linux/fs/locks.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 *  linux/fs/locks.c
   4 *
   5 *  Provide support for fcntl()'s F_GETLK, F_SETLK, and F_SETLKW calls.
   6 *  Doug Evans (dje@spiff.uucp), August 07, 1992
   7 *
   8 *  Deadlock detection added.
   9 *  FIXME: one thing isn't handled yet:
  10 *      - mandatory locks (requires lots of changes elsewhere)
  11 *  Kelly Carmichael (kelly@[142.24.8.65]), September 17, 1994.
  12 *
  13 *  Miscellaneous edits, and a total rewrite of posix_lock_file() code.
  14 *  Kai Petzke (wpp@marie.physik.tu-berlin.de), 1994
  15 *
  16 *  Converted file_lock_table to a linked list from an array, which eliminates
  17 *  the limits on how many active file locks are open.
  18 *  Chad Page (pageone@netcom.com), November 27, 1994
  19 *
  20 *  Removed dependency on file descriptors. dup()'ed file descriptors now
  21 *  get the same locks as the original file descriptors, and a close() on
  22 *  any file descriptor removes ALL the locks on the file for the current
  23 *  process. Since locks still depend on the process id, locks are inherited
  24 *  after an exec() but not after a fork(). This agrees with POSIX, and both
  25 *  BSD and SVR4 practice.
  26 *  Andy Walker (andy@lysaker.kvaerner.no), February 14, 1995
  27 *
  28 *  Scrapped free list which is redundant now that we allocate locks
  29 *  dynamically with kmalloc()/kfree().
  30 *  Andy Walker (andy@lysaker.kvaerner.no), February 21, 1995
  31 *
  32 *  Implemented two lock personalities - FL_FLOCK and FL_POSIX.
  33 *
  34 *  FL_POSIX locks are created with calls to fcntl() and lockf() through the
  35 *  fcntl() system call. They have the semantics described above.
  36 *
  37 *  FL_FLOCK locks are created with calls to flock(), through the flock()
  38 *  system call, which is new. Old C libraries implement flock() via fcntl()
  39 *  and will continue to use the old, broken implementation.
  40 *
  41 *  FL_FLOCK locks follow the 4.4 BSD flock() semantics. They are associated
  42 *  with a file pointer (filp). As a result they can be shared by a parent
  43 *  process and its children after a fork(). They are removed when the last
  44 *  file descriptor referring to the file pointer is closed (unless explicitly
  45 *  unlocked).
  46 *
  47 *  FL_FLOCK locks never deadlock, an existing lock is always removed before
  48 *  upgrading from shared to exclusive (or vice versa). When this happens
  49 *  any processes blocked by the current lock are woken up and allowed to
  50 *  run before the new lock is applied.
  51 *  Andy Walker (andy@lysaker.kvaerner.no), June 09, 1995
  52 *
  53 *  Removed some race conditions in flock_lock_file(), marked other possible
  54 *  races. Just grep for FIXME to see them.
  55 *  Dmitry Gorodchanin (pgmdsg@ibi.com), February 09, 1996.
  56 *
  57 *  Addressed Dmitry's concerns. Deadlock checking no longer recursive.
  58 *  Lock allocation changed to GFP_ATOMIC as we can't afford to sleep
  59 *  once we've checked for blocking and deadlocking.
  60 *  Andy Walker (andy@lysaker.kvaerner.no), April 03, 1996.
  61 *
  62 *  Initial implementation of mandatory locks. SunOS turned out to be
  63 *  a rotten model, so I implemented the "obvious" semantics.
  64 *  See 'Documentation/filesystems/mandatory-locking.rst' for details.
  65 *  Andy Walker (andy@lysaker.kvaerner.no), April 06, 1996.
  66 *
  67 *  Don't allow mandatory locks on mmap()'ed files. Added simple functions to
  68 *  check if a file has mandatory locks, used by mmap(), open() and creat() to
  69 *  see if system call should be rejected. Ref. HP-UX/SunOS/Solaris Reference
  70 *  Manual, Section 2.
  71 *  Andy Walker (andy@lysaker.kvaerner.no), April 09, 1996.
  72 *
  73 *  Tidied up block list handling. Added '/proc/locks' interface.
  74 *  Andy Walker (andy@lysaker.kvaerner.no), April 24, 1996.
  75 *
  76 *  Fixed deadlock condition for pathological code that mixes calls to
  77 *  flock() and fcntl().
  78 *  Andy Walker (andy@lysaker.kvaerner.no), April 29, 1996.
  79 *
  80 *  Allow only one type of locking scheme (FL_POSIX or FL_FLOCK) to be in use
  81 *  for a given file at a time. Changed the CONFIG_LOCK_MANDATORY scheme to
  82 *  guarantee sensible behaviour in the case where file system modules might
  83 *  be compiled with different options than the kernel itself.
  84 *  Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
  85 *
  86 *  Added a couple of missing wake_up() calls. Thanks to Thomas Meckel
  87 *  (Thomas.Meckel@mni.fh-giessen.de) for spotting this.
  88 *  Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
  89 *
  90 *  Changed FL_POSIX locks to use the block list in the same way as FL_FLOCK
  91 *  locks. Changed process synchronisation to avoid dereferencing locks that
  92 *  have already been freed.
  93 *  Andy Walker (andy@lysaker.kvaerner.no), Sep 21, 1996.
  94 *
  95 *  Made the block list a circular list to minimise searching in the list.
  96 *  Andy Walker (andy@lysaker.kvaerner.no), Sep 25, 1996.
  97 *
  98 *  Made mandatory locking a mount option. Default is not to allow mandatory
  99 *  locking.
 100 *  Andy Walker (andy@lysaker.kvaerner.no), Oct 04, 1996.
 101 *
 102 *  Some adaptations for NFS support.
 103 *  Olaf Kirch (okir@monad.swb.de), Dec 1996,
 104 *
 105 *  Fixed /proc/locks interface so that we can't overrun the buffer we are handed.
 106 *  Andy Walker (andy@lysaker.kvaerner.no), May 12, 1997.
 107 *
 108 *  Use slab allocator instead of kmalloc/kfree.
 109 *  Use generic list implementation from <linux/list.h>.
 110 *  Sped up posix_locks_deadlock by only considering blocked locks.
 111 *  Matthew Wilcox <willy@debian.org>, March, 2000.
 112 *
 113 *  Leases and LOCK_MAND
 114 *  Matthew Wilcox <willy@debian.org>, June, 2000.
 115 *  Stephen Rothwell <sfr@canb.auug.org.au>, June, 2000.
 116 *
 117 * Locking conflicts and dependencies:
 118 * If multiple threads attempt to lock the same byte (or flock the same file)
 119 * only one can be granted the lock, and other must wait their turn.
 120 * The first lock has been "applied" or "granted", the others are "waiting"
 121 * and are "blocked" by the "applied" lock..
 122 *
 123 * Waiting and applied locks are all kept in trees whose properties are:
 124 *
 125 *      - the root of a tree may be an applied or waiting lock.
 126 *      - every other node in the tree is a waiting lock that
 127 *        conflicts with every ancestor of that node.
 128 *
 129 * Every such tree begins life as a waiting singleton which obviously
 130 * satisfies the above properties.
 131 *
 132 * The only ways we modify trees preserve these properties:
 133 *
 134 *      1. We may add a new leaf node, but only after first verifying that it
 135 *         conflicts with all of its ancestors.
 136 *      2. We may remove the root of a tree, creating a new singleton
 137 *         tree from the root and N new trees rooted in the immediate
 138 *         children.
 139 *      3. If the root of a tree is not currently an applied lock, we may
 140 *         apply it (if possible).
 141 *      4. We may upgrade the root of the tree (either extend its range,
 142 *         or upgrade its entire range from read to write).
 143 *
 144 * When an applied lock is modified in a way that reduces or downgrades any
 145 * part of its range, we remove all its children (2 above).  This particularly
 146 * happens when a lock is unlocked.
 147 *
 148 * For each of those child trees we "wake up" the thread which is
 149 * waiting for the lock so it can continue handling as follows: if the
 150 * root of the tree applies, we do so (3).  If it doesn't, it must
 151 * conflict with some applied lock.  We remove (wake up) all of its children
 152 * (2), and add it is a new leaf to the tree rooted in the applied
 153 * lock (1).  We then repeat the process recursively with those
 154 * children.
 155 *
 156 */
 157
 158#include <linux/capability.h>
 159#include <linux/file.h>
 160#include <linux/fdtable.h>
 161#include <linux/fs.h>
 162#include <linux/init.h>
 163#include <linux/security.h>
 164#include <linux/slab.h>
 165#include <linux/syscalls.h>
 166#include <linux/time.h>
 167#include <linux/rcupdate.h>
 168#include <linux/pid_namespace.h>
 169#include <linux/hashtable.h>
 170#include <linux/percpu.h>
 171
 172#define CREATE_TRACE_POINTS
 173#include <trace/events/filelock.h>
 174
 175#include <linux/uaccess.h>
 176
 177#define IS_POSIX(fl)    (fl->fl_flags & FL_POSIX)
 178#define IS_FLOCK(fl)    (fl->fl_flags & FL_FLOCK)
 179#define IS_LEASE(fl)    (fl->fl_flags & (FL_LEASE|FL_DELEG|FL_LAYOUT))
 180#define IS_OFDLCK(fl)   (fl->fl_flags & FL_OFDLCK)
 181#define IS_REMOTELCK(fl)        (fl->fl_pid <= 0)
 182
 183static bool lease_breaking(struct file_lock *fl)
 184{
 185        return fl->fl_flags & (FL_UNLOCK_PENDING | FL_DOWNGRADE_PENDING);
 186}
 187
 188static int target_leasetype(struct file_lock *fl)
 189{
 190        if (fl->fl_flags & FL_UNLOCK_PENDING)
 191                return F_UNLCK;
 192        if (fl->fl_flags & FL_DOWNGRADE_PENDING)
 193                return F_RDLCK;
 194        return fl->fl_type;
 195}
 196
 197int leases_enable = 1;
 198int lease_break_time = 45;
 199
 200/*
 201 * The global file_lock_list is only used for displaying /proc/locks, so we
 202 * keep a list on each CPU, with each list protected by its own spinlock.
 203 * Global serialization is done using file_rwsem.
 204 *
 205 * Note that alterations to the list also require that the relevant flc_lock is
 206 * held.
 207 */
 208struct file_lock_list_struct {
 209        spinlock_t              lock;
 210        struct hlist_head       hlist;
 211};
 212static DEFINE_PER_CPU(struct file_lock_list_struct, file_lock_list);
 213DEFINE_STATIC_PERCPU_RWSEM(file_rwsem);
 214
 215
 216/*
 217 * The blocked_hash is used to find POSIX lock loops for deadlock detection.
 218 * It is protected by blocked_lock_lock.
 219 *
 220 * We hash locks by lockowner in order to optimize searching for the lock a
 221 * particular lockowner is waiting on.
 222 *
 223 * FIXME: make this value scale via some heuristic? We generally will want more
 224 * buckets when we have more lockowners holding locks, but that's a little
 225 * difficult to determine without knowing what the workload will look like.
 226 */
 227#define BLOCKED_HASH_BITS       7
 228static DEFINE_HASHTABLE(blocked_hash, BLOCKED_HASH_BITS);
 229
 230/*
 231 * This lock protects the blocked_hash. Generally, if you're accessing it, you
 232 * want to be holding this lock.
 233 *
 234 * In addition, it also protects the fl->fl_blocked_requests list, and the
 235 * fl->fl_blocker pointer for file_lock structures that are acting as lock
 236 * requests (in contrast to those that are acting as records of acquired locks).
 237 *
 238 * Note that when we acquire this lock in order to change the above fields,
 239 * we often hold the flc_lock as well. In certain cases, when reading the fields
 240 * protected by this lock, we can skip acquiring it iff we already hold the
 241 * flc_lock.
 242 */
 243static DEFINE_SPINLOCK(blocked_lock_lock);
 244
 245static struct kmem_cache *flctx_cache __read_mostly;
 246static struct kmem_cache *filelock_cache __read_mostly;
 247
 248static struct file_lock_context *
 249locks_get_lock_context(struct inode *inode, int type)
 250{
 251        struct file_lock_context *ctx;
 252
 253        /* paired with cmpxchg() below */
 254        ctx = smp_load_acquire(&inode->i_flctx);
 255        if (likely(ctx) || type == F_UNLCK)
 256                goto out;
 257
 258        ctx = kmem_cache_alloc(flctx_cache, GFP_KERNEL);
 259        if (!ctx)
 260                goto out;
 261
 262        spin_lock_init(&ctx->flc_lock);
 263        INIT_LIST_HEAD(&ctx->flc_flock);
 264        INIT_LIST_HEAD(&ctx->flc_posix);
 265        INIT_LIST_HEAD(&ctx->flc_lease);
 266
 267        /*
 268         * Assign the pointer if it's not already assigned. If it is, then
 269         * free the context we just allocated.
