fanotify.c

   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	case FANOTIFY_EVENT_TYPE_MNT:
 170		return false;
 171	default:
 172		WARN_ON_ONCE(1);
 173	}
 174
 175	return false;
 176}
 177
 178/* Limit event merges to limit CPU overhead per event */
 179#define FANOTIFY_MAX_MERGE_EVENTS 128
 180
 181/* and the list better be locked by something too! */
 182static int fanotify_merge(struct fsnotify_group *group,
 183			  struct fsnotify_event *event)
 184{
 185	struct fanotify_event *old, *new = FANOTIFY_E(event);
 186	unsigned int bucket = fanotify_event_hash_bucket(group, new);
 187	struct hlist_head *hlist = &group->fanotify_data.merge_hash[bucket];
 188	int i = 0;
 189
 190	pr_debug("%s: group=%p event=%p bucket=%u\n", __func__,
 191		 group, event, bucket);
 192
 193	/*
 194	 * Don't merge a permission event with any other event so that we know
 195	 * the event structure we have created in fanotify_handle_event() is the
 196	 * one we should check for permission response.
 197	 */
 198	if (fanotify_is_perm_event(new->mask))
 199		return 0;
 200
 201	hlist_for_each_entry(old, hlist, merge_list) {
 202		if (++i > FANOTIFY_MAX_MERGE_EVENTS)
 203			break;
 204		if (fanotify_should_merge(old, new)) {
 205			old->mask |= new->mask;
 206
 207			if (fanotify_is_error_event(old->mask))
 208				FANOTIFY_EE(old)->err_count++;
 209
 210			return 1;
 211		}
 212	}
 213
 214	return 0;
 215}
 216
 217/*
 218 * Wait for response to permission event. The function also takes care of
 219 * freeing the permission event (or offloads that in case the wait is canceled
 220 * by a signal). The function returns 0 in case access got allowed by userspace,
 221 * -EPERM in case userspace disallowed the access, and -ERESTARTSYS in case
 222 * the wait got interrupted by a signal.
 223 */
 224static int fanotify_get_response(struct fsnotify_group *group,
 225				 struct fanotify_perm_event *event,
 226				 struct fsnotify_iter_info *iter_info)
 227{
 228	int ret, errno;
 229
 230	pr_debug("%s: group=%p event=%p\n", __func__, group, event);
 231
 232	ret = wait_event_state(group->fanotify_data.access_waitq,
 233				  event->state == FAN_EVENT_ANSWERED,
 234				  (TASK_KILLABLE|TASK_FREEZABLE));
 235
 236	/* Signal pending? */
 237	if (ret < 0) {
 238		spin_lock(&group->notification_lock);
 239		/* Event reported to userspace and no answer yet? */
 240		if (event->state == FAN_EVENT_REPORTED) {
 241			/* Event will get freed once userspace answers to it */
 242			event->state = FAN_EVENT_CANCELED;
 243			spin_unlock(&group->notification_lock);
 244			return ret;
 245		}
 246		/* Event not yet reported? Just remove it. */
 247		if (event->state == FAN_EVENT_INIT) {
 248			fsnotify_remove_queued_event(group, &event->fae.fse);
 249			/* Permission events are not supposed to be hashed */
 250			WARN_ON_ONCE(!hlist_unhashed(&event->fae.merge_list));
 251		}
 252		/*
 253		 * Event may be also answered in case signal delivery raced
 254		 * with wakeup. In that case we have nothing to do besides
 255		 * freeing the event and reporting error.
 256		 */
 257		spin_unlock(&group->notification_lock);
 258		goto out;
 259	}
 260
 261	/* userspace responded, convert to something usable */
 262	switch (event->response & FANOTIFY_RESPONSE_ACCESS) {
 263	case FAN_ALLOW:
 264		ret = 0;
 265		break;
 266	case FAN_DENY:
 267		/* Check custom errno from pre-content events */
 268		errno = fanotify_get_response_errno(event->response);
 269		if (errno) {
 270			ret = -errno;
 271			break;
 272		}
 273		fallthrough;
 274	default:
 275		ret = -EPERM;
 276	}
 277
 278	/* Check if the response should be audited */
 279	if (event->response & FAN_AUDIT) {
 280		u32 response = event->response &
 281			(FANOTIFY_RESPONSE_ACCESS | FANOTIFY_RESPONSE_FLAGS);
 282		audit_fanotify(response & ~FAN_AUDIT, &event->audit_rule);
 283	}
 284
 285	pr_debug("%s: group=%p event=%p about to return ret=%d\n", __func__,
 286		 group, event, ret);
 287out:
 288	fsnotify_destroy_event(group, &event->fae.fse);
 289
 290	return ret;
 291}
 292
 293/*
 294 * This function returns a mask for an event that only contains the flags
 295 * that have been specifically requested by the user. Flags that may have
 296 * been included within the event mask, but have not been explicitly
 297 * requested by the user, will not be present in the returned mask.
