fs/lockd/svclock.c

   1 /*
   2  * linux/fs/lockd/svclock.c
   3  *
   4  * Handling of server-side locks, mostly of the blocked variety.
   5  * This is the ugliest part of lockd because we tread on very thin ice.
   6  * GRANT and CANCEL calls may get stuck, meet in mid-flight, etc.
   7  * IMNSHO introducing the grant callback into the NLM protocol was one
   8  * of the worst ideas Sun ever had. Except maybe for the idea of doing
   9  * NFS file locking at all.
  10  *
  11  * I'm trying hard to avoid race conditions by protecting most accesses
  12  * to a file's list of blocked locks through a semaphore. The global
  13  * list of blocked locks is not protected in this fashion however.
  14  * Therefore, some functions (such as the RPC callback for the async grant
  15  * call) move blocked locks towards the head of the list *while some other
  16  * process might be traversing it*. This should not be a problem in
  17  * practice, because this will only cause functions traversing the list
  18  * to visit some blocks twice.
  19  *
  20  * Copyright (C) 1996, Olaf Kirch <okir@monad.swb.de>
  21  */
  22
  23 #include <linux/types.h>
  24 #include <linux/errno.h>
  25 #include <linux/kernel.h>
  26 #include <linux/sched.h>
  27 #include <linux/smp_lock.h>
  28 #include <linux/sunrpc/clnt.h>
  29 #include <linux/sunrpc/svc.h>
  30 #include <linux/lockd/nlm.h>
  31 #include <linux/lockd/lockd.h>
  32 #include <linux/kthread.h>
  33
  34 #define NLMDBG_FACILITY         NLMDBG_SVCLOCK
  35
  36 #ifdef CONFIG_LOCKD_V4
  37 #define nlm_deadlock    nlm4_deadlock
  38 #else
  39 #define nlm_deadlock    nlm_lck_denied
  40 #endif
  41
  42 static void nlmsvc_release_block(struct nlm_block *block);
  43 static void     nlmsvc_insert_block(struct nlm_block *block, unsigned long);
  44 static void     nlmsvc_remove_block(struct nlm_block *block);
  45
  46 static int nlmsvc_setgrantargs(struct nlm_rqst *call, struct nlm_lock *lock);
  47 static void nlmsvc_freegrantargs(struct nlm_rqst *call);
  48 static const struct rpc_call_ops nlmsvc_grant_ops;
  49
  50 /*
  51  * The list of blocked locks to retry
  52  */
  53 static LIST_HEAD(nlm_blocked);
  54
  55 /*
  56  * Insert a blocked lock into the global list
  57  */
  58 static void
  59 nlmsvc_insert_block(struct nlm_block *block, unsigned long when)
  60 {
  61         struct nlm_block *b;
  62         struct list_head *pos;
  63
  64         dprintk("lockd: nlmsvc_insert_block(%p, %ld)\n", block, when);
  65         if (list_empty(&block->b_list)) {
  66                 kref_get(&block->b_count);
  67         } else {
  68                 list_del_init(&block->b_list);
  69         }
  70
  71         pos = &nlm_blocked;
  72         if (when != NLM_NEVER) {
  73                 if ((when += jiffies) == NLM_NEVER)
  74                         when ++;
  75                 list_for_each(pos, &nlm_blocked) {
  76                         b = list_entry(pos, struct nlm_block, b_list);
  77                         if (time_after(b->b_when,when) || b->b_when == NLM_NEVER)
  78                                 break;
  79                 }
  80                 /* On normal exit from the loop, pos == &nlm_blocked,
  81                  * so we will be adding to the end of the list - good
  82                  */
  83         }
  84
  85         list_add_tail(&block->b_list, pos);
  86         block->b_when = when;
  87 }
  88
  89 /*
  90  * Remove a block from the global list
  91  */
  92 static inline void
  93 nlmsvc_remove_block(struct nlm_block *block)
  94 {
  95         if (!list_empty(&block->b_list)) {
  96                 list_del_init(&block->b_list);
  97                 nlmsvc_release_block(block);
  98         }
  99 }
 100
 101 /*
 102  * Find a block for a given lock
 103  */
 104 static struct nlm_block *
 105 nlmsvc_lookup_block(struct nlm_file *file, struct nlm_lock *lock)
 106 {
 107         struct nlm_block        *block;
 108         struct file_lock        *fl;
 109
 110         dprintk("lockd: nlmsvc_lookup_block f=%p pd=%d %Ld-%Ld ty=%d\n",
 111                                 file, lock->fl.fl_pid,
 112                                 (long long)lock->fl.fl_start,
 113                                 (long long)lock->fl.fl_end, lock->fl.fl_type);
 114         list_for_each_entry(block, &nlm_blocked, b_list) {
 115                 fl = &block->b_call->a_args.lock.fl;
 116                 dprintk("lockd: check f=%p pd=%d %Ld-%Ld ty=%d cookie=%s\n",
 117                                 block->b_file, fl->fl_pid,
 118                                 (long long)fl->fl_start,
 119                                 (long long)fl->fl_end, fl->fl_type,
 120                                 nlmdbg_cookie2a(&block->b_call->a_args.cookie));
 121                 if (block->b_file == file && nlm_compare_locks(fl, &lock->fl)) {
 122                         kref_get(&block->b_count);
 123                         return block;
 124                 }
 125         }
 126
 127         return NULL;
 128 }
 129
 130 static inline int nlm_cookie_match(struct nlm_cookie *a, struct nlm_cookie *b)
 131 {
 132         if (a->len != b->len)
 133                 return 0;
 134         if (memcmp(a->data, b->data, a->len))
 135                 return 0;
 136         return 1;
 137 }
 138
 139 /*
 140  * Find a block with a given NLM cookie.
