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
  20static 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
  25static 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
  31static 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
  37static 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
  47static 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
  57static 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
  68static 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
 103static 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
 119static 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
 129static 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! */
 180static 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 */
 222static 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;
 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	default:
 266		ret = -EPERM;
 267	}
 268
 269	/* Check if the response should be audited */
 270	if (event->response & FAN_AUDIT)
 271		audit_fanotify(event->response & ~FAN_AUDIT,
 272			       &event->audit_rule);
 273
 274	pr_debug("%s: group=%p event=%p about to return ret=%d\n", __func__,
 275		 group, event, ret);
 276out:
 277	fsnotify_destroy_event(group, &event->fae.fse);
 278
 279	return ret;
 280}
 
 281
 282/*
 283 * This function returns a mask for an event that only contains the flags
 284 * that have been specifically requested by the user. Flags that may have
 285 * been included within the event mask, but have not been explicitly
 286 * requested by the user, will not be present in the returned mask.
 287 */
 288static u32 fanotify_group_event_mask(struct fsnotify_group *group,
 289				     struct fsnotify_iter_info *iter_info,
 290				     u32 *match_mask, u32 event_mask,
 291				     const void *data, int data_type,
 292				     struct inode *dir)
 293{
 294	__u32 marks_mask = 0, marks_ignore_mask = 0;
 295	__u32 test_mask, user_mask = FANOTIFY_OUTGOING_EVENTS |
 296				     FANOTIFY_EVENT_FLAGS;
 297	const struct path *path = fsnotify_data_path(data, data_type);
 298	unsigned int fid_mode = FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS);
 299	struct fsnotify_mark *mark;
 300	bool ondir = event_mask & FAN_ONDIR;
 301	int type;
 302
 303	pr_debug("%s: report_mask=%x mask=%x data=%p data_type=%d\n",
 304		 __func__, iter_info->report_mask, event_mask, data, data_type);
 305
 306	if (!fid_mode) {
 307		/* Do we have path to open a file descriptor? */
 308		if (!path)
 309			return 0;
 310		/* Path type events are only relevant for files and dirs */
 311		if (!d_is_reg(path->dentry) && !d_can_lookup(path->dentry))
 312			return 0;
 313	} else if (!(fid_mode & FAN_REPORT_FID)) {
 314		/* Do we have a directory inode to report? */
 315		if (!dir && !ondir)
 316			return 0;
 317	}
 318
 319	fsnotify_foreach_iter_mark_type(iter_info, mark, type) {
 320		/*
 321		 * Apply ignore mask depending on event flags in ignore mask.
 322		 */
 323		marks_ignore_mask |=
 324			fsnotify_effective_ignore_mask(mark, ondir, type);
 325
 326		/*
 327		 * Send the event depending on event flags in mark mask.
 328		 */
 329		if (!fsnotify_mask_applicable(mark->mask, ondir, type))
 330			continue;
 331
 332		marks_mask |= mark->mask;
 333
 334		/* Record the mark types of this group that matched the event */
 335		*match_mask |= 1U << type;
 336	}
 337
 338	test_mask = event_mask & marks_mask & ~marks_ignore_mask;
 339
 340	/*
 341	 * For dirent modification events (create/delete/move) that do not carry
 342	 * the child entry name information, we report FAN_ONDIR for mkdir/rmdir
 343	 * so user can differentiate them from creat/unlink.
 344	 *
 345	 * For backward compatibility and consistency, do not report FAN_ONDIR
 346	 * to user in legacy fanotify mode (reporting fd) and report FAN_ONDIR
 347	 * to user in fid mode for all event types.
 348	 *
 349	 * We never report FAN_EVENT_ON_CHILD to user, but we do pass it in to
 350	 * fanotify_alloc_event() when group is reporting fid as indication
 351	 * that event happened on child.
 352	 */
 353	if (fid_mode) {
 354		/* Do not report event flags without any event */
 355		if (!(test_mask & ~FANOTIFY_EVENT_FLAGS))
 356			return 0;
 357	} else {
 358		user_mask &= ~FANOTIFY_EVENT_FLAGS;
 359	}
 360
 361	return test_mask & user_mask;
 362}
 363
 364/*
 365 * Check size needed to encode fanotify_fh.
