify_mark *fsn_mark = NULL;
 543        __u32 removed;
 544
 545        fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
 546        if (!fsn_mark)
 547                return -ENOENT;
 548
 549        removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags);
 550        fsnotify_put_mark(fsn_mark);
 551        if (removed & real_mount(mnt)->mnt_fsnotify_mask)
 552                fsnotify_recalc_vfsmount_mask(mnt);
 553
 554        return 0;
 555}
 556
 557static int fanotify_remove_inode_mark(struct fsnotify_group *group,
 558                                      struct inode *inode, __u32 mask,
 559                                      unsigned int flags)
 560{
 561        struct fsnotify_mark *fsn_mark = NULL;
 562        __u32 removed;
 563
 564        fsn_mark = fsnotify_find_inode_mark(group, inode);
 565        if (!fsn_mark)
 566                return -ENOENT;
 567
 568        removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags);
 569        /* matches the fsnotify_find_inode_mark() */
 570        fsnotify_put_mark(fsn_mark);
 571        if (removed & inode->i_fsnotify_mask)
 572                fsnotify_recalc_inode_mask(inode);
 573
 574        return 0;
 575}
 576
 577static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
 578                                       __u32 mask,
 579                                       unsigned int flags)
 580{
 581        __u32 oldmask = -1;
 582
 583        spin_lock(&fsn_mark->lock);
 584        if (!(flags & FAN_MARK_IGNORED_MASK)) {
 585                oldmask = fsn_mark->mask;
 586                fsnotify_set_mark_mask_locked(fsn_mark, (oldmask | mask));
 587        } else {
 588                __u32 tmask = fsn_mark->ignored_mask | mask;
 589                fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
 590                if (flags & FAN_MARK_IGNORED_SURV_MODIFY)
 591                        fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
 592        }
 593
 594        if (!(flags & FAN_MARK_ONDIR)) {
 595                __u32 tmask = fsn_mark->ignored_mask | FAN_ONDIR;
 596                fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
 597        }
 598
 599        spin_unlock(&fsn_mark->lock);
 600
 601        return mask & ~oldmask;
 602}
 603
 604static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
 605                                      struct vfsmount *mnt, __u32 mask,
 606                                      unsigned int flags)
 607{
 608        struct fsnotify_mark *fsn_mark;
 609        __u32 added;
 610        int ret = 0;
 611
 612        fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
 613        if (!fsn_mark) {
 614                if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
 615                        return -ENOSPC;
 616
 617                fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
 618                if (!fsn_mark)
 619                        return -ENOMEM;
 620
 621                fsnotify_init_mark(fsn_mark, fanotify_free_mark);
 622                ret = fsnotify_add_mark(fsn_mark, group, NULL, mnt, 0);
 623                if (ret)
 624                        goto err;
 625        }
 626        added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
 627
 628        if (added & ~real_mount(mnt)->mnt_fsnotify_mask)
 629                fsnotify_recalc_vfsmount_mask(mnt);
 630err:
 631        fsnotify_put_mark(fsn_mark);
 632        return ret;
 633}
 634
 635static int fanotify_add_inode_mark(struct fsnotify_group *group,
 636                                   struct inode *inode, __u32 mask,
 637                                   unsigned int flags)
 638{
 639        struct fsnotify_mark *fsn_mark;
 640        __u32 added;
 641        int ret = 0;
 642
 643        pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
 644
 645        /*
 646         * If some other task has this inode open for write we should not add
 647         * an ignored mark, unless that ignored mark is supposed to survive
 648         * modification changes anyway.
