Linux Audio

Check our new training course

Loading...
v4.17
   1/*
   2 *  linux/fs/locks.c
   3 *
   4 *  Provide support for fcntl()'s F_GETLK, F_SETLK, and F_SETLKW calls.
   5 *  Doug Evans (dje@spiff.uucp), August 07, 1992
   6 *
   7 *  Deadlock detection added.
   8 *  FIXME: one thing isn't handled yet:
   9 *	- mandatory locks (requires lots of changes elsewhere)
  10 *  Kelly Carmichael (kelly@[142.24.8.65]), September 17, 1994.
  11 *
  12 *  Miscellaneous edits, and a total rewrite of posix_lock_file() code.
  13 *  Kai Petzke (wpp@marie.physik.tu-berlin.de), 1994
  14 *  
  15 *  Converted file_lock_table to a linked list from an array, which eliminates
  16 *  the limits on how many active file locks are open.
  17 *  Chad Page (pageone@netcom.com), November 27, 1994
  18 * 
  19 *  Removed dependency on file descriptors. dup()'ed file descriptors now
  20 *  get the same locks as the original file descriptors, and a close() on
  21 *  any file descriptor removes ALL the locks on the file for the current
  22 *  process. Since locks still depend on the process id, locks are inherited
  23 *  after an exec() but not after a fork(). This agrees with POSIX, and both
  24 *  BSD and SVR4 practice.
  25 *  Andy Walker (andy@lysaker.kvaerner.no), February 14, 1995
  26 *
  27 *  Scrapped free list which is redundant now that we allocate locks
  28 *  dynamically with kmalloc()/kfree().
  29 *  Andy Walker (andy@lysaker.kvaerner.no), February 21, 1995
  30 *
  31 *  Implemented two lock personalities - FL_FLOCK and FL_POSIX.
  32 *
  33 *  FL_POSIX locks are created with calls to fcntl() and lockf() through the
  34 *  fcntl() system call. They have the semantics described above.
  35 *
  36 *  FL_FLOCK locks are created with calls to flock(), through the flock()
  37 *  system call, which is new. Old C libraries implement flock() via fcntl()
  38 *  and will continue to use the old, broken implementation.
  39 *
  40 *  FL_FLOCK locks follow the 4.4 BSD flock() semantics. They are associated
  41 *  with a file pointer (filp). As a result they can be shared by a parent
  42 *  process and its children after a fork(). They are removed when the last
  43 *  file descriptor referring to the file pointer is closed (unless explicitly
  44 *  unlocked). 
  45 *
  46 *  FL_FLOCK locks never deadlock, an existing lock is always removed before
  47 *  upgrading from shared to exclusive (or vice versa). When this happens
  48 *  any processes blocked by the current lock are woken up and allowed to
  49 *  run before the new lock is applied.
  50 *  Andy Walker (andy@lysaker.kvaerner.no), June 09, 1995
  51 *
  52 *  Removed some race conditions in flock_lock_file(), marked other possible
  53 *  races. Just grep for FIXME to see them. 
  54 *  Dmitry Gorodchanin (pgmdsg@ibi.com), February 09, 1996.
  55 *
  56 *  Addressed Dmitry's concerns. Deadlock checking no longer recursive.
  57 *  Lock allocation changed to GFP_ATOMIC as we can't afford to sleep
  58 *  once we've checked for blocking and deadlocking.
  59 *  Andy Walker (andy@lysaker.kvaerner.no), April 03, 1996.
  60 *
  61 *  Initial implementation of mandatory locks. SunOS turned out to be
  62 *  a rotten model, so I implemented the "obvious" semantics.
  63 *  See 'Documentation/filesystems/mandatory-locking.txt' for details.
  64 *  Andy Walker (andy@lysaker.kvaerner.no), April 06, 1996.
  65 *
  66 *  Don't allow mandatory locks on mmap()'ed files. Added simple functions to
  67 *  check if a file has mandatory locks, used by mmap(), open() and creat() to
  68 *  see if system call should be rejected. Ref. HP-UX/SunOS/Solaris Reference
  69 *  Manual, Section 2.
  70 *  Andy Walker (andy@lysaker.kvaerner.no), April 09, 1996.
  71 *
  72 *  Tidied up block list handling. Added '/proc/locks' interface.
  73 *  Andy Walker (andy@lysaker.kvaerner.no), April 24, 1996.
  74 *
  75 *  Fixed deadlock condition for pathological code that mixes calls to
  76 *  flock() and fcntl().
  77 *  Andy Walker (andy@lysaker.kvaerner.no), April 29, 1996.
  78 *
  79 *  Allow only one type of locking scheme (FL_POSIX or FL_FLOCK) to be in use
  80 *  for a given file at a time. Changed the CONFIG_LOCK_MANDATORY scheme to
  81 *  guarantee sensible behaviour in the case where file system modules might
  82 *  be compiled with different options than the kernel itself.
  83 *  Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
  84 *
  85 *  Added a couple of missing wake_up() calls. Thanks to Thomas Meckel
  86 *  (Thomas.Meckel@mni.fh-giessen.de) for spotting this.
  87 *  Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
  88 *
  89 *  Changed FL_POSIX locks to use the block list in the same way as FL_FLOCK
  90 *  locks. Changed process synchronisation to avoid dereferencing locks that
  91 *  have already been freed.
  92 *  Andy Walker (andy@lysaker.kvaerner.no), Sep 21, 1996.
  93 *
  94 *  Made the block list a circular list to minimise searching in the list.
  95 *  Andy Walker (andy@lysaker.kvaerner.no), Sep 25, 1996.
  96 *
  97 *  Made mandatory locking a mount option. Default is not to allow mandatory
  98 *  locking.
  99 *  Andy Walker (andy@lysaker.kvaerner.no), Oct 04, 1996.
 100 *
 101 *  Some adaptations for NFS support.
 102 *  Olaf Kirch (okir@monad.swb.de), Dec 1996,
 103 *
 104 *  Fixed /proc/locks interface so that we can't overrun the buffer we are handed.
 105 *  Andy Walker (andy@lysaker.kvaerner.no), May 12, 1997.
 106 *
 107 *  Use slab allocator instead of kmalloc/kfree.
 108 *  Use generic list implementation from <linux/list.h>.
 109 *  Sped up posix_locks_deadlock by only considering blocked locks.
 110 *  Matthew Wilcox <willy@debian.org>, March, 2000.
 111 *
 112 *  Leases and LOCK_MAND
 113 *  Matthew Wilcox <willy@debian.org>, June, 2000.
 114 *  Stephen Rothwell <sfr@canb.auug.org.au>, June, 2000.
 115 */
 116
 117#include <linux/capability.h>
 118#include <linux/file.h>
 119#include <linux/fdtable.h>
 120#include <linux/fs.h>
 121#include <linux/init.h>
 
 122#include <linux/security.h>
 123#include <linux/slab.h>
 124#include <linux/syscalls.h>
 125#include <linux/time.h>
 126#include <linux/rcupdate.h>
 127#include <linux/pid_namespace.h>
 128#include <linux/hashtable.h>
 129#include <linux/percpu.h>
 130
 131#define CREATE_TRACE_POINTS
 132#include <trace/events/filelock.h>
 133
 134#include <linux/uaccess.h>
 135
 136#define IS_POSIX(fl)	(fl->fl_flags & FL_POSIX)
 137#define IS_FLOCK(fl)	(fl->fl_flags & FL_FLOCK)
 138#define IS_LEASE(fl)	(fl->fl_flags & (FL_LEASE|FL_DELEG|FL_LAYOUT))
 139#define IS_OFDLCK(fl)	(fl->fl_flags & FL_OFDLCK)
 140#define IS_REMOTELCK(fl)	(fl->fl_pid <= 0)
 141
 142static inline bool is_remote_lock(struct file *filp)
 143{
 144	return likely(!(filp->f_path.dentry->d_sb->s_flags & SB_NOREMOTELOCK));
 145}
 146
 147static bool lease_breaking(struct file_lock *fl)
 148{
 149	return fl->fl_flags & (FL_UNLOCK_PENDING | FL_DOWNGRADE_PENDING);
 150}
 151
 152static int target_leasetype(struct file_lock *fl)
 153{
 154	if (fl->fl_flags & FL_UNLOCK_PENDING)
 155		return F_UNLCK;
 156	if (fl->fl_flags & FL_DOWNGRADE_PENDING)
 157		return F_RDLCK;
 158	return fl->fl_type;
 159}
 160
 161int leases_enable = 1;
 162int lease_break_time = 45;
 163
 164/*
 165 * The global file_lock_list is only used for displaying /proc/locks, so we
 166 * keep a list on each CPU, with each list protected by its own spinlock.
 167 * Global serialization is done using file_rwsem.
 168 *
 169 * Note that alterations to the list also require that the relevant flc_lock is
 170 * held.
 171 */
 172struct file_lock_list_struct {
 173	spinlock_t		lock;
 174	struct hlist_head	hlist;
 175};
 176static DEFINE_PER_CPU(struct file_lock_list_struct, file_lock_list);
 177DEFINE_STATIC_PERCPU_RWSEM(file_rwsem);
 178
 179/*
 180 * The blocked_hash is used to find POSIX lock loops for deadlock detection.
 181 * It is protected by blocked_lock_lock.
 182 *
 183 * We hash locks by lockowner in order to optimize searching for the lock a
 184 * particular lockowner is waiting on.
 185 *
 186 * FIXME: make this value scale via some heuristic? We generally will want more
 187 * buckets when we have more lockowners holding locks, but that's a little
 188 * difficult to determine without knowing what the workload will look like.
 189 */
 190#define BLOCKED_HASH_BITS	7
 191static DEFINE_HASHTABLE(blocked_hash, BLOCKED_HASH_BITS);
 192
 193/*
 194 * This lock protects the blocked_hash. Generally, if you're accessing it, you
 195 * want to be holding this lock.
 196 *
 197 * In addition, it also protects the fl->fl_block list, and the fl->fl_next
 198 * pointer for file_lock structures that are acting as lock requests (in
 199 * contrast to those that are acting as records of acquired locks).
 200 *
 201 * Note that when we acquire this lock in order to change the above fields,
 202 * we often hold the flc_lock as well. In certain cases, when reading the fields
 203 * protected by this lock, we can skip acquiring it iff we already hold the
 204 * flc_lock.
 205 *
 206 * In particular, adding an entry to the fl_block list requires that you hold
 207 * both the flc_lock and the blocked_lock_lock (acquired in that order).
 208 * Deleting an entry from the list however only requires the file_lock_lock.
 209 */
 210static DEFINE_SPINLOCK(blocked_lock_lock);
 211
 212static struct kmem_cache *flctx_cache __read_mostly;
 213static struct kmem_cache *filelock_cache __read_mostly;
 214
 215static struct file_lock_context *
 216locks_get_lock_context(struct inode *inode, int type)
 217{
 218	struct file_lock_context *ctx;
 219
 220	/* paired with cmpxchg() below */
 221	ctx = smp_load_acquire(&inode->i_flctx);
 222	if (likely(ctx) || type == F_UNLCK)
 223		goto out;
 224
 225	ctx = kmem_cache_alloc(flctx_cache, GFP_KERNEL);
 226	if (!ctx)
 227		goto out;
 228
 229	spin_lock_init(&ctx->flc_lock);
 230	INIT_LIST_HEAD(&ctx->flc_flock);
 231	INIT_LIST_HEAD(&ctx->flc_posix);
 232	INIT_LIST_HEAD(&ctx->flc_lease);
 233
 234	/*
 235	 * Assign the pointer if it's not already assigned. If it is, then
 236	 * free the context we just allocated.
 237	 */
 238	if (cmpxchg(&inode->i_flctx, NULL, ctx)) {
 239		kmem_cache_free(flctx_cache, ctx);
 240		ctx = smp_load_acquire(&inode->i_flctx);
 241	}
 242out:
 243	trace_locks_get_lock_context(inode, type, ctx);
 244	return ctx;
 245}
 246
 247static void
 248locks_dump_ctx_list(struct list_head *list, char *list_type)
 249{
 250	struct file_lock *fl;
 251
 252	list_for_each_entry(fl, list, fl_list) {
 253		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);
 254	}
 255}
 256
 257static void
 258locks_check_ctx_lists(struct inode *inode)
 259{
 260	struct file_lock_context *ctx = inode->i_flctx;
 261
 262	if (unlikely(!list_empty(&ctx->flc_flock) ||
 263		     !list_empty(&ctx->flc_posix) ||
 264		     !list_empty(&ctx->flc_lease))) {
 265		pr_warn("Leaked locks on dev=0x%x:0x%x ino=0x%lx:\n",
 266			MAJOR(inode->i_sb->s_dev), MINOR(inode->i_sb->s_dev),
 267			inode->i_ino);
 268		locks_dump_ctx_list(&ctx->flc_flock, "FLOCK");
 269		locks_dump_ctx_list(&ctx->flc_posix, "POSIX");
 270		locks_dump_ctx_list(&ctx->flc_lease, "LEASE");
 271	}
 272}
 
 273
 274static void
 275locks_check_ctx_file_list(struct file *filp, struct list_head *list,
 276				char *list_type)
 277{
 278	struct file_lock *fl;
 279	struct inode *inode = locks_inode(filp);
 280
 281	list_for_each_entry(fl, list, fl_list)
 282		if (fl->fl_file == filp)
 283			pr_warn("Leaked %s lock on dev=0x%x:0x%x ino=0x%lx "
 284				" fl_owner=%p fl_flags=0x%x fl_type=0x%x fl_pid=%u\n",
 285				list_type, MAJOR(inode->i_sb->s_dev),
 286				MINOR(inode->i_sb->s_dev), inode->i_ino,
 287				fl->fl_owner, fl->fl_flags, fl->fl_type, fl->fl_pid);
 288}
 
 289
 290void
 291locks_free_lock_context(struct inode *inode)
 292{
 293	struct file_lock_context *ctx = inode->i_flctx;
 294
 295	if (unlikely(ctx)) {
 296		locks_check_ctx_lists(inode);
 297		kmem_cache_free(flctx_cache, ctx);
 298	}
 299}
 300
 301static void locks_init_lock_heads(struct file_lock *fl)
 302{
 303	INIT_HLIST_NODE(&fl->fl_link);
 304	INIT_LIST_HEAD(&fl->fl_list);
 305	INIT_LIST_HEAD(&fl->fl_block);
 306	init_waitqueue_head(&fl->fl_wait);
 307}
 308
 309/* Allocate an empty lock structure. */
 310struct file_lock *locks_alloc_lock(void)
 311{
 312	struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
 313
 314	if (fl)
 315		locks_init_lock_heads(fl);
 316
 317	return fl;
 318}
 319EXPORT_SYMBOL_GPL(locks_alloc_lock);
 320
 321void locks_release_private(struct file_lock *fl)
 322{
 323	if (fl->fl_ops) {
 324		if (fl->fl_ops->fl_release_private)
 325			fl->fl_ops->fl_release_private(fl);
 326		fl->fl_ops = NULL;
 327	}
 328
 329	if (fl->fl_lmops) {
 330		if (fl->fl_lmops->lm_put_owner) {
 331			fl->fl_lmops->lm_put_owner(fl->fl_owner);
 332			fl->fl_owner = NULL;
 333		}
 334		fl->fl_lmops = NULL;
 335	}
 
 336}
 337EXPORT_SYMBOL_GPL(locks_release_private);
 338
 339/* Free a lock which is not in use. */
 340void locks_free_lock(struct file_lock *fl)
 341{
 342	BUG_ON(waitqueue_active(&fl->fl_wait));
 343	BUG_ON(!list_empty(&fl->fl_list));
 344	BUG_ON(!list_empty(&fl->fl_block));
 345	BUG_ON(!hlist_unhashed(&fl->fl_link));
 346
 347	locks_release_private(fl);
 348	kmem_cache_free(filelock_cache, fl);
 349}
 350EXPORT_SYMBOL(locks_free_lock);
 351
 352static void
 353locks_dispose_list(struct list_head *dispose)
 354{
 355	struct file_lock *fl;
 356
 357	while (!list_empty(dispose)) {
 358		fl = list_first_entry(dispose, struct file_lock, fl_list);
 359		list_del_init(&fl->fl_list);
 360		locks_free_lock(fl);
 361	}
 362}
 363
 364void locks_init_lock(struct file_lock *fl)
 365{
 366	memset(fl, 0, sizeof(struct file_lock));
 367	locks_init_lock_heads(fl);
 368}
 369
 370EXPORT_SYMBOL(locks_init_lock);
 371
 
 
 
 
 
 
 
 
 
 
 
 372/*
 373 * Initialize a new lock from an existing file_lock structure.
 374 */
 375void locks_copy_conflock(struct file_lock *new, struct file_lock *fl)
 376{
 377	new->fl_owner = fl->fl_owner;
 378	new->fl_pid = fl->fl_pid;
 379	new->fl_file = NULL;
 380	new->fl_flags = fl->fl_flags;
 381	new->fl_type = fl->fl_type;
 382	new->fl_start = fl->fl_start;
 383	new->fl_end = fl->fl_end;
 384	new->fl_lmops = fl->fl_lmops;
 385	new->fl_ops = NULL;
 386
 387	if (fl->fl_lmops) {
 388		if (fl->fl_lmops->lm_get_owner)
 389			fl->fl_lmops->lm_get_owner(fl->fl_owner);
 390	}
 391}
 392EXPORT_SYMBOL(locks_copy_conflock);
 393
 394void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
 395{
 396	/* "new" must be a freshly-initialized lock */
 397	WARN_ON_ONCE(new->fl_ops);
 398
 399	locks_copy_conflock(new, fl);
 400
 
 401	new->fl_file = fl->fl_file;
 402	new->fl_ops = fl->fl_ops;
 
 403
 404	if (fl->fl_ops) {
 405		if (fl->fl_ops->fl_copy_lock)
 406			fl->fl_ops->fl_copy_lock(new, fl);
 407	}
 408}
 409
 410EXPORT_SYMBOL(locks_copy_lock);
 411
 412static inline int flock_translate_cmd(int cmd) {
 413	if (cmd & LOCK_MAND)
 414		return cmd & (LOCK_MAND | LOCK_RW);
 415	switch (cmd) {
 416	case LOCK_SH:
 417		return F_RDLCK;
 418	case LOCK_EX:
 419		return F_WRLCK;
 420	case LOCK_UN:
 421		return F_UNLCK;
 422	}
 423	return -EINVAL;
 424}
 425
 426/* Fill in a file_lock structure with an appropriate FLOCK lock. */
 427static struct file_lock *
 428flock_make_lock(struct file *filp, unsigned int cmd)
 429{
 430	struct file_lock *fl;
 431	int type = flock_translate_cmd(cmd);
 432
 433	if (type < 0)
 434		return ERR_PTR(type);
 435	
 436	fl = locks_alloc_lock();
 437	if (fl == NULL)
 438		return ERR_PTR(-ENOMEM);
 439
 440	fl->fl_file = filp;
 441	fl->fl_owner = filp;
 442	fl->fl_pid = current->tgid;
 443	fl->fl_flags = FL_FLOCK;
 444	fl->fl_type = type;
 445	fl->fl_end = OFFSET_MAX;
 446	
 447	return fl;
 
 448}
 449
 450static int assign_type(struct file_lock *fl, long type)
 451{
 452	switch (type) {
 453	case F_RDLCK:
 454	case F_WRLCK:
 455	case F_UNLCK:
 456		fl->fl_type = type;
 457		break;
 458	default:
 459		return -EINVAL;
 460	}
 461	return 0;
 462}
 463
 464static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
 465				 struct flock64 *l)
 
 
 
 466{
 
 
 467	switch (l->l_whence) {
 468	case SEEK_SET:
 469		fl->fl_start = 0;
 470		break;
 471	case SEEK_CUR:
 472		fl->fl_start = filp->f_pos;
 473		break;
 474	case SEEK_END:
 475		fl->fl_start = i_size_read(file_inode(filp));
 476		break;
 477	default:
 478		return -EINVAL;
 479	}
 480	if (l->l_start > OFFSET_MAX - fl->fl_start)
 481		return -EOVERFLOW;
 482	fl->fl_start += l->l_start;
 483	if (fl->fl_start < 0)
 484		return -EINVAL;
 485
 486	/* POSIX-1996 leaves the case l->l_len < 0 undefined;
 487	   POSIX-2001 defines it. */
 
 
 
 
 488	if (l->l_len > 0) {
 489		if (l->l_len - 1 > OFFSET_MAX - fl->fl_start)
 490			return -EOVERFLOW;
 491		fl->fl_end = fl->fl_start + l->l_len - 1;
 492
 493	} else if (l->l_len < 0) {
 494		if (fl->fl_start + l->l_len < 0)
 
 
 
 495			return -EINVAL;
 496		fl->fl_end = fl->fl_start - 1;
 497		fl->fl_start += l->l_len;
 498	} else
 499		fl->fl_end = OFFSET_MAX;
 500
 501	fl->fl_owner = current->files;
 502	fl->fl_pid = current->tgid;
 503	fl->fl_file = filp;
 504	fl->fl_flags = FL_POSIX;
 505	fl->fl_ops = NULL;
 506	fl->fl_lmops = NULL;
 507
 508	return assign_type(fl, l->l_type);
 509}
 510
 511/* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
 512 * style lock.
 513 */
 514static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
 515			       struct flock *l)
 516{
 517	struct flock64 ll = {
 518		.l_type = l->l_type,
 519		.l_whence = l->l_whence,
 520		.l_start = l->l_start,
 521		.l_len = l->l_len,
 522	};
 
 
 
 
 
 
 
 
 
 523
 524	return flock64_to_posix_lock(filp, fl, &ll);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 525}
 
 526
 527/* default lease lock manager operations */
 528static bool
 529lease_break_callback(struct file_lock *fl)
 530{
 531	kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
 532	return false;
 533}
 534
 535static void
 536lease_setup(struct file_lock *fl, void **priv)
 537{
 538	struct file *filp = fl->fl_file;
 539	struct fasync_struct *fa = *priv;
 540
 541	/*
 542	 * fasync_insert_entry() returns the old entry if any. If there was no
 543	 * old entry, then it used "priv" and inserted it into the fasync list.
 544	 * Clear the pointer to indicate that it shouldn't be freed.
 545	 */
 546	if (!fasync_insert_entry(fa->fa_fd, filp, &fl->fl_fasync, fa))
 547		*priv = NULL;
 548
 549	__f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
 
