fs/notify/fanotify/fanotify.c

/* [<][>][^][v][top][bottom][index][help] */
This source file includes following definitions.
should_merge
fanotify_merge
fanotify_get_response
fanotify_group_event_mask
fanotify_encode_fid
fanotify_fid_inode
fanotify_alloc_event
fanotify_get_fsid
fanotify_handle_event
fanotify_free_group_priv
fanotify_free_event
fanotify_free_mark
   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/fanotify.h>
   3 #include <linux/fdtable.h>
   4 #include <linux/fsnotify_backend.h>
   5 #include <linux/init.h>
   6 #include <linux/jiffies.h>
   7 #include <linux/kernel.h> /* UINT_MAX */
   8 #include <linux/mount.h>
   9 #include <linux/sched.h>
  10 #include <linux/sched/user.h>
  11 #include <linux/sched/signal.h>
  12 #include <linux/types.h>
  13 #include <linux/wait.h>
  14 #include <linux/audit.h>
  15 #include <linux/sched/mm.h>
  16 #include <linux/statfs.h>
  17 
  18 #include "fanotify.h"
  19 
  20 static bool should_merge(struct fsnotify_event *old_fsn,
  21                          struct fsnotify_event *new_fsn)
  22 {
  23         struct fanotify_event *old, *new;
  24 
  25         pr_debug("%s: old=%p new=%p\n", __func__, old_fsn, new_fsn);
  26         old = FANOTIFY_E(old_fsn);
  27         new = FANOTIFY_E(new_fsn);
  28 
  29         if (old_fsn->objectid != new_fsn->objectid || old->pid != new->pid ||
  30             old->fh_type != new->fh_type || old->fh_len != new->fh_len)
  31                 return false;
  32 
  33         if (fanotify_event_has_path(old)) {
  34                 return old->path.mnt == new->path.mnt &&
  35                         old->path.dentry == new->path.dentry;
  36         } else if (fanotify_event_has_fid(old)) {
  37                 /*
  38                  * We want to merge many dirent events in the same dir (i.e.
  39                  * creates/unlinks/renames), but we do not want to merge dirent
  40                  * events referring to subdirs with dirent events referring to
  41                  * non subdirs, otherwise, user won't be able to tell from a
  42                  * mask FAN_CREATE|FAN_DELETE|FAN_ONDIR if it describes mkdir+
  43                  * unlink pair or rmdir+create pair of events.
  44                  */
  45                 return (old->mask & FS_ISDIR) == (new->mask & FS_ISDIR) &&
  46                         fanotify_fid_equal(&old->fid, &new->fid, old->fh_len);
  47         }
  48 
  49         /* Do not merge events if we failed to encode fid */
  50         return false;
  51 }
  52 
  53 /* and the list better be locked by something too! */
  54 static int fanotify_merge(struct list_head *list, struct fsnotify_event *event)
  55 {
  56         struct fsnotify_event *test_event;
  57         struct fanotify_event *new;
  58 
  59         pr_debug("%s: list=%p event=%p\n", __func__, list, event);
  60         new = FANOTIFY_E(event);
  61 
  62         /*
  63          * Don't merge a permission event with any other event so that we know
  64          * the event structure we have created in fanotify_handle_event() is the
  65          * one we should check for permission response.
  66          */
  67         if (fanotify_is_perm_event(new->mask))
  68                 return 0;
  69 
  70         list_for_each_entry_reverse(test_event, list, list) {
  71                 if (should_merge(test_event, event)) {
  72                         FANOTIFY_E(test_event)->mask |= new->mask;
  73                         return 1;
  74                 }
  75         }
  76 
  77         return 0;
  78 }
  79 
  80 /*
  81  * Wait for response to permission event. The function also takes care of
  82  * freeing the permission event (or offloads that in case the wait is canceled
  83  * by a signal). The function returns 0 in case access got allowed by userspace,
  84  * -EPERM in case userspace disallowed the access, and -ERESTARTSYS in case
  85  * the wait got interrupted by a signal.