 141  */
 142 static inline struct nlm_block *
 143 nlmsvc_find_block(struct nlm_cookie *cookie)
 144 {
 145         struct nlm_block *block;
 146
 147         list_for_each_entry(block, &nlm_blocked, b_list) {
 148                 if (nlm_cookie_match(&block->b_call->a_args.cookie,cookie))
 149                         goto found;
 150         }
 151
 152         return NULL;
 153
 154 found:
 155         dprintk("nlmsvc_find_block(%s): block=%p\n", nlmdbg_cookie2a(cookie), block);
 156         kref_get(&block->b_count);
 157         return block;
 158 }
 159
 160 /*
 161  * Create a block and initialize it.
 162  *
 163  * Note: we explicitly set the cookie of the grant reply to that of
 164  * the blocked lock request. The spec explicitly mentions that the client
 165  * should _not_ rely on the callback containing the same cookie as the
 166  * request, but (as I found out later) that's because some implementations
 167  * do just this. Never mind the standards comittees, they support our
 168  * logging industries.
 169  *
 170  * 10 years later: I hope we can safely ignore these old and broken
 171  * clients by now. Let's fix this so we can uniquely identify an incoming
 172  * GRANTED_RES message by cookie, without having to rely on the client's IP
 173  * address. --okir
 174  */
 175 static struct nlm_block *
 176 nlmsvc_create_block(struct svc_rqst *rqstp, struct nlm_host *host,
 177                     struct nlm_file *file, struct nlm_lock *lock,
 178                     struct nlm_cookie *cookie)
 179 {
 180         struct nlm_block        *block;
 181         struct nlm_rqst         *call = NULL;
 182
 183         nlm_get_host(host);
 184         call = nlm_alloc_call(host);
 185         if (call == NULL)
 186                 return NULL;
 187
 188         /* Allocate memory for block, and initialize arguments */
 189         block = kzalloc(sizeof(*block), GFP_KERNEL);
 190         if (block == NULL)
 191                 goto failed;
 192         kref_init(&block->b_count);
 193         INIT_LIST_HEAD(&block->b_list);
 194         INIT_LIST_HEAD(&block->b_flist);
 195
 196         if (!nlmsvc_setgrantargs(call, lock))
 197                 goto failed_free;
 198
 199         /* Set notifier function for VFS, and init args */
 200         call->a_args.lock.fl.fl_flags |= FL_SLEEP;
 201         call->a_args.lock.fl.fl_lmops = &nlmsvc_lock_operations;
 202         nlmclnt_next_cookie(&call->a_args.cookie);
 203
 204         dprintk("lockd: created block %p...\n", block);
 205
 206         /* Create and initialize the block */
 207         block->b_daemon = rqstp->rq_server;
 208         block->b_host   = host;
 209         block->b_file   = file;
 210         block->b_fl = NULL;
 211         file->f_count++;
 212
 213         /* Add to file's list of blocks */
 214         list_add(&block->b_flist, &file->f_blocks);
 215
 216         /* Set up RPC arguments for callback */
 217         block->b_call = call;
 218         call->a_flags   = RPC_TASK_ASYNC;
 219         call->a_block = block;
 220
 221         return block;
 222
 223 failed_free:
 224         kfree(block);
 225 failed:
 226         nlm_release_call(call);
 227         return NULL;
 228 }
 229
 230 /*
 231  * Delete a block.
 232  * It is the caller's responsibility to check whether the file
 233  * can be closed hereafter.
