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)
 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);