aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorJoel Becker <joel.becker@oracle.com>2008-02-01 18:02:36 -0500
committerMark Fasheh <mfasheh@suse.com>2008-04-18 11:56:05 -0400
commite3dad42bf993a0f24eb6e46152356c9b119c15e8 (patch)
treedf0419dc5e4387db7e6a6e0118bb5b9d01d8c0b2
parent553aa7e408eac402c00b67ddfa7aec13fe1f3a33 (diff)
ocfs2: Create ocfs2_stack_operations and split out the o2cb stack.
Define the ocfs2_stack_operations structure. Build o2cb_stack_ops from all of the o2cb-specific stack functions. Change the generic stack glue functions to call the stack_ops instead of the o2cb functions directly. The o2cb functions are moved to stack_o2cb.c. The headers are cleaned up to where only needed headers are included. In this code, stackglue.c and stack_o2cb.c refer to some shared extern variables. When they become modules, that will change. Signed-off-by: Joel Becker <joel.becker@oracle.com> Signed-off-by: Mark Fasheh <mfasheh@suse.com>
-rw-r--r--fs/ocfs2/Makefile1
-rw-r--r--fs/ocfs2/stack_o2cb.c395
-rw-r--r--fs/ocfs2/stackglue.c385
-rw-r--r--fs/ocfs2/stackglue.h123
4 files changed, 532 insertions, 372 deletions
diff --git a/fs/ocfs2/Makefile b/fs/ocfs2/Makefile
index 3ba64af26951..8e86195fedbf 100644
--- a/fs/ocfs2/Makefile
+++ b/fs/ocfs2/Makefile
@@ -25,6 +25,7 @@ ocfs2-objs := \
25 resize.o \ 25 resize.o \
26 slot_map.o \ 26 slot_map.o \
27 stackglue.o \ 27 stackglue.o \
28 stack_o2cb.o \
28 suballoc.o \ 29 suballoc.o \
29 super.o \ 30 super.o \
30 symlink.o \ 31 symlink.o \
diff --git a/fs/ocfs2/stack_o2cb.c b/fs/ocfs2/stack_o2cb.c
new file mode 100644
index 000000000000..c9bc3541a33f
--- /dev/null
+++ b/fs/ocfs2/stack_o2cb.c
@@ -0,0 +1,395 @@
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/crc32.h>
21#include <linux/kmod.h>
22
23/* Needed for AOP_TRUNCATED_PAGE in mlog_errno() */
24#include <linux/fs.h>
25
26#include "cluster/masklog.h"
27#include "cluster/nodemanager.h"
28#include "cluster/heartbeat.h"
29
30#include "stackglue.h"
31
32struct o2dlm_private {
33 struct dlm_eviction_cb op_eviction_cb;
34};
35
36/* These should be identical */
37#if (DLM_LOCK_IV != LKM_IVMODE)
38# error Lock modes do not match
39#endif
40#if (DLM_LOCK_NL != LKM_NLMODE)
41# error Lock modes do not match
42#endif
43#if (DLM_LOCK_CR != LKM_CRMODE)
44# error Lock modes do not match
45#endif
46#if (DLM_LOCK_CW != LKM_CWMODE)
47# error Lock modes do not match
48#endif
49#if (DLM_LOCK_PR != LKM_PRMODE)
50# error Lock modes do not match
51#endif
52#if (DLM_LOCK_PW != LKM_PWMODE)
53# error Lock modes do not match
54#endif
55#if (DLM_LOCK_EX != LKM_EXMODE)
56# error Lock modes do not match
57#endif
58static inline int mode_to_o2dlm(int mode)
59{
60 BUG_ON(mode > LKM_MAXMODE);
61
62 return mode;
63}
64
65#define map_flag(_generic, _o2dlm) \
66 if (flags & (_generic)) { \
67 flags &= ~(_generic); \
68 o2dlm_flags |= (_o2dlm); \
69 }
70static int flags_to_o2dlm(u32 flags)
71{
72 int o2dlm_flags = 0;
73
74 map_flag(DLM_LKF_NOQUEUE, LKM_NOQUEUE);
75 map_flag(DLM_LKF_CANCEL, LKM_CANCEL);
76 map_flag(DLM_LKF_CONVERT, LKM_CONVERT);
77 map_flag(DLM_LKF_VALBLK, LKM_VALBLK);
78 map_flag(DLM_LKF_IVVALBLK, LKM_INVVALBLK);
79 map_flag(DLM_LKF_ORPHAN, LKM_ORPHAN);
80 map_flag(DLM_LKF_FORCEUNLOCK, LKM_FORCE);
81 map_flag(DLM_LKF_TIMEOUT, LKM_TIMEOUT);
82 map_flag(DLM_LKF_LOCAL, LKM_LOCAL);
83
84 /* map_flag() should have cleared every flag passed in */
85 BUG_ON(flags != 0);
86
87 return o2dlm_flags;
88}
89#undef map_flag
90
91/*
92 * Map an o2dlm status to standard errno values.
