1 // SPDX-License-Identifier: GPL-2.0-only
2 /* -*- mode: c; c-basic-offset: 8; -*-
3 * vim: noexpandtab sw=8 ts=8 sts=0:
7 * Code which interfaces ocfs2 with the o2cb stack.
9 * Copyright (C) 2007 Oracle. All rights reserved.
12 #include <linux/kernel.h>
13 #include <linux/crc32.h>
14 #include <linux/slab.h>
15 #include <linux/module.h>
17 /* Needed for AOP_TRUNCATED_PAGE in mlog_errno() */
20 #include "cluster/masklog.h"
21 #include "cluster/nodemanager.h"
22 #include "cluster/heartbeat.h"
23 #include "cluster/tcp.h"
25 #include "stackglue.h"
27 struct o2dlm_private
{
28 struct dlm_eviction_cb op_eviction_cb
;
31 static struct ocfs2_stack_plugin o2cb_stack
;
33 /* These should be identical */
34 #if (DLM_LOCK_IV != LKM_IVMODE)
35 # error Lock modes do not match
37 #if (DLM_LOCK_NL != LKM_NLMODE)
38 # error Lock modes do not match
40 #if (DLM_LOCK_CR != LKM_CRMODE)
41 # error Lock modes do not match
43 #if (DLM_LOCK_CW != LKM_CWMODE)
44 # error Lock modes do not match
46 #if (DLM_LOCK_PR != LKM_PRMODE)
47 # error Lock modes do not match
49 #if (DLM_LOCK_PW != LKM_PWMODE)
50 # error Lock modes do not match
52 #if (DLM_LOCK_EX != LKM_EXMODE)
53 # error Lock modes do not match
55 static inline int mode_to_o2dlm(int mode
)
57 BUG_ON(mode
> LKM_MAXMODE
);
62 #define map_flag(_generic, _o2dlm) \
63 if (flags & (_generic)) { \
64 flags &= ~(_generic); \
65 o2dlm_flags |= (_o2dlm); \
67 static int flags_to_o2dlm(u32 flags
)
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
);
81 /* map_flag() should have cleared every flag passed in */
89 * Map an o2dlm status to standard errno values.
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.
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!
99 * DLM_NOTQUEUED: -EAGAIN
100 * DLM_CANCELGRANT: -EBUSY
101 * DLM_CANCEL: -DLM_ECANCEL
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
,
149 static int dlm_status_to_errno(enum dlm_status status
)
151 BUG_ON(status
< 0 || status
>= ARRAY_SIZE(status_map
));
153 return status_map
[status
];
156 static void o2dlm_lock_ast_wrapper(void *astarg
)
158 struct ocfs2_dlm_lksb
*lksb
= astarg
;
160 lksb
->lksb_conn
->cc_proto
->lp_lock_ast(lksb
);
163 static void o2dlm_blocking_ast_wrapper(void *astarg
, int level
)
165 struct ocfs2_dlm_lksb
*lksb
= astarg
;
167 lksb
->lksb_conn
->cc_proto
->lp_blocking_ast(lksb
, level
);
170 static void o2dlm_unlock_ast_wrapper(void *astarg
, enum dlm_status status
)
172 struct ocfs2_dlm_lksb
*lksb
= astarg
;
173 int error
= dlm_status_to_errno(status
);
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
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
186 if (status
== DLM_CANCELGRANT
)
189 lksb
->lksb_conn
->cc_proto
->lp_unlock_ast(lksb
, error
);
192 static int o2cb_dlm_lock(struct ocfs2_cluster_connection
*conn
,
194 struct ocfs2_dlm_lksb
*lksb
,
197 unsigned int namelen
)
199 enum dlm_status status
;
200 int o2dlm_mode
= mode_to_o2dlm(mode
);
201 int o2dlm_flags
= flags_to_o2dlm(flags
);
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
);
212 static int o2cb_dlm_unlock(struct ocfs2_cluster_connection
*conn
,
213 struct ocfs2_dlm_lksb
*lksb
,
216 enum dlm_status status
;
217 int o2dlm_flags
= flags_to_o2dlm(flags
);
220 status
= dlmunlock(conn
->cc_lockspace
, &lksb
->lksb_o2dlm
,
221 o2dlm_flags
, o2dlm_unlock_ast_wrapper
, lksb
);
222 ret
= dlm_status_to_errno(status
);
226 static int o2cb_dlm_lock_status(struct ocfs2_dlm_lksb
*lksb
)
228 return dlm_status_to_errno(lksb
->lksb_o2dlm
.status
);
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
236 static int o2cb_dlm_lvb_valid(struct ocfs2_dlm_lksb
*lksb
)
241 static void *o2cb_dlm_lvb(struct ocfs2_dlm_lksb
*lksb
)
243 return (void *)(lksb
->lksb_o2dlm
.lvb
);
246 static void o2cb_dump_lksb(struct ocfs2_dlm_lksb
*lksb
)
248 dlm_print_one_lock(lksb
->lksb_o2dlm
.lockid
);
252 * Check if this node is heartbeating and is connected to all other
253 * heartbeating nodes.
