1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * stack_o2cb.c
4 *
5 * Code which interfaces ocfs2 with the o2cb stack.
6 *
7 * Copyright (C) 2007 Oracle. All rights reserved.
8 */
9
10 #include <linux/kernel.h>
11 #include <linux/crc32.h>
12 #include <linux/slab.h>
13 #include <linux/module.h>
14
15 /* Needed for AOP_TRUNCATED_PAGE in mlog_errno() */
16 #include <linux/fs.h>
17
18 #include "cluster/masklog.h"
19 #include "cluster/nodemanager.h"
20 #include "cluster/heartbeat.h"
21 #include "cluster/tcp.h"
22
23 #include "stackglue.h"
24
25 struct o2dlm_private {
26 struct dlm_eviction_cb op_eviction_cb;
27 };
28
29 static struct ocfs2_stack_plugin o2cb_stack;
30
31 /* These should be identical */
32 #if (DLM_LOCK_IV != LKM_IVMODE)
33 # error Lock modes do not match
34 #endif
35 #if (DLM_LOCK_NL != LKM_NLMODE)
36 # error Lock modes do not match
37 #endif
38 #if (DLM_LOCK_CR != LKM_CRMODE)
39 # error Lock modes do not match
40 #endif
41 #if (DLM_LOCK_CW != LKM_CWMODE)
42 # error Lock modes do not match
43 #endif
44 #if (DLM_LOCK_PR != LKM_PRMODE)
45 # error Lock modes do not match
46 #endif
47 #if (DLM_LOCK_PW != LKM_PWMODE)
48 # error Lock modes do not match
49 #endif
50 #if (DLM_LOCK_EX != LKM_EXMODE)
51 # error Lock modes do not match
52 #endif
mode_to_o2dlm(int mode)53 static inline int mode_to_o2dlm(int mode)
54 {
55 BUG_ON(mode > LKM_MAXMODE);
56
57 return mode;
58 }
59
flags_to_o2dlm(u32 flags)60 static int flags_to_o2dlm(u32 flags)
61 {
62 int o2dlm_flags = 0;
63
64 if (flags & DLM_LKF_NOQUEUE)
65 o2dlm_flags |= LKM_NOQUEUE;
66 if (flags & DLM_LKF_CANCEL)
67 o2dlm_flags |= LKM_CANCEL;
68 if (flags & DLM_LKF_CONVERT)
69 o2dlm_flags |= LKM_CONVERT;
70 if (flags & DLM_LKF_VALBLK)
71 o2dlm_flags |= LKM_VALBLK;
72 if (flags & DLM_LKF_IVVALBLK)
73 o2dlm_flags |= LKM_INVVALBLK;
74 if (flags & DLM_LKF_ORPHAN)
75 o2dlm_flags |= LKM_ORPHAN;
76 if (flags & DLM_LKF_FORCEUNLOCK)
77 o2dlm_flags |= LKM_FORCE;
78 if (flags & DLM_LKF_TIMEOUT)
79 o2dlm_flags |= LKM_TIMEOUT;
80 if (flags & DLM_LKF_LOCAL)
81 o2dlm_flags |= LKM_LOCAL;
82
83 return o2dlm_flags;
84 }
85
86 /*
87 * Map an o2dlm status to standard errno values.
88 *
89 * o2dlm only uses a handful of these, and returns even fewer to the
90 * caller. Still, we try to assign sane values to each error.
91 *
92 * The following value pairs have special meanings to dlmglue, thus
93 * the right hand side needs to stay unique - never duplicate the
94 * mapping elsewhere in the table!
