svclock.c

   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * linux/fs/lockd/svclock.c
   4 *
   5 * Handling of server-side locks, mostly of the blocked variety.
   6 * This is the ugliest part of lockd because we tread on very thin ice.
   7 * GRANT and CANCEL calls may get stuck, meet in mid-flight, etc.
   8 * IMNSHO introducing the grant callback into the NLM protocol was one
   9 * of the worst ideas Sun ever had. Except maybe for the idea of doing
  10 * NFS file locking at all.
  11 *
  12 * I'm trying hard to avoid race conditions by protecting most accesses
  13 * to a file's list of blocked locks through a semaphore. The global
  14 * list of blocked locks is not protected in this fashion however.
  15 * Therefore, some functions (such as the RPC callback for the async grant
  16 * call) move blocked locks towards the head of the list *while some other
  17 * process might be traversing it*. This should not be a problem in
  18 * practice, because this will only cause functions traversing the list
  19 * to visit some blocks twice.
  20 *
  21 * Copyright (C) 1996, Olaf Kirch <okir@monad.swb.de>
  22 */
  23
  24#include <linux/types.h>
  25#include <linux/slab.h>
  26#include <linux/errno.h>
  27#include <linux/kernel.h>
  28#include <linux/sched.h>
  29#include <linux/sunrpc/clnt.h>
  30#include <linux/sunrpc/svc_xprt.h>
  31#include <linux/lockd/nlm.h>
  32#include <linux/lockd/lockd.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
  42static void nlmsvc_release_block(struct nlm_block *block);
  43static void	nlmsvc_insert_block(struct nlm_block *block, unsigned long);
  44static void	nlmsvc_remove_block(struct nlm_block *block);
  45
  46static int nlmsvc_setgrantargs(struct nlm_rqst *call, struct nlm_lock *lock);
  47static void nlmsvc_freegrantargs(struct nlm_rqst *call);
  48static const struct rpc_call_ops nlmsvc_grant_ops;
  49
  50/*
  51 * The list of blocked locks to retry
  52 */
  53static LIST_HEAD(nlm_blocked);
  54static DEFINE_SPINLOCK(nlm_blocked_lock);
  55
  56#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
  57static const char *nlmdbg_cookie2a(const struct nlm_cookie *cookie)
  58{
  59	/*
  60	 * We can get away with a static buffer because this is only called
  61	 * from lockd, which is single-threaded.
  62	 */
  63	static char buf[2*NLM_MAXCOOKIELEN+1];
  64	unsigned int i, len = sizeof(buf);
  65	char *p = buf;
  66
  67	len--;	/* allow for trailing \0 */
  68	if (len < 3)
  69		return "???";
  70	for (i = 0 ; i < cookie->len ; i++) {
  71		if (len < 2) {
  72			strcpy(p-3, "...");
  73			break;
  74		}
  75		sprintf(p, "%02x", cookie->data[i]);
  76		p += 2;
  77		len -= 2;
  78	}
  79	*p = '\0';
  80
  81	return buf;
  82}
  83#endif
  84
  85/*
  86 * Insert a blocked lock into the global list
  87 */
  88static void
  89nlmsvc_insert_block_locked(struct nlm_block *block, unsigned long when)
  90{
  91	struct nlm_block *b;
  92	struct list_head *pos;
  93
  94	dprintk("lockd: nlmsvc_insert_block(%p, %ld)\n", block, when);
  95	if (list_empty(&block->b_list)) {
  96		kref_get(&block->b_count);
  97	} else {
  98		list_del_init(&block->b_list);
  99	}
 100
 101	pos = &nlm_blocked;
 102	if (when != NLM_NEVER) {
 103		if ((when += jiffies) == NLM_NEVER)
 104			when ++;
 105		list_for_each(pos, &nlm_blocked) {
 106			b = list_entry(pos, struct nlm_block, b_list);
 107			if (time_after(b->b_when,when) || b->b_when == NLM_NEVER)
 108				break;
 109		}
 110		/* On normal exit from the loop, pos == &nlm_blocked,
 111		 * so we will be adding to the end of the list - good
 112		 */
 113	}
 114
 115	list_add_tail(&block->b_list, pos);
 116	block->b_when = when;
 117}
 118
 119static void nlmsvc_insert_block(struct nlm_block *block, unsigned long when)
 120{
 121	spin_lock(&nlm_blocked_lock);
 122	nlmsvc_insert_block_locked(block, when);
 123	spin_unlock(&nlm_blocked_lock);
 124}
 125
 126/*
 127 * Remove a block from the global list
 128 */
 129static inline void
 130nlmsvc_remove_block(struct nlm_block *block)
 131{
 132	spin_lock(&nlm_blocked_lock);
 133	if (!list_empty(&block->b_list)) {
 134		list_del_init(&block->b_list);
 135		spin_unlock(&nlm_blocked_lock);
 136		nlmsvc_release_block(block);
 137		return;
 138	}
 139	spin_unlock(&nlm_blocked_lock);
 140}
 141
 142/*
 143 * Find a block for a given lock
 144 */
 145static struct nlm_block *
 146nlmsvc_lookup_block(struct nlm_file *file, struct nlm_lock *lock)
 147{
 148	struct nlm_block	*block;
 149	struct file_lock	*fl;
 150
 151	dprintk("lockd: nlmsvc_lookup_block f=%p pd=%d %Ld-%Ld ty=%d\n",
 152				file, lock->fl.c.flc_pid,
 153				(long long)lock->fl.fl_start,
 154				(long long)lock->fl.fl_end,
 155				lock->fl.c.flc_type);
 156	spin_lock(&nlm_blocked_lock);
 157	list_for_each_entry(block, &nlm_blocked, b_list) {
 158		fl = &block->b_call->a_args.lock.fl;
 159		dprintk("lockd: check f=%p pd=%d %Ld-%Ld ty=%d cookie=%s\n",
 160				block->b_file, fl->c.flc_pid,
 161				(long long)fl->fl_start,
 162				(long long)fl->fl_end, fl->c.flc_type,
 163				nlmdbg_cookie2a(&block->b_call->a_args.cookie));
 164		if (block->b_file == file && nlm_compare_locks(fl, &lock->fl)) {
 165			kref_get(&block->b_count);
 166			spin_unlock(&nlm_blocked_lock);
 167			return block;
 168		}
 169	}
 170	spin_unlock(&nlm_blocked_lock);
 171
 172	return NULL;
 173}
 174
 175static inline int nlm_cookie_match(struct nlm_cookie *a, struct nlm_cookie *b)
 176{
 177	if (a->len != b->len)
 178		return 0;
 179	if (memcmp(a->data, b->data, a->len))
 180		return 0;
 181	return 1;
 182}
 183
 184/*
 185 * Find a block with a given NLM cookie.
