stackglue.c 12.2 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
/* -*- mode: c; c-basic-offset: 8; -*-
 * vim: noexpandtab sw=8 ts=8 sts=0:
 *
 * stackglue.c
 *
 * Code which implements an OCFS2 specific interface to underlying
 * cluster stacks.
 *
 * Copyright (C) 2007 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, version 2.
 *
 * 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.
 */

21 22
#include <linux/slab.h>
#include <linux/crc32.h>
23
#include <linux/kmod.h>
24 25 26 27

/* Needed for AOP_TRUNCATED_PAGE in mlog_errno() */
#include <linux/fs.h>

28
#include "cluster/masklog.h"
29
#include "cluster/nodemanager.h"
30
#include "cluster/heartbeat.h"
31

32 33 34 35
#include "stackglue.h"

static struct ocfs2_locking_protocol *lproto;

36 37 38 39
struct o2dlm_private {
	struct dlm_eviction_cb op_eviction_cb;
};

40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94
/* These should be identical */
#if (DLM_LOCK_IV != LKM_IVMODE)
# error Lock modes do not match
#endif
#if (DLM_LOCK_NL != LKM_NLMODE)
# error Lock modes do not match
#endif
#if (DLM_LOCK_CR != LKM_CRMODE)
# error Lock modes do not match
#endif
#if (DLM_LOCK_CW != LKM_CWMODE)
# error Lock modes do not match
#endif
#if (DLM_LOCK_PR != LKM_PRMODE)
# error Lock modes do not match
#endif
#if (DLM_LOCK_PW != LKM_PWMODE)
# error Lock modes do not match
#endif
#if (DLM_LOCK_EX != LKM_EXMODE)
# error Lock modes do not match
#endif
static inline int mode_to_o2dlm(int mode)
{
	BUG_ON(mode > LKM_MAXMODE);

	return mode;
}

#define map_flag(_generic, _o2dlm)		\
	if (flags & (_generic)) {		\
		flags &= ~(_generic);		\
		o2dlm_flags |= (_o2dlm);	\
	}
static int flags_to_o2dlm(u32 flags)
{
	int o2dlm_flags = 0;

	map_flag(DLM_LKF_NOQUEUE, LKM_NOQUEUE);
	map_flag(DLM_LKF_CANCEL, LKM_CANCEL);
	map_flag(DLM_LKF_CONVERT, LKM_CONVERT);
	map_flag(DLM_LKF_VALBLK, LKM_VALBLK);
	map_flag(DLM_LKF_IVVALBLK, LKM_INVVALBLK);
	map_flag(DLM_LKF_ORPHAN, LKM_ORPHAN);
	map_flag(DLM_LKF_FORCEUNLOCK, LKM_FORCE);
	map_flag(DLM_LKF_TIMEOUT, LKM_TIMEOUT);
	map_flag(DLM_LKF_LOCAL, LKM_LOCAL);

	/* map_flag() should have cleared every flag passed in */
	BUG_ON(flags != 0);

