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/module.h>
  23
  24 /* Needed for AOP_TRUNCATED_PAGE in mlog_errno() */
  25 #include <linux/fs.h>
  26
  27 #include "cluster/masklog.h"
  28 #include "cluster/nodemanager.h"
  29 #include "cluster/heartbeat.h"
  30
  31 #include "stackglue.h"
  32
  33 struct o2dlm_private {
  34         struct dlm_eviction_cb op_eviction_cb;
  35 };
  36
  37 static struct ocfs2_stack_plugin o2cb_stack;
  38
  39 /* These should be identical */
  40 #if (DLM_LOCK_IV != LKM_IVMODE)
  41 # error Lock modes do not match
  42 #endif
  43 #if (DLM_LOCK_NL != LKM_NLMODE)
  44 # error Lock modes do not match
  45 #endif
  46 #if (DLM_LOCK_CR != LKM_CRMODE)
  47 # error Lock modes do not match
  48 #endif
  49 #if (DLM_LOCK_CW != LKM_CWMODE)
  50 # error Lock modes do not match
  51 #endif
  52 #if (DLM_LOCK_PR != LKM_PRMODE)
  53 # error Lock modes do not match
  54 #endif
  55 #if (DLM_LOCK_PW != LKM_PWMODE)
  56 # error Lock modes do not match
  57 #endif
  58 #if (DLM_LOCK_EX != LKM_EXMODE)
  59 # error Lock modes do not match
  60 #endif
  61 static inline int mode_to_o2dlm(int mode)
  62 {
  63         BUG_ON(mode > LKM_MAXMODE);
  64
  65         return mode;
  66 }
  67
  68 #define map_flag(_generic, _o2dlm)              \
  69         if (flags & (_generic)) {               \
  70                 flags &= ~(_generic);           \
  71                 o2dlm_flags |= (_o2dlm);        \
  72         }
  73 static int flags_to_o2dlm(u32 flags)
  74 {
  75         int o2dlm_flags = 0;
  76
  77         map_flag(DLM_LKF_NOQUEUE, LKM_NOQUEUE);
  78         map_flag(DLM_LKF_CANCEL, LKM_CANCEL);
  79         map_flag(DLM_LKF_CONVERT, LKM_CONVERT);
  80         map_flag(DLM_LKF_VALBLK, LKM_VALBLK);
  81         map_flag(DLM_LKF_IVVALBLK, LKM_INVVALBLK);
  82         map_flag(DLM_LKF_ORPHAN, LKM_ORPHAN);
  83         map_flag(DLM_LKF_FORCEUNLOCK, LKM_FORCE);
  84         map_flag(DLM_LKF_TIMEOUT, LKM_TIMEOUT);
  85         map_flag(DLM_LKF_LOCAL, LKM_LOCAL);
  86
  87         /* map_flag() should have cleared every flag passed in */
  88         BUG_ON(flags != 0);
  89
  90         return o2dlm_flags;
  91 }
  92 #undef map_flag
  93
  94 /*
  95  * Map an o2dlm status to standard errno values.
  96  *
  97  * o2dlm only uses a handful of these, and returns even fewer to the
  98  * caller. Still, we try to assign sane values to each error.
  99  *
 100  * The following value pairs have special meanings to dlmglue, thus
 101  * the right hand side needs to stay unique - never duplicate the
 102  * mapping elsewhere in the table!