  86  */
  87 static int fanotify_get_response(struct fsnotify_group *group,
  88                                  struct fanotify_perm_event *event,
  89                                  struct fsnotify_iter_info *iter_info)
  90 {
  91         int ret;
  92 
  93         pr_debug("%s: group=%p event=%p\n", __func__, group, event);
  94 
  95         ret = wait_event_killable(group->fanotify_data.access_waitq,
  96                                   event->state == FAN_EVENT_ANSWERED);
  97         /* Signal pending? */
  98         if (ret < 0) {
  99                 spin_lock(&group->notification_lock);
 100                 /* Event reported to userspace and no answer yet? */
 101                 if (event->state == FAN_EVENT_REPORTED) {
 102                         /* Event will get freed once userspace answers to it */
 103                         event->state = FAN_EVENT_CANCELED;
 104                         spin_unlock(&group->notification_lock);
 105                         return ret;
 106                 }
 107                 /* Event not yet reported? Just remove it. */
 108                 if (event->state == FAN_EVENT_INIT)
 109                         fsnotify_remove_queued_event(group, &event->fae.fse);
 110                 /*
 111                  * Event may be also answered in case signal delivery raced
 112                  * with wakeup. In that case we have nothing to do besides
 113                  * freeing the event and reporting error.
 114                  */
 115                 spin_unlock(&group->notification_lock);
 116                 goto out;
 117         }
 118 
 119         /* userspace responded, convert to something usable */
 120         switch (event->response & ~FAN_AUDIT) {
 121         case FAN_ALLOW:
 122                 ret = 0;
 123                 break;
 124         case FAN_DENY:
 125         default:
 126                 ret = -EPERM;
 127         }
 128 
 129         /* Check if the response should be audited */
 130         if (event->response & FAN_AUDIT)
 131                 audit_fanotify(event->response & ~FAN_AUDIT);
 132 
 133         pr_debug("%s: group=%p event=%p about to return ret=%d\n", __func__,
 134                  group, event, ret);
 135 out:
 136         fsnotify_destroy_event(group, &event->fae.fse);
 137 
 138         return ret;
 139 }
 140 
 141 /*
 142  * This function returns a mask for an event that only contains the flags
 143  * that have been specifically requested by the user. Flags that may have
 144  * been included within the event mask, but have not been explicitly
 145  * requested by the user, will not be present in the returned mask.
 146  */
 147 static u32 fanotify_group_event_mask(struct fsnotify_group *group,
 148                                      struct fsnotify_iter_info *iter_info,
 149                                      u32 event_mask, const void *data,
 150                                      int data_type)
 151 {
 152         __u32 marks_mask = 0, marks_ignored_mask = 0;
 153         __u32 test_mask, user_mask = FANOTIFY_OUTGOING_EVENTS;
 154         const struct path *path = data;
 155         struct fsnotify_mark *mark;
 156         int type;
 157 
 158         pr_debug("%s: report_mask=%x mask=%x data=%p data_type=%d\n",
 159                  __func__, iter_info->report_mask, event_mask, data, data_type);
 160 
 161         if (!FAN_GROUP_FLAG(group, FAN_REPORT_FID)) {
 162                 /* Do we have path to open a file descriptor? */
 163                 if (data_type != FSNOTIFY_EVENT_PATH)
 164                         return 0;
 165                 /* Path type events are only relevant for files and dirs */
 166                 if (!d_is_reg(path->dentry) && !d_can_lookup(path->dentry))
 167                         return 0;
 168         }
 169 
 170         fsnotify_foreach_obj_type(type) {
 171                 if (!fsnotify_iter_should_report_type(iter_info, type))
 172                         continue;
 173                 mark = iter_info->marks[type];
 174 
 175                 /* Apply ignore mask regardless of ISDIR and ON_CHILD flags */
 176                 marks_ignored_mask |= mark->ignored_mask;
 177 
 178                 /*
 179                  * If the event is on dir and this mark doesn't care about
 180                  * events on dir, don't send it!
 181                  */
 182                 if (event_mask & FS_ISDIR && !(mark->mask & FS_ISDIR))
 183                         continue;
 184 
 185                 /*
 186                  * If the event is for a child and this mark doesn't care about
 187                  * events on a child, don't send it!
 188                  */
 189                 if (event_mask & FS_EVENT_ON_CHILD &&
 190                     (type != FSNOTIFY_OBJ_TYPE_INODE ||
 191                      !(mark->mask & FS_EVENT_ON_CHILD)))
 192                         continue;
 193 
 194                 marks_mask |= mark->mask;
 195         }
 196 
 197         test_mask = event_mask & marks_mask & ~marks_ignored_mask;
 198 
 199         /*
 200          * dirent modification events (create/delete/move) do not carry the
 201          * child entry name/inode information. Instead, we report FAN_ONDIR
 202          * for mkdir/rmdir so user can differentiate them from creat/unlink.
 203          *
 204          * For backward compatibility and consistency, do not report FAN_ONDIR
 205          * to user in legacy fanotify mode (reporting fd) and report FAN_ONDIR
 206          * to user in FAN_REPORT_FID mode for all event types.