	return o2dlm_flags;
}
#undef map_flag

95 96 97 98 99 100 101 102 103 104 105 106
/*
 * Map an o2dlm status to standard errno values.
 *
 * o2dlm only uses a handful of these, and returns even fewer to the
 * caller. Still, we try to assign sane values to each error.
 *
 * The following value pairs have special meanings to dlmglue, thus
 * the right hand side needs to stay unique - never duplicate the
 * mapping elsewhere in the table!
 *
 * DLM_NORMAL:		0
 * DLM_NOTQUEUED:	-EAGAIN
107 108
 * DLM_CANCELGRANT:	-EBUSY
 * DLM_CANCEL:		-DLM_ECANCEL
109 110 111 112 113 114 115
 */
/* Keep in sync with dlmapi.h */
static int status_map[] = {
	[DLM_NORMAL]			= 0,		/* Success */
	[DLM_GRANTED]			= -EINVAL,
	[DLM_DENIED]			= -EACCES,
	[DLM_DENIED_NOLOCKS]		= -EACCES,
116
	[DLM_WORKING]			= -EACCES,
117 118 119 120 121
	[DLM_BLOCKED]			= -EINVAL,
	[DLM_BLOCKED_ORPHAN]		= -EINVAL,
	[DLM_DENIED_GRACE_PERIOD]	= -EACCES,
	[DLM_SYSERR]			= -ENOMEM,	/* It is what it is */
	[DLM_NOSUPPORT]			= -EPROTO,
122
	[DLM_CANCELGRANT]		= -EBUSY,	/* Cancel after grant */
123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139
	[DLM_IVLOCKID]			= -EINVAL,
	[DLM_SYNC]			= -EINVAL,
	[DLM_BADTYPE]			= -EINVAL,
	[DLM_BADRESOURCE]		= -EINVAL,
	[DLM_MAXHANDLES]		= -ENOMEM,
	[DLM_NOCLINFO]			= -EINVAL,
	[DLM_NOLOCKMGR]			= -EINVAL,
	[DLM_NOPURGED]			= -EINVAL,
	[DLM_BADARGS]			= -EINVAL,
	[DLM_VOID]			= -EINVAL,
	[DLM_NOTQUEUED]			= -EAGAIN,	/* Trylock failed */
	[DLM_IVBUFLEN]			= -EINVAL,
	[DLM_CVTUNGRANT]		= -EPERM,
	[DLM_BADPARAM]			= -EINVAL,
	[DLM_VALNOTVALID]		= -EINVAL,
	[DLM_REJECTED]			= -EPERM,
	[DLM_ABORT]			= -EINVAL,
140
	[DLM_CANCEL]			= -DLM_ECANCEL,	/* Successful cancel */
141 142 143 144 145 146 147 148 149 150 151 152 153 154
	[DLM_IVRESHANDLE]		= -EINVAL,
	[DLM_DEADLOCK]			= -EDEADLK,
	[DLM_DENIED_NOASTS]		= -EINVAL,
	[DLM_FORWARD]			= -EINVAL,
	[DLM_TIMEOUT]			= -ETIMEDOUT,
	[DLM_IVGROUPID]			= -EINVAL,
	[DLM_VERS_CONFLICT]		= -EOPNOTSUPP,
	[DLM_BAD_DEVICE_PATH]		= -ENOENT,
	[DLM_NO_DEVICE_PERMISSION]	= -EPERM,
	[DLM_NO_CONTROL_DEVICE]		= -ENOENT,
	[DLM_RECOVERING]		= -ENOTCONN,
	[DLM_MIGRATING]			= -ERESTART,
	[DLM_MAXSTATS]			= -EINVAL,
};
155

156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178
static int dlm_status_to_errno(enum dlm_status status)
{
	BUG_ON(status > (sizeof(status_map) / sizeof(status_map[0])));

	return status_map[status];
}

static void o2dlm_lock_ast_wrapper(void *astarg)
{
	BUG_ON(lproto == NULL);

	lproto->lp_lock_ast(astarg);
}

static void o2dlm_blocking_ast_wrapper(void *astarg, int level)
{
	BUG_ON(lproto == NULL);

	lproto->lp_blocking_ast(astarg, level);
}

static void o2dlm_unlock_ast_wrapper(void *astarg, enum dlm_status status)
{
179
	int error = dlm_status_to_errno(status);
180 181 182 183 184 185 186 187

	BUG_ON(lproto == NULL);

	/*
	 * In o2dlm, you can get both the lock_ast() for the lock being
	 * granted and the unlock_ast() for the CANCEL failing.  A
	 * successful cancel sends DLM_NORMAL here.  If the
	 * lock grant happened before the cancel arrived, you get
188
	 * DLM_CANCELGRANT.
189
	 *
190 191 192
	 * There's no need for the double-ast.  If we see DLM_CANCELGRANT,
	 * we just ignore it.  We expect the lock_ast() to handle the
	 * granted lock.
193
	 */
194 195
	if (status == DLM_CANCELGRANT)
		return;
196 197 198 199

	lproto->lp_unlock_ast(astarg, error);
}

200 201 202 203 204 205 206
static int o2cb_dlm_lock(struct ocfs2_cluster_connection *conn,
			 int mode,
			 union ocfs2_dlm_lksb *lksb,
			 u32 flags,
			 void *name,
			 unsigned int namelen,
			 void *astarg)
207
{
208
	enum dlm_status status;
209 210
	int o2dlm_mode = mode_to_o2dlm(mode);
	int o2dlm_flags = flags_to_o2dlm(flags);
211
	int ret;
212

213 214
	status = dlmlock(conn->cc_lockspace, o2dlm_mode, &lksb->lksb_o2dlm,
			 o2dlm_flags, name, namelen,
215 216
			 o2dlm_lock_ast_wrapper, astarg,
			 o2dlm_blocking_ast_wrapper);
217 218
	ret = dlm_status_to_errno(status);
	return ret;
219 220
}

221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238
int ocfs2_dlm_lock(struct ocfs2_cluster_connection *conn,
		   int mode,
		   union ocfs2_dlm_lksb *lksb,
		   u32 flags,
		   void *name,
		   unsigned int namelen,
		   void *astarg)
{
	BUG_ON(lproto == NULL);

	return o2cb_dlm_lock(conn, mode, lksb, flags,
			     name, namelen, astarg);
}

static int o2cb_dlm_unlock(struct ocfs2_cluster_connection *conn,
			   union ocfs2_dlm_lksb *lksb,
			   u32 flags,
			   void *astarg)
239
{
240
	enum dlm_status status;
241
	int o2dlm_flags = flags_to_o2dlm(flags);
242
	int ret;
243