 649         */
 650        if ((flags & FAN_MARK_IGNORED_MASK) &&
 651            !(flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
 652            (atomic_read(&inode->i_writecount) > 0))
 653                return 0;
 654
 655        fsn_mark = fsnotify_find_inode_mark(group, inode);
 656        if (!fsn_mark) {
 657                if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
 658                        return -ENOSPC;
 659
 660                fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
 661                if (!fsn_mark)
 662                        return -ENOMEM;
 663
 664                fsnotify_init_mark(fsn_mark, fanotify_free_mark);
 665                ret = fsnotify_add_mark(fsn_mark, group, inode, NULL, 0);
 666                if (ret)
 667                        goto err;
 668        }
 669        added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
 670
 671        if (added & ~inode->i_fsnotify_mask)
 672                fsnotify_recalc_inode_mask(inode);
 673err:
 674        fsnotify_put_mark(fsn_mark);
 675        return ret;
 676}
 677
 678/* fanotify syscalls */
 679SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
 680{
 681        struct fsnotify_group *group;
 682        int f_flags, fd;
 683        struct user_struct *user;
 684
 685        pr_debug("%s: flags=%d event_f_flags=%d\n",
 686                __func__, flags, event_f_flags);
 687
 688        if (!capable(CAP_SYS_ADMIN))
 689                return -EPERM;
 690
 691        if (flags & ~FAN_ALL_INIT_FLAGS)
 692                return -EINVAL;
 693
 694        user = get_current_user();
 695        if (atomic_read(&user->fanotify_listeners) > FANOTIFY_DEFAULT_MAX_LISTENERS) {
 696                free_uid(user);
 697                return -EMFILE;
 698        }
 699
 700        f_flags = O_RDWR | FMODE_NONOTIFY;
 701        if (flags & FAN_CLOEXEC)
 702                f_flags |= O_CLOEXEC;
 703        if (flags & FAN_NONBLOCK)
 704                f_flags |= O_NONBLOCK;
 705
 706        /* fsnotify_alloc_group takes a ref.  Dropped in fanotify_release */
 707        group = fsnotify_alloc_group(&fanotify_fsnotify_ops);
 708        if (IS_ERR(group)) {
 709                free_uid(user);
 710                return PTR_ERR(group);
 711        }
 712
 713        group->fanotify_data.user = user;
 714        atomic_inc(&user->fanotify_listeners);
 715
 716        group->fanotify_data.f_flags = event_f_flags;
 717#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 718        mutex_init(&group->fanotify_data.access_mutex);
 719        init_waitqueue_head(&group->fanotify_data.access_waitq);
 720        INIT_LIST_HEAD(&group->fanotify_data.access_list);
 721        atomic_set(&group->fanotify_data.bypass_perm, 0);
 722#endif
 723        switch (flags & FAN_ALL_CLASS_BITS) {
 724        case FAN_CLASS_NOTIF:
 725                group->priority = FS_PRIO_0;
 726                break;
 727        case FAN_CLASS_CONTENT:
 728                group->priority = FS_PRIO_1;
 729                break;
 730        case FAN_CLASS_PRE_CONTENT:
 731                group->priority = FS_PRIO_2;
 732                break;
 733        default:
 734                fd = -EINVAL;
 735                goto out_put_group;
 736        }
 737
 738        if (flags & FAN_UNLIMITED_QUEUE) {
 739                fd = -EPERM;
 740                if (!capable(CAP_SYS_ADMIN))
 741                        goto out_put_group;
 742                group->max_events = UINT_MAX;
 743        } else {
 744                group->max_events = FANOTIFY_DEFAULT_MAX_EVENTS;
 745        }
 746
 747        if (flags & FAN_UNLIMITED_MARKS) {
 748                fd = -EPERM;
 749                if (!capable(CAP_SYS_ADMIN))
 750                        goto out_put_group;
 751                group->fanotify_data.max_marks = UINT_MAX;
 752        } else {
 753                group->fanotify_data.max_marks = FANOTIFY_DEFAULT_MAX_MARKS;
 754        }
 755
 756        fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
 757        if (fd < 0)
 758                goto out_put_group;
 759
 760        return fd;
 761
 762out_put_group:
 763        fsnotify_put_group(group);
 764        return fd;
 765}
 766
 767SYSCALL_DEFINE(fanotify_mark)(int fanotify_fd, unsigned int flags,
 768                              __u64 mask, int dfd,
 769                              const char  __user * pathname)
 770{
 771        struct inode *inode = NULL;
 772        struct vfsmount *mnt = NULL;
 773        struct fsnotify_group *group;
