1 // SPDX-License-Identifier: GPL-2.0 2 #include <linux/fanotify.h> 3 #include <linux/fsnotify_backend.h> 4 #include <linux/init.h> 5 #include <linux/jiffies.h> 6 #include <linux/kernel.h> /* UINT_MAX */ 7 #include <linux/mount.h> 8 #include <linux/sched.h> 9 #include <linux/sched/user.h> 10 #include <linux/sched/signal.h> 11 #include <linux/types.h> 12 #include <linux/wait.h> 13 #include <linux/audit.h> 14 #include <linux/sched/mm.h> 15 #include <linux/statfs.h> 16 #include <linux/stringhash.h> 17 18 #include "fanotify.h" 19 20 static bool fanotify_path_equal(const struct path *p1, const struct path *p2) 21 { 22 return p1->mnt == p2->mnt && p1->dentry == p2->dentry; 23 } 24 25 static unsigned int fanotify_hash_path(const struct path *path) 26 { 27 return hash_ptr(path->dentry, FANOTIFY_EVENT_HASH_BITS) ^ 28 hash_ptr(path->mnt, FANOTIFY_EVENT_HASH_BITS); 29 } 30 31 static unsigned int fanotify_hash_fsid(__kernel_fsid_t *fsid) 32 { 33 return hash_32(fsid->val[0], FANOTIFY_EVENT_HASH_BITS) ^ 34 hash_32(fsid->val[1], FANOTIFY_EVENT_HASH_BITS); 35 } 36 37 static bool fanotify_fh_equal(struct fanotify_fh *fh1, 38 struct fanotify_fh *fh2) 39 { 40 if (fh1->type != fh2->type || fh1->len != fh2->len) 41 return false; 42 43 return !fh1->len || 44 !memcmp(fanotify_fh_buf(fh1), fanotify_fh_buf(fh2), fh1->len); 45 } 46 47 static unsigned int fanotify_hash_fh(struct fanotify_fh *fh) 48 { 49 long salt = (long)fh->type | (long)fh->len << 8; 50 51 /* 52 * full_name_hash() works long by long, so it handles fh buf optimally. 53 */ 54 return full_name_hash((void *)salt, fanotify_fh_buf(fh), fh->len); 55 } 56 57 static bool fanotify_fid_event_equal(struct fanotify_fid_event *ffe1, 58 struct fanotify_fid_event *ffe2) 59 { 60 /* Do not merge fid events without object fh */ 61 if (!ffe1->object_fh.len) 62 return false; 63 64 return fanotify_fsid_equal(&ffe1->fsid, &ffe2->fsid) && 65 fanotify_fh_equal(&ffe1->object_fh, &ffe2->object_fh); 66 } 67 68 static bool fanotify_info_equal(struct fanotify_info *info1, 69 struct fanotify_info *info2) 70 { 71 if (info1->dir_fh_totlen != info2->dir_fh_totlen || 72 info1->dir2_fh_totlen != info2->dir2_fh_totlen || 73 info1->file_fh_totlen != info2->file_fh_totlen || 74 info1->name_len != info2->name_len || 75 info1->name2_len != info2->name2_len) 76 return false; 77 78 if (info1->dir_fh_totlen && 79 !fanotify_fh_equal(fanotify_info_dir_fh(info1), 80 fanotify_info_dir_fh(info2))) 81 return false; 82 83 if (info1->dir2_fh_totlen && 84 !fanotify_fh_equal(fanotify_info_dir2_fh(info1), 85 fanotify_info_dir2_fh(info2))) 86 return false; 87 88 if (info1->file_fh_totlen && 89 !fanotify_fh_equal(fanotify_info_file_fh(info1), 90 fanotify_info_file_fh(info2))) 91 return false; 92 93 if (info1->name_len && 94 memcmp(fanotify_info_name(info1), fanotify_info_name(info2), 95 info1->name_len)) 96 return false; 97 98 return !info1->name2_len || 99 !memcmp(fanotify_info_name2(info1), fanotify_info_name2(info2), 100 info1->name2_len); 101 } 102 103 static bool fanotify_name_event_equal(struct fanotify_name_event *fne1, 104 struct fanotify_name_event *fne2) 105 { 106 struct fanotify_info *info1 = &fne1->info; 107 struct fanotify_info *info2 = &fne2->info; 108 109 /* Do not merge name events without dir fh */ 110 if (!info1->dir_fh_totlen) 111 return false; 112 113 if (!fanotify_fsid_equal(&fne1->fsid, &fne2->fsid)) 114 return false; 115 116 return fanotify_info_equal(info1, info2); 117 } 118 119 static bool fanotify_error_event_equal(struct fanotify_error_event *fee1, 120 struct fanotify_error_event *fee2) 121 { 122 /* Error events against the same file system are always merged. */ 123 if (!fanotify_fsid_equal(&fee1->fsid, &fee2->fsid)) 124 return false; 125 126 return true; 127 } 128 129 static bool fanotify_should_merge(struct fanotify_event *old, 130 struct fanotify_event *new) 131 { 132 pr_debug("%s: old=%p new=%p\n", __func__, old, new); 133 134 if (old->hash != new->hash || 135 old->type != new->type || old->pid != new->pid) 136 return false; 137 138 /* 139 * We want to merge many dirent events in the same dir (i.e. 140 * creates/unlinks/renames), but we do not want to merge dirent 141 * events referring to subdirs with dirent events referring to 142 * non subdirs, otherwise, user won't be able to tell from a 143 * mask FAN_CREATE|FAN_DELETE|FAN_ONDIR if it describes mkdir+ 144 * unlink pair or rmdir+create pair of events. 