 270         */
 271        if (cmpxchg(&inode->i_flctx, NULL, ctx)) {
 272                kmem_cache_free(flctx_cache, ctx);
 273                ctx = smp_load_acquire(&inode->i_flctx);
 274        }
 275out:
 276        trace_locks_get_lock_context(inode, type, ctx);
 277        return ctx;
 278}
 279
 280static void
 281locks_dump_ctx_list(struct list_head *list, char *list_type)
 282{
 283        struct file_lock *fl;
 284
 285        list_for_each_entry(fl, list, fl_list) {
 286                pr_warn("%s: fl_owner=%p fl_flags=0x%x fl_type=0x%x fl_pid=%u\n", list_type, fl->fl_owner, fl->fl_flags, fl->fl_type, fl->fl_pid);
 287        }
 288}
 289
 290static void
 291locks_check_ctx_lists(struct inode *inode)
 292{
 293        struct file_lock_context *ctx = inode->i_flctx;
 294
 295        if (unlikely(!list_empty(&ctx->flc_flock) ||
 296                     !list_empty(&ctx->flc_posix) ||
 297                     !list_empty(&ctx->flc_lease))) {
 298                pr_warn("Leaked locks on dev=0x%x:0x%x ino=0x%lx:\n",
 299                        MAJOR(inode->i_sb->s_dev), MINOR(inode->i_sb->s_dev),
 300                        inode->i_ino);
 301                locks_dump_ctx_list(&ctx->flc_flock, "FLOCK");
 302                locks_dump_ctx_list(&ctx->flc_posix, "POSIX");
 303                locks_dump_ctx_list(&ctx->flc_lease, "LEASE");
 304        }
 305}
 306
 307static void
 308locks_check_ctx_file_list(struct file *filp, struct list_head *list,
 309                                char *list_type)
 310{
 311        struct file_lock *fl;
 312        struct inode *inode = locks_inode(filp);
 313
 314        list_for_each_entry(fl, list, fl_list)
 315                if (fl->fl_file == filp)
 316                        pr_warn("Leaked %s lock on dev=0x%x:0x%x ino=0x%lx "
 317                                " fl_owner=%p fl_flags=0x%x fl_type=0x%x fl_pid=%u\n",
 318                                list_type, MAJOR(inode->i_sb->s_dev),
 319                                MINOR(inode->i_sb->s_dev), inode->i_ino,
 320                                fl->fl_owner, fl->fl_flags, fl->fl_type, fl->fl_pid);
 321}
 322
 323void
 324locks_free_lock_context(struct inode *inode)
 325{
 326        struct file_lock_context *ctx = inode->i_flctx;
 327
 328        if (unlikely(ctx)) {
 329                locks_check_ctx_lists(inode);
 330                kmem_cache_free(flctx_cache, ctx);
 331        }
 332}
 333
 334static void locks_init_lock_heads(struct file_lock *fl)
 335{
 336        INIT_HLIST_NODE(&fl->fl_link);
 337        INIT_LIST_HEAD(&fl->fl_list);
 338        INIT_LIST_HEAD(&fl->fl_blocked_requests);
 339        INIT_LIST_HEAD(&fl->fl_blocked_member);
 340        init_waitqueue_head(&fl->fl_wait);
 341}
 342
 343/* Allocate an empty lock structure. */
 344struct file_lock *locks_alloc_lock(void)
 345{
 346        struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
 347
 348        if (fl)
 349                locks_init_lock_heads(fl);
 350
 351        return fl;
 352}
 353EXPORT_SYMBOL_GPL(locks_alloc_lock);
 354
 355void locks_release_private(struct file_lock *fl)
 356{
 357        BUG_ON(waitqueue_active(&fl->fl_wait));
 358        BUG_ON(!list_empty(&fl->fl_list));
 359        BUG_ON(!list_empty(&fl->fl_blocked_requests));
 360        BUG_ON(!list_empty(&fl->fl_blocked_member));
 361        BUG_ON(!hlist_unhashed(&fl->fl_link));
 362
 363        if (fl->fl_ops) {
 364                if (fl->fl_ops->fl_release_private)
 365                        fl->fl_ops->fl_release_private(fl);
 366                fl->fl_ops = NULL;
 367        }
 368
 369        if (fl->fl_lmops) {
 370                if (fl->fl_lmops->lm_put_owner) {
 371                        fl->fl_lmops->lm_put_owner(fl->fl_owner);
 372                        fl->fl_owner = NULL;
 373                }
 374                fl->fl_lmops = NULL;
 375        }
 376}
 377EXPORT_SYMBOL_GPL(locks_release_private);
 378
 379/* Free a lock which is not in use. */
 380void locks_free_lock(struct file_lock *fl)
 381{
 382        locks_release_private(fl);
 383        kmem_cache_free(filelock_cache, fl);
 384}
 385EXPORT_SYMBOL(locks_free_lock);
 386
 387static void
 388locks_dispose_list(struct list_head *dispose)
 389{
 390        struct file_lock *fl;
 391
 392        while (!list_empty(dispose)) {
 393                fl = list_first_entry(dispose, struct file_lock, fl_list);
 394                list_del_init(&fl->fl_list);
 395                locks_free_lock(fl);
 396        }
 397}
 398
 399void locks_init_lock(struct file_lock *fl)
 400{
 401        memset(fl, 0, sizeof(struct file_lock));
 402        locks_init_lock_heads(fl);
 403}
 404EXPORT_SYMBOL(locks_init_lock);
 405
 406/*
 407 * Initialize a new lock from an existing file_lock structure.
 408 */
 409void locks_copy_conflock(struct file_lock *new, struct file_lock *fl)
 410{
 411        new->fl_owner = fl->fl_owner;
 412        new->fl_pid = fl->fl_pid;
 413        new->fl_file = NULL;
 414        new->fl_flags = fl->fl_flags;
 415        new->fl_type = fl->fl_type;
 416        new->fl_start = fl->fl_start;
 417        new->fl_end = fl->fl_end;
 418        new->fl_lmops = fl->fl_lmops;
 419        new->fl_ops = NULL;
 420
 421        if (fl->fl_lmops) {
 422                if (fl->fl_lmops->lm_get_owner)
 423                        fl->fl_lmops->lm_get_owner(fl->fl_owner);
 424        }
 425}
 426EXPORT_SYMBOL(locks_copy_conflock);
 427
 428void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
 429{
 430        /* "new" must be a freshly-initialized lock */
 431        WARN_ON_ONCE(new->fl_ops);
 432
 433        locks_copy_conflock(new, fl);
 434
 435        new->fl_file = fl->fl_file;
 436        new->fl_ops = fl->fl_ops;
 437
 438        if (fl->fl_ops) {
 439                if (fl->fl_ops->fl_copy_lock)
 440                        fl->fl_ops->fl_copy_lock(new, fl);
 441        }
 442}
 443EXPORT_SYMBOL(locks_copy_lock);
 444
 445static void locks_move_blocks(struct file_lock *new, struct file_lock *fl)
 446{
 447        struct file_lock *f;
 448
 449        /*
 450         * As ctx->flc_lock is held, new requests cannot be added to
 451         * ->fl_blocked_requests, so we don't need a lock to check if it
 452         * is empty.
 453         */
 454        if (list_empty(&fl->fl_blocked_requests))
 455                return;
 456        spin_lock(&blocked_lock_lock);
 457        list_splice_init(&fl->fl_blocked_requests, &new->fl_blocked_requests);
 458        list_for_each_entry(f, &new->fl_blocked_requests, fl_blocked_member)
 459                f->fl_blocker = new;
 460        spin_unlock(&blocked_lock_lock);
 461}
 462
 463static inline int flock_translate_cmd(int cmd) {
 464        if (cmd & LOCK_MAND)
 465                return cmd & (LOCK_MAND | LOCK_RW);
 466        switch (cmd) {
 467        case LOCK_SH:
 468                return F_RDLCK;
 469        case LOCK_EX:
 470                return F_WRLCK;
 471        case LOCK_UN:
 472                return F_UNLCK;
 473        }
 474        return -EINVAL;
 475}
 476
 477/* Fill in a file_lock structure with an appropriate FLOCK lock. */
 478static struct file_lock *
 479flock_make_lock(struct file *filp, unsigned int cmd, struct file_lock *fl)
 480{
 481        int type = flock_translate_cmd(cmd);
 482
 483        if (type < 0)
 484                return ERR_PTR(type);
 485
 486        if (fl == NULL) {
 487                fl = locks_alloc_lock();
 488                if (fl == NULL)
 489                        return ERR_PTR(-ENOMEM);
 490        } else {
 491                locks_init_lock(fl);
 492        }
 493
 494        fl->fl_file = filp;
 495        fl->fl_owner = filp;
 496        fl->fl_pid = current->tgid;
 497        fl->fl_flags = FL_FLOCK;
 498        fl->fl_type = type;
 499        fl->fl_end = OFFSET_MAX;
 500
 501        return fl;
 502}
 503
 504static int assign_type(struct file_lock *fl, long type)
 505{
 506        switch (type) {
 507        case F_RDLCK:
 508        case F_WRLCK:
 509        case F_UNLCK:
 510                fl->fl_type = type;
 511                break;
 512        default:
 513                return -EINVAL;
 514        }
 515        return 0;
 516}
 517
 518static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
 519                                 struct flock64 *l)
 520{
 521        switch (l->l_whence) {
 522        case SEEK_SET:
 523                fl->fl_start = 0;
 524                break;
 525        case SEEK_CUR:
 526                fl->fl_start = filp->f_pos;
 527                break;
 528        case SEEK_END:
 529                fl->fl_start = i_size_read(file_inode(filp));
 530                break;
 531        default:
 532                return -EINVAL;
 533        }
 534        if (l->l_start > OFFSET_MAX - fl->fl_start)
 535                return -EOVERFLOW;
 536        fl->fl_start += l->l_start;
 537        if (fl->fl_start < 0)
 538                return -EINVAL;
 539
 540        /* POSIX-1996 leaves the case l->l_len < 0 undefined;
 541           POSIX-2001 defines it. */
 542        if (l->l_len > 0) {
 543                if (l->l_len - 1 > OFFSET_MAX - fl->fl_start)
 544                        return -EOVERFLOW;
 545                fl->fl_end = fl->fl_start + (l->l_len - 1);
 546
 547        } else if (l->l_len < 0) {
 548                if (fl->fl_start + l->l_len < 0)
 549                        return -EINVAL;
 550                fl->fl_end = fl->fl_start - 1;
 551                fl->fl_start += l->l_len;
 552        } else
 553                fl->fl_end = OFFSET_MAX;
 554
 555        fl->fl_owner = current->files;
 556        fl->fl_pid = current->tgid;
 557        fl->fl_file = filp;
 558        fl->fl_flags = FL_POSIX;
 559        fl->fl_ops = NULL;
 560        fl->fl_lmops = NULL;
 561
 562        return assign_type(fl, l->l_type);
 563}
 564
 565/* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
 566 * style lock.
 567 */
 568static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
 569                               struct flock *l)
 570{
 571        struct flock64 ll = {
 572                .l_type = l->l_type,
 573                .l_whence = l->l_whence,
 574                .l_start = l->l_start,
 575                .l_len = l->l_len,
 576        };
 577
 578        return flock64_to_posix_lock(filp, fl, &ll);
 579}
 580
 581/* default lease lock manager operations */
 582static bool
 583lease_break_callback(struct file_lock *fl)
 584{
 585        kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
 586        return false;
 587}
 588
 589static void
 590lease_setup(struct file_lock *fl, void **priv)
 591{
 592        struct file *filp = fl->fl_file;
 593        struct fasync_struct *fa = *priv;
 594
 595        /*
 596         * fasync_insert_entry() returns the old entry if any. If there was no
 597         * old entry, then it used "priv" and inserted it into the fasync list.
 598         * Clear the pointer to indicate that it shouldn't be freed.
 599         */
 600        if (!fasync_insert_entry(fa->fa_fd, filp, &fl->fl_fasync, fa))
 601                *priv = NULL;
 602
 603        __f_setown(filp, task_pid(current), PIDTYPE_TGID, 0);
 604}
 605
 606static const struct lock_manager_operations lease_manager_ops = {
 607        .lm_break = lease_break_callback,
 608        .lm_change = lease_modify,
 609        .lm_setup = lease_setup,
 610};
 611
 612/*
 613 * Initialize a lease, use the default lock manager operations
 614 */
 615static int lease_init(struct file *filp, long type, struct file_lock *fl)
 616{
 617        if (assign_type(fl, type) != 0)
 618                return -EINVAL;
 619
 620        fl->fl_owner = filp;
 621        fl->fl_pid = current->tgid;
 622
 623        fl->fl_file = filp;
 624        fl->fl_flags = FL_LEASE;
 625        fl->fl_start = 0;
 626        fl->fl_end = OFFSET_MAX;
 627        fl->fl_ops = NULL;
 628        fl->fl_lmops = &lease_manager_ops;
 629        return 0;
 630}
 631
 632/* Allocate a file_lock initialised to this type of lease */
 633static struct file_lock *lease_alloc(struct file *filp, long type)
 634{
 635        struct file_lock *fl = locks_alloc_lock();
 636        int error = -ENOMEM;
 637
 638        if (fl == NULL)
 639                return ERR_PTR(error);
 640
 641        error = lease_init(filp, type, fl);
 642        if (error) {
 643                locks_free_lock(fl);
 644                return ERR_PTR(error);
 645        }
 646        return fl;
 647}
 648
 649/* Check if two locks overlap each other.
 650 */
 651static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
 652{
 653        return ((fl1->fl_end >= fl2->fl_start) &&
 654                (fl2->fl_end >= fl1->fl_start));
 655}
 656
 657/*
 658 * Check whether two locks have the same owner.
 659 */
 660static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
 661{
 662        return fl1->fl_owner == fl2->fl_owner;
 663}
 664
 665/* Must be called with the flc_lock held! */
 666static void locks_insert_global_locks(struct file_lock *fl)
 667{
 668        struct file_lock_list_struct *fll = this_cpu_ptr(&file_lock_list);
 669
 670        percpu_rwsem_assert_held(&file_rwsem);
 671
 672        spin_lock(&fll->lock);
 673        fl->fl_link_cpu = smp_processor_id();
 674        hlist_add_head(&fl->fl_link, &fll->hlist);
 675        spin_unlock(&fll->lock);
 676}
 677
 678/* Must be called with the flc_lock held! */
 679static void locks_delete_global_locks(struct file_lock *fl)
 680{
 681        struct file_lock_list_struct *fll;
 682
 683        percpu_rwsem_assert_held(&file_rwsem);
 684
 685        /*
 686         * Avoid taking lock if already unhashed. This is safe since this check
 687         * is done while holding the flc_lock, and new insertions into the list
 688         * also require that it be held.