 186 */
 187static inline struct nlm_block *
 188nlmsvc_find_block(struct nlm_cookie *cookie)
 189{
 190	struct nlm_block *block;
 191
 192	spin_lock(&nlm_blocked_lock);
 193	list_for_each_entry(block, &nlm_blocked, b_list) {
 194		if (nlm_cookie_match(&block->b_call->a_args.cookie,cookie))
 195			goto found;
 196	}
 197	spin_unlock(&nlm_blocked_lock);
 198
 199	return NULL;
 200
 201found:
 202	dprintk("nlmsvc_find_block(%s): block=%p\n", nlmdbg_cookie2a(cookie), block);
 203	kref_get(&block->b_count);
 204	spin_unlock(&nlm_blocked_lock);
 205	return block;
 206}
 207
 208/*
 209 * Create a block and initialize it.
 210 *
 211 * Note: we explicitly set the cookie of the grant reply to that of
 212 * the blocked lock request. The spec explicitly mentions that the client
 213 * should _not_ rely on the callback containing the same cookie as the
 214 * request, but (as I found out later) that's because some implementations
 215 * do just this. Never mind the standards comittees, they support our
 216 * logging industries.
 217 *
 218 * 10 years later: I hope we can safely ignore these old and broken
 219 * clients by now. Let's fix this so we can uniquely identify an incoming
 220 * GRANTED_RES message by cookie, without having to rely on the client's IP
 221 * address. --okir
 222 */
 223static struct nlm_block *
 224nlmsvc_create_block(struct svc_rqst *rqstp, struct nlm_host *host,
 225		    struct nlm_file *file, struct nlm_lock *lock,
 226		    struct nlm_cookie *cookie)
 227{
 228	struct nlm_block	*block;
 229	struct nlm_rqst		*call = NULL;
 230
 231	call = nlm_alloc_call(host);
 232	if (call == NULL)
 233		return NULL;
 234
 235	/* Allocate memory for block, and initialize arguments */
 236	block = kzalloc(sizeof(*block), GFP_KERNEL);
 237	if (block == NULL)
 238		goto failed;
 239	kref_init(&block->b_count);
 240	INIT_LIST_HEAD(&block->b_list);
 241	INIT_LIST_HEAD(&block->b_flist);
 242
 243	if (!nlmsvc_setgrantargs(call, lock))
 244		goto failed_free;
 245
 246	/* Set notifier function for VFS, and init args */
 247	call->a_args.lock.fl.c.flc_flags |= FL_SLEEP;
 248	call->a_args.lock.fl.fl_lmops = &nlmsvc_lock_operations;
 249	nlmclnt_next_cookie(&call->a_args.cookie);
 250
 251	dprintk("lockd: created block %p...\n", block);
 252
 253	/* Create and initialize the block */
 254	block->b_daemon = rqstp->rq_server;
 255	block->b_host   = host;
 256	block->b_file   = file;
 257	file->f_count++;
 258
 259	/* Add to file's list of blocks */
 260	list_add(&block->b_flist, &file->f_blocks);
 261
 262	/* Set up RPC arguments for callback */
 263	block->b_call = call;
 264	call->a_flags   = RPC_TASK_ASYNC;
 265	call->a_block = block;
 266
 267	return block;
 268
 269failed_free:
 270	kfree(block);
 271failed:
 272	nlmsvc_release_call(call);
 273	return NULL;
 274}
 275
 276/*
 277 * Delete a block.
 278 * It is the caller's responsibility to check whether the file
 279 * can be closed hereafter.