 366 *
 367 * Return size of encoded fh without fanotify_fh header.
 368 * Return 0 on failure to encode.
 369 */
 370static int fanotify_encode_fh_len(struct inode *inode)
 371{
 372	int dwords = 0;
 373	int fh_len;
 374
 375	if (!inode)
 376		return 0;
 377
 378	exportfs_encode_fid(inode, NULL, &dwords);
 379	fh_len = dwords << 2;
 380
 381	/*
 382	 * struct fanotify_error_event might be preallocated and is
 383	 * limited to MAX_HANDLE_SZ.  This should never happen, but
 384	 * safeguard by forcing an invalid file handle.
 385	 */
 386	if (WARN_ON_ONCE(fh_len > MAX_HANDLE_SZ))
 387		return 0;
 388
 389	return fh_len;
 390}
 391
 392/*
 393 * Encode fanotify_fh.
 394 *
 395 * Return total size of encoded fh including fanotify_fh header.
 396 * Return 0 on failure to encode.
 397 */
 398static int fanotify_encode_fh(struct fanotify_fh *fh, struct inode *inode,
 399			      unsigned int fh_len, unsigned int *hash,
 400			      gfp_t gfp)
 401{
 402	int dwords, type = 0;
 403	char *ext_buf = NULL;
 404	void *buf = fh->buf;
 405	int err;
 406
 407	fh->type = FILEID_ROOT;
 408	fh->len = 0;
 409	fh->flags = 0;
 410
 411	/*
 412	 * Invalid FHs are used by FAN_FS_ERROR for errors not
 413	 * linked to any inode. The f_handle won't be reported
 414	 * back to userspace.
 415	 */
 416	if (!inode)
 417		goto out;
 418
 419	/*
 420	 * !gpf means preallocated variable size fh, but fh_len could
 421	 * be zero in that case if encoding fh len failed.
 422	 */
 423	err = -ENOENT;
 424	if (fh_len < 4 || WARN_ON_ONCE(fh_len % 4) || fh_len > MAX_HANDLE_SZ)
 425		goto out_err;
 426
 427	/* No external buffer in a variable size allocated fh */
 428	if (gfp && fh_len > FANOTIFY_INLINE_FH_LEN) {
 429		/* Treat failure to allocate fh as failure to encode fh */
 430		err = -ENOMEM;
 431		ext_buf = kmalloc(fh_len, gfp);
 432		if (!ext_buf)
 433			goto out_err;
 434
 435		*fanotify_fh_ext_buf_ptr(fh) = ext_buf;
 436		buf = ext_buf;
 437		fh->flags |= FANOTIFY_FH_FLAG_EXT_BUF;
 438	}
 439
 440	dwords = fh_len >> 2;
 441	type = exportfs_encode_fid(inode, buf, &dwords);
 442	err = -EINVAL;
 443	if (type <= 0 || type == FILEID_INVALID || fh_len != dwords << 2)
 444		goto out_err;
 445
 446	fh->type = type;
 447	fh->len = fh_len;
 448
 449out:
 450	/*
 451	 * Mix fh into event merge key.  Hash might be NULL in case of
 452	 * unhashed FID events (i.e. FAN_FS_ERROR).
 453	 */
 454	if (hash)
 455		*hash ^= fanotify_hash_fh(fh);
 456
 457	return FANOTIFY_FH_HDR_LEN + fh_len;
 458
 459out_err:
 460	pr_warn_ratelimited("fanotify: failed to encode fid (type=%d, len=%d, err=%i)\n",
 461			    type, fh_len, err);
 462	kfree(ext_buf);
 463	*fanotify_fh_ext_buf_ptr(fh) = NULL;
 464	/* Report the event without a file identifier on encode error */
 465	fh->type = FILEID_INVALID;
 466	fh->len = 0;
 467	return 0;
 468}
 469
 470/*
 471 * FAN_REPORT_FID is ambiguous in that it reports the fid of the child for
 472 * some events and the fid of the parent for create/delete/move events.
 473 *
 474 * With the FAN_REPORT_TARGET_FID flag, the fid of the child is reported
 475 * also in create/delete/move events in addition to the fid of the parent
 476 * and the name of the child.