 298 */
 299static u32 fanotify_group_event_mask(struct fsnotify_group *group,
 300				     struct fsnotify_iter_info *iter_info,
 301				     u32 *match_mask, u32 event_mask,
 302				     const void *data, int data_type,
 303				     struct inode *dir)
 304{
 305	__u32 marks_mask = 0, marks_ignore_mask = 0;
 306	__u32 test_mask, user_mask = FANOTIFY_OUTGOING_EVENTS |
 307				     FANOTIFY_EVENT_FLAGS;
 308	const struct path *path = fsnotify_data_path(data, data_type);
 309	unsigned int fid_mode = FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS);
 310	struct fsnotify_mark *mark;
 311	bool ondir = event_mask & FAN_ONDIR;
 312	int type;
 313
 314	pr_debug("%s: report_mask=%x mask=%x data=%p data_type=%d\n",
 315		 __func__, iter_info->report_mask, event_mask, data, data_type);
 316
 317	if (FAN_GROUP_FLAG(group, FAN_REPORT_MNT)) {
 318		if (data_type != FSNOTIFY_EVENT_MNT)
 319			return 0;
 320	} else if (!fid_mode) {
 321		/* Do we have path to open a file descriptor? */
 322		if (!path)
 323			return 0;
 324		/* Path type events are only relevant for files and dirs */
 325		if (!d_is_reg(path->dentry) && !d_can_lookup(path->dentry))
 326			return 0;
 327	} else if (!(fid_mode & FAN_REPORT_FID)) {
 328		/* Do we have a directory inode to report? */
 329		if (!dir && !ondir)
 330			return 0;
 331	}
 332
 333	fsnotify_foreach_iter_mark_type(iter_info, mark, type) {
 334		/*
 335		 * Apply ignore mask depending on event flags in ignore mask.
 336		 */
 337		marks_ignore_mask |=
 338			fsnotify_effective_ignore_mask(mark, ondir, type);
 339
 340		/*
 341		 * Send the event depending on event flags in mark mask.
 342		 */
 343		if (!fsnotify_mask_applicable(mark->mask, ondir, type))
 344			continue;
 345
 346		marks_mask |= mark->mask;
 347
 348		/* Record the mark types of this group that matched the event */
 349		*match_mask |= 1U << type;
 350	}
 351
 352	test_mask = event_mask & marks_mask & ~marks_ignore_mask;
 353
 354	/*
 355	 * For dirent modification events (create/delete/move) that do not carry
 356	 * the child entry name information, we report FAN_ONDIR for mkdir/rmdir
 357	 * so user can differentiate them from creat/unlink.
 358	 *
 359	 * For backward compatibility and consistency, do not report FAN_ONDIR
 360	 * to user in legacy fanotify mode (reporting fd) and report FAN_ONDIR
 361	 * to user in fid mode for all event types.
 362	 *
 363	 * We never report FAN_EVENT_ON_CHILD to user, but we do pass it in to
 364	 * fanotify_alloc_event() when group is reporting fid as indication
 365	 * that event happened on child.
 366	 */
 367	if (fid_mode) {
 368		/* Do not report event flags without any event */
 369		if (!(test_mask & ~FANOTIFY_EVENT_FLAGS))
 370			return 0;
 371	} else {
 372		user_mask &= ~FANOTIFY_EVENT_FLAGS;
 373	}
 374
 375	return test_mask & user_mask;
 376}
 377
 378/*
 379 * Check size needed to encode fanotify_fh.
 380 *
 381 * Return size of encoded fh without fanotify_fh header.
 382 * Return 0 on failure to encode.
 383 */
 384static int fanotify_encode_fh_len(struct inode *inode)
 385{
 386	int dwords = 0;
 387	int fh_len;
 388
 389	if (!inode)
 390		return 0;
 391
 392	exportfs_encode_fid(inode, NULL, &dwords);
 393	fh_len = dwords << 2;
 394
 395	/*
 396	 * struct fanotify_error_event might be preallocated and is
 397	 * limited to MAX_HANDLE_SZ.  This should never happen, but
 398	 * safeguard by forcing an invalid file handle.
 399	 */
 400	if (WARN_ON_ONCE(fh_len > MAX_HANDLE_SZ))
 401		return 0;
 402
 403	return fh_len;
 404}
 405
 406/*
 407 * Encode fanotify_fh.
 408 *
 409 * Return total size of encoded fh including fanotify_fh header.
 410 * Return 0 on failure to encode.
 411 */
 412static int fanotify_encode_fh(struct fanotify_fh *fh, struct inode *inode,
 413			      unsigned int fh_len, unsigned int *hash,
 414			      gfp_t gfp)
 415{
 416	int dwords, type = 0;
 417	char *ext_buf = NULL;
 418	void *buf = fh + 1;
 419	int err;
 420
 421	fh->type = FILEID_ROOT;
 422	fh->len = 0;
 423	fh->flags = 0;
 424
 425	/*
 426	 * Invalid FHs are used by FAN_FS_ERROR for errors not
 427	 * linked to any inode. The f_handle won't be reported
 428	 * back to userspace.
