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
 331 /*
 332  * Deferred lock request handling for non-blocking lock
 333  */
 334 static __be32
 335 nlmsvc_defer_lock_rqst(struct svc_rqst *rqstp, struct nlm_block *block)
 336 {
 337         __be32 status = nlm_lck_denied_nolocks;
 338
 339         block->b_flags |= B_QUEUED;
 340
 341         nlmsvc_insert_block(block, NLM_TIMEOUT);
 342
 343         block->b_cache_req = &rqstp->rq_chandle;
 344         if (rqstp->rq_chandle.defer) {
 345                 block->b_deferred_req =
 346                         rqstp->rq_chandle.defer(block->b_cache_req);
 347                 if (block->b_deferred_req != NULL)
 348                         status = nlm_drop_reply;
 349         }
 350         dprintk("lockd: nlmsvc_defer_lock_rqst block %p flags %d status %d\n",
 351                 block, block->b_flags, ntohl(status));
 352
 353         return status;
 354 }
 355
 356 /*
 357  * Attempt to establish a lock, and if it can't be granted, block it
 358  * if required.
 359  */
 360 __be32
 361 nlmsvc_lock(struct svc_rqst *rqstp, struct nlm_file *file,
 362             struct nlm_host *host, struct nlm_lock *lock, int wait,
 363             struct nlm_cookie *cookie, int reclaim)
 364 {
 365         struct nlm_block        *block = NULL;
 366         int                     error;
 367         __be32                  ret;
 368
 369         dprintk("lockd: nlmsvc_lock(%s/%ld, ty=%d, pi=%d, %Ld-%Ld, bl=%d)\n",
 370                                 file->f_file->f_path.dentry->d_inode->i_sb->s_id,
 371                                 file->f_file->f_path.dentry->d_inode->i_ino,
 372                                 lock->fl.fl_type, lock->fl.fl_pid,
 373                                 (long long)lock->fl.fl_start,
 374                                 (long long)lock->fl.fl_end,
 375                                 wait);
 376
 377         /* Lock file against concurrent access */
 378         mutex_lock(&file->f_mutex);
 379         /* Get existing block (in case client is busy-waiting)
 380          * or create new block
 381          */
 382         block = nlmsvc_lookup_block(file, lock);
 383         if (block == NULL) {
 384                 block = nlmsvc_create_block(rqstp, host, file, lock, cookie);
 385                 ret = nlm_lck_denied_nolocks;
 386                 if (block == NULL)
 387                         goto out;
 388                 lock = &block->b_call->a_args.lock;
 389         } else
 390                 lock->fl.fl_flags &= ~FL_SLEEP;
 391
 392         if (block->b_flags & B_QUEUED) {
 393                 dprintk("lockd: nlmsvc_lock deferred block %p flags %d\n",
 394                                                         block, block->b_flags);
 395                 if (block->b_granted) {
 396                         nlmsvc_unlink_block(block);
 397                         ret = nlm_granted;
 398                         goto out;
 399                 }
 400                 if (block->b_flags & B_TIMED_OUT) {
 401                         nlmsvc_unlink_block(block);
 402                         ret = nlm_lck_denied;
 403                         goto out;
 404                 }
 405                 ret = nlm_drop_reply;
 406                 goto out;
 407         }
 408
 409         if (locks_in_grace() && !reclaim) {
 410                 ret = nlm_lck_denied_grace_period;
 411                 goto out;
 412         }
 413         if (reclaim && !locks_in_grace()) {
 414                 ret = nlm_lck_denied_grace_period;
 415                 goto out;
 416         }
 417
 418         if (!wait)
 419                 lock->fl.fl_flags &= ~FL_SLEEP;
 420         error = vfs_lock_file(file->f_file, F_SETLK, &lock->fl, NULL);
 421         lock->fl.fl_flags &= ~FL_SLEEP;
 422
 423         dprintk("lockd: vfs_lock_file returned %d\n", error);
 424         switch (error) {
 425                 case 0:
 426                         ret = nlm_granted;
 427                         goto out;
 428                 case -EAGAIN:
 429                         ret = nlm_lck_denied;
 430                         goto out;
 431                 case FILE_LOCK_DEFERRED:
 432                         if (wait)
 433                                 break;
 434                         /* Filesystem lock operation is in progress
 435                            Add it to the queue waiting for callback */
 436                         ret = nlmsvc_defer_lock_rqst(rqstp, block);
 437                         goto out;
 438                 case -EDEADLK:
 439                         ret = nlm_deadlock;
 440                         goto out;
 441                 default:                        /* includes ENOLCK */
 442                         ret = nlm_lck_denied_nolocks;
 443                         goto out;
 444         }
 445
 446         ret = nlm_lck_blocked;
 447
 448         /* Append to list of blocked */
 449         nlmsvc_insert_block(block, NLM_NEVER);
 450 out:
 451         mutex_unlock(&file->f_mutex);
 452         nlmsvc_release_block(block);
 453         dprintk("lockd: nlmsvc_lock returned %u\n", ret);
 454         return ret;
 455 }
 456
 457 /*
 458  * Test for presence of a conflicting lock.