 477 */
 478static inline bool fanotify_report_child_fid(unsigned int fid_mode, u32 mask)
 479{
 480	if (mask & ALL_FSNOTIFY_DIRENT_EVENTS)
 481		return (fid_mode & FAN_REPORT_TARGET_FID);
 482
 483	return (fid_mode & FAN_REPORT_FID) && !(mask & FAN_ONDIR);
 484}
 485
 486/*
 487 * The inode to use as identifier when reporting fid depends on the event
 488 * and the group flags.
 489 *
 490 * With the group flag FAN_REPORT_TARGET_FID, always report the child fid.
 491 *
 492 * Without the group flag FAN_REPORT_TARGET_FID, report the modified directory
 493 * fid on dirent events and the child fid otherwise.
 494 *
 495 * For example:
 496 * FS_ATTRIB reports the child fid even if reported on a watched parent.
 497 * FS_CREATE reports the modified dir fid without FAN_REPORT_TARGET_FID.
 498 *       and reports the created child fid with FAN_REPORT_TARGET_FID.
 499 */
 500static struct inode *fanotify_fid_inode(u32 event_mask, const void *data,
 501					int data_type, struct inode *dir,
 502					unsigned int fid_mode)
 503{
 504	if ((event_mask & ALL_FSNOTIFY_DIRENT_EVENTS) &&
 505	    !(fid_mode & FAN_REPORT_TARGET_FID))
 506		return dir;
 507
 508	return fsnotify_data_inode(data, data_type);
 509}
 510
 511/*
 512 * The inode to use as identifier when reporting dir fid depends on the event.
 513 * Report the modified directory inode on dirent modification events.
 514 * Report the "victim" inode if "victim" is a directory.
 515 * Report the parent inode if "victim" is not a directory and event is
 516 * reported to parent.
 517 * Otherwise, do not report dir fid.
 518 */
 519static struct inode *fanotify_dfid_inode(u32 event_mask, const void *data,
 520					 int data_type, struct inode *dir)
 521{
 522	struct inode *inode = fsnotify_data_inode(data, data_type);
 523
 524	if (event_mask & ALL_FSNOTIFY_DIRENT_EVENTS)
 525		return dir;
 526
 527	if (inode && S_ISDIR(inode->i_mode))
 528		return inode;
 529
 530	return dir;
 531}
 532
 533static struct fanotify_event *fanotify_alloc_path_event(const struct path *path,
 534							unsigned int *hash,
 535							gfp_t gfp)
 536{
 537	struct fanotify_path_event *pevent;
 538
 539	pevent = kmem_cache_alloc(fanotify_path_event_cachep, gfp);
 540	if (!pevent)
 541		return NULL;
 542
 543	pevent->fae.type = FANOTIFY_EVENT_TYPE_PATH;
 544	pevent->path = *path;
 545	*hash ^= fanotify_hash_path(path);
 546	path_get(path);
 547
 548	return &pevent->fae;
 549}
 550
 551static struct fanotify_event *fanotify_alloc_perm_event(const struct path *path,
 552							gfp_t gfp)
 553{
 554	struct fanotify_perm_event *pevent;
 555
 556	pevent = kmem_cache_alloc(fanotify_perm_event_cachep, gfp);
 557	if (!pevent)
 558		return NULL;
 559
 560	pevent->fae.type = FANOTIFY_EVENT_TYPE_PATH_PERM;
 561	pevent->response = 0;
 562	pevent->hdr.type = FAN_RESPONSE_INFO_NONE;
 563	pevent->hdr.pad = 0;
 564	pevent->hdr.len = 0;
 565	pevent->state = FAN_EVENT_INIT;
 566	pevent->path = *path;
 567	path_get(path);
 568
 569	return &pevent->fae;
 570}
 571
 572static struct fanotify_event *fanotify_alloc_fid_event(struct inode *id,
 573						       __kernel_fsid_t *fsid,
 574						       unsigned int *hash,
 575						       gfp_t gfp)
 576{
 577	struct fanotify_fid_event *ffe;
 578
 579	ffe = kmem_cache_alloc(fanotify_fid_event_cachep, gfp);
 580	if (!ffe)
 581		return NULL;
 582
 583	ffe->fae.type = FANOTIFY_EVENT_TYPE_FID;
 584	ffe->fsid = *fsid;
 585	*hash ^= fanotify_hash_fsid(fsid);
 586	fanotify_encode_fh(&ffe->object_fh, id, fanotify_encode_fh_len(id),
 587			   hash, gfp);