 550}
 551
 552static const struct lock_manager_operations lease_manager_ops = {
 553	.lm_break = lease_break_callback,
 
 554	.lm_change = lease_modify,
 555	.lm_setup = lease_setup,
 556};
 557
 558/*
 559 * Initialize a lease, use the default lock manager operations
 560 */
 561static int lease_init(struct file *filp, long type, struct file_lock *fl)
 562{
 563	if (assign_type(fl, type) != 0)
 564		return -EINVAL;
 565
 566	fl->fl_owner = filp;
 567	fl->fl_pid = current->tgid;
 568
 569	fl->fl_file = filp;
 570	fl->fl_flags = FL_LEASE;
 571	fl->fl_start = 0;
 572	fl->fl_end = OFFSET_MAX;
 573	fl->fl_ops = NULL;
 574	fl->fl_lmops = &lease_manager_ops;
 575	return 0;
 576}
 577
 578/* Allocate a file_lock initialised to this type of lease */
 579static struct file_lock *lease_alloc(struct file *filp, long type)
 580{
 581	struct file_lock *fl = locks_alloc_lock();
 582	int error = -ENOMEM;
 583
 584	if (fl == NULL)
 585		return ERR_PTR(error);
 586
 587	error = lease_init(filp, type, fl);
 588	if (error) {
 589		locks_free_lock(fl);
 590		return ERR_PTR(error);
 591	}
 592	return fl;
 593}
 594
 595/* Check if two locks overlap each other.
 596 */
 597static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
 598{
 599	return ((fl1->fl_end >= fl2->fl_start) &&
 600		(fl2->fl_end >= fl1->fl_start));
 601}
 602
 603/*
 604 * Check whether two locks have the same owner.
 605 */
 606static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
 607{
 608	if (fl1->fl_lmops && fl1->fl_lmops->lm_compare_owner)
 609		return fl2->fl_lmops == fl1->fl_lmops &&
 610			fl1->fl_lmops->lm_compare_owner(fl1, fl2);
 611	return fl1->fl_owner == fl2->fl_owner;
 612}
 613
 614/* Must be called with the flc_lock held! */
 615static void locks_insert_global_locks(struct file_lock *fl)
 616{
 617	struct file_lock_list_struct *fll = this_cpu_ptr(&file_lock_list);
 618
 619	percpu_rwsem_assert_held(&file_rwsem);
 620
 621	spin_lock(&fll->lock);
 622	fl->fl_link_cpu = smp_processor_id();
 623	hlist_add_head(&fl->fl_link, &fll->hlist);
 624	spin_unlock(&fll->lock);
 625}
 626
 627/* Must be called with the flc_lock held! */
 628static void locks_delete_global_locks(struct file_lock *fl)
 629{
 630	struct file_lock_list_struct *fll;
 631
 632	percpu_rwsem_assert_held(&file_rwsem);
 633
 634	/*
 635	 * Avoid taking lock if already unhashed. This is safe since this check
 636	 * is done while holding the flc_lock, and new insertions into the list
 637	 * also require that it be held.
 638	 */
 639	if (hlist_unhashed(&fl->fl_link))
 640		return;
 641
 642	fll = per_cpu_ptr(&file_lock_list, fl->fl_link_cpu);
 643	spin_lock(&fll->lock);
 644	hlist_del_init(&fl->fl_link);
 645	spin_unlock(&fll->lock);
 646}
 647
 648static unsigned long
 649posix_owner_key(struct file_lock *fl)
 650{
 651	if (fl->fl_lmops && fl->fl_lmops->lm_owner_key)
 652		return fl->fl_lmops->lm_owner_key(fl);
 653	return (unsigned long)fl->fl_owner;
 654}
 655
 656static void locks_insert_global_blocked(struct file_lock *waiter)
 657{
 658	lockdep_assert_held(&blocked_lock_lock);
 659
 660	hash_add(blocked_hash, &waiter->fl_link, posix_owner_key(waiter));
 661}
 662
 663static void locks_delete_global_blocked(struct file_lock *waiter)
 664{
 665	lockdep_assert_held(&blocked_lock_lock);
 666
 667	hash_del(&waiter->fl_link);
 668}
 669
 670/* Remove waiter from blocker's block list.
 671 * When blocker ends up pointing to itself then the list is empty.
 672 *
 673 * Must be called with blocked_lock_lock held.
 674 */
 675static void __locks_delete_block(struct file_lock *waiter)
 676{
 677	locks_delete_global_blocked(waiter);
 678	list_del_init(&waiter->fl_block);
 
 679	waiter->fl_next = NULL;
 680}
 681
 
 
 682static void locks_delete_block(struct file_lock *waiter)
 683{
 684	spin_lock(&blocked_lock_lock);
 685	__locks_delete_block(waiter);
 686	spin_unlock(&blocked_lock_lock);
 687}
 688
 689/* Insert waiter into blocker's block list.
 690 * We use a circular list so that processes can be easily woken up in
 691 * the order they blocked. The documentation doesn't require this but
 692 * it seems like the reasonable thing to do.
 693 *
 694 * Must be called with both the flc_lock and blocked_lock_lock held. The
 695 * fl_block list itself is protected by the blocked_lock_lock, but by ensuring
 696 * that the flc_lock is also held on insertions we can avoid taking the
 697 * blocked_lock_lock in some cases when we see that the fl_block list is empty.
 698 */
 699static void __locks_insert_block(struct file_lock *blocker,
 700					struct file_lock *waiter)
 701{
 702	BUG_ON(!list_empty(&waiter->fl_block));
 703	waiter->fl_next = blocker;
 704	list_add_tail(&waiter->fl_block, &blocker->fl_block);
 705	if (IS_POSIX(blocker) && !IS_OFDLCK(blocker))
 706		locks_insert_global_blocked(waiter);
 707}
 708
 709/* Must be called with flc_lock held. */
 710static void locks_insert_block(struct file_lock *blocker,
 711					struct file_lock *waiter)
 712{
 713	spin_lock(&blocked_lock_lock);
 714	__locks_insert_block(blocker, waiter);
 715	spin_unlock(&blocked_lock_lock);
 716}
 717
 718/*
 719 * Wake up processes blocked waiting for blocker.
 720 *
 721 * Must be called with the inode->flc_lock held!
 722 */
 723static void locks_wake_up_blocks(struct file_lock *blocker)
 724{
 725	/*
 726	 * Avoid taking global lock if list is empty. This is safe since new
 727	 * blocked requests are only added to the list under the flc_lock, and
 728	 * the flc_lock is always held here. Note that removal from the fl_block
 729	 * list does not require the flc_lock, so we must recheck list_empty()
 730	 * after acquiring the blocked_lock_lock.
 731	 */
 732	if (list_empty(&blocker->fl_block))
 733		return;
 734
 735	spin_lock(&blocked_lock_lock);
 736	while (!list_empty(&blocker->fl_block)) {
 737		struct file_lock *waiter;
 738
 739		waiter = list_first_entry(&blocker->fl_block,
 740				struct file_lock, fl_block);
 741		__locks_delete_block(waiter);
 742		if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
 743			waiter->fl_lmops->lm_notify(waiter);
 744		else
 745			wake_up(&waiter->fl_wait);
 746	}
 747	spin_unlock(&blocked_lock_lock);
 748}
 749
 750static void
 751locks_insert_lock_ctx(struct file_lock *fl, struct list_head *before)
 
 
 752{
 753	list_add_tail(&fl->fl_list, before);
 754	locks_insert_global_locks(fl);
 755}
 756
 757static void
 758locks_unlink_lock_ctx(struct file_lock *fl)
 759{
 760	locks_delete_global_locks(fl);
 761	list_del_init(&fl->fl_list);
 762	locks_wake_up_blocks(fl);
 763}
 764
 765static void
 766locks_delete_lock_ctx(struct file_lock *fl, struct list_head *dispose)
 
 
 
 
 
 767{
 768	locks_unlink_lock_ctx(fl);
 769	if (dispose)
 770		list_add(&fl->fl_list, dispose);
 771	else
 772		locks_free_lock(fl);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 773}
 774
 775/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
 776 * checks for shared/exclusive status of overlapping locks.
 777 */
 778static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
 779{
 780	if (sys_fl->fl_type == F_WRLCK)
 781		return 1;
 782	if (caller_fl->fl_type == F_WRLCK)
 783		return 1;
 784	return 0;
 785}
 786
 787/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
 788 * checking before calling the locks_conflict().
 789 */
 790static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
 791{
 792	/* POSIX locks owned by the same process do not conflict with
 793	 * each other.
 794	 */
 795	if (posix_same_owner(caller_fl, sys_fl))
 796		return (0);
 797
 798	/* Check whether they overlap */
 799	if (!locks_overlap(caller_fl, sys_fl))
 800		return 0;
 801
 802	return (locks_conflict(caller_fl, sys_fl));
 803}
 804
 805/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
 806 * checking before calling the locks_conflict().
 807 */
 808static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
 809{
 810	/* FLOCK locks referring to the same filp do not conflict with
 811	 * each other.
 812	 */
 813	if (caller_fl->fl_file == sys_fl->fl_file)
 814		return (0);
 815	if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
 816		return 0;
 817
 818	return (locks_conflict(caller_fl, sys_fl));
 819}
 820
 821void
 822posix_test_lock(struct file *filp, struct file_lock *fl)
 823{
 824	struct file_lock *cfl;
 825	struct file_lock_context *ctx;
 826	struct inode *inode = locks_inode(filp);
 827
 828	ctx = smp_load_acquire(&inode->i_flctx);
 829	if (!ctx || list_empty_careful(&ctx->flc_posix)) {
 830		fl->fl_type = F_UNLCK;
 831		return;
 832	}
 833
 834	spin_lock(&ctx->flc_lock);
 835	list_for_each_entry(cfl, &ctx->flc_posix, fl_list) {
 836		if (posix_locks_conflict(fl, cfl)) {
 837			locks_copy_conflock(fl, cfl);
 838			goto out;
 839		}
 840	}
 841	fl->fl_type = F_UNLCK;
 842out:
 843	spin_unlock(&ctx->flc_lock);
 
 
 
 
 844	return;
 845}
 846EXPORT_SYMBOL(posix_test_lock);
 847
 848/*
 849 * Deadlock detection:
 850 *
 851 * We attempt to detect deadlocks that are due purely to posix file
 852 * locks.
 853 *
 854 * We assume that a task can be waiting for at most one lock at a time.
 855 * So for any acquired lock, the process holding that lock may be
 856 * waiting on at most one other lock.  That lock in turns may be held by
 857 * someone waiting for at most one other lock.  Given a requested lock
 858 * caller_fl which is about to wait for a conflicting lock block_fl, we
 859 * follow this chain of waiters to ensure we are not about to create a
 860 * cycle.
 861 *
 862 * Since we do this before we ever put a process to sleep on a lock, we
 863 * are ensured that there is never a cycle; that is what guarantees that
 864 * the while() loop in posix_locks_deadlock() eventually completes.
 865 *
 866 * Note: the above assumption may not be true when handling lock
 867 * requests from a broken NFS client. It may also fail in the presence
 868 * of tasks (such as posix threads) sharing the same open file table.
 869 * To handle those cases, we just bail out after a few iterations.
 870 *
 871 * For FL_OFDLCK locks, the owner is the filp, not the files_struct.
 872 * Because the owner is not even nominally tied to a thread of
 873 * execution, the deadlock detection below can't reasonably work well. Just
 874 * skip it for those.
 875 *
 876 * In principle, we could do a more limited deadlock detection on FL_OFDLCK
 877 * locks that just checks for the case where two tasks are attempting to
 878 * upgrade from read to write locks on the same inode.
 879 */
 880
 881#define MAX_DEADLK_ITERATIONS 10
 882
 883/* Find a lock that the owner of the given block_fl is blocking on. */
 884static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
 885{
 886	struct file_lock *fl;
 887
 888	hash_for_each_possible(blocked_hash, fl, fl_link, posix_owner_key(block_fl)) {
 889		if (posix_same_owner(fl, block_fl))
 890			return fl->fl_next;
 891	}
 892	return NULL;
 893}
 894
 895/* Must be called with the blocked_lock_lock held! */
 896static int posix_locks_deadlock(struct file_lock *caller_fl,
 897				struct file_lock *block_fl)
 898{
 899	int i = 0;
 900
 901	lockdep_assert_held(&blocked_lock_lock);
 902
 903	/*
 904	 * This deadlock detector can't reasonably detect deadlocks with
 905	 * FL_OFDLCK locks, since they aren't owned by a process, per-se.
 906	 */
 907	if (IS_OFDLCK(caller_fl))
 908		return 0;
 909
 910	while ((block_fl = what_owner_is_waiting_for(block_fl))) {
 911		if (i++ > MAX_DEADLK_ITERATIONS)
 912			return 0;
 913		if (posix_same_owner(caller_fl, block_fl))
 914			return 1;
 915	}
 916	return 0;
 917}
 918
 919/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
 920 * after any leases, but before any posix locks.
 921 *
 922 * Note that if called with an FL_EXISTS argument, the caller may determine
 923 * whether or not a lock was successfully freed by testing the return
 924 * value for -ENOENT.
 925 */
 926static int flock_lock_inode(struct inode *inode, struct file_lock *request)
 927{
 928	struct file_lock *new_fl = NULL;
 929	struct file_lock *fl;
 930	struct file_lock_context *ctx;
 931	int error = 0;
 932	bool found = false;
 933	LIST_HEAD(dispose);
 934
 935	ctx = locks_get_lock_context(inode, request->fl_type);
 936	if (!ctx) {
 937		if (request->fl_type != F_UNLCK)
 938			return -ENOMEM;
 939		return (request->fl_flags & FL_EXISTS) ? -ENOENT : 0;
 940	}
 941
 942	if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
 943		new_fl = locks_alloc_lock();
 944		if (!new_fl)
 945			return -ENOMEM;
 946	}
 947
 948	percpu_down_read_preempt_disable(&file_rwsem);
 949	spin_lock(&ctx->flc_lock);
 950	if (request->fl_flags & FL_ACCESS)
 951		goto find_conflict;
 952
 953	list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
 954		if (request->fl_file != fl->fl_file)
 
 
 
 
 
 955			continue;
 956		if (request->fl_type == fl->fl_type)
 957			goto out;
 958		found = true;
 959		locks_delete_lock_ctx(fl, &dispose);
 960		break;
 961	}
 962
 963	if (request->fl_type == F_UNLCK) {
 964		if ((request->fl_flags & FL_EXISTS) && !found)
 965			error = -ENOENT;
 966		goto out;
 967	}
 968
 
 
 
 
 
 
 
 
 
 
 969find_conflict:
 970	list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
 
 
 
 
 
 971		if (!flock_locks_conflict(request, fl))
 972			continue;
 973		error = -EAGAIN;
 974		if (!(request->fl_flags & FL_SLEEP))
 975			goto out;
 976		error = FILE_LOCK_DEFERRED;
 977		locks_insert_block(fl, request);
 978		goto out;
 979	}
 980	if (request->fl_flags & FL_ACCESS)
 981		goto out;
 982	locks_copy_lock(new_fl, request);
 983	locks_insert_lock_ctx(new_fl, &ctx->flc_flock);
 984	new_fl = NULL;
 985	error = 0;
 986
 987out:
 988	spin_unlock(&ctx->flc_lock);
 989	percpu_up_read_preempt_enable(&file_rwsem);
 990	if (new_fl)
 991		locks_free_lock(new_fl);
 992	locks_dispose_list(&dispose);
 993	return error;
 994}
 995
 996static int posix_lock_inode(struct inode *inode, struct file_lock *request,
 997			    struct file_lock *conflock)
 998{
 999	struct file_lock *fl, *tmp;
1000	struct file_lock *new_fl = NULL;
1001	struct file_lock *new_fl2 = NULL;
1002	struct file_lock *left = NULL;
1003	struct file_lock *right = NULL;
1004	struct file_lock_context *ctx;
1005	int error;
1006	bool added = false;
1007	LIST_HEAD(dispose);
1008
1009	ctx = locks_get_lock_context(inode, request->fl_type);
1010	if (!ctx)
1011		return (request->fl_type == F_UNLCK) ? 0 : -ENOMEM;
1012
1013	/*
1014	 * We may need two file_lock structures for this operation,
1015	 * so we get them in advance to avoid races.
1016	 *
1017	 * In some cases we can be sure, that no new locks will be needed
1018	 */
1019	if (!(request->fl_flags & FL_ACCESS) &&
1020	    (request->fl_type != F_UNLCK ||
1021	     request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
1022		new_fl = locks_alloc_lock();
1023		new_fl2 = locks_alloc_lock();
1024	}
1025
1026	percpu_down_read_preempt_disable(&file_rwsem);
1027	spin_lock(&ctx->flc_lock);
1028	/*
1029	 * New lock request. Walk all POSIX locks and look for conflicts. If
1030	 * there are any, either return error or put the request on the
1031	 * blocker's list of waiters and the global blocked_hash.
1032	 */
1033	if (request->fl_type != F_UNLCK) {
1034		list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
 
 
 
1035			if (!posix_locks_conflict(request, fl))
1036				continue;
1037			if (conflock)
1038				locks_copy_conflock(conflock, fl);
1039			error = -EAGAIN;
1040			if (!(request->fl_flags & FL_SLEEP))
1041				goto out;
1042			/*
1043			 * Deadlock detection and insertion into the blocked
1044			 * locks list must be done while holding the same lock!
1045			 */
1046			error = -EDEADLK;
1047			spin_lock(&blocked_lock_lock);
1048			if (likely(!posix_locks_deadlock(request, fl))) {
1049				error = FILE_LOCK_DEFERRED;
1050				__locks_insert_block(fl, request);
1051			}
1052			spin_unlock(&blocked_lock_lock);
1053			goto out;
1054  		}
1055  	}
1056
1057	/* If we're just looking for a conflict, we're done. */
1058	error = 0;
1059	if (request->fl_flags & FL_ACCESS)
1060		goto out;
1061
1062	/* Find the first old lock with the same owner as the new lock */
1063	list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1064		if (posix_same_owner(request, fl))
1065			break;
1066	}
1067
1068	/* Process locks with this owner. */
1069	list_for_each_entry_safe_from(fl, tmp, &ctx->flc_posix, fl_list) {
1070		if (!posix_same_owner(request, fl))
1071			break;
 
1072
1073		/* Detect adjacent or overlapping regions (if same lock type) */
 
 
 
1074		if (request->fl_type == fl->fl_type) {
1075			/* In all comparisons of start vs end, use
1076			 * "start - 1" rather than "end + 1". If end
1077			 * is OFFSET_MAX, end + 1 will become negative.
1078			 */
1079			if (fl->fl_end < request->fl_start - 1)
1080				continue;
1081			/* If the next lock in the list has entirely bigger
1082			 * addresses than the new one, insert the lock here.
1083			 */
1084			if (fl->fl_start - 1 > request->fl_end)
1085				break;
1086
1087			/* If we come here, the new and old lock are of the
1088			 * same type and adjacent or overlapping. Make one
1089			 * lock yielding from the lower start address of both
1090			 * locks to the higher end address.
1091			 */
1092			if (fl->fl_start > request->fl_start)
1093				fl->fl_start = request->fl_start;
1094			else
1095				request->fl_start = fl->fl_start;
1096			if (fl->fl_end < request->fl_end)
1097				fl->fl_end = request->fl_end;
1098			else
1099				request->fl_end = fl->fl_end;
1100			if (added) {
1101				locks_delete_lock_ctx(fl, &dispose);
1102				continue;
1103			}
1104			request = fl;
1105			added = true;
1106		} else {
 
1107			/* Processing for different lock types is a bit
1108			 * more complex.
1109			 */
1110			if (fl->fl_end < request->fl_start)
1111				continue;
1112			if (fl->fl_start > request->fl_end)
1113				break;
1114			if (request->fl_type == F_UNLCK)
1115				added = true;
1116			if (fl->fl_start < request->fl_start)
1117				left = fl;
1118			/* If the next lock in the list has a higher end
1119			 * address than the new one, insert the new one here.
1120			 */
1121			if (fl->fl_end > request->fl_end) {
1122				right = fl;
1123				break;
1124			}
1125			if (fl->fl_start >= request->fl_start) {
1126				/* The new lock completely replaces an old
1127				 * one (This may happen several times).
1128				 */
1129				if (added) {
1130					locks_delete_lock_ctx(fl, &dispose);
1131					continue;
1132				}
1133				/*
1134				 * Replace the old lock with new_fl, and
1135				 * remove the old one. It's safe to do the
1136				 * insert here since we know that we won't be
1137				 * using new_fl later, and that the lock is
1138				 * just replacing an existing lock.
1139				 */
1140				error = -ENOLCK;
1141				if (!new_fl)
1142					goto out;
1143				locks_copy_lock(new_fl, request);
1144				request = new_fl;
1145				new_fl = NULL;
1146				locks_insert_lock_ctx(request, &fl->fl_list);
1147				locks_delete_lock_ctx(fl, &dispose);
1148				added = true;
1149			}
1150		}
 
 
 
 
1151	}
1152
1153	/*
1154	 * The above code only modifies existing locks in case of merging or
1155	 * replacing. If new lock(s) need to be inserted all modifications are
1156	 * done below this, so it's safe yet to bail out.
 