 234  */
 235 static int nlmsvc_unlink_block(struct nlm_block *block)
 236 {
 237         int status;
 238         dprintk("lockd: unlinking block %p...\n", block);
 239
 240         /* Remove block from list */
 241         status = posix_unblock_lock(block->b_file->f_file, &block->b_call->a_args.lock.fl);
 242         nlmsvc_remove_block(block);
 243         return status;
 244 }
 245
 246 static void nlmsvc_free_block(struct kref *kref)
 247 {
 248         struct nlm_block *block = container_of(kref, struct nlm_block, b_count);
 249         struct nlm_file         *file = block->b_file;
 250
 251         dprintk("lockd: freeing block %p...\n", block);
 252
 253         /* Remove block from file's list of blocks */
 254         mutex_lock(&file->f_mutex);
 255         list_del_init(&block->b_flist);
 256         mutex_unlock(&file->f_mutex);
 257
 258         nlmsvc_freegrantargs(block->b_call);
 259         nlm_release_call(block->b_call);
 260         nlm_release_file(block->b_file);
 261         kfree(block->b_fl);
 262         kfree(block);
 263 }
 264
 265 static void nlmsvc_release_block(struct nlm_block *block)
 266 {
 267         if (block != NULL)
 268                 kref_put(&block->b_count, nlmsvc_free_block);
 269 }
 270
 271 /*
 272  * Loop over all blocks and delete blocks held by
 273  * a matching host.
 274  */
 275 void nlmsvc_traverse_blocks(struct nlm_host *host,
 276                         struct nlm_file *file,
 277                         nlm_host_match_fn_t match)
 278 {
 279         struct nlm_block *block, *next;
 280
 281 restart:
 282         mutex_lock(&file->f_mutex);
 283         list_for_each_entry_safe(block, next, &file->f_blocks, b_flist) {
 284                 if (!match(block->b_host, host))
 285                         continue;
 286                 /* Do not destroy blocks that are not on
 287                  * the global retry list - why? */
 288                 if (list_empty(&block->b_list))
 289                         continue;
 290                 kref_get(&block->b_count);
 291                 mutex_unlock(&file->f_mutex);
 292                 nlmsvc_unlink_block(block);
 293                 nlmsvc_release_block(block);
 294                 goto restart;
 295         }
 296         mutex_unlock(&file->f_mutex);
 297 }
 298
 299 /*
 300  * Initialize arguments for GRANTED call. The nlm_rqst structure
 301  * has been cleared already.
 302  */
 303 static int nlmsvc_setgrantargs(struct nlm_rqst *call, struct nlm_lock *lock)
 304 {
 305         locks_copy_lock(&call->a_args.lock.fl, &lock->fl);
 306         memcpy(&call->a_args.lock.fh, &lock->fh, sizeof(call->a_args.lock.fh));
 307         call->a_args.lock.caller = utsname()->nodename;
 308         call->a_args.lock.oh.len = lock->oh.len;
 309
 310         /* set default data area */
 311         call->a_args.lock.oh.data = call->a_owner;
 312         call->a_args.lock.svid = lock->fl.fl_pid;
 313
 314         if (lock->oh.len > NLMCLNT_OHSIZE) {
 315                 void *data = kmalloc(lock->oh.len, GFP_KERNEL);
 316                 if (!data)
 317                         return 0;
 318                 call->a_args.lock.oh.data = (u8 *) data;
 319         }
 320
 321         memcpy(call->a_args.lock.oh.data, lock->oh.data, lock->oh.len);
 322         return 1;
 323 }
 324
 325 static void nlmsvc_freegrantargs(struct nlm_rqst *call)
 326 {
 327         if (call->a_args.lock.oh.data != call->a_owner)
 328                 kfree(call->a_args.lock.oh.data);
 329
 330         locks_release_private(&call->a_args.lock.fl);
 331 }
 332
 333 /*
 334  * Deferred lock request handling for non-blocking lock
 335  */
 336 static __be32
 337 nlmsvc_defer_lock_rqst(struct svc_rqst *rqstp, struct nlm_block *block)
 338 {
 339         __be32 status = nlm_lck_denied_nolocks;
 340
 341         block->b_flags |= B_QUEUED;
 342
 343         nlmsvc_insert_block(block, NLM_TIMEOUT);
 344
 345         block->b_cache_req = &rqstp->rq_chandle;
 346         if (rqstp->rq_chandle.defer) {
 347                 block->b_deferred_req =
 348                         rqstp->rq_chandle.defer(block->b_cache_req);
 349                 if (block->b_deferred_req != NULL)
 350                         status = nlm_drop_reply;
 351         }
 352         dprintk("lockd: nlmsvc_defer_lock_rqst block %p flags %d status %d\n",
 353                 block, block->b_flags, ntohl(status));
 354
 355         return status;
 356 }
 357
 358 /*
 359  * Attempt to establish a lock, and if it can't be granted, block it
 360  * if required.