93 *
94 * o2dlm only uses a handful of these, and returns even fewer to the
95 * caller. Still, we try to assign sane values to each error.
96 *
97 * The following value pairs have special meanings to dlmglue, thus
98 * the right hand side needs to stay unique - never duplicate the
99 * mapping elsewhere in the table!
100 *
101 * DLM_NORMAL: 0
102 * DLM_NOTQUEUED: -EAGAIN
103 * DLM_CANCELGRANT: -EBUSY
104 * DLM_CANCEL: -DLM_ECANCEL
105 */
106/* Keep in sync with dlmapi.h */
107static int status_map[] = {
108 [DLM_NORMAL] = 0, /* Success */
109 [DLM_GRANTED] = -EINVAL,
110 [DLM_DENIED] = -EACCES,
111 [DLM_DENIED_NOLOCKS] = -EACCES,
112 [DLM_WORKING] = -EACCES,
113 [DLM_BLOCKED] = -EINVAL,
114 [DLM_BLOCKED_ORPHAN] = -EINVAL,
115 [DLM_DENIED_GRACE_PERIOD] = -EACCES,
116 [DLM_SYSERR] = -ENOMEM, /* It is what it is */
117 [DLM_NOSUPPORT] = -EPROTO,
118 [DLM_CANCELGRANT] = -EBUSY, /* Cancel after grant */
119 [DLM_IVLOCKID] = -EINVAL,
120 [DLM_SYNC] = -EINVAL,
121 [DLM_BADTYPE] = -EINVAL,
122 [DLM_BADRESOURCE] = -EINVAL,
123 [DLM_MAXHANDLES] = -ENOMEM,
124 [DLM_NOCLINFO] = -EINVAL,
125 [DLM_NOLOCKMGR] = -EINVAL,
126 [DLM_NOPURGED] = -EINVAL,
127 [DLM_BADARGS] = -EINVAL,
128 [DLM_VOID] = -EINVAL,
129 [DLM_NOTQUEUED] = -EAGAIN, /* Trylock failed */
130 [DLM_IVBUFLEN] = -EINVAL,
131 [DLM_CVTUNGRANT] = -EPERM,
132 [DLM_BADPARAM] = -EINVAL,
133 [DLM_VALNOTVALID] = -EINVAL,
134 [DLM_REJECTED] = -EPERM,
135 [DLM_ABORT] = -EINVAL,
136 [DLM_CANCEL] = -DLM_ECANCEL, /* Successful cancel */
137 [DLM_IVRESHANDLE] = -EINVAL,
138 [DLM_DEADLOCK] = -EDEADLK,
139 [DLM_DENIED_NOASTS] = -EINVAL,
140 [DLM_FORWARD] = -EINVAL,
141 [DLM_TIMEOUT] = -ETIMEDOUT,
142 [DLM_IVGROUPID] = -EINVAL,
143 [DLM_VERS_CONFLICT] = -EOPNOTSUPP,
144 [DLM_BAD_DEVICE_PATH] = -ENOENT,
145 [DLM_NO_DEVICE_PERMISSION] = -EPERM,
146 [DLM_NO_CONTROL_DEVICE] = -ENOENT,
147 [DLM_RECOVERING] = -ENOTCONN,
148 [DLM_MIGRATING] = -ERESTART,
149 [DLM_MAXSTATS] = -EINVAL,
150};
151
152static int dlm_status_to_errno(enum dlm_status status)
153{
154 BUG_ON(status > (sizeof(status_map) / sizeof(status_map[0])));
155
156 return status_map[status];
157}
158
159static void o2dlm_lock_ast_wrapper(void *astarg)
160{
161 BUG_ON(stack_glue_lproto == NULL);
162
163 stack_glue_lproto->lp_lock_ast(astarg);
164}
165
166static void o2dlm_blocking_ast_wrapper(void *astarg, int level)
167{
168 BUG_ON(stack_glue_lproto == NULL);
169
170 stack_glue_lproto->lp_blocking_ast(astarg, level);
171}
172
173static void o2dlm_unlock_ast_wrapper(void *astarg, enum dlm_status status)
174{
175 int error = dlm_status_to_errno(status);
176
177 BUG_ON(stack_glue_lproto == NULL);
178
179 /*
180 * In o2dlm, you can get both the lock_ast() for the lock being
181 * granted and the unlock_ast() for the CANCEL failing. A
182 * successful cancel sends DLM_NORMAL here. If the
183 * lock grant happened before the cancel arrived, you get
184 * DLM_CANCELGRANT.
185 *
186 * There's no need for the double-ast. If we see DLM_CANCELGRANT,
187 * we just ignore it. We expect the lock_ast() to handle the
188 * granted lock.