 207          */
 208         if (FAN_GROUP_FLAG(group, FAN_REPORT_FID)) {
 209                 /* Do not report FAN_ONDIR without any event */
 210                 if (!(test_mask & ~FAN_ONDIR))
 211                         return 0;
 212         } else {
 213                 user_mask &= ~FAN_ONDIR;
 214         }
 215 
 216         return test_mask & user_mask;
 217 }
 218 
 219 static int fanotify_encode_fid(struct fanotify_event *event,
 220                                struct inode *inode, gfp_t gfp,
 221                                __kernel_fsid_t *fsid)
 222 {
 223         struct fanotify_fid *fid = &event->fid;
 224         int dwords, bytes = 0;
 225         int err, type;
 226 
 227         fid->ext_fh = NULL;
 228         dwords = 0;
 229         err = -ENOENT;
 230         type = exportfs_encode_inode_fh(inode, NULL, &dwords, NULL);
 231         if (!dwords)
 232                 goto out_err;
 233 
 234         bytes = dwords << 2;
 235         if (bytes > FANOTIFY_INLINE_FH_LEN) {
 236                 /* Treat failure to allocate fh as failure to allocate event */
 237                 err = -ENOMEM;
 238                 fid->ext_fh = kmalloc(bytes, gfp);
 239                 if (!fid->ext_fh)
 240                         goto out_err;
 241         }
 242 
 243         type = exportfs_encode_inode_fh(inode, fanotify_fid_fh(fid, bytes),
 244                                         &dwords, NULL);
 245         err = -EINVAL;
 246         if (!type || type == FILEID_INVALID || bytes != dwords << 2)
 247                 goto out_err;
 248 
 249         fid->fsid = *fsid;
 250         event->fh_len = bytes;
 251 
 252         return type;
 253 
 254 out_err:
 255         pr_warn_ratelimited("fanotify: failed to encode fid (fsid=%x.%x, "
 256                             "type=%d, bytes=%d, err=%i)\n",
 257                             fsid->val[0], fsid->val[1], type, bytes, err);
 258         kfree(fid->ext_fh);
 259         fid->ext_fh = NULL;
 260         event->fh_len = 0;
 261 
 262         return FILEID_INVALID;
 263 }
 264 
 265 /*
 266  * The inode to use as identifier when reporting fid depends on the event.
 267  * Report the modified directory inode on dirent modification events.
 268  * Report the "victim" inode otherwise.
 269  * For example:
 270  * FS_ATTRIB reports the child inode even if reported on a watched parent.
 271  * FS_CREATE reports the modified dir inode and not the created inode.
 272  */
 273 static struct inode *fanotify_fid_inode(struct inode *to_tell, u32 event_mask,
 274                                         const void *data, int data_type)
 275 {
 276         if (event_mask & ALL_FSNOTIFY_DIRENT_EVENTS)
 277                 return to_tell;
 278         else if (data_type == FSNOTIFY_EVENT_INODE)
 279                 return (struct inode *)data;
 280         else if (data_type == FSNOTIFY_EVENT_PATH)
 281                 return d_inode(((struct path *)data)->dentry);
 282         return NULL;
 283 }
 284 
 285 struct fanotify_event *fanotify_alloc_event(struct fsnotify_group *group,
 286                                             struct inode *inode, u32 mask,
 287                                             const void *data, int data_type,
 288                                             __kernel_fsid_t *fsid)
 289 {
 290         struct fanotify_event *event = NULL;
 291         gfp_t gfp = GFP_KERNEL_ACCOUNT;
 292         struct inode *id = fanotify_fid_inode(inode, mask, data, data_type);
 293 
 294         /*
 295          * For queues with unlimited length lost events are not expected and
 296          * can possibly have security implications. Avoid losing events when
 297          * memory is short. For the limited size queues, avoid OOM killer in the
 298          * target monitoring memcg as it may have security repercussion.
 299          */
 300         if (group->max_events == UINT_MAX)
 301                 gfp |= __GFP_NOFAIL;
 302         else
 303                 gfp |= __GFP_RETRY_MAYFAIL;
 304 
 305         /* Whoever is interested in the event, pays for the allocation. */
 306         memalloc_use_memcg(group->memcg);
 307 
 308         if (fanotify_is_perm_event(mask)) {
 309                 struct fanotify_perm_event *pevent;
 310 
 311                 pevent = kmem_cache_alloc(fanotify_perm_event_cachep, gfp);
 312                 if (!pevent)
 313                         goto out;
 314                 event = &pevent->fae;
 315                 pevent->response = 0;
 316                 pevent->state = FAN_EVENT_INIT;
 317                 goto init;
 318         }
 319         event = kmem_cache_alloc(fanotify_event_cachep, gfp);
 320         if (!event)
 321                 goto out;
 322 init: __maybe_unused
 323         /*
 324          * Use the victim inode instead of the watching inode as the id for
 325          * event queue, so event reported on parent is merged with event
 326          * reported on child when both directory and child watches exist.