 429	 */
 430	if (!inode)
 431		goto out;
 432
 433	/*
 434	 * !gpf means preallocated variable size fh, but fh_len could
 435	 * be zero in that case if encoding fh len failed.
 436	 */
 437	err = -ENOENT;
 438	if (fh_len < 4 || WARN_ON_ONCE(fh_len % 4) || fh_len > MAX_HANDLE_SZ)
 439		goto out_err;
 440
 441	/* No external buffer in a variable size allocated fh */
 442	if (gfp && fh_len > FANOTIFY_INLINE_FH_LEN) {
 443		/* Treat failure to allocate fh as failure to encode fh */
 444		err = -ENOMEM;
 445		ext_buf = kmalloc(fh_len, gfp);
 446		if (!ext_buf)
 447			goto out_err;
 448
 449		*fanotify_fh_ext_buf_ptr(fh) = ext_buf;
 450		buf = ext_buf;
 451		fh->flags |= FANOTIFY_FH_FLAG_EXT_BUF;
 452	}
 453
 454	dwords = fh_len >> 2;
 455	type = exportfs_encode_fid(inode, buf, &dwords);
 456	err = -EINVAL;
 457	/*
 458	 * Unlike file_handle, type and len of struct fanotify_fh are u8.
 459	 * Traditionally, filesystem return handle_type < 0xff, but there
 460	 * is no enforecement for that in vfs.
 461	 */
 462	BUILD_BUG_ON(MAX_HANDLE_SZ > 0xff || FILEID_INVALID > 0xff);
 463	if (type <= 0 || type >= FILEID_INVALID || fh_len != dwords << 2)
 464		goto out_err;
 465
 466	fh->type = type;
 467	fh->len = fh_len;
 468
 469out:
 470	/*
 471	 * Mix fh into event merge key.  Hash might be NULL in case of
 472	 * unhashed FID events (i.e. FAN_FS_ERROR).
 473	 */
 474	if (hash)
 475		*hash ^= fanotify_hash_fh(fh);
 476
 477	return FANOTIFY_FH_HDR_LEN + fh_len;
 478
 479out_err:
 480	pr_warn_ratelimited("fanotify: failed to encode fid (type=%d, len=%d, err=%i)\n",
 481			    type, fh_len, err);
 482	kfree(ext_buf);
 483	*fanotify_fh_ext_buf_ptr(fh) = NULL;
 484	/* Report the event without a file identifier on encode error */
 485	fh->type = FILEID_INVALID;
 486	fh->len = 0;
 487	return 0;
 488}
 489
 490/*
 491 * FAN_REPORT_FID is ambiguous in that it reports the fid of the child for
 492 * some events and the fid of the parent for create/delete/move events.
 493 *
 494 * With the FAN_REPORT_TARGET_FID flag, the fid of the child is reported
 495 * also in create/delete/move events in addition to the fid of the parent
 496 * and the name of the child.
 497 */
 498static inline bool fanotify_report_child_fid(unsigned int fid_mode, u32 mask)
 499{
 500	if (mask & ALL_FSNOTIFY_DIRENT_EVENTS)
 501		return (fid_mode & FAN_REPORT_TARGET_FID);
 502
 503	return (fid_mode & FAN_REPORT_FID) && !(mask & FAN_ONDIR);
 504}
 505
 506/*
 507 * The inode to use as identifier when reporting fid depends on the event
 508 * and the group flags.
 509 *
 510 * With the group flag FAN_REPORT_TARGET_FID, always report the child fid.
 511 *
 512 * Without the group flag FAN_REPORT_TARGET_FID, report the modified directory
 513 * fid on dirent events and the child fid otherwise.
 514 *
 515 * For example:
 516 * FS_ATTRIB reports the child fid even if reported on a watched parent.
 517 * FS_CREATE reports the modified dir fid without FAN_REPORT_TARGET_FID.
 518 *       and reports the created child fid with FAN_REPORT_TARGET_FID.
 519 */
 520static struct inode *fanotify_fid_inode(u32 event_mask, const void *data,
 521					int data_type, struct inode *dir,
 522					unsigned int fid_mode)
 523{
 524	if ((event_mask & ALL_FSNOTIFY_DIRENT_EVENTS) &&
 525	    !(fid_mode & FAN_REPORT_TARGET_FID))
 526		return dir;
 527
 528	return fsnotify_data_inode(data, data_type);
 529}
 530
 531/*
 532 * The inode to use as identifier when reporting dir fid depends on the event.
 533 * Report the modified directory inode on dirent modification events.
 534 * Report the "victim" inode if "victim" is a directory.
 535 * Report the parent inode if "victim" is not a directory and event is
 536 * reported to parent.
 537 * Otherwise, do not report dir fid.