 103  *
 104  * DLM_NORMAL:          0
 105  * DLM_NOTQUEUED:       -EAGAIN
 106  * DLM_CANCELGRANT:     -EBUSY
 107  * DLM_CANCEL:          -DLM_ECANCEL
 108  */
 109 /* Keep in sync with dlmapi.h */
 110 static int status_map[] = {
 111         [DLM_NORMAL]                    = 0,            /* Success */
 112         [DLM_GRANTED]                   = -EINVAL,
 113         [DLM_DENIED]                    = -EACCES,
 114         [DLM_DENIED_NOLOCKS]            = -EACCES,
 115         [DLM_WORKING]                   = -EACCES,
 116         [DLM_BLOCKED]                   = -EINVAL,
 117         [DLM_BLOCKED_ORPHAN]            = -EINVAL,
 118         [DLM_DENIED_GRACE_PERIOD]       = -EACCES,
 119         [DLM_SYSERR]                    = -ENOMEM,      /* It is what it is */
 120         [DLM_NOSUPPORT]                 = -EPROTO,
 121         [DLM_CANCELGRANT]               = -EBUSY,       /* Cancel after grant */
 122         [DLM_IVLOCKID]                  = -EINVAL,
 123         [DLM_SYNC]                      = -EINVAL,
 124         [DLM_BADTYPE]                   = -EINVAL,
 125         [DLM_BADRESOURCE]               = -EINVAL,
 126         [DLM_MAXHANDLES]                = -ENOMEM,
 127         [DLM_NOCLINFO]                  = -EINVAL,
 128         [DLM_NOLOCKMGR]                 = -EINVAL,
 129         [DLM_NOPURGED]                  = -EINVAL,
 130         [DLM_BADARGS]                   = -EINVAL,
 131         [DLM_VOID]                      = -EINVAL,
 132         [DLM_NOTQUEUED]                 = -EAGAIN,      /* Trylock failed */
 133         [DLM_IVBUFLEN]                  = -EINVAL,
 134         [DLM_CVTUNGRANT]                = -EPERM,
 135         [DLM_BADPARAM]                  = -EINVAL,
 136         [DLM_VALNOTVALID]               = -EINVAL,
 137         [DLM_REJECTED]                  = -EPERM,
 138         [DLM_ABORT]                     = -EINVAL,
 139         [DLM_CANCEL]                    = -DLM_ECANCEL, /* Successful cancel */
 140         [DLM_IVRESHANDLE]               = -EINVAL,
 141         [DLM_DEADLOCK]                  = -EDEADLK,
 142         [DLM_DENIED_NOASTS]             = -EINVAL,
 143         [DLM_FORWARD]                   = -EINVAL,
 144         [DLM_TIMEOUT]                   = -ETIMEDOUT,
 145         [DLM_IVGROUPID]                 = -EINVAL,
 146         [DLM_VERS_CONFLICT]             = -EOPNOTSUPP,
 147         [DLM_BAD_DEVICE_PATH]           = -ENOENT,
 148         [DLM_NO_DEVICE_PERMISSION]      = -EPERM,
 149         [DLM_NO_CONTROL_DEVICE]         = -ENOENT,
 150         [DLM_RECOVERING]                = -ENOTCONN,
 151         [DLM_MIGRATING]                 = -ERESTART,
 152         [DLM_MAXSTATS]                  = -EINVAL,
 153 };
 154
 155 static int dlm_status_to_errno(enum dlm_status status)
 156 {
 157         BUG_ON(status < 0 || status >= ARRAY_SIZE(status_map));
 158
 159         return status_map[status];
 160 }
 161
 162 static void o2dlm_lock_ast_wrapper(void *astarg)
 163 {
 164         BUG_ON(o2cb_stack.sp_proto == NULL);
 165
 166         o2cb_stack.sp_proto->lp_lock_ast(astarg);
 167 }
 168
 169 static void o2dlm_blocking_ast_wrapper(void *astarg, int level)
 170 {
 171         BUG_ON(o2cb_stack.sp_proto == NULL);
 172
 173         o2cb_stack.sp_proto->lp_blocking_ast(astarg, level);
 174 }
 175
 176 static void o2dlm_unlock_ast_wrapper(void *astarg, enum dlm_status status)
 177 {
 178         int error = dlm_status_to_errno(status);
 179
 180         BUG_ON(o2cb_stack.sp_proto == NULL);
 181
 182         /*
 183          * In o2dlm, you can get both the lock_ast() for the lock being
 184          * granted and the unlock_ast() for the CANCEL failing.  A
 185          * successful cancel sends DLM_NORMAL here.  If the
 186          * lock grant happened before the cancel arrived, you get
 187          * DLM_CANCELGRANT.