 689         */
 690        if (hlist_unhashed(&fl->fl_link))
 691                return;
 692
 693        fll = per_cpu_ptr(&file_lock_list, fl->fl_link_cpu);
 694        spin_lock(&fll->lock);
 695        hlist_del_init(&fl->fl_link);
 696        spin_unlock(&fll->lock);
 697}
 698
 699static unsigned long
 700posix_owner_key(struct file_lock *fl)
 701{
 702        return (unsigned long)fl->fl_owner;
 703}
 704
 705static void locks_insert_global_blocked(struct file_lock *waiter)
 706{
 707        lockdep_assert_held(&blocked_lock_lock);
 708
 709        hash_add(blocked_hash, &waiter->fl_link, posix_owner_key(waiter));
 710}
 711
 712static void locks_delete_global_blocked(struct file_lock *waiter)
 713{
 714        lockdep_assert_held(&blocked_lock_lock);
 715
 716        hash_del(&waiter->fl_link);
 717}
 718
 719/* Remove waiter from blocker's block list.
 720 * When blocker ends up pointing to itself then the list is empty.
 721 *
 722 * Must be called with blocked_lock_lock held.
 723 */
 724static void __locks_delete_block(struct file_lock *waiter)
 725{
 726        locks_delete_global_blocked(waiter);
 727        list_del_init(&waiter->fl_blocked_member);
 728}
 729
 730static void __locks_wake_up_blocks(struct file_lock *blocker)
 731{
 732        while (!list_empty(&blocker->fl_blocked_requests)) {
 733                struct file_lock *waiter;
 734
 735                waiter = list_first_entry(&blocker->fl_blocked_requests,
 736                                          struct file_lock, fl_blocked_member);
 737                __locks_delete_block(waiter);
 738                if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
 739                        waiter->fl_lmops->lm_notify(waiter);
 740                else
 741                        wake_up(&waiter->fl_wait);
 742
 743                /*
 744                 * The setting of fl_blocker to NULL marks the "done"
 745                 * point in deleting a block. Paired with acquire at the top
 746                 * of locks_delete_block().
 747                 */
 748                smp_store_release(&waiter->fl_blocker, NULL);
 749        }
 750}
 751
 752/**
 753 *      locks_delete_block - stop waiting for a file lock
 754 *      @waiter: the lock which was waiting
 755 *
 756 *      lockd/nfsd need to disconnect the lock while working on it.
 757 */
 758int locks_delete_block(struct file_lock *waiter)
 759{
 760        int status = -ENOENT;
 761
 762        /*
 763         * If fl_blocker is NULL, it won't be set again as this thread "owns"
 764         * the lock and is the only one that might try to claim the lock.
 765         *
 766         * We use acquire/release to manage fl_blocker so that we can
 767         * optimize away taking the blocked_lock_lock in many cases.
 768         *
 769         * The smp_load_acquire guarantees two things:
 770         *
 771         * 1/ that fl_blocked_requests can be tested locklessly. If something
 772         * was recently added to that list it must have been in a locked region
 773         * *before* the locked region when fl_blocker was set to NULL.
 774         *
 775         * 2/ that no other thread is accessing 'waiter', so it is safe to free
 776         * it.  __locks_wake_up_blocks is careful not to touch waiter after
 777         * fl_blocker is released.
 778         *
 779         * If a lockless check of fl_blocker shows it to be NULL, we know that
 780         * no new locks can be inserted into its fl_blocked_requests list, and
 781         * can avoid doing anything further if the list is empty.
 782         */
 783        if (!smp_load_acquire(&waiter->fl_blocker) &&
 784            list_empty(&waiter->fl_blocked_requests))
 785                return status;
 786
 787        spin_lock(&blocked_lock_lock);
 788        if (waiter->fl_blocker)
 789                status = 0;
 790        __locks_wake_up_blocks(waiter);
 791        __locks_delete_block(waiter);
 792
 793        /*
 794         * The setting of fl_blocker to NULL marks the "done" point in deleting
 795         * a block. Paired with acquire at the top of this function.
 796         */
 797        smp_store_release(&waiter->fl_blocker, NULL);
 798        spin_unlock(&blocked_lock_lock);
 799        return status;
 800}
 801EXPORT_SYMBOL(locks_delete_block);
 802
 803/* Insert waiter into blocker's block list.
 804 * We use a circular list so that processes can be easily woken up in
 805 * the order they blocked. The documentation doesn't require this but
 806 * it seems like the reasonable thing to do.
 807 *
 808 * Must be called with both the flc_lock and blocked_lock_lock held. The
 809 * fl_blocked_requests list itself is protected by the blocked_lock_lock,
 810 * but by ensuring that the flc_lock is also held on insertions we can avoid
 811 * taking the blocked_lock_lock in some cases when we see that the
 812 * fl_blocked_requests list is empty.
 813 *
 814 * Rather than just adding to the list, we check for conflicts with any existing
 815 * waiters, and add beneath any waiter that blocks the new waiter.
 816 * Thus wakeups don't happen until needed.
 817 */
 818static void __locks_insert_block(struct file_lock *blocker,
 819                                 struct file_lock *waiter,
 820                                 bool conflict(struct file_lock *,
 821                                               struct file_lock *))
 822{
 823        struct file_lock *fl;
 824        BUG_ON(!list_empty(&waiter->fl_blocked_member));
 825
 826new_blocker:
 827        list_for_each_entry(fl, &blocker->fl_blocked_requests, fl_blocked_member)
 828                if (conflict(fl, waiter)) {
 829                        blocker =  fl;
 830                        goto new_blocker;
 831                }
 832        waiter->fl_blocker = blocker;
 833        list_add_tail(&waiter->fl_blocked_member, &blocker->fl_blocked_requests);
 834        if (IS_POSIX(blocker) && !IS_OFDLCK(blocker))
 835                locks_insert_global_blocked(waiter);
 836
 837        /* The requests in waiter->fl_blocked are known to conflict with
 838         * waiter, but might not conflict with blocker, or the requests
 839         * and lock which block it.  So they all need to be woken.
 840         */
 841        __locks_wake_up_blocks(waiter);
 842}
 843
 844/* Must be called with flc_lock held. */
 845static void locks_insert_block(struct file_lock *blocker,
 846                               struct file_lock *waiter,
 847                               bool conflict(struct file_lock *,
 848                                             struct file_lock *))
 849{
 850        spin_lock(&blocked_lock_lock);
 851        __locks_insert_block(blocker, waiter, conflict);
 852        spin_unlock(&blocked_lock_lock);
 853}
 854
 855/*
 856 * Wake up processes blocked waiting for blocker.
 857 *
 858 * Must be called with the inode->flc_lock held!
 859 */
 860static void locks_wake_up_blocks(struct file_lock *blocker)
 861{
 862        /*
 863         * Avoid taking global lock if list is empty. This is safe since new
 864         * blocked requests are only added to the list under the flc_lock, and
 865         * the flc_lock is always held here. Note that removal from the
 866         * fl_blocked_requests list does not require the flc_lock, so we must
 867         * recheck list_empty() after acquiring the blocked_lock_lock.
 868         */
 869        if (list_empty(&blocker->fl_blocked_requests))
 870                return;
 871
 872        spin_lock(&blocked_lock_lock);
 873        __locks_wake_up_blocks(blocker);
 874        spin_unlock(&blocked_lock_lock);
 875}
 876
 877static void
 878locks_insert_lock_ctx(struct file_lock *fl, struct list_head *before)
 879{
 880        list_add_tail(&fl->fl_list, before);
 881        locks_insert_global_locks(fl);
 882}
 883
 884static void
 885locks_unlink_lock_ctx(struct file_lock *fl)
 886{
 887        locks_delete_global_locks(fl);
 888        list_del_init(&fl->fl_list);
 889        locks_wake_up_blocks(fl);
 890}
 891
 892static void
 893locks_delete_lock_ctx(struct file_lock *fl, struct list_head *dispose)
 894{
 895        locks_unlink_lock_ctx(fl);
 896        if (dispose)
 897                list_add(&fl->fl_list, dispose);
 898        else
 899                locks_free_lock(fl);
 900}
 901
 902/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
 903 * checks for shared/exclusive status of overlapping locks.
 904 */
 905static bool locks_conflict(struct file_lock *caller_fl,
 906                           struct file_lock *sys_fl)
 907{
 908        if (sys_fl->fl_type == F_WRLCK)
 909                return true;
 910        if (caller_fl->fl_type == F_WRLCK)
 911                return true;
 912        return false;
 913}
 914
 915/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
 916 * checking before calling the locks_conflict().
 917 */
 918static bool posix_locks_conflict(struct file_lock *caller_fl,
 919                                 struct file_lock *sys_fl)
 920{
 921        /* POSIX locks owned by the same process do not conflict with
 922         * each other.
 923         */
 924        if (posix_same_owner(caller_fl, sys_fl))
 925                return false;
 926
 927        /* Check whether they overlap */
 928        if (!locks_overlap(caller_fl, sys_fl))
 929                return false;
 930
 931        return locks_conflict(caller_fl, sys_fl);
 932}
 933
 934/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
 935 * checking before calling the locks_conflict().
 936 */
 937static bool flock_locks_conflict(struct file_lock *caller_fl,
 938                                 struct file_lock *sys_fl)
 939{
 940        /* FLOCK locks referring to the same filp do not conflict with
 941         * each other.
 942         */
 943        if (caller_fl->fl_file == sys_fl->fl_file)
 944                return false;
 945        if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
 946                return false;
 947
 948        return locks_conflict(caller_fl, sys_fl);
 949}
 950
 951void
 952posix_test_lock(struct file *filp, struct file_lock *fl)
 953{
 954        struct file_lock *cfl;
 955        struct file_lock_context *ctx;
 956        struct inode *inode = locks_inode(filp);
 957
 958        ctx = smp_load_acquire(&inode->i_flctx);
 959        if (!ctx || list_empty_careful(&ctx->flc_posix)) {
 960                fl->fl_type = F_UNLCK;
 961                return;
 962        }
 963
 964        spin_lock(&ctx->flc_lock);
 965        list_for_each_entry(cfl, &ctx->flc_posix, fl_list) {
 966                if (posix_locks_conflict(fl, cfl)) {
 967                        locks_copy_conflock(fl, cfl);
 968                        goto out;
 969                }
 970        }
 971        fl->fl_type = F_UNLCK;
 972out:
 973        spin_unlock(&ctx->flc_lock);
 974        return;
 975}
 976EXPORT_SYMBOL(posix_test_lock);
 977
 978/*
 979 * Deadlock detection:
 980 *
 981 * We attempt to detect deadlocks that are due purely to posix file
 982 * locks.
 983 *
 984 * We assume that a task can be waiting for at most one lock at a time.
 985 * So for any acquired lock, the process holding that lock may be
 986 * waiting on at most one other lock.  That lock in turns may be held by
 987 * someone waiting for at most one other lock.  Given a requested lock
 988 * caller_fl which is about to wait for a conflicting lock block_fl, we
 989 * follow this chain of waiters to ensure we are not about to create a
 990 * cycle.
 991 *
 992 * Since we do this before we ever put a process to sleep on a lock, we
 993 * are ensured that there is never a cycle; that is what guarantees that
 994 * the while() loop in posix_locks_deadlock() eventually completes.
 995 *
 996 * Note: the above assumption may not be true when handling lock
 997 * requests from a broken NFS client. It may also fail in the presence
 998 * of tasks (such as posix threads) sharing the same open file table.
 999 * To handle those cases, we just bail out after a few iterations.
1000 *
1001 * For FL_OFDLCK locks, the owner is the filp, not the files_struct.
1002 * Because the owner is not even nominally tied to a thread of
1003 * execution, the deadlock detection below can't reasonably work well. Just
1004 * skip it for those.
1005 *
1006 * In principle, we could do a more limited deadlock detection on FL_OFDLCK
1007 * locks that just checks for the case where two tasks are attempting to
1008 * upgrade from read to write locks on the same inode.
1009 */
1010
1011#define MAX_DEADLK_ITERATIONS 10
1012
1013/* Find a lock that the owner of the given block_fl is blocking on. */
1014static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
1015{
1016        struct file_lock *fl;
1017
1018        hash_for_each_possible(blocked_hash, fl, fl_link, posix_owner_key(block_fl)) {
1019                if (posix_same_owner(fl, block_fl)) {
1020                        while (fl->fl_blocker)
1021                                fl = fl->fl_blocker;
1022                        return fl;
1023                }
1024        }
1025        return NULL;
1026}
1027
1028/* Must be called with the blocked_lock_lock held! */
1029static int posix_locks_deadlock(struct file_lock *caller_fl,
1030                                struct file_lock *block_fl)
1031{
1032        int i = 0;
1033
1034        lockdep_assert_held(&blocked_lock_lock);
1035
1036        /*
1037         * This deadlock detector can't reasonably detect deadlocks with
1038         * FL_OFDLCK locks, since they aren't owned by a process, per-se.
1039         */
1040        if (IS_OFDLCK(caller_fl))
1041                return 0;
1042
1043        while ((block_fl = what_owner_is_waiting_for(block_fl))) {
1044                if (i++ > MAX_DEADLK_ITERATIONS)
1045                        return 0;
1046                if (posix_same_owner(caller_fl, block_fl))
1047                        return 1;
1048        }
1049        return 0;
1050}
1051
1052/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
1053 * after any leases, but before any posix locks.
1054 *
1055 * Note that if called with an FL_EXISTS argument, the caller may determine
1056 * whether or not a lock was successfully freed by testing the return
1057 * value for -ENOENT.
