linux/fs/notify/inotify/inotify_user.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * fs/inotify_user.c - inotify support for userspace
   4 *
   5 * Authors:
   6 *      John McCutchan  <ttb@tentacle.dhs.org>
   7 *      Robert Love     <rml@novell.com>
   8 *
   9 * Copyright (C) 2005 John McCutchan
  10 * Copyright 2006 Hewlett-Packard Development Company, L.P.
  11 *
  12 * Copyright (C) 2009 Eric Paris <Red Hat Inc>
  13 * inotify was largely rewriten to make use of the fsnotify infrastructure
  14 */
  15
  16#include <linux/file.h>
  17#include <linux/fs.h> /* struct inode */
  18#include <linux/fsnotify_backend.h>
  19#include <linux/idr.h>
  20#include <linux/init.h> /* fs_initcall */
  21#include <linux/inotify.h>
  22#include <linux/kernel.h> /* roundup() */
  23#include <linux/namei.h> /* LOOKUP_FOLLOW */
  24#include <linux/sched/signal.h>
  25#include <linux/slab.h> /* struct kmem_cache */
  26#include <linux/syscalls.h>
  27#include <linux/types.h>
  28#include <linux/anon_inodes.h>
  29#include <linux/uaccess.h>
  30#include <linux/poll.h>
  31#include <linux/wait.h>
  32#include <linux/memcontrol.h>
  33#include <linux/security.h>
  34
  35#include "inotify.h"
  36#include "../fdinfo.h"
  37
  38#include <asm/ioctls.h>
  39
  40/* configurable via /proc/sys/fs/inotify/ */
  41static int inotify_max_queued_events __read_mostly;
  42
  43struct kmem_cache *inotify_inode_mark_cachep __read_mostly;
  44
  45#ifdef CONFIG_SYSCTL
  46
  47#include <linux/sysctl.h>
  48
  49struct ctl_table inotify_table[] = {
  50        {
  51                .procname       = "max_user_instances",
  52                .data           = &init_user_ns.ucount_max[UCOUNT_INOTIFY_INSTANCES],
  53                .maxlen         = sizeof(int),
  54                .mode           = 0644,
  55                .proc_handler   = proc_dointvec_minmax,
  56                .extra1         = SYSCTL_ZERO,
  57        },
  58        {
  59                .procname       = "max_user_watches",
  60                .data           = &init_user_ns.ucount_max[UCOUNT_INOTIFY_WATCHES],
  61                .maxlen         = sizeof(int),
  62                .mode           = 0644,
  63                .proc_handler   = proc_dointvec_minmax,
  64                .extra1         = SYSCTL_ZERO,
  65        },
  66        {
  67                .procname       = "max_queued_events",
  68                .data           = &inotify_max_queued_events,
  69                .maxlen         = sizeof(int),
  70                .mode           = 0644,
  71                .proc_handler   = proc_dointvec_minmax,
  72                .extra1         = SYSCTL_ZERO
  73        },
  74        { }
  75};
  76#endif /* CONFIG_SYSCTL */
  77
  78static inline __u32 inotify_arg_to_mask(struct inode *inode, u32 arg)
  79{
  80        __u32 mask;
  81
  82        /*
  83         * Everything should accept their own ignored and should receive events
  84         * when the inode is unmounted.  All directories care about children.
  85         */
  86        mask = (FS_IN_IGNORED | FS_UNMOUNT);
  87        if (S_ISDIR(inode->i_mode))
  88                mask |= FS_EVENT_ON_CHILD;
  89
  90        /* mask off the flags used to open the fd */
  91        mask |= (arg & (IN_ALL_EVENTS | IN_ONESHOT | IN_EXCL_UNLINK));
  92
  93        return mask;
  94}
  95
  96static inline u32 inotify_mask_to_arg(__u32 mask)
  97{
  98        return mask & (IN_ALL_EVENTS | IN_ISDIR | IN_UNMOUNT | IN_IGNORED |
  99                       IN_Q_OVERFLOW);
 100}
 101
 102/* intofiy userspace file descriptor functions */
 103static __poll_t inotify_poll(struct file *file, poll_table *wait)
 104{
 105        struct fsnotify_group *group = file->private_data;
 106        __poll_t ret = 0;
 107
 108        poll_wait(file, &group->notification_waitq, wait);
 109        spin_lock(&group->notification_lock);
 110        if (!fsnotify_notify_queue_is_empty(group))
 111                ret = EPOLLIN | EPOLLRDNORM;
 112        spin_unlock(&group->notification_lock);
 113
 114        return ret;
 115}
 116
 117static int round_event_name_len(struct fsnotify_event *fsn_event)
 118{
 119        struct inotify_event_info *event;
 120
 121        event = INOTIFY_E(fsn_event);
 122        if (!event->name_len)
 123                return 0;
 124        return roundup(event->name_len + 1, sizeof(struct inotify_event));
 125}
 126
 127/*
 128 * Get an inotify_kernel_event if one exists and is small
 129 * enough to fit in "count". Return an error pointer if
 130 * not large enough.