 188          *
 189          * There's no need for the double-ast.  If we see DLM_CANCELGRANT,
 190          * we just ignore it.  We expect the lock_ast() to handle the
 191          * granted lock.
 192          */
 193         if (status == DLM_CANCELGRANT)
 194                 return;
 195
 196         o2cb_stack.sp_proto->lp_unlock_ast(astarg, error);
 197 }
 198
 199 static int o2cb_dlm_lock(struct ocfs2_cluster_connection *conn,
 200                          int mode,
 201                          union ocfs2_dlm_lksb *lksb,
 202                          u32 flags,
 203                          void *name,
 204                          unsigned int namelen,
 205                          void *astarg)
 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, astarg,
 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                            union ocfs2_dlm_lksb *lksb,
 222                            u32 flags,
 223                            void *astarg)
 224 {
 225         enum dlm_status status;
 226         int o2dlm_flags = flags_to_o2dlm(flags);
 227         int ret;
 228
 229         status = dlmunlock(conn->cc_lockspace, &lksb->lksb_o2dlm,
 230                            o2dlm_flags, o2dlm_unlock_ast_wrapper, astarg);
 231         ret = dlm_status_to_errno(status);
 232         return ret;
 233 }
 234
 235 static int o2cb_dlm_lock_status(union ocfs2_dlm_lksb *lksb)
 236 {
 237         return dlm_status_to_errno(lksb->lksb_o2dlm.status);
 238 }
 239
 240 /*
 241  * o2dlm aways has a "valid" LVB. If the dlm loses track of the LVB
 242  * contents, it will zero out the LVB.  Thus the caller can always trust
 243  * the contents.
 244  */
 245 static int o2cb_dlm_lvb_valid(union ocfs2_dlm_lksb *lksb)
 246 {
 247         return 1;
 248 }
 249
 250 static void *o2cb_dlm_lvb(union ocfs2_dlm_lksb *lksb)
 251 {
 252         return (void *)(lksb->lksb_o2dlm.lvb);
 253 }
 254
 255 static void o2cb_dump_lksb(union ocfs2_dlm_lksb *lksb)
 256 {
 257         dlm_print_one_lock(lksb->lksb_o2dlm.lockid);
 258 }
 259
 260 /*
 261  * Called from the dlm when it's about to evict a node. This is how the
 262  * classic stack signals node death.
 263  */
 264 static void o2dlm_eviction_cb(int node_num, void *data)
 265 {
 266         struct ocfs2_cluster_connection *conn = data;
 267
 268         mlog(ML_NOTICE, "o2dlm has evicted node %d from group %.*s\n",
 269              node_num, conn->cc_namelen, conn->cc_name);
 270
 271         conn->cc_recovery_handler(node_num, conn->cc_recovery_data);
 272 }
 273
 274 static int o2cb_cluster_connect(struct ocfs2_cluster_connection *conn)
 275 {
 276         int rc = 0;
 277         u32 dlm_key;
 278         struct dlm_ctxt *dlm;
 279         struct o2dlm_private *priv;
 280         struct dlm_protocol_version dlm_version;
 281
 282         BUG_ON(conn == NULL);
 283         BUG_ON(o2cb_stack.sp_proto == NULL);
 284
 285         /* for now we only have one cluster/node, make sure we see it
 286          * in the heartbeat universe */
 287         if (!o2hb_check_local_node_heartbeating()) {
 288                 rc = -EINVAL;
 289                 goto out;
 290         }
 291
 292         priv = kzalloc(sizeof(struct o2dlm_private), GFP_KERNEL);
 293         if (!priv) {
 294                 rc = -ENOMEM;