1058 */
1059static int flock_lock_inode(struct inode *inode, struct file_lock *request)
1060{
1061        struct file_lock *new_fl = NULL;
1062        struct file_lock *fl;
1063        struct file_lock_context *ctx;
1064        int error = 0;
1065        bool found = false;
1066        LIST_HEAD(dispose);
1067
1068        ctx = locks_get_lock_context(inode, request->fl_type);
1069        if (!ctx) {
1070                if (request->fl_type != F_UNLCK)
1071                        return -ENOMEM;
1072                return (request->fl_flags & FL_EXISTS) ? -ENOENT : 0;
1073        }
1074
1075        if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
1076                new_fl = locks_alloc_lock();
1077                if (!new_fl)
1078                        return -ENOMEM;
1079        }
1080
1081        percpu_down_read(&file_rwsem);
1082        spin_lock(&ctx->flc_lock);
1083        if (request->fl_flags & FL_ACCESS)
1084                goto find_conflict;
1085
1086        list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
1087                if (request->fl_file != fl->fl_file)
1088                        continue;
1089                if (request->fl_type == fl->fl_type)
1090                        goto out;
1091                found = true;
1092                locks_delete_lock_ctx(fl, &dispose);
1093                break;
1094        }
1095
1096        if (request->fl_type == F_UNLCK) {
1097                if ((request->fl_flags & FL_EXISTS) && !found)
1098                        error = -ENOENT;
1099                goto out;
1100        }
1101
1102find_conflict:
1103        list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
1104                if (!flock_locks_conflict(request, fl))
1105                        continue;
1106                error = -EAGAIN;
1107                if (!(request->fl_flags & FL_SLEEP))
1108                        goto out;
1109                error = FILE_LOCK_DEFERRED;
1110                locks_insert_block(fl, request, flock_locks_conflict);
1111                goto out;
1112        }
1113        if (request->fl_flags & FL_ACCESS)
1114                goto out;
1115        locks_copy_lock(new_fl, request);
1116        locks_move_blocks(new_fl, request);
1117        locks_insert_lock_ctx(new_fl, &ctx->flc_flock);
1118        new_fl = NULL;
1119        error = 0;
1120
1121out:
1122        spin_unlock(&ctx->flc_lock);
1123        percpu_up_read(&file_rwsem);
1124        if (new_fl)
1125                locks_free_lock(new_fl);
1126        locks_dispose_list(&dispose);
1127        trace_flock_lock_inode(inode, request, error);
1128        return error;
1129}
1130
1131static int posix_lock_inode(struct inode *inode, struct file_lock *request,
1132                            struct file_lock *conflock)
1133{
1134        struct file_lock *fl, *tmp;
1135        struct file_lock *new_fl = NULL;
1136        struct file_lock *new_fl2 = NULL;
1137        struct file_lock *left = NULL;
1138        struct file_lock *right = NULL;
1139        struct file_lock_context *ctx;
1140        int error;
1141        bool added = false;
1142        LIST_HEAD(dispose);
1143
1144        ctx = locks_get_lock_context(inode, request->fl_type);
1145        if (!ctx)
1146                return (request->fl_type == F_UNLCK) ? 0 : -ENOMEM;
1147
1148        /*
1149         * We may need two file_lock structures for this operation,
1150         * so we get them in advance to avoid races.
1151         *
1152         * In some cases we can be sure, that no new locks will be needed
1153         */
1154        if (!(request->fl_flags & FL_ACCESS) &&
1155            (request->fl_type != F_UNLCK ||
1156             request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
1157                new_fl = locks_alloc_lock();
1158                new_fl2 = locks_alloc_lock();
1159        }
1160
1161        percpu_down_read(&file_rwsem);
1162        spin_lock(&ctx->flc_lock);
1163        /*
1164         * New lock request. Walk all POSIX locks and look for conflicts. If
1165         * there are any, either return error or put the request on the
1166         * blocker's list of waiters and the global blocked_hash.
1167         */
1168        if (request->fl_type != F_UNLCK) {
1169                list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1170                        if (!posix_locks_conflict(request, fl))
1171                                continue;
1172                        if (conflock)
1173                                locks_copy_conflock(conflock, fl);
1174                        error = -EAGAIN;
1175                        if (!(request->fl_flags & FL_SLEEP))
1176                                goto out;
1177                        /*
1178                         * Deadlock detection and insertion into the blocked
1179                         * locks list must be done while holding the same lock!
1180                         */
1181                        error = -EDEADLK;
1182                        spin_lock(&blocked_lock_lock);
1183                        /*
1184                         * Ensure that we don't find any locks blocked on this
1185                         * request during deadlock detection.
1186                         */
1187                        __locks_wake_up_blocks(request);
1188                        if (likely(!posix_locks_deadlock(request, fl))) {
1189                                error = FILE_LOCK_DEFERRED;
1190                                __locks_insert_block(fl, request,
1191                                                     posix_locks_conflict);
1192                        }
1193                        spin_unlock(&blocked_lock_lock);
1194                        goto out;
1195                }
1196        }
1197
1198        /* If we're just looking for a conflict, we're done. */
1199        error = 0;
1200        if (request->fl_flags & FL_ACCESS)
1201                goto out;
1202
1203        /* Find the first old lock with the same owner as the new lock */
1204        list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1205                if (posix_same_owner(request, fl))
1206                        break;
1207        }
1208
1209        /* Process locks with this owner. */
1210        list_for_each_entry_safe_from(fl, tmp, &ctx->flc_posix, fl_list) {
1211                if (!posix_same_owner(request, fl))
1212                        break;
1213
1214                /* Detect adjacent or overlapping regions (if same lock type) */
1215                if (request->fl_type == fl->fl_type) {
1216                        /* In all comparisons of start vs end, use
1217                         * "start - 1" rather than "end + 1". If end
1218                         * is OFFSET_MAX, end + 1 will become negative.
1219                         */
1220                        if (fl->fl_end < request->fl_start - 1)
1221                                continue;
1222                        /* If the next lock in the list has entirely bigger
1223                         * addresses than the new one, insert the lock here.
1224                         */
1225                        if (fl->fl_start - 1 > request->fl_end)
1226                                break;
1227
1228                        /* If we come here, the new and old lock are of the
1229                         * same type and adjacent or overlapping. Make one
1230                         * lock yielding from the lower start address of both
1231                         * locks to the higher end address.
1232                         */
1233                        if (fl->fl_start > request->fl_start)
1234                                fl->fl_start = request->fl_start;
1235                        else
1236                                request->fl_start = fl->fl_start;
1237                        if (fl->fl_end < request->fl_end)
1238                                fl->fl_end = request->fl_end;
1239                        else
1240                                request->fl_end = fl->fl_end;
1241                        if (added) {
1242                                locks_delete_lock_ctx(fl, &dispose);
1243                                continue;
1244                        }
1245                        request = fl;
1246                        added = true;
1247                } else {
1248                        /* Processing for different lock types is a bit
1249                         * more complex.
1250                         */
1251                        if (fl->fl_end < request->fl_start)
1252                                continue;
1253                        if (fl->fl_start > request->fl_end)
1254                                break;
1255                        if (request->fl_type == F_UNLCK)
1256                                added = true;
1257                        if (fl->fl_start < request->fl_start)
1258                                left = fl;
1259                        /* If the next lock in the list has a higher end
1260                         * address than the new one, insert the new one here.
1261                         */
1262                        if (fl->fl_end > request->fl_end) {
1263                                right = fl;
1264                                break;
1265                        }
1266                        if (fl->fl_start >= request->fl_start) {
1267                                /* The new lock completely replaces an old
1268                                 * one (This may happen several times).
1269                                 */
1270                                if (added) {
1271                                        locks_delete_lock_ctx(fl, &dispose);
1272                                        continue;
1273                                }
1274                                /*
1275                                 * Replace the old lock with new_fl, and
1276                                 * remove the old one. It's safe to do the
1277                                 * insert here since we know that we won't be
1278                                 * using new_fl later, and that the lock is
1279                                 * just replacing an existing lock.
1280                                 */
1281                                error = -ENOLCK;
1282                                if (!new_fl)
1283                                        goto out;
1284                                locks_copy_lock(new_fl, request);
1285                                locks_move_blocks(new_fl, request);
1286                                request = new_fl;
1287                                new_fl = NULL;
1288                                locks_insert_lock_ctx(request, &fl->fl_list);
1289                                locks_delete_lock_ctx(fl, &dispose);
1290                                added = true;
1291                        }
1292                }
1293        }
1294
1295        /*
1296         * The above code only modifies existing locks in case of merging or
1297         * replacing. If new lock(s) need to be inserted all modifications are
1298         * done below this, so it's safe yet to bail out.
1299         */
1300        error = -ENOLCK; /* "no luck" */
1301        if (right && left == right && !new_fl2)
1302                goto out;
1303
1304        error = 0;
1305        if (!added) {
1306                if (request->fl_type == F_UNLCK) {
1307                        if (request->fl_flags & FL_EXISTS)
1308                                error = -ENOENT;
1309                        goto out;
1310                }
1311
1312                if (!new_fl) {
1313                        error = -ENOLCK;
1314                        goto out;
1315                }
1316                locks_copy_lock(new_fl, request);
1317                locks_move_blocks(new_fl, request);
1318                locks_insert_lock_ctx(new_fl, &fl->fl_list);
1319                fl = new_fl;
1320                new_fl = NULL;
1321        }
1322        if (right) {
1323                if (left == right) {
1324                        /* The new lock breaks the old one in two pieces,
1325                         * so we have to use the second new lock.
1326                         */
1327                        left = new_fl2;
1328                        new_fl2 = NULL;
1329                        locks_copy_lock(left, right);
1330                        locks_insert_lock_ctx(left, &fl->fl_list);
1331                }
1332                right->fl_start = request->fl_end + 1;
1333                locks_wake_up_blocks(right);
1334        }
1335        if (left) {
1336                left->fl_end = request->fl_start - 1;
1337                locks_wake_up_blocks(left);
1338        }
1339 out:
1340        spin_unlock(&ctx->flc_lock);
1341        percpu_up_read(&file_rwsem);
1342        /*
1343         * Free any unused locks.
1344         */
1345        if (new_fl)
1346                locks_free_lock(new_fl);
1347        if (new_fl2)
1348                locks_free_lock(new_fl2);
1349        locks_dispose_list(&dispose);
1350        trace_posix_lock_inode(inode, request, error);
1351
1352        return error;
1353}
1354
1355/**
1356 * posix_lock_file - Apply a POSIX-style lock to a file
1357 * @filp: The file to apply the lock to
1358 * @fl: The lock to be applied
1359 * @conflock: Place to return a copy of the conflicting lock, if found.
1360 *
1361 * Add a POSIX style lock to a file.
1362 * We merge adjacent & overlapping locks whenever possible.
1363 * POSIX locks are sorted by owner task, then by starting address
1364 *
1365 * Note that if called with an FL_EXISTS argument, the caller may determine
1366 * whether or not a lock was successfully freed by testing the return
1367 * value for -ENOENT.
1368 */
1369int posix_lock_file(struct file *filp, struct file_lock *fl,
1370                        struct file_lock *conflock)
1371{
1372        return posix_lock_inode(locks_inode(filp), fl, conflock);
1373}
1374EXPORT_SYMBOL(posix_lock_file);
1375
1376/**
1377 * posix_lock_inode_wait - Apply a POSIX-style lock to a file
1378 * @inode: inode of file to which lock request should be applied
1379 * @fl: The lock to be applied
1380 *
1381 * Apply a POSIX style lock request to an inode.
1382 */
1383static int posix_lock_inode_wait(struct inode *inode, struct file_lock *fl)
1384{
1385        int error;
1386        might_sleep ();
1387        for (;;) {
1388                error = posix_lock_inode(inode, fl, NULL);
1389                if (error != FILE_LOCK_DEFERRED)
1390                        break;
1391                error = wait_event_interruptible(fl->fl_wait,
1392                                        list_empty(&fl->fl_blocked_member));
1393                if (error)
1394                        break;
1395        }
1396        locks_delete_block(fl);
1397        return error;
1398}
1399
1400#ifdef CONFIG_MANDATORY_FILE_LOCKING
1401/**
1402 * locks_mandatory_locked - Check for an active lock
1403 * @file: the file to check
1404 *
1405 * Searches the inode's list of locks to find any POSIX locks which conflict.
1406 * This function is called from locks_verify_locked() only.
1407 */
1408int locks_mandatory_locked(struct file *file)
1409{
1410        int ret;
1411        struct inode *inode = locks_inode(file);
1412        struct file_lock_context *ctx;
1413        struct file_lock *fl;
1414
1415        ctx = smp_load_acquire(&inode->i_flctx);
1416        if (!ctx || list_empty_careful(&ctx->flc_posix))
1417                return 0;
1418
1419        /*
1420         * Search the lock list for this inode for any POSIX locks.
1421         */
1422        spin_lock(&ctx->flc_lock);
1423        ret = 0;
1424        list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1425                if (fl->fl_owner != current->files &&
1426                    fl->fl_owner != file) {
1427                        ret = -EAGAIN;
1428                        break;
1429                }
1430        }
1431        spin_unlock(&ctx->flc_lock);
1432        return ret;
1433}
1434
1435/**
1436 * locks_mandatory_area - Check for a conflicting lock
1437 * @inode:      the file to check
1438 * @filp:       how the file was opened (if it was)
1439 * @start:      first byte in the file to check
1440 * @end:        lastbyte in the file to check
1441 * @type:       %F_WRLCK for a write lock, else %F_RDLCK
1442 *
1443 * Searches the inode's list of locks to find any POSIX locks which conflict.
1444 */
1445int locks_mandatory_area(struct inode *inode, struct file *filp, loff_t start,
1446                         loff_t end, unsigned char type)
1447{
1448        struct file_lock fl;
1449        int error;
1450        bool sleep = false;
1451
1452        locks_init_lock(&fl);
1453        fl.fl_pid = current->tgid;
1454        fl.fl_file = filp;
1455        fl.fl_flags = FL_POSIX | FL_ACCESS;
1456        if (filp && !(filp->f_flags & O_NONBLOCK))
1457                sleep = true;
1458        fl.fl_type = type;
1459        fl.fl_start = start;
1460        fl.fl_end = end;
1461
1462        for (;;) {
1463                if (filp) {
1464                        fl.fl_owner = filp;
1465                        fl.fl_flags &= ~FL_SLEEP;
1466                        error = posix_lock_inode(inode, &fl, NULL);
1467                        if (!error)
1468                                break;
1469                }
1470
1471                if (sleep)
1472                        fl.fl_flags |= FL_SLEEP;
1473                fl.fl_owner = current->files;
1474                error = posix_lock_inode(inode, &fl, NULL);
1475                if (error != FILE_LOCK_DEFERRED)
1476                        break;
1477                error = wait_event_interruptible(fl.fl_wait,
1478                                        list_empty(&fl.fl_blocked_member));
1479                if (!error) {
1480                        /*
1481                         * If we've been sleeping someone might have
1482                         * changed the permissions behind our back.