145 */ 146 if ((old->mask & FS_ISDIR) != (new->mask & FS_ISDIR)) 147 return false; 148 149 /* 150 * FAN_RENAME event is reported with special info record types, 151 * so we cannot merge it with other events. 152 */ 153 if ((old->mask & FAN_RENAME) != (new->mask & FAN_RENAME)) 154 return false; 155 156 switch (old->type) { 157 case FANOTIFY_EVENT_TYPE_PATH: 158 return fanotify_path_equal(fanotify_event_path(old), 159 fanotify_event_path(new)); 160 case FANOTIFY_EVENT_TYPE_FID: 161 return fanotify_fid_event_equal(FANOTIFY_FE(old), 162 FANOTIFY_FE(new)); 163 case FANOTIFY_EVENT_TYPE_FID_NAME: 164 return fanotify_name_event_equal(FANOTIFY_NE(old), 165 FANOTIFY_NE(new)); 166 case FANOTIFY_EVENT_TYPE_FS_ERROR: 167 return fanotify_error_event_equal(FANOTIFY_EE(old), 168 FANOTIFY_EE(new)); 169 default: 170 WARN_ON_ONCE(1); 171 } 172 173 return false; 174 } 175 176 /* Limit event merges to limit CPU overhead per event */ 177 #define FANOTIFY_MAX_MERGE_EVENTS 128 178 179 /* and the list better be locked by something too! */ 180 static int fanotify_merge(struct fsnotify_group *group, 181 struct fsnotify_event *event) 182 { 183 struct fanotify_event *old, *new = FANOTIFY_E(event); 184 unsigned int bucket = fanotify_event_hash_bucket(group, new); 185 struct hlist_head *hlist = &group->fanotify_data.merge_hash[bucket]; 186 int i = 0; 187 188 pr_debug("%s: group=%p event=%p bucket=%u\n", __func__, 189 group, event, bucket); 190 191 /* 192 * Don't merge a permission event with any other event so that we know 193 * the event structure we have created in fanotify_handle_event() is the 194 * one we should check for permission response. 195 */ 196 if (fanotify_is_perm_event(new->mask)) 197 return 0; 198 199 hlist_for_each_entry(old, hlist, merge_list) { 200 if (++i > FANOTIFY_MAX_MERGE_EVENTS) 201 break; 202 if (fanotify_should_merge(old, new)) { 203 old->mask |= new->mask; 204 205 if (fanotify_is_error_event(old->mask)) 206 FANOTIFY_EE(old)->err_count++; 207 208 return 1; 209 } 210 } 211 212 return 0; 213 } 214 215 /* 216 * Wait for response to permission event. The function also takes care of 217 * freeing the permission event (or offloads that in case the wait is canceled 218 * by a signal). The function returns 0 in case access got allowed by userspace, 219 * -EPERM in case userspace disallowed the access, and -ERESTARTSYS in case 220 * the wait got interrupted by a signal. 221 */ 222 static int fanotify_get_response(struct fsnotify_group *group, 223 struct fanotify_perm_event *event, 224 struct fsnotify_iter_info *iter_info) 225 { 226 int ret, errno; 227 228 pr_debug("%s: group=%p event=%p\n", __func__, group, event); 229 230 ret = wait_event_state(group->fanotify_data.access_waitq, 231 event->state == FAN_EVENT_ANSWERED, 232 (TASK_KILLABLE|TASK_FREEZABLE)); 233 234 /* Signal pending? */ 235 if (ret < 0) { 236 spin_lock(&group->notification_lock); 237 /* Event reported to userspace and no answer yet? */ 238 if (event->state == FAN_EVENT_REPORTED) { 239 /* Event will get freed once userspace answers to it */ 240 event->state = FAN_EVENT_CANCELED; 241 spin_unlock(&group->notification_lock); 242 return ret; 243 } 244 /* Event not yet reported? Just remove it. */ 245 if (event->state == FAN_EVENT_INIT) { 246 fsnotify_remove_queued_event(group, &event->fae.fse); 247 /* Permission events are not supposed to be hashed */ 248 WARN_ON_ONCE(!hlist_unhashed(&event->fae.merge_list)); 249 } 250 /* 251 * Event may be also answered in case signal delivery raced 252 * with wakeup. In that case we have nothing to do besides 253 * freeing the event and reporting error. 