 588
 589	return &ffe->fae;
 590}
 591
 592static struct fanotify_event *fanotify_alloc_name_event(struct inode *dir,
 593							__kernel_fsid_t *fsid,
 594							const struct qstr *name,
 595							struct inode *child,
 596							struct dentry *moved,
 597							unsigned int *hash,
 598							gfp_t gfp)
 599{
 600	struct fanotify_name_event *fne;
 601	struct fanotify_info *info;
 602	struct fanotify_fh *dfh, *ffh;
 603	struct inode *dir2 = moved ? d_inode(moved->d_parent) : NULL;
 604	const struct qstr *name2 = moved ? &moved->d_name : NULL;
 605	unsigned int dir_fh_len = fanotify_encode_fh_len(dir);
 606	unsigned int dir2_fh_len = fanotify_encode_fh_len(dir2);
 607	unsigned int child_fh_len = fanotify_encode_fh_len(child);
 608	unsigned long name_len = name ? name->len : 0;
 609	unsigned long name2_len = name2 ? name2->len : 0;
 610	unsigned int len, size;
 611
 612	/* Reserve terminating null byte even for empty name */
 613	size = sizeof(*fne) + name_len + name2_len + 2;
 614	if (dir_fh_len)
 615		size += FANOTIFY_FH_HDR_LEN + dir_fh_len;
 616	if (dir2_fh_len)
 617		size += FANOTIFY_FH_HDR_LEN + dir2_fh_len;
 618	if (child_fh_len)
 619		size += FANOTIFY_FH_HDR_LEN + child_fh_len;
 620	fne = kmalloc(size, gfp);
 621	if (!fne)
 622		return NULL;
 623
 624	fne->fae.type = FANOTIFY_EVENT_TYPE_FID_NAME;
 625	fne->fsid = *fsid;
 626	*hash ^= fanotify_hash_fsid(fsid);
 627	info = &fne->info;
 628	fanotify_info_init(info);
 629	if (dir_fh_len) {
 630		dfh = fanotify_info_dir_fh(info);
 631		len = fanotify_encode_fh(dfh, dir, dir_fh_len, hash, 0);
 632		fanotify_info_set_dir_fh(info, len);
 633	}
 634	if (dir2_fh_len) {
 635		dfh = fanotify_info_dir2_fh(info);
 636		len = fanotify_encode_fh(dfh, dir2, dir2_fh_len, hash, 0);
 637		fanotify_info_set_dir2_fh(info, len);
 638	}
 639	if (child_fh_len) {
 640		ffh = fanotify_info_file_fh(info);
 641		len = fanotify_encode_fh(ffh, child, child_fh_len, hash, 0);
 642		fanotify_info_set_file_fh(info, len);
 643	}
 644	if (name_len) {
 645		fanotify_info_copy_name(info, name);
 646		*hash ^= full_name_hash((void *)name_len, name->name, name_len);
 647	}
 648	if (name2_len) {
 649		fanotify_info_copy_name2(info, name2);
 650		*hash ^= full_name_hash((void *)name2_len, name2->name,
 651					name2_len);
 652	}
 653
 654	pr_debug("%s: size=%u dir_fh_len=%u child_fh_len=%u name_len=%u name='%.*s'\n",
 655		 __func__, size, dir_fh_len, child_fh_len,
 656		 info->name_len, info->name_len, fanotify_info_name(info));
 657
 658	if (dir2_fh_len) {
 659		pr_debug("%s: dir2_fh_len=%u name2_len=%u name2='%.*s'\n",
 660			 __func__, dir2_fh_len, info->name2_len,
 661			 info->name2_len, fanotify_info_name2(info));
 662	}
 663
 664	return &fne->fae;
 665}
 666
 667static struct fanotify_event *fanotify_alloc_error_event(
 668						struct fsnotify_group *group,
 669						__kernel_fsid_t *fsid,
 670						const void *data, int data_type,
 671						unsigned int *hash)
 672{
 673	struct fs_error_report *report =
 674			fsnotify_data_error_report(data, data_type);
 675	struct inode *inode;
 676	struct fanotify_error_event *fee;
 677	int fh_len;
 678
 679	if (WARN_ON_ONCE(!report))
 680		return NULL;
 681
 682	fee = mempool_alloc(&group->fanotify_data.error_events_pool, GFP_NOFS);
 683	if (!fee)
 684		return NULL;
 685
 686	fee->fae.type = FANOTIFY_EVENT_TYPE_FS_ERROR;
 687	fee->error = report->error;
 688	fee->err_count = 1;
 689	fee->fsid = *fsid;
 690
 691	inode = report->inode;
 692	fh_len = fanotify_encode_fh_len(inode);
 693