 774        struct file *filp;
 775        struct path path;
 776        int ret, fput_needed;
 777
 778        pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
 779                 __func__, fanotify_fd, flags, dfd, pathname, mask);
 780
 781        /* we only use the lower 32 bits as of right now. */
 782        if (mask & ((__u64)0xffffffff << 32))
 783                return -EINVAL;
 784
 785        if (flags & ~FAN_ALL_MARK_FLAGS)
 786                return -EINVAL;
 787        switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
 788        case FAN_MARK_ADD:              /* fallthrough */
 789        case FAN_MARK_REMOVE:
 790                if (!mask)
 791                        return -EINVAL;
 792        case FAN_MARK_FLUSH:
 793                break;
 794        default:
 795                return -EINVAL;
 796        }
 797
 798        if (mask & FAN_ONDIR) {
 799                flags |= FAN_MARK_ONDIR;
 800                mask &= ~FAN_ONDIR;
 801        }
 802
 803#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 804        if (mask & ~(FAN_ALL_EVENTS | FAN_ALL_PERM_EVENTS | FAN_EVENT_ON_CHILD))
 805#else
 806        if (mask & ~(FAN_ALL_EVENTS | FAN_EVENT_ON_CHILD))
 807#endif
 808                return -EINVAL;
 809
 810        filp = fget_light(fanotify_fd, &fput_needed);
 811        if (unlikely(!filp))
 812                return -EBADF;
 813
 814        /* verify that this is indeed an fanotify instance */
 815        ret = -EINVAL;
 816        if (unlikely(filp->f_op != &fanotify_fops))
 817                goto fput_and_out;
 818        group = filp->private_data;
 819
 820        /*
 821         * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF.  These are not
 822         * allowed to set permissions events.
 823         */
 824        ret = -EINVAL;
 825        if (mask & FAN_ALL_PERM_EVENTS &&
 826            group->priority == FS_PRIO_0)
 827                goto fput_and_out;
 828
 829        ret = fanotify_find_path(dfd, pathname, &path, flags);
 830        if (ret)
 831                goto fput_and_out;
 832
 833        /* inode held in place by reference to path; group by fget on fd */
 834        if (!(flags & FAN_MARK_MOUNT))
 835                inode = path.dentry->d_inode;
 836        else
 837                mnt = path.mnt;
 838
 839        /* create/update an inode mark */
 840        switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
 841        case FAN_MARK_ADD:
 842                if (flags & FAN_MARK_MOUNT)
 843                        ret = fanotify_add_vfsmount_mark(group, mnt, mask, flags);
 844                else
 845                        ret = fanotify_add_inode_mark(group, inode, mask, flags);
 846                break;
 847        case FAN_MARK_REMOVE:
 848                if (flags & FAN_MARK_MOUNT)
 849                        ret = fanotify_remove_vfsmount_mark(group, mnt, mask, flags);
 850                else
 851                        ret = fanotify_remove_inode_mark(group, inode, mask, flags);
 852                break;
 853        case FAN_MARK_FLUSH:
 854                if (flags & FAN_MARK_MOUNT)
 855                        fsnotify_clear_vfsmount_marks_by_group(group);
 856                else
 857                        fsnotify_clear_inode_marks_by_group(group);
 858                break;
 859        default:
 860                ret = -EINVAL;
 861        }
 862
 863        path_put(&path);
 864fput_and_out:
 865        fput_light(filp, fput_needed);
 866        return ret;
 867}
 868
 869#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 870asmlinkage long SyS_fanotify_mark(long fanotify_fd, long flags, __u64 mask,
 871                                  long dfd, long pathname)
 872{
 873        return SYSC_fanotify_mark((int) fanotify_fd, (unsigned int) flags,
 874                                  mask, (int) dfd,
 875                                  (const char  __user *) pathname);
 876}
 877SYSCALL_ALIAS(sys_fanotify_mark, SyS_fanotify_mark);
 878#endif
 879
 880/*
 881 * fanotify_user_setup - Our initialization function.  Note that we cannot return
 882 * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
 883 * must result in panic().
 884 */
 885static int __init fanotify_user_setup(void)
 886{
 887        fanotify_mark_cache = KMEM_CACHE(fsnotify_mark, SLAB_PANIC);
 888        fanotify_response_event_cache = KMEM_CACHE(fanotify_response_event,
 889                                                   SLAB_PANIC);
 890
 891        return 0;
 892}
 893device_initcall(fanotify_user_setup);
 894