 131 *
 132 * Called with the group->notification_lock held.
 133 */
 134static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
 135                                            size_t count)
 136{
 137        size_t event_size = sizeof(struct inotify_event);
 138        struct fsnotify_event *event;
 139
 140        if (fsnotify_notify_queue_is_empty(group))
 141                return NULL;
 142
 143        event = fsnotify_peek_first_event(group);
 144
 145        pr_debug("%s: group=%p event=%p\n", __func__, group, event);
 146
 147        event_size += round_event_name_len(event);
 148        if (event_size > count)
 149                return ERR_PTR(-EINVAL);
 150
 151        /* held the notification_lock the whole time, so this is the
 152         * same event we peeked above */
 153        fsnotify_remove_first_event(group);
 154
 155        return event;
 156}
 157
 158/*
 159 * Copy an event to user space, returning how much we copied.
 160 *
 161 * We already checked that the event size is smaller than the
 162 * buffer we had in "get_one_event()" above.
 163 */
 164static ssize_t copy_event_to_user(struct fsnotify_group *group,
 165                                  struct fsnotify_event *fsn_event,
 166                                  char __user *buf)
 167{
 168        struct inotify_event inotify_event;
 169        struct inotify_event_info *event;
 170        size_t event_size = sizeof(struct inotify_event);
 171        size_t name_len;
 172        size_t pad_name_len;
 173
 174        pr_debug("%s: group=%p event=%p\n", __func__, group, fsn_event);
 175
 176        event = INOTIFY_E(fsn_event);
 177        name_len = event->name_len;
 178        /*
 179         * round up name length so it is a multiple of event_size
 180         * plus an extra byte for the terminating '\0'.
 181         */
 182        pad_name_len = round_event_name_len(fsn_event);
 183        inotify_event.len = pad_name_len;
 184        inotify_event.mask = inotify_mask_to_arg(event->mask);
 185        inotify_event.wd = event->wd;
 186        inotify_event.cookie = event->sync_cookie;
 187
 188        /* send the main event */
 189        if (copy_to_user(buf, &inotify_event, event_size))
 190                return -EFAULT;
 191
 192        buf += event_size;
 193
 194        /*
 195         * fsnotify only stores the pathname, so here we have to send the pathname
 196         * and then pad that pathname out to a multiple of sizeof(inotify_event)
 197         * with zeros.
 198         */
 199        if (pad_name_len) {
 200                /* copy the path name */
 201                if (copy_to_user(buf, event->name, name_len))
 202                        return -EFAULT;
 203                buf += name_len;
 204
 205                /* fill userspace with 0's */
 206                if (clear_user(buf, pad_name_len - name_len))
 207                        return -EFAULT;
 208                event_size += pad_name_len;
 209        }
 210
 211        return event_size;
 212}
 213
 214static ssize_t inotify_read(struct file *file, char __user *buf,
 215                            size_t count, loff_t *pos)
 216{
 217        struct fsnotify_group *group;
 218        struct fsnotify_event *kevent;
 219        char __user *start;
 220        int ret;
 221        DEFINE_WAIT_FUNC(wait, woken_wake_function);
 222
 223        start = buf;
 224        group = file->private_data;
 225
 226        add_wait_queue(&group->notification_waitq, &wait);
 227        while (1) {
 228                spin_lock(&group->notification_lock);
 229                kevent = get_one_event(group, count);
 230                spin_unlock(&group->notification_lock);
 231
 232                pr_debug("%s: group=%p kevent=%p\n", __func__, group, kevent);
 233
 234                if (kevent) {
 235                        ret = PTR_ERR(kevent);
 236                        if (IS_ERR(kevent))
 237                                break;
 238                        ret = copy_event_to_user(group, kevent, buf);
 239                        fsnotify_destroy_event(group, kevent);
 240                        if (ret < 0)
 241                                break;
 242                        buf += ret;
 243                        count -= ret;
 244                        continue;
 245                }
 246
 247                ret = -EAGAIN;
 248                if (file->f_flags & O_NONBLOCK)
