dlmglue.c 90.0 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43
/* -*- mode: c; c-basic-offset: 8; -*-
 * vim: noexpandtab sw=8 ts=8 sts=0:
 *
 * dlmglue.c
 *
 * Code which implements an OCFS2 specific interface to our DLM.
 *
 * Copyright (C) 2003, 2004 Oracle.  All rights reserved.
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public
 * License as published by the Free Software Foundation; either
 * version 2 of the License, or (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * You should have received a copy of the GNU General Public
 * License along with this program; if not, write to the
 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
 * Boston, MA 021110-1307, USA.
 */

#include <linux/types.h>
#include <linux/slab.h>
#include <linux/highmem.h>
#include <linux/mm.h>
#include <linux/crc32.h>
#include <linux/kthread.h>
#include <linux/pagemap.h>
#include <linux/debugfs.h>
#include <linux/seq_file.h>

#include <cluster/heartbeat.h>
#include <cluster/nodemanager.h>
#include <cluster/tcp.h>

#define MLOG_MASK_PREFIX ML_DLM_GLUE
#include <cluster/masklog.h>

#include "ocfs2.h"
44
#include "ocfs2_lockingver.h"
45 46

#include "alloc.h"
47
#include "dcache.h"
48 49
#include "dlmglue.h"
#include "extent_map.h"
50
#include "file.h"
51 52 53
#include "heartbeat.h"
#include "inode.h"
#include "journal.h"
54
#include "stackglue.h"
55 56 57 58 59 60 61 62 63 64 65 66 67 68
#include "slot_map.h"
#include "super.h"
#include "uptodate.h"

#include "buffer_head_io.h"

struct ocfs2_mask_waiter {
	struct list_head	mw_item;
	int			mw_status;
	struct completion	mw_complete;
	unsigned long		mw_mask;
	unsigned long		mw_goal;
};

69 70
static struct ocfs2_super *ocfs2_get_dentry_osb(struct ocfs2_lock_res *lockres);
static struct ocfs2_super *ocfs2_get_inode_osb(struct ocfs2_lock_res *lockres);
71
static struct ocfs2_super *ocfs2_get_file_osb(struct ocfs2_lock_res *lockres);
72

73
/*
74
 * Return value from ->downconvert_worker functions.
75
 *
76
 * These control the precise actions of ocfs2_unblock_lock()
77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92
 * and ocfs2_process_blocked_lock()
 *
 */
enum ocfs2_unblock_action {
	UNBLOCK_CONTINUE	= 0, /* Continue downconvert */
	UNBLOCK_CONTINUE_POST	= 1, /* Continue downconvert, fire
				      * ->post_unlock callback */
	UNBLOCK_STOP_POST	= 2, /* Do not downconvert, fire
				      * ->post_unlock() callback. */
};

struct ocfs2_unblock_ctl {
	int requeue;
	enum ocfs2_unblock_action unblock_action;
};

93 94 95 96
static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres,
					int new_level);
static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres);

97 98 99 100 101
static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
				     int blocking);

static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres,
				       int blocking);
102 103 104

static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
				     struct ocfs2_lock_res *lockres);
105

106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134

#define mlog_meta_lvb(__level, __lockres) ocfs2_dump_meta_lvb_info(__level, __PRETTY_FUNCTION__, __LINE__, __lockres)

/* This aids in debugging situations where a bad LVB might be involved. */
static void ocfs2_dump_meta_lvb_info(u64 level,
				     const char *function,
				     unsigned int line,
				     struct ocfs2_lock_res *lockres)
{
	struct ocfs2_meta_lvb *lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb;

	mlog(level, "LVB information for %s (called from %s:%u):\n",
	     lockres->l_name, function, line);
	mlog(level, "version: %u, clusters: %u, generation: 0x%x\n",
	     lvb->lvb_version, be32_to_cpu(lvb->lvb_iclusters),
	     be32_to_cpu(lvb->lvb_igeneration));
	mlog(level, "size: %llu, uid %u, gid %u, mode 0x%x\n",
	     (unsigned long long)be64_to_cpu(lvb->lvb_isize),
	     be32_to_cpu(lvb->lvb_iuid), be32_to_cpu(lvb->lvb_igid),
	     be16_to_cpu(lvb->lvb_imode));
	mlog(level, "nlink %u, atime_packed 0x%llx, ctime_packed 0x%llx, "
	     "mtime_packed 0x%llx iattr 0x%x\n", be16_to_cpu(lvb->lvb_inlink),
	     (long long)be64_to_cpu(lvb->lvb_iatime_packed),
	     (long long)be64_to_cpu(lvb->lvb_ictime_packed),
	     (long long)be64_to_cpu(lvb->lvb_imtime_packed),
	     be32_to_cpu(lvb->lvb_iattr));
}


135 136 137 138
/*
 * OCFS2 Lock Resource Operations
 *
 * These fine tune the behavior of the generic dlmglue locking infrastructure.
139 140 141 142 143 144 145 146
 *
 * The most basic of lock types can point ->l_priv to their respective
 * struct ocfs2_super and allow the default actions to manage things.
 *
 * Right now, each lock type also needs to implement an init function,
 * and trivial lock/unlock wrappers. ocfs2_simple_drop_lockres()
 * should be called when the lock is no longer needed (i.e., object
 * destruction time).
147
 */
148
struct ocfs2_lock_res_ops {
149 150 151 152 153
	/*
	 * Translate an ocfs2_lock_res * into an ocfs2_super *. Define
	 * this callback if ->l_priv is not an ocfs2_super pointer
	 */
	struct ocfs2_super * (*get_osb)(struct ocfs2_lock_res *);
154

155
	/*
156 157 158 159
	 * Optionally called in the downconvert thread after a
	 * successful downconvert. The lockres will not be referenced
	 * after this callback is called, so it is safe to free
	 * memory, etc.
160 161 162 163
	 *
	 * The exact semantics of when this is called are controlled
	 * by ->downconvert_worker()
	 */
164
	void (*post_unlock)(struct ocfs2_super *, struct ocfs2_lock_res *);
165

166 167 168 169 170 171 172 173 174 175 176 177
	/*
	 * Allow a lock type to add checks to determine whether it is
	 * safe to downconvert a lock. Return 0 to re-queue the
	 * downconvert at a later time, nonzero to continue.
	 *
	 * For most locks, the default checks that there are no
	 * incompatible holders are sufficient.
	 *
	 * Called with the lockres spinlock held.
	 */
	int (*check_downconvert)(struct ocfs2_lock_res *, int);

178 179 180 181 182 183 184 185 186 187 188
	/*
	 * Allows a lock type to populate the lock value block. This
	 * is called on downconvert, and when we drop a lock.
	 *
	 * Locks that want to use this should set LOCK_TYPE_USES_LVB
	 * in the flags field.
	 *
	 * Called with the lockres spinlock held.
	 */
	void (*set_lvb)(struct ocfs2_lock_res *);

189 190 191 192 193 194 195 196 197 198 199
	/*
	 * Called from the downconvert thread when it is determined
	 * that a lock will be downconverted. This is called without
	 * any locks held so the function can do work that might
	 * schedule (syncing out data, etc).
	 *
	 * This should return any one of the ocfs2_unblock_action
	 * values, depending on what it wants the thread to do.
	 */
	int (*downconvert_worker)(struct ocfs2_lock_res *, int);

200 201 202 203 204
	/*
	 * LOCK_TYPE_* flags which describe the specific requirements
	 * of a lock type. Descriptions of each individual flag follow.
	 */
	int flags;
205 206
};

207 208 209 210 211 212 213 214 215 216
/*
 * Some locks want to "refresh" potentially stale data when a
 * meaningful (PRMODE or EXMODE) lock level is first obtained. If this
 * flag is set, the OCFS2_LOCK_NEEDS_REFRESH flag will be set on the
 * individual lockres l_flags member from the ast function. It is
 * expected that the locking wrapper will clear the
 * OCFS2_LOCK_NEEDS_REFRESH flag when done.
 */
#define LOCK_TYPE_REQUIRES_REFRESH 0x1

217
/*
218 219
 * Indicate that a lock type makes use of the lock value block. The
 * ->set_lvb lock type callback must be defined.
220 221 222
 */
#define LOCK_TYPE_USES_LVB		0x2

223
static struct ocfs2_lock_res_ops ocfs2_inode_rw_lops = {
224
	.get_osb	= ocfs2_get_inode_osb,
225
	.flags		= 0,
226 227
};

228
static struct ocfs2_lock_res_ops ocfs2_inode_inode_lops = {
229
	.get_osb	= ocfs2_get_inode_osb,
230 231
	.check_downconvert = ocfs2_check_meta_downconvert,
	.set_lvb	= ocfs2_set_meta_lvb,
232
	.downconvert_worker = ocfs2_data_convert_worker,
233
	.flags		= LOCK_TYPE_REQUIRES_REFRESH|LOCK_TYPE_USES_LVB,
234 235 236
};

static struct ocfs2_lock_res_ops ocfs2_super_lops = {
237
	.flags		= LOCK_TYPE_REQUIRES_REFRESH,
238 239 240
};

static struct ocfs2_lock_res_ops ocfs2_rename_lops = {
241
	.flags		= 0,
242 243
};

244
static struct ocfs2_lock_res_ops ocfs2_dentry_lops = {
245
	.get_osb	= ocfs2_get_dentry_osb,
246
	.post_unlock	= ocfs2_dentry_post_unlock,
247
	.downconvert_worker = ocfs2_dentry_convert_worker,
248
	.flags		= 0,
249 250
};

251 252 253 254 255
static struct ocfs2_lock_res_ops ocfs2_inode_open_lops = {
	.get_osb	= ocfs2_get_inode_osb,
	.flags		= 0,
};

256 257 258 259 260
static struct ocfs2_lock_res_ops ocfs2_flock_lops = {
	.get_osb	= ocfs2_get_file_osb,
	.flags		= 0,
};

261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285
/*
 * This is the filesystem locking protocol version.
 *
 * Whenever the filesystem does new things with locks (adds or removes a
 * lock, orders them differently, does different things underneath a lock),
 * the version must be changed.  The protocol is negotiated when joining
 * the dlm domain.  A node may join the domain if its major version is
 * identical to all other nodes and its minor version is greater than
 * or equal to all other nodes.  When its minor version is greater than
 * the other nodes, it will run at the minor version specified by the
 * other nodes.
 *
 * If a locking change is made that will not be compatible with older
 * versions, the major number must be increased and the minor version set
 * to zero.  If a change merely adds a behavior that can be disabled when
 * speaking to older versions, the minor version must be increased.  If a
 * change adds a fully backwards compatible change (eg, LVB changes that
 * are just ignored by older versions), the version does not need to be
 * updated.
 */
const struct dlm_protocol_version ocfs2_locking_protocol = {
	.pv_major = OCFS2_LOCKING_PROTOCOL_MAJOR,
	.pv_minor = OCFS2_LOCKING_PROTOCOL_MINOR,
};

286 287 288
static inline int ocfs2_is_inode_lock(struct ocfs2_lock_res *lockres)
{
	return lockres->l_type == OCFS2_LOCK_TYPE_META ||
289 290
		lockres->l_type == OCFS2_LOCK_TYPE_RW ||
		lockres->l_type == OCFS2_LOCK_TYPE_OPEN;
291 292 293 294 295 296 297 298 299
}

static inline struct inode *ocfs2_lock_res_inode(struct ocfs2_lock_res *lockres)
{
	BUG_ON(!ocfs2_is_inode_lock(lockres));

	return (struct inode *) lockres->l_priv;
}

300 301 302 303 304 305 306
static inline struct ocfs2_dentry_lock *ocfs2_lock_res_dl(struct ocfs2_lock_res *lockres)
{
	BUG_ON(lockres->l_type != OCFS2_LOCK_TYPE_DENTRY);

	return (struct ocfs2_dentry_lock *)lockres->l_priv;
}

307 308 309 310 311 312 313 314
static inline struct ocfs2_super *ocfs2_get_lockres_osb(struct ocfs2_lock_res *lockres)
{
	if (lockres->l_ops->get_osb)
		return lockres->l_ops->get_osb(lockres);

	return (struct ocfs2_super *)lockres->l_priv;
}

315 316 317
static int ocfs2_lock_create(struct ocfs2_super *osb,
			     struct ocfs2_lock_res *lockres,
			     int level,
318
			     u32 dlm_flags);
319 320 321 322 323 324 325 326 327 328 329 330 331
static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres,
						     int wanted);
static void ocfs2_cluster_unlock(struct ocfs2_super *osb,
				 struct ocfs2_lock_res *lockres,
				 int level);
static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres);
static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres);
static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres);
static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres, int level);
static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
					struct ocfs2_lock_res *lockres);
static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres,
						int convert);
332 333 334
#define ocfs2_log_dlm_error(_func, _err, _lockres) do {			\
	mlog(ML_ERROR, "DLM error %d while calling %s on resource %s\n", \
	     _err, _func, _lockres->l_name);				\
335
} while (0)
336 337 338
static int ocfs2_downconvert_thread(void *arg);
static void ocfs2_downconvert_on_unlock(struct ocfs2_super *osb,
					struct ocfs2_lock_res *lockres);
339
static int ocfs2_inode_lock_update(struct inode *inode,
340 341 342
				  struct buffer_head **bh);
static void ocfs2_drop_osb_locks(struct ocfs2_super *osb);
static inline int ocfs2_highest_compat_lock_level(int level);
343 344 345 346 347 348 349 350 351 352 353
static void ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres,
				      int new_level);
static int ocfs2_downconvert_lock(struct ocfs2_super *osb,
				  struct ocfs2_lock_res *lockres,
				  int new_level,
				  int lvb);
static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb,
				        struct ocfs2_lock_res *lockres);
static int ocfs2_cancel_convert(struct ocfs2_super *osb,
				struct ocfs2_lock_res *lockres);

354 355 356 357 358 359 360 361 362 363 364 365

static void ocfs2_build_lock_name(enum ocfs2_lock_type type,
				  u64 blkno,
				  u32 generation,
				  char *name)
{
	int len;

	mlog_entry_void();

	BUG_ON(type >= OCFS2_NUM_LOCK_TYPES);

366 367 368
	len = snprintf(name, OCFS2_LOCK_ID_MAX_LEN, "%c%s%016llx%08x",
		       ocfs2_lock_type_char(type), OCFS2_LOCK_ID_PAD,
		       (long long)blkno, generation);
369 370 371 372 373 374 375 376

