Btrfs: fix space leak when we fail to reserve metadata space
[linux-2.6.git] / fs / ocfs2 / stack_o2cb.c
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/kernel.h>
21 #include <linux/crc32.h>
22 #include <linux/slab.h>
23 #include <linux/module.h>
24
25 /* Needed for AOP_TRUNCATED_PAGE in mlog_errno() */
26 #include <linux/fs.h>
27
28 #include "cluster/masklog.h"
29 #include "cluster/nodemanager.h"
30 #include "cluster/heartbeat.h"
31 #include "cluster/tcp.h"
32
33 #include "stackglue.h"
34
35 struct o2dlm_private {
36         struct dlm_eviction_cb op_eviction_cb;
37 };
38
39 static struct ocfs2_stack_plugin o2cb_stack;
40
41 /* These should be identical */
42 #if (DLM_LOCK_IV != LKM_IVMODE)
43 # error Lock modes do not match
44 #endif
45 #if (DLM_LOCK_NL != LKM_NLMODE)
46 # error Lock modes do not match
47 #endif
48 #if (DLM_LOCK_CR != LKM_CRMODE)
49 # error Lock modes do not match
50 #endif
51 #if (DLM_LOCK_CW != LKM_CWMODE)
52 # error Lock modes do not match
53 #endif
54 #if (DLM_LOCK_PR != LKM_PRMODE)
55 # error Lock modes do not match
56 #endif
57 #if (DLM_LOCK_PW != LKM_PWMODE)
58 # error Lock modes do not match
59 #endif
60 #if (DLM_LOCK_EX != LKM_EXMODE)
61 # error Lock modes do not match
62 #endif
63 static inline int mode_to_o2dlm(int mode)
64 {
65         BUG_ON(mode > LKM_MAXMODE);
66
67         return mode;
68 }
69
70 #define map_flag(_generic, _o2dlm)              \
71         if (flags & (_generic)) {               \
72                 flags &= ~(_generic);           \
73                 o2dlm_flags |= (_o2dlm);        \
74         }
75 static int flags_to_o2dlm(u32 flags)
76 {
77         int o2dlm_flags = 0;
78
79         map_flag(DLM_LKF_NOQUEUE, LKM_NOQUEUE);
80         map_flag(DLM_LKF_CANCEL, LKM_CANCEL);
81         map_flag(DLM_LKF_CONVERT, LKM_CONVERT);
82         map_flag(DLM_LKF_VALBLK, LKM_VALBLK);
83         map_flag(DLM_LKF_IVVALBLK, LKM_INVVALBLK);
84         map_flag(DLM_LKF_ORPHAN, LKM_ORPHAN);
85         map_flag(DLM_LKF_FORCEUNLOCK, LKM_FORCE);
86         map_flag(DLM_LKF_TIMEOUT, LKM_TIMEOUT);
87         map_flag(DLM_LKF_LOCAL, LKM_LOCAL);
88
89         /* map_flag() should have cleared every flag passed in */
90         BUG_ON(flags != 0);
91
92         return o2dlm_flags;
93 }
94 #undef map_flag
95
96 /*
97  * Map an o2dlm status to standard errno values.
98  *
99  * o2dlm only uses a handful of these, and returns even fewer to the
100  * caller. Still, we try to assign sane values to each error.
101  *
102  * The following value pairs have special meanings to dlmglue, thus
103  * the right hand side needs to stay unique - never duplicate the
104  * mapping elsewhere in the table!