1483                         */
1484                        if (__mandatory_lock(inode))
1485                                continue;
1486                }
1487
1488                break;
1489        }
1490        locks_delete_block(&fl);
1491
1492        return error;
1493}
1494EXPORT_SYMBOL(locks_mandatory_area);
1495#endif /* CONFIG_MANDATORY_FILE_LOCKING */
1496
1497static void lease_clear_pending(struct file_lock *fl, int arg)
1498{
1499        switch (arg) {
1500        case F_UNLCK:
1501                fl->fl_flags &= ~FL_UNLOCK_PENDING;
1502                fallthrough;
1503        case F_RDLCK:
1504                fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1505        }
1506}
1507
1508/* We already had a lease on this file; just change its type */
1509int lease_modify(struct file_lock *fl, int arg, struct list_head *dispose)
1510{
1511        int error = assign_type(fl, arg);
1512
1513        if (error)
1514                return error;
1515        lease_clear_pending(fl, arg);
1516        locks_wake_up_blocks(fl);
1517        if (arg == F_UNLCK) {
1518                struct file *filp = fl->fl_file;
1519
1520                f_delown(filp);
1521                filp->f_owner.signum = 0;
1522                fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
1523                if (fl->fl_fasync != NULL) {
1524                        printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1525                        fl->fl_fasync = NULL;
1526                }
1527                locks_delete_lock_ctx(fl, dispose);
1528        }
1529        return 0;
1530}
1531EXPORT_SYMBOL(lease_modify);
1532
1533static bool past_time(unsigned long then)
1534{
1535        if (!then)
1536                /* 0 is a special value meaning "this never expires": */
1537                return false;
1538        return time_after(jiffies, then);
1539}
1540
1541static void time_out_leases(struct inode *inode, struct list_head *dispose)
1542{
1543        struct file_lock_context *ctx = inode->i_flctx;
1544        struct file_lock *fl, *tmp;
1545
1546        lockdep_assert_held(&ctx->flc_lock);
1547
1548        list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
1549                trace_time_out_leases(inode, fl);
1550                if (past_time(fl->fl_downgrade_time))
1551                        lease_modify(fl, F_RDLCK, dispose);
1552                if (past_time(fl->fl_break_time))
1553                        lease_modify(fl, F_UNLCK, dispose);
1554        }
1555}
1556
1557static bool leases_conflict(struct file_lock *lease, struct file_lock *breaker)
1558{
1559        bool rc;
1560
1561        if (lease->fl_lmops->lm_breaker_owns_lease
1562                        && lease->fl_lmops->lm_breaker_owns_lease(lease))
1563                return false;
1564        if ((breaker->fl_flags & FL_LAYOUT) != (lease->fl_flags & FL_LAYOUT)) {
1565                rc = false;
1566                goto trace;
1567        }
1568        if ((breaker->fl_flags & FL_DELEG) && (lease->fl_flags & FL_LEASE)) {
1569                rc = false;
1570                goto trace;
1571        }
1572
1573        rc = locks_conflict(breaker, lease);
1574trace:
1575        trace_leases_conflict(rc, lease, breaker);
1576        return rc;
1577}
1578
1579static bool
1580any_leases_conflict(struct inode *inode, struct file_lock *breaker)
1581{
1582        struct file_lock_context *ctx = inode->i_flctx;
1583        struct file_lock *fl;
1584
1585        lockdep_assert_held(&ctx->flc_lock);
1586
1587        list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1588                if (leases_conflict(fl, breaker))
1589                        return true;
1590        }
1591        return false;
1592}
1593
1594/**
1595 *      __break_lease   -       revoke all outstanding leases on file
1596 *      @inode: the inode of the file to return
1597 *      @mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR:
1598 *          break all leases
1599 *      @type: FL_LEASE: break leases and delegations; FL_DELEG: break
1600 *          only delegations
1601 *
1602 *      break_lease (inlined for speed) has checked there already is at least
1603 *      some kind of lock (maybe a lease) on this file.  Leases are broken on
1604 *      a call to open() or truncate().  This function can sleep unless you
1605 *      specified %O_NONBLOCK to your open().
1606 */
1607int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
1608{
1609        int error = 0;
1610        struct file_lock_context *ctx;
1611        struct file_lock *new_fl, *fl, *tmp;
1612        unsigned long break_time;
1613        int want_write = (mode & O_ACCMODE) != O_RDONLY;
1614        LIST_HEAD(dispose);
1615
1616        new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
1617        if (IS_ERR(new_fl))
1618                return PTR_ERR(new_fl);
1619        new_fl->fl_flags = type;
1620
1621        /* typically we will check that ctx is non-NULL before calling */
1622        ctx = smp_load_acquire(&inode->i_flctx);
1623        if (!ctx) {
1624                WARN_ON_ONCE(1);
1625                goto free_lock;
1626        }
1627
1628        percpu_down_read(&file_rwsem);
1629        spin_lock(&ctx->flc_lock);
1630
1631        time_out_leases(inode, &dispose);
1632
1633        if (!any_leases_conflict(inode, new_fl))
1634                goto out;
1635
1636        break_time = 0;
1637        if (lease_break_time > 0) {
1638                break_time = jiffies + lease_break_time * HZ;
1639                if (break_time == 0)
1640                        break_time++;   /* so that 0 means no break time */
1641        }
1642
1643        list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
1644                if (!leases_conflict(fl, new_fl))
1645                        continue;
1646                if (want_write) {
1647                        if (fl->fl_flags & FL_UNLOCK_PENDING)
1648                                continue;
1649                        fl->fl_flags |= FL_UNLOCK_PENDING;
1650                        fl->fl_break_time = break_time;
1651                } else {
1652                        if (lease_breaking(fl))
1653                                continue;
1654                        fl->fl_flags |= FL_DOWNGRADE_PENDING;
1655                        fl->fl_downgrade_time = break_time;
1656                }
1657                if (fl->fl_lmops->lm_break(fl))
1658                        locks_delete_lock_ctx(fl, &dispose);
1659        }
1660
1661        if (list_empty(&ctx->flc_lease))
1662                goto out;
1663
1664        if (mode & O_NONBLOCK) {
1665                trace_break_lease_noblock(inode, new_fl);
1666                error = -EWOULDBLOCK;
1667                goto out;
1668        }
1669
1670restart:
1671        fl = list_first_entry(&ctx->flc_lease, struct file_lock, fl_list);
1672        break_time = fl->fl_break_time;
1673        if (break_time != 0)
1674                break_time -= jiffies;
1675        if (break_time == 0)
1676                break_time++;
1677        locks_insert_block(fl, new_fl, leases_conflict);
1678        trace_break_lease_block(inode, new_fl);
1679        spin_unlock(&ctx->flc_lock);
1680        percpu_up_read(&file_rwsem);
1681
1682        locks_dispose_list(&dispose);
1683        error = wait_event_interruptible_timeout(new_fl->fl_wait,
1684                                        list_empty(&new_fl->fl_blocked_member),
1685                                        break_time);
1686
1687        percpu_down_read(&file_rwsem);
1688        spin_lock(&ctx->flc_lock);
1689        trace_break_lease_unblock(inode, new_fl);
1690        locks_delete_block(new_fl);
1691        if (error >= 0) {
1692                /*
1693                 * Wait for the next conflicting lease that has not been
1694                 * broken yet
1695                 */
1696                if (error == 0)
1697                        time_out_leases(inode, &dispose);
1698                if (any_leases_conflict(inode, new_fl))
1699                        goto restart;
1700                error = 0;
1701        }
1702out:
1703        spin_unlock(&ctx->flc_lock);
1704        percpu_up_read(&file_rwsem);
1705        locks_dispose_list(&dispose);
1706free_lock:
1707        locks_free_lock(new_fl);
1708        return error;
1709}
1710EXPORT_SYMBOL(__break_lease);
1711
1712/**
1713 *      lease_get_mtime - update modified time of an inode with exclusive lease
1714 *      @inode: the inode
1715 *      @time:  pointer to a timespec which contains the last modified time
1716 *
1717 * This is to force NFS clients to flush their caches for files with
1718 * exclusive leases.  The justification is that if someone has an
1719 * exclusive lease, then they could be modifying it.
1720 */
1721void lease_get_mtime(struct inode *inode, struct timespec64 *time)
1722{
1723        bool has_lease = false;
1724        struct file_lock_context *ctx;
1725        struct file_lock *fl;
1726
1727        ctx = smp_load_acquire(&inode->i_flctx);
1728        if (ctx && !list_empty_careful(&ctx->flc_lease)) {
1729                spin_lock(&ctx->flc_lock);
1730                fl = list_first_entry_or_null(&ctx->flc_lease,
1731                                              struct file_lock, fl_list);
1732                if (fl && (fl->fl_type == F_WRLCK))
1733                        has_lease = true;
1734                spin_unlock(&ctx->flc_lock);
1735        }
1736
1737        if (has_lease)
1738                *time = current_time(inode);
1739}
1740EXPORT_SYMBOL(lease_get_mtime);
1741
1742/**
1743 *      fcntl_getlease - Enquire what lease is currently active
1744 *      @filp: the file
1745 *
1746 *      The value returned by this function will be one of
1747 *      (if no lease break is pending):
1748 *
1749 *      %F_RDLCK to indicate a shared lease is held.
1750 *
1751 *      %F_WRLCK to indicate an exclusive lease is held.
1752 *
1753 *      %F_UNLCK to indicate no lease is held.
1754 *
1755 *      (if a lease break is pending):
1756 *
1757 *      %F_RDLCK to indicate an exclusive lease needs to be
1758 *              changed to a shared lease (or removed).
1759 *
1760 *      %F_UNLCK to indicate the lease needs to be removed.
1761 *
1762 *      XXX: sfr & willy disagree over whether F_INPROGRESS
1763 *      should be returned to userspace.
1764 */
1765int fcntl_getlease(struct file *filp)
1766{
1767        struct file_lock *fl;
1768        struct inode *inode = locks_inode(filp);
1769        struct file_lock_context *ctx;
1770        int type = F_UNLCK;
1771        LIST_HEAD(dispose);
1772
1773        ctx = smp_load_acquire(&inode->i_flctx);
1774        if (ctx && !list_empty_careful(&ctx->flc_lease)) {
1775                percpu_down_read(&file_rwsem);
1776                spin_lock(&ctx->flc_lock);
1777                time_out_leases(inode, &dispose);
1778                list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1779                        if (fl->fl_file != filp)
1780                                continue;
1781                        type = target_leasetype(fl);
1782                        break;
1783                }
1784                spin_unlock(&ctx->flc_lock);
1785                percpu_up_read(&file_rwsem);
1786
1787                locks_dispose_list(&dispose);
1788        }
1789        return type;
1790}
1791
1792/**
1793 * check_conflicting_open - see if the given file points to an inode that has
1794 *                          an existing open that would conflict with the
1795 *                          desired lease.
1796 * @filp:       file to check
1797 * @arg:        type of lease that we're trying to acquire
1798 * @flags:      current lock flags
1799 *
1800 * Check to see if there's an existing open fd on this file that would
1801 * conflict with the lease we're trying to set.
1802 */
1803static int
1804check_conflicting_open(struct file *filp, const long arg, int flags)
1805{
1806        struct inode *inode = locks_inode(filp);
1807        int self_wcount = 0, self_rcount = 0;
1808
1809        if (flags & FL_LAYOUT)
1810                return 0;
1811        if (flags & FL_DELEG)
1812                /* We leave these checks to the caller */
1813                return 0;
1814
1815        if (arg == F_RDLCK)
1816                return inode_is_open_for_write(inode) ? -EAGAIN : 0;
1817        else if (arg != F_WRLCK)
1818                return 0;
1819
1820        /*
1821         * Make sure that only read/write count is from lease requestor.
1822         * Note that this will result in denying write leases when i_writecount
1823         * is negative, which is what we want.  (We shouldn't grant write leases
1824         * on files open for execution.)
1825         */
1826        if (filp->f_mode & FMODE_WRITE)
1827                self_wcount = 1;
1828        else if (filp->f_mode & FMODE_READ)
1829                self_rcount = 1;
1830
1831        if (atomic_read(&inode->i_writecount) != self_wcount ||
1832            atomic_read(&inode->i_readcount) != self_rcount)
1833                return -EAGAIN;
1834
1835        return 0;
1836}
1837
1838static int
1839generic_add_lease(struct file *filp, long arg, struct file_lock **flp, void **priv)
1840{
1841        struct file_lock *fl, *my_fl = NULL, *lease;
1842        struct inode *inode = locks_inode(filp);
1843        struct file_lock_context *ctx;
1844        bool is_deleg = (*flp)->fl_flags & FL_DELEG;
1845        int error;
1846        LIST_HEAD(dispose);
1847
1848        lease = *flp;
1849        trace_generic_add_lease(inode, lease);
1850
1851        /* Note that arg is never F_UNLCK here */
1852        ctx = locks_get_lock_context(inode, arg);
1853        if (!ctx)
1854                return -ENOMEM;
1855
1856        /*
1857         * In the delegation case we need mutual exclusion with
1858         * a number of operations that take the i_mutex.  We trylock
1859         * because delegations are an optional optimization, and if
1860         * there's some chance of a conflict--we'd rather not
1861         * bother, maybe that's a sign this just isn't a good file to
1862         * hand out a delegation on.