 538 */
 539static struct inode *fanotify_dfid_inode(u32 event_mask, const void *data,
 540					 int data_type, struct inode *dir)
 541{
 542	struct inode *inode = fsnotify_data_inode(data, data_type);
 543
 544	if (event_mask & ALL_FSNOTIFY_DIRENT_EVENTS)
 545		return dir;
 546
 547	if (inode && S_ISDIR(inode->i_mode))
 548		return inode;
 549
 550	return dir;
 551}
 552
 553static struct fanotify_event *fanotify_alloc_path_event(const struct path *path,
 554							unsigned int *hash,
 555							gfp_t gfp)
 556{
 557	struct fanotify_path_event *pevent;
 558
 559	pevent = kmem_cache_alloc(fanotify_path_event_cachep, gfp);
 560	if (!pevent)
 561		return NULL;
 562
 563	pevent->fae.type = FANOTIFY_EVENT_TYPE_PATH;
 564	pevent->path = *path;
 565	*hash ^= fanotify_hash_path(path);
 566	path_get(path);
 567
 568	return &pevent->fae;
 569}
 570
 571static struct fanotify_event *fanotify_alloc_mnt_event(u64 mnt_id, gfp_t gfp)
 572{
 573	struct fanotify_mnt_event *pevent;
 574
 575	pevent = kmem_cache_alloc(fanotify_mnt_event_cachep, gfp);
 576	if (!pevent)
 577		return NULL;
 578
 579	pevent->fae.type = FANOTIFY_EVENT_TYPE_MNT;
 580	pevent->mnt_id = mnt_id;
 581
 582	return &pevent->fae;
 583}
 584
 585static struct fanotify_event *fanotify_alloc_perm_event(const void *data,
 586							int data_type,
 587							gfp_t gfp)
 588{
 589	const struct path *path = fsnotify_data_path(data, data_type);
 590	const struct file_range *range =
 591			    fsnotify_data_file_range(data, data_type);
 592	struct fanotify_perm_event *pevent;
 593
 594	pevent = kmem_cache_alloc(fanotify_perm_event_cachep, gfp);
 595	if (!pevent)
 596		return NULL;
 597
 598	pevent->fae.type = FANOTIFY_EVENT_TYPE_PATH_PERM;
 599	pevent->response = 0;
 600	pevent->hdr.type = FAN_RESPONSE_INFO_NONE;
 601	pevent->hdr.pad = 0;
 602	pevent->hdr.len = 0;
 603	pevent->state = FAN_EVENT_INIT;
 604	pevent->path = *path;
 605	/* NULL ppos means no range info */
 606	pevent->ppos = range ? &range->pos : NULL;
 607	pevent->count = range ? range->count : 0;
 608	path_get(path);
 609
 610	return &pevent->fae;
 611}
 612
 613static struct fanotify_event *fanotify_alloc_fid_event(struct inode *id,
 614						       __kernel_fsid_t *fsid,
 615						       unsigned int *hash,
 616						       gfp_t gfp)
 617{
 618	struct fanotify_fid_event *ffe;
 619
 620	ffe = kmem_cache_alloc(fanotify_fid_event_cachep, gfp);
 621	if (!ffe)
 622		return NULL;
 623
 624	ffe->fae.type = FANOTIFY_EVENT_TYPE_FID;
 625	ffe->fsid = *fsid;
 626	*hash ^= fanotify_hash_fsid(fsid);
 627	fanotify_encode_fh(&ffe->object_fh, id, fanotify_encode_fh_len(id),
 628			   hash, gfp);
 629
 630	return &ffe->fae;
 631}
 632
 633static struct fanotify_event *fanotify_alloc_name_event(struct inode *dir,
 634							__kernel_fsid_t *fsid,
 635							const struct qstr *name,
 636							struct inode *child,
 637							struct dentry *moved,
 638							unsigned int *hash,
 639							gfp_t gfp)
 640{
 641	struct fanotify_name_event *fne;
 642	struct fanotify_info *info;
 643	struct fanotify_fh *dfh, *ffh;
 644	struct inode *dir2 = moved ? d_inode(moved->d_parent) : NULL;
 645	const struct qstr *name2 = moved ? &moved->d_name : NULL;
 646	unsigned int dir_fh_len = fanotify_encode_fh_len(dir);
 647	unsigned int dir2_fh_len = fanotify_encode_fh_len(dir2);
 648	unsigned int child_fh_len = fanotify_encode_fh_len(child);
 649	unsigned long name_len = name ? name->len : 0;
 650	unsigned long name2_len = name2 ? name2->len : 0;
 651	unsigned int len, size;
 652
 653	/* Reserve terminating null byte even for empty name */
 654	size = sizeof(*fne) + name_len + name2_len + 2;
 655	if (dir_fh_len)
 656		size += FANOTIFY_FH_HDR_LEN + dir_fh_len;
 657	if (dir2_fh_len)
 658		size += FANOTIFY_FH_HDR_LEN + dir2_fh_len;
 659	if (child_fh_len)
 660		size += FANOTIFY_FH_HDR_LEN + child_fh_len;
 661	fne = kmalloc(size, gfp);
 662	if (!fne)
 663		return NULL;
 664