 361  */
 362 __be32
 363 nlmsvc_lock(struct svc_rqst *rqstp, struct nlm_file *file,
 364             struct nlm_host *host, struct nlm_lock *lock, int wait,
 365             struct nlm_cookie *cookie, int reclaim)
 366 {
 367         struct nlm_block        *block = NULL;
 368         int                     error;
 369         __be32                  ret;
 370
 371         dprintk("lockd: nlmsvc_lock(%s/%ld, ty=%d, pi=%d, %Ld-%Ld, bl=%d)\n",
 372                                 file->f_file->f_path.dentry->d_inode->i_sb->s_id,
 373                                 file->f_file->f_path.dentry->d_inode->i_ino,
 374                                 lock->fl.fl_type, lock->fl.fl_pid,
 375                                 (long long)lock->fl.fl_start,
 376                                 (long long)lock->fl.fl_end,
 377                                 wait);
 378
 379         /* Lock file against concurrent access */
 380         mutex_lock(&file->f_mutex);
 381         /* Get existing block (in case client is busy-waiting)
 382          * or create new block
 383          */
 384         block = nlmsvc_lookup_block(file, lock);
 385         if (block == NULL) {
 386                 block = nlmsvc_create_block(rqstp, host, file, lock, cookie);
 387                 ret = nlm_lck_denied_nolocks;
 388                 if (block == NULL)
 389                         goto out;
 390                 lock = &block->b_call->a_args.lock;
 391         } else
 392                 lock->fl.fl_flags &= ~FL_SLEEP;
 393
 394         if (block->b_flags & B_QUEUED) {
 395                 dprintk("lockd: nlmsvc_lock deferred block %p flags %d\n",
 396                                                         block, block->b_flags);
 397                 if (block->b_granted) {
 398                         nlmsvc_unlink_block(block);
 399                         ret = nlm_granted;
 400                         goto out;
 401                 }
 402                 if (block->b_flags & B_TIMED_OUT) {
 403                         nlmsvc_unlink_block(block);
 404                         ret = nlm_lck_denied;
 405                         goto out;
 406                 }
 407                 ret = nlm_drop_reply;
 408                 goto out;
 409         }
 410
 411         if (locks_in_grace() && !reclaim) {
 412                 ret = nlm_lck_denied_grace_period;
 413                 goto out;
 414         }
 415         if (reclaim && !locks_in_grace()) {
 416                 ret = nlm_lck_denied_grace_period;
 417                 goto out;
 418         }
 419
 420         if (!wait)
 421                 lock->fl.fl_flags &= ~FL_SLEEP;
 422         error = vfs_lock_file(file->f_file, F_SETLK, &lock->fl, NULL);
 423         lock->fl.fl_flags &= ~FL_SLEEP;
 424
 425         dprintk("lockd: vfs_lock_file returned %d\n", error);
 426         switch (error) {
 427                 case 0:
 428                         ret = nlm_granted;
 429                         goto out;
 430                 case -EAGAIN:
 431                         /*
 432                          * If this is a blocking request for an
 433                          * already pending lock request then we need
 434                          * to put it back on lockd's block list
 435                          */
 436                         if (wait)
 437                                 break;
 438                         ret = nlm_lck_denied;
 439                         goto out;
 440                 case FILE_LOCK_DEFERRED:
 441                         if (wait)
 442                                 break;
 443                         /* Filesystem lock operation is in progress
 444                            Add it to the queue waiting for callback */
 445                         ret = nlmsvc_defer_lock_rqst(rqstp, block);
 446                         goto out;
 447                 case -EDEADLK:
 448                         ret = nlm_deadlock;
 449                         goto out;
 450                 default:                        /* includes ENOLCK */
 451                         ret = nlm_lck_denied_nolocks;
 452                         goto out;
 453         }
 454
 455         ret = nlm_lck_blocked;
 456
 457         /* Append to list of blocked */
 458         nlmsvc_insert_block(block, NLM_NEVER);
 459 out:
 460         mutex_unlock(&file->f_mutex);
 461         nlmsvc_release_block(block);
 462         dprintk("lockd: nlmsvc_lock returned %u\n", ret);
 463         return ret;
 464 }
 465
 466 /*
 467  * Test for presence of a conflicting lock.
