blob: bd805411a8566f035b0c4e983e8018748eec08f6 [file] [log] [blame]
Joel Becker24ef1812008-01-29 17:37:32 -08001/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * stackglue.c
5 *
6 * Code which implements an OCFS2 specific interface to underlying
7 * cluster stacks.
8 *
9 * Copyright (C) 2007 Oracle. All rights reserved.
10 *
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public
13 * License as published by the Free Software Foundation, version 2.
14 *
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
19 */
20
Joel Becker4670c462008-02-01 14:39:35 -080021#include <linux/slab.h>
22#include <linux/crc32.h>
Joel Becker6953b4c2008-01-29 16:59:56 -080023#include <linux/kmod.h>
Joel Becker4670c462008-02-01 14:39:35 -080024
25/* Needed for AOP_TRUNCATED_PAGE in mlog_errno() */
26#include <linux/fs.h>
27
Joel Becker7431cd72008-02-01 12:15:37 -080028#include "cluster/masklog.h"
Joel Becker19fdb622008-01-30 15:38:24 -080029#include "cluster/nodemanager.h"
Joel Becker6953b4c2008-01-29 16:59:56 -080030#include "cluster/heartbeat.h"
Joel Becker19fdb622008-01-30 15:38:24 -080031
Joel Becker24ef1812008-01-29 17:37:32 -080032#include "stackglue.h"
33
34static struct ocfs2_locking_protocol *lproto;
35
Joel Becker4670c462008-02-01 14:39:35 -080036struct o2dlm_private {
37 struct dlm_eviction_cb op_eviction_cb;
38};
39
Joel Beckerbd3e7612008-02-01 12:14:57 -080040/* These should be identical */
41#if (DLM_LOCK_IV != LKM_IVMODE)
42# error Lock modes do not match
43#endif
44#if (DLM_LOCK_NL != LKM_NLMODE)
45# error Lock modes do not match
46#endif
47#if (DLM_LOCK_CR != LKM_CRMODE)
48# error Lock modes do not match
49#endif
50#if (DLM_LOCK_CW != LKM_CWMODE)
51# error Lock modes do not match
52#endif
53#if (DLM_LOCK_PR != LKM_PRMODE)
54# error Lock modes do not match
55#endif
56#if (DLM_LOCK_PW != LKM_PWMODE)
57# error Lock modes do not match
58#endif
59#if (DLM_LOCK_EX != LKM_EXMODE)
60# error Lock modes do not match
61#endif
62static inline int mode_to_o2dlm(int mode)
63{
64 BUG_ON(mode > LKM_MAXMODE);
65
66 return mode;
67}
68
69#define map_flag(_generic, _o2dlm) \
70 if (flags & (_generic)) { \
71 flags &= ~(_generic); \
72 o2dlm_flags |= (_o2dlm); \
73 }
74static int flags_to_o2dlm(u32 flags)
75{
76 int o2dlm_flags = 0;
77
78 map_flag(DLM_LKF_NOQUEUE, LKM_NOQUEUE);
79 map_flag(DLM_LKF_CANCEL, LKM_CANCEL);
80 map_flag(DLM_LKF_CONVERT, LKM_CONVERT);
81 map_flag(DLM_LKF_VALBLK, LKM_VALBLK);
82 map_flag(DLM_LKF_IVVALBLK, LKM_INVVALBLK);
83 map_flag(DLM_LKF_ORPHAN, LKM_ORPHAN);
84 map_flag(DLM_LKF_FORCEUNLOCK, LKM_FORCE);
85 map_flag(DLM_LKF_TIMEOUT, LKM_TIMEOUT);
86 map_flag(DLM_LKF_LOCAL, LKM_LOCAL);
87
88 /* map_flag() should have cleared every flag passed in */
89 BUG_ON(flags != 0);
90
91 return o2dlm_flags;
92}
93#undef map_flag
94
Joel Becker7431cd72008-02-01 12:15:37 -080095/*
96 * Map an o2dlm status to standard errno values.
97 *
98 * o2dlm only uses a handful of these, and returns even fewer to the
99 * caller. Still, we try to assign sane values to each error.
