1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 *  linux/fs/locks.c
   4 *
   5 *  Provide support for fcntl()'s F_GETLK, F_SETLK, and F_SETLKW calls.
   6 *  Doug Evans (dje@spiff.uucp), August 07, 1992
 
 
   7 *
   8 *  Deadlock detection added.
   9 *  FIXME: one thing isn't handled yet:
  10 *	- mandatory locks (requires lots of changes elsewhere)
  11 *  Kelly Carmichael (kelly@[142.24.8.65]), September 17, 1994.
  12 *
  13 *  Miscellaneous edits, and a total rewrite of posix_lock_file() code.
  14 *  Kai Petzke (wpp@marie.physik.tu-berlin.de), 1994
  15 *
  16 *  Converted file_lock_table to a linked list from an array, which eliminates
  17 *  the limits on how many active file locks are open.
  18 *  Chad Page (pageone@netcom.com), November 27, 1994
  19 *
  20 *  Removed dependency on file descriptors. dup()'ed file descriptors now
  21 *  get the same locks as the original file descriptors, and a close() on
  22 *  any file descriptor removes ALL the locks on the file for the current
  23 *  process. Since locks still depend on the process id, locks are inherited
  24 *  after an exec() but not after a fork(). This agrees with POSIX, and both
  25 *  BSD and SVR4 practice.
  26 *  Andy Walker (andy@lysaker.kvaerner.no), February 14, 1995
  27 *
  28 *  Scrapped free list which is redundant now that we allocate locks
  29 *  dynamically with kmalloc()/kfree().
  30 *  Andy Walker (andy@lysaker.kvaerner.no), February 21, 1995
  31 *
  32 *  Implemented two lock personalities - FL_FLOCK and FL_POSIX.
  33 *
  34 *  FL_POSIX locks are created with calls to fcntl() and lockf() through the
  35 *  fcntl() system call. They have the semantics described above.
  36 *
  37 *  FL_FLOCK locks are created with calls to flock(), through the flock()
  38 *  system call, which is new. Old C libraries implement flock() via fcntl()
  39 *  and will continue to use the old, broken implementation.
  40 *
  41 *  FL_FLOCK locks follow the 4.4 BSD flock() semantics. They are associated
  42 *  with a file pointer (filp). As a result they can be shared by a parent
  43 *  process and its children after a fork(). They are removed when the last
  44 *  file descriptor referring to the file pointer is closed (unless explicitly
  45 *  unlocked).
  46 *
  47 *  FL_FLOCK locks never deadlock, an existing lock is always removed before
  48 *  upgrading from shared to exclusive (or vice versa). When this happens
  49 *  any processes blocked by the current lock are woken up and allowed to
  50 *  run before the new lock is applied.
  51 *  Andy Walker (andy@lysaker.kvaerner.no), June 09, 1995
  52 *
  53 *  Removed some race conditions in flock_lock_file(), marked other possible
  54 *  races. Just grep for FIXME to see them.
  55 *  Dmitry Gorodchanin (pgmdsg@ibi.com), February 09, 1996.
  56 *
  57 *  Addressed Dmitry's concerns. Deadlock checking no longer recursive.
  58 *  Lock allocation changed to GFP_ATOMIC as we can't afford to sleep
  59 *  once we've checked for blocking and deadlocking.
  60 *  Andy Walker (andy@lysaker.kvaerner.no), April 03, 1996.
  61 *
  62 *  Initial implementation of mandatory locks. SunOS turned out to be
  63 *  a rotten model, so I implemented the "obvious" semantics.
  64 *  See 'Documentation/filesystems/mandatory-locking.txt' for details.
  65 *  Andy Walker (andy@lysaker.kvaerner.no), April 06, 1996.
  66 *
  67 *  Don't allow mandatory locks on mmap()'ed files. Added simple functions to
  68 *  check if a file has mandatory locks, used by mmap(), open() and creat() to
  69 *  see if system call should be rejected. Ref. HP-UX/SunOS/Solaris Reference
  70 *  Manual, Section 2.
  71 *  Andy Walker (andy@lysaker.kvaerner.no), April 09, 1996.
  72 *
  73 *  Tidied up block list handling. Added '/proc/locks' interface.
  74 *  Andy Walker (andy@lysaker.kvaerner.no), April 24, 1996.
  75 *
  76 *  Fixed deadlock condition for pathological code that mixes calls to
  77 *  flock() and fcntl().
  78 *  Andy Walker (andy@lysaker.kvaerner.no), April 29, 1996.
  79 *
  80 *  Allow only one type of locking scheme (FL_POSIX or FL_FLOCK) to be in use
  81 *  for a given file at a time. Changed the CONFIG_LOCK_MANDATORY scheme to
  82 *  guarantee sensible behaviour in the case where file system modules might
  83 *  be compiled with different options than the kernel itself.
  84 *  Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
  85 *
  86 *  Added a couple of missing wake_up() calls. Thanks to Thomas Meckel
  87 *  (Thomas.Meckel@mni.fh-giessen.de) for spotting this.
  88 *  Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
  89 *
  90 *  Changed FL_POSIX locks to use the block list in the same way as FL_FLOCK
  91 *  locks. Changed process synchronisation to avoid dereferencing locks that
  92 *  have already been freed.
  93 *  Andy Walker (andy@lysaker.kvaerner.no), Sep 21, 1996.
  94 *
  95 *  Made the block list a circular list to minimise searching in the list.
  96 *  Andy Walker (andy@lysaker.kvaerner.no), Sep 25, 1996.
  97 *
  98 *  Made mandatory locking a mount option. Default is not to allow mandatory
  99 *  locking.
 100 *  Andy Walker (andy@lysaker.kvaerner.no), Oct 04, 1996.
 101 *
 102 *  Some adaptations for NFS support.
 103 *  Olaf Kirch (okir@monad.swb.de), Dec 1996,
 104 *
 105 *  Fixed /proc/locks interface so that we can't overrun the buffer we are handed.
 106 *  Andy Walker (andy@lysaker.kvaerner.no), May 12, 1997.
 107 *
 108 *  Use slab allocator instead of kmalloc/kfree.
 109 *  Use generic list implementation from <linux/list.h>.
 110 *  Sped up posix_locks_deadlock by only considering blocked locks.
 111 *  Matthew Wilcox <willy@debian.org>, March, 2000.
 112 *
 113 *  Leases and LOCK_MAND
 114 *  Matthew Wilcox <willy@debian.org>, June, 2000.
 115 *  Stephen Rothwell <sfr@canb.auug.org.au>, June, 2000.
 116 *
 117 * Locking conflicts and dependencies:
 118 * If multiple threads attempt to lock the same byte (or flock the same file)
 119 * only one can be granted the lock, and other must wait their turn.
 120 * The first lock has been "applied" or "granted", the others are "waiting"
 121 * and are "blocked" by the "applied" lock..
 122 *
 123 * Waiting and applied locks are all kept in trees whose properties are:
 124 *
 125 *	- the root of a tree may be an applied or waiting lock.
 126 *	- every other node in the tree is a waiting lock that
 127 *	  conflicts with every ancestor of that node.
 128 *
 129 * Every such tree begins life as a waiting singleton which obviously
 130 * satisfies the above properties.
 131 *
 132 * The only ways we modify trees preserve these properties:
 133 *
 134 *	1. We may add a new leaf node, but only after first verifying that it
 135 *	   conflicts with all of its ancestors.
 136 *	2. We may remove the root of a tree, creating a new singleton
 137 *	   tree from the root and N new trees rooted in the immediate
 138 *	   children.
 139 *	3. If the root of a tree is not currently an applied lock, we may
 140 *	   apply it (if possible).
 141 *	4. We may upgrade the root of the tree (either extend its range,
 142 *	   or upgrade its entire range from read to write).
 143 *
 144 * When an applied lock is modified in a way that reduces or downgrades any
 145 * part of its range, we remove all its children (2 above).  This particularly
 146 * happens when a lock is unlocked.
 147 *
 148 * For each of those child trees we "wake up" the thread which is
 149 * waiting for the lock so it can continue handling as follows: if the
 150 * root of the tree applies, we do so (3).  If it doesn't, it must
 151 * conflict with some applied lock.  We remove (wake up) all of its children
 152 * (2), and add it is a new leaf to the tree rooted in the applied
 153 * lock (1).  We then repeat the process recursively with those
 154 * children.
 155 *
 156 */
 157
 158#include <linux/capability.h>
 159#include <linux/file.h>
 160#include <linux/fdtable.h>
 
 161#include <linux/fs.h>
 162#include <linux/init.h>
 163#include <linux/security.h>
 164#include <linux/slab.h>
 165#include <linux/syscalls.h>
 166#include <linux/time.h>
 167#include <linux/rcupdate.h>
 168#include <linux/pid_namespace.h>
 169#include <linux/hashtable.h>
 170#include <linux/percpu.h>
 
 171
 172#define CREATE_TRACE_POINTS
 173#include <trace/events/filelock.h>
 174
 175#include <linux/uaccess.h>
 176
 177#define IS_POSIX(fl)	(fl->fl_flags & FL_POSIX)
 178#define IS_FLOCK(fl)	(fl->fl_flags & FL_FLOCK)
 179#define IS_LEASE(fl)	(fl->fl_flags & (FL_LEASE|FL_DELEG|FL_LAYOUT))
 180#define IS_OFDLCK(fl)	(fl->fl_flags & FL_OFDLCK)
 181#define IS_REMOTELCK(fl)	(fl->fl_pid <= 0)
 182
 183static bool lease_breaking(struct file_lock *fl)
 184{
 185	return fl->fl_flags & (FL_UNLOCK_PENDING | FL_DOWNGRADE_PENDING);
 186}
 187
 188static int target_leasetype(struct file_lock *fl)
 189{
 190	if (fl->fl_flags & FL_UNLOCK_PENDING)
 
 
 
 
 
 191		return F_UNLCK;
 192	if (fl->fl_flags & FL_DOWNGRADE_PENDING)
 193		return F_RDLCK;
 194	return fl->fl_type;
 195}
 196
 197int leases_enable = 1;
 198int lease_break_time = 45;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 199
 200/*
 201 * The global file_lock_list is only used for displaying /proc/locks, so we
 202 * keep a list on each CPU, with each list protected by its own spinlock.
 203 * Global serialization is done using file_rwsem.
 204 *
 205 * Note that alterations to the list also require that the relevant flc_lock is
 206 * held.
 207 */
 208struct file_lock_list_struct {
 209	spinlock_t		lock;
 210	struct hlist_head	hlist;
 211};
 212static DEFINE_PER_CPU(struct file_lock_list_struct, file_lock_list);
 213DEFINE_STATIC_PERCPU_RWSEM(file_rwsem);
 214
 215
 216/*
 217 * The blocked_hash is used to find POSIX lock loops for deadlock detection.
 218 * It is protected by blocked_lock_lock.
 219 *
 220 * We hash locks by lockowner in order to optimize searching for the lock a
 221 * particular lockowner is waiting on.
 222 *
 223 * FIXME: make this value scale via some heuristic? We generally will want more
 224 * buckets when we have more lockowners holding locks, but that's a little
 225 * difficult to determine without knowing what the workload will look like.
 226 */
 227#define BLOCKED_HASH_BITS	7
 228static DEFINE_HASHTABLE(blocked_hash, BLOCKED_HASH_BITS);
 229
 230/*
 231 * This lock protects the blocked_hash. Generally, if you're accessing it, you
 232 * want to be holding this lock.
 233 *
 234 * In addition, it also protects the fl->fl_blocked_requests list, and the
 235 * fl->fl_blocker pointer for file_lock structures that are acting as lock
 236 * requests (in contrast to those that are acting as records of acquired locks).
 237 *
 238 * Note that when we acquire this lock in order to change the above fields,
 239 * we often hold the flc_lock as well. In certain cases, when reading the fields
 240 * protected by this lock, we can skip acquiring it iff we already hold the
 241 * flc_lock.
 242 */
 243static DEFINE_SPINLOCK(blocked_lock_lock);
 244
 245static struct kmem_cache *flctx_cache __read_mostly;
 246static struct kmem_cache *filelock_cache __read_mostly;
 
 247
 248static struct file_lock_context *
 249locks_get_lock_context(struct inode *inode, int type)
 250{
 251	struct file_lock_context *ctx;
 252
 253	/* paired with cmpxchg() below */
 254	ctx = smp_load_acquire(&inode->i_flctx);
 255	if (likely(ctx) || type == F_UNLCK)
 256		goto out;
 257
 258	ctx = kmem_cache_alloc(flctx_cache, GFP_KERNEL);
 259	if (!ctx)
 260		goto out;
 261
 262	spin_lock_init(&ctx->flc_lock);
 263	INIT_LIST_HEAD(&ctx->flc_flock);
 264	INIT_LIST_HEAD(&ctx->flc_posix);
 265	INIT_LIST_HEAD(&ctx->flc_lease);
 266
 267	/*
 268	 * Assign the pointer if it's not already assigned. If it is, then
 269	 * free the context we just allocated.
 270	 */
 271	if (cmpxchg(&inode->i_flctx, NULL, ctx)) {
 272		kmem_cache_free(flctx_cache, ctx);
 273		ctx = smp_load_acquire(&inode->i_flctx);
 274	}
 275out:
 276	trace_locks_get_lock_context(inode, type, ctx);
 277	return ctx;
 278}
 279
 280static void
 281locks_dump_ctx_list(struct list_head *list, char *list_type)
 282{
 283	struct file_lock *fl;
 284
 285	list_for_each_entry(fl, list, fl_list) {
 286		pr_warn("%s: fl_owner=%p fl_flags=0x%x fl_type=0x%x fl_pid=%u\n", list_type, fl->fl_owner, fl->fl_flags, fl->fl_type, fl->fl_pid);
 287	}
 