	BUG_ON(len != (OCFS2_LOCK_ID_MAX_LEN - 1));

	mlog(0, "built lock resource with name: %s\n", name);

	mlog_exit_void();
}

377
static DEFINE_SPINLOCK(ocfs2_dlm_tracking_lock);
378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406

static void ocfs2_add_lockres_tracking(struct ocfs2_lock_res *res,
				       struct ocfs2_dlm_debug *dlm_debug)
{
	mlog(0, "Add tracking for lockres %s\n", res->l_name);

	spin_lock(&ocfs2_dlm_tracking_lock);
	list_add(&res->l_debug_list, &dlm_debug->d_lockres_tracking);
	spin_unlock(&ocfs2_dlm_tracking_lock);
}

static void ocfs2_remove_lockres_tracking(struct ocfs2_lock_res *res)
{
	spin_lock(&ocfs2_dlm_tracking_lock);
	if (!list_empty(&res->l_debug_list))
		list_del_init(&res->l_debug_list);
	spin_unlock(&ocfs2_dlm_tracking_lock);
}

static void ocfs2_lock_res_init_common(struct ocfs2_super *osb,
				       struct ocfs2_lock_res *res,
				       enum ocfs2_lock_type type,
				       struct ocfs2_lock_res_ops *ops,
				       void *priv)
{
	res->l_type          = type;
	res->l_ops           = ops;
	res->l_priv          = priv;

407 408 409
	res->l_level         = DLM_LOCK_IV;
	res->l_requested     = DLM_LOCK_IV;
	res->l_blocking      = DLM_LOCK_IV;
410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429
	res->l_action        = OCFS2_AST_INVALID;
	res->l_unlock_action = OCFS2_UNLOCK_INVALID;

	res->l_flags         = OCFS2_LOCK_INITIALIZED;

	ocfs2_add_lockres_tracking(res, osb->osb_dlm_debug);
}

void ocfs2_lock_res_init_once(struct ocfs2_lock_res *res)
{
	/* This also clears out the lock status block */
	memset(res, 0, sizeof(struct ocfs2_lock_res));
	spin_lock_init(&res->l_lock);
	init_waitqueue_head(&res->l_event);
	INIT_LIST_HEAD(&res->l_blocked_list);
	INIT_LIST_HEAD(&res->l_mask_waiters);
}

void ocfs2_inode_lock_res_init(struct ocfs2_lock_res *res,
			       enum ocfs2_lock_type type,
430
			       unsigned int generation,
431 432 433 434 435 436 437 438 439
			       struct inode *inode)
{
	struct ocfs2_lock_res_ops *ops;

	switch(type) {
		case OCFS2_LOCK_TYPE_RW:
			ops = &ocfs2_inode_rw_lops;
			break;
		case OCFS2_LOCK_TYPE_META:
440
			ops = &ocfs2_inode_inode_lops;
441
			break;
442 443 444
		case OCFS2_LOCK_TYPE_OPEN:
			ops = &ocfs2_inode_open_lops;
			break;
445 446 447 448 449 450
		default:
			mlog_bug_on_msg(1, "type: %d\n", type);
			ops = NULL; /* thanks, gcc */
			break;
	};

451
	ocfs2_build_lock_name(type, OCFS2_I(inode)->ip_blkno,
452
			      generation, res->l_name);
453 454 455
	ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), res, type, ops, inode);
}

456 457 458 459 460 461 462
static struct ocfs2_super *ocfs2_get_inode_osb(struct ocfs2_lock_res *lockres)
{
	struct inode *inode = ocfs2_lock_res_inode(lockres);

	return OCFS2_SB(inode->i_sb);
}

463 464 465 466 467 468 469
static struct ocfs2_super *ocfs2_get_file_osb(struct ocfs2_lock_res *lockres)
{
	struct ocfs2_file_private *fp = lockres->l_priv;

	return OCFS2_SB(fp->fp_file->f_mapping->host->i_sb);
}

470 471 472 473 474 475 476 477 478 479
static __u64 ocfs2_get_dentry_lock_ino(struct ocfs2_lock_res *lockres)
{
	__be64 inode_blkno_be;

	memcpy(&inode_blkno_be, &lockres->l_name[OCFS2_DENTRY_LOCK_INO_START],
	       sizeof(__be64));

	return be64_to_cpu(inode_blkno_be);
}

480 481 482 483 484 485 486
static struct ocfs2_super *ocfs2_get_dentry_osb(struct ocfs2_lock_res *lockres)
{
	struct ocfs2_dentry_lock *dl = lockres->l_priv;

	return OCFS2_SB(dl->dl_inode->i_sb);
}

487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524
void ocfs2_dentry_lock_res_init(struct ocfs2_dentry_lock *dl,
				u64 parent, struct inode *inode)
{
	int len;
	u64 inode_blkno = OCFS2_I(inode)->ip_blkno;
	__be64 inode_blkno_be = cpu_to_be64(inode_blkno);
	struct ocfs2_lock_res *lockres = &dl->dl_lockres;

	ocfs2_lock_res_init_once(lockres);

	/*
	 * Unfortunately, the standard lock naming scheme won't work
	 * here because we have two 16 byte values to use. Instead,
	 * we'll stuff the inode number as a binary value. We still
	 * want error prints to show something without garbling the
	 * display, so drop a null byte in there before the inode
	 * number. A future version of OCFS2 will likely use all
	 * binary lock names. The stringified names have been a
	 * tremendous aid in debugging, but now that the debugfs
	 * interface exists, we can mangle things there if need be.
	 *
	 * NOTE: We also drop the standard "pad" value (the total lock
	 * name size stays the same though - the last part is all
	 * zeros due to the memset in ocfs2_lock_res_init_once()
	 */
	len = snprintf(lockres->l_name, OCFS2_DENTRY_LOCK_INO_START,
		       "%c%016llx",
		       ocfs2_lock_type_char(OCFS2_LOCK_TYPE_DENTRY),
		       (long long)parent);

	BUG_ON(len != (OCFS2_DENTRY_LOCK_INO_START - 1));

	memcpy(&lockres->l_name[OCFS2_DENTRY_LOCK_INO_START], &inode_blkno_be,
	       sizeof(__be64));

	ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres,
				   OCFS2_LOCK_TYPE_DENTRY, &ocfs2_dentry_lops,
				   dl);
525 526 527 528 529 530 531 532
}

static void ocfs2_super_lock_res_init(struct ocfs2_lock_res *res,
				      struct ocfs2_super *osb)
{
	/* Superblock lockres doesn't come from a slab so we call init
	 * once on it manually.  */
	ocfs2_lock_res_init_once(res);
533 534
	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_SUPER, OCFS2_SUPER_BLOCK_BLKNO,
			      0, res->l_name);
535 536 537 538 539 540 541 542 543 544
	ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_SUPER,
				   &ocfs2_super_lops, osb);
}

static void ocfs2_rename_lock_res_init(struct ocfs2_lock_res *res,
				       struct ocfs2_super *osb)
{
	/* Rename lockres doesn't come from a slab so we call init
	 * once on it manually.  */
	ocfs2_lock_res_init_once(res);
545 546
	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_RENAME, 0, 0, res->l_name);
	ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_RENAME,
547 548 549
				   &ocfs2_rename_lops, osb);
}

550 551 552 553 554 555 556 557 558 559 560 561 562 563 564
void ocfs2_file_lock_res_init(struct ocfs2_lock_res *lockres,
			      struct ocfs2_file_private *fp)
{
	struct inode *inode = fp->fp_file->f_mapping->host;
	struct ocfs2_inode_info *oi = OCFS2_I(inode);

	ocfs2_lock_res_init_once(lockres);
	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_FLOCK, oi->ip_blkno,
			      inode->i_generation, lockres->l_name);
	ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres,
				   OCFS2_LOCK_TYPE_FLOCK, &ocfs2_flock_lops,
				   fp);
	lockres->l_flags |= OCFS2_LOCK_NOCACHE;
}

565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604
void ocfs2_lock_res_free(struct ocfs2_lock_res *res)
{
	mlog_entry_void();

	if (!(res->l_flags & OCFS2_LOCK_INITIALIZED))
		return;

	ocfs2_remove_lockres_tracking(res);

	mlog_bug_on_msg(!list_empty(&res->l_blocked_list),
			"Lockres %s is on the blocked list\n",
			res->l_name);
	mlog_bug_on_msg(!list_empty(&res->l_mask_waiters),
			"Lockres %s has mask waiters pending\n",
			res->l_name);
	mlog_bug_on_msg(spin_is_locked(&res->l_lock),
			"Lockres %s is locked\n",
			res->l_name);
	mlog_bug_on_msg(res->l_ro_holders,
			"Lockres %s has %u ro holders\n",
			res->l_name, res->l_ro_holders);
	mlog_bug_on_msg(res->l_ex_holders,
			"Lockres %s has %u ex holders\n",
			res->l_name, res->l_ex_holders);

	/* Need to clear out the lock status block for the dlm */
	memset(&res->l_lksb, 0, sizeof(res->l_lksb));

	res->l_flags = 0UL;
	mlog_exit_void();
}

static inline void ocfs2_inc_holders(struct ocfs2_lock_res *lockres,
				     int level)
{
	mlog_entry_void();

	BUG_ON(!lockres);

	switch(level) {
605
	case DLM_LOCK_EX:
606 607
		lockres->l_ex_holders++;
		break;
608
	case DLM_LOCK_PR:
609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625
		lockres->l_ro_holders++;
		break;
	default:
		BUG();
	}

	mlog_exit_void();
}

static inline void ocfs2_dec_holders(struct ocfs2_lock_res *lockres,
				     int level)
{
	mlog_entry_void();

	BUG_ON(!lockres);

	switch(level) {
626
	case DLM_LOCK_EX:
627 628 629
		BUG_ON(!lockres->l_ex_holders);
		lockres->l_ex_holders--;
		break;
630
	case DLM_LOCK_PR:
631 632 633 634 635 636 637 638 639 640 641 642 643 644
		BUG_ON(!lockres->l_ro_holders);
		lockres->l_ro_holders--;
		break;
	default:
		BUG();
	}
	mlog_exit_void();
}

/* WARNING: This function lives in a world where the only three lock
 * levels are EX, PR, and NL. It *will* have to be adjusted when more
 * lock types are added. */
static inline int ocfs2_highest_compat_lock_level(int level)
{
645
	int new_level = DLM_LOCK_EX;
646

647 648 649 650
	if (level == DLM_LOCK_EX)
		new_level = DLM_LOCK_NL;
	else if (level == DLM_LOCK_PR)
		new_level = DLM_LOCK_PR;
651 652 653 654 655 656
	return new_level;
}

static void lockres_set_flags(struct ocfs2_lock_res *lockres,
			      unsigned long newflags)
{
657
	struct ocfs2_mask_waiter *mw, *tmp;
658 659 660 661 662

 	assert_spin_locked(&lockres->l_lock);

	lockres->l_flags = newflags;

663
	list_for_each_entry_safe(mw, tmp, &lockres->l_mask_waiters, mw_item) {
664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688
		if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
			continue;

		list_del_init(&mw->mw_item);
		mw->mw_status = 0;
		complete(&mw->mw_complete);
	}
}
static void lockres_or_flags(struct ocfs2_lock_res *lockres, unsigned long or)
{
	lockres_set_flags(lockres, lockres->l_flags | or);
}
static void lockres_clear_flags(struct ocfs2_lock_res *lockres,
				unsigned long clear)
{
	lockres_set_flags(lockres, lockres->l_flags & ~clear);
}

static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres)
{
	mlog_entry_void();

	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED));
	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
689
	BUG_ON(lockres->l_blocking <= DLM_LOCK_NL);
690 691 692 693

	lockres->l_level = lockres->l_requested;
	if (lockres->l_level <=
	    ocfs2_highest_compat_lock_level(lockres->l_blocking)) {
694
		lockres->l_blocking = DLM_LOCK_NL;
695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712
		lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED);
	}
	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);

	mlog_exit_void();
}

static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres)
{
	mlog_entry_void();

	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED));

	/* Convert from RO to EX doesn't really need anything as our
	 * information is already up to data. Convert from NL to
	 * *anything* however should mark ourselves as needing an
	 * update */
713
	if (lockres->l_level == DLM_LOCK_NL &&
714
	    lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
715 716 717 718 719 720 721 722 723 724 725 726
		lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);

	lockres->l_level = lockres->l_requested;
	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);

	mlog_exit_void();
}

static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres)
{
	mlog_entry_void();

727
	BUG_ON((!(lockres->l_flags & OCFS2_LOCK_BUSY)));
728 729
	BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);

730
	if (lockres->l_requested > DLM_LOCK_NL &&
731 732
	    !(lockres->l_flags & OCFS2_LOCK_LOCAL) &&
	    lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767
		lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);

	lockres->l_level = lockres->l_requested;
	lockres_or_flags(lockres, OCFS2_LOCK_ATTACHED);
	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);

	mlog_exit_void();
}

static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres,
				     int level)
{
	int needs_downconvert = 0;
	mlog_entry_void();

	assert_spin_locked(&lockres->l_lock);

	lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED);

	if (level > lockres->l_blocking) {
		/* only schedule a downconvert if we haven't already scheduled
		 * one that goes low enough to satisfy the level we're
		 * blocking.  this also catches the case where we get
		 * duplicate BASTs */
		if (ocfs2_highest_compat_lock_level(level) <
		    ocfs2_highest_compat_lock_level(lockres->l_blocking))
			needs_downconvert = 1;

		lockres->l_blocking = level;
	}

	mlog_exit(needs_downconvert);
	return needs_downconvert;
}

768
static void ocfs2_blocking_ast(void *opaque, int level)
769
{
770 771
	struct ocfs2_lock_res *lockres = opaque;
	struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
772 773 774
	int needs_downconvert;
	unsigned long flags;

775
	BUG_ON(level <= DLM_LOCK_NL);
776

777 778 779 780
	mlog(0, "BAST fired for lockres %s, blocking %d, level %d type %s\n",
	     lockres->l_name, level, lockres->l_level,
	     ocfs2_lock_type_string(lockres->l_type));

781 782 783 784 785 786 787
	/*
	 * We can skip the bast for locks which don't enable caching -
	 * they'll be dropped at the earliest possible time anyway.
	 */
	if (lockres->l_flags & OCFS2_LOCK_NOCACHE)
		return;

788 789 790 791 792 793
	spin_lock_irqsave(&lockres->l_lock, flags);
	needs_downconvert = ocfs2_generic_handle_bast(lockres, level);
	if (needs_downconvert)
		ocfs2_schedule_blocked_lock(osb, lockres);
	spin_unlock_irqrestore(&lockres->l_lock, flags);

794 795
	wake_up(&lockres->l_event);