100 *
101 * The following value pairs have special meanings to dlmglue, thus
102 * the right hand side needs to stay unique - never duplicate the
103 * mapping elsewhere in the table!
104 *
105 * DLM_NORMAL: 0
106 * DLM_NOTQUEUED: -EAGAIN
Joel Beckerde551242008-02-01 14:45:08 -0800107 * DLM_CANCELGRANT: -EBUSY
108 * DLM_CANCEL: -DLM_ECANCEL
Joel Becker7431cd72008-02-01 12:15:37 -0800109 */
110/* Keep in sync with dlmapi.h */
111static int status_map[] = {
112 [DLM_NORMAL] = 0, /* Success */
113 [DLM_GRANTED] = -EINVAL,
114 [DLM_DENIED] = -EACCES,
115 [DLM_DENIED_NOLOCKS] = -EACCES,
Joel Beckerde551242008-02-01 14:45:08 -0800116 [DLM_WORKING] = -EACCES,
Joel Becker7431cd72008-02-01 12:15:37 -0800117 [DLM_BLOCKED] = -EINVAL,
118 [DLM_BLOCKED_ORPHAN] = -EINVAL,
119 [DLM_DENIED_GRACE_PERIOD] = -EACCES,
120 [DLM_SYSERR] = -ENOMEM, /* It is what it is */
121 [DLM_NOSUPPORT] = -EPROTO,
Joel Beckerde551242008-02-01 14:45:08 -0800122 [DLM_CANCELGRANT] = -EBUSY, /* Cancel after grant */
Joel Becker7431cd72008-02-01 12:15:37 -0800123 [DLM_IVLOCKID] = -EINVAL,
124 [DLM_SYNC] = -EINVAL,
125 [DLM_BADTYPE] = -EINVAL,
126 [DLM_BADRESOURCE] = -EINVAL,
127 [DLM_MAXHANDLES] = -ENOMEM,
128 [DLM_NOCLINFO] = -EINVAL,
129 [DLM_NOLOCKMGR] = -EINVAL,
130 [DLM_NOPURGED] = -EINVAL,
131 [DLM_BADARGS] = -EINVAL,
132 [DLM_VOID] = -EINVAL,
133 [DLM_NOTQUEUED] = -EAGAIN, /* Trylock failed */
134 [DLM_IVBUFLEN] = -EINVAL,
135 [DLM_CVTUNGRANT] = -EPERM,
136 [DLM_BADPARAM] = -EINVAL,
137 [DLM_VALNOTVALID] = -EINVAL,
138 [DLM_REJECTED] = -EPERM,
139 [DLM_ABORT] = -EINVAL,
Joel Beckerde551242008-02-01 14:45:08 -0800140 [DLM_CANCEL] = -DLM_ECANCEL, /* Successful cancel */
Joel Becker7431cd72008-02-01 12:15:37 -0800141 [DLM_IVRESHANDLE] = -EINVAL,
142 [DLM_DEADLOCK] = -EDEADLK,
143 [DLM_DENIED_NOASTS] = -EINVAL,
144 [DLM_FORWARD] = -EINVAL,
145 [DLM_TIMEOUT] = -ETIMEDOUT,
146 [DLM_IVGROUPID] = -EINVAL,
147 [DLM_VERS_CONFLICT] = -EOPNOTSUPP,
148 [DLM_BAD_DEVICE_PATH] = -ENOENT,
149 [DLM_NO_DEVICE_PERMISSION] = -EPERM,
150 [DLM_NO_CONTROL_DEVICE] = -ENOENT,
151 [DLM_RECOVERING] = -ENOTCONN,
152 [DLM_MIGRATING] = -ERESTART,
153 [DLM_MAXSTATS] = -EINVAL,
154};
Joel Beckerde551242008-02-01 14:45:08 -0800155
Joel Becker7431cd72008-02-01 12:15:37 -0800156static int dlm_status_to_errno(enum dlm_status status)
157{
158 BUG_ON(status > (sizeof(status_map) / sizeof(status_map[0])));
159
160 return status_map[status];
161}
162
163static void o2dlm_lock_ast_wrapper(void *astarg)
164{
165 BUG_ON(lproto == NULL);
166
167 lproto->lp_lock_ast(astarg);
168}
169
170static void o2dlm_blocking_ast_wrapper(void *astarg, int level)
171{
172 BUG_ON(lproto == NULL);
173
174 lproto->lp_blocking_ast(astarg, level);
175}
176
177static void o2dlm_unlock_ast_wrapper(void *astarg, enum dlm_status status)
178{
Joel Beckerde551242008-02-01 14:45:08 -0800179 int error = dlm_status_to_errno(status);
Joel Becker7431cd72008-02-01 12:15:37 -0800180
181 BUG_ON(lproto == NULL);
182
183 /*
Joel Becker7431cd72008-02-01 12:15:37 -0800184 * 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
Joel Beckerde551242008-02-01 14:45:08 -0800188 * DLM_CANCELGRANT.