 288}
 289
 290static void
 291locks_check_ctx_lists(struct inode *inode)
 292{
 293	struct file_lock_context *ctx = inode->i_flctx;
 294
 295	if (unlikely(!list_empty(&ctx->flc_flock) ||
 296		     !list_empty(&ctx->flc_posix) ||
 297		     !list_empty(&ctx->flc_lease))) {
 298		pr_warn("Leaked locks on dev=0x%x:0x%x ino=0x%lx:\n",
 299			MAJOR(inode->i_sb->s_dev), MINOR(inode->i_sb->s_dev),
 300			inode->i_ino);
 301		locks_dump_ctx_list(&ctx->flc_flock, "FLOCK");
 302		locks_dump_ctx_list(&ctx->flc_posix, "POSIX");
 303		locks_dump_ctx_list(&ctx->flc_lease, "LEASE");
 304	}
 305}
 306
 307static void
 308locks_check_ctx_file_list(struct file *filp, struct list_head *list,
 309				char *list_type)
 310{
 311	struct file_lock *fl;
 312	struct inode *inode = locks_inode(filp);
 313
 314	list_for_each_entry(fl, list, fl_list)
 315		if (fl->fl_file == filp)
 316			pr_warn("Leaked %s lock on dev=0x%x:0x%x ino=0x%lx "
 317				" fl_owner=%p fl_flags=0x%x fl_type=0x%x fl_pid=%u\n",
 318				list_type, MAJOR(inode->i_sb->s_dev),
 319				MINOR(inode->i_sb->s_dev), inode->i_ino,
 320				fl->fl_owner, fl->fl_flags, fl->fl_type, fl->fl_pid);
 
 321}
 322
 323void
 324locks_free_lock_context(struct inode *inode)
 325{
 326	struct file_lock_context *ctx = inode->i_flctx;
 327
 328	if (unlikely(ctx)) {
 329		locks_check_ctx_lists(inode);
 330		kmem_cache_free(flctx_cache, ctx);
 331	}
 332}
 333
 334static void locks_init_lock_heads(struct file_lock *fl)
 335{
 336	INIT_HLIST_NODE(&fl->fl_link);
 337	INIT_LIST_HEAD(&fl->fl_list);
 338	INIT_LIST_HEAD(&fl->fl_blocked_requests);
 339	INIT_LIST_HEAD(&fl->fl_blocked_member);
 340	init_waitqueue_head(&fl->fl_wait);
 341}
 342
 343/* Allocate an empty lock structure. */
 344struct file_lock *locks_alloc_lock(void)
 345{
 346	struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
 347
 348	if (fl)
 349		locks_init_lock_heads(fl);
 350
 351	return fl;
 352}
 353EXPORT_SYMBOL_GPL(locks_alloc_lock);
 354
 
 
 
 
 
 
 
 
 
 
 
 
 355void locks_release_private(struct file_lock *fl)
 356{
 357	BUG_ON(waitqueue_active(&fl->fl_wait));
 358	BUG_ON(!list_empty(&fl->fl_list));
 359	BUG_ON(!list_empty(&fl->fl_blocked_requests));
 360	BUG_ON(!list_empty(&fl->fl_blocked_member));
 361	BUG_ON(!hlist_unhashed(&fl->fl_link));
 
 
 362
 363	if (fl->fl_ops) {
 364		if (fl->fl_ops->fl_release_private)
 365			fl->fl_ops->fl_release_private(fl);
 366		fl->fl_ops = NULL;
 367	}
 368
 369	if (fl->fl_lmops) {
 370		if (fl->fl_lmops->lm_put_owner) {
 371			fl->fl_lmops->lm_put_owner(fl->fl_owner);
 372			fl->fl_owner = NULL;
 373		}
 374		fl->fl_lmops = NULL;
 375	}
 376}
 377EXPORT_SYMBOL_GPL(locks_release_private);
 378
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 379/* Free a lock which is not in use. */
 380void locks_free_lock(struct file_lock *fl)
 381{
 382	locks_release_private(fl);
 383	kmem_cache_free(filelock_cache, fl);
 384}
 385EXPORT_SYMBOL(locks_free_lock);
 386
 
 
 
 
 
 
 
 387static void
 388locks_dispose_list(struct list_head *dispose)
 389{
 390	struct file_lock *fl;
 391
 392	while (!list_empty(dispose)) {
 393		fl = list_first_entry(dispose, struct file_lock, fl_list);
 394		list_del_init(&fl->fl_list);
 395		locks_free_lock(fl);
 
 
 
 396	}
 397}
 398
 399void locks_init_lock(struct file_lock *fl)
 400{
 401	memset(fl, 0, sizeof(struct file_lock));
 402	locks_init_lock_heads(fl);
 403}
 404EXPORT_SYMBOL(locks_init_lock);
 405
 
 
 
 
 
 
 
 406/*
 407 * Initialize a new lock from an existing file_lock structure.
 408 */
 409void locks_copy_conflock(struct file_lock *new, struct file_lock *fl)
 410{
 411	new->fl_owner = fl->fl_owner;
 412	new->fl_pid = fl->fl_pid;
 413	new->fl_file = NULL;
 414	new->fl_flags = fl->fl_flags;
 415	new->fl_type = fl->fl_type;
 416	new->fl_start = fl->fl_start;
 417	new->fl_end = fl->fl_end;
 418	new->fl_lmops = fl->fl_lmops;
 419	new->fl_ops = NULL;
 420
 421	if (fl->fl_lmops) {
 422		if (fl->fl_lmops->lm_get_owner)
 423			fl->fl_lmops->lm_get_owner(fl->fl_owner);
 424	}
 425}
 426EXPORT_SYMBOL(locks_copy_conflock);
 427
 428void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
 429{
 430	/* "new" must be a freshly-initialized lock */
 431	WARN_ON_ONCE(new->fl_ops);
 432
 433	locks_copy_conflock(new, fl);
 434
 435	new->fl_file = fl->fl_file;
 436	new->fl_ops = fl->fl_ops;
 437
 438	if (fl->fl_ops) {
 439		if (fl->fl_ops->fl_copy_lock)
 440			fl->fl_ops->fl_copy_lock(new, fl);
 441	}
 442}
 443EXPORT_SYMBOL(locks_copy_lock);
 444
 445static void locks_move_blocks(struct file_lock *new, struct file_lock *fl)
 446{
 447	struct file_lock *f;
 448
 449	/*
 450	 * As ctx->flc_lock is held, new requests cannot be added to
 451	 * ->fl_blocked_requests, so we don't need a lock to check if it
 452	 * is empty.
 453	 */
 454	if (list_empty(&fl->fl_blocked_requests))
 455		return;
 456	spin_lock(&blocked_lock_lock);
 457	list_splice_init(&fl->fl_blocked_requests, &new->fl_blocked_requests);
 458	list_for_each_entry(f, &new->fl_blocked_requests, fl_blocked_member)
 459		f->fl_blocker = new;
 
 
 460	spin_unlock(&blocked_lock_lock);
 461}
 462
 463static inline int flock_translate_cmd(int cmd) {
 464	if (cmd & LOCK_MAND)
 465		return cmd & (LOCK_MAND | LOCK_RW);
 466	switch (cmd) {
 467	case LOCK_SH:
 468		return F_RDLCK;
 469	case LOCK_EX:
 470		return F_WRLCK;
 471	case LOCK_UN:
 472		return F_UNLCK;
 473	}
 474	return -EINVAL;
 475}
 476
 477/* Fill in a file_lock structure with an appropriate FLOCK lock. */
 478static struct file_lock *
 479flock_make_lock(struct file *filp, unsigned int cmd, struct file_lock *fl)
 480{
 481	int type = flock_translate_cmd(cmd);
 482
 483	if (type < 0)
 484		return ERR_PTR(type);
 485
 486	if (fl == NULL) {
 487		fl = locks_alloc_lock();
 488		if (fl == NULL)
 489			return ERR_PTR(-ENOMEM);
 490	} else {
 491		locks_init_lock(fl);
 492	}
 493
 494	fl->fl_file = filp;
 495	fl->fl_owner = filp;
 496	fl->fl_pid = current->tgid;
 497	fl->fl_flags = FL_FLOCK;
 498	fl->fl_type = type;
 499	fl->fl_end = OFFSET_MAX;
 500
 501	return fl;
 502}
 503
 504static int assign_type(struct file_lock *fl, long type)
 505{
 506	switch (type) {
 507	case F_RDLCK:
 508	case F_WRLCK:
 509	case F_UNLCK:
 510		fl->fl_type = type;
 511		break;
 512	default:
 513		return -EINVAL;
 514	}
 515	return 0;
 516}
 517
 518static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
 519				 struct flock64 *l)
 520{
 521	switch (l->l_whence) {
 522	case SEEK_SET:
 523		fl->fl_start = 0;
 524		break;
 525	case SEEK_CUR:
 526		fl->fl_start = filp->f_pos;
 527		break;
 528	case SEEK_END:
 529		fl->fl_start = i_size_read(file_inode(filp));
 530		break;
 531	default:
 532		return -EINVAL;
 533	}
 534	if (l->l_start > OFFSET_MAX - fl->fl_start)
 535		return -EOVERFLOW;
 536	fl->fl_start += l->l_start;
 537	if (fl->fl_start < 0)
 538		return -EINVAL;
 539
 540	/* POSIX-1996 leaves the case l->l_len < 0 undefined;
 541	   POSIX-2001 defines it. */
 542	if (l->l_len > 0) {
 543		if (l->l_len - 1 > OFFSET_MAX - fl->fl_start)
 544			return -EOVERFLOW;
 545		fl->fl_end = fl->fl_start + l->l_len - 1;
 546
 547	} else if (l->l_len < 0) {
 548		if (fl->fl_start + l->l_len < 0)
 549			return -EINVAL;
 550		fl->fl_end = fl->fl_start - 1;
 551		fl->fl_start += l->l_len;
 552	} else
 553		fl->fl_end = OFFSET_MAX;
 554
 555	fl->fl_owner = current->files;
 556	fl->fl_pid = current->tgid;
 557	fl->fl_file = filp;
 558	fl->fl_flags = FL_POSIX;
 559	fl->fl_ops = NULL;
 560	fl->fl_lmops = NULL;
 561
 562	return assign_type(fl, l->l_type);
 563}
 564
 565/* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
 566 * style lock.
 567 */
 568static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
 569			       struct flock *l)
 570{
 571	struct flock64 ll = {
 572		.l_type = l->l_type,
 573		.l_whence = l->l_whence,
 574		.l_start = l->l_start,
 575		.l_len = l->l_len,
 576	};
 577
 578	return flock64_to_posix_lock(filp, fl, &ll);
 579}
 580
 581/* default lease lock manager operations */
 582static bool
 583lease_break_callback(struct file_lock *fl)
 584{
 585	kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
 586	return false;
 587}
 588
 589static void
 590lease_setup(struct file_lock *fl, void **priv)
 591{
 592	struct file *filp = fl->fl_file;
 593	struct fasync_struct *fa = *priv;
 594
 595	/*
 596	 * fasync_insert_entry() returns the old entry if any. If there was no
 597	 * old entry, then it used "priv" and inserted it into the fasync list.
 598	 * Clear the pointer to indicate that it shouldn't be freed.
 599	 */
 600	if (!fasync_insert_entry(fa->fa_fd, filp, &fl->fl_fasync, fa))
 601		*priv = NULL;
 602
 603	__f_setown(filp, task_pid(current), PIDTYPE_TGID, 0);
 604}
 605
 606static const struct lock_manager_operations lease_manager_ops = {
 607	.lm_break = lease_break_callback,
 608	.lm_change = lease_modify,
 609	.lm_setup = lease_setup,
 610};
 611
 612/*
 613 * Initialize a lease, use the default lock manager operations
 614 */
 615static int lease_init(struct file *filp, long type, struct file_lock *fl)
 616{
 617	if (assign_type(fl, type) != 0)
 618		return -EINVAL;
 619
 620	fl->fl_owner = filp;
 621	fl->fl_pid = current->tgid;
 622
 623	fl->fl_file = filp;
 624	fl->fl_flags = FL_LEASE;
 625	fl->fl_start = 0;
 626	fl->fl_end = OFFSET_MAX;
 627	fl->fl_ops = NULL;
 628	fl->fl_lmops = &lease_manager_ops;
 629	return 0;
 630}
 631
 632/* Allocate a file_lock initialised to this type of lease */
 633static struct file_lock *lease_alloc(struct file *filp, long type)
 634{
 635	struct file_lock *fl = locks_alloc_lock();
 636	int error = -ENOMEM;
 637
 638	if (fl == NULL)
 639		return ERR_PTR(error);
 640
 641	error = lease_init(filp, type, fl);
 642	if (error) {
 643		locks_free_lock(fl);
 644		return ERR_PTR(error);
 645	}
 646	return fl;
 647}
 648
 649/* Check if two locks overlap each other.
 650 */
 651static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
 652{
 653	return ((fl1->fl_end >= fl2->fl_start) &&
 654		(fl2->fl_end >= fl1->fl_start));
 655}
 656
 657/*
 658 * Check whether two locks have the same owner.
 659 */
 660static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
 661{
 662	return fl1->fl_owner == fl2->fl_owner;
 663}
 664
 665/* Must be called with the flc_lock held! */
 666static void locks_insert_global_locks(struct file_lock *fl)
 667{
 668	struct file_lock_list_struct *fll = this_cpu_ptr(&file_lock_list);
 669
 670	percpu_rwsem_assert_held(&file_rwsem);
 671
 672	spin_lock(&fll->lock);
 673	fl->fl_link_cpu = smp_processor_id();
 674	hlist_add_head(&fl->fl_link, &fll->hlist);
 675	spin_unlock(&fll->lock);
 676}
 677
 678/* Must be called with the flc_lock held! */
 679static void locks_delete_global_locks(struct file_lock *fl)
 680{
 681	struct file_lock_list_struct *fll;
 682
 683	percpu_rwsem_assert_held(&file_rwsem);
 684
 685	/*
 686	 * Avoid taking lock if already unhashed. This is safe since this check
 687	 * is done while holding the flc_lock, and new insertions into the list
 688	 * also require that it be held.
 689	 */
 690	if (hlist_unhashed(&fl->fl_link))
 691		return;
 692
 693	fll = per_cpu_ptr(&file_lock_list, fl->fl_link_cpu);
 694	spin_lock(&fll->lock);
 695	hlist_del_init(&fl->fl_link);
 696	spin_unlock(&fll->lock);
 697}
 698
 699static unsigned long
 700posix_owner_key(struct file_lock *fl)
 701{
 702	return (unsigned long)fl->fl_owner;
 703}
 704
 705static void locks_insert_global_blocked(struct file_lock *waiter)
 706{
 707	lockdep_assert_held(&blocked_lock_lock);
 708
 709	hash_add(blocked_hash, &waiter->fl_link, posix_owner_key(waiter));
 710}
 711
 712static void locks_delete_global_blocked(struct file_lock *waiter)
 713{
 714	lockdep_assert_held(&blocked_lock_lock);
 715
 716	hash_del(&waiter->fl_link);
 717}
 718
 719/* Remove waiter from blocker's block list.
 720 * When blocker ends up pointing to itself then the list is empty.
 721 *
 722 * Must be called with blocked_lock_lock held.
 723 */
 724static void __locks_delete_block(struct file_lock *waiter)
 725{
 726	locks_delete_global_blocked(waiter);
 727	list_del_init(&waiter->fl_blocked_member);
 728	waiter->fl_blocker = NULL;
 729}
 730
 731static void __locks_wake_up_blocks(struct file_lock *blocker)
 732{
 733	while (!list_empty(&blocker->fl_blocked_requests)) {
 734		struct file_lock *waiter;
 