796
	ocfs2_wake_downconvert_thread(osb);
797 798
}

799
static void ocfs2_locking_ast(void *opaque)
800
{
801
	struct ocfs2_lock_res *lockres = opaque;
802 803 804 805 806 807 808 809 810 811 812 813 814 815 816
	struct dlm_lockstatus *lksb = &lockres->l_lksb;
	unsigned long flags;

	spin_lock_irqsave(&lockres->l_lock, flags);

	if (lksb->status != DLM_NORMAL) {
		mlog(ML_ERROR, "lockres %s: lksb status value of %u!\n",
		     lockres->l_name, lksb->status);
		spin_unlock_irqrestore(&lockres->l_lock, flags);
		return;
	}

	switch(lockres->l_action) {
	case OCFS2_AST_ATTACH:
		ocfs2_generic_handle_attach_action(lockres);
817
		lockres_clear_flags(lockres, OCFS2_LOCK_LOCAL);
818 819 820 821 822 823 824 825
		break;
	case OCFS2_AST_CONVERT:
		ocfs2_generic_handle_convert_action(lockres);
		break;
	case OCFS2_AST_DOWNCONVERT:
		ocfs2_generic_handle_downconvert_action(lockres);
		break;
	default:
826 827 828 829
		mlog(ML_ERROR, "lockres %s: ast fired with invalid action: %u "
		     "lockres flags = 0x%lx, unlock action: %u\n",
		     lockres->l_name, lockres->l_action, lockres->l_flags,
		     lockres->l_unlock_action);
830 831 832 833 834 835 836 837
		BUG();
	}

	/* set it to something invalid so if we get called again we
	 * can catch it. */
	lockres->l_action = OCFS2_AST_INVALID;

	wake_up(&lockres->l_event);
838
	spin_unlock_irqrestore(&lockres->l_lock, flags);
839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865
}

static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres,
						int convert)
{
	unsigned long flags;

	mlog_entry_void();
	spin_lock_irqsave(&lockres->l_lock, flags);
	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
	if (convert)
		lockres->l_action = OCFS2_AST_INVALID;
	else
		lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
	spin_unlock_irqrestore(&lockres->l_lock, flags);

	wake_up(&lockres->l_event);
	mlog_exit_void();
}

/* Note: If we detect another process working on the lock (i.e.,
 * OCFS2_LOCK_BUSY), we'll bail out returning 0. It's up to the caller
 * to do the right thing in that case.
 */
static int ocfs2_lock_create(struct ocfs2_super *osb,
			     struct ocfs2_lock_res *lockres,
			     int level,
866
			     u32 dlm_flags)
867 868 869 870 871 872
{
	int ret = 0;
	unsigned long flags;

	mlog_entry_void();

873
	mlog(0, "lock %s, level = %d, flags = %u\n", lockres->l_name, level,
874 875 876 877 878 879 880 881 882 883 884 885 886 887
	     dlm_flags);

	spin_lock_irqsave(&lockres->l_lock, flags);
	if ((lockres->l_flags & OCFS2_LOCK_ATTACHED) ||
	    (lockres->l_flags & OCFS2_LOCK_BUSY)) {
		spin_unlock_irqrestore(&lockres->l_lock, flags);
		goto bail;
	}

	lockres->l_action = OCFS2_AST_ATTACH;
	lockres->l_requested = level;
	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
	spin_unlock_irqrestore(&lockres->l_lock, flags);

888 889 890 891 892 893 894 895 896
	ret = ocfs2_dlm_lock(osb->dlm,
			     level,
			     &lockres->l_lksb,
			     dlm_flags,
			     lockres->l_name,
			     OCFS2_LOCK_ID_MAX_LEN - 1,
			     lockres);
	if (ret) {
		ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
897 898 899
		ocfs2_recover_from_dlm_error(lockres, 1);
	}

900
	mlog(0, "lock %s, return from ocfs2_dlm_lock\n", lockres->l_name);
901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994

bail:
	mlog_exit(ret);
	return ret;
}

static inline int ocfs2_check_wait_flag(struct ocfs2_lock_res *lockres,
					int flag)
{
	unsigned long flags;
	int ret;

	spin_lock_irqsave(&lockres->l_lock, flags);
	ret = lockres->l_flags & flag;
	spin_unlock_irqrestore(&lockres->l_lock, flags);

	return ret;
}

static inline void ocfs2_wait_on_busy_lock(struct ocfs2_lock_res *lockres)

{
	wait_event(lockres->l_event,
		   !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_BUSY));
}

static inline void ocfs2_wait_on_refreshing_lock(struct ocfs2_lock_res *lockres)

{
	wait_event(lockres->l_event,
		   !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_REFRESHING));
}

/* predict what lock level we'll be dropping down to on behalf
 * of another node, and return true if the currently wanted
 * level will be compatible with it. */
static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres,
						     int wanted)
{
	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));

	return wanted <= ocfs2_highest_compat_lock_level(lockres->l_blocking);
}

static void ocfs2_init_mask_waiter(struct ocfs2_mask_waiter *mw)
{
	INIT_LIST_HEAD(&mw->mw_item);
	init_completion(&mw->mw_complete);
}

static int ocfs2_wait_for_mask(struct ocfs2_mask_waiter *mw)
{
	wait_for_completion(&mw->mw_complete);
	/* Re-arm the completion in case we want to wait on it again */
	INIT_COMPLETION(mw->mw_complete);
	return mw->mw_status;
}

static void lockres_add_mask_waiter(struct ocfs2_lock_res *lockres,
				    struct ocfs2_mask_waiter *mw,
				    unsigned long mask,
				    unsigned long goal)
{
	BUG_ON(!list_empty(&mw->mw_item));

	assert_spin_locked(&lockres->l_lock);

	list_add_tail(&mw->mw_item, &lockres->l_mask_waiters);
	mw->mw_mask = mask;
	mw->mw_goal = goal;
}

/* returns 0 if the mw that was removed was already satisfied, -EBUSY
 * if the mask still hadn't reached its goal */
static int lockres_remove_mask_waiter(struct ocfs2_lock_res *lockres,
				      struct ocfs2_mask_waiter *mw)
{
	unsigned long flags;
	int ret = 0;

	spin_lock_irqsave(&lockres->l_lock, flags);
	if (!list_empty(&mw->mw_item)) {
		if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
			ret = -EBUSY;

		list_del_init(&mw->mw_item);
		init_completion(&mw->mw_complete);
	}
	spin_unlock_irqrestore(&lockres->l_lock, flags);

	return ret;

}

995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009
static int ocfs2_wait_for_mask_interruptible(struct ocfs2_mask_waiter *mw,
					     struct ocfs2_lock_res *lockres)
{
	int ret;

	ret = wait_for_completion_interruptible(&mw->mw_complete);
	if (ret)
		lockres_remove_mask_waiter(lockres, mw);
	else
		ret = mw->mw_status;
	/* Re-arm the completion in case we want to wait on it again */
	INIT_COMPLETION(mw->mw_complete);
	return ret;
}

1010 1011 1012
static int ocfs2_cluster_lock(struct ocfs2_super *osb,
			      struct ocfs2_lock_res *lockres,
			      int level,
1013
			      u32 lkm_flags,
1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024
			      int arg_flags)
{
	struct ocfs2_mask_waiter mw;
	int wait, catch_signals = !(osb->s_mount_opt & OCFS2_MOUNT_NOINTR);
	int ret = 0; /* gcc doesn't realize wait = 1 guarantees ret is set */
	unsigned long flags;

	mlog_entry_void();

	ocfs2_init_mask_waiter(&mw);

1025
	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
1026
		lkm_flags |= DLM_LKF_VALBLK;
1027

1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067
again:
	wait = 0;

	if (catch_signals && signal_pending(current)) {
		ret = -ERESTARTSYS;
		goto out;
	}

	spin_lock_irqsave(&lockres->l_lock, flags);

	mlog_bug_on_msg(lockres->l_flags & OCFS2_LOCK_FREEING,
			"Cluster lock called on freeing lockres %s! flags "
			"0x%lx\n", lockres->l_name, lockres->l_flags);

	/* We only compare against the currently granted level
	 * here. If the lock is blocked waiting on a downconvert,
	 * we'll get caught below. */
	if (lockres->l_flags & OCFS2_LOCK_BUSY &&
	    level > lockres->l_level) {
		/* is someone sitting in dlm_lock? If so, wait on
		 * them. */
		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
		wait = 1;
		goto unlock;
	}

	if (lockres->l_flags & OCFS2_LOCK_BLOCKED &&
	    !ocfs2_may_continue_on_blocked_lock(lockres, level)) {
		/* is the lock is currently blocked on behalf of
		 * another node */
		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BLOCKED, 0);
		wait = 1;
		goto unlock;
	}

	if (level > lockres->l_level) {
		if (lockres->l_action != OCFS2_AST_INVALID)
			mlog(ML_ERROR, "lockres %s has action %u pending\n",
			     lockres->l_name, lockres->l_action);

1068 1069
		if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
			lockres->l_action = OCFS2_AST_ATTACH;
1070
			lkm_flags &= ~DLM_LKF_CONVERT;
1071 1072
		} else {
			lockres->l_action = OCFS2_AST_CONVERT;
1073
			lkm_flags |= DLM_LKF_CONVERT;
1074 1075
		}

1076 1077 1078 1079
		lockres->l_requested = level;
		lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
		spin_unlock_irqrestore(&lockres->l_lock, flags);

1080 1081
		BUG_ON(level == DLM_LOCK_IV);
		BUG_ON(level == DLM_LOCK_NL);
1082 1083 1084 1085 1086

		mlog(0, "lock %s, convert from %d to level = %d\n",
		     lockres->l_name, lockres->l_level, level);

		/* call dlm_lock to upgrade lock now */
1087 1088 1089 1090 1091 1092 1093 1094 1095 1096
		ret = ocfs2_dlm_lock(osb->dlm,
				     level,
				     &lockres->l_lksb,
				     lkm_flags,
				     lockres->l_name,
				     OCFS2_LOCK_ID_MAX_LEN - 1,
				     lockres);
		if (ret) {
			if (!(lkm_flags & DLM_LKF_NOQUEUE) ||
			    (ret != -EAGAIN)) {
1097
				ocfs2_log_dlm_error("ocfs2_dlm_lock",
1098
						    ret, lockres);
1099 1100 1101 1102 1103
			}
			ocfs2_recover_from_dlm_error(lockres, 1);
			goto out;
		}

1104
		mlog(0, "lock %s, successfull return from ocfs2_dlm_lock\n",
1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157
		     lockres->l_name);

		/* At this point we've gone inside the dlm and need to
		 * complete our work regardless. */
		catch_signals = 0;

		/* wait for busy to clear and carry on */
		goto again;
	}

	/* Ok, if we get here then we're good to go. */
	ocfs2_inc_holders(lockres, level);

	ret = 0;
unlock:
	spin_unlock_irqrestore(&lockres->l_lock, flags);
out:
	/*
	 * This is helping work around a lock inversion between the page lock
	 * and dlm locks.  One path holds the page lock while calling aops
	 * which block acquiring dlm locks.  The voting thread holds dlm
	 * locks while acquiring page locks while down converting data locks.
	 * This block is helping an aop path notice the inversion and back
	 * off to unlock its page lock before trying the dlm lock again.
	 */
	if (wait && arg_flags & OCFS2_LOCK_NONBLOCK &&
	    mw.mw_mask & (OCFS2_LOCK_BUSY|OCFS2_LOCK_BLOCKED)) {
		wait = 0;
		if (lockres_remove_mask_waiter(lockres, &mw))
			ret = -EAGAIN;
		else
			goto again;
	}
	if (wait) {
		ret = ocfs2_wait_for_mask(&mw);
		if (ret == 0)
			goto again;
		mlog_errno(ret);
	}

	mlog_exit(ret);
	return ret;
}

static void ocfs2_cluster_unlock(struct ocfs2_super *osb,
				 struct ocfs2_lock_res *lockres,
				 int level)
{
	unsigned long flags;

	mlog_entry_void();
	spin_lock_irqsave(&lockres->l_lock, flags);
	ocfs2_dec_holders(lockres, level);
1158
	ocfs2_downconvert_on_unlock(osb, lockres);
1159 1160 1161 1162
	spin_unlock_irqrestore(&lockres->l_lock, flags);
	mlog_exit_void();
}

1163 1164 1165 1166
static int ocfs2_create_new_lock(struct ocfs2_super *osb,
				 struct ocfs2_lock_res *lockres,
				 int ex,
				 int local)
1167
{
1168
	int level =  ex ? DLM_LOCK_EX : DLM_LOCK_PR;
1169
	unsigned long flags;
1170
	u32 lkm_flags = local ? DLM_LKF_LOCAL : 0;
1171 1172 1173 1174 1175 1176

	spin_lock_irqsave(&lockres->l_lock, flags);
	BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
	lockres_or_flags(lockres, OCFS2_LOCK_LOCAL);
	spin_unlock_irqrestore(&lockres->l_lock, flags);

1177
	return ocfs2_lock_create(osb, lockres, level, lkm_flags);
1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188
}

/* Grants us an EX lock on the data and metadata resources, skipping
 * the normal cluster directory lookup. Use this ONLY on newly created
 * inodes which other nodes can't possibly see, and which haven't been
 * hashed in the inode hash yet. This can give us a good performance
 * increase as it'll skip the network broadcast normally associated
 * with creating a new lock resource. */
int ocfs2_create_new_inode_locks(struct inode *inode)
{
	int ret;
1189
	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1190 1191 1192 1193 1194 1195

	BUG_ON(!inode);
	BUG_ON(!ocfs2_inode_is_new(inode));

	mlog_entry_void();

1196
	mlog(0, "Inode %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
1197 1198 1199 1200 1201 1202 1203 1204 1205

	/* NOTE: That we don't increment any of the holder counts, nor
	 * do we add anything to a journal handle. Since this is
	 * supposed to be a new inode which the cluster doesn't know
	 * about yet, there is no need to.  As far as the LVB handling
	 * is concerned, this is basically like acquiring an EX lock
	 * on a resource which has an invalid one -- we'll set it
	 * valid when we release the EX. */

1206
	ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_rw_lockres, 1, 1);
1207 1208 1209 1210 1211
	if (ret) {
		mlog_errno(ret);
		goto bail;
	}

1212
	/*
1213
	 * We don't want to use DLM_LKF_LOCAL on a meta data lock as they
1214 1215
	 * don't use a generation in their lock names.
	 */
1216
	ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_inode_lockres, 1, 0);
1217 1218 1219 1220 1221
	if (ret) {
		mlog_errno(ret);
		goto bail;
	}

1222 1223 1224 1225 1226 1227
	ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_open_lockres, 0, 0);
	if (ret) {
		mlog_errno(ret);
		goto bail;
	}

1228 1229 1230 1231 1232 1233 1234 1235 1236
bail:
	mlog_exit(ret);
	return ret;
}

int ocfs2_rw_lock(struct inode *inode, int write)
{
	int status, level;
	struct ocfs2_lock_res *lockres;
S
Sunil Mushran 已提交
1237
	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1238 1239 1240 1241 1242

	BUG_ON(!inode);

	mlog_entry_void();