254 */ 255 spin_unlock(&group->notification_lock); 256 goto out; 257 } 258 259 /* userspace responded, convert to something usable */ 260 switch (event->response & FANOTIFY_RESPONSE_ACCESS) { 261 case FAN_ALLOW: 262 ret = 0; 263 break; 264 case FAN_DENY: 265 /* Check custom errno from pre-content events */ 266 errno = fanotify_get_response_errno(event->response); 267 if (errno) { 268 ret = -errno; 269 break; 270 } 271 fallthrough; 272 default: 273 ret = -EPERM; 274 } 275 276 /* Check if the response should be audited */ 277 if (event->response & FAN_AUDIT) { 278 u32 response = event->response & 279 (FANOTIFY_RESPONSE_ACCESS | FANOTIFY_RESPONSE_FLAGS); 280 audit_fanotify(response & ~FAN_AUDIT, &event->audit_rule); 281 } 282 283 pr_debug("%s: group=%p event=%p about to return ret=%d\n", __func__, 284 group, event, ret); 285 out: 286 fsnotify_destroy_event(group, &event->fae.fse); 287 288 return ret; 289 } 290 291 /* 292 * This function returns a mask for an event that only contains the flags 293 * that have been specifically requested by the user. Flags that may have 294 * been included within the event mask, but have not been explicitly 295 * requested by the user, will not be present in the returned mask. 296 */ 297 static u32 fanotify_group_event_mask(struct fsnotify_group *group, 298 struct fsnotify_iter_info *iter_info, 299 u32 *match_mask, u32 event_mask, 300 const void *data, int data_type, 301 struct inode *dir) 302 { 303 __u32 marks_mask = 0, marks_ignore_mask = 0; 304 __u32 test_mask, user_mask = FANOTIFY_OUTGOING_EVENTS | 305 FANOTIFY_EVENT_FLAGS; 306 const struct path *path = fsnotify_data_path(data, data_type); 307 unsigned int fid_mode = FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS); 308 struct fsnotify_mark *mark; 309 bool ondir = event_mask & FAN_ONDIR; 310 int type; 311 312 pr_debug("%s: report_mask=%x mask=%x data=%p data_type=%d\n", 313 __func__, iter_info->report_mask, event_mask, data, data_type); 314 315 if (!fid_mode) { 316 /* Do we have path to open a file descriptor? */ 317 if (!path) 318 return 0; 319 /* Path type events are only relevant for files and dirs */ 320 if (!d_is_reg(path->dentry) && !d_can_lookup(path->dentry)) 321 return 0; 322 } else if (!(fid_mode & FAN_REPORT_FID)) { 323 /* Do we have a directory inode to report? */ 324 if (!dir && !ondir) 325 return 0; 326 } 327 328 fsnotify_foreach_iter_mark_type(iter_info, mark, type) { 329 /* 330 * Apply ignore mask depending on event flags in ignore mask. 331 */ 332 marks_ignore_mask |= 333 fsnotify_effective_ignore_mask(mark, ondir, type); 334 335 /* 336 * Send the event depending on event flags in mark mask. 337 */ 338 if (!fsnotify_mask_applicable(mark->mask, ondir, type)) 339 continue; 340 341 marks_mask |= mark->mask; 342 343 /* Record the mark types of this group that matched the event */ 344 *match_mask |= 1U << type; 345 } 346 347 test_mask = event_mask & marks_mask & ~marks_ignore_mask; 348 349 /* 350 * For dirent modification events (create/delete/move) that do not carry 351 * the child entry name information, we report FAN_ONDIR for mkdir/rmdir 352 * so user can differentiate them from creat/unlink. 353 * 354 * For backward compatibility and consistency, do not report FAN_ONDIR 355 * to user in legacy fanotify mode (reporting fd) and report FAN_ONDIR 356 * to user in fid mode for all event types. 357 * 358 * We never report FAN_EVENT_ON_CHILD to user, but we do pass it in to 359 * fanotify_alloc_event() when group is reporting fid as indication 360 * that event happened on child. 361 */ 362 if (fid_mode) { 363 /* Do not report event flags without any event */ 364 if (!(test_mask & ~FANOTIFY_EVENT_FLAGS)) 365 return 0; 366 } else { 367 user_mask &= ~FANOTIFY_EVENT_FLAGS; 368 } 369 370 return test_mask & user_mask; 371 } 372 373 /* 374 * Check size needed to encode fanotify_fh. 375 * 376 * Return size of encoded fh without fanotify_fh header. 377 * Return 0 on failure to encode. 378 */ 379 static int fanotify_encode_fh_len(struct inode *inode) 380 { 381 int dwords = 0; 382 int fh_len; 383 384 if (!inode) 385 return 0; 386 387 exportfs_encode_fid(inode, NULL, &dwords); 388 fh_len = dwords << 2; 389 390 /* 391 * struct fanotify_error_event might be preallocated and is 392 * limited to MAX_HANDLE_SZ. This should never happen, but 393 * safeguard by forcing an invalid file handle. 394 */ 395 if (WARN_ON_ONCE(fh_len > MAX_HANDLE_SZ)) 396 return 0; 397 398 return fh_len; 399 } 400 401 /* 402 * Encode fanotify_fh. 403 * 404 * Return total size of encoded fh including fanotify_fh header. 405 * Return 0 on failure to encode. 