 459  */
 460 __be32
 461 nlmsvc_testlock(struct svc_rqst *rqstp, struct nlm_file *file,
 462                 struct nlm_host *host, struct nlm_lock *lock,
 463                 struct nlm_lock *conflock, struct nlm_cookie *cookie)
 464 {
 465         struct nlm_block        *block = NULL;
 466         int                     error;
 467         __be32                  ret;
 468
 469         dprintk("lockd: nlmsvc_testlock(%s/%ld, ty=%d, %Ld-%Ld)\n",
 470                                 file->f_file->f_path.dentry->d_inode->i_sb->s_id,
 471                                 file->f_file->f_path.dentry->d_inode->i_ino,
 472                                 lock->fl.fl_type,
 473                                 (long long)lock->fl.fl_start,
 474                                 (long long)lock->fl.fl_end);
 475
 476         /* Get existing block (in case client is busy-waiting) */
 477         block = nlmsvc_lookup_block(file, lock);
 478
 479         if (block == NULL) {
 480                 struct file_lock *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
 481
 482                 if (conf == NULL)
 483                         return nlm_granted;
 484                 block = nlmsvc_create_block(rqstp, host, file, lock, cookie);
 485                 if (block == NULL) {
 486                         kfree(conf);
 487                         return nlm_granted;
 488                 }
 489                 block->b_fl = conf;
 490         }
 491         if (block->b_flags & B_QUEUED) {
 492                 dprintk("lockd: nlmsvc_testlock deferred block %p flags %d fl %p\n",
 493                         block, block->b_flags, block->b_fl);
 494                 if (block->b_flags & B_TIMED_OUT) {
 495                         nlmsvc_unlink_block(block);
 496                         ret = nlm_lck_denied;
 497                         goto out;
 498                 }
 499                 if (block->b_flags & B_GOT_CALLBACK) {
 500                         nlmsvc_unlink_block(block);
 501                         if (block->b_fl != NULL
 502                                         && block->b_fl->fl_type != F_UNLCK) {
 503                                 lock->fl = *block->b_fl;
 504                                 goto conf_lock;
 505                         } else {
 506                                 ret = nlm_granted;
 507                                 goto out;
 508                         }
 509                 }
 510                 ret = nlm_drop_reply;
 511                 goto out;
 512         }
 513
 514         if (locks_in_grace()) {
 515                 ret = nlm_lck_denied_grace_period;
 516                 goto out;
 517         }
 518         error = vfs_test_lock(file->f_file, &lock->fl);
 519         if (error == FILE_LOCK_DEFERRED) {
 520                 ret = nlmsvc_defer_lock_rqst(rqstp, block);
 521                 goto out;
 522         }
 523         if (error) {
 524                 ret = nlm_lck_denied_nolocks;
 525                 goto out;
 526         }
 527         if (lock->fl.fl_type == F_UNLCK) {
 528                 ret = nlm_granted;
 529                 goto out;
 530         }
 531
 532 conf_lock:
 533         dprintk("lockd: conflicting lock(ty=%d, %Ld-%Ld)\n",
 534                 lock->fl.fl_type, (long long)lock->fl.fl_start,
 535                 (long long)lock->fl.fl_end);
 536         conflock->caller = "somehost";  /* FIXME */
 537         conflock->len = strlen(conflock->caller);
 538         conflock->oh.len = 0;           /* don't return OH info */
 539         conflock->svid = lock->fl.fl_pid;
 540         conflock->fl.fl_type = lock->fl.fl_type;
 541         conflock->fl.fl_start = lock->fl.fl_start;
 542         conflock->fl.fl_end = lock->fl.fl_end;
 543         ret = nlm_lck_denied;
 544 out:
 545         if (block)
 546                 nlmsvc_release_block(block);
 547         return ret;
 548 }
 549
 550 /*
 551  * Remove a lock.