 694	/* Bad fh_len. Fallback to using an invalid fh. Should never happen. */
 695	if (!fh_len && inode)
 696		inode = NULL;
 697
 698	fanotify_encode_fh(&fee->object_fh, inode, fh_len, NULL, 0);
 699
 700	*hash ^= fanotify_hash_fsid(fsid);
 701
 702	return &fee->fae;
 703}
 704
 705static struct fanotify_event *fanotify_alloc_event(
 706				struct fsnotify_group *group,
 707				u32 mask, const void *data, int data_type,
 708				struct inode *dir, const struct qstr *file_name,
 709				__kernel_fsid_t *fsid, u32 match_mask)
 710{
 711	struct fanotify_event *event = NULL;
 712	gfp_t gfp = GFP_KERNEL_ACCOUNT;
 713	unsigned int fid_mode = FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS);
 714	struct inode *id = fanotify_fid_inode(mask, data, data_type, dir,
 715					      fid_mode);
 716	struct inode *dirid = fanotify_dfid_inode(mask, data, data_type, dir);
 717	const struct path *path = fsnotify_data_path(data, data_type);
 718	struct mem_cgroup *old_memcg;
 719	struct dentry *moved = NULL;
 720	struct inode *child = NULL;
 721	bool name_event = false;
 722	unsigned int hash = 0;
 723	bool ondir = mask & FAN_ONDIR;
 724	struct pid *pid;
 725
 726	if ((fid_mode & FAN_REPORT_DIR_FID) && dirid) {
 727		/*
 728		 * For certain events and group flags, report the child fid
 729		 * in addition to reporting the parent fid and maybe child name.
 730		 */
 731		if (fanotify_report_child_fid(fid_mode, mask) && id != dirid)
 732			child = id;
 733
 734		id = dirid;
 735
 736		/*
 737		 * We record file name only in a group with FAN_REPORT_NAME
 738		 * and when we have a directory inode to report.
 739		 *
 740		 * For directory entry modification event, we record the fid of
 741		 * the directory and the name of the modified entry.
 742		 *
 743		 * For event on non-directory that is reported to parent, we
 744		 * record the fid of the parent and the name of the child.
 745		 *
 746		 * Even if not reporting name, we need a variable length
 747		 * fanotify_name_event if reporting both parent and child fids.
 748		 */
 749		if (!(fid_mode & FAN_REPORT_NAME)) {
 750			name_event = !!child;
 751			file_name = NULL;
 752		} else if ((mask & ALL_FSNOTIFY_DIRENT_EVENTS) || !ondir) {
 753			name_event = true;
 754		}
 755
 756		/*
 757		 * In the special case of FAN_RENAME event, use the match_mask
 758		 * to determine if we need to report only the old parent+name,
 759		 * only the new parent+name or both.
 760		 * 'dirid' and 'file_name' are the old parent+name and
 761		 * 'moved' has the new parent+name.
 762		 */
 763		if (mask & FAN_RENAME) {
 764			bool report_old, report_new;
 765
 766			if (WARN_ON_ONCE(!match_mask))
 767				return NULL;
 768
 769			/* Report both old and new parent+name if sb watching */
 770			report_old = report_new =
 771				match_mask & (1U << FSNOTIFY_ITER_TYPE_SB);
 772			report_old |=
 773				match_mask & (1U << FSNOTIFY_ITER_TYPE_INODE);
 774			report_new |=
 775				match_mask & (1U << FSNOTIFY_ITER_TYPE_INODE2);
 776
 777			if (!report_old) {
 778				/* Do not report old parent+name */
 779				dirid = NULL;
 780				file_name = NULL;
 781			}
 782			if (report_new) {
 783				/* Report new parent+name */
 784				moved = fsnotify_data_dentry(data, data_type);
 785			}
 786		}
 787	}
 788
 789	/*
 790	 * For queues with unlimited length lost events are not expected and
 791	 * can possibly have security implications. Avoid losing events when
 792	 * memory is short. For the limited size queues, avoid OOM killer in the
 793	 * target monitoring memcg as it may have security repercussion.