406 */ 407 static int fanotify_encode_fh(struct fanotify_fh *fh, struct inode *inode, 408 unsigned int fh_len, unsigned int *hash, 409 gfp_t gfp) 410 { 411 int dwords, type = 0; 412 char *ext_buf = NULL; 413 void *buf = fh->buf; 414 int err; 415 416 fh->type = FILEID_ROOT; 417 fh->len = 0; 418 fh->flags = 0; 419 420 /* 421 * Invalid FHs are used by FAN_FS_ERROR for errors not 422 * linked to any inode. The f_handle won't be reported 423 * back to userspace. 424 */ 425 if (!inode) 426 goto out; 427 428 /* 429 * !gpf means preallocated variable size fh, but fh_len could 430 * be zero in that case if encoding fh len failed. 431 */ 432 err = -ENOENT; 433 if (fh_len < 4 || WARN_ON_ONCE(fh_len % 4) || fh_len > MAX_HANDLE_SZ) 434 goto out_err; 435 436 /* No external buffer in a variable size allocated fh */ 437 if (gfp && fh_len > FANOTIFY_INLINE_FH_LEN) { 438 /* Treat failure to allocate fh as failure to encode fh */ 439 err = -ENOMEM; 440 ext_buf = kmalloc(fh_len, gfp); 441 if (!ext_buf) 442 goto out_err; 443 444 *fanotify_fh_ext_buf_ptr(fh) = ext_buf; 445 buf = ext_buf; 446 fh->flags |= FANOTIFY_FH_FLAG_EXT_BUF; 447 } 448 449 dwords = fh_len >> 2; 450 type = exportfs_encode_fid(inode, buf, &dwords); 451 err = -EINVAL; 452 if (type <= 0 || type == FILEID_INVALID || fh_len != dwords << 2) 453 goto out_err; 454 455 fh->type = type; 456 fh->len = fh_len; 457 458 out: 459 /* 460 * Mix fh into event merge key. Hash might be NULL in case of 461 * unhashed FID events (i.e. FAN_FS_ERROR). 462 */ 463 if (hash) 464 *hash ^= fanotify_hash_fh(fh); 465 466 return FANOTIFY_FH_HDR_LEN + fh_len; 467 468 out_err: 469 pr_warn_ratelimited("fanotify: failed to encode fid (type=%d, len=%d, err=%i)\n", 470 type, fh_len, err); 471 kfree(ext_buf); 472 *fanotify_fh_ext_buf_ptr(fh) = NULL; 473 /* Report the event without a file identifier on encode error */ 474 fh->type = FILEID_INVALID; 475 fh->len = 0; 476 return 0; 477 } 478 479 /* 480 * FAN_REPORT_FID is ambiguous in that it reports the fid of the child for 481 * some events and the fid of the parent for create/delete/move events. 482 * 483 * With the FAN_REPORT_TARGET_FID flag, the fid of the child is reported 484 * also in create/delete/move events in addition to the fid of the parent 485 * and the name of the child. 486 */ 487 static inline bool fanotify_report_child_fid(unsigned int fid_mode, u32 mask) 488 { 489 if (mask & ALL_FSNOTIFY_DIRENT_EVENTS) 490 return (fid_mode & FAN_REPORT_TARGET_FID); 491 492 return (fid_mode & FAN_REPORT_FID) && !(mask & FAN_ONDIR); 493 } 494 495 /* 496 * The inode to use as identifier when reporting fid depends on the event 497 * and the group flags. 498 * 499 * With the group flag FAN_REPORT_TARGET_FID, always report the child fid. 500 * 501 * Without the group flag FAN_REPORT_TARGET_FID, report the modified directory 502 * fid on dirent events and the child fid otherwise. 503 * 504 * For example: 505 * FS_ATTRIB reports the child fid even if reported on a watched parent. 506 * FS_CREATE reports the modified dir fid without FAN_REPORT_TARGET_FID. 507 * and reports the created child fid with FAN_REPORT_TARGET_FID. 508 */ 509 static struct inode *fanotify_fid_inode(u32 event_mask, const void *data, 510 int data_type, struct inode *dir, 511 unsigned int fid_mode) 512 { 513 if ((event_mask & ALL_FSNOTIFY_DIRENT_EVENTS) && 514 !(fid_mode & FAN_REPORT_TARGET_FID)) 515 return dir; 516 517 return fsnotify_data_inode(data, data_type); 518 } 519 520 /* 521 * The inode to use as identifier when reporting dir fid depends on the event. 522 * Report the modified directory inode on dirent modification events. 523 * Report the "victim" inode if "victim" is a directory. 524 * Report the parent inode if "victim" is not a directory and event is 525 * reported to parent. 