1863         */
1864        if (is_deleg && !inode_trylock(inode))
1865                return -EAGAIN;
1866
1867        if (is_deleg && arg == F_WRLCK) {
1868                /* Write delegations are not currently supported: */
1869                inode_unlock(inode);
1870                WARN_ON_ONCE(1);
1871                return -EINVAL;
1872        }
1873
1874        percpu_down_read(&file_rwsem);
1875        spin_lock(&ctx->flc_lock);
1876        time_out_leases(inode, &dispose);
1877        error = check_conflicting_open(filp, arg, lease->fl_flags);
1878        if (error)
1879                goto out;
1880
1881        /*
1882         * At this point, we know that if there is an exclusive
1883         * lease on this file, then we hold it on this filp
1884         * (otherwise our open of this file would have blocked).
1885         * And if we are trying to acquire an exclusive lease,
1886         * then the file is not open by anyone (including us)
1887         * except for this filp.
1888         */
1889        error = -EAGAIN;
1890        list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1891                if (fl->fl_file == filp &&
1892                    fl->fl_owner == lease->fl_owner) {
1893                        my_fl = fl;
1894                        continue;
1895                }
1896
1897                /*
1898                 * No exclusive leases if someone else has a lease on
1899                 * this file:
1900                 */
1901                if (arg == F_WRLCK)
1902                        goto out;
1903                /*
1904                 * Modifying our existing lease is OK, but no getting a
1905                 * new lease if someone else is opening for write:
1906                 */
1907                if (fl->fl_flags & FL_UNLOCK_PENDING)
1908                        goto out;
1909        }
1910
1911        if (my_fl != NULL) {
1912                lease = my_fl;
1913                error = lease->fl_lmops->lm_change(lease, arg, &dispose);
1914                if (error)
1915                        goto out;
1916                goto out_setup;
1917        }
1918
1919        error = -EINVAL;
1920        if (!leases_enable)
1921                goto out;
1922
1923        locks_insert_lock_ctx(lease, &ctx->flc_lease);
1924        /*
1925         * The check in break_lease() is lockless. It's possible for another
1926         * open to race in after we did the earlier check for a conflicting
1927         * open but before the lease was inserted. Check again for a
1928         * conflicting open and cancel the lease if there is one.
1929         *
1930         * We also add a barrier here to ensure that the insertion of the lock
1931         * precedes these checks.
1932         */
1933        smp_mb();
1934        error = check_conflicting_open(filp, arg, lease->fl_flags);
1935        if (error) {
1936                locks_unlink_lock_ctx(lease);
1937                goto out;
1938        }
1939
1940out_setup:
1941        if (lease->fl_lmops->lm_setup)
1942                lease->fl_lmops->lm_setup(lease, priv);
1943out:
1944        spin_unlock(&ctx->flc_lock);
1945        percpu_up_read(&file_rwsem);
1946        locks_dispose_list(&dispose);
1947        if (is_deleg)
1948                inode_unlock(inode);
1949        if (!error && !my_fl)
1950                *flp = NULL;
1951        return error;
1952}
1953
1954static int generic_delete_lease(struct file *filp, void *owner)
1955{
1956        int error = -EAGAIN;
1957        struct file_lock *fl, *victim = NULL;
1958        struct inode *inode = locks_inode(filp);
1959        struct file_lock_context *ctx;
1960        LIST_HEAD(dispose);
1961
1962        ctx = smp_load_acquire(&inode->i_flctx);
1963        if (!ctx) {
1964                trace_generic_delete_lease(inode, NULL);
1965                return error;
1966        }
1967
1968        percpu_down_read(&file_rwsem);
1969        spin_lock(&ctx->flc_lock);
1970        list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1971                if (fl->fl_file == filp &&
1972                    fl->fl_owner == owner) {
1973                        victim = fl;
1974                        break;
1975                }
1976        }
1977        trace_generic_delete_lease(inode, victim);
1978        if (victim)
1979                error = fl->fl_lmops->lm_change(victim, F_UNLCK, &dispose);
1980        spin_unlock(&ctx->flc_lock);
1981        percpu_up_read(&file_rwsem);
1982        locks_dispose_list(&dispose);
1983        return error;
1984}
1985
1986/**
1987 *      generic_setlease        -       sets a lease on an open file
1988 *      @filp:  file pointer
1989 *      @arg:   type of lease to obtain
1990 *      @flp:   input - file_lock to use, output - file_lock inserted
1991 *      @priv:  private data for lm_setup (may be NULL if lm_setup
1992 *              doesn't require it)
1993 *
1994 *      The (input) flp->fl_lmops->lm_break function is required
1995 *      by break_lease().
1996 */
1997int generic_setlease(struct file *filp, long arg, struct file_lock **flp,
1998                        void **priv)
1999{
2000        struct inode *inode = locks_inode(filp);
2001        int error;
2002
2003        if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
2004                return -EACCES;
2005        if (!S_ISREG(inode->i_mode))
2006                return -EINVAL;
2007        error = security_file_lock(filp, arg);
2008        if (error)
2009                return error;
2010
2011        switch (arg) {
2012        case F_UNLCK:
2013                return generic_delete_lease(filp, *priv);
2014        case F_RDLCK:
2015        case F_WRLCK:
2016                if (!(*flp)->fl_lmops->lm_break) {
2017                        WARN_ON_ONCE(1);
2018                        return -ENOLCK;
2019                }
2020
2021                return generic_add_lease(filp, arg, flp, priv);
2022        default:
2023                return -EINVAL;
2024        }
2025}
2026EXPORT_SYMBOL(generic_setlease);
2027
2028#if IS_ENABLED(CONFIG_SRCU)
2029/*
2030 * Kernel subsystems can register to be notified on any attempt to set
2031 * a new lease with the lease_notifier_chain. This is used by (e.g.) nfsd
2032 * to close files that it may have cached when there is an attempt to set a
2033 * conflicting lease.
2034 */
2035static struct srcu_notifier_head lease_notifier_chain;
2036
2037static inline void
2038lease_notifier_chain_init(void)
2039{
2040        srcu_init_notifier_head(&lease_notifier_chain);
2041}
2042
2043static inline void
2044setlease_notifier(long arg, struct file_lock *lease)
2045{
2046        if (arg != F_UNLCK)
2047                srcu_notifier_call_chain(&lease_notifier_chain, arg, lease);
2048}
2049
2050int lease_register_notifier(struct notifier_block *nb)
2051{
2052        return srcu_notifier_chain_register(&lease_notifier_chain, nb);
2053}
2054EXPORT_SYMBOL_GPL(lease_register_notifier);
2055
2056void lease_unregister_notifier(struct notifier_block *nb)
2057{
2058        srcu_notifier_chain_unregister(&lease_notifier_chain, nb);
2059}
2060EXPORT_SYMBOL_GPL(lease_unregister_notifier);
2061
2062#else /* !IS_ENABLED(CONFIG_SRCU) */
2063static inline void
2064lease_notifier_chain_init(void)
2065{
2066}
2067
2068static inline void
2069setlease_notifier(long arg, struct file_lock *lease)
2070{
2071}
2072
2073int lease_register_notifier(struct notifier_block *nb)
2074{
2075        return 0;
2076}
2077EXPORT_SYMBOL_GPL(lease_register_notifier);
2078
2079void lease_unregister_notifier(struct notifier_block *nb)
2080{
2081}
2082EXPORT_SYMBOL_GPL(lease_unregister_notifier);
2083
2084#endif /* IS_ENABLED(CONFIG_SRCU) */
2085
2086/**
2087 * vfs_setlease        -       sets a lease on an open file
2088 * @filp:       file pointer
2089 * @arg:        type of lease to obtain
2090 * @lease:      file_lock to use when adding a lease
2091 * @priv:       private info for lm_setup when adding a lease (may be
2092 *              NULL if lm_setup doesn't require it)
2093 *
2094 * Call this to establish a lease on the file. The "lease" argument is not
2095 * used for F_UNLCK requests and may be NULL. For commands that set or alter
2096 * an existing lease, the ``(*lease)->fl_lmops->lm_break`` operation must be
2097 * set; if not, this function will return -ENOLCK (and generate a scary-looking
2098 * stack trace).
2099 *
2100 * The "priv" pointer is passed directly to the lm_setup function as-is. It
2101 * may be NULL if the lm_setup operation doesn't require it.
2102 */
2103int
2104vfs_setlease(struct file *filp, long arg, struct file_lock **lease, void **priv)
2105{
2106        if (lease)
2107                setlease_notifier(arg, *lease);
2108        if (filp->f_op->setlease)
2109                return filp->f_op->setlease(filp, arg, lease, priv);
2110        else
2111                return generic_setlease(filp, arg, lease, priv);
2112}
2113EXPORT_SYMBOL_GPL(vfs_setlease);
2114
2115static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
2116{
2117        struct file_lock *fl;
2118        struct fasync_struct *new;
2119        int error;
2120
2121        fl = lease_alloc(filp, arg);
2122        if (IS_ERR(fl))
2123                return PTR_ERR(fl);
2124
2125        new = fasync_alloc();
2126        if (!new) {
2127                locks_free_lock(fl);
2128                return -ENOMEM;
2129        }
2130        new->fa_fd = fd;
2131
2132        error = vfs_setlease(filp, arg, &fl, (void **)&new);
2133        if (fl)
2134                locks_free_lock(fl);
2135        if (new)
2136                fasync_free(new);
2137        return error;
2138}
2139
2140/**
2141 *      fcntl_setlease  -       sets a lease on an open file
2142 *      @fd: open file descriptor
2143 *      @filp: file pointer
2144 *      @arg: type of lease to obtain
2145 *
2146 *      Call this fcntl to establish a lease on the file.
2147 *      Note that you also need to call %F_SETSIG to
2148 *      receive a signal when the lease is broken.
2149 */
2150int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
2151{
2152        if (arg == F_UNLCK)
2153                return vfs_setlease(filp, F_UNLCK, NULL, (void **)&filp);
2154        return do_fcntl_add_lease(fd, filp, arg);
2155}
2156
2157/**
2158 * flock_lock_inode_wait - Apply a FLOCK-style lock to a file
2159 * @inode: inode of the file to apply to
2160 * @fl: The lock to be applied
2161 *
2162 * Apply a FLOCK style lock request to an inode.
2163 */
2164static int flock_lock_inode_wait(struct inode *inode, struct file_lock *fl)
2165{
2166        int error;
2167        might_sleep();
2168        for (;;) {
2169                error = flock_lock_inode(inode, fl);
2170                if (error != FILE_LOCK_DEFERRED)
2171                        break;
2172                error = wait_event_interruptible(fl->fl_wait,
2173                                list_empty(&fl->fl_blocked_member));
2174                if (error)
2175                        break;
2176        }
2177        locks_delete_block(fl);
2178        return error;
2179}
2180
2181/**
2182 * locks_lock_inode_wait - Apply a lock to an inode
2183 * @inode: inode of the file to apply to
2184 * @fl: The lock to be applied
2185 *
2186 * Apply a POSIX or FLOCK style lock request to an inode.
2187 */
2188int locks_lock_inode_wait(struct inode *inode, struct file_lock *fl)
2189{
2190        int res = 0;
2191        switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
2192                case FL_POSIX:
2193                        res = posix_lock_inode_wait(inode, fl);
2194                        break;
2195                case FL_FLOCK:
2196                        res = flock_lock_inode_wait(inode, fl);
2197                        break;
2198                default:
2199                        BUG();
2200        }
2201        return res;
2202}
2203EXPORT_SYMBOL(locks_lock_inode_wait);
2204
2205/**
2206 *      sys_flock: - flock() system call.
2207 *      @fd: the file descriptor to lock.
2208 *      @cmd: the type of lock to apply.
2209 *
2210 *      Apply a %FL_FLOCK style lock to an open file descriptor.
2211 *      The @cmd can be one of:
2212 *
2213 *      - %LOCK_SH -- a shared lock.
2214 *      - %LOCK_EX -- an exclusive lock.
2215 *      - %LOCK_UN -- remove an existing lock.
2216 *      - %LOCK_MAND -- a 'mandatory' flock.
2217 *        This exists to emulate Windows Share Modes.
2218 *
2219 *      %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
2220 *      processes read and write access respectively.
2221 */
2222SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
2223{
2224        struct fd f = fdget(fd);
2225        struct file_lock *lock;
2226        int can_sleep, unlock;
2227        int error;
2228
2229        error = -EBADF;
2230        if (!f.file)
2231                goto out;
2232
2233        can_sleep = !(cmd & LOCK_NB);
2234        cmd &= ~LOCK_NB;
2235        unlock = (cmd == LOCK_UN);
2236
2237        if (!unlock && !(cmd & LOCK_MAND) &&
2238            !(f.file->f_mode & (FMODE_READ|FMODE_WRITE)))
2239                goto out_putf;
2240
2241        lock = flock_make_lock(f.file, cmd, NULL);
2242        if (IS_ERR(lock)) {
2243                error = PTR_ERR(lock);
2244                goto out_putf;
2245        }
2246
2247        if (can_sleep)
2248                lock->fl_flags |= FL_SLEEP;
2249
2250        error = security_file_lock(f.file, lock->fl_type);
2251        if (error)
2252                goto out_free;
2253
2254        if (f.file->f_op->flock)
2255                error = f.file->f_op->flock(f.file,
2256                                          (can_sleep) ? F_SETLKW : F_SETLK,
2257                                          lock);
2258        else
2259                error = locks_lock_file_wait(f.file, lock);
2260
2261 out_free:
2262        locks_free_lock(lock);
2263
2264 out_putf:
2265        fdput(f);
2266 out:
2267        return error;
2268}
2269
2270/**
2271 * vfs_test_lock - test file byte range lock
2272 * @filp: The file to test lock for
2273 * @fl: The lock to test; also used to hold result
2274 *
2275 * Returns -ERRNO on failure.  Indicates presence of conflicting lock by
2276 * setting conf->fl_type to something other than F_UNLCK.