189 */
190 if (status == DLM_CANCELGRANT)
191 return;
192
193 stack_glue_lproto->lp_unlock_ast(astarg, error);
194}
195
196static int o2cb_dlm_lock(struct ocfs2_cluster_connection *conn,
197 int mode,
198 union ocfs2_dlm_lksb *lksb,
199 u32 flags,
200 void *name,
201 unsigned int namelen,
202 void *astarg)
203{
204 enum dlm_status status;
205 int o2dlm_mode = mode_to_o2dlm(mode);
206 int o2dlm_flags = flags_to_o2dlm(flags);
207 int ret;
208
209 status = dlmlock(conn->cc_lockspace, o2dlm_mode, &lksb->lksb_o2dlm,
210 o2dlm_flags, name, namelen,
211 o2dlm_lock_ast_wrapper, astarg,
212 o2dlm_blocking_ast_wrapper);
213 ret = dlm_status_to_errno(status);
214 return ret;
215}
216
217static int o2cb_dlm_unlock(struct ocfs2_cluster_connection *conn,
218 union ocfs2_dlm_lksb *lksb,
219 u32 flags,
220 void *astarg)
221{
222 enum dlm_status status;
223 int o2dlm_flags = flags_to_o2dlm(flags);
224 int ret;
225
226 status = dlmunlock(conn->cc_lockspace, &lksb->lksb_o2dlm,
227 o2dlm_flags, o2dlm_unlock_ast_wrapper, astarg);
228 ret = dlm_status_to_errno(status);
229 return ret;
230}
231
232static int o2cb_dlm_lock_status(union ocfs2_dlm_lksb *lksb)
233{
234 return dlm_status_to_errno(lksb->lksb_o2dlm.status);
235}
236
237static void *o2cb_dlm_lvb(union ocfs2_dlm_lksb *lksb)
238{
239 return (void *)(lksb->lksb_o2dlm.lvb);
240}
241
242static void o2cb_dump_lksb(union ocfs2_dlm_lksb *lksb)
243{
244 dlm_print_one_lock(lksb->lksb_o2dlm.lockid);
245}
246
247/*
248 * Called from the dlm when it's about to evict a node. This is how the
249 * classic stack signals node death.
250 */
251static void o2dlm_eviction_cb(int node_num, void *data)
252{
253 struct ocfs2_cluster_connection *conn = data;
254
255 mlog(ML_NOTICE, "o2dlm has evicted node %d from group %.*s\n",
256 node_num, conn->cc_namelen, conn->cc_name);
257
258 conn->cc_recovery_handler(node_num, conn->cc_recovery_data);
259}
260
261static int o2cb_cluster_connect(struct ocfs2_cluster_connection *conn)
262{
263 int rc = 0;
264 u32 dlm_key;
265 struct dlm_ctxt *dlm;
266 struct o2dlm_private *priv;
267 struct dlm_protocol_version dlm_version;
268
269 BUG_ON(conn == NULL);
270
271 /* for now we only have one cluster/node, make sure we see it
272 * in the heartbeat universe */
273 if (!o2hb_check_local_node_heartbeating()) {
274 rc = -EINVAL;
275 goto out;
276 }
277
278 priv = kzalloc(sizeof(struct o2dlm_private), GFP_KERNEL);
279 if (!priv) {
280 rc = -ENOMEM;
281 goto out_free;
282 }
283
284 /* This just fills the structure in. It is safe to pass conn. */
285 dlm_setup_eviction_cb(&priv->op_eviction_cb, o2dlm_eviction_cb,
286 conn);
287
288 conn->cc_private = priv;
289
290 /* used by the dlm code to make message headers unique, each
291 * node in this domain must agree on this. */
292 dlm_key = crc32_le(0, conn->cc_name, conn->cc_namelen);
293 dlm_version.pv_major = conn->cc_version.pv_major;
294 dlm_version.pv_minor = conn->cc_version.pv_minor;
295
296 dlm = dlm_register_domain(conn->cc_name, dlm_key, &dlm_version);
297 if (IS_ERR(dlm)) {
298 rc = PTR_ERR(dlm);
299 mlog_errno(rc);
300 goto out_free;
301 }
302
303 conn->cc_version.pv_major = dlm_version.pv_major;
304 conn->cc_version.pv_minor = dlm_version.pv_minor;
305 conn->cc_lockspace = dlm;
306
307 dlm_register_eviction_cb(dlm, &priv->op_eviction_cb);
308
309out_free:
310 if (rc && conn->cc_private)
311 kfree(conn->cc_private);
312
313out:
314 return rc;
315}
316
317static int o2cb_cluster_disconnect(struct ocfs2_cluster_connection *conn)
318{
319 struct dlm_ctxt *dlm = conn->cc_lockspace;
320 struct o2dlm_private *priv = conn->cc_private;
321
322 dlm_unregister_eviction_cb(&priv->op_eviction_cb);
323 conn->cc_private = NULL;
324 kfree(priv);
325
326 dlm_unregister_domain(dlm);
327 conn->cc_lockspace = NULL;
328
329 return 0;
330}
331
332static void o2hb_stop(const char *group)
333{
334 int ret;
335 char *argv[5], *envp[3];
336
337 argv[0] = (char *)o2nm_get_hb_ctl_path();
338 argv[1] = "-K";
339 argv[2] = "-u";
340 argv[3] = (char *)group;
341 argv[4] = NULL;
342
343 mlog(0, "Run: %s %s %s %s\n", argv[0], argv[1], argv[2], argv[3]);
344
345 /* minimal command environment taken from cpu_run_sbin_hotplug */
346 envp[0] = "HOME=/";
347 envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
348 envp[2] = NULL;
349
350 ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
351 if (ret < 0)
352 mlog_errno(ret);
353}
354
355/*
356 * Hangup is a hack for tools compatibility. Older ocfs2-tools software
357 * expects the filesystem to call "ocfs2_hb_ctl" during unmount. This
358 * happens regardless of whether the DLM got started, so we can't do it
359 * in ocfs2_cluster_disconnect(). We bring the o2hb_stop() function into
360 * the glue and provide a "hangup" API for super.c to call.