1243 1244
	mlog(0, "inode %llu take %s RW lock\n",
	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
1245 1246
	     write ? "EXMODE" : "PRMODE");

S
Sunil Mushran 已提交
1247 1248 1249
	if (ocfs2_mount_local(osb))
		return 0;

1250 1251
	lockres = &OCFS2_I(inode)->ip_rw_lockres;

1252
	level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264

	status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres, level, 0,
				    0);
	if (status < 0)
		mlog_errno(status);

	mlog_exit(status);
	return status;
}

void ocfs2_rw_unlock(struct inode *inode, int write)
{
1265
	int level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
1266
	struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_rw_lockres;
S
Sunil Mushran 已提交
1267
	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1268 1269 1270

	mlog_entry_void();

1271 1272
	mlog(0, "inode %llu drop %s RW lock\n",
	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
1273 1274
	     write ? "EXMODE" : "PRMODE");

S
Sunil Mushran 已提交
1275 1276
	if (!ocfs2_mount_local(osb))
		ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level);
1277 1278 1279 1280

	mlog_exit_void();
}

1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302
/*
 * ocfs2_open_lock always get PR mode lock.
 */
int ocfs2_open_lock(struct inode *inode)
{
	int status = 0;
	struct ocfs2_lock_res *lockres;
	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);

	BUG_ON(!inode);

	mlog_entry_void();

	mlog(0, "inode %llu take PRMODE open lock\n",
	     (unsigned long long)OCFS2_I(inode)->ip_blkno);

	if (ocfs2_mount_local(osb))
		goto out;

	lockres = &OCFS2_I(inode)->ip_open_lockres;

	status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres,
1303
				    DLM_LOCK_PR, 0, 0);
1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330
	if (status < 0)
		mlog_errno(status);

out:
	mlog_exit(status);
	return status;
}

int ocfs2_try_open_lock(struct inode *inode, int write)
{
	int status = 0, level;
	struct ocfs2_lock_res *lockres;
	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);

	BUG_ON(!inode);

	mlog_entry_void();

	mlog(0, "inode %llu try to take %s open lock\n",
	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
	     write ? "EXMODE" : "PRMODE");

	if (ocfs2_mount_local(osb))
		goto out;

	lockres = &OCFS2_I(inode)->ip_open_lockres;

1331
	level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
1332 1333 1334

	/*
	 * The file system may already holding a PRMODE/EXMODE open lock.
1335
	 * Since we pass DLM_LKF_NOQUEUE, the request won't block waiting on
1336 1337 1338 1339
	 * other nodes and the -EAGAIN will indicate to the caller that
	 * this inode is still in use.
	 */
	status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres,
1340
				    level, DLM_LKF_NOQUEUE, 0);
1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364

out:
	mlog_exit(status);
	return status;
}

/*
 * ocfs2_open_unlock unlock PR and EX mode open locks.
 */
void ocfs2_open_unlock(struct inode *inode)
{
	struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_open_lockres;
	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);

	mlog_entry_void();

	mlog(0, "inode %llu drop open lock\n",
	     (unsigned long long)OCFS2_I(inode)->ip_blkno);

	if (ocfs2_mount_local(osb))
		goto out;

	if(lockres->l_ro_holders)
		ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres,
1365
				     DLM_LOCK_PR);
1366 1367
	if(lockres->l_ex_holders)
		ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres,
1368
				     DLM_LOCK_EX);
1369 1370 1371 1372 1373

out:
	mlog_exit_void();
}

1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454
static int ocfs2_flock_handle_signal(struct ocfs2_lock_res *lockres,
				     int level)
{
	int ret;
	struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
	unsigned long flags;
	struct ocfs2_mask_waiter mw;

	ocfs2_init_mask_waiter(&mw);

retry_cancel:
	spin_lock_irqsave(&lockres->l_lock, flags);
	if (lockres->l_flags & OCFS2_LOCK_BUSY) {
		ret = ocfs2_prepare_cancel_convert(osb, lockres);
		if (ret) {
			spin_unlock_irqrestore(&lockres->l_lock, flags);
			ret = ocfs2_cancel_convert(osb, lockres);
			if (ret < 0) {
				mlog_errno(ret);
				goto out;
			}
			goto retry_cancel;
		}
		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
		spin_unlock_irqrestore(&lockres->l_lock, flags);

		ocfs2_wait_for_mask(&mw);
		goto retry_cancel;
	}

	ret = -ERESTARTSYS;
	/*
	 * We may still have gotten the lock, in which case there's no
	 * point to restarting the syscall.
	 */
	if (lockres->l_level == level)
		ret = 0;

	mlog(0, "Cancel returning %d. flags: 0x%lx, level: %d, act: %d\n", ret,
	     lockres->l_flags, lockres->l_level, lockres->l_action);

	spin_unlock_irqrestore(&lockres->l_lock, flags);

out:
	return ret;
}

/*
 * ocfs2_file_lock() and ocfs2_file_unlock() map to a single pair of
 * flock() calls. The locking approach this requires is sufficiently
 * different from all other cluster lock types that we implement a
 * seperate path to the "low-level" dlm calls. In particular:
 *
 * - No optimization of lock levels is done - we take at exactly
 *   what's been requested.
 *
 * - No lock caching is employed. We immediately downconvert to
 *   no-lock at unlock time. This also means flock locks never go on
 *   the blocking list).
 *
 * - Since userspace can trivially deadlock itself with flock, we make
 *   sure to allow cancellation of a misbehaving applications flock()
 *   request.
 *
 * - Access to any flock lockres doesn't require concurrency, so we
 *   can simplify the code by requiring the caller to guarantee
 *   serialization of dlmglue flock calls.
 */
int ocfs2_file_lock(struct file *file, int ex, int trylock)
{
	int ret, level = ex ? LKM_EXMODE : LKM_PRMODE;
	unsigned int lkm_flags = trylock ? LKM_NOQUEUE : 0;
	unsigned long flags;
	struct ocfs2_file_private *fp = file->private_data;
	struct ocfs2_lock_res *lockres = &fp->fp_flock;
	struct ocfs2_super *osb = OCFS2_SB(file->f_mapping->host->i_sb);
	struct ocfs2_mask_waiter mw;

	ocfs2_init_mask_waiter(&mw);

	if ((lockres->l_flags & OCFS2_LOCK_BUSY) ||
1455
	    (lockres->l_level > DLM_LOCK_NL)) {
1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493
		mlog(ML_ERROR,
		     "File lock \"%s\" has busy or locked state: flags: 0x%lx, "
		     "level: %u\n", lockres->l_name, lockres->l_flags,
		     lockres->l_level);
		return -EINVAL;
	}

	spin_lock_irqsave(&lockres->l_lock, flags);
	if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
		spin_unlock_irqrestore(&lockres->l_lock, flags);

		/*
		 * Get the lock at NLMODE to start - that way we
		 * can cancel the upconvert request if need be.
		 */
		ret = ocfs2_lock_create(osb, lockres, LKM_NLMODE, 0);
		if (ret < 0) {
			mlog_errno(ret);
			goto out;
		}

		ret = ocfs2_wait_for_mask(&mw);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
		spin_lock_irqsave(&lockres->l_lock, flags);
	}

	lockres->l_action = OCFS2_AST_CONVERT;
	lkm_flags |= LKM_CONVERT;
	lockres->l_requested = level;
	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);

	lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
	spin_unlock_irqrestore(&lockres->l_lock, flags);

1494 1495 1496
	ret = ocfs2_dlm_lock(osb->dlm, level, &lockres->l_lksb, lkm_flags,
			     lockres->l_name, OCFS2_LOCK_ID_MAX_LEN - 1,
			     lockres);
1497 1498
	if (ret) {
		if (!trylock || (ret != -EAGAIN)) {
1499
			ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560
			ret = -EINVAL;
		}

		ocfs2_recover_from_dlm_error(lockres, 1);
		lockres_remove_mask_waiter(lockres, &mw);
		goto out;
	}

	ret = ocfs2_wait_for_mask_interruptible(&mw, lockres);
	if (ret == -ERESTARTSYS) {
		/*
		 * Userspace can cause deadlock itself with
		 * flock(). Current behavior locally is to allow the
		 * deadlock, but abort the system call if a signal is
		 * received. We follow this example, otherwise a
		 * poorly written program could sit in kernel until
		 * reboot.
		 *
		 * Handling this is a bit more complicated for Ocfs2
		 * though. We can't exit this function with an
		 * outstanding lock request, so a cancel convert is
		 * required. We intentionally overwrite 'ret' - if the
		 * cancel fails and the lock was granted, it's easier
		 * to just bubble sucess back up to the user.
		 */
		ret = ocfs2_flock_handle_signal(lockres, level);
	}

out:

	mlog(0, "Lock: \"%s\" ex: %d, trylock: %d, returns: %d\n",
	     lockres->l_name, ex, trylock, ret);
	return ret;
}

void ocfs2_file_unlock(struct file *file)
{
	int ret;
	unsigned long flags;
	struct ocfs2_file_private *fp = file->private_data;
	struct ocfs2_lock_res *lockres = &fp->fp_flock;
	struct ocfs2_super *osb = OCFS2_SB(file->f_mapping->host->i_sb);
	struct ocfs2_mask_waiter mw;

	ocfs2_init_mask_waiter(&mw);

	if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED))
		return;

	if (lockres->l_level == LKM_NLMODE)
		return;

	mlog(0, "Unlock: \"%s\" flags: 0x%lx, level: %d, act: %d\n",
	     lockres->l_name, lockres->l_flags, lockres->l_level,
	     lockres->l_action);

	spin_lock_irqsave(&lockres->l_lock, flags);
	/*
	 * Fake a blocking ast for the downconvert code.
	 */
	lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED);
1561
	lockres->l_blocking = DLM_LOCK_EX;
1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577

	ocfs2_prepare_downconvert(lockres, LKM_NLMODE);
	lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
	spin_unlock_irqrestore(&lockres->l_lock, flags);

	ret = ocfs2_downconvert_lock(osb, lockres, LKM_NLMODE, 0);
	if (ret) {
		mlog_errno(ret);
		return;
	}

	ret = ocfs2_wait_for_mask(&mw);
	if (ret)
		mlog_errno(ret);
}

1578 1579
static void ocfs2_downconvert_on_unlock(struct ocfs2_super *osb,
					struct ocfs2_lock_res *lockres)
1580 1581 1582 1583 1584 1585
{
	int kick = 0;

	mlog_entry_void();

	/* If we know that another node is waiting on our lock, kick
1586
	 * the downconvert thread * pre-emptively when we reach a release
1587 1588 1589
	 * condition. */
	if (lockres->l_flags & OCFS2_LOCK_BLOCKED) {
		switch(lockres->l_blocking) {
1590
		case DLM_LOCK_EX:
1591 1592 1593
			if (!lockres->l_ex_holders && !lockres->l_ro_holders)
				kick = 1;
			break;
1594
		case DLM_LOCK_PR:
1595 1596 1597 1598 1599 1600 1601 1602 1603
			if (!lockres->l_ex_holders)
				kick = 1;
			break;
		default:
			BUG();
		}
	}

	if (kick)
1604
		ocfs2_wake_downconvert_thread(osb);
1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627

	mlog_exit_void();
}

#define OCFS2_SEC_BITS   34
#define OCFS2_SEC_SHIFT  (64 - 34)
#define OCFS2_NSEC_MASK  ((1ULL << OCFS2_SEC_SHIFT) - 1)

/* LVB only has room for 64 bits of time here so we pack it for
 * now. */
static u64 ocfs2_pack_timespec(struct timespec *spec)
{
	u64 res;
	u64 sec = spec->tv_sec;
	u32 nsec = spec->tv_nsec;

	res = (sec << OCFS2_SEC_SHIFT) | (nsec & OCFS2_NSEC_MASK);

	return res;
}

/* Call this with the lockres locked. I am reasonably sure we don't
 * need ip_lock in this function as anyone who would be changing those
1628
 * values is supposed to be blocked in ocfs2_inode_lock right now. */
1629 1630 1631
static void __ocfs2_stuff_meta_lvb(struct inode *inode)
{
	struct ocfs2_inode_info *oi = OCFS2_I(inode);
1632
	struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
1633 1634 1635 1636 1637 1638
	struct ocfs2_meta_lvb *lvb;

	mlog_entry_void();

	lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb;

1639 1640 1641 1642 1643 1644 1645 1646 1647 1648
	/*
	 * Invalidate the LVB of a deleted inode - this way other
	 * nodes are forced to go to disk and discover the new inode
	 * status.
	 */
	if (oi->ip_flags & OCFS2_INODE_DELETED) {
		lvb->lvb_version = 0;
		goto out;
	}

1649
	lvb->lvb_version   = OCFS2_LVB_VERSION;
1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661
	lvb->lvb_isize	   = cpu_to_be64(i_size_read(inode));
	lvb->lvb_iclusters = cpu_to_be32(oi->ip_clusters);
	lvb->lvb_iuid      = cpu_to_be32(inode->i_uid);
	lvb->lvb_igid      = cpu_to_be32(inode->i_gid);
	lvb->lvb_imode     = cpu_to_be16(inode->i_mode);
	lvb->lvb_inlink    = cpu_to_be16(inode->i_nlink);
	lvb->lvb_iatime_packed  =
		cpu_to_be64(ocfs2_pack_timespec(&inode->i_atime));
	lvb->lvb_ictime_packed =
		cpu_to_be64(ocfs2_pack_timespec(&inode->i_ctime));
	lvb->lvb_imtime_packed =
		cpu_to_be64(ocfs2_pack_timespec(&inode->i_mtime));
1662
	lvb->lvb_iattr    = cpu_to_be32(oi->ip_attr);
1663
	lvb->lvb_idynfeatures = cpu_to_be16(oi->ip_dyn_features);
1664
	lvb->lvb_igeneration = cpu_to_be32(inode->i_generation);
1665

1666
out:
1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681
	mlog_meta_lvb(0, lockres);

	mlog_exit_void();
}

static void ocfs2_unpack_timespec(struct timespec *spec,
				  u64 packed_time)
{
	spec->tv_sec = packed_time >> OCFS2_SEC_SHIFT;
	spec->tv_nsec = packed_time & OCFS2_NSEC_MASK;
}

static void ocfs2_refresh_inode_from_lvb(struct inode *inode)
{
	struct ocfs2_inode_info *oi = OCFS2_I(inode);
1682
	struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695
	struct ocfs2_meta_lvb *lvb;

	mlog_entry_void();

	mlog_meta_lvb(0, lockres);

	lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb;

	/* We're safe here without the lockres lock... */
	spin_lock(&oi->ip_lock);
	oi->ip_clusters = be32_to_cpu(lvb->lvb_iclusters);
	i_size_write(inode, be64_to_cpu(lvb->lvb_isize));