2277 */
2278int vfs_test_lock(struct file *filp, struct file_lock *fl)
2279{
2280        if (filp->f_op->lock)
2281                return filp->f_op->lock(filp, F_GETLK, fl);
2282        posix_test_lock(filp, fl);
2283        return 0;
2284}
2285EXPORT_SYMBOL_GPL(vfs_test_lock);
2286
2287/**
2288 * locks_translate_pid - translate a file_lock's fl_pid number into a namespace
2289 * @fl: The file_lock who's fl_pid should be translated
2290 * @ns: The namespace into which the pid should be translated
2291 *
2292 * Used to tranlate a fl_pid into a namespace virtual pid number
2293 */
2294static pid_t locks_translate_pid(struct file_lock *fl, struct pid_namespace *ns)
2295{
2296        pid_t vnr;
2297        struct pid *pid;
2298
2299        if (IS_OFDLCK(fl))
2300                return -1;
2301        if (IS_REMOTELCK(fl))
2302                return fl->fl_pid;
2303        /*
2304         * If the flock owner process is dead and its pid has been already
2305         * freed, the translation below won't work, but we still want to show
2306         * flock owner pid number in init pidns.
2307         */
2308        if (ns == &init_pid_ns)
2309                return (pid_t)fl->fl_pid;
2310
2311        rcu_read_lock();
2312        pid = find_pid_ns(fl->fl_pid, &init_pid_ns);
2313        vnr = pid_nr_ns(pid, ns);
2314        rcu_read_unlock();
2315        return vnr;
2316}
2317
2318static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
2319{
2320        flock->l_pid = locks_translate_pid(fl, task_active_pid_ns(current));
2321#if BITS_PER_LONG == 32
2322        /*
2323         * Make sure we can represent the posix lock via
2324         * legacy 32bit flock.
2325         */
2326        if (fl->fl_start > OFFT_OFFSET_MAX)
2327                return -EOVERFLOW;
2328        if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
2329                return -EOVERFLOW;
2330#endif
2331        flock->l_start = fl->fl_start;
2332        flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2333                fl->fl_end - fl->fl_start + 1;
2334        flock->l_whence = 0;
2335        flock->l_type = fl->fl_type;
2336        return 0;
2337}
2338
2339#if BITS_PER_LONG == 32
2340static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
2341{
2342        flock->l_pid = locks_translate_pid(fl, task_active_pid_ns(current));
2343        flock->l_start = fl->fl_start;
2344        flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2345                fl->fl_end - fl->fl_start + 1;
2346        flock->l_whence = 0;
2347        flock->l_type = fl->fl_type;
2348}
2349#endif
2350
2351/* Report the first existing lock that would conflict with l.
2352 * This implements the F_GETLK command of fcntl().
2353 */
2354int fcntl_getlk(struct file *filp, unsigned int cmd, struct flock *flock)
2355{
2356        struct file_lock *fl;
2357        int error;
2358
2359        fl = locks_alloc_lock();
2360        if (fl == NULL)
2361                return -ENOMEM;
2362        error = -EINVAL;
2363        if (flock->l_type != F_RDLCK && flock->l_type != F_WRLCK)
2364                goto out;
2365
2366        error = flock_to_posix_lock(filp, fl, flock);
2367        if (error)
2368                goto out;
2369
2370        if (cmd == F_OFD_GETLK) {
2371                error = -EINVAL;
2372                if (flock->l_pid != 0)
2373                        goto out;
2374
2375                fl->fl_flags |= FL_OFDLCK;
2376                fl->fl_owner = filp;
2377        }
2378
2379        error = vfs_test_lock(filp, fl);
2380        if (error)
2381                goto out;
2382
2383        flock->l_type = fl->fl_type;
2384        if (fl->fl_type != F_UNLCK) {
2385                error = posix_lock_to_flock(flock, fl);
2386                if (error)
2387                        goto out;
2388        }
2389out:
2390        locks_free_lock(fl);
2391        return error;
2392}
2393
2394/**
2395 * vfs_lock_file - file byte range lock
2396 * @filp: The file to apply the lock to
2397 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
2398 * @fl: The lock to be applied
2399 * @conf: Place to return a copy of the conflicting lock, if found.
2400 *
2401 * A caller that doesn't care about the conflicting lock may pass NULL
2402 * as the final argument.
2403 *
2404 * If the filesystem defines a private ->lock() method, then @conf will
2405 * be left unchanged; so a caller that cares should initialize it to
2406 * some acceptable default.
2407 *
2408 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
2409 * locks, the ->lock() interface may return asynchronously, before the lock has
2410 * been granted or denied by the underlying filesystem, if (and only if)
2411 * lm_grant is set. Callers expecting ->lock() to return asynchronously
2412 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
2413 * the request is for a blocking lock. When ->lock() does return asynchronously,
2414 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
2415 * request completes.
2416 * If the request is for non-blocking lock the file system should return
2417 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
2418 * with the result. If the request timed out the callback routine will return a
2419 * nonzero return code and the file system should release the lock. The file
2420 * system is also responsible to keep a corresponding posix lock when it
2421 * grants a lock so the VFS can find out which locks are locally held and do
2422 * the correct lock cleanup when required.
2423 * The underlying filesystem must not drop the kernel lock or call
2424 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
2425 * return code.
2426 */
2427int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
2428{
2429        if (filp->f_op->lock)
2430                return filp->f_op->lock(filp, cmd, fl);
2431        else
2432                return posix_lock_file(filp, fl, conf);
2433}
2434EXPORT_SYMBOL_GPL(vfs_lock_file);
2435
2436static int do_lock_file_wait(struct file *filp, unsigned int cmd,
2437                             struct file_lock *fl)
2438{
2439        int error;
2440
2441        error = security_file_lock(filp, fl->fl_type);
2442        if (error)
2443                return error;
2444
2445        for (;;) {
2446                error = vfs_lock_file(filp, cmd, fl, NULL);
2447                if (error != FILE_LOCK_DEFERRED)
2448                        break;
2449                error = wait_event_interruptible(fl->fl_wait,
2450                                        list_empty(&fl->fl_blocked_member));
2451                if (error)
2452                        break;
2453        }
2454        locks_delete_block(fl);
2455
2456        return error;
2457}
2458
2459/* Ensure that fl->fl_file has compatible f_mode for F_SETLK calls */
2460static int
2461check_fmode_for_setlk(struct file_lock *fl)
2462{
2463        switch (fl->fl_type) {
2464        case F_RDLCK:
2465                if (!(fl->fl_file->f_mode & FMODE_READ))
2466                        return -EBADF;
2467                break;
2468        case F_WRLCK:
2469                if (!(fl->fl_file->f_mode & FMODE_WRITE))
2470                        return -EBADF;
2471        }
2472        return 0;
2473}
2474
2475/* Apply the lock described by l to an open file descriptor.
2476 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2477 */
2478int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
2479                struct flock *flock)
2480{
2481        struct file_lock *file_lock = locks_alloc_lock();
2482        struct inode *inode = locks_inode(filp);
2483        struct file *f;
2484        int error;
2485
2486        if (file_lock == NULL)
2487                return -ENOLCK;
2488
2489        /* Don't allow mandatory locks on files that may be memory mapped
2490         * and shared.
2491         */
2492        if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
2493                error = -EAGAIN;
2494                goto out;
2495        }
2496
2497        error = flock_to_posix_lock(filp, file_lock, flock);
2498        if (error)
2499                goto out;
2500
2501        error = check_fmode_for_setlk(file_lock);
2502        if (error)
2503                goto out;
2504
2505        /*
2506         * If the cmd is requesting file-private locks, then set the
2507         * FL_OFDLCK flag and override the owner.
2508         */
2509        switch (cmd) {
2510        case F_OFD_SETLK:
2511                error = -EINVAL;
2512                if (flock->l_pid != 0)
2513                        goto out;
2514
2515                cmd = F_SETLK;
2516                file_lock->fl_flags |= FL_OFDLCK;
2517                file_lock->fl_owner = filp;
2518                break;
2519        case F_OFD_SETLKW:
2520                error = -EINVAL;
2521                if (flock->l_pid != 0)
2522                        goto out;
2523
2524                cmd = F_SETLKW;
2525                file_lock->fl_flags |= FL_OFDLCK;
2526                file_lock->fl_owner = filp;
2527                fallthrough;
2528        case F_SETLKW:
2529                file_lock->fl_flags |= FL_SLEEP;
2530        }
2531
2532        error = do_lock_file_wait(filp, cmd, file_lock);
2533
2534        /*
2535         * Attempt to detect a close/fcntl race and recover by releasing the
2536         * lock that was just acquired. There is no need to do that when we're
2537         * unlocking though, or for OFD locks.
2538         */
2539        if (!error && file_lock->fl_type != F_UNLCK &&
2540            !(file_lock->fl_flags & FL_OFDLCK)) {
2541                struct files_struct *files = current->files;
2542                /*
2543                 * We need that spin_lock here - it prevents reordering between
2544                 * update of i_flctx->flc_posix and check for it done in
2545                 * close(). rcu_read_lock() wouldn't do.
2546                 */
2547                spin_lock(&files->file_lock);
2548                f = files_lookup_fd_locked(files, fd);
2549                spin_unlock(&files->file_lock);
2550                if (f != filp) {
2551                        file_lock->fl_type = F_UNLCK;
2552                        error = do_lock_file_wait(filp, cmd, file_lock);
2553                        WARN_ON_ONCE(error);
2554                        error = -EBADF;
2555                }
2556        }
2557out:
2558        trace_fcntl_setlk(inode, file_lock, error);
2559        locks_free_lock(file_lock);
2560        return error;
2561}
2562
2563#if BITS_PER_LONG == 32
2564/* Report the first existing lock that would conflict with l.
2565 * This implements the F_GETLK command of fcntl().
2566 */
2567int fcntl_getlk64(struct file *filp, unsigned int cmd, struct flock64 *flock)
2568{
2569        struct file_lock *fl;
2570        int error;
2571
2572        fl = locks_alloc_lock();
2573        if (fl == NULL)
2574                return -ENOMEM;
2575
2576        error = -EINVAL;
2577        if (flock->l_type != F_RDLCK && flock->l_type != F_WRLCK)
2578                goto out;
2579
2580        error = flock64_to_posix_lock(filp, fl, flock);
2581        if (error)
2582                goto out;
2583
2584        if (cmd == F_OFD_GETLK) {
2585                error = -EINVAL;
2586                if (flock->l_pid != 0)
2587                        goto out;
2588
2589                cmd = F_GETLK64;
2590                fl->fl_flags |= FL_OFDLCK;
2591                fl->fl_owner = filp;
2592        }
2593
2594        error = vfs_test_lock(filp, fl);
2595        if (error)
2596                goto out;
2597
2598        flock->l_type = fl->fl_type;
2599        if (fl->fl_type != F_UNLCK)
2600                posix_lock_to_flock64(flock, fl);
2601
2602out:
2603        locks_free_lock(fl);
2604        return error;
2605}
2606
2607/* Apply the lock described by l to an open file descriptor.
2608 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2609 */
2610int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
2611                struct flock64 *flock)
2612{
2613        struct file_lock *file_lock = locks_alloc_lock();
2614        struct inode *inode = locks_inode(filp);
2615        struct file *f;
2616        int error;
2617
2618        if (file_lock == NULL)
2619                return -ENOLCK;
2620
2621        /* Don't allow mandatory locks on files that may be memory mapped
2622         * and shared.
2623         */
2624        if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
2625                error = -EAGAIN;
2626                goto out;
2627        }
2628
2629        error = flock64_to_posix_lock(filp, file_lock, flock);
2630        if (error)
2631                goto out;
2632
2633        error = check_fmode_for_setlk(file_lock);
2634        if (error)
2635                goto out;
2636
2637        /*
2638         * If the cmd is requesting file-private locks, then set the
2639         * FL_OFDLCK flag and override the owner.
2640         */
2641        switch (cmd) {
2642        case F_OFD_SETLK:
2643                error = -EINVAL;
2644                if (flock->l_pid != 0)
2645                        goto out;
2646
2647                cmd = F_SETLK64;
2648                file_lock->fl_flags |= FL_OFDLCK;
2649                file_lock->fl_owner = filp;
2650                break;
2651        case F_OFD_SETLKW:
2652                error = -EINVAL;
2653                if (flock->l_pid != 0)
2654                        goto out;
2655
2656                cmd = F_SETLKW64;
2657                file_lock->fl_flags |= FL_OFDLCK;
2658                file_lock->fl_owner = filp;
2659                fallthrough;
2660        case F_SETLKW64:
2661                file_lock->fl_flags |= FL_SLEEP;
2662        }
2663
2664        error = do_lock_file_wait(filp, cmd, file_lock);
2665
2666        /*
2667         * Attempt to detect a close/fcntl race and recover by releasing the
2668         * lock that was just acquired. There is no need to do that when we're
2669         * unlocking though, or for OFD locks.
2670         */
2671        if (!error && file_lock->fl_type != F_UNLCK &&
2672            !(file_lock->fl_flags & FL_OFDLCK)) {
2673                struct files_struct *files = current->files;
2674                /*
2675                 * We need that spin_lock here - it prevents reordering between
2676                 * update of i_flctx->flc_posix and check for it done in
2677                 * close(). rcu_read_lock() wouldn't do.
2678                 */
2679                spin_lock(&files->file_lock);
2680                f = files_lookup_fd_locked(files, fd);
2681                spin_unlock(&files->file_lock);
2682                if (f != filp) {
2683                        file_lock->fl_type = F_UNLCK;
2684                        error = do_lock_file_wait(filp, cmd, file_lock);
2685                        WARN_ON_ONCE(error);
2686                        error = -EBADF;
2687                }
2688        }
2689out:
2690        locks_free_lock(file_lock);
2691        return error;
2692}
2693#endif /* BITS_PER_LONG == 32 */
2694
2695/*
2696 * This function is called when the file is being removed
2697 * from the task's fd array.  POSIX locks belonging to this task
2698 * are deleted at this time.