 794	 */
 795	if (group->max_events == UINT_MAX)
 796		gfp |= __GFP_NOFAIL;
 797	else
 798		gfp |= __GFP_RETRY_MAYFAIL;
 799
 800	/* Whoever is interested in the event, pays for the allocation. */
 801	old_memcg = set_active_memcg(group->memcg);
 802
 803	if (fanotify_is_perm_event(mask)) {
 804		event = fanotify_alloc_perm_event(path, gfp);
 805	} else if (fanotify_is_error_event(mask)) {
 806		event = fanotify_alloc_error_event(group, fsid, data,
 807						   data_type, &hash);
 808	} else if (name_event && (file_name || moved || child)) {
 809		event = fanotify_alloc_name_event(dirid, fsid, file_name, child,
 810						  moved, &hash, gfp);
 811	} else if (fid_mode) {
 812		event = fanotify_alloc_fid_event(id, fsid, &hash, gfp);
 813	} else {
 814		event = fanotify_alloc_path_event(path, &hash, gfp);
 815	}
 816
 817	if (!event)
 818		goto out;
 819
 820	if (FAN_GROUP_FLAG(group, FAN_REPORT_TID))
 821		pid = get_pid(task_pid(current));
 822	else
 823		pid = get_pid(task_tgid(current));
 824
 825	/* Mix event info, FAN_ONDIR flag and pid into event merge key */
 826	hash ^= hash_long((unsigned long)pid | ondir, FANOTIFY_EVENT_HASH_BITS);
 827	fanotify_init_event(event, hash, mask);
 828	event->pid = pid;
 829
 830out:
 831	set_active_memcg(old_memcg);
 832	return event;
 833}
 834
 835/*
 836 * Get cached fsid of the filesystem containing the object from any mark.
 837 * All marks are supposed to have the same fsid, but we do not verify that here.
 838 */
 839static __kernel_fsid_t fanotify_get_fsid(struct fsnotify_iter_info *iter_info)
 840{
 841	struct fsnotify_mark *mark;
 842	int type;
 843	__kernel_fsid_t fsid = {};
 844
 845	fsnotify_foreach_iter_mark_type(iter_info, mark, type) {
 846		if (!(mark->flags & FSNOTIFY_MARK_FLAG_HAS_FSID))
 847			continue;
 848		fsid = FANOTIFY_MARK(mark)->fsid;
 849		if (!(mark->flags & FSNOTIFY_MARK_FLAG_WEAK_FSID) &&
 850		    WARN_ON_ONCE(!fsid.val[0] && !fsid.val[1]))
 851			continue;
 852		return fsid;
 853	}
 854
 855	return fsid;
 856}
 857
 858/*
 859 * Add an event to hash table for faster merge.