 280 */
 281static int nlmsvc_unlink_block(struct nlm_block *block)
 282{
 283	int status;
 284	dprintk("lockd: unlinking block %p...\n", block);
 285
 286	/* Remove block from list */
 287	status = locks_delete_block(&block->b_call->a_args.lock.fl);
 288	nlmsvc_remove_block(block);
 289	return status;
 290}
 291
 292static void nlmsvc_free_block(struct kref *kref)
 293{
 294	struct nlm_block *block = container_of(kref, struct nlm_block, b_count);
 295	struct nlm_file		*file = block->b_file;
 296
 297	dprintk("lockd: freeing block %p...\n", block);
 298
 299	/* Remove block from file's list of blocks */
 300	list_del_init(&block->b_flist);
 301	mutex_unlock(&file->f_mutex);
 302
 303	nlmsvc_freegrantargs(block->b_call);
 304	nlmsvc_release_call(block->b_call);
 305	nlm_release_file(block->b_file);
 306	kfree(block);
 307}
 308
 309static void nlmsvc_release_block(struct nlm_block *block)
 310{
 311	if (block != NULL)
 312		kref_put_mutex(&block->b_count, nlmsvc_free_block, &block->b_file->f_mutex);
 313}
 314
 315/*
 316 * Loop over all blocks and delete blocks held by
 317 * a matching host.
 318 */
 319void nlmsvc_traverse_blocks(struct nlm_host *host,
 320			struct nlm_file *file,
 321			nlm_host_match_fn_t match)
 322{
 323	struct nlm_block *block, *next;
 324
 325restart:
 326	mutex_lock(&file->f_mutex);
 327	spin_lock(&nlm_blocked_lock);
 328	list_for_each_entry_safe(block, next, &file->f_blocks, b_flist) {
 329		if (!match(block->b_host, host))
 330			continue;
 331		/* Do not destroy blocks that are not on
 332		 * the global retry list - why? */
 333		if (list_empty(&block->b_list))
 334			continue;
 335		kref_get(&block->b_count);
 336		spin_unlock(&nlm_blocked_lock);
 337		mutex_unlock(&file->f_mutex);
 338		nlmsvc_unlink_block(block);
 339		nlmsvc_release_block(block);
 340		goto restart;
 341	}
 342	spin_unlock(&nlm_blocked_lock);
 343	mutex_unlock(&file->f_mutex);
 344}
 345
 346static struct nlm_lockowner *
 347nlmsvc_get_lockowner(struct nlm_lockowner *lockowner)
 348{
 349	refcount_inc(&lockowner->count);
 350	return lockowner;
 351}
 352
 353void nlmsvc_put_lockowner(struct nlm_lockowner *lockowner)
 354{
 355	if (!refcount_dec_and_lock(&lockowner->count, &lockowner->host->h_lock))
 356		return;
 357	list_del(&lockowner->list);
 358	spin_unlock(&lockowner->host->h_lock);
 359	nlmsvc_release_host(lockowner->host);
 360	kfree(lockowner);
 361}
 362
 363static struct nlm_lockowner *__nlmsvc_find_lockowner(struct nlm_host *host, pid_t pid)
 364{
 365	struct nlm_lockowner *lockowner;
 366	list_for_each_entry(lockowner, &host->h_lockowners, list) {
 367		if (lockowner->pid != pid)
 368			continue;
 369		return nlmsvc_get_lockowner(lockowner);
 370	}
 371	return NULL;
 372}
 373
 374static struct nlm_lockowner *nlmsvc_find_lockowner(struct nlm_host *host, pid_t pid)
 375{
 376	struct nlm_lockowner *res, *new = NULL;
 377
 378	spin_lock(&host->h_lock);
 379	res = __nlmsvc_find_lockowner(host, pid);
 380
 381	if (res == NULL) {
 382		spin_unlock(&host->h_lock);
 383		new = kmalloc(sizeof(*res), GFP_KERNEL);
 384		spin_lock(&host->h_lock);
 385		res = __nlmsvc_find_lockowner(host, pid);
 386		if (res == NULL && new != NULL) {
 387			res = new;
 388			/* fs/locks.c will manage the refcount through lock_ops */
 389			refcount_set(&new->count, 1);
 390			new->pid = pid;
 391			new->host = nlm_get_host(host);
 392			list_add(&new->list, &host->h_lockowners);
 393			new = NULL;
 394		}
 395	}
 396
 397	spin_unlock(&host->h_lock);
 398	kfree(new);
 399	return res;
 400}
 401
 402void
 403nlmsvc_release_lockowner(struct nlm_lock *lock)
 404{
 405	if (lock->fl.c.flc_owner)
 406		nlmsvc_put_lockowner(lock->fl.c.flc_owner);
 407}
 408
 409void nlmsvc_locks_init_private(struct file_lock *fl, struct nlm_host *host,
 410						pid_t pid)
 411{
 412	fl->c.flc_owner = nlmsvc_find_lockowner(host, pid);
 413}
 414
 415/*
 416 * Initialize arguments for GRANTED call. The nlm_rqst structure
 417 * has been cleared already.