 665	fne->fae.type = FANOTIFY_EVENT_TYPE_FID_NAME;
 666	fne->fsid = *fsid;
 667	*hash ^= fanotify_hash_fsid(fsid);
 668	info = &fne->info;
 669	fanotify_info_init(info);
 670	if (dir_fh_len) {
 671		dfh = fanotify_info_dir_fh(info);
 672		len = fanotify_encode_fh(dfh, dir, dir_fh_len, hash, 0);
 673		fanotify_info_set_dir_fh(info, len);
 674	}
 675	if (dir2_fh_len) {
 676		dfh = fanotify_info_dir2_fh(info);
 677		len = fanotify_encode_fh(dfh, dir2, dir2_fh_len, hash, 0);
 678		fanotify_info_set_dir2_fh(info, len);
 679	}
 680	if (child_fh_len) {
 681		ffh = fanotify_info_file_fh(info);
 682		len = fanotify_encode_fh(ffh, child, child_fh_len, hash, 0);
 683		fanotify_info_set_file_fh(info, len);
 684	}
 685	if (name_len) {
 686		fanotify_info_copy_name(info, name);
 687		*hash ^= full_name_hash((void *)name_len, name->name, name_len);
 688	}
 689	if (name2_len) {
 690		fanotify_info_copy_name2(info, name2);
 691		*hash ^= full_name_hash((void *)name2_len, name2->name,
 692					name2_len);
 693	}
 694
 695	pr_debug("%s: size=%u dir_fh_len=%u child_fh_len=%u name_len=%u name='%.*s'\n",
 696		 __func__, size, dir_fh_len, child_fh_len,
 697		 info->name_len, info->name_len, fanotify_info_name(info));
 698
 699	if (dir2_fh_len) {
 700		pr_debug("%s: dir2_fh_len=%u name2_len=%u name2='%.*s'\n",
 701			 __func__, dir2_fh_len, info->name2_len,
 702			 info->name2_len, fanotify_info_name2(info));
 703	}
 704
 705	return &fne->fae;
 706}
 707
 708static struct fanotify_event *fanotify_alloc_error_event(
 709						struct fsnotify_group *group,
 710						__kernel_fsid_t *fsid,
 711						const void *data, int data_type,
 712						unsigned int *hash)
 713{
 714	struct fs_error_report *report =
 715			fsnotify_data_error_report(data, data_type);
 716	struct inode *inode;
 717	struct fanotify_error_event *fee;
 718	int fh_len;
 719
 720	if (WARN_ON_ONCE(!report))
 721		return NULL;
 722
 723	fee = mempool_alloc(&group->fanotify_data.error_events_pool, GFP_NOFS);
 724	if (!fee)
 725		return NULL;
 726
 727	fee->fae.type = FANOTIFY_EVENT_TYPE_FS_ERROR;
 728	fee->error = report->error;
 729	fee->err_count = 1;
 730	fee->fsid = *fsid;
 731
 732	inode = report->inode;
 733	fh_len = fanotify_encode_fh_len(inode);
 734
 735	/* Bad fh_len. Fallback to using an invalid fh. Should never happen. */
 736	if (!fh_len && inode)
 737		inode = NULL;
 738
 739	fanotify_encode_fh(&fee->object_fh, inode, fh_len, NULL, 0);
 740
 741	*hash ^= fanotify_hash_fsid(fsid);
 742
 743	return &fee->fae;
 744}
 745
 746static struct fanotify_event *fanotify_alloc_event(
 747				struct fsnotify_group *group,
 748				u32 mask, const void *data, int data_type,
 749				struct inode *dir, const struct qstr *file_name,
 750				__kernel_fsid_t *fsid, u32 match_mask)
 751{
 752	struct fanotify_event *event = NULL;
 753	gfp_t gfp = GFP_KERNEL_ACCOUNT;
 754	unsigned int fid_mode = FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS);
 755	struct inode *id = fanotify_fid_inode(mask, data, data_type, dir,
 756					      fid_mode);
 757	struct inode *dirid = fanotify_dfid_inode(mask, data, data_type, dir);
 758	const struct path *path = fsnotify_data_path(data, data_type);
 759	u64 mnt_id = fsnotify_data_mnt_id(data, data_type);
 760	struct mem_cgroup *old_memcg;
 761	struct dentry *moved = NULL;
 762	struct inode *child = NULL;
 763	bool name_event = false;
 764	unsigned int hash = 0;
 765	bool ondir = mask & FAN_ONDIR;
 766	struct pid *pid;
 767
 768	if ((fid_mode & FAN_REPORT_DIR_FID) && dirid) {
 769		/*
 770		 * For certain events and group flags, report the child fid
 771		 * in addition to reporting the parent fid and maybe child name.
 772		 */
 773		if (fanotify_report_child_fid(fid_mode, mask) && id != dirid)
 774			child = id;
 775
 776		id = dirid;
 777
 778		/*
 779		 * We record file name only in a group with FAN_REPORT_NAME
 780		 * and when we have a directory inode to report.
