xref: /openbmc/linux/fs/ocfs2/stack_o2cb.c (revision a06c488d)
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * stack_o2cb.c
5  *
6  * Code which interfaces ocfs2 with the o2cb stack.
7  *
8  * Copyright (C) 2007 Oracle.  All rights reserved.
9  *
10  * This program is free software; you can redistribute it and/or
11  * modify it under the terms of the GNU General Public
12  * License as published by the Free Software Foundation, version 2.
13  *
14  * This program is distributed in the hope that it will be useful,
15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
17  * General Public License for more details.
18  */
19 
20 #include <linux/kernel.h>
21 #include <linux/crc32.h>
22 #include <linux/slab.h>
23 #include <linux/module.h>
24 
25 /* Needed for AOP_TRUNCATED_PAGE in mlog_errno() */
26 #include <linux/fs.h>
27 
28 #include "cluster/masklog.h"
29 #include "cluster/nodemanager.h"
30 #include "cluster/heartbeat.h"
31 #include "cluster/tcp.h"
32 
33 #include "stackglue.h"
34 
35 struct o2dlm_private {
36 	struct dlm_eviction_cb op_eviction_cb;
37 };
38 
39 static struct ocfs2_stack_plugin o2cb_stack;
40 
41 /* These should be identical */
42 #if (DLM_LOCK_IV != LKM_IVMODE)
43 # error Lock modes do not match
44 #endif
45 #if (DLM_LOCK_NL != LKM_NLMODE)
46 # error Lock modes do not match
47 #endif
48 #if (DLM_LOCK_CR != LKM_CRMODE)
49 # error Lock modes do not match
50 #endif
51 #if (DLM_LOCK_CW != LKM_CWMODE)
52 # error Lock modes do not match
53 #endif
54 #if (DLM_LOCK_PR != LKM_PRMODE)
55 # error Lock modes do not match
56 #endif
57 #if (DLM_LOCK_PW != LKM_PWMODE)
58 # error Lock modes do not match
59 #endif
60 #if (DLM_LOCK_EX != LKM_EXMODE)
61 # error Lock modes do not match
62 #endif
63 static inline int mode_to_o2dlm(int mode)
64 {
65 	BUG_ON(mode > LKM_MAXMODE);
66 
67 	return mode;
68 }
69 
70 #define map_flag(_generic, _o2dlm)		\
71 	if (flags & (_generic)) {		\
72 		flags &= ~(_generic);		\
73 		o2dlm_flags |= (_o2dlm);	\
74 	}
75 static int flags_to_o2dlm(u32 flags)
76 {
77 	int o2dlm_flags = 0;
78 
79 	map_flag(DLM_LKF_NOQUEUE, LKM_NOQUEUE);
80 	map_flag(DLM_LKF_CANCEL, LKM_CANCEL);
81 	map_flag(DLM_LKF_CONVERT, LKM_CONVERT);
82 	map_flag(DLM_LKF_VALBLK, LKM_VALBLK);
83 	map_flag(DLM_LKF_IVVALBLK, LKM_INVVALBLK);
84 	map_flag(DLM_LKF_ORPHAN, LKM_ORPHAN);
85 	map_flag(DLM_LKF_FORCEUNLOCK, LKM_FORCE);
86 	map_flag(DLM_LKF_TIMEOUT, LKM_TIMEOUT);
87 	map_flag(DLM_LKF_LOCAL, LKM_LOCAL);
88 
89 	/* map_flag() should have cleared every flag passed in */
90 	BUG_ON(flags != 0);
91 
92 	return o2dlm_flags;
93 }
94 #undef map_flag
95 
96 /*
97  * Map an o2dlm status to standard errno values.
98  *
99  * o2dlm only uses a handful of these, and returns even fewer to the
100  * caller. Still, we try to assign sane values to each error.
101  *
102  * The following value pairs have special meanings to dlmglue, thus
103  * the right hand side needs to stay unique - never duplicate the
104  * mapping elsewhere in the table!