 418 */
 419static int nlmsvc_setgrantargs(struct nlm_rqst *call, struct nlm_lock *lock)
 420{
 421	locks_copy_lock(&call->a_args.lock.fl, &lock->fl);
 422	memcpy(&call->a_args.lock.fh, &lock->fh, sizeof(call->a_args.lock.fh));
 423	call->a_args.lock.caller = utsname()->nodename;
 424	call->a_args.lock.oh.len = lock->oh.len;
 425
 426	/* set default data area */
 427	call->a_args.lock.oh.data = call->a_owner;
 428	call->a_args.lock.svid = ((struct nlm_lockowner *) lock->fl.c.flc_owner)->pid;
 429
 430	if (lock->oh.len > NLMCLNT_OHSIZE) {
 431		void *data = kmalloc(lock->oh.len, GFP_KERNEL);
 432		if (!data)
 433			return 0;
 434		call->a_args.lock.oh.data = (u8 *) data;
 435	}
 436
 437	memcpy(call->a_args.lock.oh.data, lock->oh.data, lock->oh.len);
 438	return 1;
 439}
 440
 441static void nlmsvc_freegrantargs(struct nlm_rqst *call)
 442{
 443	if (call->a_args.lock.oh.data != call->a_owner)
 444		kfree(call->a_args.lock.oh.data);
 445
 446	locks_release_private(&call->a_args.lock.fl);
 447}
 448
 449/*
 450 * Deferred lock request handling for non-blocking lock
 451 */
 452static __be32
 453nlmsvc_defer_lock_rqst(struct svc_rqst *rqstp, struct nlm_block *block)
 454{
 455	__be32 status = nlm_lck_denied_nolocks;
 456
 457	block->b_flags |= B_QUEUED;
 458
 459	nlmsvc_insert_block(block, NLM_TIMEOUT);
 460
 461	block->b_cache_req = &rqstp->rq_chandle;
 462	if (rqstp->rq_chandle.defer) {
 463		block->b_deferred_req =
 464			rqstp->rq_chandle.defer(block->b_cache_req);
 465		if (block->b_deferred_req != NULL)
 466			status = nlm_drop_reply;
 467	}
 468	dprintk("lockd: nlmsvc_defer_lock_rqst block %p flags %d status %d\n",
 469		block, block->b_flags, ntohl(status));
 470
 471	return status;
 472}
 473
 474/*
 475 * Attempt to establish a lock, and if it can't be granted, block it
 476 * if required.
 477 */
 478__be32
 479nlmsvc_lock(struct svc_rqst *rqstp, struct nlm_file *file,
 480	    struct nlm_host *host, struct nlm_lock *lock, int wait,
 481	    struct nlm_cookie *cookie, int reclaim)
 482{
 483	struct inode		*inode __maybe_unused = nlmsvc_file_inode(file);
 484	struct nlm_block	*block = NULL;
 485	int			error;
 486	int			mode;
 487	int			async_block = 0;
 488	__be32			ret;
 489
 490	dprintk("lockd: nlmsvc_lock(%s/%ld, ty=%d, pi=%d, %Ld-%Ld, bl=%d)\n",
 491				inode->i_sb->s_id, inode->i_ino,
 492				lock->fl.c.flc_type,
 493				lock->fl.c.flc_pid,
 494				(long long)lock->fl.fl_start,
 495				(long long)lock->fl.fl_end,
 496				wait);
 497
 498	if (nlmsvc_file_cannot_lock(file))
 499		return nlm_lck_denied_nolocks;
 500
 501	if (!locks_can_async_lock(nlmsvc_file_file(file)->f_op)) {
 502		async_block = wait;
 503		wait = 0;
 504	}
 505
 506	/* Lock file against concurrent access */
 507	mutex_lock(&file->f_mutex);
 508	/* Get existing block (in case client is busy-waiting)
 509	 * or create new block
 510	 */
 511	block = nlmsvc_lookup_block(file, lock);
 512	if (block == NULL) {
 513		block = nlmsvc_create_block(rqstp, host, file, lock, cookie);
 514		ret = nlm_lck_denied_nolocks;
 515		if (block == NULL)
 516			goto out;
 517		lock = &block->b_call->a_args.lock;
 518	} else
 519		lock->fl.c.flc_flags &= ~FL_SLEEP;
 520
 521	if (block->b_flags & B_QUEUED) {
 522		dprintk("lockd: nlmsvc_lock deferred block %p flags %d\n",
 523							block, block->b_flags);
 524		if (block->b_granted) {
 525			nlmsvc_unlink_block(block);
 526			ret = nlm_granted;
 527			goto out;
 528		}
 529		if (block->b_flags & B_TIMED_OUT) {
 530			nlmsvc_unlink_block(block);
 531			ret = nlm_lck_denied;
 532			goto out;
 533		}
 534		ret = nlm_drop_reply;
 535		goto out;
 536	}
 537
 538	if (locks_in_grace(SVC_NET(rqstp)) && !reclaim) {
 539		ret = nlm_lck_denied_grace_period;
 540		goto out;
 541	}
 542	if (reclaim && !locks_in_grace(SVC_NET(rqstp))) {
 543		ret = nlm_lck_denied_grace_period;
 544		goto out;
 545	}
 546
 547	spin_lock(&nlm_blocked_lock);
 548	/*
 549	 * If this is a lock request for an already pending
 550	 * lock request we return nlm_lck_blocked without calling
 551	 * vfs_lock_file() again. Otherwise we have two pending
 552	 * requests on the underlaying ->lock() implementation but
 553	 * only one nlm_block to being granted by lm_grant().