 552  * This implies a CANCEL call: We send a GRANT_MSG, the client replies
 553  * with a GRANT_RES call which gets lost, and calls UNLOCK immediately
 554  * afterwards. In this case the block will still be there, and hence
 555  * must be removed.
 556  */
 557 __be32
 558 nlmsvc_unlock(struct nlm_file *file, struct nlm_lock *lock)
 559 {
 560         int     error;
 561
 562         dprintk("lockd: nlmsvc_unlock(%s/%ld, pi=%d, %Ld-%Ld)\n",
 563                                 file->f_file->f_path.dentry->d_inode->i_sb->s_id,
 564                                 file->f_file->f_path.dentry->d_inode->i_ino,
 565                                 lock->fl.fl_pid,
 566                                 (long long)lock->fl.fl_start,
 567                                 (long long)lock->fl.fl_end);
 568
 569         /* First, cancel any lock that might be there */
 570         nlmsvc_cancel_blocked(file, lock);
 571
 572         lock->fl.fl_type = F_UNLCK;
 573         error = vfs_lock_file(file->f_file, F_SETLK, &lock->fl, NULL);
 574
 575         return (error < 0)? nlm_lck_denied_nolocks : nlm_granted;
 576 }
 577
 578 /*
 579  * Cancel a previously blocked request.
 580  *
 581  * A cancel request always overrides any grant that may currently
 582  * be in progress.
 583  * The calling procedure must check whether the file can be closed.
 584  */
 585 __be32
 586 nlmsvc_cancel_blocked(struct nlm_file *file, struct nlm_lock *lock)
 587 {
 588         struct nlm_block        *block;
 589         int status = 0;
 590
 591         dprintk("lockd: nlmsvc_cancel(%s/%ld, pi=%d, %Ld-%Ld)\n",
 592                                 file->f_file->f_path.dentry->d_inode->i_sb->s_id,
 593                                 file->f_file->f_path.dentry->d_inode->i_ino,
 594                                 lock->fl.fl_pid,
 595                                 (long long)lock->fl.fl_start,
 596                                 (long long)lock->fl.fl_end);
 597
 598         if (locks_in_grace())
 599                 return nlm_lck_denied_grace_period;
 600
 601         mutex_lock(&file->f_mutex);
 602         block = nlmsvc_lookup_block(file, lock);
 603         mutex_unlock(&file->f_mutex);
 604         if (block != NULL) {
 605                 vfs_cancel_lock(block->b_file->f_file,
 606                                 &block->b_call->a_args.lock.fl);
 607                 status = nlmsvc_unlink_block(block);
 608                 nlmsvc_release_block(block);
 609         }
 610         return status ? nlm_lck_denied : nlm_granted;
 611 }
 612
 613 /*
 614  * This is a callback from the filesystem for VFS file lock requests.
 615  * It will be used if fl_grant is defined and the filesystem can not
 616  * respond to the request immediately.
 617  * For GETLK request it will copy the reply to the nlm_block.