361 *
362 * Other stacks will eventually provide a NULL ->hangup() pointer.
363 */
364static void o2cb_cluster_hangup(const char *group, int grouplen)
365{
366 o2hb_stop(group);
367}
368
369static int o2cb_cluster_this_node(unsigned int *node)
370{
371 int node_num;
372
373 node_num = o2nm_this_node();
374 if (node_num == O2NM_INVALID_NODE_NUM)
375 return -ENOENT;
376
377 if (node_num >= O2NM_MAX_NODES)
378 return -EOVERFLOW;
379
380 *node = node_num;
381 return 0;
382}
383
384struct ocfs2_stack_operations o2cb_stack_ops = {
385 .connect = o2cb_cluster_connect,
386 .disconnect = o2cb_cluster_disconnect,
387 .hangup = o2cb_cluster_hangup,
388 .this_node = o2cb_cluster_this_node,
389 .dlm_lock = o2cb_dlm_lock,
390 .dlm_unlock = o2cb_dlm_unlock,
391 .lock_status = o2cb_dlm_lock_status,
392 .lock_lvb = o2cb_dlm_lvb,
393 .dump_lksb = o2cb_dump_lksb,
394};
395
diff --git a/fs/ocfs2/stackglue.c b/fs/ocfs2/stackglue.c
index e35dde6217f5..e197367b6bd6 100644
--- a/fs/ocfs2/stackglue.c
+++ b/fs/ocfs2/stackglue.c
@@ -19,204 +19,17 @@
19 */ 19 */
20 20
21#include <linux/slab.h> 21#include <linux/slab.h>
22#include <linux/crc32.h>
23#include <linux/kmod.h> 22#include <linux/kmod.h>
24 23
25/* Needed for AOP_TRUNCATED_PAGE in mlog_errno() */ 24/* Needed for AOP_TRUNCATED_PAGE in mlog_errno() */
26#include <linux/fs.h> 25#include <linux/fs.h>
27 26
28#include "cluster/masklog.h" 27#include "cluster/masklog.h"
29#include "cluster/nodemanager.h"
30#include "cluster/heartbeat.h"
31 28
32#include "stackglue.h" 29#include "stackglue.h"
33 30
34static struct ocfs2_locking_protocol *lproto; 31struct ocfs2_locking_protocol *stack_glue_lproto;
35
36struct o2dlm_private {
37 struct dlm_eviction_cb op_eviction_cb;
38};
39
40/* 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
95/*
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
107 * DLM_CANCELGRANT: -EBUSY
108 * DLM_CANCEL: -DLM_ECANCEL
109 */
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,
116 [DLM_WORKING] = -EACCES,
117 [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,
122 [DLM_CANCELGRANT] = -EBUSY, /* Cancel after grant */
123 [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,
140 [DLM_CANCEL] = -DLM_ECANCEL, /* Successful cancel */
141 [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};
155
156static int dlm_status_to_errno(enum dlm_status status)
157{
158 BUG_ON(status > (sizeof(status_map) / sizeof(status_map[0])));
159 32
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{
179 int error = dlm_status_to_errno(status);
180
181 BUG_ON(lproto == NULL);
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 lproto->lp_unlock_ast(astarg, error);
198}
199
200static int o2cb_dlm_lock(struct ocfs2_cluster_connection *conn,
201 int mode,
202 union ocfs2_dlm_lksb *lksb,
203 u32 flags,
204 void *name,
205 unsigned int namelen,
206 void *astarg)
207{
208 enum dlm_status status;
209 int o2dlm_mode = mode_to_o2dlm(mode);
210 int o2dlm_flags = flags_to_o2dlm(flags);
211 int ret;
212
213 status = dlmlock(conn->cc_lockspace, o2dlm_mode, &lksb->lksb_o2dlm,