 781		 *
 782		 * For directory entry modification event, we record the fid of
 783		 * the directory and the name of the modified entry.
 784		 *
 785		 * For event on non-directory that is reported to parent, we
 786		 * record the fid of the parent and the name of the child.
 787		 *
 788		 * Even if not reporting name, we need a variable length
 789		 * fanotify_name_event if reporting both parent and child fids.
 790		 */
 791		if (!(fid_mode & FAN_REPORT_NAME)) {
 792			name_event = !!child;
 793			file_name = NULL;
 794		} else if ((mask & ALL_FSNOTIFY_DIRENT_EVENTS) || !ondir) {
 795			name_event = true;
 796		}
 797
 798		/*
 799		 * In the special case of FAN_RENAME event, use the match_mask
 800		 * to determine if we need to report only the old parent+name,
 801		 * only the new parent+name or both.
 802		 * 'dirid' and 'file_name' are the old parent+name and
 803		 * 'moved' has the new parent+name.
 804		 */
 805		if (mask & FAN_RENAME) {
 806			bool report_old, report_new;
 807
 808			if (WARN_ON_ONCE(!match_mask))
 809				return NULL;
 810
 811			/* Report both old and new parent+name if sb watching */
 812			report_old = report_new =
 813				match_mask & (1U << FSNOTIFY_ITER_TYPE_SB);
 814			report_old |=
 815				match_mask & (1U << FSNOTIFY_ITER_TYPE_INODE);
 816			report_new |=
 817				match_mask & (1U << FSNOTIFY_ITER_TYPE_INODE2);
 818
 819			if (!report_old) {
 820				/* Do not report old parent+name */
 821				dirid = NULL;
 822				file_name = NULL;
 823			}
 824			if (report_new) {
 825				/* Report new parent+name */
 826				moved = fsnotify_data_dentry(data, data_type);
 827			}
 828		}
 829	}
 830
 831	/*
 832	 * For queues with unlimited length lost events are not expected and
 833	 * can possibly have security implications. Avoid losing events when
 834	 * memory is short. For the limited size queues, avoid OOM killer in the
 835	 * target monitoring memcg as it may have security repercussion.
 836	 */
 837	if (group->max_events == UINT_MAX)
 838		gfp |= __GFP_NOFAIL;
 839	else
 840		gfp |= __GFP_RETRY_MAYFAIL;
 841
 842	/* Whoever is interested in the event, pays for the allocation. */
 843	old_memcg = set_active_memcg(group->memcg);
 844
 845	if (fanotify_is_perm_event(mask)) {
 846		event = fanotify_alloc_perm_event(data, data_type, gfp);
 847	} else if (fanotify_is_error_event(mask)) {
 848		event = fanotify_alloc_error_event(group, fsid, data,
 849						   data_type, &hash);
 850	} else if (name_event && (file_name || moved || child)) {
 851		event = fanotify_alloc_name_event(dirid, fsid, file_name, child,
 852						  moved, &hash, gfp);
 853	} else if (fid_mode) {
 854		event = fanotify_alloc_fid_event(id, fsid, &hash, gfp);
 855	} else if (path) {
 856		event = fanotify_alloc_path_event(path, &hash, gfp);
 857	} else if (mnt_id) {
 858		event = fanotify_alloc_mnt_event(mnt_id, gfp);
 859	} else {
 860		WARN_ON_ONCE(1);
 861	}
 862
 863	if (!event)
 864		goto out;
 865
 866	if (FAN_GROUP_FLAG(group, FAN_REPORT_TID))
 867		pid = get_pid(task_pid(current));
 868	else
 869		pid = get_pid(task_tgid(current));
 870
 871	/* Mix event info, FAN_ONDIR flag and pid into event merge key */
 872	hash ^= hash_long((unsigned long)pid | ondir, FANOTIFY_EVENT_HASH_BITS);
 873	fanotify_init_event(event, hash, mask);
 874	event->pid = pid;
 875
 876out:
 877	set_active_memcg(old_memcg);
 878	return event;
 879}
 880
 881/*
 882 * Get cached fsid of the filesystem containing the object from any mark.
 883 * All marks are supposed to have the same fsid, but we do not verify that here.
 884 */
 885static __kernel_fsid_t fanotify_get_fsid(struct fsnotify_iter_info *iter_info)
 886{
 887	struct fsnotify_mark *mark;
 888	int type;
 889	__kernel_fsid_t fsid = {};
 890
 891	fsnotify_foreach_iter_mark_type(iter_info, mark, type) {
 892		if (!(mark->flags & FSNOTIFY_MARK_FLAG_HAS_FSID))
 893			continue;
 894		fsid = FANOTIFY_MARK(mark)->fsid;
 895		if (!(mark->flags & FSNOTIFY_MARK_FLAG_WEAK_FSID) &&
 896		    WARN_ON_ONCE(!fsid.val[0] && !fsid.val[1]))
 897			continue;
 898		return fsid;
 899	}
 900
 901	return fsid;
 902}
 903
 904/*
 905 * Add an event to hash table for faster merge.