95 *
96 * DLM_NORMAL: 0
97 * DLM_NOTQUEUED: -EAGAIN
98 * DLM_CANCELGRANT: -EBUSY
99 * DLM_CANCEL: -DLM_ECANCEL
100 */
101 /* Keep in sync with dlmapi.h */
102 static int status_map[] = {
103 [DLM_NORMAL] = 0, /* Success */
104 [DLM_GRANTED] = -EINVAL,
105 [DLM_DENIED] = -EACCES,
106 [DLM_DENIED_NOLOCKS] = -EACCES,
107 [DLM_WORKING] = -EACCES,
108 [DLM_BLOCKED] = -EINVAL,
109 [DLM_BLOCKED_ORPHAN] = -EINVAL,
110 [DLM_DENIED_GRACE_PERIOD] = -EACCES,
111 [DLM_SYSERR] = -ENOMEM, /* It is what it is */
112 [DLM_NOSUPPORT] = -EPROTO,
113 [DLM_CANCELGRANT] = -EBUSY, /* Cancel after grant */
114 [DLM_IVLOCKID] = -EINVAL,
115 [DLM_SYNC] = -EINVAL,
116 [DLM_BADTYPE] = -EINVAL,
117 [DLM_BADRESOURCE] = -EINVAL,
118 [DLM_MAXHANDLES] = -ENOMEM,
119 [DLM_NOCLINFO] = -EINVAL,
120 [DLM_NOLOCKMGR] = -EINVAL,
121 [DLM_NOPURGED] = -EINVAL,
122 [DLM_BADARGS] = -EINVAL,
123 [DLM_VOID] = -EINVAL,
124 [DLM_NOTQUEUED] = -EAGAIN, /* Trylock failed */
125 [DLM_IVBUFLEN] = -EINVAL,
126 [DLM_CVTUNGRANT] = -EPERM,
127 [DLM_BADPARAM] = -EINVAL,
128 [DLM_VALNOTVALID] = -EINVAL,
129 [DLM_REJECTED] = -EPERM,
130 [DLM_ABORT] = -EINVAL,
131 [DLM_CANCEL] = -DLM_ECANCEL, /* Successful cancel */
132 [DLM_IVRESHANDLE] = -EINVAL,
133 [DLM_DEADLOCK] = -EDEADLK,
134 [DLM_DENIED_NOASTS] = -EINVAL,
135 [DLM_FORWARD] = -EINVAL,
136 [DLM_TIMEOUT] = -ETIMEDOUT,
137 [DLM_IVGROUPID] = -EINVAL,
138 [DLM_VERS_CONFLICT] = -EOPNOTSUPP,
139 [DLM_BAD_DEVICE_PATH] = -ENOENT,
140 [DLM_NO_DEVICE_PERMISSION] = -EPERM,
141 [DLM_NO_CONTROL_DEVICE] = -ENOENT,
142 [DLM_RECOVERING] = -ENOTCONN,
143 [DLM_MIGRATING] = -ERESTART,
144 [DLM_MAXSTATS] = -EINVAL,
145 };
146
dlm_status_to_errno(enum dlm_status status)147 static int dlm_status_to_errno(enum dlm_status status)
148 {
149 BUG_ON(status < 0 || status >= ARRAY_SIZE(status_map));
150
151 return status_map[status];
152 }
153
o2dlm_lock_ast_wrapper(void * astarg)154 static void o2dlm_lock_ast_wrapper(void *astarg)
155 {
156 struct ocfs2_dlm_lksb *lksb = astarg;
157
158 lksb->lksb_conn->cc_proto->lp_lock_ast(lksb);
159 }
160
o2dlm_blocking_ast_wrapper(void * astarg,int level)161 static void o2dlm_blocking_ast_wrapper(void *astarg, int level)
162 {
163 struct ocfs2_dlm_lksb *lksb = astarg;
164
165 lksb->lksb_conn->cc_proto->lp_blocking_ast(lksb, level);
166 }
167
o2dlm_unlock_ast_wrapper(void * astarg,enum dlm_status status)168 static void o2dlm_unlock_ast_wrapper(void *astarg, enum dlm_status status)
169 {
170 struct ocfs2_dlm_lksb *lksb = astarg;
171 int error = dlm_status_to_errno(status);
172
173 /*
174 * In o2dlm, you can get both the lock_ast() for the lock being
175 * granted and the unlock_ast() for the CANCEL failing. A
176 * successful cancel sends DLM_NORMAL here. If the
177 * lock grant happened before the cancel arrived, you get
178 * DLM_CANCELGRANT.
179 *
180 * There's no need for the double-ast. If we see DLM_CANCELGRANT,
181 * we just ignore it. We expect the lock_ast() to handle the
182 * granted lock.
183 */
184 if (status == DLM_CANCELGRANT)
185 return;
186
187 lksb->lksb_conn->cc_proto->lp_unlock_ast(lksb, error);
188 }
189
o2cb_dlm_lock(struct ocfs2_cluster_connection * conn,int mode,struct ocfs2_dlm_lksb * lksb,u32 flags,void * name,unsigned int namelen)190 static int o2cb_dlm_lock(struct ocfs2_cluster_connection *conn,
191 int mode,
192 struct ocfs2_dlm_lksb *lksb,
193 u32 flags,
194 void *name,
195 unsigned int namelen)
196 {
197 enum dlm_status status;
198 int o2dlm_mode = mode_to_o2dlm(mode);
199 int o2dlm_flags = flags_to_o2dlm(flags);
200 int ret;
201
202 status = dlmlock(conn->cc_lockspace, o2dlm_mode, &lksb->lksb_o2dlm,
203 o2dlm_flags, name, namelen,
204 o2dlm_lock_ast_wrapper, lksb,
205 o2dlm_blocking_ast_wrapper);
206 ret = dlm_status_to_errno(status);
207 return ret;
208 }
209
o2cb_dlm_unlock(struct ocfs2_cluster_connection * conn,struct ocfs2_dlm_lksb * lksb,u32 flags)210 static int o2cb_dlm_unlock(struct ocfs2_cluster_connection *conn,
211 struct ocfs2_dlm_lksb *lksb,
212 u32 flags)
213 {
214 enum dlm_status status;
215 int o2dlm_flags = flags_to_o2dlm(flags);
216 int ret;
217
218 status = dlmunlock(conn->cc_lockspace, &lksb->lksb_o2dlm,
219 o2dlm_flags, o2dlm_unlock_ast_wrapper, lksb);
220 ret = dlm_status_to_errno(status);
221 return ret;
222 }
223
o2cb_dlm_lock_status(struct ocfs2_dlm_lksb * lksb)224 static int o2cb_dlm_lock_status(struct ocfs2_dlm_lksb *lksb)
225 {
226 return dlm_status_to_errno(lksb->lksb_o2dlm.status);
227 }
228
229 /*
230 * o2dlm aways has a "valid" LVB. If the dlm loses track of the LVB
231 * contents, it will zero out the LVB. Thus the caller can always trust
232 * the contents.