Joel Becker7431cd72008-02-01 12:15:37 -0800189 *
Joel Beckerde551242008-02-01 14:45:08 -0800190 * 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.
Joel Becker7431cd72008-02-01 12:15:37 -0800193 */
Joel Beckerde551242008-02-01 14:45:08 -0800194 if (status == DLM_CANCELGRANT)
195 return;
Joel Becker7431cd72008-02-01 12:15:37 -0800196
197 lproto->lp_unlock_ast(astarg, error);
198}
199
Joel Becker4670c462008-02-01 14:39:35 -0800200int ocfs2_dlm_lock(struct ocfs2_cluster_connection *conn,
Joel Becker24ef1812008-01-29 17:37:32 -0800201 int mode,
Joel Becker8f2c9c12008-02-01 12:16:57 -0800202 union ocfs2_dlm_lksb *lksb,
Joel Becker24ef1812008-01-29 17:37:32 -0800203 u32 flags,
204 void *name,
205 unsigned int namelen,
206 void *astarg)
207{
Joel Becker7431cd72008-02-01 12:15:37 -0800208 enum dlm_status status;
Joel Beckerbd3e7612008-02-01 12:14:57 -0800209 int o2dlm_mode = mode_to_o2dlm(mode);
210 int o2dlm_flags = flags_to_o2dlm(flags);
Joel Becker7431cd72008-02-01 12:15:37 -0800211 int ret;
Joel Beckerbd3e7612008-02-01 12:14:57 -0800212
Joel Becker24ef1812008-01-29 17:37:32 -0800213 BUG_ON(lproto == NULL);
Joel Beckerbd3e7612008-02-01 12:14:57 -0800214
Joel Becker4670c462008-02-01 14:39:35 -0800215 status = dlmlock(conn->cc_lockspace, o2dlm_mode, &lksb->lksb_o2dlm,
216 o2dlm_flags, name, namelen,
Joel Becker8f2c9c12008-02-01 12:16:57 -0800217 o2dlm_lock_ast_wrapper, astarg,
218 o2dlm_blocking_ast_wrapper);
Joel Becker7431cd72008-02-01 12:15:37 -0800219 ret = dlm_status_to_errno(status);
220 return ret;
Joel Becker24ef1812008-01-29 17:37:32 -0800221}
222
Joel Becker4670c462008-02-01 14:39:35 -0800223int ocfs2_dlm_unlock(struct ocfs2_cluster_connection *conn,
Joel Becker8f2c9c12008-02-01 12:16:57 -0800224 union ocfs2_dlm_lksb *lksb,
Joel Becker24ef1812008-01-29 17:37:32 -0800225 u32 flags,
226 void *astarg)
227{
Joel Becker7431cd72008-02-01 12:15:37 -0800228 enum dlm_status status;
Joel Beckerbd3e7612008-02-01 12:14:57 -0800229 int o2dlm_flags = flags_to_o2dlm(flags);
Joel Becker7431cd72008-02-01 12:15:37 -0800230 int ret;
Joel Beckerbd3e7612008-02-01 12:14:57 -0800231
Joel Becker24ef1812008-01-29 17:37:32 -0800232 BUG_ON(lproto == NULL);
233
Joel Becker4670c462008-02-01 14:39:35 -0800234 status = dlmunlock(conn->cc_lockspace, &lksb->lksb_o2dlm,
235 o2dlm_flags, o2dlm_unlock_ast_wrapper, astarg);
Joel Becker7431cd72008-02-01 12:15:37 -0800236 ret = dlm_status_to_errno(status);
237 return ret;
Joel Becker24ef1812008-01-29 17:37:32 -0800238}
239
Joel Becker8f2c9c12008-02-01 12:16:57 -0800240int ocfs2_dlm_lock_status(union ocfs2_dlm_lksb *lksb)
241{
242 return dlm_status_to_errno(lksb->lksb_o2dlm.status);
243}
244
245/*
246 * Why don't we cast to ocfs2_meta_lvb? The "clean" answer is that we
247 * don't cast at the glue level. The real answer is that the header
248 * ordering is nigh impossible.