 906 */
 907static void fanotify_insert_event(struct fsnotify_group *group,
 908				  struct fsnotify_event *fsn_event)
 909{
 910	struct fanotify_event *event = FANOTIFY_E(fsn_event);
 911	unsigned int bucket = fanotify_event_hash_bucket(group, event);
 912	struct hlist_head *hlist = &group->fanotify_data.merge_hash[bucket];
 913
 914	assert_spin_locked(&group->notification_lock);
 915
 916	if (!fanotify_is_hashed_event(event->mask))
 917		return;
 918
 919	pr_debug("%s: group=%p event=%p bucket=%u\n", __func__,
 920		 group, event, bucket);
 921
 922	hlist_add_head(&event->merge_list, hlist);
 923}
 924
 925static int fanotify_handle_event(struct fsnotify_group *group, u32 mask,
 926				 const void *data, int data_type,
 927				 struct inode *dir,
 928				 const struct qstr *file_name, u32 cookie,
 929				 struct fsnotify_iter_info *iter_info)
 930{
 931	int ret = 0;
 932	struct fanotify_event *event;
 933	struct fsnotify_event *fsn_event;
 934	__kernel_fsid_t fsid = {};
 935	u32 match_mask = 0;
 936
 937	BUILD_BUG_ON(FAN_ACCESS != FS_ACCESS);
 938	BUILD_BUG_ON(FAN_MODIFY != FS_MODIFY);
 939	BUILD_BUG_ON(FAN_ATTRIB != FS_ATTRIB);
 940	BUILD_BUG_ON(FAN_CLOSE_NOWRITE != FS_CLOSE_NOWRITE);
 941	BUILD_BUG_ON(FAN_CLOSE_WRITE != FS_CLOSE_WRITE);
 942	BUILD_BUG_ON(FAN_OPEN != FS_OPEN);
 943	BUILD_BUG_ON(FAN_MOVED_TO != FS_MOVED_TO);
 944	BUILD_BUG_ON(FAN_MOVED_FROM != FS_MOVED_FROM);
 945	BUILD_BUG_ON(FAN_CREATE != FS_CREATE);
 946	BUILD_BUG_ON(FAN_DELETE != FS_DELETE);
 947	BUILD_BUG_ON(FAN_DELETE_SELF != FS_DELETE_SELF);
 948	BUILD_BUG_ON(FAN_MOVE_SELF != FS_MOVE_SELF);
 949	BUILD_BUG_ON(FAN_EVENT_ON_CHILD != FS_EVENT_ON_CHILD);
 950	BUILD_BUG_ON(FAN_Q_OVERFLOW != FS_Q_OVERFLOW);
 951	BUILD_BUG_ON(FAN_OPEN_PERM != FS_OPEN_PERM);
 952	BUILD_BUG_ON(FAN_ACCESS_PERM != FS_ACCESS_PERM);
 953	BUILD_BUG_ON(FAN_ONDIR != FS_ISDIR);
 954	BUILD_BUG_ON(FAN_OPEN_EXEC != FS_OPEN_EXEC);
 955	BUILD_BUG_ON(FAN_OPEN_EXEC_PERM != FS_OPEN_EXEC_PERM);
 956	BUILD_BUG_ON(FAN_FS_ERROR != FS_ERROR);
 957	BUILD_BUG_ON(FAN_RENAME != FS_RENAME);
 958	BUILD_BUG_ON(FAN_PRE_ACCESS != FS_PRE_ACCESS);
 959
 960	BUILD_BUG_ON(HWEIGHT32(ALL_FANOTIFY_EVENT_BITS) != 24);
 961
 962	mask = fanotify_group_event_mask(group, iter_info, &match_mask,
 963					 mask, data, data_type, dir);
 964	if (!mask)
 965		return 0;
 966
 967	pr_debug("%s: group=%p mask=%x report_mask=%x\n", __func__,
 968		 group, mask, match_mask);
 969
 970	if (fanotify_is_perm_event(mask)) {
 971		/*
 972		 * fsnotify_prepare_user_wait() fails if we race with mark
 973		 * deletion.  Just let the operation pass in that case.
 974		 */
 975		if (!fsnotify_prepare_user_wait(iter_info))
 976			return 0;
 977	}
 978
 979	if (FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS))
 980		fsid = fanotify_get_fsid(iter_info);
 981
 982	event = fanotify_alloc_event(group, mask, data, data_type, dir,
 983				     file_name, &fsid, match_mask);
 984	ret = -ENOMEM;
 985	if (unlikely(!event)) {
 986		/*
 987		 * We don't queue overflow events for permission events as
 988		 * there the access is denied and so no event is in fact lost.