214 o2dlm_flags, name, namelen,
215 o2dlm_lock_ast_wrapper, astarg,
216 o2dlm_blocking_ast_wrapper);
217 ret = dlm_status_to_errno(status);
218 return ret;
219}
220 33
221int ocfs2_dlm_lock(struct ocfs2_cluster_connection *conn, 34int ocfs2_dlm_lock(struct ocfs2_cluster_connection *conn,
222 int mode, 35 int mode,
@@ -226,25 +39,10 @@ int ocfs2_dlm_lock(struct ocfs2_cluster_connection *conn,
226 unsigned int namelen, 39 unsigned int namelen,
227 void *astarg) 40 void *astarg)
228{ 41{
229 BUG_ON(lproto == NULL); 42 BUG_ON(stack_glue_lproto == NULL);
230
231 return o2cb_dlm_lock(conn, mode, lksb, flags,
232 name, namelen, astarg);
233}
234
235static int o2cb_dlm_unlock(struct ocfs2_cluster_connection *conn,
236 union ocfs2_dlm_lksb *lksb,
237 u32 flags,
238 void *astarg)
239{
240 enum dlm_status status;
241 int o2dlm_flags = flags_to_o2dlm(flags);
242 int ret;
243 43
244 status = dlmunlock(conn->cc_lockspace, &lksb->lksb_o2dlm, 44 return o2cb_stack_ops.dlm_lock(conn, mode, lksb, flags,
245 o2dlm_flags, o2dlm_unlock_ast_wrapper, astarg); 45 name, namelen, astarg);
246 ret = dlm_status_to_errno(status);
247 return ret;
248} 46}
249 47
250int ocfs2_dlm_unlock(struct ocfs2_cluster_connection *conn, 48int ocfs2_dlm_unlock(struct ocfs2_cluster_connection *conn,
@@ -252,19 +50,14 @@ int ocfs2_dlm_unlock(struct ocfs2_cluster_connection *conn,
252 u32 flags, 50 u32 flags,
253 void *astarg) 51 void *astarg)
254{ 52{
255 BUG_ON(lproto == NULL); 53 BUG_ON(stack_glue_lproto == NULL);
256 54
257 return o2cb_dlm_unlock(conn, lksb, flags, astarg); 55 return o2cb_stack_ops.dlm_unlock(conn, lksb, flags, astarg);
258}
259
260static int o2cb_dlm_lock_status(union ocfs2_dlm_lksb *lksb)
261{
262 return dlm_status_to_errno(lksb->lksb_o2dlm.status);
263} 56}
264 57
265int ocfs2_dlm_lock_status(union ocfs2_dlm_lksb *lksb) 58int ocfs2_dlm_lock_status(union ocfs2_dlm_lksb *lksb)
266{ 59{
267 return o2cb_dlm_lock_status(lksb); 60 return o2cb_stack_ops.lock_status(lksb);
268} 61}
269 62
270/* 63/*
@@ -272,94 +65,14 @@ int ocfs2_dlm_lock_status(union ocfs2_dlm_lksb *lksb)
272 * don't cast at the glue level. The real answer is that the header 65 * don't cast at the glue level. The real answer is that the header
273 * ordering is nigh impossible. 66 * ordering is nigh impossible.
274 */ 67 */
275static void *o2cb_dlm_lvb(union ocfs2_dlm_lksb *lksb)
276{
277 return (void *)(lksb->lksb_o2dlm.lvb);
278}
279
280void *ocfs2_dlm_lvb(union ocfs2_dlm_lksb *lksb) 68void *ocfs2_dlm_lvb(union ocfs2_dlm_lksb *lksb)
281{ 69{
282 return o2cb_dlm_lvb(lksb); 70 return o2cb_stack_ops.lock_lvb(lksb);
283}
284
285static void o2cb_dlm_dump_lksb(union ocfs2_dlm_lksb *lksb)
286{
287 dlm_print_one_lock(lksb->lksb_o2dlm.lockid);
288} 71}
289 72
290void ocfs2_dlm_dump_lksb(union ocfs2_dlm_lksb *lksb) 73void ocfs2_dlm_dump_lksb(union ocfs2_dlm_lksb *lksb)
291{ 74{
292 o2cb_dlm_dump_lksb(lksb); 75 o2cb_stack_ops.dump_lksb(lksb);
293}
294
295/*
296 * Called from the dlm when it's about to evict a node. This is how the
297 * classic stack signals node death.