 
 
 
 735
 736		waiter = list_first_entry(&blocker->fl_blocked_requests,
 737					  struct file_lock, fl_blocked_member);
 738		__locks_delete_block(waiter);
 739		if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
 740			waiter->fl_lmops->lm_notify(waiter);
 741		else
 742			wake_up(&waiter->fl_wait);
 
 
 
 
 
 
 
 743	}
 744}
 745
 746/**
 747 *	locks_delete_lock - stop waiting for a file lock
 748 *	@waiter: the lock which was waiting
 749 *
 750 *	lockd/nfsd need to disconnect the lock while working on it.
 751 */
 752int locks_delete_block(struct file_lock *waiter)
 753{
 754	int status = -ENOENT;
 755
 756	/*
 757	 * If fl_blocker is NULL, it won't be set again as this thread
 758	 * "owns" the lock and is the only one that might try to claim
 759	 * the lock.  So it is safe to test fl_blocker locklessly.
 760	 * Also if fl_blocker is NULL, this waiter is not listed on
 761	 * fl_blocked_requests for some lock, so no other request can
 762	 * be added to the list of fl_blocked_requests for this
 763	 * request.  So if fl_blocker is NULL, it is safe to
 764	 * locklessly check if fl_blocked_requests is empty.  If both
 765	 * of these checks succeed, there is no need to take the lock.
 
 
 
 
 
 
 
 
 
 
 766	 */
 767	if (waiter->fl_blocker == NULL &&
 768	    list_empty(&waiter->fl_blocked_requests))
 769		return status;
 
 770	spin_lock(&blocked_lock_lock);
 771	if (waiter->fl_blocker)
 772		status = 0;
 773	__locks_wake_up_blocks(waiter);
 774	__locks_delete_block(waiter);
 
 
 
 
 
 
 775	spin_unlock(&blocked_lock_lock);
 776	return status;
 777}
 
 
 
 
 
 
 
 
 
 
 
 778EXPORT_SYMBOL(locks_delete_block);
 779
 780/* Insert waiter into blocker's block list.
 781 * We use a circular list so that processes can be easily woken up in
 782 * the order they blocked. The documentation doesn't require this but
 783 * it seems like the reasonable thing to do.
 784 *
 785 * Must be called with both the flc_lock and blocked_lock_lock held. The
 786 * fl_blocked_requests list itself is protected by the blocked_lock_lock,
 787 * but by ensuring that the flc_lock is also held on insertions we can avoid
 788 * taking the blocked_lock_lock in some cases when we see that the
 789 * fl_blocked_requests list is empty.
 790 *
 791 * Rather than just adding to the list, we check for conflicts with any existing
 792 * waiters, and add beneath any waiter that blocks the new waiter.
 793 * Thus wakeups don't happen until needed.
 794 */
 795static void __locks_insert_block(struct file_lock *blocker,
 796				 struct file_lock *waiter,
 797				 bool conflict(struct file_lock *,
 798					       struct file_lock *))
 799{
 800	struct file_lock *fl;
 801	BUG_ON(!list_empty(&waiter->fl_blocked_member));
 802
 
 803new_blocker:
 804	list_for_each_entry(fl, &blocker->fl_blocked_requests, fl_blocked_member)
 805		if (conflict(fl, waiter)) {
 806			blocker =  fl;
 807			goto new_blocker;
 808		}
 809	waiter->fl_blocker = blocker;
 810	list_add_tail(&waiter->fl_blocked_member, &blocker->fl_blocked_requests);
 811	if (IS_POSIX(blocker) && !IS_OFDLCK(blocker))
 
 
 812		locks_insert_global_blocked(waiter);
 813
 814	/* The requests in waiter->fl_blocked are known to conflict with
 815	 * waiter, but might not conflict with blocker, or the requests
 816	 * and lock which block it.  So they all need to be woken.
 817	 */
 818	__locks_wake_up_blocks(waiter);
 819}
 820
 821/* Must be called with flc_lock held. */
 822static void locks_insert_block(struct file_lock *blocker,
 823			       struct file_lock *waiter,
 824			       bool conflict(struct file_lock *,
 825					     struct file_lock *))
 826{
 827	spin_lock(&blocked_lock_lock);
 828	__locks_insert_block(blocker, waiter, conflict);
 829	spin_unlock(&blocked_lock_lock);
 830}
 831
 832/*
 833 * Wake up processes blocked waiting for blocker.
 834 *
 835 * Must be called with the inode->flc_lock held!
 836 */
 837static void locks_wake_up_blocks(struct file_lock *blocker)
 838{
 839	/*
 840	 * Avoid taking global lock if list is empty. This is safe since new
 841	 * blocked requests are only added to the list under the flc_lock, and
 842	 * the flc_lock is always held here. Note that removal from the
 843	 * fl_blocked_requests list does not require the flc_lock, so we must
 844	 * recheck list_empty() after acquiring the blocked_lock_lock.
 845	 */
 846	if (list_empty(&blocker->fl_blocked_requests))
 847		return;
 848
 849	spin_lock(&blocked_lock_lock);
 850	__locks_wake_up_blocks(blocker);
 851	spin_unlock(&blocked_lock_lock);
 852}
 853
 854static void
 855locks_insert_lock_ctx(struct file_lock *fl, struct list_head *before)
 856{
 857	list_add_tail(&fl->fl_list, before);
 858	locks_insert_global_locks(fl);
 859}
 860
 861static void
 862locks_unlink_lock_ctx(struct file_lock *fl)
 863{
 864	locks_delete_global_locks(fl);
 865	list_del_init(&fl->fl_list);
 866	locks_wake_up_blocks(fl);
 867}
 868
 869static void
 870locks_delete_lock_ctx(struct file_lock *fl, struct list_head *dispose)
 871{
 872	locks_unlink_lock_ctx(fl);
 873	if (dispose)
 874		list_add(&fl->fl_list, dispose);
 875	else
 876		locks_free_lock(fl);
 877}
 878
 879/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
 880 * checks for shared/exclusive status of overlapping locks.
 881 */
 882static bool locks_conflict(struct file_lock *caller_fl,
 883			   struct file_lock *sys_fl)
 884{
 885	if (sys_fl->fl_type == F_WRLCK)
 886		return true;
 887	if (caller_fl->fl_type == F_WRLCK)
 888		return true;
 889	return false;
 890}
 891
 892/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
 893 * checking before calling the locks_conflict().
 894 */
 895static bool posix_locks_conflict(struct file_lock *caller_fl,
 896				 struct file_lock *sys_fl)
 897{
 
 
 
 898	/* POSIX locks owned by the same process do not conflict with
 899	 * each other.
 900	 */
 901	if (posix_same_owner(caller_fl, sys_fl))
 902		return false;
 903
 904	/* Check whether they overlap */
 905	if (!locks_overlap(caller_fl, sys_fl))
 906		return false;
 907
 908	return locks_conflict(caller_fl, sys_fl);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 909}
 910
 911/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
 912 * checking before calling the locks_conflict().
 913 */
 914static bool flock_locks_conflict(struct file_lock *caller_fl,
 915				 struct file_lock *sys_fl)
 916{
 917	/* FLOCK locks referring to the same filp do not conflict with
 918	 * each other.
 919	 */
 920	if (caller_fl->fl_file == sys_fl->fl_file)
 921		return false;
 922	if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
 923		return false;
 924
 925	return locks_conflict(caller_fl, sys_fl);
 926}
 927
 928void
 929posix_test_lock(struct file *filp, struct file_lock *fl)
 930{
 931	struct file_lock *cfl;
 932	struct file_lock_context *ctx;
 933	struct inode *inode = locks_inode(filp);
 
 
 934
 935	ctx = smp_load_acquire(&inode->i_flctx);
 936	if (!ctx || list_empty_careful(&ctx->flc_posix)) {
 937		fl->fl_type = F_UNLCK;
 938		return;
 939	}
 940
 
 941	spin_lock(&ctx->flc_lock);
 942	list_for_each_entry(cfl, &ctx->flc_posix, fl_list) {
 943		if (posix_locks_conflict(fl, cfl)) {
 944			locks_copy_conflock(fl, cfl);
 945			goto out;
 
 
 
 
 
 
 
 
 946		}
 
 
 947	}
 948	fl->fl_type = F_UNLCK;
 949out:
 950	spin_unlock(&ctx->flc_lock);
 951	return;
 952}
 953EXPORT_SYMBOL(posix_test_lock);
 954
 955/*
 956 * Deadlock detection:
 957 *
 958 * We attempt to detect deadlocks that are due purely to posix file
 959 * locks.
 960 *
 961 * We assume that a task can be waiting for at most one lock at a time.
 962 * So for any acquired lock, the process holding that lock may be
 963 * waiting on at most one other lock.  That lock in turns may be held by
 964 * someone waiting for at most one other lock.  Given a requested lock
 965 * caller_fl which is about to wait for a conflicting lock block_fl, we
 966 * follow this chain of waiters to ensure we are not about to create a
 967 * cycle.
 968 *
 969 * Since we do this before we ever put a process to sleep on a lock, we
 970 * are ensured that there is never a cycle; that is what guarantees that
 971 * the while() loop in posix_locks_deadlock() eventually completes.
 972 *
 973 * Note: the above assumption may not be true when handling lock
 974 * requests from a broken NFS client. It may also fail in the presence
 975 * of tasks (such as posix threads) sharing the same open file table.
 976 * To handle those cases, we just bail out after a few iterations.
 977 *
 978 * For FL_OFDLCK locks, the owner is the filp, not the files_struct.
 979 * Because the owner is not even nominally tied to a thread of
 980 * execution, the deadlock detection below can't reasonably work well. Just
 981 * skip it for those.
 982 *
 983 * In principle, we could do a more limited deadlock detection on FL_OFDLCK
 984 * locks that just checks for the case where two tasks are attempting to
 985 * upgrade from read to write locks on the same inode.
 986 */
 987
 988#define MAX_DEADLK_ITERATIONS 10
 989
 990/* Find a lock that the owner of the given block_fl is blocking on. */
 991static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
 992{
 993	struct file_lock *fl;
 994
 995	hash_for_each_possible(blocked_hash, fl, fl_link, posix_owner_key(block_fl)) {
 996		if (posix_same_owner(fl, block_fl)) {
 997			while (fl->fl_blocker)
 998				fl = fl->fl_blocker;
 999			return fl;
1000		}
1001	}
1002	return NULL;
1003}
1004
1005/* Must be called with the blocked_lock_lock held! */
1006static int posix_locks_deadlock(struct file_lock *caller_fl,
1007				struct file_lock *block_fl)
1008{
 