1696
	oi->ip_attr = be32_to_cpu(lvb->lvb_iattr);
1697
	oi->ip_dyn_features = be16_to_cpu(lvb->lvb_idynfeatures);
1698 1699
	ocfs2_set_inode_flags(inode);

1700 1701 1702 1703
	/* fast-symlinks are a special case */
	if (S_ISLNK(inode->i_mode) && !oi->ip_clusters)
		inode->i_blocks = 0;
	else
1704
		inode->i_blocks = ocfs2_inode_sector_count(inode);
1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720

	inode->i_uid     = be32_to_cpu(lvb->lvb_iuid);
	inode->i_gid     = be32_to_cpu(lvb->lvb_igid);
	inode->i_mode    = be16_to_cpu(lvb->lvb_imode);
	inode->i_nlink   = be16_to_cpu(lvb->lvb_inlink);
	ocfs2_unpack_timespec(&inode->i_atime,
			      be64_to_cpu(lvb->lvb_iatime_packed));
	ocfs2_unpack_timespec(&inode->i_mtime,
			      be64_to_cpu(lvb->lvb_imtime_packed));
	ocfs2_unpack_timespec(&inode->i_ctime,
			      be64_to_cpu(lvb->lvb_ictime_packed));
	spin_unlock(&oi->ip_lock);

	mlog_exit_void();
}

1721 1722
static inline int ocfs2_meta_lvb_is_trustable(struct inode *inode,
					      struct ocfs2_lock_res *lockres)
1723 1724 1725
{
	struct ocfs2_meta_lvb *lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb;

1726 1727
	if (lvb->lvb_version == OCFS2_LVB_VERSION
	    && be32_to_cpu(lvb->lvb_igeneration) == inode->i_generation)
1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789
		return 1;
	return 0;
}

/* Determine whether a lock resource needs to be refreshed, and
 * arbitrate who gets to refresh it.
 *
 *   0 means no refresh needed.
 *
 *   > 0 means you need to refresh this and you MUST call
 *   ocfs2_complete_lock_res_refresh afterwards. */
static int ocfs2_should_refresh_lock_res(struct ocfs2_lock_res *lockres)
{
	unsigned long flags;
	int status = 0;

	mlog_entry_void();

refresh_check:
	spin_lock_irqsave(&lockres->l_lock, flags);
	if (!(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) {
		spin_unlock_irqrestore(&lockres->l_lock, flags);
		goto bail;
	}

	if (lockres->l_flags & OCFS2_LOCK_REFRESHING) {
		spin_unlock_irqrestore(&lockres->l_lock, flags);

		ocfs2_wait_on_refreshing_lock(lockres);
		goto refresh_check;
	}

	/* Ok, I'll be the one to refresh this lock. */
	lockres_or_flags(lockres, OCFS2_LOCK_REFRESHING);
	spin_unlock_irqrestore(&lockres->l_lock, flags);

	status = 1;
bail:
	mlog_exit(status);
	return status;
}

/* If status is non zero, I'll mark it as not being in refresh
 * anymroe, but i won't clear the needs refresh flag. */
static inline void ocfs2_complete_lock_res_refresh(struct ocfs2_lock_res *lockres,
						   int status)
{
	unsigned long flags;
	mlog_entry_void();

	spin_lock_irqsave(&lockres->l_lock, flags);
	lockres_clear_flags(lockres, OCFS2_LOCK_REFRESHING);
	if (!status)
		lockres_clear_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
	spin_unlock_irqrestore(&lockres->l_lock, flags);

	wake_up(&lockres->l_event);

	mlog_exit_void();
}

/* may or may not return a bh if it went to disk. */
1790
static int ocfs2_inode_lock_update(struct inode *inode,
1791 1792 1793 1794
				  struct buffer_head **bh)
{
	int status = 0;
	struct ocfs2_inode_info *oi = OCFS2_I(inode);
1795
	struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
1796
	struct ocfs2_dinode *fe;
S
Sunil Mushran 已提交
1797
	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1798 1799 1800

	mlog_entry_void();

1801 1802 1803
	if (ocfs2_mount_local(osb))
		goto bail;

1804 1805
	spin_lock(&oi->ip_lock);
	if (oi->ip_flags & OCFS2_INODE_DELETED) {
1806
		mlog(0, "Orphaned inode %llu was deleted while we "
1807
		     "were waiting on a lock. ip_flags = 0x%x\n",
1808
		     (unsigned long long)oi->ip_blkno, oi->ip_flags);
1809 1810 1811 1812 1813 1814
		spin_unlock(&oi->ip_lock);
		status = -ENOENT;
		goto bail;
	}
	spin_unlock(&oi->ip_lock);

1815 1816
	if (!ocfs2_should_refresh_lock_res(lockres))
		goto bail;
1817 1818 1819 1820 1821

	/* This will discard any caching information we might have had
	 * for the inode metadata. */
	ocfs2_metadata_cache_purge(inode);

M
Mark Fasheh 已提交
1822 1823
	ocfs2_extent_map_trunc(inode, 0);

1824
	if (ocfs2_meta_lvb_is_trustable(inode, lockres)) {
1825 1826
		mlog(0, "Trusting LVB on inode %llu\n",
		     (unsigned long long)oi->ip_blkno);
1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853
		ocfs2_refresh_inode_from_lvb(inode);
	} else {
		/* Boo, we have to go to disk. */
		/* read bh, cast, ocfs2_refresh_inode */
		status = ocfs2_read_block(OCFS2_SB(inode->i_sb), oi->ip_blkno,
					  bh, OCFS2_BH_CACHED, inode);
		if (status < 0) {
			mlog_errno(status);
			goto bail_refresh;
		}
		fe = (struct ocfs2_dinode *) (*bh)->b_data;

		/* This is a good chance to make sure we're not
		 * locking an invalid object.
		 *
		 * We bug on a stale inode here because we checked
		 * above whether it was wiped from disk. The wiping
		 * node provides a guarantee that we receive that
		 * message and can mark the inode before dropping any
		 * locks associated with it. */
		if (!OCFS2_IS_VALID_DINODE(fe)) {
			OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
			status = -EIO;
			goto bail_refresh;
		}
		mlog_bug_on_msg(inode->i_generation !=
				le32_to_cpu(fe->i_generation),
1854
				"Invalid dinode %llu disk generation: %u "
1855
				"inode->i_generation: %u\n",
1856 1857
				(unsigned long long)oi->ip_blkno,
				le32_to_cpu(fe->i_generation),
1858 1859 1860
				inode->i_generation);
		mlog_bug_on_msg(le64_to_cpu(fe->i_dtime) ||
				!(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)),
1861 1862 1863
				"Stale dinode %llu dtime: %llu flags: 0x%x\n",
				(unsigned long long)oi->ip_blkno,
				(unsigned long long)le64_to_cpu(fe->i_dtime),
1864 1865 1866 1867 1868 1869 1870
				le32_to_cpu(fe->i_flags));

		ocfs2_refresh_inode(inode, fe);
	}

	status = 0;
bail_refresh:
1871
	ocfs2_complete_lock_res_refresh(lockres, status);
1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906
bail:
	mlog_exit(status);
	return status;
}

static int ocfs2_assign_bh(struct inode *inode,
			   struct buffer_head **ret_bh,
			   struct buffer_head *passed_bh)
{
	int status;

	if (passed_bh) {
		/* Ok, the update went to disk for us, use the
		 * returned bh. */
		*ret_bh = passed_bh;
		get_bh(*ret_bh);

		return 0;
	}

	status = ocfs2_read_block(OCFS2_SB(inode->i_sb),
				  OCFS2_I(inode)->ip_blkno,
				  ret_bh,
				  OCFS2_BH_CACHED,
				  inode);
	if (status < 0)
		mlog_errno(status);

	return status;
}

/*
 * returns < 0 error if the callback will never be called, otherwise
 * the result of the lock will be communicated via the callback.
 */
1907
int ocfs2_inode_lock_full(struct inode *inode,
1908 1909 1910 1911
			 struct buffer_head **ret_bh,
			 int ex,
			 int arg_flags)
{
1912 1913
	int status, level, acquired;
	u32 dlm_flags;
S
Sunil Mushran 已提交
1914
	struct ocfs2_lock_res *lockres = NULL;
1915 1916 1917 1918 1919 1920 1921
	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
	struct buffer_head *local_bh = NULL;

	BUG_ON(!inode);

	mlog_entry_void();

1922 1923
	mlog(0, "inode %llu, take %s META lock\n",
	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935
	     ex ? "EXMODE" : "PRMODE");

	status = 0;
	acquired = 0;
	/* We'll allow faking a readonly metadata lock for
	 * rodevices. */
	if (ocfs2_is_hard_readonly(osb)) {
		if (ex)
			status = -EROFS;
		goto bail;
	}

S
Sunil Mushran 已提交
1936 1937 1938
	if (ocfs2_mount_local(osb))
		goto local;

1939
	if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
1940
		ocfs2_wait_for_recovery(osb);
1941

1942
	lockres = &OCFS2_I(inode)->ip_inode_lockres;
1943
	level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
1944 1945
	dlm_flags = 0;
	if (arg_flags & OCFS2_META_LOCK_NOQUEUE)
1946
		dlm_flags |= DLM_LKF_NOQUEUE;
1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962

	status = ocfs2_cluster_lock(osb, lockres, level, dlm_flags, arg_flags);
	if (status < 0) {
		if (status != -EAGAIN && status != -EIOCBRETRY)
			mlog_errno(status);
		goto bail;
	}

	/* Notify the error cleanup path to drop the cluster lock. */
	acquired = 1;

	/* We wait twice because a node may have died while we were in
	 * the lower dlm layers. The second time though, we've
	 * committed to owning this lock so we don't allow signals to
	 * abort the operation. */
	if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
1963
		ocfs2_wait_for_recovery(osb);
1964

S
Sunil Mushran 已提交
1965
local:
1966 1967 1968 1969 1970 1971 1972 1973
	/*
	 * We only see this flag if we're being called from
	 * ocfs2_read_locked_inode(). It means we're locking an inode
	 * which hasn't been populated yet, so clear the refresh flag
	 * and let the caller handle it.
	 */
	if (inode->i_state & I_NEW) {
		status = 0;
S
Sunil Mushran 已提交
1974 1975
		if (lockres)
			ocfs2_complete_lock_res_refresh(lockres, 0);
1976 1977 1978
		goto bail;
	}

1979
	/* This is fun. The caller may want a bh back, or it may
1980
	 * not. ocfs2_inode_lock_update definitely wants one in, but
1981 1982 1983
	 * may or may not read one, depending on what's in the
	 * LVB. The result of all of this is that we've *only* gone to
	 * disk if we have to, so the complexity is worthwhile. */
1984
	status = ocfs2_inode_lock_update(inode, &local_bh);
1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005
	if (status < 0) {
		if (status != -ENOENT)
			mlog_errno(status);
		goto bail;
	}

	if (ret_bh) {
		status = ocfs2_assign_bh(inode, ret_bh, local_bh);
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}
	}

bail:
	if (status < 0) {
		if (ret_bh && (*ret_bh)) {
			brelse(*ret_bh);
			*ret_bh = NULL;
		}
		if (acquired)
2006
			ocfs2_inode_unlock(inode, ex);
2007 2008 2009 2010 2011 2012 2013 2014 2015 2016
	}

	if (local_bh)
		brelse(local_bh);

	mlog_exit(status);
	return status;
}

/*
2017 2018 2019
 * This is working around a lock inversion between tasks acquiring DLM
 * locks while holding a page lock and the downconvert thread which
 * blocks dlm lock acquiry while acquiring page locks.
2020 2021 2022 2023 2024
 *
 * ** These _with_page variantes are only intended to be called from aop
 * methods that hold page locks and return a very specific *positive* error
 * code that aop methods pass up to the VFS -- test for errors with != 0. **
 *
2025 2026 2027 2028 2029 2030
 * The DLM is called such that it returns -EAGAIN if it would have
 * blocked waiting for the downconvert thread.  In that case we unlock
 * our page so the downconvert thread can make progress.  Once we've
 * done this we have to return AOP_TRUNCATED_PAGE so the aop method
 * that called us can bubble that back up into the VFS who will then
 * immediately retry the aop call.
2031 2032 2033 2034 2035 2036 2037
 *
 * We do a blocking lock and immediate unlock before returning, though, so that
 * the lock has a great chance of being cached on this node by the time the VFS
 * calls back to retry the aop.    This has a potential to livelock as nodes
 * ping locks back and forth, but that's a risk we're willing to take to avoid
 * the lock inversion simply.
 */
2038
int ocfs2_inode_lock_with_page(struct inode *inode,
2039 2040 2041 2042 2043 2044
			      struct buffer_head **ret_bh,
			      int ex,
			      struct page *page)
{
	int ret;

2045
	ret = ocfs2_inode_lock_full(inode, ret_bh, ex, OCFS2_LOCK_NONBLOCK);
2046 2047
	if (ret == -EAGAIN) {
		unlock_page(page);
2048 2049
		if (ocfs2_inode_lock(inode, ret_bh, ex) == 0)
			ocfs2_inode_unlock(inode, ex);
2050 2051 2052 2053 2054 2055
		ret = AOP_TRUNCATED_PAGE;
	}

	return ret;
}

2056
int ocfs2_inode_lock_atime(struct inode *inode,
2057 2058 2059 2060 2061 2062
			  struct vfsmount *vfsmnt,
			  int *level)
{
	int ret;

	mlog_entry_void();
2063
	ret = ocfs2_inode_lock(inode, NULL, 0);
2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075
	if (ret < 0) {
		mlog_errno(ret);
		return ret;
	}

	/*
	 * If we should update atime, we will get EX lock,
	 * otherwise we just get PR lock.
	 */
	if (ocfs2_should_update_atime(inode, vfsmnt)) {
		struct buffer_head *bh = NULL;

2076 2077
		ocfs2_inode_unlock(inode, 0);
		ret = ocfs2_inode_lock(inode, &bh, 1);
2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093
		if (ret < 0) {
			mlog_errno(ret);
			return ret;
		}
		*level = 1;
		if (ocfs2_should_update_atime(inode, vfsmnt))
			ocfs2_update_inode_atime(inode, bh);
		if (bh)
			brelse(bh);
	} else
		*level = 0;

	mlog_exit(ret);
	return ret;
}

2094
void ocfs2_inode_unlock(struct inode *inode,
2095 2096
		       int ex)
{
2097
	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2098
	struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_inode_lockres;
S
Sunil Mushran 已提交
2099
	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2100 2101 2102

	mlog_entry_void();

2103 2104
	mlog(0, "inode %llu drop %s META lock\n",
	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
2105 2106
	     ex ? "EXMODE" : "PRMODE");