298 */
299static void o2dlm_eviction_cb(int node_num, void *data)
300{
301 struct ocfs2_cluster_connection *conn = data;
302
303 mlog(ML_NOTICE, "o2dlm has evicted node %d from group %.*s\n",
304 node_num, conn->cc_namelen, conn->cc_name);
305
306 conn->cc_recovery_handler(node_num, conn->cc_recovery_data);
307}
308
309static int o2cb_cluster_connect(struct ocfs2_cluster_connection *conn)
310{
311 int rc = 0;
312 u32 dlm_key;
313 struct dlm_ctxt *dlm;
314 struct o2dlm_private *priv;
315 struct dlm_protocol_version dlm_version;
316
317 BUG_ON(conn == NULL);
318
319 /* for now we only have one cluster/node, make sure we see it
320 * in the heartbeat universe */
321 if (!o2hb_check_local_node_heartbeating()) {
322 rc = -EINVAL;
323 goto out;
324 }
325
326 priv = kzalloc(sizeof(struct o2dlm_private), GFP_KERNEL);
327 if (!priv) {
328 rc = -ENOMEM;
329 goto out_free;
330 }
331
332 /* This just fills the structure in. It is safe to pass conn. */
333 dlm_setup_eviction_cb(&priv->op_eviction_cb, o2dlm_eviction_cb,
334 conn);
335
336 conn->cc_private = priv;
337
338 /* used by the dlm code to make message headers unique, each
339 * node in this domain must agree on this. */
340 dlm_key = crc32_le(0, conn->cc_name, conn->cc_namelen);
341 dlm_version.pv_major = conn->cc_version.pv_major;
342 dlm_version.pv_minor = conn->cc_version.pv_minor;
343
344 dlm = dlm_register_domain(conn->cc_name, dlm_key, &dlm_version);
345 if (IS_ERR(dlm)) {
346 rc = PTR_ERR(dlm);
347 mlog_errno(rc);
348 goto out_free;
349 }
350
351 conn->cc_version.pv_major = dlm_version.pv_major;
352 conn->cc_version.pv_minor = dlm_version.pv_minor;
353 conn->cc_lockspace = dlm;
354
355 dlm_register_eviction_cb(dlm, &priv->op_eviction_cb);
356
357out_free:
358 if (rc && conn->cc_private)
359 kfree(conn->cc_private);
360
361out:
362 return rc;
363} 76}
364 77
365int ocfs2_cluster_connect(const char *group, 78int ocfs2_cluster_connect(const char *group,
@@ -394,9 +107,9 @@ int ocfs2_cluster_connect(const char *group,
394 new_conn->cc_recovery_data = recovery_data; 107 new_conn->cc_recovery_data = recovery_data;
395 108
396 /* Start the new connection at our maximum compatibility level */ 109 /* Start the new connection at our maximum compatibility level */
397 new_conn->cc_version = lproto->lp_max_version; 110 new_conn->cc_version = stack_glue_lproto->lp_max_version;
398 111
399 rc = o2cb_cluster_connect(new_conn); 112 rc = o2cb_stack_ops.connect(new_conn);
400 if (rc) { 113 if (rc) {
401 mlog_errno(rc); 114 mlog_errno(rc);
402 goto out_free; 115 goto out_free;
@@ -412,29 +125,13 @@ out:
412 return rc; 125 return rc;
413} 126}
414 127
415
416static int o2cb_cluster_disconnect(struct ocfs2_cluster_connection *conn)
417{
418 struct dlm_ctxt *dlm = conn->cc_lockspace;
419 struct o2dlm_private *priv = conn->cc_private;
420
421 dlm_unregister_eviction_cb(&priv->op_eviction_cb);
422 conn->cc_private = NULL;
423 kfree(priv);
424
425 dlm_unregister_domain(dlm);
426 conn->cc_lockspace = NULL;
427
428 return 0;
429}
430
431int ocfs2_cluster_disconnect(struct ocfs2_cluster_connection *conn) 128int ocfs2_cluster_disconnect(struct ocfs2_cluster_connection *conn)
432{ 129{
433 int ret; 130 int ret;
434 131
435 BUG_ON(conn == NULL); 132 BUG_ON(conn == NULL);
436 133
437 ret = o2cb_cluster_disconnect(conn); 134 ret = o2cb_stack_ops.disconnect(conn);
438 135
439 /* XXX Should we free it anyway? */ 136 /* XXX Should we free it anyway? */
440 if (!ret) 137 if (!ret)
@@ -443,75 +140,23 @@ int ocfs2_cluster_disconnect(struct ocfs2_cluster_connection *conn)
443 return ret; 140 return ret;
444} 141}
445 142
446static void o2hb_stop(const char *group)
447{
448 int ret;
449 char *argv[5], *envp[3];
450
451 argv[0] = (char *)o2nm_get_hb_ctl_path();
452 argv[1] = "-K";
453 argv[2] = "-u";
454 argv[3] = (char *)group;
455 argv[4] = NULL;
456
457 mlog(0, "Run: %s %s %s %s\n", argv[0], argv[1], argv[2], argv[3]);
458
459 /* minimal command environment taken from cpu_run_sbin_hotplug */
460 envp[0] = "HOME=/";
461 envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
462 envp[2] = NULL;
463
464 ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
465 if (ret < 0)
466 mlog_errno(ret);
467}
468
469/*
470 * Hangup is a hack for tools compatibility. Older ocfs2-tools software
471 * expects the filesystem to call "ocfs2_hb_ctl" during unmount. This
472 * happens regardless of whether the DLM got started, so we can't do it
473 * in ocfs2_cluster_disconnect(). We bring the o2hb_stop() function into
474 * the glue and provide a "hangup" API for super.c to call.