 554	 */
 555	if (locks_can_async_lock(nlmsvc_file_file(file)->f_op) &&
 556	    !list_empty(&block->b_list)) {
 557		spin_unlock(&nlm_blocked_lock);
 558		ret = nlm_lck_blocked;
 559		goto out;
 560	}
 561
 562	/* Append to list of blocked */
 563	nlmsvc_insert_block_locked(block, NLM_NEVER);
 564	spin_unlock(&nlm_blocked_lock);
 565
 566	if (!wait)
 567		lock->fl.c.flc_flags &= ~FL_SLEEP;
 568	mode = lock_to_openmode(&lock->fl);
 569	error = vfs_lock_file(file->f_file[mode], F_SETLK, &lock->fl, NULL);
 570	lock->fl.c.flc_flags &= ~FL_SLEEP;
 571
 572	dprintk("lockd: vfs_lock_file returned %d\n", error);
 573	switch (error) {
 574		case 0:
 575			nlmsvc_remove_block(block);
 576			ret = nlm_granted;
 577			goto out;
 578		case -EAGAIN:
 579			if (!wait)
 580				nlmsvc_remove_block(block);
 581			ret = async_block ? nlm_lck_blocked : nlm_lck_denied;
 582			goto out;
 583		case FILE_LOCK_DEFERRED:
 584			if (wait)
 585				break;
 586			/* Filesystem lock operation is in progress
 587			   Add it to the queue waiting for callback */
 588			ret = nlmsvc_defer_lock_rqst(rqstp, block);
 589			goto out;
 590		case -EDEADLK:
 591			nlmsvc_remove_block(block);
 592			ret = nlm_deadlock;
 593			goto out;
 594		default:			/* includes ENOLCK */
 595			nlmsvc_remove_block(block);
 596			ret = nlm_lck_denied_nolocks;
 597			goto out;
 598	}
 599
 600	ret = nlm_lck_blocked;
 601out:
 602	mutex_unlock(&file->f_mutex);
 603	nlmsvc_release_block(block);
 604	dprintk("lockd: nlmsvc_lock returned %u\n", ret);
 605	return ret;
 606}
 607
 608/*
 609 * Test for presence of a conflicting lock.
 610 */
 611__be32
 612nlmsvc_testlock(struct svc_rqst *rqstp, struct nlm_file *file,
 613		struct nlm_host *host, struct nlm_lock *lock,
 614		struct nlm_lock *conflock)
 615{
 616	int			error;
 617	int			mode;
 618	__be32			ret;
 619
 620	dprintk("lockd: nlmsvc_testlock(%s/%ld, ty=%d, %Ld-%Ld)\n",
 621				nlmsvc_file_inode(file)->i_sb->s_id,
 622				nlmsvc_file_inode(file)->i_ino,
 623				lock->fl.c.flc_type,
 624				(long long)lock->fl.fl_start,
 625				(long long)lock->fl.fl_end);
 626
 627	if (nlmsvc_file_cannot_lock(file))
 628		return nlm_lck_denied_nolocks;
 629
 630	if (locks_in_grace(SVC_NET(rqstp))) {
 631		ret = nlm_lck_denied_grace_period;
 632		goto out;
 633	}
 634
 635	mode = lock_to_openmode(&lock->fl);
 636	locks_init_lock(&conflock->fl);
 637	/* vfs_test_lock only uses start, end, and owner, but tests flc_file */
 638	conflock->fl.c.flc_file = lock->fl.c.flc_file;
 639	conflock->fl.fl_start = lock->fl.fl_start;
 640	conflock->fl.fl_end = lock->fl.fl_end;
 641	conflock->fl.c.flc_owner = lock->fl.c.flc_owner;
 642	error = vfs_test_lock(file->f_file[mode], &conflock->fl);
 643	if (error) {
 644		/* We can't currently deal with deferred test requests */
 645		if (error == FILE_LOCK_DEFERRED)
 646			WARN_ON_ONCE(1);
 647
 648		ret = nlm_lck_denied_nolocks;
 649		goto out;
 650	}
 651
 652	if (conflock->fl.c.flc_type == F_UNLCK) {
 653		ret = nlm_granted;
 654		goto out;
 655	}
 656
 657	dprintk("lockd: conflicting lock(ty=%d, %Ld-%Ld)\n",
 658		conflock->fl.c.flc_type, (long long)conflock->fl.fl_start,
 659		(long long)conflock->fl.fl_end);
 660	conflock->caller = "somehost";	/* FIXME */
 661	conflock->len = strlen(conflock->caller);
 662	conflock->oh.len = 0;		/* don't return OH info */
 663	conflock->svid = conflock->fl.c.flc_pid;
 664	locks_release_private(&conflock->fl);
 665
 666	ret = nlm_lck_denied;
 667out:
 668	return ret;
 669}
 670
 671/*
 672 * Remove a lock.
 673 * This implies a CANCEL call: We send a GRANT_MSG, the client replies
 674 * with a GRANT_RES call which gets lost, and calls UNLOCK immediately
 675 * afterwards. In this case the block will still be there, and hence
 676 * must be removed.
