1 // SPDX-License-Identifier: GPL-2.0-only
2 /* -*- mode: c; c-basic-offset: 8; -*-
3 * vim: noexpandtab sw=8 ts=8 sts=0:
4 *
5 * stack_o2cb.c
6 *
7 * Code which interfaces ocfs2 with the o2cb stack.
8 *
9 * Copyright (C) 2007 Oracle. All rights reserved.
10 */
11
12 #include <linux/kernel.h>
13 #include <linux/crc32.h>
14 #include <linux/slab.h>
15 #include <linux/module.h>
16
17 /* Needed for AOP_TRUNCATED_PAGE in mlog_errno() */
18 #include <linux/fs.h>
19
20 #include "cluster/masklog.h"
21 #include "cluster/nodemanager.h"
22 #include "cluster/heartbeat.h"
23 #include "cluster/tcp.h"
24
25 #include "stackglue.h"
26
27 struct o2dlm_private {
28 struct dlm_eviction_cb op_eviction_cb;
29 };
30
31 static struct ocfs2_stack_plugin o2cb_stack;
32
33 /* These should be identical */
34 #if (DLM_LOCK_IV != LKM_IVMODE)
35 # error Lock modes do not match
36 #endif
37 #if (DLM_LOCK_NL != LKM_NLMODE)
38 # error Lock modes do not match
39 #endif
40 #if (DLM_LOCK_CR != LKM_CRMODE)
41 # error Lock modes do not match
42 #endif
43 #if (DLM_LOCK_CW != LKM_CWMODE)
44 # error Lock modes do not match
45 #endif
46 #if (DLM_LOCK_PR != LKM_PRMODE)
47 # error Lock modes do not match
48 #endif
49 #if (DLM_LOCK_PW != LKM_PWMODE)
50 # error Lock modes do not match
51 #endif
52 #if (DLM_LOCK_EX != LKM_EXMODE)
53 # error Lock modes do not match
54 #endif
mode_to_o2dlm(int mode)55 static inline int mode_to_o2dlm(int mode)
56 {
57 BUG_ON(mode > LKM_MAXMODE);
58
59 return mode;
60 }
61
62 #define map_flag(_generic, _o2dlm) \
63 if (flags & (_generic)) { \
64 flags &= ~(_generic); \
65 o2dlm_flags |= (_o2dlm); \
66 }
flags_to_o2dlm(u32 flags)67 static int flags_to_o2dlm(u32 flags)
68 {
69 int o2dlm_flags = 0;
70
71 map_flag(DLM_LKF_NOQUEUE, LKM_NOQUEUE);
72 map_flag(DLM_LKF_CANCEL, LKM_CANCEL);
73 map_flag(DLM_LKF_CONVERT, LKM_CONVERT);
74 map_flag(DLM_LKF_VALBLK, LKM_VALBLK);
75 map_flag(DLM_LKF_IVVALBLK, LKM_INVVALBLK);
76 map_flag(DLM_LKF_ORPHAN, LKM_ORPHAN);
77 map_flag(DLM_LKF_FORCEUNLOCK, LKM_FORCE);
78 map_flag(DLM_LKF_TIMEOUT, LKM_TIMEOUT);
79 map_flag(DLM_LKF_LOCAL, LKM_LOCAL);
80
81 /* map_flag() should have cleared every flag passed in */
82 BUG_ON(flags != 0);
83
84 return o2dlm_flags;
85 }
86 #undef map_flag
87
88 /*
89 * Map an o2dlm status to standard errno values.
90 *
91 * o2dlm only uses a handful of these, and returns even fewer to the
92 * caller. Still, we try to assign sane values to each error.
93 *
94 * The following value pairs have special meanings to dlmglue, thus
95 * the right hand side needs to stay unique - never duplicate the
96 * mapping elsewhere in the table!