255 static int o2cb_cluster_check(void)
259 unsigned long hbmap
[BITS_TO_LONGS(O2NM_MAX_NODES
)];
260 unsigned long netmap
[BITS_TO_LONGS(O2NM_MAX_NODES
)];
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");
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.
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"));
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
)))
290 if (i
< O2CB_MAP_STABILIZE_COUNT
- 1)
294 printk(KERN_ERR
"o2cb: This node could not connect to nodes:");
296 while ((i
= find_next_bit(hbmap
, O2NM_MAX_NODES
,
297 i
+ 1)) < O2NM_MAX_NODES
) {
298 if (!test_bit(i
, netmap
))
307 * Called from the dlm when it's about to evict a node. This is how the
308 * classic stack signals node death.
310 static void o2dlm_eviction_cb(int node_num
, void *data
)
312 struct ocfs2_cluster_connection
*conn
= data
;
314 printk(KERN_NOTICE
"o2cb: o2dlm has evicted node %d from domain %.*s\n",
315 node_num
, conn
->cc_namelen
, conn
->cc_name
);
317 conn
->cc_recovery_handler(node_num
, conn
->cc_recovery_data
);
320 static int o2cb_cluster_connect(struct ocfs2_cluster_connection
*conn
)
324 struct dlm_ctxt
*dlm
;
325 struct o2dlm_private
*priv
;
326 struct dlm_protocol_version fs_version
;
328 BUG_ON(conn
== NULL
);
329 BUG_ON(conn
->cc_proto
== NULL
);
331 /* Ensure cluster stack is up and all nodes are connected */
332 rc
= o2cb_cluster_check();
334 printk(KERN_ERR
"o2cb: Cluster check failed. Fix errors "
335 "before retrying.\n");
339 priv
= kzalloc(sizeof(struct o2dlm_private
), GFP_KERNEL
);
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
,
349 conn
->cc_private
= priv
;
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
;
357 dlm
= dlm_register_domain(conn
->cc_name
, dlm_key
, &fs_version
);
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
;
368 dlm_register_eviction_cb(dlm
, &priv
->op_eviction_cb
);
372 kfree(conn
->cc_private
);
378 static int o2cb_cluster_disconnect(struct ocfs2_cluster_connection
*conn
)
380 struct dlm_ctxt
*dlm
= conn
->cc_lockspace
;
381 struct o2dlm_private
*priv
= conn
->cc_private
;
383 dlm_unregister_eviction_cb(&priv
->op_eviction_cb
);
384 conn
->cc_private
= NULL
;
387 dlm_unregister_domain(dlm
);
388 conn
->cc_lockspace
= NULL
;
393 static int o2cb_cluster_this_node(struct ocfs2_cluster_connection
*conn
,
398 node_num
= o2nm_this_node();
399 if (node_num
== O2NM_INVALID_NODE_NUM
)
402 if (node_num
>= O2NM_MAX_NODES
)
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
,
421 static struct ocfs2_stack_plugin o2cb_stack
= {
423 .sp_ops
= &o2cb_stack_ops
,
424 .sp_owner
= THIS_MODULE
,
427 static int __init
o2cb_stack_init(void)
429 return ocfs2_stack_glue_register(&o2cb_stack
);
432 static void __exit
o2cb_stack_exit(void)
434 ocfs2_stack_glue_unregister(&o2cb_stack
);
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
);