1157	 */
1158	error = -ENOLCK; /* "no luck" */
1159	if (right && left == right && !new_fl2)
1160		goto out;
1161
1162	error = 0;
1163	if (!added) {
1164		if (request->fl_type == F_UNLCK) {
1165			if (request->fl_flags & FL_EXISTS)
1166				error = -ENOENT;
1167			goto out;
1168		}
1169
1170		if (!new_fl) {
1171			error = -ENOLCK;
1172			goto out;
1173		}
1174		locks_copy_lock(new_fl, request);
1175		locks_insert_lock_ctx(new_fl, &fl->fl_list);
1176		fl = new_fl;
1177		new_fl = NULL;
1178	}
1179	if (right) {
1180		if (left == right) {
1181			/* The new lock breaks the old one in two pieces,
1182			 * so we have to use the second new lock.
1183			 */
1184			left = new_fl2;
1185			new_fl2 = NULL;
1186			locks_copy_lock(left, right);
1187			locks_insert_lock_ctx(left, &fl->fl_list);
1188		}
1189		right->fl_start = request->fl_end + 1;
1190		locks_wake_up_blocks(right);
1191	}
1192	if (left) {
1193		left->fl_end = request->fl_start - 1;
1194		locks_wake_up_blocks(left);
1195	}
1196 out:
1197	spin_unlock(&ctx->flc_lock);
1198	percpu_up_read_preempt_enable(&file_rwsem);
1199	/*
1200	 * Free any unused locks.
1201	 */
1202	if (new_fl)
1203		locks_free_lock(new_fl);
1204	if (new_fl2)
1205		locks_free_lock(new_fl2);
1206	locks_dispose_list(&dispose);
1207	trace_posix_lock_inode(inode, request, error);
1208
1209	return error;
1210}
1211
1212/**
1213 * posix_lock_file - Apply a POSIX-style lock to a file
1214 * @filp: The file to apply the lock to
1215 * @fl: The lock to be applied
1216 * @conflock: Place to return a copy of the conflicting lock, if found.
1217 *
1218 * Add a POSIX style lock to a file.
1219 * We merge adjacent & overlapping locks whenever possible.
1220 * POSIX locks are sorted by owner task, then by starting address
1221 *
1222 * Note that if called with an FL_EXISTS argument, the caller may determine
1223 * whether or not a lock was successfully freed by testing the return
1224 * value for -ENOENT.
1225 */
1226int posix_lock_file(struct file *filp, struct file_lock *fl,
1227			struct file_lock *conflock)
1228{
1229	return posix_lock_inode(locks_inode(filp), fl, conflock);
1230}
1231EXPORT_SYMBOL(posix_lock_file);
1232
1233/**
1234 * posix_lock_inode_wait - Apply a POSIX-style lock to a file
1235 * @inode: inode of file to which lock request should be applied
1236 * @fl: The lock to be applied
1237 *
1238 * Apply a POSIX style lock request to an inode.
 
 
1239 */
1240static int posix_lock_inode_wait(struct inode *inode, struct file_lock *fl)
1241{
1242	int error;
1243	might_sleep ();
1244	for (;;) {
1245		error = posix_lock_inode(inode, fl, NULL);
1246		if (error != FILE_LOCK_DEFERRED)
1247			break;
1248		error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1249		if (!error)
1250			continue;
1251
1252		locks_delete_block(fl);
1253		break;
1254	}
1255	return error;
1256}
 
1257
1258#ifdef CONFIG_MANDATORY_FILE_LOCKING
1259/**
1260 * locks_mandatory_locked - Check for an active lock
1261 * @file: the file to check
1262 *
1263 * Searches the inode's list of locks to find any POSIX locks which conflict.
1264 * This function is called from locks_verify_locked() only.
1265 */
1266int locks_mandatory_locked(struct file *file)
1267{
1268	int ret;
1269	struct inode *inode = locks_inode(file);
1270	struct file_lock_context *ctx;
1271	struct file_lock *fl;
1272
1273	ctx = smp_load_acquire(&inode->i_flctx);
1274	if (!ctx || list_empty_careful(&ctx->flc_posix))
1275		return 0;
1276
1277	/*
1278	 * Search the lock list for this inode for any POSIX locks.
1279	 */
1280	spin_lock(&ctx->flc_lock);
1281	ret = 0;
1282	list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1283		if (fl->fl_owner != current->files &&
1284		    fl->fl_owner != file) {
1285			ret = -EAGAIN;
1286			break;
1287		}
1288	}
1289	spin_unlock(&ctx->flc_lock);
1290	return ret;
1291}
1292
1293/**
1294 * locks_mandatory_area - Check for a conflicting lock
1295 * @inode:	the file to check
 
 
1296 * @filp:       how the file was opened (if it was)
1297 * @start:	first byte in the file to check
1298 * @end:	lastbyte in the file to check
1299 * @type:	%F_WRLCK for a write lock, else %F_RDLCK
1300 *
1301 * Searches the inode's list of locks to find any POSIX locks which conflict.
 
 
1302 */
1303int locks_mandatory_area(struct inode *inode, struct file *filp, loff_t start,
1304			 loff_t end, unsigned char type)
 
1305{
1306	struct file_lock fl;
1307	int error;
1308	bool sleep = false;
1309
1310	locks_init_lock(&fl);
 
1311	fl.fl_pid = current->tgid;
1312	fl.fl_file = filp;
1313	fl.fl_flags = FL_POSIX | FL_ACCESS;
1314	if (filp && !(filp->f_flags & O_NONBLOCK))
1315		sleep = true;
1316	fl.fl_type = type;
1317	fl.fl_start = start;
1318	fl.fl_end = end;
1319
1320	for (;;) {
1321		if (filp) {
1322			fl.fl_owner = filp;
1323			fl.fl_flags &= ~FL_SLEEP;
1324			error = posix_lock_inode(inode, &fl, NULL);
1325			if (!error)
1326				break;
1327		}
1328
1329		if (sleep)
1330			fl.fl_flags |= FL_SLEEP;
1331		fl.fl_owner = current->files;
1332		error = posix_lock_inode(inode, &fl, NULL);
1333		if (error != FILE_LOCK_DEFERRED)
1334			break;
1335		error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1336		if (!error) {
1337			/*
1338			 * If we've been sleeping someone might have
1339			 * changed the permissions behind our back.
1340			 */
1341			if (__mandatory_lock(inode))
1342				continue;
1343		}
1344
1345		locks_delete_block(&fl);
1346		break;
1347	}
1348
1349	return error;
1350}
1351
1352EXPORT_SYMBOL(locks_mandatory_area);
1353#endif /* CONFIG_MANDATORY_FILE_LOCKING */
1354
1355static void lease_clear_pending(struct file_lock *fl, int arg)
1356{
1357	switch (arg) {
1358	case F_UNLCK:
1359		fl->fl_flags &= ~FL_UNLOCK_PENDING;
1360		/* fall through: */
1361	case F_RDLCK:
1362		fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1363	}
1364}
1365
1366/* We already had a lease on this file; just change its type */
1367int lease_modify(struct file_lock *fl, int arg, struct list_head *dispose)
1368{
 
1369	int error = assign_type(fl, arg);
1370
1371	if (error)
1372		return error;
1373	lease_clear_pending(fl, arg);
1374	locks_wake_up_blocks(fl);
1375	if (arg == F_UNLCK) {
1376		struct file *filp = fl->fl_file;
1377
1378		f_delown(filp);
1379		filp->f_owner.signum = 0;
1380		fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
1381		if (fl->fl_fasync != NULL) {
1382			printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1383			fl->fl_fasync = NULL;
1384		}
1385		locks_delete_lock_ctx(fl, dispose);
1386	}
1387	return 0;
1388}
1389EXPORT_SYMBOL(lease_modify);
1390
1391static bool past_time(unsigned long then)
1392{
1393	if (!then)
1394		/* 0 is a special value meaning "this never expires": */
1395		return false;
1396	return time_after(jiffies, then);
1397}
1398
1399static void time_out_leases(struct inode *inode, struct list_head *dispose)
1400{
1401	struct file_lock_context *ctx = inode->i_flctx;
1402	struct file_lock *fl, *tmp;
1403
1404	lockdep_assert_held(&ctx->flc_lock);
1405
1406	list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
1407		trace_time_out_leases(inode, fl);
1408		if (past_time(fl->fl_downgrade_time))
1409			lease_modify(fl, F_RDLCK, dispose);
1410		if (past_time(fl->fl_break_time))
1411			lease_modify(fl, F_UNLCK, dispose);
1412	}
1413}
1414
1415static bool leases_conflict(struct file_lock *lease, struct file_lock *breaker)
1416{
1417	if ((breaker->fl_flags & FL_LAYOUT) != (lease->fl_flags & FL_LAYOUT))
1418		return false;
1419	if ((breaker->fl_flags & FL_DELEG) && (lease->fl_flags & FL_LEASE))
1420		return false;
1421	return locks_conflict(breaker, lease);
1422}
1423
1424static bool
1425any_leases_conflict(struct inode *inode, struct file_lock *breaker)
1426{
1427	struct file_lock_context *ctx = inode->i_flctx;
1428	struct file_lock *fl;
1429
1430	lockdep_assert_held(&ctx->flc_lock);
1431
1432	list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1433		if (leases_conflict(fl, breaker))
1434			return true;
 
 
 
 
 
1435	}
1436	return false;
1437}
1438
1439/**
1440 *	__break_lease	-	revoke all outstanding leases on file
1441 *	@inode: the inode of the file to return
1442 *	@mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR:
1443 *	    break all leases
1444 *	@type: FL_LEASE: break leases and delegations; FL_DELEG: break
1445 *	    only delegations
1446 *
1447 *	break_lease (inlined for speed) has checked there already is at least
1448 *	some kind of lock (maybe a lease) on this file.  Leases are broken on
1449 *	a call to open() or truncate().  This function can sleep unless you
1450 *	specified %O_NONBLOCK to your open().
1451 */
1452int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
1453{
1454	int error = 0;
1455	struct file_lock_context *ctx;
1456	struct file_lock *new_fl, *fl, *tmp;
1457	unsigned long break_time;
 
1458	int want_write = (mode & O_ACCMODE) != O_RDONLY;
1459	LIST_HEAD(dispose);
1460
1461	new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
1462	if (IS_ERR(new_fl))
1463		return PTR_ERR(new_fl);
1464	new_fl->fl_flags = type;
1465
1466	/* typically we will check that ctx is non-NULL before calling */
1467	ctx = smp_load_acquire(&inode->i_flctx);
1468	if (!ctx) {
1469		WARN_ON_ONCE(1);
1470		return error;
1471	}
1472
1473	percpu_down_read_preempt_disable(&file_rwsem);
1474	spin_lock(&ctx->flc_lock);
1475
1476	time_out_leases(inode, &dispose);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1477
1478	if (!any_leases_conflict(inode, new_fl))
 
 
1479		goto out;
 
1480
1481	break_time = 0;
1482	if (lease_break_time > 0) {
1483		break_time = jiffies + lease_break_time * HZ;
1484		if (break_time == 0)
1485			break_time++;	/* so that 0 means no break time */
1486	}
1487
1488	list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
1489		if (!leases_conflict(fl, new_fl))
1490			continue;
1491		if (want_write) {
1492			if (fl->fl_flags & FL_UNLOCK_PENDING)
1493				continue;
1494			fl->fl_flags |= FL_UNLOCK_PENDING;
1495			fl->fl_break_time = break_time;
1496		} else {
1497			if (lease_breaking(fl))
1498				continue;
1499			fl->fl_flags |= FL_DOWNGRADE_PENDING;
1500			fl->fl_downgrade_time = break_time;
1501		}
1502		if (fl->fl_lmops->lm_break(fl))
1503			locks_delete_lock_ctx(fl, &dispose);
1504	}
1505
1506	if (list_empty(&ctx->flc_lease))
1507		goto out;
1508
1509	if (mode & O_NONBLOCK) {
1510		trace_break_lease_noblock(inode, new_fl);
1511		error = -EWOULDBLOCK;
1512		goto out;
1513	}
1514
1515restart:
1516	fl = list_first_entry(&ctx->flc_lease, struct file_lock, fl_list);
1517	break_time = fl->fl_break_time;
1518	if (break_time != 0)
1519		break_time -= jiffies;
1520	if (break_time == 0)
1521		break_time++;
1522	locks_insert_block(fl, new_fl);
1523	trace_break_lease_block(inode, new_fl);
1524	spin_unlock(&ctx->flc_lock);
1525	percpu_up_read_preempt_enable(&file_rwsem);
1526
1527	locks_dispose_list(&dispose);
1528	error = wait_event_interruptible_timeout(new_fl->fl_wait,
1529						!new_fl->fl_next, break_time);
1530
1531	percpu_down_read_preempt_disable(&file_rwsem);
1532	spin_lock(&ctx->flc_lock);
1533	trace_break_lease_unblock(inode, new_fl);
1534	locks_delete_block(new_fl);
1535	if (error >= 0) {
1536		/*
1537		 * Wait for the next conflicting lease that has not been
1538		 * broken yet
1539		 */
1540		if (error == 0)
1541			time_out_leases(inode, &dispose);
1542		if (any_leases_conflict(inode, new_fl))
1543			goto restart;
 
 
 
 
1544		error = 0;
1545	}
 
1546out:
1547	spin_unlock(&ctx->flc_lock);
1548	percpu_up_read_preempt_enable(&file_rwsem);
1549	locks_dispose_list(&dispose);
1550	locks_free_lock(new_fl);
1551	return error;
1552}
1553
1554EXPORT_SYMBOL(__break_lease);
1555
1556/**
1557 *	lease_get_mtime - update modified time of an inode with exclusive lease
1558 *	@inode: the inode
1559 *      @time:  pointer to a timespec which contains the last modified time
1560 *
1561 * This is to force NFS clients to flush their caches for files with
1562 * exclusive leases.  The justification is that if someone has an
1563 * exclusive lease, then they could be modifying it.
1564 */
1565void lease_get_mtime(struct inode *inode, struct timespec *time)
1566{
1567	bool has_lease = false;
1568	struct file_lock_context *ctx;
1569	struct file_lock *fl;
1570
1571	ctx = smp_load_acquire(&inode->i_flctx);
1572	if (ctx && !list_empty_careful(&ctx->flc_lease)) {
1573		spin_lock(&ctx->flc_lock);
1574		fl = list_first_entry_or_null(&ctx->flc_lease,
1575					      struct file_lock, fl_list);
1576		if (fl && (fl->fl_type == F_WRLCK))
1577			has_lease = true;
1578		spin_unlock(&ctx->flc_lock);
1579	}
1580
1581	if (has_lease)
1582		*time = current_time(inode);
1583}
1584
1585EXPORT_SYMBOL(lease_get_mtime);
1586
1587/**
1588 *	fcntl_getlease - Enquire what lease is currently active
1589 *	@filp: the file
1590 *
1591 *	The value returned by this function will be one of
1592 *	(if no lease break is pending):
1593 *
1594 *	%F_RDLCK to indicate a shared lease is held.
1595 *
1596 *	%F_WRLCK to indicate an exclusive lease is held.
1597 *
1598 *	%F_UNLCK to indicate no lease is held.
1599 *
1600 *	(if a lease break is pending):
1601 *
1602 *	%F_RDLCK to indicate an exclusive lease needs to be
1603 *		changed to a shared lease (or removed).
1604 *
1605 *	%F_UNLCK to indicate the lease needs to be removed.
1606 *
1607 *	XXX: sfr & willy disagree over whether F_INPROGRESS
1608 *	should be returned to userspace.
1609 */
1610int fcntl_getlease(struct file *filp)
1611{
1612	struct file_lock *fl;
1613	struct inode *inode = locks_inode(filp);
1614	struct file_lock_context *ctx;
1615	int type = F_UNLCK;
1616	LIST_HEAD(dispose);
1617
1618	ctx = smp_load_acquire(&inode->i_flctx);
1619	if (ctx && !list_empty_careful(&ctx->flc_lease)) {
1620		percpu_down_read_preempt_disable(&file_rwsem);
1621		spin_lock(&ctx->flc_lock);
1622		time_out_leases(inode, &dispose);
1623		list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1624			if (fl->fl_file != filp)
1625				continue;
1626			type = target_leasetype(fl);
1627			break;
1628		}
1629		spin_unlock(&ctx->flc_lock);
1630		percpu_up_read_preempt_enable(&file_rwsem);
1631
1632		locks_dispose_list(&dispose);
1633	}
 
1634	return type;
1635}
1636
1637/**
1638 * check_conflicting_open - see if the given dentry points to a file that has
1639 * 			    an existing open that would conflict with the
1640 * 			    desired lease.
1641 * @dentry:	dentry to check
1642 * @arg:	type of lease that we're trying to acquire
1643 * @flags:	current lock flags
1644 *
1645 * Check to see if there's an existing open fd on this file that would
1646 * conflict with the lease we're trying to set.
 
 
1647 */
1648static int
1649check_conflicting_open(const struct dentry *dentry, const long arg, int flags)
1650{
1651	int ret = 0;
1652	struct inode *inode = dentry->d_inode;
1653
1654	if (flags & FL_LAYOUT)
1655		return 0;
1656
1657	if ((arg == F_RDLCK) &&
1658	    (atomic_read(&d_real_inode(dentry)->i_writecount) > 0))
1659		return -EAGAIN;
1660
1661	if ((arg == F_WRLCK) && ((d_count(dentry) > 1) ||
1662	    (atomic_read(&inode->i_count) > 1)))
1663		ret = -EAGAIN;
1664
1665	return ret;
1666}
1667
1668static int
1669generic_add_lease(struct file *filp, long arg, struct file_lock **flp, void **priv)
1670{
1671	struct file_lock *fl, *my_fl = NULL, *lease;
1672	struct dentry *dentry = filp->f_path.dentry;
1673	struct inode *inode = dentry->d_inode;
1674	struct file_lock_context *ctx;
1675	bool is_deleg = (*flp)->fl_flags & FL_DELEG;
1676	int error;
1677	LIST_HEAD(dispose);
1678
1679	lease = *flp;
1680	trace_generic_add_lease(inode, lease);
1681
1682	/* Note that arg is never F_UNLCK here */
1683	ctx = locks_get_lock_context(inode, arg);
1684	if (!ctx)
1685		return -ENOMEM;
 
 
 
 
 
1686
1687	/*
1688	 * In the delegation case we need mutual exclusion with
1689	 * a number of operations that take the i_mutex.  We trylock
1690	 * because delegations are an optional optimization, and if
1691	 * there's some chance of a conflict--we'd rather not
1692	 * bother, maybe that's a sign this just isn't a good file to
1693	 * hand out a delegation on.
1694	 */
1695	if (is_deleg && !inode_trylock(inode))
1696		return -EAGAIN;
1697
1698	if (is_deleg && arg == F_WRLCK) {
1699		/* Write delegations are not currently supported: */
1700		inode_unlock(inode);
1701		WARN_ON_ONCE(1);
1702		return -EINVAL;
1703	}
1704
1705	percpu_down_read_preempt_disable(&file_rwsem);
1706	spin_lock(&ctx->flc_lock);
1707	time_out_leases(inode, &dispose);
1708	error = check_conflicting_open(dentry, arg, lease->fl_flags);
1709	if (error)
1710		goto out;
 
 
 
1711
1712	/*
1713	 * At this point, we know that if there is an exclusive
1714	 * lease on this file, then we hold it on this filp
1715	 * (otherwise our open of this file would have blocked).
1716	 * And if we are trying to acquire an exclusive lease,
1717	 * then the file is not open by anyone (including us)
1718	 * except for this filp.
1719	 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1720	error = -EAGAIN;
1721	list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1722		if (fl->fl_file == filp &&
1723		    fl->fl_owner == lease->fl_owner) {
1724			my_fl = fl;
1725			continue;
1726		}
1727
1728		/*
1729		 * No exclusive leases if someone else has a lease on
1730		 * this file:
1731		 */
1732		if (arg == F_WRLCK)
1733			goto out;
1734		/*
1735		 * Modifying our existing lease is OK, but no getting a
1736		 * new lease if someone else is opening for write:
1737		 */
1738		if (fl->fl_flags & FL_UNLOCK_PENDING)
1739			goto out;
1740	}
1741
1742	if (my_fl != NULL) {
1743		lease = my_fl;
1744		error = lease->fl_lmops->lm_change(lease, arg, &dispose);
1745		if (error)
1746			goto out;
1747		goto out_setup;
1748	}
1749
1750	error = -EINVAL;
1751	if (!leases_enable)
1752		goto out;
1753
1754	locks_insert_lock_ctx(lease, &ctx->flc_lease);
1755	/*
1756	 * The check in break_lease() is lockless. It's possible for another
1757	 * open to race in after we did the earlier check for a conflicting
1758	 * open but before the lease was inserted. Check again for a
1759	 * conflicting open and cancel the lease if there is one.
1760	 *
1761	 * We also add a barrier here to ensure that the insertion of the lock
1762	 * precedes these checks.
1763	 */
1764	smp_mb();
1765	error = check_conflicting_open(dentry, arg, lease->fl_flags);
1766	if (error) {
1767		locks_unlink_lock_ctx(lease);
1768		goto out;
1769	}
1770
1771out_setup:
1772	if (lease->fl_lmops->lm_setup)
1773		lease->fl_lmops->lm_setup(lease, priv);
1774out:
1775	spin_unlock(&ctx->flc_lock);
1776	percpu_up_read_preempt_enable(&file_rwsem);
1777	locks_dispose_list(&dispose);
1778	if (is_deleg)
1779		inode_unlock(inode);
1780	if (!error && !my_fl)
1781		*flp = NULL;
1782	return error;
1783}
 
1784
1785static int generic_delete_lease(struct file *filp, void *owner)
1786{
1787	int error = -EAGAIN;
1788	struct file_lock *fl, *victim = NULL;
1789	struct inode *inode = locks_inode(filp);
1790	struct file_lock_context *ctx;
1791	LIST_HEAD(dispose);
1792
1793	ctx = smp_load_acquire(&inode->i_flctx);
1794	if (!ctx) {
1795		trace_generic_delete_lease(inode, NULL);
1796		return error;
1797	}
1798
1799	percpu_down_read_preempt_disable(&file_rwsem);
1800	spin_lock(&ctx->flc_lock);
1801	list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1802		if (fl->fl_file == filp &&
1803		    fl->fl_owner == owner) {
1804			victim = fl;
1805			break;
1806		}
1807	}
1808	trace_generic_delete_lease(inode, victim);
1809	if (victim)
1810		error = fl->fl_lmops->lm_change(victim, F_UNLCK, &dispose);
1811	spin_unlock(&ctx->flc_lock);
1812	percpu_up_read_preempt_enable(&file_rwsem);
1813	locks_dispose_list(&dispose);
1814	return error;
1815}
1816
1817/**
1818 *	generic_setlease	-	sets a lease on an open file
1819 *	@filp:	file pointer
1820 *	@arg:	type of lease to obtain
1821 *	@flp:	input - file_lock to use, output - file_lock inserted
1822 *	@priv:	private data for lm_setup (may be NULL if lm_setup
1823 *		doesn't require it)
1824 *
1825 *	The (input) flp->fl_lmops->lm_break function is required
1826 *	by break_lease().
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1827 */
1828int generic_setlease(struct file *filp, long arg, struct file_lock **flp,
1829			void **priv)
1830{
1831	struct inode *inode = locks_inode(filp);
1832	int error;
1833
1834	if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
1835		return -EACCES;
1836	if (!S_ISREG(inode->i_mode))
1837		return -EINVAL;
1838	error = security_file_lock(filp, arg);
1839	if (error)
1840		return error;
1841
1842	switch (arg) {
1843	case F_UNLCK:
1844		return generic_delete_lease(filp, *priv);
1845	case F_RDLCK:
1846	case F_WRLCK:
1847		if (!(*flp)->fl_lmops->lm_break) {
1848			WARN_ON_ONCE(1);
1849			return -ENOLCK;
1850		}
1851
1852		return generic_add_lease(filp, arg, flp, priv);
1853	default:
1854		return -EINVAL;
1855	}
1856}
1857EXPORT_SYMBOL(generic_setlease);
1858
1859/**
1860 * vfs_setlease        -       sets a lease on an open file
1861 * @filp:	file pointer
1862 * @arg:	type of lease to obtain
1863 * @lease:	file_lock to use when adding a lease
1864 * @priv:	private info for lm_setup when adding a lease (may be
1865 * 		NULL if lm_setup doesn't require it)
1866 *
1867 * Call this to establish a lease on the file. The "lease" argument is not
1868 * used for F_UNLCK requests and may be NULL. For commands that set or alter
1869 * an existing lease, the ``(*lease)->fl_lmops->lm_break`` operation must be
1870 * set; if not, this function will return -ENOLCK (and generate a scary-looking
1871 * stack trace).
1872 *
1873 * The "priv" pointer is passed directly to the lm_setup function as-is. It
1874 * may be NULL if the lm_setup operation doesn't require it.
1875 */
1876int
1877vfs_setlease(struct file *filp, long arg, struct file_lock **lease, void **priv)
1878{
1879	if (filp->f_op->setlease && is_remote_lock(filp))
1880		return filp->f_op->setlease(filp, arg, lease, priv);
1881	else
1882		return generic_setlease(filp, arg, lease, priv);
 