97 *
98 * DLM_NORMAL: 0
99 * DLM_NOTQUEUED: -EAGAIN
100 * DLM_CANCELGRANT: -EBUSY
101 * DLM_CANCEL: -DLM_ECANCEL
102 */
103 /* Keep in sync with dlmapi.h */
104 static int status_map[] = {
105 [DLM_NORMAL] = 0, /* Success */
106 [DLM_GRANTED] = -EINVAL,
107 [DLM_DENIED] = -EACCES,
108 [DLM_DENIED_NOLOCKS] = -EACCES,
109 [DLM_WORKING] = -EACCES,
110 [DLM_BLOCKED] = -EINVAL,
111 [DLM_BLOCKED_ORPHAN] = -EINVAL,
112 [DLM_DENIED_GRACE_PERIOD] = -EACCES,
113 [DLM_SYSERR] = -ENOMEM, /* It is what it is */
114 [DLM_NOSUPPORT] = -EPROTO,
115 [DLM_CANCELGRANT] = -EBUSY, /* Cancel after grant */
116 [DLM_IVLOCKID] = -EINVAL,
117 [DLM_SYNC] = -EINVAL,
118 [DLM_BADTYPE] = -EINVAL,
119 [DLM_BADRESOURCE] = -EINVAL,
120 [DLM_MAXHANDLES] = -ENOMEM,
121 [DLM_NOCLINFO] = -EINVAL,
122 [DLM_NOLOCKMGR] = -EINVAL,
123 [DLM_NOPURGED] = -EINVAL,
124 [DLM_BADARGS] = -EINVAL,
125 [DLM_VOID] = -EINVAL,
126 [DLM_NOTQUEUED] = -EAGAIN, /* Trylock failed */
127 [DLM_IVBUFLEN] = -EINVAL,
128 [DLM_CVTUNGRANT] = -EPERM,
129 [DLM_BADPARAM] = -EINVAL,
130 [DLM_VALNOTVALID] = -EINVAL,
131 [DLM_REJECTED] = -EPERM,
132 [DLM_ABORT] = -EINVAL,
133 [DLM_CANCEL] = -DLM_ECANCEL, /* Successful cancel */
134 [DLM_IVRESHANDLE] = -EINVAL,
135 [DLM_DEADLOCK] = -EDEADLK,
136 [DLM_DENIED_NOASTS] = -EINVAL,
137 [DLM_FORWARD] = -EINVAL,
138 [DLM_TIMEOUT] = -ETIMEDOUT,
139 [DLM_IVGROUPID] = -EINVAL,
140 [DLM_VERS_CONFLICT] = -EOPNOTSUPP,
141 [DLM_BAD_DEVICE_PATH] = -ENOENT,
142 [DLM_NO_DEVICE_PERMISSION] = -EPERM,
143 [DLM_NO_CONTROL_DEVICE] = -ENOENT,
144 [DLM_RECOVERING] = -ENOTCONN,
145 [DLM_MIGRATING] = -ERESTART,
146 [DLM_MAXSTATS] = -EINVAL,
147 };
148
dlm_status_to_errno(enum dlm_status status)149 static int dlm_status_to_errno(enum dlm_status status)
150 {
151 BUG_ON(status < 0 || status >= ARRAY_SIZE(status_map));
152
153 return status_map[status];
154 }
155
o2dlm_lock_ast_wrapper(void * astarg)156 static void o2dlm_lock_ast_wrapper(void *astarg)
157 {
158 struct ocfs2_dlm_lksb *lksb = astarg;
159
160 lksb->lksb_conn->cc_proto->lp_lock_ast(lksb);
161 }
162
o2dlm_blocking_ast_wrapper(void * astarg,int level)163 static void o2dlm_blocking_ast_wrapper(void *astarg, int level)
164 {
165 struct ocfs2_dlm_lksb *lksb = astarg;
166
167 lksb->lksb_conn->cc_proto->lp_blocking_ast(lksb, level);
168 }
169
o2dlm_unlock_ast_wrapper(void * astarg,enum dlm_status status)170 static void o2dlm_unlock_ast_wrapper(void *astarg, enum dlm_status status)
171 {
172 struct ocfs2_dlm_lksb *lksb = astarg;
173 int error = dlm_status_to_errno(status);
174
175 /*
176 * In o2dlm, you can get both the lock_ast() for the lock being
177 * granted and the unlock_ast() for the CANCEL failing. A
178 * successful cancel sends DLM_NORMAL here. If the
179 * lock grant happened before the cancel arrived, you get
180 * DLM_CANCELGRANT.
181 *
182 * There's no need for the double-ast. If we see DLM_CANCELGRANT,
183 * we just ignore it. We expect the lock_ast() to handle the
184 * granted lock.