526 * Otherwise, do not report dir fid. 527 */ 528 static struct inode *fanotify_dfid_inode(u32 event_mask, const void *data, 529 int data_type, struct inode *dir) 530 { 531 struct inode *inode = fsnotify_data_inode(data, data_type); 532 533 if (event_mask & ALL_FSNOTIFY_DIRENT_EVENTS) 534 return dir; 535 536 if (inode && S_ISDIR(inode->i_mode)) 537 return inode; 538 539 return dir; 540 } 541 542 static struct fanotify_event *fanotify_alloc_path_event(const struct path *path, 543 unsigned int *hash, 544 gfp_t gfp) 545 { 546 struct fanotify_path_event *pevent; 547 548 pevent = kmem_cache_alloc(fanotify_path_event_cachep, gfp); 549 if (!pevent) 550 return NULL; 551 552 pevent->fae.type = FANOTIFY_EVENT_TYPE_PATH; 553 pevent->path = *path; 554 *hash ^= fanotify_hash_path(path); 555 path_get(path); 556 557 return &pevent->fae; 558 } 559 560 static struct fanotify_event *fanotify_alloc_perm_event(const void *data, 561 int data_type, 562 gfp_t gfp) 563 { 564 const struct path *path = fsnotify_data_path(data, data_type); 565 const struct file_range *range = 566 fsnotify_data_file_range(data, data_type); 567 struct fanotify_perm_event *pevent; 568 569 pevent = kmem_cache_alloc(fanotify_perm_event_cachep, gfp); 570 if (!pevent) 571 return NULL; 572 573 pevent->fae.type = FANOTIFY_EVENT_TYPE_PATH_PERM; 574 pevent->response = 0; 575 pevent->hdr.type = FAN_RESPONSE_INFO_NONE; 576 pevent->hdr.pad = 0; 577 pevent->hdr.len = 0; 578 pevent->state = FAN_EVENT_INIT; 579 pevent->path = *path; 580 /* NULL ppos means no range info */ 581 pevent->ppos = range ? &range->pos : NULL; 582 pevent->count = range ? range->count : 0; 583 path_get(path); 584 585 return &pevent->fae; 586 } 587 588 static struct fanotify_event *fanotify_alloc_fid_event(struct inode *id, 589 __kernel_fsid_t *fsid, 590 unsigned int *hash, 591 gfp_t gfp) 592 { 593 struct fanotify_fid_event *ffe; 594 595 ffe = kmem_cache_alloc(fanotify_fid_event_cachep, gfp); 596 if (!ffe) 597 return NULL; 598 599 ffe->fae.type = FANOTIFY_EVENT_TYPE_FID; 600 ffe->fsid = *fsid; 601 *hash ^= fanotify_hash_fsid(fsid); 602 fanotify_encode_fh(&ffe->object_fh, id, fanotify_encode_fh_len(id), 603 hash, gfp); 604 605 return &ffe->fae; 606 } 607 608 static struct fanotify_event *fanotify_alloc_name_event(struct inode *dir, 609 __kernel_fsid_t *fsid, 610 const struct qstr *name, 611 struct inode *child, 612 struct dentry *moved, 613 unsigned int *hash, 614 gfp_t gfp) 615 { 616 struct fanotify_name_event *fne; 617 struct fanotify_info *info; 618 struct fanotify_fh *dfh, *ffh; 619 struct inode *dir2 = moved ? d_inode(moved->d_parent) : NULL; 620 const struct qstr *name2 = moved ? &moved->d_name : NULL; 621 unsigned int dir_fh_len = fanotify_encode_fh_len(dir); 622 unsigned int dir2_fh_len = fanotify_encode_fh_len(dir2); 623 unsigned int child_fh_len = fanotify_encode_fh_len(child); 624 unsigned long name_len = name ? name->len : 0; 625 unsigned long name2_len = name2 ? name2->len : 0; 626 unsigned int len, size; 627 628 /* Reserve terminating null byte even for empty name */ 629 size = sizeof(*fne) + name_len + name2_len + 2; 630 if (dir_fh_len) 631 size += FANOTIFY_FH_HDR_LEN + dir_fh_len; 632 if (dir2_fh_len) 633 size += FANOTIFY_FH_HDR_LEN + dir2_fh_len; 634 if (child_fh_len) 635 size += FANOTIFY_FH_HDR_LEN + child_fh_len; 636 fne = kmalloc(size, gfp); 637 if (!fne) 638 return NULL; 639 640 fne->fae.type = FANOTIFY_EVENT_TYPE_FID_NAME; 641 fne->fsid = *fsid; 642 *hash ^= fanotify_hash_fsid(fsid); 643 info = &fne->info; 644 fanotify_info_init(info); 645 if (dir_fh_len) { 646 dfh = fanotify_info_dir_fh(info); 647 len = fanotify_encode_fh(dfh, dir, dir_fh_len, hash, 0); 648 fanotify_info_set_dir_fh(info, len); 649 } 650 if (dir2_fh_len) { 651 dfh = fanotify_info_dir2_fh(info); 652 len = fanotify_encode_fh(dfh, dir2, dir2_fh_len, hash, 0); 653 fanotify_info_set_dir2_fh(info, len); 654 } 655 if (child_fh_len) { 656 ffh = fanotify_info_file_fh(info); 657 len = fanotify_encode_fh(ffh, child, child_fh_len, hash, 0); 658 fanotify_info_set_file_fh(info, len); 659 } 660 if (name_len) { 661 fanotify_info_copy_name(info, name); 662 *hash ^= full_name_hash((void *)name_len, name->name, name_len); 663 } 664 if (name2_len) { 665 fanotify_info_copy_name2(info, name2); 666 *hash ^= full_name_hash((void *)name2_len, name2->name, 667 name2_len); 668 } 669 670 pr_debug("%s: size=%u dir_fh_len=%u child_fh_len=%u name_len=%u name='%.*s'\n", 671 __func__, size, dir_fh_len, child_fh_len, 672 info->name_len, info->name_len, fanotify_info_name(info)); 673 674 if (dir2_fh_len) { 675 pr_debug("%s: dir2_fh_len=%u name2_len=%u name2='%.