1883}
1884EXPORT_SYMBOL_GPL(vfs_setlease);
1885
1886static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
1887{
1888	struct file_lock *fl;
1889	struct fasync_struct *new;
1890	int error;
1891
1892	fl = lease_alloc(filp, arg);
1893	if (IS_ERR(fl))
1894		return PTR_ERR(fl);
1895
1896	new = fasync_alloc();
1897	if (!new) {
1898		locks_free_lock(fl);
1899		return -ENOMEM;
1900	}
1901	new->fa_fd = fd;
1902
1903	error = vfs_setlease(filp, arg, &fl, (void **)&new);
1904	if (fl)
 
 
 
 
 
1905		locks_free_lock(fl);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1906	if (new)
1907		fasync_free(new);
1908	return error;
1909}
1910
1911/**
1912 *	fcntl_setlease	-	sets a lease on an open file
1913 *	@fd: open file descriptor
1914 *	@filp: file pointer
1915 *	@arg: type of lease to obtain
1916 *
1917 *	Call this fcntl to establish a lease on the file.
1918 *	Note that you also need to call %F_SETSIG to
1919 *	receive a signal when the lease is broken.
1920 */
1921int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1922{
1923	if (arg == F_UNLCK)
1924		return vfs_setlease(filp, F_UNLCK, NULL, (void **)&filp);
1925	return do_fcntl_add_lease(fd, filp, arg);
1926}
1927
1928/**
1929 * flock_lock_inode_wait - Apply a FLOCK-style lock to a file
1930 * @inode: inode of the file to apply to
1931 * @fl: The lock to be applied
1932 *
1933 * Apply a FLOCK style lock request to an inode.
1934 */
1935static int flock_lock_inode_wait(struct inode *inode, struct file_lock *fl)
1936{
1937	int error;
1938	might_sleep();
1939	for (;;) {
1940		error = flock_lock_inode(inode, fl);
1941		if (error != FILE_LOCK_DEFERRED)
1942			break;
1943		error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1944		if (!error)
1945			continue;
1946
1947		locks_delete_block(fl);
1948		break;
1949	}
1950	return error;
1951}
1952
1953/**
1954 * locks_lock_inode_wait - Apply a lock to an inode
1955 * @inode: inode of the file to apply to
1956 * @fl: The lock to be applied
1957 *
1958 * Apply a POSIX or FLOCK style lock request to an inode.
1959 */
1960int locks_lock_inode_wait(struct inode *inode, struct file_lock *fl)
1961{
1962	int res = 0;
1963	switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
1964		case FL_POSIX:
1965			res = posix_lock_inode_wait(inode, fl);
1966			break;
1967		case FL_FLOCK:
1968			res = flock_lock_inode_wait(inode, fl);
1969			break;
1970		default:
1971			BUG();
1972	}
1973	return res;
1974}
1975EXPORT_SYMBOL(locks_lock_inode_wait);
1976
1977/**
1978 *	sys_flock: - flock() system call.
1979 *	@fd: the file descriptor to lock.
1980 *	@cmd: the type of lock to apply.
1981 *
1982 *	Apply a %FL_FLOCK style lock to an open file descriptor.
1983 *	The @cmd can be one of:
 
 
 
 
 
 
1984 *
1985 *	- %LOCK_SH -- a shared lock.
1986 *	- %LOCK_EX -- an exclusive lock.
1987 *	- %LOCK_UN -- remove an existing lock.
1988 *	- %LOCK_MAND -- a 'mandatory' flock.
1989 *	  This exists to emulate Windows Share Modes.
1990 *
1991 *	%LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1992 *	processes read and write access respectively.
1993 */
1994SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
1995{
1996	struct fd f = fdget(fd);
1997	struct file_lock *lock;
1998	int can_sleep, unlock;
1999	int error;
2000
2001	error = -EBADF;
2002	if (!f.file)
 
2003		goto out;
2004
2005	can_sleep = !(cmd & LOCK_NB);
2006	cmd &= ~LOCK_NB;
2007	unlock = (cmd == LOCK_UN);
2008
2009	if (!unlock && !(cmd & LOCK_MAND) &&
2010	    !(f.file->f_mode & (FMODE_READ|FMODE_WRITE)))
2011		goto out_putf;
2012
2013	lock = flock_make_lock(f.file, cmd);
2014	if (IS_ERR(lock)) {
2015		error = PTR_ERR(lock);
2016		goto out_putf;
2017	}
2018
2019	if (can_sleep)
2020		lock->fl_flags |= FL_SLEEP;
2021
2022	error = security_file_lock(f.file, lock->fl_type);
2023	if (error)
2024		goto out_free;
2025
2026	if (f.file->f_op->flock && is_remote_lock(f.file))
2027		error = f.file->f_op->flock(f.file,
2028					  (can_sleep) ? F_SETLKW : F_SETLK,
2029					  lock);
2030	else
2031		error = locks_lock_file_wait(f.file, lock);
2032
2033 out_free:
2034	locks_free_lock(lock);
2035
2036 out_putf:
2037	fdput(f);
2038 out:
2039	return error;
2040}
2041
2042/**
2043 * vfs_test_lock - test file byte range lock
2044 * @filp: The file to test lock for
2045 * @fl: The lock to test; also used to hold result
2046 *
2047 * Returns -ERRNO on failure.  Indicates presence of conflicting lock by
2048 * setting conf->fl_type to something other than F_UNLCK.
2049 */
2050int vfs_test_lock(struct file *filp, struct file_lock *fl)
2051{
2052	if (filp->f_op->lock && is_remote_lock(filp))
2053		return filp->f_op->lock(filp, F_GETLK, fl);
2054	posix_test_lock(filp, fl);
2055	return 0;
2056}
2057EXPORT_SYMBOL_GPL(vfs_test_lock);
2058
2059/**
2060 * locks_translate_pid - translate a file_lock's fl_pid number into a namespace
2061 * @fl: The file_lock who's fl_pid should be translated
2062 * @ns: The namespace into which the pid should be translated
2063 *
2064 * Used to tranlate a fl_pid into a namespace virtual pid number
2065 */
2066static pid_t locks_translate_pid(struct file_lock *fl, struct pid_namespace *ns)
2067{
2068	pid_t vnr;
2069	struct pid *pid;
2070
2071	if (IS_OFDLCK(fl))
2072		return -1;
2073	if (IS_REMOTELCK(fl))
2074		return fl->fl_pid;
2075
2076	rcu_read_lock();
2077	pid = find_pid_ns(fl->fl_pid, &init_pid_ns);
2078	vnr = pid_nr_ns(pid, ns);
2079	rcu_read_unlock();
2080	return vnr;
2081}
2082
2083static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
2084{
2085	flock->l_pid = locks_translate_pid(fl, task_active_pid_ns(current));
2086#if BITS_PER_LONG == 32
2087	/*
2088	 * Make sure we can represent the posix lock via
2089	 * legacy 32bit flock.
2090	 */
2091	if (fl->fl_start > OFFT_OFFSET_MAX)
2092		return -EOVERFLOW;
2093	if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
2094		return -EOVERFLOW;
2095#endif
2096	flock->l_start = fl->fl_start;
2097	flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2098		fl->fl_end - fl->fl_start + 1;
2099	flock->l_whence = 0;
2100	flock->l_type = fl->fl_type;
2101	return 0;
2102}
2103
2104#if BITS_PER_LONG == 32
2105static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
2106{
2107	flock->l_pid = locks_translate_pid(fl, task_active_pid_ns(current));
2108	flock->l_start = fl->fl_start;
2109	flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2110		fl->fl_end - fl->fl_start + 1;
2111	flock->l_whence = 0;
2112	flock->l_type = fl->fl_type;
2113}
2114#endif
2115
2116/* Report the first existing lock that would conflict with l.
2117 * This implements the F_GETLK command of fcntl().
2118 */
2119int fcntl_getlk(struct file *filp, unsigned int cmd, struct flock *flock)
2120{
2121	struct file_lock *fl;
 
2122	int error;
2123
2124	fl = locks_alloc_lock();
2125	if (fl == NULL)
2126		return -ENOMEM;
2127	error = -EINVAL;
2128	if (flock->l_type != F_RDLCK && flock->l_type != F_WRLCK)
2129		goto out;
2130
2131	error = flock_to_posix_lock(filp, fl, flock);
2132	if (error)
2133		goto out;
2134
2135	if (cmd == F_OFD_GETLK) {
2136		error = -EINVAL;
2137		if (flock->l_pid != 0)
2138			goto out;
2139
2140		cmd = F_GETLK;
2141		fl->fl_flags |= FL_OFDLCK;
2142		fl->fl_owner = filp;
2143	}
2144
2145	error = vfs_test_lock(filp, fl);
2146	if (error)
2147		goto out;
2148 
2149	flock->l_type = fl->fl_type;
2150	if (fl->fl_type != F_UNLCK) {
2151		error = posix_lock_to_flock(flock, fl);
2152		if (error)
2153			goto out;
2154	}
 
 
 
2155out:
2156	locks_free_lock(fl);
2157	return error;
2158}
2159
2160/**
2161 * vfs_lock_file - file byte range lock
2162 * @filp: The file to apply the lock to
2163 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
2164 * @fl: The lock to be applied
2165 * @conf: Place to return a copy of the conflicting lock, if found.
2166 *
2167 * A caller that doesn't care about the conflicting lock may pass NULL
2168 * as the final argument.
2169 *
2170 * If the filesystem defines a private ->lock() method, then @conf will
2171 * be left unchanged; so a caller that cares should initialize it to
2172 * some acceptable default.
2173 *
2174 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
2175 * locks, the ->lock() interface may return asynchronously, before the lock has
2176 * been granted or denied by the underlying filesystem, if (and only if)
2177 * lm_grant is set. Callers expecting ->lock() to return asynchronously
2178 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
2179 * the request is for a blocking lock. When ->lock() does return asynchronously,
2180 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
2181 * request completes.
2182 * If the request is for non-blocking lock the file system should return
2183 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
2184 * with the result. If the request timed out the callback routine will return a
2185 * nonzero return code and the file system should release the lock. The file
2186 * system is also responsible to keep a corresponding posix lock when it
2187 * grants a lock so the VFS can find out which locks are locally held and do
2188 * the correct lock cleanup when required.
2189 * The underlying filesystem must not drop the kernel lock or call
2190 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
2191 * return code.
2192 */
2193int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
2194{
2195	if (filp->f_op->lock && is_remote_lock(filp))
2196		return filp->f_op->lock(filp, cmd, fl);
2197	else
2198		return posix_lock_file(filp, fl, conf);
2199}
2200EXPORT_SYMBOL_GPL(vfs_lock_file);
2201
2202static int do_lock_file_wait(struct file *filp, unsigned int cmd,
2203			     struct file_lock *fl)
2204{
2205	int error;
2206
2207	error = security_file_lock(filp, fl->fl_type);
2208	if (error)
2209		return error;
2210
2211	for (;;) {
2212		error = vfs_lock_file(filp, cmd, fl, NULL);
2213		if (error != FILE_LOCK_DEFERRED)
2214			break;
2215		error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
2216		if (!error)
2217			continue;
2218
2219		locks_delete_block(fl);
2220		break;
2221	}
2222
2223	return error;
2224}
2225
2226/* Ensure that fl->fl_file has compatible f_mode for F_SETLK calls */
2227static int
2228check_fmode_for_setlk(struct file_lock *fl)
2229{
2230	switch (fl->fl_type) {
2231	case F_RDLCK:
2232		if (!(fl->fl_file->f_mode & FMODE_READ))
2233			return -EBADF;
2234		break;
2235	case F_WRLCK:
2236		if (!(fl->fl_file->f_mode & FMODE_WRITE))
2237			return -EBADF;
2238	}
2239	return 0;
2240}
2241
2242/* Apply the lock described by l to an open file descriptor.
2243 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2244 */
2245int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
2246		struct flock *flock)
2247{
2248	struct file_lock *file_lock = locks_alloc_lock();
2249	struct inode *inode = locks_inode(filp);
 
2250	struct file *f;
2251	int error;
2252
2253	if (file_lock == NULL)
2254		return -ENOLCK;
2255
 
 
 
 
 
 
 
 
 
2256	/* Don't allow mandatory locks on files that may be memory mapped
2257	 * and shared.
2258	 */
2259	if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
2260		error = -EAGAIN;
2261		goto out;
2262	}
2263
2264	error = flock_to_posix_lock(filp, file_lock, flock);
2265	if (error)
2266		goto out;
2267
2268	error = check_fmode_for_setlk(file_lock);
2269	if (error)
2270		goto out;
2271
2272	/*
2273	 * If the cmd is requesting file-private locks, then set the
2274	 * FL_OFDLCK flag and override the owner.
2275	 */
2276	switch (cmd) {
2277	case F_OFD_SETLK:
2278		error = -EINVAL;
2279		if (flock->l_pid != 0)
2280			goto out;
2281
2282		cmd = F_SETLK;
2283		file_lock->fl_flags |= FL_OFDLCK;
2284		file_lock->fl_owner = filp;
2285		break;
2286	case F_OFD_SETLKW:
2287		error = -EINVAL;
2288		if (flock->l_pid != 0)
2289			goto out;
2290
2291		cmd = F_SETLKW;
2292		file_lock->fl_flags |= FL_OFDLCK;
2293		file_lock->fl_owner = filp;
2294		/* Fallthrough */
2295	case F_SETLKW:
2296		file_lock->fl_flags |= FL_SLEEP;
2297	}
2298
2299	error = do_lock_file_wait(filp, cmd, file_lock);
2300
2301	/*
2302	 * Attempt to detect a close/fcntl race and recover by releasing the
2303	 * lock that was just acquired. There is no need to do that when we're
2304	 * unlocking though, or for OFD locks.
2305	 */
2306	if (!error && file_lock->fl_type != F_UNLCK &&
2307	    !(file_lock->fl_flags & FL_OFDLCK)) {
2308		/*
2309		 * We need that spin_lock here - it prevents reordering between
2310		 * update of i_flctx->flc_posix and check for it done in
2311		 * close(). rcu_read_lock() wouldn't do.
2312		 */
2313		spin_lock(&current->files->file_lock);
2314		f = fcheck(fd);
2315		spin_unlock(&current->files->file_lock);
2316		if (f != filp) {
2317			file_lock->fl_type = F_UNLCK;
2318			error = do_lock_file_wait(filp, cmd, file_lock);
2319			WARN_ON_ONCE(error);
2320			error = -EBADF;
2321		}
2322	}
 
2323out:
2324	trace_fcntl_setlk(inode, file_lock, error);
2325	locks_free_lock(file_lock);
2326	return error;
2327}
2328
2329#if BITS_PER_LONG == 32
2330/* Report the first existing lock that would conflict with l.
2331 * This implements the F_GETLK command of fcntl().
2332 */
2333int fcntl_getlk64(struct file *filp, unsigned int cmd, struct flock64 *flock)
2334{
2335	struct file_lock *fl;
 
2336	int error;
2337
2338	fl = locks_alloc_lock();
2339	if (fl == NULL)
2340		return -ENOMEM;
2341
2342	error = -EINVAL;
2343	if (flock->l_type != F_RDLCK && flock->l_type != F_WRLCK)
2344		goto out;
2345
2346	error = flock64_to_posix_lock(filp, fl, flock);
2347	if (error)
2348		goto out;
2349
2350	if (cmd == F_OFD_GETLK) {
2351		error = -EINVAL;
2352		if (flock->l_pid != 0)
2353			goto out;
2354
2355		cmd = F_GETLK64;
2356		fl->fl_flags |= FL_OFDLCK;
2357		fl->fl_owner = filp;
2358	}
2359
2360	error = vfs_test_lock(filp, fl);
2361	if (error)
2362		goto out;
2363
2364	flock->l_type = fl->fl_type;
2365	if (fl->fl_type != F_UNLCK)
2366		posix_lock_to_flock64(flock, fl);
2367
 
 
 
 
2368out:
2369	locks_free_lock(fl);
2370	return error;
2371}
2372
2373/* Apply the lock described by l to an open file descriptor.
2374 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2375 */
2376int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
2377		struct flock64 *flock)
2378{
2379	struct file_lock *file_lock = locks_alloc_lock();
2380	struct inode *inode = locks_inode(filp);
 
2381	struct file *f;
2382	int error;
2383
2384	if (file_lock == NULL)
2385		return -ENOLCK;
2386
 
 
 
 
 
 
 
 
 
2387	/* Don't allow mandatory locks on files that may be memory mapped
2388	 * and shared.
2389	 */
2390	if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
2391		error = -EAGAIN;
2392		goto out;
2393	}
2394
2395	error = flock64_to_posix_lock(filp, file_lock, flock);
2396	if (error)
2397		goto out;
2398
2399	error = check_fmode_for_setlk(file_lock);
2400	if (error)
2401		goto out;
2402
2403	/*
2404	 * If the cmd is requesting file-private locks, then set the
2405	 * FL_OFDLCK flag and override the owner.
2406	 */
2407	switch (cmd) {
2408	case F_OFD_SETLK:
2409		error = -EINVAL;
2410		if (flock->l_pid != 0)
2411			goto out;
2412
2413		cmd = F_SETLK64;
2414		file_lock->fl_flags |= FL_OFDLCK;
2415		file_lock->fl_owner = filp;
2416		break;
2417	case F_OFD_SETLKW:
2418		error = -EINVAL;
2419		if (flock->l_pid != 0)
2420			goto out;
2421
2422		cmd = F_SETLKW64;
2423		file_lock->fl_flags |= FL_OFDLCK;
2424		file_lock->fl_owner = filp;
2425		/* Fallthrough */
2426	case F_SETLKW64:
2427		file_lock->fl_flags |= FL_SLEEP;
2428	}
2429
2430	error = do_lock_file_wait(filp, cmd, file_lock);
2431
2432	/*
2433	 * Attempt to detect a close/fcntl race and recover by releasing the
2434	 * lock that was just acquired. There is no need to do that when we're
2435	 * unlocking though, or for OFD locks.
2436	 */
2437	if (!error && file_lock->fl_type != F_UNLCK &&
2438	    !(file_lock->fl_flags & FL_OFDLCK)) {
2439		/*
2440		 * We need that spin_lock here - it prevents reordering between
2441		 * update of i_flctx->flc_posix and check for it done in
2442		 * close(). rcu_read_lock() wouldn't do.
2443		 */
2444		spin_lock(&current->files->file_lock);
2445		f = fcheck(fd);
2446		spin_unlock(&current->files->file_lock);
2447		if (f != filp) {
2448			file_lock->fl_type = F_UNLCK;
2449			error = do_lock_file_wait(filp, cmd, file_lock);
2450			WARN_ON_ONCE(error);
2451			error = -EBADF;
2452		}
2453	}
 
2454out:
2455	locks_free_lock(file_lock);
2456	return error;
2457}
2458#endif /* BITS_PER_LONG == 32 */
2459
2460/*
2461 * This function is called when the file is being removed
2462 * from the task's fd array.  POSIX locks belonging to this task
2463 * are deleted at this time.
2464 */
2465void locks_remove_posix(struct file *filp, fl_owner_t owner)
2466{
2467	int error;
2468	struct inode *inode = locks_inode(filp);
2469	struct file_lock lock;
2470	struct file_lock_context *ctx;
2471
2472	/*
2473	 * If there are no locks held on this file, we don't need to call
2474	 * posix_lock_file().  Another process could be setting a lock on this
2475	 * file at the same time, but we wouldn't remove that lock anyway.
2476	 */
2477	ctx =  smp_load_acquire(&inode->i_flctx);
2478	if (!ctx || list_empty(&ctx->flc_posix))
2479		return;
2480
2481	lock.fl_type = F_UNLCK;
2482	lock.fl_flags = FL_POSIX | FL_CLOSE;
2483	lock.fl_start = 0;
2484	lock.fl_end = OFFSET_MAX;
2485	lock.fl_owner = owner;
2486	lock.fl_pid = current->tgid;
2487	lock.fl_file = filp;
2488	lock.fl_ops = NULL;
2489	lock.fl_lmops = NULL;
2490
2491	error = vfs_lock_file(filp, F_SETLK, &lock, NULL);
2492
2493	if (lock.fl_ops && lock.fl_ops->fl_release_private)
2494		lock.fl_ops->fl_release_private(&lock);
2495	trace_locks_remove_posix(inode, &lock, error);
2496}
2497
2498EXPORT_SYMBOL(locks_remove_posix);
2499
2500/* The i_flctx must be valid when calling into here */
2501static void
2502locks_remove_flock(struct file *filp, struct file_lock_context *flctx)
2503{
2504	struct file_lock fl = {
2505		.fl_owner = filp,
2506		.fl_pid = current->tgid,
2507		.fl_file = filp,
2508		.fl_flags = FL_FLOCK | FL_CLOSE,
2509		.fl_type = F_UNLCK,
2510		.fl_end = OFFSET_MAX,
2511	};
2512	struct inode *inode = locks_inode(filp);
2513
2514	if (list_empty(&flctx->flc_flock))
2515		return;
2516
2517	if (filp->f_op->flock && is_remote_lock(filp))
2518		filp->f_op->flock(filp, F_SETLKW, &fl);
2519	else
2520		flock_lock_inode(inode, &fl);
2521
2522	if (fl.fl_ops && fl.fl_ops->fl_release_private)
2523		fl.fl_ops->fl_release_private(&fl);
2524}
2525
2526/* The i_flctx must be valid when calling into here */
2527static void
2528locks_remove_lease(struct file *filp, struct file_lock_context *ctx)
2529{
2530	struct file_lock *fl, *tmp;
2531	LIST_HEAD(dispose);
2532
2533	if (list_empty(&ctx->flc_lease))
2534		return;
2535
2536	percpu_down_read_preempt_disable(&file_rwsem);
2537	spin_lock(&ctx->flc_lock);
2538	list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list)
2539		if (filp == fl->fl_file)
2540			lease_modify(fl, F_UNLCK, &dispose);
2541	spin_unlock(&ctx->flc_lock);
2542	percpu_up_read_preempt_enable(&file_rwsem);
2543
2544	locks_dispose_list(&dispose);
2545}
2546
2547/*
2548 * This function is called on the last close of an open file.
2549 */
2550void locks_remove_file(struct file *filp)
2551{
2552	struct file_lock_context *ctx;
 