S
Sunil Mushran 已提交
2107 2108
	if (!ocfs2_is_hard_readonly(OCFS2_SB(inode->i_sb)) &&
	    !ocfs2_mount_local(osb))
2109 2110 2111 2112 2113 2114 2115 2116
		ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level);

	mlog_exit_void();
}

int ocfs2_super_lock(struct ocfs2_super *osb,
		     int ex)
{
S
Sunil Mushran 已提交
2117
	int status = 0;
2118
	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2119 2120 2121 2122 2123 2124 2125
	struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;

	mlog_entry_void();

	if (ocfs2_is_hard_readonly(osb))
		return -EROFS;

S
Sunil Mushran 已提交
2126 2127 2128
	if (ocfs2_mount_local(osb))
		goto bail;

2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144
	status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

	/* The super block lock path is really in the best position to
	 * know when resources covered by the lock need to be
	 * refreshed, so we do it here. Of course, making sense of
	 * everything is up to the caller :) */
	status = ocfs2_should_refresh_lock_res(lockres);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}
	if (status) {
2145
		status = ocfs2_refresh_slot_info(osb);
2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159

		ocfs2_complete_lock_res_refresh(lockres, status);

		if (status < 0)
			mlog_errno(status);
	}
bail:
	mlog_exit(status);
	return status;
}

void ocfs2_super_unlock(struct ocfs2_super *osb,
			int ex)
{
2160
	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2161 2162
	struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;

S
Sunil Mushran 已提交
2163 2164
	if (!ocfs2_mount_local(osb))
		ocfs2_cluster_unlock(osb, lockres, level);
2165 2166 2167 2168 2169 2170 2171 2172 2173 2174
}

int ocfs2_rename_lock(struct ocfs2_super *osb)
{
	int status;
	struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;

	if (ocfs2_is_hard_readonly(osb))
		return -EROFS;

S
Sunil Mushran 已提交
2175 2176 2177
	if (ocfs2_mount_local(osb))
		return 0;

2178
	status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_EX, 0, 0);
2179 2180 2181 2182 2183 2184 2185 2186 2187 2188
	if (status < 0)
		mlog_errno(status);

	return status;
}

void ocfs2_rename_unlock(struct ocfs2_super *osb)
{
	struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;

S
Sunil Mushran 已提交
2189
	if (!ocfs2_mount_local(osb))
2190
		ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX);
2191 2192
}

2193 2194 2195
int ocfs2_dentry_lock(struct dentry *dentry, int ex)
{
	int ret;
2196
	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2197 2198 2199 2200 2201 2202 2203 2204
	struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
	struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);

	BUG_ON(!dl);

	if (ocfs2_is_hard_readonly(osb))
		return -EROFS;

S
Sunil Mushran 已提交
2205 2206 2207
	if (ocfs2_mount_local(osb))
		return 0;

2208 2209 2210 2211 2212 2213 2214 2215 2216
	ret = ocfs2_cluster_lock(osb, &dl->dl_lockres, level, 0, 0);
	if (ret < 0)
		mlog_errno(ret);

	return ret;
}

void ocfs2_dentry_unlock(struct dentry *dentry, int ex)
{
2217
	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2218 2219 2220
	struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
	struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);

S
Sunil Mushran 已提交
2221 2222
	if (!ocfs2_mount_local(osb))
		ocfs2_cluster_unlock(osb, &dl->dl_lockres, level);
2223 2224
}

2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354
/* Reference counting of the dlm debug structure. We want this because
 * open references on the debug inodes can live on after a mount, so
 * we can't rely on the ocfs2_super to always exist. */
static void ocfs2_dlm_debug_free(struct kref *kref)
{
	struct ocfs2_dlm_debug *dlm_debug;

	dlm_debug = container_of(kref, struct ocfs2_dlm_debug, d_refcnt);

	kfree(dlm_debug);
}

void ocfs2_put_dlm_debug(struct ocfs2_dlm_debug *dlm_debug)
{
	if (dlm_debug)
		kref_put(&dlm_debug->d_refcnt, ocfs2_dlm_debug_free);
}

static void ocfs2_get_dlm_debug(struct ocfs2_dlm_debug *debug)
{
	kref_get(&debug->d_refcnt);
}

struct ocfs2_dlm_debug *ocfs2_new_dlm_debug(void)
{
	struct ocfs2_dlm_debug *dlm_debug;

	dlm_debug = kmalloc(sizeof(struct ocfs2_dlm_debug), GFP_KERNEL);
	if (!dlm_debug) {
		mlog_errno(-ENOMEM);
		goto out;
	}

	kref_init(&dlm_debug->d_refcnt);
	INIT_LIST_HEAD(&dlm_debug->d_lockres_tracking);
	dlm_debug->d_locking_state = NULL;
out:
	return dlm_debug;
}

/* Access to this is arbitrated for us via seq_file->sem. */
struct ocfs2_dlm_seq_priv {
	struct ocfs2_dlm_debug *p_dlm_debug;
	struct ocfs2_lock_res p_iter_res;
	struct ocfs2_lock_res p_tmp_res;
};

static struct ocfs2_lock_res *ocfs2_dlm_next_res(struct ocfs2_lock_res *start,
						 struct ocfs2_dlm_seq_priv *priv)
{
	struct ocfs2_lock_res *iter, *ret = NULL;
	struct ocfs2_dlm_debug *dlm_debug = priv->p_dlm_debug;

	assert_spin_locked(&ocfs2_dlm_tracking_lock);

	list_for_each_entry(iter, &start->l_debug_list, l_debug_list) {
		/* discover the head of the list */
		if (&iter->l_debug_list == &dlm_debug->d_lockres_tracking) {
			mlog(0, "End of list found, %p\n", ret);
			break;
		}

		/* We track our "dummy" iteration lockres' by a NULL
		 * l_ops field. */
		if (iter->l_ops != NULL) {
			ret = iter;
			break;
		}
	}

	return ret;
}

static void *ocfs2_dlm_seq_start(struct seq_file *m, loff_t *pos)
{
	struct ocfs2_dlm_seq_priv *priv = m->private;
	struct ocfs2_lock_res *iter;

	spin_lock(&ocfs2_dlm_tracking_lock);
	iter = ocfs2_dlm_next_res(&priv->p_iter_res, priv);
	if (iter) {
		/* Since lockres' have the lifetime of their container
		 * (which can be inodes, ocfs2_supers, etc) we want to
		 * copy this out to a temporary lockres while still
		 * under the spinlock. Obviously after this we can't
		 * trust any pointers on the copy returned, but that's
		 * ok as the information we want isn't typically held
		 * in them. */
		priv->p_tmp_res = *iter;
		iter = &priv->p_tmp_res;
	}
	spin_unlock(&ocfs2_dlm_tracking_lock);

	return iter;
}

static void ocfs2_dlm_seq_stop(struct seq_file *m, void *v)
{
}

static void *ocfs2_dlm_seq_next(struct seq_file *m, void *v, loff_t *pos)
{
	struct ocfs2_dlm_seq_priv *priv = m->private;
	struct ocfs2_lock_res *iter = v;
	struct ocfs2_lock_res *dummy = &priv->p_iter_res;

	spin_lock(&ocfs2_dlm_tracking_lock);
	iter = ocfs2_dlm_next_res(iter, priv);
	list_del_init(&dummy->l_debug_list);
	if (iter) {
		list_add(&dummy->l_debug_list, &iter->l_debug_list);
		priv->p_tmp_res = *iter;
		iter = &priv->p_tmp_res;
	}
	spin_unlock(&ocfs2_dlm_tracking_lock);

	return iter;
}

/* So that debugfs.ocfs2 can determine which format is being used */
#define OCFS2_DLM_DEBUG_STR_VERSION 1
static int ocfs2_dlm_seq_show(struct seq_file *m, void *v)
{
	int i;
	char *lvb;
	struct ocfs2_lock_res *lockres = v;

	if (!lockres)
		return -EINVAL;

2355 2356 2357 2358 2359 2360 2361 2362 2363 2364
	seq_printf(m, "0x%x\t", OCFS2_DLM_DEBUG_STR_VERSION);

	if (lockres->l_type == OCFS2_LOCK_TYPE_DENTRY)
		seq_printf(m, "%.*s%08x\t", OCFS2_DENTRY_LOCK_INO_START - 1,
			   lockres->l_name,
			   (unsigned int)ocfs2_get_dentry_lock_ino(lockres));
	else
		seq_printf(m, "%.*s\t", OCFS2_LOCK_ID_MAX_LEN, lockres->l_name);

	seq_printf(m, "%d\t"
2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390
		   "0x%lx\t"
		   "0x%x\t"
		   "0x%x\t"
		   "%u\t"
		   "%u\t"
		   "%d\t"
		   "%d\t",
		   lockres->l_level,
		   lockres->l_flags,
		   lockres->l_action,
		   lockres->l_unlock_action,
		   lockres->l_ro_holders,
		   lockres->l_ex_holders,
		   lockres->l_requested,
		   lockres->l_blocking);

	/* Dump the raw LVB */
	lvb = lockres->l_lksb.lvb;
	for(i = 0; i < DLM_LVB_LEN; i++)
		seq_printf(m, "0x%x\t", lvb[i]);

	/* End the line */
	seq_printf(m, "\n");
	return 0;
}

2391
static const struct seq_operations ocfs2_dlm_seq_ops = {
2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421
	.start =	ocfs2_dlm_seq_start,
	.stop =		ocfs2_dlm_seq_stop,
	.next =		ocfs2_dlm_seq_next,
	.show =		ocfs2_dlm_seq_show,
};

static int ocfs2_dlm_debug_release(struct inode *inode, struct file *file)
{
	struct seq_file *seq = (struct seq_file *) file->private_data;
	struct ocfs2_dlm_seq_priv *priv = seq->private;
	struct ocfs2_lock_res *res = &priv->p_iter_res;

	ocfs2_remove_lockres_tracking(res);
	ocfs2_put_dlm_debug(priv->p_dlm_debug);
	return seq_release_private(inode, file);
}

static int ocfs2_dlm_debug_open(struct inode *inode, struct file *file)
{
	int ret;
	struct ocfs2_dlm_seq_priv *priv;
	struct seq_file *seq;
	struct ocfs2_super *osb;

	priv = kzalloc(sizeof(struct ocfs2_dlm_seq_priv), GFP_KERNEL);
	if (!priv) {
		ret = -ENOMEM;
		mlog_errno(ret);
		goto out;
	}
2422
	osb = inode->i_private;
2423 2424 2425 2426 2427 2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442 2443
	ocfs2_get_dlm_debug(osb->osb_dlm_debug);
	priv->p_dlm_debug = osb->osb_dlm_debug;
	INIT_LIST_HEAD(&priv->p_iter_res.l_debug_list);

	ret = seq_open(file, &ocfs2_dlm_seq_ops);
	if (ret) {
		kfree(priv);
		mlog_errno(ret);
		goto out;
	}

	seq = (struct seq_file *) file->private_data;
	seq->private = priv;

	ocfs2_add_lockres_tracking(&priv->p_iter_res,
				   priv->p_dlm_debug);

out:
	return ret;
}

2444
static const struct file_operations ocfs2_dlm_debug_fops = {
2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484
	.open =		ocfs2_dlm_debug_open,
	.release =	ocfs2_dlm_debug_release,
	.read =		seq_read,
	.llseek =	seq_lseek,
};

static int ocfs2_dlm_init_debug(struct ocfs2_super *osb)
{
	int ret = 0;
	struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;

	dlm_debug->d_locking_state = debugfs_create_file("locking_state",
							 S_IFREG|S_IRUSR,
							 osb->osb_debug_root,
							 osb,
							 &ocfs2_dlm_debug_fops);
	if (!dlm_debug->d_locking_state) {
		ret = -EINVAL;
		mlog(ML_ERROR,
		     "Unable to create locking state debugfs file.\n");
		goto out;
	}

	ocfs2_get_dlm_debug(dlm_debug);
out:
	return ret;
}

static void ocfs2_dlm_shutdown_debug(struct ocfs2_super *osb)
{
	struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;

	if (dlm_debug) {
		debugfs_remove(dlm_debug->d_locking_state);
		ocfs2_put_dlm_debug(dlm_debug);
	}
}

int ocfs2_dlm_init(struct ocfs2_super *osb)
{
S
Sunil Mushran 已提交
2485
	int status = 0;
2486
	u32 dlm_key;
S
Sunil Mushran 已提交
2487
	struct dlm_ctxt *dlm = NULL;
2488 2489 2490

	mlog_entry_void();

S
Sunil Mushran 已提交
2491 2492 2493
	if (ocfs2_mount_local(osb))
		goto local;

2494 2495 2496 2497 2498 2499
	status = ocfs2_dlm_init_debug(osb);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

2500 2501 2502 2503 2504
	/* launch downconvert thread */
	osb->dc_task = kthread_run(ocfs2_downconvert_thread, osb, "ocfs2dc");
	if (IS_ERR(osb->dc_task)) {
		status = PTR_ERR(osb->dc_task);
		osb->dc_task = NULL;
2505 2506 2507 2508 2509 2510 2511 2512 2513
		mlog_errno(status);
		goto bail;
	}

	/* used by the dlm code to make message headers unique, each
	 * node in this domain must agree on this. */
	dlm_key = crc32_le(0, osb->uuid_str, strlen(osb->uuid_str));

	/* for now, uuid == domain */
2514 2515
	dlm = dlm_register_domain(osb->uuid_str, dlm_key,
				  &osb->osb_locking_proto);
2516 2517 2518 2519 2520 2521
	if (IS_ERR(dlm)) {
		status = PTR_ERR(dlm);
		mlog_errno(status);
		goto bail;
	}

S
Sunil Mushran 已提交
2522 2523 2524
	dlm_register_eviction_cb(dlm, &osb->osb_eviction_cb);

local:
2525 2526 2527 2528 2529 2530 2531 2532 2533
	ocfs2_super_lock_res_init(&osb->osb_super_lockres, osb);
	ocfs2_rename_lock_res_init(&osb->osb_rename_lockres, osb);

	osb->dlm = dlm;

	status = 0;
bail:
	if (status < 0) {
		ocfs2_dlm_shutdown_debug(osb);
2534 2535
		if (osb->dc_task)
			kthread_stop(osb->dc_task);
2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549
	}

	mlog_exit(status);
	return status;
}

void ocfs2_dlm_shutdown(struct ocfs2_super *osb)
{
	mlog_entry_void();

	dlm_unregister_eviction_cb(&osb->osb_eviction_cb);

	ocfs2_drop_osb_locks(osb);