 
1009	int i = 0;
1010
1011	lockdep_assert_held(&blocked_lock_lock);
1012
1013	/*
1014	 * This deadlock detector can't reasonably detect deadlocks with
1015	 * FL_OFDLCK locks, since they aren't owned by a process, per-se.
1016	 */
1017	if (IS_OFDLCK(caller_fl))
1018		return 0;
1019
1020	while ((block_fl = what_owner_is_waiting_for(block_fl))) {
1021		if (i++ > MAX_DEADLK_ITERATIONS)
1022			return 0;
1023		if (posix_same_owner(caller_fl, block_fl))
1024			return 1;
1025	}
1026	return 0;
1027}
1028
1029/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
1030 * after any leases, but before any posix locks.
1031 *
1032 * Note that if called with an FL_EXISTS argument, the caller may determine
1033 * whether or not a lock was successfully freed by testing the return
1034 * value for -ENOENT.
1035 */
1036static int flock_lock_inode(struct inode *inode, struct file_lock *request)
1037{
1038	struct file_lock *new_fl = NULL;
1039	struct file_lock *fl;
1040	struct file_lock_context *ctx;
1041	int error = 0;
1042	bool found = false;
1043	LIST_HEAD(dispose);
1044
1045	ctx = locks_get_lock_context(inode, request->fl_type);
1046	if (!ctx) {
1047		if (request->fl_type != F_UNLCK)
1048			return -ENOMEM;
1049		return (request->fl_flags & FL_EXISTS) ? -ENOENT : 0;
1050	}
1051
1052	if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
1053		new_fl = locks_alloc_lock();
1054		if (!new_fl)
1055			return -ENOMEM;
1056	}
1057
1058	percpu_down_read(&file_rwsem);
1059	spin_lock(&ctx->flc_lock);
1060	if (request->fl_flags & FL_ACCESS)
1061		goto find_conflict;
1062
1063	list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
1064		if (request->fl_file != fl->fl_file)
1065			continue;
1066		if (request->fl_type == fl->fl_type)
1067			goto out;
1068		found = true;
1069		locks_delete_lock_ctx(fl, &dispose);
1070		break;
1071	}
1072
1073	if (request->fl_type == F_UNLCK) {
1074		if ((request->fl_flags & FL_EXISTS) && !found)
1075			error = -ENOENT;
1076		goto out;
1077	}
1078
1079find_conflict:
1080	list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
1081		if (!flock_locks_conflict(request, fl))
1082			continue;
1083		error = -EAGAIN;
1084		if (!(request->fl_flags & FL_SLEEP))
1085			goto out;
1086		error = FILE_LOCK_DEFERRED;
1087		locks_insert_block(fl, request, flock_locks_conflict);
1088		goto out;
1089	}
1090	if (request->fl_flags & FL_ACCESS)
1091		goto out;
1092	locks_copy_lock(new_fl, request);
1093	locks_move_blocks(new_fl, request);
1094	locks_insert_lock_ctx(new_fl, &ctx->flc_flock);
1095	new_fl = NULL;
1096	error = 0;
1097
1098out:
1099	spin_unlock(&ctx->flc_lock);
1100	percpu_up_read(&file_rwsem);
1101	if (new_fl)
1102		locks_free_lock(new_fl);
1103	locks_dispose_list(&dispose);
1104	trace_flock_lock_inode(inode, request, error);
1105	return error;
1106}
1107
1108static int posix_lock_inode(struct inode *inode, struct file_lock *request,
1109			    struct file_lock *conflock)
1110{
1111	struct file_lock *fl, *tmp;
1112	struct file_lock *new_fl = NULL;
1113	struct file_lock *new_fl2 = NULL;
1114	struct file_lock *left = NULL;
1115	struct file_lock *right = NULL;
1116	struct file_lock_context *ctx;
1117	int error;
1118	bool added = false;
1119	LIST_HEAD(dispose);
 
 
1120
1121	ctx = locks_get_lock_context(inode, request->fl_type);
1122	if (!ctx)
1123		return (request->fl_type == F_UNLCK) ? 0 : -ENOMEM;
1124
1125	/*
1126	 * We may need two file_lock structures for this operation,
1127	 * so we get them in advance to avoid races.
1128	 *
1129	 * In some cases we can be sure, that no new locks will be needed
1130	 */
1131	if (!(request->fl_flags & FL_ACCESS) &&
1132	    (request->fl_type != F_UNLCK ||
1133	     request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
1134		new_fl = locks_alloc_lock();
1135		new_fl2 = locks_alloc_lock();
1136	}
1137
 
1138	percpu_down_read(&file_rwsem);
1139	spin_lock(&ctx->flc_lock);
1140	/*
1141	 * New lock request. Walk all POSIX locks and look for conflicts. If
1142	 * there are any, either return error or put the request on the
1143	 * blocker's list of waiters and the global blocked_hash.
1144	 */
1145	if (request->fl_type != F_UNLCK) {
1146		list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1147			if (!posix_locks_conflict(request, fl))
1148				continue;
 
 
 
 
 
 
 
 
 
 
 
1149			if (conflock)
1150				locks_copy_conflock(conflock, fl);
1151			error = -EAGAIN;
1152			if (!(request->fl_flags & FL_SLEEP))
1153				goto out;
1154			/*
1155			 * Deadlock detection and insertion into the blocked
1156			 * locks list must be done while holding the same lock!
1157			 */
1158			error = -EDEADLK;
1159			spin_lock(&blocked_lock_lock);
1160			/*
1161			 * Ensure that we don't find any locks blocked on this
1162			 * request during deadlock detection.
1163			 */
1164			__locks_wake_up_blocks(request);
1165			if (likely(!posix_locks_deadlock(request, fl))) {
1166				error = FILE_LOCK_DEFERRED;
1167				__locks_insert_block(fl, request,
1168						     posix_locks_conflict);
1169			}
1170			spin_unlock(&blocked_lock_lock);
1171			goto out;
1172		}
1173	}
1174
1175	/* If we're just looking for a conflict, we're done. */
1176	error = 0;
1177	if (request->fl_flags & FL_ACCESS)
1178		goto out;
1179
1180	/* Find the first old lock with the same owner as the new lock */
1181	list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1182		if (posix_same_owner(request, fl))
1183			break;
1184	}
1185
1186	/* Process locks with this owner. */
1187	list_for_each_entry_safe_from(fl, tmp, &ctx->flc_posix, fl_list) {
1188		if (!posix_same_owner(request, fl))
1189			break;
1190
1191		/* Detect adjacent or overlapping regions (if same lock type) */
1192		if (request->fl_type == fl->fl_type) {
1193			/* In all comparisons of start vs end, use
1194			 * "start - 1" rather than "end + 1". If end
1195			 * is OFFSET_MAX, end + 1 will become negative.
1196			 */
1197			if (fl->fl_end < request->fl_start - 1)
1198				continue;
1199			/* If the next lock in the list has entirely bigger
1200			 * addresses than the new one, insert the lock here.
1201			 */
1202			if (fl->fl_start - 1 > request->fl_end)
1203				break;
1204
1205			/* If we come here, the new and old lock are of the
1206			 * same type and adjacent or overlapping. Make one
1207			 * lock yielding from the lower start address of both
1208			 * locks to the higher end address.
1209			 */
1210			if (fl->fl_start > request->fl_start)
1211				fl->fl_start = request->fl_start;
1212			else
1213				request->fl_start = fl->fl_start;
1214			if (fl->fl_end < request->fl_end)
1215				fl->fl_end = request->fl_end;
1216			else
1217				request->fl_end = fl->fl_end;
1218			if (added) {
1219				locks_delete_lock_ctx(fl, &dispose);
1220				continue;
1221			}
1222			request = fl;
1223			added = true;
1224		} else {
1225			/* Processing for different lock types is a bit
1226			 * more complex.
1227			 */
1228			if (fl->fl_end < request->fl_start)
1229				continue;
1230			if (fl->fl_start > request->fl_end)
1231				break;
1232			if (request->fl_type == F_UNLCK)
1233				added = true;
1234			if (fl->fl_start < request->fl_start)
1235				left = fl;
1236			/* If the next lock in the list has a higher end
1237			 * address than the new one, insert the new one here.
1238			 */
1239			if (fl->fl_end > request->fl_end) {
1240				right = fl;
1241				break;
1242			}
1243			if (fl->fl_start >= request->fl_start) {
1244				/* The new lock completely replaces an old
1245				 * one (This may happen several times).
1246				 */
1247				if (added) {
1248					locks_delete_lock_ctx(fl, &dispose);
1249					continue;
1250				}
1251				/*
1252				 * Replace the old lock with new_fl, and
1253				 * remove the old one. It's safe to do the
1254				 * insert here since we know that we won't be
1255				 * using new_fl later, and that the lock is
1256				 * just replacing an existing lock.
1257				 */
1258				error = -ENOLCK;
1259				if (!new_fl)
1260					goto out;
1261				locks_copy_lock(new_fl, request);
 
1262				request = new_fl;
1263				new_fl = NULL;
1264				locks_insert_lock_ctx(request, &fl->fl_list);
1265				locks_delete_lock_ctx(fl, &dispose);
 
1266				added = true;
1267			}
1268		}
1269	}
1270
1271	/*
1272	 * The above code only modifies existing locks in case of merging or
1273	 * replacing. If new lock(s) need to be inserted all modifications are
1274	 * done below this, so it's safe yet to bail out.
1275	 */
1276	error = -ENOLCK; /* "no luck" */
1277	if (right && left == right && !new_fl2)
1278		goto out;
1279
1280	error = 0;
1281	if (!added) {
1282		if (request->fl_type == F_UNLCK) {
1283			if (request->fl_flags & FL_EXISTS)
1284				error = -ENOENT;
1285			goto out;
1286		}
1287
1288		if (!new_fl) {
1289			error = -ENOLCK;
1290			goto out;
1291		}
1292		locks_copy_lock(new_fl, request);
1293		locks_move_blocks(new_fl, request);
1294		locks_insert_lock_ctx(new_fl, &fl->fl_list);
1295		fl = new_fl;
1296		new_fl = NULL;
1297	}
1298	if (right) {
1299		if (left == right) {
1300			/* The new lock breaks the old one in two pieces,
1301			 * so we have to use the second new lock.
1302			 */
1303			left = new_fl2;
1304			new_fl2 = NULL;
1305			locks_copy_lock(left, right);
1306			locks_insert_lock_ctx(left, &fl->fl_list);
1307		}
1308		right->fl_start = request->fl_end + 1;
1309		locks_wake_up_blocks(right);
1310	}
1311	if (left) {
1312		left->fl_end = request->fl_start - 1;
1313		locks_wake_up_blocks(left);
1314	}
1315 out:
 
1316	spin_unlock(&ctx->flc_lock);
1317	percpu_up_read(&file_rwsem);
1318	/*
1319	 * Free any unused locks.
1320	 */
1321	if (new_fl)
1322		locks_free_lock(new_fl);
1323	if (new_fl2)
1324		locks_free_lock(new_fl2);
1325	locks_dispose_list(&dispose);
1326	trace_posix_lock_inode(inode, request, error);
1327
1328	return error;
1329}
1330
1331/**
1332 * posix_lock_file - Apply a POSIX-style lock to a file
1333 * @filp: The file to apply the lock to
1334 * @fl: The lock to be applied
1335 * @conflock: Place to return a copy of the conflicting lock, if found.
1336 *
1337 * Add a POSIX style lock to a file.
1338 * We merge adjacent & overlapping locks whenever possible.
1339 * POSIX locks are sorted by owner task, then by starting address
1340 *
1341 * Note that if called with an FL_EXISTS argument, the caller may determine
1342 * whether or not a lock was successfully freed by testing the return
1343 * value for -ENOENT.
1344 */
1345int posix_lock_file(struct file *filp, struct file_lock *fl,
1346			struct file_lock *conflock)
1347{
1348	return posix_lock_inode(locks_inode(filp), fl, conflock);
1349}
1350EXPORT_SYMBOL(posix_lock_file);
1351
1352/**
1353 * posix_lock_inode_wait - Apply a POSIX-style lock to a file
1354 * @inode: inode of file to which lock request should be applied
1355 * @fl: The lock to be applied
1356 *
1357 * Apply a POSIX style lock request to an inode.
1358 */
1359static int posix_lock_inode_wait(struct inode *inode, struct file_lock *fl)
1360{
1361	int error;
1362	might_sleep ();
1363	for (;;) {
1364		error = posix_lock_inode(inode, fl, NULL);
1365		if (error != FILE_LOCK_DEFERRED)
1366			break;
1367		error = wait_event_interruptible(fl->fl_wait, !fl->fl_blocker);
 