2699 */
2700void locks_remove_posix(struct file *filp, fl_owner_t owner)
2701{
2702        int error;
2703        struct inode *inode = locks_inode(filp);
2704        struct file_lock lock;
2705        struct file_lock_context *ctx;
2706
2707        /*
2708         * If there are no locks held on this file, we don't need to call
2709         * posix_lock_file().  Another process could be setting a lock on this
2710         * file at the same time, but we wouldn't remove that lock anyway.
2711         */
2712        ctx =  smp_load_acquire(&inode->i_flctx);
2713        if (!ctx || list_empty(&ctx->flc_posix))
2714                return;
2715
2716        locks_init_lock(&lock);
2717        lock.fl_type = F_UNLCK;
2718        lock.fl_flags = FL_POSIX | FL_CLOSE;
2719        lock.fl_start = 0;
2720        lock.fl_end = OFFSET_MAX;
2721        lock.fl_owner = owner;
2722        lock.fl_pid = current->tgid;
2723        lock.fl_file = filp;
2724        lock.fl_ops = NULL;
2725        lock.fl_lmops = NULL;
2726
2727        error = vfs_lock_file(filp, F_SETLK, &lock, NULL);
2728
2729        if (lock.fl_ops && lock.fl_ops->fl_release_private)
2730                lock.fl_ops->fl_release_private(&lock);
2731        trace_locks_remove_posix(inode, &lock, error);
2732}
2733EXPORT_SYMBOL(locks_remove_posix);
2734
2735/* The i_flctx must be valid when calling into here */
2736static void
2737locks_remove_flock(struct file *filp, struct file_lock_context *flctx)
2738{
2739        struct file_lock fl;
2740        struct inode *inode = locks_inode(filp);
2741
2742        if (list_empty(&flctx->flc_flock))
2743                return;
2744
2745        flock_make_lock(filp, LOCK_UN, &fl);
2746        fl.fl_flags |= FL_CLOSE;
2747
2748        if (filp->f_op->flock)
2749                filp->f_op->flock(filp, F_SETLKW, &fl);
2750        else
2751                flock_lock_inode(inode, &fl);
2752
2753        if (fl.fl_ops && fl.fl_ops->fl_release_private)
2754                fl.fl_ops->fl_release_private(&fl);
2755}
2756
2757/* The i_flctx must be valid when calling into here */
2758static void
2759locks_remove_lease(struct file *filp, struct file_lock_context *ctx)
2760{
2761        struct file_lock *fl, *tmp;
2762        LIST_HEAD(dispose);
2763
2764        if (list_empty(&ctx->flc_lease))
2765                return;
2766
2767        percpu_down_read(&file_rwsem);
2768        spin_lock(&ctx->flc_lock);
2769        list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list)
2770                if (filp == fl->fl_file)
2771                        lease_modify(fl, F_UNLCK, &dispose);
2772        spin_unlock(&ctx->flc_lock);
2773        percpu_up_read(&file_rwsem);
2774
2775        locks_dispose_list(&dispose);
2776}
2777
2778/*
2779 * This function is called on the last close of an open file.
2780 */
2781void locks_remove_file(struct file *filp)
2782{
2783        struct file_lock_context *ctx;
2784
2785        ctx = smp_load_acquire(&locks_inode(filp)->i_flctx);
2786        if (!ctx)
2787                return;
2788
2789        /* remove any OFD locks */
2790        locks_remove_posix(filp, filp);
2791
2792        /* remove flock locks */
2793        locks_remove_flock(filp, ctx);
2794
2795        /* remove any leases */
2796        locks_remove_lease(filp, ctx);
2797
2798        spin_lock(&ctx->flc_lock);
2799        locks_check_ctx_file_list(filp, &ctx->flc_posix, "POSIX");
2800        locks_check_ctx_file_list(filp, &ctx->flc_flock, "FLOCK");
2801        locks_check_ctx_file_list(filp, &ctx->flc_lease, "LEASE");
2802        spin_unlock(&ctx->flc_lock);
2803}
2804
2805/**
2806 * vfs_cancel_lock - file byte range unblock lock
2807 * @filp: The file to apply the unblock to
2808 * @fl: The lock to be unblocked
2809 *
2810 * Used by lock managers to cancel blocked requests
2811 */
2812int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2813{
2814        if (filp->f_op->lock)
2815                return filp->f_op->lock(filp, F_CANCELLK, fl);
2816        return 0;
2817}
2818EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2819
2820#ifdef CONFIG_PROC_FS
2821#include <linux/proc_fs.h>
2822#include <linux/seq_file.h>
2823
2824struct locks_iterator {
2825        int     li_cpu;
2826        loff_t  li_pos;
2827};
2828
2829static void lock_get_status(struct seq_file *f, struct file_lock *fl,
2830                            loff_t id, char *pfx, int repeat)
2831{
2832        struct inode *inode = NULL;
2833        unsigned int fl_pid;
2834        struct pid_namespace *proc_pidns = proc_pid_ns(file_inode(f->file)->i_sb);
2835
2836        fl_pid = locks_translate_pid(fl, proc_pidns);
2837        /*
2838         * If lock owner is dead (and pid is freed) or not visible in current
2839         * pidns, zero is shown as a pid value. Check lock info from
2840         * init_pid_ns to get saved lock pid value.
2841         */
2842
2843        if (fl->fl_file != NULL)
2844                inode = locks_inode(fl->fl_file);
2845
2846        seq_printf(f, "%lld: ", id);
2847
2848        if (repeat)
2849                seq_printf(f, "%*s", repeat - 1 + (int)strlen(pfx), pfx);
2850
2851        if (IS_POSIX(fl)) {
2852                if (fl->fl_flags & FL_ACCESS)
2853                        seq_puts(f, "ACCESS");
2854                else if (IS_OFDLCK(fl))
2855                        seq_puts(f, "OFDLCK");
2856                else
2857                        seq_puts(f, "POSIX ");
2858
2859                seq_printf(f, " %s ",
2860                             (inode == NULL) ? "*NOINODE*" :
2861                             mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
2862        } else if (IS_FLOCK(fl)) {
2863                if (fl->fl_type & LOCK_MAND) {
2864                        seq_puts(f, "FLOCK  MSNFS     ");
2865                } else {
2866                        seq_puts(f, "FLOCK  ADVISORY  ");
2867                }
2868        } else if (IS_LEASE(fl)) {
2869                if (fl->fl_flags & FL_DELEG)
2870                        seq_puts(f, "DELEG  ");
2871                else
2872                        seq_puts(f, "LEASE  ");
2873
2874                if (lease_breaking(fl))
2875                        seq_puts(f, "BREAKING  ");
2876                else if (fl->fl_file)
2877                        seq_puts(f, "ACTIVE    ");
2878                else
2879                        seq_puts(f, "BREAKER   ");
2880        } else {
2881                seq_puts(f, "UNKNOWN UNKNOWN  ");
2882        }
2883        if (fl->fl_type & LOCK_MAND) {
2884                seq_printf(f, "%s ",
2885                               (fl->fl_type & LOCK_READ)
2886                               ? (fl->fl_type & LOCK_WRITE) ? "RW   " : "READ "
2887                               : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2888        } else {
2889                int type = IS_LEASE(fl) ? target_leasetype(fl) : fl->fl_type;
2890
2891                seq_printf(f, "%s ", (type == F_WRLCK) ? "WRITE" :
2892                                     (type == F_RDLCK) ? "READ" : "UNLCK");
2893        }
2894        if (inode) {
2895                /* userspace relies on this representation of dev_t */
2896                seq_printf(f, "%d %02x:%02x:%lu ", fl_pid,
2897                                MAJOR(inode->i_sb->s_dev),
2898                                MINOR(inode->i_sb->s_dev), inode->i_ino);
2899        } else {
2900                seq_printf(f, "%d <none>:0 ", fl_pid);
2901        }
2902        if (IS_POSIX(fl)) {
2903                if (fl->fl_end == OFFSET_MAX)
2904                        seq_printf(f, "%Ld EOF\n", fl->fl_start);
2905                else
2906                        seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
2907        } else {
2908                seq_puts(f, "0 EOF\n");
2909        }
2910}
2911
2912static struct file_lock *get_next_blocked_member(struct file_lock *node)
2913{
2914        struct file_lock *tmp;
2915
2916        /* NULL node or root node */
2917        if (node == NULL || node->fl_blocker == NULL)
2918                return NULL;
2919
2920        /* Next member in the linked list could be itself */
2921        tmp = list_next_entry(node, fl_blocked_member);
2922        if (list_entry_is_head(tmp, &node->fl_blocker->fl_blocked_requests, fl_blocked_member)
2923                || tmp == node) {
2924                return NULL;
2925        }
2926
2927        return tmp;
2928}
2929
2930static int locks_show(struct seq_file *f, void *v)
2931{
2932        struct locks_iterator *iter = f->private;
2933        struct file_lock *cur, *tmp;
2934        struct pid_namespace *proc_pidns = proc_pid_ns(file_inode(f->file)->i_sb);
2935        int level = 0;
2936
2937        cur = hlist_entry(v, struct file_lock, fl_link);
2938
2939        if (locks_translate_pid(cur, proc_pidns) == 0)
2940                return 0;
2941
2942        /* View this crossed linked list as a binary tree, the first member of fl_blocked_requests
2943         * is the left child of current node, the next silibing in fl_blocked_member is the
2944         * right child, we can alse get the parent of current node from fl_blocker, so this
2945         * question becomes traversal of a binary tree
2946         */
2947        while (cur != NULL) {
2948                if (level)
2949                        lock_get_status(f, cur, iter->li_pos, "-> ", level);
2950                else
2951                        lock_get_status(f, cur, iter->li_pos, "", level);
2952
2953                if (!list_empty(&cur->fl_blocked_requests)) {
2954                        /* Turn left */
2955                        cur = list_first_entry_or_null(&cur->fl_blocked_requests,
2956                                struct file_lock, fl_blocked_member);
2957                        level++;
2958                } else {
2959                        /* Turn right */
2960                        tmp = get_next_blocked_member(cur);
2961                        /* Fall back to parent node */
2962                        while (tmp == NULL && cur->fl_blocker != NULL) {
2963                                cur = cur->fl_blocker;
2964                                level--;
2965                                tmp = get_next_blocked_member(cur);
2966                        }
2967                        cur = tmp;
2968                }
2969        }
2970
2971        return 0;
2972}
2973
2974static void __show_fd_locks(struct seq_file *f,
2975                        struct list_head *head, int *id,
2976                        struct file *filp, struct files_struct *files)
2977{
2978        struct file_lock *fl;
2979
2980        list_for_each_entry(fl, head, fl_list) {
2981
2982                if (filp != fl->fl_file)
2983                        continue;
2984                if (fl->fl_owner != files &&
2985                    fl->fl_owner != filp)
2986                        continue;
2987
2988                (*id)++;
2989                seq_puts(f, "lock:\t");
2990                lock_get_status(f, fl, *id, "", 0);
2991        }
2992}
2993
2994void show_fd_locks(struct seq_file *f,
2995                  struct file *filp, struct files_struct *files)
2996{
2997        struct inode *inode = locks_inode(filp);
2998        struct file_lock_context *ctx;
2999        int id = 0;
3000
3001        ctx = smp_load_acquire(&inode->i_flctx);
3002        if (!ctx)
3003                return;
3004
3005        spin_lock(&ctx->flc_lock);
3006        __show_fd_locks(f, &ctx->flc_flock, &id, filp, files);
3007        __show_fd_locks(f, &ctx->flc_posix, &id, filp, files);
3008        __show_fd_locks(f, &ctx->flc_lease, &id, filp, files);
3009        spin_unlock(&ctx->flc_lock);
3010}
3011
3012static void *locks_start(struct seq_file *f, loff_t *pos)
3013        __acquires(&blocked_lock_lock)
3014{
3015        struct locks_iterator *iter = f->private;
3016
3017        iter->li_pos = *pos + 1;
3018        percpu_down_write(&file_rwsem);
3019        spin_lock(&blocked_lock_lock);
3020        return seq_hlist_start_percpu(&file_lock_list.hlist, &iter->li_cpu, *pos);
3021}
3022
3023static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
3024{
3025        struct locks_iterator *iter = f->private;
3026
3027        ++iter->li_pos;
3028        return seq_hlist_next_percpu(v, &file_lock_list.hlist, &iter->li_cpu, pos);
3029}
3030
3031static void locks_stop(struct seq_file *f, void *v)
3032        __releases(&blocked_lock_lock)
3033{
3034        spin_unlock(&blocked_lock_lock);
3035        percpu_up_write(&file_rwsem);
3036}
3037
3038static const struct seq_operations locks_seq_operations = {
3039        .start  = locks_start,
3040        .next   = locks_next,
3041        .stop   = locks_stop,
3042        .show   = locks_show,
3043};
3044
3045static int __init proc_locks_init(void)
3046{
3047        proc_create_seq_private("locks", 0, NULL, &locks_seq_operations,
3048                        sizeof(struct locks_iterator), NULL);
3049        return 0;
3050}
3051fs_initcall(proc_locks_init);
3052#endif
3053
3054static int __init filelock_init(void)
3055{
3056        int i;
3057
3058        flctx_cache = kmem_cache_create("file_lock_ctx",
3059                        sizeof(struct file_lock_context), 0, SLAB_PANIC, NULL);
3060
3061        filelock_cache = kmem_cache_create("file_lock_cache",
3062                        sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
3063
3064        for_each_possible_cpu(i) {
3065                struct file_lock_list_struct *fll = per_cpu_ptr(&file_lock_list, i);
3066
3067                spin_lock_init(&fll->lock);
3068                INIT_HLIST_HEAD(&fll->hlist);
3069        }
3070
3071        lease_notifier_chain_init();
3072        return 0;
3073}
3074core_initcall(filelock_init);
3075