 
2553
2554	ctx = smp_load_acquire(&locks_inode(filp)->i_flctx);
2555	if (!ctx)
2556		return;
2557
2558	/* remove any OFD locks */
2559	locks_remove_posix(filp, filp);
2560
2561	/* remove flock locks */
2562	locks_remove_flock(filp, ctx);
 
 
 
 
 
 
 
2563
2564	/* remove any leases */
2565	locks_remove_lease(filp, ctx);
2566
2567	spin_lock(&ctx->flc_lock);
2568	locks_check_ctx_file_list(filp, &ctx->flc_posix, "POSIX");
2569	locks_check_ctx_file_list(filp, &ctx->flc_flock, "FLOCK");
2570	locks_check_ctx_file_list(filp, &ctx->flc_lease, "LEASE");
2571	spin_unlock(&ctx->flc_lock);
 
 
 
 
 
 
 
 
 
 
 
2572}
2573
2574/**
2575 *	posix_unblock_lock - stop waiting for a file lock
 
2576 *	@waiter: the lock which was waiting
2577 *
2578 *	lockd needs to block waiting for locks.
2579 */
2580int
2581posix_unblock_lock(struct file_lock *waiter)
2582{
2583	int status = 0;
2584
2585	spin_lock(&blocked_lock_lock);
2586	if (waiter->fl_next)
2587		__locks_delete_block(waiter);
2588	else
2589		status = -ENOENT;
2590	spin_unlock(&blocked_lock_lock);
2591	return status;
2592}
 
2593EXPORT_SYMBOL(posix_unblock_lock);
2594
2595/**
2596 * vfs_cancel_lock - file byte range unblock lock
2597 * @filp: The file to apply the unblock to
2598 * @fl: The lock to be unblocked
2599 *
2600 * Used by lock managers to cancel blocked requests
2601 */
2602int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2603{
2604	if (filp->f_op->lock && is_remote_lock(filp))
2605		return filp->f_op->lock(filp, F_CANCELLK, fl);
2606	return 0;
2607}
2608
2609EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2610
2611#ifdef CONFIG_PROC_FS
2612#include <linux/proc_fs.h>
2613#include <linux/seq_file.h>
2614
2615struct locks_iterator {
2616	int	li_cpu;
2617	loff_t	li_pos;
2618};
2619
2620static void lock_get_status(struct seq_file *f, struct file_lock *fl,
2621			    loff_t id, char *pfx)
2622{
2623	struct inode *inode = NULL;
2624	unsigned int fl_pid;
2625	struct pid_namespace *proc_pidns = file_inode(f->file)->i_sb->s_fs_info;
2626
2627	fl_pid = locks_translate_pid(fl, proc_pidns);
2628	/*
2629	 * If there isn't a fl_pid don't display who is waiting on
2630	 * the lock if we are called from locks_show, or if we are
2631	 * called from __show_fd_info - skip lock entirely
2632	 */
2633	if (fl_pid == 0)
2634		return;
2635
2636	if (fl->fl_file != NULL)
2637		inode = locks_inode(fl->fl_file);
2638
2639	seq_printf(f, "%lld:%s ", id, pfx);
2640	if (IS_POSIX(fl)) {
2641		if (fl->fl_flags & FL_ACCESS)
2642			seq_puts(f, "ACCESS");
2643		else if (IS_OFDLCK(fl))
2644			seq_puts(f, "OFDLCK");
2645		else
2646			seq_puts(f, "POSIX ");
2647
2648		seq_printf(f, " %s ",
2649			     (inode == NULL) ? "*NOINODE*" :
2650			     mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
2651	} else if (IS_FLOCK(fl)) {
2652		if (fl->fl_type & LOCK_MAND) {
2653			seq_puts(f, "FLOCK  MSNFS     ");
2654		} else {
2655			seq_puts(f, "FLOCK  ADVISORY  ");
2656		}
2657	} else if (IS_LEASE(fl)) {
2658		if (fl->fl_flags & FL_DELEG)
2659			seq_puts(f, "DELEG  ");
2660		else
2661			seq_puts(f, "LEASE  ");
2662
2663		if (lease_breaking(fl))
2664			seq_puts(f, "BREAKING  ");
2665		else if (fl->fl_file)
2666			seq_puts(f, "ACTIVE    ");
2667		else
2668			seq_puts(f, "BREAKER   ");
2669	} else {
2670		seq_puts(f, "UNKNOWN UNKNOWN  ");
2671	}
2672	if (fl->fl_type & LOCK_MAND) {
2673		seq_printf(f, "%s ",
2674			       (fl->fl_type & LOCK_READ)
2675			       ? (fl->fl_type & LOCK_WRITE) ? "RW   " : "READ "
2676			       : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2677	} else {
2678		seq_printf(f, "%s ",
2679			       (lease_breaking(fl))
2680			       ? (fl->fl_type == F_UNLCK) ? "UNLCK" : "READ "
2681			       : (fl->fl_type == F_WRLCK) ? "WRITE" : "READ ");
2682	}
2683	if (inode) {
2684		/* userspace relies on this representation of dev_t */
 
 
 
 
2685		seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
2686				MAJOR(inode->i_sb->s_dev),
2687				MINOR(inode->i_sb->s_dev), inode->i_ino);
 
2688	} else {
2689		seq_printf(f, "%d <none>:0 ", fl_pid);
2690	}
2691	if (IS_POSIX(fl)) {
2692		if (fl->fl_end == OFFSET_MAX)
2693			seq_printf(f, "%Ld EOF\n", fl->fl_start);
2694		else
2695			seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
2696	} else {
2697		seq_puts(f, "0 EOF\n");
2698	}
2699}
2700
2701static int locks_show(struct seq_file *f, void *v)
2702{
2703	struct locks_iterator *iter = f->private;
2704	struct file_lock *fl, *bfl;
2705	struct pid_namespace *proc_pidns = file_inode(f->file)->i_sb->s_fs_info;
2706
2707	fl = hlist_entry(v, struct file_lock, fl_link);
2708
2709	if (locks_translate_pid(fl, proc_pidns) == 0)
2710		return 0;
2711
2712	lock_get_status(f, fl, iter->li_pos, "");
2713
2714	list_for_each_entry(bfl, &fl->fl_block, fl_block)
2715		lock_get_status(f, bfl, iter->li_pos, " ->");
2716
2717	return 0;
2718}
2719
2720static void __show_fd_locks(struct seq_file *f,
2721			struct list_head *head, int *id,
2722			struct file *filp, struct files_struct *files)
2723{
2724	struct file_lock *fl;
2725
2726	list_for_each_entry(fl, head, fl_list) {
2727
2728		if (filp != fl->fl_file)
2729			continue;
2730		if (fl->fl_owner != files &&
2731		    fl->fl_owner != filp)
2732			continue;
2733
2734		(*id)++;
2735		seq_puts(f, "lock:\t");
2736		lock_get_status(f, fl, *id, "");
2737	}
2738}
2739
2740void show_fd_locks(struct seq_file *f,
2741		  struct file *filp, struct files_struct *files)
2742{
2743	struct inode *inode = locks_inode(filp);
2744	struct file_lock_context *ctx;
2745	int id = 0;
2746
2747	ctx = smp_load_acquire(&inode->i_flctx);
2748	if (!ctx)
2749		return;
2750
2751	spin_lock(&ctx->flc_lock);
2752	__show_fd_locks(f, &ctx->flc_flock, &id, filp, files);
2753	__show_fd_locks(f, &ctx->flc_posix, &id, filp, files);
2754	__show_fd_locks(f, &ctx->flc_lease, &id, filp, files);
2755	spin_unlock(&ctx->flc_lock);
2756}
2757
2758static void *locks_start(struct seq_file *f, loff_t *pos)
2759	__acquires(&blocked_lock_lock)
2760{
2761	struct locks_iterator *iter = f->private;
2762
2763	iter->li_pos = *pos + 1;
2764	percpu_down_write(&file_rwsem);
2765	spin_lock(&blocked_lock_lock);
2766	return seq_hlist_start_percpu(&file_lock_list.hlist, &iter->li_cpu, *pos);
2767}
2768
2769static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2770{
2771	struct locks_iterator *iter = f->private;
2772
2773	++iter->li_pos;
2774	return seq_hlist_next_percpu(v, &file_lock_list.hlist, &iter->li_cpu, pos);
2775}
2776
2777static void locks_stop(struct seq_file *f, void *v)
2778	__releases(&blocked_lock_lock)
2779{
2780	spin_unlock(&blocked_lock_lock);
2781	percpu_up_write(&file_rwsem);
2782}
2783
2784static const struct seq_operations locks_seq_operations = {
2785	.start	= locks_start,
2786	.next	= locks_next,
2787	.stop	= locks_stop,
2788	.show	= locks_show,
2789};
2790
2791static int locks_open(struct inode *inode, struct file *filp)
2792{
2793	return seq_open_private(filp, &locks_seq_operations,
2794					sizeof(struct locks_iterator));
2795}
2796
2797static const struct file_operations proc_locks_operations = {
2798	.open		= locks_open,
2799	.read		= seq_read,
2800	.llseek		= seq_lseek,
2801	.release	= seq_release_private,
2802};
2803
2804static int __init proc_locks_init(void)
2805{
2806	proc_create("locks", 0, NULL, &proc_locks_operations);
2807	return 0;
2808}
2809fs_initcall(proc_locks_init);
2810#endif
2811
2812static int __init filelock_init(void)
 
 
 
 
 
 
 
 
 
 
 
 
 
2813{
2814	int i;
2815
2816	flctx_cache = kmem_cache_create("file_lock_ctx",
2817			sizeof(struct file_lock_context), 0, SLAB_PANIC, NULL);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2818
2819	filelock_cache = kmem_cache_create("file_lock_cache",
2820			sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2821
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2822
2823	for_each_possible_cpu(i) {
2824		struct file_lock_list_struct *fll = per_cpu_ptr(&file_lock_list, i);
2825
2826		spin_lock_init(&fll->lock);
2827		INIT_HLIST_HEAD(&fll->hlist);
2828	}
 
2829
2830	return 0;
2831}
2832
2833core_initcall(filelock_init);
v3.1
   1/*
   2 *  linux/fs/locks.c
   3 *
   4 *  Provide support for fcntl()'s F_GETLK, F_SETLK, and F_SETLKW calls.
   5 *  Doug Evans (dje@spiff.uucp), August 07, 1992
   6 *
   7 *  Deadlock detection added.
   8 *  FIXME: one thing isn't handled yet:
   9 *	- mandatory locks (requires lots of changes elsewhere)
  10 *  Kelly Carmichael (kelly@[142.24.8.65]), September 17, 1994.
  11 *
  12 *  Miscellaneous edits, and a total rewrite of posix_lock_file() code.
  13 *  Kai Petzke (wpp@marie.physik.tu-berlin.de), 1994
  14 *  
  15 *  Converted file_lock_table to a linked list from an array, which eliminates
  16 *  the limits on how many active file locks are open.
  17 *  Chad Page (pageone@netcom.com), November 27, 1994
  18 * 
  19 *  Removed dependency on file descriptors. dup()'ed file descriptors now
  20 *  get the same locks as the original file descriptors, and a close() on
  21 *  any file descriptor removes ALL the locks on the file for the current
  22 *  process. Since locks still depend on the process id, locks are inherited
  23 *  after an exec() but not after a fork(). This agrees with POSIX, and both
  24 *  BSD and SVR4 practice.
  25 *  Andy Walker (andy@lysaker.kvaerner.no), February 14, 1995
  26 *
  27 *  Scrapped free list which is redundant now that we allocate locks
  28 *  dynamically with kmalloc()/kfree().
  29 *  Andy Walker (andy@lysaker.kvaerner.no), February 21, 1995
  30 *
  31 *  Implemented two lock personalities - FL_FLOCK and FL_POSIX.
  32 *
  33 *  FL_POSIX locks are created with calls to fcntl() and lockf() through the
  34 *  fcntl() system call. They have the semantics described above.
  35 *
  36 *  FL_FLOCK locks are created with calls to flock(), through the flock()
  37 *  system call, which is new. Old C libraries implement flock() via fcntl()
  38 *  and will continue to use the old, broken implementation.
  39 *
  40 *  FL_FLOCK locks follow the 4.4 BSD flock() semantics. They are associated
  41 *  with a file pointer (filp). As a result they can be shared by a parent
  42 *  process and its children after a fork(). They are removed when the last
  43 *  file descriptor referring to the file pointer is closed (unless explicitly
  44 *  unlocked). 
  45 *
  46 *  FL_FLOCK locks never deadlock, an existing lock is always removed before
  47 *  upgrading from shared to exclusive (or vice versa). When this happens
  48 *  any processes blocked by the current lock are woken up and allowed to
  49 *  run before the new lock is applied.
  50 *  Andy Walker (andy@lysaker.kvaerner.no), June 09, 1995
  51 *
  52 *  Removed some race conditions in flock_lock_file(), marked other possible
  53 *  races. Just grep for FIXME to see them. 
  54 *  Dmitry Gorodchanin (pgmdsg@ibi.com), February 09, 1996.
  55 *
  56 *  Addressed Dmitry's concerns. Deadlock checking no longer recursive.
  57 *  Lock allocation changed to GFP_ATOMIC as we can't afford to sleep
  58 *  once we've checked for blocking and deadlocking.
  59 *  Andy Walker (andy@lysaker.kvaerner.no), April 03, 1996.
  60 *
  61 *  Initial implementation of mandatory locks. SunOS turned out to be
  62 *  a rotten model, so I implemented the "obvious" semantics.
  63 *  See 'Documentation/mandatory.txt' for details.
  64 *  Andy Walker (andy@lysaker.kvaerner.no), April 06, 1996.
  65 *
  66 *  Don't allow mandatory locks on mmap()'ed files. Added simple functions to
  67 *  check if a file has mandatory locks, used by mmap(), open() and creat() to
  68 *  see if system call should be rejected. Ref. HP-UX/SunOS/Solaris Reference
  69 *  Manual, Section 2.
  70 *  Andy Walker (andy@lysaker.kvaerner.no), April 09, 1996.
  71 *
  72 *  Tidied up block list handling. Added '/proc/locks' interface.
  73 *  Andy Walker (andy@lysaker.kvaerner.no), April 24, 1996.
  74 *
  75 *  Fixed deadlock condition for pathological code that mixes calls to
  76 *  flock() and fcntl().
  77 *  Andy Walker (andy@lysaker.kvaerner.no), April 29, 1996.
  78 *
  79 *  Allow only one type of locking scheme (FL_POSIX or FL_FLOCK) to be in use
  80 *  for a given file at a time. Changed the CONFIG_LOCK_MANDATORY scheme to
  81 *  guarantee sensible behaviour in the case where file system modules might
  82 *  be compiled with different options than the kernel itself.
  83 *  Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
  84 *
  85 *  Added a couple of missing wake_up() calls. Thanks to Thomas Meckel
  86 *  (Thomas.Meckel@mni.fh-giessen.de) for spotting this.
  87 *  Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
  88 *
  89 *  Changed FL_POSIX locks to use the block list in the same way as FL_FLOCK
  90 *  locks. Changed process synchronisation to avoid dereferencing locks that
  91 *  have already been freed.
  92 *  Andy Walker (andy@lysaker.kvaerner.no), Sep 21, 1996.
  93 *
  94 *  Made the block list a circular list to minimise searching in the list.
  95 *  Andy Walker (andy@lysaker.kvaerner.no), Sep 25, 1996.
  96 *
  97 *  Made mandatory locking a mount option. Default is not to allow mandatory
  98 *  locking.
  99 *  Andy Walker (andy@lysaker.kvaerner.no), Oct 04, 1996.
 100 *
 101 *  Some adaptations for NFS support.
 102 *  Olaf Kirch (okir@monad.swb.de), Dec 1996,
 103 *
 104 *  Fixed /proc/locks interface so that we can't overrun the buffer we are handed.
 105 *  Andy Walker (andy@lysaker.kvaerner.no), May 12, 1997.
 106 *
 107 *  Use slab allocator instead of kmalloc/kfree.
 108 *  Use generic list implementation from <linux/list.h>.
 109 *  Sped up posix_locks_deadlock by only considering blocked locks.
 110 *  Matthew Wilcox <willy@debian.org>, March, 2000.
 111 *
 112 *  Leases and LOCK_MAND
 113 *  Matthew Wilcox <willy@debian.org>, June, 2000.
 114 *  Stephen Rothwell <sfr@canb.auug.org.au>, June, 2000.
 115 */
 116
 117#include <linux/capability.h>
 118#include <linux/file.h>
 119#include <linux/fdtable.h>
 120#include <linux/fs.h>
 121#include <linux/init.h>
 122#include <linux/module.h>
 123#include <linux/security.h>
 124#include <linux/slab.h>
 125#include <linux/syscalls.h>
 126#include <linux/time.h>
 127#include <linux/rcupdate.h>
 128#include <linux/pid_namespace.h>
 
 
 129
 130#include <asm/uaccess.h>
 
 
 
 131
 132#define IS_POSIX(fl)	(fl->fl_flags & FL_POSIX)
 133#define IS_FLOCK(fl)	(fl->fl_flags & FL_FLOCK)
 134#define IS_LEASE(fl)	(fl->fl_flags & FL_LEASE)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 135
 136int leases_enable = 1;
 137int lease_break_time = 45;
 138
 139#define for_each_lock(inode, lockp) \
 140	for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
 
 
 
 
 
 
 
 
 
 
 
 
 141
 142static LIST_HEAD(file_lock_list);
 143static LIST_HEAD(blocked_list);
 144static DEFINE_SPINLOCK(file_lock_lock);
 
 
 
 
 
 
 
 
 
 
 145
 146/*
 147 * Protects the two list heads above, plus the inode->i_flock list
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 148 */
 149void lock_flocks(void)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 150{
 151	spin_lock(&file_lock_lock);
 
 
 
 
 
 
 
 
 
 
 
 152}
 153EXPORT_SYMBOL_GPL(lock_flocks);
 154
 155void unlock_flocks(void)
 
 
 156{
 157	spin_unlock(&file_lock_lock);
 
 
 
 
 
 
 
 
 
 158}
 159EXPORT_SYMBOL_GPL(unlock_flocks);
 160
 161static struct kmem_cache *filelock_cache __read_mostly;
 
 
 
 
 
 
 
 
 
 162
 163static void locks_init_lock_heads(struct file_lock *fl)
 164{
 165	INIT_LIST_HEAD(&fl->fl_link);
 
 166	INIT_LIST_HEAD(&fl->fl_block);
 167	init_waitqueue_head(&fl->fl_wait);
 168}
 169
 170/* Allocate an empty lock structure. */
 171struct file_lock *locks_alloc_lock(void)
 172{
 173	struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
 174
 175	if (fl)
 176		locks_init_lock_heads(fl);
 177
 178	return fl;
 179}
 180EXPORT_SYMBOL_GPL(locks_alloc_lock);
 181
 182void locks_release_private(struct file_lock *fl)
 183{
 184	if (fl->fl_ops) {
 185		if (fl->fl_ops->fl_release_private)
 186			fl->fl_ops->fl_release_private(fl);
 187		fl->fl_ops = NULL;
 188	}
 
 189	if (fl->fl_lmops) {
 190		if (fl->fl_lmops->lm_release_private)
 191			fl->fl_lmops->lm_release_private(fl);
 
 
 192		fl->fl_lmops = NULL;
 193	}
 194
 195}
 196EXPORT_SYMBOL_GPL(locks_release_private);
 197
 198/* Free a lock which is not in use. */
 199void locks_free_lock(struct file_lock *fl)
 200{
 201	BUG_ON(waitqueue_active(&fl->fl_wait));
 
 202	BUG_ON(!list_empty(&fl->fl_block));
 203	BUG_ON(!list_empty(&fl->fl_link));
 204
 205	locks_release_private(fl);
 206	kmem_cache_free(filelock_cache, fl);
 207}
 208EXPORT_SYMBOL(locks_free_lock);
 209
 
 
 
 
 
 
 
 
 
 
 
 
 210void locks_init_lock(struct file_lock *fl)
 211{
 212	memset(fl, 0, sizeof(struct file_lock));
 213	locks_init_lock_heads(fl);
 214}
 215
 216EXPORT_SYMBOL(locks_init_lock);
 217
 218static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
 219{
 220	if (fl->fl_ops) {
 221		if (fl->fl_ops->fl_copy_lock)
 222			fl->fl_ops->fl_copy_lock(new, fl);
 223		new->fl_ops = fl->fl_ops;
 224	}
 225	if (fl->fl_lmops)
 226		new->fl_lmops = fl->fl_lmops;
 227}
 228
 229/*
 230 * Initialize a new lock from an existing file_lock structure.
 231 */
 232void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
 233{
 234	new->fl_owner = fl->fl_owner;
 235	new->fl_pid = fl->fl_pid;
 236	new->fl_file = NULL;
 237	new->fl_flags = fl->fl_flags;
 238	new->fl_type = fl->fl_type;
 239	new->fl_start = fl->fl_start;
 240	new->fl_end = fl->fl_end;
 