 249                        break;
 250                ret = -ERESTARTSYS;
 251                if (signal_pending(current))
 252                        break;
 253
 254                if (start != buf)
 255                        break;
 256
 257                wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
 258        }
 259        remove_wait_queue(&group->notification_waitq, &wait);
 260
 261        if (start != buf && ret != -EFAULT)
 262                ret = buf - start;
 263        return ret;
 264}
 265
 266static int inotify_release(struct inode *ignored, struct file *file)
 267{
 268        struct fsnotify_group *group = file->private_data;
 269
 270        pr_debug("%s: group=%p\n", __func__, group);
 271
 272        /* free this group, matching get was inotify_init->fsnotify_obtain_group */
 273        fsnotify_destroy_group(group);
 274
 275        return 0;
 276}
 277
 278static long inotify_ioctl(struct file *file, unsigned int cmd,
 279                          unsigned long arg)
 280{
 281        struct fsnotify_group *group;
 282        struct fsnotify_event *fsn_event;
 283        void __user *p;
 284        int ret = -ENOTTY;
 285        size_t send_len = 0;
 286
 287        group = file->private_data;
 288        p = (void __user *) arg;
 289
 290        pr_debug("%s: group=%p cmd=%u\n", __func__, group, cmd);
 291
 292        switch (cmd) {
 293        case FIONREAD:
 294                spin_lock(&group->notification_lock);
 295                list_for_each_entry(fsn_event, &group->notification_list,
 296                                    list) {
 297                        send_len += sizeof(struct inotify_event);
 298                        send_len += round_event_name_len(fsn_event);
 299                }
 300                spin_unlock(&group->notification_lock);
 301                ret = put_user(send_len, (int __user *) p);
 302                break;
 303#ifdef CONFIG_CHECKPOINT_RESTORE
 304        case INOTIFY_IOC_SETNEXTWD:
 305                ret = -EINVAL;
 306                if (arg >= 1 && arg <= INT_MAX) {
 307                        struct inotify_group_private_data *data;
 308
 309                        data = &group->inotify_data;
 310                        spin_lock(&data->idr_lock);
 311                        idr_set_cursor(&data->idr, (unsigned int)arg);
 312                        spin_unlock(&data->idr_lock);
 313                        ret = 0;
 314                }
 315                break;
 316#endif /* CONFIG_CHECKPOINT_RESTORE */
 317        }
 318
 319        return ret;
 320}
 321
 322static const struct file_operations inotify_fops = {
 323        .show_fdinfo    = inotify_show_fdinfo,
 324        .poll           = inotify_poll,
 325        .read           = inotify_read,
 326        .fasync         = fsnotify_fasync,
 327        .release        = inotify_release,
 328        .unlocked_ioctl = inotify_ioctl,
 329        .compat_ioctl   = inotify_ioctl,
 330        .llseek         = noop_llseek,
 331};
 332
 333
 334/*
 335 * find_inode - resolve a user-given path to a specific inode
 336 */
 337static int inotify_find_inode(const char __user *dirname, struct path *path,
 338                                                unsigned int flags, __u64 mask)
 339{
 340        int error;
 341
 342        error = user_path_at(AT_FDCWD, dirname, flags, path);
 343        if (error)
 344                return error;
 345        /* you can only watch an inode if you have read permissions on it */
 346        error = inode_permission(path->dentry->d_inode, MAY_READ);
 347        if (error) {
 348                path_put(path);
 349                return error;
 350        }
 351        error = security_path_notify(path, mask,
 352                                FSNOTIFY_OBJ_TYPE_INODE);
 353        if (error)
 354                path_put(path);
 355
 356        return error;
 357}
 358
 359static int inotify_add_to_idr(struct idr *idr, spinlock_t *idr_lock,
 360                              struct inotify_inode_mark *i_mark)
 361{
 362        int ret;
 363
 364        idr_preload(GFP_KERNEL);
 365        spin_lock(idr_lock);
 366
 367        ret = idr_alloc_cyclic(idr, i_mark, 1, 0, GFP_NOWAIT);
 368        if (ret >= 0) {
 369                /* we added the mark to the idr, take a reference */
 370                i_mark->wd = ret;