105  *
106  * DLM_NORMAL:		0
107  * DLM_NOTQUEUED:	-EAGAIN
108  * DLM_CANCELGRANT:	-EBUSY
109  * DLM_CANCEL:		-DLM_ECANCEL
110  */
111 /* Keep in sync with dlmapi.h */
112 static int status_map[] = {
113 	[DLM_NORMAL]			= 0,		/* Success */
114 	[DLM_GRANTED]			= -EINVAL,
115 	[DLM_DENIED]			= -EACCES,
116 	[DLM_DENIED_NOLOCKS]		= -EACCES,
117 	[DLM_WORKING]			= -EACCES,
118 	[DLM_BLOCKED]			= -EINVAL,
119 	[DLM_BLOCKED_ORPHAN]		= -EINVAL,
120 	[DLM_DENIED_GRACE_PERIOD]	= -EACCES,
121 	[DLM_SYSERR]			= -ENOMEM,	/* It is what it is */
122 	[DLM_NOSUPPORT]			= -EPROTO,
123 	[DLM_CANCELGRANT]		= -EBUSY,	/* Cancel after grant */
124 	[DLM_IVLOCKID]			= -EINVAL,
125 	[DLM_SYNC]			= -EINVAL,
126 	[DLM_BADTYPE]			= -EINVAL,
127 	[DLM_BADRESOURCE]		= -EINVAL,
128 	[DLM_MAXHANDLES]		= -ENOMEM,
129 	[DLM_NOCLINFO]			= -EINVAL,
130 	[DLM_NOLOCKMGR]			= -EINVAL,
131 	[DLM_NOPURGED]			= -EINVAL,
132 	[DLM_BADARGS]			= -EINVAL,
133 	[DLM_VOID]			= -EINVAL,
134 	[DLM_NOTQUEUED]			= -EAGAIN,	/* Trylock failed */
135 	[DLM_IVBUFLEN]			= -EINVAL,
136 	[DLM_CVTUNGRANT]		= -EPERM,
137 	[DLM_BADPARAM]			= -EINVAL,
138 	[DLM_VALNOTVALID]		= -EINVAL,
139 	[DLM_REJECTED]			= -EPERM,
140 	[DLM_ABORT]			= -EINVAL,
141 	[DLM_CANCEL]			= -DLM_ECANCEL,	/* Successful cancel */
142 	[DLM_IVRESHANDLE]		= -EINVAL,
143 	[DLM_DEADLOCK]			= -EDEADLK,
144 	[DLM_DENIED_NOASTS]		= -EINVAL,
145 	[DLM_FORWARD]			= -EINVAL,
146 	[DLM_TIMEOUT]			= -ETIMEDOUT,
147 	[DLM_IVGROUPID]			= -EINVAL,
148 	[DLM_VERS_CONFLICT]		= -EOPNOTSUPP,
149 	[DLM_BAD_DEVICE_PATH]		= -ENOENT,
150 	[DLM_NO_DEVICE_PERMISSION]	= -EPERM,
151 	[DLM_NO_CONTROL_DEVICE]		= -ENOENT,
152 	[DLM_RECOVERING]		= -ENOTCONN,
153 	[DLM_MIGRATING]			= -ERESTART,
154 	[DLM_MAXSTATS]			= -EINVAL,
155 };
156 
157 static int dlm_status_to_errno(enum dlm_status status)
158 {
159 	BUG_ON(status < 0 || status >= ARRAY_SIZE(status_map));
160 
161 	return status_map[status];
162 }
163 
164 static void o2dlm_lock_ast_wrapper(void *astarg)
165 {
166 	struct ocfs2_dlm_lksb *lksb = astarg;
167 
168 	lksb->lksb_conn->cc_proto->lp_lock_ast(lksb);
169 }
170 
171 static void o2dlm_blocking_ast_wrapper(void *astarg, int level)
172 {
173 	struct ocfs2_dlm_lksb *lksb = astarg;
174 
175 	lksb->lksb_conn->cc_proto->lp_blocking_ast(lksb, level);
176 }
177 
178 static void o2dlm_unlock_ast_wrapper(void *astarg, enum dlm_status status)
179 {
180 	struct ocfs2_dlm_lksb *lksb = astarg;
181 	int error = dlm_status_to_errno(status);
182 
183 	/*
184 	 * In o2dlm, you can get both the lock_ast() for the lock being
185 	 * granted and the unlock_ast() for the CANCEL failing.  A
186 	 * successful cancel sends DLM_NORMAL here.  If the
187 	 * lock grant happened before the cancel arrived, you get
188 	 * DLM_CANCELGRANT.
189 	 *
190 	 * There's no need for the double-ast.  If we see DLM_CANCELGRANT,
191 	 * we just ignore it.  We expect the lock_ast() to handle the
192 	 * granted lock.