 241	new->fl_ops = NULL;
 242	new->fl_lmops = NULL;
 
 
 
 
 243}
 244EXPORT_SYMBOL(__locks_copy_lock);
 245
 246void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
 247{
 248	locks_release_private(new);
 
 
 
 249
 250	__locks_copy_lock(new, fl);
 251	new->fl_file = fl->fl_file;
 252	new->fl_ops = fl->fl_ops;
 253	new->fl_lmops = fl->fl_lmops;
 254
 255	locks_copy_private(new, fl);
 
 
 
 256}
 257
 258EXPORT_SYMBOL(locks_copy_lock);
 259
 260static inline int flock_translate_cmd(int cmd) {
 261	if (cmd & LOCK_MAND)
 262		return cmd & (LOCK_MAND | LOCK_RW);
 263	switch (cmd) {
 264	case LOCK_SH:
 265		return F_RDLCK;
 266	case LOCK_EX:
 267		return F_WRLCK;
 268	case LOCK_UN:
 269		return F_UNLCK;
 270	}
 271	return -EINVAL;
 272}
 273
 274/* Fill in a file_lock structure with an appropriate FLOCK lock. */
 275static int flock_make_lock(struct file *filp, struct file_lock **lock,
 276		unsigned int cmd)
 277{
 278	struct file_lock *fl;
 279	int type = flock_translate_cmd(cmd);
 
 280	if (type < 0)
 281		return type;
 282	
 283	fl = locks_alloc_lock();
 284	if (fl == NULL)
 285		return -ENOMEM;
 286
 287	fl->fl_file = filp;
 
 288	fl->fl_pid = current->tgid;
 289	fl->fl_flags = FL_FLOCK;
 290	fl->fl_type = type;
 291	fl->fl_end = OFFSET_MAX;
 292	
 293	*lock = fl;
 294	return 0;
 295}
 296
 297static int assign_type(struct file_lock *fl, int type)
 298{
 299	switch (type) {
 300	case F_RDLCK:
 301	case F_WRLCK:
 302	case F_UNLCK:
 303		fl->fl_type = type;
 304		break;
 305	default:
 306		return -EINVAL;
 307	}
 308	return 0;
 309}
 310
 311/* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
 312 * style lock.
 313 */
 314static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
 315			       struct flock *l)
 316{
 317	off_t start, end;
 318
 319	switch (l->l_whence) {
 320	case SEEK_SET:
 321		start = 0;
 322		break;
 323	case SEEK_CUR:
 324		start = filp->f_pos;
 325		break;
 326	case SEEK_END:
 327		start = i_size_read(filp->f_path.dentry->d_inode);
 328		break;
 329	default:
 330		return -EINVAL;
 331	}
 
 
 
 
 
 332
 333	/* POSIX-1996 leaves the case l->l_len < 0 undefined;
 334	   POSIX-2001 defines it. */
 335	start += l->l_start;
 336	if (start < 0)
 337		return -EINVAL;
 338	fl->fl_end = OFFSET_MAX;
 339	if (l->l_len > 0) {
 340		end = start + l->l_len - 1;
 341		fl->fl_end = end;
 
 
 342	} else if (l->l_len < 0) {
 343		end = start - 1;
 344		fl->fl_end = end;
 345		start += l->l_len;
 346		if (start < 0)
 347			return -EINVAL;
 348	}
 349	fl->fl_start = start;	/* we record the absolute position */
 350	if (fl->fl_end < fl->fl_start)
 351		return -EOVERFLOW;
 352	
 353	fl->fl_owner = current->files;
 354	fl->fl_pid = current->tgid;
 355	fl->fl_file = filp;
 356	fl->fl_flags = FL_POSIX;
 357	fl->fl_ops = NULL;
 358	fl->fl_lmops = NULL;
 359
 360	return assign_type(fl, l->l_type);
 361}
 362
 363#if BITS_PER_LONG == 32
 364static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
 365				 struct flock64 *l)
 
 
 366{
 367	loff_t start;
 368
 369	switch (l->l_whence) {
 370	case SEEK_SET:
 371		start = 0;
 372		break;
 373	case SEEK_CUR:
 374		start = filp->f_pos;
 375		break;
 376	case SEEK_END:
 377		start = i_size_read(filp->f_path.dentry->d_inode);
 378		break;
 379	default:
 380		return -EINVAL;
 381	}
 382
 383	start += l->l_start;
 384	if (start < 0)
 385		return -EINVAL;
 386	fl->fl_end = OFFSET_MAX;
 387	if (l->l_len > 0) {
 388		fl->fl_end = start + l->l_len - 1;
 389	} else if (l->l_len < 0) {
 390		fl->fl_end = start - 1;
 391		start += l->l_len;
 392		if (start < 0)
 393			return -EINVAL;
 394	}
 395	fl->fl_start = start;	/* we record the absolute position */
 396	if (fl->fl_end < fl->fl_start)
 397		return -EOVERFLOW;
 398	
 399	fl->fl_owner = current->files;
 400	fl->fl_pid = current->tgid;
 401	fl->fl_file = filp;
 402	fl->fl_flags = FL_POSIX;
 403	fl->fl_ops = NULL;
 404	fl->fl_lmops = NULL;
 405
 406	return assign_type(fl, l->l_type);
 407}
 408#endif
 409
 410/* default lease lock manager operations */
 411static void lease_break_callback(struct file_lock *fl)
 
 412{
 413	kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
 
 414}
 415
 416static void lease_release_private_callback(struct file_lock *fl)
 
 417{
 418	if (!fl->fl_file)
 419		return;
 
 
 
 
 
 
 
 
 420
 421	f_delown(fl->fl_file);
 422	fl->fl_file->f_owner.signum = 0;
 423}
 424
 425static const struct lock_manager_operations lease_manager_ops = {
 426	.lm_break = lease_break_callback,
 427	.lm_release_private = lease_release_private_callback,
 428	.lm_change = lease_modify,
 
 429};
 430
 431/*
 432 * Initialize a lease, use the default lock manager operations
 433 */
 434static int lease_init(struct file *filp, int type, struct file_lock *fl)
 435 {
 436	if (assign_type(fl, type) != 0)
 437		return -EINVAL;
 438
 439	fl->fl_owner = current->files;
 440	fl->fl_pid = current->tgid;
 441
 442	fl->fl_file = filp;
 443	fl->fl_flags = FL_LEASE;
 444	fl->fl_start = 0;
 445	fl->fl_end = OFFSET_MAX;
 446	fl->fl_ops = NULL;
 447	fl->fl_lmops = &lease_manager_ops;
 448	return 0;
 449}
 450
 451/* Allocate a file_lock initialised to this type of lease */
 452static struct file_lock *lease_alloc(struct file *filp, int type)
 453{
 454	struct file_lock *fl = locks_alloc_lock();
 455	int error = -ENOMEM;
 456
 457	if (fl == NULL)
 458		return ERR_PTR(error);
 459
 460	error = lease_init(filp, type, fl);
 461	if (error) {
 462		locks_free_lock(fl);
 463		return ERR_PTR(error);
 464	}
 465	return fl;
 466}
 467
 468/* Check if two locks overlap each other.
 469 */
 470static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
 471{
 472	return ((fl1->fl_end >= fl2->fl_start) &&
 473		(fl2->fl_end >= fl1->fl_start));
 474}
 475
 476/*
 477 * Check whether two locks have the same owner.
 478 */
 479static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
 480{
 481	if (fl1->fl_lmops && fl1->fl_lmops->lm_compare_owner)
 482		return fl2->fl_lmops == fl1->fl_lmops &&
 483			fl1->fl_lmops->lm_compare_owner(fl1, fl2);
 484	return fl1->fl_owner == fl2->fl_owner;
 485}
 486
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 487/* Remove waiter from blocker's block list.
 488 * When blocker ends up pointing to itself then the list is empty.
 
 
 489 */
 490static void __locks_delete_block(struct file_lock *waiter)
 491{
 
 492	list_del_init(&waiter->fl_block);
 493	list_del_init(&waiter->fl_link);
 494	waiter->fl_next = NULL;
 495}
 496
 497/*
 498 */
 499static void locks_delete_block(struct file_lock *waiter)
 500{
 501	lock_flocks();
 502	__locks_delete_block(waiter);
 503	unlock_flocks();
 504}
 505
 506/* Insert waiter into blocker's block list.
 507 * We use a circular list so that processes can be easily woken up in
 508 * the order they blocked. The documentation doesn't require this but
 509 * it seems like the reasonable thing to do.
 
 
 
 
 
 510 */
 511static void locks_insert_block(struct file_lock *blocker, 
 512			       struct file_lock *waiter)
 513{
 514	BUG_ON(!list_empty(&waiter->fl_block));
 
 515	list_add_tail(&waiter->fl_block, &blocker->fl_block);
 516	waiter->fl_next = blocker;
 517	if (IS_POSIX(blocker))
 518		list_add(&waiter->fl_link, &blocked_list);
 
 
 
 
 
 
 
 
 519}
 520
 521/* Wake up processes blocked waiting for blocker.
 522 * If told to wait then schedule the processes until the block list
 523 * is empty, otherwise empty the block list ourselves.
 
 524 */
 525static void locks_wake_up_blocks(struct file_lock *blocker)
 526{
 
 
 
 
 
 
 
 
 
 
 
 527	while (!list_empty(&blocker->fl_block)) {
 528		struct file_lock *waiter;
 529
 530		waiter = list_first_entry(&blocker->fl_block,
 531				struct file_lock, fl_block);
 532		__locks_delete_block(waiter);
 533		if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
 534			waiter->fl_lmops->lm_notify(waiter);
 535		else
 536			wake_up(&waiter->fl_wait);
 537	}
 
 538}
 539
 540/* Insert file lock fl into an inode's lock list at the position indicated
 541 * by pos. At the same time add the lock to the global file lock list.
 542 */
 543static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
 544{
 545	list_add(&fl->fl_link, &file_lock_list);
 
 
 546
 547	fl->fl_nspid = get_pid(task_tgid(current));
 548
 549	/* insert into file's list */
 550	fl->fl_next = *pos;
 551	*pos = fl;
 
 552}
 553
 554/*
 555 * Delete a lock and then free it.
 556 * Wake up processes that are blocked waiting for this lock,
 557 * notify the FS that the lock has been cleared and
 558 * finally free the lock.
 559 */
 560static void locks_delete_lock(struct file_lock **thisfl_p)
 561{
 562	struct file_lock *fl = *thisfl_p;
 563
 564	*thisfl_p = fl->fl_next;
 565	fl->fl_next = NULL;
 566	list_del_init(&fl->fl_link);
 567
 568	fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
 569	if (fl->fl_fasync != NULL) {
 570		printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
 571		fl->fl_fasync = NULL;
 572	}
 573
 574	if (fl->fl_nspid) {
 575		put_pid(fl->fl_nspid);
 576		fl->fl_nspid = NULL;
 577	}
 578
 579	locks_wake_up_blocks(fl);
 580	locks_free_lock(fl);
 581}
 582
 583/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
 584 * checks for shared/exclusive status of overlapping locks.
 585 */
 586static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
 587{
 588	if (sys_fl->fl_type == F_WRLCK)
 589		return 1;
 590	if (caller_fl->fl_type == F_WRLCK)
 591		return 1;
 592	return 0;
 593}
 594
 595/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
 596 * checking before calling the locks_conflict().
 597 */
 598static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
 599{
 600	/* POSIX locks owned by the same process do not conflict with
 601	 * each other.
 602	 */
 603	if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
 604		return (0);
 605
 606	/* Check whether they overlap */
 607	if (!locks_overlap(caller_fl, sys_fl))
 608		return 0;
 609
 610	return (locks_conflict(caller_fl, sys_fl));
 611}
 612
 613/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
 614 * checking before calling the locks_conflict().
 615 */
 616static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
 617{
 618	/* FLOCK locks referring to the same filp do not conflict with
 619	 * each other.
 620	 */
 621	if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
 622		return (0);
 623	if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
 624		return 0;
 625
 626	return (locks_conflict(caller_fl, sys_fl));
 627}
 628
 629void
 630posix_test_lock(struct file *filp, struct file_lock *fl)
 631{
 632	struct file_lock *cfl;
 
 
 633
 634	lock_flocks();
 635	for (cfl = filp->f_path.dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
 636		if (!IS_POSIX(cfl))
 637			continue;
 638		if (posix_locks_conflict(fl, cfl))
 639			break;
 
 
 
 
 
 
 640	}
 641	if (cfl) {
 642		__locks_copy_lock(fl, cfl);
 643		if (cfl->fl_nspid)
 644			fl->fl_pid = pid_vnr(cfl->fl_nspid);
 645	} else
 646		fl->fl_type = F_UNLCK;
 647	unlock_flocks();
 648	return;
 649}
 650EXPORT_SYMBOL(posix_test_lock);
 651
 652/*
 653 * Deadlock detection:
 654 *
 655 * We attempt to detect deadlocks that are due purely to posix file
 656 * locks.
 657 *
 658 * We assume that a task can be waiting for at most one lock at a time.
 659 * So for any acquired lock, the process holding that lock may be
 660 * waiting on at most one other lock.  That lock in turns may be held by
 661 * someone waiting for at most one other lock.  Given a requested lock
 662 * caller_fl which is about to wait for a conflicting lock block_fl, we
 663 * follow this chain of waiters to ensure we are not about to create a
 664 * cycle.
 665 *
 666 * Since we do this before we ever put a process to sleep on a lock, we
 667 * are ensured that there is never a cycle; that is what guarantees that
 668 * the while() loop in posix_locks_deadlock() eventually completes.
 669 *
 670 * Note: the above assumption may not be true when handling lock
 671 * requests from a broken NFS client. It may also fail in the presence
 672 * of tasks (such as posix threads) sharing the same open file table.
 
 673 *
 674 * To handle those cases, we just bail out after a few iterations.
 
 
 
 
 
 
 
 675 */
 676
 677#define MAX_DEADLK_ITERATIONS 10
 678
 679/* Find a lock that the owner of the given block_fl is blocking on. */
 680static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
 681{
 682	struct file_lock *fl;
 683
 684	list_for_each_entry(fl, &blocked_list, fl_link) {
 685		if (posix_same_owner(fl, block_fl))
 686			return fl->fl_next;
 687	}
 688	return NULL;
 689}
 690
 
 691static int posix_locks_deadlock(struct file_lock *caller_fl,
 692				struct file_lock *block_fl)
 693{
 694	int i = 0;
 695
 
 
 
 
 
 
 
 
 
 696	while ((block_fl = what_owner_is_waiting_for(block_fl))) {
 697		if (i++ > MAX_DEADLK_ITERATIONS)
 698			return 0;
 699		if (posix_same_owner(caller_fl, block_fl))
 700			return 1;
 701	}
 702	return 0;
 703}
 704
 705/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
 706 * after any leases, but before any posix locks.
 707 *
 708 * Note that if called with an FL_EXISTS argument, the caller may determine
 709 * whether or not a lock was successfully freed by testing the return
 710 * value for -ENOENT.
 711 */
 712static int flock_lock_file(struct file *filp, struct file_lock *request)
 713{
 714	struct file_lock *new_fl = NULL;
 715	struct file_lock **before;
 716	struct inode * inode = filp->f_path.dentry->d_inode;
 717	int error = 0;
 718	int found = 0;
 
 
 
 
 
 
 
 
 719
 720	if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
 721		new_fl = locks_alloc_lock();
 722		if (!new_fl)
 723			return -ENOMEM;
 724	}
 725
 726	lock_flocks();
 
 727	if (request->fl_flags & FL_ACCESS)
 728		goto find_conflict;
 729
 730	for_each_lock(inode, before) {
 731		struct file_lock *fl = *before;
 732		if (IS_POSIX(fl))
 733			break;
 734		if (IS_LEASE(fl))
 735			continue;
 736		if (filp != fl->fl_file)
 737			continue;
 738		if (request->fl_type == fl->fl_type)
 739			goto out;
 740		found = 1;
 741		locks_delete_lock(before);
 742		break;
 743	}
 744
 745	if (request->fl_type == F_UNLCK) {
 746		if ((request->fl_flags & FL_EXISTS) && !found)
 747			error = -ENOENT;
 748		goto out;
 749	}
 750
 751	/*
 752	 * If a higher-priority process was blocked on the old file lock,
 753	 * give it the opportunity to lock the file.
 754	 */
 755	if (found) {
 756		unlock_flocks();
 757		cond_resched();
 758		lock_flocks();
 759	}
 760
 761find_conflict:
 762	for_each_lock(inode, before) {
 763		struct file_lock *fl = *before;
 764		if (IS_POSIX(fl))
 765			break;
 766		if (IS_LEASE(fl))
 767			continue;
 768		if (!flock_locks_conflict(request, fl))
 769			continue;
 770		error = -EAGAIN;
 771		if (!(request->fl_flags & FL_SLEEP))
 772			goto out;
 773		error = FILE_LOCK_DEFERRED;
 774		locks_insert_block(fl, request);
 775		goto out;
 776	}
 777	if (request->fl_flags & FL_ACCESS)
 778		goto out;
 779	locks_copy_lock(new_fl, request);
 780	locks_insert_lock(before, new_fl);
 781	new_fl = NULL;
 782	error = 0;
 783
 784out:
 785	unlock_flocks();
 
 786	if (new_fl)
 787		locks_free_lock(new_fl);
 
 788	return error;
 789}
 790
 791static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
 
 792{
 793	struct file_lock *fl;
 794	struct file_lock *new_fl = NULL;
 795	struct file_lock *new_fl2 = NULL;
 796	struct file_lock *left = NULL;
 797	struct file_lock *right = NULL;
 798	struct file_lock **before;
 799	int error, added = 0;
 
 
 
 
 
 
 800
 801	/*
 802	 * We may need two file_lock structures for this operation,
 803	 * so we get them in advance to avoid races.
 804	 *
 805	 * In some cases we can be sure, that no new locks will be needed
 806	 */
 807	if (!(request->fl_flags & FL_ACCESS) &&
 808	    (request->fl_type != F_UNLCK ||
 809	     request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
 810		new_fl = locks_alloc_lock();
 811		new_fl2 = locks_alloc_lock();
 812	}
 813
 814	lock_flocks();
 
 
 
 
 
 
 815	if (request->fl_type != F_UNLCK) {
 816		for_each_lock(inode, before) {
 817			fl = *before;
 818			if (!IS_POSIX(fl))
 819				continue;
 820			if (!posix_locks_conflict(request, fl))
 821				continue;
 822			if (conflock)
 823				__locks_copy_lock(conflock, fl);
 824			error = -EAGAIN;
 825			if (!(request->fl_flags & FL_SLEEP))
 826				goto out;
 
 
 
 
 827			error = -EDEADLK;
 828			if (posix_locks_deadlock(request, fl))
 829				goto out;
 830			error = FILE_LOCK_DEFERRED;
 831			locks_insert_block(fl, request);
 
 
 832			goto out;
 833  		}
 834  	}
 835
 836	/* If we're just looking for a conflict, we're done. */
 837	error = 0;
 838	if (request->fl_flags & FL_ACCESS)
 839		goto out;
 840
 841	/*
 842	 * Find the first old lock with the same owner as the new lock.
 843	 */
 844	
 845	before = &inode->i_flock;
 846
 847	/* First skip locks owned by other processes.  */
 848	while ((fl = *before) && (!IS_POSIX(fl) ||
 849				  !posix_same_owner(request, fl))) {
 850		before = &fl->fl_next;
 851	}
 852
 853	/* Process locks with this owner.  */
 854	while ((fl = *before) && posix_same_owner(request, fl)) {
 855		/* Detect adjacent or overlapping regions (if same lock type)
 856		 */
 857		if (request->fl_type == fl->fl_type) {
 858			/* In all comparisons of start vs end, use
 859			 * "start - 1" rather than "end + 1". If end
 860			 * is OFFSET_MAX, end + 1 will become negative.
 861			 */
 862			if (fl->fl_end < request->fl_start - 1)
 863				goto next_lock;
 864			/* If the next lock in the list has entirely bigger
 865			 * addresses than the new one, insert the lock here.
 866			 */
 867			if (fl->fl_start - 1 > request->fl_end)
 868				break;
 869
 870			/* If we come here, the new and old lock are of the
 871			 * same type and adjacent or overlapping. Make one
 872			 * lock yielding from the lower start address of both
 873			 * locks to the higher end address.
 874			 */
 875			if (fl->fl_start > request->fl_start)
 876				fl->fl_start = request->fl_start;
 877			else
 878				request->fl_start = fl->fl_start;
 879			if (fl->fl_end < request->fl_end)
 880				fl->fl_end = request->fl_end;
 881			else
 882				request->fl_end = fl->fl_end;
 883			if (added) {
 884				locks_delete_lock(before);
 885				continue;
 886			}
 887			request = fl;
 888			added = 1;
 889		}
 890		else {
 891			/* Processing for different lock types is a bit
 892			 * more complex.
 893			 */
 894			if (fl->fl_end < request->fl_start)
 895				goto next_lock;
 896			if (fl->fl_start > request->fl_end)
 897				break;
 898			if (request->fl_type == F_UNLCK)
 899				added = 1;
 900			if (fl->fl_start < request->fl_start)
 901				left = fl;
 902			/* If the next lock in the list has a higher end
 903			 * address than the new one, insert the new one here.
 904			 */
 905			if (fl->fl_end > request->fl_end) {
 906				right = fl;
 907				break;
 908			}
 909			if (fl->fl_start >= request->fl_start) {
 910				/* The new lock completely replaces an old
 911				 * one (This may happen several times).
 912				 */
 913				if (added) {
 914					locks_delete_lock(before);
 915					continue;
 916				}
 917				/* Replace the old lock with the new one.
 918				 * Wake up anybody waiting for the old one,
 919				 * as the change in lock type might satisfy
 920				 * their needs.
 