*s'\n", 676 __func__, dir2_fh_len, info->name2_len, 677 info->name2_len, fanotify_info_name2(info)); 678 } 679 680 return &fne->fae; 681 } 682 683 static struct fanotify_event *fanotify_alloc_error_event( 684 struct fsnotify_group *group, 685 __kernel_fsid_t *fsid, 686 const void *data, int data_type, 687 unsigned int *hash) 688 { 689 struct fs_error_report *report = 690 fsnotify_data_error_report(data, data_type); 691 struct inode *inode; 692 struct fanotify_error_event *fee; 693 int fh_len; 694 695 if (WARN_ON_ONCE(!report)) 696 return NULL; 697 698 fee = mempool_alloc(&group->fanotify_data.error_events_pool, GFP_NOFS); 699 if (!fee) 700 return NULL; 701 702 fee->fae.type = FANOTIFY_EVENT_TYPE_FS_ERROR; 703 fee->error = report->error; 704 fee->err_count = 1; 705 fee->fsid = *fsid; 706 707 inode = report->inode; 708 fh_len = fanotify_encode_fh_len(inode); 709 710 /* Bad fh_len. Fallback to using an invalid fh. Should never happen. */ 711 if (!fh_len && inode) 712 inode = NULL; 713 714 fanotify_encode_fh(&fee->object_fh, inode, fh_len, NULL, 0); 715 716 *hash ^= fanotify_hash_fsid(fsid); 717 718 return &fee->fae; 719 } 720 721 static struct fanotify_event *fanotify_alloc_event( 722 struct fsnotify_group *group, 723 u32 mask, const void *data, int data_type, 724 struct inode *dir, const struct qstr *file_name, 725 __kernel_fsid_t *fsid, u32 match_mask) 726 { 727 struct fanotify_event *event = NULL; 728 gfp_t gfp = GFP_KERNEL_ACCOUNT; 729 unsigned int fid_mode = FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS); 730 struct inode *id = fanotify_fid_inode(mask, data, data_type, dir, 731 fid_mode); 732 struct inode *dirid = fanotify_dfid_inode(mask, data, data_type, dir); 733 const struct path *path = fsnotify_data_path(data, data_type); 734 struct mem_cgroup *old_memcg; 735 struct dentry *moved = NULL; 736 struct inode *child = NULL; 737 bool name_event = false; 738 unsigned int hash = 0; 739 bool ondir = mask & FAN_ONDIR; 740 struct pid *pid; 741 742 if ((fid_mode & FAN_REPORT_DIR_FID) && dirid) { 743 /* 744 * For certain events and group flags, report the child fid 745 * in addition to reporting the parent fid and maybe child name. 746 */ 747 if (fanotify_report_child_fid(fid_mode, mask) && id != dirid) 748 child = id; 749 750 id = dirid; 751 752 /* 753 * We record file name only in a group with FAN_REPORT_NAME 754 * and when we have a directory inode to report. 755 * 756 * For directory entry modification event, we record the fid of 757 * the directory and the name of the modified entry. 758 * 759 * For event on non-directory that is reported to parent, we 760 * record the fid of the parent and the name of the child. 761 * 762 * Even if not reporting name, we need a variable length 763 * fanotify_name_event if reporting both parent and child fids. 764 */ 765 if (!(fid_mode & FAN_REPORT_NAME)) { 766 name_event = !!child; 767 file_name = NULL; 768 } else if ((mask & ALL_FSNOTIFY_DIRENT_EVENTS) || !ondir) { 769 name_event = true; 770 } 771 772 /* 773 * In the special case of FAN_RENAME event, use the match_mask 774 * to determine if we need to report only the old parent+name, 775 * only the new parent+name or both. 776 * 'dirid' and 'file_name' are the old parent+name and 777 * 'moved' has the new parent+name. 778 */ 779 if (mask & FAN_RENAME) { 780 bool report_old, report_new; 781 782 if (WARN_ON_ONCE(!match_mask)) 783 return NULL; 784 785 /* Report both old and new parent+name if sb watching */ 786 report_old = report_new = 787 match_mask & (1U << FSNOTIFY_ITER_TYPE_SB); 788 report_old |= 789 match_mask & (1U << FSNOTIFY_ITER_TYPE_INODE); 790 report_new |= 791 match_mask & (1U << FSNOTIFY_ITER_TYPE_INODE2); 792 793 if (!report_old) { 794 /* Do not report old parent+name */ 795 dirid = NULL; 796 file_name = NULL; 797 } 798 if (report_new) { 799 /* Report new parent+name */ 800 moved = fsnotify_data_dentry(data, data_type); 801 } 802 } 803 } 804 805 /* 806 * For queues with unlimited length lost events are not expected and 807 * can possibly have security implications. Avoid losing events when 808 * memory is short. For the limited size queues, avoid OOM killer in the 809 * target monitoring memcg as it may have security repercussion. 