 468  */
 469 __be32
 470 nlmsvc_testlock(struct svc_rqst *rqstp, struct nlm_file *file,
 471                 struct nlm_host *host, struct nlm_lock *lock,
 472                 struct nlm_lock *conflock, struct nlm_cookie *cookie)
 473 {
 474         struct nlm_block        *block = NULL;
 475         int                     error;
 476         __be32                  ret;
 477
 478         dprintk("lockd: nlmsvc_testlock(%s/%ld, ty=%d, %Ld-%Ld)\n",
 479                                 file->f_file->f_path.dentry->d_inode->i_sb->s_id,
 480                                 file->f_file->f_path.dentry->d_inode->i_ino,
 481                                 lock->fl.fl_type,
 482                                 (long long)lock->fl.fl_start,
 483                                 (long long)lock->fl.fl_end);
 484
 485         /* Get existing block (in case client is busy-waiting) */
 486         block = nlmsvc_lookup_block(file, lock);
 487
 488         if (block == NULL) {
 489                 struct file_lock *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
 490
 491                 if (conf == NULL)
 492                         return nlm_granted;
 493                 block = nlmsvc_create_block(rqstp, host, file, lock, cookie);
 494                 if (block == NULL) {
 495                         kfree(conf);
 496                         return nlm_granted;
 497                 }
 498                 block->b_fl = conf;
 499         }
 500         if (block->b_flags & B_QUEUED) {
 501                 dprintk("lockd: nlmsvc_testlock deferred block %p flags %d fl %p\n",
 502                         block, block->b_flags, block->b_fl);
 503                 if (block->b_flags & B_TIMED_OUT) {
 504                         nlmsvc_unlink_block(block);
 505                         ret = nlm_lck_denied;
 506                         goto out;
 507                 }
 508                 if (block->b_flags & B_GOT_CALLBACK) {
 509                         nlmsvc_unlink_block(block);
 510                         if (block->b_fl != NULL
 511                                         && block->b_fl->fl_type != F_UNLCK) {
 512                                 lock->fl = *block->b_fl;
 513                                 goto conf_lock;
 514                         } else {
 515                                 ret = nlm_granted;
 516                                 goto out;
 517                         }
 518                 }
 519                 ret = nlm_drop_reply;
 520                 goto out;
 521         }
 522
 523         if (locks_in_grace()) {
 524                 ret = nlm_lck_denied_grace_period;
 525                 goto out;
 526         }
 527         error = vfs_test_lock(file->f_file, &lock->fl);
 528         if (error == FILE_LOCK_DEFERRED) {
 529                 ret = nlmsvc_defer_lock_rqst(rqstp, block);
 530                 goto out;
 531         }
 532         if (error) {
 533                 ret = nlm_lck_denied_nolocks;
 534                 goto out;
 535         }
 536         if (lock->fl.fl_type == F_UNLCK) {
 537                 ret = nlm_granted;
 538                 goto out;
 539         }
 540
 541 conf_lock:
 542         dprintk("lockd: conflicting lock(ty=%d, %Ld-%Ld)\n",
 543                 lock->fl.fl_type, (long long)lock->fl.fl_start,
 544                 (long long)lock->fl.fl_end);
 545         conflock->caller = "somehost";  /* FIXME */
 546         conflock->len = strlen(conflock->caller);
 547         conflock->oh.len = 0;           /* don't return OH info */
 548         conflock->svid = lock->fl.fl_pid;
 549         conflock->fl.fl_type = lock->fl.fl_type;
 550         conflock->fl.fl_start = lock->fl.fl_start;
 551         conflock->fl.fl_end = lock->fl.fl_end;
 552         ret = nlm_lck_denied;
 553 out:
 554         if (block)
 555                 nlmsvc_release_block(block);
 556         return ret;
 557 }
 558
 559 /*
 560  * Remove a lock.
 561  * This implies a CANCEL call: We send a GRANT_MSG, the client replies
 562  * with a GRANT_RES call which gets lost, and calls UNLOCK immediately
 563  * afterwards. In this case the block will still be there, and hence
 564  * must be removed.
 565  */
 566 __be32
 567 nlmsvc_unlock(struct nlm_file *file, struct nlm_lock *lock)
 568 {
 569         int     error;
 570
 571         dprintk("lockd: nlmsvc_unlock(%s/%ld, pi=%d, %Ld-%Ld)\n",
 572                                 file->f_file->f_path.dentry->d_inode->i_sb->s_id,
 573                                 file->f_file->f_path.dentry->d_inode->i_ino,
 574                                 lock->fl.fl_pid,
 575                                 (long long)lock->fl.fl_start,
 576                                 (long long)lock->fl.fl_end);
 577
 578         /* First, cancel any lock that might be there */
 579         nlmsvc_cancel_blocked(file, lock);
 580
 581         lock->fl.fl_type = F_UNLCK;
 582         error = vfs_lock_file(file->f_file, F_SETLK, &lock->fl, NULL);
 583
 584         return (error < 0)? nlm_lck_denied_nolocks : nlm_granted;
 585 }
 586
 587 /*
 588  * Cancel a previously blocked request.