 
 921				 */
 922				locks_wake_up_blocks(fl);
 923				fl->fl_start = request->fl_start;
 924				fl->fl_end = request->fl_end;
 925				fl->fl_type = request->fl_type;
 926				locks_release_private(fl);
 927				locks_copy_private(fl, request);
 928				request = fl;
 929				added = 1;
 
 930			}
 931		}
 932		/* Go on to next lock.
 933		 */
 934	next_lock:
 935		before = &fl->fl_next;
 936	}
 937
 938	/*
 939	 * The above code only modifies existing locks in case of
 940	 * merging or replacing.  If new lock(s) need to be inserted
 941	 * all modifications are done bellow this, so it's safe yet to
 942	 * bail out.
 943	 */
 944	error = -ENOLCK; /* "no luck" */
 945	if (right && left == right && !new_fl2)
 946		goto out;
 947
 948	error = 0;
 949	if (!added) {
 950		if (request->fl_type == F_UNLCK) {
 951			if (request->fl_flags & FL_EXISTS)
 952				error = -ENOENT;
 953			goto out;
 954		}
 955
 956		if (!new_fl) {
 957			error = -ENOLCK;
 958			goto out;
 959		}
 960		locks_copy_lock(new_fl, request);
 961		locks_insert_lock(before, new_fl);
 
 962		new_fl = NULL;
 963	}
 964	if (right) {
 965		if (left == right) {
 966			/* The new lock breaks the old one in two pieces,
 967			 * so we have to use the second new lock.
 968			 */
 969			left = new_fl2;
 970			new_fl2 = NULL;
 971			locks_copy_lock(left, right);
 972			locks_insert_lock(before, left);
 973		}
 974		right->fl_start = request->fl_end + 1;
 975		locks_wake_up_blocks(right);
 976	}
 977	if (left) {
 978		left->fl_end = request->fl_start - 1;
 979		locks_wake_up_blocks(left);
 980	}
 981 out:
 982	unlock_flocks();
 
 983	/*
 984	 * Free any unused locks.
 985	 */
 986	if (new_fl)
 987		locks_free_lock(new_fl);
 988	if (new_fl2)
 989		locks_free_lock(new_fl2);
 
 
 
 990	return error;
 991}
 992
 993/**
 994 * posix_lock_file - Apply a POSIX-style lock to a file
 995 * @filp: The file to apply the lock to
 996 * @fl: The lock to be applied
 997 * @conflock: Place to return a copy of the conflicting lock, if found.
 998 *
 999 * Add a POSIX style lock to a file.
1000 * We merge adjacent & overlapping locks whenever possible.
1001 * POSIX locks are sorted by owner task, then by starting address
1002 *
1003 * Note that if called with an FL_EXISTS argument, the caller may determine
1004 * whether or not a lock was successfully freed by testing the return
1005 * value for -ENOENT.
1006 */
1007int posix_lock_file(struct file *filp, struct file_lock *fl,
1008			struct file_lock *conflock)
1009{
1010	return __posix_lock_file(filp->f_path.dentry->d_inode, fl, conflock);
1011}
1012EXPORT_SYMBOL(posix_lock_file);
1013
1014/**
1015 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1016 * @filp: The file to apply the lock to
1017 * @fl: The lock to be applied
1018 *
1019 * Add a POSIX style lock to a file.
1020 * We merge adjacent & overlapping locks whenever possible.
1021 * POSIX locks are sorted by owner task, then by starting address
1022 */
1023int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1024{
1025	int error;
1026	might_sleep ();
1027	for (;;) {
1028		error = posix_lock_file(filp, fl, NULL);
1029		if (error != FILE_LOCK_DEFERRED)
1030			break;
1031		error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1032		if (!error)
1033			continue;
1034
1035		locks_delete_block(fl);
1036		break;
1037	}
1038	return error;
1039}
1040EXPORT_SYMBOL(posix_lock_file_wait);
1041
 
1042/**
1043 * locks_mandatory_locked - Check for an active lock
1044 * @inode: the file to check
1045 *
1046 * Searches the inode's list of locks to find any POSIX locks which conflict.
1047 * This function is called from locks_verify_locked() only.
1048 */
1049int locks_mandatory_locked(struct inode *inode)
1050{
1051	fl_owner_t owner = current->files;
 
 
1052	struct file_lock *fl;
1053
 
 
 
 
1054	/*
1055	 * Search the lock list for this inode for any POSIX locks.
1056	 */
1057	lock_flocks();
1058	for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1059		if (!IS_POSIX(fl))
1060			continue;
1061		if (fl->fl_owner != owner)
 
1062			break;
 
1063	}
1064	unlock_flocks();
1065	return fl ? -EAGAIN : 0;
1066}
1067
1068/**
1069 * locks_mandatory_area - Check for a conflicting lock
1070 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1071 *		for shared
1072 * @inode:      the file to check
1073 * @filp:       how the file was opened (if it was)
1074 * @offset:     start of area to check
1075 * @count:      length of area to check
 
1076 *
1077 * Searches the inode's list of locks to find any POSIX locks which conflict.
1078 * This function is called from rw_verify_area() and
1079 * locks_verify_truncate().
1080 */
1081int locks_mandatory_area(int read_write, struct inode *inode,
1082			 struct file *filp, loff_t offset,
1083			 size_t count)
1084{
1085	struct file_lock fl;
1086	int error;
 
1087
1088	locks_init_lock(&fl);
1089	fl.fl_owner = current->files;
1090	fl.fl_pid = current->tgid;
1091	fl.fl_file = filp;
1092	fl.fl_flags = FL_POSIX | FL_ACCESS;
1093	if (filp && !(filp->f_flags & O_NONBLOCK))
1094		fl.fl_flags |= FL_SLEEP;
1095	fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1096	fl.fl_start = offset;
1097	fl.fl_end = offset + count - 1;
1098
1099	for (;;) {
1100		error = __posix_lock_file(inode, &fl, NULL);
 
 
 
 
 
 
 
 
 
 
 
1101		if (error != FILE_LOCK_DEFERRED)
1102			break;
1103		error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1104		if (!error) {
1105			/*
1106			 * If we've been sleeping someone might have
1107			 * changed the permissions behind our back.
1108			 */
1109			if (__mandatory_lock(inode))
1110				continue;
1111		}
1112
1113		locks_delete_block(&fl);
1114		break;
1115	}
1116
1117	return error;
1118}
1119
1120EXPORT_SYMBOL(locks_mandatory_area);
 
 
 
 
 
 
 
 
 
 
 
 
1121
1122/* We already had a lease on this file; just change its type */
1123int lease_modify(struct file_lock **before, int arg)
1124{
1125	struct file_lock *fl = *before;
1126	int error = assign_type(fl, arg);
1127
1128	if (error)
1129		return error;
 
1130	locks_wake_up_blocks(fl);
1131	if (arg == F_UNLCK)
1132		locks_delete_lock(before);
 
 
 
 
 
 
 
 
 
 
1133	return 0;
1134}
 
1135
1136EXPORT_SYMBOL(lease_modify);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1137
1138static void time_out_leases(struct inode *inode)
 
1139{
1140	struct file_lock **before;
1141	struct file_lock *fl;
1142
1143	before = &inode->i_flock;
1144	while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1145		if ((fl->fl_break_time == 0)
1146				|| time_before(jiffies, fl->fl_break_time)) {
1147			before = &fl->fl_next;
1148			continue;
1149		}
1150		lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1151		if (fl == *before)	/* lease_modify may have freed fl */
1152			before = &fl->fl_next;
1153	}
 
1154}
1155
1156/**
1157 *	__break_lease	-	revoke all outstanding leases on file
1158 *	@inode: the inode of the file to return
1159 *	@mode: the open mode (read or write)
 
 
 
1160 *
1161 *	break_lease (inlined for speed) has checked there already is at least
1162 *	some kind of lock (maybe a lease) on this file.  Leases are broken on
1163 *	a call to open() or truncate().  This function can sleep unless you
1164 *	specified %O_NONBLOCK to your open().
1165 */
1166int __break_lease(struct inode *inode, unsigned int mode)
1167{
1168	int error = 0, future;
1169	struct file_lock *new_fl, *flock;
1170	struct file_lock *fl;
1171	unsigned long break_time;
1172	int i_have_this_lease = 0;
1173	int want_write = (mode & O_ACCMODE) != O_RDONLY;
 
1174
1175	new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
 
 
 
 
 
 
 
 
 
 
1176
1177	lock_flocks();
 
1178
1179	time_out_leases(inode);
1180
1181	flock = inode->i_flock;
1182	if ((flock == NULL) || !IS_LEASE(flock))
1183		goto out;
1184
1185	for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1186		if (fl->fl_owner == current->files)
1187			i_have_this_lease = 1;
1188
1189	if (want_write) {
1190		/* If we want write access, we have to revoke any lease. */
1191		future = F_UNLCK | F_INPROGRESS;
1192	} else if (flock->fl_type & F_INPROGRESS) {
1193		/* If the lease is already being broken, we just leave it */
1194		future = flock->fl_type;
1195	} else if (flock->fl_type & F_WRLCK) {
1196		/* Downgrade the exclusive lease to a read-only lease. */
1197		future = F_RDLCK | F_INPROGRESS;
1198	} else {
1199		/* the existing lease was read-only, so we can read too. */
1200		goto out;
1201	}
1202
1203	if (IS_ERR(new_fl) && !i_have_this_lease
1204			&& ((mode & O_NONBLOCK) == 0)) {
1205		error = PTR_ERR(new_fl);
1206		goto out;
1207	}
1208
1209	break_time = 0;
1210	if (lease_break_time > 0) {
1211		break_time = jiffies + lease_break_time * HZ;
1212		if (break_time == 0)
1213			break_time++;	/* so that 0 means no break time */
1214	}
1215
1216	for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1217		if (fl->fl_type != future) {
1218			fl->fl_type = future;
 
 
 
 
1219			fl->fl_break_time = break_time;
1220			/* lease must have lmops break callback */
1221			fl->fl_lmops->lm_break(fl);
 
 
 
1222		}
 
 
1223	}
1224
1225	if (i_have_this_lease || (mode & O_NONBLOCK)) {
 
 
 
 
1226		error = -EWOULDBLOCK;
1227		goto out;
1228	}
1229
1230restart:
1231	break_time = flock->fl_break_time;
1232	if (break_time != 0) {
 
1233		break_time -= jiffies;
1234		if (break_time == 0)
1235			break_time++;
1236	}
1237	locks_insert_block(flock, new_fl);
1238	unlock_flocks();
 
 
 
1239	error = wait_event_interruptible_timeout(new_fl->fl_wait,
1240						!new_fl->fl_next, break_time);
1241	lock_flocks();
1242	__locks_delete_block(new_fl);
 
 
 
1243	if (error >= 0) {
 
 
 
 
1244		if (error == 0)
1245			time_out_leases(inode);
1246		/* Wait for the next lease that has not been broken yet */
1247		for (flock = inode->i_flock; flock && IS_LEASE(flock);
1248				flock = flock->fl_next) {
1249			if (flock->fl_type & F_INPROGRESS)
1250				goto restart;
1251		}
1252		error = 0;
1253	}
1254
1255out:
1256	unlock_flocks();
1257	if (!IS_ERR(new_fl))
1258		locks_free_lock(new_fl);
 
1259	return error;
1260}
1261
1262EXPORT_SYMBOL(__break_lease);
1263
1264/**
1265 *	lease_get_mtime - get the last modified time of an inode
1266 *	@inode: the inode
1267 *      @time:  pointer to a timespec which will contain the last modified time
1268 *
1269 * This is to force NFS clients to flush their caches for files with
1270 * exclusive leases.  The justification is that if someone has an
1271 * exclusive lease, then they could be modifying it.
1272 */
1273void lease_get_mtime(struct inode *inode, struct timespec *time)
1274{
1275	struct file_lock *flock = inode->i_flock;
1276	if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1277		*time = current_fs_time(inode->i_sb);
1278	else
1279		*time = inode->i_mtime;
 
 
 
 
 
 
 
 
 
 
 
1280}
1281
1282EXPORT_SYMBOL(lease_get_mtime);
1283
1284/**
1285 *	fcntl_getlease - Enquire what lease is currently active
1286 *	@filp: the file
1287 *
1288 *	The value returned by this function will be one of
1289 *	(if no lease break is pending):
1290 *
1291 *	%F_RDLCK to indicate a shared lease is held.
1292 *
1293 *	%F_WRLCK to indicate an exclusive lease is held.
1294 *
1295 *	%F_UNLCK to indicate no lease is held.
1296 *
1297 *	(if a lease break is pending):
1298 *
1299 *	%F_RDLCK to indicate an exclusive lease needs to be
1300 *		changed to a shared lease (or removed).
1301 *
1302 *	%F_UNLCK to indicate the lease needs to be removed.
1303 *
1304 *	XXX: sfr & willy disagree over whether F_INPROGRESS
1305 *	should be returned to userspace.
1306 */
1307int fcntl_getlease(struct file *filp)
1308{
1309	struct file_lock *fl;
 
 
1310	int type = F_UNLCK;
 
1311
1312	lock_flocks();
1313	time_out_leases(filp->f_path.dentry->d_inode);
1314	for (fl = filp->f_path.dentry->d_inode->i_flock; fl && IS_LEASE(fl);
1315			fl = fl->fl_next) {
1316		if (fl->fl_file == filp) {
1317			type = fl->fl_type & ~F_INPROGRESS;
 
 
 
1318			break;
1319		}
 
 
 
 
1320	}
1321	unlock_flocks();
1322	return type;
1323}
1324
1325/**
1326 *	generic_setlease	-	sets a lease on an open file
1327 *	@filp: file pointer
1328 *	@arg: type of lease to obtain
1329 *	@flp: input - file_lock to use, output - file_lock inserted
 
 
1330 *
1331 *	The (input) flp->fl_lmops->lm_break function is required
1332 *	by break_lease().
1333 *
1334 *	Called with file_lock_lock held.
1335 */
1336int generic_setlease(struct file *filp, long arg, struct file_lock **flp)
 
1337{
1338	struct file_lock *fl, **before, **my_before = NULL, *lease;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1339	struct dentry *dentry = filp->f_path.dentry;
1340	struct inode *inode = dentry->d_inode;
1341	int error, rdlease_count = 0, wrlease_count = 0;
 
 
 
1342
1343	lease = *flp;
 
1344
1345	error = -EACCES;
1346	if ((current_fsuid() != inode->i_uid) && !capable(CAP_LEASE))
1347		goto out;
1348	error = -EINVAL;
1349	if (!S_ISREG(inode->i_mode))
1350		goto out;
1351	error = security_file_lock(filp, arg);
1352	if (error)
1353		goto out;
1354
1355	time_out_leases(inode);
 
 
 
 
 
 
 
 
 
1356
1357	BUG_ON(!(*flp)->fl_lmops->lm_break);
 
 
 
 
 
1358
1359	if (arg != F_UNLCK) {
1360		error = -EAGAIN;
1361		if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1362			goto out;
1363		if ((arg == F_WRLCK)
1364		    && ((dentry->d_count > 1)
1365			|| (atomic_read(&inode->i_count) > 1)))
1366			goto out;
1367	}
1368
1369	/*
1370	 * At this point, we know that if there is an exclusive
1371	 * lease on this file, then we hold it on this filp
1372	 * (otherwise our open of this file would have blocked).
1373	 * And if we are trying to acquire an exclusive lease,
1374	 * then the file is not open by anyone (including us)
1375	 * except for this filp.
1376	 */
1377	for (before = &inode->i_flock;
1378			((fl = *before) != NULL) && IS_LEASE(fl);
1379			before = &fl->fl_next) {
1380		if (fl->fl_file == filp)
1381			my_before = before;
1382		else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1383			/*
1384			 * Someone is in the process of opening this
1385			 * file for writing so we may not take an
1386			 * exclusive lease on it.
1387			 */
1388			wrlease_count++;
1389		else
1390			rdlease_count++;
1391	}
1392
1393	error = -EAGAIN;
1394	if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1395	    (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1396		goto out;
 
 
 
1397
1398	if (my_before != NULL) {
1399		error = lease->fl_lmops->lm_change(my_before, arg);
1400		if (!error)
1401			*flp = *my_before;
1402		goto out;
 
 
 
 
 
 
 
1403	}
1404
1405	if (arg == F_UNLCK)
1406		goto out;
 
 
 
 
 
1407
1408	error = -EINVAL;
1409	if (!leases_enable)
1410		goto out;
1411
1412	locks_insert_lock(before, lease);
1413	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1414
 
 
 
1415out:
 
 
 
 
 
 
 
1416	return error;
1417}
1418EXPORT_SYMBOL(generic_setlease);
1419
1420static int __vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1421{
1422	if (filp->f_op && filp->f_op->setlease)
1423		return filp->f_op->setlease(filp, arg, lease);
1424	else
1425		return generic_setlease(filp, arg, lease);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1426}
1427
1428/**
1429 *	vfs_setlease        -       sets a lease on an open file
1430 *	@filp: file pointer
1431 *	@arg: type of lease to obtain
1432 *	@lease: file_lock to use
 
 
1433 *
1434 *	Call this to establish a lease on the file.
1435 *	The (*lease)->fl_lmops->lm_break operation must be set; if not,
1436 *	break_lease will oops!
1437 *
1438 *	This will call the filesystem's setlease file method, if
1439 *	defined.  Note that there is no getlease method; instead, the
1440 *	filesystem setlease method should call back to setlease() to
1441 *	add a lease to the inode's lease list, where fcntl_getlease() can
1442 *	find it.  Since fcntl_getlease() only reports whether the current
1443 *	task holds a lease, a cluster filesystem need only do this for
1444 *	leases held by processes on this node.
1445 *
1446 *	There is also no break_lease method; filesystems that
1447 *	handle their own leases should break leases themselves from the
1448 *	filesystem's open, create, and (on truncate) setattr methods.
1449 *
1450 *	Warning: the only current setlease methods exist only to disable
1451 *	leases in certain cases.  More vfs changes may be required to
1452 *	allow a full filesystem lease implementation.
1453 */
1454
1455int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1456{
 
1457	int error;
1458
1459	lock_flocks();
1460	error = __vfs_setlease(filp, arg, lease);
1461	unlock_flocks();
 
 
 
 
1462
1463	return error;
 
 
 
 
 
 
 
 
 
 
 
 
 
1464}
1465EXPORT_SYMBOL_GPL(vfs_setlease);
1466
1467static int do_fcntl_delete_lease(struct file *filp)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1468{
1469	struct file_lock fl, *flp = &fl;
1470
1471	lease_init(filp, F_UNLCK, flp);
1472
1473	return vfs_setlease(filp, F_UNLCK, &flp);
1474}
 
1475
1476static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
1477{
1478	struct file_lock *fl, *ret;
1479	struct fasync_struct *new;
1480	int error;
1481
1482	fl = lease_alloc(filp, arg);
1483	if (IS_ERR(fl))
1484		return PTR_ERR(fl);
1485
1486	new = fasync_alloc();
1487	if (!new) {
1488		locks_free_lock(fl);
1489		return -ENOMEM;
1490	}
1491	ret = fl;
1492	lock_flocks();
1493	error = __vfs_setlease(filp, arg, &ret);
1494	if (error) {
1495		unlock_flocks();
1496		locks_free_lock(fl);
1497		goto out_free_fasync;
1498	}
1499	if (ret != fl)
1500		locks_free_lock(fl);
1501
1502	/*
1503	 * fasync_insert_entry() returns the old entry if any.
1504	 * If there was no old entry, then it used 'new' and
1505	 * inserted it into the fasync list. Clear new so that
1506	 * we don't release it here.
1507	 */
1508	if (!fasync_insert_entry(fd, filp, &ret->fl_fasync, new))
1509		new = NULL;
1510
1511	error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
1512	unlock_flocks();
1513
1514out_free_fasync:
1515	if (new)
1516		fasync_free(new);
1517	return error;
1518}
1519
1520/**
1521 *	fcntl_setlease	-	sets a lease on an open file
1522 *	@fd: open file descriptor
1523 *	@filp: file pointer
1524 *	@arg: type of lease to obtain
1525 *
1526 *	Call this fcntl to establish a lease on the file.
1527 *	Note that you also need to call %F_SETSIG to
1528 *	receive a signal when the lease is broken.
1529 */
1530int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1531{
1532	if (arg == F_UNLCK)
1533		return do_fcntl_delete_lease(filp);
1534	return do_fcntl_add_lease(fd, filp, arg);
1535}
1536
1537/**
1538 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1539 * @filp: The file to apply the lock to
1540 * @fl: The lock to be applied
1541 *
1542 * Add a FLOCK style lock to a file.
1543 */
1544int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1545{
1546	int error;
1547	might_sleep();
1548	for (;;) {
1549		error = flock_lock_file(filp, fl);
1550		if (error != FILE_LOCK_DEFERRED)
1551			break;
1552		error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1553		if (!error)
1554			continue;
1555
1556		locks_delete_block(fl);
1557		break;
1558	}
1559	return error;
1560}
1561
1562EXPORT_SYMBOL(flock_lock_file_wait);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1563
1564/**
1565 *	sys_flock: - flock() system call.
1566 *	@fd: the file descriptor to lock.
1567 *	@cmd: the type of lock to apply.
1568 *
1569 *	Apply a %FL_FLOCK style lock to an open file descriptor.
1570 *	The @cmd can be one of
1571 *
1572 *	%LOCK_SH -- a shared lock.
1573 *
1574 *	%LOCK_EX -- an exclusive lock.
1575 *
1576 *	%LOCK_UN -- remove an existing lock.
1577 *
1578 *	%LOCK_MAND -- a `mandatory' flock.  This exists to emulate Windows Share Modes.
 