 677 */
 678__be32
 679nlmsvc_unlock(struct net *net, struct nlm_file *file, struct nlm_lock *lock)
 680{
 681	int	error = 0;
 682
 683	dprintk("lockd: nlmsvc_unlock(%s/%ld, pi=%d, %Ld-%Ld)\n",
 684				nlmsvc_file_inode(file)->i_sb->s_id,
 685				nlmsvc_file_inode(file)->i_ino,
 686				lock->fl.c.flc_pid,
 687				(long long)lock->fl.fl_start,
 688				(long long)lock->fl.fl_end);
 689
 690	if (nlmsvc_file_cannot_lock(file))
 691		return nlm_lck_denied_nolocks;
 692
 693	/* First, cancel any lock that might be there */
 694	nlmsvc_cancel_blocked(net, file, lock);
 695
 696	lock->fl.c.flc_type = F_UNLCK;
 697	lock->fl.c.flc_file = file->f_file[O_RDONLY];
 698	if (lock->fl.c.flc_file)
 699		error = vfs_lock_file(lock->fl.c.flc_file, F_SETLK,
 700					&lock->fl, NULL);
 701	lock->fl.c.flc_file = file->f_file[O_WRONLY];
 702	if (lock->fl.c.flc_file)
 703		error |= vfs_lock_file(lock->fl.c.flc_file, F_SETLK,
 704					&lock->fl, NULL);
 705
 706	return (error < 0)? nlm_lck_denied_nolocks : nlm_granted;
 707}
 708
 709/*
 710 * Cancel a previously blocked request.
 711 *
 712 * A cancel request always overrides any grant that may currently
 713 * be in progress.
 714 * The calling procedure must check whether the file can be closed.
 715 */
 716__be32
 717nlmsvc_cancel_blocked(struct net *net, struct nlm_file *file, struct nlm_lock *lock)
 718{
 719	struct nlm_block	*block;
 720	int status = 0;
 721	int mode;
 722
 723	dprintk("lockd: nlmsvc_cancel(%s/%ld, pi=%d, %Ld-%Ld)\n",
 724				nlmsvc_file_inode(file)->i_sb->s_id,
 725				nlmsvc_file_inode(file)->i_ino,
 726				lock->fl.c.flc_pid,
 727				(long long)lock->fl.fl_start,
 728				(long long)lock->fl.fl_end);
 729
 730	if (nlmsvc_file_cannot_lock(file))
 731		return nlm_lck_denied_nolocks;
 732
 733	if (locks_in_grace(net))
 734		return nlm_lck_denied_grace_period;
 735
 736	mutex_lock(&file->f_mutex);
 737	block = nlmsvc_lookup_block(file, lock);
 738	mutex_unlock(&file->f_mutex);
 739	if (block != NULL) {
 740		struct file_lock *fl = &block->b_call->a_args.lock.fl;
 741
 742		mode = lock_to_openmode(fl);
 743		vfs_cancel_lock(block->b_file->f_file[mode], fl);
 744		status = nlmsvc_unlink_block(block);
 745		nlmsvc_release_block(block);
 746	}
 747	return status ? nlm_lck_denied : nlm_granted;
 748}
 749
 750/*
 751 * This is a callback from the filesystem for VFS file lock requests.
 752 * It will be used if lm_grant is defined and the filesystem can not
 753 * respond to the request immediately.
 754 * For SETLK or SETLKW request it will get the local posix lock.
 755 * In all cases it will move the block to the head of nlm_blocked q where
 756 * nlmsvc_retry_blocked() can send back a reply for SETLKW or revisit the
 757 * deferred rpc for GETLK and SETLK.
 758 */
 759static void
 760nlmsvc_update_deferred_block(struct nlm_block *block, int result)
 761{
 762	block->b_flags |= B_GOT_CALLBACK;
 763	if (result == 0)
 764		block->b_granted = 1;
 765	else
 766		block->b_flags |= B_TIMED_OUT;
 767}
 768
 769static int nlmsvc_grant_deferred(struct file_lock *fl, int result)
 770{
 771	struct nlm_block *block;
 772	int rc = -ENOENT;
 773
 774	spin_lock(&nlm_blocked_lock);
 775	list_for_each_entry(block, &nlm_blocked, b_list) {
 776		if (nlm_compare_locks(&block->b_call->a_args.lock.fl, fl)) {
 777			dprintk("lockd: nlmsvc_notify_blocked block %p flags %d\n",
 778							block, block->b_flags);
 779			if (block->b_flags & B_QUEUED) {
 780				if (block->b_flags & B_TIMED_OUT) {
 781					rc = -ENOLCK;
 782					break;
 783				}
 784				nlmsvc_update_deferred_block(block, result);
 785			} else if (result == 0)
 786				block->b_granted = 1;
 787
 788			nlmsvc_insert_block_locked(block, 0);
 789			svc_wake_up(block->b_daemon);
 790			rc = 0;
 791			break;
 792		}
 793	}
 794	spin_unlock(&nlm_blocked_lock);
 795	if (rc == -ENOENT)
 796		printk(KERN_WARNING "lockd: grant for unknown block\n");
 797	return rc;
 798}
 799
 800/*
 801 * Unblock a blocked lock request. This is a callback invoked from the
 802 * VFS layer when a lock on which we blocked is removed.
 803 *
 804 * This function doesn't grant the blocked lock instantly, but rather moves
 805 * the block to the head of nlm_blocked where it can be picked up by lockd.