233 */
o2cb_dlm_lvb_valid(struct ocfs2_dlm_lksb * lksb)234 static int o2cb_dlm_lvb_valid(struct ocfs2_dlm_lksb *lksb)
235 {
236 return 1;
237 }
238
o2cb_dlm_lvb(struct ocfs2_dlm_lksb * lksb)239 static void *o2cb_dlm_lvb(struct ocfs2_dlm_lksb *lksb)
240 {
241 return (void *)(lksb->lksb_o2dlm.lvb);
242 }
243
o2cb_dump_lksb(struct ocfs2_dlm_lksb * lksb)244 static void o2cb_dump_lksb(struct ocfs2_dlm_lksb *lksb)
245 {
246 dlm_print_one_lock(lksb->lksb_o2dlm.lockid);
247 }
248
249 /*
250 * Check if this node is heartbeating and is connected to all other
251 * heartbeating nodes.
252 */
o2cb_cluster_check(void)253 static int o2cb_cluster_check(void)
254 {
255 u8 node_num;
256 int i;
257 unsigned long hbmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
258 unsigned long netmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
259
260 node_num = o2nm_this_node();
261 if (node_num == O2NM_MAX_NODES) {
262 printk(KERN_ERR "o2cb: This node has not been configured.\n");
263 return -EINVAL;
264 }
265
266 /*
267 * o2dlm expects o2net sockets to be created. If not, then
268 * dlm_join_domain() fails with a stack of errors which are both cryptic
269 * and incomplete. The idea here is to detect upfront whether we have
270 * managed to connect to all nodes or not. If not, then list the nodes
271 * to allow the user to check the configuration (incorrect IP, firewall,
272 * etc.) Yes, this is racy. But its not the end of the world.
273 */
274 #define O2CB_MAP_STABILIZE_COUNT 60
275 for (i = 0; i < O2CB_MAP_STABILIZE_COUNT; ++i) {
276 o2hb_fill_node_map(hbmap, sizeof(hbmap));
277 if (!test_bit(node_num, hbmap)) {
278 printk(KERN_ERR "o2cb: %s heartbeat has not been "
279 "started.\n", (o2hb_global_heartbeat_active() ?
280 "Global" : "Local"));
281 return -EINVAL;
282 }
283 o2net_fill_node_map(netmap, sizeof(netmap));
284 /* Force set the current node to allow easy compare */
285 set_bit(node_num, netmap);
286 if (!memcmp(hbmap, netmap, sizeof(hbmap)))
287 return 0;
288 if (i < O2CB_MAP_STABILIZE_COUNT - 1)
289 msleep(1000);
290 }
291
292 printk(KERN_ERR "o2cb: This node could not connect to nodes:");
293 i = -1;
294 while ((i = find_next_bit(hbmap, O2NM_MAX_NODES,
295 i + 1)) < O2NM_MAX_NODES) {
296 if (!test_bit(i, netmap))
297 printk(" %u", i);
298 }
299 printk(".\n");
300
301 return -ENOTCONN;
302 }
303
304 /*
305 * Called from the dlm when it's about to evict a node. This is how the
306 * classic stack signals node death.