1368		if (error)
1369			break;
1370	}
1371	locks_delete_block(fl);
1372	return error;
1373}
1374
1375#ifdef CONFIG_MANDATORY_FILE_LOCKING
1376/**
1377 * locks_mandatory_locked - Check for an active lock
1378 * @file: the file to check
1379 *
1380 * Searches the inode's list of locks to find any POSIX locks which conflict.
1381 * This function is called from locks_verify_locked() only.
1382 */
1383int locks_mandatory_locked(struct file *file)
1384{
1385	int ret;
1386	struct inode *inode = locks_inode(file);
1387	struct file_lock_context *ctx;
1388	struct file_lock *fl;
1389
1390	ctx = smp_load_acquire(&inode->i_flctx);
1391	if (!ctx || list_empty_careful(&ctx->flc_posix))
1392		return 0;
1393
1394	/*
1395	 * Search the lock list for this inode for any POSIX locks.
1396	 */
1397	spin_lock(&ctx->flc_lock);
1398	ret = 0;
1399	list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1400		if (fl->fl_owner != current->files &&
1401		    fl->fl_owner != file) {
1402			ret = -EAGAIN;
1403			break;
1404		}
1405	}
1406	spin_unlock(&ctx->flc_lock);
1407	return ret;
1408}
1409
1410/**
1411 * locks_mandatory_area - Check for a conflicting lock
1412 * @inode:	the file to check
1413 * @filp:       how the file was opened (if it was)
1414 * @start:	first byte in the file to check
1415 * @end:	lastbyte in the file to check
1416 * @type:	%F_WRLCK for a write lock, else %F_RDLCK
1417 *
1418 * Searches the inode's list of locks to find any POSIX locks which conflict.
1419 */
1420int locks_mandatory_area(struct inode *inode, struct file *filp, loff_t start,
1421			 loff_t end, unsigned char type)
1422{
1423	struct file_lock fl;
1424	int error;
1425	bool sleep = false;
1426
1427	locks_init_lock(&fl);
1428	fl.fl_pid = current->tgid;
1429	fl.fl_file = filp;
1430	fl.fl_flags = FL_POSIX | FL_ACCESS;
1431	if (filp && !(filp->f_flags & O_NONBLOCK))
1432		sleep = true;
1433	fl.fl_type = type;
1434	fl.fl_start = start;
1435	fl.fl_end = end;
1436
1437	for (;;) {
1438		if (filp) {
1439			fl.fl_owner = filp;
1440			fl.fl_flags &= ~FL_SLEEP;
1441			error = posix_lock_inode(inode, &fl, NULL);
1442			if (!error)
1443				break;
1444		}
1445
1446		if (sleep)
1447			fl.fl_flags |= FL_SLEEP;
1448		fl.fl_owner = current->files;
1449		error = posix_lock_inode(inode, &fl, NULL);
1450		if (error != FILE_LOCK_DEFERRED)
1451			break;
1452		error = wait_event_interruptible(fl.fl_wait, !fl.fl_blocker);
1453		if (!error) {
1454			/*
1455			 * If we've been sleeping someone might have
1456			 * changed the permissions behind our back.
1457			 */
1458			if (__mandatory_lock(inode))
1459				continue;
1460		}
1461
1462		break;
1463	}
1464	locks_delete_block(&fl);
1465
1466	return error;
1467}
1468EXPORT_SYMBOL(locks_mandatory_area);
1469#endif /* CONFIG_MANDATORY_FILE_LOCKING */
1470
1471static void lease_clear_pending(struct file_lock *fl, int arg)
1472{
1473	switch (arg) {
1474	case F_UNLCK:
1475		fl->fl_flags &= ~FL_UNLOCK_PENDING;
1476		/* fall through */
1477	case F_RDLCK:
1478		fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1479	}
1480}
1481
1482/* We already had a lease on this file; just change its type */
1483int lease_modify(struct file_lock *fl, int arg, struct list_head *dispose)
1484{
1485	int error = assign_type(fl, arg);
1486
1487	if (error)
1488		return error;
1489	lease_clear_pending(fl, arg);
1490	locks_wake_up_blocks(fl);
1491	if (arg == F_UNLCK) {
1492		struct file *filp = fl->fl_file;
1493
1494		f_delown(filp);
1495		filp->f_owner.signum = 0;
1496		fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
1497		if (fl->fl_fasync != NULL) {
1498			printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1499			fl->fl_fasync = NULL;
1500		}
1501		locks_delete_lock_ctx(fl, dispose);
1502	}
1503	return 0;
1504}
1505EXPORT_SYMBOL(lease_modify);
1506
1507static bool past_time(unsigned long then)
1508{
1509	if (!then)
1510		/* 0 is a special value meaning "this never expires": */
1511		return false;
1512	return time_after(jiffies, then);
1513}
1514
1515static void time_out_leases(struct inode *inode, struct list_head *dispose)
1516{
1517	struct file_lock_context *ctx = inode->i_flctx;
1518	struct file_lock *fl, *tmp;
1519
1520	lockdep_assert_held(&ctx->flc_lock);
1521
1522	list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
1523		trace_time_out_leases(inode, fl);
1524		if (past_time(fl->fl_downgrade_time))
1525			lease_modify(fl, F_RDLCK, dispose);
1526		if (past_time(fl->fl_break_time))
1527			lease_modify(fl, F_UNLCK, dispose);
1528	}
1529}
1530
1531static bool leases_conflict(struct file_lock *lease, struct file_lock *breaker)
1532{
1533	bool rc;
 
 
1534
1535	if ((breaker->fl_flags & FL_LAYOUT) != (lease->fl_flags & FL_LAYOUT)) {
 
 
 
1536		rc = false;
1537		goto trace;
1538	}
1539	if ((breaker->fl_flags & FL_DELEG) && (lease->fl_flags & FL_LEASE)) {
1540		rc = false;
1541		goto trace;
1542	}
1543
1544	rc = locks_conflict(breaker, lease);
1545trace:
1546	trace_leases_conflict(rc, lease, breaker);
1547	return rc;
1548}
1549
1550static bool
1551any_leases_conflict(struct inode *inode, struct file_lock *breaker)
1552{
1553	struct file_lock_context *ctx = inode->i_flctx;
1554	struct file_lock *fl;
1555
1556	lockdep_assert_held(&ctx->flc_lock);
1557
1558	list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1559		if (leases_conflict(fl, breaker))
1560			return true;
1561	}
1562	return false;
1563}
1564
1565/**
1566 *	__break_lease	-	revoke all outstanding leases on file
1567 *	@inode: the inode of the file to return
1568 *	@mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR:
1569 *	    break all leases
1570 *	@type: FL_LEASE: break leases and delegations; FL_DELEG: break
1571 *	    only delegations
1572 *
1573 *	break_lease (inlined for speed) has checked there already is at least
1574 *	some kind of lock (maybe a lease) on this file.  Leases are broken on
1575 *	a call to open() or truncate().  This function can sleep unless you
1576 *	specified %O_NONBLOCK to your open().
1577 */
1578int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
1579{
1580	int error = 0;
1581	struct file_lock_context *ctx;
1582	struct file_lock *new_fl, *fl, *tmp;
1583	unsigned long break_time;
1584	int want_write = (mode & O_ACCMODE) != O_RDONLY;
1585	LIST_HEAD(dispose);
1586
1587	new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
1588	if (IS_ERR(new_fl))
1589		return PTR_ERR(new_fl);
1590	new_fl->fl_flags = type;
1591
1592	/* typically we will check that ctx is non-NULL before calling */
1593	ctx = smp_load_acquire(&inode->i_flctx);
1594	if (!ctx) {
1595		WARN_ON_ONCE(1);
1596		goto free_lock;
1597	}
1598
1599	percpu_down_read(&file_rwsem);
1600	spin_lock(&ctx->flc_lock);
1601
1602	time_out_leases(inode, &dispose);
1603
1604	if (!any_leases_conflict(inode, new_fl))
1605		goto out;
1606
1607	break_time = 0;
1608	if (lease_break_time > 0) {
1609		break_time = jiffies + lease_break_time * HZ;
1610		if (break_time == 0)
1611			break_time++;	/* so that 0 means no break time */
1612	}
1613
1614	list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
1615		if (!leases_conflict(fl, new_fl))
1616			continue;
1617		if (want_write) {
1618			if (fl->fl_flags & FL_UNLOCK_PENDING)
1619				continue;
1620			fl->fl_flags |= FL_UNLOCK_PENDING;
1621			fl->fl_break_time = break_time;
1622		} else {
1623			if (lease_breaking(fl))
1624				continue;
1625			fl->fl_flags |= FL_DOWNGRADE_PENDING;
1626			fl->fl_downgrade_time = break_time;
1627		}
1628		if (fl->fl_lmops->lm_break(fl))
1629			locks_delete_lock_ctx(fl, &dispose);
1630	}
1631
1632	if (list_empty(&ctx->flc_lease))
1633		goto out;
1634
1635	if (mode & O_NONBLOCK) {
1636		trace_break_lease_noblock(inode, new_fl);
1637		error = -EWOULDBLOCK;
1638		goto out;
1639	}
1640
1641restart:
1642	fl = list_first_entry(&ctx->flc_lease, struct file_lock, fl_list);
1643	break_time = fl->fl_break_time;
1644	if (break_time != 0)
1645		break_time -= jiffies;
1646	if (break_time == 0)
1647		break_time++;
1648	locks_insert_block(fl, new_fl, leases_conflict);
1649	trace_break_lease_block(inode, new_fl);
1650	spin_unlock(&ctx->flc_lock);
1651	percpu_up_read(&file_rwsem);
1652
1653	locks_dispose_list(&dispose);
1654	error = wait_event_interruptible_timeout(new_fl->fl_wait,
1655						!new_fl->fl_blocker, break_time);
 
1656
1657	percpu_down_read(&file_rwsem);
1658	spin_lock(&ctx->flc_lock);
1659	trace_break_lease_unblock(inode, new_fl);
1660	locks_delete_block(new_fl);
1661	if (error >= 0) {
1662		/*
1663		 * Wait for the next conflicting lease that has not been
1664		 * broken yet
1665		 */
1666		if (error == 0)
1667			time_out_leases(inode, &dispose);
1668		if (any_leases_conflict(inode, new_fl))
1669			goto restart;
1670		error = 0;
1671	}
1672out:
1673	spin_unlock(&ctx->flc_lock);
1674	percpu_up_read(&file_rwsem);
1675	locks_dispose_list(&dispose);
1676free_lock:
1677	locks_free_lock(new_fl);
1678	return error;
1679}
1680EXPORT_SYMBOL(__break_lease);
1681
1682/**
1683 *	lease_get_mtime - update modified time of an inode with exclusive lease
1684 *	@inode: the inode
1685 *      @time:  pointer to a timespec which contains the last modified time
1686 *
1687 * This is to force NFS clients to flush their caches for files with
1688 * exclusive leases.  The justification is that if someone has an
1689 * exclusive lease, then they could be modifying it.
1690 */
1691void lease_get_mtime(struct inode *inode, struct timespec64 *time)
1692{
1693	bool has_lease = false;
1694	struct file_lock_context *ctx;
1695	struct file_lock *fl;
1696
1697	ctx = smp_load_acquire(&inode->i_flctx);
1698	if (ctx && !list_empty_careful(&ctx->flc_lease)) {
1699		spin_lock(&ctx->flc_lock);
1700		fl = list_first_entry_or_null(&ctx->flc_lease,
1701					      struct file_lock, fl_list);
1702		if (fl && (fl->fl_type == F_WRLCK))
1703			has_lease = true;
1704		spin_unlock(&ctx->flc_lock);
1705	}
1706
1707	if (has_lease)
1708		*time = current_time(inode);
1709}
1710EXPORT_SYMBOL(lease_get_mtime);
1711
1712/**
1713 *	fcntl_getlease - Enquire what lease is currently active
1714 *	@filp: the file
1715 *
1716 *	The value returned by this function will be one of
1717 *	(if no lease break is pending):
1718 *
1719 *	%F_RDLCK to indicate a shared lease is held.
1720 *
1721 *	%F_WRLCK to indicate an exclusive lease is held.
1722 *
1723 *	%F_UNLCK to indicate no lease is held.
1724 *
1725 *	(if a lease break is pending):
1726 *
1727 *	%F_RDLCK to indicate an exclusive lease needs to be
1728 *		changed to a shared lease (or removed).
1729 *
1730 *	%F_UNLCK to indicate the lease needs to be removed.
1731 *
1732 *	XXX: sfr & willy disagree over whether F_INPROGRESS
1733 *	should be returned to userspace.
1734 */
1735int fcntl_getlease(struct file *filp)
1736{
1737	struct file_lock *fl;
1738	struct inode *inode = locks_inode(filp);
1739	struct file_lock_context *ctx;
1740	int type = F_UNLCK;
1741	LIST_HEAD(dispose);
1742
1743	ctx = smp_load_acquire(&inode->i_flctx);
1744	if (ctx && !list_empty_careful(&ctx->flc_lease)) {
1745		percpu_down_read(&file_rwsem);
1746		spin_lock(&ctx->flc_lock);
1747		time_out_leases(inode, &dispose);
1748		list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1749			if (fl->fl_file != filp)
1750				continue;
1751			type = target_leasetype(fl);
1752			break;
1753		}
1754		spin_unlock(&ctx->flc_lock);
1755		percpu_up_read(&file_rwsem);
1756
1757		locks_dispose_list(&dispose);
1758	}
1759	return type;
1760}
1761
1762/**
1763 * check_conflicting_open - see if the given file points to an inode that has
1764 *			    an existing open that would conflict with the
1765 *			    desired lease.
1766 * @filp:	file to check
1767 * @arg:	type of lease that we're trying to acquire
1768 * @flags:	current lock flags
1769 *
1770 * Check to see if there's an existing open fd on this file that would
1771 * conflict with the lease we're trying to set.
1772 */
1773static int
1774check_conflicting_open(struct file *filp, const long arg, int flags)
1775{
1776	struct inode *inode = locks_inode(filp);
1777	int self_wcount = 0, self_rcount = 0;
1778
1779	if (flags & FL_LAYOUT)
1780		return 0;
 
 
 
1781
1782	if (arg == F_RDLCK)
1783		return inode_is_open_for_write(inode) ? -EAGAIN : 0;
1784	else if (arg != F_WRLCK)
1785		return 0;
1786
1787	/*
1788	 * Make sure that only read/write count is from lease requestor.
1789	 * Note that this will result in denying write leases when i_writecount
1790	 * is negative, which is what we want.  (We shouldn't grant write leases
1791	 * on files open for execution.)
1792	 */
1793	if (filp->f_mode & FMODE_WRITE)
1794		self_wcount = 1;
1795	else if (filp->f_mode & FMODE_READ)
1796		self_rcount = 1;
1797
1798	if (atomic_read(&inode->i_writecount) != self_wcount ||
1799	    atomic_read(&inode->i_readcount) != self_rcount)
1800		return -EAGAIN;
1801
1802	return 0;
1803}
1804
1805static int
1806generic_add_lease(struct file *filp, long arg, struct file_lock **flp, void **priv)
1807{
1808	struct file_lock *fl, *my_fl = NULL, *lease;
1809	struct inode *inode = locks_inode(filp);
1810	struct file_lock_context *ctx;
1811	bool is_deleg = (*flp)->fl_flags & FL_DELEG;
1812	int error;
1813	LIST_HEAD(dispose);
1814
1815	lease = *flp;
1816	trace_generic_add_lease(inode, lease);
1817
 