 806 */
 807static void
 808nlmsvc_notify_blocked(struct file_lock *fl)
 809{
 810	struct nlm_block	*block;
 811
 812	dprintk("lockd: VFS unblock notification for block %p\n", fl);
 813	spin_lock(&nlm_blocked_lock);
 814	list_for_each_entry(block, &nlm_blocked, b_list) {
 815		if (nlm_compare_locks(&block->b_call->a_args.lock.fl, fl)) {
 816			nlmsvc_insert_block_locked(block, 0);
 817			spin_unlock(&nlm_blocked_lock);
 818			svc_wake_up(block->b_daemon);
 819			return;
 820		}
 821	}
 822	spin_unlock(&nlm_blocked_lock);
 823	printk(KERN_WARNING "lockd: notification for unknown block!\n");
 824}
 825
 826static fl_owner_t nlmsvc_get_owner(fl_owner_t owner)
 827{
 828	return nlmsvc_get_lockowner(owner);
 829}
 830
 831static void nlmsvc_put_owner(fl_owner_t owner)
 832{
 833	nlmsvc_put_lockowner(owner);
 834}
 835
 836const struct lock_manager_operations nlmsvc_lock_operations = {
 837	.lm_notify = nlmsvc_notify_blocked,
 838	.lm_grant = nlmsvc_grant_deferred,
 839	.lm_get_owner = nlmsvc_get_owner,
 840	.lm_put_owner = nlmsvc_put_owner,
 841};
 842
 843/*
 844 * Try to claim a lock that was previously blocked.
 845 *
 846 * Note that we use both the RPC_GRANTED_MSG call _and_ an async
 847 * RPC thread when notifying the client. This seems like overkill...
 848 * Here's why:
 849 *  -	we don't want to use a synchronous RPC thread, otherwise
 850 *	we might find ourselves hanging on a dead portmapper.
 851 *  -	Some lockd implementations (e.g. HP) don't react to
 852 *	RPC_GRANTED calls; they seem to insist on RPC_GRANTED_MSG calls.
 853 */
 854static void
 855nlmsvc_grant_blocked(struct nlm_block *block)
 856{
 857	struct nlm_file		*file = block->b_file;
 858	struct nlm_lock		*lock = &block->b_call->a_args.lock;
 859	int			mode;
 860	int			error;
 861	loff_t			fl_start, fl_end;
 862
 863	dprintk("lockd: grant blocked lock %p\n", block);
 864
 865	kref_get(&block->b_count);
 866
 867	/* Unlink block request from list */
 868	nlmsvc_unlink_block(block);
 869
 870	/* If b_granted is true this means we've been here before.
 871	 * Just retry the grant callback, possibly refreshing the RPC
 872	 * binding */
 873	if (block->b_granted) {
 874		nlm_rebind_host(block->b_host);
 875		goto callback;
 876	}
 877
 878	/* Try the lock operation again */
 879	/* vfs_lock_file() can mangle fl_start and fl_end, but we need
 880	 * them unchanged for the GRANT_MSG
 881	 */
 882	lock->fl.c.flc_flags |= FL_SLEEP;
 883	fl_start = lock->fl.fl_start;
 884	fl_end = lock->fl.fl_end;
 885	mode = lock_to_openmode(&lock->fl);
 886	error = vfs_lock_file(file->f_file[mode], F_SETLK, &lock->fl, NULL);
 887	lock->fl.c.flc_flags &= ~FL_SLEEP;
 888	lock->fl.fl_start = fl_start;
 889	lock->fl.fl_end = fl_end;
 890
 891	switch (error) {
 892	case 0:
 893		break;
 894	case FILE_LOCK_DEFERRED:
 895		dprintk("lockd: lock still blocked error %d\n", error);
 896		nlmsvc_insert_block(block, NLM_NEVER);
 897		nlmsvc_release_block(block);
 898		return;
 899	default:
 900		printk(KERN_WARNING "lockd: unexpected error %d in %s!\n",
 901				-error, __func__);
 902		nlmsvc_insert_block(block, 10 * HZ);
 903		nlmsvc_release_block(block);
 904		return;
 905	}
 906
 907callback:
 908	/* Lock was granted by VFS. */
 909	dprintk("lockd: GRANTing blocked lock.\n");
 910	block->b_granted = 1;
 911
 912	/* keep block on the list, but don't reattempt until the RPC
 913	 * completes or the submission fails
 914	 */
 915	nlmsvc_insert_block(block, NLM_NEVER);
 916
 917	/* Call the client -- use a soft RPC task since nlmsvc_retry_blocked
 918	 * will queue up a new one if this one times out
 919	 */
 920	error = nlm_async_call(block->b_call, NLMPROC_GRANTED_MSG,
 921				&nlmsvc_grant_ops);
 922
 923	/* RPC submission failed, wait a bit and retry */
 924	if (error < 0)
 925		nlmsvc_insert_block(block, 10 * HZ);
 926}
 927
 928/*
 929 * This is the callback from the RPC layer when the NLM_GRANTED_MSG
 930 * RPC call has succeeded or timed out.
 931 * Like all RPC callbacks, it is invoked by the rpciod process, so it
 932 * better not sleep. Therefore, we put the blocked lock on the nlm_blocked
 933 * chain once more in order to have it removed by lockd itself (which can
 934 * then sleep on the file semaphore without disrupting e.g. the nfs client).
 935 */
 936static void nlmsvc_grant_callback(struct rpc_task *task, void *data)
 937{
 938	struct nlm_rqst		*call = data;
 939	struct nlm_block	*block = call->a_block;
 940	unsigned long		timeout;
 941
 942	dprintk("lockd: GRANT_MSG RPC callback\n");
 943
 944	spin_lock(&nlm_blocked_lock);
 945	/* if the block is not on a list at this point then it has
 946	 * been invalidated. Don't try to requeue it.