2550 2551 2552
	if (osb->dc_task) {
		kthread_stop(osb->dc_task);
		osb->dc_task = NULL;
2553 2554 2555 2556 2557 2558 2559 2560 2561 2562 2563 2564 2565
	}

	ocfs2_lock_res_free(&osb->osb_super_lockres);
	ocfs2_lock_res_free(&osb->osb_rename_lockres);

	dlm_unregister_domain(osb->dlm);
	osb->dlm = NULL;

	ocfs2_dlm_shutdown_debug(osb);

	mlog_exit_void();
}

2566
static void ocfs2_unlock_ast(void *opaque, int error)
2567 2568 2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581
{
	struct ocfs2_lock_res *lockres = opaque;
	unsigned long flags;

	mlog_entry_void();

	mlog(0, "UNLOCK AST called on lock %s, action = %d\n", lockres->l_name,
	     lockres->l_unlock_action);

	spin_lock_irqsave(&lockres->l_lock, flags);
	/* We tried to cancel a convert request, but it was already
	 * granted. All we want to do here is clear our unlock
	 * state. The wake_up call done at the bottom is redundant
	 * (ocfs2_prepare_cancel_convert doesn't sleep on this) but doesn't
	 * hurt anything anyway */
2582
	if (error == -DLM_ECANCEL &&
2583 2584 2585 2586 2587 2588 2589 2590 2591
	    lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) {
		mlog(0, "Got cancelgrant for %s\n", lockres->l_name);

		/* We don't clear the busy flag in this case as it
		 * should have been cleared by the ast which the dlm
		 * has called. */
		goto complete_unlock;
	}

2592 2593 2594 2595
	/* DLM_EUNLOCK is the success code for unlock */
	if (error != -DLM_EUNLOCK) {
		mlog(ML_ERROR, "Dlm passes error %d for lock %s, "
		     "unlock_action %d\n", error, lockres->l_name,
2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606
		     lockres->l_unlock_action);
		spin_unlock_irqrestore(&lockres->l_lock, flags);
		return;
	}

	switch(lockres->l_unlock_action) {
	case OCFS2_UNLOCK_CANCEL_CONVERT:
		mlog(0, "Cancel convert success for %s\n", lockres->l_name);
		lockres->l_action = OCFS2_AST_INVALID;
		break;
	case OCFS2_UNLOCK_DROP_LOCK:
2607
		lockres->l_level = DLM_LOCK_IV;
2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623
		break;
	default:
		BUG();
	}

	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
complete_unlock:
	lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
	spin_unlock_irqrestore(&lockres->l_lock, flags);

	wake_up(&lockres->l_event);

	mlog_exit_void();
}

static int ocfs2_drop_lock(struct ocfs2_super *osb,
2624
			   struct ocfs2_lock_res *lockres)
2625
{
2626
	int ret;
2627
	unsigned long flags;
2628
	u32 lkm_flags = 0;
2629 2630 2631 2632 2633

	/* We didn't get anywhere near actually using this lockres. */
	if (!(lockres->l_flags & OCFS2_LOCK_INITIALIZED))
		goto out;

2634
	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
2635
		lkm_flags |= DLM_LKF_VALBLK;
2636

2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658
	spin_lock_irqsave(&lockres->l_lock, flags);

	mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_FREEING),
			"lockres %s, flags 0x%lx\n",
			lockres->l_name, lockres->l_flags);

	while (lockres->l_flags & OCFS2_LOCK_BUSY) {
		mlog(0, "waiting on busy lock \"%s\": flags = %lx, action = "
		     "%u, unlock_action = %u\n",
		     lockres->l_name, lockres->l_flags, lockres->l_action,
		     lockres->l_unlock_action);

		spin_unlock_irqrestore(&lockres->l_lock, flags);

		/* XXX: Today we just wait on any busy
		 * locks... Perhaps we need to cancel converts in the
		 * future? */
		ocfs2_wait_on_busy_lock(lockres);

		spin_lock_irqsave(&lockres->l_lock, flags);
	}

2659 2660
	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) {
		if (lockres->l_flags & OCFS2_LOCK_ATTACHED &&
2661
		    lockres->l_level == DLM_LOCK_EX &&
2662 2663 2664
		    !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
			lockres->l_ops->set_lvb(lockres);
	}
2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689

	if (lockres->l_flags & OCFS2_LOCK_BUSY)
		mlog(ML_ERROR, "destroying busy lock: \"%s\"\n",
		     lockres->l_name);
	if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
		mlog(0, "destroying blocked lock: \"%s\"\n", lockres->l_name);

	if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
		spin_unlock_irqrestore(&lockres->l_lock, flags);
		goto out;
	}

	lockres_clear_flags(lockres, OCFS2_LOCK_ATTACHED);

	/* make sure we never get here while waiting for an ast to
	 * fire. */
	BUG_ON(lockres->l_action != OCFS2_AST_INVALID);

	/* is this necessary? */
	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
	lockres->l_unlock_action = OCFS2_UNLOCK_DROP_LOCK;
	spin_unlock_irqrestore(&lockres->l_lock, flags);

	mlog(0, "lock %s\n", lockres->l_name);

2690 2691 2692 2693
	ret = ocfs2_dlm_unlock(osb->dlm, &lockres->l_lksb, lkm_flags,
			       lockres);
	if (ret) {
		ocfs2_log_dlm_error("ocfs2_dlm_unlock", ret, lockres);
2694 2695 2696 2697
		mlog(ML_ERROR, "lockres flags: %lu\n", lockres->l_flags);
		dlm_print_one_lock(lockres->l_lksb.lockid);
		BUG();
	}
2698
	mlog(0, "lock %s, successfull return from ocfs2_dlm_unlock\n",
2699 2700 2701 2702 2703 2704 2705 2706 2707 2708
	     lockres->l_name);

	ocfs2_wait_on_busy_lock(lockres);
out:
	mlog_exit(0);
	return 0;
}

/* Mark the lockres as being dropped. It will no longer be
 * queued if blocking, but we still may have to wait on it
2709
 * being dequeued from the downconvert thread before we can consider
2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737
 * it safe to drop. 
 *
 * You can *not* attempt to call cluster_lock on this lockres anymore. */
void ocfs2_mark_lockres_freeing(struct ocfs2_lock_res *lockres)
{
	int status;
	struct ocfs2_mask_waiter mw;
	unsigned long flags;

	ocfs2_init_mask_waiter(&mw);

	spin_lock_irqsave(&lockres->l_lock, flags);
	lockres->l_flags |= OCFS2_LOCK_FREEING;
	while (lockres->l_flags & OCFS2_LOCK_QUEUED) {
		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_QUEUED, 0);
		spin_unlock_irqrestore(&lockres->l_lock, flags);

		mlog(0, "Waiting on lockres %s\n", lockres->l_name);

		status = ocfs2_wait_for_mask(&mw);
		if (status)
			mlog_errno(status);

		spin_lock_irqsave(&lockres->l_lock, flags);
	}
	spin_unlock_irqrestore(&lockres->l_lock, flags);
}

2738 2739
void ocfs2_simple_drop_lockres(struct ocfs2_super *osb,
			       struct ocfs2_lock_res *lockres)
2740
{
2741
	int ret;
2742

2743
	ocfs2_mark_lockres_freeing(lockres);
2744
	ret = ocfs2_drop_lock(osb, lockres);
2745 2746 2747
	if (ret)
		mlog_errno(ret);
}
2748

2749 2750 2751 2752
static void ocfs2_drop_osb_locks(struct ocfs2_super *osb)
{
	ocfs2_simple_drop_lockres(osb, &osb->osb_super_lockres);
	ocfs2_simple_drop_lockres(osb, &osb->osb_rename_lockres);
2753 2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764
}

int ocfs2_drop_inode_locks(struct inode *inode)
{
	int status, err;

	mlog_entry_void();

	/* No need to call ocfs2_mark_lockres_freeing here -
	 * ocfs2_clear_inode has done it for us. */

	err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
2765
			      &OCFS2_I(inode)->ip_open_lockres);
2766 2767 2768 2769 2770 2771
	if (err < 0)
		mlog_errno(err);

	status = err;

	err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
2772
			      &OCFS2_I(inode)->ip_inode_lockres);
2773 2774 2775 2776 2777 2778
	if (err < 0)
		mlog_errno(err);
	if (err < 0 && !status)
		status = err;

	err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
2779
			      &OCFS2_I(inode)->ip_rw_lockres);
2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793
	if (err < 0)
		mlog_errno(err);
	if (err < 0 && !status)
		status = err;

	mlog_exit(status);
	return status;
}

static void ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres,
				      int new_level)
{
	assert_spin_locked(&lockres->l_lock);

2794
	BUG_ON(lockres->l_blocking <= DLM_LOCK_NL);
2795 2796

	if (lockres->l_level <= new_level) {
2797
		mlog(ML_ERROR, "lockres->l_level (%d) <= new_level (%d)\n",
2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814
		     lockres->l_level, new_level);
		BUG();
	}

	mlog(0, "lock %s, new_level = %d, l_blocking = %d\n",
	     lockres->l_name, new_level, lockres->l_blocking);

	lockres->l_action = OCFS2_AST_DOWNCONVERT;
	lockres->l_requested = new_level;
	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
}

static int ocfs2_downconvert_lock(struct ocfs2_super *osb,
				  struct ocfs2_lock_res *lockres,
				  int new_level,
				  int lvb)
{
2815 2816
	int ret;
	u32 dlm_flags = DLM_LKF_CONVERT;
2817 2818 2819 2820

	mlog_entry_void();

	if (lvb)
2821
		dlm_flags |= DLM_LKF_VALBLK;
2822

2823 2824 2825 2826 2827 2828 2829 2830 2831
	ret = ocfs2_dlm_lock(osb->dlm,
			     new_level,
			     &lockres->l_lksb,
			     dlm_flags,
			     lockres->l_name,
			     OCFS2_LOCK_ID_MAX_LEN - 1,
			     lockres);
	if (ret) {
		ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
2832 2833 2834 2835 2836 2837 2838 2839 2840 2841
		ocfs2_recover_from_dlm_error(lockres, 1);
		goto bail;
	}

	ret = 0;
bail:
	mlog_exit(ret);
	return ret;
}

2842
/* returns 1 when the caller should unlock and call ocfs2_dlm_unlock */
2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881
static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb,
				        struct ocfs2_lock_res *lockres)
{
	assert_spin_locked(&lockres->l_lock);

	mlog_entry_void();
	mlog(0, "lock %s\n", lockres->l_name);

	if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) {
		/* If we're already trying to cancel a lock conversion
		 * then just drop the spinlock and allow the caller to
		 * requeue this lock. */

		mlog(0, "Lockres %s, skip convert\n", lockres->l_name);
		return 0;
	}

	/* were we in a convert when we got the bast fire? */
	BUG_ON(lockres->l_action != OCFS2_AST_CONVERT &&
	       lockres->l_action != OCFS2_AST_DOWNCONVERT);
	/* set things up for the unlockast to know to just
	 * clear out the ast_action and unset busy, etc. */
	lockres->l_unlock_action = OCFS2_UNLOCK_CANCEL_CONVERT;

	mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_BUSY),
			"lock %s, invalid flags: 0x%lx\n",
			lockres->l_name, lockres->l_flags);

	return 1;
}

static int ocfs2_cancel_convert(struct ocfs2_super *osb,
				struct ocfs2_lock_res *lockres)
{
	int ret;

	mlog_entry_void();
	mlog(0, "lock %s\n", lockres->l_name);

2882 2883 2884 2885
	ret = ocfs2_dlm_unlock(osb->dlm, &lockres->l_lksb,
			       DLM_LKF_CANCEL, lockres);
	if (ret) {
		ocfs2_log_dlm_error("ocfs2_dlm_unlock", ret, lockres);
2886 2887 2888
		ocfs2_recover_from_dlm_error(lockres, 0);
	}

2889
	mlog(0, "lock %s return from ocfs2_dlm_unlock\n", lockres->l_name);
2890 2891 2892 2893 2894

	mlog_exit(ret);
	return ret;
}

2895 2896 2897
static int ocfs2_unblock_lock(struct ocfs2_super *osb,
			      struct ocfs2_lock_res *lockres,
			      struct ocfs2_unblock_ctl *ctl)
2898 2899 2900 2901 2902
{
	unsigned long flags;
	int blocking;
	int new_level;
	int ret = 0;
2903
	int set_lvb = 0;
2904 2905 2906 2907 2908 2909 2910 2911 2912

	mlog_entry_void();

	spin_lock_irqsave(&lockres->l_lock, flags);

	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));

recheck:
	if (lockres->l_flags & OCFS2_LOCK_BUSY) {
2913
		ctl->requeue = 1;
2914 2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925
		ret = ocfs2_prepare_cancel_convert(osb, lockres);
		spin_unlock_irqrestore(&lockres->l_lock, flags);
		if (ret) {
			ret = ocfs2_cancel_convert(osb, lockres);
			if (ret < 0)
				mlog_errno(ret);
		}
		goto leave;
	}

	/* if we're blocking an exclusive and we have *any* holders,
	 * then requeue. */
2926
	if ((lockres->l_blocking == DLM_LOCK_EX)
2927 2928
	    && (lockres->l_ex_holders || lockres->l_ro_holders))
		goto leave_requeue;
2929 2930 2931

	/* If it's a PR we're blocking, then only
	 * requeue if we've got any EX holders */
2932
	if (lockres->l_blocking == DLM_LOCK_PR &&
2933 2934 2935 2936 2937 2938 2939 2940 2941 2942
	    lockres->l_ex_holders)
		goto leave_requeue;

	/*
	 * Can we get a lock in this state if the holder counts are
	 * zero? The meta data unblock code used to check this.
	 */
	if ((lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
	    && (lockres->l_flags & OCFS2_LOCK_REFRESHING))
		goto leave_requeue;
2943

2944 2945 2946 2947 2948 2949
	new_level = ocfs2_highest_compat_lock_level(lockres->l_blocking);

	if (lockres->l_ops->check_downconvert
	    && !lockres->l_ops->check_downconvert(lockres, new_level))
		goto leave_requeue;

2950 2951 2952
	/* If we get here, then we know that there are no more
	 * incompatible holders (and anyone asking for an incompatible
	 * lock is blocked). We can now downconvert the lock */
2953
	if (!lockres->l_ops->downconvert_worker)
2954 2955 2956 2957 2958 2959 2960 2961 2962
		goto downconvert;

	/* Some lockres types want to do a bit of work before
	 * downconverting a lock. Allow that here. The worker function
	 * may sleep, so we save off a copy of what we're blocking as
	 * it may change while we're not holding the spin lock. */
	blocking = lockres->l_blocking;
	spin_unlock_irqrestore(&lockres->l_lock, flags);

2963
	ctl->unblock_action = lockres->l_ops->downconvert_worker(lockres, blocking);
2964 2965 2966

	if (ctl->unblock_action == UNBLOCK_STOP_POST)
		goto leave;
2967 2968 2969 2970 2971 2972 2973 2974 2975

	spin_lock_irqsave(&lockres->l_lock, flags);
	if (blocking != lockres->l_blocking) {
		/* If this changed underneath us, then we can't drop
		 * it just yet. */
		goto recheck;
	}

downconvert:
2976
	ctl->requeue = 0;
2977

2978
	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) {
2979
		if (lockres->l_level == DLM_LOCK_EX)
2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991
			set_lvb = 1;