810 */ 811 if (group->max_events == UINT_MAX) 812 gfp |= __GFP_NOFAIL; 813 else 814 gfp |= __GFP_RETRY_MAYFAIL; 815 816 /* Whoever is interested in the event, pays for the allocation. */ 817 old_memcg = set_active_memcg(group->memcg); 818 819 if (fanotify_is_perm_event(mask)) { 820 event = fanotify_alloc_perm_event(data, data_type, gfp); 821 } else if (fanotify_is_error_event(mask)) { 822 event = fanotify_alloc_error_event(group, fsid, data, 823 data_type, &hash); 824 } else if (name_event && (file_name || moved || child)) { 825 event = fanotify_alloc_name_event(dirid, fsid, file_name, child, 826 moved, &hash, gfp); 827 } else if (fid_mode) { 828 event = fanotify_alloc_fid_event(id, fsid, &hash, gfp); 829 } else { 830 event = fanotify_alloc_path_event(path, &hash, gfp); 831 } 832 833 if (!event) 834 goto out; 835 836 if (FAN_GROUP_FLAG(group, FAN_REPORT_TID)) 837 pid = get_pid(task_pid(current)); 838 else 839 pid = get_pid(task_tgid(current)); 840 841 /* Mix event info, FAN_ONDIR flag and pid into event merge key */ 842 hash ^= hash_long((unsigned long)pid | ondir, FANOTIFY_EVENT_HASH_BITS); 843 fanotify_init_event(event, hash, mask); 844 event->pid = pid; 845 846 out: 847 set_active_memcg(old_memcg); 848 return event; 849 } 850 851 /* 852 * Get cached fsid of the filesystem containing the object from any mark. 853 * All marks are supposed to have the same fsid, but we do not verify that here. 854 */ 855 static __kernel_fsid_t fanotify_get_fsid(struct fsnotify_iter_info *iter_info) 856 { 857 struct fsnotify_mark *mark; 858 int type; 859 __kernel_fsid_t fsid = {}; 860 861 fsnotify_foreach_iter_mark_type(iter_info, mark, type) { 862 if (!(mark->flags & FSNOTIFY_MARK_FLAG_HAS_FSID)) 863 continue; 864 fsid = FANOTIFY_MARK(mark)->fsid; 865 if (!(mark->flags & FSNOTIFY_MARK_FLAG_WEAK_FSID) && 866 WARN_ON_ONCE(!fsid.val[0] && !fsid.val[1])) 867 continue; 868 return fsid; 869 } 870 871 return fsid; 872 } 873 874 /* 875 * Add an event to hash table for faster merge. 876 */ 877 static void fanotify_insert_event(struct fsnotify_group *group, 878 struct fsnotify_event *fsn_event) 879 { 880 struct fanotify_event *event = FANOTIFY_E(fsn_event); 881 unsigned int bucket = fanotify_event_hash_bucket(group, event); 882 struct hlist_head *hlist = &group->fanotify_data.merge_hash[bucket]; 883 884 assert_spin_locked(&group->notification_lock); 885 886 if (!fanotify_is_hashed_event(event->mask)) 887 return; 888 889 pr_debug("%s: group=%p event=%p bucket=%u\n", __func__, 890 group, event, bucket); 891 892 hlist_add_head(&event->merge_list, hlist); 893 } 894 895 static int fanotify_handle_event(struct fsnotify_group *group, u32 mask, 896 const void *data, int data_type, 897 struct inode *dir, 898 const struct qstr *file_name, u32 cookie, 899 struct fsnotify_iter_info *iter_info) 900 { 901 int ret = 0; 902 struct fanotify_event *event; 903 struct fsnotify_event *fsn_event; 904 __kernel_fsid_t fsid = {}; 905 u32 match_mask = 0; 906 907 BUILD_BUG_ON(FAN_ACCESS != FS_ACCESS); 908 BUILD_BUG_ON(FAN_MODIFY != FS_MODIFY); 909 BUILD_BUG_ON(FAN_ATTRIB != FS_ATTRIB); 910 BUILD_BUG_ON(FAN_CLOSE_NOWRITE != FS_CLOSE_NOWRITE); 911 BUILD_BUG_ON(FAN_CLOSE_WRITE != FS_CLOSE_WRITE); 912 BUILD_BUG_ON(FAN_OPEN != FS_OPEN); 913 BUILD_BUG_ON(FAN_MOVED_TO != FS_MOVED_TO); 914 BUILD_BUG_ON(FAN_MOVED_FROM != FS_MOVED_FROM); 915 BUILD_BUG_ON(FAN_CREATE != FS_CREATE); 916 BUILD_BUG_ON(FAN_DELETE != FS_DELETE); 917 BUILD_BUG_ON(FAN_DELETE_SELF != FS_DELETE_SELF); 918 BUILD_BUG_ON(FAN_MOVE_SELF != FS_MOVE_SELF); 919 BUILD_BUG_ON(FAN_EVENT_ON_CHILD != FS_EVENT_ON_CHILD); 920 BUILD_BUG_ON(FAN_Q_OVERFLOW != FS_Q_OVERFLOW); 921 BUILD_BUG_ON(FAN_OPEN_PERM != FS_OPEN_PERM); 922 BUILD_BUG_ON(FAN_ACCESS_PERM != FS_ACCESS_PERM); 923 BUILD_BUG_ON(FAN_ONDIR != FS_ISDIR); 924 BUILD_BUG_ON(FAN_OPEN_EXEC != FS_OPEN_EXEC); 925 BUILD_BUG_ON(FAN_OPEN_EXEC_PERM != FS_OPEN_EXEC_PERM); 926 BUILD_BUG_ON(FAN_FS_ERROR != FS_ERROR); 927 BUILD_BUG_ON(FAN_RENAME != FS_RENAME); 928 BUILD_BUG_ON(FAN_PRE_ACCESS != FS_PRE_ACCESS); 929 930 BUILD_BUG_ON(HWEIGHT32(ALL_FANOTIFY_EVENT_BITS) != 22); 931 932 mask = fanotify_group_event_mask(group, iter_info, &match_mask, 933 mask, data, data_type, dir); 934 if (!mask) 935 return 0; 936 937 pr_debug("%s: group=%p mask=%x report_mask=%x\n", __func__, 938 group, mask, match_mask); 939 940 if (fanotify_is_perm_event(mask)) { 941 /* 942 * fsnotify_prepare_user_wait() fails if we race