193 	 */
194 	if (status == DLM_CANCELGRANT)
195 		return;
196 
197 	lksb->lksb_conn->cc_proto->lp_unlock_ast(lksb, error);
198 }
199 
200 static int o2cb_dlm_lock(struct ocfs2_cluster_connection *conn,
201 			 int mode,
202 			 struct ocfs2_dlm_lksb *lksb,
203 			 u32 flags,
204 			 void *name,
205 			 unsigned int namelen)
206 {
207 	enum dlm_status status;
208 	int o2dlm_mode = mode_to_o2dlm(mode);
209 	int o2dlm_flags = flags_to_o2dlm(flags);
210 	int ret;
211 
212 	status = dlmlock(conn->cc_lockspace, o2dlm_mode, &lksb->lksb_o2dlm,
213 			 o2dlm_flags, name, namelen,
214 			 o2dlm_lock_ast_wrapper, lksb,
215 			 o2dlm_blocking_ast_wrapper);
216 	ret = dlm_status_to_errno(status);
217 	return ret;
218 }
219 
220 static int o2cb_dlm_unlock(struct ocfs2_cluster_connection *conn,
221 			   struct ocfs2_dlm_lksb *lksb,
222 			   u32 flags)
223 {
224 	enum dlm_status status;
225 	int o2dlm_flags = flags_to_o2dlm(flags);
226 	int ret;
227 
228 	status = dlmunlock(conn->cc_lockspace, &lksb->lksb_o2dlm,
229 			   o2dlm_flags, o2dlm_unlock_ast_wrapper, lksb);
230 	ret = dlm_status_to_errno(status);
231 	return ret;
232 }
233 
234 static int o2cb_dlm_lock_status(struct ocfs2_dlm_lksb *lksb)
235 {
236 	return dlm_status_to_errno(lksb->lksb_o2dlm.status);
237 }
238 
239 /*
240  * o2dlm aways has a "valid" LVB. If the dlm loses track of the LVB
241  * contents, it will zero out the LVB.  Thus the caller can always trust
242  * the contents.
243  */
244 static int o2cb_dlm_lvb_valid(struct ocfs2_dlm_lksb *lksb)
245 {
246 	return 1;
247 }
248 
249 static void *o2cb_dlm_lvb(struct ocfs2_dlm_lksb *lksb)
250 {
251 	return (void *)(lksb->lksb_o2dlm.lvb);
252 }
253 
254 static void o2cb_dump_lksb(struct ocfs2_dlm_lksb *lksb)
255 {
256 	dlm_print_one_lock(lksb->lksb_o2dlm.lockid);
257 }
258 
259 /*
260  * Check if this node is heartbeating and is connected to all other
261  * heartbeating nodes.
262  */
263 static int o2cb_cluster_check(void)
264 {
265 	u8 node_num;
266 	int i;
267 	unsigned long hbmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
268 	unsigned long netmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
269 
270 	node_num = o2nm_this_node();
271 	if (node_num == O2NM_MAX_NODES) {
272 		printk(KERN_ERR "o2cb: This node has not been configured.\n");
273 		return -EINVAL;
274 	}
275 
276 	/*
277 	 * o2dlm expects o2net sockets to be created. If not, then
278 	 * dlm_join_domain() fails with a stack of errors which are both cryptic
279 	 * and incomplete. The idea here is to detect upfront whether we have
280 	 * managed to connect to all nodes or not. If not, then list the nodes
281 	 * to allow the user to check the configuration (incorrect IP, firewall,
282 	 * etc.) Yes, this is racy. But its not the end of the world.
283 	 */
284 #define	O2CB_MAP_STABILIZE_COUNT	60
285 	for (i = 0; i < O2CB_MAP_STABILIZE_COUNT; ++i) {
286 		o2hb_fill_node_map(hbmap, sizeof(hbmap));
287 		if (!test_bit(node_num, hbmap)) {
288 			printk(KERN_ERR "o2cb: %s heartbeat has not been "
289 			       "started.\n", (o2hb_global_heartbeat_active() ?
290 					      "Global" : "Local"));
291 			return -EINVAL;
292 		}
293 		o2net_fill_node_map(netmap, sizeof(netmap));
294 		/* Force set the current node to allow easy compare */
295 		set_bit(node_num, netmap);
296 		if (!memcmp(hbmap, netmap, sizeof(hbmap)))
297 			return 0;
298 		if (i < O2CB_MAP_STABILIZE_COUNT - 1)
299 			msleep(1000);
300 	}
301 
302 	printk(KERN_ERR "o2cb: This node could not connect to nodes:");
303 	i = -1;
304 	while ((i = find_next_bit(hbmap, O2NM_MAX_NODES,
305 				  i + 1)) < O2NM_MAX_NODES) {
306 		if (!test_bit(i, netmap))
307 			printk(" %u", i);
308 	}
309 	printk(".\n");
310 
311 	return -ENOTCONN;
312 }
313 
314 /*
315  * Called from the dlm when it's about to evict a node. This is how the
316  * classic stack signals node death.