 327          */
 328         fsnotify_init_event(&event->fse, (unsigned long)id);
 329         event->mask = mask;
 330         if (FAN_GROUP_FLAG(group, FAN_REPORT_TID))
 331                 event->pid = get_pid(task_pid(current));
 332         else
 333                 event->pid = get_pid(task_tgid(current));
 334         event->fh_len = 0;
 335         if (id && FAN_GROUP_FLAG(group, FAN_REPORT_FID)) {
 336                 /* Report the event without a file identifier on encode error */
 337                 event->fh_type = fanotify_encode_fid(event, id, gfp, fsid);
 338         } else if (data_type == FSNOTIFY_EVENT_PATH) {
 339                 event->fh_type = FILEID_ROOT;
 340                 event->path = *((struct path *)data);
 341                 path_get(&event->path);
 342         } else {
 343                 event->fh_type = FILEID_INVALID;
 344                 event->path.mnt = NULL;
 345                 event->path.dentry = NULL;
 346         }
 347 out:
 348         memalloc_unuse_memcg();
 349         return event;
 350 }
 351 
 352 /*
 353  * Get cached fsid of the filesystem containing the object from any connector.
 354  * All connectors are supposed to have the same fsid, but we do not verify that
 355  * here.
 356  */
 357 static __kernel_fsid_t fanotify_get_fsid(struct fsnotify_iter_info *iter_info)
 358 {
 359         int type;
 360         __kernel_fsid_t fsid = {};
 361 
 362         fsnotify_foreach_obj_type(type) {
 363                 struct fsnotify_mark_connector *conn;
 364 
 365                 if (!fsnotify_iter_should_report_type(iter_info, type))
 366                         continue;
 367 
 368                 conn = READ_ONCE(iter_info->marks[type]->connector);
 369                 /* Mark is just getting destroyed or created? */
 370                 if (!conn)
 371                         continue;
 372                 if (!(conn->flags & FSNOTIFY_CONN_FLAG_HAS_FSID))
 373                         continue;
 374                 /* Pairs with smp_wmb() in fsnotify_add_mark_list() */
 375                 smp_rmb();
 376                 fsid = conn->fsid;
 377                 if (WARN_ON_ONCE(!fsid.val[0] && !fsid.val[1]))
 378                         continue;
 379                 return fsid;
 380         }
 381 
 382         return fsid;
 383 }
 384 
 385 static int fanotify_handle_event(struct fsnotify_group *group,
 386                                  struct inode *inode,
 387                                  u32 mask, const void *data, int data_type,
 388                                  const struct qstr *file_name, u32 cookie,
 389                                  struct fsnotify_iter_info *iter_info)
 390 {
 391         int ret = 0;
 392         struct fanotify_event *event;
 393         struct fsnotify_event *fsn_event;
 394         __kernel_fsid_t fsid = {};
 395 
 396         BUILD_BUG_ON(FAN_ACCESS != FS_ACCESS);
 397         BUILD_BUG_ON(FAN_MODIFY != FS_MODIFY);
 398         BUILD_BUG_ON(FAN_ATTRIB != FS_ATTRIB);
 399         BUILD_BUG_ON(FAN_CLOSE_NOWRITE != FS_CLOSE_NOWRITE);
 400         BUILD_BUG_ON(FAN_CLOSE_WRITE != FS_CLOSE_WRITE);
 401         BUILD_BUG_ON(FAN_OPEN != FS_OPEN);
 402         BUILD_BUG_ON(FAN_MOVED_TO != FS_MOVED_TO);
 403         BUILD_BUG_ON(FAN_MOVED_FROM != FS_MOVED_FROM);
 404         BUILD_BUG_ON(FAN_CREATE != FS_CREATE);
 405         BUILD_BUG_ON(FAN_DELETE != FS_DELETE);
 406         BUILD_BUG_ON(FAN_DELETE_SELF != FS_DELETE_SELF);
 407         BUILD_BUG_ON(FAN_MOVE_SELF != FS_MOVE_SELF);
 408         BUILD_BUG_ON(FAN_EVENT_ON_CHILD != FS_EVENT_ON_CHILD);
 409         BUILD_BUG_ON(FAN_Q_OVERFLOW != FS_Q_OVERFLOW);
 410         BUILD_BUG_ON(FAN_OPEN_PERM != FS_OPEN_PERM);
 411         BUILD_BUG_ON(FAN_ACCESS_PERM != FS_ACCESS_PERM);
 412         BUILD_BUG_ON(FAN_ONDIR != FS_ISDIR);
 413         BUILD_BUG_ON(FAN_OPEN_EXEC != FS_OPEN_EXEC);
 414         BUILD_BUG_ON(FAN_OPEN_EXEC_PERM != FS_OPEN_EXEC_PERM);
 415 
 416         BUILD_BUG_ON(HWEIGHT32(ALL_FANOTIFY_EVENT_BITS) != 19);
 417 
 418         mask = fanotify_group_event_mask(group, iter_info, mask, data,
 419                                          data_type);
 420         if (!mask)
 421                 return 0;
 422 
 423         pr_debug("%s: group=%p inode=%p mask=%x\n", __func__, group, inode,
 424                  mask);
 425 
 426         if (fanotify_is_perm_event(mask)) {
 427                 /*
 428                  * fsnotify_prepare_user_wait() fails if we race with mark
 429                  * deletion.  Just let the operation pass in that case.