105  *
106  * DLM_NORMAL:          0
107  * DLM_NOTQUEUED:       -EAGAIN
108  * DLM_CANCELGRANT:     -EBUSY
109  * DLM_CANCEL:          -DLM_ECANCEL
110  */
111 /* Keep in sync with dlmapi.h */
112 static int status_map[] = {
113         [DLM_NORMAL]                    = 0,            /* Success */
114         [DLM_GRANTED]                   = -EINVAL,
115         [DLM_DENIED]                    = -EACCES,
116         [DLM_DENIED_NOLOCKS]            = -EACCES,
117         [DLM_WORKING]                   = -EACCES,
118         [DLM_BLOCKED]                   = -EINVAL,
119         [DLM_BLOCKED_ORPHAN]            = -EINVAL,
120         [DLM_DENIED_GRACE_PERIOD]       = -EACCES,
121         [DLM_SYSERR]                    = -ENOMEM,      /* It is what it is */
122         [DLM_NOSUPPORT]                 = -EPROTO,
123         [DLM_CANCELGRANT]               = -EBUSY,       /* Cancel after grant */
124         [DLM_IVLOCKID]                  = -EINVAL,
125         [DLM_SYNC]                      = -EINVAL,
126         [DLM_BADTYPE]                   = -EINVAL,
127         [DLM_BADRESOURCE]               = -EINVAL,
128         [DLM_MAXHANDLES]                = -ENOMEM,
129         [DLM_NOCLINFO]                  = -EINVAL,
130         [DLM_NOLOCKMGR]                 = -EINVAL,
131         [DLM_NOPURGED]                  = -EINVAL,
132         [DLM_BADARGS]                   = -EINVAL,
133         [DLM_VOID]                      = -EINVAL,
134         [DLM_NOTQUEUED]                 = -EAGAIN,      /* Trylock failed */
135         [DLM_IVBUFLEN]                  = -EINVAL,
136         [DLM_CVTUNGRANT]                = -EPERM,
137         [DLM_BADPARAM]                  = -EINVAL,
138         [DLM_VALNOTVALID]               = -EINVAL,
139         [DLM_REJECTED]                  = -EPERM,
140         [DLM_ABORT]                     = -EINVAL,
141         [DLM_CANCEL]                    = -DLM_ECANCEL, /* Successful cancel */
142         [DLM_IVRESHANDLE]               = -EINVAL,
143         [DLM_DEADLOCK]                  = -EDEADLK,
144         [DLM_DENIED_NOASTS]             = -EINVAL,
145         [DLM_FORWARD]                   = -EINVAL,
146         [DLM_TIMEOUT]                   = -ETIMEDOUT,
147         [DLM_IVGROUPID]                 = -EINVAL,
148         [DLM_VERS_CONFLICT]             = -EOPNOTSUPP,
149         [DLM_BAD_DEVICE_PATH]           = -ENOENT,
150         [DLM_NO_DEVICE_PERMISSION]      = -EPERM,
151         [DLM_NO_CONTROL_DEVICE]         = -ENOENT,
152         [DLM_RECOVERING]                = -ENOTCONN,
153         [DLM_MIGRATING]                 = -ERESTART,
154         [DLM_MAXSTATS]                  = -EINVAL,
155 };
156
157 static int dlm_status_to_errno(enum dlm_status status)
158 {
159         BUG_ON(status < 0 || status >= ARRAY_SIZE(status_map));
160
161         return status_map[status];
162 }
163
164 static void o2dlm_lock_ast_wrapper(void *astarg)
165 {
166         struct ocfs2_dlm_lksb *lksb = astarg;
167
168         lksb->lksb_conn->cc_proto->lp_lock_ast(lksb);
169 }
170
171 static void o2dlm_blocking_ast_wrapper(void *astarg, int level)
172 {
173         struct ocfs2_dlm_lksb *lksb = astarg;
174
175         lksb->lksb_conn->cc_proto->lp_blocking_ast(lksb, level);
176 }
177
178 static void o2dlm_unlock_ast_wrapper(void *astarg, enum dlm_status status)
179 {
180         struct ocfs2_dlm_lksb *lksb = astarg;
181         int error = dlm_status_to_errno(status);
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         lksb->lksb_conn->cc_proto->lp_unlock_ast(lksb, error);
198 }
199
200 static int o2cb_dlm_lock(struct ocfs2_cluster_connection *conn,
201                          int mode,
202                          struct ocfs2_dlm_lksb *lksb,
203                          u32 flags,
204                          void *name,
205                          unsigned int namelen)
206 {
207         enum dlm_status status;
208         int o2dlm_mode = mode_to_o2dlm(mode);
209         int o2dlm_flags = flags_to_o2dlm(flags);
210         int ret;
211
212         status = dlmlock(conn->cc_lockspace, o2dlm_mode, &lksb->lksb_o2dlm,
213                          o2dlm_flags, name, namelen,
214                          o2dlm_lock_ast_wrapper, lksb,
215                          o2dlm_blocking_ast_wrapper);
216         ret = dlm_status_to_errno(status);
217         return ret;
218 }
219
220 static int o2cb_dlm_unlock(struct ocfs2_cluster_connection *conn,
221                            struct ocfs2_dlm_lksb *lksb,
222                            u32 flags)
223 {
224         enum dlm_status status;
225         int o2dlm_flags = flags_to_o2dlm(flags);
226         int ret;
227
228         status = dlmunlock(conn->cc_lockspace, &lksb->lksb_o2dlm,
229                            o2dlm_flags, o2dlm_unlock_ast_wrapper, lksb);
230         ret = dlm_status_to_errno(status);
231         return ret;
232 }
233
234 static int o2cb_dlm_lock_status(struct ocfs2_dlm_lksb *lksb)
235 {
236         return dlm_status_to_errno(lksb->lksb_o2dlm.status);
237 }
238
239 /*
240  * o2dlm aways has a "valid" LVB. If the dlm loses track of the LVB
241  * contents, it will zero out the LVB.  Thus the caller can always trust
242  * the contents.