 618  * For SETLK or SETLKW request it will get the local posix lock.
 619  * In all cases it will move the block to the head of nlm_blocked q where
 620  * nlmsvc_retry_blocked() can send back a reply for SETLKW or revisit the
 621  * deferred rpc for GETLK and SETLK.
 622  */
 623 static void
 624 nlmsvc_update_deferred_block(struct nlm_block *block, struct file_lock *conf,
 625                              int result)
 626 {
 627         block->b_flags |= B_GOT_CALLBACK;
 628         if (result == 0)
 629                 block->b_granted = 1;
 630         else
 631                 block->b_flags |= B_TIMED_OUT;
 632         if (conf) {
 633                 if (block->b_fl)
 634                         __locks_copy_lock(block->b_fl, conf);
 635         }
 636 }
 637
 638 static int nlmsvc_grant_deferred(struct file_lock *fl, struct file_lock *conf,
 639                                         int result)
 640 {
 641         struct nlm_block *block;
 642         int rc = -ENOENT;
 643
 644         lock_kernel();
 645         list_for_each_entry(block, &nlm_blocked, b_list) {
 646                 if (nlm_compare_locks(&block->b_call->a_args.lock.fl, fl)) {
 647                         dprintk("lockd: nlmsvc_notify_blocked block %p flags %d\n",
 648                                                         block, block->b_flags);
 649                         if (block->b_flags & B_QUEUED) {
 650                                 if (block->b_flags & B_TIMED_OUT) {
 651                                         rc = -ENOLCK;
 652                                         break;
 653                                 }
 654                                 nlmsvc_update_deferred_block(block, conf, result);
 655                         } else if (result == 0)
 656                                 block->b_granted = 1;
 657
 658                         nlmsvc_insert_block(block, 0);
 659                         svc_wake_up(block->b_daemon);
 660                         rc = 0;
 661                         break;
 662                 }
 663         }
 664         unlock_kernel();
 665         if (rc == -ENOENT)
 666                 printk(KERN_WARNING "lockd: grant for unknown block\n");
 667         return rc;
 668 }
 669
 670 /*
 671  * Unblock a blocked lock request. This is a callback invoked from the
 672  * VFS layer when a lock on which we blocked is removed.
 673  *
 674  * This function doesn't grant the blocked lock instantly, but rather moves
 675  * the block to the head of nlm_blocked where it can be picked up by lockd.
 676  */
 677 static void
 678 nlmsvc_notify_blocked(struct file_lock *fl)
 679 {
 680         struct nlm_block        *block;
 681
 682         dprintk("lockd: VFS unblock notification for block %p\n", fl);
 683         list_for_each_entry(block, &nlm_blocked, b_list) {
 684                 if (nlm_compare_locks(&block->b_call->a_args.lock.fl, fl)) {
 685                         nlmsvc_insert_block(block, 0);
 686                         svc_wake_up(block->b_daemon);
 687                         return;
 688                 }
 689         }
 690
 691         printk(KERN_WARNING "lockd: notification for unknown block!\n");
 692 }
 693
 694 static int nlmsvc_same_owner(struct file_lock *fl1, struct file_lock *fl2)
 695 {
 696         return fl1->fl_owner == fl2->fl_owner && fl1->fl_pid == fl2->fl_pid;
 697 }
 698
 699 struct lock_manager_operations nlmsvc_lock_operations = {
 700         .fl_compare_owner = nlmsvc_same_owner,
 701         .fl_notify = nlmsvc_notify_blocked,
 702         .fl_grant = nlmsvc_grant_deferred,
 703 };
 704
 705 /*
 706  * Try to claim a lock that was previously blocked.
 707  *
 708  * Note that we use both the RPC_GRANTED_MSG call _and_ an async
 709  * RPC thread when notifying the client. This seems like overkill...
 710  * Here's why:
 711  *  -   we don't want to use a synchronous RPC thread, otherwise
 712  *      we might find ourselves hanging on a dead portmapper.