 989		 */
 990		if (!fanotify_is_perm_event(mask))
 991			fsnotify_queue_overflow(group);
 992		goto finish;
 993	}
 994
 995	fsn_event = &event->fse;
 996	ret = fsnotify_insert_event(group, fsn_event, fanotify_merge,
 997				    fanotify_insert_event);
 998	if (ret) {
 999		/* Permission events shouldn't be merged */
1000		BUG_ON(ret == 1 && mask & FANOTIFY_PERM_EVENTS);
1001		/* Our event wasn't used in the end. Free it. */
1002		fsnotify_destroy_event(group, fsn_event);
1003
1004		ret = 0;
1005	} else if (fanotify_is_perm_event(mask)) {
1006		ret = fanotify_get_response(group, FANOTIFY_PERM(event),
1007					    iter_info);
1008	}
1009finish:
1010	if (fanotify_is_perm_event(mask))
1011		fsnotify_finish_user_wait(iter_info);
1012
1013	return ret;
1014}
1015
1016static void fanotify_free_group_priv(struct fsnotify_group *group)
1017{
1018	put_user_ns(group->user_ns);
1019	kfree(group->fanotify_data.merge_hash);
1020	if (group->fanotify_data.ucounts)
1021		dec_ucount(group->fanotify_data.ucounts,
1022			   UCOUNT_FANOTIFY_GROUPS);
1023
1024	if (mempool_initialized(&group->fanotify_data.error_events_pool))
1025		mempool_exit(&group->fanotify_data.error_events_pool);
1026}
1027
1028static void fanotify_free_path_event(struct fanotify_event *event)
1029{
1030	path_put(fanotify_event_path(event));
1031	kmem_cache_free(fanotify_path_event_cachep, FANOTIFY_PE(event));
1032}
1033
1034static void fanotify_free_perm_event(struct fanotify_event *event)
1035{
1036	path_put(fanotify_event_path(event));
1037	kmem_cache_free(fanotify_perm_event_cachep, FANOTIFY_PERM(event));
1038}
1039
1040static void fanotify_free_fid_event(struct fanotify_event *event)
1041{
1042	struct fanotify_fid_event *ffe = FANOTIFY_FE(event);
1043
1044	if (fanotify_fh_has_ext_buf(&ffe->object_fh))
1045		kfree(fanotify_fh_ext_buf(&ffe->object_fh));
1046	kmem_cache_free(fanotify_fid_event_cachep, ffe);
1047}
1048
1049static void fanotify_free_name_event(struct fanotify_event *event)
1050{
1051	kfree(FANOTIFY_NE(event));
1052}
1053
1054static void fanotify_free_error_event(struct fsnotify_group *group,
1055				      struct fanotify_event *event)
1056{
1057	struct fanotify_error_event *fee = FANOTIFY_EE(event);
1058
1059	mempool_free(fee, &group->fanotify_data.error_events_pool);
1060}
1061
1062static void fanotify_free_mnt_event(struct fanotify_event *event)
1063{
1064	kmem_cache_free(fanotify_mnt_event_cachep, FANOTIFY_ME(event));
1065}
1066
1067static void fanotify_free_event(struct fsnotify_group *group,
1068				struct fsnotify_event *fsn_event)
1069{
1070	struct fanotify_event *event;
1071
1072	event = FANOTIFY_E(fsn_event);
1073	put_pid(event->pid);
1074	switch (event->type) {
1075	case FANOTIFY_EVENT_TYPE_PATH:
1076		fanotify_free_path_event(event);
1077		break;
1078	case FANOTIFY_EVENT_TYPE_PATH_PERM:
1079		fanotify_free_perm_event(event);
1080		break;
1081	case FANOTIFY_EVENT_TYPE_FID:
1082		fanotify_free_fid_event(event);
1083		break;
1084	case FANOTIFY_EVENT_TYPE_FID_NAME:
1085		fanotify_free_name_event(event);
1086		break;
1087	case FANOTIFY_EVENT_TYPE_OVERFLOW:
1088		kfree(event);
1089		break;
1090	case FANOTIFY_EVENT_TYPE_FS_ERROR:
1091		fanotify_free_error_event(group, event);
1092		break;
1093	case FANOTIFY_EVENT_TYPE_MNT:
1094		fanotify_free_mnt_event(event);
1095		break;
1096	default:
1097		WARN_ON_ONCE(1);
1098	}
1099}
1100
1101static void fanotify_freeing_mark(struct fsnotify_mark *mark,
1102				  struct fsnotify_group *group)
1103{
1104	if (!FAN_GROUP_FLAG(group, FAN_UNLIMITED_MARKS))
1105		dec_ucount(group->fanotify_data.ucounts, UCOUNT_FANOTIFY_MARKS);
1106}
1107
1108static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
1109{
1110	kmem_cache_free(fanotify_mark_cache, FANOTIFY_MARK(fsn_mark));
1111}
1112
1113const struct fsnotify_ops fanotify_fsnotify_ops = {
1114	.handle_event = fanotify_handle_event,
1115	.free_group_priv = fanotify_free_group_priv,
1116	.free_event = fanotify_free_event,
1117	.freeing_mark = fanotify_freeing_mark,
1118	.free_mark = fanotify_free_mark,
1119};