 589  *
 590  * A cancel request always overrides any grant that may currently
 591  * be in progress.
 592  * The calling procedure must check whether the file can be closed.
 593  */
 594 __be32
 595 nlmsvc_cancel_blocked(struct nlm_file *file, struct nlm_lock *lock)
 596 {
 597         struct nlm_block        *block;
 598         int status = 0;
 599
 600         dprintk("lockd: nlmsvc_cancel(%s/%ld, pi=%d, %Ld-%Ld)\n",
 601                                 file->f_file->f_path.dentry->d_inode->i_sb->s_id,
 602                                 file->f_file->f_path.dentry->d_inode->i_ino,
 603                                 lock->fl.fl_pid,
 604                                 (long long)lock->fl.fl_start,
 605                                 (long long)lock->fl.fl_end);
 606
 607         if (locks_in_grace())
 608                 return nlm_lck_denied_grace_period;
 609
 610         mutex_lock(&file->f_mutex);
 611         block = nlmsvc_lookup_block(file, lock);
 612         mutex_unlock(&file->f_mutex);
 613         if (block != NULL) {
 614                 vfs_cancel_lock(block->b_file->f_file,
 615                                 &block->b_call->a_args.lock.fl);
 616                 status = nlmsvc_unlink_block(block);
 617                 nlmsvc_release_block(block);
 618         }
 619         return status ? nlm_lck_denied : nlm_granted;
 620 }
 621
 622 /*
 623  * This is a callback from the filesystem for VFS file lock requests.
 624  * It will be used if fl_grant is defined and the filesystem can not
 625  * respond to the request immediately.
 626  * For GETLK request it will copy the reply to the nlm_block.
 627  * For SETLK or SETLKW request it will get the local posix lock.
 628  * In all cases it will move the block to the head of nlm_blocked q where
 629  * nlmsvc_retry_blocked() can send back a reply for SETLKW or revisit the
 630  * deferred rpc for GETLK and SETLK.
 631  */
 632 static void
 633 nlmsvc_update_deferred_block(struct nlm_block *block, struct file_lock *conf,
 634                              int result)
 635 {
 636         block->b_flags |= B_GOT_CALLBACK;
 637         if (result == 0)
 638                 block->b_granted = 1;
 639         else
 640                 block->b_flags |= B_TIMED_OUT;
 641         if (conf) {
 642                 if (block->b_fl)
 643                         __locks_copy_lock(block->b_fl, conf);
 644         }
 645 }
 646
 647 static int nlmsvc_grant_deferred(struct file_lock *fl, struct file_lock *conf,
 648                                         int result)
 649 {
 650         struct nlm_block *block;
 651         int rc = -ENOENT;
 652
 653         lock_kernel();
 654         list_for_each_entry(block, &nlm_blocked, b_list) {
 655                 if (nlm_compare_locks(&block->b_call->a_args.lock.fl, fl)) {
 656                         dprintk("lockd: nlmsvc_notify_blocked block %p flags %d\n",
 657                                                         block, block->b_flags);
 658                         if (block->b_flags & B_QUEUED) {
 659                                 if (block->b_flags & B_TIMED_OUT) {
 660                                         rc = -ENOLCK;
 661                                         break;
 662                                 }
 663                                 nlmsvc_update_deferred_block(block, conf, result);
 664                         } else if (result == 0)
 665                                 block->b_granted = 1;
 666
 667                         nlmsvc_insert_block(block, 0);
 668                         svc_wake_up(block->b_daemon);
 669                         rc = 0;
 670                         break;
 671                 }
 672         }
 673         unlock_kernel();
 674         if (rc == -ENOENT)
 675                 printk(KERN_WARNING "lockd: grant for unknown block\n");
 676         return rc;
 677 }
 678
 679 /*
 680  * Unblock a blocked lock request. This is a callback invoked from the
 681  * VFS layer when a lock on which we blocked is removed.
 682  *
 683  * This function doesn't grant the blocked lock instantly, but rather moves
 684  * the block to the head of nlm_blocked where it can be picked up by lockd.