 713  *  -   Some lockd implementations (e.g. HP) don't react to
 714  *      RPC_GRANTED calls; they seem to insist on RPC_GRANTED_MSG calls.
 715  */
 716 static void
 717 nlmsvc_grant_blocked(struct nlm_block *block)
 718 {
 719         struct nlm_file         *file = block->b_file;
 720         struct nlm_lock         *lock = &block->b_call->a_args.lock;
 721         int                     error;
 722
 723         dprintk("lockd: grant blocked lock %p\n", block);
 724
 725         kref_get(&block->b_count);
 726
 727         /* Unlink block request from list */
 728         nlmsvc_unlink_block(block);
 729
 730         /* If b_granted is true this means we've been here before.
 731          * Just retry the grant callback, possibly refreshing the RPC
 732          * binding */
 733         if (block->b_granted) {
 734                 nlm_rebind_host(block->b_host);
 735                 goto callback;
 736         }
 737
 738         /* Try the lock operation again */
 739         lock->fl.fl_flags |= FL_SLEEP;
 740         error = vfs_lock_file(file->f_file, F_SETLK, &lock->fl, NULL);
 741         lock->fl.fl_flags &= ~FL_SLEEP;
 742
 743         switch (error) {
 744         case 0:
 745                 break;
 746         case FILE_LOCK_DEFERRED:
 747                 dprintk("lockd: lock still blocked error %d\n", error);
 748                 nlmsvc_insert_block(block, NLM_NEVER);
 749                 nlmsvc_release_block(block);
 750                 return;
 751         default:
 752                 printk(KERN_WARNING "lockd: unexpected error %d in %s!\n",
 753                                 -error, __func__);
 754                 nlmsvc_insert_block(block, 10 * HZ);
 755                 nlmsvc_release_block(block);
 756                 return;
 757         }
 758
 759 callback:
 760         /* Lock was granted by VFS. */
 761         dprintk("lockd: GRANTing blocked lock.\n");
 762         block->b_granted = 1;
 763
 764         /* keep block on the list, but don't reattempt until the RPC
 765          * completes or the submission fails
 766          */
 767         nlmsvc_insert_block(block, NLM_NEVER);
 768
 769         /* Call the client -- use a soft RPC task since nlmsvc_retry_blocked
 770          * will queue up a new one if this one times out
 771          */
 772         error = nlm_async_call(block->b_call, NLMPROC_GRANTED_MSG,
 773                                 &nlmsvc_grant_ops);
 774
 775         /* RPC submission failed, wait a bit and retry */
 776         if (error < 0)
 777                 nlmsvc_insert_block(block, 10 * HZ);
 778 }
 779
 780 /*
 781  * This is the callback from the RPC layer when the NLM_GRANTED_MSG
 782  * RPC call has succeeded or timed out.
 783  * Like all RPC callbacks, it is invoked by the rpciod process, so it
 784  * better not sleep. Therefore, we put the blocked lock on the nlm_blocked
 785  * chain once more in order to have it removed by lockd itself (which can
 786  * then sleep on the file semaphore without disrupting e.g. the nfs client).
 787  */
 788 static void nlmsvc_grant_callback(struct rpc_task *task, void *data)
 789 {
 790         struct nlm_rqst         *call = data;
 791         struct nlm_block        *block = call->a_block;
 792         unsigned long           timeout;
 793
 794         dprintk("lockd: GRANT_MSG RPC callback\n");
 795
 796         lock_kernel();
 797         /* if the block is not on a list at this point then it has
 798          * been invalidated. Don't try to requeue it.
 799          *
 800          * FIXME: it's possible that the block is removed from the list
 801          * after this check but before the nlmsvc_insert_block. In that
 802          * case it will be added back. Perhaps we need better locking
 803          * for nlm_blocked?