243  */
244 static int o2cb_dlm_lvb_valid(struct ocfs2_dlm_lksb *lksb)
245 {
246         return 1;
247 }
248
249 static void *o2cb_dlm_lvb(struct ocfs2_dlm_lksb *lksb)
250 {
251         return (void *)(lksb->lksb_o2dlm.lvb);
252 }
253
254 static void o2cb_dump_lksb(struct ocfs2_dlm_lksb *lksb)
255 {
256         dlm_print_one_lock(lksb->lksb_o2dlm.lockid);
257 }
258
259 /*
260  * Check if this node is heartbeating and is connected to all other
261  * heartbeating nodes.
262  */
263 static int o2cb_cluster_check(void)
264 {
265         u8 node_num;
266         int i;
267         unsigned long hbmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
268         unsigned long netmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
269
270         node_num = o2nm_this_node();
271         if (node_num == O2NM_MAX_NODES) {
272                 printk(KERN_ERR "o2cb: This node has not been configured.\n");
273                 return -EINVAL;
274         }
275
276         /*
277          * o2dlm expects o2net sockets to be created. If not, then
278          * dlm_join_domain() fails with a stack of errors which are both cryptic
279          * and incomplete. The idea here is to detect upfront whether we have
280          * managed to connect to all nodes or not. If not, then list the nodes
281          * to allow the user to check the configuration (incorrect IP, firewall,
282          * etc.) Yes, this is racy. But its not the end of the world.
283          */
284 #define O2CB_MAP_STABILIZE_COUNT        60
285         for (i = 0; i < O2CB_MAP_STABILIZE_COUNT; ++i) {
286                 o2hb_fill_node_map(hbmap, sizeof(hbmap));
287                 if (!test_bit(node_num, hbmap)) {
288                         printk(KERN_ERR "o2cb: %s heartbeat has not been "
289                                "started.\n", (o2hb_global_heartbeat_active() ?
290                                               "Global" : "Local"));
291                         return -EINVAL;
292                 }
293                 o2net_fill_node_map(netmap, sizeof(netmap));
294                 /* Force set the current node to allow easy compare */
295                 set_bit(node_num, netmap);
296                 if (!memcmp(hbmap, netmap, sizeof(hbmap)))
297                         return 0;
298                 if (i < O2CB_MAP_STABILIZE_COUNT)
299                         msleep(1000);
300         }
301
302         printk(KERN_ERR "o2cb: This node could not connect to nodes:");
303         i = -1;
304         while ((i = find_next_bit(hbmap, O2NM_MAX_NODES,
305                                   i + 1)) < O2NM_MAX_NODES) {
306                 if (!test_bit(i, netmap))
307                         printk(" %u", i);
308         }
309         printk(".\n");
310
311         return -ENOTCONN;
312 }
313
314 /*
315  * Called from the dlm when it's about to evict a node. This is how the
316  * classic stack signals node death.