 860 */
 861static void fanotify_insert_event(struct fsnotify_group *group,
 862				  struct fsnotify_event *fsn_event)
 863{
 864	struct fanotify_event *event = FANOTIFY_E(fsn_event);
 865	unsigned int bucket = fanotify_event_hash_bucket(group, event);
 866	struct hlist_head *hlist = &group->fanotify_data.merge_hash[bucket];
 867
 868	assert_spin_locked(&group->notification_lock);
 869
 870	if (!fanotify_is_hashed_event(event->mask))
 871		return;
 872
 873	pr_debug("%s: group=%p event=%p bucket=%u\n", __func__,
 874		 group, event, bucket);
 875
 876	hlist_add_head(&event->merge_list, hlist);
 877}
 878
 879static int fanotify_handle_event(struct fsnotify_group *group, u32 mask,
 880				 const void *data, int data_type,
 881				 struct inode *dir,
 882				 const struct qstr *file_name, u32 cookie,
 883				 struct fsnotify_iter_info *iter_info)
 884{
 885	int ret = 0;
 886	struct fanotify_event *event;
 887	struct fsnotify_event *fsn_event;
 888	__kernel_fsid_t fsid = {};
 889	u32 match_mask = 0;
 890
 891	BUILD_BUG_ON(FAN_ACCESS != FS_ACCESS);
 892	BUILD_BUG_ON(FAN_MODIFY != FS_MODIFY);
 893	BUILD_BUG_ON(FAN_ATTRIB != FS_ATTRIB);
 894	BUILD_BUG_ON(FAN_CLOSE_NOWRITE != FS_CLOSE_NOWRITE);
 895	BUILD_BUG_ON(FAN_CLOSE_WRITE != FS_CLOSE_WRITE);
 896	BUILD_BUG_ON(FAN_OPEN != FS_OPEN);
 897	BUILD_BUG_ON(FAN_MOVED_TO != FS_MOVED_TO);
 898	BUILD_BUG_ON(FAN_MOVED_FROM != FS_MOVED_FROM);
 899	BUILD_BUG_ON(FAN_CREATE != FS_CREATE);
 900	BUILD_BUG_ON(FAN_DELETE != FS_DELETE);
 901	BUILD_BUG_ON(FAN_DELETE_SELF != FS_DELETE_SELF);
 902	BUILD_BUG_ON(FAN_MOVE_SELF != FS_MOVE_SELF);
 903	BUILD_BUG_ON(FAN_EVENT_ON_CHILD != FS_EVENT_ON_CHILD);
 904	BUILD_BUG_ON(FAN_Q_OVERFLOW != FS_Q_OVERFLOW);
 905	BUILD_BUG_ON(FAN_OPEN_PERM != FS_OPEN_PERM);
 906	BUILD_BUG_ON(FAN_ACCESS_PERM != FS_ACCESS_PERM);
 907	BUILD_BUG_ON(FAN_ONDIR != FS_ISDIR);
 908	BUILD_BUG_ON(FAN_OPEN_EXEC != FS_OPEN_EXEC);
 909	BUILD_BUG_ON(FAN_OPEN_EXEC_PERM != FS_OPEN_EXEC_PERM);
 910	BUILD_BUG_ON(FAN_FS_ERROR != FS_ERROR);
 911	BUILD_BUG_ON(FAN_RENAME != FS_RENAME);
 912
 913	BUILD_BUG_ON(HWEIGHT32(ALL_FANOTIFY_EVENT_BITS) != 21);
 914
 915	mask = fanotify_group_event_mask(group, iter_info, &match_mask,
 916					 mask, data, data_type, dir);
 917	if (!mask)
 918		return 0;
 919
 920	pr_debug("%s: group=%p mask=%x report_mask=%x\n", __func__,
 921		 group, mask, match_mask);
 922
 923	if (fanotify_is_perm_event(mask)) {
 924		/*
 925		 * fsnotify_prepare_user_wait() fails if we race with mark
 926		 * deletion.  Just let the operation pass in that case.
 927		 */
 928		if (!fsnotify_prepare_user_wait(iter_info))
 929			return 0;
 930	}
 931
 932	if (FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS))
 933		fsid = fanotify_get_fsid(iter_info);
 934
 935	event = fanotify_alloc_event(group, mask, data, data_type, dir,
 936				     file_name, &fsid, match_mask);
 937	ret = -ENOMEM;
 938	if (unlikely(!event)) {
 939		/*
 940		 * We don't queue overflow events for permission events as
 941		 * there the access is denied and so no event is in fact lost.