 
 
 
1818	/* Note that arg is never F_UNLCK here */
1819	ctx = locks_get_lock_context(inode, arg);
1820	if (!ctx)
1821		return -ENOMEM;
1822
1823	/*
1824	 * In the delegation case we need mutual exclusion with
1825	 * a number of operations that take the i_mutex.  We trylock
1826	 * because delegations are an optional optimization, and if
1827	 * there's some chance of a conflict--we'd rather not
1828	 * bother, maybe that's a sign this just isn't a good file to
1829	 * hand out a delegation on.
1830	 */
1831	if (is_deleg && !inode_trylock(inode))
1832		return -EAGAIN;
1833
1834	if (is_deleg && arg == F_WRLCK) {
1835		/* Write delegations are not currently supported: */
1836		inode_unlock(inode);
1837		WARN_ON_ONCE(1);
1838		return -EINVAL;
1839	}
1840
1841	percpu_down_read(&file_rwsem);
1842	spin_lock(&ctx->flc_lock);
1843	time_out_leases(inode, &dispose);
1844	error = check_conflicting_open(filp, arg, lease->fl_flags);
1845	if (error)
1846		goto out;
1847
1848	/*
1849	 * At this point, we know that if there is an exclusive
1850	 * lease on this file, then we hold it on this filp
1851	 * (otherwise our open of this file would have blocked).
1852	 * And if we are trying to acquire an exclusive lease,
1853	 * then the file is not open by anyone (including us)
1854	 * except for this filp.
1855	 */
1856	error = -EAGAIN;
1857	list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1858		if (fl->fl_file == filp &&
1859		    fl->fl_owner == lease->fl_owner) {
1860			my_fl = fl;
1861			continue;
1862		}
1863
1864		/*
1865		 * No exclusive leases if someone else has a lease on
1866		 * this file:
1867		 */
1868		if (arg == F_WRLCK)
1869			goto out;
1870		/*
1871		 * Modifying our existing lease is OK, but no getting a
1872		 * new lease if someone else is opening for write:
1873		 */
1874		if (fl->fl_flags & FL_UNLOCK_PENDING)
1875			goto out;
1876	}
1877
1878	if (my_fl != NULL) {
1879		lease = my_fl;
1880		error = lease->fl_lmops->lm_change(lease, arg, &dispose);
1881		if (error)
1882			goto out;
1883		goto out_setup;
1884	}
1885
1886	error = -EINVAL;
1887	if (!leases_enable)
1888		goto out;
1889
1890	locks_insert_lock_ctx(lease, &ctx->flc_lease);
1891	/*
1892	 * The check in break_lease() is lockless. It's possible for another
1893	 * open to race in after we did the earlier check for a conflicting
1894	 * open but before the lease was inserted. Check again for a
1895	 * conflicting open and cancel the lease if there is one.
1896	 *
1897	 * We also add a barrier here to ensure that the insertion of the lock
1898	 * precedes these checks.
1899	 */
1900	smp_mb();
1901	error = check_conflicting_open(filp, arg, lease->fl_flags);
1902	if (error) {
1903		locks_unlink_lock_ctx(lease);
1904		goto out;
1905	}
1906
1907out_setup:
1908	if (lease->fl_lmops->lm_setup)
1909		lease->fl_lmops->lm_setup(lease, priv);
1910out:
1911	spin_unlock(&ctx->flc_lock);
1912	percpu_up_read(&file_rwsem);
1913	locks_dispose_list(&dispose);
1914	if (is_deleg)
1915		inode_unlock(inode);
1916	if (!error && !my_fl)
1917		*flp = NULL;
1918	return error;
1919}
1920
1921static int generic_delete_lease(struct file *filp, void *owner)
1922{
1923	int error = -EAGAIN;
1924	struct file_lock *fl, *victim = NULL;
1925	struct inode *inode = locks_inode(filp);
1926	struct file_lock_context *ctx;
1927	LIST_HEAD(dispose);
1928
1929	ctx = smp_load_acquire(&inode->i_flctx);
1930	if (!ctx) {
1931		trace_generic_delete_lease(inode, NULL);
1932		return error;
1933	}
1934
1935	percpu_down_read(&file_rwsem);
1936	spin_lock(&ctx->flc_lock);
1937	list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1938		if (fl->fl_file == filp &&
1939		    fl->fl_owner == owner) {
1940			victim = fl;
1941			break;
1942		}
1943	}
1944	trace_generic_delete_lease(inode, victim);
1945	if (victim)
1946		error = fl->fl_lmops->lm_change(victim, F_UNLCK, &dispose);
1947	spin_unlock(&ctx->flc_lock);
1948	percpu_up_read(&file_rwsem);
1949	locks_dispose_list(&dispose);
1950	return error;
1951}
1952
1953/**
1954 *	generic_setlease	-	sets a lease on an open file
1955 *	@filp:	file pointer
1956 *	@arg:	type of lease to obtain
1957 *	@flp:	input - file_lock to use, output - file_lock inserted
1958 *	@priv:	private data for lm_setup (may be NULL if lm_setup
1959 *		doesn't require it)
1960 *
1961 *	The (input) flp->fl_lmops->lm_break function is required
1962 *	by break_lease().
1963 */
1964int generic_setlease(struct file *filp, long arg, struct file_lock **flp,
1965			void **priv)
1966{
1967	struct inode *inode = locks_inode(filp);
1968	int error;
1969
1970	if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
1971		return -EACCES;
1972	if (!S_ISREG(inode->i_mode))
1973		return -EINVAL;
1974	error = security_file_lock(filp, arg);
1975	if (error)
1976		return error;
1977
1978	switch (arg) {
1979	case F_UNLCK:
1980		return generic_delete_lease(filp, *priv);
1981	case F_RDLCK:
1982	case F_WRLCK:
1983		if (!(*flp)->fl_lmops->lm_break) {
1984			WARN_ON_ONCE(1);
1985			return -ENOLCK;
1986		}
1987
1988		return generic_add_lease(filp, arg, flp, priv);
1989	default:
1990		return -EINVAL;
1991	}
1992}
1993EXPORT_SYMBOL(generic_setlease);
1994
1995#if IS_ENABLED(CONFIG_SRCU)
1996/*
1997 * Kernel subsystems can register to be notified on any attempt to set
1998 * a new lease with the lease_notifier_chain. This is used by (e.g.) nfsd
1999 * to close files that it may have cached when there is an attempt to set a
2000 * conflicting lease.
2001 */
2002static struct srcu_notifier_head lease_notifier_chain;
2003
2004static inline void
2005lease_notifier_chain_init(void)
2006{
2007	srcu_init_notifier_head(&lease_notifier_chain);
2008}
2009
2010static inline void
2011setlease_notifier(long arg, struct file_lock *lease)
2012{
2013	if (arg != F_UNLCK)
2014		srcu_notifier_call_chain(&lease_notifier_chain, arg, lease);
2015}
2016
2017int lease_register_notifier(struct notifier_block *nb)
2018{
2019	return srcu_notifier_chain_register(&lease_notifier_chain, nb);
2020}
2021EXPORT_SYMBOL_GPL(lease_register_notifier);
2022
2023void lease_unregister_notifier(struct notifier_block *nb)
2024{
2025	srcu_notifier_chain_unregister(&lease_notifier_chain, nb);
2026}
2027EXPORT_SYMBOL_GPL(lease_unregister_notifier);
2028
2029#else /* !IS_ENABLED(CONFIG_SRCU) */
2030static inline void
2031lease_notifier_chain_init(void)
2032{
2033}
2034
2035static inline void
2036setlease_notifier(long arg, struct file_lock *lease)
2037{
2038}
2039
2040int lease_register_notifier(struct notifier_block *nb)
2041{
2042	return 0;
2043}
2044EXPORT_SYMBOL_GPL(lease_register_notifier);
2045
2046void lease_unregister_notifier(struct notifier_block *nb)
 
2047{
 
 
 
 
 
 
2048}
2049EXPORT_SYMBOL_GPL(lease_unregister_notifier);
2050
2051#endif /* IS_ENABLED(CONFIG_SRCU) */
2052
2053/**
2054 * vfs_setlease        -       sets a lease on an open file
2055 * @filp:	file pointer
2056 * @arg:	type of lease to obtain
2057 * @lease:	file_lock to use when adding a lease
2058 * @priv:	private info for lm_setup when adding a lease (may be
2059 *		NULL if lm_setup doesn't require it)
2060 *
2061 * Call this to establish a lease on the file. The "lease" argument is not
2062 * used for F_UNLCK requests and may be NULL. For commands that set or alter
2063 * an existing lease, the ``(*lease)->fl_lmops->lm_break`` operation must be
2064 * set; if not, this function will return -ENOLCK (and generate a scary-looking
2065 * stack trace).
2066 *
2067 * The "priv" pointer is passed directly to the lm_setup function as-is. It
2068 * may be NULL if the lm_setup operation doesn't require it.
2069 */
2070int
2071vfs_setlease(struct file *filp, long arg, struct file_lock **lease, void **priv)
2072{
2073	if (lease)
2074		setlease_notifier(arg, *lease);
2075	if (filp->f_op->setlease)
2076		return filp->f_op->setlease(filp, arg, lease, priv);
2077	else
2078		return generic_setlease(filp, arg, lease, priv);
 
 
 
 
 
 
2079}
2080EXPORT_SYMBOL_GPL(vfs_setlease);
2081
2082static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
2083{
2084	struct file_lock *fl;
2085	struct fasync_struct *new;
2086	int error;
2087
2088	fl = lease_alloc(filp, arg);
2089	if (IS_ERR(fl))
2090		return PTR_ERR(fl);
2091
2092	new = fasync_alloc();
2093	if (!new) {
2094		locks_free_lock(fl);
2095		return -ENOMEM;
2096	}
2097	new->fa_fd = fd;
2098
2099	error = vfs_setlease(filp, arg, &fl, (void **)&new);
2100	if (fl)
2101		locks_free_lock(fl);
2102	if (new)
2103		fasync_free(new);
2104	return error;
2105}
2106
2107/**
2108 *	fcntl_setlease	-	sets a lease on an open file
2109 *	@fd: open file descriptor
2110 *	@filp: file pointer
2111 *	@arg: type of lease to obtain
2112 *
2113 *	Call this fcntl to establish a lease on the file.
2114 *	Note that you also need to call %F_SETSIG to
2115 *	receive a signal when the lease is broken.
2116 */
2117int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
2118{
2119	if (arg == F_UNLCK)
2120		return vfs_setlease(filp, F_UNLCK, NULL, (void **)&filp);
2121	return do_fcntl_add_lease(fd, filp, arg);
2122}
2123
2124/**
2125 * flock_lock_inode_wait - Apply a FLOCK-style lock to a file
2126 * @inode: inode of the file to apply to
2127 * @fl: The lock to be applied
2128 *
2129 * Apply a FLOCK style lock request to an inode.
2130 */
2131static int flock_lock_inode_wait(struct inode *inode, struct file_lock *fl)
2132{
2133	int error;
2134	might_sleep();
2135	for (;;) {
2136		error = flock_lock_inode(inode, fl);
2137		if (error != FILE_LOCK_DEFERRED)
2138			break;
2139		error = wait_event_interruptible(fl->fl_wait, !fl->fl_blocker);
 
2140		if (error)
2141			break;
2142	}
2143	locks_delete_block(fl);
2144	return error;
2145}
2146
2147/**
2148 * locks_lock_inode_wait - Apply a lock to an inode
2149 * @inode: inode of the file to apply to
2150 * @fl: The lock to be applied
2151 *
2152 * Apply a POSIX or FLOCK style lock request to an inode.
2153 */
2154int locks_lock_inode_wait(struct inode *inode, struct file_lock *fl)
2155{
2156	int res = 0;
2157	switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
2158		case FL_POSIX:
2159			res = posix_lock_inode_wait(inode, fl);
2160			break;
2161		case FL_FLOCK:
2162			res = flock_lock_inode_wait(inode, fl);
2163			break;
2164		default:
2165			BUG();
2166	}
2167	return res;
2168}
2169EXPORT_SYMBOL(locks_lock_inode_wait);
2170
2171/**
2172 *	sys_flock: - flock() system call.
2173 *	@fd: the file descriptor to lock.
2174 *	@cmd: the type of lock to apply.
2175 *
2176 *	Apply a %FL_FLOCK style lock to an open file descriptor.
2177 *	The @cmd can be one of:
2178 *
2179 *	- %LOCK_SH -- a shared lock.
2180 *	- %LOCK_EX -- an exclusive lock.
2181 *	- %LOCK_UN -- remove an existing lock.
2182 *	- %LOCK_MAND -- a 'mandatory' flock.
2183 *	  This exists to emulate Windows Share Modes.
2184 *
2185 *	%LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
2186 *	processes read and write access respectively.
2187 */
2188SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
2189{
2190	struct fd f = fdget(fd);
2191	struct file_lock *lock;
2192	int can_sleep, unlock;
2193	int error;
2194
2195	error = -EBADF;
2196	if (!f.file)
2197		goto out;
 
 
 
 
 
 
 
 
 
2198
2199	can_sleep = !(cmd & LOCK_NB);
2200	cmd &= ~LOCK_NB;
2201	unlock = (cmd == LOCK_UN);
2202
2203	if (!unlock && !(cmd & LOCK_MAND) &&
2204	    !(f.file->f_mode & (FMODE_READ|FMODE_WRITE)))
2205		goto out_putf;
2206
2207	lock = flock_make_lock(f.file, cmd, NULL);
2208	if (IS_ERR(lock)) {
2209		error = PTR_ERR(lock);
2210		goto out_putf;
2211	}
2212
2213	if (can_sleep)
2214		lock->fl_flags |= FL_SLEEP;
 