 371                fsnotify_get_mark(&i_mark->fsn_mark);
 372        }
 373
 374        spin_unlock(idr_lock);
 375        idr_preload_end();
 376        return ret < 0 ? ret : 0;
 377}
 378
 379static struct inotify_inode_mark *inotify_idr_find_locked(struct fsnotify_group *group,
 380                                                                int wd)
 381{
 382        struct idr *idr = &group->inotify_data.idr;
 383        spinlock_t *idr_lock = &group->inotify_data.idr_lock;
 384        struct inotify_inode_mark *i_mark;
 385
 386        assert_spin_locked(idr_lock);
 387
 388        i_mark = idr_find(idr, wd);
 389        if (i_mark) {
 390                struct fsnotify_mark *fsn_mark = &i_mark->fsn_mark;
 391
 392                fsnotify_get_mark(fsn_mark);
 393                /* One ref for being in the idr, one ref we just took */
 394                BUG_ON(refcount_read(&fsn_mark->refcnt) < 2);
 395        }
 396
 397        return i_mark;
 398}
 399
 400static struct inotify_inode_mark *inotify_idr_find(struct fsnotify_group *group,
 401                                                         int wd)
 402{
 403        struct inotify_inode_mark *i_mark;
 404        spinlock_t *idr_lock = &group->inotify_data.idr_lock;
 405
 406        spin_lock(idr_lock);
 407        i_mark = inotify_idr_find_locked(group, wd);
 408        spin_unlock(idr_lock);
 409
 410        return i_mark;
 411}
 412
 413/*
 414 * Remove the mark from the idr (if present) and drop the reference
 415 * on the mark because it was in the idr.
 416 */
 417static void inotify_remove_from_idr(struct fsnotify_group *group,
 418                                    struct inotify_inode_mark *i_mark)
 419{
 420        struct idr *idr = &group->inotify_data.idr;
 421        spinlock_t *idr_lock = &group->inotify_data.idr_lock;
 422        struct inotify_inode_mark *found_i_mark = NULL;
 423        int wd;
 424
 425        spin_lock(idr_lock);
 426        wd = i_mark->wd;
 427
 428        /*
 429         * does this i_mark think it is in the idr?  we shouldn't get called
 430         * if it wasn't....
 431         */
 432        if (wd == -1) {
 433                WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p\n",
 434                        __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group);
 435                goto out;
 436        }
 437
 438        /* Lets look in the idr to see if we find it */
 439        found_i_mark = inotify_idr_find_locked(group, wd);
 440        if (unlikely(!found_i_mark)) {
 441                WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p\n",
 442                        __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group);
 443                goto out;
 444        }
 445
 446        /*
 447         * We found an mark in the idr at the right wd, but it's
 448         * not the mark we were told to remove.  eparis seriously
 449         * fucked up somewhere.
 450         */
 451        if (unlikely(found_i_mark != i_mark)) {
 452                WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p "
 453                        "found_i_mark=%p found_i_mark->wd=%d "
 454                        "found_i_mark->group=%p\n", __func__, i_mark,
 455                        i_mark->wd, i_mark->fsn_mark.group, found_i_mark,
 456                        found_i_mark->wd, found_i_mark->fsn_mark.group);
 457                goto out;
 458        }
 459
 460        /*
 461         * One ref for being in the idr
 462         * one ref grabbed by inotify_idr_find
 463         */
 464        if (unlikely(refcount_read(&i_mark->fsn_mark.refcnt) < 2)) {
 465                printk(KERN_ERR "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p\n",
 466                         __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group);
 467                /* we can't really recover with bad ref cnting.. */
 468                BUG();
 469        }
 470
 471        idr_remove(idr, wd);
 472        /* Removed from the idr, drop that ref. */
 473        fsnotify_put_mark(&i_mark->fsn_mark);
 474out:
 475        i_mark->wd = -1;
 476        spin_unlock(idr_lock);
 477        /* match the ref taken by inotify_idr_find_locked() */
 478        if (found_i_mark)
 479                fsnotify_put_mark(&found_i_mark->fsn_mark);
 480}
 481
 482/*
 483 * Send IN_IGNORED for this wd, remove this wd from the idr.