249 */
250void *ocfs2_dlm_lvb(union ocfs2_dlm_lksb *lksb)
251{
252 return (void *)(lksb->lksb_o2dlm.lvb);
253}
Joel Becker24ef1812008-01-29 17:37:32 -0800254
Joel Beckercf0acdc2008-01-29 16:59:55 -0800255void ocfs2_dlm_dump_lksb(union ocfs2_dlm_lksb *lksb)
256{
257 dlm_print_one_lock(lksb->lksb_o2dlm.lockid);
258}
259
Joel Becker4670c462008-02-01 14:39:35 -0800260/*
261 * Called from the dlm when it's about to evict a node. This is how the
262 * classic stack signals node death.
263 */
264static void o2dlm_eviction_cb(int node_num, void *data)
265{
266 struct ocfs2_cluster_connection *conn = data;
267
268 mlog(ML_NOTICE, "o2dlm has evicted node %d from group %.*s\n",
269 node_num, conn->cc_namelen, conn->cc_name);
270
271 conn->cc_recovery_handler(node_num, conn->cc_recovery_data);
272}
273
274int ocfs2_cluster_connect(const char *group,
275 int grouplen,
276 void (*recovery_handler)(int node_num,
277 void *recovery_data),
278 void *recovery_data,
279 struct ocfs2_cluster_connection **conn)
280{
281 int rc = 0;
282 struct ocfs2_cluster_connection *new_conn;
283 u32 dlm_key;
284 struct dlm_ctxt *dlm;
285 struct o2dlm_private *priv;
286 struct dlm_protocol_version dlm_version;
287
288 BUG_ON(group == NULL);
289 BUG_ON(conn == NULL);
290 BUG_ON(recovery_handler == NULL);
291
292 if (grouplen > GROUP_NAME_MAX) {
293 rc = -EINVAL;
294 goto out;
295 }
296
Joel Becker6953b4c2008-01-29 16:59:56 -0800297 /* for now we only have one cluster/node, make sure we see it
298 * in the heartbeat universe */
299 if (!o2hb_check_local_node_heartbeating()) {
300 rc = -EINVAL;
301 goto out;
302 }
303
Joel Becker4670c462008-02-01 14:39:35 -0800304 new_conn = kzalloc(sizeof(struct ocfs2_cluster_connection),
305 GFP_KERNEL);
306 if (!new_conn) {
307 rc = -ENOMEM;
308 goto out;
309 }
310
311 memcpy(new_conn->cc_name, group, grouplen);
312 new_conn->cc_namelen = grouplen;
313 new_conn->cc_recovery_handler = recovery_handler;
314 new_conn->cc_recovery_data = recovery_data;
315
316 /* Start the new connection at our maximum compatibility level */
317 new_conn->cc_version = lproto->lp_max_version;
318
319 priv = kzalloc(sizeof(struct o2dlm_private), GFP_KERNEL);
320 if (!priv) {
321 rc = -ENOMEM;
322 goto out_free;
323 }
324
325 /* This just fills the structure in. It is safe to use new_conn. */
326 dlm_setup_eviction_cb(&priv->op_eviction_cb, o2dlm_eviction_cb,
327 new_conn);
328
329 new_conn->cc_private = priv;
330
331 /* used by the dlm code to make message headers unique, each
332 * node in this domain must agree on this. */
333 dlm_key = crc32_le(0, group, grouplen);
334 dlm_version.pv_major = new_conn->cc_version.pv_major;
335 dlm_version.pv_minor = new_conn->cc_version.pv_minor;
336