 
 
 
1579 *
1580 *	%LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1581 *	processes read and write access respectively.
1582 */
1583SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
1584{
1585	struct file *filp;
1586	struct file_lock *lock;
1587	int can_sleep, unlock;
1588	int error;
1589
1590	error = -EBADF;
1591	filp = fget(fd);
1592	if (!filp)
1593		goto out;
1594
1595	can_sleep = !(cmd & LOCK_NB);
1596	cmd &= ~LOCK_NB;
1597	unlock = (cmd == LOCK_UN);
1598
1599	if (!unlock && !(cmd & LOCK_MAND) &&
1600	    !(filp->f_mode & (FMODE_READ|FMODE_WRITE)))
1601		goto out_putf;
1602
1603	error = flock_make_lock(filp, &lock, cmd);
1604	if (error)
 
1605		goto out_putf;
 
 
1606	if (can_sleep)
1607		lock->fl_flags |= FL_SLEEP;
1608
1609	error = security_file_lock(filp, lock->fl_type);
1610	if (error)
1611		goto out_free;
1612
1613	if (filp->f_op && filp->f_op->flock)
1614		error = filp->f_op->flock(filp,
1615					  (can_sleep) ? F_SETLKW : F_SETLK,
1616					  lock);
1617	else
1618		error = flock_lock_file_wait(filp, lock);
1619
1620 out_free:
1621	locks_free_lock(lock);
1622
1623 out_putf:
1624	fput(filp);
1625 out:
1626	return error;
1627}
1628
1629/**
1630 * vfs_test_lock - test file byte range lock
1631 * @filp: The file to test lock for
1632 * @fl: The lock to test; also used to hold result
1633 *
1634 * Returns -ERRNO on failure.  Indicates presence of conflicting lock by
1635 * setting conf->fl_type to something other than F_UNLCK.
1636 */
1637int vfs_test_lock(struct file *filp, struct file_lock *fl)
1638{
1639	if (filp->f_op && filp->f_op->lock)
1640		return filp->f_op->lock(filp, F_GETLK, fl);
1641	posix_test_lock(filp, fl);
1642	return 0;
1643}
1644EXPORT_SYMBOL_GPL(vfs_test_lock);
1645
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1646static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1647{
1648	flock->l_pid = fl->fl_pid;
1649#if BITS_PER_LONG == 32
1650	/*
1651	 * Make sure we can represent the posix lock via
1652	 * legacy 32bit flock.
1653	 */
1654	if (fl->fl_start > OFFT_OFFSET_MAX)
1655		return -EOVERFLOW;
1656	if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1657		return -EOVERFLOW;
1658#endif
1659	flock->l_start = fl->fl_start;
1660	flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1661		fl->fl_end - fl->fl_start + 1;
1662	flock->l_whence = 0;
1663	flock->l_type = fl->fl_type;
1664	return 0;
1665}
1666
1667#if BITS_PER_LONG == 32
1668static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1669{
1670	flock->l_pid = fl->fl_pid;
1671	flock->l_start = fl->fl_start;
1672	flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1673		fl->fl_end - fl->fl_start + 1;
1674	flock->l_whence = 0;
1675	flock->l_type = fl->fl_type;
1676}
1677#endif
1678
1679/* Report the first existing lock that would conflict with l.
1680 * This implements the F_GETLK command of fcntl().
1681 */
1682int fcntl_getlk(struct file *filp, struct flock __user *l)
1683{
1684	struct file_lock file_lock;
1685	struct flock flock;
1686	int error;
1687
1688	error = -EFAULT;
1689	if (copy_from_user(&flock, l, sizeof(flock)))
1690		goto out;
1691	error = -EINVAL;
1692	if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1693		goto out;
1694
1695	error = flock_to_posix_lock(filp, &file_lock, &flock);
1696	if (error)
1697		goto out;
1698
1699	error = vfs_test_lock(filp, &file_lock);
 
 
 
 
 
 
 
 
 
 
1700	if (error)
1701		goto out;
1702 
1703	flock.l_type = file_lock.fl_type;
1704	if (file_lock.fl_type != F_UNLCK) {
1705		error = posix_lock_to_flock(&flock, &file_lock);
1706		if (error)
1707			goto out;
1708	}
1709	error = -EFAULT;
1710	if (!copy_to_user(l, &flock, sizeof(flock)))
1711		error = 0;
1712out:
 
1713	return error;
1714}
1715
1716/**
1717 * vfs_lock_file - file byte range lock
1718 * @filp: The file to apply the lock to
1719 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1720 * @fl: The lock to be applied
1721 * @conf: Place to return a copy of the conflicting lock, if found.
1722 *
1723 * A caller that doesn't care about the conflicting lock may pass NULL
1724 * as the final argument.
1725 *
1726 * If the filesystem defines a private ->lock() method, then @conf will
1727 * be left unchanged; so a caller that cares should initialize it to
1728 * some acceptable default.
1729 *
1730 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1731 * locks, the ->lock() interface may return asynchronously, before the lock has
1732 * been granted or denied by the underlying filesystem, if (and only if)
1733 * lm_grant is set. Callers expecting ->lock() to return asynchronously
1734 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1735 * the request is for a blocking lock. When ->lock() does return asynchronously,
1736 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
1737 * request completes.
1738 * If the request is for non-blocking lock the file system should return
1739 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
1740 * with the result. If the request timed out the callback routine will return a
1741 * nonzero return code and the file system should release the lock. The file
1742 * system is also responsible to keep a corresponding posix lock when it
1743 * grants a lock so the VFS can find out which locks are locally held and do
1744 * the correct lock cleanup when required.
1745 * The underlying filesystem must not drop the kernel lock or call
1746 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
1747 * return code.
1748 */
1749int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
1750{
1751	if (filp->f_op && filp->f_op->lock)
1752		return filp->f_op->lock(filp, cmd, fl);
1753	else
1754		return posix_lock_file(filp, fl, conf);
1755}
1756EXPORT_SYMBOL_GPL(vfs_lock_file);
1757
1758static int do_lock_file_wait(struct file *filp, unsigned int cmd,
1759			     struct file_lock *fl)
1760{
1761	int error;
1762
1763	error = security_file_lock(filp, fl->fl_type);
1764	if (error)
1765		return error;
1766
1767	for (;;) {
1768		error = vfs_lock_file(filp, cmd, fl, NULL);
1769		if (error != FILE_LOCK_DEFERRED)
1770			break;
1771		error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1772		if (!error)
1773			continue;
1774
1775		locks_delete_block(fl);
1776		break;
1777	}
1778
1779	return error;
1780}
1781
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1782/* Apply the lock described by l to an open file descriptor.
1783 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1784 */
1785int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1786		struct flock __user *l)
1787{
1788	struct file_lock *file_lock = locks_alloc_lock();
1789	struct flock flock;
1790	struct inode *inode;
1791	struct file *f;
1792	int error;
1793
1794	if (file_lock == NULL)
1795		return -ENOLCK;
1796
1797	/*
1798	 * This might block, so we do it before checking the inode.
1799	 */
1800	error = -EFAULT;
1801	if (copy_from_user(&flock, l, sizeof(flock)))
1802		goto out;
1803
1804	inode = filp->f_path.dentry->d_inode;
1805
1806	/* Don't allow mandatory locks on files that may be memory mapped
1807	 * and shared.
1808	 */
1809	if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
1810		error = -EAGAIN;
1811		goto out;
1812	}
1813
1814again:
1815	error = flock_to_posix_lock(filp, file_lock, &flock);
 
 
 
1816	if (error)
1817		goto out;
1818	if (cmd == F_SETLKW) {
1819		file_lock->fl_flags |= FL_SLEEP;
1820	}
1821	
1822	error = -EBADF;
1823	switch (flock.l_type) {
1824	case F_RDLCK:
1825		if (!(filp->f_mode & FMODE_READ))
 
1826			goto out;
 
 
 
 
1827		break;
1828	case F_WRLCK:
1829		if (!(filp->f_mode & FMODE_WRITE))
 
1830			goto out;
1831		break;
1832	case F_UNLCK:
1833		break;
1834	default:
1835		error = -EINVAL;
1836		goto out;
 
1837	}
1838
1839	error = do_lock_file_wait(filp, cmd, file_lock);
1840
1841	/*
1842	 * Attempt to detect a close/fcntl race and recover by
1843	 * releasing the lock that was just acquired.
 
1844	 */
1845	/*
1846	 * we need that spin_lock here - it prevents reordering between
1847	 * update of inode->i_flock and check for it done in close().
1848	 * rcu_read_lock() wouldn't do.
1849	 */
1850	spin_lock(&current->files->file_lock);
1851	f = fcheck(fd);
1852	spin_unlock(&current->files->file_lock);
1853	if (!error && f != filp && flock.l_type != F_UNLCK) {
1854		flock.l_type = F_UNLCK;
1855		goto again;
 
 
 
 
 
1856	}
1857
1858out:
 
1859	locks_free_lock(file_lock);
1860	return error;
1861}
1862
1863#if BITS_PER_LONG == 32
1864/* Report the first existing lock that would conflict with l.
1865 * This implements the F_GETLK command of fcntl().
1866 */
1867int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1868{
1869	struct file_lock file_lock;
1870	struct flock64 flock;
1871	int error;
1872
1873	error = -EFAULT;
1874	if (copy_from_user(&flock, l, sizeof(flock)))
1875		goto out;
 
1876	error = -EINVAL;
1877	if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1878		goto out;
1879
1880	error = flock64_to_posix_lock(filp, &file_lock, &flock);
1881	if (error)
1882		goto out;
1883
1884	error = vfs_test_lock(filp, &file_lock);
 
 
 
 
 
 
 
 
 
 
1885	if (error)
1886		goto out;
1887
1888	flock.l_type = file_lock.fl_type;
1889	if (file_lock.fl_type != F_UNLCK)
1890		posix_lock_to_flock64(&flock, &file_lock);
1891
1892	error = -EFAULT;
1893	if (!copy_to_user(l, &flock, sizeof(flock)))
1894		error = 0;
1895  
1896out:
 
1897	return error;
1898}
1899
1900/* Apply the lock described by l to an open file descriptor.
1901 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1902 */
1903int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1904		struct flock64 __user *l)
1905{
1906	struct file_lock *file_lock = locks_alloc_lock();
1907	struct flock64 flock;
1908	struct inode *inode;
1909	struct file *f;
1910	int error;
1911
1912	if (file_lock == NULL)
1913		return -ENOLCK;
1914
1915	/*
1916	 * This might block, so we do it before checking the inode.
1917	 */
1918	error = -EFAULT;
1919	if (copy_from_user(&flock, l, sizeof(flock)))
1920		goto out;
1921
1922	inode = filp->f_path.dentry->d_inode;
1923
1924	/* Don't allow mandatory locks on files that may be memory mapped
1925	 * and shared.
1926	 */
1927	if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
1928		error = -EAGAIN;
1929		goto out;
1930	}
1931
1932again:
1933	error = flock64_to_posix_lock(filp, file_lock, &flock);
 
 
 
1934	if (error)
1935		goto out;
1936	if (cmd == F_SETLKW64) {
1937		file_lock->fl_flags |= FL_SLEEP;
1938	}
1939	
1940	error = -EBADF;
1941	switch (flock.l_type) {
1942	case F_RDLCK:
1943		if (!(filp->f_mode & FMODE_READ))
 
1944			goto out;
 
 
 
 
1945		break;
1946	case F_WRLCK:
1947		if (!(filp->f_mode & FMODE_WRITE))
 
1948			goto out;
1949		break;
1950	case F_UNLCK:
1951		break;
1952	default:
1953		error = -EINVAL;
1954		goto out;
 
1955	}
1956
1957	error = do_lock_file_wait(filp, cmd, file_lock);
1958
1959	/*
1960	 * Attempt to detect a close/fcntl race and recover by
1961	 * releasing the lock that was just acquired.
 
1962	 */
1963	spin_lock(&current->files->file_lock);
1964	f = fcheck(fd);
1965	spin_unlock(&current->files->file_lock);
1966	if (!error && f != filp && flock.l_type != F_UNLCK) {
1967		flock.l_type = F_UNLCK;
1968		goto again;
 
 
 
 
 
 
 
 
 
 
1969	}
1970
1971out:
1972	locks_free_lock(file_lock);
1973	return error;
1974}
1975#endif /* BITS_PER_LONG == 32 */
1976
1977/*
1978 * This function is called when the file is being removed
1979 * from the task's fd array.  POSIX locks belonging to this task
1980 * are deleted at this time.
1981 */
1982void locks_remove_posix(struct file *filp, fl_owner_t owner)
1983{
 
 
1984	struct file_lock lock;
 
1985
1986	/*
1987	 * If there are no locks held on this file, we don't need to call
1988	 * posix_lock_file().  Another process could be setting a lock on this
1989	 * file at the same time, but we wouldn't remove that lock anyway.
1990	 */
1991	if (!filp->f_path.dentry->d_inode->i_flock)
 
1992		return;
1993
1994	lock.fl_type = F_UNLCK;
1995	lock.fl_flags = FL_POSIX | FL_CLOSE;
1996	lock.fl_start = 0;
1997	lock.fl_end = OFFSET_MAX;
1998	lock.fl_owner = owner;
1999	lock.fl_pid = current->tgid;
2000	lock.fl_file = filp;
2001	lock.fl_ops = NULL;
2002	lock.fl_lmops = NULL;
2003
2004	vfs_lock_file(filp, F_SETLK, &lock, NULL);
2005
2006	if (lock.fl_ops && lock.fl_ops->fl_release_private)
2007		lock.fl_ops->fl_release_private(&lock);
 
2008}
2009
2010EXPORT_SYMBOL(locks_remove_posix);
2011
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2012/*
2013 * This function is called on the last close of an open file.
2014 */
2015void locks_remove_flock(struct file *filp)
2016{
2017	struct inode * inode = filp->f_path.dentry->d_inode;
2018	struct file_lock *fl;
2019	struct file_lock **before;
2020
2021	if (!inode->i_flock)
 
2022		return;
2023
2024	if (filp->f_op && filp->f_op->flock) {
2025		struct file_lock fl = {
2026			.fl_pid = current->tgid,
2027			.fl_file = filp,
2028			.fl_flags = FL_FLOCK,
2029			.fl_type = F_UNLCK,
2030			.fl_end = OFFSET_MAX,
2031		};
2032		filp->f_op->flock(filp, F_SETLKW, &fl);
2033		if (fl.fl_ops && fl.fl_ops->fl_release_private)
2034			fl.fl_ops->fl_release_private(&fl);
2035	}
2036
2037	lock_flocks();
2038	before = &inode->i_flock;
2039
2040	while ((fl = *before) != NULL) {
2041		if (fl->fl_file == filp) {
2042			if (IS_FLOCK(fl)) {
2043				locks_delete_lock(before);
2044				continue;
2045			}
2046			if (IS_LEASE(fl)) {
2047				lease_modify(before, F_UNLCK);
2048				continue;
2049			}
2050			/* What? */
2051			BUG();
2052 		}
2053		before = &fl->fl_next;
2054	}
2055	unlock_flocks();
2056}
2057
2058/**
2059 *	posix_unblock_lock - stop waiting for a file lock
2060 *      @filp:   how the file was opened
2061 *	@waiter: the lock which was waiting
2062 *
2063 *	lockd needs to block waiting for locks.
2064 */
2065int
2066posix_unblock_lock(struct file *filp, struct file_lock *waiter)
2067{
2068	int status = 0;
2069
2070	lock_flocks();
2071	if (waiter->fl_next)
2072		__locks_delete_block(waiter);
2073	else
2074		status = -ENOENT;
2075	unlock_flocks();
2076	return status;
2077}
2078
2079EXPORT_SYMBOL(posix_unblock_lock);
2080
2081/**
2082 * vfs_cancel_lock - file byte range unblock lock
2083 * @filp: The file to apply the unblock to
2084 * @fl: The lock to be unblocked
2085 *
2086 * Used by lock managers to cancel blocked requests
2087 */
2088int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2089{
2090	if (filp->f_op && filp->f_op->lock)
2091		return filp->f_op->lock(filp, F_CANCELLK, fl);
2092	return 0;
2093}
2094
2095EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2096
2097#ifdef CONFIG_PROC_FS
2098#include <linux/proc_fs.h>
2099#include <linux/seq_file.h>
2100
 
 
 
 
 
2101static void lock_get_status(struct seq_file *f, struct file_lock *fl,
2102			    loff_t id, char *pfx)
2103{
2104	struct inode *inode = NULL;
2105	unsigned int fl_pid;
 
2106
2107	if (fl->fl_nspid)
2108		fl_pid = pid_vnr(fl->fl_nspid);
2109	else
2110		fl_pid = fl->fl_pid;
 
 
 
 
2111
2112	if (fl->fl_file != NULL)
2113		inode = fl->fl_file->f_path.dentry->d_inode;
2114
2115	seq_printf(f, "%lld:%s ", id, pfx);
2116	if (IS_POSIX(fl)) {
2117		seq_printf(f, "%6s %s ",
2118			     (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
 
 
 
 
 
 
2119			     (inode == NULL) ? "*NOINODE*" :
2120			     mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
2121	} else if (IS_FLOCK(fl)) {
2122		if (fl->fl_type & LOCK_MAND) {
2123			seq_printf(f, "FLOCK  MSNFS     ");
2124		} else {
2125			seq_printf(f, "FLOCK  ADVISORY  ");
2126		}
2127	} else if (IS_LEASE(fl)) {
2128		seq_printf(f, "LEASE  ");
2129		if (fl->fl_type & F_INPROGRESS)
2130			seq_printf(f, "BREAKING  ");
 
 
 
 
2131		else if (fl->fl_file)
2132			seq_printf(f, "ACTIVE    ");
2133		else
2134			seq_printf(f, "BREAKER   ");
2135	} else {
2136		seq_printf(f, "UNKNOWN UNKNOWN  ");
2137	}
2138	if (fl->fl_type & LOCK_MAND) {
2139		seq_printf(f, "%s ",
2140			       (fl->fl_type & LOCK_READ)
2141			       ? (fl->fl_type & LOCK_WRITE) ? "RW   " : "READ "
2142			       : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2143	} else {
2144		seq_printf(f, "%s ",
2145			       (fl->fl_type & F_INPROGRESS)
2146			       ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2147			       : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2148	}
2149	if (inode) {
2150#ifdef WE_CAN_BREAK_LSLK_NOW
2151		seq_printf(f, "%d %s:%ld ", fl_pid,
2152				inode->i_sb->s_id, inode->i_ino);
2153#else
2154		/* userspace relies on this representation of dev_t ;-( */
2155		seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
2156				MAJOR(inode->i_sb->s_dev),
2157				MINOR(inode->i_sb->s_dev), inode->i_ino);
2158#endif
2159	} else {
2160		seq_printf(f, "%d <none>:0 ", fl_pid);
2161	}
2162	if (IS_POSIX(fl)) {
2163		if (fl->fl_end == OFFSET_MAX)
2164			seq_printf(f, "%Ld EOF\n", fl->fl_start);
2165		else
2166			seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
2167	} else {
2168		seq_printf(f, "0 EOF\n");
2169	}
2170}
2171
2172static int locks_show(struct seq_file *f, void *v)
2173{
 
2174	struct file_lock *fl, *bfl;
 
2175
2176	fl = list_entry(v, struct file_lock, fl_link);
2177
2178	lock_get_status(f, fl, *((loff_t *)f->private), "");
 
 
 
2179
2180	list_for_each_entry(bfl, &fl->fl_block, fl_block)
2181		lock_get_status(f, bfl, *((loff_t *)f->private), " ->");
2182
2183	return 0;
2184}
2185
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2186static void *locks_start(struct seq_file *f, loff_t *pos)
 
2187{
2188	loff_t *p = f->private;
2189
2190	lock_flocks();
2191	*p = (*pos + 1);
2192	return seq_list_start(&file_lock_list, *pos);
 
2193}
2194
2195static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2196{
2197	loff_t *p = f->private;
2198	++*p;
2199	return seq_list_next(v, &file_lock_list, pos);
 
2200}
2201
2202static void locks_stop(struct seq_file *f, void *v)
 
2203{
2204	unlock_flocks();
 
2205}
2206
2207static const struct seq_operations locks_seq_operations = {
2208	.start	= locks_start,
2209	.next	= locks_next,
2210	.stop	= locks_stop,
2211	.show	= locks_show,
2212};
2213
2214static int locks_open(struct inode *inode, struct file *filp)
2215{
2216	return seq_open_private(filp, &locks_seq_operations, sizeof(loff_t));
 
2217}
2218
2219static const struct file_operations proc_locks_operations = {
2220	.open		= locks_open,
2221	.read		= seq_read,
2222	.llseek		= seq_lseek,
2223	.release	= seq_release_private,
2224};
2225
2226static int __init proc_locks_init(void)
2227{
2228	proc_create("locks", 0, NULL, &proc_locks_operations);
2229	return 0;
2230}
2231module_init(proc_locks_init);
2232#endif
2233
2234/**
2235 *	lock_may_read - checks that the region is free of locks
2236 *	@inode: the inode that is being read
2237 *	@start: the first byte to read
2238 *	@len: the number of bytes to read
2239 *
2240 *	Emulates Windows locking requirements.  Whole-file
2241 *	mandatory locks (share modes) can prohibit a read and
2242 *	byte-range POSIX locks can prohibit a read if they overlap.
2243 *
2244 *	N.B. this function is only ever called
2245 *	from knfsd and ownership of locks is never checked.
2246 */
2247int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2248{
2249	struct file_lock *fl;
2250	int result = 1;
2251	lock_flocks();
2252	for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2253		if (IS_POSIX(fl)) {
2254			if (fl->fl_type == F_RDLCK)
2255				continue;
2256			if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2257				continue;
2258		} else if (IS_FLOCK(fl)) {
2259			if (!(fl->fl_type & LOCK_MAND))
2260				continue;
2261			if (fl->fl_type & LOCK_READ)
2262				continue;
2263		} else
2264			continue;
2265		result = 0;
2266		break;
2267	}
2268	unlock_flocks();
2269	return result;
2270}
2271
2272EXPORT_SYMBOL(lock_may_read);
 
2273
2274/**
2275 *	lock_may_write - checks that the region is free of locks
2276 *	@inode: the inode that is being written
2277 *	@start: the first byte to write
2278 *	@len: the number of bytes to write
2279 *
2280 *	Emulates Windows locking requirements.  Whole-file
2281 *	mandatory locks (share modes) can prohibit a write and
2282 *	byte-range POSIX locks can prohibit a write if they overlap.
2283 *
2284 *	N.B. this function is only ever called
2285 *	from knfsd and ownership of locks is never checked.
2286 */
2287int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2288{
2289	struct file_lock *fl;
2290	int result = 1;
2291	lock_flocks();
2292	for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2293		if (IS_POSIX(fl)) {
2294			if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2295				continue;
2296		} else if (IS_FLOCK(fl)) {
2297			if (!(fl->fl_type & LOCK_MAND))
2298				continue;
2299			if (fl->fl_type & LOCK_WRITE)
2300				continue;
2301		} else
2302			continue;
2303		result = 0;
2304		break;
2305	}
2306	unlock_flocks();
2307	return result;
2308}
2309
2310EXPORT_SYMBOL(lock_may_write);
 
2311
2312static int __init filelock_init(void)
2313{
2314	filelock_cache = kmem_cache_create("file_lock_cache",
2315			sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2316
2317	return 0;
2318}
2319
2320core_initcall(filelock_init);