 484 */
 485void inotify_ignored_and_remove_idr(struct fsnotify_mark *fsn_mark,
 486                                    struct fsnotify_group *group)
 487{
 488        struct inotify_inode_mark *i_mark;
 489        struct fsnotify_iter_info iter_info = { };
 490
 491        fsnotify_iter_set_report_type_mark(&iter_info, FSNOTIFY_OBJ_TYPE_INODE,
 492                                           fsn_mark);
 493
 494        /* Queue ignore event for the watch */
 495        inotify_handle_event(group, FS_IN_IGNORED, NULL, FSNOTIFY_EVENT_NONE,
 496                             NULL, NULL, 0, &iter_info);
 497
 498        i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
 499        /* remove this mark from the idr */
 500        inotify_remove_from_idr(group, i_mark);
 501
 502        dec_inotify_watches(group->inotify_data.ucounts);
 503}
 504
 505static int inotify_update_existing_watch(struct fsnotify_group *group,
 506                                         struct inode *inode,
 507                                         u32 arg)
 508{
 509        struct fsnotify_mark *fsn_mark;
 510        struct inotify_inode_mark *i_mark;
 511        __u32 old_mask, new_mask;
 512        __u32 mask;
 513        int add = (arg & IN_MASK_ADD);
 514        int create = (arg & IN_MASK_CREATE);
 515        int ret;
 516
 517        mask = inotify_arg_to_mask(inode, arg);
 518
 519        fsn_mark = fsnotify_find_mark(&inode->i_fsnotify_marks, group);
 520        if (!fsn_mark)
 521                return -ENOENT;
 522        else if (create) {
 523                ret = -EEXIST;
 524                goto out;
 525        }
 526
 527        i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
 528
 529        spin_lock(&fsn_mark->lock);
 530        old_mask = fsn_mark->mask;
 531        if (add)
 532                fsn_mark->mask |= mask;
 533        else
 534                fsn_mark->mask = mask;
 535        new_mask = fsn_mark->mask;
 536        spin_unlock(&fsn_mark->lock);
 537
 538        if (old_mask != new_mask) {
 539                /* more bits in old than in new? */
 540                int dropped = (old_mask & ~new_mask);
 541                /* more bits in this fsn_mark than the inode's mask? */
 542                int do_inode = (new_mask & ~inode->i_fsnotify_mask);
 543
 544                /* update the inode with this new fsn_mark */
 545                if (dropped || do_inode)
 546                        fsnotify_recalc_mask(inode->i_fsnotify_marks);
 547
 548        }
 549
 550        /* return the wd */
 551        ret = i_mark->wd;
 552
 553out:
 554        /* match the get from fsnotify_find_mark() */
 555        fsnotify_put_mark(fsn_mark);
 556
 557        return ret;
 558}
 559
 560static int inotify_new_watch(struct fsnotify_group *group,
 561                             struct inode *inode,
 562                             u32 arg)
 563{
 564        struct inotify_inode_mark *tmp_i_mark;
 565        __u32 mask;
 566        int ret;
 567        struct idr *idr = &group->inotify_data.idr;
 568        spinlock_t *idr_lock = &group->inotify_data.idr_lock;
 569
 570        mask = inotify_arg_to_mask(inode, arg);
 571
 572        tmp_i_mark = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL);
 573        if (unlikely(!tmp_i_mark))
 574                return -ENOMEM;
 575
 576        fsnotify_init_mark(&tmp_i_mark->fsn_mark, group);
 577        tmp_i_mark->fsn_mark.mask = mask;
 578        tmp_i_mark->wd = -1;
 579
 580        ret = inotify_add_to_idr(idr, idr_lock, tmp_i_mark);
 581        if (ret)
 582                goto out_err;
 583
 584        /* increment the number of watches the user has */
 585        if (!inc_inotify_watches(group->inotify_data.ucounts)) {
 586                inotify_remove_from_idr(group, tmp_i_mark);
 587                ret = -ENOSPC;
 588                goto out_err;
 589        }
 590
 591        /* we are on the idr, now get on the inode */
 592        ret = fsnotify_add_inode_mark_locked(&tmp_i_mark->fsn_mark, inode, 0);
 593        if (ret) {
 594                /* we failed to get on the inode, get off the idr */
 595                inotify_remove_from_idr(group, tmp_i_mark);
 596                goto out_err;