317  */
318 static void o2dlm_eviction_cb(int node_num, void *data)
319 {
320         struct ocfs2_cluster_connection *conn = data;
321
322         printk(KERN_NOTICE "o2cb: o2dlm has evicted node %d from domain %.*s\n",
323                node_num, conn->cc_namelen, conn->cc_name);
324
325         conn->cc_recovery_handler(node_num, conn->cc_recovery_data);
326 }
327
328 static int o2cb_cluster_connect(struct ocfs2_cluster_connection *conn)
329 {
330         int rc = 0;
331         u32 dlm_key;
332         struct dlm_ctxt *dlm;
333         struct o2dlm_private *priv;
334         struct dlm_protocol_version fs_version;
335
336         BUG_ON(conn == NULL);
337         BUG_ON(conn->cc_proto == NULL);
338
339         /* Ensure cluster stack is up and all nodes are connected */
340         rc = o2cb_cluster_check();
341         if (rc) {
342                 printk(KERN_ERR "o2cb: Cluster check failed. Fix errors "
343                        "before retrying.\n");
344                 goto out;
345         }
346
347         priv = kzalloc(sizeof(struct o2dlm_private), GFP_KERNEL);
348         if (!priv) {
349                 rc = -ENOMEM;
350                 goto out_free;
351         }
352
353         /* This just fills the structure in.  It is safe to pass conn. */
354         dlm_setup_eviction_cb(&priv->op_eviction_cb, o2dlm_eviction_cb,
355                               conn);
356
357         conn->cc_private = priv;
358
359         /* used by the dlm code to make message headers unique, each
360          * node in this domain must agree on this. */
361         dlm_key = crc32_le(0, conn->cc_name, conn->cc_namelen);
362         fs_version.pv_major = conn->cc_version.pv_major;
363         fs_version.pv_minor = conn->cc_version.pv_minor;
364
365         dlm = dlm_register_domain(conn->cc_name, dlm_key, &fs_version);
366         if (IS_ERR(dlm)) {
367                 rc = PTR_ERR(dlm);
368                 mlog_errno(rc);
369                 goto out_free;
370         }
371
372         conn->cc_version.pv_major = fs_version.pv_major;
373         conn->cc_version.pv_minor = fs_version.pv_minor;
374         conn->cc_lockspace = dlm;
375
376         dlm_register_eviction_cb(dlm, &priv->op_eviction_cb);
377
378 out_free:
379         if (rc && conn->cc_private)
380                 kfree(conn->cc_private);
381
382 out:
383         return rc;
384 }
385
386 static int o2cb_cluster_disconnect(struct ocfs2_cluster_connection *conn)
387 {
388         struct dlm_ctxt *dlm = conn->cc_lockspace;
389         struct o2dlm_private *priv = conn->cc_private;
390
391         dlm_unregister_eviction_cb(&priv->op_eviction_cb);
392         conn->cc_private = NULL;
393         kfree(priv);
394
395         dlm_unregister_domain(dlm);
396         conn->cc_lockspace = NULL;
397
398         return 0;
399 }
400
401 static int o2cb_cluster_this_node(unsigned int *node)
402 {
403         int node_num;
404
405         node_num = o2nm_this_node();
406         if (node_num == O2NM_INVALID_NODE_NUM)
407                 return -ENOENT;
408
409         if (node_num >= O2NM_MAX_NODES)
410                 return -EOVERFLOW;
411
412         *node = node_num;
413         return 0;
414 }
415
416 static struct ocfs2_stack_operations o2cb_stack_ops = {
417         .connect        = o2cb_cluster_connect,
418         .disconnect     = o2cb_cluster_disconnect,
419         .this_node      = o2cb_cluster_this_node,
420         .dlm_lock       = o2cb_dlm_lock,
421         .dlm_unlock     = o2cb_dlm_unlock,
422         .lock_status    = o2cb_dlm_lock_status,
423         .lvb_valid      = o2cb_dlm_lvb_valid,
424         .lock_lvb       = o2cb_dlm_lvb,
425         .dump_lksb      = o2cb_dump_lksb,
426 };
427
428 static struct ocfs2_stack_plugin o2cb_stack = {
429         .sp_name        = "o2cb",
430         .sp_ops         = &o2cb_stack_ops,
431         .sp_owner       = THIS_MODULE,
432 };
433
434 static int __init o2cb_stack_init(void)
435 {
436         return ocfs2_stack_glue_register(&o2cb_stack);
437 }
438
439 static void __exit o2cb_stack_exit(void)
440 {
441         ocfs2_stack_glue_unregister(&o2cb_stack);
442 }
443
444 MODULE_AUTHOR("Oracle");
445 MODULE_DESCRIPTION("ocfs2 driver for the classic o2cb stack");
446 MODULE_LICENSE("GPL");
447 module_init(o2cb_stack_init);
448 module_exit(o2cb_stack_exit);