244 245
	status = dlmunlock(conn->cc_lockspace, &lksb->lksb_o2dlm,
			   o2dlm_flags, o2dlm_unlock_ast_wrapper, astarg);
246 247
	ret = dlm_status_to_errno(status);
	return ret;
248 249
}

250 251 252 253 254 255 256 257 258 259 260
int ocfs2_dlm_unlock(struct ocfs2_cluster_connection *conn,
		     union ocfs2_dlm_lksb *lksb,
		     u32 flags,
		     void *astarg)
{
	BUG_ON(lproto == NULL);

	return o2cb_dlm_unlock(conn, lksb, flags, astarg);
}

static int o2cb_dlm_lock_status(union ocfs2_dlm_lksb *lksb)
261 262 263 264
{
	return dlm_status_to_errno(lksb->lksb_o2dlm.status);
}

265 266 267 268 269
int ocfs2_dlm_lock_status(union ocfs2_dlm_lksb *lksb)
{
	return o2cb_dlm_lock_status(lksb);
}

270 271 272 273 274
/*
 * Why don't we cast to ocfs2_meta_lvb?  The "clean" answer is that we
 * don't cast at the glue level.  The real answer is that the header
 * ordering is nigh impossible.
 */
275
static void *o2cb_dlm_lvb(union ocfs2_dlm_lksb *lksb)
276 277 278
{
	return (void *)(lksb->lksb_o2dlm.lvb);
}
279

280 281 282 283 284 285
void *ocfs2_dlm_lvb(union ocfs2_dlm_lksb *lksb)
{
	return o2cb_dlm_lvb(lksb);
}

static void o2cb_dlm_dump_lksb(union ocfs2_dlm_lksb *lksb)
286 287 288 289
{
	dlm_print_one_lock(lksb->lksb_o2dlm.lockid);
}

290 291 292 293 294
void ocfs2_dlm_dump_lksb(union ocfs2_dlm_lksb *lksb)
{
	o2cb_dlm_dump_lksb(lksb);
}

295 296 297 298 299 300 301 302 303 304 305 306 307 308
/*
 * Called from the dlm when it's about to evict a node. This is how the
 * classic stack signals node death.
 */
static void o2dlm_eviction_cb(int node_num, void *data)
{
	struct ocfs2_cluster_connection *conn = data;

	mlog(ML_NOTICE, "o2dlm has evicted node %d from group %.*s\n",
	     node_num, conn->cc_namelen, conn->cc_name);

	conn->cc_recovery_handler(node_num, conn->cc_recovery_data);
}

309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364
static int o2cb_cluster_connect(struct ocfs2_cluster_connection *conn)
{
	int rc = 0;
	u32 dlm_key;
	struct dlm_ctxt *dlm;
	struct o2dlm_private *priv;
	struct dlm_protocol_version dlm_version;

	BUG_ON(conn == NULL);

	/* for now we only have one cluster/node, make sure we see it
	 * in the heartbeat universe */
	if (!o2hb_check_local_node_heartbeating()) {
		rc = -EINVAL;
		goto out;
	}

	priv = kzalloc(sizeof(struct o2dlm_private), GFP_KERNEL);
	if (!priv) {
		rc = -ENOMEM;
		goto out_free;
	}

	/* This just fills the structure in.  It is safe to pass conn. */
	dlm_setup_eviction_cb(&priv->op_eviction_cb, o2dlm_eviction_cb,
			      conn);

	conn->cc_private = priv;

	/* used by the dlm code to make message headers unique, each
	 * node in this domain must agree on this. */
	dlm_key = crc32_le(0, conn->cc_name, conn->cc_namelen);
	dlm_version.pv_major = conn->cc_version.pv_major;
	dlm_version.pv_minor = conn->cc_version.pv_minor;

	dlm = dlm_register_domain(conn->cc_name, dlm_key, &dlm_version);
	if (IS_ERR(dlm)) {
		rc = PTR_ERR(dlm);
		mlog_errno(rc);
		goto out_free;
	}

	conn->cc_version.pv_major = dlm_version.pv_major;
	conn->cc_version.pv_minor = dlm_version.pv_minor;
	conn->cc_lockspace = dlm;

	dlm_register_eviction_cb(dlm, &priv->op_eviction_cb);

out_free:
	if (rc && conn->cc_private)
		kfree(conn->cc_private);

out:
	return rc;
}

365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398
int ocfs2_cluster_connect(const char *group,
			  int grouplen,
			  void (*recovery_handler)(int node_num,
						   void *recovery_data),
			  void *recovery_data,
			  struct ocfs2_cluster_connection **conn)
{
	int rc = 0;
	struct ocfs2_cluster_connection *new_conn;