475 *
476 * Other stacks will eventually provide a NULL ->hangup() pointer.
477 */
478static void o2cb_cluster_hangup(const char *group, int grouplen)
479{
480 o2hb_stop(group);
481}
482
483void ocfs2_cluster_hangup(const char *group, int grouplen) 143void ocfs2_cluster_hangup(const char *group, int grouplen)
484{ 144{
485 BUG_ON(group == NULL); 145 BUG_ON(group == NULL);
486 BUG_ON(group[grouplen] != '\0'); 146 BUG_ON(group[grouplen] != '\0');
487 147
488 o2cb_cluster_hangup(group, grouplen); 148 o2cb_stack_ops.hangup(group, grouplen);
489}
490
491static int o2cb_cluster_this_node(unsigned int *node)
492{
493 int node_num;
494
495 node_num = o2nm_this_node();
496 if (node_num == O2NM_INVALID_NODE_NUM)
497 return -ENOENT;
498
499 if (node_num >= O2NM_MAX_NODES)
500 return -EOVERFLOW;
501
502 *node = node_num;
503 return 0;
504} 149}
505 150
506int ocfs2_cluster_this_node(unsigned int *node) 151int ocfs2_cluster_this_node(unsigned int *node)
507{ 152{
508 return o2cb_cluster_this_node(node); 153 return o2cb_stack_ops.this_node(node);
509} 154}
510 155
511void ocfs2_stack_glue_set_locking_protocol(struct ocfs2_locking_protocol *proto) 156void ocfs2_stack_glue_set_locking_protocol(struct ocfs2_locking_protocol *proto)
512{ 157{
513 BUG_ON(proto != NULL); 158 BUG_ON(proto != NULL);
514 159
515 lproto = proto; 160 stack_glue_lproto = proto;
516} 161}
517 162
diff --git a/fs/ocfs2/stackglue.h b/fs/ocfs2/stackglue.h
index decb147106fd..083632215dc5 100644
--- a/fs/ocfs2/stackglue.h
+++ b/fs/ocfs2/stackglue.h
@@ -25,6 +25,8 @@
25#include <linux/list.h> 25#include <linux/list.h>
26#include <linux/dlmconstants.h> 26#include <linux/dlmconstants.h>
27 27
28#include "dlm/dlmapi.h"
29
28/* 30/*
29 * dlmconstants.h does not have a LOCAL flag. We hope to remove it 31 * dlmconstants.h does not have a LOCAL flag. We hope to remove it
30 * some day, but right now we need it. Let's fake it. This value is larger 32 * some day, but right now we need it. Let's fake it. This value is larger
@@ -39,13 +41,18 @@
39#define GROUP_NAME_MAX 64 41#define GROUP_NAME_MAX 64
40 42
41 43
42#include "dlm/dlmapi.h" 44/*
43 45 * ocfs2_protocol_version changes when ocfs2 does something different in
46 * its inter-node behavior. See dlmglue.c for more information.
47 */
44struct ocfs2_protocol_version { 48struct ocfs2_protocol_version {
45 u8 pv_major; 49 u8 pv_major;
46 u8 pv_minor; 50 u8 pv_minor;
47}; 51};
48 52
53/*
54 * The ocfs2_locking_protocol defines the handlers called on ocfs2's behalf.
55 */
49struct ocfs2_locking_protocol { 56struct ocfs2_locking_protocol {
50 struct ocfs2_protocol_version lp_max_version; 57 struct ocfs2_protocol_version lp_max_version;
51 void (*lp_lock_ast)(void *astarg); 58 void (*lp_lock_ast)(void *astarg);
@@ -53,10 +60,20 @@ struct ocfs2_locking_protocol {
53 void (*lp_unlock_ast)(void *astarg, int error); 60 void (*lp_unlock_ast)(void *astarg, int error);
54}; 61};
55 62
63/*
64 * A union of all lock status structures. We define it here so that the
65 * size of the union is known. Lock status structures are embedded in
66 * ocfs2 inodes.
67 */
56union ocfs2_dlm_lksb { 68union ocfs2_dlm_lksb {
57 struct dlm_lockstatus lksb_o2dlm; 69 struct dlm_lockstatus lksb_o2dlm;
58}; 70};
59 71
72/*
73 * A cluster connection. Mostly opaque to ocfs2, the connection holds
74 * state for the underlying stack. ocfs2 does use cc_version to determine
75 * locking compatibility.