 685  */
 686 static void
 687 nlmsvc_notify_blocked(struct file_lock *fl)
 688 {
 689         struct nlm_block        *block;
 690
 691         dprintk("lockd: VFS unblock notification for block %p\n", fl);
 692         list_for_each_entry(block, &nlm_blocked, b_list) {
 693                 if (nlm_compare_locks(&block->b_call->a_args.lock.fl, fl)) {
 694                         nlmsvc_insert_block(block, 0);
 695                         svc_wake_up(block->b_daemon);
 696                         return;
 697                 }
 698         }
 699
 700         printk(KERN_WARNING "lockd: notification for unknown block!\n");
 701 }
 702
 703 static int nlmsvc_same_owner(struct file_lock *fl1, struct file_lock *fl2)
 704 {
 705         return fl1->fl_owner == fl2->fl_owner && fl1->fl_pid == fl2->fl_pid;
 706 }
 707
 708 struct lock_manager_operations nlmsvc_lock_operations = {
 709         .fl_compare_owner = nlmsvc_same_owner,
 710         .fl_notify = nlmsvc_notify_blocked,
 711         .fl_grant = nlmsvc_grant_deferred,
 712 };
 713
 714 /*
 715  * Try to claim a lock that was previously blocked.
 716  *
 717  * Note that we use both the RPC_GRANTED_MSG call _and_ an async
 718  * RPC thread when notifying the client. This seems like overkill...
 719  * Here's why:
 720  *  -   we don't want to use a synchronous RPC thread, otherwise
 721  *      we might find ourselves hanging on a dead portmapper.
 722  *  -   Some lockd implementations (e.g. HP) don't react to
 723  *      RPC_GRANTED calls; they seem to insist on RPC_GRANTED_MSG calls.
 724  */
 725 static void
 726 nlmsvc_grant_blocked(struct nlm_block *block)
 727 {
 728         struct nlm_file         *file = block->b_file;
 729         struct nlm_lock         *lock = &block->b_call->a_args.lock;
 730         int                     error;
 731
 732         dprintk("lockd: grant blocked lock %p\n", block);
 733
 734         kref_get(&block->b_count);
 735
 736         /* Unlink block request from list */
 737         nlmsvc_unlink_block(block);
 738
 739         /* If b_granted is true this means we've been here before.
 740          * Just retry the grant callback, possibly refreshing the RPC
 741          * binding */
 742         if (block->b_granted) {
 743                 nlm_rebind_host(block->b_host);
 744                 goto callback;
 745         }
 746
 747         /* Try the lock operation again */
 748         lock->fl.fl_flags |= FL_SLEEP;
 749         error = vfs_lock_file(file->f_file, F_SETLK, &lock->fl, NULL);
 750         lock->fl.fl_flags &= ~FL_SLEEP;
 751
 752         switch (error) {
 753         case 0:
 754                 break;
 755         case FILE_LOCK_DEFERRED:
 756                 dprintk("lockd: lock still blocked error %d\n", error);
 757                 nlmsvc_insert_block(block, NLM_NEVER);
 758                 nlmsvc_release_block(block);
 759                 return;
 760         default:
 761                 printk(KERN_WARNING "lockd: unexpected error %d in %s!\n",
 762                                 -error, __func__);
 763                 nlmsvc_insert_block(block, 10 * HZ);
 764                 nlmsvc_release_block(block);
 765                 return;
 766         }
 767
 768 callback:
 769         /* Lock was granted by VFS. */
 770         dprintk("lockd: GRANTing blocked lock.\n");
 771         block->b_granted = 1;
 772
 773         /* keep block on the list, but don't reattempt until the RPC
 774          * completes or the submission fails
 775          */
 776         nlmsvc_insert_block(block, NLM_NEVER);
 777
 778         /* Call the client -- use a soft RPC task since nlmsvc_retry_blocked
 779          * will queue up a new one if this one times out
 780          */
 781         error = nlm_async_call(block->b_call, NLMPROC_GRANTED_MSG,
 782                                 &nlmsvc_grant_ops);
 783
 784         /* RPC submission failed, wait a bit and retry */
 785         if (error < 0)
 786                 nlmsvc_insert_block(block, 10 * HZ);
 787 }
 788
 789 /*
 790  * This is the callback from the RPC layer when the NLM_GRANTED_MSG
 791  * RPC call has succeeded or timed out.
 792  * Like all RPC callbacks, it is invoked by the rpciod process, so it
 793  * better not sleep. Therefore, we put the blocked lock on the nlm_blocked
 794  * chain once more in order to have it removed by lockd itself (which can
 795  * then sleep on the file semaphore without disrupting e.g. the nfs client).
 796  */
 797 static void nlmsvc_grant_callback(struct rpc_task *task, void *data)
 798 {
 799         struct nlm_rqst         *call = data;
 800         struct nlm_block        *block = call->a_block;
 801         unsigned long           timeout;
 802
 803         dprintk("lockd: GRANT_MSG RPC callback\n");
 804
 805         lock_kernel();
 806         /* if the block is not on a list at this point then it has
 807          * been invalidated. Don't try to requeue it.