with mark 943 * deletion. Just let the operation pass in that case. 944 */ 945 if (!fsnotify_prepare_user_wait(iter_info)) 946 return 0; 947 } 948 949 if (FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS)) 950 fsid = fanotify_get_fsid(iter_info); 951 952 event = fanotify_alloc_event(group, mask, data, data_type, dir, 953 file_name, &fsid, match_mask); 954 ret = -ENOMEM; 955 if (unlikely(!event)) { 956 /* 957 * We don't queue overflow events for permission events as 958 * there the access is denied and so no event is in fact lost. 959 */ 960 if (!fanotify_is_perm_event(mask)) 961 fsnotify_queue_overflow(group); 962 goto finish; 963 } 964 965 fsn_event = &event->fse; 966 ret = fsnotify_insert_event(group, fsn_event, fanotify_merge, 967 fanotify_insert_event); 968 if (ret) { 969 /* Permission events shouldn't be merged */ 970 BUG_ON(ret == 1 && mask & FANOTIFY_PERM_EVENTS); 971 /* Our event wasn't used in the end. Free it. */ 972 fsnotify_destroy_event(group, fsn_event); 973 974 ret = 0; 975 } else if (fanotify_is_perm_event(mask)) { 976 ret = fanotify_get_response(group, FANOTIFY_PERM(event), 977 iter_info); 978 } 979 finish: 980 if (fanotify_is_perm_event(mask)) 981 fsnotify_finish_user_wait(iter_info); 982 983 return ret; 984 } 985 986 static void fanotify_free_group_priv(struct fsnotify_group *group) 987 { 988 kfree(group->fanotify_data.merge_hash); 989 if (group->fanotify_data.ucounts) 990 dec_ucount(group->fanotify_data.ucounts, 991 UCOUNT_FANOTIFY_GROUPS); 992 993 if (mempool_initialized(&group->fanotify_data.error_events_pool)) 994 mempool_exit(&group->fanotify_data.error_events_pool); 995 } 996 997 static void fanotify_free_path_event(struct fanotify_event *event) 998 { 999 path_put(fanotify_event_path(event)); 1000 kmem_cache_free(fanotify_path_event_cachep, FANOTIFY_PE(event)); 1001 } 1002 1003 static void fanotify_free_perm_event(struct fanotify_event *event) 1004 { 1005 path_put(fanotify_event_path(event)); 1006 kmem_cache_free(fanotify_perm_event_cachep, FANOTIFY_PERM(event)); 1007 } 1008 1009 static void fanotify_free_fid_event(struct fanotify_event *event) 1010 { 1011 struct fanotify_fid_event *ffe = FANOTIFY_FE(event); 1012 1013 if (fanotify_fh_has_ext_buf(&ffe->object_fh)) 1014 kfree(fanotify_fh_ext_buf(&ffe->object_fh)); 1015 kmem_cache_free(fanotify_fid_event_cachep, ffe); 1016 } 1017 1018 static void fanotify_free_name_event(struct fanotify_event *event) 1019 { 1020 kfree(FANOTIFY_NE(event)); 1021 } 1022 1023 static void fanotify_free_error_event(struct fsnotify_group *group, 1024 struct fanotify_event *event) 1025 { 1026 struct fanotify_error_event *fee = FANOTIFY_EE(event); 1027 1028 mempool_free(fee, &group->fanotify_data.error_events_pool); 1029 } 1030 1031 static void fanotify_free_event(struct fsnotify_group *group, 1032 struct fsnotify_event *fsn_event) 1033 { 1034 struct fanotify_event *event; 1035 1036 event = FANOTIFY_E(fsn_event); 1037 put_pid(event->pid); 1038 switch (event->type) { 1039 case FANOTIFY_EVENT_TYPE_PATH: 1040 fanotify_free_path_event(event); 1041 break; 1042 case FANOTIFY_EVENT_TYPE_PATH_PERM: 1043 fanotify_free_perm_event(event); 1044 break; 1045 case FANOTIFY_EVENT_TYPE_FID: 1046 fanotify_free_fid_event(event); 1047 break; 1048 case FANOTIFY_EVENT_TYPE_FID_NAME: 1049 fanotify_free_name_event(event); 1050 break; 1051 case FANOTIFY_EVENT_TYPE_OVERFLOW: 1052 kfree(event); 1053 break; 1054 case FANOTIFY_EVENT_TYPE_FS_ERROR: 1055 fanotify_free_error_event(group, event); 1056 break; 1057 default: 1058 WARN_ON_ONCE(1); 1059 } 1060 } 1061 1062 static void fanotify_freeing_mark(struct fsnotify_mark *mark, 1063 struct fsnotify_group *group) 1064 { 1065 if (!FAN_GROUP_FLAG(group, FAN_UNLIMITED_MARKS)) 1066 dec_ucount(group->fanotify_data.ucounts, UCOUNT_FANOTIFY_MARKS); 1067 } 1068 1069 static void fanotify_free_mark(struct fsnotify_mark *fsn_mark) 1070 { 1071 kmem_cache_free(fanotify_mark_cache, FANOTIFY_MARK(fsn_mark)); 1072 } 1073 1074 const struct fsnotify_ops fanotify_fsnotify_ops = { 1075 .handle_event = fanotify_handle_event, 1076 .free_group_priv = fanotify_free_group_priv, 1077 .free_event = fanotify_free_event, 1078 .freeing_mark = fanotify_freeing_mark, 1079 .free_mark = fanotify_free_mark, 1080 }; 1081