 942		 */
 943		if (!fanotify_is_perm_event(mask))
 944			fsnotify_queue_overflow(group);
 945		goto finish;
 946	}
 
 947
 948	fsn_event = &event->fse;
 949	ret = fsnotify_insert_event(group, fsn_event, fanotify_merge,
 950				    fanotify_insert_event);
 951	if (ret) {
 952		/* Permission events shouldn't be merged */
 953		BUG_ON(ret == 1 && mask & FANOTIFY_PERM_EVENTS);
 954		/* Our event wasn't used in the end. Free it. */
 955		fsnotify_destroy_event(group, fsn_event);
 956
 957		ret = 0;
 958	} else if (fanotify_is_perm_event(mask)) {
 959		ret = fanotify_get_response(group, FANOTIFY_PERM(event),
 960					    iter_info);
 961	}
 962finish:
 963	if (fanotify_is_perm_event(mask))
 964		fsnotify_finish_user_wait(iter_info);
 965
 966	return ret;
 967}
 968
 969static void fanotify_free_group_priv(struct fsnotify_group *group)
 970{
 971	kfree(group->fanotify_data.merge_hash);
 972	if (group->fanotify_data.ucounts)
 973		dec_ucount(group->fanotify_data.ucounts,
 974			   UCOUNT_FANOTIFY_GROUPS);
 975
 976	if (mempool_initialized(&group->fanotify_data.error_events_pool))
 977		mempool_exit(&group->fanotify_data.error_events_pool);
 978}
 979
 980static void fanotify_free_path_event(struct fanotify_event *event)
 981{
 982	path_put(fanotify_event_path(event));
 983	kmem_cache_free(fanotify_path_event_cachep, FANOTIFY_PE(event));
 984}
 985
 986static void fanotify_free_perm_event(struct fanotify_event *event)
 987{
 988	path_put(fanotify_event_path(event));
 989	kmem_cache_free(fanotify_perm_event_cachep, FANOTIFY_PERM(event));
 990}
 991
 992static void fanotify_free_fid_event(struct fanotify_event *event)
 993{
 994	struct fanotify_fid_event *ffe = FANOTIFY_FE(event);
 995
 996	if (fanotify_fh_has_ext_buf(&ffe->object_fh))
 997		kfree(fanotify_fh_ext_buf(&ffe->object_fh));
 998	kmem_cache_free(fanotify_fid_event_cachep, ffe);
 999}
1000
1001static void fanotify_free_name_event(struct fanotify_event *event)
1002{
1003	kfree(FANOTIFY_NE(event));
1004}
1005
1006static void fanotify_free_error_event(struct fsnotify_group *group,
1007				      struct fanotify_event *event)
1008{
1009	struct fanotify_error_event *fee = FANOTIFY_EE(event);
1010
1011	mempool_free(fee, &group->fanotify_data.error_events_pool);
1012}
 
1013
1014static void fanotify_free_event(struct fsnotify_group *group,
1015				struct fsnotify_event *fsn_event)
1016{
1017	struct fanotify_event *event;
1018
1019	event = FANOTIFY_E(fsn_event);
1020	put_pid(event->pid);
1021	switch (event->type) {
1022	case FANOTIFY_EVENT_TYPE_PATH:
1023		fanotify_free_path_event(event);
1024		break;
1025	case FANOTIFY_EVENT_TYPE_PATH_PERM:
1026		fanotify_free_perm_event(event);
1027		break;
1028	case FANOTIFY_EVENT_TYPE_FID:
1029		fanotify_free_fid_event(event);
1030		break;
1031	case FANOTIFY_EVENT_TYPE_FID_NAME:
1032		fanotify_free_name_event(event);
1033		break;
1034	case FANOTIFY_EVENT_TYPE_OVERFLOW:
1035		kfree(event);
1036		break;
1037	case FANOTIFY_EVENT_TYPE_FS_ERROR:
1038		fanotify_free_error_event(group, event);
1039		break;
1040	default:
1041		WARN_ON_ONCE(1);
1042	}
1043}
1044
1045static void fanotify_freeing_mark(struct fsnotify_mark *mark,
1046				  struct fsnotify_group *group)
1047{
1048	if (!FAN_GROUP_FLAG(group, FAN_UNLIMITED_MARKS))
1049		dec_ucount(group->fanotify_data.ucounts, UCOUNT_FANOTIFY_MARKS);
1050}
1051
1052static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
1053{
1054	kmem_cache_free(fanotify_mark_cache, FANOTIFY_MARK(fsn_mark));
1055}
1056
1057const struct fsnotify_ops fanotify_fsnotify_ops = {
1058	.handle_event = fanotify_handle_event,
 
1059	.free_group_priv = fanotify_free_group_priv,
1060	.free_event = fanotify_free_event,
1061	.freeing_mark = fanotify_freeing_mark,
1062	.free_mark = fanotify_free_mark,
1063};