 947	 *
 948	 * FIXME: it's possible that the block is removed from the list
 949	 * after this check but before the nlmsvc_insert_block. In that
 950	 * case it will be added back. Perhaps we need better locking
 951	 * for nlm_blocked?
 952	 */
 953	if (list_empty(&block->b_list))
 954		goto out;
 955
 956	/* Technically, we should down the file semaphore here. Since we
 957	 * move the block towards the head of the queue only, no harm
 958	 * can be done, though. */
 959	if (task->tk_status < 0) {
 960		/* RPC error: Re-insert for retransmission */
 961		timeout = 10 * HZ;
 962	} else {
 963		/* Call was successful, now wait for client callback */
 964		timeout = 60 * HZ;
 965	}
 966	nlmsvc_insert_block_locked(block, timeout);
 967	svc_wake_up(block->b_daemon);
 968out:
 969	spin_unlock(&nlm_blocked_lock);
 970}
 971
 972/*
 973 * FIXME: nlmsvc_release_block() grabs a mutex.  This is not allowed for an
 974 * .rpc_release rpc_call_op
 975 */
 976static void nlmsvc_grant_release(void *data)
 977{
 978	struct nlm_rqst		*call = data;
 979	nlmsvc_release_block(call->a_block);
 980}
 981
 982static const struct rpc_call_ops nlmsvc_grant_ops = {
 983	.rpc_call_done = nlmsvc_grant_callback,
 984	.rpc_release = nlmsvc_grant_release,
 985};
 986
 987/*
 988 * We received a GRANT_RES callback. Try to find the corresponding
 989 * block.
 990 */
 991void
 992nlmsvc_grant_reply(struct nlm_cookie *cookie, __be32 status)
 993{
 994	struct nlm_block	*block;
 995	struct file_lock	*fl;
 996	int			error;
 997
 998	dprintk("grant_reply: looking for cookie %x, s=%d\n",
 999		*(unsigned int *)(cookie->data), status);
1000	if (!(block = nlmsvc_find_block(cookie)))
1001		return;
1002
1003	switch (status) {
1004	case nlm_lck_denied_grace_period:
1005		/* Try again in a couple of seconds */
1006		nlmsvc_insert_block(block, 10 * HZ);
1007		break;
1008	case nlm_lck_denied:
1009		/* Client doesn't want it, just unlock it */
1010		nlmsvc_unlink_block(block);
1011		fl = &block->b_call->a_args.lock.fl;
1012		fl->c.flc_type = F_UNLCK;
1013		error = vfs_lock_file(fl->c.flc_file, F_SETLK, fl, NULL);
1014		if (error)
1015			pr_warn("lockd: unable to unlock lock rejected by client!\n");
1016		break;
1017	default:
1018		/*
1019		 * Either it was accepted or the status makes no sense
1020		 * just unlink it either way.
1021		 */
1022		nlmsvc_unlink_block(block);
1023	}
1024	nlmsvc_release_block(block);
1025}
1026
1027/* Helper function to handle retry of a deferred block.
1028 * If it is a blocking lock, call grant_blocked.
1029 * For a non-blocking lock or test lock, revisit the request.
1030 */
1031static void
1032retry_deferred_block(struct nlm_block *block)
1033{
1034	if (!(block->b_flags & B_GOT_CALLBACK))
1035		block->b_flags |= B_TIMED_OUT;
1036	nlmsvc_insert_block(block, NLM_TIMEOUT);
1037	dprintk("revisit block %p flags %d\n",	block, block->b_flags);
1038	if (block->b_deferred_req) {
1039		block->b_deferred_req->revisit(block->b_deferred_req, 0);
1040		block->b_deferred_req = NULL;
1041	}
1042}
1043
1044/*
1045 * Retry all blocked locks that have been notified. This is where lockd
1046 * picks up locks that can be granted, or grant notifications that must
1047 * be retransmitted.
1048 */
1049void
1050nlmsvc_retry_blocked(struct svc_rqst *rqstp)
1051{
1052	unsigned long	timeout = MAX_SCHEDULE_TIMEOUT;
1053	struct nlm_block *block;
1054
1055	spin_lock(&nlm_blocked_lock);
1056	while (!list_empty(&nlm_blocked) && !svc_thread_should_stop(rqstp)) {
1057		block = list_entry(nlm_blocked.next, struct nlm_block, b_list);
1058
1059		if (block->b_when == NLM_NEVER)
1060			break;
1061		if (time_after(block->b_when, jiffies)) {
1062			timeout = block->b_when - jiffies;
1063			break;
1064		}
1065		spin_unlock(&nlm_blocked_lock);
1066
1067		dprintk("nlmsvc_retry_blocked(%p, when=%ld)\n",
1068			block, block->b_when);
1069		if (block->b_flags & B_QUEUED) {
1070			dprintk("nlmsvc_retry_blocked delete block (%p, granted=%d, flags=%d)\n",
1071				block, block->b_granted, block->b_flags);
1072			retry_deferred_block(block);
1073		} else
1074			nlmsvc_grant_blocked(block);
1075		spin_lock(&nlm_blocked_lock);
1076	}
1077	spin_unlock(&nlm_blocked_lock);
1078
1079	if (timeout < MAX_SCHEDULE_TIMEOUT)
1080		mod_timer(&nlmsvc_retry, jiffies + timeout);
1081}