317  */
318 static void o2dlm_eviction_cb(int node_num, void *data)
319 {
320 	struct ocfs2_cluster_connection *conn = data;
321 
322 	printk(KERN_NOTICE "o2cb: o2dlm has evicted node %d from domain %.*s\n",
323 	       node_num, conn->cc_namelen, conn->cc_name);
324 
325 	conn->cc_recovery_handler(node_num, conn->cc_recovery_data);
326 }
327 
328 static int o2cb_cluster_connect(struct ocfs2_cluster_connection *conn)
329 {
330 	int rc = 0;
331 	u32 dlm_key;
332 	struct dlm_ctxt *dlm;
333 	struct o2dlm_private *priv;
334 	struct dlm_protocol_version fs_version;
335 
336 	BUG_ON(conn == NULL);
337 	BUG_ON(conn->cc_proto == NULL);
338 
339 	/* Ensure cluster stack is up and all nodes are connected */
340 	rc = o2cb_cluster_check();
341 	if (rc) {
342 		printk(KERN_ERR "o2cb: Cluster check failed. Fix errors "
343 		       "before retrying.\n");
344 		goto out;
345 	}
346 
347 	priv = kzalloc(sizeof(struct o2dlm_private), GFP_KERNEL);
348 	if (!priv) {
349 		rc = -ENOMEM;
350 		goto out_free;
351 	}
352 
353 	/* This just fills the structure in.  It is safe to pass conn. */
354 	dlm_setup_eviction_cb(&priv->op_eviction_cb, o2dlm_eviction_cb,
355 			      conn);
356 
357 	conn->cc_private = priv;
358 
359 	/* used by the dlm code to make message headers unique, each
360 	 * node in this domain must agree on this. */
361 	dlm_key = crc32_le(0, conn->cc_name, conn->cc_namelen);
362 	fs_version.pv_major = conn->cc_version.pv_major;
363 	fs_version.pv_minor = conn->cc_version.pv_minor;
364 
365 	dlm = dlm_register_domain(conn->cc_name, dlm_key, &fs_version);
366 	if (IS_ERR(dlm)) {
367 		rc = PTR_ERR(dlm);
368 		mlog_errno(rc);
369 		goto out_free;
370 	}
371 
372 	conn->cc_version.pv_major = fs_version.pv_major;
373 	conn->cc_version.pv_minor = fs_version.pv_minor;
374 	conn->cc_lockspace = dlm;
375 
376 	dlm_register_eviction_cb(dlm, &priv->op_eviction_cb);
377 
378 out_free:
379 	if (rc)
380 		kfree(conn->cc_private);
381 
382 out:
383 	return rc;
384 }
385 
386 static int o2cb_cluster_disconnect(struct ocfs2_cluster_connection *conn)
387 {
388 	struct dlm_ctxt *dlm = conn->cc_lockspace;
389 	struct o2dlm_private *priv = conn->cc_private;
390 
391 	dlm_unregister_eviction_cb(&priv->op_eviction_cb);
392 	conn->cc_private = NULL;
393 	kfree(priv);
394 
395 	dlm_unregister_domain(dlm);
396 	conn->cc_lockspace = NULL;
397 
398 	return 0;
399 }
400 
401 static int o2cb_cluster_this_node(struct ocfs2_cluster_connection *conn,
402 				  unsigned int *node)
403 {
404 	int node_num;
405 
406 	node_num = o2nm_this_node();
407 	if (node_num == O2NM_INVALID_NODE_NUM)
408 		return -ENOENT;
409 
410 	if (node_num >= O2NM_MAX_NODES)
411 		return -EOVERFLOW;
412 
413 	*node = node_num;
414 	return 0;
415 }
416 
417 static struct ocfs2_stack_operations o2cb_stack_ops = {
418 	.connect	= o2cb_cluster_connect,
419 	.disconnect	= o2cb_cluster_disconnect,
420 	.this_node	= o2cb_cluster_this_node,
421 	.dlm_lock	= o2cb_dlm_lock,
422 	.dlm_unlock	= o2cb_dlm_unlock,
423 	.lock_status	= o2cb_dlm_lock_status,
424 	.lvb_valid	= o2cb_dlm_lvb_valid,
425 	.lock_lvb	= o2cb_dlm_lvb,
426 	.dump_lksb	= o2cb_dump_lksb,
427 };
428 
429 static struct ocfs2_stack_plugin o2cb_stack = {
430 	.sp_name	= "o2cb",
431 	.sp_ops		= &o2cb_stack_ops,
432 	.sp_owner	= THIS_MODULE,
433 };
434 
435 static int __init o2cb_stack_init(void)
436 {
437 	return ocfs2_stack_glue_register(&o2cb_stack);
438 }
439 
440 static void __exit o2cb_stack_exit(void)
441 {
442 	ocfs2_stack_glue_unregister(&o2cb_stack);
443 }
444 
445 MODULE_AUTHOR("Oracle");
446 MODULE_DESCRIPTION("ocfs2 driver for the classic o2cb stack");
447 MODULE_LICENSE("GPL");
448 module_init(o2cb_stack_init);
449 module_exit(o2cb_stack_exit);
450