76 */
60struct ocfs2_cluster_connection { 77struct ocfs2_cluster_connection {
61 char cc_name[GROUP_NAME_MAX]; 78 char cc_name[GROUP_NAME_MAX];
62 int cc_namelen; 79 int cc_namelen;
@@ -67,6 +84,106 @@ struct ocfs2_cluster_connection {
67 void *cc_private; 84 void *cc_private;
68}; 85};
69 86
87/*
88 * Each cluster stack implements the stack operations structure. Not used
89 * in the ocfs2 code, the stackglue code translates generic cluster calls
90 * into stack operations.
91 */
92struct ocfs2_stack_operations {
93 /*
94 * The fs code calls ocfs2_cluster_connect() to attach a new
95 * filesystem to the cluster stack. The ->connect() op is passed
96 * an ocfs2_cluster_connection with the name and recovery field
97 * filled in.
98 *
99 * The stack must set up any notification mechanisms and create
100 * the filesystem lockspace in the DLM. The lockspace should be
101 * stored on cc_lockspace. Any other information can be stored on
102 * cc_private.
103 *
104 * ->connect() must not return until it is guaranteed that
105 *
106 * - Node down notifications for the filesystem will be recieved
107 * and passed to conn->cc_recovery_handler().
108 * - Locking requests for the filesystem will be processed.
109 */
110 int (*connect)(struct ocfs2_cluster_connection *conn);
111
112 /*
113 * The fs code calls ocfs2_cluster_disconnect() when a filesystem
114 * no longer needs cluster services. All DLM locks have been
115 * dropped, and recovery notification is being ignored by the
116 * fs code. The stack must disengage from the DLM and discontinue
117 * recovery notification.
118 *
119 * Once ->disconnect() has returned, the connection structure will
120 * be freed. Thus, a stack must not return from ->disconnect()
121 * until it will no longer reference the conn pointer.
122 */
123 int (*disconnect)(struct ocfs2_cluster_connection *conn);
124
125 /*
126 * ocfs2_cluster_hangup() exists for compatibility with older
127 * ocfs2 tools. Only the classic stack really needs it. As such
128 * ->hangup() is not required of all stacks. See the comment by
129 * ocfs2_cluster_hangup() for more details.
130 */
131 void (*hangup)(const char *group, int grouplen);
132
133 /*
134 * ->this_node() returns the cluster's unique identifier for the
135 * local node.
136 */
137 int (*this_node)(unsigned int *node);
138
139 /*
140 * Call the underlying dlm lock function. The ->dlm_lock()
141 * callback should convert the flags and mode as appropriate.
142 *
143 * ast and bast functions are not part of the call because the
144 * stack will likely want to wrap ast and bast calls before passing
145 * them to stack->sp_proto.
146 */
147 int (*dlm_lock)(struct ocfs2_cluster_connection *conn,
148 int mode,
149 union ocfs2_dlm_lksb *lksb,
150 u32 flags,
151 void *name,
152 unsigned int namelen,
153 void *astarg);
154
155 /*
156 * Call the underlying dlm unlock function. The ->dlm_unlock()
157 * function should convert the flags as appropriate.
158 *
159 * The unlock ast is not passed, as the stack will want to wrap
160 * it before calling stack->sp_proto->lp_unlock_ast().
161 */
162 int (*dlm_unlock)(struct ocfs2_cluster_connection *conn,
163 union ocfs2_dlm_lksb *lksb,
164 u32 flags,
165 void *astarg);
166
167 /*
168 * Return the status of the current lock status block. The fs
169 * code should never dereference the union. The ->lock_status()
170 * callback pulls out the stack-specific lksb, converts the status
171 * to a proper errno, and returns it.
172 */
173 int (*lock_status)(union ocfs2_dlm_lksb *lksb);
174
175 /*
176 * Pull the lvb pointer off of the stack-specific lksb.
177 */
178 void *(*lock_lvb)(union ocfs2_dlm_lksb *lksb);
179
180 /*
181 * This is an optoinal debugging hook. If provided, the
182 * stack can dump debugging information about this lock.
183 */
184 void (*dump_lksb)(union ocfs2_dlm_lksb *lksb);
185};
186
70int ocfs2_cluster_connect(const char *group, 187int ocfs2_cluster_connect(const char *group,
71 int grouplen, 188 int grouplen,
72 void (*recovery_handler)(int node_num, 189 void (*recovery_handler)(int node_num,
@@ -95,4 +212,6 @@ void ocfs2_dlm_dump_lksb(union ocfs2_dlm_lksb *lksb);
95 212
96void ocfs2_stack_glue_set_locking_protocol(struct ocfs2_locking_protocol *proto); 213void ocfs2_stack_glue_set_locking_protocol(struct ocfs2_locking_protocol *proto);
97 214
215extern struct ocfs2_locking_protocol *stack_glue_lproto;
216extern struct ocfs2_stack_operations o2cb_stack_ops;
98#endif /* STACKGLUE_H */ 217#endif /* STACKGLUE_H */