337 dlm = dlm_register_domain(group, dlm_key, &dlm_version);
338 if (IS_ERR(dlm)) {
339 rc = PTR_ERR(dlm);
340 mlog_errno(rc);
341 goto out_free;
342 }
343
344 new_conn->cc_version.pv_major = dlm_version.pv_major;
345 new_conn->cc_version.pv_minor = dlm_version.pv_minor;
346 new_conn->cc_lockspace = dlm;
347
348 dlm_register_eviction_cb(dlm, &priv->op_eviction_cb);
349
350 *conn = new_conn;
351
352out_free:
353 if (rc) {
354 kfree(new_conn->cc_private);
355 kfree(new_conn);
356 }
357
358out:
359 return rc;
360}
361
Joel Becker6953b4c2008-01-29 16:59:56 -0800362
Joel Becker4670c462008-02-01 14:39:35 -0800363int ocfs2_cluster_disconnect(struct ocfs2_cluster_connection *conn)
364{
365 struct dlm_ctxt *dlm = conn->cc_lockspace;
366 struct o2dlm_private *priv = conn->cc_private;
367
368 dlm_unregister_eviction_cb(&priv->op_eviction_cb);
369 dlm_unregister_domain(dlm);
370
371 kfree(priv);
372 kfree(conn);
373
374 return 0;
375}
376
Joel Becker6953b4c2008-01-29 16:59:56 -0800377static void o2hb_stop(const char *group)
378{
379 int ret;
380 char *argv[5], *envp[3];
381
382 argv[0] = (char *)o2nm_get_hb_ctl_path();
383 argv[1] = "-K";
384 argv[2] = "-u";
385 argv[3] = (char *)group;
386 argv[4] = NULL;
387
388 mlog(0, "Run: %s %s %s %s\n", argv[0], argv[1], argv[2], argv[3]);
389
390 /* minimal command environment taken from cpu_run_sbin_hotplug */
391 envp[0] = "HOME=/";
392 envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
393 envp[2] = NULL;
394
395 ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
396 if (ret < 0)
397 mlog_errno(ret);
398}
399
400/*
401 * Hangup is a hack for tools compatibility. Older ocfs2-tools software
402 * expects the filesystem to call "ocfs2_hb_ctl" during unmount. This
403 * happens regardless of whether the DLM got started, so we can't do it
404 * in ocfs2_cluster_disconnect(). We bring the o2hb_stop() function into
405 * the glue and provide a "hangup" API for super.c to call.
406 *
407 * Other stacks will eventually provide a NULL ->hangup() pointer.
408 */
409void ocfs2_cluster_hangup(const char *group, int grouplen)
410{
411 BUG_ON(group == NULL);
412 BUG_ON(group[grouplen] != '\0');
413
414 o2hb_stop(group);
415}
416
Joel Becker19fdb622008-01-30 15:38:24 -0800417int ocfs2_cluster_this_node(unsigned int *node)
418{
419 int node_num;
420
421 node_num = o2nm_this_node();
422 if (node_num == O2NM_INVALID_NODE_NUM)
423 return -ENOENT;
424
425 if (node_num >= O2NM_MAX_NODES)
426 return -EOVERFLOW;
427
428 *node = node_num;
429 return 0;
430}
431
Joel Becker24ef1812008-01-29 17:37:32 -0800432void o2cb_get_stack(struct ocfs2_locking_protocol *proto)
433{
434 BUG_ON(proto == NULL);
435
436 lproto = proto;
437}
438
439void o2cb_put_stack(void)
440{
441 lproto = NULL;
442}