185 */
186 if (status == DLM_CANCELGRANT)
187 return;
188
189 lksb->lksb_conn->cc_proto->lp_unlock_ast(lksb, error);
190 }
191
o2cb_dlm_lock(struct ocfs2_cluster_connection * conn,int mode,struct ocfs2_dlm_lksb * lksb,u32 flags,void * name,unsigned int namelen)192 static int o2cb_dlm_lock(struct ocfs2_cluster_connection *conn,
193 int mode,
194 struct ocfs2_dlm_lksb *lksb,
195 u32 flags,
196 void *name,
197 unsigned int namelen)
198 {
199 enum dlm_status status;
200 int o2dlm_mode = mode_to_o2dlm(mode);
201 int o2dlm_flags = flags_to_o2dlm(flags);
202 int ret;
203
204 status = dlmlock(conn->cc_lockspace, o2dlm_mode, &lksb->lksb_o2dlm,
205 o2dlm_flags, name, namelen,
206 o2dlm_lock_ast_wrapper, lksb,
207 o2dlm_blocking_ast_wrapper);
208 ret = dlm_status_to_errno(status);
209 return ret;
210 }
211
o2cb_dlm_unlock(struct ocfs2_cluster_connection * conn,struct ocfs2_dlm_lksb * lksb,u32 flags)212 static int o2cb_dlm_unlock(struct ocfs2_cluster_connection *conn,
213 struct ocfs2_dlm_lksb *lksb,
214 u32 flags)
215 {
216 enum dlm_status status;
217 int o2dlm_flags = flags_to_o2dlm(flags);
218 int ret;
219
220 status = dlmunlock(conn->cc_lockspace, &lksb->lksb_o2dlm,
221 o2dlm_flags, o2dlm_unlock_ast_wrapper, lksb);
222 ret = dlm_status_to_errno(status);
223 return ret;
224 }
225
o2cb_dlm_lock_status(struct ocfs2_dlm_lksb * lksb)226 static int o2cb_dlm_lock_status(struct ocfs2_dlm_lksb *lksb)
227 {
228 return dlm_status_to_errno(lksb->lksb_o2dlm.status);
229 }
230
231 /*
232 * o2dlm aways has a "valid" LVB. If the dlm loses track of the LVB
233 * contents, it will zero out the LVB. Thus the caller can always trust
234 * the contents.
235 */
o2cb_dlm_lvb_valid(struct ocfs2_dlm_lksb * lksb)236 static int o2cb_dlm_lvb_valid(struct ocfs2_dlm_lksb *lksb)
237 {
238 return 1;
239 }
240
o2cb_dlm_lvb(struct ocfs2_dlm_lksb * lksb)241 static void *o2cb_dlm_lvb(struct ocfs2_dlm_lksb *lksb)
242 {
243 return (void *)(lksb->lksb_o2dlm.lvb);
244 }
245
o2cb_dump_lksb(struct ocfs2_dlm_lksb * lksb)246 static void o2cb_dump_lksb(struct ocfs2_dlm_lksb *lksb)
247 {
248 dlm_print_one_lock(lksb->lksb_o2dlm.lockid);
249 }
250
251 /*
252 * Check if this node is heartbeating and is connected to all other
253 * heartbeating nodes.
254 */
o2cb_cluster_check(void)255 static int o2cb_cluster_check(void)
256 {
257 u8 node_num;
258 int i;
259 unsigned long hbmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
260 unsigned long netmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
261
262 node_num = o2nm_this_node();
263 if (node_num == O2NM_MAX_NODES) {
264 printk(KERN_ERR "o2cb: This node has not been configured.\n");
265 return -EINVAL;
266 }
267
268 /*
269 * o2dlm expects o2net sockets to be created. If not, then
270 * dlm_join_domain() fails with a stack of errors which are both cryptic
271 * and incomplete. The idea here is to detect upfront whether we have
272 * managed to connect to all nodes or not. If not, then list the nodes
273 * to allow the user to check the configuration (incorrect IP, firewall,
274 * etc.) Yes, this is racy. But its not the end of the world.
275 */
276 #define O2CB_MAP_STABILIZE_COUNT 60
277 for (i = 0; i < O2CB_MAP_STABILIZE_COUNT; ++i) {
278 o2hb_fill_node_map(hbmap, sizeof(hbmap));
279 if (!test_bit(node_num, hbmap)) {
280 printk(KERN_ERR "o2cb: %s heartbeat has not been "
281 "started.\n", (o2hb_global_heartbeat_active() ?
282 "Global" : "Local"));
283 return -EINVAL;
284 }
285 o2net_fill_node_map(netmap, sizeof(netmap));
286 /* Force set the current node to allow easy compare */
287 set_bit(node_num, netmap);
288 if (!memcmp(hbmap, netmap, sizeof(hbmap)))
289 return 0;
290 if (i < O2CB_MAP_STABILIZE_COUNT - 1)
291 msleep(1000);
292 }
293
294 printk(KERN_ERR "o2cb: This node could not connect to nodes:");
295 i = -1;
296 while ((i = find_next_bit(hbmap, O2NM_MAX_NODES,
297 i + 1)) < O2NM_MAX_NODES) {
298 if (!test_bit(i, netmap))
299 printk(" %u", i);
300 }
301 printk(".\n");
302
303 return -ENOTCONN;
304 }
305
306 /*
307 * Called from the dlm when it's about to evict a node. This is how the
308 * classic stack signals node death.