 597        }
 598
 599
 600        /* return the watch descriptor for this new mark */
 601        ret = tmp_i_mark->wd;
 602
 603out_err:
 604        /* match the ref from fsnotify_init_mark() */
 605        fsnotify_put_mark(&tmp_i_mark->fsn_mark);
 606
 607        return ret;
 608}
 609
 610static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg)
 611{
 612        int ret = 0;
 613
 614        mutex_lock(&group->mark_mutex);
 615        /* try to update and existing watch with the new arg */
 616        ret = inotify_update_existing_watch(group, inode, arg);
 617        /* no mark present, try to add a new one */
 618        if (ret == -ENOENT)
 619                ret = inotify_new_watch(group, inode, arg);
 620        mutex_unlock(&group->mark_mutex);
 621
 622        return ret;
 623}
 624
 625static struct fsnotify_group *inotify_new_group(unsigned int max_events)
 626{
 627        struct fsnotify_group *group;
 628        struct inotify_event_info *oevent;
 629
 630        group = fsnotify_alloc_group(&inotify_fsnotify_ops);
 631        if (IS_ERR(group))
 632                return group;
 633
 634        oevent = kmalloc(sizeof(struct inotify_event_info), GFP_KERNEL);
 635        if (unlikely(!oevent)) {
 636                fsnotify_destroy_group(group);
 637                return ERR_PTR(-ENOMEM);
 638        }
 639        group->overflow_event = &oevent->fse;
 640        fsnotify_init_event(group->overflow_event, 0);
 641        oevent->mask = FS_Q_OVERFLOW;
 642        oevent->wd = -1;
 643        oevent->sync_cookie = 0;
 644        oevent->name_len = 0;
 645
 646        group->max_events = max_events;
 647        group->memcg = get_mem_cgroup_from_mm(current->mm);
 648
 649        spin_lock_init(&group->inotify_data.idr_lock);
 650        idr_init(&group->inotify_data.idr);
 651        group->inotify_data.ucounts = inc_ucount(current_user_ns(),
 652                                                 current_euid(),
 653                                                 UCOUNT_INOTIFY_INSTANCES);
 654
 655        if (!group->inotify_data.ucounts) {
 656                fsnotify_destroy_group(group);
 657                return ERR_PTR(-EMFILE);
 658        }
 659
 660        return group;
 661}
 662
 663
 664/* inotify syscalls */
 665static int do_inotify_init(int flags)
 666{
 667        struct fsnotify_group *group;
 668        int ret;
 669
 670        /* Check the IN_* constants for consistency.  */
 671        BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC);
 672        BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK);
 673
 674        if (flags & ~(IN_CLOEXEC | IN_NONBLOCK))
 675                return -EINVAL;
 676
 677        /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
 678        group = inotify_new_group(inotify_max_queued_events);
 679        if (IS_ERR(group))
 680                return PTR_ERR(group);
 681
 682        ret = anon_inode_getfd("inotify", &inotify_fops, group,
 683                                  O_RDONLY | flags);
 684        if (ret < 0)
 685                fsnotify_destroy_group(group);
 686
 687        return ret;
 688}
 689
 690SYSCALL_DEFINE1(inotify_init1, int, flags)
 691{
 692        return do_inotify_init(flags);
 693}
 694
 695SYSCALL_DEFINE0(inotify_init)
 696{
 697        return do_inotify_init(0);
 698}
 699
 700SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname,
 701                u32, mask)
 702{
 703        struct fsnotify_group *group;
 704        struct inode *inode;
 705        struct path path;
 706        struct fd f;
 707        int ret;
 708        unsigned flags = 0;
 709
 710        /*
 711         * We share a lot of code with fs/dnotify.  We also share
 712         * the bit layout between inotify's IN_* and the fsnotify
 713         * FS_*.  This check ensures that only the inotify IN_*
 714         * bits get passed in and set in watches/events.
 715         */
 716        if (unlikely(mask & ~ALL_INOTIFY_BITS))
 717                return -EINVAL;
 718        /*
 719         * Require at least one valid bit set in the mask.
 720         * Without _something_ set, we would have no events to
 721         * watch for.