 
2215
2216	error = security_file_lock(f.file, lock->fl_type);
2217	if (error)
2218		goto out_free;
2219
2220	if (f.file->f_op->flock)
2221		error = f.file->f_op->flock(f.file,
2222					  (can_sleep) ? F_SETLKW : F_SETLK,
2223					  lock);
2224	else
2225		error = locks_lock_file_wait(f.file, lock);
2226
2227 out_free:
2228	locks_free_lock(lock);
 
 
 
 
2229
2230 out_putf:
2231	fdput(f);
2232 out:
2233	return error;
2234}
2235
2236/**
2237 * vfs_test_lock - test file byte range lock
2238 * @filp: The file to test lock for
2239 * @fl: The lock to test; also used to hold result
2240 *
2241 * Returns -ERRNO on failure.  Indicates presence of conflicting lock by
2242 * setting conf->fl_type to something other than F_UNLCK.
2243 */
2244int vfs_test_lock(struct file *filp, struct file_lock *fl)
2245{
 
2246	if (filp->f_op->lock)
2247		return filp->f_op->lock(filp, F_GETLK, fl);
2248	posix_test_lock(filp, fl);
2249	return 0;
2250}
2251EXPORT_SYMBOL_GPL(vfs_test_lock);
2252
2253/**
2254 * locks_translate_pid - translate a file_lock's fl_pid number into a namespace
2255 * @fl: The file_lock who's fl_pid should be translated
2256 * @ns: The namespace into which the pid should be translated
2257 *
2258 * Used to tranlate a fl_pid into a namespace virtual pid number
2259 */
2260static pid_t locks_translate_pid(struct file_lock *fl, struct pid_namespace *ns)
2261{
2262	pid_t vnr;
2263	struct pid *pid;
2264
2265	if (IS_OFDLCK(fl))
2266		return -1;
2267	if (IS_REMOTELCK(fl))
2268		return fl->fl_pid;
 
 
 
2269	/*
2270	 * If the flock owner process is dead and its pid has been already
2271	 * freed, the translation below won't work, but we still want to show
2272	 * flock owner pid number in init pidns.
2273	 */
2274	if (ns == &init_pid_ns)
2275		return (pid_t)fl->fl_pid;
2276
2277	rcu_read_lock();
2278	pid = find_pid_ns(fl->fl_pid, &init_pid_ns);
2279	vnr = pid_nr_ns(pid, ns);
2280	rcu_read_unlock();
2281	return vnr;
2282}
2283
2284static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
2285{
2286	flock->l_pid = locks_translate_pid(fl, task_active_pid_ns(current));
2287#if BITS_PER_LONG == 32
2288	/*
2289	 * Make sure we can represent the posix lock via
2290	 * legacy 32bit flock.
2291	 */
2292	if (fl->fl_start > OFFT_OFFSET_MAX)
2293		return -EOVERFLOW;
2294	if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
2295		return -EOVERFLOW;
2296#endif
2297	flock->l_start = fl->fl_start;
2298	flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2299		fl->fl_end - fl->fl_start + 1;
2300	flock->l_whence = 0;
2301	flock->l_type = fl->fl_type;
2302	return 0;
2303}
2304
2305#if BITS_PER_LONG == 32
2306static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
2307{
2308	flock->l_pid = locks_translate_pid(fl, task_active_pid_ns(current));
2309	flock->l_start = fl->fl_start;
2310	flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2311		fl->fl_end - fl->fl_start + 1;
2312	flock->l_whence = 0;
2313	flock->l_type = fl->fl_type;
2314}
2315#endif
2316
2317/* Report the first existing lock that would conflict with l.
2318 * This implements the F_GETLK command of fcntl().
2319 */
2320int fcntl_getlk(struct file *filp, unsigned int cmd, struct flock *flock)
2321{
2322	struct file_lock *fl;
2323	int error;
2324
2325	fl = locks_alloc_lock();
2326	if (fl == NULL)
2327		return -ENOMEM;
2328	error = -EINVAL;
2329	if (flock->l_type != F_RDLCK && flock->l_type != F_WRLCK)
 
2330		goto out;
2331
2332	error = flock_to_posix_lock(filp, fl, flock);
2333	if (error)
2334		goto out;
2335
2336	if (cmd == F_OFD_GETLK) {
2337		error = -EINVAL;
2338		if (flock->l_pid != 0)
2339			goto out;
2340
2341		cmd = F_GETLK;
2342		fl->fl_flags |= FL_OFDLCK;
2343		fl->fl_owner = filp;
2344	}
2345
2346	error = vfs_test_lock(filp, fl);
2347	if (error)
2348		goto out;
2349
2350	flock->l_type = fl->fl_type;
2351	if (fl->fl_type != F_UNLCK) {
2352		error = posix_lock_to_flock(flock, fl);
2353		if (error)
2354			goto out;
2355	}
2356out:
2357	locks_free_lock(fl);
2358	return error;
2359}
2360
2361/**
2362 * vfs_lock_file - file byte range lock
2363 * @filp: The file to apply the lock to
2364 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
2365 * @fl: The lock to be applied
2366 * @conf: Place to return a copy of the conflicting lock, if found.
2367 *
2368 * A caller that doesn't care about the conflicting lock may pass NULL
2369 * as the final argument.
2370 *
2371 * If the filesystem defines a private ->lock() method, then @conf will
2372 * be left unchanged; so a caller that cares should initialize it to
2373 * some acceptable default.
2374 *
2375 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
2376 * locks, the ->lock() interface may return asynchronously, before the lock has
2377 * been granted or denied by the underlying filesystem, if (and only if)
2378 * lm_grant is set. Callers expecting ->lock() to return asynchronously
2379 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
2380 * the request is for a blocking lock. When ->lock() does return asynchronously,
2381 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
2382 * request completes.
 
 
2383 * If the request is for non-blocking lock the file system should return
2384 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
2385 * with the result. If the request timed out the callback routine will return a
2386 * nonzero return code and the file system should release the lock. The file
2387 * system is also responsible to keep a corresponding posix lock when it
2388 * grants a lock so the VFS can find out which locks are locally held and do
2389 * the correct lock cleanup when required.
2390 * The underlying filesystem must not drop the kernel lock or call
2391 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
2392 * return code.
2393 */
2394int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
2395{
 
2396	if (filp->f_op->lock)
2397		return filp->f_op->lock(filp, cmd, fl);
2398	else
2399		return posix_lock_file(filp, fl, conf);
2400}
2401EXPORT_SYMBOL_GPL(vfs_lock_file);
2402
2403static int do_lock_file_wait(struct file *filp, unsigned int cmd,
2404			     struct file_lock *fl)
2405{
2406	int error;
2407
2408	error = security_file_lock(filp, fl->fl_type);
2409	if (error)
2410		return error;
2411
2412	for (;;) {
2413		error = vfs_lock_file(filp, cmd, fl, NULL);
2414		if (error != FILE_LOCK_DEFERRED)
2415			break;
2416		error = wait_event_interruptible(fl->fl_wait, !fl->fl_blocker);
 
2417		if (error)
2418			break;
2419	}
2420	locks_delete_block(fl);
2421
2422	return error;
2423}
2424
2425/* Ensure that fl->fl_file has compatible f_mode for F_SETLK calls */
2426static int
2427check_fmode_for_setlk(struct file_lock *fl)
2428{
2429	switch (fl->fl_type) {
2430	case F_RDLCK:
2431		if (!(fl->fl_file->f_mode & FMODE_READ))
2432			return -EBADF;
2433		break;
2434	case F_WRLCK:
2435		if (!(fl->fl_file->f_mode & FMODE_WRITE))
2436			return -EBADF;
2437	}
2438	return 0;
2439}
2440
2441/* Apply the lock described by l to an open file descriptor.
2442 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2443 */
2444int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
2445		struct flock *flock)
2446{
2447	struct file_lock *file_lock = locks_alloc_lock();
2448	struct inode *inode = locks_inode(filp);
2449	struct file *f;
2450	int error;
2451
2452	if (file_lock == NULL)
2453		return -ENOLCK;
2454
2455	/* Don't allow mandatory locks on files that may be memory mapped
2456	 * and shared.
2457	 */
2458	if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
2459		error = -EAGAIN;
2460		goto out;
2461	}
2462
2463	error = flock_to_posix_lock(filp, file_lock, flock);
2464	if (error)
2465		goto out;
2466
2467	error = check_fmode_for_setlk(file_lock);
2468	if (error)
2469		goto out;
2470
2471	/*
2472	 * If the cmd is requesting file-private locks, then set the
2473	 * FL_OFDLCK flag and override the owner.
2474	 */
2475	switch (cmd) {
2476	case F_OFD_SETLK:
2477		error = -EINVAL;
2478		if (flock->l_pid != 0)
2479			goto out;
2480
2481		cmd = F_SETLK;
2482		file_lock->fl_flags |= FL_OFDLCK;
2483		file_lock->fl_owner = filp;
2484		break;
2485	case F_OFD_SETLKW:
2486		error = -EINVAL;
2487		if (flock->l_pid != 0)
2488			goto out;
2489
2490		cmd = F_SETLKW;
2491		file_lock->fl_flags |= FL_OFDLCK;
2492		file_lock->fl_owner = filp;
2493		/* Fallthrough */
2494	case F_SETLKW:
2495		file_lock->fl_flags |= FL_SLEEP;
2496	}
2497
2498	error = do_lock_file_wait(filp, cmd, file_lock);
2499
2500	/*
2501	 * Attempt to detect a close/fcntl race and recover by releasing the
2502	 * lock that was just acquired. There is no need to do that when we're
 
2503	 * unlocking though, or for OFD locks.
2504	 */
2505	if (!error && file_lock->fl_type != F_UNLCK &&
2506	    !(file_lock->fl_flags & FL_OFDLCK)) {
 
2507		/*
2508		 * We need that spin_lock here - it prevents reordering between
2509		 * update of i_flctx->flc_posix and check for it done in
2510		 * close(). rcu_read_lock() wouldn't do.
2511		 */
2512		spin_lock(&current->files->file_lock);
2513		f = fcheck(fd);
2514		spin_unlock(&current->files->file_lock);
2515		if (f != filp) {
2516			file_lock->fl_type = F_UNLCK;
2517			error = do_lock_file_wait(filp, cmd, file_lock);
2518			WARN_ON_ONCE(error);
2519			error = -EBADF;
2520		}
2521	}
2522out:
2523	trace_fcntl_setlk(inode, file_lock, error);
2524	locks_free_lock(file_lock);
2525	return error;
2526}
2527
2528#if BITS_PER_LONG == 32
2529/* Report the first existing lock that would conflict with l.
2530 * This implements the F_GETLK command of fcntl().
2531 */
2532int fcntl_getlk64(struct file *filp, unsigned int cmd, struct flock64 *flock)
2533{
2534	struct file_lock *fl;
2535	int error;
2536
2537	fl = locks_alloc_lock();
2538	if (fl == NULL)
2539		return -ENOMEM;
2540
2541	error = -EINVAL;
2542	if (flock->l_type != F_RDLCK && flock->l_type != F_WRLCK)
 
2543		goto out;
2544
2545	error = flock64_to_posix_lock(filp, fl, flock);
2546	if (error)
2547		goto out;
2548
2549	if (cmd == F_OFD_GETLK) {
2550		error = -EINVAL;
2551		if (flock->l_pid != 0)
2552			goto out;
2553
2554		cmd = F_GETLK64;
2555		fl->fl_flags |= FL_OFDLCK;
2556		fl->fl_owner = filp;
2557	}
2558
2559	error = vfs_test_lock(filp, fl);
2560	if (error)
2561		goto out;
2562
2563	flock->l_type = fl->fl_type;
2564	if (fl->fl_type != F_UNLCK)
2565		posix_lock_to_flock64(flock, fl);
2566
2567out:
2568	locks_free_lock(fl);
2569	return error;
2570}
2571
2572/* Apply the lock described by l to an open file descriptor.
2573 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2574 */
2575int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
2576		struct flock64 *flock)
2577{
2578	struct file_lock *file_lock = locks_alloc_lock();
2579	struct inode *inode = locks_inode(filp);
2580	struct file *f;
2581	int error;
2582
2583	if (file_lock == NULL)
2584		return -ENOLCK;
2585
2586	/* Don't allow mandatory locks on files that may be memory mapped
2587	 * and shared.
2588	 */
2589	if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
2590		error = -EAGAIN;
2591		goto out;
2592	}
2593
2594	error = flock64_to_posix_lock(filp, file_lock, flock);
2595	if (error)
2596		goto out;
2597
2598	error = check_fmode_for_setlk(file_lock);
2599	if (error)
2600		goto out;
2601
2602	/*
2603	 * If the cmd is requesting file-private locks, then set the
2604	 * FL_OFDLCK flag and override the owner.
2605	 */
2606	switch (cmd) {
2607	case F_OFD_SETLK:
2608		error = -EINVAL;
2609		if (flock->l_pid != 0)
2610			goto out;
2611
2612		cmd = F_SETLK64;
2613		file_lock->fl_flags |= FL_OFDLCK;
2614		file_lock->fl_owner = filp;
2615		break;
2616	case F_OFD_SETLKW:
2617		error = -EINVAL;
2618		if (flock->l_pid != 0)
2619			goto out;
2620
2621		cmd = F_SETLKW64;
2622		file_lock->fl_flags |= FL_OFDLCK;
2623		file_lock->fl_owner = filp;
2624		/* Fallthrough */
2625	case F_SETLKW64:
2626		file_lock->fl_flags |= FL_SLEEP;
2627	}
2628
2629	error = do_lock_file_wait(filp, cmd, file_lock);
2630
2631	/*
2632	 * Attempt to detect a close/fcntl race and recover by releasing the
2633	 * lock that was just acquired. There is no need to do that when we're
 