 430                  */
 431                 if (!fsnotify_prepare_user_wait(iter_info))
 432                         return 0;
 433         }
 434 
 435         if (FAN_GROUP_FLAG(group, FAN_REPORT_FID)) {
 436                 fsid = fanotify_get_fsid(iter_info);
 437                 /* Racing with mark destruction or creation? */
 438                 if (!fsid.val[0] && !fsid.val[1])
 439                         return 0;
 440         }
 441 
 442         event = fanotify_alloc_event(group, inode, mask, data, data_type,
 443                                      &fsid);
 444         ret = -ENOMEM;
 445         if (unlikely(!event)) {
 446                 /*
 447                  * We don't queue overflow events for permission events as
 448                  * there the access is denied and so no event is in fact lost.
 449                  */
 450                 if (!fanotify_is_perm_event(mask))
 451                         fsnotify_queue_overflow(group);
 452                 goto finish;
 453         }
 454 
 455         fsn_event = &event->fse;
 456         ret = fsnotify_add_event(group, fsn_event, fanotify_merge);
 457         if (ret) {
 458                 /* Permission events shouldn't be merged */
 459                 BUG_ON(ret == 1 && mask & FANOTIFY_PERM_EVENTS);
 460                 /* Our event wasn't used in the end. Free it. */
 461                 fsnotify_destroy_event(group, fsn_event);
 462 
 463                 ret = 0;
 464         } else if (fanotify_is_perm_event(mask)) {
 465                 ret = fanotify_get_response(group, FANOTIFY_PE(fsn_event),
 466                                             iter_info);
 467         }
 468 finish:
 469         if (fanotify_is_perm_event(mask))
 470                 fsnotify_finish_user_wait(iter_info);
 471 
 472         return ret;
 473 }
 474 
 475 static void fanotify_free_group_priv(struct fsnotify_group *group)
 476 {
 477         struct user_struct *user;
 478 
 479         user = group->fanotify_data.user;
 480         atomic_dec(&user->fanotify_listeners);
 481         free_uid(user);
 482 }
 483 
 484 static void fanotify_free_event(struct fsnotify_event *fsn_event)
 485 {
 486         struct fanotify_event *event;
 487 
 488         event = FANOTIFY_E(fsn_event);
 489         if (fanotify_event_has_path(event))
 490                 path_put(&event->path);
 491         else if (fanotify_event_has_ext_fh(event))
 492                 kfree(event->fid.ext_fh);
 493         put_pid(event->pid);
 494         if (fanotify_is_perm_event(event->mask)) {
 495                 kmem_cache_free(fanotify_perm_event_cachep,
 496                                 FANOTIFY_PE(fsn_event));
 497                 return;
 498         }
 499         kmem_cache_free(fanotify_event_cachep, event);
 500 }
 501 
 502 static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
 503 {
 504         kmem_cache_free(fanotify_mark_cache, fsn_mark);
 505 }
 506 
 507 const struct fsnotify_ops fanotify_fsnotify_ops = {
 508         .handle_event = fanotify_handle_event,
 509         .free_group_priv = fanotify_free_group_priv,
 510         .free_event = fanotify_free_event,
 511         .free_mark = fanotify_free_mark,
 512 };
/* [<][>][^][v][top][bottom][index][help] */
root/fs/notify/fanotify/fanotify.c

DEFINITIONS