	BUG_ON(group == NULL);
	BUG_ON(conn == NULL);
	BUG_ON(recovery_handler == NULL);

	if (grouplen > GROUP_NAME_MAX) {
		rc = -EINVAL;
		goto out;
	}

	new_conn = kzalloc(sizeof(struct ocfs2_cluster_connection),
			   GFP_KERNEL);
	if (!new_conn) {
		rc = -ENOMEM;
		goto out;
	}

	memcpy(new_conn->cc_name, group, grouplen);
	new_conn->cc_namelen = grouplen;
	new_conn->cc_recovery_handler = recovery_handler;
	new_conn->cc_recovery_data = recovery_data;

	/* Start the new connection at our maximum compatibility level */
	new_conn->cc_version = lproto->lp_max_version;

399 400
	rc = o2cb_cluster_connect(new_conn);
	if (rc) {
401 402 403 404 405 406 407
		mlog_errno(rc);
		goto out_free;
	}

	*conn = new_conn;

out_free:
408
	if (rc)
409 410 411 412 413 414
		kfree(new_conn);

out:
	return rc;
}

415

416
static int o2cb_cluster_disconnect(struct ocfs2_cluster_connection *conn)
417 418 419 420 421
{
	struct dlm_ctxt *dlm = conn->cc_lockspace;
	struct o2dlm_private *priv = conn->cc_private;

	dlm_unregister_eviction_cb(&priv->op_eviction_cb);
422
	conn->cc_private = NULL;
423
	kfree(priv);
424 425 426

	dlm_unregister_domain(dlm);
	conn->cc_lockspace = NULL;
427 428 429 430

	return 0;
}

431 432 433 434 435 436 437 438 439 440 441 442 443 444 445
int ocfs2_cluster_disconnect(struct ocfs2_cluster_connection *conn)
{
	int ret;

	BUG_ON(conn == NULL);

	ret = o2cb_cluster_disconnect(conn);

	/* XXX Should we free it anyway? */
	if (!ret)
		kfree(conn);

	return ret;
}

446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477
static void o2hb_stop(const char *group)
{
	int ret;
	char *argv[5], *envp[3];

	argv[0] = (char *)o2nm_get_hb_ctl_path();
	argv[1] = "-K";
	argv[2] = "-u";
	argv[3] = (char *)group;
	argv[4] = NULL;

	mlog(0, "Run: %s %s %s %s\n", argv[0], argv[1], argv[2], argv[3]);

	/* minimal command environment taken from cpu_run_sbin_hotplug */
	envp[0] = "HOME=/";
	envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
	envp[2] = NULL;

	ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
	if (ret < 0)
		mlog_errno(ret);
}

/*
 * Hangup is a hack for tools compatibility.  Older ocfs2-tools software
 * expects the filesystem to call "ocfs2_hb_ctl" during unmount.  This
 * happens regardless of whether the DLM got started, so we can't do it
 * in ocfs2_cluster_disconnect().  We bring the o2hb_stop() function into
 * the glue and provide a "hangup" API for super.c to call.
 *
 * Other stacks will eventually provide a NULL ->hangup() pointer.
 */
478 479 480 481 482
static void o2cb_cluster_hangup(const char *group, int grouplen)
{
	o2hb_stop(group);
}

483 484 485 486 487
void ocfs2_cluster_hangup(const char *group, int grouplen)
{
	BUG_ON(group == NULL);
	BUG_ON(group[grouplen] != '\0');

488
	o2cb_cluster_hangup(group, grouplen);
489 490
}

491
static int o2cb_cluster_this_node(unsigned int *node)
492 493 494 495 496 497 498 499 500 501 502 503 504 505
{
	int node_num;

	node_num = o2nm_this_node();
	if (node_num == O2NM_INVALID_NODE_NUM)
		return -ENOENT;

	if (node_num >= O2NM_MAX_NODES)
		return -EOVERFLOW;

	*node = node_num;
	return 0;
}

506 507 508 509 510
int ocfs2_cluster_this_node(unsigned int *node)
{
	return o2cb_cluster_this_node(node);
}

511
void ocfs2_stack_glue_set_locking_protocol(struct ocfs2_locking_protocol *proto)
512
{
513
	BUG_ON(proto != NULL);
514 515 516 517

	lproto = proto;
}