 295                 goto out_free;
 296         }
 297
 298         /* This just fills the structure in.  It is safe to pass conn. */
 299         dlm_setup_eviction_cb(&priv->op_eviction_cb, o2dlm_eviction_cb,
 300                               conn);
 301
 302         conn->cc_private = priv;
 303
 304         /* used by the dlm code to make message headers unique, each
 305          * node in this domain must agree on this. */
 306         dlm_key = crc32_le(0, conn->cc_name, conn->cc_namelen);
 307         dlm_version.pv_major = conn->cc_version.pv_major;
 308         dlm_version.pv_minor = conn->cc_version.pv_minor;
 309
 310         dlm = dlm_register_domain(conn->cc_name, dlm_key, &dlm_version);
 311         if (IS_ERR(dlm)) {
 312                 rc = PTR_ERR(dlm);
 313                 mlog_errno(rc);
 314                 goto out_free;
 315         }
 316
 317         conn->cc_version.pv_major = dlm_version.pv_major;
 318         conn->cc_version.pv_minor = dlm_version.pv_minor;
 319         conn->cc_lockspace = dlm;
 320
 321         dlm_register_eviction_cb(dlm, &priv->op_eviction_cb);
 322
 323 out_free:
 324         if (rc && conn->cc_private)
 325                 kfree(conn->cc_private);
 326
 327 out:
 328         return rc;
 329 }
 330
 331 static int o2cb_cluster_disconnect(struct ocfs2_cluster_connection *conn)
 332 {
 333         struct dlm_ctxt *dlm = conn->cc_lockspace;
 334         struct o2dlm_private *priv = conn->cc_private;
 335
 336         dlm_unregister_eviction_cb(&priv->op_eviction_cb);
 337         conn->cc_private = NULL;
 338         kfree(priv);
 339
 340         dlm_unregister_domain(dlm);
 341         conn->cc_lockspace = NULL;
 342
 343         return 0;
 344 }
 345
 346 static int o2cb_cluster_this_node(unsigned int *node)
 347 {
 348         int node_num;
 349
 350         node_num = o2nm_this_node();
 351         if (node_num == O2NM_INVALID_NODE_NUM)
 352                 return -ENOENT;
 353
 354         if (node_num >= O2NM_MAX_NODES)
 355                 return -EOVERFLOW;
 356
 357         *node = node_num;
 358         return 0;
 359 }
 360
 361 static struct ocfs2_stack_operations o2cb_stack_ops = {
 362         .connect        = o2cb_cluster_connect,
 363         .disconnect     = o2cb_cluster_disconnect,
 364         .this_node      = o2cb_cluster_this_node,
 365         .dlm_lock       = o2cb_dlm_lock,
 366         .dlm_unlock     = o2cb_dlm_unlock,
 367         .lock_status    = o2cb_dlm_lock_status,
 368         .lvb_valid      = o2cb_dlm_lvb_valid,
 369         .lock_lvb       = o2cb_dlm_lvb,
 370         .dump_lksb      = o2cb_dump_lksb,
 371 };
 372
 373 static struct ocfs2_stack_plugin o2cb_stack = {
 374         .sp_name        = "o2cb",
 375         .sp_ops         = &o2cb_stack_ops,
 376         .sp_owner       = THIS_MODULE,
 377 };
 378
 379 static int __init o2cb_stack_init(void)
 380 {
 381         return ocfs2_stack_glue_register(&o2cb_stack);
 382 }
 383
 384 static void __exit o2cb_stack_exit(void)
 385 {
 386         ocfs2_stack_glue_unregister(&o2cb_stack);
 387 }
 388
 389 MODULE_AUTHOR("Oracle");
 390 MODULE_DESCRIPTION("ocfs2 driver for the classic o2cb stack");
 391 MODULE_LICENSE("GPL");
 392 module_init(o2cb_stack_init);
 393 module_exit(o2cb_stack_exit);