2634	 * unlocking though, or for OFD locks.
2635	 */
2636	if (!error && file_lock->fl_type != F_UNLCK &&
2637	    !(file_lock->fl_flags & FL_OFDLCK)) {
 
2638		/*
2639		 * We need that spin_lock here - it prevents reordering between
2640		 * update of i_flctx->flc_posix and check for it done in
2641		 * close(). rcu_read_lock() wouldn't do.
2642		 */
2643		spin_lock(&current->files->file_lock);
2644		f = fcheck(fd);
2645		spin_unlock(&current->files->file_lock);
2646		if (f != filp) {
2647			file_lock->fl_type = F_UNLCK;
2648			error = do_lock_file_wait(filp, cmd, file_lock);
2649			WARN_ON_ONCE(error);
2650			error = -EBADF;
2651		}
2652	}
2653out:
2654	locks_free_lock(file_lock);
2655	return error;
2656}
2657#endif /* BITS_PER_LONG == 32 */
2658
2659/*
2660 * This function is called when the file is being removed
2661 * from the task's fd array.  POSIX locks belonging to this task
2662 * are deleted at this time.
2663 */
2664void locks_remove_posix(struct file *filp, fl_owner_t owner)
2665{
2666	int error;
2667	struct inode *inode = locks_inode(filp);
2668	struct file_lock lock;
2669	struct file_lock_context *ctx;
2670
2671	/*
2672	 * If there are no locks held on this file, we don't need to call
2673	 * posix_lock_file().  Another process could be setting a lock on this
2674	 * file at the same time, but we wouldn't remove that lock anyway.
2675	 */
2676	ctx =  smp_load_acquire(&inode->i_flctx);
2677	if (!ctx || list_empty(&ctx->flc_posix))
2678		return;
2679
2680	locks_init_lock(&lock);
2681	lock.fl_type = F_UNLCK;
2682	lock.fl_flags = FL_POSIX | FL_CLOSE;
2683	lock.fl_start = 0;
2684	lock.fl_end = OFFSET_MAX;
2685	lock.fl_owner = owner;
2686	lock.fl_pid = current->tgid;
2687	lock.fl_file = filp;
2688	lock.fl_ops = NULL;
2689	lock.fl_lmops = NULL;
2690
2691	error = vfs_lock_file(filp, F_SETLK, &lock, NULL);
2692
2693	if (lock.fl_ops && lock.fl_ops->fl_release_private)
2694		lock.fl_ops->fl_release_private(&lock);
2695	trace_locks_remove_posix(inode, &lock, error);
2696}
2697EXPORT_SYMBOL(locks_remove_posix);
2698
2699/* The i_flctx must be valid when calling into here */
2700static void
2701locks_remove_flock(struct file *filp, struct file_lock_context *flctx)
2702{
2703	struct file_lock fl;
2704	struct inode *inode = locks_inode(filp);
2705
2706	if (list_empty(&flctx->flc_flock))
2707		return;
2708
2709	flock_make_lock(filp, LOCK_UN, &fl);
2710	fl.fl_flags |= FL_CLOSE;
2711
2712	if (filp->f_op->flock)
2713		filp->f_op->flock(filp, F_SETLKW, &fl);
2714	else
2715		flock_lock_inode(inode, &fl);
2716
2717	if (fl.fl_ops && fl.fl_ops->fl_release_private)
2718		fl.fl_ops->fl_release_private(&fl);
2719}
2720
2721/* The i_flctx must be valid when calling into here */
2722static void
2723locks_remove_lease(struct file *filp, struct file_lock_context *ctx)
2724{
2725	struct file_lock *fl, *tmp;
2726	LIST_HEAD(dispose);
2727
2728	if (list_empty(&ctx->flc_lease))
2729		return;
2730
2731	percpu_down_read(&file_rwsem);
2732	spin_lock(&ctx->flc_lock);
2733	list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list)
2734		if (filp == fl->fl_file)
2735			lease_modify(fl, F_UNLCK, &dispose);
2736	spin_unlock(&ctx->flc_lock);
2737	percpu_up_read(&file_rwsem);
2738
2739	locks_dispose_list(&dispose);
2740}
2741
2742/*
2743 * This function is called on the last close of an open file.
2744 */
2745void locks_remove_file(struct file *filp)
2746{
2747	struct file_lock_context *ctx;
2748
2749	ctx = smp_load_acquire(&locks_inode(filp)->i_flctx);
2750	if (!ctx)
2751		return;
2752
2753	/* remove any OFD locks */
2754	locks_remove_posix(filp, filp);
2755
2756	/* remove flock locks */
2757	locks_remove_flock(filp, ctx);
2758
2759	/* remove any leases */
2760	locks_remove_lease(filp, ctx);
2761
2762	spin_lock(&ctx->flc_lock);
2763	locks_check_ctx_file_list(filp, &ctx->flc_posix, "POSIX");
2764	locks_check_ctx_file_list(filp, &ctx->flc_flock, "FLOCK");
2765	locks_check_ctx_file_list(filp, &ctx->flc_lease, "LEASE");
2766	spin_unlock(&ctx->flc_lock);
2767}
2768
2769/**
2770 * vfs_cancel_lock - file byte range unblock lock
2771 * @filp: The file to apply the unblock to
2772 * @fl: The lock to be unblocked
2773 *
2774 * Used by lock managers to cancel blocked requests
2775 */
2776int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2777{
 
2778	if (filp->f_op->lock)
2779		return filp->f_op->lock(filp, F_CANCELLK, fl);
2780	return 0;
2781}
2782EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2783
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2784#ifdef CONFIG_PROC_FS
2785#include <linux/proc_fs.h>
2786#include <linux/seq_file.h>
2787
2788struct locks_iterator {
2789	int	li_cpu;
2790	loff_t	li_pos;
2791};
2792
2793static void lock_get_status(struct seq_file *f, struct file_lock *fl,
2794			    loff_t id, char *pfx)
2795{
2796	struct inode *inode = NULL;
2797	unsigned int fl_pid;
2798	struct pid_namespace *proc_pidns = file_inode(f->file)->i_sb->s_fs_info;
 
 
 
 
2799
2800	fl_pid = locks_translate_pid(fl, proc_pidns);
2801	/*
2802	 * If lock owner is dead (and pid is freed) or not visible in current
2803	 * pidns, zero is shown as a pid value. Check lock info from
2804	 * init_pid_ns to get saved lock pid value.
2805	 */
 
 
2806
2807	if (fl->fl_file != NULL)
2808		inode = locks_inode(fl->fl_file);
2809
2810	seq_printf(f, "%lld:%s ", id, pfx);
2811	if (IS_POSIX(fl)) {
2812		if (fl->fl_flags & FL_ACCESS)
 
 
2813			seq_puts(f, "ACCESS");
2814		else if (IS_OFDLCK(fl))
2815			seq_puts(f, "OFDLCK");
2816		else
2817			seq_puts(f, "POSIX ");
2818
2819		seq_printf(f, " %s ",
2820			     (inode == NULL) ? "*NOINODE*" :
2821			     mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
2822	} else if (IS_FLOCK(fl)) {
2823		if (fl->fl_type & LOCK_MAND) {
2824			seq_puts(f, "FLOCK  MSNFS     ");
2825		} else {
2826			seq_puts(f, "FLOCK  ADVISORY  ");
2827		}
2828	} else if (IS_LEASE(fl)) {
2829		if (fl->fl_flags & FL_DELEG)
2830			seq_puts(f, "DELEG  ");
2831		else
2832			seq_puts(f, "LEASE  ");
2833
2834		if (lease_breaking(fl))
2835			seq_puts(f, "BREAKING  ");
2836		else if (fl->fl_file)
2837			seq_puts(f, "ACTIVE    ");
2838		else
2839			seq_puts(f, "BREAKER   ");
2840	} else {
2841		seq_puts(f, "UNKNOWN UNKNOWN  ");
2842	}
2843	if (fl->fl_type & LOCK_MAND) {
2844		seq_printf(f, "%s ",
2845			       (fl->fl_type & LOCK_READ)
2846			       ? (fl->fl_type & LOCK_WRITE) ? "RW   " : "READ "
2847			       : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2848	} else {
2849		int type = IS_LEASE(fl) ? target_leasetype(fl) : fl->fl_type;
2850
2851		seq_printf(f, "%s ", (type == F_WRLCK) ? "WRITE" :
2852				     (type == F_RDLCK) ? "READ" : "UNLCK");
2853	}
2854	if (inode) {
2855		/* userspace relies on this representation of dev_t */
2856		seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
2857				MAJOR(inode->i_sb->s_dev),
2858				MINOR(inode->i_sb->s_dev), inode->i_ino);
2859	} else {
2860		seq_printf(f, "%d <none>:0 ", fl_pid);
2861	}
2862	if (IS_POSIX(fl)) {
2863		if (fl->fl_end == OFFSET_MAX)
2864			seq_printf(f, "%Ld EOF\n", fl->fl_start);
2865		else
2866			seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
2867	} else {
2868		seq_puts(f, "0 EOF\n");
2869	}
2870}
2871
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2872static int locks_show(struct seq_file *f, void *v)
2873{
2874	struct locks_iterator *iter = f->private;
2875	struct file_lock *fl, *bfl;
2876	struct pid_namespace *proc_pidns = file_inode(f->file)->i_sb->s_fs_info;
 
2877
2878	fl = hlist_entry(v, struct file_lock, fl_link);
2879
2880	if (locks_translate_pid(fl, proc_pidns) == 0)
2881		return 0;
2882
2883	lock_get_status(f, fl, iter->li_pos, "");
 
 
 
 
 
 
 
 
 
2884
2885	list_for_each_entry(bfl, &fl->fl_blocked_requests, fl_blocked_member)
2886		lock_get_status(f, bfl, iter->li_pos, " ->");
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2887
2888	return 0;
2889}
2890
2891static void __show_fd_locks(struct seq_file *f,
2892			struct list_head *head, int *id,
2893			struct file *filp, struct files_struct *files)
2894{
2895	struct file_lock *fl;
2896
2897	list_for_each_entry(fl, head, fl_list) {
2898
2899		if (filp != fl->fl_file)
2900			continue;
2901		if (fl->fl_owner != files &&
2902		    fl->fl_owner != filp)
2903			continue;
2904
2905		(*id)++;
2906		seq_puts(f, "lock:\t");
2907		lock_get_status(f, fl, *id, "");
2908	}
2909}
2910
2911void show_fd_locks(struct seq_file *f,
2912		  struct file *filp, struct files_struct *files)
2913{
2914	struct inode *inode = locks_inode(filp);
2915	struct file_lock_context *ctx;
2916	int id = 0;
2917
2918	ctx = smp_load_acquire(&inode->i_flctx);
2919	if (!ctx)
2920		return;
2921
2922	spin_lock(&ctx->flc_lock);
2923	__show_fd_locks(f, &ctx->flc_flock, &id, filp, files);
2924	__show_fd_locks(f, &ctx->flc_posix, &id, filp, files);
2925	__show_fd_locks(f, &ctx->flc_lease, &id, filp, files);
2926	spin_unlock(&ctx->flc_lock);
2927}
2928
2929static void *locks_start(struct seq_file *f, loff_t *pos)
2930	__acquires(&blocked_lock_lock)
2931{
2932	struct locks_iterator *iter = f->private;
2933
2934	iter->li_pos = *pos + 1;
2935	percpu_down_write(&file_rwsem);
2936	spin_lock(&blocked_lock_lock);
2937	return seq_hlist_start_percpu(&file_lock_list.hlist, &iter->li_cpu, *pos);
2938}
2939
2940static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2941{
2942	struct locks_iterator *iter = f->private;
2943
2944	++iter->li_pos;
2945	return seq_hlist_next_percpu(v, &file_lock_list.hlist, &iter->li_cpu, pos);
2946}
2947
2948static void locks_stop(struct seq_file *f, void *v)
2949	__releases(&blocked_lock_lock)
2950{
2951	spin_unlock(&blocked_lock_lock);
2952	percpu_up_write(&file_rwsem);
2953}
2954
2955static const struct seq_operations locks_seq_operations = {
2956	.start	= locks_start,
2957	.next	= locks_next,
2958	.stop	= locks_stop,
2959	.show	= locks_show,
2960};
2961
2962static int __init proc_locks_init(void)
2963{
2964	proc_create_seq_private("locks", 0, NULL, &locks_seq_operations,
2965			sizeof(struct locks_iterator), NULL);
2966	return 0;
2967}
2968fs_initcall(proc_locks_init);
2969#endif
2970
2971static int __init filelock_init(void)
2972{
2973	int i;
2974
2975	flctx_cache = kmem_cache_create("file_lock_ctx",
2976			sizeof(struct file_lock_context), 0, SLAB_PANIC, NULL);
2977
2978	filelock_cache = kmem_cache_create("file_lock_cache",
2979			sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
 
 
 
2980
2981	for_each_possible_cpu(i) {
2982		struct file_lock_list_struct *fll = per_cpu_ptr(&file_lock_list, i);
2983
2984		spin_lock_init(&fll->lock);
2985		INIT_HLIST_HEAD(&fll->hlist);
2986	}
2987
2988	lease_notifier_chain_init();
2989	return 0;
2990}
2991core_initcall(filelock_init);