 722         */
 723        if (unlikely(!(mask & ALL_INOTIFY_BITS)))
 724                return -EINVAL;
 725
 726        f = fdget(fd);
 727        if (unlikely(!f.file))
 728                return -EBADF;
 729
 730        /* IN_MASK_ADD and IN_MASK_CREATE don't make sense together */
 731        if (unlikely((mask & IN_MASK_ADD) && (mask & IN_MASK_CREATE))) {
 732                ret = -EINVAL;
 733                goto fput_and_out;
 734        }
 735
 736        /* verify that this is indeed an inotify instance */
 737        if (unlikely(f.file->f_op != &inotify_fops)) {
 738                ret = -EINVAL;
 739                goto fput_and_out;
 740        }
 741
 742        if (!(mask & IN_DONT_FOLLOW))
 743                flags |= LOOKUP_FOLLOW;
 744        if (mask & IN_ONLYDIR)
 745                flags |= LOOKUP_DIRECTORY;
 746
 747        ret = inotify_find_inode(pathname, &path, flags,
 748                        (mask & IN_ALL_EVENTS));
 749        if (ret)
 750                goto fput_and_out;
 751
 752        /* inode held in place by reference to path; group by fget on fd */
 753        inode = path.dentry->d_inode;
 754        group = f.file->private_data;
 755
 756        /* create/update an inode mark */
 757        ret = inotify_update_watch(group, inode, mask);
 758        path_put(&path);
 759fput_and_out:
 760        fdput(f);
 761        return ret;
 762}
 763
 764SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
 765{
 766        struct fsnotify_group *group;
 767        struct inotify_inode_mark *i_mark;
 768        struct fd f;
 769        int ret = -EINVAL;
 770
 771        f = fdget(fd);
 772        if (unlikely(!f.file))
 773                return -EBADF;
 774
 775        /* verify that this is indeed an inotify instance */
 776        if (unlikely(f.file->f_op != &inotify_fops))
 777                goto out;
 778
 779        group = f.file->private_data;
 780
 781        i_mark = inotify_idr_find(group, wd);
 782        if (unlikely(!i_mark))
 783                goto out;
 784
 785        ret = 0;
 786
 787        fsnotify_destroy_mark(&i_mark->fsn_mark, group);
 788
 789        /* match ref taken by inotify_idr_find */
 790        fsnotify_put_mark(&i_mark->fsn_mark);
 791
 792out:
 793        fdput(f);
 794        return ret;
 795}
 796
 797/*
 798 * inotify_user_setup - Our initialization function.  Note that we cannot return
 799 * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
 800 * must result in panic().
 801 */
 802static int __init inotify_user_setup(void)
 803{
 804        BUILD_BUG_ON(IN_ACCESS != FS_ACCESS);
 805        BUILD_BUG_ON(IN_MODIFY != FS_MODIFY);
 806        BUILD_BUG_ON(IN_ATTRIB != FS_ATTRIB);
 807        BUILD_BUG_ON(IN_CLOSE_WRITE != FS_CLOSE_WRITE);
 808        BUILD_BUG_ON(IN_CLOSE_NOWRITE != FS_CLOSE_NOWRITE);
 809        BUILD_BUG_ON(IN_OPEN != FS_OPEN);
 810        BUILD_BUG_ON(IN_MOVED_FROM != FS_MOVED_FROM);
 811        BUILD_BUG_ON(IN_MOVED_TO != FS_MOVED_TO);
 812        BUILD_BUG_ON(IN_CREATE != FS_CREATE);
 813        BUILD_BUG_ON(IN_DELETE != FS_DELETE);
 814        BUILD_BUG_ON(IN_DELETE_SELF != FS_DELETE_SELF);
 815        BUILD_BUG_ON(IN_MOVE_SELF != FS_MOVE_SELF);
 816        BUILD_BUG_ON(IN_UNMOUNT != FS_UNMOUNT);
 817        BUILD_BUG_ON(IN_Q_OVERFLOW != FS_Q_OVERFLOW);
 818        BUILD_BUG_ON(IN_IGNORED != FS_IN_IGNORED);
 819        BUILD_BUG_ON(IN_EXCL_UNLINK != FS_EXCL_UNLINK);
 820        BUILD_BUG_ON(IN_ISDIR != FS_ISDIR);
 821        BUILD_BUG_ON(IN_ONESHOT != FS_IN_ONESHOT);
 822
 823        BUILD_BUG_ON(HWEIGHT32(ALL_INOTIFY_BITS) != 22);
 824
 825        inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark,
 826                                               SLAB_PANIC|SLAB_ACCOUNT);
 827
 828        inotify_max_queued_events = 16384;
 829        init_user_ns.ucount_max[UCOUNT_INOTIFY_INSTANCES] = 128;
 830        init_user_ns.ucount_max[UCOUNT_INOTIFY_WATCHES] = 8192;
 831
 832        return 0;
 833}
 834fs_initcall(inotify_user_setup);
 835