307 */
o2dlm_eviction_cb(int node_num,void * data)308 static void o2dlm_eviction_cb(int node_num, void *data)
309 {
310 struct ocfs2_cluster_connection *conn = data;
311
312 printk(KERN_NOTICE "o2cb: o2dlm has evicted node %d from domain %.*s\n",
313 node_num, conn->cc_namelen, conn->cc_name);
314
315 conn->cc_recovery_handler(node_num, conn->cc_recovery_data);
316 }
317
o2cb_cluster_connect(struct ocfs2_cluster_connection * conn)318 static int o2cb_cluster_connect(struct ocfs2_cluster_connection *conn)
319 {
320 int rc = 0;
321 u32 dlm_key;
322 struct dlm_ctxt *dlm;
323 struct o2dlm_private *priv;
324 struct dlm_protocol_version fs_version;
325
326 BUG_ON(conn == NULL);
327 BUG_ON(conn->cc_proto == NULL);
328
329 /* Ensure cluster stack is up and all nodes are connected */
330 rc = o2cb_cluster_check();
331 if (rc) {
332 printk(KERN_ERR "o2cb: Cluster check failed. Fix errors "
333 "before retrying.\n");
334 goto out;
335 }
336
337 priv = kzalloc(sizeof(struct o2dlm_private), GFP_KERNEL);
338 if (!priv) {
339 rc = -ENOMEM;
340 goto out_free;
341 }
342
343 /* This just fills the structure in. It is safe to pass conn. */
344 dlm_setup_eviction_cb(&priv->op_eviction_cb, o2dlm_eviction_cb,
345 conn);
346
347 conn->cc_private = priv;
348
349 /* used by the dlm code to make message headers unique, each
350 * node in this domain must agree on this. */
351 dlm_key = crc32_le(0, conn->cc_name, conn->cc_namelen);
352 fs_version.pv_major = conn->cc_version.pv_major;
353 fs_version.pv_minor = conn->cc_version.pv_minor;
354
355 dlm = dlm_register_domain(conn->cc_name, dlm_key, &fs_version);
356 if (IS_ERR(dlm)) {
357 rc = PTR_ERR(dlm);
358 mlog_errno(rc);
359 goto out_free;
360 }
361
362 conn->cc_version.pv_major = fs_version.pv_major;
363 conn->cc_version.pv_minor = fs_version.pv_minor;
364 conn->cc_lockspace = dlm;
365
366 dlm_register_eviction_cb(dlm, &priv->op_eviction_cb);
367
368 out_free:
369 if (rc)
370 kfree(conn->cc_private);
371
372 out:
373 return rc;
374 }
375
o2cb_cluster_disconnect(struct ocfs2_cluster_connection * conn)376 static int o2cb_cluster_disconnect(struct ocfs2_cluster_connection *conn)
377 {
378 struct dlm_ctxt *dlm = conn->cc_lockspace;
379 struct o2dlm_private *priv = conn->cc_private;
380
381 dlm_unregister_eviction_cb(&priv->op_eviction_cb);
382 conn->cc_private = NULL;
383 kfree(priv);
384
385 dlm_unregister_domain(dlm);
386 conn->cc_lockspace = NULL;
387
388 return 0;
389 }
390
o2cb_cluster_this_node(struct ocfs2_cluster_connection * conn,unsigned int * node)391 static int o2cb_cluster_this_node(struct ocfs2_cluster_connection *conn,
392 unsigned int *node)
393 {
394 int node_num;
395
396 node_num = o2nm_this_node();
397 if (node_num == O2NM_INVALID_NODE_NUM)
398 return -ENOENT;
399
400 if (node_num >= O2NM_MAX_NODES)
401 return -EOVERFLOW;
402
403 *node = node_num;
404 return 0;
405 }
406
407 static struct ocfs2_stack_operations o2cb_stack_ops = {
408 .connect = o2cb_cluster_connect,
409 .disconnect = o2cb_cluster_disconnect,
410 .this_node = o2cb_cluster_this_node,
411 .dlm_lock = o2cb_dlm_lock,
412 .dlm_unlock = o2cb_dlm_unlock,
413 .lock_status = o2cb_dlm_lock_status,
414 .lvb_valid = o2cb_dlm_lvb_valid,
415 .lock_lvb = o2cb_dlm_lvb,
416 .dump_lksb = o2cb_dump_lksb,
417 };
418
419 static struct ocfs2_stack_plugin o2cb_stack = {
420 .sp_name = "o2cb",
421 .sp_ops = &o2cb_stack_ops,
422 .sp_owner = THIS_MODULE,
423 };
424
o2cb_stack_init(void)425 static int __init o2cb_stack_init(void)
426 {
427 return ocfs2_stack_glue_register(&o2cb_stack);
428 }
429
o2cb_stack_exit(void)430 static void __exit o2cb_stack_exit(void)
431 {
432 ocfs2_stack_glue_unregister(&o2cb_stack);
433 }
434
435 MODULE_AUTHOR("Oracle");
436 MODULE_DESCRIPTION("ocfs2 driver for the classic o2cb stack");
437 MODULE_LICENSE("GPL");
438 module_init(o2cb_stack_init);
439 module_exit(o2cb_stack_exit);
440