 808          *
 809          * FIXME: it's possible that the block is removed from the list
 810          * after this check but before the nlmsvc_insert_block. In that
 811          * case it will be added back. Perhaps we need better locking
 812          * for nlm_blocked?
 813          */
 814         if (list_empty(&block->b_list))
 815                 goto out;
 816
 817         /* Technically, we should down the file semaphore here. Since we
 818          * move the block towards the head of the queue only, no harm
 819          * can be done, though. */
 820         if (task->tk_status < 0) {
 821                 /* RPC error: Re-insert for retransmission */
 822                 timeout = 10 * HZ;
 823         } else {
 824                 /* Call was successful, now wait for client callback */
 825                 timeout = 60 * HZ;
 826         }
 827         nlmsvc_insert_block(block, timeout);
 828         svc_wake_up(block->b_daemon);
 829 out:
 830         unlock_kernel();
 831 }
 832
 833 static void nlmsvc_grant_release(void *data)
 834 {
 835         struct nlm_rqst         *call = data;
 836
 837         lock_kernel();
 838         nlmsvc_release_block(call->a_block);
 839         unlock_kernel();
 840 }
 841
 842 static const struct rpc_call_ops nlmsvc_grant_ops = {
 843         .rpc_call_done = nlmsvc_grant_callback,
 844         .rpc_release = nlmsvc_grant_release,
 845 };
 846
 847 /*
 848  * We received a GRANT_RES callback. Try to find the corresponding
 849  * block.
 850  */
 851 void
 852 nlmsvc_grant_reply(struct nlm_cookie *cookie, __be32 status)
 853 {
 854         struct nlm_block        *block;
 855
 856         dprintk("grant_reply: looking for cookie %x, s=%d \n",
 857                 *(unsigned int *)(cookie->data), status);
 858         if (!(block = nlmsvc_find_block(cookie)))
 859                 return;
 860
 861         if (block) {
 862                 if (status == nlm_lck_denied_grace_period) {
 863                         /* Try again in a couple of seconds */
 864                         nlmsvc_insert_block(block, 10 * HZ);
 865                 } else {
 866                         /* Lock is now held by client, or has been rejected.
 867                          * In both cases, the block should be removed. */
 868                         nlmsvc_unlink_block(block);
 869                 }
 870         }
 871         nlmsvc_release_block(block);
 872 }
 873
 874 /* Helper function to handle retry of a deferred block.
 875  * If it is a blocking lock, call grant_blocked.
 876  * For a non-blocking lock or test lock, revisit the request.
 877  */
 878 static void
 879 retry_deferred_block(struct nlm_block *block)
 880 {
 881         if (!(block->b_flags & B_GOT_CALLBACK))
 882                 block->b_flags |= B_TIMED_OUT;
 883         nlmsvc_insert_block(block, NLM_TIMEOUT);
 884         dprintk("revisit block %p flags %d\n",  block, block->b_flags);
 885         if (block->b_deferred_req) {
 886                 block->b_deferred_req->revisit(block->b_deferred_req, 0);
 887                 block->b_deferred_req = NULL;
 888         }
 889 }
 890
 891 /*
 892  * Retry all blocked locks that have been notified. This is where lockd
 893  * picks up locks that can be granted, or grant notifications that must
 894  * be retransmitted.
 895  */
 896 unsigned long
 897 nlmsvc_retry_blocked(void)
 898 {
 899         unsigned long   timeout = MAX_SCHEDULE_TIMEOUT;
 900         struct nlm_block *block;
 901
 902         while (!list_empty(&nlm_blocked) && !kthread_should_stop()) {
 903                 block = list_entry(nlm_blocked.next, struct nlm_block, b_list);
 904
 905                 if (block->b_when == NLM_NEVER)
 906                         break;
 907                 if (time_after(block->b_when, jiffies)) {
 908                         timeout = block->b_when - jiffies;
 909                         break;
 910                 }
 911
 912                 dprintk("nlmsvc_retry_blocked(%p, when=%ld)\n",
 913                         block, block->b_when);
 914                 if (block->b_flags & B_QUEUED) {
 915                         dprintk("nlmsvc_retry_blocked delete block (%p, granted=%d, flags=%d)\n",
 916                                 block, block->b_granted, block->b_flags);
 917                         retry_deferred_block(block);
 918                 } else
 919                         nlmsvc_grant_blocked(block);
 920         }
 921
 922         return timeout;
 923 }