309 */
o2dlm_eviction_cb(int node_num,void * data)310 static void o2dlm_eviction_cb(int node_num, void *data)
311 {
312 struct ocfs2_cluster_connection *conn = data;
313
314 printk(KERN_NOTICE "o2cb: o2dlm has evicted node %d from domain %.*s\n",
315 node_num, conn->cc_namelen, conn->cc_name);
316
317 conn->cc_recovery_handler(node_num, conn->cc_recovery_data);
318 }
319
o2cb_cluster_connect(struct ocfs2_cluster_connection * conn)320 static int o2cb_cluster_connect(struct ocfs2_cluster_connection *conn)
321 {
322 int rc = 0;
323 u32 dlm_key;
324 struct dlm_ctxt *dlm;
325 struct o2dlm_private *priv;
326 struct dlm_protocol_version fs_version;
327
328 BUG_ON(conn == NULL);
329 BUG_ON(conn->cc_proto == NULL);
330
331 /* Ensure cluster stack is up and all nodes are connected */
332 rc = o2cb_cluster_check();
333 if (rc) {
334 printk(KERN_ERR "o2cb: Cluster check failed. Fix errors "
335 "before retrying.\n");
336 goto out;
337 }
338
339 priv = kzalloc(sizeof(struct o2dlm_private), GFP_KERNEL);
340 if (!priv) {
341 rc = -ENOMEM;
342 goto out_free;
343 }
344
345 /* This just fills the structure in. It is safe to pass conn. */
346 dlm_setup_eviction_cb(&priv->op_eviction_cb, o2dlm_eviction_cb,
347 conn);
348
349 conn->cc_private = priv;
350
351 /* used by the dlm code to make message headers unique, each
352 * node in this domain must agree on this. */
353 dlm_key = crc32_le(0, conn->cc_name, conn->cc_namelen);
354 fs_version.pv_major = conn->cc_version.pv_major;
355 fs_version.pv_minor = conn->cc_version.pv_minor;
356
357 dlm = dlm_register_domain(conn->cc_name, dlm_key, &fs_version);
358 if (IS_ERR(dlm)) {
359 rc = PTR_ERR(dlm);
360 mlog_errno(rc);
361 goto out_free;
362 }
363
364 conn->cc_version.pv_major = fs_version.pv_major;
365 conn->cc_version.pv_minor = fs_version.pv_minor;
366 conn->cc_lockspace = dlm;
367
368 dlm_register_eviction_cb(dlm, &priv->op_eviction_cb);
369
370 out_free:
371 if (rc)
372 kfree(conn->cc_private);
373
374 out:
375 return rc;
376 }
377
o2cb_cluster_disconnect(struct ocfs2_cluster_connection * conn)378 static int o2cb_cluster_disconnect(struct ocfs2_cluster_connection *conn)
379 {
380 struct dlm_ctxt *dlm = conn->cc_lockspace;
381 struct o2dlm_private *priv = conn->cc_private;
382
383 dlm_unregister_eviction_cb(&priv->op_eviction_cb);
384 conn->cc_private = NULL;
385 kfree(priv);
386
387 dlm_unregister_domain(dlm);
388 conn->cc_lockspace = NULL;
389
390 return 0;
391 }
392
o2cb_cluster_this_node(struct ocfs2_cluster_connection * conn,unsigned int * node)393 static int o2cb_cluster_this_node(struct ocfs2_cluster_connection *conn,
394 unsigned int *node)
395 {
396 int node_num;
397
398 node_num = o2nm_this_node();
399 if (node_num == O2NM_INVALID_NODE_NUM)
400 return -ENOENT;
401
402 if (node_num >= O2NM_MAX_NODES)
403 return -EOVERFLOW;
404
405 *node = node_num;
406 return 0;
407 }
408
409 static struct ocfs2_stack_operations o2cb_stack_ops = {
410 .connect = o2cb_cluster_connect,
411 .disconnect = o2cb_cluster_disconnect,
412 .this_node = o2cb_cluster_this_node,
413 .dlm_lock = o2cb_dlm_lock,
414 .dlm_unlock = o2cb_dlm_unlock,
415 .lock_status = o2cb_dlm_lock_status,
416 .lvb_valid = o2cb_dlm_lvb_valid,
417 .lock_lvb = o2cb_dlm_lvb,
418 .dump_lksb = o2cb_dump_lksb,
419 };
420
421 static struct ocfs2_stack_plugin o2cb_stack = {
422 .sp_name = "o2cb",
423 .sp_ops = &o2cb_stack_ops,
424 .sp_owner = THIS_MODULE,
425 };
426
o2cb_stack_init(void)427 static int __init o2cb_stack_init(void)
428 {
429 return ocfs2_stack_glue_register(&o2cb_stack);
430 }
431
o2cb_stack_exit(void)432 static void __exit o2cb_stack_exit(void)
433 {
434 ocfs2_stack_glue_unregister(&o2cb_stack);
435 }
436
437 MODULE_AUTHOR("Oracle");
438 MODULE_DESCRIPTION("ocfs2 driver for the classic o2cb stack");
439 MODULE_LICENSE("GPL");
440 module_init(o2cb_stack_init);
441 module_exit(o2cb_stack_exit);
442