		/*
		 * We only set the lvb if the lock has been fully
		 * refreshed - otherwise we risk setting stale
		 * data. Otherwise, there's no need to actually clear
		 * out the lvb here as it's value is still valid.
		 */
		if (set_lvb && !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
			lockres->l_ops->set_lvb(lockres);
	}

2992 2993
	ocfs2_prepare_downconvert(lockres, new_level);
	spin_unlock_irqrestore(&lockres->l_lock, flags);
2994
	ret = ocfs2_downconvert_lock(osb, lockres, new_level, set_lvb);
2995 2996 2997
leave:
	mlog_exit(ret);
	return ret;
2998 2999 3000 3001 3002 3003 3004

leave_requeue:
	spin_unlock_irqrestore(&lockres->l_lock, flags);
	ctl->requeue = 1;

	mlog_exit(0);
	return 0;
3005 3006
}

3007 3008
static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
				     int blocking)
3009 3010 3011 3012 3013 3014 3015
{
	struct inode *inode;
	struct address_space *mapping;

       	inode = ocfs2_lock_res_inode(lockres);
	mapping = inode->i_mapping;

3016
	if (!S_ISREG(inode->i_mode))
3017 3018
		goto out;

3019 3020 3021 3022 3023 3024 3025 3026 3027
	/*
	 * We need this before the filemap_fdatawrite() so that it can
	 * transfer the dirty bit from the PTE to the
	 * page. Unfortunately this means that even for EX->PR
	 * downconverts, we'll lose our mappings and have to build
	 * them up again.
	 */
	unmap_mapping_range(mapping, 0, 0, 0);

3028
	if (filemap_fdatawrite(mapping)) {
3029 3030
		mlog(ML_ERROR, "Could not sync inode %llu for downconvert!",
		     (unsigned long long)OCFS2_I(inode)->ip_blkno);
3031 3032
	}
	sync_mapping_buffers(mapping);
3033
	if (blocking == DLM_LOCK_EX) {
3034 3035 3036 3037 3038 3039 3040 3041 3042 3043
		truncate_inode_pages(mapping, 0);
	} else {
		/* We only need to wait on the I/O if we're not also
		 * truncating pages because truncate_inode_pages waits
		 * for us above. We don't truncate pages if we're
		 * blocking anything < EXMODE because we want to keep
		 * them around in that case. */
		filemap_fdatawait(mapping);
	}

3044
out:
3045
	return UNBLOCK_CONTINUE;
3046 3047
}

3048 3049 3050 3051 3052 3053
static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres,
					int new_level)
{
	struct inode *inode = ocfs2_lock_res_inode(lockres);
	int checkpointed = ocfs2_inode_fully_checkpointed(inode);

3054 3055
	BUG_ON(new_level != DLM_LOCK_NL && new_level != DLM_LOCK_PR);
	BUG_ON(lockres->l_level != DLM_LOCK_EX && !checkpointed);
3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070

	if (checkpointed)
		return 1;

	ocfs2_start_checkpoint(OCFS2_SB(inode->i_sb));
	return 0;
}

static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres)
{
	struct inode *inode = ocfs2_lock_res_inode(lockres);

	__ocfs2_stuff_meta_lvb(inode);
}

3071 3072
/*
 * Does the final reference drop on our dentry lock. Right now this
3073
 * happens in the downconvert thread, but we could choose to simplify the
3074 3075 3076 3077 3078 3079 3080 3081 3082 3083 3084 3085 3086 3087 3088 3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114 3115 3116 3117 3118
 * dlmglue API and push these off to the ocfs2_wq in the future.
 */
static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
				     struct ocfs2_lock_res *lockres)
{
	struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
	ocfs2_dentry_lock_put(osb, dl);
}

/*
 * d_delete() matching dentries before the lock downconvert.
 *
 * At this point, any process waiting to destroy the
 * dentry_lock due to last ref count is stopped by the
 * OCFS2_LOCK_QUEUED flag.
 *
 * We have two potential problems
 *
 * 1) If we do the last reference drop on our dentry_lock (via dput)
 *    we'll wind up in ocfs2_release_dentry_lock(), waiting on
 *    the downconvert to finish. Instead we take an elevated
 *    reference and push the drop until after we've completed our
 *    unblock processing.
 *
 * 2) There might be another process with a final reference,
 *    waiting on us to finish processing. If this is the case, we
 *    detect it and exit out - there's no more dentries anyway.
 */
static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres,
				       int blocking)
{
	struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
	struct ocfs2_inode_info *oi = OCFS2_I(dl->dl_inode);
	struct dentry *dentry;
	unsigned long flags;
	int extra_ref = 0;

	/*
	 * This node is blocking another node from getting a read
	 * lock. This happens when we've renamed within a
	 * directory. We've forced the other nodes to d_delete(), but
	 * we never actually dropped our lock because it's still
	 * valid. The downconvert code will retain a PR for this node,
	 * so there's no further work to do.
	 */
3119
	if (blocking == DLM_LOCK_PR)
3120 3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131 3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162 3163 3164 3165 3166 3167 3168 3169 3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185 3186 3187 3188 3189 3190 3191 3192
		return UNBLOCK_CONTINUE;

	/*
	 * Mark this inode as potentially orphaned. The code in
	 * ocfs2_delete_inode() will figure out whether it actually
	 * needs to be freed or not.
	 */
	spin_lock(&oi->ip_lock);
	oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
	spin_unlock(&oi->ip_lock);

	/*
	 * Yuck. We need to make sure however that the check of
	 * OCFS2_LOCK_FREEING and the extra reference are atomic with
	 * respect to a reference decrement or the setting of that
	 * flag.
	 */
	spin_lock_irqsave(&lockres->l_lock, flags);
	spin_lock(&dentry_attach_lock);
	if (!(lockres->l_flags & OCFS2_LOCK_FREEING)
	    && dl->dl_count) {
		dl->dl_count++;
		extra_ref = 1;
	}
	spin_unlock(&dentry_attach_lock);
	spin_unlock_irqrestore(&lockres->l_lock, flags);

	mlog(0, "extra_ref = %d\n", extra_ref);

	/*
	 * We have a process waiting on us in ocfs2_dentry_iput(),
	 * which means we can't have any more outstanding
	 * aliases. There's no need to do any more work.
	 */
	if (!extra_ref)
		return UNBLOCK_CONTINUE;

	spin_lock(&dentry_attach_lock);
	while (1) {
		dentry = ocfs2_find_local_alias(dl->dl_inode,
						dl->dl_parent_blkno, 1);
		if (!dentry)
			break;
		spin_unlock(&dentry_attach_lock);

		mlog(0, "d_delete(%.*s);\n", dentry->d_name.len,
		     dentry->d_name.name);

		/*
		 * The following dcache calls may do an
		 * iput(). Normally we don't want that from the
		 * downconverting thread, but in this case it's ok
		 * because the requesting node already has an
		 * exclusive lock on the inode, so it can't be queued
		 * for a downconvert.
		 */
		d_delete(dentry);
		dput(dentry);

		spin_lock(&dentry_attach_lock);
	}
	spin_unlock(&dentry_attach_lock);

	/*
	 * If we are the last holder of this dentry lock, there is no
	 * reason to downconvert so skip straight to the unlock.
	 */
	if (dl->dl_count == 1)
		return UNBLOCK_STOP_POST;

	return UNBLOCK_CONTINUE_POST;
}

3193 3194 3195 3196 3197 3198 3199 3200 3201 3202 3203 3204 3205 3206 3207 3208 3209
static struct ocfs2_locking_protocol lproto = {
	.lp_lock_ast		= ocfs2_locking_ast,
	.lp_blocking_ast	= ocfs2_blocking_ast,
	.lp_unlock_ast		= ocfs2_unlock_ast,
};

/* This interface isn't the final one, hence the less-than-perfect names */
void dlmglue_init_stack(void)
{
	o2cb_get_stack(&lproto);
}

void dlmglue_exit_stack(void)
{
	o2cb_put_stack();
}

3210 3211
static void ocfs2_process_blocked_lock(struct ocfs2_super *osb,
				       struct ocfs2_lock_res *lockres)
3212 3213
{
	int status;
3214
	struct ocfs2_unblock_ctl ctl = {0, 0,};
3215 3216 3217 3218 3219 3220 3221 3222 3223 3224 3225 3226 3227 3228
	unsigned long flags;

	/* Our reference to the lockres in this function can be
	 * considered valid until we remove the OCFS2_LOCK_QUEUED
	 * flag. */

	mlog_entry_void();

	BUG_ON(!lockres);
	BUG_ON(!lockres->l_ops);

	mlog(0, "lockres %s blocked.\n", lockres->l_name);

	/* Detect whether a lock has been marked as going away while
3229
	 * the downconvert thread was processing other things. A lock can
3230 3231 3232 3233 3234 3235 3236 3237
	 * still be marked with OCFS2_LOCK_FREEING after this check,
	 * but short circuiting here will still save us some
	 * performance. */
	spin_lock_irqsave(&lockres->l_lock, flags);
	if (lockres->l_flags & OCFS2_LOCK_FREEING)
		goto unqueue;
	spin_unlock_irqrestore(&lockres->l_lock, flags);

3238
	status = ocfs2_unblock_lock(osb, lockres, &ctl);
3239 3240 3241 3242 3243
	if (status < 0)
		mlog_errno(status);

	spin_lock_irqsave(&lockres->l_lock, flags);
unqueue:
3244
	if (lockres->l_flags & OCFS2_LOCK_FREEING || !ctl.requeue) {
3245 3246 3247 3248 3249
		lockres_clear_flags(lockres, OCFS2_LOCK_QUEUED);
	} else
		ocfs2_schedule_blocked_lock(osb, lockres);

	mlog(0, "lockres %s, requeue = %s.\n", lockres->l_name,
3250
	     ctl.requeue ? "yes" : "no");
3251 3252
	spin_unlock_irqrestore(&lockres->l_lock, flags);

3253 3254 3255 3256
	if (ctl.unblock_action != UNBLOCK_CONTINUE
	    && lockres->l_ops->post_unlock)
		lockres->l_ops->post_unlock(osb, lockres);

3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270 3271 3272 3273 3274 3275 3276 3277
	mlog_exit_void();
}

static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
					struct ocfs2_lock_res *lockres)
{
	mlog_entry_void();

	assert_spin_locked(&lockres->l_lock);

	if (lockres->l_flags & OCFS2_LOCK_FREEING) {
		/* Do not schedule a lock for downconvert when it's on
		 * the way to destruction - any nodes wanting access
		 * to the resource will get it soon. */
		mlog(0, "Lockres %s won't be scheduled: flags 0x%lx\n",
		     lockres->l_name, lockres->l_flags);
		return;
	}

	lockres_or_flags(lockres, OCFS2_LOCK_QUEUED);

3278
	spin_lock(&osb->dc_task_lock);
3279 3280 3281 3282 3283
	if (list_empty(&lockres->l_blocked_list)) {
		list_add_tail(&lockres->l_blocked_list,
			      &osb->blocked_lock_list);
		osb->blocked_lock_count++;
	}
3284
	spin_unlock(&osb->dc_task_lock);
3285 3286 3287

	mlog_exit_void();
}
3288 3289 3290 3291 3292 3293 3294 3295 3296 3297 3298 3299 3300 3301 3302 3303 3304 3305 3306 3307 3308 3309 3310 3311 3312 3313 3314 3315 3316 3317 3318 3319 3320 3321 3322 3323 3324 3325 3326 3327 3328 3329 3330 3331 3332 3333 3334 3335 3336 3337 3338 3339 3340 3341 3342 3343 3344 3345 3346

static void ocfs2_downconvert_thread_do_work(struct ocfs2_super *osb)
{
	unsigned long processed;
	struct ocfs2_lock_res *lockres;

	mlog_entry_void();

	spin_lock(&osb->dc_task_lock);
	/* grab this early so we know to try again if a state change and
	 * wake happens part-way through our work  */
	osb->dc_work_sequence = osb->dc_wake_sequence;

	processed = osb->blocked_lock_count;
	while (processed) {
		BUG_ON(list_empty(&osb->blocked_lock_list));

		lockres = list_entry(osb->blocked_lock_list.next,
				     struct ocfs2_lock_res, l_blocked_list);
		list_del_init(&lockres->l_blocked_list);
		osb->blocked_lock_count--;
		spin_unlock(&osb->dc_task_lock);

		BUG_ON(!processed);
		processed--;

		ocfs2_process_blocked_lock(osb, lockres);

		spin_lock(&osb->dc_task_lock);
	}
	spin_unlock(&osb->dc_task_lock);

	mlog_exit_void();
}

static int ocfs2_downconvert_thread_lists_empty(struct ocfs2_super *osb)
{
	int empty = 0;

	spin_lock(&osb->dc_task_lock);
	if (list_empty(&osb->blocked_lock_list))
		empty = 1;

	spin_unlock(&osb->dc_task_lock);
	return empty;
}

static int ocfs2_downconvert_thread_should_wake(struct ocfs2_super *osb)
{
	int should_wake = 0;

	spin_lock(&osb->dc_task_lock);
	if (osb->dc_work_sequence != osb->dc_wake_sequence)
		should_wake = 1;
	spin_unlock(&osb->dc_task_lock);

	return should_wake;
}

3347
static int ocfs2_downconvert_thread(void *arg)
3348 3349 3350 3351 3352 3353 3354 3355 3356 3357 3358 3359 3360 3361 3362 3363 3364 3365 3366 3367 3368 3369 3370 3371 3372 3373 3374 3375 3376 3377 3378
{
	int status = 0;
	struct ocfs2_super *osb = arg;

	/* only quit once we've been asked to stop and there is no more
	 * work available */
	while (!(kthread_should_stop() &&
		ocfs2_downconvert_thread_lists_empty(osb))) {

		wait_event_interruptible(osb->dc_event,
					 ocfs2_downconvert_thread_should_wake(osb) ||
					 kthread_should_stop());

		mlog(0, "downconvert_thread: awoken\n");

		ocfs2_downconvert_thread_do_work(osb);
	}

	osb->dc_task = NULL;
	return status;
}

void ocfs2_wake_downconvert_thread(struct ocfs2_super *osb)
{
	spin_lock(&osb->dc_task_lock);
	/* make sure the voting thread gets a swipe at whatever changes
	 * the caller may have made to the voting state */
	osb->dc_wake_sequence++;
	spin_unlock(&osb->dc_task_lock);
	wake_up(&osb->dc_event);
}
反馈
建议
客服 返回
顶部