 804          */
 805         if (list_empty(&block->b_list))
 806                 goto out;
 807
 808         /* Technically, we should down the file semaphore here. Since we
 809          * move the block towards the head of the queue only, no harm
 810          * can be done, though. */
 811         if (task->tk_status < 0) {
 812                 /* RPC error: Re-insert for retransmission */
 813                 timeout = 10 * HZ;
 814         } else {
 815                 /* Call was successful, now wait for client callback */
 816                 timeout = 60 * HZ;
 817         }
 818         nlmsvc_insert_block(block, timeout);
 819         svc_wake_up(block->b_daemon);
 820 out:
 821         unlock_kernel();
 822 }
 823
 824 static void nlmsvc_grant_release(void *data)
 825 {
 826         struct nlm_rqst         *call = data;
 827
 828         lock_kernel();
 829         nlmsvc_release_block(call->a_block);
 830         unlock_kernel();
 831 }
 832
 833 static const struct rpc_call_ops nlmsvc_grant_ops = {
 834         .rpc_call_done = nlmsvc_grant_callback,
 835         .rpc_release = nlmsvc_grant_release,
 836 };
 837
 838 /*
 839  * We received a GRANT_RES callback. Try to find the corresponding
 840  * block.
 841  */
 842 void
 843 nlmsvc_grant_reply(struct nlm_cookie *cookie, __be32 status)
 844 {
 845         struct nlm_block        *block;
 846
 847         dprintk("grant_reply: looking for cookie %x, s=%d \n",
 848                 *(unsigned int *)(cookie->data), status);
 849         if (!(block = nlmsvc_find_block(cookie)))
 850                 return;
 851
 852         if (block) {
 853                 if (status == nlm_lck_denied_grace_period) {
 854                         /* Try again in a couple of seconds */
 855                         nlmsvc_insert_block(block, 10 * HZ);
 856                 } else {
 857                         /* Lock is now held by client, or has been rejected.
 858                          * In both cases, the block should be removed. */
 859                         nlmsvc_unlink_block(block);
 860                 }
 861         }
 862         nlmsvc_release_block(block);
 863 }
 864
 865 /* Helper function to handle retry of a deferred block.
 866  * If it is a blocking lock, call grant_blocked.
 867  * For a non-blocking lock or test lock, revisit the request.
 868  */
 869 static void
 870 retry_deferred_block(struct nlm_block *block)
 871 {
 872         if (!(block->b_flags & B_GOT_CALLBACK))
 873                 block->b_flags |= B_TIMED_OUT;
 874         nlmsvc_insert_block(block, NLM_TIMEOUT);
 875         dprintk("revisit block %p flags %d\n",  block, block->b_flags);
 876         if (block->b_deferred_req) {
 877                 block->b_deferred_req->revisit(block->b_deferred_req, 0);
 878                 block->b_deferred_req = NULL;
 879         }
 880 }
 881
 882 /*
 883  * Retry all blocked locks that have been notified. This is where lockd
 884  * picks up locks that can be granted, or grant notifications that must
 885  * be retransmitted.
 886  */
 887 unsigned long
 888 nlmsvc_retry_blocked(void)
 889 {
 890         unsigned long   timeout = MAX_SCHEDULE_TIMEOUT;
 891         struct nlm_block *block;
 892
 893         while (!list_empty(&nlm_blocked) && !kthread_should_stop()) {
 894                 block = list_entry(nlm_blocked.next, struct nlm_block, b_list);
 895
 896                 if (block->b_when == NLM_NEVER)
 897                         break;
 898                 if (time_after(block->b_when, jiffies)) {
 899                         timeout = block->b_when - jiffies;
 900                         break;
 901                 }
 902
 903                 dprintk("nlmsvc_retry_blocked(%p, when=%ld)\n",
 904                         block, block->b_when);
 905                 if (block->b_flags & B_QUEUED) {
 906                         dprintk("nlmsvc_retry_blocked delete block (%p, granted=%d, flags=%d)\n",
 907                                 block, block->b_granted, block->b_flags);
 908                         retry_deferred_block(block);
 909                 } else
 910                         nlmsvc_grant_blocked(block);
 911         }
 912
 913         return timeout;
 914 }